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

   1 /*****************************************************************************
   2  * encoder.c: top-level encoder functions
   3  *****************************************************************************
   4  * Copyright (C) 2003-2011 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 <math.h>
  29
  30 #include "common/common.h"
  31
  32 #include "set.h"
  33 #include "analyse.h"
  34 #include "ratecontrol.h"
  35 #include "macroblock.h"
  36 #include "me.h"
  37
  38 #if HAVE_VISUALIZE
  39 #include "common/visualize.h"
  40 #endif
  41
  42 //#define DEBUG_MB_TYPE
  43
  44 #define bs_write_ue bs_write_ue_big
  45
  46 static int x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
  47                                    x264_nal_t **pp_nal, int *pi_nal,
  48                                    x264_picture_t *pic_out );
  49
  50 /****************************************************************************
  51  *
  52  ******************************* x264 libs **********************************
  53  *
  54  ****************************************************************************/
  55 static double x264_psnr( double sqe, double size )
  56 {
  57     double mse = sqe / (PIXEL_MAX*PIXEL_MAX * size);
  58     if( mse <= 0.0000000001 ) /* Max 100dB */
  59         return 100;
  60
  61     return -10.0 * log10( mse );
  62 }
  63
  64 static double x264_ssim( double ssim )
  65 {
  66     return -10.0 * log10( 1 - ssim );
  67 }
  68
  69 static void x264_frame_dump( x264_t *h )
  70 {
  71     FILE *f = fopen( h->param.psz_dump_yuv, "r+b" );
  72     if( !f )
  73         return;
  74     /* Write the frame in display order */
  75     fseek( f, (uint64_t)h->fdec->i_frame * h->param.i_height * h->param.i_width * 3/2 * sizeof(pixel), SEEK_SET );
  76     for( int y = 0; y < h->param.i_height; y++ )
  77         fwrite( &h->fdec->plane[0][y*h->fdec->i_stride[0]], sizeof(pixel), h->param.i_width, f );
  78     int cw = h->param.i_width>>1;
  79     int ch = h->param.i_height>>1;
  80     pixel *planeu = x264_malloc( (cw*ch*2+32)*sizeof(pixel) );
  81     pixel *planev = planeu + cw*ch + 16;
  82     h->mc.plane_copy_deinterleave( planeu, cw, planev, cw, h->fdec->plane[1], h->fdec->i_stride[1], cw, ch );
  83     fwrite( planeu, 1, cw*ch*sizeof(pixel), f );
  84     fwrite( planev, 1, cw*ch*sizeof(pixel), f );
  85     x264_free( planeu );
  86     fclose( f );
  87 }
  88
  89
  90 /* Fill "default" values */
  91 static void x264_slice_header_init( x264_t *h, x264_slice_header_t *sh,
  92                                     x264_sps_t *sps, x264_pps_t *pps,
  93                                     int i_idr_pic_id, int i_frame, int i_qp )
  94 {
  95     x264_param_t *param = &h->param;
  96
  97     /* First we fill all fields */
  98     sh->sps = sps;
  99     sh->pps = pps;
 100
 101     sh->i_first_mb  = 0;
 102     sh->i_last_mb   = h->mb.i_mb_count - 1;
 103     sh->i_pps_id    = pps->i_id;
 104
 105     sh->i_frame_num = i_frame;
 106
 107     sh->b_mbaff = h->param.b_interlaced;
 108     sh->b_field_pic = 0;    /* no field support for now */
 109     sh->b_bottom_field = 0; /* not yet used */
 110
 111     sh->i_idr_pic_id = i_idr_pic_id;
 112
 113     /* poc stuff, fixed later */
 114     sh->i_poc = 0;
 115     sh->i_delta_poc_bottom = 0;
 116     sh->i_delta_poc[0] = 0;
 117     sh->i_delta_poc[1] = 0;
 118
 119     sh->i_redundant_pic_cnt = 0;
 120
 121     h->mb.b_direct_auto_write = h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO
 122                                 && h->param.i_bframe
 123                                 && ( h->param.rc.b_stat_write || !h->param.rc.b_stat_read );
 124
 125     if( !h->mb.b_direct_auto_read && sh->i_type == SLICE_TYPE_B )
 126     {
 127         if( h->fref[1][0]->i_poc_l0ref0 == h->fref[0][0]->i_poc )
 128         {
 129             if( h->mb.b_direct_auto_write )
 130                 sh->b_direct_spatial_mv_pred = ( h->stat.i_direct_score[1] > h->stat.i_direct_score[0] );
 131             else
 132                 sh->b_direct_spatial_mv_pred = ( param->analyse.i_direct_mv_pred == X264_DIRECT_PRED_SPATIAL );
 133         }
 134         else
 135         {
 136             h->mb.b_direct_auto_write = 0;
 137             sh->b_direct_spatial_mv_pred = 1;
 138         }
 139     }
 140     /* else b_direct_spatial_mv_pred was read from the 2pass statsfile */
 141
 142     sh->b_num_ref_idx_override = 0;
 143     sh->i_num_ref_idx_l0_active = 1;
 144     sh->i_num_ref_idx_l1_active = 1;
 145
 146     sh->b_ref_pic_list_reordering[0] = h->b_ref_reorder[0];
 147     sh->b_ref_pic_list_reordering[1] = h->b_ref_reorder[1];
 148
 149     /* If the ref list isn't in the default order, construct reordering header */
 150     for( int list = 0; list < 2; list++ )
 151     {
 152         if( sh->b_ref_pic_list_reordering[list] )
 153         {
 154             int pred_frame_num = i_frame;
 155             for( int i = 0; i < h->i_ref[list]; i++ )
 156             {
 157                 int diff = h->fref[list][i]->i_frame_num - pred_frame_num;
 158                 sh->ref_pic_list_order[list][i].idc = ( diff > 0 );
 159                 sh->ref_pic_list_order[list][i].arg = (abs(diff) - 1) & ((1 << sps->i_log2_max_frame_num) - 1);
 160                 pred_frame_num = h->fref[list][i]->i_frame_num;
 161             }
 162         }
 163     }
 164
 165     sh->i_cabac_init_idc = param->i_cabac_init_idc;
 166
 167     sh->i_qp = SPEC_QP(i_qp);
 168     sh->i_qp_delta = sh->i_qp - pps->i_pic_init_qp;
 169     sh->b_sp_for_swidth = 0;
 170     sh->i_qs_delta = 0;
 171
 172     int deblock_thresh = i_qp + 2 * X264_MIN(param->i_deblocking_filter_alphac0, param->i_deblocking_filter_beta);
 173     /* If effective qp <= 15, deblocking would have no effect anyway */
 174     if( param->b_deblocking_filter && (h->mb.b_variable_qp || 15 < deblock_thresh ) )
 175         sh->i_disable_deblocking_filter_idc = param->b_sliced_threads ? 2 : 0;
 176     else
 177         sh->i_disable_deblocking_filter_idc = 1;
 178     sh->i_alpha_c0_offset = param->i_deblocking_filter_alphac0 << 1;
 179     sh->i_beta_offset = param->i_deblocking_filter_beta << 1;
 180 }
 181
 182 static void x264_slice_header_write( bs_t *s, x264_slice_header_t *sh, int i_nal_ref_idc )
 183 {
 184     if( sh->b_mbaff )
 185     {
 186         assert( sh->i_first_mb % (2*sh->sps->i_mb_width) == 0 );
 187         bs_write_ue( s, sh->i_first_mb >> 1 );
 188     }
 189     else
 190         bs_write_ue( s, sh->i_first_mb );
 191
 192     bs_write_ue( s, sh->i_type + 5 );   /* same type things */
 193     bs_write_ue( s, sh->i_pps_id );
 194     bs_write( s, sh->sps->i_log2_max_frame_num, sh->i_frame_num & ((1<<sh->sps->i_log2_max_frame_num)-1) );
 195
 196     if( !sh->sps->b_frame_mbs_only )
 197     {
 198         bs_write1( s, sh->b_field_pic );
 199         if( sh->b_field_pic )
 200             bs_write1( s, sh->b_bottom_field );
 201     }
 202
 203     if( sh->i_idr_pic_id >= 0 ) /* NAL IDR */
 204         bs_write_ue( s, sh->i_idr_pic_id );
 205
 206     if( sh->sps->i_poc_type == 0 )
 207     {
 208         bs_write( s, sh->sps->i_log2_max_poc_lsb, sh->i_poc & ((1<<sh->sps->i_log2_max_poc_lsb)-1) );
 209         if( sh->pps->b_pic_order && !sh->b_field_pic )
 210             bs_write_se( s, sh->i_delta_poc_bottom );
 211     }
 212     else if( sh->sps->i_poc_type == 1 && !sh->sps->b_delta_pic_order_always_zero )
 213     {
 214         bs_write_se( s, sh->i_delta_poc[0] );
 215         if( sh->pps->b_pic_order && !sh->b_field_pic )
 216             bs_write_se( s, sh->i_delta_poc[1] );
 217     }
 218
 219     if( sh->pps->b_redundant_pic_cnt )
 220         bs_write_ue( s, sh->i_redundant_pic_cnt );
 221
 222     if( sh->i_type == SLICE_TYPE_B )
 223         bs_write1( s, sh->b_direct_spatial_mv_pred );
 224
 225     if( sh->i_type == SLICE_TYPE_P || sh->i_type == SLICE_TYPE_B )
 226     {
 227         bs_write1( s, sh->b_num_ref_idx_override );
 228         if( sh->b_num_ref_idx_override )
 229         {
 230             bs_write_ue( s, sh->i_num_ref_idx_l0_active - 1 );
 231             if( sh->i_type == SLICE_TYPE_B )
 232                 bs_write_ue( s, sh->i_num_ref_idx_l1_active - 1 );
 233         }
 234     }
 235
 236     /* ref pic list reordering */
 237     if( sh->i_type != SLICE_TYPE_I )
 238     {
 239         bs_write1( s, sh->b_ref_pic_list_reordering[0] );
 240         if( sh->b_ref_pic_list_reordering[0] )
 241         {
 242             for( int i = 0; i < sh->i_num_ref_idx_l0_active; i++ )
 243             {
 244                 bs_write_ue( s, sh->ref_pic_list_order[0][i].idc );
 245                 bs_write_ue( s, sh->ref_pic_list_order[0][i].arg );
 246             }
 247             bs_write_ue( s, 3 );
 248         }
 249     }
 250     if( sh->i_type == SLICE_TYPE_B )
 251     {
 252         bs_write1( s, sh->b_ref_pic_list_reordering[1] );
 253         if( sh->b_ref_pic_list_reordering[1] )
 254         {
 255             for( int i = 0; i < sh->i_num_ref_idx_l1_active; i++ )
 256             {
 257                 bs_write_ue( s, sh->ref_pic_list_order[1][i].idc );
 258                 bs_write_ue( s, sh->ref_pic_list_order[1][i].arg );
 259             }
 260             bs_write_ue( s, 3 );
 261         }
 262     }
 263
 264     if( sh->pps->b_weighted_pred && sh->i_type == SLICE_TYPE_P )
 265     {
 266         /* pred_weight_table() */
 267         bs_write_ue( s, sh->weight[0][0].i_denom );
 268         bs_write_ue( s, sh->weight[0][1].i_denom );
 269         for( int i = 0; i < sh->i_num_ref_idx_l0_active; i++ )
 270         {
 271             int luma_weight_l0_flag = !!sh->weight[i][0].weightfn;
 272             int chroma_weight_l0_flag = !!sh->weight[i][1].weightfn || !!sh->weight[i][2].weightfn;
 273             bs_write1( s, luma_weight_l0_flag );
 274             if( luma_weight_l0_flag )
 275             {
 276                 bs_write_se( s, sh->weight[i][0].i_scale );
 277                 bs_write_se( s, sh->weight[i][0].i_offset );
 278             }
 279             bs_write1( s, chroma_weight_l0_flag );
 280             if( chroma_weight_l0_flag )
 281             {
 282                 for( int j = 1; j < 3; j++ )
 283                 {
 284                     bs_write_se( s, sh->weight[i][j].i_scale );
 285                     bs_write_se( s, sh->weight[i][j].i_offset );
 286                 }
 287             }
 288         }
 289     }
 290     else if( sh->pps->b_weighted_bipred == 1 && sh->i_type == SLICE_TYPE_B )
 291     {
 292       /* TODO */
 293     }
 294
 295     if( i_nal_ref_idc != 0 )
 296     {
 297         if( sh->i_idr_pic_id >= 0 )
 298         {
 299             bs_write1( s, 0 );  /* no output of prior pics flag */
 300             bs_write1( s, 0 );  /* long term reference flag */
 301         }
 302         else
 303         {
 304             bs_write1( s, sh->i_mmco_command_count > 0 ); /* adaptive_ref_pic_marking_mode_flag */
 305             if( sh->i_mmco_command_count > 0 )
 306             {
 307                 for( int i = 0; i < sh->i_mmco_command_count; i++ )
 308                 {
 309                     bs_write_ue( s, 1 ); /* mark short term ref as unused */
 310                     bs_write_ue( s, sh->mmco[i].i_difference_of_pic_nums - 1 );
 311                 }
 312                 bs_write_ue( s, 0 ); /* end command list */
 313             }
 314         }
 315     }
 316
 317     if( sh->pps->b_cabac && sh->i_type != SLICE_TYPE_I )
 318         bs_write_ue( s, sh->i_cabac_init_idc );
 319
 320     bs_write_se( s, sh->i_qp_delta );      /* slice qp delta */
 321
 322     if( sh->pps->b_deblocking_filter_control )
 323     {
 324         bs_write_ue( s, sh->i_disable_deblocking_filter_idc );
 325         if( sh->i_disable_deblocking_filter_idc != 1 )
 326         {
 327             bs_write_se( s, sh->i_alpha_c0_offset >> 1 );
 328             bs_write_se( s, sh->i_beta_offset >> 1 );
 329         }
 330     }
 331 }
 332
 333 /* If we are within a reasonable distance of the end of the memory allocated for the bitstream, */
 334 /* reallocate, adding an arbitrary amount of space (100 kilobytes). */
 335 static int x264_bitstream_check_buffer( x264_t *h )
 336 {
 337     uint8_t *bs_bak = h->out.p_bitstream;
 338     if( (h->param.b_cabac && (h->cabac.p_end - h->cabac.p < 2500)) ||
 339         (h->out.bs.p_end - h->out.bs.p < 2500) )
 340     {
 341         h->out.i_bitstream += 100000;
 342         CHECKED_MALLOC( h->out.p_bitstream, h->out.i_bitstream );
 343         h->mc.memcpy_aligned( h->out.p_bitstream, bs_bak, (h->out.i_bitstream - 100000) & ~15 );
 344         intptr_t delta = h->out.p_bitstream - bs_bak;
 345
 346         h->out.bs.p_start += delta;
 347         h->out.bs.p += delta;
 348         h->out.bs.p_end = h->out.p_bitstream + h->out.i_bitstream;
 349
 350         h->cabac.p_start += delta;
 351         h->cabac.p += delta;
 352         h->cabac.p_end = h->out.p_bitstream + h->out.i_bitstream;
 353
 354         for( int i = 0; i <= h->out.i_nal; i++ )
 355             h->out.nal[i].p_payload += delta;
 356         x264_free( bs_bak );
 357     }
 358     return 0;
 359 fail:
 360     x264_free( bs_bak );
 361     return -1;
 362 }
 363
 364 #if HAVE_THREAD
 365 static void x264_encoder_thread_init( x264_t *h )
 366 {
 367     if( h->param.i_sync_lookahead )
 368         x264_lower_thread_priority( 10 );
 369
 370 #if HAVE_MMX
 371     /* Misalign mask has to be set separately for each thread. */
 372     if( h->param.cpu&X264_CPU_SSE_MISALIGN )
 373         x264_cpu_mask_misalign_sse();
 374 #endif
 375 }
 376 #endif
 377
 378 /****************************************************************************
 379  *
 380  ****************************************************************************
 381  ****************************** External API*********************************
 382  ****************************************************************************
 383  *
 384  ****************************************************************************/
 385
 386 static int x264_validate_parameters( x264_t *h )
 387 {
 388 #if HAVE_MMX
 389 #ifdef __SSE__
 390     if( !(x264_cpu_detect() & X264_CPU_SSE) )
 391     {
 392         x264_log( h, X264_LOG_ERROR, "your cpu does not support SSE1, but x264 was compiled with asm support\n");
 393 #else
 394     if( !(x264_cpu_detect() & X264_CPU_MMXEXT) )
 395     {
 396         x264_log( h, X264_LOG_ERROR, "your cpu does not support MMXEXT, but x264 was compiled with asm support\n");
 397 #endif
 398         x264_log( h, X264_LOG_ERROR, "to run x264, recompile without asm support (configure --disable-asm)\n");
 399         return -1;
 400     }
 401 #endif
 402     if( h->param.i_width <= 0 || h->param.i_height <= 0 )
 403     {
 404         x264_log( h, X264_LOG_ERROR, "invalid width x height (%dx%d)\n",
 405                   h->param.i_width, h->param.i_height );
 406         return -1;
 407     }
 408
 409     if( h->param.i_width % 2 || h->param.i_height % 2 )
 410     {
 411         x264_log( h, X264_LOG_ERROR, "width or height not divisible by 2 (%dx%d)\n",
 412                   h->param.i_width, h->param.i_height );
 413         return -1;
 414     }
 415     int i_csp = h->param.i_csp & X264_CSP_MASK;
 416     if( i_csp <= X264_CSP_NONE || i_csp >= X264_CSP_MAX )
 417     {
 418         x264_log( h, X264_LOG_ERROR, "invalid CSP (only I420/YV12/NV12 supported)\n" );
 419         return -1;
 420     }
 421
 422     if( (h->param.crop_rect.i_left + h->param.crop_rect.i_right ) >= h->param.i_width ||
 423         (h->param.crop_rect.i_top  + h->param.crop_rect.i_bottom) >= h->param.i_height )
 424     {
 425         x264_log( h, X264_LOG_ERROR, "invalid crop-rect %u,%u,%u,%u\n", h->param.crop_rect.i_left,
 426                   h->param.crop_rect.i_top, h->param.crop_rect.i_right,  h->param.crop_rect.i_bottom );
 427         return -1;
 428     }
 429
 430     if( h->param.i_threads == X264_THREADS_AUTO )
 431         h->param.i_threads = x264_cpu_num_processors() * (h->param.b_sliced_threads?2:3)/2;
 432     h->param.i_threads = x264_clip3( h->param.i_threads, 1, X264_THREAD_MAX );
 433     if( h->param.i_threads > 1 )
 434     {
 435 #if !HAVE_THREAD
 436         x264_log( h, X264_LOG_WARNING, "not compiled with thread support!\n");
 437         h->param.i_threads = 1;
 438 #endif
 439         /* Avoid absurdly small thread slices as they can reduce performance
 440          * and VBV compliance.  Capped at an arbitrary 4 rows per thread. */
 441         if( h->param.b_sliced_threads )
 442         {
 443             int max_threads = (h->param.i_height+15)/16 / 4;
 444             h->param.i_threads = X264_MIN( h->param.i_threads, max_threads );
 445         }
 446     }
 447     else
 448         h->param.b_sliced_threads = 0;
 449     h->i_thread_frames = h->param.b_sliced_threads ? 1 : h->param.i_threads;
 450     if( h->i_thread_frames > 1 )
 451         h->param.nalu_process = NULL;
 452
 453     h->param.i_keyint_max = x264_clip3( h->param.i_keyint_max, 1, X264_KEYINT_MAX_INFINITE );
 454     if( h->param.i_keyint_max == 1 )
 455     {
 456         h->param.b_intra_refresh = 0;
 457         h->param.analyse.i_weighted_pred = 0;
 458     }
 459
 460     if( h->param.b_interlaced )
 461     {
 462         if( h->param.analyse.i_me_method >= X264_ME_ESA )
 463         {
 464             x264_log( h, X264_LOG_WARNING, "interlace + me=esa is not implemented\n" );
 465             h->param.analyse.i_me_method = X264_ME_UMH;
 466         }
 467         if( h->param.analyse.i_weighted_pred > 0 )
 468         {
 469             x264_log( h, X264_LOG_WARNING, "interlace + weightp is not implemented\n" );
 470             h->param.analyse.i_weighted_pred = X264_WEIGHTP_NONE;
 471         }
 472     }
 473
 474     h->param.i_frame_packing = x264_clip3( h->param.i_frame_packing, -1, 5 );
 475
 476     /* Detect default ffmpeg settings and terminate with an error. */
 477     {
 478         int score = 0;
 479         score += h->param.analyse.i_me_range == 0;
 480         score += h->param.rc.i_qp_step == 3;
 481         score += h->param.i_keyint_max == 12;
 482         score += h->param.rc.i_qp_min == 2;
 483         score += h->param.rc.i_qp_max == 31;
 484         score += h->param.rc.f_qcompress == 0.5;
 485         score += fabs(h->param.rc.f_ip_factor - 1.25) < 0.01;
 486         score += fabs(h->param.rc.f_pb_factor - 1.25) < 0.01;
 487         score += h->param.analyse.inter == 0 && h->param.analyse.i_subpel_refine == 8;
 488         if( score >= 5 )
 489         {
 490             x264_log( h, X264_LOG_ERROR, "broken ffmpeg default settings detected\n" );
 491             x264_log( h, X264_LOG_ERROR, "use an encoding preset (e.g. -vpre medium)\n" );
 492             x264_log( h, X264_LOG_ERROR, "preset usage: -vpre <speed> -vpre <profile>\n" );
 493             x264_log( h, X264_LOG_ERROR, "speed presets are listed in x264 --help\n" );
 494             x264_log( h, X264_LOG_ERROR, "profile is optional; x264 defaults to high\n" );
 495             return -1;
 496         }
 497     }
 498
 499     if( h->param.rc.i_rc_method < 0 || h->param.rc.i_rc_method > 2 )
 500     {
 501         x264_log( h, X264_LOG_ERROR, "no ratecontrol method specified\n" );
 502         return -1;
 503     }
 504     h->param.rc.f_rf_constant = x264_clip3f( h->param.rc.f_rf_constant, -QP_BD_OFFSET, 51 );
 505     h->param.rc.i_qp_constant = x264_clip3( h->param.rc.i_qp_constant, 0, QP_MAX );
 506     if( h->param.rc.i_rc_method == X264_RC_CRF )
 507     {
 508         h->param.rc.i_qp_constant = h->param.rc.f_rf_constant + QP_BD_OFFSET;
 509         h->param.rc.i_bitrate = 0;
 510     }
 511     if( (h->param.rc.i_rc_method == X264_RC_CQP || h->param.rc.i_rc_method == X264_RC_CRF)
 512         && h->param.rc.i_qp_constant == 0 )
 513     {
 514         h->mb.b_lossless = 1;
 515         h->param.i_cqm_preset = X264_CQM_FLAT;
 516         h->param.psz_cqm_file = NULL;
 517         h->param.rc.i_rc_method = X264_RC_CQP;
 518         h->param.rc.f_ip_factor = 1;
 519         h->param.rc.f_pb_factor = 1;
 520         h->param.analyse.b_psnr = 0;
 521         h->param.analyse.b_ssim = 0;
 522         h->param.analyse.i_chroma_qp_offset = 0;
 523         h->param.analyse.i_trellis = 0;
 524         h->param.analyse.b_fast_pskip = 0;
 525         h->param.analyse.i_noise_reduction = 0;
 526         h->param.analyse.b_psy = 0;
 527         h->param.i_bframe = 0;
 528         /* 8x8dct is not useful without RD in CAVLC lossless */
 529         if( !h->param.b_cabac && h->param.analyse.i_subpel_refine < 6 )
 530             h->param.analyse.b_transform_8x8 = 0;
 531     }
 532     if( h->param.rc.i_rc_method == X264_RC_CQP )
 533     {
 534         float qp_p = h->param.rc.i_qp_constant;
 535         float qp_i = qp_p - 6*log2f( h->param.rc.f_ip_factor );
 536         float qp_b = qp_p + 6*log2f( h->param.rc.f_pb_factor );
 537         h->param.rc.i_qp_min = x264_clip3( (int)(X264_MIN3( qp_p, qp_i, qp_b )), 0, QP_MAX );
 538         h->param.rc.i_qp_max = x264_clip3( (int)(X264_MAX3( qp_p, qp_i, qp_b ) + .999), 0, QP_MAX );
 539         h->param.rc.i_aq_mode = 0;
 540         h->param.rc.b_mb_tree = 0;
 541     }
 542     h->param.rc.i_qp_max = x264_clip3( h->param.rc.i_qp_max, 0, QP_MAX );
 543     h->param.rc.i_qp_min = x264_clip3( h->param.rc.i_qp_min, 0, h->param.rc.i_qp_max );
 544     if( h->param.rc.i_vbv_buffer_size )
 545     {
 546         if( h->param.rc.i_rc_method == X264_RC_CQP )
 547         {
 548             x264_log( h, X264_LOG_WARNING, "VBV is incompatible with constant QP, ignored.\n" );
 549             h->param.rc.i_vbv_max_bitrate = 0;
 550             h->param.rc.i_vbv_buffer_size = 0;
 551         }
 552         else if( h->param.rc.i_vbv_max_bitrate == 0 )
 553         {
 554             if( h->param.rc.i_rc_method == X264_RC_ABR )
 555             {
 556                 x264_log( h, X264_LOG_WARNING, "VBV maxrate unspecified, assuming CBR\n" );
 557                 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate;
 558             }
 559             else
 560             {
 561                 x264_log( h, X264_LOG_WARNING, "VBV bufsize set but maxrate unspecified, ignored\n" );
 562                 h->param.rc.i_vbv_buffer_size = 0;
 563             }
 564         }
 565         else if( h->param.rc.i_vbv_max_bitrate < h->param.rc.i_bitrate &&
 566                  h->param.rc.i_rc_method == X264_RC_ABR )
 567         {
 568             x264_log( h, X264_LOG_WARNING, "max bitrate less than average bitrate, assuming CBR\n" );
 569             h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate;
 570         }
 571     }
 572     else if( h->param.rc.i_vbv_max_bitrate )
 573     {
 574         x264_log( h, X264_LOG_WARNING, "VBV maxrate specified, but no bufsize, ignored\n" );
 575         h->param.rc.i_vbv_max_bitrate = 0;
 576     }
 577
 578     if( h->param.b_interlaced && h->param.i_slice_max_size )
 579     {
 580         x264_log( h, X264_LOG_WARNING, "interlaced + slice-max-size is not implemented\n" );
 581         h->param.i_slice_max_size = 0;
 582     }
 583     if( h->param.b_interlaced && h->param.i_slice_max_mbs )
 584     {
 585         x264_log( h, X264_LOG_WARNING, "interlaced + slice-max-mbs is not implemented\n" );
 586         h->param.i_slice_max_mbs = 0;
 587     }
 588     int max_slices = (h->param.i_height+((16<<h->param.b_interlaced)-1))/(16<<h->param.b_interlaced);
 589     if( h->param.b_sliced_threads )
 590         h->param.i_slice_count = x264_clip3( h->param.i_threads, 0, max_slices );
 591     else
 592     {
 593         h->param.i_slice_count = x264_clip3( h->param.i_slice_count, 0, max_slices );
 594         h->param.i_slice_max_size = X264_MAX( h->param.i_slice_max_size, 0 );
 595         h->param.i_slice_max_mbs = X264_MAX( h->param.i_slice_max_mbs, 0 );
 596         if( h->param.i_slice_max_mbs || h->param.i_slice_max_size )
 597             h->param.i_slice_count = 0;
 598     }
 599
 600     h->param.i_frame_reference = x264_clip3( h->param.i_frame_reference, 1, X264_REF_MAX );
 601     h->param.i_dpb_size = x264_clip3( h->param.i_dpb_size, 1, X264_REF_MAX );
 602     if( h->param.i_scenecut_threshold < 0 )
 603         h->param.i_scenecut_threshold = 0;
 604     if( !h->param.analyse.i_subpel_refine && h->param.analyse.i_direct_mv_pred > X264_DIRECT_PRED_SPATIAL )
 605     {
 606         x264_log( h, X264_LOG_WARNING, "subme=0 + direct=temporal is not supported\n" );
 607         h->param.analyse.i_direct_mv_pred = X264_DIRECT_PRED_SPATIAL;
 608     }
 609     h->param.i_bframe = x264_clip3( h->param.i_bframe, 0, X264_MIN( X264_BFRAME_MAX, h->param.i_keyint_max-1 ) );
 610     h->param.i_open_gop = x264_clip3( h->param.i_open_gop, X264_OPEN_GOP_NONE, X264_OPEN_GOP_BLURAY );
 611     h->param.i_bframe_bias = x264_clip3( h->param.i_bframe_bias, -90, 100 );
 612     if( h->param.i_bframe <= 1 )
 613         h->param.i_bframe_pyramid = X264_B_PYRAMID_NONE;
 614     h->param.i_bframe_pyramid = x264_clip3( h->param.i_bframe_pyramid, X264_B_PYRAMID_NONE, X264_B_PYRAMID_NORMAL );
 615     if( !h->param.i_bframe )
 616     {
 617         h->param.i_bframe_adaptive = X264_B_ADAPT_NONE;
 618         h->param.analyse.i_direct_mv_pred = 0;
 619         h->param.analyse.b_weighted_bipred = 0;
 620         h->param.i_open_gop = X264_OPEN_GOP_NONE;
 621     }
 622     if( h->param.b_intra_refresh && h->param.i_bframe_pyramid == X264_B_PYRAMID_NORMAL )
 623     {
 624         x264_log( h, X264_LOG_WARNING, "b-pyramid normal + intra-refresh is not supported\n" );
 625         h->param.i_bframe_pyramid = X264_B_PYRAMID_STRICT;
 626     }
 627     if( h->param.b_intra_refresh && (h->param.i_frame_reference > 1 || h->param.i_dpb_size > 1) )
 628     {
 629         x264_log( h, X264_LOG_WARNING, "ref > 1 + intra-refresh is not supported\n" );
 630         h->param.i_frame_reference = 1;
 631         h->param.i_dpb_size = 1;
 632     }
 633     if( h->param.b_intra_refresh && h->param.i_open_gop )
 634     {
 635         x264_log( h, X264_LOG_WARNING, "intra-refresh is not compatible with open-gop\n" );
 636         h->param.i_open_gop = X264_OPEN_GOP_NONE;
 637     }
 638     float fps = h->param.i_fps_num > 0 && h->param.i_fps_den > 0 ? (float) h->param.i_fps_num / h->param.i_fps_den : 25.0;
 639     if( h->param.i_keyint_min == X264_KEYINT_MIN_AUTO )
 640         h->param.i_keyint_min = X264_MIN( h->param.i_keyint_max / 10, fps );
 641     h->param.i_keyint_min = x264_clip3( h->param.i_keyint_min, 1, h->param.i_keyint_max/2+1 );
 642     h->param.rc.i_lookahead = x264_clip3( h->param.rc.i_lookahead, 0, X264_LOOKAHEAD_MAX );
 643     {
 644         int maxrate = X264_MAX( h->param.rc.i_vbv_max_bitrate, h->param.rc.i_bitrate );
 645         float bufsize = maxrate ? (float)h->param.rc.i_vbv_buffer_size / maxrate : 0;
 646         h->param.rc.i_lookahead = X264_MIN( h->param.rc.i_lookahead, X264_MAX( h->param.i_keyint_max, bufsize*fps ) );
 647     }
 648
 649     if( !h->param.i_timebase_num || !h->param.i_timebase_den || !(h->param.b_vfr_input || h->param.b_pulldown) )
 650     {
 651         h->param.i_timebase_num = h->param.i_fps_den;
 652         h->param.i_timebase_den = h->param.i_fps_num;
 653     }
 654
 655     h->param.rc.f_qcompress = x264_clip3f( h->param.rc.f_qcompress, 0.0, 1.0 );
 656     if( h->param.i_keyint_max == 1 || h->param.rc.f_qcompress == 1 )
 657         h->param.rc.b_mb_tree = 0;
 658     if( (!h->param.b_intra_refresh && h->param.i_keyint_max != X264_KEYINT_MAX_INFINITE) &&
 659         !h->param.rc.i_lookahead && h->param.rc.b_mb_tree )
 660     {
 661         x264_log( h, X264_LOG_WARNING, "lookaheadless mb-tree requires intra refresh or infinite keyint\n" );
 662         h->param.rc.b_mb_tree = 0;
 663     }
 664     if( h->param.rc.b_stat_read )
 665         h->param.rc.i_lookahead = 0;
 666 #if HAVE_THREAD
 667     if( h->param.i_sync_lookahead < 0 )
 668         h->param.i_sync_lookahead = h->param.i_bframe + 1;
 669     h->param.i_sync_lookahead = X264_MIN( h->param.i_sync_lookahead, X264_LOOKAHEAD_MAX );
 670     if( h->param.rc.b_stat_read || h->i_thread_frames == 1 )
 671         h->param.i_sync_lookahead = 0;
 672 #else
 673     h->param.i_sync_lookahead = 0;
 674 #endif
 675
 676     h->param.i_deblocking_filter_alphac0 = x264_clip3( h->param.i_deblocking_filter_alphac0, -6, 6 );
 677     h->param.i_deblocking_filter_beta    = x264_clip3( h->param.i_deblocking_filter_beta, -6, 6 );
 678     h->param.analyse.i_luma_deadzone[0] = x264_clip3( h->param.analyse.i_luma_deadzone[0], 0, 32 );
 679     h->param.analyse.i_luma_deadzone[1] = x264_clip3( h->param.analyse.i_luma_deadzone[1], 0, 32 );
 680
 681     h->param.i_cabac_init_idc = x264_clip3( h->param.i_cabac_init_idc, 0, 2 );
 682
 683     if( h->param.i_cqm_preset < X264_CQM_FLAT || h->param.i_cqm_preset > X264_CQM_CUSTOM )
 684         h->param.i_cqm_preset = X264_CQM_FLAT;
 685
 686     if( h->param.analyse.i_me_method < X264_ME_DIA ||
 687         h->param.analyse.i_me_method > X264_ME_TESA )
 688         h->param.analyse.i_me_method = X264_ME_HEX;
 689     if( h->param.analyse.i_me_range < 4 )
 690         h->param.analyse.i_me_range = 4;
 691     if( h->param.analyse.i_me_range > 16 && h->param.analyse.i_me_method <= X264_ME_HEX )
 692         h->param.analyse.i_me_range = 16;
 693     if( h->param.analyse.i_me_method == X264_ME_TESA &&
 694         (h->mb.b_lossless || h->param.analyse.i_subpel_refine <= 1) )
 695         h->param.analyse.i_me_method = X264_ME_ESA;
 696     h->param.analyse.i_subpel_refine = x264_clip3( h->param.analyse.i_subpel_refine, 0, 10 );
 697     h->param.analyse.b_mixed_references = h->param.analyse.b_mixed_references && h->param.i_frame_reference > 1;
 698     h->param.analyse.inter &= X264_ANALYSE_PSUB16x16|X264_ANALYSE_PSUB8x8|X264_ANALYSE_BSUB16x16|
 699                               X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
 700     h->param.analyse.intra &= X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
 701     if( !(h->param.analyse.inter & X264_ANALYSE_PSUB16x16) )
 702         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
 703     if( !h->param.analyse.b_transform_8x8 )
 704     {
 705         h->param.analyse.inter &= ~X264_ANALYSE_I8x8;
 706         h->param.analyse.intra &= ~X264_ANALYSE_I8x8;
 707     }
 708     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -12, 12);
 709     h->param.analyse.i_trellis = x264_clip3( h->param.analyse.i_trellis, 0, 2 );
 710     if( !h->param.analyse.b_psy )
 711     {
 712         h->param.analyse.f_psy_rd = 0;
 713         h->param.analyse.f_psy_trellis = 0;
 714     }
 715     if( !h->param.analyse.i_trellis )
 716         h->param.analyse.f_psy_trellis = 0;
 717     h->param.analyse.f_psy_rd = x264_clip3f( h->param.analyse.f_psy_rd, 0, 10 );
 718     h->param.analyse.f_psy_trellis = x264_clip3f( h->param.analyse.f_psy_trellis, 0, 10 );
 719     if( h->param.analyse.i_subpel_refine < 6 )
 720         h->param.analyse.f_psy_rd = 0;
 721     h->mb.i_psy_rd = FIX8( h->param.analyse.f_psy_rd );
 722     /* Psy RDO increases overall quantizers to improve the quality of luma--this indirectly hurts chroma quality */
 723     /* so we lower the chroma QP offset to compensate */
 724     /* This can be triggered repeatedly on multiple calls to parameter_validate, but since encoding
 725      * uses the pps chroma qp offset not the param chroma qp offset, this is not a problem. */
 726     if( h->mb.i_psy_rd )
 727         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_rd < 0.25 ? 1 : 2;
 728     h->mb.i_psy_trellis = FIX8( h->param.analyse.f_psy_trellis / 4 );
 729     /* Psy trellis has a similar effect. */
 730     if( h->mb.i_psy_trellis )
 731         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_trellis < 0.25 ? 1 : 2;
 732     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -12, 12);
 733     h->param.rc.i_aq_mode = x264_clip3( h->param.rc.i_aq_mode, 0, 2 );
 734     h->param.rc.f_aq_strength = x264_clip3f( h->param.rc.f_aq_strength, 0, 3 );
 735     if( h->param.rc.f_aq_strength == 0 )
 736         h->param.rc.i_aq_mode = 0;
 737     /* MB-tree requires AQ to be on, even if the strength is zero. */
 738     if( !h->param.rc.i_aq_mode && h->param.rc.b_mb_tree )
 739     {
 740         h->param.rc.i_aq_mode = 1;
 741         h->param.rc.f_aq_strength = 0;
 742     }
 743     h->param.analyse.i_noise_reduction = x264_clip3( h->param.analyse.i_noise_reduction, 0, 1<<16 );
 744     if( h->param.analyse.i_subpel_refine == 10 && (h->param.analyse.i_trellis != 2 || !h->param.rc.i_aq_mode) )
 745         h->param.analyse.i_subpel_refine = 9;
 746
 747     {
 748         const x264_level_t *l = x264_levels;
 749         if( h->param.i_level_idc < 0 )
 750         {
 751             int maxrate_bak = h->param.rc.i_vbv_max_bitrate;
 752             if( h->param.rc.i_rc_method == X264_RC_ABR && h->param.rc.i_vbv_buffer_size <= 0 )
 753                 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate * 2;
 754             h->sps = h->sps_array;
 755             x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
 756             do h->param.i_level_idc = l->level_idc;
 757                 while( l[1].level_idc && x264_validate_levels( h, 0 ) && l++ );
 758             h->param.rc.i_vbv_max_bitrate = maxrate_bak;
 759         }
 760         else
 761         {
 762             while( l->level_idc && l->level_idc != h->param.i_level_idc )
 763                 l++;
 764             if( l->level_idc == 0 )
 765             {
 766                 x264_log( h, X264_LOG_ERROR, "invalid level_idc: %d\n", h->param.i_level_idc );
 767                 return -1;
 768             }
 769         }
 770         if( h->param.analyse.i_mv_range <= 0 )
 771             h->param.analyse.i_mv_range = l->mv_range >> h->param.b_interlaced;
 772         else
 773             h->param.analyse.i_mv_range = x264_clip3(h->param.analyse.i_mv_range, 32, 512 >> h->param.b_interlaced);
 774     }
 775
 776     h->param.analyse.i_weighted_pred = x264_clip3( h->param.analyse.i_weighted_pred, X264_WEIGHTP_NONE, X264_WEIGHTP_SMART );
 777     if( !h->param.analyse.i_weighted_pred && h->param.rc.b_mb_tree && h->param.analyse.b_psy && !h->param.b_interlaced )
 778         h->param.analyse.i_weighted_pred = X264_WEIGHTP_FAKE;
 779
 780     if( h->i_thread_frames > 1 )
 781     {
 782         int r = h->param.analyse.i_mv_range_thread;
 783         int r2;
 784         if( r <= 0 )
 785         {
 786             // half of the available space is reserved and divided evenly among the threads,
 787             // the rest is allocated to whichever thread is far enough ahead to use it.
 788             // reserving more space increases quality for some videos, but costs more time
 789             // in thread synchronization.
 790             int max_range = (h->param.i_height + X264_THREAD_HEIGHT) / h->i_thread_frames - X264_THREAD_HEIGHT;
 791             r = max_range / 2;
 792         }
 793         r = X264_MAX( r, h->param.analyse.i_me_range );
 794         r = X264_MIN( r, h->param.analyse.i_mv_range );
 795         // round up to use the whole mb row
 796         r2 = (r & ~15) + ((-X264_THREAD_HEIGHT) & 15);
 797         if( r2 < r )
 798             r2 += 16;
 799         x264_log( h, X264_LOG_DEBUG, "using mv_range_thread = %d\n", r2 );
 800         h->param.analyse.i_mv_range_thread = r2;
 801     }
 802
 803     if( h->param.rc.f_qblur < 0 )
 804         h->param.rc.f_qblur = 0;
 805     if( h->param.rc.f_complexity_blur < 0 )
 806         h->param.rc.f_complexity_blur = 0;
 807
 808     h->param.i_sps_id &= 31;
 809
 810     if( h->param.i_log_level < X264_LOG_INFO )
 811     {
 812         h->param.analyse.b_psnr = 0;
 813         h->param.analyse.b_ssim = 0;
 814     }
 815
 816     if( h->param.b_interlaced )
 817         h->param.b_pic_struct = 1;
 818
 819     if( h->param.i_nal_hrd && !h->param.rc.i_vbv_buffer_size )
 820     {
 821         x264_log( h, X264_LOG_WARNING, "NAL HRD parameters require VBV parameters\n" );
 822         h->param.i_nal_hrd = X264_NAL_HRD_NONE;
 823     }
 824
 825     if( h->param.i_nal_hrd == X264_NAL_HRD_CBR &&
 826        (h->param.rc.i_bitrate != h->param.rc.i_vbv_max_bitrate || !h->param.rc.i_vbv_max_bitrate) )
 827     {
 828         x264_log( h, X264_LOG_WARNING, "CBR HRD requires constant bitrate\n" );
 829         h->param.i_nal_hrd = X264_NAL_HRD_VBR;
 830     }
 831
 832     /* ensure the booleans are 0 or 1 so they can be used in math */
 833 #define BOOLIFY(x) h->param.x = !!h->param.x
 834     BOOLIFY( b_cabac );
 835     BOOLIFY( b_constrained_intra );
 836     BOOLIFY( b_deblocking_filter );
 837     BOOLIFY( b_deterministic );
 838     BOOLIFY( b_sliced_threads );
 839     BOOLIFY( b_interlaced );
 840     BOOLIFY( b_intra_refresh );
 841     BOOLIFY( b_visualize );
 842     BOOLIFY( b_aud );
 843     BOOLIFY( b_repeat_headers );
 844     BOOLIFY( b_annexb );
 845     BOOLIFY( b_vfr_input );
 846     BOOLIFY( b_pic_struct );
 847     BOOLIFY( b_fake_interlaced );
 848     BOOLIFY( analyse.b_transform_8x8 );
 849     BOOLIFY( analyse.b_weighted_bipred );
 850     BOOLIFY( analyse.b_chroma_me );
 851     BOOLIFY( analyse.b_mixed_references );
 852     BOOLIFY( analyse.b_fast_pskip );
 853     BOOLIFY( analyse.b_dct_decimate );
 854     BOOLIFY( analyse.b_psy );
 855     BOOLIFY( analyse.b_psnr );
 856     BOOLIFY( analyse.b_ssim );
 857     BOOLIFY( rc.b_stat_write );
 858     BOOLIFY( rc.b_stat_read );
 859     BOOLIFY( rc.b_mb_tree );
 860 #undef BOOLIFY
 861
 862     return 0;
 863 }
 864
 865 static void mbcmp_init( x264_t *h )
 866 {
 867     int satd = !h->mb.b_lossless && h->param.analyse.i_subpel_refine > 1;
 868     memcpy( h->pixf.mbcmp, satd ? h->pixf.satd : h->pixf.sad_aligned, sizeof(h->pixf.mbcmp) );
 869     memcpy( h->pixf.mbcmp_unaligned, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.mbcmp_unaligned) );
 870     h->pixf.intra_mbcmp_x3_16x16 = satd ? h->pixf.intra_satd_x3_16x16 : h->pixf.intra_sad_x3_16x16;
 871     h->pixf.intra_mbcmp_x3_8x8c = satd ? h->pixf.intra_satd_x3_8x8c : h->pixf.intra_sad_x3_8x8c;
 872     h->pixf.intra_mbcmp_x3_8x8 = satd ? h->pixf.intra_sa8d_x3_8x8 : h->pixf.intra_sad_x3_8x8;
 873     h->pixf.intra_mbcmp_x3_4x4 = satd ? h->pixf.intra_satd_x3_4x4 : h->pixf.intra_sad_x3_4x4;
 874     satd &= h->param.analyse.i_me_method == X264_ME_TESA;
 875     memcpy( h->pixf.fpelcmp, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.fpelcmp) );
 876     memcpy( h->pixf.fpelcmp_x3, satd ? h->pixf.satd_x3 : h->pixf.sad_x3, sizeof(h->pixf.fpelcmp_x3) );
 877     memcpy( h->pixf.fpelcmp_x4, satd ? h->pixf.satd_x4 : h->pixf.sad_x4, sizeof(h->pixf.fpelcmp_x4) );
 878 }
 879
 880 static void x264_set_aspect_ratio( x264_t *h, x264_param_t *param, int initial )
 881 {
 882     /* VUI */
 883     if( param->vui.i_sar_width > 0 && param->vui.i_sar_height > 0 )
 884     {
 885         uint32_t i_w = param->vui.i_sar_width;
 886         uint32_t i_h = param->vui.i_sar_height;
 887         uint32_t old_w = h->param.vui.i_sar_width;
 888         uint32_t old_h = h->param.vui.i_sar_height;
 889
 890         x264_reduce_fraction( &i_w, &i_h );
 891
 892         while( i_w > 65535 || i_h > 65535 )
 893         {
 894             i_w /= 2;
 895             i_h /= 2;
 896         }
 897
 898         x264_reduce_fraction( &i_w, &i_h );
 899
 900         if( i_w != old_w || i_h != old_h || initial )
 901         {
 902             h->param.vui.i_sar_width = 0;
 903             h->param.vui.i_sar_height = 0;
 904             if( i_w == 0 || i_h == 0 )
 905                 x264_log( h, X264_LOG_WARNING, "cannot create valid sample aspect ratio\n" );
 906             else
 907             {
 908                 x264_log( h, initial?X264_LOG_INFO:X264_LOG_DEBUG, "using SAR=%d/%d\n", i_w, i_h );
 909                 h->param.vui.i_sar_width = i_w;
 910                 h->param.vui.i_sar_height = i_h;
 911             }
 912         }
 913     }
 914 }
 915
 916 /****************************************************************************
 917  * x264_encoder_open:
 918  ****************************************************************************/
 919 x264_t *x264_encoder_open( x264_param_t *param )
 920 {
 921     x264_t *h;
 922     char buf[1000], *p;
 923     int qp, i_slicetype_length;
 924
 925     CHECKED_MALLOCZERO( h, sizeof(x264_t) );
 926
 927     /* Create a copy of param */
 928     memcpy( &h->param, param, sizeof(x264_param_t) );
 929
 930     if( param->param_free )
 931         param->param_free( param );
 932
 933     if( x264_threading_init() )
 934     {
 935         x264_log( h, X264_LOG_ERROR, "unable to initialize threading\n" );
 936         goto fail;
 937     }
 938
 939     if( x264_validate_parameters( h ) < 0 )
 940         goto fail;
 941
 942     if( h->param.psz_cqm_file )
 943         if( x264_cqm_parse_file( h, h->param.psz_cqm_file ) < 0 )
 944             goto fail;
 945
 946     if( h->param.rc.psz_stat_out )
 947         h->param.rc.psz_stat_out = strdup( h->param.rc.psz_stat_out );
 948     if( h->param.rc.psz_stat_in )
 949         h->param.rc.psz_stat_in = strdup( h->param.rc.psz_stat_in );
 950
 951     x264_set_aspect_ratio( h, &h->param, 1 );
 952
 953     x264_reduce_fraction( &h->param.i_fps_num, &h->param.i_fps_den );
 954     x264_reduce_fraction( &h->param.i_timebase_num, &h->param.i_timebase_den );
 955
 956     /* Init x264_t */
 957     h->i_frame = -1;
 958     h->i_frame_num = 0;
 959     h->i_idr_pic_id = 0;
 960
 961     if( (uint64_t)h->param.i_timebase_den * 2 > UINT32_MAX )
 962     {
 963         x264_log( h, X264_LOG_ERROR, "Effective timebase denominator %u exceeds H.264 maximum\n", h->param.i_timebase_den );
 964         goto fail;
 965     }
 966
 967     h->sps = &h->sps_array[0];
 968     x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
 969
 970     h->pps = &h->pps_array[0];
 971     x264_pps_init( h->pps, h->param.i_sps_id, &h->param, h->sps );
 972
 973     x264_validate_levels( h, 1 );
 974
 975     h->chroma_qp_table = i_chroma_qp_table + 12 + h->pps->i_chroma_qp_index_offset;
 976
 977     if( x264_cqm_init( h ) < 0 )
 978         goto fail;
 979
 980     h->mb.i_mb_width = h->sps->i_mb_width;
 981     h->mb.i_mb_height = h->sps->i_mb_height;
 982     h->mb.i_mb_count = h->mb.i_mb_width * h->mb.i_mb_height;
 983
 984     /* Init frames. */
 985     if( h->param.i_bframe_adaptive == X264_B_ADAPT_TRELLIS && !h->param.rc.b_stat_read )
 986         h->frames.i_delay = X264_MAX(h->param.i_bframe,3)*4;
 987     else
 988         h->frames.i_delay = h->param.i_bframe;
 989     if( h->param.rc.b_mb_tree || h->param.rc.i_vbv_buffer_size )
 990         h->frames.i_delay = X264_MAX( h->frames.i_delay, h->param.rc.i_lookahead );
 991     i_slicetype_length = h->frames.i_delay;
 992     h->frames.i_delay += h->i_thread_frames - 1;
 993     h->frames.i_delay += h->param.i_sync_lookahead;
 994     h->frames.i_delay += h->param.b_vfr_input && (h->param.rc.i_rc_method == X264_RC_ABR || h->param.rc.b_stat_write
 995                                                  || h->param.rc.i_vbv_buffer_size);
 996     h->frames.i_bframe_delay = h->param.i_bframe ? (h->param.i_bframe_pyramid ? 2 : 1) : 0;
 997
 998     h->frames.i_max_ref0 = h->param.i_frame_reference;
 999     h->frames.i_max_ref1 = X264_MIN( h->sps->vui.i_num_reorder_frames, h->param.i_frame_reference );
1000     h->frames.i_max_dpb  = h->sps->vui.i_max_dec_frame_buffering;
1001     h->frames.b_have_lowres = !h->param.rc.b_stat_read
1002         && ( h->param.rc.i_rc_method == X264_RC_ABR
1003           || h->param.rc.i_rc_method == X264_RC_CRF
1004           || h->param.i_bframe_adaptive
1005           || h->param.i_scenecut_threshold
1006           || h->param.rc.b_mb_tree
1007           || h->param.analyse.i_weighted_pred );
1008     h->frames.b_have_lowres |= h->param.rc.b_stat_read && h->param.rc.i_vbv_buffer_size > 0;
1009     h->frames.b_have_sub8x8_esa = !!(h->param.analyse.inter & X264_ANALYSE_PSUB8x8);
1010
1011     h->frames.i_last_idr =
1012     h->frames.i_last_keyframe = - h->param.i_keyint_max;
1013     h->frames.i_input    = 0;
1014     h->frames.i_largest_pts = h->frames.i_second_largest_pts = -1;
1015     h->frames.i_poc_last_open_gop = -1;
1016
1017     CHECKED_MALLOCZERO( h->frames.unused[0], (h->frames.i_delay + 3) * sizeof(x264_frame_t *) );
1018     /* Allocate room for max refs plus a few extra just in case. */
1019     CHECKED_MALLOCZERO( h->frames.unused[1], (h->i_thread_frames + X264_REF_MAX + 4) * sizeof(x264_frame_t *) );
1020     CHECKED_MALLOCZERO( h->frames.current, (h->param.i_sync_lookahead + h->param.i_bframe
1021                         + h->i_thread_frames + 3) * sizeof(x264_frame_t *) );
1022     if( h->param.analyse.i_weighted_pred > 0 )
1023         CHECKED_MALLOCZERO( h->frames.blank_unused, h->i_thread_frames * 4 * sizeof(x264_frame_t *) );
1024     h->i_ref[0] = h->i_ref[1] = 0;
1025     h->i_cpb_delay = h->i_coded_fields = h->i_disp_fields = 0;
1026     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);
1027     h->i_disp_fields_last_frame = -1;
1028     x264_rdo_init();
1029
1030     /* init CPU functions */
1031     x264_predict_16x16_init( h->param.cpu, h->predict_16x16 );
1032     x264_predict_8x8c_init( h->param.cpu, h->predict_8x8c );
1033     x264_predict_8x8_init( h->param.cpu, h->predict_8x8, &h->predict_8x8_filter );
1034     x264_predict_4x4_init( h->param.cpu, h->predict_4x4 );
1035     if( !h->param.b_cabac )
1036         x264_init_vlc_tables();
1037     x264_pixel_init( h->param.cpu, &h->pixf );
1038     x264_dct_init( h->param.cpu, &h->dctf );
1039     x264_zigzag_init( h->param.cpu, &h->zigzagf, h->param.b_interlaced );
1040     x264_mc_init( h->param.cpu, &h->mc );
1041     x264_quant_init( h, h->param.cpu, &h->quantf );
1042     x264_deblock_init( h->param.cpu, &h->loopf );
1043     x264_bitstream_init( h->param.cpu, &h->bsf );
1044     x264_dct_init_weights();
1045
1046     mbcmp_init( h );
1047
1048     p = buf + sprintf( buf, "using cpu capabilities:" );
1049     for( int i = 0; x264_cpu_names[i].flags; i++ )
1050     {
1051         if( !strcmp(x264_cpu_names[i].name, "SSE2")
1052             && h->param.cpu & (X264_CPU_SSE2_IS_FAST|X264_CPU_SSE2_IS_SLOW) )
1053             continue;
1054         if( !strcmp(x264_cpu_names[i].name, "SSE3")
1055             && (h->param.cpu & X264_CPU_SSSE3 || !(h->param.cpu & X264_CPU_CACHELINE_64)) )
1056             continue;
1057         if( !strcmp(x264_cpu_names[i].name, "SSE4.1")
1058             && (h->param.cpu & X264_CPU_SSE42) )
1059             continue;
1060         if( (h->param.cpu & x264_cpu_names[i].flags) == x264_cpu_names[i].flags
1061             && (!i || x264_cpu_names[i].flags != x264_cpu_names[i-1].flags) )
1062             p += sprintf( p, " %s", x264_cpu_names[i].name );
1063     }
1064     if( !h->param.cpu )
1065         p += sprintf( p, " none!" );
1066     x264_log( h, X264_LOG_INFO, "%s\n", buf );
1067
1068     int qp_max = h->param.rc.i_qp_max == QP_MAX_SPEC ? QP_MAX : h->param.rc.i_qp_max;
1069     for( qp = X264_MIN( h->param.rc.i_qp_min, QP_MAX_SPEC ); qp <= qp_max; qp++ )
1070         if( x264_analyse_init_costs( h, qp ) )
1071             goto fail;
1072     if( x264_analyse_init_costs( h, X264_LOOKAHEAD_QP ) )
1073         goto fail;
1074
1075     static const uint16_t cost_mv_correct[7] = { 24, 47, 95, 189, 379, 757, 1515 };
1076     /* Checks for known miscompilation issues. */
1077     if( h->cost_mv[X264_LOOKAHEAD_QP][2013] != cost_mv_correct[BIT_DEPTH-8] )
1078     {
1079         x264_log( h, X264_LOG_ERROR, "MV cost test failed: x264 has been miscompiled!\n" );
1080         goto fail;
1081     }
1082
1083     /* Must be volatile or else GCC will optimize it out. */
1084     volatile int temp = 392;
1085     if( x264_clz( temp ) != 23 )
1086     {
1087         x264_log( h, X264_LOG_ERROR, "CLZ test failed: x264 has been miscompiled!\n" );
1088 #if ARCH_X86 || ARCH_X86_64
1089         x264_log( h, X264_LOG_ERROR, "Are you attempting to run an SSE4a-targeted build on a CPU that\n" );
1090         x264_log( h, X264_LOG_ERROR, "doesn't support it?\n" );
1091 #endif
1092         goto fail;
1093     }
1094
1095     h->out.i_nal = 0;
1096     h->out.i_bitstream = X264_MAX( 1000000, h->param.i_width * h->param.i_height * 4
1097         * ( h->param.rc.i_rc_method == X264_RC_ABR ? pow( 0.95, h->param.rc.i_qp_min )
1098           : pow( 0.95, h->param.rc.i_qp_constant ) * X264_MAX( 1, h->param.rc.f_ip_factor )));
1099
1100     CHECKED_MALLOC( h->nal_buffer, h->out.i_bitstream * 3/2 + 4 );
1101     h->nal_buffer_size = h->out.i_bitstream * 3/2 + 4;
1102
1103     if( h->param.i_threads > 1 &&
1104         x264_threadpool_init( &h->threadpool, h->param.i_threads, (void*)x264_encoder_thread_init, h ) )
1105         goto fail;
1106
1107     h->thread[0] = h;
1108     for( int i = 1; i < h->param.i_threads + !!h->param.i_sync_lookahead; i++ )
1109         CHECKED_MALLOC( h->thread[i], sizeof(x264_t) );
1110
1111     for( int i = 0; i < h->param.i_threads; i++ )
1112     {
1113         int init_nal_count = h->param.i_slice_count + 3;
1114         int allocate_threadlocal_data = !h->param.b_sliced_threads || !i;
1115         if( i > 0 )
1116             *h->thread[i] = *h;
1117
1118         if( allocate_threadlocal_data )
1119         {
1120             h->thread[i]->fdec = x264_frame_pop_unused( h, 1 );
1121             if( !h->thread[i]->fdec )
1122                 goto fail;
1123         }
1124         else
1125             h->thread[i]->fdec = h->thread[0]->fdec;
1126
1127         CHECKED_MALLOC( h->thread[i]->out.p_bitstream, h->out.i_bitstream );
1128         /* Start each thread with room for init_nal_count NAL units; it'll realloc later if needed. */
1129         CHECKED_MALLOC( h->thread[i]->out.nal, init_nal_count*sizeof(x264_nal_t) );
1130         h->thread[i]->out.i_nals_allocated = init_nal_count;
1131
1132         if( allocate_threadlocal_data && x264_macroblock_cache_allocate( h->thread[i] ) < 0 )
1133             goto fail;
1134     }
1135
1136     if( x264_lookahead_init( h, i_slicetype_length ) )
1137         goto fail;
1138
1139     for( int i = 0; i < h->param.i_threads; i++ )
1140         if( x264_macroblock_thread_allocate( h->thread[i], 0 ) < 0 )
1141             goto fail;
1142
1143     if( x264_ratecontrol_new( h ) < 0 )
1144         goto fail;
1145
1146     if( h->param.i_nal_hrd )
1147     {
1148         x264_log( h, X264_LOG_DEBUG, "HRD bitrate: %i bits/sec\n", h->sps->vui.hrd.i_bit_rate_unscaled );
1149         x264_log( h, X264_LOG_DEBUG, "CPB size: %i bits\n", h->sps->vui.hrd.i_cpb_size_unscaled );
1150     }
1151
1152     if( h->param.psz_dump_yuv )
1153     {
1154         /* create or truncate the reconstructed video file */
1155         FILE *f = fopen( h->param.psz_dump_yuv, "w" );
1156         if( !f )
1157         {
1158             x264_log( h, X264_LOG_ERROR, "dump_yuv: can't write to %s\n", h->param.psz_dump_yuv );
1159             goto fail;
1160         }
1161         else if( !x264_is_regular_file( f ) )
1162         {
1163             x264_log( h, X264_LOG_ERROR, "dump_yuv: incompatible with non-regular file %s\n", h->param.psz_dump_yuv );
1164             goto fail;
1165         }
1166         fclose( f );
1167     }
1168
1169     const char *profile = h->sps->i_profile_idc == PROFILE_BASELINE ? "Constrained Baseline" :
1170                           h->sps->i_profile_idc == PROFILE_MAIN ? "Main" :
1171                           h->sps->i_profile_idc == PROFILE_HIGH ? "High" :
1172                           h->sps->i_profile_idc == PROFILE_HIGH10 ? (h->sps->b_constraint_set3 == 1 ? "High 10 Intra" : "High 10") :
1173                           "High 4:4:4 Predictive";
1174     char level[4];
1175     snprintf( level, sizeof(level), "%d.%d", h->sps->i_level_idc/10, h->sps->i_level_idc%10 );
1176     if( h->sps->i_level_idc == 9 || ( h->sps->i_level_idc == 11 && h->sps->b_constraint_set3 ) )
1177         strcpy( level, "1b" );
1178
1179     if( h->sps->i_profile_idc < PROFILE_HIGH10 )
1180     {
1181         x264_log( h, X264_LOG_INFO, "profile %s, level %s\n",
1182             profile, level );
1183     }
1184     else
1185     {
1186         x264_log( h, X264_LOG_INFO, "profile %s, level %s, bit depth %d\n",
1187             profile, level, BIT_DEPTH );
1188     }
1189
1190     return h;
1191 fail:
1192     x264_free( h );
1193     return NULL;
1194 }
1195
1196 /****************************************************************************
1197  * x264_encoder_reconfig:
1198  ****************************************************************************/
1199 int x264_encoder_reconfig( x264_t *h, x264_param_t *param )
1200 {
1201     int rc_reconfig = 0;
1202     h = h->thread[h->thread[0]->i_thread_phase];
1203     x264_set_aspect_ratio( h, param, 0 );
1204 #define COPY(var) h->param.var = param->var
1205     COPY( i_frame_reference ); // but never uses more refs than initially specified
1206     COPY( i_bframe_bias );
1207     if( h->param.i_scenecut_threshold )
1208         COPY( i_scenecut_threshold ); // can't turn it on or off, only vary the threshold
1209     COPY( b_deblocking_filter );
1210     COPY( i_deblocking_filter_alphac0 );
1211     COPY( i_deblocking_filter_beta );
1212     COPY( i_frame_packing );
1213     COPY( analyse.inter );
1214     COPY( analyse.intra );
1215     COPY( analyse.i_direct_mv_pred );
1216     /* Scratch buffer prevents me_range from being increased for esa/tesa */
1217     if( h->param.analyse.i_me_method < X264_ME_ESA || param->analyse.i_me_range < h->param.analyse.i_me_range )
1218         COPY( analyse.i_me_range );
1219     COPY( analyse.i_noise_reduction );
1220     /* We can't switch out of subme=0 during encoding. */
1221     if( h->param.analyse.i_subpel_refine )
1222         COPY( analyse.i_subpel_refine );
1223     COPY( analyse.i_trellis );
1224     COPY( analyse.b_chroma_me );
1225     COPY( analyse.b_dct_decimate );
1226     COPY( analyse.b_fast_pskip );
1227     COPY( analyse.b_mixed_references );
1228     COPY( analyse.f_psy_rd );
1229     COPY( analyse.f_psy_trellis );
1230     COPY( crop_rect );
1231     // can only twiddle these if they were enabled to begin with:
1232     if( h->param.analyse.i_me_method >= X264_ME_ESA || param->analyse.i_me_method < X264_ME_ESA )
1233         COPY( analyse.i_me_method );
1234     if( h->param.analyse.i_me_method >= X264_ME_ESA && !h->frames.b_have_sub8x8_esa )
1235         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
1236     if( h->pps->b_transform_8x8_mode )
1237         COPY( analyse.b_transform_8x8 );
1238     if( h->frames.i_max_ref1 > 1 )
1239         COPY( i_bframe_pyramid );
1240     COPY( i_slice_max_size );
1241     COPY( i_slice_max_mbs );
1242     COPY( i_slice_count );
1243     COPY( b_tff );
1244
1245     /* VBV can't be turned on if it wasn't on to begin with */
1246     if( h->param.rc.i_vbv_max_bitrate > 0 && h->param.rc.i_vbv_buffer_size > 0 &&
1247           param->rc.i_vbv_max_bitrate > 0 &&   param->rc.i_vbv_buffer_size > 0 )
1248     {
1249         COPY( rc.i_vbv_max_bitrate );
1250         COPY( rc.i_vbv_buffer_size );
1251         COPY( rc.i_bitrate );
1252         rc_reconfig = 1;
1253     }
1254     if( h->param.rc.f_rf_constant != param->rc.f_rf_constant )
1255     {
1256         COPY( rc.f_rf_constant );
1257         rc_reconfig = 1;
1258     }
1259     if( h->param.rc.f_rf_constant_max != param->rc.f_rf_constant_max )
1260     {
1261         COPY( rc.f_rf_constant_max );
1262         rc_reconfig = 1;
1263     }
1264
1265 #undef COPY
1266
1267     mbcmp_init( h );
1268
1269     int ret = x264_validate_parameters( h );
1270
1271     /* Supported reconfiguration options (1-pass only):
1272      * vbv-maxrate
1273      * vbv-bufsize
1274      * crf
1275      * bitrate (CBR only) */
1276     if( !ret && rc_reconfig )
1277         x264_ratecontrol_init_reconfigurable( h, 0 );
1278
1279     return ret;
1280 }
1281
1282 /****************************************************************************
1283  * x264_encoder_parameters:
1284  ****************************************************************************/
1285 void x264_encoder_parameters( x264_t *h, x264_param_t *param )
1286 {
1287     memcpy( param, &h->thread[h->i_thread_phase]->param, sizeof(x264_param_t) );
1288 }
1289
1290 /* internal usage */
1291 static void x264_nal_start( x264_t *h, int i_type, int i_ref_idc )
1292 {
1293     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1294
1295     nal->i_ref_idc        = i_ref_idc;
1296     nal->i_type           = i_type;
1297     nal->b_long_startcode = 1;
1298
1299     nal->i_payload= 0;
1300     nal->p_payload= &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8];
1301 }
1302
1303 /* if number of allocated nals is not enough, re-allocate a larger one. */
1304 static int x264_nal_check_buffer( x264_t *h )
1305 {
1306     if( h->out.i_nal >= h->out.i_nals_allocated )
1307     {
1308         x264_nal_t *new_out = x264_malloc( sizeof(x264_nal_t) * (h->out.i_nals_allocated*2) );
1309         if( !new_out )
1310             return -1;
1311         memcpy( new_out, h->out.nal, sizeof(x264_nal_t) * (h->out.i_nals_allocated) );
1312         x264_free( h->out.nal );
1313         h->out.nal = new_out;
1314         h->out.i_nals_allocated *= 2;
1315     }
1316     return 0;
1317 }
1318
1319 static int x264_nal_end( x264_t *h )
1320 {
1321     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1322     nal->i_payload = &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8] - nal->p_payload;
1323     if( h->param.nalu_process )
1324         h->param.nalu_process( h, nal );
1325     h->out.i_nal++;
1326
1327     return x264_nal_check_buffer( h );
1328 }
1329
1330 static int x264_encoder_encapsulate_nals( x264_t *h, int start )
1331 {
1332     int nal_size = 0, previous_nal_size = 0;
1333
1334     if( h->param.nalu_process )
1335     {
1336         for( int i = start; i < h->out.i_nal; i++ )
1337             nal_size += h->out.nal[i].i_payload;
1338         return nal_size;
1339     }
1340
1341     for( int i = 0; i < start; i++ )
1342         previous_nal_size += h->out.nal[i].i_payload;
1343
1344     for( int i = start; i < h->out.i_nal; i++ )
1345         nal_size += h->out.nal[i].i_payload;
1346
1347     /* Worst-case NAL unit escaping: reallocate the buffer if it's too small. */
1348     if( h->nal_buffer_size < nal_size * 3/2 + h->out.i_nal * 4 )
1349     {
1350         uint8_t *buf = x264_malloc( nal_size * 2 + h->out.i_nal * 4 );
1351         if( !buf )
1352             return -1;
1353         if( previous_nal_size )
1354             memcpy( buf, h->nal_buffer, previous_nal_size );
1355         x264_free( h->nal_buffer );
1356         h->nal_buffer = buf;
1357     }
1358
1359     uint8_t *nal_buffer = h->nal_buffer + previous_nal_size;
1360
1361     for( int i = start; i < h->out.i_nal; i++ )
1362     {
1363         h->out.nal[i].b_long_startcode = !i || h->out.nal[i].i_type == NAL_SPS || h->out.nal[i].i_type == NAL_PPS;
1364         x264_nal_encode( h, nal_buffer, &h->out.nal[i] );
1365         nal_buffer += h->out.nal[i].i_payload;
1366     }
1367
1368     x264_emms();
1369
1370     return nal_buffer - (h->nal_buffer + previous_nal_size);
1371 }
1372
1373 /****************************************************************************
1374  * x264_encoder_headers:
1375  ****************************************************************************/
1376 int x264_encoder_headers( x264_t *h, x264_nal_t **pp_nal, int *pi_nal )
1377 {
1378     int frame_size = 0;
1379     /* init bitstream context */
1380     h->out.i_nal = 0;
1381     bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
1382
1383     /* Write SEI, SPS and PPS. */
1384
1385     /* generate sequence parameters */
1386     x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
1387     x264_sps_write( &h->out.bs, h->sps );
1388     if( x264_nal_end( h ) )
1389         return -1;
1390
1391     /* generate picture parameters */
1392     x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
1393     x264_pps_write( &h->out.bs, h->pps );
1394     if( x264_nal_end( h ) )
1395         return -1;
1396
1397     /* identify ourselves */
1398     x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
1399     if( x264_sei_version_write( h, &h->out.bs ) )
1400         return -1;
1401     if( x264_nal_end( h ) )
1402         return -1;
1403
1404     frame_size = x264_encoder_encapsulate_nals( h, 0 );
1405
1406     /* now set output*/
1407     *pi_nal = h->out.i_nal;
1408     *pp_nal = &h->out.nal[0];
1409     h->out.i_nal = 0;
1410
1411     return frame_size;
1412 }
1413
1414 /* Check to see whether we have chosen a reference list ordering different
1415  * from the standard's default. */
1416 static inline void x264_reference_check_reorder( x264_t *h )
1417 {
1418     /* The reorder check doesn't check for missing frames, so just
1419      * force a reorder if one of the reference list is corrupt. */
1420     for( int i = 0; h->frames.reference[i]; i++ )
1421         if( h->frames.reference[i]->b_corrupt )
1422         {
1423             h->b_ref_reorder[0] = 1;
1424             return;
1425         }
1426     for( int list = 0; list <= (h->sh.i_type == SLICE_TYPE_B); list++ )
1427         for( int i = 0; i < h->i_ref[list] - 1; i++ )
1428         {
1429             int framenum_diff = h->fref[list][i+1]->i_frame_num - h->fref[list][i]->i_frame_num;
1430             int poc_diff = h->fref[list][i+1]->i_poc - h->fref[list][i]->i_poc;
1431             /* P and B-frames use different default orders. */
1432             if( h->sh.i_type == SLICE_TYPE_P ? framenum_diff > 0 : list == 1 ? poc_diff < 0 : poc_diff > 0 )
1433             {
1434                 h->b_ref_reorder[list] = 1;
1435                 return;
1436             }
1437         }
1438 }
1439
1440 /* return -1 on failure, else return the index of the new reference frame */
1441 int x264_weighted_reference_duplicate( x264_t *h, int i_ref, const x264_weight_t *w )
1442 {
1443     int i = h->i_ref[0];
1444     int j = 1;
1445     x264_frame_t *newframe;
1446     if( i <= 1 ) /* empty list, definitely can't duplicate frame */
1447         return -1;
1448
1449     //Duplication is only used in X264_WEIGHTP_SMART
1450     if( h->param.analyse.i_weighted_pred != X264_WEIGHTP_SMART )
1451         return -1;
1452
1453     /* Duplication is a hack to compensate for crappy rounding in motion compensation.
1454      * With high bit depth, it's not worth doing, so turn it off except in the case of
1455      * unweighted dupes. */
1456     if( BIT_DEPTH > 8 && w != weight_none )
1457         return -1;
1458
1459     newframe = x264_frame_pop_blank_unused( h );
1460     if( !newframe )
1461         return -1;
1462
1463     //FIXME: probably don't need to copy everything
1464     *newframe = *h->fref[0][i_ref];
1465     newframe->i_reference_count = 1;
1466     newframe->orig = h->fref[0][i_ref];
1467     newframe->b_duplicate = 1;
1468     memcpy( h->fenc->weight[j], w, sizeof(h->fenc->weight[i]) );
1469
1470     /* shift the frames to make space for the dupe. */
1471     h->b_ref_reorder[0] = 1;
1472     if( h->i_ref[0] < X264_REF_MAX )
1473         ++h->i_ref[0];
1474     h->fref[0][X264_REF_MAX-1] = NULL;
1475     x264_frame_unshift( &h->fref[0][j], newframe );
1476
1477     return j;
1478 }
1479
1480 static void x264_weighted_pred_init( x264_t *h )
1481 {
1482     /* for now no analysis and set all weights to nothing */
1483     for( int i_ref = 0; i_ref < h->i_ref[0]; i_ref++ )
1484         h->fenc->weighted[i_ref] = h->fref[0][i_ref]->filtered[0];
1485
1486     // FIXME: This only supports weighting of one reference frame
1487     // and duplicates of that frame.
1488     h->fenc->i_lines_weighted = 0;
1489
1490     for( int i_ref = 0; i_ref < (h->i_ref[0] << h->sh.b_mbaff); i_ref++ )
1491         for( int i = 0; i < 3; i++ )
1492             h->sh.weight[i_ref][i].weightfn = NULL;
1493
1494
1495     if( h->sh.i_type != SLICE_TYPE_P || h->param.analyse.i_weighted_pred <= 0 )
1496         return;
1497
1498     int i_padv = PADV << h->param.b_interlaced;
1499     int denom = -1;
1500     int weightplane[2] = { 0, 0 };
1501     int buffer_next = 0;
1502     for( int i = 0; i < 3; i++ )
1503     {
1504         for( int j = 0; j < h->i_ref[0]; j++ )
1505         {
1506             if( h->fenc->weight[j][i].weightfn )
1507             {
1508                 h->sh.weight[j][i] = h->fenc->weight[j][i];
1509                 // if weight is useless, don't write it to stream
1510                 if( h->sh.weight[j][i].i_scale == 1<<h->sh.weight[j][i].i_denom && h->sh.weight[j][i].i_offset == 0 )
1511                     h->sh.weight[j][i].weightfn = NULL;
1512                 else
1513                 {
1514                     if( !weightplane[!!i] )
1515                     {
1516                         weightplane[!!i] = 1;
1517                         h->sh.weight[0][!!i].i_denom = denom = h->sh.weight[j][i].i_denom;
1518                         assert( x264_clip3( denom, 0, 7 ) == denom );
1519                     }
1520
1521                     assert( h->sh.weight[j][i].i_denom == denom );
1522                     if( !i )
1523                     {
1524                         h->fenc->weighted[j] = h->mb.p_weight_buf[buffer_next++] + h->fenc->i_stride[0] * i_padv + PADH;
1525                         //scale full resolution frame
1526                         if( h->param.i_threads == 1 )
1527                         {
1528                             pixel *src = h->fref[0][j]->filtered[0] - h->fref[0][j]->i_stride[0]*i_padv - PADH;
1529                             pixel *dst = h->fenc->weighted[j] - h->fenc->i_stride[0]*i_padv - PADH;
1530                             int stride = h->fenc->i_stride[0];
1531                             int width = h->fenc->i_width[0] + PADH*2;
1532                             int height = h->fenc->i_lines[0] + i_padv*2;
1533                             x264_weight_scale_plane( h, dst, stride, src, stride, width, height, &h->sh.weight[j][0] );
1534                             h->fenc->i_lines_weighted = height;
1535                         }
1536                     }
1537                 }
1538             }
1539         }
1540     }
1541
1542     if( weightplane[1] )
1543         for( int i = 0; i < h->i_ref[0]; i++ )
1544         {
1545             if( h->sh.weight[i][1].weightfn && !h->sh.weight[i][2].weightfn )
1546             {
1547                 h->sh.weight[i][2].i_scale = 1 << h->sh.weight[0][1].i_denom;
1548                 h->sh.weight[i][2].i_offset = 0;
1549             }
1550             else if( h->sh.weight[i][2].weightfn && !h->sh.weight[i][1].weightfn )
1551             {
1552                 h->sh.weight[i][1].i_scale = 1 << h->sh.weight[0][1].i_denom;
1553                 h->sh.weight[i][1].i_offset = 0;
1554             }
1555         }
1556
1557     if( !weightplane[0] )
1558         h->sh.weight[0][0].i_denom = 0;
1559     if( !weightplane[1] )
1560         h->sh.weight[0][1].i_denom = 0;
1561     h->sh.weight[0][2].i_denom = h->sh.weight[0][1].i_denom;
1562 }
1563
1564 static inline int x264_reference_distance( x264_t *h, x264_frame_t *frame )
1565 {
1566     if( h->param.i_frame_packing == 5 )
1567         return abs((h->fenc->i_frame&~1) - (frame->i_frame&~1)) +
1568                   ((h->fenc->i_frame&1) != (frame->i_frame&1));
1569     else
1570         return abs(h->fenc->i_frame - frame->i_frame);
1571 }
1572
1573 static inline void x264_reference_build_list( x264_t *h, int i_poc )
1574 {
1575     int b_ok;
1576
1577     /* build ref list 0/1 */
1578     h->mb.pic.i_fref[0] = h->i_ref[0] = 0;
1579     h->mb.pic.i_fref[1] = h->i_ref[1] = 0;
1580     if( h->sh.i_type == SLICE_TYPE_I )
1581         return;
1582
1583     for( int i = 0; h->frames.reference[i]; i++ )
1584     {
1585         if( h->frames.reference[i]->b_corrupt )
1586             continue;
1587         if( h->frames.reference[i]->i_poc < i_poc )
1588             h->fref[0][h->i_ref[0]++] = h->frames.reference[i];
1589         else if( h->frames.reference[i]->i_poc > i_poc )
1590             h->fref[1][h->i_ref[1]++] = h->frames.reference[i];
1591     }
1592
1593     /* Order reference lists by distance from the current frame. */
1594     for( int list = 0; list < 2; list++ )
1595     {
1596         h->fref_nearest[list] = h->fref[list][0];
1597         do
1598         {
1599             b_ok = 1;
1600             for( int i = 0; i < h->i_ref[list] - 1; i++ )
1601             {
1602                 if( list ? h->fref[list][i+1]->i_poc < h->fref_nearest[list]->i_poc
1603                          : h->fref[list][i+1]->i_poc > h->fref_nearest[list]->i_poc )
1604                     h->fref_nearest[list] = h->fref[list][i+1];
1605                 if( x264_reference_distance( h, h->fref[list][i] ) > x264_reference_distance( h, h->fref[list][i+1] ) )
1606                 {
1607                     XCHG( x264_frame_t*, h->fref[list][i], h->fref[list][i+1] );
1608                     b_ok = 0;
1609                     break;
1610                 }
1611             }
1612         } while( !b_ok );
1613     }
1614
1615     if( h->sh.i_mmco_remove_from_end )
1616         for( int i = h->i_ref[0]-1; i >= h->i_ref[0] - h->sh.i_mmco_remove_from_end; i-- )
1617         {
1618             int diff = h->i_frame_num - h->fref[0][i]->i_frame_num;
1619             h->sh.mmco[h->sh.i_mmco_command_count].i_poc = h->fref[0][i]->i_poc;
1620             h->sh.mmco[h->sh.i_mmco_command_count++].i_difference_of_pic_nums = diff;
1621         }
1622
1623     x264_reference_check_reorder( h );
1624
1625     h->i_ref[1] = X264_MIN( h->i_ref[1], h->frames.i_max_ref1 );
1626     h->i_ref[0] = X264_MIN( h->i_ref[0], h->frames.i_max_ref0 );
1627     h->i_ref[0] = X264_MIN( h->i_ref[0], h->param.i_frame_reference ); // if reconfig() has lowered the limit
1628
1629     /* add duplicates */
1630     if( h->fenc->i_type == X264_TYPE_P )
1631     {
1632         int idx = -1;
1633         if( h->param.analyse.i_weighted_pred >= X264_WEIGHTP_SIMPLE )
1634         {
1635             x264_weight_t w[3];
1636             w[1].weightfn = w[2].weightfn = NULL;
1637             if( h->param.rc.b_stat_read )
1638                 x264_ratecontrol_set_weights( h, h->fenc );
1639
1640             if( !h->fenc->weight[0][0].weightfn )
1641             {
1642                 h->fenc->weight[0][0].i_denom = 0;
1643                 SET_WEIGHT( w[0], 1, 1, 0, -1 );
1644                 idx = x264_weighted_reference_duplicate( h, 0, w );
1645             }
1646             else
1647             {
1648                 if( h->fenc->weight[0][0].i_scale == 1<<h->fenc->weight[0][0].i_denom )
1649                 {
1650                     SET_WEIGHT( h->fenc->weight[0][0], 1, 1, 0, h->fenc->weight[0][0].i_offset );
1651                 }
1652                 x264_weighted_reference_duplicate( h, 0, weight_none );
1653                 if( h->fenc->weight[0][0].i_offset > -128 )
1654                 {
1655                     w[0] = h->fenc->weight[0][0];
1656                     w[0].i_offset--;
1657                     h->mc.weight_cache( h, &w[0] );
1658                     idx = x264_weighted_reference_duplicate( h, 0, w );
1659                 }
1660             }
1661         }
1662         h->mb.ref_blind_dupe = idx;
1663     }
1664
1665     assert( h->i_ref[0] + h->i_ref[1] <= X264_REF_MAX );
1666     h->mb.pic.i_fref[0] = h->i_ref[0];
1667     h->mb.pic.i_fref[1] = h->i_ref[1];
1668 }
1669
1670 static void x264_fdec_filter_row( x264_t *h, int mb_y, int b_inloop )
1671 {
1672     /* mb_y is the mb to be encoded next, not the mb to be filtered here */
1673     int b_hpel = h->fdec->b_kept_as_ref;
1674     int b_deblock = h->sh.i_disable_deblocking_filter_idc != 1;
1675     int b_end = mb_y == h->i_threadslice_end;
1676     int b_measure_quality = 1;
1677     int min_y = mb_y - (1 << h->sh.b_mbaff);
1678     int b_start = min_y == h->i_threadslice_start;
1679     int max_y = b_end ? h->i_threadslice_end : mb_y;
1680     b_deblock &= b_hpel || h->param.psz_dump_yuv;
1681     if( h->param.b_sliced_threads && b_start && min_y && !b_inloop )
1682     {
1683         b_deblock = 0;         /* We already deblocked on the inloop pass. */
1684         b_measure_quality = 0; /* We already measured quality on the inloop pass. */
1685     }
1686     if( mb_y & h->sh.b_mbaff )
1687         return;
1688     if( min_y < h->i_threadslice_start )
1689         return;
1690
1691     if( b_deblock )
1692         for( int y = min_y; y < max_y; y += (1 << h->sh.b_mbaff) )
1693             x264_frame_deblock_row( h, y );
1694
1695     if( b_hpel )
1696     {
1697         int end = mb_y == h->mb.i_mb_height;
1698         x264_frame_expand_border( h, h->fdec, min_y, end );
1699         if( h->param.analyse.i_subpel_refine )
1700         {
1701             x264_frame_filter( h, h->fdec, min_y, end );
1702             x264_frame_expand_border_filtered( h, h->fdec, min_y, end );
1703         }
1704     }
1705
1706     if( h->i_thread_frames > 1 && h->fdec->b_kept_as_ref )
1707         x264_frame_cond_broadcast( h->fdec, mb_y*16 + (b_end ? 10000 : -(X264_THREAD_HEIGHT << h->sh.b_mbaff)) );
1708
1709     min_y = min_y*16 - 8 * !b_start;
1710     max_y = b_end ? X264_MIN( h->i_threadslice_end*16 , h->param.i_height ) : mb_y*16 - 8;
1711
1712     if( b_measure_quality )
1713     {
1714         if( h->param.analyse.b_psnr )
1715         {
1716             uint64_t ssd_y = x264_pixel_ssd_wxh( &h->pixf,
1717                 h->fdec->plane[0] + min_y * h->fdec->i_stride[0], h->fdec->i_stride[0],
1718                 h->fenc->plane[0] + min_y * h->fenc->i_stride[0], h->fenc->i_stride[0],
1719                 h->param.i_width, max_y-min_y );
1720             uint64_t ssd_u, ssd_v;
1721             x264_pixel_ssd_nv12( &h->pixf,
1722                 h->fdec->plane[1] + (min_y>>1) * h->fdec->i_stride[1], h->fdec->i_stride[1],
1723                 h->fenc->plane[1] + (min_y>>1) * h->fenc->i_stride[1], h->fenc->i_stride[1],
1724                 h->param.i_width>>1, (max_y-min_y)>>1, &ssd_u, &ssd_v );
1725             h->stat.frame.i_ssd[0] += ssd_y;
1726             h->stat.frame.i_ssd[1] += ssd_u;
1727             h->stat.frame.i_ssd[2] += ssd_v;
1728         }
1729
1730         if( h->param.analyse.b_ssim )
1731         {
1732             x264_emms();
1733             /* offset by 2 pixels to avoid alignment of ssim blocks with dct blocks,
1734              * and overlap by 4 */
1735             min_y += b_start ? 2 : -6;
1736             h->stat.frame.f_ssim +=
1737                 x264_pixel_ssim_wxh( &h->pixf,
1738                     h->fdec->plane[0] + 2+min_y*h->fdec->i_stride[0], h->fdec->i_stride[0],
1739                     h->fenc->plane[0] + 2+min_y*h->fenc->i_stride[0], h->fenc->i_stride[0],
1740                     h->param.i_width-2, max_y-min_y, h->scratch_buffer );
1741         }
1742     }
1743 }
1744
1745 static inline int x264_reference_update( x264_t *h )
1746 {
1747     if( !h->fdec->b_kept_as_ref )
1748     {
1749         if( h->i_thread_frames > 1 )
1750         {
1751             x264_frame_push_unused( h, h->fdec );
1752             h->fdec = x264_frame_pop_unused( h, 1 );
1753             if( !h->fdec )
1754                 return -1;
1755         }
1756         return 0;
1757     }
1758
1759     /* apply mmco from previous frame. */
1760     for( int i = 0; i < h->sh.i_mmco_command_count; i++ )
1761         for( int j = 0; h->frames.reference[j]; j++ )
1762             if( h->frames.reference[j]->i_poc == h->sh.mmco[i].i_poc )
1763                 x264_frame_push_unused( h, x264_frame_shift( &h->frames.reference[j] ) );
1764
1765     /* move frame in the buffer */
1766     x264_frame_push( h->frames.reference, h->fdec );
1767     if( h->frames.reference[h->sps->i_num_ref_frames] )
1768         x264_frame_push_unused( h, x264_frame_shift( h->frames.reference ) );
1769     h->fdec = x264_frame_pop_unused( h, 1 );
1770     if( !h->fdec )
1771         return -1;
1772     return 0;
1773 }
1774
1775 static inline void x264_reference_reset( x264_t *h )
1776 {
1777     while( h->frames.reference[0] )
1778         x264_frame_push_unused( h, x264_frame_pop( h->frames.reference ) );
1779     h->fdec->i_poc =
1780     h->fenc->i_poc = 0;
1781 }
1782
1783 static inline void x264_reference_hierarchy_reset( x264_t *h )
1784 {
1785     int ref;
1786     int b_hasdelayframe = 0;
1787
1788     /* look for delay frames -- chain must only contain frames that are disposable */
1789     for( int i = 0; h->frames.current[i] && IS_DISPOSABLE( h->frames.current[i]->i_type ); i++ )
1790         b_hasdelayframe |= h->frames.current[i]->i_coded
1791                         != h->frames.current[i]->i_frame + h->sps->vui.i_num_reorder_frames;
1792
1793     /* This function must handle b-pyramid and clear frames for open-gop */
1794     if( h->param.i_bframe_pyramid != X264_B_PYRAMID_STRICT && !b_hasdelayframe && h->frames.i_poc_last_open_gop == -1 )
1795         return;
1796
1797     /* Remove last BREF. There will never be old BREFs in the
1798      * dpb during a BREF decode when pyramid == STRICT */
1799     for( ref = 0; h->frames.reference[ref]; ref++ )
1800     {
1801         if( ( h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT
1802             && h->frames.reference[ref]->i_type == X264_TYPE_BREF )
1803             || ( h->frames.reference[ref]->i_poc < h->frames.i_poc_last_open_gop
1804             && h->sh.i_type != SLICE_TYPE_B ) )
1805         {
1806             int diff = h->i_frame_num - h->frames.reference[ref]->i_frame_num;
1807             h->sh.mmco[h->sh.i_mmco_command_count].i_difference_of_pic_nums = diff;
1808             h->sh.mmco[h->sh.i_mmco_command_count++].i_poc = h->frames.reference[ref]->i_poc;
1809             x264_frame_push_unused( h, x264_frame_shift( &h->frames.reference[ref] ) );
1810             h->b_ref_reorder[0] = 1;
1811             ref--;
1812         }
1813     }
1814
1815     /* Prepare room in the dpb for the delayed display time of the later b-frame's */
1816     if( h->param.i_bframe_pyramid )
1817         h->sh.i_mmco_remove_from_end = X264_MAX( ref + 2 - h->frames.i_max_dpb, 0 );
1818 }
1819
1820 static inline void x264_slice_init( x264_t *h, int i_nal_type, int i_global_qp )
1821 {
1822     /* ------------------------ Create slice header  ----------------------- */
1823     if( i_nal_type == NAL_SLICE_IDR )
1824     {
1825         x264_slice_header_init( h, &h->sh, h->sps, h->pps, h->i_idr_pic_id, h->i_frame_num, i_global_qp );
1826
1827         /* alternate id */
1828         h->i_idr_pic_id ^= 1;
1829     }
1830     else
1831     {
1832         x264_slice_header_init( h, &h->sh, h->sps, h->pps, -1, h->i_frame_num, i_global_qp );
1833
1834         h->sh.i_num_ref_idx_l0_active = h->i_ref[0] <= 0 ? 1 : h->i_ref[0];
1835         h->sh.i_num_ref_idx_l1_active = h->i_ref[1] <= 0 ? 1 : h->i_ref[1];
1836         if( h->sh.i_num_ref_idx_l0_active != h->pps->i_num_ref_idx_l0_default_active ||
1837             (h->sh.i_type == SLICE_TYPE_B && h->sh.i_num_ref_idx_l1_active != h->pps->i_num_ref_idx_l1_default_active) )
1838         {
1839             h->sh.b_num_ref_idx_override = 1;
1840         }
1841     }
1842
1843     h->fdec->i_frame_num = h->sh.i_frame_num;
1844
1845     if( h->sps->i_poc_type == 0 )
1846     {
1847         h->sh.i_poc = h->fdec->i_poc;
1848         if( h->param.b_interlaced )
1849         {
1850             h->sh.i_delta_poc_bottom = h->param.b_tff ? 1 : -1;
1851             h->sh.i_poc += h->sh.i_delta_poc_bottom == -1;
1852         }
1853         else
1854             h->sh.i_delta_poc_bottom = 0;
1855         h->fdec->i_delta_poc[0] = h->sh.i_delta_poc_bottom == -1;
1856         h->fdec->i_delta_poc[1] = h->sh.i_delta_poc_bottom ==  1;
1857     }
1858     else if( h->sps->i_poc_type == 1 )
1859     {
1860         /* FIXME TODO FIXME */
1861     }
1862     else
1863     {
1864         /* Nothing to do ? */
1865     }
1866
1867     x264_macroblock_slice_init( h );
1868 }
1869
1870 static int x264_slice_write( x264_t *h )
1871 {
1872     int i_skip;
1873     int mb_xy, i_mb_x, i_mb_y;
1874     int i_skip_bak = 0; /* Shut up GCC. */
1875     bs_t UNINIT(bs_bak);
1876     x264_cabac_t cabac_bak;
1877     uint8_t cabac_prevbyte_bak = 0; /* Shut up GCC. */
1878     int mv_bits_bak = 0;
1879     int tex_bits_bak = 0;
1880     /* NALUs other than the first use a 3-byte startcode.
1881      * Add one extra byte for the rbsp, and one more for the final CABAC putbyte.
1882      * Then add an extra 5 bytes just in case, to account for random NAL escapes and
1883      * other inaccuracies. */
1884     int overhead_guess = (NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal)) + 1 + h->param.b_cabac + 5;
1885     int slice_max_size = h->param.i_slice_max_size > 0 ? (h->param.i_slice_max_size-overhead_guess)*8 : 0;
1886     int starting_bits = bs_pos(&h->out.bs);
1887     int b_deblock = h->sh.i_disable_deblocking_filter_idc != 1;
1888     int b_hpel = h->fdec->b_kept_as_ref;
1889     b_deblock &= b_hpel || h->param.psz_dump_yuv;
1890     bs_realign( &h->out.bs );
1891
1892     /* Slice */
1893     x264_nal_start( h, h->i_nal_type, h->i_nal_ref_idc );
1894     h->out.nal[h->out.i_nal].i_first_mb = h->sh.i_first_mb;
1895
1896     /* Slice header */
1897     x264_macroblock_thread_init( h );
1898
1899     /* If this isn't the first slice in the threadslice, set the slice QP
1900      * equal to the last QP in the previous slice for more accurate
1901      * CABAC initialization. */
1902     if( h->sh.i_first_mb != h->i_threadslice_start * h->mb.i_mb_width )
1903     {
1904         h->sh.i_qp = h->mb.i_last_qp;
1905         h->sh.i_qp_delta = h->sh.i_qp - h->pps->i_pic_init_qp;
1906     }
1907
1908     x264_slice_header_write( &h->out.bs, &h->sh, h->i_nal_ref_idc );
1909     if( h->param.b_cabac )
1910     {
1911         /* alignment needed */
1912         bs_align_1( &h->out.bs );
1913
1914         /* init cabac */
1915         x264_cabac_context_init( &h->cabac, h->sh.i_type, x264_clip3( h->sh.i_qp-QP_BD_OFFSET, 0, 51 ), h->sh.i_cabac_init_idc );
1916         x264_cabac_encode_init ( &h->cabac, h->out.bs.p, h->out.bs.p_end );
1917     }
1918     h->mb.i_last_qp = h->sh.i_qp;
1919     h->mb.i_last_dqp = 0;
1920
1921     i_mb_y = h->sh.i_first_mb / h->mb.i_mb_width;
1922     i_mb_x = h->sh.i_first_mb % h->mb.i_mb_width;
1923     i_skip = 0;
1924
1925     while( (mb_xy = i_mb_x + i_mb_y * h->mb.i_mb_width) <= h->sh.i_last_mb )
1926     {
1927         int mb_spos = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
1928
1929         if( x264_bitstream_check_buffer( h ) )
1930             return -1;
1931
1932         if( slice_max_size )
1933         {
1934             mv_bits_bak = h->stat.frame.i_mv_bits;
1935             tex_bits_bak = h->stat.frame.i_tex_bits;
1936             /* We don't need the contexts because flushing the CABAC encoder has no context
1937              * dependency and macroblocks are only re-encoded in the case where a slice is
1938              * ended (and thus the content of all contexts are thrown away). */
1939             if( h->param.b_cabac )
1940             {
1941                 memcpy( &cabac_bak, &h->cabac, offsetof(x264_cabac_t, f8_bits_encoded) );
1942                 /* x264's CABAC writer modifies the previous byte during carry, so it has to be
1943                  * backed up. */
1944                 cabac_prevbyte_bak = h->cabac.p[-1];
1945             }
1946             else
1947             {
1948                 bs_bak = h->out.bs;
1949                 i_skip_bak = i_skip;
1950             }
1951         }
1952
1953         if( i_mb_x == 0 && !h->mb.b_reencode_mb )
1954             x264_fdec_filter_row( h, i_mb_y, 1 );
1955
1956         /* load cache */
1957         x264_macroblock_cache_load( h, i_mb_x, i_mb_y );
1958
1959         x264_macroblock_analyse( h );
1960
1961         /* encode this macroblock -> be careful it can change the mb type to P_SKIP if needed */
1962         x264_macroblock_encode( h );
1963
1964         if( h->param.b_cabac )
1965         {
1966             if( mb_xy > h->sh.i_first_mb && !(h->sh.b_mbaff && (i_mb_y&1)) )
1967                 x264_cabac_encode_terminal( &h->cabac );
1968
1969             if( IS_SKIP( h->mb.i_type ) )
1970                 x264_cabac_mb_skip( h, 1 );
1971             else
1972             {
1973                 if( h->sh.i_type != SLICE_TYPE_I )
1974                     x264_cabac_mb_skip( h, 0 );
1975                 x264_macroblock_write_cabac( h, &h->cabac );
1976             }
1977         }
1978         else
1979         {
1980             if( IS_SKIP( h->mb.i_type ) )
1981                 i_skip++;
1982             else
1983             {
1984                 if( h->sh.i_type != SLICE_TYPE_I )
1985                 {
1986                     bs_write_ue( &h->out.bs, i_skip );  /* skip run */
1987                     i_skip = 0;
1988                 }
1989                 x264_macroblock_write_cavlc( h );
1990             }
1991         }
1992
1993         int total_bits = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
1994         int mb_size = total_bits - mb_spos;
1995
1996         if( slice_max_size )
1997         {
1998             /* Count the skip run, just in case. */
1999             if( !h->param.b_cabac )
2000                 total_bits += bs_size_ue_big( i_skip );
2001             /* HACK: we assume no more than 3 bytes of NALU escaping, but
2002              * this can fail in CABAC streams with an extremely large number of identical
2003              * blocks in sequence (e.g. all-black intra blocks).
2004              * Thus, every 64 blocks, pretend we've used a byte.
2005              * For reference, a seqeuence of identical empty-CBP i16x16 blocks will use
2006              * one byte after 26 macroblocks, assuming a perfectly adapted CABAC.
2007              * That's 78 macroblocks to generate the 3-byte sequence to trigger an escape. */
2008             else if( ((mb_xy - h->sh.i_first_mb) & 63) == 63 )
2009                 slice_max_size -= 8;
2010             /* We'll just re-encode this last macroblock if we go over the max slice size. */
2011             if( total_bits - starting_bits > slice_max_size && !h->mb.b_reencode_mb )
2012             {
2013                 if( mb_xy != h->sh.i_first_mb )
2014                 {
2015                     h->stat.frame.i_mv_bits = mv_bits_bak;
2016                     h->stat.frame.i_tex_bits = tex_bits_bak;
2017                     if( h->param.b_cabac )
2018                     {
2019                         memcpy( &h->cabac, &cabac_bak, offsetof(x264_cabac_t, f8_bits_encoded) );
2020                         h->cabac.p[-1] = cabac_prevbyte_bak;
2021                     }
2022                     else
2023                     {
2024                         h->out.bs = bs_bak;
2025                         i_skip = i_skip_bak;
2026                     }
2027                     h->mb.b_reencode_mb = 1;
2028                     h->sh.i_last_mb = mb_xy-1;
2029                     break;
2030                 }
2031                 else
2032                 {
2033                     h->sh.i_last_mb = mb_xy;
2034                     h->mb.b_reencode_mb = 0;
2035                 }
2036             }
2037             else
2038                 h->mb.b_reencode_mb = 0;
2039         }
2040
2041 #if HAVE_VISUALIZE
2042         if( h->param.b_visualize )
2043             x264_visualize_mb( h );
2044 #endif
2045
2046         /* save cache */
2047         x264_macroblock_cache_save( h );
2048
2049         /* accumulate mb stats */
2050         h->stat.frame.i_mb_count[h->mb.i_type]++;
2051
2052         int b_intra = IS_INTRA( h->mb.i_type );
2053         if( h->param.i_log_level >= X264_LOG_INFO || h->param.rc.b_stat_write )
2054         {
2055             if( !b_intra && !IS_SKIP( h->mb.i_type ) && !IS_DIRECT( h->mb.i_type ) )
2056             {
2057                 if( h->mb.i_partition != D_8x8 )
2058                         h->stat.frame.i_mb_partition[h->mb.i_partition] += 4;
2059                     else
2060                         for( int i = 0; i < 4; i++ )
2061                             h->stat.frame.i_mb_partition[h->mb.i_sub_partition[i]] ++;
2062                 if( h->param.i_frame_reference > 1 )
2063                     for( int i_list = 0; i_list <= (h->sh.i_type == SLICE_TYPE_B); i_list++ )
2064                         for( int i = 0; i < 4; i++ )
2065                         {
2066                             int i_ref = h->mb.cache.ref[i_list][ x264_scan8[4*i] ];
2067                             if( i_ref >= 0 )
2068                                 h->stat.frame.i_mb_count_ref[i_list][i_ref] ++;
2069                         }
2070             }
2071         }
2072
2073         if( h->param.i_log_level >= X264_LOG_INFO )
2074         {
2075             if( h->mb.i_cbp_luma | h->mb.i_cbp_chroma )
2076             {
2077                 int cbpsum = (h->mb.i_cbp_luma&1) + ((h->mb.i_cbp_luma>>1)&1)
2078                            + ((h->mb.i_cbp_luma>>2)&1) + (h->mb.i_cbp_luma>>3);
2079                 h->stat.frame.i_mb_cbp[!b_intra + 0] += cbpsum;
2080                 h->stat.frame.i_mb_cbp[!b_intra + 2] += !!h->mb.i_cbp_chroma;
2081                 h->stat.frame.i_mb_cbp[!b_intra + 4] += h->mb.i_cbp_chroma >> 1;
2082             }
2083             if( h->mb.i_cbp_luma && !b_intra )
2084             {
2085                 h->stat.frame.i_mb_count_8x8dct[0] ++;
2086                 h->stat.frame.i_mb_count_8x8dct[1] += h->mb.b_transform_8x8;
2087             }
2088             if( b_intra && h->mb.i_type != I_PCM )
2089             {
2090                 if( h->mb.i_type == I_16x16 )
2091                     h->stat.frame.i_mb_pred_mode[0][h->mb.i_intra16x16_pred_mode]++;
2092                 else if( h->mb.i_type == I_8x8 )
2093                     for( int i = 0; i < 16; i += 4 )
2094                         h->stat.frame.i_mb_pred_mode[1][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
2095                 else //if( h->mb.i_type == I_4x4 )
2096                     for( int i = 0; i < 16; i++ )
2097                         h->stat.frame.i_mb_pred_mode[2][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
2098                 h->stat.frame.i_mb_pred_mode[3][x264_mb_pred_mode8x8c_fix[h->mb.i_chroma_pred_mode]]++;
2099             }
2100         }
2101
2102         /* calculate deblock strength values (actual deblocking is done per-row along with hpel) */
2103         if( b_deblock )
2104         {
2105             int mvy_limit = 4 >> h->sh.b_mbaff;
2106             uint8_t (*bs)[4][4] = h->deblock_strength[h->mb.i_mb_y&1][h->mb.i_mb_x];
2107             x264_macroblock_cache_load_deblock( h );
2108             if( IS_INTRA( h->mb.type[h->mb.i_mb_xy] ) )
2109                 memset( bs, 3, 2*4*4*sizeof(uint8_t) );
2110             else
2111                 h->loopf.deblock_strength( h->mb.cache.non_zero_count, h->mb.cache.ref, h->mb.cache.mv,
2112                                            bs, mvy_limit, h->sh.i_type == SLICE_TYPE_B );
2113         }
2114
2115         x264_ratecontrol_mb( h, mb_size );
2116
2117         if( h->sh.b_mbaff )
2118         {
2119             i_mb_x += i_mb_y & 1;
2120             i_mb_y ^= i_mb_x < h->mb.i_mb_width;
2121         }
2122         else
2123             i_mb_x++;
2124         if( i_mb_x == h->mb.i_mb_width )
2125         {
2126             i_mb_y++;
2127             i_mb_x = 0;
2128         }
2129     }
2130     h->out.nal[h->out.i_nal].i_last_mb = h->sh.i_last_mb;
2131
2132     if( h->param.b_cabac )
2133     {
2134         x264_cabac_encode_flush( h, &h->cabac );
2135         h->out.bs.p = h->cabac.p;
2136     }
2137     else
2138     {
2139         if( i_skip > 0 )
2140             bs_write_ue( &h->out.bs, i_skip );  /* last skip run */
2141         /* rbsp_slice_trailing_bits */
2142         bs_rbsp_trailing( &h->out.bs );
2143         bs_flush( &h->out.bs );
2144     }
2145     if( x264_nal_end( h ) )
2146         return -1;
2147
2148     if( h->sh.i_last_mb == (h->i_threadslice_end * h->mb.i_mb_width - 1) )
2149     {
2150         h->stat.frame.i_misc_bits = bs_pos( &h->out.bs )
2151                                   + (h->out.i_nal*NALU_OVERHEAD * 8)
2152                                   - h->stat.frame.i_tex_bits
2153                                   - h->stat.frame.i_mv_bits;
2154         x264_fdec_filter_row( h, h->i_threadslice_end, 1 );
2155     }
2156
2157     return 0;
2158 }
2159
2160 static void x264_thread_sync_context( x264_t *dst, x264_t *src )
2161 {
2162     if( dst == src )
2163         return;
2164
2165     // reference counting
2166     for( x264_frame_t **f = src->frames.reference; *f; f++ )
2167         (*f)->i_reference_count++;
2168     for( x264_frame_t **f = dst->frames.reference; *f; f++ )
2169         x264_frame_push_unused( src, *f );
2170     src->fdec->i_reference_count++;
2171     x264_frame_push_unused( src, dst->fdec );
2172
2173     // copy everything except the per-thread pointers and the constants.
2174     memcpy( &dst->i_frame, &src->i_frame, offsetof(x264_t, mb.type) - offsetof(x264_t, i_frame) );
2175     dst->param = src->param;
2176     dst->stat = src->stat;
2177 }
2178
2179 static void x264_thread_sync_stat( x264_t *dst, x264_t *src )
2180 {
2181     if( dst == src )
2182         return;
2183     memcpy( &dst->stat.i_frame_count, &src->stat.i_frame_count, sizeof(dst->stat) - sizeof(dst->stat.frame) );
2184 }
2185
2186 static void *x264_slices_write( x264_t *h )
2187 {
2188     int i_slice_num = 0;
2189     int last_thread_mb = h->sh.i_last_mb;
2190
2191 #if HAVE_VISUALIZE
2192     if( h->param.b_visualize )
2193         if( x264_visualize_init( h ) )
2194             return (void *)-1;
2195 #endif
2196
2197     /* init stats */
2198     memset( &h->stat.frame, 0, sizeof(h->stat.frame) );
2199     h->mb.b_reencode_mb = 0;
2200     while( h->sh.i_first_mb <= last_thread_mb )
2201     {
2202         h->sh.i_last_mb = last_thread_mb;
2203         if( h->param.i_slice_max_mbs )
2204             h->sh.i_last_mb = h->sh.i_first_mb + h->param.i_slice_max_mbs - 1;
2205         else if( h->param.i_slice_count && !h->param.b_sliced_threads )
2206         {
2207             int height = h->mb.i_mb_height >> h->param.b_interlaced;
2208             int width = h->mb.i_mb_width << h->param.b_interlaced;
2209             i_slice_num++;
2210             h->sh.i_last_mb = (height * i_slice_num + h->param.i_slice_count/2) / h->param.i_slice_count * width - 1;
2211         }
2212         h->sh.i_last_mb = X264_MIN( h->sh.i_last_mb, last_thread_mb );
2213         if( x264_stack_align( x264_slice_write, h ) )
2214             return (void *)-1;
2215         h->sh.i_first_mb = h->sh.i_last_mb + 1;
2216     }
2217
2218 #if HAVE_VISUALIZE
2219     if( h->param.b_visualize )
2220     {
2221         x264_visualize_show( h );
2222         x264_visualize_close( h );
2223     }
2224 #endif
2225
2226     return (void *)0;
2227 }
2228
2229 static int x264_threaded_slices_write( x264_t *h )
2230 {
2231     /* set first/last mb and sync contexts */
2232     for( int i = 0; i < h->param.i_threads; i++ )
2233     {
2234         x264_t *t = h->thread[i];
2235         if( i )
2236         {
2237             t->param = h->param;
2238             memcpy( &t->i_frame, &h->i_frame, offsetof(x264_t, rc) - offsetof(x264_t, i_frame) );
2239         }
2240         int height = h->mb.i_mb_height >> h->param.b_interlaced;
2241         t->i_threadslice_start = ((height *  i    + h->param.i_slice_count/2) / h->param.i_threads) << h->param.b_interlaced;
2242         t->i_threadslice_end   = ((height * (i+1) + h->param.i_slice_count/2) / h->param.i_threads) << h->param.b_interlaced;
2243         t->sh.i_first_mb = t->i_threadslice_start * h->mb.i_mb_width;
2244         t->sh.i_last_mb  =   t->i_threadslice_end * h->mb.i_mb_width - 1;
2245     }
2246
2247     x264_stack_align( x264_analyse_weight_frame, h, h->mb.i_mb_height*16 + 16 );
2248
2249     x264_threads_distribute_ratecontrol( h );
2250
2251     /* dispatch */
2252     for( int i = 0; i < h->param.i_threads; i++ )
2253     {
2254         x264_threadpool_run( h->threadpool, (void*)x264_slices_write, h->thread[i] );
2255         h->thread[i]->b_thread_active = 1;
2256     }
2257     for( int i = 0; i < h->param.i_threads; i++ )
2258     {
2259         h->thread[i]->b_thread_active = 0;
2260         if( (intptr_t)x264_threadpool_wait( h->threadpool, h->thread[i] ) )
2261             return -1;
2262     }
2263
2264     /* Go back and fix up the hpel on the borders between slices. */
2265     for( int i = 1; i < h->param.i_threads; i++ )
2266     {
2267         x264_fdec_filter_row( h->thread[i], h->thread[i]->i_threadslice_start + 1, 0 );
2268         if( h->sh.b_mbaff )
2269             x264_fdec_filter_row( h->thread[i], h->thread[i]->i_threadslice_start + 2, 0 );
2270     }
2271
2272     x264_threads_merge_ratecontrol( h );
2273
2274     for( int i = 1; i < h->param.i_threads; i++ )
2275     {
2276         x264_t *t = h->thread[i];
2277         for( int j = 0; j < t->out.i_nal; j++ )
2278         {
2279             h->out.nal[h->out.i_nal] = t->out.nal[j];
2280             h->out.i_nal++;
2281             x264_nal_check_buffer( h );
2282         }
2283         /* All entries in stat.frame are ints except for ssd/ssim. */
2284         for( int j = 0; j < (offsetof(x264_t,stat.frame.i_ssd) - offsetof(x264_t,stat.frame.i_mv_bits)) / sizeof(int); j++ )
2285             ((int*)&h->stat.frame)[j] += ((int*)&t->stat.frame)[j];
2286         for( int j = 0; j < 3; j++ )
2287             h->stat.frame.i_ssd[j] += t->stat.frame.i_ssd[j];
2288         h->stat.frame.f_ssim += t->stat.frame.f_ssim;
2289     }
2290
2291     return 0;
2292 }
2293
2294 void x264_encoder_intra_refresh( x264_t *h )
2295 {
2296     h = h->thread[h->i_thread_phase];
2297     h->b_queued_intra_refresh = 1;
2298 }
2299
2300 int x264_encoder_invalidate_reference( x264_t *h, int64_t pts )
2301 {
2302     if( h->param.i_bframe )
2303     {
2304         x264_log( h, X264_LOG_ERROR, "x264_encoder_invalidate_reference is not supported with B-frames enabled\n" );
2305         return -1;
2306     }
2307     if( h->param.b_intra_refresh )
2308     {
2309         x264_log( h, X264_LOG_ERROR, "x264_encoder_invalidate_reference is not supported with intra refresh enabled\n" );
2310         return -1;
2311     }
2312     h = h->thread[h->i_thread_phase];
2313     if( pts >= h->i_last_idr_pts )
2314     {
2315         for( int i = 0; h->frames.reference[i]; i++ )
2316             if( pts <= h->frames.reference[i]->i_pts )
2317                 h->frames.reference[i]->b_corrupt = 1;
2318         if( pts <= h->fdec->i_pts )
2319             h->fdec->b_corrupt = 1;
2320     }
2321     return 0;
2322 }
2323
2324 /****************************************************************************
2325  * x264_encoder_encode:
2326  *  XXX: i_poc   : is the poc of the current given picture
2327  *       i_frame : is the number of the frame being coded
2328  *  ex:  type frame poc
2329  *       I      0   2*0
2330  *       P      1   2*3
2331  *       B      2   2*1
2332  *       B      3   2*2
2333  *       P      4   2*6
2334  *       B      5   2*4
2335  *       B      6   2*5
2336  ****************************************************************************/
2337 int     x264_encoder_encode( x264_t *h,
2338                              x264_nal_t **pp_nal, int *pi_nal,
2339                              x264_picture_t *pic_in,
2340                              x264_picture_t *pic_out )
2341 {
2342     x264_t *thread_current, *thread_prev, *thread_oldest;
2343     int i_nal_type, i_nal_ref_idc, i_global_qp;
2344     int overhead = NALU_OVERHEAD;
2345
2346     if( h->i_thread_frames > 1 )
2347     {
2348         thread_prev    = h->thread[ h->i_thread_phase ];
2349         h->i_thread_phase = (h->i_thread_phase + 1) % h->i_thread_frames;
2350         thread_current = h->thread[ h->i_thread_phase ];
2351         thread_oldest  = h->thread[ (h->i_thread_phase + 1) % h->i_thread_frames ];
2352         x264_thread_sync_context( thread_current, thread_prev );
2353         x264_thread_sync_ratecontrol( thread_current, thread_prev, thread_oldest );
2354         h = thread_current;
2355     }
2356     else
2357     {
2358         thread_current =
2359         thread_oldest  = h;
2360     }
2361 #if HAVE_MMX
2362     if( h->i_thread_frames == 1 && h->param.cpu&X264_CPU_SSE_MISALIGN )
2363         x264_cpu_mask_misalign_sse();
2364 #endif
2365
2366     // ok to call this before encoding any frames, since the initial values of fdec have b_kept_as_ref=0
2367     if( x264_reference_update( h ) )
2368         return -1;
2369     h->fdec->i_lines_completed = -1;
2370
2371     /* no data out */
2372     *pi_nal = 0;
2373     *pp_nal = NULL;
2374
2375     /* ------------------- Setup new frame from picture -------------------- */
2376     if( pic_in != NULL )
2377     {
2378         /* 1: Copy the picture to a frame and move it to a buffer */
2379         x264_frame_t *fenc = x264_frame_pop_unused( h, 0 );
2380         if( !fenc )
2381             return -1;
2382
2383         if( x264_frame_copy_picture( h, fenc, pic_in ) < 0 )
2384             return -1;
2385
2386         if( h->param.i_width != 16 * h->mb.i_mb_width ||
2387             h->param.i_height != 16 * h->mb.i_mb_height )
2388             x264_frame_expand_border_mod16( h, fenc );
2389
2390         fenc->i_frame = h->frames.i_input++;
2391
2392         if( fenc->i_frame == 0 )
2393             h->frames.i_first_pts = fenc->i_pts;
2394         if( h->frames.i_bframe_delay && fenc->i_frame == h->frames.i_bframe_delay )
2395             h->frames.i_bframe_delay_time = fenc->i_pts - h->frames.i_first_pts;
2396
2397         if( h->param.b_vfr_input && fenc->i_pts <= h->frames.i_largest_pts )
2398             x264_log( h, X264_LOG_WARNING, "non-strictly-monotonic PTS\n" );
2399
2400         h->frames.i_second_largest_pts = h->frames.i_largest_pts;
2401         h->frames.i_largest_pts = fenc->i_pts;
2402
2403         if( (fenc->i_pic_struct < PIC_STRUCT_AUTO) || (fenc->i_pic_struct > PIC_STRUCT_TRIPLE) )
2404             fenc->i_pic_struct = PIC_STRUCT_AUTO;
2405
2406         if( fenc->i_pic_struct == PIC_STRUCT_AUTO )
2407         {
2408             int b_interlaced = fenc->param ? fenc->param->b_interlaced : h->param.b_interlaced;
2409             if( b_interlaced )
2410             {
2411                 int b_tff = fenc->param ? fenc->param->b_tff : h->param.b_tff;
2412                 fenc->i_pic_struct = b_tff ? PIC_STRUCT_TOP_BOTTOM : PIC_STRUCT_BOTTOM_TOP;
2413             }
2414             else
2415                 fenc->i_pic_struct = PIC_STRUCT_PROGRESSIVE;
2416         }
2417
2418         if( h->param.rc.b_mb_tree && h->param.rc.b_stat_read )
2419         {
2420             if( x264_macroblock_tree_read( h, fenc, pic_in->prop.quant_offsets ) )
2421                 return -1;
2422         }
2423         else
2424             x264_stack_align( x264_adaptive_quant_frame, h, fenc, pic_in->prop.quant_offsets );
2425
2426         if( pic_in->prop.quant_offsets_free )
2427             pic_in->prop.quant_offsets_free( pic_in->prop.quant_offsets );
2428
2429         if( h->frames.b_have_lowres )
2430             x264_frame_init_lowres( h, fenc );
2431
2432         /* 2: Place the frame into the queue for its slice type decision */
2433         x264_lookahead_put_frame( h, fenc );
2434
2435         if( h->frames.i_input <= h->frames.i_delay + 1 - h->i_thread_frames )
2436         {
2437             /* Nothing yet to encode, waiting for filling of buffers */
2438             pic_out->i_type = X264_TYPE_AUTO;
2439             return 0;
2440         }
2441     }
2442     else
2443     {
2444         /* signal kills for lookahead thread */
2445         x264_pthread_mutex_lock( &h->lookahead->ifbuf.mutex );
2446         h->lookahead->b_exit_thread = 1;
2447         x264_pthread_cond_broadcast( &h->lookahead->ifbuf.cv_fill );
2448         x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
2449     }
2450
2451     h->i_frame++;
2452     /* 3: The picture is analyzed in the lookahead */
2453     if( !h->frames.current[0] )
2454         x264_lookahead_get_frames( h );
2455
2456     if( !h->frames.current[0] && x264_lookahead_is_empty( h ) )
2457         return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
2458
2459     /* ------------------- Get frame to be encoded ------------------------- */
2460     /* 4: get picture to encode */
2461     h->fenc = x264_frame_shift( h->frames.current );
2462     if( h->i_frame == h->i_thread_frames - 1 )
2463         h->i_reordered_pts_delay = h->fenc->i_reordered_pts;
2464     if( h->fenc->param )
2465     {
2466         x264_encoder_reconfig( h, h->fenc->param );
2467         if( h->fenc->param->param_free )
2468             h->fenc->param->param_free( h->fenc->param );
2469     }
2470
2471     if( !IS_X264_TYPE_I( h->fenc->i_type ) )
2472     {
2473         int valid_refs_left = 0;
2474         for( int i = 0; h->frames.reference[i]; i++ )
2475             if( !h->frames.reference[i]->b_corrupt )
2476                 valid_refs_left++;
2477         /* No valid reference frames left: force an IDR. */
2478         if( !valid_refs_left )
2479         {
2480             h->fenc->b_keyframe = 1;
2481             h->fenc->i_type = X264_TYPE_IDR;
2482         }
2483     }
2484
2485     if( h->fenc->b_keyframe )
2486     {
2487         h->frames.i_last_keyframe = h->fenc->i_frame;
2488         if( h->fenc->i_type == X264_TYPE_IDR )
2489         {
2490             h->i_frame_num = 0;
2491             h->frames.i_last_idr = h->fenc->i_frame;
2492         }
2493     }
2494     h->sh.i_mmco_command_count =
2495     h->sh.i_mmco_remove_from_end = 0;
2496     h->b_ref_reorder[0] =
2497     h->b_ref_reorder[1] = 0;
2498     h->fdec->i_poc =
2499     h->fenc->i_poc = 2 * ( h->fenc->i_frame - X264_MAX( h->frames.i_last_idr, 0 ) );
2500
2501     /* ------------------- Setup frame context ----------------------------- */
2502     /* 5: Init data dependent of frame type */
2503     if( h->fenc->i_type == X264_TYPE_IDR )
2504     {
2505         /* reset ref pictures */
2506         i_nal_type    = NAL_SLICE_IDR;
2507         i_nal_ref_idc = NAL_PRIORITY_HIGHEST;
2508         h->sh.i_type = SLICE_TYPE_I;
2509         x264_reference_reset( h );
2510         h->frames.i_poc_last_open_gop = -1;
2511     }
2512     else if( h->fenc->i_type == X264_TYPE_I )
2513     {
2514         i_nal_type    = NAL_SLICE;
2515         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
2516         h->sh.i_type = SLICE_TYPE_I;
2517         x264_reference_hierarchy_reset( h );
2518         if( h->param.i_open_gop )
2519             h->frames.i_poc_last_open_gop = h->fenc->b_keyframe ? h->fenc->i_poc : -1;
2520     }
2521     else if( h->fenc->i_type == X264_TYPE_P )
2522     {
2523         i_nal_type    = NAL_SLICE;
2524         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
2525         h->sh.i_type = SLICE_TYPE_P;
2526         x264_reference_hierarchy_reset( h );
2527         h->frames.i_poc_last_open_gop = -1;
2528     }
2529     else if( h->fenc->i_type == X264_TYPE_BREF )
2530     {
2531         i_nal_type    = NAL_SLICE;
2532         i_nal_ref_idc = h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT ? NAL_PRIORITY_LOW : NAL_PRIORITY_HIGH;
2533         h->sh.i_type = SLICE_TYPE_B;
2534         x264_reference_hierarchy_reset( h );
2535     }
2536     else    /* B frame */
2537     {
2538         i_nal_type    = NAL_SLICE;
2539         i_nal_ref_idc = NAL_PRIORITY_DISPOSABLE;
2540         h->sh.i_type = SLICE_TYPE_B;
2541     }
2542
2543     h->fdec->i_type = h->fenc->i_type;
2544     h->fdec->i_frame = h->fenc->i_frame;
2545     h->fenc->b_kept_as_ref =
2546     h->fdec->b_kept_as_ref = i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE && h->param.i_keyint_max > 1;
2547
2548     h->fdec->i_pts = h->fenc->i_pts;
2549     if( h->frames.i_bframe_delay )
2550     {
2551         int64_t *prev_reordered_pts = thread_current->frames.i_prev_reordered_pts;
2552         h->fdec->i_dts = h->i_frame > h->frames.i_bframe_delay
2553                        ? prev_reordered_pts[ (h->i_frame - h->frames.i_bframe_delay) % h->frames.i_bframe_delay ]
2554                        : h->fenc->i_reordered_pts - h->frames.i_bframe_delay_time;
2555         prev_reordered_pts[ h->i_frame % h->frames.i_bframe_delay ] = h->fenc->i_reordered_pts;
2556     }
2557     else
2558         h->fdec->i_dts = h->fenc->i_reordered_pts;
2559     if( h->fenc->i_type == X264_TYPE_IDR )
2560         h->i_last_idr_pts = h->fdec->i_pts;
2561
2562     /* ------------------- Init                ----------------------------- */
2563     /* build ref list 0/1 */
2564     x264_reference_build_list( h, h->fdec->i_poc );
2565
2566     /* ---------------------- Write the bitstream -------------------------- */
2567     /* Init bitstream context */
2568     if( h->param.b_sliced_threads )
2569     {
2570         for( int i = 0; i < h->param.i_threads; i++ )
2571         {
2572             bs_init( &h->thread[i]->out.bs, h->thread[i]->out.p_bitstream, h->thread[i]->out.i_bitstream );
2573             h->thread[i]->out.i_nal = 0;
2574         }
2575     }
2576     else
2577     {
2578         bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
2579         h->out.i_nal = 0;
2580     }
2581
2582     if( h->param.b_aud )
2583     {
2584         int pic_type;
2585
2586         if( h->sh.i_type == SLICE_TYPE_I )
2587             pic_type = 0;
2588         else if( h->sh.i_type == SLICE_TYPE_P )
2589             pic_type = 1;
2590         else if( h->sh.i_type == SLICE_TYPE_B )
2591             pic_type = 2;
2592         else
2593             pic_type = 7;
2594
2595         x264_nal_start( h, NAL_AUD, NAL_PRIORITY_DISPOSABLE );
2596         bs_write( &h->out.bs, 3, pic_type );
2597         bs_rbsp_trailing( &h->out.bs );
2598         if( x264_nal_end( h ) )
2599             return -1;
2600         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2601     }
2602
2603     h->i_nal_type = i_nal_type;
2604     h->i_nal_ref_idc = i_nal_ref_idc;
2605
2606     if( h->param.b_intra_refresh )
2607     {
2608         if( IS_X264_TYPE_I( h->fenc->i_type ) )
2609         {
2610             h->fdec->i_frames_since_pir = 0;
2611             h->b_queued_intra_refresh = 0;
2612             /* PIR is currently only supported with ref == 1, so any intra frame effectively refreshes
2613              * the whole frame and counts as an intra refresh. */
2614             h->fdec->f_pir_position = h->mb.i_mb_width;
2615         }
2616         else if( h->fenc->i_type == X264_TYPE_P )
2617         {
2618             int pocdiff = (h->fdec->i_poc - h->fref[0][0]->i_poc)/2;
2619             float increment = X264_MAX( ((float)h->mb.i_mb_width-1) / h->param.i_keyint_max, 1 );
2620             h->fdec->f_pir_position = h->fref[0][0]->f_pir_position;
2621             h->fdec->i_frames_since_pir = h->fref[0][0]->i_frames_since_pir + pocdiff;
2622             if( h->fdec->i_frames_since_pir >= h->param.i_keyint_max ||
2623                 (h->b_queued_intra_refresh && h->fdec->f_pir_position + 0.5 >= h->mb.i_mb_width) )
2624             {
2625                 h->fdec->f_pir_position = 0;
2626                 h->fdec->i_frames_since_pir = 0;
2627                 h->b_queued_intra_refresh = 0;
2628                 h->fenc->b_keyframe = 1;
2629             }
2630             h->fdec->i_pir_start_col = h->fdec->f_pir_position+0.5;
2631             h->fdec->f_pir_position += increment * pocdiff;
2632             h->fdec->i_pir_end_col = h->fdec->f_pir_position+0.5;
2633             /* If our intra refresh has reached the right side of the frame, we're done. */
2634             if( h->fdec->i_pir_end_col >= h->mb.i_mb_width - 1 )
2635                 h->fdec->f_pir_position = h->mb.i_mb_width;
2636         }
2637     }
2638
2639     if( h->fenc->b_keyframe )
2640     {
2641         /* Write SPS and PPS */
2642         if( h->param.b_repeat_headers )
2643         {
2644             /* generate sequence parameters */
2645             x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
2646             x264_sps_write( &h->out.bs, h->sps );
2647             if( x264_nal_end( h ) )
2648                 return -1;
2649             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
2650
2651             /* generate picture parameters */
2652             x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
2653             x264_pps_write( &h->out.bs, h->pps );
2654             if( x264_nal_end( h ) )
2655                 return -1;
2656             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
2657         }
2658
2659         /* buffering period sei is written in x264_encoder_frame_end */
2660     }
2661
2662     /* write extra sei */
2663     for( int i = 0; i < h->fenc->extra_sei.num_payloads; i++ )
2664     {
2665         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2666         x264_sei_write( &h->out.bs, h->fenc->extra_sei.payloads[i].payload, h->fenc->extra_sei.payloads[i].payload_size,
2667                         h->fenc->extra_sei.payloads[i].payload_type );
2668         if( x264_nal_end( h ) )
2669             return -1;
2670         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2671         if( h->fenc->extra_sei.sei_free && h->fenc->extra_sei.payloads[i].payload )
2672             h->fenc->extra_sei.sei_free( h->fenc->extra_sei.payloads[i].payload );
2673     }
2674
2675     if( h->fenc->extra_sei.sei_free && h->fenc->extra_sei.payloads )
2676         h->fenc->extra_sei.sei_free( h->fenc->extra_sei.payloads );
2677
2678     if( h->fenc->b_keyframe )
2679     {
2680         if( h->param.b_repeat_headers && h->fenc->i_frame == 0 )
2681         {
2682             /* identify ourself */
2683             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2684             if( x264_sei_version_write( h, &h->out.bs ) )
2685                 return -1;
2686             if( x264_nal_end( h ) )
2687                 return -1;
2688             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2689         }
2690
2691         if( h->fenc->i_type != X264_TYPE_IDR )
2692         {
2693             int time_to_recovery = h->param.i_open_gop ? 0 : X264_MIN( h->mb.i_mb_width - 1, h->param.i_keyint_max ) + h->param.i_bframe - 1;
2694             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2695             x264_sei_recovery_point_write( h, &h->out.bs, time_to_recovery );
2696             if( x264_nal_end( h ) )
2697                 return -1;
2698             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2699         }
2700
2701         if ( h->param.i_frame_packing >= 0 )
2702         {
2703             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2704             x264_sei_frame_packing_write( h, &h->out.bs );
2705             if( x264_nal_end( h ) )
2706                 return -1;
2707             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2708         }
2709     }
2710
2711     /* generate sei pic timing */
2712     if( h->sps->vui.b_pic_struct_present || h->sps->vui.b_nal_hrd_parameters_present )
2713     {
2714         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2715         x264_sei_pic_timing_write( h, &h->out.bs );
2716         if( x264_nal_end( h ) )
2717             return -1;
2718         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2719     }
2720
2721     if( h->fenc->b_keyframe && h->param.b_intra_refresh )
2722         h->i_cpb_delay_pir_offset = h->fenc->i_cpb_delay;
2723
2724     /* Init the rate control */
2725     /* FIXME: Include slice header bit cost. */
2726     x264_ratecontrol_start( h, h->fenc->i_qpplus1, overhead*8 );
2727     i_global_qp = x264_ratecontrol_qp( h );
2728
2729     pic_out->i_qpplus1 =
2730     h->fdec->i_qpplus1 = i_global_qp + 1;
2731
2732     if( h->param.rc.b_stat_read && h->sh.i_type != SLICE_TYPE_I )
2733     {
2734         x264_reference_build_list_optimal( h );
2735         x264_reference_check_reorder( h );
2736     }
2737
2738     if( h->i_ref[0] )
2739         h->fdec->i_poc_l0ref0 = h->fref[0][0]->i_poc;
2740
2741     /* ------------------------ Create slice header  ----------------------- */
2742     x264_slice_init( h, i_nal_type, i_global_qp );
2743
2744     /*------------------------- Weights -------------------------------------*/
2745     if( h->sh.i_type == SLICE_TYPE_B )
2746         x264_macroblock_bipred_init( h );
2747
2748     x264_weighted_pred_init( h );
2749
2750     if( i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE )
2751         h->i_frame_num++;
2752
2753     /* Write frame */
2754     h->i_threadslice_start = 0;
2755     h->i_threadslice_end = h->mb.i_mb_height;
2756     if( h->i_thread_frames > 1 )
2757     {
2758         x264_threadpool_run( h->threadpool, (void*)x264_slices_write, h );
2759         h->b_thread_active = 1;
2760     }
2761     else if( h->param.b_sliced_threads )
2762     {
2763         if( x264_threaded_slices_write( h ) )
2764             return -1;
2765     }
2766     else
2767         if( (intptr_t)x264_slices_write( h ) )
2768             return -1;
2769
2770     return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
2771 }
2772
2773 static int x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
2774                                    x264_nal_t **pp_nal, int *pi_nal,
2775                                    x264_picture_t *pic_out )
2776 {
2777     char psz_message[80];
2778
2779     if( h->b_thread_active )
2780     {
2781         h->b_thread_active = 0;
2782         if( (intptr_t)x264_threadpool_wait( h->threadpool, h ) )
2783             return -1;
2784     }
2785     if( !h->out.i_nal )
2786     {
2787         pic_out->i_type = X264_TYPE_AUTO;
2788         return 0;
2789     }
2790
2791     x264_emms();
2792     /* generate sei buffering period and insert it into place */
2793     if( h->fenc->b_keyframe && h->sps->vui.b_nal_hrd_parameters_present )
2794     {
2795         x264_hrd_fullness( h );
2796         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2797         x264_sei_buffering_period_write( h, &h->out.bs );
2798         if( x264_nal_end( h ) )
2799            return -1;
2800         /* buffering period sei must follow AUD, SPS and PPS and precede all other SEIs */
2801         int idx = 0;
2802         while( h->out.nal[idx].i_type == NAL_AUD ||
2803                h->out.nal[idx].i_type == NAL_SPS ||
2804                h->out.nal[idx].i_type == NAL_PPS )
2805             idx++;
2806         x264_nal_t nal_tmp = h->out.nal[h->out.i_nal-1];
2807         memmove( &h->out.nal[idx+1], &h->out.nal[idx], (h->out.i_nal-idx-1)*sizeof(x264_nal_t) );
2808         h->out.nal[idx] = nal_tmp;
2809     }
2810
2811     int frame_size = x264_encoder_encapsulate_nals( h, 0 );
2812
2813     /* Set output picture properties */
2814     pic_out->i_type = h->fenc->i_type;
2815
2816     pic_out->b_keyframe = h->fenc->b_keyframe;
2817
2818     pic_out->i_pts = h->fdec->i_pts;
2819     pic_out->i_dts = h->fdec->i_dts;
2820
2821     if( pic_out->i_pts < pic_out->i_dts )
2822         x264_log( h, X264_LOG_WARNING, "invalid DTS: PTS is less than DTS\n" );
2823
2824     pic_out->img.i_csp = X264_CSP_NV12;
2825 #if HIGH_BIT_DEPTH
2826     pic_out->img.i_csp |= X264_CSP_HIGH_DEPTH;
2827 #endif
2828     pic_out->img.i_plane = h->fdec->i_plane;
2829     for( int i = 0; i < 2; i++ )
2830     {
2831         pic_out->img.i_stride[i] = h->fdec->i_stride[i] * sizeof(pixel);
2832         pic_out->img.plane[i] = (uint8_t*)h->fdec->plane[i];
2833     }
2834
2835     x264_frame_push_unused( thread_current, h->fenc );
2836
2837     /* ---------------------- Update encoder state ------------------------- */
2838
2839     /* update rc */
2840     int filler = 0;
2841     if( x264_ratecontrol_end( h, frame_size * 8, &filler ) < 0 )
2842         return -1;
2843
2844     pic_out->hrd_timing = h->fenc->hrd_timing;
2845
2846     while( filler > 0 )
2847     {
2848         int f, overhead;
2849         overhead = (FILLER_OVERHEAD - h->param.b_annexb);
2850         if( h->param.i_slice_max_size && filler > h->param.i_slice_max_size )
2851         {
2852             int next_size = filler - h->param.i_slice_max_size;
2853             int overflow = X264_MAX( overhead - next_size, 0 );
2854             f = h->param.i_slice_max_size - overhead - overflow;
2855         }
2856         else
2857             f = X264_MAX( 0, filler - overhead );
2858
2859         x264_nal_start( h, NAL_FILLER, NAL_PRIORITY_DISPOSABLE );
2860         x264_filler_write( h, &h->out.bs, f );
2861         if( x264_nal_end( h ) )
2862             return -1;
2863         int total_size = x264_encoder_encapsulate_nals( h, h->out.i_nal-1 );
2864         frame_size += total_size;
2865         filler -= total_size;
2866     }
2867
2868     /* End bitstream, set output  */
2869     *pi_nal = h->out.i_nal;
2870     *pp_nal = h->out.nal;
2871
2872     h->out.i_nal = 0;
2873
2874     x264_noise_reduction_update( h );
2875
2876     /* ---------------------- Compute/Print statistics --------------------- */
2877     x264_thread_sync_stat( h, h->thread[0] );
2878
2879     /* Slice stat */
2880     h->stat.i_frame_count[h->sh.i_type]++;
2881     h->stat.i_frame_size[h->sh.i_type] += frame_size;
2882     h->stat.f_frame_qp[h->sh.i_type] += h->fdec->f_qp_avg_aq;
2883
2884     for( int i = 0; i < X264_MBTYPE_MAX; i++ )
2885         h->stat.i_mb_count[h->sh.i_type][i] += h->stat.frame.i_mb_count[i];
2886     for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
2887         h->stat.i_mb_partition[h->sh.i_type][i] += h->stat.frame.i_mb_partition[i];
2888     for( int i = 0; i < 2; i++ )
2889         h->stat.i_mb_count_8x8dct[i] += h->stat.frame.i_mb_count_8x8dct[i];
2890     for( int i = 0; i < 6; i++ )
2891         h->stat.i_mb_cbp[i] += h->stat.frame.i_mb_cbp[i];
2892     for( int i = 0; i < 4; i++ )
2893         for( int j = 0; j < 13; j++ )
2894             h->stat.i_mb_pred_mode[i][j] += h->stat.frame.i_mb_pred_mode[i][j];
2895     if( h->sh.i_type != SLICE_TYPE_I )
2896         for( int i_list = 0; i_list < 2; i_list++ )
2897             for( int i = 0; i < X264_REF_MAX*2; i++ )
2898                 h->stat.i_mb_count_ref[h->sh.i_type][i_list][i] += h->stat.frame.i_mb_count_ref[i_list][i];
2899     if( h->sh.i_type == SLICE_TYPE_P && h->param.analyse.i_weighted_pred >= X264_WEIGHTP_SIMPLE )
2900     {
2901         h->stat.i_wpred[0] += !!h->sh.weight[0][0].weightfn;
2902         h->stat.i_wpred[1] += !!h->sh.weight[0][1].weightfn || !!h->sh.weight[0][2].weightfn;
2903     }
2904     if( h->sh.i_type == SLICE_TYPE_B )
2905     {
2906         h->stat.i_direct_frames[ h->sh.b_direct_spatial_mv_pred ] ++;
2907         if( h->mb.b_direct_auto_write )
2908         {
2909             //FIXME somewhat arbitrary time constants
2910             if( h->stat.i_direct_score[0] + h->stat.i_direct_score[1] > h->mb.i_mb_count )
2911                 for( int i = 0; i < 2; i++ )
2912                     h->stat.i_direct_score[i] = h->stat.i_direct_score[i] * 9/10;
2913             for( int i = 0; i < 2; i++ )
2914                 h->stat.i_direct_score[i] += h->stat.frame.i_direct_score[i];
2915         }
2916     }
2917     else
2918         h->stat.i_consecutive_bframes[h->fenc->i_bframes]++;
2919
2920     psz_message[0] = '\0';
2921     double dur = h->fenc->f_duration;
2922     h->stat.f_frame_duration[h->sh.i_type] += dur;
2923     if( h->param.analyse.b_psnr )
2924     {
2925         int64_t ssd[3] =
2926         {
2927             h->stat.frame.i_ssd[0],
2928             h->stat.frame.i_ssd[1],
2929             h->stat.frame.i_ssd[2],
2930         };
2931
2932         h->stat.f_ssd_global[h->sh.i_type]   += dur * (ssd[0] + ssd[1] + ssd[2]);
2933         h->stat.f_psnr_average[h->sh.i_type] += dur * x264_psnr( ssd[0] + ssd[1] + ssd[2], 3 * h->param.i_width * h->param.i_height / 2 );
2934         h->stat.f_psnr_mean_y[h->sh.i_type]  += dur * x264_psnr( ssd[0], h->param.i_width * h->param.i_height );
2935         h->stat.f_psnr_mean_u[h->sh.i_type]  += dur * x264_psnr( ssd[1], h->param.i_width * h->param.i_height / 4 );
2936         h->stat.f_psnr_mean_v[h->sh.i_type]  += dur * x264_psnr( ssd[2], h->param.i_width * h->param.i_height / 4 );
2937
2938         snprintf( psz_message, 80, " PSNR Y:%5.2f U:%5.2f V:%5.2f",
2939                   x264_psnr( ssd[0], h->param.i_width * h->param.i_height ),
2940                   x264_psnr( ssd[1], h->param.i_width * h->param.i_height / 4),
2941                   x264_psnr( ssd[2], h->param.i_width * h->param.i_height / 4) );
2942     }
2943
2944     if( h->param.analyse.b_ssim )
2945     {
2946         double ssim_y = h->stat.frame.f_ssim
2947                       / (((h->param.i_width-6)>>2) * ((h->param.i_height-6)>>2));
2948         h->stat.f_ssim_mean_y[h->sh.i_type] += ssim_y * dur;
2949         snprintf( psz_message + strlen(psz_message), 80 - strlen(psz_message),
2950                   " SSIM Y:%.5f", ssim_y );
2951     }
2952     psz_message[79] = '\0';
2953
2954     x264_log( h, X264_LOG_DEBUG,
2955                   "frame=%4d QP=%.2f NAL=%d Slice:%c Poc:%-3d I:%-4d P:%-4d SKIP:%-4d size=%d bytes%s\n",
2956               h->i_frame,
2957               h->fdec->f_qp_avg_aq,
2958               h->i_nal_ref_idc,
2959               h->sh.i_type == SLICE_TYPE_I ? 'I' : (h->sh.i_type == SLICE_TYPE_P ? 'P' : 'B' ),
2960               h->fdec->i_poc,
2961               h->stat.frame.i_mb_count_i,
2962               h->stat.frame.i_mb_count_p,
2963               h->stat.frame.i_mb_count_skip,
2964               frame_size,
2965               psz_message );
2966
2967     // keep stats all in one place
2968     x264_thread_sync_stat( h->thread[0], h );
2969     // for the use of the next frame
2970     x264_thread_sync_stat( thread_current, h );
2971
2972 #ifdef DEBUG_MB_TYPE
2973 {
2974     static const char mb_chars[] = { 'i', 'i', 'I', 'C', 'P', '8', 'S',
2975         'D', '<', 'X', 'B', 'X', '>', 'B', 'B', 'B', 'B', '8', 'S' };
2976     for( int mb_xy = 0; mb_xy < h->mb.i_mb_width * h->mb.i_mb_height; mb_xy++ )
2977     {
2978         if( h->mb.type[mb_xy] < X264_MBTYPE_MAX && h->mb.type[mb_xy] >= 0 )
2979             fprintf( stderr, "%c ", mb_chars[ h->mb.type[mb_xy] ] );
2980         else
2981             fprintf( stderr, "? " );
2982
2983         if( (mb_xy+1) % h->mb.i_mb_width == 0 )
2984             fprintf( stderr, "\n" );
2985     }
2986 }
2987 #endif
2988
2989     /* Remove duplicates, must be done near the end as breaks h->fref0 array
2990      * by freeing some of its pointers. */
2991     for( int i = 0; i < h->i_ref[0]; i++ )
2992         if( h->fref[0][i] && h->fref[0][i]->b_duplicate )
2993         {
2994             x264_frame_push_blank_unused( h, h->fref[0][i] );
2995             h->fref[0][i] = 0;
2996         }
2997
2998     if( h->param.psz_dump_yuv )
2999         x264_frame_dump( h );
3000     x264_emms();
3001
3002     return frame_size;
3003 }
3004
3005 static void x264_print_intra( int64_t *i_mb_count, double i_count, int b_print_pcm, char *intra )
3006 {
3007     intra += sprintf( intra, "I16..4%s: %4.1f%% %4.1f%% %4.1f%%",
3008         b_print_pcm ? "..PCM" : "",
3009         i_mb_count[I_16x16]/ i_count,
3010         i_mb_count[I_8x8]  / i_count,
3011         i_mb_count[I_4x4]  / i_count );
3012     if( b_print_pcm )
3013         sprintf( intra, " %4.1f%%", i_mb_count[I_PCM]  / i_count );
3014 }
3015
3016 /****************************************************************************
3017  * x264_encoder_close:
3018  ****************************************************************************/
3019 void    x264_encoder_close  ( x264_t *h )
3020 {
3021     int64_t i_yuv_size = 3 * h->param.i_width * h->param.i_height / 2;
3022     int64_t i_mb_count_size[2][7] = {{0}};
3023     char buf[200];
3024     int b_print_pcm = h->stat.i_mb_count[SLICE_TYPE_I][I_PCM]
3025                    || h->stat.i_mb_count[SLICE_TYPE_P][I_PCM]
3026                    || h->stat.i_mb_count[SLICE_TYPE_B][I_PCM];
3027
3028     x264_lookahead_delete( h );
3029
3030     if( h->param.i_threads > 1 )
3031         x264_threadpool_delete( h->threadpool );
3032     if( h->i_thread_frames > 1 )
3033     {
3034         for( int i = 0; i < h->i_thread_frames; i++ )
3035             if( h->thread[i]->b_thread_active )
3036             {
3037                 assert( h->thread[i]->fenc->i_reference_count == 1 );
3038                 x264_frame_delete( h->thread[i]->fenc );
3039             }
3040
3041         x264_t *thread_prev = h->thread[h->i_thread_phase];
3042         x264_thread_sync_ratecontrol( h, thread_prev, h );
3043         x264_thread_sync_ratecontrol( thread_prev, thread_prev, h );
3044         h->i_frame = thread_prev->i_frame + 1 - h->i_thread_frames;
3045     }
3046     h->i_frame++;
3047
3048     /* Slices used and PSNR */
3049     for( int i = 0; i < 3; i++ )
3050     {
3051         static const uint8_t slice_order[] = { SLICE_TYPE_I, SLICE_TYPE_P, SLICE_TYPE_B };
3052         int i_slice = slice_order[i];
3053
3054         if( h->stat.i_frame_count[i_slice] > 0 )
3055         {
3056             int i_count = h->stat.i_frame_count[i_slice];
3057             double dur =  h->stat.f_frame_duration[i_slice];
3058             if( h->param.analyse.b_psnr )
3059             {
3060                 x264_log( h, X264_LOG_INFO,
3061                           "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",
3062                           slice_type_to_char[i_slice],
3063                           i_count,
3064                           h->stat.f_frame_qp[i_slice] / i_count,
3065                           (double)h->stat.i_frame_size[i_slice] / i_count,
3066                           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,
3067                           h->stat.f_psnr_average[i_slice] / dur,
3068                           x264_psnr( h->stat.f_ssd_global[i_slice], dur * i_yuv_size ) );
3069             }
3070             else
3071             {
3072                 x264_log( h, X264_LOG_INFO,
3073                           "frame %c:%-5d Avg QP:%5.2f  size:%6.0f\n",
3074                           slice_type_to_char[i_slice],
3075                           i_count,
3076                           h->stat.f_frame_qp[i_slice] / i_count,
3077                           (double)h->stat.i_frame_size[i_slice] / i_count );
3078             }
3079         }
3080     }
3081     if( h->param.i_bframe && h->stat.i_frame_count[SLICE_TYPE_B] )
3082     {
3083         char *p = buf;
3084         int den = 0;
3085         // weight by number of frames (including the I/P-frames) that are in a sequence of N B-frames
3086         for( int i = 0; i <= h->param.i_bframe; i++ )
3087             den += (i+1) * h->stat.i_consecutive_bframes[i];
3088         for( int i = 0; i <= h->param.i_bframe; i++ )
3089             p += sprintf( p, " %4.1f%%", 100. * (i+1) * h->stat.i_consecutive_bframes[i] / den );
3090         x264_log( h, X264_LOG_INFO, "consecutive B-frames:%s\n", buf );
3091     }
3092
3093     for( int i_type = 0; i_type < 2; i_type++ )
3094         for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
3095         {
3096             if( i == D_DIRECT_8x8 ) continue; /* direct is counted as its own type */
3097             i_mb_count_size[i_type][x264_mb_partition_pixel_table[i]] += h->stat.i_mb_partition[i_type][i];
3098         }
3099
3100     /* MB types used */
3101     if( h->stat.i_frame_count[SLICE_TYPE_I] > 0 )
3102     {
3103         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_I];
3104         double i_count = h->stat.i_frame_count[SLICE_TYPE_I] * h->mb.i_mb_count / 100.0;
3105         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
3106         x264_log( h, X264_LOG_INFO, "mb I  %s\n", buf );
3107     }
3108     if( h->stat.i_frame_count[SLICE_TYPE_P] > 0 )
3109     {
3110         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_P];
3111         double i_count = h->stat.i_frame_count[SLICE_TYPE_P] * h->mb.i_mb_count / 100.0;
3112         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_P];
3113         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
3114         x264_log( h, X264_LOG_INFO,
3115                   "mb P  %s  P16..4: %4.1f%% %4.1f%% %4.1f%% %4.1f%% %4.1f%%    skip:%4.1f%%\n",
3116                   buf,
3117                   i_mb_size[PIXEL_16x16] / (i_count*4),
3118                   (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
3119                   i_mb_size[PIXEL_8x8] / (i_count*4),
3120                   (i_mb_size[PIXEL_8x4] + i_mb_size[PIXEL_4x8]) / (i_count*4),
3121                   i_mb_size[PIXEL_4x4] / (i_count*4),
3122                   i_mb_count[P_SKIP] / i_count );
3123     }
3124     if( h->stat.i_frame_count[SLICE_TYPE_B] > 0 )
3125     {
3126         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_B];
3127         double i_count = h->stat.i_frame_count[SLICE_TYPE_B] * h->mb.i_mb_count / 100.0;
3128         double i_mb_list_count;
3129         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_B];
3130         int64_t list_count[3] = {0}; /* 0 == L0, 1 == L1, 2 == BI */
3131         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
3132         for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
3133             for( int j = 0; j < 2; j++ )
3134             {
3135                 int l0 = x264_mb_type_list_table[i][0][j];
3136                 int l1 = x264_mb_type_list_table[i][1][j];
3137                 if( l0 || l1 )
3138                     list_count[l1+l0*l1] += h->stat.i_mb_count[SLICE_TYPE_B][i] * 2;
3139             }
3140         list_count[0] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L0_8x8];
3141         list_count[1] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L1_8x8];
3142         list_count[2] += h->stat.i_mb_partition[SLICE_TYPE_B][D_BI_8x8];
3143         i_mb_count[B_DIRECT] += (h->stat.i_mb_partition[SLICE_TYPE_B][D_DIRECT_8x8]+2)/4;
3144         i_mb_list_count = (list_count[0] + list_count[1] + list_count[2]) / 100.0;
3145         x264_log( h, X264_LOG_INFO,
3146                   "mb B  %s  B16..8: %4.1f%% %4.1f%% %4.1f%%  direct:%4.1f%%  skip:%4.1f%%  L0:%4.1f%% L1:%4.1f%% BI:%4.1f%%\n",
3147                   buf,
3148                   i_mb_size[PIXEL_16x16] / (i_count*4),
3149                   (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
3150                   i_mb_size[PIXEL_8x8] / (i_count*4),
3151                   i_mb_count[B_DIRECT] / i_count,
3152                   i_mb_count[B_SKIP]   / i_count,
3153                   list_count[0] / i_mb_list_count,
3154                   list_count[1] / i_mb_list_count,
3155                   list_count[2] / i_mb_list_count );
3156     }
3157
3158     x264_ratecontrol_summary( h );
3159
3160     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 )
3161     {
3162 #define SUM3(p) (p[SLICE_TYPE_I] + p[SLICE_TYPE_P] + p[SLICE_TYPE_B])
3163 #define SUM3b(p,o) (p[SLICE_TYPE_I][o] + p[SLICE_TYPE_P][o] + p[SLICE_TYPE_B][o])
3164         int64_t i_i8x8 = SUM3b( h->stat.i_mb_count, I_8x8 );
3165         int64_t i_intra = i_i8x8 + SUM3b( h->stat.i_mb_count, I_4x4 )
3166                                  + SUM3b( h->stat.i_mb_count, I_16x16 );
3167         int64_t i_all_intra = i_intra + SUM3b( h->stat.i_mb_count, I_PCM);
3168         const int i_count = h->stat.i_frame_count[SLICE_TYPE_I] +
3169                             h->stat.i_frame_count[SLICE_TYPE_P] +
3170                             h->stat.i_frame_count[SLICE_TYPE_B];
3171         const double duration = h->stat.f_frame_duration[SLICE_TYPE_I] +
3172                                 h->stat.f_frame_duration[SLICE_TYPE_P] +
3173                                 h->stat.f_frame_duration[SLICE_TYPE_B];
3174         int64_t i_mb_count = (int64_t)i_count * h->mb.i_mb_count;
3175         float f_bitrate = SUM3(h->stat.i_frame_size) / duration / 125;
3176
3177         if( h->pps->b_transform_8x8_mode )
3178         {
3179             buf[0] = 0;
3180             if( h->stat.i_mb_count_8x8dct[0] )
3181                 sprintf( buf, " inter:%.1f%%", 100. * h->stat.i_mb_count_8x8dct[1] / h->stat.i_mb_count_8x8dct[0] );
3182             x264_log( h, X264_LOG_INFO, "8x8 transform intra:%.1f%%%s\n", 100. * i_i8x8 / i_intra, buf );
3183         }
3184
3185         if( (h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO ||
3186             (h->stat.i_direct_frames[0] && h->stat.i_direct_frames[1]))
3187             && h->stat.i_frame_count[SLICE_TYPE_B] )
3188         {
3189             x264_log( h, X264_LOG_INFO, "direct mvs  spatial:%.1f%% temporal:%.1f%%\n",
3190                       h->stat.i_direct_frames[1] * 100. / h->stat.i_frame_count[SLICE_TYPE_B],
3191                       h->stat.i_direct_frames[0] * 100. / h->stat.i_frame_count[SLICE_TYPE_B] );
3192         }
3193
3194         buf[0] = 0;
3195         if( i_mb_count != i_all_intra )
3196             sprintf( buf, " inter: %.1f%% %.1f%% %.1f%%",
3197                      h->stat.i_mb_cbp[1] * 100.0 / ((i_mb_count - i_all_intra)*4),
3198                      h->stat.i_mb_cbp[3] * 100.0 / ((i_mb_count - i_all_intra)  ),
3199                      h->stat.i_mb_cbp[5] * 100.0 / ((i_mb_count - i_all_intra)) );
3200         x264_log( h, X264_LOG_INFO, "coded y,uvDC,uvAC intra: %.1f%% %.1f%% %.1f%%%s\n",
3201                   h->stat.i_mb_cbp[0] * 100.0 / (i_all_intra*4),
3202                   h->stat.i_mb_cbp[2] * 100.0 / (i_all_intra  ),
3203                   h->stat.i_mb_cbp[4] * 100.0 / (i_all_intra  ), buf );
3204
3205         int64_t fixed_pred_modes[4][9] = {{0}};
3206         int64_t sum_pred_modes[4] = {0};
3207         for( int i = 0; i <= I_PRED_16x16_DC_128; i++ )
3208         {
3209             fixed_pred_modes[0][x264_mb_pred_mode16x16_fix[i]] += h->stat.i_mb_pred_mode[0][i];
3210             sum_pred_modes[0] += h->stat.i_mb_pred_mode[0][i];
3211         }
3212         if( sum_pred_modes[0] )
3213             x264_log( h, X264_LOG_INFO, "i16 v,h,dc,p: %2.0f%% %2.0f%% %2.0f%% %2.0f%%\n",
3214                       fixed_pred_modes[0][0] * 100.0 / sum_pred_modes[0],
3215                       fixed_pred_modes[0][1] * 100.0 / sum_pred_modes[0],
3216                       fixed_pred_modes[0][2] * 100.0 / sum_pred_modes[0],
3217                       fixed_pred_modes[0][3] * 100.0 / sum_pred_modes[0] );
3218         for( int i = 1; i <= 2; i++ )
3219         {
3220             for( int j = 0; j <= I_PRED_8x8_DC_128; j++ )
3221             {
3222                 fixed_pred_modes[i][x264_mb_pred_mode4x4_fix(j)] += h->stat.i_mb_pred_mode[i][j];
3223                 sum_pred_modes[i] += h->stat.i_mb_pred_mode[i][j];
3224             }
3225             if( sum_pred_modes[i] )
3226                 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,
3227                           fixed_pred_modes[i][0] * 100.0 / sum_pred_modes[i],
3228                           fixed_pred_modes[i][1] * 100.0 / sum_pred_modes[i],
3229                           fixed_pred_modes[i][2] * 100.0 / sum_pred_modes[i],
3230                           fixed_pred_modes[i][3] * 100.0 / sum_pred_modes[i],
3231                           fixed_pred_modes[i][4] * 100.0 / sum_pred_modes[i],
3232                           fixed_pred_modes[i][5] * 100.0 / sum_pred_modes[i],
3233                           fixed_pred_modes[i][6] * 100.0 / sum_pred_modes[i],
3234                           fixed_pred_modes[i][7] * 100.0 / sum_pred_modes[i],
3235                           fixed_pred_modes[i][8] * 100.0 / sum_pred_modes[i] );
3236         }
3237         for( int i = 0; i <= I_PRED_CHROMA_DC_128; i++ )
3238         {
3239             fixed_pred_modes[3][x264_mb_pred_mode8x8c_fix[i]] += h->stat.i_mb_pred_mode[3][i];
3240             sum_pred_modes[3] += h->stat.i_mb_pred_mode[3][i];
3241         }
3242         if( sum_pred_modes[3] )
3243             x264_log( h, X264_LOG_INFO, "i8c dc,h,v,p: %2.0f%% %2.0f%% %2.0f%% %2.0f%%\n",
3244                       fixed_pred_modes[3][0] * 100.0 / sum_pred_modes[3],
3245                       fixed_pred_modes[3][1] * 100.0 / sum_pred_modes[3],
3246                       fixed_pred_modes[3][2] * 100.0 / sum_pred_modes[3],
3247                       fixed_pred_modes[3][3] * 100.0 / sum_pred_modes[3] );
3248
3249         if( h->param.analyse.i_weighted_pred >= X264_WEIGHTP_SIMPLE && h->stat.i_frame_count[SLICE_TYPE_P] > 0 )
3250             x264_log( h, X264_LOG_INFO, "Weighted P-Frames: Y:%.1f%% UV:%.1f%%\n",
3251                       h->stat.i_wpred[0] * 100.0 / h->stat.i_frame_count[SLICE_TYPE_P],
3252                       h->stat.i_wpred[1] * 100.0 / h->stat.i_frame_count[SLICE_TYPE_P] );
3253
3254         for( int i_list = 0; i_list < 2; i_list++ )
3255             for( int i_slice = 0; i_slice < 2; i_slice++ )
3256             {
3257                 char *p = buf;
3258                 int64_t i_den = 0;
3259                 int i_max = 0;
3260                 for( int i = 0; i < X264_REF_MAX*2; i++ )
3261                     if( h->stat.i_mb_count_ref[i_slice][i_list][i] )
3262                     {
3263                         i_den += h->stat.i_mb_count_ref[i_slice][i_list][i];
3264                         i_max = i;
3265                     }
3266                 if( i_max == 0 )
3267                     continue;
3268                 for( int i = 0; i <= i_max; i++ )
3269                     p += sprintf( p, " %4.1f%%", 100. * h->stat.i_mb_count_ref[i_slice][i_list][i] / i_den );
3270                 x264_log( h, X264_LOG_INFO, "ref %c L%d:%s\n", "PB"[i_slice], i_list, buf );
3271             }
3272
3273         if( h->param.analyse.b_ssim )
3274         {
3275             float ssim = SUM3( h->stat.f_ssim_mean_y ) / duration;
3276             x264_log( h, X264_LOG_INFO, "SSIM Mean Y:%.7f (%6.3fdb)\n", ssim, x264_ssim( ssim ) );
3277         }
3278         if( h->param.analyse.b_psnr )
3279         {
3280             x264_log( h, X264_LOG_INFO,
3281                       "PSNR Mean Y:%6.3f U:%6.3f V:%6.3f Avg:%6.3f Global:%6.3f kb/s:%.2f\n",
3282                       SUM3( h->stat.f_psnr_mean_y ) / duration,
3283                       SUM3( h->stat.f_psnr_mean_u ) / duration,
3284                       SUM3( h->stat.f_psnr_mean_v ) / duration,
3285                       SUM3( h->stat.f_psnr_average ) / duration,
3286                       x264_psnr( SUM3( h->stat.f_ssd_global ), duration * i_yuv_size ),
3287                       f_bitrate );
3288         }
3289         else
3290             x264_log( h, X264_LOG_INFO, "kb/s:%.2f\n", f_bitrate );
3291     }
3292
3293     /* rc */
3294     x264_ratecontrol_delete( h );
3295
3296     /* param */
3297     if( h->param.rc.psz_stat_out )
3298         free( h->param.rc.psz_stat_out );
3299     if( h->param.rc.psz_stat_in )
3300         free( h->param.rc.psz_stat_in );
3301
3302     x264_cqm_delete( h );
3303     x264_free( h->nal_buffer );
3304     x264_analyse_free_costs( h );
3305
3306     if( h->i_thread_frames > 1)
3307         h = h->thread[h->i_thread_phase];
3308
3309     /* frames */
3310     x264_frame_delete_list( h->frames.unused[0] );
3311     x264_frame_delete_list( h->frames.unused[1] );
3312     x264_frame_delete_list( h->frames.current );
3313     x264_frame_delete_list( h->frames.blank_unused );
3314
3315     h = h->thread[0];
3316
3317     for( int i = h->param.i_threads - 1; i >= 0; i-- )
3318     {
3319         x264_frame_t **frame;
3320
3321         if( !h->param.b_sliced_threads || i == 0 )
3322         {
3323             for( frame = h->thread[i]->frames.reference; *frame; frame++ )
3324             {
3325                 assert( (*frame)->i_reference_count > 0 );
3326                 (*frame)->i_reference_count--;
3327                 if( (*frame)->i_reference_count == 0 )
3328                     x264_frame_delete( *frame );
3329             }
3330             frame = &h->thread[i]->fdec;
3331             if( *frame )
3332             {
3333                 assert( (*frame)->i_reference_count > 0 );
3334                 (*frame)->i_reference_count--;
3335                 if( (*frame)->i_reference_count == 0 )
3336                     x264_frame_delete( *frame );
3337             }
3338             x264_macroblock_cache_free( h->thread[i] );
3339         }
3340         x264_macroblock_thread_free( h->thread[i], 0 );
3341         x264_free( h->thread[i]->out.p_bitstream );
3342         x264_free( h->thread[i]->out.nal);
3343         x264_free( h->thread[i] );
3344     }
3345 }
3346
3347 int x264_encoder_delayed_frames( x264_t *h )
3348 {
3349     int delayed_frames = 0;
3350     if( h->i_thread_frames > 1 )
3351     {
3352         for( int i = 0; i < h->i_thread_frames; i++ )
3353             delayed_frames += h->thread[i]->b_thread_active;
3354         h = h->thread[h->i_thread_phase];
3355     }
3356     for( int i = 0; h->frames.current[i]; i++ )
3357         delayed_frames++;
3358     x264_pthread_mutex_lock( &h->lookahead->ofbuf.mutex );
3359     x264_pthread_mutex_lock( &h->lookahead->ifbuf.mutex );
3360     x264_pthread_mutex_lock( &h->lookahead->next.mutex );
3361     delayed_frames += h->lookahead->ifbuf.i_size + h->lookahead->next.i_size + h->lookahead->ofbuf.i_size;
3362     x264_pthread_mutex_unlock( &h->lookahead->next.mutex );
3363     x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
3364     x264_pthread_mutex_unlock( &h->lookahead->ofbuf.mutex );
3365     return delayed_frames;
3366 }
3367
3368 int x264_encoder_maximum_delayed_frames( x264_t *h )
3369 {
3370     return h->frames.i_delay;
3371 }