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

   1 /*****************************************************************************
   2  * x264: h264 encoder
   3  *****************************************************************************
   4  * Copyright (C) 2003-2008 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
  25 #include <math.h>
  26
  27 #include "common/common.h"
  28
  29 #include "set.h"
  30 #include "analyse.h"
  31 #include "ratecontrol.h"
  32 #include "macroblock.h"
  33 #include "me.h"
  34
  35 #if HAVE_VISUALIZE
  36 #include "common/visualize.h"
  37 #endif
  38
  39 //#define DEBUG_MB_TYPE
  40
  41 #define bs_write_ue bs_write_ue_big
  42
  43 static int x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
  44                                    x264_nal_t **pp_nal, int *pi_nal,
  45                                    x264_picture_t *pic_out );
  46
  47 /****************************************************************************
  48  *
  49  ******************************* x264 libs **********************************
  50  *
  51  ****************************************************************************/
  52 static float x264_psnr( int64_t i_sqe, int64_t i_size )
  53 {
  54     double f_mse = (double)i_sqe / (PIXEL_MAX*PIXEL_MAX * (double)i_size);
  55     if( f_mse <= 0.0000000001 ) /* Max 100dB */
  56         return 100;
  57
  58     return -10.0 * log10( f_mse );
  59 }
  60
  61 static float x264_ssim( float ssim )
  62 {
  63     return -10.0 * log10( 1 - ssim );
  64 }
  65
  66 static void x264_frame_dump( x264_t *h )
  67 {
  68     FILE *f = fopen( h->param.psz_dump_yuv, "r+b" );
  69     if( !f )
  70         return;
  71     /* Write the frame in display order */
  72     fseek( f, (uint64_t)h->fdec->i_frame * h->param.i_height * h->param.i_width * 3/2 * sizeof(pixel), SEEK_SET );
  73     for( int y = 0; y < h->param.i_height; y++ )
  74         fwrite( &h->fdec->plane[0][y*h->fdec->i_stride[0]], sizeof(pixel), h->param.i_width, f );
  75     int cw = h->param.i_width>>1;
  76     int ch = h->param.i_height>>1;
  77     pixel *planeu = x264_malloc( cw*ch*2*sizeof(pixel) );
  78     pixel *planev = planeu + cw*ch;
  79     h->mc.plane_copy_deinterleave( planeu, cw, planev, cw, h->fdec->plane[1], h->fdec->i_stride[1], cw, ch );
  80     fwrite( planeu, 1, cw*ch*2*sizeof(pixel), f );
  81     x264_free( planeu );
  82     fclose( f );
  83 }
  84
  85
  86 /* Fill "default" values */
  87 static void x264_slice_header_init( x264_t *h, x264_slice_header_t *sh,
  88                                     x264_sps_t *sps, x264_pps_t *pps,
  89                                     int i_idr_pic_id, int i_frame, int i_qp )
  90 {
  91     x264_param_t *param = &h->param;
  92
  93     /* First we fill all fields */
  94     sh->sps = sps;
  95     sh->pps = pps;
  96
  97     sh->i_first_mb  = 0;
  98     sh->i_last_mb   = h->mb.i_mb_count - 1;
  99     sh->i_pps_id    = pps->i_id;
 100
 101     sh->i_frame_num = i_frame;
 102
 103     sh->b_mbaff = h->param.b_interlaced;
 104     sh->b_field_pic = 0;    /* no field support for now */
 105     sh->b_bottom_field = 0; /* not yet used */
 106
 107     sh->i_idr_pic_id = i_idr_pic_id;
 108
 109     /* poc stuff, fixed later */
 110     sh->i_poc = 0;
 111     sh->i_delta_poc_bottom = 0;
 112     sh->i_delta_poc[0] = 0;
 113     sh->i_delta_poc[1] = 0;
 114
 115     sh->i_redundant_pic_cnt = 0;
 116
 117     h->mb.b_direct_auto_write = h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO
 118                                 && h->param.i_bframe
 119                                 && ( h->param.rc.b_stat_write || !h->param.rc.b_stat_read );
 120
 121     if( !h->mb.b_direct_auto_read && sh->i_type == SLICE_TYPE_B )
 122     {
 123         if( h->fref1[0]->i_poc_l0ref0 == h->fref0[0]->i_poc )
 124         {
 125             if( h->mb.b_direct_auto_write )
 126                 sh->b_direct_spatial_mv_pred = ( h->stat.i_direct_score[1] > h->stat.i_direct_score[0] );
 127             else
 128                 sh->b_direct_spatial_mv_pred = ( param->analyse.i_direct_mv_pred == X264_DIRECT_PRED_SPATIAL );
 129         }
 130         else
 131         {
 132             h->mb.b_direct_auto_write = 0;
 133             sh->b_direct_spatial_mv_pred = 1;
 134         }
 135     }
 136     /* else b_direct_spatial_mv_pred was read from the 2pass statsfile */
 137
 138     sh->b_num_ref_idx_override = 0;
 139     sh->i_num_ref_idx_l0_active = 1;
 140     sh->i_num_ref_idx_l1_active = 1;
 141
 142     sh->b_ref_pic_list_reordering_l0 = h->b_ref_reorder[0];
 143     sh->b_ref_pic_list_reordering_l1 = h->b_ref_reorder[1];
 144
 145     /* If the ref list isn't in the default order, construct reordering header */
 146     /* List1 reordering isn't needed yet */
 147     if( sh->b_ref_pic_list_reordering_l0 )
 148     {
 149         int pred_frame_num = i_frame;
 150         for( int i = 0; i < h->i_ref0; i++ )
 151         {
 152             int diff = h->fref0[i]->i_frame_num - pred_frame_num;
 153             sh->ref_pic_list_order[0][i].idc = ( diff > 0 );
 154             sh->ref_pic_list_order[0][i].arg = (abs(diff) - 1) & ((1 << sps->i_log2_max_frame_num) - 1);
 155             pred_frame_num = h->fref0[i]->i_frame_num;
 156         }
 157     }
 158
 159     sh->i_cabac_init_idc = param->i_cabac_init_idc;
 160
 161     sh->i_qp = i_qp;
 162     sh->i_qp_delta = i_qp - pps->i_pic_init_qp;
 163     sh->b_sp_for_swidth = 0;
 164     sh->i_qs_delta = 0;
 165
 166     int deblock_thresh = i_qp + 2 * X264_MIN(param->i_deblocking_filter_alphac0, param->i_deblocking_filter_beta);
 167     /* If effective qp <= 15, deblocking would have no effect anyway */
 168     if( param->b_deblocking_filter && (h->mb.b_variable_qp || 15 < deblock_thresh ) )
 169         sh->i_disable_deblocking_filter_idc = param->b_sliced_threads ? 2 : 0;
 170     else
 171         sh->i_disable_deblocking_filter_idc = 1;
 172     sh->i_alpha_c0_offset = param->i_deblocking_filter_alphac0 << 1;
 173     sh->i_beta_offset = param->i_deblocking_filter_beta << 1;
 174 }
 175
 176 static void x264_slice_header_write( bs_t *s, x264_slice_header_t *sh, int i_nal_ref_idc )
 177 {
 178     if( sh->b_mbaff )
 179     {
 180         assert( sh->i_first_mb % (2*sh->sps->i_mb_width) == 0 );
 181         bs_write_ue( s, sh->i_first_mb >> 1 );
 182     }
 183     else
 184         bs_write_ue( s, sh->i_first_mb );
 185
 186     bs_write_ue( s, sh->i_type + 5 );   /* same type things */
 187     bs_write_ue( s, sh->i_pps_id );
 188     bs_write( s, sh->sps->i_log2_max_frame_num, sh->i_frame_num & ((1<<sh->sps->i_log2_max_frame_num)-1) );
 189
 190     if( !sh->sps->b_frame_mbs_only )
 191     {
 192         bs_write1( s, sh->b_field_pic );
 193         if( sh->b_field_pic )
 194             bs_write1( s, sh->b_bottom_field );
 195     }
 196
 197     if( sh->i_idr_pic_id >= 0 ) /* NAL IDR */
 198         bs_write_ue( s, sh->i_idr_pic_id );
 199
 200     if( sh->sps->i_poc_type == 0 )
 201     {
 202         bs_write( s, sh->sps->i_log2_max_poc_lsb, sh->i_poc & ((1<<sh->sps->i_log2_max_poc_lsb)-1) );
 203         if( sh->pps->b_pic_order && !sh->b_field_pic )
 204             bs_write_se( s, sh->i_delta_poc_bottom );
 205     }
 206     else if( sh->sps->i_poc_type == 1 && !sh->sps->b_delta_pic_order_always_zero )
 207     {
 208         bs_write_se( s, sh->i_delta_poc[0] );
 209         if( sh->pps->b_pic_order && !sh->b_field_pic )
 210             bs_write_se( s, sh->i_delta_poc[1] );
 211     }
 212
 213     if( sh->pps->b_redundant_pic_cnt )
 214         bs_write_ue( s, sh->i_redundant_pic_cnt );
 215
 216     if( sh->i_type == SLICE_TYPE_B )
 217         bs_write1( s, sh->b_direct_spatial_mv_pred );
 218
 219     if( sh->i_type == SLICE_TYPE_P || sh->i_type == SLICE_TYPE_B )
 220     {
 221         bs_write1( s, sh->b_num_ref_idx_override );
 222         if( sh->b_num_ref_idx_override )
 223         {
 224             bs_write_ue( s, sh->i_num_ref_idx_l0_active - 1 );
 225             if( sh->i_type == SLICE_TYPE_B )
 226                 bs_write_ue( s, sh->i_num_ref_idx_l1_active - 1 );
 227         }
 228     }
 229
 230     /* ref pic list reordering */
 231     if( sh->i_type != SLICE_TYPE_I )
 232     {
 233         bs_write1( s, sh->b_ref_pic_list_reordering_l0 );
 234         if( sh->b_ref_pic_list_reordering_l0 )
 235         {
 236             for( int i = 0; i < sh->i_num_ref_idx_l0_active; i++ )
 237             {
 238                 bs_write_ue( s, sh->ref_pic_list_order[0][i].idc );
 239                 bs_write_ue( s, sh->ref_pic_list_order[0][i].arg );
 240             }
 241             bs_write_ue( s, 3 );
 242         }
 243     }
 244     if( sh->i_type == SLICE_TYPE_B )
 245     {
 246         bs_write1( s, sh->b_ref_pic_list_reordering_l1 );
 247         if( sh->b_ref_pic_list_reordering_l1 )
 248         {
 249             for( int i = 0; i < sh->i_num_ref_idx_l1_active; i++ )
 250             {
 251                 bs_write_ue( s, sh->ref_pic_list_order[1][i].idc );
 252                 bs_write_ue( s, sh->ref_pic_list_order[1][i].arg );
 253             }
 254             bs_write_ue( s, 3 );
 255         }
 256     }
 257
 258     if( sh->pps->b_weighted_pred && sh->i_type == SLICE_TYPE_P )
 259     {
 260         /* pred_weight_table() */
 261         bs_write_ue( s, sh->weight[0][0].i_denom );
 262         bs_write_ue( s, sh->weight[0][1].i_denom );
 263         for( int i = 0; i < sh->i_num_ref_idx_l0_active; i++ )
 264         {
 265             int luma_weight_l0_flag = !!sh->weight[i][0].weightfn;
 266             int chroma_weight_l0_flag = !!sh->weight[i][1].weightfn || !!sh->weight[i][2].weightfn;
 267             bs_write1( s, luma_weight_l0_flag );
 268             if( luma_weight_l0_flag )
 269             {
 270                 bs_write_se( s, sh->weight[i][0].i_scale );
 271                 bs_write_se( s, sh->weight[i][0].i_offset );
 272             }
 273             bs_write1( s, chroma_weight_l0_flag );
 274             if( chroma_weight_l0_flag )
 275             {
 276                 for( int j = 1; j < 3; j++ )
 277                 {
 278                     bs_write_se( s, sh->weight[i][j].i_scale );
 279                     bs_write_se( s, sh->weight[i][j].i_offset );
 280                 }
 281             }
 282         }
 283     }
 284     else if( sh->pps->b_weighted_bipred == 1 && sh->i_type == SLICE_TYPE_B )
 285     {
 286       /* TODO */
 287     }
 288
 289     if( i_nal_ref_idc != 0 )
 290     {
 291         if( sh->i_idr_pic_id >= 0 )
 292         {
 293             bs_write1( s, 0 );  /* no output of prior pics flag */
 294             bs_write1( s, 0 );  /* long term reference flag */
 295         }
 296         else
 297         {
 298             bs_write1( s, sh->i_mmco_command_count > 0 ); /* adaptive_ref_pic_marking_mode_flag */
 299             if( sh->i_mmco_command_count > 0 )
 300             {
 301                 for( int i = 0; i < sh->i_mmco_command_count; i++ )
 302                 {
 303                     bs_write_ue( s, 1 ); /* mark short term ref as unused */
 304                     bs_write_ue( s, sh->mmco[i].i_difference_of_pic_nums - 1 );
 305                 }
 306                 bs_write_ue( s, 0 ); /* end command list */
 307             }
 308         }
 309     }
 310
 311     if( sh->pps->b_cabac && sh->i_type != SLICE_TYPE_I )
 312         bs_write_ue( s, sh->i_cabac_init_idc );
 313
 314     bs_write_se( s, sh->i_qp_delta );      /* slice qp delta */
 315
 316     if( sh->pps->b_deblocking_filter_control )
 317     {
 318         bs_write_ue( s, sh->i_disable_deblocking_filter_idc );
 319         if( sh->i_disable_deblocking_filter_idc != 1 )
 320         {
 321             bs_write_se( s, sh->i_alpha_c0_offset >> 1 );
 322             bs_write_se( s, sh->i_beta_offset >> 1 );
 323         }
 324     }
 325 }
 326
 327 /* If we are within a reasonable distance of the end of the memory allocated for the bitstream, */
 328 /* reallocate, adding an arbitrary amount of space (100 kilobytes). */
 329 static int x264_bitstream_check_buffer( x264_t *h )
 330 {
 331     uint8_t *bs_bak = h->out.p_bitstream;
 332     if( (h->param.b_cabac && (h->cabac.p_end - h->cabac.p < 2500)) ||
 333         (h->out.bs.p_end - h->out.bs.p < 2500) )
 334     {
 335         h->out.i_bitstream += 100000;
 336         CHECKED_MALLOC( h->out.p_bitstream, h->out.i_bitstream );
 337         h->mc.memcpy_aligned( h->out.p_bitstream, bs_bak, (h->out.i_bitstream - 100000) & ~15 );
 338         intptr_t delta = h->out.p_bitstream - bs_bak;
 339
 340         h->out.bs.p_start += delta;
 341         h->out.bs.p += delta;
 342         h->out.bs.p_end = h->out.p_bitstream + h->out.i_bitstream;
 343
 344         h->cabac.p_start += delta;
 345         h->cabac.p += delta;
 346         h->cabac.p_end = h->out.p_bitstream + h->out.i_bitstream;
 347
 348         for( int i = 0; i <= h->out.i_nal; i++ )
 349             h->out.nal[i].p_payload += delta;
 350         x264_free( bs_bak );
 351     }
 352     return 0;
 353 fail:
 354     x264_free( bs_bak );
 355     return -1;
 356 }
 357
 358 #if HAVE_PTHREAD
 359 static void x264_encoder_thread_init( x264_t *h )
 360 {
 361     if( h->param.i_sync_lookahead )
 362         x264_lower_thread_priority( 10 );
 363
 364 #if HAVE_MMX
 365     /* Misalign mask has to be set separately for each thread. */
 366     if( h->param.cpu&X264_CPU_SSE_MISALIGN )
 367         x264_cpu_mask_misalign_sse();
 368 #endif
 369 }
 370 #endif
 371
 372 /****************************************************************************
 373  *
 374  ****************************************************************************
 375  ****************************** External API*********************************
 376  ****************************************************************************
 377  *
 378  ****************************************************************************/
 379
 380 static int x264_validate_parameters( x264_t *h )
 381 {
 382 #if HAVE_MMX
 383 #ifdef __SSE__
 384     if( !(x264_cpu_detect() & X264_CPU_SSE) )
 385     {
 386         x264_log( h, X264_LOG_ERROR, "your cpu does not support SSE1, but x264 was compiled with asm support\n");
 387 #else
 388     if( !(x264_cpu_detect() & X264_CPU_MMXEXT) )
 389     {
 390         x264_log( h, X264_LOG_ERROR, "your cpu does not support MMXEXT, but x264 was compiled with asm support\n");
 391 #endif
 392         x264_log( h, X264_LOG_ERROR, "to run x264, recompile without asm support (configure --disable-asm)\n");
 393         return -1;
 394     }
 395 #endif
 396     if( h->param.i_width <= 0 || h->param.i_height <= 0 )
 397     {
 398         x264_log( h, X264_LOG_ERROR, "invalid width x height (%dx%d)\n",
 399                   h->param.i_width, h->param.i_height );
 400         return -1;
 401     }
 402
 403     if( h->param.i_width % 2 || h->param.i_height % 2 )
 404     {
 405         x264_log( h, X264_LOG_ERROR, "width or height not divisible by 2 (%dx%d)\n",
 406                   h->param.i_width, h->param.i_height );
 407         return -1;
 408     }
 409     int i_csp = h->param.i_csp & X264_CSP_MASK;
 410     if( i_csp <= X264_CSP_NONE || i_csp >= X264_CSP_MAX )
 411     {
 412         x264_log( h, X264_LOG_ERROR, "invalid CSP (only I420/YV12/NV12 supported)\n" );
 413         return -1;
 414     }
 415
 416     if( h->param.i_threads == X264_THREADS_AUTO )
 417         h->param.i_threads = x264_cpu_num_processors() * (h->param.b_sliced_threads?2:3)/2;
 418     h->param.i_threads = x264_clip3( h->param.i_threads, 1, X264_THREAD_MAX );
 419     if( h->param.i_threads > 1 )
 420     {
 421 #if !HAVE_PTHREAD
 422         x264_log( h, X264_LOG_WARNING, "not compiled with pthread support!\n");
 423         h->param.i_threads = 1;
 424 #endif
 425         /* Avoid absurdly small thread slices as they can reduce performance
 426          * and VBV compliance.  Capped at an arbitrary 4 rows per thread. */
 427         if( h->param.b_sliced_threads )
 428         {
 429             int max_threads = (h->param.i_height+15)/16 / 4;
 430             h->param.i_threads = X264_MIN( h->param.i_threads, max_threads );
 431         }
 432     }
 433     else
 434         h->param.b_sliced_threads = 0;
 435     h->i_thread_frames = h->param.b_sliced_threads ? 1 : h->param.i_threads;
 436     if( h->i_thread_frames > 1 )
 437         h->param.nalu_process = NULL;
 438
 439     if( h->param.b_interlaced )
 440     {
 441         if( h->param.analyse.i_me_method >= X264_ME_ESA )
 442         {
 443             x264_log( h, X264_LOG_WARNING, "interlace + me=esa is not implemented\n" );
 444             h->param.analyse.i_me_method = X264_ME_UMH;
 445         }
 446         if( h->param.analyse.i_weighted_pred > 0 )
 447         {
 448             x264_log( h, X264_LOG_WARNING, "interlace + weightp is not implemented\n" );
 449             h->param.analyse.i_weighted_pred = X264_WEIGHTP_NONE;
 450         }
 451     }
 452
 453     /* Detect default ffmpeg settings and terminate with an error. */
 454     {
 455         int score = 0;
 456         score += h->param.analyse.i_me_range == 0;
 457         score += h->param.rc.i_qp_step == 3;
 458         score += h->param.i_keyint_max == 12;
 459         score += h->param.rc.i_qp_min == 2;
 460         score += h->param.rc.i_qp_max == 31;
 461         score += h->param.rc.f_qcompress == 0.5;
 462         score += fabs(h->param.rc.f_ip_factor - 1.25) < 0.01;
 463         score += fabs(h->param.rc.f_pb_factor - 1.25) < 0.01;
 464         score += h->param.analyse.inter == 0 && h->param.analyse.i_subpel_refine == 8;
 465         if( score >= 5 )
 466         {
 467             x264_log( h, X264_LOG_ERROR, "broken ffmpeg default settings detected\n" );
 468             x264_log( h, X264_LOG_ERROR, "use an encoding preset (vpre)\n" );
 469             return -1;
 470         }
 471     }
 472
 473     if( h->param.rc.i_rc_method < 0 || h->param.rc.i_rc_method > 2 )
 474     {
 475         x264_log( h, X264_LOG_ERROR, "no ratecontrol method specified\n" );
 476         return -1;
 477     }
 478     h->param.rc.f_rf_constant = x264_clip3f( h->param.rc.f_rf_constant, 0, QP_MAX );
 479     h->param.rc.i_qp_constant = x264_clip3( h->param.rc.i_qp_constant, 0, QP_MAX );
 480     if( h->param.rc.i_rc_method == X264_RC_CRF )
 481     {
 482         h->param.rc.i_qp_constant = h->param.rc.f_rf_constant;
 483         h->param.rc.i_bitrate = 0;
 484     }
 485     if( (h->param.rc.i_rc_method == X264_RC_CQP || h->param.rc.i_rc_method == X264_RC_CRF)
 486         && h->param.rc.i_qp_constant == 0 )
 487     {
 488         h->mb.b_lossless = 1;
 489         h->param.i_cqm_preset = X264_CQM_FLAT;
 490         h->param.psz_cqm_file = NULL;
 491         h->param.rc.i_rc_method = X264_RC_CQP;
 492         h->param.rc.f_ip_factor = 1;
 493         h->param.rc.f_pb_factor = 1;
 494         h->param.analyse.b_psnr = 0;
 495         h->param.analyse.b_ssim = 0;
 496         h->param.analyse.i_chroma_qp_offset = 0;
 497         h->param.analyse.i_trellis = 0;
 498         h->param.analyse.b_fast_pskip = 0;
 499         h->param.analyse.i_noise_reduction = 0;
 500         h->param.analyse.b_psy = 0;
 501         h->param.i_bframe = 0;
 502         /* 8x8dct is not useful at all in CAVLC lossless */
 503         if( !h->param.b_cabac )
 504             h->param.analyse.b_transform_8x8 = 0;
 505     }
 506     if( h->param.rc.i_rc_method == X264_RC_CQP )
 507     {
 508         float qp_p = h->param.rc.i_qp_constant;
 509         float qp_i = qp_p - 6*log2f( h->param.rc.f_ip_factor );
 510         float qp_b = qp_p + 6*log2f( h->param.rc.f_pb_factor );
 511         h->param.rc.i_qp_min = x264_clip3( (int)(X264_MIN3( qp_p, qp_i, qp_b )), 0, QP_MAX );
 512         h->param.rc.i_qp_max = x264_clip3( (int)(X264_MAX3( qp_p, qp_i, qp_b ) + .999), 0, QP_MAX );
 513         h->param.rc.i_aq_mode = 0;
 514         h->param.rc.b_mb_tree = 0;
 515     }
 516     h->param.rc.i_qp_max = x264_clip3( h->param.rc.i_qp_max, 0, QP_MAX );
 517     h->param.rc.i_qp_min = x264_clip3( h->param.rc.i_qp_min, 0, h->param.rc.i_qp_max );
 518     if( h->param.rc.i_vbv_buffer_size )
 519     {
 520         if( h->param.rc.i_rc_method == X264_RC_CQP )
 521         {
 522             x264_log( h, X264_LOG_WARNING, "VBV is incompatible with constant QP, ignored.\n" );
 523             h->param.rc.i_vbv_max_bitrate = 0;
 524             h->param.rc.i_vbv_buffer_size = 0;
 525         }
 526         else if( h->param.rc.i_vbv_max_bitrate == 0 )
 527         {
 528             if( h->param.rc.i_rc_method == X264_RC_ABR )
 529             {
 530                 x264_log( h, X264_LOG_WARNING, "VBV maxrate unspecified, assuming CBR\n" );
 531                 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate;
 532             }
 533             else
 534             {
 535                 x264_log( h, X264_LOG_WARNING, "VBV bufsize set but maxrate unspecified, ignored\n" );
 536                 h->param.rc.i_vbv_buffer_size = 0;
 537             }
 538         }
 539         else if( h->param.rc.i_vbv_max_bitrate < h->param.rc.i_bitrate &&
 540                  h->param.rc.i_rc_method == X264_RC_ABR )
 541         {
 542             x264_log( h, X264_LOG_WARNING, "max bitrate less than average bitrate, assuming CBR\n" );
 543             h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate;
 544         }
 545     }
 546     else if( h->param.rc.i_vbv_max_bitrate )
 547     {
 548         x264_log( h, X264_LOG_WARNING, "VBV maxrate specified, but no bufsize, ignored\n" );
 549         h->param.rc.i_vbv_max_bitrate = 0;
 550     }
 551
 552     if( h->param.b_interlaced && h->param.i_slice_max_size )
 553     {
 554         x264_log( h, X264_LOG_WARNING, "interlaced + slice-max-size is not implemented\n" );
 555         h->param.i_slice_max_size = 0;
 556     }
 557     if( h->param.b_interlaced && h->param.i_slice_max_mbs )
 558     {
 559         x264_log( h, X264_LOG_WARNING, "interlaced + slice-max-mbs is not implemented\n" );
 560         h->param.i_slice_max_mbs = 0;
 561     }
 562     int max_slices = (h->param.i_height+((16<<h->param.b_interlaced)-1))/(16<<h->param.b_interlaced);
 563     if( h->param.b_sliced_threads )
 564         h->param.i_slice_count = x264_clip3( h->param.i_threads, 0, max_slices );
 565     else
 566     {
 567         h->param.i_slice_count = x264_clip3( h->param.i_slice_count, 0, max_slices );
 568         h->param.i_slice_max_size = X264_MAX( h->param.i_slice_max_size, 0 );
 569         h->param.i_slice_max_mbs = X264_MAX( h->param.i_slice_max_mbs, 0 );
 570         if( h->param.i_slice_max_mbs || h->param.i_slice_max_size )
 571             h->param.i_slice_count = 0;
 572     }
 573
 574     h->param.i_frame_reference = x264_clip3( h->param.i_frame_reference, 1, 16 );
 575     h->param.i_dpb_size = x264_clip3( h->param.i_dpb_size, 1, 16 );
 576     h->param.i_keyint_max = x264_clip3( h->param.i_keyint_max, 1, X264_KEYINT_MAX_INFINITE );
 577     if( h->param.i_scenecut_threshold < 0 )
 578         h->param.i_scenecut_threshold = 0;
 579     if( !h->param.analyse.i_subpel_refine && h->param.analyse.i_direct_mv_pred > X264_DIRECT_PRED_SPATIAL )
 580     {
 581         x264_log( h, X264_LOG_WARNING, "subme=0 + direct=temporal is not supported\n" );
 582         h->param.analyse.i_direct_mv_pred = X264_DIRECT_PRED_SPATIAL;
 583     }
 584     h->param.i_bframe = x264_clip3( h->param.i_bframe, 0, X264_MIN( X264_BFRAME_MAX, h->param.i_keyint_max-1 ) );
 585     h->param.i_open_gop = x264_clip3( h->param.i_open_gop, X264_OPEN_GOP_NONE, X264_OPEN_GOP_BLURAY );
 586     if( h->param.i_keyint_max == 1 )
 587         h->param.b_intra_refresh = 0;
 588     h->param.i_bframe_bias = x264_clip3( h->param.i_bframe_bias, -90, 100 );
 589     if( h->param.i_bframe <= 1 )
 590         h->param.i_bframe_pyramid = X264_B_PYRAMID_NONE;
 591     h->param.i_bframe_pyramid = x264_clip3( h->param.i_bframe_pyramid, X264_B_PYRAMID_NONE, X264_B_PYRAMID_NORMAL );
 592     if( !h->param.i_bframe )
 593     {
 594         h->param.i_bframe_adaptive = X264_B_ADAPT_NONE;
 595         h->param.analyse.i_direct_mv_pred = 0;
 596         h->param.analyse.b_weighted_bipred = 0;
 597         h->param.i_open_gop = X264_OPEN_GOP_NONE;
 598     }
 599     if( h->param.b_intra_refresh && h->param.i_bframe_pyramid == X264_B_PYRAMID_NORMAL )
 600     {
 601         x264_log( h, X264_LOG_WARNING, "b-pyramid normal + intra-refresh is not supported\n" );
 602         h->param.i_bframe_pyramid = X264_B_PYRAMID_STRICT;
 603     }
 604     if( h->param.b_intra_refresh && (h->param.i_frame_reference > 1 || h->param.i_dpb_size > 1) )
 605     {
 606         x264_log( h, X264_LOG_WARNING, "ref > 1 + intra-refresh is not supported\n" );
 607         h->param.i_frame_reference = 1;
 608         h->param.i_dpb_size = 1;
 609     }
 610     if( h->param.b_intra_refresh && h->param.i_open_gop )
 611     {
 612         x264_log( h, X264_LOG_WARNING, "intra-refresh is not compatible with open-gop\n" );
 613         h->param.i_open_gop = X264_OPEN_GOP_NONE;
 614     }
 615     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;
 616     if( h->param.i_keyint_min == X264_KEYINT_MIN_AUTO )
 617         h->param.i_keyint_min = X264_MIN( h->param.i_keyint_max / 10, fps );
 618     h->param.i_keyint_min = x264_clip3( h->param.i_keyint_min, 1, h->param.i_keyint_max/2+1 );
 619     h->param.rc.i_lookahead = x264_clip3( h->param.rc.i_lookahead, 0, X264_LOOKAHEAD_MAX );
 620     {
 621         int maxrate = X264_MAX( h->param.rc.i_vbv_max_bitrate, h->param.rc.i_bitrate );
 622         float bufsize = maxrate ? (float)h->param.rc.i_vbv_buffer_size / maxrate : 0;
 623         h->param.rc.i_lookahead = X264_MIN( h->param.rc.i_lookahead, X264_MAX( h->param.i_keyint_max, bufsize*fps ) );
 624     }
 625
 626     if( !h->param.i_timebase_num || !h->param.i_timebase_den )
 627     {
 628         h->param.i_timebase_num = h->param.i_fps_den;
 629         h->param.i_timebase_den = h->param.i_fps_num;
 630     }
 631
 632     h->param.rc.f_qcompress = x264_clip3f( h->param.rc.f_qcompress, 0.0, 1.0 );
 633     if( h->param.i_keyint_max == 1 || h->param.rc.f_qcompress == 1 )
 634         h->param.rc.b_mb_tree = 0;
 635     if( (!h->param.b_intra_refresh && h->param.i_keyint_max != X264_KEYINT_MAX_INFINITE) &&
 636         !h->param.rc.i_lookahead && h->param.rc.b_mb_tree )
 637     {
 638         x264_log( h, X264_LOG_WARNING, "lookaheadless mb-tree requires intra refresh or infinite keyint\n" );
 639         h->param.rc.b_mb_tree = 0;
 640     }
 641     if( h->param.rc.b_stat_read )
 642         h->param.rc.i_lookahead = 0;
 643 #if HAVE_PTHREAD
 644     if( h->param.i_sync_lookahead < 0 )
 645         h->param.i_sync_lookahead = h->param.i_bframe + 1;
 646     h->param.i_sync_lookahead = X264_MIN( h->param.i_sync_lookahead, X264_LOOKAHEAD_MAX );
 647     if( h->param.rc.b_stat_read || h->i_thread_frames == 1 )
 648         h->param.i_sync_lookahead = 0;
 649 #else
 650     h->param.i_sync_lookahead = 0;
 651 #endif
 652
 653     h->param.i_deblocking_filter_alphac0 = x264_clip3( h->param.i_deblocking_filter_alphac0, -6, 6 );
 654     h->param.i_deblocking_filter_beta    = x264_clip3( h->param.i_deblocking_filter_beta, -6, 6 );
 655     h->param.analyse.i_luma_deadzone[0] = x264_clip3( h->param.analyse.i_luma_deadzone[0], 0, 32 );
 656     h->param.analyse.i_luma_deadzone[1] = x264_clip3( h->param.analyse.i_luma_deadzone[1], 0, 32 );
 657
 658     h->param.i_cabac_init_idc = x264_clip3( h->param.i_cabac_init_idc, 0, 2 );
 659
 660     if( h->param.i_cqm_preset < X264_CQM_FLAT || h->param.i_cqm_preset > X264_CQM_CUSTOM )
 661         h->param.i_cqm_preset = X264_CQM_FLAT;
 662
 663     if( h->param.analyse.i_me_method < X264_ME_DIA ||
 664         h->param.analyse.i_me_method > X264_ME_TESA )
 665         h->param.analyse.i_me_method = X264_ME_HEX;
 666     if( h->param.analyse.i_me_range < 4 )
 667         h->param.analyse.i_me_range = 4;
 668     if( h->param.analyse.i_me_range > 16 && h->param.analyse.i_me_method <= X264_ME_HEX )
 669         h->param.analyse.i_me_range = 16;
 670     if( h->param.analyse.i_me_method == X264_ME_TESA &&
 671         (h->mb.b_lossless || h->param.analyse.i_subpel_refine <= 1) )
 672         h->param.analyse.i_me_method = X264_ME_ESA;
 673     h->param.analyse.i_subpel_refine = x264_clip3( h->param.analyse.i_subpel_refine, 0, 10 );
 674     h->param.analyse.b_mixed_references = h->param.analyse.b_mixed_references && h->param.i_frame_reference > 1;
 675     h->param.analyse.inter &= X264_ANALYSE_PSUB16x16|X264_ANALYSE_PSUB8x8|X264_ANALYSE_BSUB16x16|
 676                               X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
 677     h->param.analyse.intra &= X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
 678     if( !(h->param.analyse.inter & X264_ANALYSE_PSUB16x16) )
 679         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
 680     if( !h->param.analyse.b_transform_8x8 )
 681     {
 682         h->param.analyse.inter &= ~X264_ANALYSE_I8x8;
 683         h->param.analyse.intra &= ~X264_ANALYSE_I8x8;
 684     }
 685     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -12, 12);
 686     if( !h->param.b_cabac )
 687         h->param.analyse.i_trellis = 0;
 688     h->param.analyse.i_trellis = x264_clip3( h->param.analyse.i_trellis, 0, 2 );
 689     if( !h->param.analyse.b_psy )
 690     {
 691         h->param.analyse.f_psy_rd = 0;
 692         h->param.analyse.f_psy_trellis = 0;
 693     }
 694     if( !h->param.analyse.i_trellis )
 695         h->param.analyse.f_psy_trellis = 0;
 696     h->param.analyse.f_psy_rd = x264_clip3f( h->param.analyse.f_psy_rd, 0, 10 );
 697     h->param.analyse.f_psy_trellis = x264_clip3f( h->param.analyse.f_psy_trellis, 0, 10 );
 698     if( h->param.analyse.i_subpel_refine < 6 )
 699         h->param.analyse.f_psy_rd = 0;
 700     h->mb.i_psy_rd = FIX8( h->param.analyse.f_psy_rd );
 701     /* Psy RDO increases overall quantizers to improve the quality of luma--this indirectly hurts chroma quality */
 702     /* so we lower the chroma QP offset to compensate */
 703     /* This can be triggered repeatedly on multiple calls to parameter_validate, but since encoding
 704      * uses the pps chroma qp offset not the param chroma qp offset, this is not a problem. */
 705     if( h->mb.i_psy_rd )
 706         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_rd < 0.25 ? 1 : 2;
 707     h->mb.i_psy_trellis = FIX8( h->param.analyse.f_psy_trellis / 4 );
 708     /* Psy trellis has a similar effect. */
 709     if( h->mb.i_psy_trellis )
 710         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_trellis < 0.25 ? 1 : 2;
 711     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -12, 12);
 712     h->param.rc.i_aq_mode = x264_clip3( h->param.rc.i_aq_mode, 0, 2 );
 713     h->param.rc.f_aq_strength = x264_clip3f( h->param.rc.f_aq_strength, 0, 3 );
 714     if( h->param.rc.f_aq_strength == 0 )
 715         h->param.rc.i_aq_mode = 0;
 716     /* MB-tree requires AQ to be on, even if the strength is zero. */
 717     if( !h->param.rc.i_aq_mode && h->param.rc.b_mb_tree )
 718     {
 719         h->param.rc.i_aq_mode = 1;
 720         h->param.rc.f_aq_strength = 0;
 721     }
 722     h->param.analyse.i_noise_reduction = x264_clip3( h->param.analyse.i_noise_reduction, 0, 1<<16 );
 723     if( h->param.analyse.i_subpel_refine == 10 && (h->param.analyse.i_trellis != 2 || !h->param.rc.i_aq_mode) )
 724         h->param.analyse.i_subpel_refine = 9;
 725
 726     {
 727         const x264_level_t *l = x264_levels;
 728         if( h->param.i_level_idc < 0 )
 729         {
 730             int maxrate_bak = h->param.rc.i_vbv_max_bitrate;
 731             if( h->param.rc.i_rc_method == X264_RC_ABR && h->param.rc.i_vbv_buffer_size <= 0 )
 732                 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate * 2;
 733             h->sps = h->sps_array;
 734             x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
 735             do h->param.i_level_idc = l->level_idc;
 736                 while( l[1].level_idc && x264_validate_levels( h, 0 ) && l++ );
 737             h->param.rc.i_vbv_max_bitrate = maxrate_bak;
 738         }
 739         else
 740         {
 741             while( l->level_idc && l->level_idc != h->param.i_level_idc )
 742                 l++;
 743             if( l->level_idc == 0 )
 744             {
 745                 x264_log( h, X264_LOG_ERROR, "invalid level_idc: %d\n", h->param.i_level_idc );
 746                 return -1;
 747             }
 748         }
 749         if( h->param.analyse.i_mv_range <= 0 )
 750             h->param.analyse.i_mv_range = l->mv_range >> h->param.b_interlaced;
 751         else
 752             h->param.analyse.i_mv_range = x264_clip3(h->param.analyse.i_mv_range, 32, 512 >> h->param.b_interlaced);
 753     }
 754
 755     h->param.analyse.i_weighted_pred = x264_clip3( h->param.analyse.i_weighted_pred, 0, X264_WEIGHTP_SMART );
 756     if( !h->param.analyse.i_weighted_pred && h->param.rc.b_mb_tree && h->param.analyse.b_psy && !h->param.b_interlaced )
 757         h->param.analyse.i_weighted_pred = X264_WEIGHTP_FAKE;
 758
 759     if( h->i_thread_frames > 1 )
 760     {
 761         int r = h->param.analyse.i_mv_range_thread;
 762         int r2;
 763         if( r <= 0 )
 764         {
 765             // half of the available space is reserved and divided evenly among the threads,
 766             // the rest is allocated to whichever thread is far enough ahead to use it.
 767             // reserving more space increases quality for some videos, but costs more time
 768             // in thread synchronization.
 769             int max_range = (h->param.i_height + X264_THREAD_HEIGHT) / h->i_thread_frames - X264_THREAD_HEIGHT;
 770             r = max_range / 2;
 771         }
 772         r = X264_MAX( r, h->param.analyse.i_me_range );
 773         r = X264_MIN( r, h->param.analyse.i_mv_range );
 774         // round up to use the whole mb row
 775         r2 = (r & ~15) + ((-X264_THREAD_HEIGHT) & 15);
 776         if( r2 < r )
 777             r2 += 16;
 778         x264_log( h, X264_LOG_DEBUG, "using mv_range_thread = %d\n", r2 );
 779         h->param.analyse.i_mv_range_thread = r2;
 780     }
 781
 782     if( h->param.rc.f_qblur < 0 )
 783         h->param.rc.f_qblur = 0;
 784     if( h->param.rc.f_complexity_blur < 0 )
 785         h->param.rc.f_complexity_blur = 0;
 786
 787     h->param.i_sps_id &= 31;
 788
 789     if( h->param.i_log_level < X264_LOG_INFO )
 790     {
 791         h->param.analyse.b_psnr = 0;
 792         h->param.analyse.b_ssim = 0;
 793     }
 794
 795     if( h->param.b_interlaced )
 796         h->param.b_pic_struct = 1;
 797
 798     if( h->param.i_nal_hrd && !h->param.rc.i_vbv_buffer_size )
 799     {
 800         x264_log( h, X264_LOG_WARNING, "NAL HRD parameters require VBV parameters\n" );
 801         h->param.i_nal_hrd = X264_NAL_HRD_NONE;
 802     }
 803
 804     if( h->param.i_nal_hrd == X264_NAL_HRD_CBR &&
 805        (h->param.rc.i_bitrate != h->param.rc.i_vbv_max_bitrate || !h->param.rc.i_vbv_max_bitrate) )
 806     {
 807         x264_log( h, X264_LOG_WARNING, "CBR HRD requires constant bitrate\n" );
 808         h->param.i_nal_hrd = X264_NAL_HRD_VBR;
 809     }
 810
 811     /* ensure the booleans are 0 or 1 so they can be used in math */
 812 #define BOOLIFY(x) h->param.x = !!h->param.x
 813     BOOLIFY( b_cabac );
 814     BOOLIFY( b_constrained_intra );
 815     BOOLIFY( b_deblocking_filter );
 816     BOOLIFY( b_deterministic );
 817     BOOLIFY( b_sliced_threads );
 818     BOOLIFY( b_interlaced );
 819     BOOLIFY( b_intra_refresh );
 820     BOOLIFY( b_visualize );
 821     BOOLIFY( b_aud );
 822     BOOLIFY( b_repeat_headers );
 823     BOOLIFY( b_annexb );
 824     BOOLIFY( b_vfr_input );
 825     BOOLIFY( b_pic_struct );
 826     BOOLIFY( b_fake_interlaced );
 827     BOOLIFY( analyse.b_transform_8x8 );
 828     BOOLIFY( analyse.b_weighted_bipred );
 829     BOOLIFY( analyse.b_chroma_me );
 830     BOOLIFY( analyse.b_mixed_references );
 831     BOOLIFY( analyse.b_fast_pskip );
 832     BOOLIFY( analyse.b_dct_decimate );
 833     BOOLIFY( analyse.b_psy );
 834     BOOLIFY( analyse.b_psnr );
 835     BOOLIFY( analyse.b_ssim );
 836     BOOLIFY( rc.b_stat_write );
 837     BOOLIFY( rc.b_stat_read );
 838     BOOLIFY( rc.b_mb_tree );
 839 #undef BOOLIFY
 840
 841     return 0;
 842 }
 843
 844 static void mbcmp_init( x264_t *h )
 845 {
 846     int satd = !h->mb.b_lossless && h->param.analyse.i_subpel_refine > 1;
 847     memcpy( h->pixf.mbcmp, satd ? h->pixf.satd : h->pixf.sad_aligned, sizeof(h->pixf.mbcmp) );
 848     memcpy( h->pixf.mbcmp_unaligned, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.mbcmp_unaligned) );
 849     h->pixf.intra_mbcmp_x3_16x16 = satd ? h->pixf.intra_satd_x3_16x16 : h->pixf.intra_sad_x3_16x16;
 850     h->pixf.intra_mbcmp_x3_8x8c = satd ? h->pixf.intra_satd_x3_8x8c : h->pixf.intra_sad_x3_8x8c;
 851     h->pixf.intra_mbcmp_x3_8x8 = satd ? h->pixf.intra_sa8d_x3_8x8 : h->pixf.intra_sad_x3_8x8;
 852     h->pixf.intra_mbcmp_x3_4x4 = satd ? h->pixf.intra_satd_x3_4x4 : h->pixf.intra_sad_x3_4x4;
 853     satd &= h->param.analyse.i_me_method == X264_ME_TESA;
 854     memcpy( h->pixf.fpelcmp, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.fpelcmp) );
 855     memcpy( h->pixf.fpelcmp_x3, satd ? h->pixf.satd_x3 : h->pixf.sad_x3, sizeof(h->pixf.fpelcmp_x3) );
 856     memcpy( h->pixf.fpelcmp_x4, satd ? h->pixf.satd_x4 : h->pixf.sad_x4, sizeof(h->pixf.fpelcmp_x4) );
 857 }
 858
 859 static void x264_set_aspect_ratio( x264_t *h, x264_param_t *param, int initial )
 860 {
 861     /* VUI */
 862     if( param->vui.i_sar_width > 0 && param->vui.i_sar_height > 0 )
 863     {
 864         uint32_t i_w = param->vui.i_sar_width;
 865         uint32_t i_h = param->vui.i_sar_height;
 866         uint32_t old_w = h->param.vui.i_sar_width;
 867         uint32_t old_h = h->param.vui.i_sar_height;
 868
 869         x264_reduce_fraction( &i_w, &i_h );
 870
 871         while( i_w > 65535 || i_h > 65535 )
 872         {
 873             i_w /= 2;
 874             i_h /= 2;
 875         }
 876
 877         x264_reduce_fraction( &i_w, &i_h );
 878
 879         if( i_w != old_w || i_h != old_h || initial )
 880         {
 881             h->param.vui.i_sar_width = 0;
 882             h->param.vui.i_sar_height = 0;
 883             if( i_w == 0 || i_h == 0 )
 884                 x264_log( h, X264_LOG_WARNING, "cannot create valid sample aspect ratio\n" );
 885             else
 886             {
 887                 x264_log( h, initial?X264_LOG_INFO:X264_LOG_DEBUG, "using SAR=%d/%d\n", i_w, i_h );
 888                 h->param.vui.i_sar_width = i_w;
 889                 h->param.vui.i_sar_height = i_h;
 890             }
 891         }
 892     }
 893 }
 894
 895 /****************************************************************************
 896  * x264_encoder_open:
 897  ****************************************************************************/
 898 x264_t *x264_encoder_open( x264_param_t *param )
 899 {
 900     x264_t *h;
 901     char buf[1000], *p;
 902     int qp, i_slicetype_length;
 903
 904     CHECKED_MALLOCZERO( h, sizeof(x264_t) );
 905
 906     /* Create a copy of param */
 907     memcpy( &h->param, param, sizeof(x264_param_t) );
 908
 909     if( param->param_free )
 910         param->param_free( param );
 911
 912     if( x264_validate_parameters( h ) < 0 )
 913         goto fail;
 914
 915     if( h->param.psz_cqm_file )
 916         if( x264_cqm_parse_file( h, h->param.psz_cqm_file ) < 0 )
 917             goto fail;
 918
 919     if( h->param.rc.psz_stat_out )
 920         h->param.rc.psz_stat_out = strdup( h->param.rc.psz_stat_out );
 921     if( h->param.rc.psz_stat_in )
 922         h->param.rc.psz_stat_in = strdup( h->param.rc.psz_stat_in );
 923
 924     x264_set_aspect_ratio( h, &h->param, 1 );
 925
 926     x264_reduce_fraction( &h->param.i_fps_num, &h->param.i_fps_den );
 927     x264_reduce_fraction( &h->param.i_timebase_num, &h->param.i_timebase_den );
 928
 929     /* Init x264_t */
 930     h->i_frame = -1;
 931     h->i_frame_num = 0;
 932     h->i_idr_pic_id = 0;
 933     uint64_t new_timebase_den = h->param.i_timebase_den;
 934     if( h->param.b_dts_compress )
 935     {
 936         /* h->i_dts_compress_multiplier == h->frames.i_bframe_delay + 1 */
 937         h->i_dts_compress_multiplier = h->param.i_bframe ? (h->param.i_bframe_pyramid ? 3 : 2) : 1;
 938         if( h->i_dts_compress_multiplier != 1 )
 939         {
 940             new_timebase_den = h->param.i_timebase_den * h->i_dts_compress_multiplier;
 941             x264_log( h, X264_LOG_DEBUG, "DTS compresion changed timebase: %u/%u -> %u/%"PRIu64"\n",
 942                       h->param.i_timebase_num, h->param.i_timebase_den,
 943                       h->param.i_timebase_num, new_timebase_den );
 944         }
 945     }
 946     else
 947         h->i_dts_compress_multiplier = 1;
 948
 949     if( new_timebase_den * 2 > UINT32_MAX )
 950     {
 951         x264_log( h, X264_LOG_ERROR, "Effective timebase denominator %"PRIu64" exceeds H.264 maximum\n", new_timebase_den );
 952         goto fail;
 953     }
 954     h->param.i_timebase_den = new_timebase_den;
 955
 956     h->sps = &h->sps_array[0];
 957     x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
 958
 959     h->pps = &h->pps_array[0];
 960     x264_pps_init( h->pps, h->param.i_sps_id, &h->param, h->sps );
 961
 962     x264_validate_levels( h, 1 );
 963
 964     h->chroma_qp_table = i_chroma_qp_table + 12 + h->pps->i_chroma_qp_index_offset;
 965
 966     if( x264_cqm_init( h ) < 0 )
 967         goto fail;
 968
 969     h->mb.i_mb_width = h->sps->i_mb_width;
 970     h->mb.i_mb_height = h->sps->i_mb_height;
 971     h->mb.i_mb_count = h->mb.i_mb_width * h->mb.i_mb_height;
 972
 973     /* Init frames. */
 974     if( h->param.i_bframe_adaptive == X264_B_ADAPT_TRELLIS && !h->param.rc.b_stat_read )
 975         h->frames.i_delay = X264_MAX(h->param.i_bframe,3)*4;
 976     else
 977         h->frames.i_delay = h->param.i_bframe;
 978     if( h->param.rc.b_mb_tree || h->param.rc.i_vbv_buffer_size )
 979         h->frames.i_delay = X264_MAX( h->frames.i_delay, h->param.rc.i_lookahead );
 980     i_slicetype_length = h->frames.i_delay;
 981     h->frames.i_delay += h->i_thread_frames - 1;
 982     h->frames.i_delay += h->param.i_sync_lookahead;
 983     h->frames.i_delay += h->param.b_vfr_input && (h->param.rc.i_rc_method == X264_RC_ABR || h->param.rc.b_stat_write
 984                                                  || h->param.rc.i_vbv_buffer_size);
 985     h->frames.i_bframe_delay = h->param.i_bframe ? (h->param.i_bframe_pyramid ? 2 : 1) : 0;
 986
 987     h->frames.i_max_ref0 = h->param.i_frame_reference;
 988     h->frames.i_max_ref1 = X264_MIN( h->sps->vui.i_num_reorder_frames, h->param.i_frame_reference );
 989     h->frames.i_max_dpb  = h->sps->vui.i_max_dec_frame_buffering;
 990     h->frames.b_have_lowres = !h->param.rc.b_stat_read
 991         && ( h->param.rc.i_rc_method == X264_RC_ABR
 992           || h->param.rc.i_rc_method == X264_RC_CRF
 993           || h->param.i_bframe_adaptive
 994           || h->param.i_scenecut_threshold
 995           || h->param.rc.b_mb_tree
 996           || h->param.analyse.i_weighted_pred == X264_WEIGHTP_SMART );
 997     h->frames.b_have_lowres |= h->param.rc.b_stat_read && h->param.rc.i_vbv_buffer_size > 0;
 998     h->frames.b_have_sub8x8_esa = !!(h->param.analyse.inter & X264_ANALYSE_PSUB8x8);
 999
1000     h->frames.i_last_idr =
1001     h->frames.i_last_keyframe = - h->param.i_keyint_max;
1002     h->frames.i_input    = 0;
1003     h->frames.i_largest_pts = h->frames.i_second_largest_pts = -1;
1004     h->frames.i_poc_last_open_gop = -1;
1005
1006     CHECKED_MALLOCZERO( h->frames.unused[0], (h->frames.i_delay + 3) * sizeof(x264_frame_t *) );
1007     /* Allocate room for max refs plus a few extra just in case. */
1008     CHECKED_MALLOCZERO( h->frames.unused[1], (h->i_thread_frames + 20) * sizeof(x264_frame_t *) );
1009     CHECKED_MALLOCZERO( h->frames.current, (h->param.i_sync_lookahead + h->param.i_bframe
1010                         + h->i_thread_frames + 3) * sizeof(x264_frame_t *) );
1011     if( h->param.analyse.i_weighted_pred > 0 )
1012         CHECKED_MALLOCZERO( h->frames.blank_unused, h->i_thread_frames * 4 * sizeof(x264_frame_t *) );
1013     h->i_ref0 = 0;
1014     h->i_ref1 = 0;
1015     h->i_cpb_delay = h->i_coded_fields = h->i_disp_fields = h->i_prev_duration = 0;
1016     h->i_disp_fields_last_frame = -1;
1017     x264_rdo_init();
1018
1019     /* init CPU functions */
1020     x264_predict_16x16_init( h->param.cpu, h->predict_16x16 );
1021     x264_predict_8x8c_init( h->param.cpu, h->predict_8x8c );
1022     x264_predict_8x8_init( h->param.cpu, h->predict_8x8, &h->predict_8x8_filter );
1023     x264_predict_4x4_init( h->param.cpu, h->predict_4x4 );
1024     if( !h->param.b_cabac )
1025         x264_init_vlc_tables();
1026     x264_pixel_init( h->param.cpu, &h->pixf );
1027     x264_dct_init( h->param.cpu, &h->dctf );
1028     x264_zigzag_init( h->param.cpu, &h->zigzagf, h->param.b_interlaced );
1029     x264_mc_init( h->param.cpu, &h->mc );
1030     x264_quant_init( h, h->param.cpu, &h->quantf );
1031     x264_deblock_init( h->param.cpu, &h->loopf );
1032     x264_bitstream_init( h->param.cpu, &h->bsf );
1033     x264_dct_init_weights();
1034
1035     mbcmp_init( h );
1036
1037     p = buf + sprintf( buf, "using cpu capabilities:" );
1038     for( int i = 0; x264_cpu_names[i].flags; i++ )
1039     {
1040         if( !strcmp(x264_cpu_names[i].name, "SSE2")
1041             && h->param.cpu & (X264_CPU_SSE2_IS_FAST|X264_CPU_SSE2_IS_SLOW) )
1042             continue;
1043         if( !strcmp(x264_cpu_names[i].name, "SSE3")
1044             && (h->param.cpu & X264_CPU_SSSE3 || !(h->param.cpu & X264_CPU_CACHELINE_64)) )
1045             continue;
1046         if( !strcmp(x264_cpu_names[i].name, "SSE4.1")
1047             && (h->param.cpu & X264_CPU_SSE42) )
1048             continue;
1049         if( (h->param.cpu & x264_cpu_names[i].flags) == x264_cpu_names[i].flags
1050             && (!i || x264_cpu_names[i].flags != x264_cpu_names[i-1].flags) )
1051             p += sprintf( p, " %s", x264_cpu_names[i].name );
1052     }
1053     if( !h->param.cpu )
1054         p += sprintf( p, " none!" );
1055     x264_log( h, X264_LOG_INFO, "%s\n", buf );
1056
1057     for( qp = h->param.rc.i_qp_min; qp <= h->param.rc.i_qp_max; qp++ )
1058         if( x264_analyse_init_costs( h, qp ) )
1059             goto fail;
1060     if( x264_analyse_init_costs( h, X264_LOOKAHEAD_QP ) )
1061         goto fail;
1062
1063     static const uint16_t cost_mv_correct[7] = { 24, 47, 95, 189, 379, 757, 1515 };
1064     /* Checks for known miscompilation issues. */
1065     if( h->cost_mv[x264_lambda_tab[X264_LOOKAHEAD_QP]][2013] != cost_mv_correct[BIT_DEPTH-8] )
1066     {
1067         x264_log( h, X264_LOG_ERROR, "MV cost test failed: x264 has been miscompiled!\n" );
1068         goto fail;
1069     }
1070
1071     /* Must be volatile or else GCC will optimize it out. */
1072     volatile int temp = 392;
1073     if( x264_clz( temp ) != 23 )
1074     {
1075         x264_log( h, X264_LOG_ERROR, "CLZ test failed: x264 has been miscompiled!\n" );
1076 #if ARCH_X86 || ARCH_X86_64
1077         x264_log( h, X264_LOG_ERROR, "Are you attempting to run an SSE4a-targeted build on a CPU that\n" );
1078         x264_log( h, X264_LOG_ERROR, "doesn't support it?\n" );
1079 #endif
1080         goto fail;
1081     }
1082
1083     h->out.i_nal = 0;
1084     h->out.i_bitstream = X264_MAX( 1000000, h->param.i_width * h->param.i_height * 4
1085         * ( h->param.rc.i_rc_method == X264_RC_ABR ? pow( 0.95, h->param.rc.i_qp_min )
1086           : pow( 0.95, h->param.rc.i_qp_constant ) * X264_MAX( 1, h->param.rc.f_ip_factor )));
1087
1088     CHECKED_MALLOC( h->nal_buffer, h->out.i_bitstream * 3/2 + 4 );
1089     h->nal_buffer_size = h->out.i_bitstream * 3/2 + 4;
1090
1091     if( h->param.i_threads > 1 &&
1092         x264_threadpool_init( &h->threadpool, h->param.i_threads, (void*)x264_encoder_thread_init, h ) )
1093         goto fail;
1094
1095     h->thread[0] = h;
1096     for( int i = 1; i < h->param.i_threads + !!h->param.i_sync_lookahead; i++ )
1097         CHECKED_MALLOC( h->thread[i], sizeof(x264_t) );
1098
1099     for( int i = 0; i < h->param.i_threads; i++ )
1100     {
1101         int init_nal_count = h->param.i_slice_count + 3;
1102         int allocate_threadlocal_data = !h->param.b_sliced_threads || !i;
1103         if( i > 0 )
1104             *h->thread[i] = *h;
1105
1106         if( allocate_threadlocal_data )
1107         {
1108             h->thread[i]->fdec = x264_frame_pop_unused( h, 1 );
1109             if( !h->thread[i]->fdec )
1110                 goto fail;
1111         }
1112         else
1113             h->thread[i]->fdec = h->thread[0]->fdec;
1114
1115         CHECKED_MALLOC( h->thread[i]->out.p_bitstream, h->out.i_bitstream );
1116         /* Start each thread with room for init_nal_count NAL units; it'll realloc later if needed. */
1117         CHECKED_MALLOC( h->thread[i]->out.nal, init_nal_count*sizeof(x264_nal_t) );
1118         h->thread[i]->out.i_nals_allocated = init_nal_count;
1119
1120         if( allocate_threadlocal_data && x264_macroblock_cache_allocate( h->thread[i] ) < 0 )
1121             goto fail;
1122     }
1123
1124     if( x264_lookahead_init( h, i_slicetype_length ) )
1125         goto fail;
1126
1127     for( int i = 0; i < h->param.i_threads; i++ )
1128         if( x264_macroblock_thread_allocate( h->thread[i], 0 ) < 0 )
1129             goto fail;
1130
1131     if( x264_ratecontrol_new( h ) < 0 )
1132         goto fail;
1133
1134     if( h->param.i_nal_hrd )
1135     {
1136         x264_log( h, X264_LOG_DEBUG, "HRD bitrate: %i bits/sec\n", h->sps->vui.hrd.i_bit_rate_unscaled );
1137         x264_log( h, X264_LOG_DEBUG, "CPB size: %i bits\n", h->sps->vui.hrd.i_cpb_size_unscaled );
1138     }
1139
1140     if( h->param.psz_dump_yuv )
1141     {
1142         /* create or truncate the reconstructed video file */
1143         FILE *f = fopen( h->param.psz_dump_yuv, "w" );
1144         if( !f )
1145         {
1146             x264_log( h, X264_LOG_ERROR, "dump_yuv: can't write to %s\n", h->param.psz_dump_yuv );
1147             goto fail;
1148         }
1149         else if( !x264_is_regular_file( f ) )
1150         {
1151             x264_log( h, X264_LOG_ERROR, "dump_yuv: incompatible with non-regular file %s\n", h->param.psz_dump_yuv );
1152             goto fail;
1153         }
1154         fclose( f );
1155     }
1156
1157     const char *profile = h->sps->i_profile_idc == PROFILE_BASELINE ? "Baseline" :
1158                           h->sps->i_profile_idc == PROFILE_MAIN ? "Main" :
1159                           h->sps->i_profile_idc == PROFILE_HIGH ? "High" :
1160                           h->sps->i_profile_idc == PROFILE_HIGH10 ? "High 10" :
1161                           "High 4:4:4 Predictive";
1162
1163     if( h->sps->i_profile_idc < PROFILE_HIGH10 )
1164     {
1165         x264_log( h, X264_LOG_INFO, "profile %s, level %d.%d\n",
1166             profile, h->sps->i_level_idc/10, h->sps->i_level_idc%10 );
1167     }
1168     else
1169     {
1170         x264_log( h, X264_LOG_INFO, "profile %s, level %d.%d, bit depth %d\n",
1171             profile, h->sps->i_level_idc/10, h->sps->i_level_idc%10, BIT_DEPTH );
1172     }
1173
1174     return h;
1175 fail:
1176     x264_free( h );
1177     return NULL;
1178 }
1179
1180 /****************************************************************************
1181  * x264_encoder_reconfig:
1182  ****************************************************************************/
1183 int x264_encoder_reconfig( x264_t *h, x264_param_t *param )
1184 {
1185     int rc_reconfig = 0;
1186     h = h->thread[h->thread[0]->i_thread_phase];
1187     x264_set_aspect_ratio( h, param, 0 );
1188 #define COPY(var) h->param.var = param->var
1189     COPY( i_frame_reference ); // but never uses more refs than initially specified
1190     COPY( i_bframe_bias );
1191     if( h->param.i_scenecut_threshold )
1192         COPY( i_scenecut_threshold ); // can't turn it on or off, only vary the threshold
1193     COPY( b_deblocking_filter );
1194     COPY( i_deblocking_filter_alphac0 );
1195     COPY( i_deblocking_filter_beta );
1196     COPY( analyse.inter );
1197     COPY( analyse.intra );
1198     COPY( analyse.i_direct_mv_pred );
1199     /* Scratch buffer prevents me_range from being increased for esa/tesa */
1200     if( h->param.analyse.i_me_method < X264_ME_ESA || param->analyse.i_me_range < h->param.analyse.i_me_range )
1201         COPY( analyse.i_me_range );
1202     COPY( analyse.i_noise_reduction );
1203     /* We can't switch out of subme=0 during encoding. */
1204     if( h->param.analyse.i_subpel_refine )
1205         COPY( analyse.i_subpel_refine );
1206     COPY( analyse.i_trellis );
1207     COPY( analyse.b_chroma_me );
1208     COPY( analyse.b_dct_decimate );
1209     COPY( analyse.b_fast_pskip );
1210     COPY( analyse.b_mixed_references );
1211     COPY( analyse.f_psy_rd );
1212     COPY( analyse.f_psy_trellis );
1213     // can only twiddle these if they were enabled to begin with:
1214     if( h->param.analyse.i_me_method >= X264_ME_ESA || param->analyse.i_me_method < X264_ME_ESA )
1215         COPY( analyse.i_me_method );
1216     if( h->param.analyse.i_me_method >= X264_ME_ESA && !h->frames.b_have_sub8x8_esa )
1217         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
1218     if( h->pps->b_transform_8x8_mode )
1219         COPY( analyse.b_transform_8x8 );
1220     if( h->frames.i_max_ref1 > 1 )
1221         COPY( i_bframe_pyramid );
1222     COPY( i_slice_max_size );
1223     COPY( i_slice_max_mbs );
1224     COPY( i_slice_count );
1225
1226     /* VBV can't be turned on if it wasn't on to begin with */
1227     if( h->param.rc.i_vbv_max_bitrate > 0 && h->param.rc.i_vbv_buffer_size > 0 &&
1228           param->rc.i_vbv_max_bitrate > 0 &&   param->rc.i_vbv_buffer_size > 0 )
1229     {
1230         COPY( rc.i_vbv_max_bitrate );
1231         COPY( rc.i_vbv_buffer_size );
1232         COPY( rc.i_bitrate );
1233         rc_reconfig = 1;
1234     }
1235     if( h->param.rc.f_rf_constant != param->rc.f_rf_constant )
1236     {
1237         COPY( rc.f_rf_constant );
1238         rc_reconfig = 1;
1239     }
1240     if( h->param.rc.f_rf_constant_max != param->rc.f_rf_constant_max )
1241     {
1242         COPY( rc.f_rf_constant_max );
1243         rc_reconfig = 1;
1244     }
1245
1246 #undef COPY
1247
1248     mbcmp_init( h );
1249
1250     int ret = x264_validate_parameters( h );
1251
1252     /* Supported reconfiguration options (1-pass only):
1253      * vbv-maxrate
1254      * vbv-bufsize
1255      * crf
1256      * bitrate (CBR only) */
1257     if( !ret && rc_reconfig )
1258         x264_ratecontrol_init_reconfigurable( h, 0 );
1259
1260     return ret;
1261 }
1262
1263 /****************************************************************************
1264  * x264_encoder_parameters:
1265  ****************************************************************************/
1266 void x264_encoder_parameters( x264_t *h, x264_param_t *param )
1267 {
1268     memcpy( param, &h->thread[h->i_thread_phase]->param, sizeof(x264_param_t) );
1269 }
1270
1271 /* internal usage */
1272 static void x264_nal_start( x264_t *h, int i_type, int i_ref_idc )
1273 {
1274     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1275
1276     nal->i_ref_idc        = i_ref_idc;
1277     nal->i_type           = i_type;
1278     nal->b_long_startcode = 1;
1279
1280     nal->i_payload= 0;
1281     nal->p_payload= &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8];
1282 }
1283
1284 /* if number of allocated nals is not enough, re-allocate a larger one. */
1285 static int x264_nal_check_buffer( x264_t *h )
1286 {
1287     if( h->out.i_nal >= h->out.i_nals_allocated )
1288     {
1289         x264_nal_t *new_out = x264_malloc( sizeof(x264_nal_t) * (h->out.i_nals_allocated*2) );
1290         if( !new_out )
1291             return -1;
1292         memcpy( new_out, h->out.nal, sizeof(x264_nal_t) * (h->out.i_nals_allocated) );
1293         x264_free( h->out.nal );
1294         h->out.nal = new_out;
1295         h->out.i_nals_allocated *= 2;
1296     }
1297     return 0;
1298 }
1299
1300 static int x264_nal_end( x264_t *h )
1301 {
1302     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1303     nal->i_payload = &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8] - nal->p_payload;
1304     if( h->param.nalu_process )
1305         h->param.nalu_process( h, nal );
1306     h->out.i_nal++;
1307
1308     return x264_nal_check_buffer( h );
1309 }
1310
1311 static int x264_encoder_encapsulate_nals( x264_t *h, int start )
1312 {
1313     int nal_size = 0, previous_nal_size = 0;
1314
1315     if( h->param.nalu_process )
1316     {
1317         for( int i = start; i < h->out.i_nal; i++ )
1318             nal_size += h->out.nal[i].i_payload;
1319         return nal_size;
1320     }
1321
1322     for( int i = 0; i < start; i++ )
1323         previous_nal_size += h->out.nal[i].i_payload;
1324
1325     for( int i = start; i < h->out.i_nal; i++ )
1326         nal_size += h->out.nal[i].i_payload;
1327
1328     /* Worst-case NAL unit escaping: reallocate the buffer if it's too small. */
1329     if( h->nal_buffer_size < nal_size * 3/2 + h->out.i_nal * 4 )
1330     {
1331         uint8_t *buf = x264_malloc( nal_size * 2 + h->out.i_nal * 4 );
1332         if( !buf )
1333             return -1;
1334         if( previous_nal_size )
1335             memcpy( buf, h->nal_buffer, previous_nal_size );
1336         x264_free( h->nal_buffer );
1337         h->nal_buffer = buf;
1338     }
1339
1340     uint8_t *nal_buffer = h->nal_buffer + previous_nal_size;
1341
1342     for( int i = start; i < h->out.i_nal; i++ )
1343     {
1344         h->out.nal[i].b_long_startcode = !i || h->out.nal[i].i_type == NAL_SPS || h->out.nal[i].i_type == NAL_PPS;
1345         x264_nal_encode( h, nal_buffer, &h->out.nal[i] );
1346         nal_buffer += h->out.nal[i].i_payload;
1347     }
1348
1349     x264_emms();
1350
1351     return nal_buffer - (h->nal_buffer + previous_nal_size);
1352 }
1353
1354 /****************************************************************************
1355  * x264_encoder_headers:
1356  ****************************************************************************/
1357 int x264_encoder_headers( x264_t *h, x264_nal_t **pp_nal, int *pi_nal )
1358 {
1359     int frame_size = 0;
1360     /* init bitstream context */
1361     h->out.i_nal = 0;
1362     bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
1363
1364     /* Write SEI, SPS and PPS. */
1365
1366     /* generate sequence parameters */
1367     x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
1368     x264_sps_write( &h->out.bs, h->sps );
1369     if( x264_nal_end( h ) )
1370         return -1;
1371
1372     /* generate picture parameters */
1373     x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
1374     x264_pps_write( &h->out.bs, h->pps );
1375     if( x264_nal_end( h ) )
1376         return -1;
1377
1378     /* identify ourselves */
1379     x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
1380     if( x264_sei_version_write( h, &h->out.bs ) )
1381         return -1;
1382     if( x264_nal_end( h ) )
1383         return -1;
1384
1385     frame_size = x264_encoder_encapsulate_nals( h, 0 );
1386
1387     /* now set output*/
1388     *pi_nal = h->out.i_nal;
1389     *pp_nal = &h->out.nal[0];
1390     h->out.i_nal = 0;
1391
1392     return frame_size;
1393 }
1394
1395 /* Check to see whether we have chosen a reference list ordering different
1396  * from the standard's default. */
1397 static inline void x264_reference_check_reorder( x264_t *h )
1398 {
1399     /* The reorder check doesn't check for missing frames, so just
1400      * force a reorder if one of the reference list is corrupt. */
1401     for( int i = 0; h->frames.reference[i]; i++ )
1402         if( h->frames.reference[i]->b_corrupt )
1403         {
1404             h->b_ref_reorder[0] = 1;
1405             return;
1406         }
1407     for( int i = 0; i < h->i_ref0 - 1; i++ )
1408         /* P and B-frames use different default orders. */
1409         if( h->sh.i_type == SLICE_TYPE_P ? h->fref0[i]->i_frame_num < h->fref0[i+1]->i_frame_num
1410                                          : h->fref0[i]->i_poc < h->fref0[i+1]->i_poc )
1411         {
1412             h->b_ref_reorder[0] = 1;
1413             return;
1414         }
1415 }
1416
1417 /* return -1 on failure, else return the index of the new reference frame */
1418 int x264_weighted_reference_duplicate( x264_t *h, int i_ref, const x264_weight_t *w )
1419 {
1420     int i = h->i_ref0;
1421     int j = 1;
1422     x264_frame_t *newframe;
1423     if( i <= 1 ) /* empty list, definitely can't duplicate frame */
1424         return -1;
1425
1426     newframe = x264_frame_pop_blank_unused( h );
1427
1428     //FIXME: probably don't need to copy everything
1429     *newframe = *h->fref0[i_ref];
1430     newframe->i_reference_count = 1;
1431     newframe->orig = h->fref0[i_ref];
1432     newframe->b_duplicate = 1;
1433     memcpy( h->fenc->weight[j], w, sizeof(h->fenc->weight[i]) );
1434
1435     /* shift the frames to make space for the dupe. */
1436     h->b_ref_reorder[0] = 1;
1437     if( h->i_ref0 < 16 )
1438         ++h->i_ref0;
1439     h->fref0[15] = NULL;
1440     x264_frame_unshift( &h->fref0[j], newframe );
1441
1442     return j;
1443 }
1444
1445 static void x264_weighted_pred_init( x264_t *h )
1446 {
1447     /* for now no analysis and set all weights to nothing */
1448     for( int i_ref = 0; i_ref < h->i_ref0; i_ref++ )
1449         h->fenc->weighted[i_ref] = h->fref0[i_ref]->filtered[0];
1450
1451     // FIXME: This only supports weighting of one reference frame
1452     // and duplicates of that frame.
1453     h->fenc->i_lines_weighted = 0;
1454
1455     for( int i_ref = 0; i_ref < (h->i_ref0 << h->sh.b_mbaff); i_ref++ )
1456         for( int i = 0; i < 3; i++ )
1457             h->sh.weight[i_ref][i].weightfn = NULL;
1458
1459
1460     if( h->sh.i_type != SLICE_TYPE_P || h->param.analyse.i_weighted_pred <= 0 )
1461         return;
1462
1463     int i_padv = PADV << h->param.b_interlaced;
1464     int denom = -1;
1465     int weightluma = 0;
1466     int buffer_next = 0;
1467     //FIXME: when chroma support is added, move this into loop
1468     h->sh.weight[0][1].weightfn = h->sh.weight[0][2].weightfn = NULL;
1469     h->sh.weight[0][1].i_denom = h->sh.weight[0][2].i_denom = 0;
1470     for( int j = 0; j < h->i_ref0; j++ )
1471     {
1472         if( h->fenc->weight[j][0].weightfn )
1473         {
1474             h->sh.weight[j][0] = h->fenc->weight[j][0];
1475             // if weight is useless, don't write it to stream
1476             if( h->sh.weight[j][0].i_scale == 1<<h->sh.weight[j][0].i_denom && h->sh.weight[j][0].i_offset == 0 )
1477                 h->sh.weight[j][0].weightfn = NULL;
1478             else
1479             {
1480                 if( !weightluma )
1481                 {
1482                     weightluma = 1;
1483                     h->sh.weight[0][0].i_denom = denom = h->sh.weight[j][0].i_denom;
1484                     assert( x264_clip3( denom, 0, 7 ) == denom );
1485                 }
1486                 assert( h->sh.weight[j][0].i_denom == denom );
1487                 assert( x264_clip3( h->sh.weight[j][0].i_scale, 0, 127 ) == h->sh.weight[j][0].i_scale );
1488                 assert( x264_clip3( h->sh.weight[j][0].i_offset, -128, 127 ) == h->sh.weight[j][0].i_offset );
1489                 h->fenc->weighted[j] = h->mb.p_weight_buf[buffer_next++] +
1490                     h->fenc->i_stride[0] * i_padv + PADH;
1491             }
1492         }
1493
1494         //scale full resolution frame
1495         if( h->sh.weight[j][0].weightfn && h->param.i_threads == 1 )
1496         {
1497             pixel *src = h->fref0[j]->filtered[0] - h->fref0[j]->i_stride[0]*i_padv - PADH;
1498             pixel *dst = h->fenc->weighted[j] - h->fenc->i_stride[0]*i_padv - PADH;
1499             int stride = h->fenc->i_stride[0];
1500             int width = h->fenc->i_width[0] + PADH*2;
1501             int height = h->fenc->i_lines[0] + i_padv*2;
1502             x264_weight_scale_plane( h, dst, stride, src, stride, width, height, &h->sh.weight[j][0] );
1503             h->fenc->i_lines_weighted = height;
1504         }
1505     }
1506     if( !weightluma )
1507         h->sh.weight[0][0].i_denom = 0;
1508 }
1509
1510 static inline void x264_reference_build_list( x264_t *h, int i_poc )
1511 {
1512     int b_ok;
1513
1514     /* build ref list 0/1 */
1515     h->mb.pic.i_fref[0] = h->i_ref0 = 0;
1516     h->mb.pic.i_fref[1] = h->i_ref1 = 0;
1517     if( h->sh.i_type == SLICE_TYPE_I )
1518         return;
1519
1520     for( int i = 0; h->frames.reference[i]; i++ )
1521     {
1522         if( h->frames.reference[i]->b_corrupt )
1523             continue;
1524         if( h->frames.reference[i]->i_poc < i_poc )
1525             h->fref0[h->i_ref0++] = h->frames.reference[i];
1526         else if( h->frames.reference[i]->i_poc > i_poc )
1527             h->fref1[h->i_ref1++] = h->frames.reference[i];
1528     }
1529
1530     /* Order ref0 from higher to lower poc */
1531     do
1532     {
1533         b_ok = 1;
1534         for( int i = 0; i < h->i_ref0 - 1; i++ )
1535         {
1536             if( h->fref0[i]->i_poc < h->fref0[i+1]->i_poc )
1537             {
1538                 XCHG( x264_frame_t*, h->fref0[i], h->fref0[i+1] );
1539                 b_ok = 0;
1540                 break;
1541             }
1542         }
1543     } while( !b_ok );
1544
1545     if( h->sh.i_mmco_remove_from_end )
1546         for( int i = h->i_ref0-1; i >= h->i_ref0 - h->sh.i_mmco_remove_from_end; i-- )
1547         {
1548             int diff = h->i_frame_num - h->fref0[i]->i_frame_num;
1549             h->sh.mmco[h->sh.i_mmco_command_count].i_poc = h->fref0[i]->i_poc;
1550             h->sh.mmco[h->sh.i_mmco_command_count++].i_difference_of_pic_nums = diff;
1551         }
1552
1553     /* Order ref1 from lower to higher poc (bubble sort) for B-frame */
1554     do
1555     {
1556         b_ok = 1;
1557         for( int i = 0; i < h->i_ref1 - 1; i++ )
1558         {
1559             if( h->fref1[i]->i_poc > h->fref1[i+1]->i_poc )
1560             {
1561                 XCHG( x264_frame_t*, h->fref1[i], h->fref1[i+1] );
1562                 b_ok = 0;
1563                 break;
1564             }
1565         }
1566     } while( !b_ok );
1567
1568     x264_reference_check_reorder( h );
1569
1570     h->i_ref1 = X264_MIN( h->i_ref1, h->frames.i_max_ref1 );
1571     h->i_ref0 = X264_MIN( h->i_ref0, h->frames.i_max_ref0 );
1572     h->i_ref0 = X264_MIN( h->i_ref0, h->param.i_frame_reference ); // if reconfig() has lowered the limit
1573
1574     /* add duplicates */
1575     if( h->fenc->i_type == X264_TYPE_P )
1576     {
1577         int idx = -1;
1578         if( h->param.analyse.i_weighted_pred == X264_WEIGHTP_SMART )
1579         {
1580             x264_weight_t w[3];
1581             w[1].weightfn = w[2].weightfn = NULL;
1582             if( h->param.rc.b_stat_read )
1583                 x264_ratecontrol_set_weights( h, h->fenc );
1584
1585             if( !h->fenc->weight[0][0].weightfn )
1586             {
1587                 h->fenc->weight[0][0].i_denom = 0;
1588                 SET_WEIGHT( w[0], 1, 1, 0, -1 );
1589                 idx = x264_weighted_reference_duplicate( h, 0, w );
1590             }
1591             else
1592             {
1593                 if( h->fenc->weight[0][0].i_scale == 1<<h->fenc->weight[0][0].i_denom )
1594                 {
1595                     SET_WEIGHT( h->fenc->weight[0][0], 1, 1, 0, h->fenc->weight[0][0].i_offset );
1596                 }
1597                 x264_weighted_reference_duplicate( h, 0, weight_none );
1598                 if( h->fenc->weight[0][0].i_offset > -128 )
1599                 {
1600                     w[0] = h->fenc->weight[0][0];
1601                     w[0].i_offset--;
1602                     h->mc.weight_cache( h, &w[0] );
1603                     idx = x264_weighted_reference_duplicate( h, 0, w );
1604                 }
1605             }
1606         }
1607         else if( h->param.analyse.i_weighted_pred == X264_WEIGHTP_BLIND )
1608         {
1609             //weighted offset=-1
1610             x264_weight_t w[3];
1611             SET_WEIGHT( w[0], 1, 1, 0, -1 );
1612             h->fenc->weight[0][0].i_denom = 0;
1613             w[1].weightfn = w[2].weightfn = NULL;
1614             idx = x264_weighted_reference_duplicate( h, 0, w );
1615         }
1616         h->mb.ref_blind_dupe = idx;
1617     }
1618
1619     assert( h->i_ref0 + h->i_ref1 <= 16 );
1620     h->mb.pic.i_fref[0] = h->i_ref0;
1621     h->mb.pic.i_fref[1] = h->i_ref1;
1622 }
1623
1624 static void x264_fdec_filter_row( x264_t *h, int mb_y, int b_inloop )
1625 {
1626     /* mb_y is the mb to be encoded next, not the mb to be filtered here */
1627     int b_hpel = h->fdec->b_kept_as_ref;
1628     int b_deblock = h->sh.i_disable_deblocking_filter_idc != 1;
1629     int b_end = mb_y == h->i_threadslice_end;
1630     int b_measure_quality = 1;
1631     int min_y = mb_y - (1 << h->sh.b_mbaff);
1632     int b_start = min_y == h->i_threadslice_start;
1633     int max_y = b_end ? h->i_threadslice_end : mb_y;
1634     b_deblock &= b_hpel || h->param.psz_dump_yuv;
1635     if( h->param.b_sliced_threads && b_start && min_y && !b_inloop )
1636     {
1637         b_deblock = 0;         /* We already deblocked on the inloop pass. */
1638         b_measure_quality = 0; /* We already measured quality on the inloop pass. */
1639     }
1640     if( mb_y & h->sh.b_mbaff )
1641         return;
1642     if( min_y < h->i_threadslice_start )
1643         return;
1644
1645     if( b_deblock )
1646         for( int y = min_y; y < max_y; y += (1 << h->sh.b_mbaff) )
1647             x264_frame_deblock_row( h, y );
1648
1649     if( b_hpel )
1650     {
1651         int end = mb_y == h->mb.i_mb_height;
1652         x264_frame_expand_border( h, h->fdec, min_y, end );
1653         if( h->param.analyse.i_subpel_refine )
1654         {
1655             x264_frame_filter( h, h->fdec, min_y, end );
1656             x264_frame_expand_border_filtered( h, h->fdec, min_y, end );
1657         }
1658     }
1659
1660     if( h->i_thread_frames > 1 && h->fdec->b_kept_as_ref )
1661         x264_frame_cond_broadcast( h->fdec, mb_y*16 + (b_end ? 10000 : -(X264_THREAD_HEIGHT << h->sh.b_mbaff)) );
1662
1663     min_y = min_y*16 - 8 * !b_start;
1664     max_y = b_end ? X264_MIN( h->i_threadslice_end*16 , h->param.i_height ) : mb_y*16 - 8;
1665
1666     if( b_measure_quality )
1667     {
1668         if( h->param.analyse.b_psnr )
1669         {
1670             uint64_t ssd_y = x264_pixel_ssd_wxh( &h->pixf,
1671                 h->fdec->plane[0] + min_y * h->fdec->i_stride[0], h->fdec->i_stride[0],
1672                 h->fenc->plane[0] + min_y * h->fenc->i_stride[0], h->fenc->i_stride[0],
1673                 h->param.i_width, max_y-min_y );
1674             uint64_t ssd_uv = x264_pixel_ssd_nv12( &h->pixf,
1675                 h->fdec->plane[1] + (min_y>>1) * h->fdec->i_stride[1], h->fdec->i_stride[1],
1676                 h->fenc->plane[1] + (min_y>>1) * h->fenc->i_stride[1], h->fenc->i_stride[1],
1677                 h->param.i_width>>1, (max_y-min_y)>>1 );
1678             h->stat.frame.i_ssd[0] += ssd_y;
1679             h->stat.frame.i_ssd[1] += (uint32_t)ssd_uv;
1680             h->stat.frame.i_ssd[2] += ssd_uv>>32;
1681         }
1682
1683         if( h->param.analyse.b_ssim )
1684         {
1685             x264_emms();
1686             /* offset by 2 pixels to avoid alignment of ssim blocks with dct blocks,
1687              * and overlap by 4 */
1688             min_y += b_start ? 2 : -6;
1689             h->stat.frame.f_ssim +=
1690                 x264_pixel_ssim_wxh( &h->pixf,
1691                     h->fdec->plane[0] + 2+min_y*h->fdec->i_stride[0], h->fdec->i_stride[0],
1692                     h->fenc->plane[0] + 2+min_y*h->fenc->i_stride[0], h->fenc->i_stride[0],
1693                     h->param.i_width-2, max_y-min_y, h->scratch_buffer );
1694         }
1695     }
1696 }
1697
1698 static inline int x264_reference_update( x264_t *h )
1699 {
1700     if( !h->fdec->b_kept_as_ref )
1701     {
1702         if( h->i_thread_frames > 1 )
1703         {
1704             x264_frame_push_unused( h, h->fdec );
1705             h->fdec = x264_frame_pop_unused( h, 1 );
1706             if( !h->fdec )
1707                 return -1;
1708         }
1709         return 0;
1710     }
1711
1712     /* apply mmco from previous frame. */
1713     for( int i = 0; i < h->sh.i_mmco_command_count; i++ )
1714         for( int j = 0; h->frames.reference[j]; j++ )
1715             if( h->frames.reference[j]->i_poc == h->sh.mmco[i].i_poc )
1716                 x264_frame_push_unused( h, x264_frame_shift( &h->frames.reference[j] ) );
1717
1718     /* move frame in the buffer */
1719     x264_frame_push( h->frames.reference, h->fdec );
1720     if( h->frames.reference[h->sps->i_num_ref_frames] )
1721         x264_frame_push_unused( h, x264_frame_shift( h->frames.reference ) );
1722     h->fdec = x264_frame_pop_unused( h, 1 );
1723     if( !h->fdec )
1724         return -1;
1725     return 0;
1726 }
1727
1728 static inline void x264_reference_reset( x264_t *h )
1729 {
1730     while( h->frames.reference[0] )
1731         x264_frame_push_unused( h, x264_frame_pop( h->frames.reference ) );
1732     h->fdec->i_poc =
1733     h->fenc->i_poc = 0;
1734 }
1735
1736 static inline void x264_reference_hierarchy_reset( x264_t *h )
1737 {
1738     int ref;
1739     int b_hasdelayframe = 0;
1740
1741     /* look for delay frames -- chain must only contain frames that are disposable */
1742     for( int i = 0; h->frames.current[i] && IS_DISPOSABLE( h->frames.current[i]->i_type ); i++ )
1743         b_hasdelayframe |= h->frames.current[i]->i_coded
1744                         != h->frames.current[i]->i_frame + h->sps->vui.i_num_reorder_frames;
1745
1746     /* This function must handle b-pyramid and clear frames for open-gop */
1747     if( h->param.i_bframe_pyramid != X264_B_PYRAMID_STRICT && !b_hasdelayframe && h->frames.i_poc_last_open_gop == -1 )
1748         return;
1749
1750     /* Remove last BREF. There will never be old BREFs in the
1751      * dpb during a BREF decode when pyramid == STRICT */
1752     for( ref = 0; h->frames.reference[ref]; ref++ )
1753     {
1754         if( ( h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT
1755             && h->frames.reference[ref]->i_type == X264_TYPE_BREF )
1756             || ( h->frames.reference[ref]->i_poc < h->frames.i_poc_last_open_gop
1757             && h->sh.i_type != SLICE_TYPE_B ) )
1758         {
1759             int diff = h->i_frame_num - h->frames.reference[ref]->i_frame_num;
1760             h->sh.mmco[h->sh.i_mmco_command_count].i_difference_of_pic_nums = diff;
1761             h->sh.mmco[h->sh.i_mmco_command_count++].i_poc = h->frames.reference[ref]->i_poc;
1762             x264_frame_push_unused( h, x264_frame_shift( &h->frames.reference[ref] ) );
1763             h->b_ref_reorder[0] = 1;
1764             ref--;
1765         }
1766     }
1767
1768     /* Prepare room in the dpb for the delayed display time of the later b-frame's */
1769     if( h->param.i_bframe_pyramid )
1770         h->sh.i_mmco_remove_from_end = X264_MAX( ref + 2 - h->frames.i_max_dpb, 0 );
1771 }
1772
1773 static inline void x264_slice_init( x264_t *h, int i_nal_type, int i_global_qp )
1774 {
1775     /* ------------------------ Create slice header  ----------------------- */
1776     if( i_nal_type == NAL_SLICE_IDR )
1777     {
1778         x264_slice_header_init( h, &h->sh, h->sps, h->pps, h->i_idr_pic_id, h->i_frame_num, i_global_qp );
1779
1780         /* increment id */
1781         h->i_idr_pic_id = ( h->i_idr_pic_id + 1 ) % 65536;
1782     }
1783     else
1784     {
1785         x264_slice_header_init( h, &h->sh, h->sps, h->pps, -1, h->i_frame_num, i_global_qp );
1786
1787         h->sh.i_num_ref_idx_l0_active = h->i_ref0 <= 0 ? 1 : h->i_ref0;
1788         h->sh.i_num_ref_idx_l1_active = h->i_ref1 <= 0 ? 1 : h->i_ref1;
1789         if( h->sh.i_num_ref_idx_l0_active != h->pps->i_num_ref_idx_l0_default_active ||
1790             (h->sh.i_type == SLICE_TYPE_B && h->sh.i_num_ref_idx_l1_active != h->pps->i_num_ref_idx_l1_default_active) )
1791         {
1792             h->sh.b_num_ref_idx_override = 1;
1793         }
1794     }
1795
1796     h->fdec->i_frame_num = h->sh.i_frame_num;
1797
1798     if( h->sps->i_poc_type == 0 )
1799     {
1800         h->sh.i_poc = h->fdec->i_poc;
1801         if( h->param.b_interlaced )
1802         {
1803             h->sh.i_delta_poc_bottom = h->param.b_tff ? 1 : -1;
1804             if( h->sh.i_delta_poc_bottom == -1 )
1805                 h->sh.i_poc = h->fdec->i_poc + 1;
1806         }
1807         else
1808             h->sh.i_delta_poc_bottom = 0;
1809     }
1810     else if( h->sps->i_poc_type == 1 )
1811     {
1812         /* FIXME TODO FIXME */
1813     }
1814     else
1815     {
1816         /* Nothing to do ? */
1817     }
1818
1819     x264_macroblock_slice_init( h );
1820 }
1821
1822 static int x264_slice_write( x264_t *h )
1823 {
1824     int i_skip;
1825     int mb_xy, i_mb_x, i_mb_y;
1826     int i_skip_bak = 0; /* Shut up GCC. */
1827     bs_t bs_bak;
1828     x264_cabac_t cabac_bak;
1829     uint8_t cabac_prevbyte_bak = 0; /* Shut up GCC. */
1830     /* Assume no more than 3 bytes of NALU escaping.
1831      * NALUs other than the first use a 3-byte startcode. */
1832     int overhead_guess = (NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal)) + 3;
1833     int slice_max_size = h->param.i_slice_max_size > 0 ? (h->param.i_slice_max_size-overhead_guess)*8 : INT_MAX;
1834     int starting_bits = bs_pos(&h->out.bs);
1835     int b_deblock = h->sh.i_disable_deblocking_filter_idc != 1;
1836     int b_hpel = h->fdec->b_kept_as_ref;
1837     b_deblock &= b_hpel || h->param.psz_dump_yuv;
1838     bs_realign( &h->out.bs );
1839
1840     /* Slice */
1841     x264_nal_start( h, h->i_nal_type, h->i_nal_ref_idc );
1842     h->out.nal[h->out.i_nal].i_first_mb = h->sh.i_first_mb;
1843
1844     /* Slice header */
1845     x264_macroblock_thread_init( h );
1846
1847     /* If this isn't the first slice in the threadslice, set the slice QP
1848      * equal to the last QP in the previous slice for more accurate
1849      * CABAC initialization. */
1850     if( h->sh.i_first_mb != h->i_threadslice_start * h->mb.i_mb_width )
1851     {
1852         h->sh.i_qp = h->mb.i_last_qp;
1853         h->sh.i_qp_delta = h->sh.i_qp - h->pps->i_pic_init_qp;
1854     }
1855
1856     x264_slice_header_write( &h->out.bs, &h->sh, h->i_nal_ref_idc );
1857     if( h->param.b_cabac )
1858     {
1859         /* alignment needed */
1860         bs_align_1( &h->out.bs );
1861
1862         /* init cabac */
1863         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 );
1864         x264_cabac_encode_init ( &h->cabac, h->out.bs.p, h->out.bs.p_end );
1865     }
1866     h->mb.i_last_qp = h->sh.i_qp;
1867     h->mb.i_last_dqp = 0;
1868
1869     i_mb_y = h->sh.i_first_mb / h->mb.i_mb_width;
1870     i_mb_x = h->sh.i_first_mb % h->mb.i_mb_width;
1871     i_skip = 0;
1872
1873     while( (mb_xy = i_mb_x + i_mb_y * h->mb.i_mb_width) <= h->sh.i_last_mb )
1874     {
1875         int mb_spos = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
1876         if( h->param.i_slice_max_size > 0 )
1877         {
1878             /* We don't need the contexts because flushing the CABAC encoder has no context
1879              * dependency and macroblocks are only re-encoded in the case where a slice is
1880              * ended (and thus the content of all contexts are thrown away). */
1881             if( h->param.b_cabac )
1882             {
1883                 memcpy( &cabac_bak, &h->cabac, offsetof(x264_cabac_t, f8_bits_encoded) );
1884                 /* x264's CABAC writer modifies the previous byte during carry, so it has to be
1885                  * backed up. */
1886                 cabac_prevbyte_bak = h->cabac.p[-1];
1887             }
1888             else
1889             {
1890                 bs_bak = h->out.bs;
1891                 i_skip_bak = i_skip;
1892             }
1893         }
1894
1895         if( i_mb_x == 0 && !h->mb.b_reencode_mb )
1896             x264_fdec_filter_row( h, i_mb_y, 1 );
1897
1898         /* load cache */
1899         x264_macroblock_cache_load( h, i_mb_x, i_mb_y );
1900
1901         x264_macroblock_analyse( h );
1902
1903         /* encode this macroblock -> be careful it can change the mb type to P_SKIP if needed */
1904         x264_macroblock_encode( h );
1905
1906         if( x264_bitstream_check_buffer( h ) )
1907             return -1;
1908
1909         if( h->param.b_cabac )
1910         {
1911             if( mb_xy > h->sh.i_first_mb && !(h->sh.b_mbaff && (i_mb_y&1)) )
1912                 x264_cabac_encode_terminal( &h->cabac );
1913
1914             if( IS_SKIP( h->mb.i_type ) )
1915                 x264_cabac_mb_skip( h, 1 );
1916             else
1917             {
1918                 if( h->sh.i_type != SLICE_TYPE_I )
1919                     x264_cabac_mb_skip( h, 0 );
1920                 x264_macroblock_write_cabac( h, &h->cabac );
1921             }
1922         }
1923         else
1924         {
1925             if( IS_SKIP( h->mb.i_type ) )
1926                 i_skip++;
1927             else
1928             {
1929                 if( h->sh.i_type != SLICE_TYPE_I )
1930                 {
1931                     bs_write_ue( &h->out.bs, i_skip );  /* skip run */
1932                     i_skip = 0;
1933                 }
1934                 x264_macroblock_write_cavlc( h );
1935             }
1936         }
1937
1938         int total_bits = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
1939         int mb_size = total_bits - mb_spos;
1940
1941         /* We'll just re-encode this last macroblock if we go over the max slice size. */
1942         if( total_bits - starting_bits > slice_max_size && !h->mb.b_reencode_mb )
1943         {
1944             if( mb_xy != h->sh.i_first_mb )
1945             {
1946                 if( h->param.b_cabac )
1947                 {
1948                     memcpy( &h->cabac, &cabac_bak, offsetof(x264_cabac_t, f8_bits_encoded) );
1949                     h->cabac.p[-1] = cabac_prevbyte_bak;
1950                 }
1951                 else
1952                 {
1953                     h->out.bs = bs_bak;
1954                     i_skip = i_skip_bak;
1955                 }
1956                 h->mb.b_reencode_mb = 1;
1957                 h->sh.i_last_mb = mb_xy-1;
1958                 break;
1959             }
1960             else
1961             {
1962                 h->sh.i_last_mb = mb_xy;
1963                 h->mb.b_reencode_mb = 0;
1964             }
1965         }
1966         else
1967             h->mb.b_reencode_mb = 0;
1968
1969 #if HAVE_VISUALIZE
1970         if( h->param.b_visualize )
1971             x264_visualize_mb( h );
1972 #endif
1973
1974         /* save cache */
1975         x264_macroblock_cache_save( h );
1976
1977         /* accumulate mb stats */
1978         h->stat.frame.i_mb_count[h->mb.i_type]++;
1979
1980         int b_intra = IS_INTRA( h->mb.i_type );
1981         if( h->param.i_log_level >= X264_LOG_INFO || h->param.rc.b_stat_write )
1982         {
1983             if( !b_intra && !IS_SKIP( h->mb.i_type ) && !IS_DIRECT( h->mb.i_type ) )
1984             {
1985                 if( h->mb.i_partition != D_8x8 )
1986                         h->stat.frame.i_mb_partition[h->mb.i_partition] += 4;
1987                     else
1988                         for( int i = 0; i < 4; i++ )
1989                             h->stat.frame.i_mb_partition[h->mb.i_sub_partition[i]] ++;
1990                 if( h->param.i_frame_reference > 1 )
1991                     for( int i_list = 0; i_list <= (h->sh.i_type == SLICE_TYPE_B); i_list++ )
1992                         for( int i = 0; i < 4; i++ )
1993                         {
1994                             int i_ref = h->mb.cache.ref[i_list][ x264_scan8[4*i] ];
1995                             if( i_ref >= 0 )
1996                                 h->stat.frame.i_mb_count_ref[i_list][i_ref] ++;
1997                         }
1998             }
1999         }
2000
2001         if( h->param.i_log_level >= X264_LOG_INFO )
2002         {
2003             if( h->mb.i_cbp_luma | h->mb.i_cbp_chroma )
2004             {
2005                 int cbpsum = (h->mb.i_cbp_luma&1) + ((h->mb.i_cbp_luma>>1)&1)
2006                            + ((h->mb.i_cbp_luma>>2)&1) + (h->mb.i_cbp_luma>>3);
2007                 h->stat.frame.i_mb_cbp[!b_intra + 0] += cbpsum;
2008                 h->stat.frame.i_mb_cbp[!b_intra + 2] += !!h->mb.i_cbp_chroma;
2009                 h->stat.frame.i_mb_cbp[!b_intra + 4] += h->mb.i_cbp_chroma >> 1;
2010             }
2011             if( h->mb.i_cbp_luma && !b_intra )
2012             {
2013                 h->stat.frame.i_mb_count_8x8dct[0] ++;
2014                 h->stat.frame.i_mb_count_8x8dct[1] += h->mb.b_transform_8x8;
2015             }
2016             if( b_intra && h->mb.i_type != I_PCM )
2017             {
2018                 if( h->mb.i_type == I_16x16 )
2019                     h->stat.frame.i_mb_pred_mode[0][h->mb.i_intra16x16_pred_mode]++;
2020                 else if( h->mb.i_type == I_8x8 )
2021                     for( int i = 0; i < 16; i += 4 )
2022                         h->stat.frame.i_mb_pred_mode[1][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
2023                 else //if( h->mb.i_type == I_4x4 )
2024                     for( int i = 0; i < 16; i++ )
2025                         h->stat.frame.i_mb_pred_mode[2][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
2026                 h->stat.frame.i_mb_pred_mode[3][x264_mb_pred_mode8x8c_fix[h->mb.i_chroma_pred_mode]]++;
2027             }
2028         }
2029
2030         /* calculate deblock strength values (actual deblocking is done per-row along with hpel) */
2031         if( b_deblock )
2032         {
2033             int mvy_limit = 4 >> h->sh.b_mbaff;
2034             uint8_t (*bs)[4][4] = h->deblock_strength[h->mb.i_mb_y&h->sh.b_mbaff][h->mb.i_mb_x];
2035             x264_macroblock_cache_load_deblock( h );
2036             if( IS_INTRA( h->mb.type[h->mb.i_mb_xy] ) )
2037                 memset( bs, 3, 2*4*4*sizeof(uint8_t) );
2038             else
2039                 h->loopf.deblock_strength( h->mb.cache.non_zero_count, h->mb.cache.ref, h->mb.cache.mv,
2040                                            bs, mvy_limit, h->sh.i_type == SLICE_TYPE_B );
2041         }
2042
2043         x264_ratecontrol_mb( h, mb_size );
2044
2045         if( h->sh.b_mbaff )
2046         {
2047             i_mb_x += i_mb_y & 1;
2048             i_mb_y ^= i_mb_x < h->mb.i_mb_width;
2049         }
2050         else
2051             i_mb_x++;
2052         if( i_mb_x == h->mb.i_mb_width )
2053         {
2054             i_mb_y++;
2055             i_mb_x = 0;
2056         }
2057     }
2058     h->out.nal[h->out.i_nal].i_last_mb = h->sh.i_last_mb;
2059
2060     if( h->param.b_cabac )
2061     {
2062         x264_cabac_encode_flush( h, &h->cabac );
2063         h->out.bs.p = h->cabac.p;
2064     }
2065     else
2066     {
2067         if( i_skip > 0 )
2068             bs_write_ue( &h->out.bs, i_skip );  /* last skip run */
2069         /* rbsp_slice_trailing_bits */
2070         bs_rbsp_trailing( &h->out.bs );
2071         bs_flush( &h->out.bs );
2072     }
2073     if( x264_nal_end( h ) )
2074         return -1;
2075
2076     if( h->sh.i_last_mb == (h->i_threadslice_end * h->mb.i_mb_width - 1) )
2077     {
2078         h->stat.frame.i_misc_bits = bs_pos( &h->out.bs )
2079                                   + (h->out.i_nal*NALU_OVERHEAD * 8)
2080                                   - h->stat.frame.i_tex_bits
2081                                   - h->stat.frame.i_mv_bits;
2082         x264_fdec_filter_row( h, h->i_threadslice_end, 1 );
2083     }
2084
2085     return 0;
2086 }
2087
2088 static void x264_thread_sync_context( x264_t *dst, x264_t *src )
2089 {
2090     if( dst == src )
2091         return;
2092
2093     // reference counting
2094     for( x264_frame_t **f = src->frames.reference; *f; f++ )
2095         (*f)->i_reference_count++;
2096     for( x264_frame_t **f = dst->frames.reference; *f; f++ )
2097         x264_frame_push_unused( src, *f );
2098     src->fdec->i_reference_count++;
2099     x264_frame_push_unused( src, dst->fdec );
2100
2101     // copy everything except the per-thread pointers and the constants.
2102     memcpy( &dst->i_frame, &src->i_frame, offsetof(x264_t, mb.type) - offsetof(x264_t, i_frame) );
2103     dst->param = src->param;
2104     dst->stat = src->stat;
2105 }
2106
2107 static void x264_thread_sync_stat( x264_t *dst, x264_t *src )
2108 {
2109     if( dst == src )
2110         return;
2111     memcpy( &dst->stat.i_frame_count, &src->stat.i_frame_count, sizeof(dst->stat) - sizeof(dst->stat.frame) );
2112 }
2113
2114 static void *x264_slices_write( x264_t *h )
2115 {
2116     int i_slice_num = 0;
2117     int last_thread_mb = h->sh.i_last_mb;
2118
2119 #if HAVE_VISUALIZE
2120     if( h->param.b_visualize )
2121         if( x264_visualize_init( h ) )
2122             return (void *)-1;
2123 #endif
2124
2125     /* init stats */
2126     memset( &h->stat.frame, 0, sizeof(h->stat.frame) );
2127     h->mb.b_reencode_mb = 0;
2128     while( h->sh.i_first_mb <= last_thread_mb )
2129     {
2130         h->sh.i_last_mb = last_thread_mb;
2131         if( h->param.i_slice_max_mbs )
2132             h->sh.i_last_mb = h->sh.i_first_mb + h->param.i_slice_max_mbs - 1;
2133         else if( h->param.i_slice_count && !h->param.b_sliced_threads )
2134         {
2135             int height = h->mb.i_mb_height >> h->param.b_interlaced;
2136             int width = h->mb.i_mb_width << h->param.b_interlaced;
2137             i_slice_num++;
2138             h->sh.i_last_mb = (height * i_slice_num + h->param.i_slice_count/2) / h->param.i_slice_count * width - 1;
2139         }
2140         h->sh.i_last_mb = X264_MIN( h->sh.i_last_mb, last_thread_mb );
2141         if( x264_stack_align( x264_slice_write, h ) )
2142             return (void *)-1;
2143         h->sh.i_first_mb = h->sh.i_last_mb + 1;
2144     }
2145
2146 #if HAVE_VISUALIZE
2147     if( h->param.b_visualize )
2148     {
2149         x264_visualize_show( h );
2150         x264_visualize_close( h );
2151     }
2152 #endif
2153
2154     return (void *)0;
2155 }
2156
2157 static int x264_threaded_slices_write( x264_t *h )
2158 {
2159     /* set first/last mb and sync contexts */
2160     for( int i = 0; i < h->param.i_threads; i++ )
2161     {
2162         x264_t *t = h->thread[i];
2163         if( i )
2164         {
2165             t->param = h->param;
2166             memcpy( &t->i_frame, &h->i_frame, offsetof(x264_t, rc) - offsetof(x264_t, i_frame) );
2167         }
2168         int height = h->mb.i_mb_height >> h->param.b_interlaced;
2169         t->i_threadslice_start = ((height *  i    + h->param.i_slice_count/2) / h->param.i_threads) << h->param.b_interlaced;
2170         t->i_threadslice_end   = ((height * (i+1) + h->param.i_slice_count/2) / h->param.i_threads) << h->param.b_interlaced;
2171         t->sh.i_first_mb = t->i_threadslice_start * h->mb.i_mb_width;
2172         t->sh.i_last_mb  =   t->i_threadslice_end * h->mb.i_mb_width - 1;
2173     }
2174
2175     x264_stack_align( x264_analyse_weight_frame, h, h->mb.i_mb_height*16 + 16 );
2176
2177     x264_threads_distribute_ratecontrol( h );
2178
2179     /* dispatch */
2180     for( int i = 0; i < h->param.i_threads; i++ )
2181     {
2182         x264_threadpool_run( h->threadpool, (void*)x264_slices_write, h->thread[i] );
2183         h->thread[i]->b_thread_active = 1;
2184     }
2185     for( int i = 0; i < h->param.i_threads; i++ )
2186     {
2187         h->thread[i]->b_thread_active = 0;
2188         if( (intptr_t)x264_threadpool_wait( h->threadpool, h->thread[i] ) )
2189             return -1;
2190     }
2191
2192     /* Go back and fix up the hpel on the borders between slices. */
2193     for( int i = 1; i < h->param.i_threads; i++ )
2194     {
2195         x264_fdec_filter_row( h->thread[i], h->thread[i]->i_threadslice_start + 1, 0 );
2196         if( h->sh.b_mbaff )
2197             x264_fdec_filter_row( h->thread[i], h->thread[i]->i_threadslice_start + 2, 0 );
2198     }
2199
2200     x264_threads_merge_ratecontrol( h );
2201
2202     for( int i = 1; i < h->param.i_threads; i++ )
2203     {
2204         x264_t *t = h->thread[i];
2205         for( int j = 0; j < t->out.i_nal; j++ )
2206         {
2207             h->out.nal[h->out.i_nal] = t->out.nal[j];
2208             h->out.i_nal++;
2209             x264_nal_check_buffer( h );
2210         }
2211         /* All entries in stat.frame are ints except for ssd/ssim. */
2212         for( int j = 0; j < (offsetof(x264_t,stat.frame.i_ssd) - offsetof(x264_t,stat.frame.i_mv_bits)) / sizeof(int); j++ )
2213             ((int*)&h->stat.frame)[j] += ((int*)&t->stat.frame)[j];
2214         for( int j = 0; j < 3; j++ )
2215             h->stat.frame.i_ssd[j] += t->stat.frame.i_ssd[j];
2216         h->stat.frame.f_ssim += t->stat.frame.f_ssim;
2217     }
2218
2219     return 0;
2220 }
2221
2222 void x264_encoder_intra_refresh( x264_t *h )
2223 {
2224     h = h->thread[h->i_thread_phase];
2225     h->b_queued_intra_refresh = 1;
2226 }
2227
2228 int x264_encoder_invalidate_reference( x264_t *h, int64_t pts )
2229 {
2230     if( h->param.i_bframe )
2231     {
2232         x264_log( h, X264_LOG_ERROR, "x264_encoder_invalidate_reference is not supported with B-frames enabled\n" );
2233         return -1;
2234     }
2235     if( h->param.b_intra_refresh )
2236     {
2237         x264_log( h, X264_LOG_ERROR, "x264_encoder_invalidate_reference is not supported with intra refresh enabled\n" );
2238         return -1;
2239     }
2240     h = h->thread[h->i_thread_phase];
2241     if( pts >= h->i_last_idr_pts )
2242     {
2243         for( int i = 0; h->frames.reference[i]; i++ )
2244             if( pts <= h->frames.reference[i]->i_pts )
2245                 h->frames.reference[i]->b_corrupt = 1;
2246         if( pts <= h->fdec->i_pts )
2247             h->fdec->b_corrupt = 1;
2248     }
2249     return 0;
2250 }
2251
2252 /****************************************************************************
2253  * x264_encoder_encode:
2254  *  XXX: i_poc   : is the poc of the current given picture
2255  *       i_frame : is the number of the frame being coded
2256  *  ex:  type frame poc
2257  *       I      0   2*0
2258  *       P      1   2*3
2259  *       B      2   2*1
2260  *       B      3   2*2
2261  *       P      4   2*6
2262  *       B      5   2*4
2263  *       B      6   2*5
2264  ****************************************************************************/
2265 int     x264_encoder_encode( x264_t *h,
2266                              x264_nal_t **pp_nal, int *pi_nal,
2267                              x264_picture_t *pic_in,
2268                              x264_picture_t *pic_out )
2269 {
2270     x264_t *thread_current, *thread_prev, *thread_oldest;
2271     int i_nal_type, i_nal_ref_idc, i_global_qp;
2272     int overhead = NALU_OVERHEAD;
2273
2274     if( h->i_thread_frames > 1 )
2275     {
2276         thread_prev    = h->thread[ h->i_thread_phase ];
2277         h->i_thread_phase = (h->i_thread_phase + 1) % h->i_thread_frames;
2278         thread_current = h->thread[ h->i_thread_phase ];
2279         thread_oldest  = h->thread[ (h->i_thread_phase + 1) % h->i_thread_frames ];
2280         x264_thread_sync_context( thread_current, thread_prev );
2281         x264_thread_sync_ratecontrol( thread_current, thread_prev, thread_oldest );
2282         h = thread_current;
2283     }
2284     else
2285     {
2286         thread_current =
2287         thread_oldest  = h;
2288     }
2289 #if HAVE_MMX
2290     if( h->i_thread_frames == 1 && h->param.cpu&X264_CPU_SSE_MISALIGN )
2291         x264_cpu_mask_misalign_sse();
2292 #endif
2293
2294     // ok to call this before encoding any frames, since the initial values of fdec have b_kept_as_ref=0
2295     if( x264_reference_update( h ) )
2296         return -1;
2297     h->fdec->i_lines_completed = -1;
2298
2299     /* no data out */
2300     *pi_nal = 0;
2301     *pp_nal = NULL;
2302
2303     /* ------------------- Setup new frame from picture -------------------- */
2304     if( pic_in != NULL )
2305     {
2306         /* 1: Copy the picture to a frame and move it to a buffer */
2307         x264_frame_t *fenc = x264_frame_pop_unused( h, 0 );
2308         if( !fenc )
2309             return -1;
2310
2311         if( x264_frame_copy_picture( h, fenc, pic_in ) < 0 )
2312             return -1;
2313
2314         if( h->param.i_width != 16 * h->mb.i_mb_width ||
2315             h->param.i_height != 16 * h->mb.i_mb_height )
2316             x264_frame_expand_border_mod16( h, fenc );
2317
2318         fenc->i_frame = h->frames.i_input++;
2319
2320         if( h->frames.i_bframe_delay && fenc->i_frame == h->frames.i_bframe_delay )
2321             h->frames.i_bframe_delay_time = fenc->i_pts;
2322
2323         if( h->param.b_vfr_input && fenc->i_pts <= h->frames.i_largest_pts )
2324             x264_log( h, X264_LOG_WARNING, "non-strictly-monotonic PTS\n" );
2325
2326         h->frames.i_second_largest_pts = h->frames.i_largest_pts;
2327         h->frames.i_largest_pts = fenc->i_pts;
2328
2329         if( (fenc->i_pic_struct < PIC_STRUCT_AUTO) || (fenc->i_pic_struct > PIC_STRUCT_TRIPLE) )
2330             fenc->i_pic_struct = PIC_STRUCT_AUTO;
2331
2332         if( fenc->i_pic_struct == PIC_STRUCT_AUTO )
2333         {
2334             int b_interlaced = fenc->param ? fenc->param->b_interlaced : h->param.b_interlaced;
2335             if( b_interlaced )
2336             {
2337                 int b_tff = fenc->param ? fenc->param->b_tff : h->param.b_tff;
2338                 fenc->i_pic_struct = b_tff ? PIC_STRUCT_TOP_BOTTOM : PIC_STRUCT_BOTTOM_TOP;
2339             }
2340             else
2341                 fenc->i_pic_struct = PIC_STRUCT_PROGRESSIVE;
2342         }
2343
2344         if( h->param.rc.b_mb_tree && h->param.rc.b_stat_read )
2345         {
2346             if( x264_macroblock_tree_read( h, fenc, pic_in->prop.quant_offsets ) )
2347                 return -1;
2348         }
2349         else
2350             x264_stack_align( x264_adaptive_quant_frame, h, fenc, pic_in->prop.quant_offsets );
2351
2352         if( pic_in->prop.quant_offsets_free )
2353             pic_in->prop.quant_offsets_free( pic_in->prop.quant_offsets );
2354
2355         if( h->frames.b_have_lowres )
2356             x264_frame_init_lowres( h, fenc );
2357
2358         /* 2: Place the frame into the queue for its slice type decision */
2359         x264_lookahead_put_frame( h, fenc );
2360
2361         if( h->frames.i_input <= h->frames.i_delay + 1 - h->i_thread_frames )
2362         {
2363             /* Nothing yet to encode, waiting for filling of buffers */
2364             pic_out->i_type = X264_TYPE_AUTO;
2365             return 0;
2366         }
2367     }
2368     else
2369     {
2370         /* signal kills for lookahead thread */
2371         x264_pthread_mutex_lock( &h->lookahead->ifbuf.mutex );
2372         h->lookahead->b_exit_thread = 1;
2373         x264_pthread_cond_broadcast( &h->lookahead->ifbuf.cv_fill );
2374         x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
2375     }
2376
2377     h->i_frame++;
2378     /* 3: The picture is analyzed in the lookahead */
2379     if( !h->frames.current[0] )
2380         x264_lookahead_get_frames( h );
2381
2382     if( !h->frames.current[0] && x264_lookahead_is_empty( h ) )
2383         return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
2384
2385     /* ------------------- Get frame to be encoded ------------------------- */
2386     /* 4: get picture to encode */
2387     h->fenc = x264_frame_shift( h->frames.current );
2388     if( h->i_frame == h->i_thread_frames - 1 )
2389         h->i_reordered_pts_delay = h->fenc->i_reordered_pts;
2390     if( h->fenc->param )
2391     {
2392         x264_encoder_reconfig( h, h->fenc->param );
2393         if( h->fenc->param->param_free )
2394             h->fenc->param->param_free( h->fenc->param );
2395     }
2396
2397     if( !IS_X264_TYPE_I( h->fenc->i_type ) )
2398     {
2399         int valid_refs_left = 0;
2400         for( int i = 0; h->frames.reference[i]; i++ )
2401             if( !h->frames.reference[i]->b_corrupt )
2402                 valid_refs_left++;
2403         /* No valid reference frames left: force an IDR. */
2404         if( !valid_refs_left )
2405         {
2406             h->fenc->b_keyframe = 1;
2407             h->fenc->i_type = X264_TYPE_IDR;
2408         }
2409     }
2410
2411     if( h->fenc->b_keyframe )
2412     {
2413         h->frames.i_last_keyframe = h->fenc->i_frame;
2414         if( h->fenc->i_type == X264_TYPE_IDR )
2415         {
2416             h->i_frame_num = 0;
2417             h->frames.i_last_idr = h->fenc->i_frame;
2418         }
2419     }
2420     h->sh.i_mmco_command_count =
2421     h->sh.i_mmco_remove_from_end = 0;
2422     h->b_ref_reorder[0] =
2423     h->b_ref_reorder[1] = 0;
2424     h->fdec->i_poc =
2425     h->fenc->i_poc = 2 * ( h->fenc->i_frame - X264_MAX( h->frames.i_last_idr, 0 ) );
2426
2427     /* ------------------- Setup frame context ----------------------------- */
2428     /* 5: Init data dependent of frame type */
2429     if( h->fenc->i_type == X264_TYPE_IDR )
2430     {
2431         /* reset ref pictures */
2432         i_nal_type    = NAL_SLICE_IDR;
2433         i_nal_ref_idc = NAL_PRIORITY_HIGHEST;
2434         h->sh.i_type = SLICE_TYPE_I;
2435         x264_reference_reset( h );
2436         h->frames.i_poc_last_open_gop = -1;
2437     }
2438     else if( h->fenc->i_type == X264_TYPE_I )
2439     {
2440         i_nal_type    = NAL_SLICE;
2441         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
2442         h->sh.i_type = SLICE_TYPE_I;
2443         x264_reference_hierarchy_reset( h );
2444         if( h->param.i_open_gop )
2445             h->frames.i_poc_last_open_gop = h->fenc->b_keyframe ? h->fenc->i_poc : -1;
2446     }
2447     else if( h->fenc->i_type == X264_TYPE_P )
2448     {
2449         i_nal_type    = NAL_SLICE;
2450         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
2451         h->sh.i_type = SLICE_TYPE_P;
2452         x264_reference_hierarchy_reset( h );
2453         h->frames.i_poc_last_open_gop = -1;
2454     }
2455     else if( h->fenc->i_type == X264_TYPE_BREF )
2456     {
2457         i_nal_type    = NAL_SLICE;
2458         i_nal_ref_idc = h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT ? NAL_PRIORITY_LOW : NAL_PRIORITY_HIGH;
2459         h->sh.i_type = SLICE_TYPE_B;
2460         x264_reference_hierarchy_reset( h );
2461     }
2462     else    /* B frame */
2463     {
2464         i_nal_type    = NAL_SLICE;
2465         i_nal_ref_idc = NAL_PRIORITY_DISPOSABLE;
2466         h->sh.i_type = SLICE_TYPE_B;
2467     }
2468
2469     h->fdec->i_type = h->fenc->i_type;
2470     h->fdec->i_frame = h->fenc->i_frame;
2471     h->fenc->b_kept_as_ref =
2472     h->fdec->b_kept_as_ref = i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE && h->param.i_keyint_max > 1;
2473
2474     h->fdec->i_pts = h->fenc->i_pts *= h->i_dts_compress_multiplier;
2475     if( h->frames.i_bframe_delay )
2476     {
2477         int64_t *prev_reordered_pts = thread_current->frames.i_prev_reordered_pts;
2478         if( h->i_frame <= h->frames.i_bframe_delay )
2479         {
2480             if( h->i_dts_compress_multiplier == 1 )
2481                 h->fdec->i_dts = h->fenc->i_reordered_pts - h->frames.i_bframe_delay_time;
2482             else
2483             {
2484                 /* DTS compression */
2485                 if( h->i_frame == 1 )
2486                     thread_current->frames.i_init_delta = h->fenc->i_reordered_pts * h->i_dts_compress_multiplier;
2487                 h->fdec->i_dts = h->i_frame * thread_current->frames.i_init_delta / h->i_dts_compress_multiplier;
2488             }
2489         }
2490         else
2491             h->fdec->i_dts = prev_reordered_pts[ (h->i_frame - h->frames.i_bframe_delay) % h->frames.i_bframe_delay ];
2492         prev_reordered_pts[ h->i_frame % h->frames.i_bframe_delay ] = h->fenc->i_reordered_pts * h->i_dts_compress_multiplier;
2493     }
2494     else
2495         h->fdec->i_dts = h->fenc->i_reordered_pts;
2496     if( h->fenc->i_type == X264_TYPE_IDR )
2497         h->i_last_idr_pts = h->fdec->i_pts;
2498
2499     /* ------------------- Init                ----------------------------- */
2500     /* build ref list 0/1 */
2501     x264_reference_build_list( h, h->fdec->i_poc );
2502
2503     /* ---------------------- Write the bitstream -------------------------- */
2504     /* Init bitstream context */
2505     if( h->param.b_sliced_threads )
2506     {
2507         for( int i = 0; i < h->param.i_threads; i++ )
2508         {
2509             bs_init( &h->thread[i]->out.bs, h->thread[i]->out.p_bitstream, h->thread[i]->out.i_bitstream );
2510             h->thread[i]->out.i_nal = 0;
2511         }
2512     }
2513     else
2514     {
2515         bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
2516         h->out.i_nal = 0;
2517     }
2518
2519     if( h->param.b_aud )
2520     {
2521         int pic_type;
2522
2523         if( h->sh.i_type == SLICE_TYPE_I )
2524             pic_type = 0;
2525         else if( h->sh.i_type == SLICE_TYPE_P )
2526             pic_type = 1;
2527         else if( h->sh.i_type == SLICE_TYPE_B )
2528             pic_type = 2;
2529         else
2530             pic_type = 7;
2531
2532         x264_nal_start( h, NAL_AUD, NAL_PRIORITY_DISPOSABLE );
2533         bs_write( &h->out.bs, 3, pic_type );
2534         bs_rbsp_trailing( &h->out.bs );
2535         if( x264_nal_end( h ) )
2536             return -1;
2537         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2538     }
2539
2540     h->i_nal_type = i_nal_type;
2541     h->i_nal_ref_idc = i_nal_ref_idc;
2542
2543     if( h->param.b_intra_refresh )
2544     {
2545         if( IS_X264_TYPE_I( h->fenc->i_type ) )
2546         {
2547             h->fdec->i_frames_since_pir = 0;
2548             h->b_queued_intra_refresh = 0;
2549             /* PIR is currently only supported with ref == 1, so any intra frame effectively refreshes
2550              * the whole frame and counts as an intra refresh. */
2551             h->fdec->f_pir_position = h->mb.i_mb_width;
2552         }
2553         else if( h->fenc->i_type == X264_TYPE_P )
2554         {
2555             int pocdiff = (h->fdec->i_poc - h->fref0[0]->i_poc)/2;
2556             float increment = X264_MAX( ((float)h->mb.i_mb_width-1) / h->param.i_keyint_max, 1 );
2557             h->fdec->f_pir_position = h->fref0[0]->f_pir_position;
2558             h->fdec->i_frames_since_pir = h->fref0[0]->i_frames_since_pir + pocdiff;
2559             if( h->fdec->i_frames_since_pir >= h->param.i_keyint_max ||
2560                 (h->b_queued_intra_refresh && h->fdec->f_pir_position + 0.5 >= h->mb.i_mb_width) )
2561             {
2562                 h->fdec->f_pir_position = 0;
2563                 h->fdec->i_frames_since_pir = 0;
2564                 h->b_queued_intra_refresh = 0;
2565                 h->fenc->b_keyframe = 1;
2566             }
2567             h->fdec->i_pir_start_col = h->fdec->f_pir_position+0.5;
2568             h->fdec->f_pir_position += increment * pocdiff;
2569             h->fdec->i_pir_end_col = h->fdec->f_pir_position+0.5;
2570         }
2571     }
2572
2573     if( h->fenc->b_keyframe )
2574     {
2575         /* Write SPS and PPS */
2576         if( h->param.b_repeat_headers )
2577         {
2578             /* generate sequence parameters */
2579             x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
2580             x264_sps_write( &h->out.bs, h->sps );
2581             if( x264_nal_end( h ) )
2582                 return -1;
2583             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
2584
2585             /* generate picture parameters */
2586             x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
2587             x264_pps_write( &h->out.bs, h->pps );
2588             if( x264_nal_end( h ) )
2589                 return -1;
2590             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
2591         }
2592
2593         /* buffering period sei is written in x264_encoder_frame_end */
2594
2595         if( h->param.b_repeat_headers && h->fenc->i_frame == 0 )
2596         {
2597             /* identify ourself */
2598             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2599             if( x264_sei_version_write( h, &h->out.bs ) )
2600                 return -1;
2601             if( x264_nal_end( h ) )
2602                 return -1;
2603             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2604         }
2605
2606         if( h->fenc->i_type != X264_TYPE_IDR )
2607         {
2608             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;
2609             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2610             x264_sei_recovery_point_write( h, &h->out.bs, time_to_recovery );
2611             x264_nal_end( h );
2612             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2613         }
2614     }
2615
2616     /* generate sei pic timing */
2617     if( h->sps->vui.b_pic_struct_present || h->sps->vui.b_nal_hrd_parameters_present )
2618     {
2619         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2620         x264_sei_pic_timing_write( h, &h->out.bs );
2621         if( x264_nal_end( h ) )
2622             return -1;
2623         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2624     }
2625
2626     /* Init the rate control */
2627     /* FIXME: Include slice header bit cost. */
2628     x264_ratecontrol_start( h, h->fenc->i_qpplus1, overhead*8 );
2629     i_global_qp = x264_ratecontrol_qp( h );
2630
2631     pic_out->i_qpplus1 =
2632     h->fdec->i_qpplus1 = i_global_qp + 1;
2633
2634     if( h->param.rc.b_stat_read && h->sh.i_type != SLICE_TYPE_I )
2635     {
2636         x264_reference_build_list_optimal( h );
2637         x264_reference_check_reorder( h );
2638     }
2639
2640     if( h->i_ref0 )
2641         h->fdec->i_poc_l0ref0 = h->fref0[0]->i_poc;
2642
2643     if( h->sh.i_type == SLICE_TYPE_B )
2644         x264_macroblock_bipred_init( h );
2645
2646     /*------------------------- Weights -------------------------------------*/
2647     x264_weighted_pred_init( h );
2648
2649     /* ------------------------ Create slice header  ----------------------- */
2650     x264_slice_init( h, i_nal_type, i_global_qp );
2651
2652     if( i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE )
2653         h->i_frame_num++;
2654
2655     /* Write frame */
2656     h->i_threadslice_start = 0;
2657     h->i_threadslice_end = h->mb.i_mb_height;
2658     if( h->i_thread_frames > 1 )
2659     {
2660         x264_threadpool_run( h->threadpool, (void*)x264_slices_write, h );
2661         h->b_thread_active = 1;
2662     }
2663     else if( h->param.b_sliced_threads )
2664     {
2665         if( x264_threaded_slices_write( h ) )
2666             return -1;
2667     }
2668     else
2669         if( (intptr_t)x264_slices_write( h ) )
2670             return -1;
2671
2672     return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
2673 }
2674
2675 static int x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
2676                                    x264_nal_t **pp_nal, int *pi_nal,
2677                                    x264_picture_t *pic_out )
2678 {
2679     char psz_message[80];
2680
2681     if( h->b_thread_active )
2682     {
2683         h->b_thread_active = 0;
2684         if( (intptr_t)x264_threadpool_wait( h->threadpool, h ) )
2685             return -1;
2686     }
2687     if( !h->out.i_nal )
2688     {
2689         pic_out->i_type = X264_TYPE_AUTO;
2690         return 0;
2691     }
2692
2693     x264_emms();
2694     /* generate sei buffering period and insert it into place */
2695     if( h->fenc->b_keyframe && h->sps->vui.b_nal_hrd_parameters_present )
2696     {
2697         x264_hrd_fullness( h );
2698         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2699         x264_sei_buffering_period_write( h, &h->out.bs );
2700         if( x264_nal_end( h ) )
2701            return -1;
2702         /* buffering period sei must follow AUD, SPS and PPS and precede all other SEIs */
2703         int idx = 0;
2704         while( h->out.nal[idx].i_type == NAL_AUD ||
2705                h->out.nal[idx].i_type == NAL_SPS ||
2706                h->out.nal[idx].i_type == NAL_PPS )
2707             idx++;
2708         x264_nal_t nal_tmp = h->out.nal[h->out.i_nal-1];
2709         memmove( &h->out.nal[idx+1], &h->out.nal[idx], (h->out.i_nal-idx-1)*sizeof(x264_nal_t) );
2710         h->out.nal[idx] = nal_tmp;
2711     }
2712
2713     int frame_size = x264_encoder_encapsulate_nals( h, 0 );
2714
2715     /* Set output picture properties */
2716     pic_out->i_type = h->fenc->i_type;
2717
2718     pic_out->b_keyframe = h->fenc->b_keyframe;
2719
2720     pic_out->i_pts = h->fdec->i_pts;
2721     pic_out->i_dts = h->fdec->i_dts;
2722
2723     if( pic_out->i_pts < pic_out->i_dts )
2724         x264_log( h, X264_LOG_WARNING, "invalid DTS: PTS is less than DTS\n" );
2725
2726     pic_out->img.i_csp = X264_CSP_NV12;
2727     pic_out->img.i_plane = h->fdec->i_plane;
2728     for( int i = 0; i < 2; i++ )
2729     {
2730         pic_out->img.i_stride[i] = h->fdec->i_stride[i];
2731         // FIXME This breaks the API when pixel != uint8_t.
2732         pic_out->img.plane[i] = h->fdec->plane[i];
2733     }
2734
2735     x264_frame_push_unused( thread_current, h->fenc );
2736
2737     /* ---------------------- Update encoder state ------------------------- */
2738
2739     /* update rc */
2740     int filler = 0;
2741     if( x264_ratecontrol_end( h, frame_size * 8, &filler ) < 0 )
2742         return -1;
2743
2744     pic_out->hrd_timing = h->fenc->hrd_timing;
2745
2746     while( filler > 0 )
2747     {
2748         int f, overhead;
2749         overhead = (FILLER_OVERHEAD - h->param.b_annexb);
2750         if( h->param.i_slice_max_size && filler > h->param.i_slice_max_size )
2751         {
2752             int next_size = filler - h->param.i_slice_max_size;
2753             int overflow = X264_MAX( overhead - next_size, 0 );
2754             f = h->param.i_slice_max_size - overhead - overflow;
2755         }
2756         else
2757             f = X264_MAX( 0, filler - overhead );
2758
2759         x264_nal_start( h, NAL_FILLER, NAL_PRIORITY_DISPOSABLE );
2760         x264_filler_write( h, &h->out.bs, f );
2761         if( x264_nal_end( h ) )
2762             return -1;
2763         int total_size = x264_encoder_encapsulate_nals( h, h->out.i_nal-1 );
2764         frame_size += total_size;
2765         filler -= total_size;
2766     }
2767
2768     /* End bitstream, set output  */
2769     *pi_nal = h->out.i_nal;
2770     *pp_nal = h->out.nal;
2771
2772     h->out.i_nal = 0;
2773
2774     x264_noise_reduction_update( h );
2775
2776     /* ---------------------- Compute/Print statistics --------------------- */
2777     x264_thread_sync_stat( h, h->thread[0] );
2778
2779     /* Slice stat */
2780     h->stat.i_frame_count[h->sh.i_type]++;
2781     h->stat.i_frame_size[h->sh.i_type] += frame_size;
2782     h->stat.f_frame_qp[h->sh.i_type] += h->fdec->f_qp_avg_aq;
2783
2784     for( int i = 0; i < X264_MBTYPE_MAX; i++ )
2785         h->stat.i_mb_count[h->sh.i_type][i] += h->stat.frame.i_mb_count[i];
2786     for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
2787         h->stat.i_mb_partition[h->sh.i_type][i] += h->stat.frame.i_mb_partition[i];
2788     for( int i = 0; i < 2; i++ )
2789         h->stat.i_mb_count_8x8dct[i] += h->stat.frame.i_mb_count_8x8dct[i];
2790     for( int i = 0; i < 6; i++ )
2791         h->stat.i_mb_cbp[i] += h->stat.frame.i_mb_cbp[i];
2792     for( int i = 0; i < 4; i++ )
2793         for( int j = 0; j < 13; j++ )
2794             h->stat.i_mb_pred_mode[i][j] += h->stat.frame.i_mb_pred_mode[i][j];
2795     if( h->sh.i_type != SLICE_TYPE_I )
2796         for( int i_list = 0; i_list < 2; i_list++ )
2797             for( int i = 0; i < 32; i++ )
2798                 h->stat.i_mb_count_ref[h->sh.i_type][i_list][i] += h->stat.frame.i_mb_count_ref[i_list][i];
2799     if( h->sh.i_type == SLICE_TYPE_P )
2800     {
2801         h->stat.i_consecutive_bframes[h->fdec->i_frame - h->fref0[0]->i_frame - 1]++;
2802         if( h->param.analyse.i_weighted_pred == X264_WEIGHTP_SMART )
2803             for( int i = 0; i < 3; i++ )
2804                 for( int j = 0; j < h->i_ref0; j++ )
2805                     if( h->sh.weight[0][i].i_denom != 0 )
2806                     {
2807                         h->stat.i_wpred[i]++;
2808                         break;
2809                     }
2810     }
2811     if( h->sh.i_type == SLICE_TYPE_B )
2812     {
2813         h->stat.i_direct_frames[ h->sh.b_direct_spatial_mv_pred ] ++;
2814         if( h->mb.b_direct_auto_write )
2815         {
2816             //FIXME somewhat arbitrary time constants
2817             if( h->stat.i_direct_score[0] + h->stat.i_direct_score[1] > h->mb.i_mb_count )
2818                 for( int i = 0; i < 2; i++ )
2819                     h->stat.i_direct_score[i] = h->stat.i_direct_score[i] * 9/10;
2820             for( int i = 0; i < 2; i++ )
2821                 h->stat.i_direct_score[i] += h->stat.frame.i_direct_score[i];
2822         }
2823     }
2824
2825     psz_message[0] = '\0';
2826     if( h->param.analyse.b_psnr )
2827     {
2828         int64_t ssd[3] = {
2829             h->stat.frame.i_ssd[0],
2830             h->stat.frame.i_ssd[1],
2831             h->stat.frame.i_ssd[2],
2832         };
2833
2834         h->stat.i_ssd_global[h->sh.i_type] += ssd[0] + ssd[1] + ssd[2];
2835         h->stat.f_psnr_average[h->sh.i_type] += x264_psnr( ssd[0] + ssd[1] + ssd[2], 3 * h->param.i_width * h->param.i_height / 2 );
2836         h->stat.f_psnr_mean_y[h->sh.i_type] += x264_psnr( ssd[0], h->param.i_width * h->param.i_height );
2837         h->stat.f_psnr_mean_u[h->sh.i_type] += x264_psnr( ssd[1], h->param.i_width * h->param.i_height / 4 );
2838         h->stat.f_psnr_mean_v[h->sh.i_type] += x264_psnr( ssd[2], h->param.i_width * h->param.i_height / 4 );
2839
2840         snprintf( psz_message, 80, " PSNR Y:%5.2f U:%5.2f V:%5.2f",
2841                   x264_psnr( ssd[0], h->param.i_width * h->param.i_height ),
2842                   x264_psnr( ssd[1], h->param.i_width * h->param.i_height / 4),
2843                   x264_psnr( ssd[2], h->param.i_width * h->param.i_height / 4) );
2844     }
2845
2846     if( h->param.analyse.b_ssim )
2847     {
2848         double ssim_y = h->stat.frame.f_ssim
2849                       / (((h->param.i_width-6)>>2) * ((h->param.i_height-6)>>2));
2850         h->stat.f_ssim_mean_y[h->sh.i_type] += ssim_y;
2851         snprintf( psz_message + strlen(psz_message), 80 - strlen(psz_message),
2852                   " SSIM Y:%.5f", ssim_y );
2853     }
2854     psz_message[79] = '\0';
2855
2856     x264_log( h, X264_LOG_DEBUG,
2857                   "frame=%4d QP=%.2f NAL=%d Slice:%c Poc:%-3d I:%-4d P:%-4d SKIP:%-4d size=%d bytes%s\n",
2858               h->i_frame,
2859               h->fdec->f_qp_avg_aq,
2860               h->i_nal_ref_idc,
2861               h->sh.i_type == SLICE_TYPE_I ? 'I' : (h->sh.i_type == SLICE_TYPE_P ? 'P' : 'B' ),
2862               h->fdec->i_poc,
2863               h->stat.frame.i_mb_count_i,
2864               h->stat.frame.i_mb_count_p,
2865               h->stat.frame.i_mb_count_skip,
2866               frame_size,
2867               psz_message );
2868
2869     // keep stats all in one place
2870     x264_thread_sync_stat( h->thread[0], h );
2871     // for the use of the next frame
2872     x264_thread_sync_stat( thread_current, h );
2873
2874 #ifdef DEBUG_MB_TYPE
2875 {
2876     static const char mb_chars[] = { 'i', 'i', 'I', 'C', 'P', '8', 'S',
2877         'D', '<', 'X', 'B', 'X', '>', 'B', 'B', 'B', 'B', '8', 'S' };
2878     for( int mb_xy = 0; mb_xy < h->mb.i_mb_width * h->mb.i_mb_height; mb_xy++ )
2879     {
2880         if( h->mb.type[mb_xy] < X264_MBTYPE_MAX && h->mb.type[mb_xy] >= 0 )
2881             fprintf( stderr, "%c ", mb_chars[ h->mb.type[mb_xy] ] );
2882         else
2883             fprintf( stderr, "? " );
2884
2885         if( (mb_xy+1) % h->mb.i_mb_width == 0 )
2886             fprintf( stderr, "\n" );
2887     }
2888 }
2889 #endif
2890
2891     /* Remove duplicates, must be done near the end as breaks h->fref0 array
2892      * by freeing some of its pointers. */
2893      for( int i = 0; i < h->i_ref0; i++ )
2894          if( h->fref0[i] && h->fref0[i]->b_duplicate )
2895          {
2896              x264_frame_push_blank_unused( h, h->fref0[i] );
2897              h->fref0[i] = 0;
2898          }
2899
2900     if( h->param.psz_dump_yuv )
2901         x264_frame_dump( h );
2902
2903     return frame_size;
2904 }
2905
2906 static void x264_print_intra( int64_t *i_mb_count, double i_count, int b_print_pcm, char *intra )
2907 {
2908     intra += sprintf( intra, "I16..4%s: %4.1f%% %4.1f%% %4.1f%%",
2909         b_print_pcm ? "..PCM" : "",
2910         i_mb_count[I_16x16]/ i_count,
2911         i_mb_count[I_8x8]  / i_count,
2912         i_mb_count[I_4x4]  / i_count );
2913     if( b_print_pcm )
2914         sprintf( intra, " %4.1f%%", i_mb_count[I_PCM]  / i_count );
2915 }
2916
2917 /****************************************************************************
2918  * x264_encoder_close:
2919  ****************************************************************************/
2920 void    x264_encoder_close  ( x264_t *h )
2921 {
2922     int64_t i_yuv_size = 3 * h->param.i_width * h->param.i_height / 2;
2923     int64_t i_mb_count_size[2][7] = {{0}};
2924     char buf[200];
2925     int b_print_pcm = h->stat.i_mb_count[SLICE_TYPE_I][I_PCM]
2926                    || h->stat.i_mb_count[SLICE_TYPE_P][I_PCM]
2927                    || h->stat.i_mb_count[SLICE_TYPE_B][I_PCM];
2928
2929     x264_lookahead_delete( h );
2930
2931     if( h->param.i_threads > 1 )
2932         x264_threadpool_delete( h->threadpool );
2933     if( h->i_thread_frames > 1 )
2934     {
2935         for( int i = 0; i < h->i_thread_frames; i++ )
2936             if( h->thread[i]->b_thread_active )
2937             {
2938                 assert( h->thread[i]->fenc->i_reference_count == 1 );
2939                 x264_frame_delete( h->thread[i]->fenc );
2940             }
2941
2942         x264_t *thread_prev = h->thread[h->i_thread_phase];
2943         x264_thread_sync_ratecontrol( h, thread_prev, h );
2944         x264_thread_sync_ratecontrol( thread_prev, thread_prev, h );
2945         h->i_frame = thread_prev->i_frame + 1 - h->i_thread_frames;
2946     }
2947     h->i_frame++;
2948
2949     /* Slices used and PSNR */
2950     for( int i = 0; i < 3; i++ )
2951     {
2952         static const uint8_t slice_order[] = { SLICE_TYPE_I, SLICE_TYPE_P, SLICE_TYPE_B };
2953         int i_slice = slice_order[i];
2954
2955         if( h->stat.i_frame_count[i_slice] > 0 )
2956         {
2957             const int i_count = h->stat.i_frame_count[i_slice];
2958             if( h->param.analyse.b_psnr )
2959             {
2960                 x264_log( h, X264_LOG_INFO,
2961                           "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",
2962                           slice_type_to_char[i_slice],
2963                           i_count,
2964                           h->stat.f_frame_qp[i_slice] / i_count,
2965                           (double)h->stat.i_frame_size[i_slice] / i_count,
2966                           h->stat.f_psnr_mean_y[i_slice] / i_count, h->stat.f_psnr_mean_u[i_slice] / i_count, h->stat.f_psnr_mean_v[i_slice] / i_count,
2967                           h->stat.f_psnr_average[i_slice] / i_count,
2968                           x264_psnr( h->stat.i_ssd_global[i_slice], i_count * i_yuv_size ) );
2969             }
2970             else
2971             {
2972                 x264_log( h, X264_LOG_INFO,
2973                           "frame %c:%-5d Avg QP:%5.2f  size:%6.0f\n",
2974                           slice_type_to_char[i_slice],
2975                           i_count,
2976                           h->stat.f_frame_qp[i_slice] / i_count,
2977                           (double)h->stat.i_frame_size[i_slice] / i_count );
2978             }
2979         }
2980     }
2981     if( h->param.i_bframe && h->stat.i_frame_count[SLICE_TYPE_P] )
2982     {
2983         char *p = buf;
2984         int den = 0;
2985         // weight by number of frames (including the P-frame) that are in a sequence of N B-frames
2986         for( int i = 0; i <= h->param.i_bframe; i++ )
2987             den += (i+1) * h->stat.i_consecutive_bframes[i];
2988         for( int i = 0; i <= h->param.i_bframe; i++ )
2989             p += sprintf( p, " %4.1f%%", 100. * (i+1) * h->stat.i_consecutive_bframes[i] / den );
2990         x264_log( h, X264_LOG_INFO, "consecutive B-frames:%s\n", buf );
2991     }
2992
2993     for( int i_type = 0; i_type < 2; i_type++ )
2994         for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
2995         {
2996             if( i == D_DIRECT_8x8 ) continue; /* direct is counted as its own type */
2997             i_mb_count_size[i_type][x264_mb_partition_pixel_table[i]] += h->stat.i_mb_partition[i_type][i];
2998         }
2999
3000     /* MB types used */
3001     if( h->stat.i_frame_count[SLICE_TYPE_I] > 0 )
3002     {
3003         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_I];
3004         double i_count = h->stat.i_frame_count[SLICE_TYPE_I] * h->mb.i_mb_count / 100.0;
3005         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
3006         x264_log( h, X264_LOG_INFO, "mb I  %s\n", buf );
3007     }
3008     if( h->stat.i_frame_count[SLICE_TYPE_P] > 0 )
3009     {
3010         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_P];
3011         double i_count = h->stat.i_frame_count[SLICE_TYPE_P] * h->mb.i_mb_count / 100.0;
3012         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_P];
3013         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
3014         x264_log( h, X264_LOG_INFO,
3015                   "mb P  %s  P16..4: %4.1f%% %4.1f%% %4.1f%% %4.1f%% %4.1f%%    skip:%4.1f%%\n",
3016                   buf,
3017                   i_mb_size[PIXEL_16x16] / (i_count*4),
3018                   (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
3019                   i_mb_size[PIXEL_8x8] / (i_count*4),
3020                   (i_mb_size[PIXEL_8x4] + i_mb_size[PIXEL_4x8]) / (i_count*4),
3021                   i_mb_size[PIXEL_4x4] / (i_count*4),
3022                   i_mb_count[P_SKIP] / i_count );
3023     }
3024     if( h->stat.i_frame_count[SLICE_TYPE_B] > 0 )
3025     {
3026         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_B];
3027         double i_count = h->stat.i_frame_count[SLICE_TYPE_B] * h->mb.i_mb_count / 100.0;
3028         double i_mb_list_count;
3029         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_B];
3030         int64_t list_count[3] = {0}; /* 0 == L0, 1 == L1, 2 == BI */
3031         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
3032         for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
3033             for( int j = 0; j < 2; j++ )
3034             {
3035                 int l0 = x264_mb_type_list_table[i][0][j];
3036                 int l1 = x264_mb_type_list_table[i][1][j];
3037                 if( l0 || l1 )
3038                     list_count[l1+l0*l1] += h->stat.i_mb_count[SLICE_TYPE_B][i] * 2;
3039             }
3040         list_count[0] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L0_8x8];
3041         list_count[1] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L1_8x8];
3042         list_count[2] += h->stat.i_mb_partition[SLICE_TYPE_B][D_BI_8x8];
3043         i_mb_count[B_DIRECT] += (h->stat.i_mb_partition[SLICE_TYPE_B][D_DIRECT_8x8]+2)/4;
3044         i_mb_list_count = (list_count[0] + list_count[1] + list_count[2]) / 100.0;
3045         x264_log( h, X264_LOG_INFO,
3046                   "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",
3047                   buf,
3048                   i_mb_size[PIXEL_16x16] / (i_count*4),
3049                   (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
3050                   i_mb_size[PIXEL_8x8] / (i_count*4),
3051                   i_mb_count[B_DIRECT] / i_count,
3052                   i_mb_count[B_SKIP]   / i_count,
3053                   list_count[0] / i_mb_list_count,
3054                   list_count[1] / i_mb_list_count,
3055                   list_count[2] / i_mb_list_count );
3056     }
3057
3058     x264_ratecontrol_summary( h );
3059
3060     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 )
3061     {
3062 #define SUM3(p) (p[SLICE_TYPE_I] + p[SLICE_TYPE_P] + p[SLICE_TYPE_B])
3063 #define SUM3b(p,o) (p[SLICE_TYPE_I][o] + p[SLICE_TYPE_P][o] + p[SLICE_TYPE_B][o])
3064         int64_t i_i8x8 = SUM3b( h->stat.i_mb_count, I_8x8 );
3065         int64_t i_intra = i_i8x8 + SUM3b( h->stat.i_mb_count, I_4x4 )
3066                                  + SUM3b( h->stat.i_mb_count, I_16x16 );
3067         int64_t i_all_intra = i_intra + SUM3b( h->stat.i_mb_count, I_PCM);
3068         const int i_count = h->stat.i_frame_count[SLICE_TYPE_I] +
3069                             h->stat.i_frame_count[SLICE_TYPE_P] +
3070                             h->stat.i_frame_count[SLICE_TYPE_B];
3071         int64_t i_mb_count = i_count * h->mb.i_mb_count;
3072         float fps = (float) h->param.i_fps_num / h->param.i_fps_den;
3073         float f_bitrate;
3074         /* duration algorithm fails with one frame */
3075         if( !h->param.b_vfr_input || i_count == 1 )
3076             f_bitrate = fps * SUM3(h->stat.i_frame_size) / i_count / 125;
3077         else
3078         {
3079             float duration = (float)(2 * h->frames.i_largest_pts - h->frames.i_second_largest_pts) * h->param.i_timebase_num / h->param.i_timebase_den;
3080             f_bitrate = SUM3(h->stat.i_frame_size) / duration / 125;
3081         }
3082
3083         if( h->pps->b_transform_8x8_mode )
3084         {
3085             buf[0] = 0;
3086             if( h->stat.i_mb_count_8x8dct[0] )
3087                 sprintf( buf, " inter:%.1f%%", 100. * h->stat.i_mb_count_8x8dct[1] / h->stat.i_mb_count_8x8dct[0] );
3088             x264_log( h, X264_LOG_INFO, "8x8 transform intra:%.1f%%%s\n", 100. * i_i8x8 / i_intra, buf );
3089         }
3090
3091         if( (h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO ||
3092             (h->stat.i_direct_frames[0] && h->stat.i_direct_frames[1]))
3093             && h->stat.i_frame_count[SLICE_TYPE_B] )
3094         {
3095             x264_log( h, X264_LOG_INFO, "direct mvs  spatial:%.1f%% temporal:%.1f%%\n",
3096                       h->stat.i_direct_frames[1] * 100. / h->stat.i_frame_count[SLICE_TYPE_B],
3097                       h->stat.i_direct_frames[0] * 100. / h->stat.i_frame_count[SLICE_TYPE_B] );
3098         }
3099
3100         buf[0] = 0;
3101         if( i_mb_count != i_all_intra )
3102             sprintf( buf, " inter: %.1f%% %.1f%% %.1f%%",
3103                      h->stat.i_mb_cbp[1] * 100.0 / ((i_mb_count - i_all_intra)*4),
3104                      h->stat.i_mb_cbp[3] * 100.0 / ((i_mb_count - i_all_intra)  ),
3105                      h->stat.i_mb_cbp[5] * 100.0 / ((i_mb_count - i_all_intra)) );
3106         x264_log( h, X264_LOG_INFO, "coded y,uvDC,uvAC intra: %.1f%% %.1f%% %.1f%%%s\n",
3107                   h->stat.i_mb_cbp[0] * 100.0 / (i_all_intra*4),
3108                   h->stat.i_mb_cbp[2] * 100.0 / (i_all_intra  ),
3109                   h->stat.i_mb_cbp[4] * 100.0 / (i_all_intra  ), buf );
3110
3111         int64_t fixed_pred_modes[4][9] = {{0}};
3112         int64_t sum_pred_modes[4] = {0};
3113         for( int i = 0; i <= I_PRED_16x16_DC_128; i++ )
3114         {
3115             fixed_pred_modes[0][x264_mb_pred_mode16x16_fix[i]] += h->stat.i_mb_pred_mode[0][i];
3116             sum_pred_modes[0] += h->stat.i_mb_pred_mode[0][i];
3117         }
3118         if( sum_pred_modes[0] )
3119             x264_log( h, X264_LOG_INFO, "i16 v,h,dc,p: %2.0f%% %2.0f%% %2.0f%% %2.0f%%\n",
3120                       fixed_pred_modes[0][0] * 100.0 / sum_pred_modes[0],
3121                       fixed_pred_modes[0][1] * 100.0 / sum_pred_modes[0],
3122                       fixed_pred_modes[0][2] * 100.0 / sum_pred_modes[0],
3123                       fixed_pred_modes[0][3] * 100.0 / sum_pred_modes[0] );
3124         for( int i = 1; i <= 2; i++ )
3125         {
3126             for( int j = 0; j <= I_PRED_8x8_DC_128; j++ )
3127             {
3128                 fixed_pred_modes[i][x264_mb_pred_mode4x4_fix(j)] += h->stat.i_mb_pred_mode[i][j];
3129                 sum_pred_modes[i] += h->stat.i_mb_pred_mode[i][j];
3130             }
3131             if( sum_pred_modes[i] )
3132                 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,
3133                           fixed_pred_modes[i][0] * 100.0 / sum_pred_modes[i],
3134                           fixed_pred_modes[i][1] * 100.0 / sum_pred_modes[i],
3135                           fixed_pred_modes[i][2] * 100.0 / sum_pred_modes[i],
3136                           fixed_pred_modes[i][3] * 100.0 / sum_pred_modes[i],
3137                           fixed_pred_modes[i][4] * 100.0 / sum_pred_modes[i],
3138                           fixed_pred_modes[i][5] * 100.0 / sum_pred_modes[i],
3139                           fixed_pred_modes[i][6] * 100.0 / sum_pred_modes[i],
3140                           fixed_pred_modes[i][7] * 100.0 / sum_pred_modes[i],
3141                           fixed_pred_modes[i][8] * 100.0 / sum_pred_modes[i] );
3142         }
3143         for( int i = 0; i <= I_PRED_CHROMA_DC_128; i++ )
3144         {
3145             fixed_pred_modes[3][x264_mb_pred_mode8x8c_fix[i]] += h->stat.i_mb_pred_mode[3][i];
3146             sum_pred_modes[3] += h->stat.i_mb_pred_mode[3][i];
3147         }
3148         if( sum_pred_modes[3] )
3149             x264_log( h, X264_LOG_INFO, "i8c dc,h,v,p: %2.0f%% %2.0f%% %2.0f%% %2.0f%%\n",
3150                       fixed_pred_modes[3][0] * 100.0 / sum_pred_modes[3],
3151                       fixed_pred_modes[3][1] * 100.0 / sum_pred_modes[3],
3152                       fixed_pred_modes[3][2] * 100.0 / sum_pred_modes[3],
3153                       fixed_pred_modes[3][3] * 100.0 / sum_pred_modes[3] );
3154
3155         if( h->param.analyse.i_weighted_pred == X264_WEIGHTP_SMART && h->stat.i_frame_count[SLICE_TYPE_P] > 0 )
3156             x264_log( h, X264_LOG_INFO, "Weighted P-Frames: Y:%.1f%%\n",
3157                       h->stat.i_wpred[0] * 100.0 / h->stat.i_frame_count[SLICE_TYPE_P] );
3158
3159         for( int i_list = 0; i_list < 2; i_list++ )
3160             for( int i_slice = 0; i_slice < 2; i_slice++ )
3161             {
3162                 char *p = buf;
3163                 int64_t i_den = 0;
3164                 int i_max = 0;
3165                 for( int i = 0; i < 32; i++ )
3166                     if( h->stat.i_mb_count_ref[i_slice][i_list][i] )
3167                     {
3168                         i_den += h->stat.i_mb_count_ref[i_slice][i_list][i];
3169                         i_max = i;
3170                     }
3171                 if( i_max == 0 )
3172                     continue;
3173                 for( int i = 0; i <= i_max; i++ )
3174                     p += sprintf( p, " %4.1f%%", 100. * h->stat.i_mb_count_ref[i_slice][i_list][i] / i_den );
3175                 x264_log( h, X264_LOG_INFO, "ref %c L%d:%s\n", "PB"[i_slice], i_list, buf );
3176             }
3177
3178         if( h->param.analyse.b_ssim )
3179         {
3180             float ssim = SUM3( h->stat.f_ssim_mean_y ) / i_count;
3181             x264_log( h, X264_LOG_INFO, "SSIM Mean Y:%.7f (%6.3fdb)\n", ssim, x264_ssim( ssim ) );
3182         }
3183         if( h->param.analyse.b_psnr )
3184         {
3185             x264_log( h, X264_LOG_INFO,
3186                       "PSNR Mean Y:%6.3f U:%6.3f V:%6.3f Avg:%6.3f Global:%6.3f kb/s:%.2f\n",
3187                       SUM3( h->stat.f_psnr_mean_y ) / i_count,
3188                       SUM3( h->stat.f_psnr_mean_u ) / i_count,
3189                       SUM3( h->stat.f_psnr_mean_v ) / i_count,
3190                       SUM3( h->stat.f_psnr_average ) / i_count,
3191                       x264_psnr( SUM3( h->stat.i_ssd_global ), i_count * i_yuv_size ),
3192                       f_bitrate );
3193         }
3194         else
3195             x264_log( h, X264_LOG_INFO, "kb/s:%.2f\n", f_bitrate );
3196     }
3197
3198     /* rc */
3199     x264_ratecontrol_delete( h );
3200
3201     /* param */
3202     if( h->param.rc.psz_stat_out )
3203         free( h->param.rc.psz_stat_out );
3204     if( h->param.rc.psz_stat_in )
3205         free( h->param.rc.psz_stat_in );
3206
3207     x264_cqm_delete( h );
3208     x264_free( h->nal_buffer );
3209     x264_analyse_free_costs( h );
3210
3211     if( h->i_thread_frames > 1)
3212         h = h->thread[h->i_thread_phase];
3213
3214     /* frames */
3215     x264_frame_delete_list( h->frames.unused[0] );
3216     x264_frame_delete_list( h->frames.unused[1] );
3217     x264_frame_delete_list( h->frames.current );
3218     x264_frame_delete_list( h->frames.blank_unused );
3219
3220     h = h->thread[0];
3221
3222     for( int i = h->param.i_threads - 1; i >= 0; i-- )
3223     {
3224         x264_frame_t **frame;
3225
3226         if( !h->param.b_sliced_threads || i == 0 )
3227         {
3228             for( frame = h->thread[i]->frames.reference; *frame; frame++ )
3229             {
3230                 assert( (*frame)->i_reference_count > 0 );
3231                 (*frame)->i_reference_count--;
3232                 if( (*frame)->i_reference_count == 0 )
3233                     x264_frame_delete( *frame );
3234             }
3235             frame = &h->thread[i]->fdec;
3236             assert( (*frame)->i_reference_count > 0 );
3237             (*frame)->i_reference_count--;
3238             if( (*frame)->i_reference_count == 0 )
3239                 x264_frame_delete( *frame );
3240             x264_macroblock_cache_free( h->thread[i] );
3241         }
3242         x264_macroblock_thread_free( h->thread[i], 0 );
3243         x264_free( h->thread[i]->out.p_bitstream );
3244         x264_free( h->thread[i]->out.nal);
3245         x264_free( h->thread[i] );
3246     }
3247 }
3248
3249 /****************************************************************************
3250  * x264_encoder_delayed_frames:
3251  ****************************************************************************/
3252 int x264_encoder_delayed_frames( x264_t *h )
3253 {
3254     int delayed_frames = 0;
3255     if( h->i_thread_frames > 1 )
3256     {
3257         for( int i = 0; i < h->i_thread_frames; i++ )
3258             delayed_frames += h->thread[i]->b_thread_active;
3259         h = h->thread[h->i_thread_phase];
3260     }
3261     for( int i = 0; h->frames.current[i]; i++ )
3262         delayed_frames++;
3263     x264_pthread_mutex_lock( &h->lookahead->ofbuf.mutex );
3264     x264_pthread_mutex_lock( &h->lookahead->ifbuf.mutex );
3265     x264_pthread_mutex_lock( &h->lookahead->next.mutex );
3266     delayed_frames += h->lookahead->ifbuf.i_size + h->lookahead->next.i_size + h->lookahead->ofbuf.i_size;
3267     x264_pthread_mutex_unlock( &h->lookahead->next.mutex );
3268     x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
3269     x264_pthread_mutex_unlock( &h->lookahead->ofbuf.mutex );
3270     return delayed_frames;
3271 }