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 at all in CAVLC lossless */
 525         if( !h->param.b_cabac )
 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     if( h->param.analyse.i_weighted_pred == X264_WEIGHTP_BLIND && BIT_DEPTH > 8 )
 776         h->param.analyse.i_weighted_pred = X264_WEIGHTP_NONE;
 777
 778     if( h->i_thread_frames > 1 )
 779     {
 780         int r = h->param.analyse.i_mv_range_thread;
 781         int r2;
 782         if( r <= 0 )
 783         {
 784             // half of the available space is reserved and divided evenly among the threads,
 785             // the rest is allocated to whichever thread is far enough ahead to use it.
 786             // reserving more space increases quality for some videos, but costs more time
 787             // in thread synchronization.
 788             int max_range = (h->param.i_height + X264_THREAD_HEIGHT) / h->i_thread_frames - X264_THREAD_HEIGHT;
 789             r = max_range / 2;
 790         }
 791         r = X264_MAX( r, h->param.analyse.i_me_range );
 792         r = X264_MIN( r, h->param.analyse.i_mv_range );
 793         // round up to use the whole mb row
 794         r2 = (r & ~15) + ((-X264_THREAD_HEIGHT) & 15);
 795         if( r2 < r )
 796             r2 += 16;
 797         x264_log( h, X264_LOG_DEBUG, "using mv_range_thread = %d\n", r2 );
 798         h->param.analyse.i_mv_range_thread = r2;
 799     }
 800
 801     if( h->param.rc.f_qblur < 0 )
 802         h->param.rc.f_qblur = 0;
 803     if( h->param.rc.f_complexity_blur < 0 )
 804         h->param.rc.f_complexity_blur = 0;
 805
 806     h->param.i_sps_id &= 31;
 807
 808     if( h->param.i_log_level < X264_LOG_INFO )
 809     {
 810         h->param.analyse.b_psnr = 0;
 811         h->param.analyse.b_ssim = 0;
 812     }
 813
 814     if( h->param.b_interlaced )
 815         h->param.b_pic_struct = 1;
 816
 817     if( h->param.i_nal_hrd && !h->param.rc.i_vbv_buffer_size )
 818     {
 819         x264_log( h, X264_LOG_WARNING, "NAL HRD parameters require VBV parameters\n" );
 820         h->param.i_nal_hrd = X264_NAL_HRD_NONE;
 821     }
 822
 823     if( h->param.i_nal_hrd == X264_NAL_HRD_CBR &&
 824        (h->param.rc.i_bitrate != h->param.rc.i_vbv_max_bitrate || !h->param.rc.i_vbv_max_bitrate) )
 825     {
 826         x264_log( h, X264_LOG_WARNING, "CBR HRD requires constant bitrate\n" );
 827         h->param.i_nal_hrd = X264_NAL_HRD_VBR;
 828     }
 829
 830     /* ensure the booleans are 0 or 1 so they can be used in math */
 831 #define BOOLIFY(x) h->param.x = !!h->param.x
 832     BOOLIFY( b_cabac );
 833     BOOLIFY( b_constrained_intra );
 834     BOOLIFY( b_deblocking_filter );
 835     BOOLIFY( b_deterministic );
 836     BOOLIFY( b_sliced_threads );
 837     BOOLIFY( b_interlaced );
 838     BOOLIFY( b_intra_refresh );
 839     BOOLIFY( b_visualize );
 840     BOOLIFY( b_aud );
 841     BOOLIFY( b_repeat_headers );
 842     BOOLIFY( b_annexb );
 843     BOOLIFY( b_vfr_input );
 844     BOOLIFY( b_pic_struct );
 845     BOOLIFY( b_fake_interlaced );
 846     BOOLIFY( analyse.b_transform_8x8 );
 847     BOOLIFY( analyse.b_weighted_bipred );
 848     BOOLIFY( analyse.b_chroma_me );
 849     BOOLIFY( analyse.b_mixed_references );
 850     BOOLIFY( analyse.b_fast_pskip );
 851     BOOLIFY( analyse.b_dct_decimate );
 852     BOOLIFY( analyse.b_psy );
 853     BOOLIFY( analyse.b_psnr );
 854     BOOLIFY( analyse.b_ssim );
 855     BOOLIFY( rc.b_stat_write );
 856     BOOLIFY( rc.b_stat_read );
 857     BOOLIFY( rc.b_mb_tree );
 858 #undef BOOLIFY
 859
 860     return 0;
 861 }
 862
 863 static void mbcmp_init( x264_t *h )
 864 {
 865     int satd = !h->mb.b_lossless && h->param.analyse.i_subpel_refine > 1;
 866     memcpy( h->pixf.mbcmp, satd ? h->pixf.satd : h->pixf.sad_aligned, sizeof(h->pixf.mbcmp) );
 867     memcpy( h->pixf.mbcmp_unaligned, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.mbcmp_unaligned) );
 868     h->pixf.intra_mbcmp_x3_16x16 = satd ? h->pixf.intra_satd_x3_16x16 : h->pixf.intra_sad_x3_16x16;
 869     h->pixf.intra_mbcmp_x3_8x8c = satd ? h->pixf.intra_satd_x3_8x8c : h->pixf.intra_sad_x3_8x8c;
 870     h->pixf.intra_mbcmp_x3_8x8 = satd ? h->pixf.intra_sa8d_x3_8x8 : h->pixf.intra_sad_x3_8x8;
 871     h->pixf.intra_mbcmp_x3_4x4 = satd ? h->pixf.intra_satd_x3_4x4 : h->pixf.intra_sad_x3_4x4;
 872     satd &= h->param.analyse.i_me_method == X264_ME_TESA;
 873     memcpy( h->pixf.fpelcmp, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.fpelcmp) );
 874     memcpy( h->pixf.fpelcmp_x3, satd ? h->pixf.satd_x3 : h->pixf.sad_x3, sizeof(h->pixf.fpelcmp_x3) );
 875     memcpy( h->pixf.fpelcmp_x4, satd ? h->pixf.satd_x4 : h->pixf.sad_x4, sizeof(h->pixf.fpelcmp_x4) );
 876 }
 877
 878 static void x264_set_aspect_ratio( x264_t *h, x264_param_t *param, int initial )
 879 {
 880     /* VUI */
 881     if( param->vui.i_sar_width > 0 && param->vui.i_sar_height > 0 )
 882     {
 883         uint32_t i_w = param->vui.i_sar_width;
 884         uint32_t i_h = param->vui.i_sar_height;
 885         uint32_t old_w = h->param.vui.i_sar_width;
 886         uint32_t old_h = h->param.vui.i_sar_height;
 887
 888         x264_reduce_fraction( &i_w, &i_h );
 889
 890         while( i_w > 65535 || i_h > 65535 )
 891         {
 892             i_w /= 2;
 893             i_h /= 2;
 894         }
 895
 896         x264_reduce_fraction( &i_w, &i_h );
 897
 898         if( i_w != old_w || i_h != old_h || initial )
 899         {
 900             h->param.vui.i_sar_width = 0;
 901             h->param.vui.i_sar_height = 0;
 902             if( i_w == 0 || i_h == 0 )
 903                 x264_log( h, X264_LOG_WARNING, "cannot create valid sample aspect ratio\n" );
 904             else
 905             {
 906                 x264_log( h, initial?X264_LOG_INFO:X264_LOG_DEBUG, "using SAR=%d/%d\n", i_w, i_h );
 907                 h->param.vui.i_sar_width = i_w;
 908                 h->param.vui.i_sar_height = i_h;
 909             }
 910         }
 911     }
 912 }
 913
 914 /****************************************************************************
 915  * x264_encoder_open:
 916  ****************************************************************************/
 917 x264_t *x264_encoder_open( x264_param_t *param )
 918 {
 919     x264_t *h;
 920     char buf[1000], *p;
 921     int qp, i_slicetype_length;
 922
 923     CHECKED_MALLOCZERO( h, sizeof(x264_t) );
 924
 925     /* Create a copy of param */
 926     memcpy( &h->param, param, sizeof(x264_param_t) );
 927
 928     if( param->param_free )
 929         param->param_free( param );
 930
 931     if( x264_validate_parameters( h ) < 0 )
 932         goto fail;
 933
 934     if( h->param.psz_cqm_file )
 935         if( x264_cqm_parse_file( h, h->param.psz_cqm_file ) < 0 )
 936             goto fail;
 937
 938     if( h->param.rc.psz_stat_out )
 939         h->param.rc.psz_stat_out = strdup( h->param.rc.psz_stat_out );
 940     if( h->param.rc.psz_stat_in )
 941         h->param.rc.psz_stat_in = strdup( h->param.rc.psz_stat_in );
 942
 943     x264_set_aspect_ratio( h, &h->param, 1 );
 944
 945     x264_reduce_fraction( &h->param.i_fps_num, &h->param.i_fps_den );
 946     x264_reduce_fraction( &h->param.i_timebase_num, &h->param.i_timebase_den );
 947
 948     /* Init x264_t */
 949     h->i_frame = -1;
 950     h->i_frame_num = 0;
 951     h->i_idr_pic_id = 0;
 952
 953     if( (uint64_t)h->param.i_timebase_den * 2 > UINT32_MAX )
 954     {
 955         x264_log( h, X264_LOG_ERROR, "Effective timebase denominator %u exceeds H.264 maximum\n", h->param.i_timebase_den );
 956         goto fail;
 957     }
 958
 959     h->sps = &h->sps_array[0];
 960     x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
 961
 962     h->pps = &h->pps_array[0];
 963     x264_pps_init( h->pps, h->param.i_sps_id, &h->param, h->sps );
 964
 965     x264_validate_levels( h, 1 );
 966
 967     h->chroma_qp_table = i_chroma_qp_table + 12 + h->pps->i_chroma_qp_index_offset;
 968
 969     if( x264_cqm_init( h ) < 0 )
 970         goto fail;
 971
 972     h->mb.i_mb_width = h->sps->i_mb_width;
 973     h->mb.i_mb_height = h->sps->i_mb_height;
 974     h->mb.i_mb_count = h->mb.i_mb_width * h->mb.i_mb_height;
 975
 976     /* Init frames. */
 977     if( h->param.i_bframe_adaptive == X264_B_ADAPT_TRELLIS && !h->param.rc.b_stat_read )
 978         h->frames.i_delay = X264_MAX(h->param.i_bframe,3)*4;
 979     else
 980         h->frames.i_delay = h->param.i_bframe;
 981     if( h->param.rc.b_mb_tree || h->param.rc.i_vbv_buffer_size )
 982         h->frames.i_delay = X264_MAX( h->frames.i_delay, h->param.rc.i_lookahead );
 983     i_slicetype_length = h->frames.i_delay;
 984     h->frames.i_delay += h->i_thread_frames - 1;
 985     h->frames.i_delay += h->param.i_sync_lookahead;
 986     h->frames.i_delay += h->param.b_vfr_input && (h->param.rc.i_rc_method == X264_RC_ABR || h->param.rc.b_stat_write
 987                                                  || h->param.rc.i_vbv_buffer_size);
 988     h->frames.i_bframe_delay = h->param.i_bframe ? (h->param.i_bframe_pyramid ? 2 : 1) : 0;
 989
 990     h->frames.i_max_ref0 = h->param.i_frame_reference;
 991     h->frames.i_max_ref1 = X264_MIN( h->sps->vui.i_num_reorder_frames, h->param.i_frame_reference );
 992     h->frames.i_max_dpb  = h->sps->vui.i_max_dec_frame_buffering;
 993     h->frames.b_have_lowres = !h->param.rc.b_stat_read
 994         && ( h->param.rc.i_rc_method == X264_RC_ABR
 995           || h->param.rc.i_rc_method == X264_RC_CRF
 996           || h->param.i_bframe_adaptive
 997           || h->param.i_scenecut_threshold
 998           || h->param.rc.b_mb_tree
 999           || h->param.analyse.i_weighted_pred == X264_WEIGHTP_SMART );
1000     h->frames.b_have_lowres |= h->param.rc.b_stat_read && h->param.rc.i_vbv_buffer_size > 0;
1001     h->frames.b_have_sub8x8_esa = !!(h->param.analyse.inter & X264_ANALYSE_PSUB8x8);
1002
1003     h->frames.i_last_idr =
1004     h->frames.i_last_keyframe = - h->param.i_keyint_max;
1005     h->frames.i_input    = 0;
1006     h->frames.i_largest_pts = h->frames.i_second_largest_pts = -1;
1007     h->frames.i_poc_last_open_gop = -1;
1008
1009     CHECKED_MALLOCZERO( h->frames.unused[0], (h->frames.i_delay + 3) * sizeof(x264_frame_t *) );
1010     /* Allocate room for max refs plus a few extra just in case. */
1011     CHECKED_MALLOCZERO( h->frames.unused[1], (h->i_thread_frames + X264_REF_MAX + 4) * sizeof(x264_frame_t *) );
1012     CHECKED_MALLOCZERO( h->frames.current, (h->param.i_sync_lookahead + h->param.i_bframe
1013                         + h->i_thread_frames + 3) * sizeof(x264_frame_t *) );
1014     if( h->param.analyse.i_weighted_pred > 0 )
1015         CHECKED_MALLOCZERO( h->frames.blank_unused, h->i_thread_frames * 4 * sizeof(x264_frame_t *) );
1016     h->i_ref0 = 0;
1017     h->i_ref1 = 0;
1018     h->i_cpb_delay = h->i_coded_fields = h->i_disp_fields = 0;
1019     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);
1020     h->i_disp_fields_last_frame = -1;
1021     x264_rdo_init();
1022
1023     /* init CPU functions */
1024     x264_predict_16x16_init( h->param.cpu, h->predict_16x16 );
1025     x264_predict_8x8c_init( h->param.cpu, h->predict_8x8c );
1026     x264_predict_8x8_init( h->param.cpu, h->predict_8x8, &h->predict_8x8_filter );
1027     x264_predict_4x4_init( h->param.cpu, h->predict_4x4 );
1028     if( !h->param.b_cabac )
1029         x264_init_vlc_tables();
1030     x264_pixel_init( h->param.cpu, &h->pixf );
1031     x264_dct_init( h->param.cpu, &h->dctf );
1032     x264_zigzag_init( h->param.cpu, &h->zigzagf, h->param.b_interlaced );
1033     x264_mc_init( h->param.cpu, &h->mc );
1034     x264_quant_init( h, h->param.cpu, &h->quantf );
1035     x264_deblock_init( h->param.cpu, &h->loopf );
1036     x264_bitstream_init( h->param.cpu, &h->bsf );
1037     x264_dct_init_weights();
1038
1039     mbcmp_init( h );
1040
1041     p = buf + sprintf( buf, "using cpu capabilities:" );
1042     for( int i = 0; x264_cpu_names[i].flags; i++ )
1043     {
1044         if( !strcmp(x264_cpu_names[i].name, "SSE2")
1045             && h->param.cpu & (X264_CPU_SSE2_IS_FAST|X264_CPU_SSE2_IS_SLOW) )
1046             continue;
1047         if( !strcmp(x264_cpu_names[i].name, "SSE3")
1048             && (h->param.cpu & X264_CPU_SSSE3 || !(h->param.cpu & X264_CPU_CACHELINE_64)) )
1049             continue;
1050         if( !strcmp(x264_cpu_names[i].name, "SSE4.1")
1051             && (h->param.cpu & X264_CPU_SSE42) )
1052             continue;
1053         if( (h->param.cpu & x264_cpu_names[i].flags) == x264_cpu_names[i].flags
1054             && (!i || x264_cpu_names[i].flags != x264_cpu_names[i-1].flags) )
1055             p += sprintf( p, " %s", x264_cpu_names[i].name );
1056     }
1057     if( !h->param.cpu )
1058         p += sprintf( p, " none!" );
1059     x264_log( h, X264_LOG_INFO, "%s\n", buf );
1060
1061     for( qp = h->param.rc.i_qp_min; qp <= h->param.rc.i_qp_max; qp++ )
1062         if( x264_analyse_init_costs( h, qp ) )
1063             goto fail;
1064     if( x264_analyse_init_costs( h, X264_LOOKAHEAD_QP ) )
1065         goto fail;
1066
1067     static const uint16_t cost_mv_correct[7] = { 24, 47, 95, 189, 379, 757, 1515 };
1068     /* Checks for known miscompilation issues. */
1069     if( h->cost_mv[x264_lambda_tab[X264_LOOKAHEAD_QP]][2013] != cost_mv_correct[BIT_DEPTH-8] )
1070     {
1071         x264_log( h, X264_LOG_ERROR, "MV cost test failed: x264 has been miscompiled!\n" );
1072         goto fail;
1073     }
1074
1075     /* Must be volatile or else GCC will optimize it out. */
1076     volatile int temp = 392;
1077     if( x264_clz( temp ) != 23 )
1078     {
1079         x264_log( h, X264_LOG_ERROR, "CLZ test failed: x264 has been miscompiled!\n" );
1080 #if ARCH_X86 || ARCH_X86_64
1081         x264_log( h, X264_LOG_ERROR, "Are you attempting to run an SSE4a-targeted build on a CPU that\n" );
1082         x264_log( h, X264_LOG_ERROR, "doesn't support it?\n" );
1083 #endif
1084         goto fail;
1085     }
1086
1087     h->out.i_nal = 0;
1088     h->out.i_bitstream = X264_MAX( 1000000, h->param.i_width * h->param.i_height * 4
1089         * ( h->param.rc.i_rc_method == X264_RC_ABR ? pow( 0.95, h->param.rc.i_qp_min )
1090           : pow( 0.95, h->param.rc.i_qp_constant ) * X264_MAX( 1, h->param.rc.f_ip_factor )));
1091
1092     CHECKED_MALLOC( h->nal_buffer, h->out.i_bitstream * 3/2 + 4 );
1093     h->nal_buffer_size = h->out.i_bitstream * 3/2 + 4;
1094
1095     if( h->param.i_threads > 1 &&
1096         x264_threadpool_init( &h->threadpool, h->param.i_threads, (void*)x264_encoder_thread_init, h ) )
1097         goto fail;
1098
1099     h->thread[0] = h;
1100     for( int i = 1; i < h->param.i_threads + !!h->param.i_sync_lookahead; i++ )
1101         CHECKED_MALLOC( h->thread[i], sizeof(x264_t) );
1102
1103     for( int i = 0; i < h->param.i_threads; i++ )
1104     {
1105         int init_nal_count = h->param.i_slice_count + 3;
1106         int allocate_threadlocal_data = !h->param.b_sliced_threads || !i;
1107         if( i > 0 )
1108             *h->thread[i] = *h;
1109
1110         if( allocate_threadlocal_data )
1111         {
1112             h->thread[i]->fdec = x264_frame_pop_unused( h, 1 );
1113             if( !h->thread[i]->fdec )
1114                 goto fail;
1115         }
1116         else
1117             h->thread[i]->fdec = h->thread[0]->fdec;
1118
1119         CHECKED_MALLOC( h->thread[i]->out.p_bitstream, h->out.i_bitstream );
1120         /* Start each thread with room for init_nal_count NAL units; it'll realloc later if needed. */
1121         CHECKED_MALLOC( h->thread[i]->out.nal, init_nal_count*sizeof(x264_nal_t) );
1122         h->thread[i]->out.i_nals_allocated = init_nal_count;
1123
1124         if( allocate_threadlocal_data && x264_macroblock_cache_allocate( h->thread[i] ) < 0 )
1125             goto fail;
1126     }
1127
1128     if( x264_lookahead_init( h, i_slicetype_length ) )
1129         goto fail;
1130
1131     for( int i = 0; i < h->param.i_threads; i++ )
1132         if( x264_macroblock_thread_allocate( h->thread[i], 0 ) < 0 )
1133             goto fail;
1134
1135     if( x264_ratecontrol_new( h ) < 0 )
1136         goto fail;
1137
1138     if( h->param.i_nal_hrd )
1139     {
1140         x264_log( h, X264_LOG_DEBUG, "HRD bitrate: %i bits/sec\n", h->sps->vui.hrd.i_bit_rate_unscaled );
1141         x264_log( h, X264_LOG_DEBUG, "CPB size: %i bits\n", h->sps->vui.hrd.i_cpb_size_unscaled );
1142     }
1143
1144     if( h->param.psz_dump_yuv )
1145     {
1146         /* create or truncate the reconstructed video file */
1147         FILE *f = fopen( h->param.psz_dump_yuv, "w" );
1148         if( !f )
1149         {
1150             x264_log( h, X264_LOG_ERROR, "dump_yuv: can't write to %s\n", h->param.psz_dump_yuv );
1151             goto fail;
1152         }
1153         else if( !x264_is_regular_file( f ) )
1154         {
1155             x264_log( h, X264_LOG_ERROR, "dump_yuv: incompatible with non-regular file %s\n", h->param.psz_dump_yuv );
1156             goto fail;
1157         }
1158         fclose( f );
1159     }
1160
1161     const char *profile = h->sps->i_profile_idc == PROFILE_BASELINE ? "Constrained Baseline" :
1162                           h->sps->i_profile_idc == PROFILE_MAIN ? "Main" :
1163                           h->sps->i_profile_idc == PROFILE_HIGH ? "High" :
1164                           h->sps->i_profile_idc == PROFILE_HIGH10 ? (h->sps->b_constraint_set3 == 1 ? "High 10 Intra" : "High 10") :
1165                           "High 4:4:4 Predictive";
1166     char level[4];
1167     snprintf( level, sizeof(level), "%d.%d", h->sps->i_level_idc/10, h->sps->i_level_idc%10 );
1168     if( h->sps->i_level_idc == 9 || ( h->sps->i_level_idc == 11 && h->sps->b_constraint_set3 ) )
1169         strcpy( level, "1b" );
1170
1171     if( h->sps->i_profile_idc < PROFILE_HIGH10 )
1172     {
1173         x264_log( h, X264_LOG_INFO, "profile %s, level %s\n",
1174             profile, level );
1175     }
1176     else
1177     {
1178         x264_log( h, X264_LOG_INFO, "profile %s, level %s, bit depth %d\n",
1179             profile, level, BIT_DEPTH );
1180     }
1181
1182     return h;
1183 fail:
1184     x264_free( h );
1185     return NULL;
1186 }
1187
1188 /****************************************************************************
1189  * x264_encoder_reconfig:
1190  ****************************************************************************/
1191 int x264_encoder_reconfig( x264_t *h, x264_param_t *param )
1192 {
1193     int rc_reconfig = 0;
1194     h = h->thread[h->thread[0]->i_thread_phase];
1195     x264_set_aspect_ratio( h, param, 0 );
1196 #define COPY(var) h->param.var = param->var
1197     COPY( i_frame_reference ); // but never uses more refs than initially specified
1198     COPY( i_bframe_bias );
1199     if( h->param.i_scenecut_threshold )
1200         COPY( i_scenecut_threshold ); // can't turn it on or off, only vary the threshold
1201     COPY( b_deblocking_filter );
1202     COPY( i_deblocking_filter_alphac0 );
1203     COPY( i_deblocking_filter_beta );
1204     COPY( analyse.inter );
1205     COPY( analyse.intra );
1206     COPY( analyse.i_direct_mv_pred );
1207     /* Scratch buffer prevents me_range from being increased for esa/tesa */
1208     if( h->param.analyse.i_me_method < X264_ME_ESA || param->analyse.i_me_range < h->param.analyse.i_me_range )
1209         COPY( analyse.i_me_range );
1210     COPY( analyse.i_noise_reduction );
1211     /* We can't switch out of subme=0 during encoding. */
1212     if( h->param.analyse.i_subpel_refine )
1213         COPY( analyse.i_subpel_refine );
1214     COPY( analyse.i_trellis );
1215     COPY( analyse.b_chroma_me );
1216     COPY( analyse.b_dct_decimate );
1217     COPY( analyse.b_fast_pskip );
1218     COPY( analyse.b_mixed_references );
1219     COPY( analyse.f_psy_rd );
1220     COPY( analyse.f_psy_trellis );
1221     COPY( crop_rect );
1222     // can only twiddle these if they were enabled to begin with:
1223     if( h->param.analyse.i_me_method >= X264_ME_ESA || param->analyse.i_me_method < X264_ME_ESA )
1224         COPY( analyse.i_me_method );
1225     if( h->param.analyse.i_me_method >= X264_ME_ESA && !h->frames.b_have_sub8x8_esa )
1226         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
1227     if( h->pps->b_transform_8x8_mode )
1228         COPY( analyse.b_transform_8x8 );
1229     if( h->frames.i_max_ref1 > 1 )
1230         COPY( i_bframe_pyramid );
1231     COPY( i_slice_max_size );
1232     COPY( i_slice_max_mbs );
1233     COPY( i_slice_count );
1234
1235     /* VBV can't be turned on if it wasn't on to begin with */
1236     if( h->param.rc.i_vbv_max_bitrate > 0 && h->param.rc.i_vbv_buffer_size > 0 &&
1237           param->rc.i_vbv_max_bitrate > 0 &&   param->rc.i_vbv_buffer_size > 0 )
1238     {
1239         COPY( rc.i_vbv_max_bitrate );
1240         COPY( rc.i_vbv_buffer_size );
1241         COPY( rc.i_bitrate );
1242         rc_reconfig = 1;
1243     }
1244     if( h->param.rc.f_rf_constant != param->rc.f_rf_constant )
1245     {
1246         COPY( rc.f_rf_constant );
1247         rc_reconfig = 1;
1248     }
1249     if( h->param.rc.f_rf_constant_max != param->rc.f_rf_constant_max )
1250     {
1251         COPY( rc.f_rf_constant_max );
1252         rc_reconfig = 1;
1253     }
1254
1255 #undef COPY
1256
1257     mbcmp_init( h );
1258
1259     int ret = x264_validate_parameters( h );
1260
1261     /* Supported reconfiguration options (1-pass only):
1262      * vbv-maxrate
1263      * vbv-bufsize
1264      * crf
1265      * bitrate (CBR only) */
1266     if( !ret && rc_reconfig )
1267         x264_ratecontrol_init_reconfigurable( h, 0 );
1268
1269     return ret;
1270 }
1271
1272 /****************************************************************************
1273  * x264_encoder_parameters:
1274  ****************************************************************************/
1275 void x264_encoder_parameters( x264_t *h, x264_param_t *param )
1276 {
1277     memcpy( param, &h->thread[h->i_thread_phase]->param, sizeof(x264_param_t) );
1278 }
1279
1280 /* internal usage */
1281 static void x264_nal_start( x264_t *h, int i_type, int i_ref_idc )
1282 {
1283     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1284
1285     nal->i_ref_idc        = i_ref_idc;
1286     nal->i_type           = i_type;
1287     nal->b_long_startcode = 1;
1288
1289     nal->i_payload= 0;
1290     nal->p_payload= &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8];
1291 }
1292
1293 /* if number of allocated nals is not enough, re-allocate a larger one. */
1294 static int x264_nal_check_buffer( x264_t *h )
1295 {
1296     if( h->out.i_nal >= h->out.i_nals_allocated )
1297     {
1298         x264_nal_t *new_out = x264_malloc( sizeof(x264_nal_t) * (h->out.i_nals_allocated*2) );
1299         if( !new_out )
1300             return -1;
1301         memcpy( new_out, h->out.nal, sizeof(x264_nal_t) * (h->out.i_nals_allocated) );
1302         x264_free( h->out.nal );
1303         h->out.nal = new_out;
1304         h->out.i_nals_allocated *= 2;
1305     }
1306     return 0;
1307 }
1308
1309 static int x264_nal_end( x264_t *h )
1310 {
1311     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1312     nal->i_payload = &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8] - nal->p_payload;
1313     if( h->param.nalu_process )
1314         h->param.nalu_process( h, nal );
1315     h->out.i_nal++;
1316
1317     return x264_nal_check_buffer( h );
1318 }
1319
1320 static int x264_encoder_encapsulate_nals( x264_t *h, int start )
1321 {
1322     int nal_size = 0, previous_nal_size = 0;
1323
1324     if( h->param.nalu_process )
1325     {
1326         for( int i = start; i < h->out.i_nal; i++ )
1327             nal_size += h->out.nal[i].i_payload;
1328         return nal_size;
1329     }
1330
1331     for( int i = 0; i < start; i++ )
1332         previous_nal_size += h->out.nal[i].i_payload;
1333
1334     for( int i = start; i < h->out.i_nal; i++ )
1335         nal_size += h->out.nal[i].i_payload;
1336
1337     /* Worst-case NAL unit escaping: reallocate the buffer if it's too small. */
1338     if( h->nal_buffer_size < nal_size * 3/2 + h->out.i_nal * 4 )
1339     {
1340         uint8_t *buf = x264_malloc( nal_size * 2 + h->out.i_nal * 4 );
1341         if( !buf )
1342             return -1;
1343         if( previous_nal_size )
1344             memcpy( buf, h->nal_buffer, previous_nal_size );
1345         x264_free( h->nal_buffer );
1346         h->nal_buffer = buf;
1347     }
1348
1349     uint8_t *nal_buffer = h->nal_buffer + previous_nal_size;
1350
1351     for( int i = start; i < h->out.i_nal; i++ )
1352     {
1353         h->out.nal[i].b_long_startcode = !i || h->out.nal[i].i_type == NAL_SPS || h->out.nal[i].i_type == NAL_PPS;
1354         x264_nal_encode( h, nal_buffer, &h->out.nal[i] );
1355         nal_buffer += h->out.nal[i].i_payload;
1356     }
1357
1358     x264_emms();
1359
1360     return nal_buffer - (h->nal_buffer + previous_nal_size);
1361 }
1362
1363 /****************************************************************************
1364  * x264_encoder_headers:
1365  ****************************************************************************/
1366 int x264_encoder_headers( x264_t *h, x264_nal_t **pp_nal, int *pi_nal )
1367 {
1368     int frame_size = 0;
1369     /* init bitstream context */
1370     h->out.i_nal = 0;
1371     bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
1372
1373     /* Write SEI, SPS and PPS. */
1374
1375     /* generate sequence parameters */
1376     x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
1377     x264_sps_write( &h->out.bs, h->sps );
1378     if( x264_nal_end( h ) )
1379         return -1;
1380
1381     /* generate picture parameters */
1382     x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
1383     x264_pps_write( &h->out.bs, h->pps );
1384     if( x264_nal_end( h ) )
1385         return -1;
1386
1387     /* identify ourselves */
1388     x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
1389     if( x264_sei_version_write( h, &h->out.bs ) )
1390         return -1;
1391     if( x264_nal_end( h ) )
1392         return -1;
1393
1394     frame_size = x264_encoder_encapsulate_nals( h, 0 );
1395
1396     /* now set output*/
1397     *pi_nal = h->out.i_nal;
1398     *pp_nal = &h->out.nal[0];
1399     h->out.i_nal = 0;
1400
1401     return frame_size;
1402 }
1403
1404 /* Check to see whether we have chosen a reference list ordering different
1405  * from the standard's default. */
1406 static inline void x264_reference_check_reorder( x264_t *h )
1407 {
1408     /* The reorder check doesn't check for missing frames, so just
1409      * force a reorder if one of the reference list is corrupt. */
1410     for( int i = 0; h->frames.reference[i]; i++ )
1411         if( h->frames.reference[i]->b_corrupt )
1412         {
1413             h->b_ref_reorder[0] = 1;
1414             return;
1415         }
1416     for( int i = 0; i < h->i_ref0 - 1; i++ )
1417         /* P and B-frames use different default orders. */
1418         if( h->sh.i_type == SLICE_TYPE_P ? h->fref0[i]->i_frame_num < h->fref0[i+1]->i_frame_num
1419                                          : h->fref0[i]->i_poc < h->fref0[i+1]->i_poc )
1420         {
1421             h->b_ref_reorder[0] = 1;
1422             return;
1423         }
1424 }
1425
1426 /* return -1 on failure, else return the index of the new reference frame */
1427 int x264_weighted_reference_duplicate( x264_t *h, int i_ref, const x264_weight_t *w )
1428 {
1429     int i = h->i_ref0;
1430     int j = 1;
1431     x264_frame_t *newframe;
1432     if( i <= 1 ) /* empty list, definitely can't duplicate frame */
1433         return -1;
1434
1435     /* Duplication is a hack to compensate for crappy rounding in motion compensation.
1436      * With high bit depth, it's not worth doing, so turn it off except in the case of
1437      * unweighted dupes. */
1438     if( BIT_DEPTH > 8 && w != weight_none )
1439         return -1;
1440
1441     newframe = x264_frame_pop_blank_unused( h );
1442
1443     //FIXME: probably don't need to copy everything
1444     *newframe = *h->fref0[i_ref];
1445     newframe->i_reference_count = 1;
1446     newframe->orig = h->fref0[i_ref];
1447     newframe->b_duplicate = 1;
1448     memcpy( h->fenc->weight[j], w, sizeof(h->fenc->weight[i]) );
1449
1450     /* shift the frames to make space for the dupe. */
1451     h->b_ref_reorder[0] = 1;
1452     if( h->i_ref0 < X264_REF_MAX )
1453         ++h->i_ref0;
1454     h->fref0[X264_REF_MAX-1] = NULL;
1455     x264_frame_unshift( &h->fref0[j], newframe );
1456
1457     return j;
1458 }
1459
1460 static void x264_weighted_pred_init( x264_t *h )
1461 {
1462     /* for now no analysis and set all weights to nothing */
1463     for( int i_ref = 0; i_ref < h->i_ref0; i_ref++ )
1464         h->fenc->weighted[i_ref] = h->fref0[i_ref]->filtered[0];
1465
1466     // FIXME: This only supports weighting of one reference frame
1467     // and duplicates of that frame.
1468     h->fenc->i_lines_weighted = 0;
1469
1470     for( int i_ref = 0; i_ref < (h->i_ref0 << h->sh.b_mbaff); i_ref++ )
1471         for( int i = 0; i < 3; i++ )
1472             h->sh.weight[i_ref][i].weightfn = NULL;
1473
1474
1475     if( h->sh.i_type != SLICE_TYPE_P || h->param.analyse.i_weighted_pred <= 0 )
1476         return;
1477
1478     int i_padv = PADV << h->param.b_interlaced;
1479     int denom = -1;
1480     int weightplane[2] = { 0, 0 };
1481     int buffer_next = 0;
1482     for( int i = 0; i < 3; i++ )
1483     {
1484         for( int j = 0; j < h->i_ref0; j++ )
1485         {
1486             if( h->fenc->weight[j][i].weightfn )
1487             {
1488                 h->sh.weight[j][i] = h->fenc->weight[j][i];
1489                 // if weight is useless, don't write it to stream
1490                 if( h->sh.weight[j][i].i_scale == 1<<h->sh.weight[j][i].i_denom && h->sh.weight[j][i].i_offset == 0 )
1491                     h->sh.weight[j][i].weightfn = NULL;
1492                 else
1493                 {
1494                     if( !weightplane[!!i] )
1495                     {
1496                         weightplane[!!i] = 1;
1497                         h->sh.weight[0][!!i].i_denom = denom = h->sh.weight[j][i].i_denom;
1498                         assert( x264_clip3( denom, 0, 7 ) == denom );
1499                     }
1500
1501                     assert( h->sh.weight[j][i].i_denom == denom );
1502                     if( !i )
1503                     {
1504                         h->fenc->weighted[j] = h->mb.p_weight_buf[buffer_next++] + h->fenc->i_stride[0] * i_padv + PADH;
1505                         //scale full resolution frame
1506                         if( h->param.i_threads == 1 )
1507                         {
1508                             pixel *src = h->fref0[j]->filtered[0] - h->fref0[j]->i_stride[0]*i_padv - PADH;
1509                             pixel *dst = h->fenc->weighted[j] - h->fenc->i_stride[0]*i_padv - PADH;
1510                             int stride = h->fenc->i_stride[0];
1511                             int width = h->fenc->i_width[0] + PADH*2;
1512                             int height = h->fenc->i_lines[0] + i_padv*2;
1513                             x264_weight_scale_plane( h, dst, stride, src, stride, width, height, &h->sh.weight[j][0] );
1514                             h->fenc->i_lines_weighted = height;
1515                         }
1516                     }
1517                 }
1518             }
1519         }
1520     }
1521
1522     if( weightplane[1] )
1523         for( int i = 0; i < h->i_ref0; i++ )
1524         {
1525             if( h->sh.weight[i][1].weightfn && !h->sh.weight[i][2].weightfn )
1526             {
1527                 h->sh.weight[i][2].i_scale = 1 << h->sh.weight[0][1].i_denom;
1528                 h->sh.weight[i][2].i_offset = 0;
1529             }
1530             else if( h->sh.weight[i][2].weightfn && !h->sh.weight[i][1].weightfn )
1531             {
1532                 h->sh.weight[i][1].i_scale = 1 << h->sh.weight[0][1].i_denom;
1533                 h->sh.weight[i][1].i_offset = 0;
1534             }
1535         }
1536
1537     if( !weightplane[0] )
1538         h->sh.weight[0][0].i_denom = 0;
1539     if( !weightplane[1] )
1540         h->sh.weight[0][1].i_denom = 0;
1541     h->sh.weight[0][2].i_denom = h->sh.weight[0][1].i_denom;
1542 }
1543
1544 static inline void x264_reference_build_list( x264_t *h, int i_poc )
1545 {
1546     int b_ok;
1547
1548     /* build ref list 0/1 */
1549     h->mb.pic.i_fref[0] = h->i_ref0 = 0;
1550     h->mb.pic.i_fref[1] = h->i_ref1 = 0;
1551     if( h->sh.i_type == SLICE_TYPE_I )
1552         return;
1553
1554     for( int i = 0; h->frames.reference[i]; i++ )
1555     {
1556         if( h->frames.reference[i]->b_corrupt )
1557             continue;
1558         if( h->frames.reference[i]->i_poc < i_poc )
1559             h->fref0[h->i_ref0++] = h->frames.reference[i];
1560         else if( h->frames.reference[i]->i_poc > i_poc )
1561             h->fref1[h->i_ref1++] = h->frames.reference[i];
1562     }
1563
1564     /* Order ref0 from higher to lower poc */
1565     do
1566     {
1567         b_ok = 1;
1568         for( int i = 0; i < h->i_ref0 - 1; i++ )
1569         {
1570             if( h->fref0[i]->i_poc < h->fref0[i+1]->i_poc )
1571             {
1572                 XCHG( x264_frame_t*, h->fref0[i], h->fref0[i+1] );
1573                 b_ok = 0;
1574                 break;
1575             }
1576         }
1577     } while( !b_ok );
1578
1579     if( h->sh.i_mmco_remove_from_end )
1580         for( int i = h->i_ref0-1; i >= h->i_ref0 - h->sh.i_mmco_remove_from_end; i-- )
1581         {
1582             int diff = h->i_frame_num - h->fref0[i]->i_frame_num;
1583             h->sh.mmco[h->sh.i_mmco_command_count].i_poc = h->fref0[i]->i_poc;
1584             h->sh.mmco[h->sh.i_mmco_command_count++].i_difference_of_pic_nums = diff;
1585         }
1586
1587     /* Order ref1 from lower to higher poc (bubble sort) for B-frame */
1588     do
1589     {
1590         b_ok = 1;
1591         for( int i = 0; i < h->i_ref1 - 1; i++ )
1592         {
1593             if( h->fref1[i]->i_poc > h->fref1[i+1]->i_poc )
1594             {
1595                 XCHG( x264_frame_t*, h->fref1[i], h->fref1[i+1] );
1596                 b_ok = 0;
1597                 break;
1598             }
1599         }
1600     } while( !b_ok );
1601
1602     x264_reference_check_reorder( h );
1603
1604     h->i_ref1 = X264_MIN( h->i_ref1, h->frames.i_max_ref1 );
1605     h->i_ref0 = X264_MIN( h->i_ref0, h->frames.i_max_ref0 );
1606     h->i_ref0 = X264_MIN( h->i_ref0, h->param.i_frame_reference ); // if reconfig() has lowered the limit
1607
1608     /* add duplicates */
1609     if( h->fenc->i_type == X264_TYPE_P )
1610     {
1611         int idx = -1;
1612         if( h->param.analyse.i_weighted_pred == X264_WEIGHTP_SMART )
1613         {
1614             x264_weight_t w[3];
1615             w[1].weightfn = w[2].weightfn = NULL;
1616             if( h->param.rc.b_stat_read )
1617                 x264_ratecontrol_set_weights( h, h->fenc );
1618
1619             if( !h->fenc->weight[0][0].weightfn )
1620             {
1621                 h->fenc->weight[0][0].i_denom = 0;
1622                 SET_WEIGHT( w[0], 1, 1, 0, -1 );
1623                 idx = x264_weighted_reference_duplicate( h, 0, w );
1624             }
1625             else
1626             {
1627                 if( h->fenc->weight[0][0].i_scale == 1<<h->fenc->weight[0][0].i_denom )
1628                 {
1629                     SET_WEIGHT( h->fenc->weight[0][0], 1, 1, 0, h->fenc->weight[0][0].i_offset );
1630                 }
1631                 x264_weighted_reference_duplicate( h, 0, weight_none );
1632                 if( h->fenc->weight[0][0].i_offset > -128 )
1633                 {
1634                     w[0] = h->fenc->weight[0][0];
1635                     w[0].i_offset--;
1636                     h->mc.weight_cache( h, &w[0] );
1637                     idx = x264_weighted_reference_duplicate( h, 0, w );
1638                 }
1639             }
1640         }
1641         else if( h->param.analyse.i_weighted_pred == X264_WEIGHTP_BLIND )
1642         {
1643             //weighted offset=-1
1644             x264_weight_t w[3];
1645             SET_WEIGHT( w[0], 1, 1, 0, -1 );
1646             h->fenc->weight[0][0].i_denom = 0;
1647             w[1].weightfn = w[2].weightfn = NULL;
1648             idx = x264_weighted_reference_duplicate( h, 0, w );
1649         }
1650         h->mb.ref_blind_dupe = idx;
1651     }
1652
1653     assert( h->i_ref0 + h->i_ref1 <= X264_REF_MAX );
1654     h->mb.pic.i_fref[0] = h->i_ref0;
1655     h->mb.pic.i_fref[1] = h->i_ref1;
1656 }
1657
1658 static void x264_fdec_filter_row( x264_t *h, int mb_y, int b_inloop )
1659 {
1660     /* mb_y is the mb to be encoded next, not the mb to be filtered here */
1661     int b_hpel = h->fdec->b_kept_as_ref;
1662     int b_deblock = h->sh.i_disable_deblocking_filter_idc != 1;
1663     int b_end = mb_y == h->i_threadslice_end;
1664     int b_measure_quality = 1;
1665     int min_y = mb_y - (1 << h->sh.b_mbaff);
1666     int b_start = min_y == h->i_threadslice_start;
1667     int max_y = b_end ? h->i_threadslice_end : mb_y;
1668     b_deblock &= b_hpel || h->param.psz_dump_yuv;
1669     if( h->param.b_sliced_threads && b_start && min_y && !b_inloop )
1670     {
1671         b_deblock = 0;         /* We already deblocked on the inloop pass. */
1672         b_measure_quality = 0; /* We already measured quality on the inloop pass. */
1673     }
1674     if( mb_y & h->sh.b_mbaff )
1675         return;
1676     if( min_y < h->i_threadslice_start )
1677         return;
1678
1679     if( b_deblock )
1680         for( int y = min_y; y < max_y; y += (1 << h->sh.b_mbaff) )
1681             x264_frame_deblock_row( h, y );
1682
1683     if( b_hpel )
1684     {
1685         int end = mb_y == h->mb.i_mb_height;
1686         x264_frame_expand_border( h, h->fdec, min_y, end );
1687         if( h->param.analyse.i_subpel_refine )
1688         {
1689             x264_frame_filter( h, h->fdec, min_y, end );
1690             x264_frame_expand_border_filtered( h, h->fdec, min_y, end );
1691         }
1692     }
1693
1694     if( h->i_thread_frames > 1 && h->fdec->b_kept_as_ref )
1695         x264_frame_cond_broadcast( h->fdec, mb_y*16 + (b_end ? 10000 : -(X264_THREAD_HEIGHT << h->sh.b_mbaff)) );
1696
1697     min_y = min_y*16 - 8 * !b_start;
1698     max_y = b_end ? X264_MIN( h->i_threadslice_end*16 , h->param.i_height ) : mb_y*16 - 8;
1699
1700     if( b_measure_quality )
1701     {
1702         if( h->param.analyse.b_psnr )
1703         {
1704             uint64_t ssd_y = x264_pixel_ssd_wxh( &h->pixf,
1705                 h->fdec->plane[0] + min_y * h->fdec->i_stride[0], h->fdec->i_stride[0],
1706                 h->fenc->plane[0] + min_y * h->fenc->i_stride[0], h->fenc->i_stride[0],
1707                 h->param.i_width, max_y-min_y );
1708             uint64_t ssd_u, ssd_v;
1709             x264_pixel_ssd_nv12( &h->pixf,
1710                 h->fdec->plane[1] + (min_y>>1) * h->fdec->i_stride[1], h->fdec->i_stride[1],
1711                 h->fenc->plane[1] + (min_y>>1) * h->fenc->i_stride[1], h->fenc->i_stride[1],
1712                 h->param.i_width>>1, (max_y-min_y)>>1, &ssd_u, &ssd_v );
1713             h->stat.frame.i_ssd[0] += ssd_y;
1714             h->stat.frame.i_ssd[1] += ssd_u;
1715             h->stat.frame.i_ssd[2] += ssd_v;
1716         }
1717
1718         if( h->param.analyse.b_ssim )
1719         {
1720             x264_emms();
1721             /* offset by 2 pixels to avoid alignment of ssim blocks with dct blocks,
1722              * and overlap by 4 */
1723             min_y += b_start ? 2 : -6;
1724             h->stat.frame.f_ssim +=
1725                 x264_pixel_ssim_wxh( &h->pixf,
1726                     h->fdec->plane[0] + 2+min_y*h->fdec->i_stride[0], h->fdec->i_stride[0],
1727                     h->fenc->plane[0] + 2+min_y*h->fenc->i_stride[0], h->fenc->i_stride[0],
1728                     h->param.i_width-2, max_y-min_y, h->scratch_buffer );
1729         }
1730     }
1731 }
1732
1733 static inline int x264_reference_update( x264_t *h )
1734 {
1735     if( !h->fdec->b_kept_as_ref )
1736     {
1737         if( h->i_thread_frames > 1 )
1738         {
1739             x264_frame_push_unused( h, h->fdec );
1740             h->fdec = x264_frame_pop_unused( h, 1 );
1741             if( !h->fdec )
1742                 return -1;
1743         }
1744         return 0;
1745     }
1746
1747     /* apply mmco from previous frame. */
1748     for( int i = 0; i < h->sh.i_mmco_command_count; i++ )
1749         for( int j = 0; h->frames.reference[j]; j++ )
1750             if( h->frames.reference[j]->i_poc == h->sh.mmco[i].i_poc )
1751                 x264_frame_push_unused( h, x264_frame_shift( &h->frames.reference[j] ) );
1752
1753     /* move frame in the buffer */
1754     x264_frame_push( h->frames.reference, h->fdec );
1755     if( h->frames.reference[h->sps->i_num_ref_frames] )
1756         x264_frame_push_unused( h, x264_frame_shift( h->frames.reference ) );
1757     h->fdec = x264_frame_pop_unused( h, 1 );
1758     if( !h->fdec )
1759         return -1;
1760     return 0;
1761 }
1762
1763 static inline void x264_reference_reset( x264_t *h )
1764 {
1765     while( h->frames.reference[0] )
1766         x264_frame_push_unused( h, x264_frame_pop( h->frames.reference ) );
1767     h->fdec->i_poc =
1768     h->fenc->i_poc = 0;
1769 }
1770
1771 static inline void x264_reference_hierarchy_reset( x264_t *h )
1772 {
1773     int ref;
1774     int b_hasdelayframe = 0;
1775
1776     /* look for delay frames -- chain must only contain frames that are disposable */
1777     for( int i = 0; h->frames.current[i] && IS_DISPOSABLE( h->frames.current[i]->i_type ); i++ )
1778         b_hasdelayframe |= h->frames.current[i]->i_coded
1779                         != h->frames.current[i]->i_frame + h->sps->vui.i_num_reorder_frames;
1780
1781     /* This function must handle b-pyramid and clear frames for open-gop */
1782     if( h->param.i_bframe_pyramid != X264_B_PYRAMID_STRICT && !b_hasdelayframe && h->frames.i_poc_last_open_gop == -1 )
1783         return;
1784
1785     /* Remove last BREF. There will never be old BREFs in the
1786      * dpb during a BREF decode when pyramid == STRICT */
1787     for( ref = 0; h->frames.reference[ref]; ref++ )
1788     {
1789         if( ( h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT
1790             && h->frames.reference[ref]->i_type == X264_TYPE_BREF )
1791             || ( h->frames.reference[ref]->i_poc < h->frames.i_poc_last_open_gop
1792             && h->sh.i_type != SLICE_TYPE_B ) )
1793         {
1794             int diff = h->i_frame_num - h->frames.reference[ref]->i_frame_num;
1795             h->sh.mmco[h->sh.i_mmco_command_count].i_difference_of_pic_nums = diff;
1796             h->sh.mmco[h->sh.i_mmco_command_count++].i_poc = h->frames.reference[ref]->i_poc;
1797             x264_frame_push_unused( h, x264_frame_shift( &h->frames.reference[ref] ) );
1798             h->b_ref_reorder[0] = 1;
1799             ref--;
1800         }
1801     }
1802
1803     /* Prepare room in the dpb for the delayed display time of the later b-frame's */
1804     if( h->param.i_bframe_pyramid )
1805         h->sh.i_mmco_remove_from_end = X264_MAX( ref + 2 - h->frames.i_max_dpb, 0 );
1806 }
1807
1808 static inline void x264_slice_init( x264_t *h, int i_nal_type, int i_global_qp )
1809 {
1810     /* ------------------------ Create slice header  ----------------------- */
1811     if( i_nal_type == NAL_SLICE_IDR )
1812     {
1813         x264_slice_header_init( h, &h->sh, h->sps, h->pps, h->i_idr_pic_id, h->i_frame_num, i_global_qp );
1814
1815         /* alternate id */
1816         h->i_idr_pic_id ^= 1;
1817     }
1818     else
1819     {
1820         x264_slice_header_init( h, &h->sh, h->sps, h->pps, -1, h->i_frame_num, i_global_qp );
1821
1822         h->sh.i_num_ref_idx_l0_active = h->i_ref0 <= 0 ? 1 : h->i_ref0;
1823         h->sh.i_num_ref_idx_l1_active = h->i_ref1 <= 0 ? 1 : h->i_ref1;
1824         if( h->sh.i_num_ref_idx_l0_active != h->pps->i_num_ref_idx_l0_default_active ||
1825             (h->sh.i_type == SLICE_TYPE_B && h->sh.i_num_ref_idx_l1_active != h->pps->i_num_ref_idx_l1_default_active) )
1826         {
1827             h->sh.b_num_ref_idx_override = 1;
1828         }
1829     }
1830
1831     h->fdec->i_frame_num = h->sh.i_frame_num;
1832
1833     if( h->sps->i_poc_type == 0 )
1834     {
1835         h->sh.i_poc = h->fdec->i_poc;
1836         if( h->param.b_interlaced )
1837         {
1838             h->sh.i_delta_poc_bottom = h->param.b_tff ? 1 : -1;
1839             if( h->sh.i_delta_poc_bottom == -1 )
1840                 h->sh.i_poc = h->fdec->i_poc + 1;
1841         }
1842         else
1843             h->sh.i_delta_poc_bottom = 0;
1844     }
1845     else if( h->sps->i_poc_type == 1 )
1846     {
1847         /* FIXME TODO FIXME */
1848     }
1849     else
1850     {
1851         /* Nothing to do ? */
1852     }
1853
1854     x264_macroblock_slice_init( h );
1855 }
1856
1857 static int x264_slice_write( x264_t *h )
1858 {
1859     int i_skip;
1860     int mb_xy, i_mb_x, i_mb_y;
1861     int i_skip_bak = 0; /* Shut up GCC. */
1862     bs_t bs_bak;
1863     x264_cabac_t cabac_bak;
1864     uint8_t cabac_prevbyte_bak = 0; /* Shut up GCC. */
1865     int mv_bits_bak = 0;
1866     int tex_bits_bak = 0;
1867     /* NALUs other than the first use a 3-byte startcode.
1868      * Add one extra byte for the rbsp, and one more for the final CABAC putbyte.
1869      * Then add an extra 5 bytes just in case, to account for random NAL escapes and
1870      * other inaccuracies. */
1871     int overhead_guess = (NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal)) + 1 + h->param.b_cabac + 5;
1872     int slice_max_size = h->param.i_slice_max_size > 0 ? (h->param.i_slice_max_size-overhead_guess)*8 : 0;
1873     int starting_bits = bs_pos(&h->out.bs);
1874     int b_deblock = h->sh.i_disable_deblocking_filter_idc != 1;
1875     int b_hpel = h->fdec->b_kept_as_ref;
1876     b_deblock &= b_hpel || h->param.psz_dump_yuv;
1877     bs_realign( &h->out.bs );
1878
1879     /* Slice */
1880     x264_nal_start( h, h->i_nal_type, h->i_nal_ref_idc );
1881     h->out.nal[h->out.i_nal].i_first_mb = h->sh.i_first_mb;
1882
1883     /* Slice header */
1884     x264_macroblock_thread_init( h );
1885
1886     /* If this isn't the first slice in the threadslice, set the slice QP
1887      * equal to the last QP in the previous slice for more accurate
1888      * CABAC initialization. */
1889     if( h->sh.i_first_mb != h->i_threadslice_start * h->mb.i_mb_width )
1890     {
1891         h->sh.i_qp = h->mb.i_last_qp;
1892         h->sh.i_qp_delta = h->sh.i_qp - h->pps->i_pic_init_qp;
1893     }
1894
1895     x264_slice_header_write( &h->out.bs, &h->sh, h->i_nal_ref_idc );
1896     if( h->param.b_cabac )
1897     {
1898         /* alignment needed */
1899         bs_align_1( &h->out.bs );
1900
1901         /* init cabac */
1902         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 );
1903         x264_cabac_encode_init ( &h->cabac, h->out.bs.p, h->out.bs.p_end );
1904     }
1905     h->mb.i_last_qp = h->sh.i_qp;
1906     h->mb.i_last_dqp = 0;
1907
1908     i_mb_y = h->sh.i_first_mb / h->mb.i_mb_width;
1909     i_mb_x = h->sh.i_first_mb % h->mb.i_mb_width;
1910     i_skip = 0;
1911
1912     while( (mb_xy = i_mb_x + i_mb_y * h->mb.i_mb_width) <= h->sh.i_last_mb )
1913     {
1914         int mb_spos = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
1915
1916         if( x264_bitstream_check_buffer( h ) )
1917             return -1;
1918
1919         if( slice_max_size )
1920         {
1921             mv_bits_bak = h->stat.frame.i_mv_bits;
1922             tex_bits_bak = h->stat.frame.i_tex_bits;
1923             /* We don't need the contexts because flushing the CABAC encoder has no context
1924              * dependency and macroblocks are only re-encoded in the case where a slice is
1925              * ended (and thus the content of all contexts are thrown away). */
1926             if( h->param.b_cabac )
1927             {
1928                 memcpy( &cabac_bak, &h->cabac, offsetof(x264_cabac_t, f8_bits_encoded) );
1929                 /* x264's CABAC writer modifies the previous byte during carry, so it has to be
1930                  * backed up. */
1931                 cabac_prevbyte_bak = h->cabac.p[-1];
1932             }
1933             else
1934             {
1935                 bs_bak = h->out.bs;
1936                 i_skip_bak = i_skip;
1937             }
1938         }
1939
1940         if( i_mb_x == 0 && !h->mb.b_reencode_mb )
1941             x264_fdec_filter_row( h, i_mb_y, 1 );
1942
1943         /* load cache */
1944         x264_macroblock_cache_load( h, i_mb_x, i_mb_y );
1945
1946         x264_macroblock_analyse( h );
1947
1948         /* encode this macroblock -> be careful it can change the mb type to P_SKIP if needed */
1949         x264_macroblock_encode( h );
1950
1951         if( h->param.b_cabac )
1952         {
1953             if( mb_xy > h->sh.i_first_mb && !(h->sh.b_mbaff && (i_mb_y&1)) )
1954                 x264_cabac_encode_terminal( &h->cabac );
1955
1956             if( IS_SKIP( h->mb.i_type ) )
1957                 x264_cabac_mb_skip( h, 1 );
1958             else
1959             {
1960                 if( h->sh.i_type != SLICE_TYPE_I )
1961                     x264_cabac_mb_skip( h, 0 );
1962                 x264_macroblock_write_cabac( h, &h->cabac );
1963             }
1964         }
1965         else
1966         {
1967             if( IS_SKIP( h->mb.i_type ) )
1968                 i_skip++;
1969             else
1970             {
1971                 if( h->sh.i_type != SLICE_TYPE_I )
1972                 {
1973                     bs_write_ue( &h->out.bs, i_skip );  /* skip run */
1974                     i_skip = 0;
1975                 }
1976                 x264_macroblock_write_cavlc( h );
1977             }
1978         }
1979
1980         int total_bits = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
1981         int mb_size = total_bits - mb_spos;
1982
1983         if( slice_max_size )
1984         {
1985             /* Count the skip run, just in case. */
1986             if( !h->param.b_cabac )
1987                 total_bits += bs_size_ue_big( i_skip );
1988             /* HACK: we assume no more than 3 bytes of NALU escaping, but
1989              * this can fail in CABAC streams with an extremely large number of identical
1990              * blocks in sequence (e.g. all-black intra blocks).
1991              * Thus, every 64 blocks, pretend we've used a byte.
1992              * For reference, a seqeuence of identical empty-CBP i16x16 blocks will use
1993              * one byte after 26 macroblocks, assuming a perfectly adapted CABAC.
1994              * That's 78 macroblocks to generate the 3-byte sequence to trigger an escape. */
1995             else if( ((mb_xy - h->sh.i_first_mb) & 63) == 63 )
1996                 slice_max_size -= 8;
1997             /* We'll just re-encode this last macroblock if we go over the max slice size. */
1998             if( total_bits - starting_bits > slice_max_size && !h->mb.b_reencode_mb )
1999             {
2000                 if( mb_xy != h->sh.i_first_mb )
2001                 {
2002                     h->stat.frame.i_mv_bits = mv_bits_bak;
2003                     h->stat.frame.i_tex_bits = tex_bits_bak;
2004                     if( h->param.b_cabac )
2005                     {
2006                         memcpy( &h->cabac, &cabac_bak, offsetof(x264_cabac_t, f8_bits_encoded) );
2007                         h->cabac.p[-1] = cabac_prevbyte_bak;
2008                     }
2009                     else
2010                     {
2011                         h->out.bs = bs_bak;
2012                         i_skip = i_skip_bak;
2013                     }
2014                     h->mb.b_reencode_mb = 1;
2015                     h->sh.i_last_mb = mb_xy-1;
2016                     break;
2017                 }
2018                 else
2019                 {
2020                     h->sh.i_last_mb = mb_xy;
2021                     h->mb.b_reencode_mb = 0;
2022                 }
2023             }
2024             else
2025                 h->mb.b_reencode_mb = 0;
2026         }
2027
2028 #if HAVE_VISUALIZE
2029         if( h->param.b_visualize )
2030             x264_visualize_mb( h );
2031 #endif
2032
2033         /* save cache */
2034         x264_macroblock_cache_save( h );
2035
2036         /* accumulate mb stats */
2037         h->stat.frame.i_mb_count[h->mb.i_type]++;
2038
2039         int b_intra = IS_INTRA( h->mb.i_type );
2040         if( h->param.i_log_level >= X264_LOG_INFO || h->param.rc.b_stat_write )
2041         {
2042             if( !b_intra && !IS_SKIP( h->mb.i_type ) && !IS_DIRECT( h->mb.i_type ) )
2043             {
2044                 if( h->mb.i_partition != D_8x8 )
2045                         h->stat.frame.i_mb_partition[h->mb.i_partition] += 4;
2046                     else
2047                         for( int i = 0; i < 4; i++ )
2048                             h->stat.frame.i_mb_partition[h->mb.i_sub_partition[i]] ++;
2049                 if( h->param.i_frame_reference > 1 )
2050                     for( int i_list = 0; i_list <= (h->sh.i_type == SLICE_TYPE_B); i_list++ )
2051                         for( int i = 0; i < 4; i++ )
2052                         {
2053                             int i_ref = h->mb.cache.ref[i_list][ x264_scan8[4*i] ];
2054                             if( i_ref >= 0 )
2055                                 h->stat.frame.i_mb_count_ref[i_list][i_ref] ++;
2056                         }
2057             }
2058         }
2059
2060         if( h->param.i_log_level >= X264_LOG_INFO )
2061         {
2062             if( h->mb.i_cbp_luma | h->mb.i_cbp_chroma )
2063             {
2064                 int cbpsum = (h->mb.i_cbp_luma&1) + ((h->mb.i_cbp_luma>>1)&1)
2065                            + ((h->mb.i_cbp_luma>>2)&1) + (h->mb.i_cbp_luma>>3);
2066                 h->stat.frame.i_mb_cbp[!b_intra + 0] += cbpsum;
2067                 h->stat.frame.i_mb_cbp[!b_intra + 2] += !!h->mb.i_cbp_chroma;
2068                 h->stat.frame.i_mb_cbp[!b_intra + 4] += h->mb.i_cbp_chroma >> 1;
2069             }
2070             if( h->mb.i_cbp_luma && !b_intra )
2071             {
2072                 h->stat.frame.i_mb_count_8x8dct[0] ++;
2073                 h->stat.frame.i_mb_count_8x8dct[1] += h->mb.b_transform_8x8;
2074             }
2075             if( b_intra && h->mb.i_type != I_PCM )
2076             {
2077                 if( h->mb.i_type == I_16x16 )
2078                     h->stat.frame.i_mb_pred_mode[0][h->mb.i_intra16x16_pred_mode]++;
2079                 else if( h->mb.i_type == I_8x8 )
2080                     for( int i = 0; i < 16; i += 4 )
2081                         h->stat.frame.i_mb_pred_mode[1][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
2082                 else //if( h->mb.i_type == I_4x4 )
2083                     for( int i = 0; i < 16; i++ )
2084                         h->stat.frame.i_mb_pred_mode[2][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
2085                 h->stat.frame.i_mb_pred_mode[3][x264_mb_pred_mode8x8c_fix[h->mb.i_chroma_pred_mode]]++;
2086             }
2087         }
2088
2089         /* calculate deblock strength values (actual deblocking is done per-row along with hpel) */
2090         if( b_deblock )
2091         {
2092             int mvy_limit = 4 >> h->sh.b_mbaff;
2093             uint8_t (*bs)[4][4] = h->deblock_strength[h->mb.i_mb_y&1][h->mb.i_mb_x];
2094             x264_macroblock_cache_load_deblock( h );
2095             if( IS_INTRA( h->mb.type[h->mb.i_mb_xy] ) )
2096                 memset( bs, 3, 2*4*4*sizeof(uint8_t) );
2097             else
2098                 h->loopf.deblock_strength( h->mb.cache.non_zero_count, h->mb.cache.ref, h->mb.cache.mv,
2099                                            bs, mvy_limit, h->sh.i_type == SLICE_TYPE_B );
2100         }
2101
2102         x264_ratecontrol_mb( h, mb_size );
2103
2104         if( h->sh.b_mbaff )
2105         {
2106             i_mb_x += i_mb_y & 1;
2107             i_mb_y ^= i_mb_x < h->mb.i_mb_width;
2108         }
2109         else
2110             i_mb_x++;
2111         if( i_mb_x == h->mb.i_mb_width )
2112         {
2113             i_mb_y++;
2114             i_mb_x = 0;
2115         }
2116     }
2117     h->out.nal[h->out.i_nal].i_last_mb = h->sh.i_last_mb;
2118
2119     if( h->param.b_cabac )
2120     {
2121         x264_cabac_encode_flush( h, &h->cabac );
2122         h->out.bs.p = h->cabac.p;
2123     }
2124     else
2125     {
2126         if( i_skip > 0 )
2127             bs_write_ue( &h->out.bs, i_skip );  /* last skip run */
2128         /* rbsp_slice_trailing_bits */
2129         bs_rbsp_trailing( &h->out.bs );
2130         bs_flush( &h->out.bs );
2131     }
2132     if( x264_nal_end( h ) )
2133         return -1;
2134
2135     if( h->sh.i_last_mb == (h->i_threadslice_end * h->mb.i_mb_width - 1) )
2136     {
2137         h->stat.frame.i_misc_bits = bs_pos( &h->out.bs )
2138                                   + (h->out.i_nal*NALU_OVERHEAD * 8)
2139                                   - h->stat.frame.i_tex_bits
2140                                   - h->stat.frame.i_mv_bits;
2141         x264_fdec_filter_row( h, h->i_threadslice_end, 1 );
2142     }
2143
2144     return 0;
2145 }
2146
2147 static void x264_thread_sync_context( x264_t *dst, x264_t *src )
2148 {
2149     if( dst == src )
2150         return;
2151
2152     // reference counting
2153     for( x264_frame_t **f = src->frames.reference; *f; f++ )
2154         (*f)->i_reference_count++;
2155     for( x264_frame_t **f = dst->frames.reference; *f; f++ )
2156         x264_frame_push_unused( src, *f );
2157     src->fdec->i_reference_count++;
2158     x264_frame_push_unused( src, dst->fdec );
2159
2160     // copy everything except the per-thread pointers and the constants.
2161     memcpy( &dst->i_frame, &src->i_frame, offsetof(x264_t, mb.type) - offsetof(x264_t, i_frame) );
2162     dst->param = src->param;
2163     dst->stat = src->stat;
2164 }
2165
2166 static void x264_thread_sync_stat( x264_t *dst, x264_t *src )
2167 {
2168     if( dst == src )
2169         return;
2170     memcpy( &dst->stat.i_frame_count, &src->stat.i_frame_count, sizeof(dst->stat) - sizeof(dst->stat.frame) );
2171 }
2172
2173 static void *x264_slices_write( x264_t *h )
2174 {
2175     int i_slice_num = 0;
2176     int last_thread_mb = h->sh.i_last_mb;
2177
2178 #if HAVE_VISUALIZE
2179     if( h->param.b_visualize )
2180         if( x264_visualize_init( h ) )
2181             return (void *)-1;
2182 #endif
2183
2184     /* init stats */
2185     memset( &h->stat.frame, 0, sizeof(h->stat.frame) );
2186     h->mb.b_reencode_mb = 0;
2187     while( h->sh.i_first_mb <= last_thread_mb )
2188     {
2189         h->sh.i_last_mb = last_thread_mb;
2190         if( h->param.i_slice_max_mbs )
2191             h->sh.i_last_mb = h->sh.i_first_mb + h->param.i_slice_max_mbs - 1;
2192         else if( h->param.i_slice_count && !h->param.b_sliced_threads )
2193         {
2194             int height = h->mb.i_mb_height >> h->param.b_interlaced;
2195             int width = h->mb.i_mb_width << h->param.b_interlaced;
2196             i_slice_num++;
2197             h->sh.i_last_mb = (height * i_slice_num + h->param.i_slice_count/2) / h->param.i_slice_count * width - 1;
2198         }
2199         h->sh.i_last_mb = X264_MIN( h->sh.i_last_mb, last_thread_mb );
2200         if( x264_stack_align( x264_slice_write, h ) )
2201             return (void *)-1;
2202         h->sh.i_first_mb = h->sh.i_last_mb + 1;
2203     }
2204
2205 #if HAVE_VISUALIZE
2206     if( h->param.b_visualize )
2207     {
2208         x264_visualize_show( h );
2209         x264_visualize_close( h );
2210     }
2211 #endif
2212
2213     return (void *)0;
2214 }
2215
2216 static int x264_threaded_slices_write( x264_t *h )
2217 {
2218     /* set first/last mb and sync contexts */
2219     for( int i = 0; i < h->param.i_threads; i++ )
2220     {
2221         x264_t *t = h->thread[i];
2222         if( i )
2223         {
2224             t->param = h->param;
2225             memcpy( &t->i_frame, &h->i_frame, offsetof(x264_t, rc) - offsetof(x264_t, i_frame) );
2226         }
2227         int height = h->mb.i_mb_height >> h->param.b_interlaced;
2228         t->i_threadslice_start = ((height *  i    + h->param.i_slice_count/2) / h->param.i_threads) << h->param.b_interlaced;
2229         t->i_threadslice_end   = ((height * (i+1) + h->param.i_slice_count/2) / h->param.i_threads) << h->param.b_interlaced;
2230         t->sh.i_first_mb = t->i_threadslice_start * h->mb.i_mb_width;
2231         t->sh.i_last_mb  =   t->i_threadslice_end * h->mb.i_mb_width - 1;
2232     }
2233
2234     x264_stack_align( x264_analyse_weight_frame, h, h->mb.i_mb_height*16 + 16 );
2235
2236     x264_threads_distribute_ratecontrol( h );
2237
2238     /* dispatch */
2239     for( int i = 0; i < h->param.i_threads; i++ )
2240     {
2241         x264_threadpool_run( h->threadpool, (void*)x264_slices_write, h->thread[i] );
2242         h->thread[i]->b_thread_active = 1;
2243     }
2244     for( int i = 0; i < h->param.i_threads; i++ )
2245     {
2246         h->thread[i]->b_thread_active = 0;
2247         if( (intptr_t)x264_threadpool_wait( h->threadpool, h->thread[i] ) )
2248             return -1;
2249     }
2250
2251     /* Go back and fix up the hpel on the borders between slices. */
2252     for( int i = 1; i < h->param.i_threads; i++ )
2253     {
2254         x264_fdec_filter_row( h->thread[i], h->thread[i]->i_threadslice_start + 1, 0 );
2255         if( h->sh.b_mbaff )
2256             x264_fdec_filter_row( h->thread[i], h->thread[i]->i_threadslice_start + 2, 0 );
2257     }
2258
2259     x264_threads_merge_ratecontrol( h );
2260
2261     for( int i = 1; i < h->param.i_threads; i++ )
2262     {
2263         x264_t *t = h->thread[i];
2264         for( int j = 0; j < t->out.i_nal; j++ )
2265         {
2266             h->out.nal[h->out.i_nal] = t->out.nal[j];
2267             h->out.i_nal++;
2268             x264_nal_check_buffer( h );
2269         }
2270         /* All entries in stat.frame are ints except for ssd/ssim. */
2271         for( int j = 0; j < (offsetof(x264_t,stat.frame.i_ssd) - offsetof(x264_t,stat.frame.i_mv_bits)) / sizeof(int); j++ )
2272             ((int*)&h->stat.frame)[j] += ((int*)&t->stat.frame)[j];
2273         for( int j = 0; j < 3; j++ )
2274             h->stat.frame.i_ssd[j] += t->stat.frame.i_ssd[j];
2275         h->stat.frame.f_ssim += t->stat.frame.f_ssim;
2276     }
2277
2278     return 0;
2279 }
2280
2281 void x264_encoder_intra_refresh( x264_t *h )
2282 {
2283     h = h->thread[h->i_thread_phase];
2284     h->b_queued_intra_refresh = 1;
2285 }
2286
2287 int x264_encoder_invalidate_reference( x264_t *h, int64_t pts )
2288 {
2289     if( h->param.i_bframe )
2290     {
2291         x264_log( h, X264_LOG_ERROR, "x264_encoder_invalidate_reference is not supported with B-frames enabled\n" );
2292         return -1;
2293     }
2294     if( h->param.b_intra_refresh )
2295     {
2296         x264_log( h, X264_LOG_ERROR, "x264_encoder_invalidate_reference is not supported with intra refresh enabled\n" );
2297         return -1;
2298     }
2299     h = h->thread[h->i_thread_phase];
2300     if( pts >= h->i_last_idr_pts )
2301     {
2302         for( int i = 0; h->frames.reference[i]; i++ )
2303             if( pts <= h->frames.reference[i]->i_pts )
2304                 h->frames.reference[i]->b_corrupt = 1;
2305         if( pts <= h->fdec->i_pts )
2306             h->fdec->b_corrupt = 1;
2307     }
2308     return 0;
2309 }
2310
2311 /****************************************************************************
2312  * x264_encoder_encode:
2313  *  XXX: i_poc   : is the poc of the current given picture
2314  *       i_frame : is the number of the frame being coded
2315  *  ex:  type frame poc
2316  *       I      0   2*0
2317  *       P      1   2*3
2318  *       B      2   2*1
2319  *       B      3   2*2
2320  *       P      4   2*6
2321  *       B      5   2*4
2322  *       B      6   2*5
2323  ****************************************************************************/
2324 int     x264_encoder_encode( x264_t *h,
2325                              x264_nal_t **pp_nal, int *pi_nal,
2326                              x264_picture_t *pic_in,
2327                              x264_picture_t *pic_out )
2328 {
2329     x264_t *thread_current, *thread_prev, *thread_oldest;
2330     int i_nal_type, i_nal_ref_idc, i_global_qp;
2331     int overhead = NALU_OVERHEAD;
2332
2333     if( h->i_thread_frames > 1 )
2334     {
2335         thread_prev    = h->thread[ h->i_thread_phase ];
2336         h->i_thread_phase = (h->i_thread_phase + 1) % h->i_thread_frames;
2337         thread_current = h->thread[ h->i_thread_phase ];
2338         thread_oldest  = h->thread[ (h->i_thread_phase + 1) % h->i_thread_frames ];
2339         x264_thread_sync_context( thread_current, thread_prev );
2340         x264_thread_sync_ratecontrol( thread_current, thread_prev, thread_oldest );
2341         h = thread_current;
2342     }
2343     else
2344     {
2345         thread_current =
2346         thread_oldest  = h;
2347     }
2348 #if HAVE_MMX
2349     if( h->i_thread_frames == 1 && h->param.cpu&X264_CPU_SSE_MISALIGN )
2350         x264_cpu_mask_misalign_sse();
2351 #endif
2352
2353     // ok to call this before encoding any frames, since the initial values of fdec have b_kept_as_ref=0
2354     if( x264_reference_update( h ) )
2355         return -1;
2356     h->fdec->i_lines_completed = -1;
2357
2358     /* no data out */
2359     *pi_nal = 0;
2360     *pp_nal = NULL;
2361
2362     /* ------------------- Setup new frame from picture -------------------- */
2363     if( pic_in != NULL )
2364     {
2365         /* 1: Copy the picture to a frame and move it to a buffer */
2366         x264_frame_t *fenc = x264_frame_pop_unused( h, 0 );
2367         if( !fenc )
2368             return -1;
2369
2370         if( x264_frame_copy_picture( h, fenc, pic_in ) < 0 )
2371             return -1;
2372
2373         if( h->param.i_width != 16 * h->mb.i_mb_width ||
2374             h->param.i_height != 16 * h->mb.i_mb_height )
2375             x264_frame_expand_border_mod16( h, fenc );
2376
2377         fenc->i_frame = h->frames.i_input++;
2378
2379         if( fenc->i_frame == 0 )
2380             h->frames.i_first_pts = fenc->i_pts;
2381         if( h->frames.i_bframe_delay && fenc->i_frame == h->frames.i_bframe_delay )
2382             h->frames.i_bframe_delay_time = fenc->i_pts - h->frames.i_first_pts;
2383
2384         if( h->param.b_vfr_input && fenc->i_pts <= h->frames.i_largest_pts )
2385             x264_log( h, X264_LOG_WARNING, "non-strictly-monotonic PTS\n" );
2386
2387         h->frames.i_second_largest_pts = h->frames.i_largest_pts;
2388         h->frames.i_largest_pts = fenc->i_pts;
2389
2390         if( (fenc->i_pic_struct < PIC_STRUCT_AUTO) || (fenc->i_pic_struct > PIC_STRUCT_TRIPLE) )
2391             fenc->i_pic_struct = PIC_STRUCT_AUTO;
2392
2393         if( fenc->i_pic_struct == PIC_STRUCT_AUTO )
2394         {
2395             int b_interlaced = fenc->param ? fenc->param->b_interlaced : h->param.b_interlaced;
2396             if( b_interlaced )
2397             {
2398                 int b_tff = fenc->param ? fenc->param->b_tff : h->param.b_tff;
2399                 fenc->i_pic_struct = b_tff ? PIC_STRUCT_TOP_BOTTOM : PIC_STRUCT_BOTTOM_TOP;
2400             }
2401             else
2402                 fenc->i_pic_struct = PIC_STRUCT_PROGRESSIVE;
2403         }
2404
2405         if( h->param.rc.b_mb_tree && h->param.rc.b_stat_read )
2406         {
2407             if( x264_macroblock_tree_read( h, fenc, pic_in->prop.quant_offsets ) )
2408                 return -1;
2409         }
2410         else
2411             x264_stack_align( x264_adaptive_quant_frame, h, fenc, pic_in->prop.quant_offsets );
2412
2413         if( pic_in->prop.quant_offsets_free )
2414             pic_in->prop.quant_offsets_free( pic_in->prop.quant_offsets );
2415
2416         if( h->frames.b_have_lowres )
2417             x264_frame_init_lowres( h, fenc );
2418
2419         /* 2: Place the frame into the queue for its slice type decision */
2420         x264_lookahead_put_frame( h, fenc );
2421
2422         if( h->frames.i_input <= h->frames.i_delay + 1 - h->i_thread_frames )
2423         {
2424             /* Nothing yet to encode, waiting for filling of buffers */
2425             pic_out->i_type = X264_TYPE_AUTO;
2426             return 0;
2427         }
2428     }
2429     else
2430     {
2431         /* signal kills for lookahead thread */
2432         x264_pthread_mutex_lock( &h->lookahead->ifbuf.mutex );
2433         h->lookahead->b_exit_thread = 1;
2434         x264_pthread_cond_broadcast( &h->lookahead->ifbuf.cv_fill );
2435         x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
2436     }
2437
2438     h->i_frame++;
2439     /* 3: The picture is analyzed in the lookahead */
2440     if( !h->frames.current[0] )
2441         x264_lookahead_get_frames( h );
2442
2443     if( !h->frames.current[0] && x264_lookahead_is_empty( h ) )
2444         return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
2445
2446     /* ------------------- Get frame to be encoded ------------------------- */
2447     /* 4: get picture to encode */
2448     h->fenc = x264_frame_shift( h->frames.current );
2449     if( h->i_frame == h->i_thread_frames - 1 )
2450         h->i_reordered_pts_delay = h->fenc->i_reordered_pts;
2451     if( h->fenc->param )
2452     {
2453         x264_encoder_reconfig( h, h->fenc->param );
2454         if( h->fenc->param->param_free )
2455             h->fenc->param->param_free( h->fenc->param );
2456     }
2457
2458     if( !IS_X264_TYPE_I( h->fenc->i_type ) )
2459     {
2460         int valid_refs_left = 0;
2461         for( int i = 0; h->frames.reference[i]; i++ )
2462             if( !h->frames.reference[i]->b_corrupt )
2463                 valid_refs_left++;
2464         /* No valid reference frames left: force an IDR. */
2465         if( !valid_refs_left )
2466         {
2467             h->fenc->b_keyframe = 1;
2468             h->fenc->i_type = X264_TYPE_IDR;
2469         }
2470     }
2471
2472     if( h->fenc->b_keyframe )
2473     {
2474         h->frames.i_last_keyframe = h->fenc->i_frame;
2475         if( h->fenc->i_type == X264_TYPE_IDR )
2476         {
2477             h->i_frame_num = 0;
2478             h->frames.i_last_idr = h->fenc->i_frame;
2479         }
2480     }
2481     h->sh.i_mmco_command_count =
2482     h->sh.i_mmco_remove_from_end = 0;
2483     h->b_ref_reorder[0] =
2484     h->b_ref_reorder[1] = 0;
2485     h->fdec->i_poc =
2486     h->fenc->i_poc = 2 * ( h->fenc->i_frame - X264_MAX( h->frames.i_last_idr, 0 ) );
2487
2488     /* ------------------- Setup frame context ----------------------------- */
2489     /* 5: Init data dependent of frame type */
2490     if( h->fenc->i_type == X264_TYPE_IDR )
2491     {
2492         /* reset ref pictures */
2493         i_nal_type    = NAL_SLICE_IDR;
2494         i_nal_ref_idc = NAL_PRIORITY_HIGHEST;
2495         h->sh.i_type = SLICE_TYPE_I;
2496         x264_reference_reset( h );
2497         h->frames.i_poc_last_open_gop = -1;
2498     }
2499     else if( h->fenc->i_type == X264_TYPE_I )
2500     {
2501         i_nal_type    = NAL_SLICE;
2502         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
2503         h->sh.i_type = SLICE_TYPE_I;
2504         x264_reference_hierarchy_reset( h );
2505         if( h->param.i_open_gop )
2506             h->frames.i_poc_last_open_gop = h->fenc->b_keyframe ? h->fenc->i_poc : -1;
2507     }
2508     else if( h->fenc->i_type == X264_TYPE_P )
2509     {
2510         i_nal_type    = NAL_SLICE;
2511         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
2512         h->sh.i_type = SLICE_TYPE_P;
2513         x264_reference_hierarchy_reset( h );
2514         h->frames.i_poc_last_open_gop = -1;
2515     }
2516     else if( h->fenc->i_type == X264_TYPE_BREF )
2517     {
2518         i_nal_type    = NAL_SLICE;
2519         i_nal_ref_idc = h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT ? NAL_PRIORITY_LOW : NAL_PRIORITY_HIGH;
2520         h->sh.i_type = SLICE_TYPE_B;
2521         x264_reference_hierarchy_reset( h );
2522     }
2523     else    /* B frame */
2524     {
2525         i_nal_type    = NAL_SLICE;
2526         i_nal_ref_idc = NAL_PRIORITY_DISPOSABLE;
2527         h->sh.i_type = SLICE_TYPE_B;
2528     }
2529
2530     h->fdec->i_type = h->fenc->i_type;
2531     h->fdec->i_frame = h->fenc->i_frame;
2532     h->fenc->b_kept_as_ref =
2533     h->fdec->b_kept_as_ref = i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE && h->param.i_keyint_max > 1;
2534
2535     h->fdec->i_pts = h->fenc->i_pts;
2536     if( h->frames.i_bframe_delay )
2537     {
2538         int64_t *prev_reordered_pts = thread_current->frames.i_prev_reordered_pts;
2539         h->fdec->i_dts = h->i_frame > h->frames.i_bframe_delay
2540                        ? prev_reordered_pts[ (h->i_frame - h->frames.i_bframe_delay) % h->frames.i_bframe_delay ]
2541                        : h->fenc->i_reordered_pts - h->frames.i_bframe_delay_time;
2542         prev_reordered_pts[ h->i_frame % h->frames.i_bframe_delay ] = h->fenc->i_reordered_pts;
2543     }
2544     else
2545         h->fdec->i_dts = h->fenc->i_reordered_pts;
2546     if( h->fenc->i_type == X264_TYPE_IDR )
2547         h->i_last_idr_pts = h->fdec->i_pts;
2548
2549     /* ------------------- Init                ----------------------------- */
2550     /* build ref list 0/1 */
2551     x264_reference_build_list( h, h->fdec->i_poc );
2552
2553     /* ---------------------- Write the bitstream -------------------------- */
2554     /* Init bitstream context */
2555     if( h->param.b_sliced_threads )
2556     {
2557         for( int i = 0; i < h->param.i_threads; i++ )
2558         {
2559             bs_init( &h->thread[i]->out.bs, h->thread[i]->out.p_bitstream, h->thread[i]->out.i_bitstream );
2560             h->thread[i]->out.i_nal = 0;
2561         }
2562     }
2563     else
2564     {
2565         bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
2566         h->out.i_nal = 0;
2567     }
2568
2569     if( h->param.b_aud )
2570     {
2571         int pic_type;
2572
2573         if( h->sh.i_type == SLICE_TYPE_I )
2574             pic_type = 0;
2575         else if( h->sh.i_type == SLICE_TYPE_P )
2576             pic_type = 1;
2577         else if( h->sh.i_type == SLICE_TYPE_B )
2578             pic_type = 2;
2579         else
2580             pic_type = 7;
2581
2582         x264_nal_start( h, NAL_AUD, NAL_PRIORITY_DISPOSABLE );
2583         bs_write( &h->out.bs, 3, pic_type );
2584         bs_rbsp_trailing( &h->out.bs );
2585         if( x264_nal_end( h ) )
2586             return -1;
2587         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2588     }
2589
2590     h->i_nal_type = i_nal_type;
2591     h->i_nal_ref_idc = i_nal_ref_idc;
2592
2593     if( h->param.b_intra_refresh )
2594     {
2595         if( IS_X264_TYPE_I( h->fenc->i_type ) )
2596         {
2597             h->fdec->i_frames_since_pir = 0;
2598             h->b_queued_intra_refresh = 0;
2599             /* PIR is currently only supported with ref == 1, so any intra frame effectively refreshes
2600              * the whole frame and counts as an intra refresh. */
2601             h->fdec->f_pir_position = h->mb.i_mb_width;
2602         }
2603         else if( h->fenc->i_type == X264_TYPE_P )
2604         {
2605             int pocdiff = (h->fdec->i_poc - h->fref0[0]->i_poc)/2;
2606             float increment = X264_MAX( ((float)h->mb.i_mb_width-1) / h->param.i_keyint_max, 1 );
2607             h->fdec->f_pir_position = h->fref0[0]->f_pir_position;
2608             h->fdec->i_frames_since_pir = h->fref0[0]->i_frames_since_pir + pocdiff;
2609             if( h->fdec->i_frames_since_pir >= h->param.i_keyint_max ||
2610                 (h->b_queued_intra_refresh && h->fdec->f_pir_position + 0.5 >= h->mb.i_mb_width) )
2611             {
2612                 h->fdec->f_pir_position = 0;
2613                 h->fdec->i_frames_since_pir = 0;
2614                 h->b_queued_intra_refresh = 0;
2615                 h->fenc->b_keyframe = 1;
2616             }
2617             h->fdec->i_pir_start_col = h->fdec->f_pir_position+0.5;
2618             h->fdec->f_pir_position += increment * pocdiff;
2619             h->fdec->i_pir_end_col = h->fdec->f_pir_position+0.5;
2620             /* If our intra refresh has reached the right side of the frame, we're done. */
2621             if( h->fdec->i_pir_end_col >= h->mb.i_mb_width - 1 )
2622                 h->fdec->f_pir_position = h->mb.i_mb_width;
2623         }
2624     }
2625
2626     if( h->fenc->b_keyframe )
2627     {
2628         /* Write SPS and PPS */
2629         if( h->param.b_repeat_headers )
2630         {
2631             /* generate sequence parameters */
2632             x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
2633             x264_sps_write( &h->out.bs, h->sps );
2634             if( x264_nal_end( h ) )
2635                 return -1;
2636             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
2637
2638             /* generate picture parameters */
2639             x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
2640             x264_pps_write( &h->out.bs, h->pps );
2641             if( x264_nal_end( h ) )
2642                 return -1;
2643             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
2644         }
2645
2646         /* buffering period sei is written in x264_encoder_frame_end */
2647     }
2648
2649     /* write extra sei */
2650     for( int i = 0; i < h->fenc->extra_sei.num_payloads; i++ )
2651     {
2652         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2653         x264_sei_write( &h->out.bs, h->fenc->extra_sei.payloads[i].payload, h->fenc->extra_sei.payloads[i].payload_size,
2654                         h->fenc->extra_sei.payloads[i].payload_type );
2655         if( x264_nal_end( h ) )
2656             return -1;
2657         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2658         if( h->fenc->extra_sei.sei_free && h->fenc->extra_sei.payloads[i].payload )
2659             h->fenc->extra_sei.sei_free( h->fenc->extra_sei.payloads[i].payload );
2660     }
2661
2662     if( h->fenc->extra_sei.sei_free && h->fenc->extra_sei.payloads )
2663         h->fenc->extra_sei.sei_free( h->fenc->extra_sei.payloads );
2664
2665     if( h->fenc->b_keyframe )
2666     {
2667         if( h->param.b_repeat_headers && h->fenc->i_frame == 0 )
2668         {
2669             /* identify ourself */
2670             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2671             if( x264_sei_version_write( h, &h->out.bs ) )
2672                 return -1;
2673             if( x264_nal_end( h ) )
2674                 return -1;
2675             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2676         }
2677
2678         if( h->fenc->i_type != X264_TYPE_IDR )
2679         {
2680             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;
2681             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2682             x264_sei_recovery_point_write( h, &h->out.bs, time_to_recovery );
2683             x264_nal_end( h );
2684             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2685         }
2686     }
2687
2688     /* generate sei pic timing */
2689     if( h->sps->vui.b_pic_struct_present || h->sps->vui.b_nal_hrd_parameters_present )
2690     {
2691         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2692         x264_sei_pic_timing_write( h, &h->out.bs );
2693         if( x264_nal_end( h ) )
2694             return -1;
2695         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2696     }
2697
2698     if( h->fenc->b_keyframe && h->param.b_intra_refresh )
2699         h->i_cpb_delay_pir_offset = h->fenc->i_cpb_delay;
2700
2701     /* Init the rate control */
2702     /* FIXME: Include slice header bit cost. */
2703     x264_ratecontrol_start( h, h->fenc->i_qpplus1, overhead*8 );
2704     i_global_qp = x264_ratecontrol_qp( h );
2705
2706     pic_out->i_qpplus1 =
2707     h->fdec->i_qpplus1 = i_global_qp + 1;
2708
2709     if( h->param.rc.b_stat_read && h->sh.i_type != SLICE_TYPE_I )
2710     {
2711         x264_reference_build_list_optimal( h );
2712         x264_reference_check_reorder( h );
2713     }
2714
2715     if( h->i_ref0 )
2716         h->fdec->i_poc_l0ref0 = h->fref0[0]->i_poc;
2717
2718     if( h->sh.i_type == SLICE_TYPE_B )
2719         x264_macroblock_bipred_init( h );
2720
2721     /*------------------------- Weights -------------------------------------*/
2722     x264_weighted_pred_init( h );
2723
2724     /* ------------------------ Create slice header  ----------------------- */
2725     x264_slice_init( h, i_nal_type, i_global_qp );
2726
2727     if( i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE )
2728         h->i_frame_num++;
2729
2730     /* Write frame */
2731     h->i_threadslice_start = 0;
2732     h->i_threadslice_end = h->mb.i_mb_height;
2733     if( h->i_thread_frames > 1 )
2734     {
2735         x264_threadpool_run( h->threadpool, (void*)x264_slices_write, h );
2736         h->b_thread_active = 1;
2737     }
2738     else if( h->param.b_sliced_threads )
2739     {
2740         if( x264_threaded_slices_write( h ) )
2741             return -1;
2742     }
2743     else
2744         if( (intptr_t)x264_slices_write( h ) )
2745             return -1;
2746
2747     return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
2748 }
2749
2750 static int x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
2751                                    x264_nal_t **pp_nal, int *pi_nal,
2752                                    x264_picture_t *pic_out )
2753 {
2754     char psz_message[80];
2755
2756     if( h->b_thread_active )
2757     {
2758         h->b_thread_active = 0;
2759         if( (intptr_t)x264_threadpool_wait( h->threadpool, h ) )
2760             return -1;
2761     }
2762     if( !h->out.i_nal )
2763     {
2764         pic_out->i_type = X264_TYPE_AUTO;
2765         return 0;
2766     }
2767
2768     x264_emms();
2769     /* generate sei buffering period and insert it into place */
2770     if( h->fenc->b_keyframe && h->sps->vui.b_nal_hrd_parameters_present )
2771     {
2772         x264_hrd_fullness( h );
2773         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2774         x264_sei_buffering_period_write( h, &h->out.bs );
2775         if( x264_nal_end( h ) )
2776            return -1;
2777         /* buffering period sei must follow AUD, SPS and PPS and precede all other SEIs */
2778         int idx = 0;
2779         while( h->out.nal[idx].i_type == NAL_AUD ||
2780                h->out.nal[idx].i_type == NAL_SPS ||
2781                h->out.nal[idx].i_type == NAL_PPS )
2782             idx++;
2783         x264_nal_t nal_tmp = h->out.nal[h->out.i_nal-1];
2784         memmove( &h->out.nal[idx+1], &h->out.nal[idx], (h->out.i_nal-idx-1)*sizeof(x264_nal_t) );
2785         h->out.nal[idx] = nal_tmp;
2786     }
2787
2788     int frame_size = x264_encoder_encapsulate_nals( h, 0 );
2789
2790     /* Set output picture properties */
2791     pic_out->i_type = h->fenc->i_type;
2792
2793     pic_out->b_keyframe = h->fenc->b_keyframe;
2794
2795     pic_out->i_pts = h->fdec->i_pts;
2796     pic_out->i_dts = h->fdec->i_dts;
2797
2798     if( pic_out->i_pts < pic_out->i_dts )
2799         x264_log( h, X264_LOG_WARNING, "invalid DTS: PTS is less than DTS\n" );
2800
2801     pic_out->img.i_csp = X264_CSP_NV12;
2802 #if HIGH_BIT_DEPTH
2803     pic_out->img.i_csp |= X264_CSP_HIGH_DEPTH;
2804 #endif
2805     pic_out->img.i_plane = h->fdec->i_plane;
2806     for( int i = 0; i < 2; i++ )
2807     {
2808         pic_out->img.i_stride[i] = h->fdec->i_stride[i] * sizeof(pixel);
2809         pic_out->img.plane[i] = (uint8_t*)h->fdec->plane[i];
2810     }
2811
2812     x264_frame_push_unused( thread_current, h->fenc );
2813
2814     /* ---------------------- Update encoder state ------------------------- */
2815
2816     /* update rc */
2817     int filler = 0;
2818     if( x264_ratecontrol_end( h, frame_size * 8, &filler ) < 0 )
2819         return -1;
2820
2821     pic_out->hrd_timing = h->fenc->hrd_timing;
2822
2823     while( filler > 0 )
2824     {
2825         int f, overhead;
2826         overhead = (FILLER_OVERHEAD - h->param.b_annexb);
2827         if( h->param.i_slice_max_size && filler > h->param.i_slice_max_size )
2828         {
2829             int next_size = filler - h->param.i_slice_max_size;
2830             int overflow = X264_MAX( overhead - next_size, 0 );
2831             f = h->param.i_slice_max_size - overhead - overflow;
2832         }
2833         else
2834             f = X264_MAX( 0, filler - overhead );
2835
2836         x264_nal_start( h, NAL_FILLER, NAL_PRIORITY_DISPOSABLE );
2837         x264_filler_write( h, &h->out.bs, f );
2838         if( x264_nal_end( h ) )
2839             return -1;
2840         int total_size = x264_encoder_encapsulate_nals( h, h->out.i_nal-1 );
2841         frame_size += total_size;
2842         filler -= total_size;
2843     }
2844
2845     /* End bitstream, set output  */
2846     *pi_nal = h->out.i_nal;
2847     *pp_nal = h->out.nal;
2848
2849     h->out.i_nal = 0;
2850
2851     x264_noise_reduction_update( h );
2852
2853     /* ---------------------- Compute/Print statistics --------------------- */
2854     x264_thread_sync_stat( h, h->thread[0] );
2855
2856     /* Slice stat */
2857     h->stat.i_frame_count[h->sh.i_type]++;
2858     h->stat.i_frame_size[h->sh.i_type] += frame_size;
2859     h->stat.f_frame_qp[h->sh.i_type] += h->fdec->f_qp_avg_aq;
2860
2861     for( int i = 0; i < X264_MBTYPE_MAX; i++ )
2862         h->stat.i_mb_count[h->sh.i_type][i] += h->stat.frame.i_mb_count[i];
2863     for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
2864         h->stat.i_mb_partition[h->sh.i_type][i] += h->stat.frame.i_mb_partition[i];
2865     for( int i = 0; i < 2; i++ )
2866         h->stat.i_mb_count_8x8dct[i] += h->stat.frame.i_mb_count_8x8dct[i];
2867     for( int i = 0; i < 6; i++ )
2868         h->stat.i_mb_cbp[i] += h->stat.frame.i_mb_cbp[i];
2869     for( int i = 0; i < 4; i++ )
2870         for( int j = 0; j < 13; j++ )
2871             h->stat.i_mb_pred_mode[i][j] += h->stat.frame.i_mb_pred_mode[i][j];
2872     if( h->sh.i_type != SLICE_TYPE_I )
2873         for( int i_list = 0; i_list < 2; i_list++ )
2874             for( int i = 0; i < X264_REF_MAX*2; i++ )
2875                 h->stat.i_mb_count_ref[h->sh.i_type][i_list][i] += h->stat.frame.i_mb_count_ref[i_list][i];
2876     if( h->sh.i_type == SLICE_TYPE_P )
2877     {
2878         h->stat.i_consecutive_bframes[h->fdec->i_frame - h->fref0[0]->i_frame - 1]++;
2879         if( h->param.analyse.i_weighted_pred == X264_WEIGHTP_SMART )
2880         {
2881             h->stat.i_wpred[0] += !!h->sh.weight[0][0].weightfn;
2882             h->stat.i_wpred[1] += !!h->sh.weight[0][1].weightfn || !!h->sh.weight[0][2].weightfn;
2883         }
2884     }
2885     if( h->sh.i_type == SLICE_TYPE_B )
2886     {
2887         h->stat.i_direct_frames[ h->sh.b_direct_spatial_mv_pred ] ++;
2888         if( h->mb.b_direct_auto_write )
2889         {
2890             //FIXME somewhat arbitrary time constants
2891             if( h->stat.i_direct_score[0] + h->stat.i_direct_score[1] > h->mb.i_mb_count )
2892                 for( int i = 0; i < 2; i++ )
2893                     h->stat.i_direct_score[i] = h->stat.i_direct_score[i] * 9/10;
2894             for( int i = 0; i < 2; i++ )
2895                 h->stat.i_direct_score[i] += h->stat.frame.i_direct_score[i];
2896         }
2897     }
2898
2899     psz_message[0] = '\0';
2900     double dur = h->fenc->f_duration;
2901     h->stat.f_frame_duration[h->sh.i_type] += dur;
2902     if( h->param.analyse.b_psnr )
2903     {
2904         int64_t ssd[3] =
2905         {
2906             h->stat.frame.i_ssd[0],
2907             h->stat.frame.i_ssd[1],
2908             h->stat.frame.i_ssd[2],
2909         };
2910
2911         h->stat.f_ssd_global[h->sh.i_type]   += dur * (ssd[0] + ssd[1] + ssd[2]);
2912         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 );
2913         h->stat.f_psnr_mean_y[h->sh.i_type]  += dur * x264_psnr( ssd[0], h->param.i_width * h->param.i_height );
2914         h->stat.f_psnr_mean_u[h->sh.i_type]  += dur * x264_psnr( ssd[1], h->param.i_width * h->param.i_height / 4 );
2915         h->stat.f_psnr_mean_v[h->sh.i_type]  += dur * x264_psnr( ssd[2], h->param.i_width * h->param.i_height / 4 );
2916
2917         snprintf( psz_message, 80, " PSNR Y:%5.2f U:%5.2f V:%5.2f",
2918                   x264_psnr( ssd[0], h->param.i_width * h->param.i_height ),
2919                   x264_psnr( ssd[1], h->param.i_width * h->param.i_height / 4),
2920                   x264_psnr( ssd[2], h->param.i_width * h->param.i_height / 4) );
2921     }
2922
2923     if( h->param.analyse.b_ssim )
2924     {
2925         double ssim_y = h->stat.frame.f_ssim
2926                       / (((h->param.i_width-6)>>2) * ((h->param.i_height-6)>>2));
2927         h->stat.f_ssim_mean_y[h->sh.i_type] += ssim_y * dur;
2928         snprintf( psz_message + strlen(psz_message), 80 - strlen(psz_message),
2929                   " SSIM Y:%.5f", ssim_y );
2930     }
2931     psz_message[79] = '\0';
2932
2933     x264_log( h, X264_LOG_DEBUG,
2934                   "frame=%4d QP=%.2f NAL=%d Slice:%c Poc:%-3d I:%-4d P:%-4d SKIP:%-4d size=%d bytes%s\n",
2935               h->i_frame,
2936               h->fdec->f_qp_avg_aq,
2937               h->i_nal_ref_idc,
2938               h->sh.i_type == SLICE_TYPE_I ? 'I' : (h->sh.i_type == SLICE_TYPE_P ? 'P' : 'B' ),
2939               h->fdec->i_poc,
2940               h->stat.frame.i_mb_count_i,
2941               h->stat.frame.i_mb_count_p,
2942               h->stat.frame.i_mb_count_skip,
2943               frame_size,
2944               psz_message );
2945
2946     // keep stats all in one place
2947     x264_thread_sync_stat( h->thread[0], h );
2948     // for the use of the next frame
2949     x264_thread_sync_stat( thread_current, h );
2950
2951 #ifdef DEBUG_MB_TYPE
2952 {
2953     static const char mb_chars[] = { 'i', 'i', 'I', 'C', 'P', '8', 'S',
2954         'D', '<', 'X', 'B', 'X', '>', 'B', 'B', 'B', 'B', '8', 'S' };
2955     for( int mb_xy = 0; mb_xy < h->mb.i_mb_width * h->mb.i_mb_height; mb_xy++ )
2956     {
2957         if( h->mb.type[mb_xy] < X264_MBTYPE_MAX && h->mb.type[mb_xy] >= 0 )
2958             fprintf( stderr, "%c ", mb_chars[ h->mb.type[mb_xy] ] );
2959         else
2960             fprintf( stderr, "? " );
2961
2962         if( (mb_xy+1) % h->mb.i_mb_width == 0 )
2963             fprintf( stderr, "\n" );
2964     }
2965 }
2966 #endif
2967
2968     /* Remove duplicates, must be done near the end as breaks h->fref0 array
2969      * by freeing some of its pointers. */
2970     for( int i = 0; i < h->i_ref0; i++ )
2971         if( h->fref0[i] && h->fref0[i]->b_duplicate )
2972         {
2973             x264_frame_push_blank_unused( h, h->fref0[i] );
2974             h->fref0[i] = 0;
2975         }
2976
2977     if( h->param.psz_dump_yuv )
2978         x264_frame_dump( h );
2979     x264_emms();
2980
2981     return frame_size;
2982 }
2983
2984 static void x264_print_intra( int64_t *i_mb_count, double i_count, int b_print_pcm, char *intra )
2985 {
2986     intra += sprintf( intra, "I16..4%s: %4.1f%% %4.1f%% %4.1f%%",
2987         b_print_pcm ? "..PCM" : "",
2988         i_mb_count[I_16x16]/ i_count,
2989         i_mb_count[I_8x8]  / i_count,
2990         i_mb_count[I_4x4]  / i_count );
2991     if( b_print_pcm )
2992         sprintf( intra, " %4.1f%%", i_mb_count[I_PCM]  / i_count );
2993 }
2994
2995 /****************************************************************************
2996  * x264_encoder_close:
2997  ****************************************************************************/
2998 void    x264_encoder_close  ( x264_t *h )
2999 {
3000     int64_t i_yuv_size = 3 * h->param.i_width * h->param.i_height / 2;
3001     int64_t i_mb_count_size[2][7] = {{0}};
3002     char buf[200];
3003     int b_print_pcm = h->stat.i_mb_count[SLICE_TYPE_I][I_PCM]
3004                    || h->stat.i_mb_count[SLICE_TYPE_P][I_PCM]
3005                    || h->stat.i_mb_count[SLICE_TYPE_B][I_PCM];
3006
3007     x264_lookahead_delete( h );
3008
3009     if( h->param.i_threads > 1 )
3010         x264_threadpool_delete( h->threadpool );
3011     if( h->i_thread_frames > 1 )
3012     {
3013         for( int i = 0; i < h->i_thread_frames; i++ )
3014             if( h->thread[i]->b_thread_active )
3015             {
3016                 assert( h->thread[i]->fenc->i_reference_count == 1 );
3017                 x264_frame_delete( h->thread[i]->fenc );
3018             }
3019
3020         x264_t *thread_prev = h->thread[h->i_thread_phase];
3021         x264_thread_sync_ratecontrol( h, thread_prev, h );
3022         x264_thread_sync_ratecontrol( thread_prev, thread_prev, h );
3023         h->i_frame = thread_prev->i_frame + 1 - h->i_thread_frames;
3024     }
3025     h->i_frame++;
3026
3027     /* Slices used and PSNR */
3028     for( int i = 0; i < 3; i++ )
3029     {
3030         static const uint8_t slice_order[] = { SLICE_TYPE_I, SLICE_TYPE_P, SLICE_TYPE_B };
3031         int i_slice = slice_order[i];
3032
3033         if( h->stat.i_frame_count[i_slice] > 0 )
3034         {
3035             int i_count = h->stat.i_frame_count[i_slice];
3036             double dur =  h->stat.f_frame_duration[i_slice];
3037             if( h->param.analyse.b_psnr )
3038             {
3039                 x264_log( h, X264_LOG_INFO,
3040                           "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",
3041                           slice_type_to_char[i_slice],
3042                           i_count,
3043                           h->stat.f_frame_qp[i_slice] / i_count,
3044                           (double)h->stat.i_frame_size[i_slice] / i_count,
3045                           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,
3046                           h->stat.f_psnr_average[i_slice] / dur,
3047                           x264_psnr( h->stat.f_ssd_global[i_slice], dur * i_yuv_size ) );
3048             }
3049             else
3050             {
3051                 x264_log( h, X264_LOG_INFO,
3052                           "frame %c:%-5d Avg QP:%5.2f  size:%6.0f\n",
3053                           slice_type_to_char[i_slice],
3054                           i_count,
3055                           h->stat.f_frame_qp[i_slice] / i_count,
3056                           (double)h->stat.i_frame_size[i_slice] / i_count );
3057             }
3058         }
3059     }
3060     if( h->param.i_bframe && h->stat.i_frame_count[SLICE_TYPE_P] )
3061     {
3062         char *p = buf;
3063         int den = 0;
3064         // weight by number of frames (including the P-frame) that are in a sequence of N B-frames
3065         for( int i = 0; i <= h->param.i_bframe; i++ )
3066             den += (i+1) * h->stat.i_consecutive_bframes[i];
3067         for( int i = 0; i <= h->param.i_bframe; i++ )
3068             p += sprintf( p, " %4.1f%%", 100. * (i+1) * h->stat.i_consecutive_bframes[i] / den );
3069         x264_log( h, X264_LOG_INFO, "consecutive B-frames:%s\n", buf );
3070     }
3071
3072     for( int i_type = 0; i_type < 2; i_type++ )
3073         for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
3074         {
3075             if( i == D_DIRECT_8x8 ) continue; /* direct is counted as its own type */
3076             i_mb_count_size[i_type][x264_mb_partition_pixel_table[i]] += h->stat.i_mb_partition[i_type][i];
3077         }
3078
3079     /* MB types used */
3080     if( h->stat.i_frame_count[SLICE_TYPE_I] > 0 )
3081     {
3082         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_I];
3083         double i_count = h->stat.i_frame_count[SLICE_TYPE_I] * h->mb.i_mb_count / 100.0;
3084         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
3085         x264_log( h, X264_LOG_INFO, "mb I  %s\n", buf );
3086     }
3087     if( h->stat.i_frame_count[SLICE_TYPE_P] > 0 )
3088     {
3089         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_P];
3090         double i_count = h->stat.i_frame_count[SLICE_TYPE_P] * h->mb.i_mb_count / 100.0;
3091         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_P];
3092         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
3093         x264_log( h, X264_LOG_INFO,
3094                   "mb P  %s  P16..4: %4.1f%% %4.1f%% %4.1f%% %4.1f%% %4.1f%%    skip:%4.1f%%\n",
3095                   buf,
3096                   i_mb_size[PIXEL_16x16] / (i_count*4),
3097                   (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
3098                   i_mb_size[PIXEL_8x8] / (i_count*4),
3099                   (i_mb_size[PIXEL_8x4] + i_mb_size[PIXEL_4x8]) / (i_count*4),
3100                   i_mb_size[PIXEL_4x4] / (i_count*4),
3101                   i_mb_count[P_SKIP] / i_count );
3102     }
3103     if( h->stat.i_frame_count[SLICE_TYPE_B] > 0 )
3104     {
3105         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_B];
3106         double i_count = h->stat.i_frame_count[SLICE_TYPE_B] * h->mb.i_mb_count / 100.0;
3107         double i_mb_list_count;
3108         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_B];
3109         int64_t list_count[3] = {0}; /* 0 == L0, 1 == L1, 2 == BI */
3110         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
3111         for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
3112             for( int j = 0; j < 2; j++ )
3113             {
3114                 int l0 = x264_mb_type_list_table[i][0][j];
3115                 int l1 = x264_mb_type_list_table[i][1][j];
3116                 if( l0 || l1 )
3117                     list_count[l1+l0*l1] += h->stat.i_mb_count[SLICE_TYPE_B][i] * 2;
3118             }
3119         list_count[0] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L0_8x8];
3120         list_count[1] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L1_8x8];
3121         list_count[2] += h->stat.i_mb_partition[SLICE_TYPE_B][D_BI_8x8];
3122         i_mb_count[B_DIRECT] += (h->stat.i_mb_partition[SLICE_TYPE_B][D_DIRECT_8x8]+2)/4;
3123         i_mb_list_count = (list_count[0] + list_count[1] + list_count[2]) / 100.0;
3124         x264_log( h, X264_LOG_INFO,
3125                   "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",
3126                   buf,
3127                   i_mb_size[PIXEL_16x16] / (i_count*4),
3128                   (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
3129                   i_mb_size[PIXEL_8x8] / (i_count*4),
3130                   i_mb_count[B_DIRECT] / i_count,
3131                   i_mb_count[B_SKIP]   / i_count,
3132                   list_count[0] / i_mb_list_count,
3133                   list_count[1] / i_mb_list_count,
3134                   list_count[2] / i_mb_list_count );
3135     }
3136
3137     x264_ratecontrol_summary( h );
3138
3139     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 )
3140     {
3141 #define SUM3(p) (p[SLICE_TYPE_I] + p[SLICE_TYPE_P] + p[SLICE_TYPE_B])
3142 #define SUM3b(p,o) (p[SLICE_TYPE_I][o] + p[SLICE_TYPE_P][o] + p[SLICE_TYPE_B][o])
3143         int64_t i_i8x8 = SUM3b( h->stat.i_mb_count, I_8x8 );
3144         int64_t i_intra = i_i8x8 + SUM3b( h->stat.i_mb_count, I_4x4 )
3145                                  + SUM3b( h->stat.i_mb_count, I_16x16 );
3146         int64_t i_all_intra = i_intra + SUM3b( h->stat.i_mb_count, I_PCM);
3147         const int i_count = h->stat.i_frame_count[SLICE_TYPE_I] +
3148                             h->stat.i_frame_count[SLICE_TYPE_P] +
3149                             h->stat.i_frame_count[SLICE_TYPE_B];
3150         const double duration = h->stat.f_frame_duration[SLICE_TYPE_I] +
3151                                 h->stat.f_frame_duration[SLICE_TYPE_P] +
3152                                 h->stat.f_frame_duration[SLICE_TYPE_B];
3153         int64_t i_mb_count = (int64_t)i_count * h->mb.i_mb_count;
3154         float f_bitrate = SUM3(h->stat.i_frame_size) / duration / 125;
3155
3156         if( h->pps->b_transform_8x8_mode )
3157         {
3158             buf[0] = 0;
3159             if( h->stat.i_mb_count_8x8dct[0] )
3160                 sprintf( buf, " inter:%.1f%%", 100. * h->stat.i_mb_count_8x8dct[1] / h->stat.i_mb_count_8x8dct[0] );
3161             x264_log( h, X264_LOG_INFO, "8x8 transform intra:%.1f%%%s\n", 100. * i_i8x8 / i_intra, buf );
3162         }
3163
3164         if( (h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO ||
3165             (h->stat.i_direct_frames[0] && h->stat.i_direct_frames[1]))
3166             && h->stat.i_frame_count[SLICE_TYPE_B] )
3167         {
3168             x264_log( h, X264_LOG_INFO, "direct mvs  spatial:%.1f%% temporal:%.1f%%\n",
3169                       h->stat.i_direct_frames[1] * 100. / h->stat.i_frame_count[SLICE_TYPE_B],
3170                       h->stat.i_direct_frames[0] * 100. / h->stat.i_frame_count[SLICE_TYPE_B] );
3171         }
3172
3173         buf[0] = 0;
3174         if( i_mb_count != i_all_intra )
3175             sprintf( buf, " inter: %.1f%% %.1f%% %.1f%%",
3176                      h->stat.i_mb_cbp[1] * 100.0 / ((i_mb_count - i_all_intra)*4),
3177                      h->stat.i_mb_cbp[3] * 100.0 / ((i_mb_count - i_all_intra)  ),
3178                      h->stat.i_mb_cbp[5] * 100.0 / ((i_mb_count - i_all_intra)) );
3179         x264_log( h, X264_LOG_INFO, "coded y,uvDC,uvAC intra: %.1f%% %.1f%% %.1f%%%s\n",
3180                   h->stat.i_mb_cbp[0] * 100.0 / (i_all_intra*4),
3181                   h->stat.i_mb_cbp[2] * 100.0 / (i_all_intra  ),
3182                   h->stat.i_mb_cbp[4] * 100.0 / (i_all_intra  ), buf );
3183
3184         int64_t fixed_pred_modes[4][9] = {{0}};
3185         int64_t sum_pred_modes[4] = {0};
3186         for( int i = 0; i <= I_PRED_16x16_DC_128; i++ )
3187         {
3188             fixed_pred_modes[0][x264_mb_pred_mode16x16_fix[i]] += h->stat.i_mb_pred_mode[0][i];
3189             sum_pred_modes[0] += h->stat.i_mb_pred_mode[0][i];
3190         }
3191         if( sum_pred_modes[0] )
3192             x264_log( h, X264_LOG_INFO, "i16 v,h,dc,p: %2.0f%% %2.0f%% %2.0f%% %2.0f%%\n",
3193                       fixed_pred_modes[0][0] * 100.0 / sum_pred_modes[0],
3194                       fixed_pred_modes[0][1] * 100.0 / sum_pred_modes[0],
3195                       fixed_pred_modes[0][2] * 100.0 / sum_pred_modes[0],
3196                       fixed_pred_modes[0][3] * 100.0 / sum_pred_modes[0] );
3197         for( int i = 1; i <= 2; i++ )
3198         {
3199             for( int j = 0; j <= I_PRED_8x8_DC_128; j++ )
3200             {
3201                 fixed_pred_modes[i][x264_mb_pred_mode4x4_fix(j)] += h->stat.i_mb_pred_mode[i][j];
3202                 sum_pred_modes[i] += h->stat.i_mb_pred_mode[i][j];
3203             }
3204             if( sum_pred_modes[i] )
3205                 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,
3206                           fixed_pred_modes[i][0] * 100.0 / sum_pred_modes[i],
3207                           fixed_pred_modes[i][1] * 100.0 / sum_pred_modes[i],
3208                           fixed_pred_modes[i][2] * 100.0 / sum_pred_modes[i],
3209                           fixed_pred_modes[i][3] * 100.0 / sum_pred_modes[i],
3210                           fixed_pred_modes[i][4] * 100.0 / sum_pred_modes[i],
3211                           fixed_pred_modes[i][5] * 100.0 / sum_pred_modes[i],
3212                           fixed_pred_modes[i][6] * 100.0 / sum_pred_modes[i],
3213                           fixed_pred_modes[i][7] * 100.0 / sum_pred_modes[i],
3214                           fixed_pred_modes[i][8] * 100.0 / sum_pred_modes[i] );
3215         }
3216         for( int i = 0; i <= I_PRED_CHROMA_DC_128; i++ )
3217         {
3218             fixed_pred_modes[3][x264_mb_pred_mode8x8c_fix[i]] += h->stat.i_mb_pred_mode[3][i];
3219             sum_pred_modes[3] += h->stat.i_mb_pred_mode[3][i];
3220         }
3221         if( sum_pred_modes[3] )
3222             x264_log( h, X264_LOG_INFO, "i8c dc,h,v,p: %2.0f%% %2.0f%% %2.0f%% %2.0f%%\n",
3223                       fixed_pred_modes[3][0] * 100.0 / sum_pred_modes[3],
3224                       fixed_pred_modes[3][1] * 100.0 / sum_pred_modes[3],
3225                       fixed_pred_modes[3][2] * 100.0 / sum_pred_modes[3],
3226                       fixed_pred_modes[3][3] * 100.0 / sum_pred_modes[3] );
3227
3228         if( h->param.analyse.i_weighted_pred == X264_WEIGHTP_SMART && h->stat.i_frame_count[SLICE_TYPE_P] > 0 )
3229             x264_log( h, X264_LOG_INFO, "Weighted P-Frames: Y:%.1f%% UV:%.1f%%\n",
3230                       h->stat.i_wpred[0] * 100.0 / h->stat.i_frame_count[SLICE_TYPE_P],
3231                       h->stat.i_wpred[1] * 100.0 / h->stat.i_frame_count[SLICE_TYPE_P] );
3232
3233         for( int i_list = 0; i_list < 2; i_list++ )
3234             for( int i_slice = 0; i_slice < 2; i_slice++ )
3235             {
3236                 char *p = buf;
3237                 int64_t i_den = 0;
3238                 int i_max = 0;
3239                 for( int i = 0; i < X264_REF_MAX*2; i++ )
3240                     if( h->stat.i_mb_count_ref[i_slice][i_list][i] )
3241                     {
3242                         i_den += h->stat.i_mb_count_ref[i_slice][i_list][i];
3243                         i_max = i;
3244                     }
3245                 if( i_max == 0 )
3246                     continue;
3247                 for( int i = 0; i <= i_max; i++ )
3248                     p += sprintf( p, " %4.1f%%", 100. * h->stat.i_mb_count_ref[i_slice][i_list][i] / i_den );
3249                 x264_log( h, X264_LOG_INFO, "ref %c L%d:%s\n", "PB"[i_slice], i_list, buf );
3250             }
3251
3252         if( h->param.analyse.b_ssim )
3253         {
3254             float ssim = SUM3( h->stat.f_ssim_mean_y ) / duration;
3255             x264_log( h, X264_LOG_INFO, "SSIM Mean Y:%.7f (%6.3fdb)\n", ssim, x264_ssim( ssim ) );
3256         }
3257         if( h->param.analyse.b_psnr )
3258         {
3259             x264_log( h, X264_LOG_INFO,
3260                       "PSNR Mean Y:%6.3f U:%6.3f V:%6.3f Avg:%6.3f Global:%6.3f kb/s:%.2f\n",
3261                       SUM3( h->stat.f_psnr_mean_y ) / duration,
3262                       SUM3( h->stat.f_psnr_mean_u ) / duration,
3263                       SUM3( h->stat.f_psnr_mean_v ) / duration,
3264                       SUM3( h->stat.f_psnr_average ) / duration,
3265                       x264_psnr( SUM3( h->stat.f_ssd_global ), duration * i_yuv_size ),
3266                       f_bitrate );
3267         }
3268         else
3269             x264_log( h, X264_LOG_INFO, "kb/s:%.2f\n", f_bitrate );
3270     }
3271
3272     /* rc */
3273     x264_ratecontrol_delete( h );
3274
3275     /* param */
3276     if( h->param.rc.psz_stat_out )
3277         free( h->param.rc.psz_stat_out );
3278     if( h->param.rc.psz_stat_in )
3279         free( h->param.rc.psz_stat_in );
3280
3281     x264_cqm_delete( h );
3282     x264_free( h->nal_buffer );
3283     x264_analyse_free_costs( h );
3284
3285     if( h->i_thread_frames > 1)
3286         h = h->thread[h->i_thread_phase];
3287
3288     /* frames */
3289     x264_frame_delete_list( h->frames.unused[0] );
3290     x264_frame_delete_list( h->frames.unused[1] );
3291     x264_frame_delete_list( h->frames.current );
3292     x264_frame_delete_list( h->frames.blank_unused );
3293
3294     h = h->thread[0];
3295
3296     for( int i = h->param.i_threads - 1; i >= 0; i-- )
3297     {
3298         x264_frame_t **frame;
3299
3300         if( !h->param.b_sliced_threads || i == 0 )
3301         {
3302             for( frame = h->thread[i]->frames.reference; *frame; frame++ )
3303             {
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             }
3309             frame = &h->thread[i]->fdec;
3310             assert( (*frame)->i_reference_count > 0 );
3311             (*frame)->i_reference_count--;
3312             if( (*frame)->i_reference_count == 0 )
3313                 x264_frame_delete( *frame );
3314             x264_macroblock_cache_free( h->thread[i] );
3315         }
3316         x264_macroblock_thread_free( h->thread[i], 0 );
3317         x264_free( h->thread[i]->out.p_bitstream );
3318         x264_free( h->thread[i]->out.nal);
3319         x264_free( h->thread[i] );
3320     }
3321 }
3322
3323 /****************************************************************************
3324  * x264_encoder_delayed_frames:
3325  ****************************************************************************/
3326 int x264_encoder_delayed_frames( x264_t *h )
3327 {
3328     int delayed_frames = 0;
3329     if( h->i_thread_frames > 1 )
3330     {
3331         for( int i = 0; i < h->i_thread_frames; i++ )
3332             delayed_frames += h->thread[i]->b_thread_active;
3333         h = h->thread[h->i_thread_phase];
3334     }
3335     for( int i = 0; h->frames.current[i]; i++ )
3336         delayed_frames++;
3337     x264_pthread_mutex_lock( &h->lookahead->ofbuf.mutex );
3338     x264_pthread_mutex_lock( &h->lookahead->ifbuf.mutex );
3339     x264_pthread_mutex_lock( &h->lookahead->next.mutex );
3340     delayed_frames += h->lookahead->ifbuf.i_size + h->lookahead->next.i_size + h->lookahead->ofbuf.i_size;
3341     x264_pthread_mutex_unlock( &h->lookahead->next.mutex );
3342     x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
3343     x264_pthread_mutex_unlock( &h->lookahead->ofbuf.mutex );
3344     return delayed_frames;
3345 }