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

   1 /*****************************************************************************
   2  * encoder.c: top-level encoder functions
   3  *****************************************************************************
   4  * Copyright (C) 2003-2010 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->fref1[0]->i_poc_l0ref0 == h->fref0[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_l0 = h->b_ref_reorder[0];
 147     sh->b_ref_pic_list_reordering_l1 = h->b_ref_reorder[1];
 148
 149     /* If the ref list isn't in the default order, construct reordering header */
 150     /* List1 reordering isn't needed yet */
 151     if( sh->b_ref_pic_list_reordering_l0 )
 152     {
 153         int pred_frame_num = i_frame;
 154         for( int i = 0; i < h->i_ref0; i++ )
 155         {
 156             int diff = h->fref0[i]->i_frame_num - pred_frame_num;
 157             sh->ref_pic_list_order[0][i].idc = ( diff > 0 );
 158             sh->ref_pic_list_order[0][i].arg = (abs(diff) - 1) & ((1 << sps->i_log2_max_frame_num) - 1);
 159             pred_frame_num = h->fref0[i]->i_frame_num;
 160         }
 161     }
 162
 163     sh->i_cabac_init_idc = param->i_cabac_init_idc;
 164
 165     sh->i_qp = i_qp;
 166     sh->i_qp_delta = i_qp - pps->i_pic_init_qp;
 167     sh->b_sp_for_swidth = 0;
 168     sh->i_qs_delta = 0;
 169
 170     int deblock_thresh = i_qp + 2 * X264_MIN(param->i_deblocking_filter_alphac0, param->i_deblocking_filter_beta);
 171     /* If effective qp <= 15, deblocking would have no effect anyway */
 172     if( param->b_deblocking_filter && (h->mb.b_variable_qp || 15 < deblock_thresh ) )
 173         sh->i_disable_deblocking_filter_idc = param->b_sliced_threads ? 2 : 0;
 174     else
 175         sh->i_disable_deblocking_filter_idc = 1;
 176     sh->i_alpha_c0_offset = param->i_deblocking_filter_alphac0 << 1;
 177     sh->i_beta_offset = param->i_deblocking_filter_beta << 1;
 178 }
 179
 180 static void x264_slice_header_write( bs_t *s, x264_slice_header_t *sh, int i_nal_ref_idc )
 181 {
 182     if( sh->b_mbaff )
 183     {
 184         assert( sh->i_first_mb % (2*sh->sps->i_mb_width) == 0 );
 185         bs_write_ue( s, sh->i_first_mb >> 1 );
 186     }
 187     else
 188         bs_write_ue( s, sh->i_first_mb );
 189
 190     bs_write_ue( s, sh->i_type + 5 );   /* same type things */
 191     bs_write_ue( s, sh->i_pps_id );
 192     bs_write( s, sh->sps->i_log2_max_frame_num, sh->i_frame_num & ((1<<sh->sps->i_log2_max_frame_num)-1) );
 193
 194     if( !sh->sps->b_frame_mbs_only )
 195     {
 196         bs_write1( s, sh->b_field_pic );
 197         if( sh->b_field_pic )
 198             bs_write1( s, sh->b_bottom_field );
 199     }
 200
 201     if( sh->i_idr_pic_id >= 0 ) /* NAL IDR */
 202         bs_write_ue( s, sh->i_idr_pic_id );
 203
 204     if( sh->sps->i_poc_type == 0 )
 205     {
 206         bs_write( s, sh->sps->i_log2_max_poc_lsb, sh->i_poc & ((1<<sh->sps->i_log2_max_poc_lsb)-1) );
 207         if( sh->pps->b_pic_order && !sh->b_field_pic )
 208             bs_write_se( s, sh->i_delta_poc_bottom );
 209     }
 210     else if( sh->sps->i_poc_type == 1 && !sh->sps->b_delta_pic_order_always_zero )
 211     {
 212         bs_write_se( s, sh->i_delta_poc[0] );
 213         if( sh->pps->b_pic_order && !sh->b_field_pic )
 214             bs_write_se( s, sh->i_delta_poc[1] );
 215     }
 216
 217     if( sh->pps->b_redundant_pic_cnt )
 218         bs_write_ue( s, sh->i_redundant_pic_cnt );
 219
 220     if( sh->i_type == SLICE_TYPE_B )
 221         bs_write1( s, sh->b_direct_spatial_mv_pred );
 222
 223     if( sh->i_type == SLICE_TYPE_P || sh->i_type == SLICE_TYPE_B )
 224     {
 225         bs_write1( s, sh->b_num_ref_idx_override );
 226         if( sh->b_num_ref_idx_override )
 227         {
 228             bs_write_ue( s, sh->i_num_ref_idx_l0_active - 1 );
 229             if( sh->i_type == SLICE_TYPE_B )
 230                 bs_write_ue( s, sh->i_num_ref_idx_l1_active - 1 );
 231         }
 232     }
 233
 234     /* ref pic list reordering */
 235     if( sh->i_type != SLICE_TYPE_I )
 236     {
 237         bs_write1( s, sh->b_ref_pic_list_reordering_l0 );
 238         if( sh->b_ref_pic_list_reordering_l0 )
 239         {
 240             for( int i = 0; i < sh->i_num_ref_idx_l0_active; i++ )
 241             {
 242                 bs_write_ue( s, sh->ref_pic_list_order[0][i].idc );
 243                 bs_write_ue( s, sh->ref_pic_list_order[0][i].arg );
 244             }
 245             bs_write_ue( s, 3 );
 246         }
 247     }
 248     if( sh->i_type == SLICE_TYPE_B )
 249     {
 250         bs_write1( s, sh->b_ref_pic_list_reordering_l1 );
 251         if( sh->b_ref_pic_list_reordering_l1 )
 252         {
 253             for( int i = 0; i < sh->i_num_ref_idx_l1_active; i++ )
 254             {
 255                 bs_write_ue( s, sh->ref_pic_list_order[1][i].idc );
 256                 bs_write_ue( s, sh->ref_pic_list_order[1][i].arg );
 257             }
 258             bs_write_ue( s, 3 );
 259         }
 260     }
 261
 262     if( sh->pps->b_weighted_pred && sh->i_type == SLICE_TYPE_P )
 263     {
 264         /* pred_weight_table() */
 265         bs_write_ue( s, sh->weight[0][0].i_denom );
 266         bs_write_ue( s, sh->weight[0][1].i_denom );
 267         for( int i = 0; i < sh->i_num_ref_idx_l0_active; i++ )
 268         {
 269             int luma_weight_l0_flag = !!sh->weight[i][0].weightfn;
 270             int chroma_weight_l0_flag = !!sh->weight[i][1].weightfn || !!sh->weight[i][2].weightfn;
 271             bs_write1( s, luma_weight_l0_flag );
 272             if( luma_weight_l0_flag )
 273             {
 274                 bs_write_se( s, sh->weight[i][0].i_scale );
 275                 bs_write_se( s, sh->weight[i][0].i_offset );
 276             }
 277             bs_write1( s, chroma_weight_l0_flag );
 278             if( chroma_weight_l0_flag )
 279             {
 280                 for( int j = 1; j < 3; j++ )
 281                 {
 282                     bs_write_se( s, sh->weight[i][j].i_scale );
 283                     bs_write_se( s, sh->weight[i][j].i_offset );
 284                 }
 285             }
 286         }
 287     }
 288     else if( sh->pps->b_weighted_bipred == 1 && sh->i_type == SLICE_TYPE_B )
 289     {
 290       /* TODO */
 291     }
 292
 293     if( i_nal_ref_idc != 0 )
 294     {
 295         if( sh->i_idr_pic_id >= 0 )
 296         {
 297             bs_write1( s, 0 );  /* no output of prior pics flag */
 298             bs_write1( s, 0 );  /* long term reference flag */
 299         }
 300         else
 301         {
 302             bs_write1( s, sh->i_mmco_command_count > 0 ); /* adaptive_ref_pic_marking_mode_flag */
 303             if( sh->i_mmco_command_count > 0 )
 304             {
 305                 for( int i = 0; i < sh->i_mmco_command_count; i++ )
 306                 {
 307                     bs_write_ue( s, 1 ); /* mark short term ref as unused */
 308                     bs_write_ue( s, sh->mmco[i].i_difference_of_pic_nums - 1 );
 309                 }
 310                 bs_write_ue( s, 0 ); /* end command list */
 311             }
 312         }
 313     }
 314
 315     if( sh->pps->b_cabac && sh->i_type != SLICE_TYPE_I )
 316         bs_write_ue( s, sh->i_cabac_init_idc );
 317
 318     bs_write_se( s, sh->i_qp_delta );      /* slice qp delta */
 319
 320     if( sh->pps->b_deblocking_filter_control )
 321     {
 322         bs_write_ue( s, sh->i_disable_deblocking_filter_idc );
 323         if( sh->i_disable_deblocking_filter_idc != 1 )
 324         {
 325             bs_write_se( s, sh->i_alpha_c0_offset >> 1 );
 326             bs_write_se( s, sh->i_beta_offset >> 1 );
 327         }
 328     }
 329 }
 330
 331 /* If we are within a reasonable distance of the end of the memory allocated for the bitstream, */
 332 /* reallocate, adding an arbitrary amount of space (100 kilobytes). */
 333 static int x264_bitstream_check_buffer( x264_t *h )
 334 {
 335     uint8_t *bs_bak = h->out.p_bitstream;
 336     if( (h->param.b_cabac && (h->cabac.p_end - h->cabac.p < 2500)) ||
 337         (h->out.bs.p_end - h->out.bs.p < 2500) )
 338     {
 339         h->out.i_bitstream += 100000;
 340         CHECKED_MALLOC( h->out.p_bitstream, h->out.i_bitstream );
 341         h->mc.memcpy_aligned( h->out.p_bitstream, bs_bak, (h->out.i_bitstream - 100000) & ~15 );
 342         intptr_t delta = h->out.p_bitstream - bs_bak;
 343
 344         h->out.bs.p_start += delta;
 345         h->out.bs.p += delta;
 346         h->out.bs.p_end = h->out.p_bitstream + h->out.i_bitstream;
 347
 348         h->cabac.p_start += delta;
 349         h->cabac.p += delta;
 350         h->cabac.p_end = h->out.p_bitstream + h->out.i_bitstream;
 351
 352         for( int i = 0; i <= h->out.i_nal; i++ )
 353             h->out.nal[i].p_payload += delta;
 354         x264_free( bs_bak );
 355     }
 356     return 0;
 357 fail:
 358     x264_free( bs_bak );
 359     return -1;
 360 }
 361
 362 #if HAVE_PTHREAD
 363 static void x264_encoder_thread_init( x264_t *h )
 364 {
 365     if( h->param.i_sync_lookahead )
 366         x264_lower_thread_priority( 10 );
 367
 368 #if HAVE_MMX
 369     /* Misalign mask has to be set separately for each thread. */
 370     if( h->param.cpu&X264_CPU_SSE_MISALIGN )
 371         x264_cpu_mask_misalign_sse();
 372 #endif
 373 }
 374 #endif
 375
 376 /****************************************************************************
 377  *
 378  ****************************************************************************
 379  ****************************** External API*********************************
 380  ****************************************************************************
 381  *
 382  ****************************************************************************/
 383
 384 static int x264_validate_parameters( x264_t *h )
 385 {
 386 #if HAVE_MMX
 387 #ifdef __SSE__
 388     if( !(x264_cpu_detect() & X264_CPU_SSE) )
 389     {
 390         x264_log( h, X264_LOG_ERROR, "your cpu does not support SSE1, but x264 was compiled with asm support\n");
 391 #else
 392     if( !(x264_cpu_detect() & X264_CPU_MMXEXT) )
 393     {
 394         x264_log( h, X264_LOG_ERROR, "your cpu does not support MMXEXT, but x264 was compiled with asm support\n");
 395 #endif
 396         x264_log( h, X264_LOG_ERROR, "to run x264, recompile without asm support (configure --disable-asm)\n");
 397         return -1;
 398     }
 399 #endif
 400     if( h->param.i_width <= 0 || h->param.i_height <= 0 )
 401     {
 402         x264_log( h, X264_LOG_ERROR, "invalid width x height (%dx%d)\n",
 403                   h->param.i_width, h->param.i_height );
 404         return -1;
 405     }
 406
 407     if( h->param.i_width % 2 || h->param.i_height % 2 )
 408     {
 409         x264_log( h, X264_LOG_ERROR, "width or height not divisible by 2 (%dx%d)\n",
 410                   h->param.i_width, h->param.i_height );
 411         return -1;
 412     }
 413     int i_csp = h->param.i_csp & X264_CSP_MASK;
 414     if( i_csp <= X264_CSP_NONE || i_csp >= X264_CSP_MAX )
 415     {
 416         x264_log( h, X264_LOG_ERROR, "invalid CSP (only I420/YV12/NV12 supported)\n" );
 417         return -1;
 418     }
 419
 420     if( (h->param.crop_rect.i_left + h->param.crop_rect.i_right ) >= h->param.i_width ||
 421         (h->param.crop_rect.i_top  + h->param.crop_rect.i_bottom) >= h->param.i_height )
 422     {
 423         x264_log( h, X264_LOG_ERROR, "invalid crop-rect %u,%u,%u,%u\n", h->param.crop_rect.i_left,
 424                   h->param.crop_rect.i_top, h->param.crop_rect.i_right,  h->param.crop_rect.i_bottom );
 425         return -1;
 426     }
 427
 428     if( h->param.i_threads == X264_THREADS_AUTO )
 429         h->param.i_threads = x264_cpu_num_processors() * (h->param.b_sliced_threads?2:3)/2;
 430     h->param.i_threads = x264_clip3( h->param.i_threads, 1, X264_THREAD_MAX );
 431     if( h->param.i_threads > 1 )
 432     {
 433 #if !HAVE_PTHREAD
 434         x264_log( h, X264_LOG_WARNING, "not compiled with pthread support!\n");
 435         h->param.i_threads = 1;
 436 #endif
 437         /* Avoid absurdly small thread slices as they can reduce performance
 438          * and VBV compliance.  Capped at an arbitrary 4 rows per thread. */
 439         if( h->param.b_sliced_threads )
 440         {
 441             int max_threads = (h->param.i_height+15)/16 / 4;
 442             h->param.i_threads = X264_MIN( h->param.i_threads, max_threads );
 443         }
 444     }
 445     else
 446         h->param.b_sliced_threads = 0;
 447     h->i_thread_frames = h->param.b_sliced_threads ? 1 : h->param.i_threads;
 448     if( h->i_thread_frames > 1 )
 449         h->param.nalu_process = NULL;
 450
 451     h->param.i_keyint_max = x264_clip3( h->param.i_keyint_max, 1, X264_KEYINT_MAX_INFINITE );
 452     if( h->param.i_keyint_max == 1 )
 453     {
 454         h->param.b_intra_refresh = 0;
 455         h->param.analyse.i_weighted_pred = 0;
 456     }
 457
 458     if( h->param.b_interlaced )
 459     {
 460         if( h->param.analyse.i_me_method >= X264_ME_ESA )
 461         {
 462             x264_log( h, X264_LOG_WARNING, "interlace + me=esa is not implemented\n" );
 463             h->param.analyse.i_me_method = X264_ME_UMH;
 464         }
 465         if( h->param.analyse.i_weighted_pred > 0 )
 466         {
 467             x264_log( h, X264_LOG_WARNING, "interlace + weightp is not implemented\n" );
 468             h->param.analyse.i_weighted_pred = X264_WEIGHTP_NONE;
 469         }
 470     }
 471
 472     /* Detect default ffmpeg settings and terminate with an error. */
 473     {
 474         int score = 0;
 475         score += h->param.analyse.i_me_range == 0;
 476         score += h->param.rc.i_qp_step == 3;
 477         score += h->param.i_keyint_max == 12;
 478         score += h->param.rc.i_qp_min == 2;
 479         score += h->param.rc.i_qp_max == 31;
 480         score += h->param.rc.f_qcompress == 0.5;
 481         score += fabs(h->param.rc.f_ip_factor - 1.25) < 0.01;
 482         score += fabs(h->param.rc.f_pb_factor - 1.25) < 0.01;
 483         score += h->param.analyse.inter == 0 && h->param.analyse.i_subpel_refine == 8;
 484         if( score >= 5 )
 485         {
 486             x264_log( h, X264_LOG_ERROR, "broken ffmpeg default settings detected\n" );
 487             x264_log( h, X264_LOG_ERROR, "use an encoding preset (e.g. -vpre medium)\n" );
 488             x264_log( h, X264_LOG_ERROR, "preset usage: -vpre <speed> -vpre <profile>\n" );
 489             x264_log( h, X264_LOG_ERROR, "speed presets are listed in x264 --help\n" );
 490             x264_log( h, X264_LOG_ERROR, "profile is optional; x264 defaults to high\n" );
 491             return -1;
 492         }
 493     }
 494
 495     if( h->param.rc.i_rc_method < 0 || h->param.rc.i_rc_method > 2 )
 496     {
 497         x264_log( h, X264_LOG_ERROR, "no ratecontrol method specified\n" );
 498         return -1;
 499     }
 500     h->param.rc.f_rf_constant = x264_clip3f( h->param.rc.f_rf_constant, -QP_BD_OFFSET, 51 );
 501     h->param.rc.i_qp_constant = x264_clip3( h->param.rc.i_qp_constant, 0, QP_MAX );
 502     if( h->param.rc.i_rc_method == X264_RC_CRF )
 503     {
 504         h->param.rc.i_qp_constant = h->param.rc.f_rf_constant + QP_BD_OFFSET;
 505         h->param.rc.i_bitrate = 0;
 506     }
 507     if( (h->param.rc.i_rc_method == X264_RC_CQP || h->param.rc.i_rc_method == X264_RC_CRF)
 508         && h->param.rc.i_qp_constant == 0 )
 509     {
 510         h->mb.b_lossless = 1;
 511         h->param.i_cqm_preset = X264_CQM_FLAT;
 512         h->param.psz_cqm_file = NULL;
 513         h->param.rc.i_rc_method = X264_RC_CQP;
 514         h->param.rc.f_ip_factor = 1;
 515         h->param.rc.f_pb_factor = 1;
 516         h->param.analyse.b_psnr = 0;
 517         h->param.analyse.b_ssim = 0;
 518         h->param.analyse.i_chroma_qp_offset = 0;
 519         h->param.analyse.i_trellis = 0;
 520         h->param.analyse.b_fast_pskip = 0;
 521         h->param.analyse.i_noise_reduction = 0;
 522         h->param.analyse.b_psy = 0;
 523         h->param.i_bframe = 0;
 524         /* 8x8dct is not useful without RD in CAVLC lossless */
 525         if( !h->param.b_cabac && h->param.analyse.i_subpel_refine < 6 )
 526             h->param.analyse.b_transform_8x8 = 0;
 527     }
 528     if( h->param.rc.i_rc_method == X264_RC_CQP )
 529     {
 530         float qp_p = h->param.rc.i_qp_constant;
 531         float qp_i = qp_p - 6*log2f( h->param.rc.f_ip_factor );
 532         float qp_b = qp_p + 6*log2f( h->param.rc.f_pb_factor );
 533         h->param.rc.i_qp_min = x264_clip3( (int)(X264_MIN3( qp_p, qp_i, qp_b )), 0, QP_MAX );
 534         h->param.rc.i_qp_max = x264_clip3( (int)(X264_MAX3( qp_p, qp_i, qp_b ) + .999), 0, QP_MAX );
 535         h->param.rc.i_aq_mode = 0;
 536         h->param.rc.b_mb_tree = 0;
 537     }
 538     h->param.rc.i_qp_max = x264_clip3( h->param.rc.i_qp_max, 0, QP_MAX );
 539     h->param.rc.i_qp_min = x264_clip3( h->param.rc.i_qp_min, 0, h->param.rc.i_qp_max );
 540     if( h->param.rc.i_vbv_buffer_size )
 541     {
 542         if( h->param.rc.i_rc_method == X264_RC_CQP )
 543         {
 544             x264_log( h, X264_LOG_WARNING, "VBV is incompatible with constant QP, ignored.\n" );
 545             h->param.rc.i_vbv_max_bitrate = 0;
 546             h->param.rc.i_vbv_buffer_size = 0;
 547         }
 548         else if( h->param.rc.i_vbv_max_bitrate == 0 )
 549         {
 550             if( h->param.rc.i_rc_method == X264_RC_ABR )
 551             {
 552                 x264_log( h, X264_LOG_WARNING, "VBV maxrate unspecified, assuming CBR\n" );
 553                 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate;
 554             }
 555             else
 556             {
 557                 x264_log( h, X264_LOG_WARNING, "VBV bufsize set but maxrate unspecified, ignored\n" );
 558                 h->param.rc.i_vbv_buffer_size = 0;
 559             }
 560         }
 561         else if( h->param.rc.i_vbv_max_bitrate < h->param.rc.i_bitrate &&
 562                  h->param.rc.i_rc_method == X264_RC_ABR )
 563         {
 564             x264_log( h, X264_LOG_WARNING, "max bitrate less than average bitrate, assuming CBR\n" );
 565             h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate;
 566         }
 567     }
 568     else if( h->param.rc.i_vbv_max_bitrate )
 569     {
 570         x264_log( h, X264_LOG_WARNING, "VBV maxrate specified, but no bufsize, ignored\n" );
 571         h->param.rc.i_vbv_max_bitrate = 0;
 572     }
 573
 574     if( h->param.b_interlaced && h->param.i_slice_max_size )
 575     {
 576         x264_log( h, X264_LOG_WARNING, "interlaced + slice-max-size is not implemented\n" );
 577         h->param.i_slice_max_size = 0;
 578     }
 579     if( h->param.b_interlaced && h->param.i_slice_max_mbs )
 580     {
 581         x264_log( h, X264_LOG_WARNING, "interlaced + slice-max-mbs is not implemented\n" );
 582         h->param.i_slice_max_mbs = 0;
 583     }
 584     int max_slices = (h->param.i_height+((16<<h->param.b_interlaced)-1))/(16<<h->param.b_interlaced);
 585     if( h->param.b_sliced_threads )
 586         h->param.i_slice_count = x264_clip3( h->param.i_threads, 0, max_slices );
 587     else
 588     {
 589         h->param.i_slice_count = x264_clip3( h->param.i_slice_count, 0, max_slices );
 590         h->param.i_slice_max_size = X264_MAX( h->param.i_slice_max_size, 0 );
 591         h->param.i_slice_max_mbs = X264_MAX( h->param.i_slice_max_mbs, 0 );
 592         if( h->param.i_slice_max_mbs || h->param.i_slice_max_size )
 593             h->param.i_slice_count = 0;
 594     }
 595
 596     h->param.i_frame_reference = x264_clip3( h->param.i_frame_reference, 1, X264_REF_MAX );
 597     h->param.i_dpb_size = x264_clip3( h->param.i_dpb_size, 1, X264_REF_MAX );
 598     if( h->param.i_scenecut_threshold < 0 )
 599         h->param.i_scenecut_threshold = 0;
 600     if( !h->param.analyse.i_subpel_refine && h->param.analyse.i_direct_mv_pred > X264_DIRECT_PRED_SPATIAL )
 601     {
 602         x264_log( h, X264_LOG_WARNING, "subme=0 + direct=temporal is not supported\n" );
 603         h->param.analyse.i_direct_mv_pred = X264_DIRECT_PRED_SPATIAL;
 604     }
 605     h->param.i_bframe = x264_clip3( h->param.i_bframe, 0, X264_MIN( X264_BFRAME_MAX, h->param.i_keyint_max-1 ) );
 606     h->param.i_open_gop = x264_clip3( h->param.i_open_gop, X264_OPEN_GOP_NONE, X264_OPEN_GOP_BLURAY );
 607     h->param.i_bframe_bias = x264_clip3( h->param.i_bframe_bias, -90, 100 );
 608     if( h->param.i_bframe <= 1 )
 609         h->param.i_bframe_pyramid = X264_B_PYRAMID_NONE;
 610     h->param.i_bframe_pyramid = x264_clip3( h->param.i_bframe_pyramid, X264_B_PYRAMID_NONE, X264_B_PYRAMID_NORMAL );
 611     if( !h->param.i_bframe )
 612     {
 613         h->param.i_bframe_adaptive = X264_B_ADAPT_NONE;
 614         h->param.analyse.i_direct_mv_pred = 0;
 615         h->param.analyse.b_weighted_bipred = 0;
 616         h->param.i_open_gop = X264_OPEN_GOP_NONE;
 617     }
 618     if( h->param.b_intra_refresh && h->param.i_bframe_pyramid == X264_B_PYRAMID_NORMAL )
 619     {
 620         x264_log( h, X264_LOG_WARNING, "b-pyramid normal + intra-refresh is not supported\n" );
 621         h->param.i_bframe_pyramid = X264_B_PYRAMID_STRICT;
 622     }
 623     if( h->param.b_intra_refresh && (h->param.i_frame_reference > 1 || h->param.i_dpb_size > 1) )
 624     {
 625         x264_log( h, X264_LOG_WARNING, "ref > 1 + intra-refresh is not supported\n" );
 626         h->param.i_frame_reference = 1;
 627         h->param.i_dpb_size = 1;
 628     }
 629     if( h->param.b_intra_refresh && h->param.i_open_gop )
 630     {
 631         x264_log( h, X264_LOG_WARNING, "intra-refresh is not compatible with open-gop\n" );
 632         h->param.i_open_gop = X264_OPEN_GOP_NONE;
 633     }
 634     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;
 635     if( h->param.i_keyint_min == X264_KEYINT_MIN_AUTO )
 636         h->param.i_keyint_min = X264_MIN( h->param.i_keyint_max / 10, fps );
 637     h->param.i_keyint_min = x264_clip3( h->param.i_keyint_min, 1, h->param.i_keyint_max/2+1 );
 638     h->param.rc.i_lookahead = x264_clip3( h->param.rc.i_lookahead, 0, X264_LOOKAHEAD_MAX );
 639     {
 640         int maxrate = X264_MAX( h->param.rc.i_vbv_max_bitrate, h->param.rc.i_bitrate );
 641         float bufsize = maxrate ? (float)h->param.rc.i_vbv_buffer_size / maxrate : 0;
 642         h->param.rc.i_lookahead = X264_MIN( h->param.rc.i_lookahead, X264_MAX( h->param.i_keyint_max, bufsize*fps ) );
 643     }
 644
 645     if( !h->param.i_timebase_num || !h->param.i_timebase_den || !(h->param.b_vfr_input || h->param.b_pulldown) )
 646     {
 647         h->param.i_timebase_num = h->param.i_fps_den;
 648         h->param.i_timebase_den = h->param.i_fps_num;
 649     }
 650
 651     h->param.rc.f_qcompress = x264_clip3f( h->param.rc.f_qcompress, 0.0, 1.0 );
 652     if( h->param.i_keyint_max == 1 || h->param.rc.f_qcompress == 1 )
 653         h->param.rc.b_mb_tree = 0;
 654     if( (!h->param.b_intra_refresh && h->param.i_keyint_max != X264_KEYINT_MAX_INFINITE) &&
 655         !h->param.rc.i_lookahead && h->param.rc.b_mb_tree )
 656     {
 657         x264_log( h, X264_LOG_WARNING, "lookaheadless mb-tree requires intra refresh or infinite keyint\n" );
 658         h->param.rc.b_mb_tree = 0;
 659     }
 660     if( h->param.rc.b_stat_read )
 661         h->param.rc.i_lookahead = 0;
 662 #if HAVE_PTHREAD
 663     if( h->param.i_sync_lookahead < 0 )
 664         h->param.i_sync_lookahead = h->param.i_bframe + 1;
 665     h->param.i_sync_lookahead = X264_MIN( h->param.i_sync_lookahead, X264_LOOKAHEAD_MAX );
 666     if( h->param.rc.b_stat_read || h->i_thread_frames == 1 )
 667         h->param.i_sync_lookahead = 0;
 668 #else
 669     h->param.i_sync_lookahead = 0;
 670 #endif
 671
 672     h->param.i_deblocking_filter_alphac0 = x264_clip3( h->param.i_deblocking_filter_alphac0, -6, 6 );
 673     h->param.i_deblocking_filter_beta    = x264_clip3( h->param.i_deblocking_filter_beta, -6, 6 );
 674     h->param.analyse.i_luma_deadzone[0] = x264_clip3( h->param.analyse.i_luma_deadzone[0], 0, 32 );
 675     h->param.analyse.i_luma_deadzone[1] = x264_clip3( h->param.analyse.i_luma_deadzone[1], 0, 32 );
 676
 677     h->param.i_cabac_init_idc = x264_clip3( h->param.i_cabac_init_idc, 0, 2 );
 678
 679     if( h->param.i_cqm_preset < X264_CQM_FLAT || h->param.i_cqm_preset > X264_CQM_CUSTOM )
 680         h->param.i_cqm_preset = X264_CQM_FLAT;
 681
 682     if( h->param.analyse.i_me_method < X264_ME_DIA ||
 683         h->param.analyse.i_me_method > X264_ME_TESA )
 684         h->param.analyse.i_me_method = X264_ME_HEX;
 685     if( h->param.analyse.i_me_range < 4 )
 686         h->param.analyse.i_me_range = 4;
 687     if( h->param.analyse.i_me_range > 16 && h->param.analyse.i_me_method <= X264_ME_HEX )
 688         h->param.analyse.i_me_range = 16;
 689     if( h->param.analyse.i_me_method == X264_ME_TESA &&
 690         (h->mb.b_lossless || h->param.analyse.i_subpel_refine <= 1) )
 691         h->param.analyse.i_me_method = X264_ME_ESA;
 692     h->param.analyse.i_subpel_refine = x264_clip3( h->param.analyse.i_subpel_refine, 0, 10 );
 693     h->param.analyse.b_mixed_references = h->param.analyse.b_mixed_references && h->param.i_frame_reference > 1;
 694     h->param.analyse.inter &= X264_ANALYSE_PSUB16x16|X264_ANALYSE_PSUB8x8|X264_ANALYSE_BSUB16x16|
 695                               X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
 696     h->param.analyse.intra &= X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
 697     if( !(h->param.analyse.inter & X264_ANALYSE_PSUB16x16) )
 698         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
 699     if( !h->param.analyse.b_transform_8x8 )
 700     {
 701         h->param.analyse.inter &= ~X264_ANALYSE_I8x8;
 702         h->param.analyse.intra &= ~X264_ANALYSE_I8x8;
 703     }
 704     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -12, 12);
 705     h->param.analyse.i_trellis = x264_clip3( h->param.analyse.i_trellis, 0, 2 );
 706     if( !h->param.analyse.b_psy )
 707     {
 708         h->param.analyse.f_psy_rd = 0;
 709         h->param.analyse.f_psy_trellis = 0;
 710     }
 711     if( !h->param.analyse.i_trellis )
 712         h->param.analyse.f_psy_trellis = 0;
 713     h->param.analyse.f_psy_rd = x264_clip3f( h->param.analyse.f_psy_rd, 0, 10 );
 714     h->param.analyse.f_psy_trellis = x264_clip3f( h->param.analyse.f_psy_trellis, 0, 10 );
 715     if( h->param.analyse.i_subpel_refine < 6 )
 716         h->param.analyse.f_psy_rd = 0;
 717     h->mb.i_psy_rd = FIX8( h->param.analyse.f_psy_rd );
 718     /* Psy RDO increases overall quantizers to improve the quality of luma--this indirectly hurts chroma quality */
 719     /* so we lower the chroma QP offset to compensate */
 720     /* This can be triggered repeatedly on multiple calls to parameter_validate, but since encoding
 721      * uses the pps chroma qp offset not the param chroma qp offset, this is not a problem. */
 722     if( h->mb.i_psy_rd )
 723         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_rd < 0.25 ? 1 : 2;
 724     h->mb.i_psy_trellis = FIX8( h->param.analyse.f_psy_trellis / 4 );
 725     /* Psy trellis has a similar effect. */
 726     if( h->mb.i_psy_trellis )
 727         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_trellis < 0.25 ? 1 : 2;
 728     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -12, 12);
 729     h->param.rc.i_aq_mode = x264_clip3( h->param.rc.i_aq_mode, 0, 2 );
 730     h->param.rc.f_aq_strength = x264_clip3f( h->param.rc.f_aq_strength, 0, 3 );
 731     if( h->param.rc.f_aq_strength == 0 )
 732         h->param.rc.i_aq_mode = 0;
 733     /* MB-tree requires AQ to be on, even if the strength is zero. */
 734     if( !h->param.rc.i_aq_mode && h->param.rc.b_mb_tree )
 735     {
 736         h->param.rc.i_aq_mode = 1;
 737         h->param.rc.f_aq_strength = 0;
 738     }
 739     h->param.analyse.i_noise_reduction = x264_clip3( h->param.analyse.i_noise_reduction, 0, 1<<16 );
 740     if( h->param.analyse.i_subpel_refine == 10 && (h->param.analyse.i_trellis != 2 || !h->param.rc.i_aq_mode) )
 741         h->param.analyse.i_subpel_refine = 9;
 742
 743     {
 744         const x264_level_t *l = x264_levels;
 745         if( h->param.i_level_idc < 0 )
 746         {
 747             int maxrate_bak = h->param.rc.i_vbv_max_bitrate;
 748             if( h->param.rc.i_rc_method == X264_RC_ABR && h->param.rc.i_vbv_buffer_size <= 0 )
 749                 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate * 2;
 750             h->sps = h->sps_array;
 751             x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
 752             do h->param.i_level_idc = l->level_idc;
 753                 while( l[1].level_idc && x264_validate_levels( h, 0 ) && l++ );
 754             h->param.rc.i_vbv_max_bitrate = maxrate_bak;
 755         }
 756         else
 757         {
 758             while( l->level_idc && l->level_idc != h->param.i_level_idc )
 759                 l++;
 760             if( l->level_idc == 0 )
 761             {
 762                 x264_log( h, X264_LOG_ERROR, "invalid level_idc: %d\n", h->param.i_level_idc );
 763                 return -1;
 764             }
 765         }
 766         if( h->param.analyse.i_mv_range <= 0 )
 767             h->param.analyse.i_mv_range = l->mv_range >> h->param.b_interlaced;
 768         else
 769             h->param.analyse.i_mv_range = x264_clip3(h->param.analyse.i_mv_range, 32, 512 >> h->param.b_interlaced);
 770     }
 771
 772     h->param.analyse.i_weighted_pred = x264_clip3( h->param.analyse.i_weighted_pred, X264_WEIGHTP_NONE, X264_WEIGHTP_SMART );
 773     if( !h->param.analyse.i_weighted_pred && h->param.rc.b_mb_tree && h->param.analyse.b_psy && !h->param.b_interlaced )
 774         h->param.analyse.i_weighted_pred = X264_WEIGHTP_FAKE;
 775
 776     if( h->i_thread_frames > 1 )
 777     {
 778         int r = h->param.analyse.i_mv_range_thread;
 779         int r2;
 780         if( r <= 0 )
 781         {
 782             // half of the available space is reserved and divided evenly among the threads,
 783             // the rest is allocated to whichever thread is far enough ahead to use it.
 784             // reserving more space increases quality for some videos, but costs more time
 785             // in thread synchronization.
 786             int max_range = (h->param.i_height + X264_THREAD_HEIGHT) / h->i_thread_frames - X264_THREAD_HEIGHT;
 787             r = max_range / 2;
 788         }
 789         r = X264_MAX( r, h->param.analyse.i_me_range );
 790         r = X264_MIN( r, h->param.analyse.i_mv_range );
 791         // round up to use the whole mb row
 792         r2 = (r & ~15) + ((-X264_THREAD_HEIGHT) & 15);
 793         if( r2 < r )
 794             r2 += 16;
 795         x264_log( h, X264_LOG_DEBUG, "using mv_range_thread = %d\n", r2 );
 796         h->param.analyse.i_mv_range_thread = r2;
 797     }
 798
 799     if( h->param.rc.f_qblur < 0 )
 800         h->param.rc.f_qblur = 0;
 801     if( h->param.rc.f_complexity_blur < 0 )
 802         h->param.rc.f_complexity_blur = 0;
 803
 804     h->param.i_sps_id &= 31;
 805
 806     if( h->param.i_log_level < X264_LOG_INFO )
 807     {
 808         h->param.analyse.b_psnr = 0;
 809         h->param.analyse.b_ssim = 0;
 810     }
 811
 812     if( h->param.b_interlaced )
 813         h->param.b_pic_struct = 1;
 814
 815     if( h->param.i_nal_hrd && !h->param.rc.i_vbv_buffer_size )
 816     {
 817         x264_log( h, X264_LOG_WARNING, "NAL HRD parameters require VBV parameters\n" );
 818         h->param.i_nal_hrd = X264_NAL_HRD_NONE;
 819     }
 820
 821     if( h->param.i_nal_hrd == X264_NAL_HRD_CBR &&
 822        (h->param.rc.i_bitrate != h->param.rc.i_vbv_max_bitrate || !h->param.rc.i_vbv_max_bitrate) )
 823     {
 824         x264_log( h, X264_LOG_WARNING, "CBR HRD requires constant bitrate\n" );
 825         h->param.i_nal_hrd = X264_NAL_HRD_VBR;
 826     }
 827
 828     /* ensure the booleans are 0 or 1 so they can be used in math */
 829 #define BOOLIFY(x) h->param.x = !!h->param.x
 830     BOOLIFY( b_cabac );
 831     BOOLIFY( b_constrained_intra );
 832     BOOLIFY( b_deblocking_filter );
 833     BOOLIFY( b_deterministic );
 834     BOOLIFY( b_sliced_threads );
 835     BOOLIFY( b_interlaced );
 836     BOOLIFY( b_intra_refresh );
 837     BOOLIFY( b_visualize );
 838     BOOLIFY( b_aud );
 839     BOOLIFY( b_repeat_headers );
 840     BOOLIFY( b_annexb );
 841     BOOLIFY( b_vfr_input );
 842     BOOLIFY( b_pic_struct );
 843     BOOLIFY( b_fake_interlaced );
 844     BOOLIFY( analyse.b_transform_8x8 );
 845     BOOLIFY( analyse.b_weighted_bipred );
 846     BOOLIFY( analyse.b_chroma_me );
 847     BOOLIFY( analyse.b_mixed_references );
 848     BOOLIFY( analyse.b_fast_pskip );
 849     BOOLIFY( analyse.b_dct_decimate );
 850     BOOLIFY( analyse.b_psy );
 851     BOOLIFY( analyse.b_psnr );
 852     BOOLIFY( analyse.b_ssim );
 853     BOOLIFY( rc.b_stat_write );
 854     BOOLIFY( rc.b_stat_read );
 855     BOOLIFY( rc.b_mb_tree );
 856 #undef BOOLIFY
 857
 858     return 0;
 859 }
 860
 861 static void mbcmp_init( x264_t *h )
 862 {
 863     int satd = !h->mb.b_lossless && h->param.analyse.i_subpel_refine > 1;
 864     memcpy( h->pixf.mbcmp, satd ? h->pixf.satd : h->pixf.sad_aligned, sizeof(h->pixf.mbcmp) );
 865     memcpy( h->pixf.mbcmp_unaligned, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.mbcmp_unaligned) );
 866     h->pixf.intra_mbcmp_x3_16x16 = satd ? h->pixf.intra_satd_x3_16x16 : h->pixf.intra_sad_x3_16x16;
 867     h->pixf.intra_mbcmp_x3_8x8c = satd ? h->pixf.intra_satd_x3_8x8c : h->pixf.intra_sad_x3_8x8c;
 868     h->pixf.intra_mbcmp_x3_8x8 = satd ? h->pixf.intra_sa8d_x3_8x8 : h->pixf.intra_sad_x3_8x8;
 869     h->pixf.intra_mbcmp_x3_4x4 = satd ? h->pixf.intra_satd_x3_4x4 : h->pixf.intra_sad_x3_4x4;
 870     satd &= h->param.analyse.i_me_method == X264_ME_TESA;
 871     memcpy( h->pixf.fpelcmp, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.fpelcmp) );
 872     memcpy( h->pixf.fpelcmp_x3, satd ? h->pixf.satd_x3 : h->pixf.sad_x3, sizeof(h->pixf.fpelcmp_x3) );
 873     memcpy( h->pixf.fpelcmp_x4, satd ? h->pixf.satd_x4 : h->pixf.sad_x4, sizeof(h->pixf.fpelcmp_x4) );
 874 }
 875
 876 static void x264_set_aspect_ratio( x264_t *h, x264_param_t *param, int initial )
 877 {
 878     /* VUI */
 879     if( param->vui.i_sar_width > 0 && param->vui.i_sar_height > 0 )
 880     {
 881         uint32_t i_w = param->vui.i_sar_width;
 882         uint32_t i_h = param->vui.i_sar_height;
 883         uint32_t old_w = h->param.vui.i_sar_width;
 884         uint32_t old_h = h->param.vui.i_sar_height;
 885
 886         x264_reduce_fraction( &i_w, &i_h );
 887
 888         while( i_w > 65535 || i_h > 65535 )
 889         {
 890             i_w /= 2;
 891             i_h /= 2;
 892         }
 893
 894         x264_reduce_fraction( &i_w, &i_h );
 895
 896         if( i_w != old_w || i_h != old_h || initial )
 897         {
 898             h->param.vui.i_sar_width = 0;
 899             h->param.vui.i_sar_height = 0;
 900             if( i_w == 0 || i_h == 0 )
 901                 x264_log( h, X264_LOG_WARNING, "cannot create valid sample aspect ratio\n" );
 902             else
 903             {
 904                 x264_log( h, initial?X264_LOG_INFO:X264_LOG_DEBUG, "using SAR=%d/%d\n", i_w, i_h );
 905                 h->param.vui.i_sar_width = i_w;
 906                 h->param.vui.i_sar_height = i_h;
 907             }
 908         }
 909     }
 910 }
 911
 912 /****************************************************************************
 913  * x264_encoder_open:
 914  ****************************************************************************/
 915 x264_t *x264_encoder_open( x264_param_t *param )
 916 {
 917     x264_t *h;
 918     char buf[1000], *p;
 919     int qp, i_slicetype_length;
 920
 921     CHECKED_MALLOCZERO( h, sizeof(x264_t) );
 922
 923     /* Create a copy of param */
 924     memcpy( &h->param, param, sizeof(x264_param_t) );
 925
 926     if( param->param_free )
 927         param->param_free( param );
 928
 929     if( x264_validate_parameters( h ) < 0 )
 930         goto fail;
 931
 932     if( h->param.psz_cqm_file )
 933         if( x264_cqm_parse_file( h, h->param.psz_cqm_file ) < 0 )
 934             goto fail;
 935
 936     if( h->param.rc.psz_stat_out )
 937         h->param.rc.psz_stat_out = strdup( h->param.rc.psz_stat_out );
 938     if( h->param.rc.psz_stat_in )
 939         h->param.rc.psz_stat_in = strdup( h->param.rc.psz_stat_in );
 940
 941     x264_set_aspect_ratio( h, &h->param, 1 );
 942
 943     x264_reduce_fraction( &h->param.i_fps_num, &h->param.i_fps_den );
 944     x264_reduce_fraction( &h->param.i_timebase_num, &h->param.i_timebase_den );
 945
 946     /* Init x264_t */
 947     h->i_frame = -1;
 948     h->i_frame_num = 0;
 949     h->i_idr_pic_id = 0;
 950
 951     if( (uint64_t)h->param.i_timebase_den * 2 > UINT32_MAX )
 952     {
 953         x264_log( h, X264_LOG_ERROR, "Effective timebase denominator %u exceeds H.264 maximum\n", h->param.i_timebase_den );
 954         goto fail;
 955     }
 956
 957     h->sps = &h->sps_array[0];
 958     x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
 959
 960     h->pps = &h->pps_array[0];
 961     x264_pps_init( h->pps, h->param.i_sps_id, &h->param, h->sps );
 962
 963     x264_validate_levels( h, 1 );
 964
 965     h->chroma_qp_table = i_chroma_qp_table + 12 + h->pps->i_chroma_qp_index_offset;
 966
 967     if( x264_cqm_init( h ) < 0 )
 968         goto fail;
 969
 970     h->mb.i_mb_width = h->sps->i_mb_width;
 971     h->mb.i_mb_height = h->sps->i_mb_height;
 972     h->mb.i_mb_count = h->mb.i_mb_width * h->mb.i_mb_height;
 973
 974     /* Init frames. */
 975     if( h->param.i_bframe_adaptive == X264_B_ADAPT_TRELLIS && !h->param.rc.b_stat_read )
 976         h->frames.i_delay = X264_MAX(h->param.i_bframe,3)*4;
 977     else
 978         h->frames.i_delay = h->param.i_bframe;
 979     if( h->param.rc.b_mb_tree || h->param.rc.i_vbv_buffer_size )
 980         h->frames.i_delay = X264_MAX( h->frames.i_delay, h->param.rc.i_lookahead );
 981     i_slicetype_length = h->frames.i_delay;
 982     h->frames.i_delay += h->i_thread_frames - 1;
 983     h->frames.i_delay += h->param.i_sync_lookahead;
 984     h->frames.i_delay += h->param.b_vfr_input && (h->param.rc.i_rc_method == X264_RC_ABR || h->param.rc.b_stat_write
 985                                                  || h->param.rc.i_vbv_buffer_size);
 986     h->frames.i_bframe_delay = h->param.i_bframe ? (h->param.i_bframe_pyramid ? 2 : 1) : 0;
 987
 988     h->frames.i_max_ref0 = h->param.i_frame_reference;
 989     h->frames.i_max_ref1 = X264_MIN( h->sps->vui.i_num_reorder_frames, h->param.i_frame_reference );
 990     h->frames.i_max_dpb  = h->sps->vui.i_max_dec_frame_buffering;
 991     h->frames.b_have_lowres = !h->param.rc.b_stat_read
 992         && ( h->param.rc.i_rc_method == X264_RC_ABR
 993           || h->param.rc.i_rc_method == X264_RC_CRF
 994           || h->param.i_bframe_adaptive
 995           || h->param.i_scenecut_threshold
 996           || h->param.rc.b_mb_tree
 997           || h->param.analyse.i_weighted_pred );
 998     h->frames.b_have_lowres |= h->param.rc.b_stat_read && h->param.rc.i_vbv_buffer_size > 0;
 999     h->frames.b_have_sub8x8_esa = !!(h->param.analyse.inter & X264_ANALYSE_PSUB8x8);
1000
1001     h->frames.i_last_idr =
1002     h->frames.i_last_keyframe = - h->param.i_keyint_max;
1003     h->frames.i_input    = 0;
1004     h->frames.i_largest_pts = h->frames.i_second_largest_pts = -1;
1005     h->frames.i_poc_last_open_gop = -1;
1006
1007     CHECKED_MALLOCZERO( h->frames.unused[0], (h->frames.i_delay + 3) * sizeof(x264_frame_t *) );
1008     /* Allocate room for max refs plus a few extra just in case. */
1009     CHECKED_MALLOCZERO( h->frames.unused[1], (h->i_thread_frames + X264_REF_MAX + 4) * sizeof(x264_frame_t *) );
1010     CHECKED_MALLOCZERO( h->frames.current, (h->param.i_sync_lookahead + h->param.i_bframe
1011                         + h->i_thread_frames + 3) * sizeof(x264_frame_t *) );
1012     if( h->param.analyse.i_weighted_pred > 0 )
1013         CHECKED_MALLOCZERO( h->frames.blank_unused, h->i_thread_frames * 4 * sizeof(x264_frame_t *) );
1014     h->i_ref0 = 0;
1015     h->i_ref1 = 0;
1016     h->i_cpb_delay = h->i_coded_fields = h->i_disp_fields = 0;
1017     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);
1018     h->i_disp_fields_last_frame = -1;
1019     x264_rdo_init();
1020
1021     /* init CPU functions */
1022     x264_predict_16x16_init( h->param.cpu, h->predict_16x16 );
1023     x264_predict_8x8c_init( h->param.cpu, h->predict_8x8c );
1024     x264_predict_8x8_init( h->param.cpu, h->predict_8x8, &h->predict_8x8_filter );
1025     x264_predict_4x4_init( h->param.cpu, h->predict_4x4 );
1026     if( !h->param.b_cabac )
1027         x264_init_vlc_tables();
1028     x264_pixel_init( h->param.cpu, &h->pixf );
1029     x264_dct_init( h->param.cpu, &h->dctf );
1030     x264_zigzag_init( h->param.cpu, &h->zigzagf, h->param.b_interlaced );
1031     x264_mc_init( h->param.cpu, &h->mc );
1032     x264_quant_init( h, h->param.cpu, &h->quantf );
1033     x264_deblock_init( h->param.cpu, &h->loopf );
1034     x264_bitstream_init( h->param.cpu, &h->bsf );
1035     x264_dct_init_weights();
1036
1037     mbcmp_init( h );
1038
1039     p = buf + sprintf( buf, "using cpu capabilities:" );
1040     for( int i = 0; x264_cpu_names[i].flags; i++ )
1041     {
1042         if( !strcmp(x264_cpu_names[i].name, "SSE2")
1043             && h->param.cpu & (X264_CPU_SSE2_IS_FAST|X264_CPU_SSE2_IS_SLOW) )
1044             continue;
1045         if( !strcmp(x264_cpu_names[i].name, "SSE3")
1046             && (h->param.cpu & X264_CPU_SSSE3 || !(h->param.cpu & X264_CPU_CACHELINE_64)) )
1047             continue;
1048         if( !strcmp(x264_cpu_names[i].name, "SSE4.1")
1049             && (h->param.cpu & X264_CPU_SSE42) )
1050             continue;
1051         if( (h->param.cpu & x264_cpu_names[i].flags) == x264_cpu_names[i].flags
1052             && (!i || x264_cpu_names[i].flags != x264_cpu_names[i-1].flags) )
1053             p += sprintf( p, " %s", x264_cpu_names[i].name );
1054     }
1055     if( !h->param.cpu )
1056         p += sprintf( p, " none!" );
1057     x264_log( h, X264_LOG_INFO, "%s\n", buf );
1058
1059     for( qp = h->param.rc.i_qp_min; qp <= h->param.rc.i_qp_max; qp++ )
1060         if( x264_analyse_init_costs( h, qp ) )
1061             goto fail;
1062     if( x264_analyse_init_costs( h, X264_LOOKAHEAD_QP ) )
1063         goto fail;
1064
1065     static const uint16_t cost_mv_correct[7] = { 24, 47, 95, 189, 379, 757, 1515 };
1066     /* Checks for known miscompilation issues. */
1067     if( h->cost_mv[x264_lambda_tab[X264_LOOKAHEAD_QP]][2013] != cost_mv_correct[BIT_DEPTH-8] )
1068     {
1069         x264_log( h, X264_LOG_ERROR, "MV cost test failed: x264 has been miscompiled!\n" );
1070         goto fail;
1071     }
1072
1073     /* Must be volatile or else GCC will optimize it out. */
1074     volatile int temp = 392;
1075     if( x264_clz( temp ) != 23 )
1076     {
1077         x264_log( h, X264_LOG_ERROR, "CLZ test failed: x264 has been miscompiled!\n" );
1078 #if ARCH_X86 || ARCH_X86_64
1079         x264_log( h, X264_LOG_ERROR, "Are you attempting to run an SSE4a-targeted build on a CPU that\n" );
1080         x264_log( h, X264_LOG_ERROR, "doesn't support it?\n" );
1081 #endif
1082         goto fail;
1083     }
1084
1085     h->out.i_nal = 0;
1086     h->out.i_bitstream = X264_MAX( 1000000, h->param.i_width * h->param.i_height * 4
1087         * ( h->param.rc.i_rc_method == X264_RC_ABR ? pow( 0.95, h->param.rc.i_qp_min )
1088           : pow( 0.95, h->param.rc.i_qp_constant ) * X264_MAX( 1, h->param.rc.f_ip_factor )));
1089
1090     CHECKED_MALLOC( h->nal_buffer, h->out.i_bitstream * 3/2 + 4 );
1091     h->nal_buffer_size = h->out.i_bitstream * 3/2 + 4;
1092
1093     if( h->param.i_threads > 1 &&
1094         x264_threadpool_init( &h->threadpool, h->param.i_threads, (void*)x264_encoder_thread_init, h ) )
1095         goto fail;
1096
1097     h->thread[0] = h;
1098     for( int i = 1; i < h->param.i_threads + !!h->param.i_sync_lookahead; i++ )
1099         CHECKED_MALLOC( h->thread[i], sizeof(x264_t) );
1100
1101     for( int i = 0; i < h->param.i_threads; i++ )
1102     {
1103         int init_nal_count = h->param.i_slice_count + 3;
1104         int allocate_threadlocal_data = !h->param.b_sliced_threads || !i;
1105         if( i > 0 )
1106             *h->thread[i] = *h;
1107
1108         if( allocate_threadlocal_data )
1109         {
1110             h->thread[i]->fdec = x264_frame_pop_unused( h, 1 );
1111             if( !h->thread[i]->fdec )
1112                 goto fail;
1113         }
1114         else
1115             h->thread[i]->fdec = h->thread[0]->fdec;
1116
1117         CHECKED_MALLOC( h->thread[i]->out.p_bitstream, h->out.i_bitstream );
1118         /* Start each thread with room for init_nal_count NAL units; it'll realloc later if needed. */
1119         CHECKED_MALLOC( h->thread[i]->out.nal, init_nal_count*sizeof(x264_nal_t) );
1120         h->thread[i]->out.i_nals_allocated = init_nal_count;
1121
1122         if( allocate_threadlocal_data && x264_macroblock_cache_allocate( h->thread[i] ) < 0 )
1123             goto fail;
1124     }
1125
1126     if( x264_lookahead_init( h, i_slicetype_length ) )
1127         goto fail;
1128
1129     for( int i = 0; i < h->param.i_threads; i++ )
1130         if( x264_macroblock_thread_allocate( h->thread[i], 0 ) < 0 )
1131             goto fail;
1132
1133     if( x264_ratecontrol_new( h ) < 0 )
1134         goto fail;
1135
1136     if( h->param.i_nal_hrd )
1137     {
1138         x264_log( h, X264_LOG_DEBUG, "HRD bitrate: %i bits/sec\n", h->sps->vui.hrd.i_bit_rate_unscaled );
1139         x264_log( h, X264_LOG_DEBUG, "CPB size: %i bits\n", h->sps->vui.hrd.i_cpb_size_unscaled );
1140     }
1141
1142     if( h->param.psz_dump_yuv )
1143     {
1144         /* create or truncate the reconstructed video file */
1145         FILE *f = fopen( h->param.psz_dump_yuv, "w" );
1146         if( !f )
1147         {
1148             x264_log( h, X264_LOG_ERROR, "dump_yuv: can't write to %s\n", h->param.psz_dump_yuv );
1149             goto fail;
1150         }
1151         else if( !x264_is_regular_file( f ) )
1152         {
1153             x264_log( h, X264_LOG_ERROR, "dump_yuv: incompatible with non-regular file %s\n", h->param.psz_dump_yuv );
1154             goto fail;
1155         }
1156         fclose( f );
1157     }
1158
1159     const char *profile = h->sps->i_profile_idc == PROFILE_BASELINE ? "Constrained Baseline" :
1160                           h->sps->i_profile_idc == PROFILE_MAIN ? "Main" :
1161                           h->sps->i_profile_idc == PROFILE_HIGH ? "High" :
1162                           h->sps->i_profile_idc == PROFILE_HIGH10 ? (h->sps->b_constraint_set3 == 1 ? "High 10 Intra" : "High 10") :
1163                           "High 4:4:4 Predictive";
1164     char level[4];
1165     snprintf( level, sizeof(level), "%d.%d", h->sps->i_level_idc/10, h->sps->i_level_idc%10 );
1166     if( h->sps->i_level_idc == 9 || ( h->sps->i_level_idc == 11 && h->sps->b_constraint_set3 ) )
1167         strcpy( level, "1b" );
1168
1169     if( h->sps->i_profile_idc < PROFILE_HIGH10 )
1170     {
1171         x264_log( h, X264_LOG_INFO, "profile %s, level %s\n",
1172             profile, level );
1173     }
1174     else
1175     {
1176         x264_log( h, X264_LOG_INFO, "profile %s, level %s, bit depth %d\n",
1177             profile, level, BIT_DEPTH );
1178     }
1179
1180     return h;
1181 fail:
1182     x264_free( h );
1183     return NULL;
1184 }
1185
1186 /****************************************************************************
1187  * x264_encoder_reconfig:
1188  ****************************************************************************/
1189 int x264_encoder_reconfig( x264_t *h, x264_param_t *param )
1190 {
1191     int rc_reconfig = 0;
1192     h = h->thread[h->thread[0]->i_thread_phase];
1193     x264_set_aspect_ratio( h, param, 0 );
1194 #define COPY(var) h->param.var = param->var
1195     COPY( i_frame_reference ); // but never uses more refs than initially specified
1196     COPY( i_bframe_bias );
1197     if( h->param.i_scenecut_threshold )
1198         COPY( i_scenecut_threshold ); // can't turn it on or off, only vary the threshold
1199     COPY( b_deblocking_filter );
1200     COPY( i_deblocking_filter_alphac0 );
1201     COPY( i_deblocking_filter_beta );
1202     COPY( analyse.inter );
1203     COPY( analyse.intra );
1204     COPY( analyse.i_direct_mv_pred );
1205     /* Scratch buffer prevents me_range from being increased for esa/tesa */
1206     if( h->param.analyse.i_me_method < X264_ME_ESA || param->analyse.i_me_range < h->param.analyse.i_me_range )
1207         COPY( analyse.i_me_range );
1208     COPY( analyse.i_noise_reduction );
1209     /* We can't switch out of subme=0 during encoding. */
1210     if( h->param.analyse.i_subpel_refine )
1211         COPY( analyse.i_subpel_refine );
1212     COPY( analyse.i_trellis );
1213     COPY( analyse.b_chroma_me );
1214     COPY( analyse.b_dct_decimate );
1215     COPY( analyse.b_fast_pskip );
1216     COPY( analyse.b_mixed_references );
1217     COPY( analyse.f_psy_rd );
1218     COPY( analyse.f_psy_trellis );
1219     COPY( crop_rect );
1220     // can only twiddle these if they were enabled to begin with:
1221     if( h->param.analyse.i_me_method >= X264_ME_ESA || param->analyse.i_me_method < X264_ME_ESA )
1222         COPY( analyse.i_me_method );
1223     if( h->param.analyse.i_me_method >= X264_ME_ESA && !h->frames.b_have_sub8x8_esa )
1224         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
1225     if( h->pps->b_transform_8x8_mode )
1226         COPY( analyse.b_transform_8x8 );
1227     if( h->frames.i_max_ref1 > 1 )
1228         COPY( i_bframe_pyramid );
1229     COPY( i_slice_max_size );
1230     COPY( i_slice_max_mbs );
1231     COPY( i_slice_count );
1232     COPY( b_tff );
1233
1234     /* VBV can't be turned on if it wasn't on to begin with */
1235     if( h->param.rc.i_vbv_max_bitrate > 0 && h->param.rc.i_vbv_buffer_size > 0 &&
1236           param->rc.i_vbv_max_bitrate > 0 &&   param->rc.i_vbv_buffer_size > 0 )
1237     {
1238         COPY( rc.i_vbv_max_bitrate );
1239         COPY( rc.i_vbv_buffer_size );
1240         COPY( rc.i_bitrate );
1241         rc_reconfig = 1;
1242     }
1243     if( h->param.rc.f_rf_constant != param->rc.f_rf_constant )
1244     {
1245         COPY( rc.f_rf_constant );
1246         rc_reconfig = 1;
1247     }
1248     if( h->param.rc.f_rf_constant_max != param->rc.f_rf_constant_max )
1249     {
1250         COPY( rc.f_rf_constant_max );
1251         rc_reconfig = 1;
1252     }
1253
1254 #undef COPY
1255
1256     mbcmp_init( h );
1257
1258     int ret = x264_validate_parameters( h );
1259
1260     /* Supported reconfiguration options (1-pass only):
1261      * vbv-maxrate
1262      * vbv-bufsize
1263      * crf
1264      * bitrate (CBR only) */
1265     if( !ret && rc_reconfig )
1266         x264_ratecontrol_init_reconfigurable( h, 0 );
1267
1268     return ret;
1269 }
1270
1271 /****************************************************************************
1272  * x264_encoder_parameters:
1273  ****************************************************************************/
1274 void x264_encoder_parameters( x264_t *h, x264_param_t *param )
1275 {
1276     memcpy( param, &h->thread[h->i_thread_phase]->param, sizeof(x264_param_t) );
1277 }
1278
1279 /* internal usage */
1280 static void x264_nal_start( x264_t *h, int i_type, int i_ref_idc )
1281 {
1282     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1283
1284     nal->i_ref_idc        = i_ref_idc;
1285     nal->i_type           = i_type;
1286     nal->b_long_startcode = 1;
1287
1288     nal->i_payload= 0;
1289     nal->p_payload= &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8];
1290 }
1291
1292 /* if number of allocated nals is not enough, re-allocate a larger one. */
1293 static int x264_nal_check_buffer( x264_t *h )
1294 {
1295     if( h->out.i_nal >= h->out.i_nals_allocated )
1296     {
1297         x264_nal_t *new_out = x264_malloc( sizeof(x264_nal_t) * (h->out.i_nals_allocated*2) );
1298         if( !new_out )
1299             return -1;
1300         memcpy( new_out, h->out.nal, sizeof(x264_nal_t) * (h->out.i_nals_allocated) );
1301         x264_free( h->out.nal );
1302         h->out.nal = new_out;
1303         h->out.i_nals_allocated *= 2;
1304     }
1305     return 0;
1306 }
1307
1308 static int x264_nal_end( x264_t *h )
1309 {
1310     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1311     nal->i_payload = &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8] - nal->p_payload;
1312     if( h->param.nalu_process )
1313         h->param.nalu_process( h, nal );
1314     h->out.i_nal++;
1315
1316     return x264_nal_check_buffer( h );
1317 }
1318
1319 static int x264_encoder_encapsulate_nals( x264_t *h, int start )
1320 {
1321     int nal_size = 0, previous_nal_size = 0;
1322
1323     if( h->param.nalu_process )
1324     {
1325         for( int i = start; i < h->out.i_nal; i++ )
1326             nal_size += h->out.nal[i].i_payload;
1327         return nal_size;
1328     }
1329
1330     for( int i = 0; i < start; i++ )
1331         previous_nal_size += h->out.nal[i].i_payload;
1332
1333     for( int i = start; i < h->out.i_nal; i++ )
1334         nal_size += h->out.nal[i].i_payload;
1335
1336     /* Worst-case NAL unit escaping: reallocate the buffer if it's too small. */
1337     if( h->nal_buffer_size < nal_size * 3/2 + h->out.i_nal * 4 )
1338     {
1339         uint8_t *buf = x264_malloc( nal_size * 2 + h->out.i_nal * 4 );
1340         if( !buf )
1341             return -1;
1342         if( previous_nal_size )
1343             memcpy( buf, h->nal_buffer, previous_nal_size );
1344         x264_free( h->nal_buffer );
1345         h->nal_buffer = buf;
1346     }
1347
1348     uint8_t *nal_buffer = h->nal_buffer + previous_nal_size;
1349
1350     for( int i = start; i < h->out.i_nal; i++ )
1351     {
1352         h->out.nal[i].b_long_startcode = !i || h->out.nal[i].i_type == NAL_SPS || h->out.nal[i].i_type == NAL_PPS;
1353         x264_nal_encode( h, nal_buffer, &h->out.nal[i] );
1354         nal_buffer += h->out.nal[i].i_payload;
1355     }
1356
1357     x264_emms();
1358
1359     return nal_buffer - (h->nal_buffer + previous_nal_size);
1360 }
1361
1362 /****************************************************************************
1363  * x264_encoder_headers:
1364  ****************************************************************************/
1365 int x264_encoder_headers( x264_t *h, x264_nal_t **pp_nal, int *pi_nal )
1366 {
1367     int frame_size = 0;
1368     /* init bitstream context */
1369     h->out.i_nal = 0;
1370     bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
1371
1372     /* Write SEI, SPS and PPS. */
1373
1374     /* generate sequence parameters */
1375     x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
1376     x264_sps_write( &h->out.bs, h->sps );
1377     if( x264_nal_end( h ) )
1378         return -1;
1379
1380     /* generate picture parameters */
1381     x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
1382     x264_pps_write( &h->out.bs, h->pps );
1383     if( x264_nal_end( h ) )
1384         return -1;
1385
1386     /* identify ourselves */
1387     x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
1388     if( x264_sei_version_write( h, &h->out.bs ) )
1389         return -1;
1390     if( x264_nal_end( h ) )
1391         return -1;
1392
1393     frame_size = x264_encoder_encapsulate_nals( h, 0 );
1394
1395     /* now set output*/
1396     *pi_nal = h->out.i_nal;
1397     *pp_nal = &h->out.nal[0];
1398     h->out.i_nal = 0;
1399
1400     return frame_size;
1401 }
1402
1403 /* Check to see whether we have chosen a reference list ordering different
1404  * from the standard's default. */
1405 static inline void x264_reference_check_reorder( x264_t *h )
1406 {
1407     /* The reorder check doesn't check for missing frames, so just
1408      * force a reorder if one of the reference list is corrupt. */
1409     for( int i = 0; h->frames.reference[i]; i++ )
1410         if( h->frames.reference[i]->b_corrupt )
1411         {
1412             h->b_ref_reorder[0] = 1;
1413             return;
1414         }
1415     for( int i = 0; i < h->i_ref0 - 1; i++ )
1416         /* P and B-frames use different default orders. */
1417         if( h->sh.i_type == SLICE_TYPE_P ? h->fref0[i]->i_frame_num < h->fref0[i+1]->i_frame_num
1418                                          : h->fref0[i]->i_poc < h->fref0[i+1]->i_poc )
1419         {
1420             h->b_ref_reorder[0] = 1;
1421             return;
1422         }
1423 }
1424
1425 /* return -1 on failure, else return the index of the new reference frame */
1426 int x264_weighted_reference_duplicate( x264_t *h, int i_ref, const x264_weight_t *w )
1427 {
1428     int i = h->i_ref0;
1429     int j = 1;
1430     x264_frame_t *newframe;
1431     if( i <= 1 ) /* empty list, definitely can't duplicate frame */
1432         return -1;
1433
1434     //Duplication is only used in X264_WEIGHTP_SMART
1435     if( h->param.analyse.i_weighted_pred != X264_WEIGHTP_SMART )
1436         return -1;
1437
1438     /* Duplication is a hack to compensate for crappy rounding in motion compensation.
1439      * With high bit depth, it's not worth doing, so turn it off except in the case of
1440      * unweighted dupes. */
1441     if( BIT_DEPTH > 8 && w != weight_none )
1442         return -1;
1443
1444     newframe = x264_frame_pop_blank_unused( h );
1445
1446     //FIXME: probably don't need to copy everything
1447     *newframe = *h->fref0[i_ref];
1448     newframe->i_reference_count = 1;
1449     newframe->orig = h->fref0[i_ref];
1450     newframe->b_duplicate = 1;
1451     memcpy( h->fenc->weight[j], w, sizeof(h->fenc->weight[i]) );
1452
1453     /* shift the frames to make space for the dupe. */
1454     h->b_ref_reorder[0] = 1;
1455     if( h->i_ref0 < X264_REF_MAX )
1456         ++h->i_ref0;
1457     h->fref0[X264_REF_MAX-1] = NULL;
1458     x264_frame_unshift( &h->fref0[j], newframe );
1459
1460     return j;
1461 }
1462
1463 static void x264_weighted_pred_init( x264_t *h )
1464 {
1465     /* for now no analysis and set all weights to nothing */
1466     for( int i_ref = 0; i_ref < h->i_ref0; i_ref++ )
1467         h->fenc->weighted[i_ref] = h->fref0[i_ref]->filtered[0];
1468
1469     // FIXME: This only supports weighting of one reference frame
1470     // and duplicates of that frame.
1471     h->fenc->i_lines_weighted = 0;
1472
1473     for( int i_ref = 0; i_ref < (h->i_ref0 << h->sh.b_mbaff); i_ref++ )
1474         for( int i = 0; i < 3; i++ )
1475             h->sh.weight[i_ref][i].weightfn = NULL;
1476
1477
1478     if( h->sh.i_type != SLICE_TYPE_P || h->param.analyse.i_weighted_pred <= 0 )
1479         return;
1480
1481     int i_padv = PADV << h->param.b_interlaced;
1482     int denom = -1;
1483     int weightplane[2] = { 0, 0 };
1484     int buffer_next = 0;
1485     for( int i = 0; i < 3; i++ )
1486     {
1487         for( int j = 0; j < h->i_ref0; j++ )
1488         {
1489             if( h->fenc->weight[j][i].weightfn )
1490             {
1491                 h->sh.weight[j][i] = h->fenc->weight[j][i];
1492                 // if weight is useless, don't write it to stream
1493                 if( h->sh.weight[j][i].i_scale == 1<<h->sh.weight[j][i].i_denom && h->sh.weight[j][i].i_offset == 0 )
1494                     h->sh.weight[j][i].weightfn = NULL;
1495                 else
1496                 {
1497                     if( !weightplane[!!i] )
1498                     {
1499                         weightplane[!!i] = 1;
1500                         h->sh.weight[0][!!i].i_denom = denom = h->sh.weight[j][i].i_denom;
1501                         assert( x264_clip3( denom, 0, 7 ) == denom );
1502                     }
1503
1504                     assert( h->sh.weight[j][i].i_denom == denom );
1505                     if( !i )
1506                     {
1507                         h->fenc->weighted[j] = h->mb.p_weight_buf[buffer_next++] + h->fenc->i_stride[0] * i_padv + PADH;
1508                         //scale full resolution frame
1509                         if( h->param.i_threads == 1 )
1510                         {
1511                             pixel *src = h->fref0[j]->filtered[0] - h->fref0[j]->i_stride[0]*i_padv - PADH;
1512                             pixel *dst = h->fenc->weighted[j] - h->fenc->i_stride[0]*i_padv - PADH;
1513                             int stride = h->fenc->i_stride[0];
1514                             int width = h->fenc->i_width[0] + PADH*2;
1515                             int height = h->fenc->i_lines[0] + i_padv*2;
1516                             x264_weight_scale_plane( h, dst, stride, src, stride, width, height, &h->sh.weight[j][0] );
1517                             h->fenc->i_lines_weighted = height;
1518                         }
1519                     }
1520                 }
1521             }
1522         }
1523     }
1524
1525     if( weightplane[1] )
1526         for( int i = 0; i < h->i_ref0; i++ )
1527         {
1528             if( h->sh.weight[i][1].weightfn && !h->sh.weight[i][2].weightfn )
1529             {
1530                 h->sh.weight[i][2].i_scale = 1 << h->sh.weight[0][1].i_denom;
1531                 h->sh.weight[i][2].i_offset = 0;
1532             }
1533             else if( h->sh.weight[i][2].weightfn && !h->sh.weight[i][1].weightfn )
1534             {
1535                 h->sh.weight[i][1].i_scale = 1 << h->sh.weight[0][1].i_denom;
1536                 h->sh.weight[i][1].i_offset = 0;
1537             }
1538         }
1539
1540     if( !weightplane[0] )
1541         h->sh.weight[0][0].i_denom = 0;
1542     if( !weightplane[1] )
1543         h->sh.weight[0][1].i_denom = 0;
1544     h->sh.weight[0][2].i_denom = h->sh.weight[0][1].i_denom;
1545 }
1546
1547 static inline void x264_reference_build_list( x264_t *h, int i_poc )
1548 {
1549     int b_ok;
1550
1551     /* build ref list 0/1 */
1552     h->mb.pic.i_fref[0] = h->i_ref0 = 0;
1553     h->mb.pic.i_fref[1] = h->i_ref1 = 0;
1554     if( h->sh.i_type == SLICE_TYPE_I )
1555         return;
1556
1557     for( int i = 0; h->frames.reference[i]; i++ )
1558     {
1559         if( h->frames.reference[i]->b_corrupt )
1560             continue;
1561         if( h->frames.reference[i]->i_poc < i_poc )
1562             h->fref0[h->i_ref0++] = h->frames.reference[i];
1563         else if( h->frames.reference[i]->i_poc > i_poc )
1564             h->fref1[h->i_ref1++] = h->frames.reference[i];
1565     }
1566
1567     /* Order ref0 from higher to lower poc */
1568     do
1569     {
1570         b_ok = 1;
1571         for( int i = 0; i < h->i_ref0 - 1; i++ )
1572         {
1573             if( h->fref0[i]->i_poc < h->fref0[i+1]->i_poc )
1574             {
1575                 XCHG( x264_frame_t*, h->fref0[i], h->fref0[i+1] );
1576                 b_ok = 0;
1577                 break;
1578             }
1579         }
1580     } while( !b_ok );
1581
1582     if( h->sh.i_mmco_remove_from_end )
1583         for( int i = h->i_ref0-1; i >= h->i_ref0 - h->sh.i_mmco_remove_from_end; i-- )
1584         {
1585             int diff = h->i_frame_num - h->fref0[i]->i_frame_num;
1586             h->sh.mmco[h->sh.i_mmco_command_count].i_poc = h->fref0[i]->i_poc;
1587             h->sh.mmco[h->sh.i_mmco_command_count++].i_difference_of_pic_nums = diff;
1588         }
1589
1590     /* Order ref1 from lower to higher poc (bubble sort) for B-frame */
1591     do
1592     {
1593         b_ok = 1;
1594         for( int i = 0; i < h->i_ref1 - 1; i++ )
1595         {
1596             if( h->fref1[i]->i_poc > h->fref1[i+1]->i_poc )
1597             {
1598                 XCHG( x264_frame_t*, h->fref1[i], h->fref1[i+1] );
1599                 b_ok = 0;
1600                 break;
1601             }
1602         }
1603     } while( !b_ok );
1604
1605     x264_reference_check_reorder( h );
1606
1607     h->i_ref1 = X264_MIN( h->i_ref1, h->frames.i_max_ref1 );
1608     h->i_ref0 = X264_MIN( h->i_ref0, h->frames.i_max_ref0 );
1609     h->i_ref0 = X264_MIN( h->i_ref0, h->param.i_frame_reference ); // if reconfig() has lowered the limit
1610
1611     /* add duplicates */
1612     if( h->fenc->i_type == X264_TYPE_P )
1613     {
1614         int idx = -1;
1615         if( h->param.analyse.i_weighted_pred >= X264_WEIGHTP_SIMPLE )
1616         {
1617             x264_weight_t w[3];
1618             w[1].weightfn = w[2].weightfn = NULL;
1619             if( h->param.rc.b_stat_read )
1620                 x264_ratecontrol_set_weights( h, h->fenc );
1621
1622             if( !h->fenc->weight[0][0].weightfn )
1623             {
1624                 h->fenc->weight[0][0].i_denom = 0;
1625                 SET_WEIGHT( w[0], 1, 1, 0, -1 );
1626                 idx = x264_weighted_reference_duplicate( h, 0, w );
1627             }
1628             else
1629             {
1630                 if( h->fenc->weight[0][0].i_scale == 1<<h->fenc->weight[0][0].i_denom )
1631                 {
1632                     SET_WEIGHT( h->fenc->weight[0][0], 1, 1, 0, h->fenc->weight[0][0].i_offset );
1633                 }
1634                 x264_weighted_reference_duplicate( h, 0, weight_none );
1635                 if( h->fenc->weight[0][0].i_offset > -128 )
1636                 {
1637                     w[0] = h->fenc->weight[0][0];
1638                     w[0].i_offset--;
1639                     h->mc.weight_cache( h, &w[0] );
1640                     idx = x264_weighted_reference_duplicate( h, 0, w );
1641                 }
1642             }
1643         }
1644         h->mb.ref_blind_dupe = idx;
1645     }
1646
1647     assert( h->i_ref0 + h->i_ref1 <= X264_REF_MAX );
1648     h->mb.pic.i_fref[0] = h->i_ref0;
1649     h->mb.pic.i_fref[1] = h->i_ref1;
1650 }
1651
1652 static void x264_fdec_filter_row( x264_t *h, int mb_y, int b_inloop )
1653 {
1654     /* mb_y is the mb to be encoded next, not the mb to be filtered here */
1655     int b_hpel = h->fdec->b_kept_as_ref;
1656     int b_deblock = h->sh.i_disable_deblocking_filter_idc != 1;
1657     int b_end = mb_y == h->i_threadslice_end;
1658     int b_measure_quality = 1;
1659     int min_y = mb_y - (1 << h->sh.b_mbaff);
1660     int b_start = min_y == h->i_threadslice_start;
1661     int max_y = b_end ? h->i_threadslice_end : mb_y;
1662     b_deblock &= b_hpel || h->param.psz_dump_yuv;
1663     if( h->param.b_sliced_threads && b_start && min_y && !b_inloop )
1664     {
1665         b_deblock = 0;         /* We already deblocked on the inloop pass. */
1666         b_measure_quality = 0; /* We already measured quality on the inloop pass. */
1667     }
1668     if( mb_y & h->sh.b_mbaff )
1669         return;
1670     if( min_y < h->i_threadslice_start )
1671         return;
1672
1673     if( b_deblock )
1674         for( int y = min_y; y < max_y; y += (1 << h->sh.b_mbaff) )
1675             x264_frame_deblock_row( h, y );
1676
1677     if( b_hpel )
1678     {
1679         int end = mb_y == h->mb.i_mb_height;
1680         x264_frame_expand_border( h, h->fdec, min_y, end );
1681         if( h->param.analyse.i_subpel_refine )
1682         {
1683             x264_frame_filter( h, h->fdec, min_y, end );
1684             x264_frame_expand_border_filtered( h, h->fdec, min_y, end );
1685         }
1686     }
1687
1688     if( h->i_thread_frames > 1 && h->fdec->b_kept_as_ref )
1689         x264_frame_cond_broadcast( h->fdec, mb_y*16 + (b_end ? 10000 : -(X264_THREAD_HEIGHT << h->sh.b_mbaff)) );
1690
1691     min_y = min_y*16 - 8 * !b_start;
1692     max_y = b_end ? X264_MIN( h->i_threadslice_end*16 , h->param.i_height ) : mb_y*16 - 8;
1693
1694     if( b_measure_quality )
1695     {
1696         if( h->param.analyse.b_psnr )
1697         {
1698             uint64_t ssd_y = x264_pixel_ssd_wxh( &h->pixf,
1699                 h->fdec->plane[0] + min_y * h->fdec->i_stride[0], h->fdec->i_stride[0],
1700                 h->fenc->plane[0] + min_y * h->fenc->i_stride[0], h->fenc->i_stride[0],
1701                 h->param.i_width, max_y-min_y );
1702             uint64_t ssd_u, ssd_v;
1703             x264_pixel_ssd_nv12( &h->pixf,
1704                 h->fdec->plane[1] + (min_y>>1) * h->fdec->i_stride[1], h->fdec->i_stride[1],
1705                 h->fenc->plane[1] + (min_y>>1) * h->fenc->i_stride[1], h->fenc->i_stride[1],
1706                 h->param.i_width>>1, (max_y-min_y)>>1, &ssd_u, &ssd_v );
1707             h->stat.frame.i_ssd[0] += ssd_y;
1708             h->stat.frame.i_ssd[1] += ssd_u;
1709             h->stat.frame.i_ssd[2] += ssd_v;
1710         }
1711
1712         if( h->param.analyse.b_ssim )
1713         {
1714             x264_emms();
1715             /* offset by 2 pixels to avoid alignment of ssim blocks with dct blocks,
1716              * and overlap by 4 */
1717             min_y += b_start ? 2 : -6;
1718             h->stat.frame.f_ssim +=
1719                 x264_pixel_ssim_wxh( &h->pixf,
1720                     h->fdec->plane[0] + 2+min_y*h->fdec->i_stride[0], h->fdec->i_stride[0],
1721                     h->fenc->plane[0] + 2+min_y*h->fenc->i_stride[0], h->fenc->i_stride[0],
1722                     h->param.i_width-2, max_y-min_y, h->scratch_buffer );
1723         }
1724     }
1725 }
1726
1727 static inline int x264_reference_update( x264_t *h )
1728 {
1729     if( !h->fdec->b_kept_as_ref )
1730     {
1731         if( h->i_thread_frames > 1 )
1732         {
1733             x264_frame_push_unused( h, h->fdec );
1734             h->fdec = x264_frame_pop_unused( h, 1 );
1735             if( !h->fdec )
1736                 return -1;
1737         }
1738         return 0;
1739     }
1740
1741     /* apply mmco from previous frame. */
1742     for( int i = 0; i < h->sh.i_mmco_command_count; i++ )
1743         for( int j = 0; h->frames.reference[j]; j++ )
1744             if( h->frames.reference[j]->i_poc == h->sh.mmco[i].i_poc )
1745                 x264_frame_push_unused( h, x264_frame_shift( &h->frames.reference[j] ) );
1746
1747     /* move frame in the buffer */
1748     x264_frame_push( h->frames.reference, h->fdec );
1749     if( h->frames.reference[h->sps->i_num_ref_frames] )
1750         x264_frame_push_unused( h, x264_frame_shift( h->frames.reference ) );
1751     h->fdec = x264_frame_pop_unused( h, 1 );
1752     if( !h->fdec )
1753         return -1;
1754     return 0;
1755 }
1756
1757 static inline void x264_reference_reset( x264_t *h )
1758 {
1759     while( h->frames.reference[0] )
1760         x264_frame_push_unused( h, x264_frame_pop( h->frames.reference ) );
1761     h->fdec->i_poc =
1762     h->fenc->i_poc = 0;
1763 }
1764
1765 static inline void x264_reference_hierarchy_reset( x264_t *h )
1766 {
1767     int ref;
1768     int b_hasdelayframe = 0;
1769
1770     /* look for delay frames -- chain must only contain frames that are disposable */
1771     for( int i = 0; h->frames.current[i] && IS_DISPOSABLE( h->frames.current[i]->i_type ); i++ )
1772         b_hasdelayframe |= h->frames.current[i]->i_coded
1773                         != h->frames.current[i]->i_frame + h->sps->vui.i_num_reorder_frames;
1774
1775     /* This function must handle b-pyramid and clear frames for open-gop */
1776     if( h->param.i_bframe_pyramid != X264_B_PYRAMID_STRICT && !b_hasdelayframe && h->frames.i_poc_last_open_gop == -1 )
1777         return;
1778
1779     /* Remove last BREF. There will never be old BREFs in the
1780      * dpb during a BREF decode when pyramid == STRICT */
1781     for( ref = 0; h->frames.reference[ref]; ref++ )
1782     {
1783         if( ( h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT
1784             && h->frames.reference[ref]->i_type == X264_TYPE_BREF )
1785             || ( h->frames.reference[ref]->i_poc < h->frames.i_poc_last_open_gop
1786             && h->sh.i_type != SLICE_TYPE_B ) )
1787         {
1788             int diff = h->i_frame_num - h->frames.reference[ref]->i_frame_num;
1789             h->sh.mmco[h->sh.i_mmco_command_count].i_difference_of_pic_nums = diff;
1790             h->sh.mmco[h->sh.i_mmco_command_count++].i_poc = h->frames.reference[ref]->i_poc;
1791             x264_frame_push_unused( h, x264_frame_shift( &h->frames.reference[ref] ) );
1792             h->b_ref_reorder[0] = 1;
1793             ref--;
1794         }
1795     }
1796
1797     /* Prepare room in the dpb for the delayed display time of the later b-frame's */
1798     if( h->param.i_bframe_pyramid )
1799         h->sh.i_mmco_remove_from_end = X264_MAX( ref + 2 - h->frames.i_max_dpb, 0 );
1800 }
1801
1802 static inline void x264_slice_init( x264_t *h, int i_nal_type, int i_global_qp )
1803 {
1804     /* ------------------------ Create slice header  ----------------------- */
1805     if( i_nal_type == NAL_SLICE_IDR )
1806     {
1807         x264_slice_header_init( h, &h->sh, h->sps, h->pps, h->i_idr_pic_id, h->i_frame_num, i_global_qp );
1808
1809         /* alternate id */
1810         h->i_idr_pic_id ^= 1;
1811     }
1812     else
1813     {
1814         x264_slice_header_init( h, &h->sh, h->sps, h->pps, -1, h->i_frame_num, i_global_qp );
1815
1816         h->sh.i_num_ref_idx_l0_active = h->i_ref0 <= 0 ? 1 : h->i_ref0;
1817         h->sh.i_num_ref_idx_l1_active = h->i_ref1 <= 0 ? 1 : h->i_ref1;
1818         if( h->sh.i_num_ref_idx_l0_active != h->pps->i_num_ref_idx_l0_default_active ||
1819             (h->sh.i_type == SLICE_TYPE_B && h->sh.i_num_ref_idx_l1_active != h->pps->i_num_ref_idx_l1_default_active) )
1820         {
1821             h->sh.b_num_ref_idx_override = 1;
1822         }
1823     }
1824
1825     h->fdec->i_frame_num = h->sh.i_frame_num;
1826
1827     if( h->sps->i_poc_type == 0 )
1828     {
1829         h->sh.i_poc = h->fdec->i_poc;
1830         if( h->param.b_interlaced )
1831         {
1832             h->sh.i_delta_poc_bottom = h->param.b_tff ? 1 : -1;
1833             if( h->sh.i_delta_poc_bottom == -1 )
1834                 h->sh.i_poc = h->fdec->i_poc + 1;
1835         }
1836         else
1837             h->sh.i_delta_poc_bottom = 0;
1838     }
1839     else if( h->sps->i_poc_type == 1 )
1840     {
1841         /* FIXME TODO FIXME */
1842     }
1843     else
1844     {
1845         /* Nothing to do ? */
1846     }
1847
1848     x264_macroblock_slice_init( h );
1849 }
1850
1851 static int x264_slice_write( x264_t *h )
1852 {
1853     int i_skip;
1854     int mb_xy, i_mb_x, i_mb_y;
1855     int i_skip_bak = 0; /* Shut up GCC. */
1856     bs_t bs_bak;
1857     x264_cabac_t cabac_bak;
1858     uint8_t cabac_prevbyte_bak = 0; /* Shut up GCC. */
1859     int mv_bits_bak = 0;
1860     int tex_bits_bak = 0;
1861     /* NALUs other than the first use a 3-byte startcode.
1862      * Add one extra byte for the rbsp, and one more for the final CABAC putbyte.
1863      * Then add an extra 5 bytes just in case, to account for random NAL escapes and
1864      * other inaccuracies. */
1865     int overhead_guess = (NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal)) + 1 + h->param.b_cabac + 5;
1866     int slice_max_size = h->param.i_slice_max_size > 0 ? (h->param.i_slice_max_size-overhead_guess)*8 : 0;
1867     int starting_bits = bs_pos(&h->out.bs);
1868     int b_deblock = h->sh.i_disable_deblocking_filter_idc != 1;
1869     int b_hpel = h->fdec->b_kept_as_ref;
1870     b_deblock &= b_hpel || h->param.psz_dump_yuv;
1871     bs_realign( &h->out.bs );
1872
1873     /* Slice */
1874     x264_nal_start( h, h->i_nal_type, h->i_nal_ref_idc );
1875     h->out.nal[h->out.i_nal].i_first_mb = h->sh.i_first_mb;
1876
1877     /* Slice header */
1878     x264_macroblock_thread_init( h );
1879
1880     /* If this isn't the first slice in the threadslice, set the slice QP
1881      * equal to the last QP in the previous slice for more accurate
1882      * CABAC initialization. */
1883     if( h->sh.i_first_mb != h->i_threadslice_start * h->mb.i_mb_width )
1884     {
1885         h->sh.i_qp = h->mb.i_last_qp;
1886         h->sh.i_qp_delta = h->sh.i_qp - h->pps->i_pic_init_qp;
1887     }
1888
1889     x264_slice_header_write( &h->out.bs, &h->sh, h->i_nal_ref_idc );
1890     if( h->param.b_cabac )
1891     {
1892         /* alignment needed */
1893         bs_align_1( &h->out.bs );
1894
1895         /* init cabac */
1896         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 );
1897         x264_cabac_encode_init ( &h->cabac, h->out.bs.p, h->out.bs.p_end );
1898     }
1899     h->mb.i_last_qp = h->sh.i_qp;
1900     h->mb.i_last_dqp = 0;
1901
1902     i_mb_y = h->sh.i_first_mb / h->mb.i_mb_width;
1903     i_mb_x = h->sh.i_first_mb % h->mb.i_mb_width;
1904     i_skip = 0;
1905
1906     while( (mb_xy = i_mb_x + i_mb_y * h->mb.i_mb_width) <= h->sh.i_last_mb )
1907     {
1908         int mb_spos = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
1909
1910         if( x264_bitstream_check_buffer( h ) )
1911             return -1;
1912
1913         if( slice_max_size )
1914         {
1915             mv_bits_bak = h->stat.frame.i_mv_bits;
1916             tex_bits_bak = h->stat.frame.i_tex_bits;
1917             /* We don't need the contexts because flushing the CABAC encoder has no context
1918              * dependency and macroblocks are only re-encoded in the case where a slice is
1919              * ended (and thus the content of all contexts are thrown away). */
1920             if( h->param.b_cabac )
1921             {
1922                 memcpy( &cabac_bak, &h->cabac, offsetof(x264_cabac_t, f8_bits_encoded) );
1923                 /* x264's CABAC writer modifies the previous byte during carry, so it has to be
1924                  * backed up. */
1925                 cabac_prevbyte_bak = h->cabac.p[-1];
1926             }
1927             else
1928             {
1929                 bs_bak = h->out.bs;
1930                 i_skip_bak = i_skip;
1931             }
1932         }
1933
1934         if( i_mb_x == 0 && !h->mb.b_reencode_mb )
1935             x264_fdec_filter_row( h, i_mb_y, 1 );
1936
1937         /* load cache */
1938         x264_macroblock_cache_load( h, i_mb_x, i_mb_y );
1939
1940         x264_macroblock_analyse( h );
1941
1942         /* encode this macroblock -> be careful it can change the mb type to P_SKIP if needed */
1943         x264_macroblock_encode( h );
1944
1945         if( h->param.b_cabac )
1946         {
1947             if( mb_xy > h->sh.i_first_mb && !(h->sh.b_mbaff && (i_mb_y&1)) )
1948                 x264_cabac_encode_terminal( &h->cabac );
1949
1950             if( IS_SKIP( h->mb.i_type ) )
1951                 x264_cabac_mb_skip( h, 1 );
1952             else
1953             {
1954                 if( h->sh.i_type != SLICE_TYPE_I )
1955                     x264_cabac_mb_skip( h, 0 );
1956                 x264_macroblock_write_cabac( h, &h->cabac );
1957             }
1958         }
1959         else
1960         {
1961             if( IS_SKIP( h->mb.i_type ) )
1962                 i_skip++;
1963             else
1964             {
1965                 if( h->sh.i_type != SLICE_TYPE_I )
1966                 {
1967                     bs_write_ue( &h->out.bs, i_skip );  /* skip run */
1968                     i_skip = 0;
1969                 }
1970                 x264_macroblock_write_cavlc( h );
1971             }
1972         }
1973
1974         int total_bits = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
1975         int mb_size = total_bits - mb_spos;
1976
1977         if( slice_max_size )
1978         {
1979             /* Count the skip run, just in case. */
1980             if( !h->param.b_cabac )
1981                 total_bits += bs_size_ue_big( i_skip );
1982             /* HACK: we assume no more than 3 bytes of NALU escaping, but
1983              * this can fail in CABAC streams with an extremely large number of identical
1984              * blocks in sequence (e.g. all-black intra blocks).
1985              * Thus, every 64 blocks, pretend we've used a byte.
1986              * For reference, a seqeuence of identical empty-CBP i16x16 blocks will use
1987              * one byte after 26 macroblocks, assuming a perfectly adapted CABAC.
1988              * That's 78 macroblocks to generate the 3-byte sequence to trigger an escape. */
1989             else if( ((mb_xy - h->sh.i_first_mb) & 63) == 63 )
1990                 slice_max_size -= 8;
1991             /* We'll just re-encode this last macroblock if we go over the max slice size. */
1992             if( total_bits - starting_bits > slice_max_size && !h->mb.b_reencode_mb )
1993             {
1994                 if( mb_xy != h->sh.i_first_mb )
1995                 {
1996                     h->stat.frame.i_mv_bits = mv_bits_bak;
1997                     h->stat.frame.i_tex_bits = tex_bits_bak;
1998                     if( h->param.b_cabac )
1999                     {
2000                         memcpy( &h->cabac, &cabac_bak, offsetof(x264_cabac_t, f8_bits_encoded) );
2001                         h->cabac.p[-1] = cabac_prevbyte_bak;
2002                     }
2003                     else
2004                     {
2005                         h->out.bs = bs_bak;
2006                         i_skip = i_skip_bak;
2007                     }
2008                     h->mb.b_reencode_mb = 1;
2009                     h->sh.i_last_mb = mb_xy-1;
2010                     break;
2011                 }
2012                 else
2013                 {
2014                     h->sh.i_last_mb = mb_xy;
2015                     h->mb.b_reencode_mb = 0;
2016                 }
2017             }
2018             else
2019                 h->mb.b_reencode_mb = 0;
2020         }
2021
2022 #if HAVE_VISUALIZE
2023         if( h->param.b_visualize )
2024             x264_visualize_mb( h );
2025 #endif
2026
2027         /* save cache */
2028         x264_macroblock_cache_save( h );
2029
2030         /* accumulate mb stats */
2031         h->stat.frame.i_mb_count[h->mb.i_type]++;
2032
2033         int b_intra = IS_INTRA( h->mb.i_type );
2034         if( h->param.i_log_level >= X264_LOG_INFO || h->param.rc.b_stat_write )
2035         {
2036             if( !b_intra && !IS_SKIP( h->mb.i_type ) && !IS_DIRECT( h->mb.i_type ) )
2037             {
2038                 if( h->mb.i_partition != D_8x8 )
2039                         h->stat.frame.i_mb_partition[h->mb.i_partition] += 4;
2040                     else
2041                         for( int i = 0; i < 4; i++ )
2042                             h->stat.frame.i_mb_partition[h->mb.i_sub_partition[i]] ++;
2043                 if( h->param.i_frame_reference > 1 )
2044                     for( int i_list = 0; i_list <= (h->sh.i_type == SLICE_TYPE_B); i_list++ )
2045                         for( int i = 0; i < 4; i++ )
2046                         {
2047                             int i_ref = h->mb.cache.ref[i_list][ x264_scan8[4*i] ];
2048                             if( i_ref >= 0 )
2049                                 h->stat.frame.i_mb_count_ref[i_list][i_ref] ++;
2050                         }
2051             }
2052         }
2053
2054         if( h->param.i_log_level >= X264_LOG_INFO )
2055         {
2056             if( h->mb.i_cbp_luma | h->mb.i_cbp_chroma )
2057             {
2058                 int cbpsum = (h->mb.i_cbp_luma&1) + ((h->mb.i_cbp_luma>>1)&1)
2059                            + ((h->mb.i_cbp_luma>>2)&1) + (h->mb.i_cbp_luma>>3);
2060                 h->stat.frame.i_mb_cbp[!b_intra + 0] += cbpsum;
2061                 h->stat.frame.i_mb_cbp[!b_intra + 2] += !!h->mb.i_cbp_chroma;
2062                 h->stat.frame.i_mb_cbp[!b_intra + 4] += h->mb.i_cbp_chroma >> 1;
2063             }
2064             if( h->mb.i_cbp_luma && !b_intra )
2065             {
2066                 h->stat.frame.i_mb_count_8x8dct[0] ++;
2067                 h->stat.frame.i_mb_count_8x8dct[1] += h->mb.b_transform_8x8;
2068             }
2069             if( b_intra && h->mb.i_type != I_PCM )
2070             {
2071                 if( h->mb.i_type == I_16x16 )
2072                     h->stat.frame.i_mb_pred_mode[0][h->mb.i_intra16x16_pred_mode]++;
2073                 else if( h->mb.i_type == I_8x8 )
2074                     for( int i = 0; i < 16; i += 4 )
2075                         h->stat.frame.i_mb_pred_mode[1][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
2076                 else //if( h->mb.i_type == I_4x4 )
2077                     for( int i = 0; i < 16; i++ )
2078                         h->stat.frame.i_mb_pred_mode[2][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
2079                 h->stat.frame.i_mb_pred_mode[3][x264_mb_pred_mode8x8c_fix[h->mb.i_chroma_pred_mode]]++;
2080             }
2081         }
2082
2083         /* calculate deblock strength values (actual deblocking is done per-row along with hpel) */
2084         if( b_deblock )
2085         {
2086             int mvy_limit = 4 >> h->sh.b_mbaff;
2087             uint8_t (*bs)[4][4] = h->deblock_strength[h->mb.i_mb_y&1][h->mb.i_mb_x];
2088             x264_macroblock_cache_load_deblock( h );
2089             if( IS_INTRA( h->mb.type[h->mb.i_mb_xy] ) )
2090                 memset( bs, 3, 2*4*4*sizeof(uint8_t) );
2091             else
2092                 h->loopf.deblock_strength( h->mb.cache.non_zero_count, h->mb.cache.ref, h->mb.cache.mv,
2093                                            bs, mvy_limit, h->sh.i_type == SLICE_TYPE_B );
2094         }
2095
2096         x264_ratecontrol_mb( h, mb_size );
2097
2098         if( h->sh.b_mbaff )
2099         {
2100             i_mb_x += i_mb_y & 1;
2101             i_mb_y ^= i_mb_x < h->mb.i_mb_width;
2102         }
2103         else
2104             i_mb_x++;
2105         if( i_mb_x == h->mb.i_mb_width )
2106         {
2107             i_mb_y++;
2108             i_mb_x = 0;
2109         }
2110     }
2111     h->out.nal[h->out.i_nal].i_last_mb = h->sh.i_last_mb;
2112
2113     if( h->param.b_cabac )
2114     {
2115         x264_cabac_encode_flush( h, &h->cabac );
2116         h->out.bs.p = h->cabac.p;
2117     }
2118     else
2119     {
2120         if( i_skip > 0 )
2121             bs_write_ue( &h->out.bs, i_skip );  /* last skip run */
2122         /* rbsp_slice_trailing_bits */
2123         bs_rbsp_trailing( &h->out.bs );
2124         bs_flush( &h->out.bs );
2125     }
2126     if( x264_nal_end( h ) )
2127         return -1;
2128
2129     if( h->sh.i_last_mb == (h->i_threadslice_end * h->mb.i_mb_width - 1) )
2130     {
2131         h->stat.frame.i_misc_bits = bs_pos( &h->out.bs )
2132                                   + (h->out.i_nal*NALU_OVERHEAD * 8)
2133                                   - h->stat.frame.i_tex_bits
2134                                   - h->stat.frame.i_mv_bits;
2135         x264_fdec_filter_row( h, h->i_threadslice_end, 1 );
2136     }
2137
2138     return 0;
2139 }
2140
2141 static void x264_thread_sync_context( x264_t *dst, x264_t *src )
2142 {
2143     if( dst == src )
2144         return;
2145
2146     // reference counting
2147     for( x264_frame_t **f = src->frames.reference; *f; f++ )
2148         (*f)->i_reference_count++;
2149     for( x264_frame_t **f = dst->frames.reference; *f; f++ )
2150         x264_frame_push_unused( src, *f );
2151     src->fdec->i_reference_count++;
2152     x264_frame_push_unused( src, dst->fdec );
2153
2154     // copy everything except the per-thread pointers and the constants.
2155     memcpy( &dst->i_frame, &src->i_frame, offsetof(x264_t, mb.type) - offsetof(x264_t, i_frame) );
2156     dst->param = src->param;
2157     dst->stat = src->stat;
2158 }
2159
2160 static void x264_thread_sync_stat( x264_t *dst, x264_t *src )
2161 {
2162     if( dst == src )
2163         return;
2164     memcpy( &dst->stat.i_frame_count, &src->stat.i_frame_count, sizeof(dst->stat) - sizeof(dst->stat.frame) );
2165 }
2166
2167 static void *x264_slices_write( x264_t *h )
2168 {
2169     int i_slice_num = 0;
2170     int last_thread_mb = h->sh.i_last_mb;
2171
2172 #if HAVE_VISUALIZE
2173     if( h->param.b_visualize )
2174         if( x264_visualize_init( h ) )
2175             return (void *)-1;
2176 #endif
2177
2178     /* init stats */
2179     memset( &h->stat.frame, 0, sizeof(h->stat.frame) );
2180     h->mb.b_reencode_mb = 0;
2181     while( h->sh.i_first_mb <= last_thread_mb )
2182     {
2183         h->sh.i_last_mb = last_thread_mb;
2184         if( h->param.i_slice_max_mbs )
2185             h->sh.i_last_mb = h->sh.i_first_mb + h->param.i_slice_max_mbs - 1;
2186         else if( h->param.i_slice_count && !h->param.b_sliced_threads )
2187         {
2188             int height = h->mb.i_mb_height >> h->param.b_interlaced;
2189             int width = h->mb.i_mb_width << h->param.b_interlaced;
2190             i_slice_num++;
2191             h->sh.i_last_mb = (height * i_slice_num + h->param.i_slice_count/2) / h->param.i_slice_count * width - 1;
2192         }
2193         h->sh.i_last_mb = X264_MIN( h->sh.i_last_mb, last_thread_mb );
2194         if( x264_stack_align( x264_slice_write, h ) )
2195             return (void *)-1;
2196         h->sh.i_first_mb = h->sh.i_last_mb + 1;
2197     }
2198
2199 #if HAVE_VISUALIZE
2200     if( h->param.b_visualize )
2201     {
2202         x264_visualize_show( h );
2203         x264_visualize_close( h );
2204     }
2205 #endif
2206
2207     return (void *)0;
2208 }
2209
2210 static int x264_threaded_slices_write( x264_t *h )
2211 {
2212     /* set first/last mb and sync contexts */
2213     for( int i = 0; i < h->param.i_threads; i++ )
2214     {
2215         x264_t *t = h->thread[i];
2216         if( i )
2217         {
2218             t->param = h->param;
2219             memcpy( &t->i_frame, &h->i_frame, offsetof(x264_t, rc) - offsetof(x264_t, i_frame) );
2220         }
2221         int height = h->mb.i_mb_height >> h->param.b_interlaced;
2222         t->i_threadslice_start = ((height *  i    + h->param.i_slice_count/2) / h->param.i_threads) << h->param.b_interlaced;
2223         t->i_threadslice_end   = ((height * (i+1) + h->param.i_slice_count/2) / h->param.i_threads) << h->param.b_interlaced;
2224         t->sh.i_first_mb = t->i_threadslice_start * h->mb.i_mb_width;
2225         t->sh.i_last_mb  =   t->i_threadslice_end * h->mb.i_mb_width - 1;
2226     }
2227
2228     x264_stack_align( x264_analyse_weight_frame, h, h->mb.i_mb_height*16 + 16 );
2229
2230     x264_threads_distribute_ratecontrol( h );
2231
2232     /* dispatch */
2233     for( int i = 0; i < h->param.i_threads; i++ )
2234     {
2235         x264_threadpool_run( h->threadpool, (void*)x264_slices_write, h->thread[i] );
2236         h->thread[i]->b_thread_active = 1;
2237     }
2238     for( int i = 0; i < h->param.i_threads; i++ )
2239     {
2240         h->thread[i]->b_thread_active = 0;
2241         if( (intptr_t)x264_threadpool_wait( h->threadpool, h->thread[i] ) )
2242             return -1;
2243     }
2244
2245     /* Go back and fix up the hpel on the borders between slices. */
2246     for( int i = 1; i < h->param.i_threads; i++ )
2247     {
2248         x264_fdec_filter_row( h->thread[i], h->thread[i]->i_threadslice_start + 1, 0 );
2249         if( h->sh.b_mbaff )
2250             x264_fdec_filter_row( h->thread[i], h->thread[i]->i_threadslice_start + 2, 0 );
2251     }
2252
2253     x264_threads_merge_ratecontrol( h );
2254
2255     for( int i = 1; i < h->param.i_threads; i++ )
2256     {
2257         x264_t *t = h->thread[i];
2258         for( int j = 0; j < t->out.i_nal; j++ )
2259         {
2260             h->out.nal[h->out.i_nal] = t->out.nal[j];
2261             h->out.i_nal++;
2262             x264_nal_check_buffer( h );
2263         }
2264         /* All entries in stat.frame are ints except for ssd/ssim. */
2265         for( int j = 0; j < (offsetof(x264_t,stat.frame.i_ssd) - offsetof(x264_t,stat.frame.i_mv_bits)) / sizeof(int); j++ )
2266             ((int*)&h->stat.frame)[j] += ((int*)&t->stat.frame)[j];
2267         for( int j = 0; j < 3; j++ )
2268             h->stat.frame.i_ssd[j] += t->stat.frame.i_ssd[j];
2269         h->stat.frame.f_ssim += t->stat.frame.f_ssim;
2270     }
2271
2272     return 0;
2273 }
2274
2275 void x264_encoder_intra_refresh( x264_t *h )
2276 {
2277     h = h->thread[h->i_thread_phase];
2278     h->b_queued_intra_refresh = 1;
2279 }
2280
2281 int x264_encoder_invalidate_reference( x264_t *h, int64_t pts )
2282 {
2283     if( h->param.i_bframe )
2284     {
2285         x264_log( h, X264_LOG_ERROR, "x264_encoder_invalidate_reference is not supported with B-frames enabled\n" );
2286         return -1;
2287     }
2288     if( h->param.b_intra_refresh )
2289     {
2290         x264_log( h, X264_LOG_ERROR, "x264_encoder_invalidate_reference is not supported with intra refresh enabled\n" );
2291         return -1;
2292     }
2293     h = h->thread[h->i_thread_phase];
2294     if( pts >= h->i_last_idr_pts )
2295     {
2296         for( int i = 0; h->frames.reference[i]; i++ )
2297             if( pts <= h->frames.reference[i]->i_pts )
2298                 h->frames.reference[i]->b_corrupt = 1;
2299         if( pts <= h->fdec->i_pts )
2300             h->fdec->b_corrupt = 1;
2301     }
2302     return 0;
2303 }
2304
2305 /****************************************************************************
2306  * x264_encoder_encode:
2307  *  XXX: i_poc   : is the poc of the current given picture
2308  *       i_frame : is the number of the frame being coded
2309  *  ex:  type frame poc
2310  *       I      0   2*0
2311  *       P      1   2*3
2312  *       B      2   2*1
2313  *       B      3   2*2
2314  *       P      4   2*6
2315  *       B      5   2*4
2316  *       B      6   2*5
2317  ****************************************************************************/
2318 int     x264_encoder_encode( x264_t *h,
2319                              x264_nal_t **pp_nal, int *pi_nal,
2320                              x264_picture_t *pic_in,
2321                              x264_picture_t *pic_out )
2322 {
2323     x264_t *thread_current, *thread_prev, *thread_oldest;
2324     int i_nal_type, i_nal_ref_idc, i_global_qp;
2325     int overhead = NALU_OVERHEAD;
2326
2327     if( h->i_thread_frames > 1 )
2328     {
2329         thread_prev    = h->thread[ h->i_thread_phase ];
2330         h->i_thread_phase = (h->i_thread_phase + 1) % h->i_thread_frames;
2331         thread_current = h->thread[ h->i_thread_phase ];
2332         thread_oldest  = h->thread[ (h->i_thread_phase + 1) % h->i_thread_frames ];
2333         x264_thread_sync_context( thread_current, thread_prev );
2334         x264_thread_sync_ratecontrol( thread_current, thread_prev, thread_oldest );
2335         h = thread_current;
2336     }
2337     else
2338     {
2339         thread_current =
2340         thread_oldest  = h;
2341     }
2342 #if HAVE_MMX
2343     if( h->i_thread_frames == 1 && h->param.cpu&X264_CPU_SSE_MISALIGN )
2344         x264_cpu_mask_misalign_sse();
2345 #endif
2346
2347     // ok to call this before encoding any frames, since the initial values of fdec have b_kept_as_ref=0
2348     if( x264_reference_update( h ) )
2349         return -1;
2350     h->fdec->i_lines_completed = -1;
2351
2352     /* no data out */
2353     *pi_nal = 0;
2354     *pp_nal = NULL;
2355
2356     /* ------------------- Setup new frame from picture -------------------- */
2357     if( pic_in != NULL )
2358     {
2359         /* 1: Copy the picture to a frame and move it to a buffer */
2360         x264_frame_t *fenc = x264_frame_pop_unused( h, 0 );
2361         if( !fenc )
2362             return -1;
2363
2364         if( x264_frame_copy_picture( h, fenc, pic_in ) < 0 )
2365             return -1;
2366
2367         if( h->param.i_width != 16 * h->mb.i_mb_width ||
2368             h->param.i_height != 16 * h->mb.i_mb_height )
2369             x264_frame_expand_border_mod16( h, fenc );
2370
2371         fenc->i_frame = h->frames.i_input++;
2372
2373         if( fenc->i_frame == 0 )
2374             h->frames.i_first_pts = fenc->i_pts;
2375         if( h->frames.i_bframe_delay && fenc->i_frame == h->frames.i_bframe_delay )
2376             h->frames.i_bframe_delay_time = fenc->i_pts - h->frames.i_first_pts;
2377
2378         if( h->param.b_vfr_input && fenc->i_pts <= h->frames.i_largest_pts )
2379             x264_log( h, X264_LOG_WARNING, "non-strictly-monotonic PTS\n" );
2380
2381         h->frames.i_second_largest_pts = h->frames.i_largest_pts;
2382         h->frames.i_largest_pts = fenc->i_pts;
2383
2384         if( (fenc->i_pic_struct < PIC_STRUCT_AUTO) || (fenc->i_pic_struct > PIC_STRUCT_TRIPLE) )
2385             fenc->i_pic_struct = PIC_STRUCT_AUTO;
2386
2387         if( fenc->i_pic_struct == PIC_STRUCT_AUTO )
2388         {
2389             int b_interlaced = fenc->param ? fenc->param->b_interlaced : h->param.b_interlaced;
2390             if( b_interlaced )
2391             {
2392                 int b_tff = fenc->param ? fenc->param->b_tff : h->param.b_tff;
2393                 fenc->i_pic_struct = b_tff ? PIC_STRUCT_TOP_BOTTOM : PIC_STRUCT_BOTTOM_TOP;
2394             }
2395             else
2396                 fenc->i_pic_struct = PIC_STRUCT_PROGRESSIVE;
2397         }
2398
2399         if( h->param.rc.b_mb_tree && h->param.rc.b_stat_read )
2400         {
2401             if( x264_macroblock_tree_read( h, fenc, pic_in->prop.quant_offsets ) )
2402                 return -1;
2403         }
2404         else
2405             x264_stack_align( x264_adaptive_quant_frame, h, fenc, pic_in->prop.quant_offsets );
2406
2407         if( pic_in->prop.quant_offsets_free )
2408             pic_in->prop.quant_offsets_free( pic_in->prop.quant_offsets );
2409
2410         if( h->frames.b_have_lowres )
2411             x264_frame_init_lowres( h, fenc );
2412
2413         /* 2: Place the frame into the queue for its slice type decision */
2414         x264_lookahead_put_frame( h, fenc );
2415
2416         if( h->frames.i_input <= h->frames.i_delay + 1 - h->i_thread_frames )
2417         {
2418             /* Nothing yet to encode, waiting for filling of buffers */
2419             pic_out->i_type = X264_TYPE_AUTO;
2420             return 0;
2421         }
2422     }
2423     else
2424     {
2425         /* signal kills for lookahead thread */
2426         x264_pthread_mutex_lock( &h->lookahead->ifbuf.mutex );
2427         h->lookahead->b_exit_thread = 1;
2428         x264_pthread_cond_broadcast( &h->lookahead->ifbuf.cv_fill );
2429         x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
2430     }
2431
2432     h->i_frame++;
2433     /* 3: The picture is analyzed in the lookahead */
2434     if( !h->frames.current[0] )
2435         x264_lookahead_get_frames( h );
2436
2437     if( !h->frames.current[0] && x264_lookahead_is_empty( h ) )
2438         return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
2439
2440     /* ------------------- Get frame to be encoded ------------------------- */
2441     /* 4: get picture to encode */
2442     h->fenc = x264_frame_shift( h->frames.current );
2443     if( h->i_frame == h->i_thread_frames - 1 )
2444         h->i_reordered_pts_delay = h->fenc->i_reordered_pts;
2445     if( h->fenc->param )
2446     {
2447         x264_encoder_reconfig( h, h->fenc->param );
2448         if( h->fenc->param->param_free )
2449             h->fenc->param->param_free( h->fenc->param );
2450     }
2451
2452     if( !IS_X264_TYPE_I( h->fenc->i_type ) )
2453     {
2454         int valid_refs_left = 0;
2455         for( int i = 0; h->frames.reference[i]; i++ )
2456             if( !h->frames.reference[i]->b_corrupt )
2457                 valid_refs_left++;
2458         /* No valid reference frames left: force an IDR. */
2459         if( !valid_refs_left )
2460         {
2461             h->fenc->b_keyframe = 1;
2462             h->fenc->i_type = X264_TYPE_IDR;
2463         }
2464     }
2465
2466     if( h->fenc->b_keyframe )
2467     {
2468         h->frames.i_last_keyframe = h->fenc->i_frame;
2469         if( h->fenc->i_type == X264_TYPE_IDR )
2470         {
2471             h->i_frame_num = 0;
2472             h->frames.i_last_idr = h->fenc->i_frame;
2473         }
2474     }
2475     h->sh.i_mmco_command_count =
2476     h->sh.i_mmco_remove_from_end = 0;
2477     h->b_ref_reorder[0] =
2478     h->b_ref_reorder[1] = 0;
2479     h->fdec->i_poc =
2480     h->fenc->i_poc = 2 * ( h->fenc->i_frame - X264_MAX( h->frames.i_last_idr, 0 ) );
2481
2482     /* ------------------- Setup frame context ----------------------------- */
2483     /* 5: Init data dependent of frame type */
2484     if( h->fenc->i_type == X264_TYPE_IDR )
2485     {
2486         /* reset ref pictures */
2487         i_nal_type    = NAL_SLICE_IDR;
2488         i_nal_ref_idc = NAL_PRIORITY_HIGHEST;
2489         h->sh.i_type = SLICE_TYPE_I;
2490         x264_reference_reset( h );
2491         h->frames.i_poc_last_open_gop = -1;
2492     }
2493     else if( h->fenc->i_type == X264_TYPE_I )
2494     {
2495         i_nal_type    = NAL_SLICE;
2496         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
2497         h->sh.i_type = SLICE_TYPE_I;
2498         x264_reference_hierarchy_reset( h );
2499         if( h->param.i_open_gop )
2500             h->frames.i_poc_last_open_gop = h->fenc->b_keyframe ? h->fenc->i_poc : -1;
2501     }
2502     else if( h->fenc->i_type == X264_TYPE_P )
2503     {
2504         i_nal_type    = NAL_SLICE;
2505         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
2506         h->sh.i_type = SLICE_TYPE_P;
2507         x264_reference_hierarchy_reset( h );
2508         h->frames.i_poc_last_open_gop = -1;
2509     }
2510     else if( h->fenc->i_type == X264_TYPE_BREF )
2511     {
2512         i_nal_type    = NAL_SLICE;
2513         i_nal_ref_idc = h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT ? NAL_PRIORITY_LOW : NAL_PRIORITY_HIGH;
2514         h->sh.i_type = SLICE_TYPE_B;
2515         x264_reference_hierarchy_reset( h );
2516     }
2517     else    /* B frame */
2518     {
2519         i_nal_type    = NAL_SLICE;
2520         i_nal_ref_idc = NAL_PRIORITY_DISPOSABLE;
2521         h->sh.i_type = SLICE_TYPE_B;
2522     }
2523
2524     h->fdec->i_type = h->fenc->i_type;
2525     h->fdec->i_frame = h->fenc->i_frame;
2526     h->fenc->b_kept_as_ref =
2527     h->fdec->b_kept_as_ref = i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE && h->param.i_keyint_max > 1;
2528
2529     h->fdec->i_pts = h->fenc->i_pts;
2530     if( h->frames.i_bframe_delay )
2531     {
2532         int64_t *prev_reordered_pts = thread_current->frames.i_prev_reordered_pts;
2533         h->fdec->i_dts = h->i_frame > h->frames.i_bframe_delay
2534                        ? prev_reordered_pts[ (h->i_frame - h->frames.i_bframe_delay) % h->frames.i_bframe_delay ]
2535                        : h->fenc->i_reordered_pts - h->frames.i_bframe_delay_time;
2536         prev_reordered_pts[ h->i_frame % h->frames.i_bframe_delay ] = h->fenc->i_reordered_pts;
2537     }
2538     else
2539         h->fdec->i_dts = h->fenc->i_reordered_pts;
2540     if( h->fenc->i_type == X264_TYPE_IDR )
2541         h->i_last_idr_pts = h->fdec->i_pts;
2542
2543     /* ------------------- Init                ----------------------------- */
2544     /* build ref list 0/1 */
2545     x264_reference_build_list( h, h->fdec->i_poc );
2546
2547     /* ---------------------- Write the bitstream -------------------------- */
2548     /* Init bitstream context */
2549     if( h->param.b_sliced_threads )
2550     {
2551         for( int i = 0; i < h->param.i_threads; i++ )
2552         {
2553             bs_init( &h->thread[i]->out.bs, h->thread[i]->out.p_bitstream, h->thread[i]->out.i_bitstream );
2554             h->thread[i]->out.i_nal = 0;
2555         }
2556     }
2557     else
2558     {
2559         bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
2560         h->out.i_nal = 0;
2561     }
2562
2563     if( h->param.b_aud )
2564     {
2565         int pic_type;
2566
2567         if( h->sh.i_type == SLICE_TYPE_I )
2568             pic_type = 0;
2569         else if( h->sh.i_type == SLICE_TYPE_P )
2570             pic_type = 1;
2571         else if( h->sh.i_type == SLICE_TYPE_B )
2572             pic_type = 2;
2573         else
2574             pic_type = 7;
2575
2576         x264_nal_start( h, NAL_AUD, NAL_PRIORITY_DISPOSABLE );
2577         bs_write( &h->out.bs, 3, pic_type );
2578         bs_rbsp_trailing( &h->out.bs );
2579         if( x264_nal_end( h ) )
2580             return -1;
2581         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2582     }
2583
2584     h->i_nal_type = i_nal_type;
2585     h->i_nal_ref_idc = i_nal_ref_idc;
2586
2587     if( h->param.b_intra_refresh )
2588     {
2589         if( IS_X264_TYPE_I( h->fenc->i_type ) )
2590         {
2591             h->fdec->i_frames_since_pir = 0;
2592             h->b_queued_intra_refresh = 0;
2593             /* PIR is currently only supported with ref == 1, so any intra frame effectively refreshes
2594              * the whole frame and counts as an intra refresh. */
2595             h->fdec->f_pir_position = h->mb.i_mb_width;
2596         }
2597         else if( h->fenc->i_type == X264_TYPE_P )
2598         {
2599             int pocdiff = (h->fdec->i_poc - h->fref0[0]->i_poc)/2;
2600             float increment = X264_MAX( ((float)h->mb.i_mb_width-1) / h->param.i_keyint_max, 1 );
2601             h->fdec->f_pir_position = h->fref0[0]->f_pir_position;
2602             h->fdec->i_frames_since_pir = h->fref0[0]->i_frames_since_pir + pocdiff;
2603             if( h->fdec->i_frames_since_pir >= h->param.i_keyint_max ||
2604                 (h->b_queued_intra_refresh && h->fdec->f_pir_position + 0.5 >= h->mb.i_mb_width) )
2605             {
2606                 h->fdec->f_pir_position = 0;
2607                 h->fdec->i_frames_since_pir = 0;
2608                 h->b_queued_intra_refresh = 0;
2609                 h->fenc->b_keyframe = 1;
2610             }
2611             h->fdec->i_pir_start_col = h->fdec->f_pir_position+0.5;
2612             h->fdec->f_pir_position += increment * pocdiff;
2613             h->fdec->i_pir_end_col = h->fdec->f_pir_position+0.5;
2614             /* If our intra refresh has reached the right side of the frame, we're done. */
2615             if( h->fdec->i_pir_end_col >= h->mb.i_mb_width - 1 )
2616                 h->fdec->f_pir_position = h->mb.i_mb_width;
2617         }
2618     }
2619
2620     if( h->fenc->b_keyframe )
2621     {
2622         /* Write SPS and PPS */
2623         if( h->param.b_repeat_headers )
2624         {
2625             /* generate sequence parameters */
2626             x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
2627             x264_sps_write( &h->out.bs, h->sps );
2628             if( x264_nal_end( h ) )
2629                 return -1;
2630             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
2631
2632             /* generate picture parameters */
2633             x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
2634             x264_pps_write( &h->out.bs, h->pps );
2635             if( x264_nal_end( h ) )
2636                 return -1;
2637             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
2638         }
2639
2640         /* buffering period sei is written in x264_encoder_frame_end */
2641     }
2642
2643     /* write extra sei */
2644     for( int i = 0; i < h->fenc->extra_sei.num_payloads; i++ )
2645     {
2646         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2647         x264_sei_write( &h->out.bs, h->fenc->extra_sei.payloads[i].payload, h->fenc->extra_sei.payloads[i].payload_size,
2648                         h->fenc->extra_sei.payloads[i].payload_type );
2649         if( x264_nal_end( h ) )
2650             return -1;
2651         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2652         if( h->fenc->extra_sei.sei_free && h->fenc->extra_sei.payloads[i].payload )
2653             h->fenc->extra_sei.sei_free( h->fenc->extra_sei.payloads[i].payload );
2654     }
2655
2656     if( h->fenc->extra_sei.sei_free && h->fenc->extra_sei.payloads )
2657         h->fenc->extra_sei.sei_free( h->fenc->extra_sei.payloads );
2658
2659     if( h->fenc->b_keyframe )
2660     {
2661         if( h->param.b_repeat_headers && h->fenc->i_frame == 0 )
2662         {
2663             /* identify ourself */
2664             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2665             if( x264_sei_version_write( h, &h->out.bs ) )
2666                 return -1;
2667             if( x264_nal_end( h ) )
2668                 return -1;
2669             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2670         }
2671
2672         if( h->fenc->i_type != X264_TYPE_IDR )
2673         {
2674             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;
2675             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2676             x264_sei_recovery_point_write( h, &h->out.bs, time_to_recovery );
2677             x264_nal_end( h );
2678             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2679         }
2680     }
2681
2682     /* generate sei pic timing */
2683     if( h->sps->vui.b_pic_struct_present || h->sps->vui.b_nal_hrd_parameters_present )
2684     {
2685         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2686         x264_sei_pic_timing_write( h, &h->out.bs );
2687         if( x264_nal_end( h ) )
2688             return -1;
2689         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2690     }
2691
2692     if( h->fenc->b_keyframe && h->param.b_intra_refresh )
2693         h->i_cpb_delay_pir_offset = h->fenc->i_cpb_delay;
2694
2695     /* Init the rate control */
2696     /* FIXME: Include slice header bit cost. */
2697     x264_ratecontrol_start( h, h->fenc->i_qpplus1, overhead*8 );
2698     i_global_qp = x264_ratecontrol_qp( h );
2699
2700     pic_out->i_qpplus1 =
2701     h->fdec->i_qpplus1 = i_global_qp + 1;
2702
2703     if( h->param.rc.b_stat_read && h->sh.i_type != SLICE_TYPE_I )
2704     {
2705         x264_reference_build_list_optimal( h );
2706         x264_reference_check_reorder( h );
2707     }
2708
2709     if( h->i_ref0 )
2710         h->fdec->i_poc_l0ref0 = h->fref0[0]->i_poc;
2711
2712     if( h->sh.i_type == SLICE_TYPE_B )
2713         x264_macroblock_bipred_init( h );
2714
2715     /*------------------------- Weights -------------------------------------*/
2716     x264_weighted_pred_init( h );
2717
2718     /* ------------------------ Create slice header  ----------------------- */
2719     x264_slice_init( h, i_nal_type, i_global_qp );
2720
2721     if( i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE )
2722         h->i_frame_num++;
2723
2724     /* Write frame */
2725     h->i_threadslice_start = 0;
2726     h->i_threadslice_end = h->mb.i_mb_height;
2727     if( h->i_thread_frames > 1 )
2728     {
2729         x264_threadpool_run( h->threadpool, (void*)x264_slices_write, h );
2730         h->b_thread_active = 1;
2731     }
2732     else if( h->param.b_sliced_threads )
2733     {
2734         if( x264_threaded_slices_write( h ) )
2735             return -1;
2736     }
2737     else
2738         if( (intptr_t)x264_slices_write( h ) )
2739             return -1;
2740
2741     return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
2742 }
2743
2744 static int x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
2745                                    x264_nal_t **pp_nal, int *pi_nal,
2746                                    x264_picture_t *pic_out )
2747 {
2748     char psz_message[80];
2749
2750     if( h->b_thread_active )
2751     {
2752         h->b_thread_active = 0;
2753         if( (intptr_t)x264_threadpool_wait( h->threadpool, h ) )
2754             return -1;
2755     }
2756     if( !h->out.i_nal )
2757     {
2758         pic_out->i_type = X264_TYPE_AUTO;
2759         return 0;
2760     }
2761
2762     x264_emms();
2763     /* generate sei buffering period and insert it into place */
2764     if( h->fenc->b_keyframe && h->sps->vui.b_nal_hrd_parameters_present )
2765     {
2766         x264_hrd_fullness( h );
2767         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2768         x264_sei_buffering_period_write( h, &h->out.bs );
2769         if( x264_nal_end( h ) )
2770            return -1;
2771         /* buffering period sei must follow AUD, SPS and PPS and precede all other SEIs */
2772         int idx = 0;
2773         while( h->out.nal[idx].i_type == NAL_AUD ||
2774                h->out.nal[idx].i_type == NAL_SPS ||
2775                h->out.nal[idx].i_type == NAL_PPS )
2776             idx++;
2777         x264_nal_t nal_tmp = h->out.nal[h->out.i_nal-1];
2778         memmove( &h->out.nal[idx+1], &h->out.nal[idx], (h->out.i_nal-idx-1)*sizeof(x264_nal_t) );
2779         h->out.nal[idx] = nal_tmp;
2780     }
2781
2782     int frame_size = x264_encoder_encapsulate_nals( h, 0 );
2783
2784     /* Set output picture properties */
2785     pic_out->i_type = h->fenc->i_type;
2786
2787     pic_out->b_keyframe = h->fenc->b_keyframe;
2788
2789     pic_out->i_pts = h->fdec->i_pts;
2790     pic_out->i_dts = h->fdec->i_dts;
2791
2792     if( pic_out->i_pts < pic_out->i_dts )
2793         x264_log( h, X264_LOG_WARNING, "invalid DTS: PTS is less than DTS\n" );
2794
2795     pic_out->img.i_csp = X264_CSP_NV12;
2796 #if HIGH_BIT_DEPTH
2797     pic_out->img.i_csp |= X264_CSP_HIGH_DEPTH;
2798 #endif
2799     pic_out->img.i_plane = h->fdec->i_plane;
2800     for( int i = 0; i < 2; i++ )
2801     {
2802         pic_out->img.i_stride[i] = h->fdec->i_stride[i] * sizeof(pixel);
2803         pic_out->img.plane[i] = (uint8_t*)h->fdec->plane[i];
2804     }
2805
2806     x264_frame_push_unused( thread_current, h->fenc );
2807
2808     /* ---------------------- Update encoder state ------------------------- */
2809
2810     /* update rc */
2811     int filler = 0;
2812     if( x264_ratecontrol_end( h, frame_size * 8, &filler ) < 0 )
2813         return -1;
2814
2815     pic_out->hrd_timing = h->fenc->hrd_timing;
2816
2817     while( filler > 0 )
2818     {
2819         int f, overhead;
2820         overhead = (FILLER_OVERHEAD - h->param.b_annexb);
2821         if( h->param.i_slice_max_size && filler > h->param.i_slice_max_size )
2822         {
2823             int next_size = filler - h->param.i_slice_max_size;
2824             int overflow = X264_MAX( overhead - next_size, 0 );
2825             f = h->param.i_slice_max_size - overhead - overflow;
2826         }
2827         else
2828             f = X264_MAX( 0, filler - overhead );
2829
2830         x264_nal_start( h, NAL_FILLER, NAL_PRIORITY_DISPOSABLE );
2831         x264_filler_write( h, &h->out.bs, f );
2832         if( x264_nal_end( h ) )
2833             return -1;
2834         int total_size = x264_encoder_encapsulate_nals( h, h->out.i_nal-1 );
2835         frame_size += total_size;
2836         filler -= total_size;
2837     }
2838
2839     /* End bitstream, set output  */
2840     *pi_nal = h->out.i_nal;
2841     *pp_nal = h->out.nal;
2842
2843     h->out.i_nal = 0;
2844
2845     x264_noise_reduction_update( h );
2846
2847     /* ---------------------- Compute/Print statistics --------------------- */
2848     x264_thread_sync_stat( h, h->thread[0] );
2849
2850     /* Slice stat */
2851     h->stat.i_frame_count[h->sh.i_type]++;
2852     h->stat.i_frame_size[h->sh.i_type] += frame_size;
2853     h->stat.f_frame_qp[h->sh.i_type] += h->fdec->f_qp_avg_aq;
2854
2855     for( int i = 0; i < X264_MBTYPE_MAX; i++ )
2856         h->stat.i_mb_count[h->sh.i_type][i] += h->stat.frame.i_mb_count[i];
2857     for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
2858         h->stat.i_mb_partition[h->sh.i_type][i] += h->stat.frame.i_mb_partition[i];
2859     for( int i = 0; i < 2; i++ )
2860         h->stat.i_mb_count_8x8dct[i] += h->stat.frame.i_mb_count_8x8dct[i];
2861     for( int i = 0; i < 6; i++ )
2862         h->stat.i_mb_cbp[i] += h->stat.frame.i_mb_cbp[i];
2863     for( int i = 0; i < 4; i++ )
2864         for( int j = 0; j < 13; j++ )
2865             h->stat.i_mb_pred_mode[i][j] += h->stat.frame.i_mb_pred_mode[i][j];
2866     if( h->sh.i_type != SLICE_TYPE_I )
2867         for( int i_list = 0; i_list < 2; i_list++ )
2868             for( int i = 0; i < X264_REF_MAX*2; i++ )
2869                 h->stat.i_mb_count_ref[h->sh.i_type][i_list][i] += h->stat.frame.i_mb_count_ref[i_list][i];
2870     if( h->sh.i_type == SLICE_TYPE_P )
2871     {
2872         h->stat.i_consecutive_bframes[h->fdec->i_frame - h->fref0[0]->i_frame - 1]++;
2873         if( h->param.analyse.i_weighted_pred >= X264_WEIGHTP_SIMPLE )
2874         {
2875             h->stat.i_wpred[0] += !!h->sh.weight[0][0].weightfn;
2876             h->stat.i_wpred[1] += !!h->sh.weight[0][1].weightfn || !!h->sh.weight[0][2].weightfn;
2877         }
2878     }
2879     if( h->sh.i_type == SLICE_TYPE_B )
2880     {
2881         h->stat.i_direct_frames[ h->sh.b_direct_spatial_mv_pred ] ++;
2882         if( h->mb.b_direct_auto_write )
2883         {
2884             //FIXME somewhat arbitrary time constants
2885             if( h->stat.i_direct_score[0] + h->stat.i_direct_score[1] > h->mb.i_mb_count )
2886                 for( int i = 0; i < 2; i++ )
2887                     h->stat.i_direct_score[i] = h->stat.i_direct_score[i] * 9/10;
2888             for( int i = 0; i < 2; i++ )
2889                 h->stat.i_direct_score[i] += h->stat.frame.i_direct_score[i];
2890         }
2891     }
2892
2893     psz_message[0] = '\0';
2894     double dur = h->fenc->f_duration;
2895     h->stat.f_frame_duration[h->sh.i_type] += dur;
2896     if( h->param.analyse.b_psnr )
2897     {
2898         int64_t ssd[3] =
2899         {
2900             h->stat.frame.i_ssd[0],
2901             h->stat.frame.i_ssd[1],
2902             h->stat.frame.i_ssd[2],
2903         };
2904
2905         h->stat.f_ssd_global[h->sh.i_type]   += dur * (ssd[0] + ssd[1] + ssd[2]);
2906         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 );
2907         h->stat.f_psnr_mean_y[h->sh.i_type]  += dur * x264_psnr( ssd[0], h->param.i_width * h->param.i_height );
2908         h->stat.f_psnr_mean_u[h->sh.i_type]  += dur * x264_psnr( ssd[1], h->param.i_width * h->param.i_height / 4 );
2909         h->stat.f_psnr_mean_v[h->sh.i_type]  += dur * x264_psnr( ssd[2], h->param.i_width * h->param.i_height / 4 );
2910
2911         snprintf( psz_message, 80, " PSNR Y:%5.2f U:%5.2f V:%5.2f",
2912                   x264_psnr( ssd[0], h->param.i_width * h->param.i_height ),
2913                   x264_psnr( ssd[1], h->param.i_width * h->param.i_height / 4),
2914                   x264_psnr( ssd[2], h->param.i_width * h->param.i_height / 4) );
2915     }
2916
2917     if( h->param.analyse.b_ssim )
2918     {
2919         double ssim_y = h->stat.frame.f_ssim
2920                       / (((h->param.i_width-6)>>2) * ((h->param.i_height-6)>>2));
2921         h->stat.f_ssim_mean_y[h->sh.i_type] += ssim_y * dur;
2922         snprintf( psz_message + strlen(psz_message), 80 - strlen(psz_message),
2923                   " SSIM Y:%.5f", ssim_y );
2924     }
2925     psz_message[79] = '\0';
2926
2927     x264_log( h, X264_LOG_DEBUG,
2928                   "frame=%4d QP=%.2f NAL=%d Slice:%c Poc:%-3d I:%-4d P:%-4d SKIP:%-4d size=%d bytes%s\n",
2929               h->i_frame,
2930               h->fdec->f_qp_avg_aq,
2931               h->i_nal_ref_idc,
2932               h->sh.i_type == SLICE_TYPE_I ? 'I' : (h->sh.i_type == SLICE_TYPE_P ? 'P' : 'B' ),
2933               h->fdec->i_poc,
2934               h->stat.frame.i_mb_count_i,
2935               h->stat.frame.i_mb_count_p,
2936               h->stat.frame.i_mb_count_skip,
2937               frame_size,
2938               psz_message );
2939
2940     // keep stats all in one place
2941     x264_thread_sync_stat( h->thread[0], h );
2942     // for the use of the next frame
2943     x264_thread_sync_stat( thread_current, h );
2944
2945 #ifdef DEBUG_MB_TYPE
2946 {
2947     static const char mb_chars[] = { 'i', 'i', 'I', 'C', 'P', '8', 'S',
2948         'D', '<', 'X', 'B', 'X', '>', 'B', 'B', 'B', 'B', '8', 'S' };
2949     for( int mb_xy = 0; mb_xy < h->mb.i_mb_width * h->mb.i_mb_height; mb_xy++ )
2950     {
2951         if( h->mb.type[mb_xy] < X264_MBTYPE_MAX && h->mb.type[mb_xy] >= 0 )
2952             fprintf( stderr, "%c ", mb_chars[ h->mb.type[mb_xy] ] );
2953         else
2954             fprintf( stderr, "? " );
2955
2956         if( (mb_xy+1) % h->mb.i_mb_width == 0 )
2957             fprintf( stderr, "\n" );
2958     }
2959 }
2960 #endif
2961
2962     /* Remove duplicates, must be done near the end as breaks h->fref0 array
2963      * by freeing some of its pointers. */
2964     for( int i = 0; i < h->i_ref0; i++ )
2965         if( h->fref0[i] && h->fref0[i]->b_duplicate )
2966         {
2967             x264_frame_push_blank_unused( h, h->fref0[i] );
2968             h->fref0[i] = 0;
2969         }
2970
2971     if( h->param.psz_dump_yuv )
2972         x264_frame_dump( h );
2973     x264_emms();
2974
2975     return frame_size;
2976 }
2977
2978 static void x264_print_intra( int64_t *i_mb_count, double i_count, int b_print_pcm, char *intra )
2979 {
2980     intra += sprintf( intra, "I16..4%s: %4.1f%% %4.1f%% %4.1f%%",
2981         b_print_pcm ? "..PCM" : "",
2982         i_mb_count[I_16x16]/ i_count,
2983         i_mb_count[I_8x8]  / i_count,
2984         i_mb_count[I_4x4]  / i_count );
2985     if( b_print_pcm )
2986         sprintf( intra, " %4.1f%%", i_mb_count[I_PCM]  / i_count );
2987 }
2988
2989 /****************************************************************************
2990  * x264_encoder_close:
2991  ****************************************************************************/
2992 void    x264_encoder_close  ( x264_t *h )
2993 {
2994     int64_t i_yuv_size = 3 * h->param.i_width * h->param.i_height / 2;
2995     int64_t i_mb_count_size[2][7] = {{0}};
2996     char buf[200];
2997     int b_print_pcm = h->stat.i_mb_count[SLICE_TYPE_I][I_PCM]
2998                    || h->stat.i_mb_count[SLICE_TYPE_P][I_PCM]
2999                    || h->stat.i_mb_count[SLICE_TYPE_B][I_PCM];
3000
3001     x264_lookahead_delete( h );
3002
3003     if( h->param.i_threads > 1 )
3004         x264_threadpool_delete( h->threadpool );
3005     if( h->i_thread_frames > 1 )
3006     {
3007         for( int i = 0; i < h->i_thread_frames; i++ )
3008             if( h->thread[i]->b_thread_active )
3009             {
3010                 assert( h->thread[i]->fenc->i_reference_count == 1 );
3011                 x264_frame_delete( h->thread[i]->fenc );
3012             }
3013
3014         x264_t *thread_prev = h->thread[h->i_thread_phase];
3015         x264_thread_sync_ratecontrol( h, thread_prev, h );
3016         x264_thread_sync_ratecontrol( thread_prev, thread_prev, h );
3017         h->i_frame = thread_prev->i_frame + 1 - h->i_thread_frames;
3018     }
3019     h->i_frame++;
3020
3021     /* Slices used and PSNR */
3022     for( int i = 0; i < 3; i++ )
3023     {
3024         static const uint8_t slice_order[] = { SLICE_TYPE_I, SLICE_TYPE_P, SLICE_TYPE_B };
3025         int i_slice = slice_order[i];
3026
3027         if( h->stat.i_frame_count[i_slice] > 0 )
3028         {
3029             int i_count = h->stat.i_frame_count[i_slice];
3030             double dur =  h->stat.f_frame_duration[i_slice];
3031             if( h->param.analyse.b_psnr )
3032             {
3033                 x264_log( h, X264_LOG_INFO,
3034                           "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",
3035                           slice_type_to_char[i_slice],
3036                           i_count,
3037                           h->stat.f_frame_qp[i_slice] / i_count,
3038                           (double)h->stat.i_frame_size[i_slice] / i_count,
3039                           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,
3040                           h->stat.f_psnr_average[i_slice] / dur,
3041                           x264_psnr( h->stat.f_ssd_global[i_slice], dur * i_yuv_size ) );
3042             }
3043             else
3044             {
3045                 x264_log( h, X264_LOG_INFO,
3046                           "frame %c:%-5d Avg QP:%5.2f  size:%6.0f\n",
3047                           slice_type_to_char[i_slice],
3048                           i_count,
3049                           h->stat.f_frame_qp[i_slice] / i_count,
3050                           (double)h->stat.i_frame_size[i_slice] / i_count );
3051             }
3052         }
3053     }
3054     if( h->param.i_bframe && h->stat.i_frame_count[SLICE_TYPE_P] )
3055     {
3056         char *p = buf;
3057         int den = 0;
3058         // weight by number of frames (including the P-frame) that are in a sequence of N B-frames
3059         for( int i = 0; i <= h->param.i_bframe; i++ )
3060             den += (i+1) * h->stat.i_consecutive_bframes[i];
3061         for( int i = 0; i <= h->param.i_bframe; i++ )
3062             p += sprintf( p, " %4.1f%%", 100. * (i+1) * h->stat.i_consecutive_bframes[i] / den );
3063         x264_log( h, X264_LOG_INFO, "consecutive B-frames:%s\n", buf );
3064     }
3065
3066     for( int i_type = 0; i_type < 2; i_type++ )
3067         for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
3068         {
3069             if( i == D_DIRECT_8x8 ) continue; /* direct is counted as its own type */
3070             i_mb_count_size[i_type][x264_mb_partition_pixel_table[i]] += h->stat.i_mb_partition[i_type][i];
3071         }
3072
3073     /* MB types used */
3074     if( h->stat.i_frame_count[SLICE_TYPE_I] > 0 )
3075     {
3076         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_I];
3077         double i_count = h->stat.i_frame_count[SLICE_TYPE_I] * h->mb.i_mb_count / 100.0;
3078         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
3079         x264_log( h, X264_LOG_INFO, "mb I  %s\n", buf );
3080     }
3081     if( h->stat.i_frame_count[SLICE_TYPE_P] > 0 )
3082     {
3083         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_P];
3084         double i_count = h->stat.i_frame_count[SLICE_TYPE_P] * h->mb.i_mb_count / 100.0;
3085         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_P];
3086         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
3087         x264_log( h, X264_LOG_INFO,
3088                   "mb P  %s  P16..4: %4.1f%% %4.1f%% %4.1f%% %4.1f%% %4.1f%%    skip:%4.1f%%\n",
3089                   buf,
3090                   i_mb_size[PIXEL_16x16] / (i_count*4),
3091                   (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
3092                   i_mb_size[PIXEL_8x8] / (i_count*4),
3093                   (i_mb_size[PIXEL_8x4] + i_mb_size[PIXEL_4x8]) / (i_count*4),
3094                   i_mb_size[PIXEL_4x4] / (i_count*4),
3095                   i_mb_count[P_SKIP] / i_count );
3096     }
3097     if( h->stat.i_frame_count[SLICE_TYPE_B] > 0 )
3098     {
3099         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_B];
3100         double i_count = h->stat.i_frame_count[SLICE_TYPE_B] * h->mb.i_mb_count / 100.0;
3101         double i_mb_list_count;
3102         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_B];
3103         int64_t list_count[3] = {0}; /* 0 == L0, 1 == L1, 2 == BI */
3104         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
3105         for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
3106             for( int j = 0; j < 2; j++ )
3107             {
3108                 int l0 = x264_mb_type_list_table[i][0][j];
3109                 int l1 = x264_mb_type_list_table[i][1][j];
3110                 if( l0 || l1 )
3111                     list_count[l1+l0*l1] += h->stat.i_mb_count[SLICE_TYPE_B][i] * 2;
3112             }
3113         list_count[0] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L0_8x8];
3114         list_count[1] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L1_8x8];
3115         list_count[2] += h->stat.i_mb_partition[SLICE_TYPE_B][D_BI_8x8];
3116         i_mb_count[B_DIRECT] += (h->stat.i_mb_partition[SLICE_TYPE_B][D_DIRECT_8x8]+2)/4;
3117         i_mb_list_count = (list_count[0] + list_count[1] + list_count[2]) / 100.0;
3118         x264_log( h, X264_LOG_INFO,
3119                   "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",
3120                   buf,
3121                   i_mb_size[PIXEL_16x16] / (i_count*4),
3122                   (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
3123                   i_mb_size[PIXEL_8x8] / (i_count*4),
3124                   i_mb_count[B_DIRECT] / i_count,
3125                   i_mb_count[B_SKIP]   / i_count,
3126                   list_count[0] / i_mb_list_count,
3127                   list_count[1] / i_mb_list_count,
3128                   list_count[2] / i_mb_list_count );
3129     }
3130
3131     x264_ratecontrol_summary( h );
3132
3133     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 )
3134     {
3135 #define SUM3(p) (p[SLICE_TYPE_I] + p[SLICE_TYPE_P] + p[SLICE_TYPE_B])
3136 #define SUM3b(p,o) (p[SLICE_TYPE_I][o] + p[SLICE_TYPE_P][o] + p[SLICE_TYPE_B][o])
3137         int64_t i_i8x8 = SUM3b( h->stat.i_mb_count, I_8x8 );
3138         int64_t i_intra = i_i8x8 + SUM3b( h->stat.i_mb_count, I_4x4 )
3139                                  + SUM3b( h->stat.i_mb_count, I_16x16 );
3140         int64_t i_all_intra = i_intra + SUM3b( h->stat.i_mb_count, I_PCM);
3141         const int i_count = h->stat.i_frame_count[SLICE_TYPE_I] +
3142                             h->stat.i_frame_count[SLICE_TYPE_P] +
3143                             h->stat.i_frame_count[SLICE_TYPE_B];
3144         const double duration = h->stat.f_frame_duration[SLICE_TYPE_I] +
3145                                 h->stat.f_frame_duration[SLICE_TYPE_P] +
3146                                 h->stat.f_frame_duration[SLICE_TYPE_B];
3147         int64_t i_mb_count = (int64_t)i_count * h->mb.i_mb_count;
3148         float f_bitrate = SUM3(h->stat.i_frame_size) / duration / 125;
3149
3150         if( h->pps->b_transform_8x8_mode )
3151         {
3152             buf[0] = 0;
3153             if( h->stat.i_mb_count_8x8dct[0] )
3154                 sprintf( buf, " inter:%.1f%%", 100. * h->stat.i_mb_count_8x8dct[1] / h->stat.i_mb_count_8x8dct[0] );
3155             x264_log( h, X264_LOG_INFO, "8x8 transform intra:%.1f%%%s\n", 100. * i_i8x8 / i_intra, buf );
3156         }
3157
3158         if( (h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO ||
3159             (h->stat.i_direct_frames[0] && h->stat.i_direct_frames[1]))
3160             && h->stat.i_frame_count[SLICE_TYPE_B] )
3161         {
3162             x264_log( h, X264_LOG_INFO, "direct mvs  spatial:%.1f%% temporal:%.1f%%\n",
3163                       h->stat.i_direct_frames[1] * 100. / h->stat.i_frame_count[SLICE_TYPE_B],
3164                       h->stat.i_direct_frames[0] * 100. / h->stat.i_frame_count[SLICE_TYPE_B] );
3165         }
3166
3167         buf[0] = 0;
3168         if( i_mb_count != i_all_intra )
3169             sprintf( buf, " inter: %.1f%% %.1f%% %.1f%%",
3170                      h->stat.i_mb_cbp[1] * 100.0 / ((i_mb_count - i_all_intra)*4),
3171                      h->stat.i_mb_cbp[3] * 100.0 / ((i_mb_count - i_all_intra)  ),
3172                      h->stat.i_mb_cbp[5] * 100.0 / ((i_mb_count - i_all_intra)) );
3173         x264_log( h, X264_LOG_INFO, "coded y,uvDC,uvAC intra: %.1f%% %.1f%% %.1f%%%s\n",
3174                   h->stat.i_mb_cbp[0] * 100.0 / (i_all_intra*4),
3175                   h->stat.i_mb_cbp[2] * 100.0 / (i_all_intra  ),
3176                   h->stat.i_mb_cbp[4] * 100.0 / (i_all_intra  ), buf );
3177
3178         int64_t fixed_pred_modes[4][9] = {{0}};
3179         int64_t sum_pred_modes[4] = {0};
3180         for( int i = 0; i <= I_PRED_16x16_DC_128; i++ )
3181         {
3182             fixed_pred_modes[0][x264_mb_pred_mode16x16_fix[i]] += h->stat.i_mb_pred_mode[0][i];
3183             sum_pred_modes[0] += h->stat.i_mb_pred_mode[0][i];
3184         }
3185         if( sum_pred_modes[0] )
3186             x264_log( h, X264_LOG_INFO, "i16 v,h,dc,p: %2.0f%% %2.0f%% %2.0f%% %2.0f%%\n",
3187                       fixed_pred_modes[0][0] * 100.0 / sum_pred_modes[0],
3188                       fixed_pred_modes[0][1] * 100.0 / sum_pred_modes[0],
3189                       fixed_pred_modes[0][2] * 100.0 / sum_pred_modes[0],
3190                       fixed_pred_modes[0][3] * 100.0 / sum_pred_modes[0] );
3191         for( int i = 1; i <= 2; i++ )
3192         {
3193             for( int j = 0; j <= I_PRED_8x8_DC_128; j++ )
3194             {
3195                 fixed_pred_modes[i][x264_mb_pred_mode4x4_fix(j)] += h->stat.i_mb_pred_mode[i][j];
3196                 sum_pred_modes[i] += h->stat.i_mb_pred_mode[i][j];
3197             }
3198             if( sum_pred_modes[i] )
3199                 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,
3200                           fixed_pred_modes[i][0] * 100.0 / sum_pred_modes[i],
3201                           fixed_pred_modes[i][1] * 100.0 / sum_pred_modes[i],
3202                           fixed_pred_modes[i][2] * 100.0 / sum_pred_modes[i],
3203                           fixed_pred_modes[i][3] * 100.0 / sum_pred_modes[i],
3204                           fixed_pred_modes[i][4] * 100.0 / sum_pred_modes[i],
3205                           fixed_pred_modes[i][5] * 100.0 / sum_pred_modes[i],
3206                           fixed_pred_modes[i][6] * 100.0 / sum_pred_modes[i],
3207                           fixed_pred_modes[i][7] * 100.0 / sum_pred_modes[i],
3208                           fixed_pred_modes[i][8] * 100.0 / sum_pred_modes[i] );
3209         }
3210         for( int i = 0; i <= I_PRED_CHROMA_DC_128; i++ )
3211         {
3212             fixed_pred_modes[3][x264_mb_pred_mode8x8c_fix[i]] += h->stat.i_mb_pred_mode[3][i];
3213             sum_pred_modes[3] += h->stat.i_mb_pred_mode[3][i];
3214         }
3215         if( sum_pred_modes[3] )
3216             x264_log( h, X264_LOG_INFO, "i8c dc,h,v,p: %2.0f%% %2.0f%% %2.0f%% %2.0f%%\n",
3217                       fixed_pred_modes[3][0] * 100.0 / sum_pred_modes[3],
3218                       fixed_pred_modes[3][1] * 100.0 / sum_pred_modes[3],
3219                       fixed_pred_modes[3][2] * 100.0 / sum_pred_modes[3],
3220                       fixed_pred_modes[3][3] * 100.0 / sum_pred_modes[3] );
3221
3222         if( h->param.analyse.i_weighted_pred >= X264_WEIGHTP_SIMPLE && h->stat.i_frame_count[SLICE_TYPE_P] > 0 )
3223             x264_log( h, X264_LOG_INFO, "Weighted P-Frames: Y:%.1f%% UV:%.1f%%\n",
3224                       h->stat.i_wpred[0] * 100.0 / h->stat.i_frame_count[SLICE_TYPE_P],
3225                       h->stat.i_wpred[1] * 100.0 / h->stat.i_frame_count[SLICE_TYPE_P] );
3226
3227         for( int i_list = 0; i_list < 2; i_list++ )
3228             for( int i_slice = 0; i_slice < 2; i_slice++ )
3229             {
3230                 char *p = buf;
3231                 int64_t i_den = 0;
3232                 int i_max = 0;
3233                 for( int i = 0; i < X264_REF_MAX*2; i++ )
3234                     if( h->stat.i_mb_count_ref[i_slice][i_list][i] )
3235                     {
3236                         i_den += h->stat.i_mb_count_ref[i_slice][i_list][i];
3237                         i_max = i;
3238                     }
3239                 if( i_max == 0 )
3240                     continue;
3241                 for( int i = 0; i <= i_max; i++ )
3242                     p += sprintf( p, " %4.1f%%", 100. * h->stat.i_mb_count_ref[i_slice][i_list][i] / i_den );
3243                 x264_log( h, X264_LOG_INFO, "ref %c L%d:%s\n", "PB"[i_slice], i_list, buf );
3244             }
3245
3246         if( h->param.analyse.b_ssim )
3247         {
3248             float ssim = SUM3( h->stat.f_ssim_mean_y ) / duration;
3249             x264_log( h, X264_LOG_INFO, "SSIM Mean Y:%.7f (%6.3fdb)\n", ssim, x264_ssim( ssim ) );
3250         }
3251         if( h->param.analyse.b_psnr )
3252         {
3253             x264_log( h, X264_LOG_INFO,
3254                       "PSNR Mean Y:%6.3f U:%6.3f V:%6.3f Avg:%6.3f Global:%6.3f kb/s:%.2f\n",
3255                       SUM3( h->stat.f_psnr_mean_y ) / duration,
3256                       SUM3( h->stat.f_psnr_mean_u ) / duration,
3257                       SUM3( h->stat.f_psnr_mean_v ) / duration,
3258                       SUM3( h->stat.f_psnr_average ) / duration,
3259                       x264_psnr( SUM3( h->stat.f_ssd_global ), duration * i_yuv_size ),
3260                       f_bitrate );
3261         }
3262         else
3263             x264_log( h, X264_LOG_INFO, "kb/s:%.2f\n", f_bitrate );
3264     }
3265
3266     /* rc */
3267     x264_ratecontrol_delete( h );
3268
3269     /* param */
3270     if( h->param.rc.psz_stat_out )
3271         free( h->param.rc.psz_stat_out );
3272     if( h->param.rc.psz_stat_in )
3273         free( h->param.rc.psz_stat_in );
3274
3275     x264_cqm_delete( h );
3276     x264_free( h->nal_buffer );
3277     x264_analyse_free_costs( h );
3278
3279     if( h->i_thread_frames > 1)
3280         h = h->thread[h->i_thread_phase];
3281
3282     /* frames */
3283     x264_frame_delete_list( h->frames.unused[0] );
3284     x264_frame_delete_list( h->frames.unused[1] );
3285     x264_frame_delete_list( h->frames.current );
3286     x264_frame_delete_list( h->frames.blank_unused );
3287
3288     h = h->thread[0];
3289
3290     for( int i = h->param.i_threads - 1; i >= 0; i-- )
3291     {
3292         x264_frame_t **frame;
3293
3294         if( !h->param.b_sliced_threads || i == 0 )
3295         {
3296             for( frame = h->thread[i]->frames.reference; *frame; frame++ )
3297             {
3298                 assert( (*frame)->i_reference_count > 0 );
3299                 (*frame)->i_reference_count--;
3300                 if( (*frame)->i_reference_count == 0 )
3301                     x264_frame_delete( *frame );
3302             }
3303             frame = &h->thread[i]->fdec;
3304             assert( (*frame)->i_reference_count > 0 );
3305             (*frame)->i_reference_count--;
3306             if( (*frame)->i_reference_count == 0 )
3307                 x264_frame_delete( *frame );
3308             x264_macroblock_cache_free( h->thread[i] );
3309         }
3310         x264_macroblock_thread_free( h->thread[i], 0 );
3311         x264_free( h->thread[i]->out.p_bitstream );
3312         x264_free( h->thread[i]->out.nal);
3313         x264_free( h->thread[i] );
3314     }
3315 }
3316
3317 int x264_encoder_delayed_frames( x264_t *h )
3318 {
3319     int delayed_frames = 0;
3320     if( h->i_thread_frames > 1 )
3321     {
3322         for( int i = 0; i < h->i_thread_frames; i++ )
3323             delayed_frames += h->thread[i]->b_thread_active;
3324         h = h->thread[h->i_thread_phase];
3325     }
3326     for( int i = 0; h->frames.current[i]; i++ )
3327         delayed_frames++;
3328     x264_pthread_mutex_lock( &h->lookahead->ofbuf.mutex );
3329     x264_pthread_mutex_lock( &h->lookahead->ifbuf.mutex );
3330     x264_pthread_mutex_lock( &h->lookahead->next.mutex );
3331     delayed_frames += h->lookahead->ifbuf.i_size + h->lookahead->next.i_size + h->lookahead->ofbuf.i_size;
3332     x264_pthread_mutex_unlock( &h->lookahead->next.mutex );
3333     x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
3334     x264_pthread_mutex_unlock( &h->lookahead->ofbuf.mutex );
3335     return delayed_frames;
3336 }
3337
3338 int x264_encoder_maximum_delayed_frames( x264_t *h )
3339 {
3340     return h->frames.i_delay;
3341 }