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

   1 /*****************************************************************************
   2  * encoder.c: top-level encoder functions
   3  *****************************************************************************
   4  * Copyright (C) 2003-2010 x264 project
   5  *
   6  * Authors: Laurent Aimar <fenrir@via.ecp.fr>
   7  *          Loren Merritt <lorenm@u.washington.edu>
   8  *          Fiona Glaser <fiona@x264.com>
   9  *
  10  * This program is free software; you can redistribute it and/or modify
  11  * it under the terms of the GNU General Public License as published by
  12  * the Free Software Foundation; either version 2 of the License, or
  13  * (at your option) any later version.
  14  *
  15  * This program is distributed in the hope that it will be useful,
  16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  18  * GNU General Public License for more details.
  19  *
  20  * You should have received a copy of the GNU General Public License
  21  * along with this program; if not, write to the Free Software
  22  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02111, USA.
  23  *
  24  * This program is also available under a commercial proprietary license.
  25  * For more information, contact us at licensing@x264.com.
  26  *****************************************************************************/
  27
  28 #include <math.h>
  29
  30 #include "common/common.h"
  31
  32 #include "set.h"
  33 #include "analyse.h"
  34 #include "ratecontrol.h"
  35 #include "macroblock.h"
  36 #include "me.h"
  37
  38 #if HAVE_VISUALIZE
  39 #include "common/visualize.h"
  40 #endif
  41
  42 //#define DEBUG_MB_TYPE
  43
  44 #define bs_write_ue bs_write_ue_big
  45
  46 static int x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
  47                                    x264_nal_t **pp_nal, int *pi_nal,
  48                                    x264_picture_t *pic_out );
  49
  50 /****************************************************************************
  51  *
  52  ******************************* x264 libs **********************************
  53  *
  54  ****************************************************************************/
  55 static float x264_psnr( int64_t i_sqe, int64_t i_size )
  56 {
  57     double f_mse = (double)i_sqe / (PIXEL_MAX*PIXEL_MAX * (double)i_size);
  58     if( f_mse <= 0.0000000001 ) /* Max 100dB */
  59         return 100;
  60
  61     return -10.0 * log10( f_mse );
  62 }
  63
  64 static float x264_ssim( float ssim )
  65 {
  66     return -10.0 * log10( 1 - ssim );
  67 }
  68
  69 static void x264_frame_dump( x264_t *h )
  70 {
  71     FILE *f = fopen( h->param.psz_dump_yuv, "r+b" );
  72     if( !f )
  73         return;
  74     /* Write the frame in display order */
  75     fseek( f, (uint64_t)h->fdec->i_frame * h->param.i_height * h->param.i_width * 3/2 * sizeof(pixel), SEEK_SET );
  76     for( int y = 0; y < h->param.i_height; y++ )
  77         fwrite( &h->fdec->plane[0][y*h->fdec->i_stride[0]], sizeof(pixel), h->param.i_width, f );
  78     int cw = h->param.i_width>>1;
  79     int ch = h->param.i_height>>1;
  80     pixel *planeu = x264_malloc( cw*ch*2*sizeof(pixel) );
  81     pixel *planev = planeu + cw*ch;
  82     h->mc.plane_copy_deinterleave( planeu, cw, planev, cw, h->fdec->plane[1], h->fdec->i_stride[1], cw, ch );
  83     fwrite( planeu, 1, cw*ch*2*sizeof(pixel), f );
  84     x264_free( planeu );
  85     fclose( f );
  86 }
  87
  88
  89 /* Fill "default" values */
  90 static void x264_slice_header_init( x264_t *h, x264_slice_header_t *sh,
  91                                     x264_sps_t *sps, x264_pps_t *pps,
  92                                     int i_idr_pic_id, int i_frame, int i_qp )
  93 {
  94     x264_param_t *param = &h->param;
  95
  96     /* First we fill all fields */
  97     sh->sps = sps;
  98     sh->pps = pps;
  99
 100     sh->i_first_mb  = 0;
 101     sh->i_last_mb   = h->mb.i_mb_count - 1;
 102     sh->i_pps_id    = pps->i_id;
 103
 104     sh->i_frame_num = i_frame;
 105
 106     sh->b_mbaff = h->param.b_interlaced;
 107     sh->b_field_pic = 0;    /* no field support for now */
 108     sh->b_bottom_field = 0; /* not yet used */
 109
 110     sh->i_idr_pic_id = i_idr_pic_id;
 111
 112     /* poc stuff, fixed later */
 113     sh->i_poc = 0;
 114     sh->i_delta_poc_bottom = 0;
 115     sh->i_delta_poc[0] = 0;
 116     sh->i_delta_poc[1] = 0;
 117
 118     sh->i_redundant_pic_cnt = 0;
 119
 120     h->mb.b_direct_auto_write = h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO
 121                                 && h->param.i_bframe
 122                                 && ( h->param.rc.b_stat_write || !h->param.rc.b_stat_read );
 123
 124     if( !h->mb.b_direct_auto_read && sh->i_type == SLICE_TYPE_B )
 125     {
 126         if( h->fref1[0]->i_poc_l0ref0 == h->fref0[0]->i_poc )
 127         {
 128             if( h->mb.b_direct_auto_write )
 129                 sh->b_direct_spatial_mv_pred = ( h->stat.i_direct_score[1] > h->stat.i_direct_score[0] );
 130             else
 131                 sh->b_direct_spatial_mv_pred = ( param->analyse.i_direct_mv_pred == X264_DIRECT_PRED_SPATIAL );
 132         }
 133         else
 134         {
 135             h->mb.b_direct_auto_write = 0;
 136             sh->b_direct_spatial_mv_pred = 1;
 137         }
 138     }
 139     /* else b_direct_spatial_mv_pred was read from the 2pass statsfile */
 140
 141     sh->b_num_ref_idx_override = 0;
 142     sh->i_num_ref_idx_l0_active = 1;
 143     sh->i_num_ref_idx_l1_active = 1;
 144
 145     sh->b_ref_pic_list_reordering_l0 = h->b_ref_reorder[0];
 146     sh->b_ref_pic_list_reordering_l1 = h->b_ref_reorder[1];
 147
 148     /* If the ref list isn't in the default order, construct reordering header */
 149     /* List1 reordering isn't needed yet */
 150     if( sh->b_ref_pic_list_reordering_l0 )
 151     {
 152         int pred_frame_num = i_frame;
 153         for( int i = 0; i < h->i_ref0; i++ )
 154         {
 155             int diff = h->fref0[i]->i_frame_num - pred_frame_num;
 156             sh->ref_pic_list_order[0][i].idc = ( diff > 0 );
 157             sh->ref_pic_list_order[0][i].arg = (abs(diff) - 1) & ((1 << sps->i_log2_max_frame_num) - 1);
 158             pred_frame_num = h->fref0[i]->i_frame_num;
 159         }
 160     }
 161
 162     sh->i_cabac_init_idc = param->i_cabac_init_idc;
 163
 164     sh->i_qp = i_qp;
 165     sh->i_qp_delta = i_qp - pps->i_pic_init_qp;
 166     sh->b_sp_for_swidth = 0;
 167     sh->i_qs_delta = 0;
 168
 169     int deblock_thresh = i_qp + 2 * X264_MIN(param->i_deblocking_filter_alphac0, param->i_deblocking_filter_beta);
 170     /* If effective qp <= 15, deblocking would have no effect anyway */
 171     if( param->b_deblocking_filter && (h->mb.b_variable_qp || 15 < deblock_thresh ) )
 172         sh->i_disable_deblocking_filter_idc = param->b_sliced_threads ? 2 : 0;
 173     else
 174         sh->i_disable_deblocking_filter_idc = 1;
 175     sh->i_alpha_c0_offset = param->i_deblocking_filter_alphac0 << 1;
 176     sh->i_beta_offset = param->i_deblocking_filter_beta << 1;
 177 }
 178
 179 static void x264_slice_header_write( bs_t *s, x264_slice_header_t *sh, int i_nal_ref_idc )
 180 {
 181     if( sh->b_mbaff )
 182     {
 183         assert( sh->i_first_mb % (2*sh->sps->i_mb_width) == 0 );
 184         bs_write_ue( s, sh->i_first_mb >> 1 );
 185     }
 186     else
 187         bs_write_ue( s, sh->i_first_mb );
 188
 189     bs_write_ue( s, sh->i_type + 5 );   /* same type things */
 190     bs_write_ue( s, sh->i_pps_id );
 191     bs_write( s, sh->sps->i_log2_max_frame_num, sh->i_frame_num & ((1<<sh->sps->i_log2_max_frame_num)-1) );
 192
 193     if( !sh->sps->b_frame_mbs_only )
 194     {
 195         bs_write1( s, sh->b_field_pic );
 196         if( sh->b_field_pic )
 197             bs_write1( s, sh->b_bottom_field );
 198     }
 199
 200     if( sh->i_idr_pic_id >= 0 ) /* NAL IDR */
 201         bs_write_ue( s, sh->i_idr_pic_id );
 202
 203     if( sh->sps->i_poc_type == 0 )
 204     {
 205         bs_write( s, sh->sps->i_log2_max_poc_lsb, sh->i_poc & ((1<<sh->sps->i_log2_max_poc_lsb)-1) );
 206         if( sh->pps->b_pic_order && !sh->b_field_pic )
 207             bs_write_se( s, sh->i_delta_poc_bottom );
 208     }
 209     else if( sh->sps->i_poc_type == 1 && !sh->sps->b_delta_pic_order_always_zero )
 210     {
 211         bs_write_se( s, sh->i_delta_poc[0] );
 212         if( sh->pps->b_pic_order && !sh->b_field_pic )
 213             bs_write_se( s, sh->i_delta_poc[1] );
 214     }
 215
 216     if( sh->pps->b_redundant_pic_cnt )
 217         bs_write_ue( s, sh->i_redundant_pic_cnt );
 218
 219     if( sh->i_type == SLICE_TYPE_B )
 220         bs_write1( s, sh->b_direct_spatial_mv_pred );
 221
 222     if( sh->i_type == SLICE_TYPE_P || sh->i_type == SLICE_TYPE_B )
 223     {
 224         bs_write1( s, sh->b_num_ref_idx_override );
 225         if( sh->b_num_ref_idx_override )
 226         {
 227             bs_write_ue( s, sh->i_num_ref_idx_l0_active - 1 );
 228             if( sh->i_type == SLICE_TYPE_B )
 229                 bs_write_ue( s, sh->i_num_ref_idx_l1_active - 1 );
 230         }
 231     }
 232
 233     /* ref pic list reordering */
 234     if( sh->i_type != SLICE_TYPE_I )
 235     {
 236         bs_write1( s, sh->b_ref_pic_list_reordering_l0 );
 237         if( sh->b_ref_pic_list_reordering_l0 )
 238         {
 239             for( int i = 0; i < sh->i_num_ref_idx_l0_active; i++ )
 240             {
 241                 bs_write_ue( s, sh->ref_pic_list_order[0][i].idc );
 242                 bs_write_ue( s, sh->ref_pic_list_order[0][i].arg );
 243             }
 244             bs_write_ue( s, 3 );
 245         }
 246     }
 247     if( sh->i_type == SLICE_TYPE_B )
 248     {
 249         bs_write1( s, sh->b_ref_pic_list_reordering_l1 );
 250         if( sh->b_ref_pic_list_reordering_l1 )
 251         {
 252             for( int i = 0; i < sh->i_num_ref_idx_l1_active; i++ )
 253             {
 254                 bs_write_ue( s, sh->ref_pic_list_order[1][i].idc );
 255                 bs_write_ue( s, sh->ref_pic_list_order[1][i].arg );
 256             }
 257             bs_write_ue( s, 3 );
 258         }
 259     }
 260
 261     if( sh->pps->b_weighted_pred && sh->i_type == SLICE_TYPE_P )
 262     {
 263         /* pred_weight_table() */
 264         bs_write_ue( s, sh->weight[0][0].i_denom );
 265         bs_write_ue( s, sh->weight[0][1].i_denom );
 266         for( int i = 0; i < sh->i_num_ref_idx_l0_active; i++ )
 267         {
 268             int luma_weight_l0_flag = !!sh->weight[i][0].weightfn;
 269             int chroma_weight_l0_flag = !!sh->weight[i][1].weightfn || !!sh->weight[i][2].weightfn;
 270             bs_write1( s, luma_weight_l0_flag );
 271             if( luma_weight_l0_flag )
 272             {
 273                 bs_write_se( s, sh->weight[i][0].i_scale );
 274                 bs_write_se( s, sh->weight[i][0].i_offset );
 275             }
 276             bs_write1( s, chroma_weight_l0_flag );
 277             if( chroma_weight_l0_flag )
 278             {
 279                 for( int j = 1; j < 3; j++ )
 280                 {
 281                     bs_write_se( s, sh->weight[i][j].i_scale );
 282                     bs_write_se( s, sh->weight[i][j].i_offset );
 283                 }
 284             }
 285         }
 286     }
 287     else if( sh->pps->b_weighted_bipred == 1 && sh->i_type == SLICE_TYPE_B )
 288     {
 289       /* TODO */
 290     }
 291
 292     if( i_nal_ref_idc != 0 )
 293     {
 294         if( sh->i_idr_pic_id >= 0 )
 295         {
 296             bs_write1( s, 0 );  /* no output of prior pics flag */
 297             bs_write1( s, 0 );  /* long term reference flag */
 298         }
 299         else
 300         {
 301             bs_write1( s, sh->i_mmco_command_count > 0 ); /* adaptive_ref_pic_marking_mode_flag */
 302             if( sh->i_mmco_command_count > 0 )
 303             {
 304                 for( int i = 0; i < sh->i_mmco_command_count; i++ )
 305                 {
 306                     bs_write_ue( s, 1 ); /* mark short term ref as unused */
 307                     bs_write_ue( s, sh->mmco[i].i_difference_of_pic_nums - 1 );
 308                 }
 309                 bs_write_ue( s, 0 ); /* end command list */
 310             }
 311         }
 312     }
 313
 314     if( sh->pps->b_cabac && sh->i_type != SLICE_TYPE_I )
 315         bs_write_ue( s, sh->i_cabac_init_idc );
 316
 317     bs_write_se( s, sh->i_qp_delta );      /* slice qp delta */
 318
 319     if( sh->pps->b_deblocking_filter_control )
 320     {
 321         bs_write_ue( s, sh->i_disable_deblocking_filter_idc );
 322         if( sh->i_disable_deblocking_filter_idc != 1 )
 323         {
 324             bs_write_se( s, sh->i_alpha_c0_offset >> 1 );
 325             bs_write_se( s, sh->i_beta_offset >> 1 );
 326         }
 327     }
 328 }
 329
 330 /* If we are within a reasonable distance of the end of the memory allocated for the bitstream, */
 331 /* reallocate, adding an arbitrary amount of space (100 kilobytes). */
 332 static int x264_bitstream_check_buffer( x264_t *h )
 333 {
 334     uint8_t *bs_bak = h->out.p_bitstream;
 335     if( (h->param.b_cabac && (h->cabac.p_end - h->cabac.p < 2500)) ||
 336         (h->out.bs.p_end - h->out.bs.p < 2500) )
 337     {
 338         h->out.i_bitstream += 100000;
 339         CHECKED_MALLOC( h->out.p_bitstream, h->out.i_bitstream );
 340         h->mc.memcpy_aligned( h->out.p_bitstream, bs_bak, (h->out.i_bitstream - 100000) & ~15 );
 341         intptr_t delta = h->out.p_bitstream - bs_bak;
 342
 343         h->out.bs.p_start += delta;
 344         h->out.bs.p += delta;
 345         h->out.bs.p_end = h->out.p_bitstream + h->out.i_bitstream;
 346
 347         h->cabac.p_start += delta;
 348         h->cabac.p += delta;
 349         h->cabac.p_end = h->out.p_bitstream + h->out.i_bitstream;
 350
 351         for( int i = 0; i <= h->out.i_nal; i++ )
 352             h->out.nal[i].p_payload += delta;
 353         x264_free( bs_bak );
 354     }
 355     return 0;
 356 fail:
 357     x264_free( bs_bak );
 358     return -1;
 359 }
 360
 361 #if HAVE_PTHREAD
 362 static void x264_encoder_thread_init( x264_t *h )
 363 {
 364     if( h->param.i_sync_lookahead )
 365         x264_lower_thread_priority( 10 );
 366
 367 #if HAVE_MMX
 368     /* Misalign mask has to be set separately for each thread. */
 369     if( h->param.cpu&X264_CPU_SSE_MISALIGN )
 370         x264_cpu_mask_misalign_sse();
 371 #endif
 372 }
 373 #endif
 374
 375 /****************************************************************************
 376  *
 377  ****************************************************************************
 378  ****************************** External API*********************************
 379  ****************************************************************************
 380  *
 381  ****************************************************************************/
 382
 383 static int x264_validate_parameters( x264_t *h )
 384 {
 385 #if HAVE_MMX
 386 #ifdef __SSE__
 387     if( !(x264_cpu_detect() & X264_CPU_SSE) )
 388     {
 389         x264_log( h, X264_LOG_ERROR, "your cpu does not support SSE1, but x264 was compiled with asm support\n");
 390 #else
 391     if( !(x264_cpu_detect() & X264_CPU_MMXEXT) )
 392     {
 393         x264_log( h, X264_LOG_ERROR, "your cpu does not support MMXEXT, but x264 was compiled with asm support\n");
 394 #endif
 395         x264_log( h, X264_LOG_ERROR, "to run x264, recompile without asm support (configure --disable-asm)\n");
 396         return -1;
 397     }
 398 #endif
 399     if( h->param.i_width <= 0 || h->param.i_height <= 0 )
 400     {
 401         x264_log( h, X264_LOG_ERROR, "invalid width x height (%dx%d)\n",
 402                   h->param.i_width, h->param.i_height );
 403         return -1;
 404     }
 405
 406     if( h->param.i_width % 2 || h->param.i_height % 2 )
 407     {
 408         x264_log( h, X264_LOG_ERROR, "width or height not divisible by 2 (%dx%d)\n",
 409                   h->param.i_width, h->param.i_height );
 410         return -1;
 411     }
 412     int i_csp = h->param.i_csp & X264_CSP_MASK;
 413     if( i_csp <= X264_CSP_NONE || i_csp >= X264_CSP_MAX )
 414     {
 415         x264_log( h, X264_LOG_ERROR, "invalid CSP (only I420/YV12/NV12 supported)\n" );
 416         return -1;
 417     }
 418
 419     if( h->param.i_threads == X264_THREADS_AUTO )
 420         h->param.i_threads = x264_cpu_num_processors() * (h->param.b_sliced_threads?2:3)/2;
 421     h->param.i_threads = x264_clip3( h->param.i_threads, 1, X264_THREAD_MAX );
 422     if( h->param.i_threads > 1 )
 423     {
 424 #if !HAVE_PTHREAD
 425         x264_log( h, X264_LOG_WARNING, "not compiled with pthread support!\n");
 426         h->param.i_threads = 1;
 427 #endif
 428         /* Avoid absurdly small thread slices as they can reduce performance
 429          * and VBV compliance.  Capped at an arbitrary 4 rows per thread. */
 430         if( h->param.b_sliced_threads )
 431         {
 432             int max_threads = (h->param.i_height+15)/16 / 4;
 433             h->param.i_threads = X264_MIN( h->param.i_threads, max_threads );
 434         }
 435     }
 436     else
 437         h->param.b_sliced_threads = 0;
 438     h->i_thread_frames = h->param.b_sliced_threads ? 1 : h->param.i_threads;
 439     if( h->i_thread_frames > 1 )
 440         h->param.nalu_process = NULL;
 441
 442     if( h->param.b_interlaced )
 443     {
 444         if( h->param.analyse.i_me_method >= X264_ME_ESA )
 445         {
 446             x264_log( h, X264_LOG_WARNING, "interlace + me=esa is not implemented\n" );
 447             h->param.analyse.i_me_method = X264_ME_UMH;
 448         }
 449         if( h->param.analyse.i_weighted_pred > 0 )
 450         {
 451             x264_log( h, X264_LOG_WARNING, "interlace + weightp is not implemented\n" );
 452             h->param.analyse.i_weighted_pred = X264_WEIGHTP_NONE;
 453         }
 454     }
 455
 456     /* Detect default ffmpeg settings and terminate with an error. */
 457     {
 458         int score = 0;
 459         score += h->param.analyse.i_me_range == 0;
 460         score += h->param.rc.i_qp_step == 3;
 461         score += h->param.i_keyint_max == 12;
 462         score += h->param.rc.i_qp_min == 2;
 463         score += h->param.rc.i_qp_max == 31;
 464         score += h->param.rc.f_qcompress == 0.5;
 465         score += fabs(h->param.rc.f_ip_factor - 1.25) < 0.01;
 466         score += fabs(h->param.rc.f_pb_factor - 1.25) < 0.01;
 467         score += h->param.analyse.inter == 0 && h->param.analyse.i_subpel_refine == 8;
 468         if( score >= 5 )
 469         {
 470             x264_log( h, X264_LOG_ERROR, "broken ffmpeg default settings detected\n" );
 471             x264_log( h, X264_LOG_ERROR, "use an encoding preset (vpre)\n" );
 472             return -1;
 473         }
 474     }
 475
 476     if( h->param.rc.i_rc_method < 0 || h->param.rc.i_rc_method > 2 )
 477     {
 478         x264_log( h, X264_LOG_ERROR, "no ratecontrol method specified\n" );
 479         return -1;
 480     }
 481     h->param.rc.f_rf_constant = x264_clip3f( h->param.rc.f_rf_constant, 0, QP_MAX );
 482     h->param.rc.i_qp_constant = x264_clip3( h->param.rc.i_qp_constant, 0, QP_MAX );
 483     if( h->param.rc.i_rc_method == X264_RC_CRF )
 484     {
 485         h->param.rc.i_qp_constant = h->param.rc.f_rf_constant;
 486         h->param.rc.i_bitrate = 0;
 487     }
 488     if( (h->param.rc.i_rc_method == X264_RC_CQP || h->param.rc.i_rc_method == X264_RC_CRF)
 489         && h->param.rc.i_qp_constant == 0 )
 490     {
 491         h->mb.b_lossless = 1;
 492         h->param.i_cqm_preset = X264_CQM_FLAT;
 493         h->param.psz_cqm_file = NULL;
 494         h->param.rc.i_rc_method = X264_RC_CQP;
 495         h->param.rc.f_ip_factor = 1;
 496         h->param.rc.f_pb_factor = 1;
 497         h->param.analyse.b_psnr = 0;
 498         h->param.analyse.b_ssim = 0;
 499         h->param.analyse.i_chroma_qp_offset = 0;
 500         h->param.analyse.i_trellis = 0;
 501         h->param.analyse.b_fast_pskip = 0;
 502         h->param.analyse.i_noise_reduction = 0;
 503         h->param.analyse.b_psy = 0;
 504         h->param.i_bframe = 0;
 505         /* 8x8dct is not useful at all in CAVLC lossless */
 506         if( !h->param.b_cabac )
 507             h->param.analyse.b_transform_8x8 = 0;
 508     }
 509     if( h->param.rc.i_rc_method == X264_RC_CQP )
 510     {
 511         float qp_p = h->param.rc.i_qp_constant;
 512         float qp_i = qp_p - 6*log2f( h->param.rc.f_ip_factor );
 513         float qp_b = qp_p + 6*log2f( h->param.rc.f_pb_factor );
 514         h->param.rc.i_qp_min = x264_clip3( (int)(X264_MIN3( qp_p, qp_i, qp_b )), 0, QP_MAX );
 515         h->param.rc.i_qp_max = x264_clip3( (int)(X264_MAX3( qp_p, qp_i, qp_b ) + .999), 0, QP_MAX );
 516         h->param.rc.i_aq_mode = 0;
 517         h->param.rc.b_mb_tree = 0;
 518     }
 519     h->param.rc.i_qp_max = x264_clip3( h->param.rc.i_qp_max, 0, QP_MAX );
 520     h->param.rc.i_qp_min = x264_clip3( h->param.rc.i_qp_min, 0, h->param.rc.i_qp_max );
 521     if( h->param.rc.i_vbv_buffer_size )
 522     {
 523         if( h->param.rc.i_rc_method == X264_RC_CQP )
 524         {
 525             x264_log( h, X264_LOG_WARNING, "VBV is incompatible with constant QP, ignored.\n" );
 526             h->param.rc.i_vbv_max_bitrate = 0;
 527             h->param.rc.i_vbv_buffer_size = 0;
 528         }
 529         else if( h->param.rc.i_vbv_max_bitrate == 0 )
 530         {
 531             if( h->param.rc.i_rc_method == X264_RC_ABR )
 532             {
 533                 x264_log( h, X264_LOG_WARNING, "VBV maxrate unspecified, assuming CBR\n" );
 534                 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate;
 535             }
 536             else
 537             {
 538                 x264_log( h, X264_LOG_WARNING, "VBV bufsize set but maxrate unspecified, ignored\n" );
 539                 h->param.rc.i_vbv_buffer_size = 0;
 540             }
 541         }
 542         else if( h->param.rc.i_vbv_max_bitrate < h->param.rc.i_bitrate &&
 543                  h->param.rc.i_rc_method == X264_RC_ABR )
 544         {
 545             x264_log( h, X264_LOG_WARNING, "max bitrate less than average bitrate, assuming CBR\n" );
 546             h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate;
 547         }
 548     }
 549     else if( h->param.rc.i_vbv_max_bitrate )
 550     {
 551         x264_log( h, X264_LOG_WARNING, "VBV maxrate specified, but no bufsize, ignored\n" );
 552         h->param.rc.i_vbv_max_bitrate = 0;
 553     }
 554
 555     if( h->param.b_interlaced && h->param.i_slice_max_size )
 556     {
 557         x264_log( h, X264_LOG_WARNING, "interlaced + slice-max-size is not implemented\n" );
 558         h->param.i_slice_max_size = 0;
 559     }
 560     if( h->param.b_interlaced && h->param.i_slice_max_mbs )
 561     {
 562         x264_log( h, X264_LOG_WARNING, "interlaced + slice-max-mbs is not implemented\n" );
 563         h->param.i_slice_max_mbs = 0;
 564     }
 565     int max_slices = (h->param.i_height+((16<<h->param.b_interlaced)-1))/(16<<h->param.b_interlaced);
 566     if( h->param.b_sliced_threads )
 567         h->param.i_slice_count = x264_clip3( h->param.i_threads, 0, max_slices );
 568     else
 569     {
 570         h->param.i_slice_count = x264_clip3( h->param.i_slice_count, 0, max_slices );
 571         h->param.i_slice_max_size = X264_MAX( h->param.i_slice_max_size, 0 );
 572         h->param.i_slice_max_mbs = X264_MAX( h->param.i_slice_max_mbs, 0 );
 573         if( h->param.i_slice_max_mbs || h->param.i_slice_max_size )
 574             h->param.i_slice_count = 0;
 575     }
 576
 577     h->param.i_frame_reference = x264_clip3( h->param.i_frame_reference, 1, X264_REF_MAX );
 578     h->param.i_dpb_size = x264_clip3( h->param.i_dpb_size, 1, X264_REF_MAX );
 579     h->param.i_keyint_max = x264_clip3( h->param.i_keyint_max, 1, X264_KEYINT_MAX_INFINITE );
 580     if( h->param.i_scenecut_threshold < 0 )
 581         h->param.i_scenecut_threshold = 0;
 582     if( !h->param.analyse.i_subpel_refine && h->param.analyse.i_direct_mv_pred > X264_DIRECT_PRED_SPATIAL )
 583     {
 584         x264_log( h, X264_LOG_WARNING, "subme=0 + direct=temporal is not supported\n" );
 585         h->param.analyse.i_direct_mv_pred = X264_DIRECT_PRED_SPATIAL;
 586     }
 587     h->param.i_bframe = x264_clip3( h->param.i_bframe, 0, X264_MIN( X264_BFRAME_MAX, h->param.i_keyint_max-1 ) );
 588     h->param.i_open_gop = x264_clip3( h->param.i_open_gop, X264_OPEN_GOP_NONE, X264_OPEN_GOP_BLURAY );
 589     if( h->param.i_keyint_max == 1 )
 590         h->param.b_intra_refresh = 0;
 591     h->param.i_bframe_bias = x264_clip3( h->param.i_bframe_bias, -90, 100 );
 592     if( h->param.i_bframe <= 1 )
 593         h->param.i_bframe_pyramid = X264_B_PYRAMID_NONE;
 594     h->param.i_bframe_pyramid = x264_clip3( h->param.i_bframe_pyramid, X264_B_PYRAMID_NONE, X264_B_PYRAMID_NORMAL );
 595     if( !h->param.i_bframe )
 596     {
 597         h->param.i_bframe_adaptive = X264_B_ADAPT_NONE;
 598         h->param.analyse.i_direct_mv_pred = 0;
 599         h->param.analyse.b_weighted_bipred = 0;
 600         h->param.i_open_gop = X264_OPEN_GOP_NONE;
 601     }
 602     if( h->param.b_intra_refresh && h->param.i_bframe_pyramid == X264_B_PYRAMID_NORMAL )
 603     {
 604         x264_log( h, X264_LOG_WARNING, "b-pyramid normal + intra-refresh is not supported\n" );
 605         h->param.i_bframe_pyramid = X264_B_PYRAMID_STRICT;
 606     }
 607     if( h->param.b_intra_refresh && (h->param.i_frame_reference > 1 || h->param.i_dpb_size > 1) )
 608     {
 609         x264_log( h, X264_LOG_WARNING, "ref > 1 + intra-refresh is not supported\n" );
 610         h->param.i_frame_reference = 1;
 611         h->param.i_dpb_size = 1;
 612     }
 613     if( h->param.b_intra_refresh && h->param.i_open_gop )
 614     {
 615         x264_log( h, X264_LOG_WARNING, "intra-refresh is not compatible with open-gop\n" );
 616         h->param.i_open_gop = X264_OPEN_GOP_NONE;
 617     }
 618     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;
 619     if( h->param.i_keyint_min == X264_KEYINT_MIN_AUTO )
 620         h->param.i_keyint_min = X264_MIN( h->param.i_keyint_max / 10, fps );
 621     h->param.i_keyint_min = x264_clip3( h->param.i_keyint_min, 1, h->param.i_keyint_max/2+1 );
 622     h->param.rc.i_lookahead = x264_clip3( h->param.rc.i_lookahead, 0, X264_LOOKAHEAD_MAX );
 623     {
 624         int maxrate = X264_MAX( h->param.rc.i_vbv_max_bitrate, h->param.rc.i_bitrate );
 625         float bufsize = maxrate ? (float)h->param.rc.i_vbv_buffer_size / maxrate : 0;
 626         h->param.rc.i_lookahead = X264_MIN( h->param.rc.i_lookahead, X264_MAX( h->param.i_keyint_max, bufsize*fps ) );
 627     }
 628
 629     if( !h->param.i_timebase_num || !h->param.i_timebase_den )
 630     {
 631         h->param.i_timebase_num = h->param.i_fps_den;
 632         h->param.i_timebase_den = h->param.i_fps_num;
 633     }
 634
 635     h->param.rc.f_qcompress = x264_clip3f( h->param.rc.f_qcompress, 0.0, 1.0 );
 636     if( h->param.i_keyint_max == 1 || h->param.rc.f_qcompress == 1 )
 637         h->param.rc.b_mb_tree = 0;
 638     if( (!h->param.b_intra_refresh && h->param.i_keyint_max != X264_KEYINT_MAX_INFINITE) &&
 639         !h->param.rc.i_lookahead && h->param.rc.b_mb_tree )
 640     {
 641         x264_log( h, X264_LOG_WARNING, "lookaheadless mb-tree requires intra refresh or infinite keyint\n" );
 642         h->param.rc.b_mb_tree = 0;
 643     }
 644     if( h->param.rc.b_stat_read )
 645         h->param.rc.i_lookahead = 0;
 646 #if HAVE_PTHREAD
 647     if( h->param.i_sync_lookahead < 0 )
 648         h->param.i_sync_lookahead = h->param.i_bframe + 1;
 649     h->param.i_sync_lookahead = X264_MIN( h->param.i_sync_lookahead, X264_LOOKAHEAD_MAX );
 650     if( h->param.rc.b_stat_read || h->i_thread_frames == 1 )
 651         h->param.i_sync_lookahead = 0;
 652 #else
 653     h->param.i_sync_lookahead = 0;
 654 #endif
 655
 656     h->param.i_deblocking_filter_alphac0 = x264_clip3( h->param.i_deblocking_filter_alphac0, -6, 6 );
 657     h->param.i_deblocking_filter_beta    = x264_clip3( h->param.i_deblocking_filter_beta, -6, 6 );
 658     h->param.analyse.i_luma_deadzone[0] = x264_clip3( h->param.analyse.i_luma_deadzone[0], 0, 32 );
 659     h->param.analyse.i_luma_deadzone[1] = x264_clip3( h->param.analyse.i_luma_deadzone[1], 0, 32 );
 660
 661     h->param.i_cabac_init_idc = x264_clip3( h->param.i_cabac_init_idc, 0, 2 );
 662
 663     if( h->param.i_cqm_preset < X264_CQM_FLAT || h->param.i_cqm_preset > X264_CQM_CUSTOM )
 664         h->param.i_cqm_preset = X264_CQM_FLAT;
 665
 666     if( h->param.analyse.i_me_method < X264_ME_DIA ||
 667         h->param.analyse.i_me_method > X264_ME_TESA )
 668         h->param.analyse.i_me_method = X264_ME_HEX;
 669     if( h->param.analyse.i_me_range < 4 )
 670         h->param.analyse.i_me_range = 4;
 671     if( h->param.analyse.i_me_range > 16 && h->param.analyse.i_me_method <= X264_ME_HEX )
 672         h->param.analyse.i_me_range = 16;
 673     if( h->param.analyse.i_me_method == X264_ME_TESA &&
 674         (h->mb.b_lossless || h->param.analyse.i_subpel_refine <= 1) )
 675         h->param.analyse.i_me_method = X264_ME_ESA;
 676     h->param.analyse.i_subpel_refine = x264_clip3( h->param.analyse.i_subpel_refine, 0, 10 );
 677     h->param.analyse.b_mixed_references = h->param.analyse.b_mixed_references && h->param.i_frame_reference > 1;
 678     h->param.analyse.inter &= X264_ANALYSE_PSUB16x16|X264_ANALYSE_PSUB8x8|X264_ANALYSE_BSUB16x16|
 679                               X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
 680     h->param.analyse.intra &= X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
 681     if( !(h->param.analyse.inter & X264_ANALYSE_PSUB16x16) )
 682         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
 683     if( !h->param.analyse.b_transform_8x8 )
 684     {
 685         h->param.analyse.inter &= ~X264_ANALYSE_I8x8;
 686         h->param.analyse.intra &= ~X264_ANALYSE_I8x8;
 687     }
 688     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -12, 12);
 689     h->param.analyse.i_trellis = x264_clip3( h->param.analyse.i_trellis, 0, 2 );
 690     if( !h->param.analyse.b_psy )
 691     {
 692         h->param.analyse.f_psy_rd = 0;
 693         h->param.analyse.f_psy_trellis = 0;
 694     }
 695     if( !h->param.analyse.i_trellis )
 696         h->param.analyse.f_psy_trellis = 0;
 697     h->param.analyse.f_psy_rd = x264_clip3f( h->param.analyse.f_psy_rd, 0, 10 );
 698     h->param.analyse.f_psy_trellis = x264_clip3f( h->param.analyse.f_psy_trellis, 0, 10 );
 699     if( h->param.analyse.i_subpel_refine < 6 )
 700         h->param.analyse.f_psy_rd = 0;
 701     h->mb.i_psy_rd = FIX8( h->param.analyse.f_psy_rd );
 702     /* Psy RDO increases overall quantizers to improve the quality of luma--this indirectly hurts chroma quality */
 703     /* so we lower the chroma QP offset to compensate */
 704     /* This can be triggered repeatedly on multiple calls to parameter_validate, but since encoding
 705      * uses the pps chroma qp offset not the param chroma qp offset, this is not a problem. */
 706     if( h->mb.i_psy_rd )
 707         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_rd < 0.25 ? 1 : 2;
 708     h->mb.i_psy_trellis = FIX8( h->param.analyse.f_psy_trellis / 4 );
 709     /* Psy trellis has a similar effect. */
 710     if( h->mb.i_psy_trellis )
 711         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_trellis < 0.25 ? 1 : 2;
 712     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -12, 12);
 713     h->param.rc.i_aq_mode = x264_clip3( h->param.rc.i_aq_mode, 0, 2 );
 714     h->param.rc.f_aq_strength = x264_clip3f( h->param.rc.f_aq_strength, 0, 3 );
 715     if( h->param.rc.f_aq_strength == 0 )
 716         h->param.rc.i_aq_mode = 0;
 717     /* MB-tree requires AQ to be on, even if the strength is zero. */
 718     if( !h->param.rc.i_aq_mode && h->param.rc.b_mb_tree )
 719     {
 720         h->param.rc.i_aq_mode = 1;
 721         h->param.rc.f_aq_strength = 0;
 722     }
 723     h->param.analyse.i_noise_reduction = x264_clip3( h->param.analyse.i_noise_reduction, 0, 1<<16 );
 724     if( h->param.analyse.i_subpel_refine == 10 && (h->param.analyse.i_trellis != 2 || !h->param.rc.i_aq_mode) )
 725         h->param.analyse.i_subpel_refine = 9;
 726
 727     {
 728         const x264_level_t *l = x264_levels;
 729         if( h->param.i_level_idc < 0 )
 730         {
 731             int maxrate_bak = h->param.rc.i_vbv_max_bitrate;
 732             if( h->param.rc.i_rc_method == X264_RC_ABR && h->param.rc.i_vbv_buffer_size <= 0 )
 733                 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate * 2;
 734             h->sps = h->sps_array;
 735             x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
 736             do h->param.i_level_idc = l->level_idc;
 737                 while( l[1].level_idc && x264_validate_levels( h, 0 ) && l++ );
 738             h->param.rc.i_vbv_max_bitrate = maxrate_bak;
 739         }
 740         else
 741         {
 742             while( l->level_idc && l->level_idc != h->param.i_level_idc )
 743                 l++;
 744             if( l->level_idc == 0 )
 745             {
 746                 x264_log( h, X264_LOG_ERROR, "invalid level_idc: %d\n", h->param.i_level_idc );
 747                 return -1;
 748             }
 749         }
 750         if( h->param.analyse.i_mv_range <= 0 )
 751             h->param.analyse.i_mv_range = l->mv_range >> h->param.b_interlaced;
 752         else
 753             h->param.analyse.i_mv_range = x264_clip3(h->param.analyse.i_mv_range, 32, 512 >> h->param.b_interlaced);
 754     }
 755
 756     h->param.analyse.i_weighted_pred = x264_clip3( h->param.analyse.i_weighted_pred, 0, X264_WEIGHTP_SMART );
 757     if( !h->param.analyse.i_weighted_pred && h->param.rc.b_mb_tree && h->param.analyse.b_psy && !h->param.b_interlaced )
 758         h->param.analyse.i_weighted_pred = X264_WEIGHTP_FAKE;
 759
 760     if( h->i_thread_frames > 1 )
 761     {
 762         int r = h->param.analyse.i_mv_range_thread;
 763         int r2;
 764         if( r <= 0 )
 765         {
 766             // half of the available space is reserved and divided evenly among the threads,
 767             // the rest is allocated to whichever thread is far enough ahead to use it.
 768             // reserving more space increases quality for some videos, but costs more time
 769             // in thread synchronization.
 770             int max_range = (h->param.i_height + X264_THREAD_HEIGHT) / h->i_thread_frames - X264_THREAD_HEIGHT;
 771             r = max_range / 2;
 772         }
 773         r = X264_MAX( r, h->param.analyse.i_me_range );
 774         r = X264_MIN( r, h->param.analyse.i_mv_range );
 775         // round up to use the whole mb row
 776         r2 = (r & ~15) + ((-X264_THREAD_HEIGHT) & 15);
 777         if( r2 < r )
 778             r2 += 16;
 779         x264_log( h, X264_LOG_DEBUG, "using mv_range_thread = %d\n", r2 );
 780         h->param.analyse.i_mv_range_thread = r2;
 781     }
 782
 783     if( h->param.rc.f_qblur < 0 )
 784         h->param.rc.f_qblur = 0;
 785     if( h->param.rc.f_complexity_blur < 0 )
 786         h->param.rc.f_complexity_blur = 0;
 787
 788     h->param.i_sps_id &= 31;
 789
 790     if( h->param.i_log_level < X264_LOG_INFO )
 791     {
 792         h->param.analyse.b_psnr = 0;
 793         h->param.analyse.b_ssim = 0;
 794     }
 795
 796     if( h->param.b_interlaced )
 797         h->param.b_pic_struct = 1;
 798
 799     if( h->param.i_nal_hrd && !h->param.rc.i_vbv_buffer_size )
 800     {
 801         x264_log( h, X264_LOG_WARNING, "NAL HRD parameters require VBV parameters\n" );
 802         h->param.i_nal_hrd = X264_NAL_HRD_NONE;
 803     }
 804
 805     if( h->param.i_nal_hrd == X264_NAL_HRD_CBR &&
 806        (h->param.rc.i_bitrate != h->param.rc.i_vbv_max_bitrate || !h->param.rc.i_vbv_max_bitrate) )
 807     {
 808         x264_log( h, X264_LOG_WARNING, "CBR HRD requires constant bitrate\n" );
 809         h->param.i_nal_hrd = X264_NAL_HRD_VBR;
 810     }
 811
 812     /* ensure the booleans are 0 or 1 so they can be used in math */
 813 #define BOOLIFY(x) h->param.x = !!h->param.x
 814     BOOLIFY( b_cabac );
 815     BOOLIFY( b_constrained_intra );
 816     BOOLIFY( b_deblocking_filter );
 817     BOOLIFY( b_deterministic );
 818     BOOLIFY( b_sliced_threads );
 819     BOOLIFY( b_interlaced );
 820     BOOLIFY( b_intra_refresh );
 821     BOOLIFY( b_visualize );
 822     BOOLIFY( b_aud );
 823     BOOLIFY( b_repeat_headers );
 824     BOOLIFY( b_annexb );
 825     BOOLIFY( b_vfr_input );
 826     BOOLIFY( b_pic_struct );
 827     BOOLIFY( b_fake_interlaced );
 828     BOOLIFY( analyse.b_transform_8x8 );
 829     BOOLIFY( analyse.b_weighted_bipred );
 830     BOOLIFY( analyse.b_chroma_me );
 831     BOOLIFY( analyse.b_mixed_references );
 832     BOOLIFY( analyse.b_fast_pskip );
 833     BOOLIFY( analyse.b_dct_decimate );
 834     BOOLIFY( analyse.b_psy );
 835     BOOLIFY( analyse.b_psnr );
 836     BOOLIFY( analyse.b_ssim );
 837     BOOLIFY( rc.b_stat_write );
 838     BOOLIFY( rc.b_stat_read );
 839     BOOLIFY( rc.b_mb_tree );
 840 #undef BOOLIFY
 841
 842     return 0;
 843 }
 844
 845 static void mbcmp_init( x264_t *h )
 846 {
 847     int satd = !h->mb.b_lossless && h->param.analyse.i_subpel_refine > 1;
 848     memcpy( h->pixf.mbcmp, satd ? h->pixf.satd : h->pixf.sad_aligned, sizeof(h->pixf.mbcmp) );
 849     memcpy( h->pixf.mbcmp_unaligned, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.mbcmp_unaligned) );
 850     h->pixf.intra_mbcmp_x3_16x16 = satd ? h->pixf.intra_satd_x3_16x16 : h->pixf.intra_sad_x3_16x16;
 851     h->pixf.intra_mbcmp_x3_8x8c = satd ? h->pixf.intra_satd_x3_8x8c : h->pixf.intra_sad_x3_8x8c;
 852     h->pixf.intra_mbcmp_x3_8x8 = satd ? h->pixf.intra_sa8d_x3_8x8 : h->pixf.intra_sad_x3_8x8;
 853     h->pixf.intra_mbcmp_x3_4x4 = satd ? h->pixf.intra_satd_x3_4x4 : h->pixf.intra_sad_x3_4x4;
 854     satd &= h->param.analyse.i_me_method == X264_ME_TESA;
 855     memcpy( h->pixf.fpelcmp, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.fpelcmp) );
 856     memcpy( h->pixf.fpelcmp_x3, satd ? h->pixf.satd_x3 : h->pixf.sad_x3, sizeof(h->pixf.fpelcmp_x3) );
 857     memcpy( h->pixf.fpelcmp_x4, satd ? h->pixf.satd_x4 : h->pixf.sad_x4, sizeof(h->pixf.fpelcmp_x4) );
 858 }
 859
 860 static void x264_set_aspect_ratio( x264_t *h, x264_param_t *param, int initial )
 861 {
 862     /* VUI */
 863     if( param->vui.i_sar_width > 0 && param->vui.i_sar_height > 0 )
 864     {
 865         uint32_t i_w = param->vui.i_sar_width;
 866         uint32_t i_h = param->vui.i_sar_height;
 867         uint32_t old_w = h->param.vui.i_sar_width;
 868         uint32_t old_h = h->param.vui.i_sar_height;
 869
 870         x264_reduce_fraction( &i_w, &i_h );
 871
 872         while( i_w > 65535 || i_h > 65535 )
 873         {
 874             i_w /= 2;
 875             i_h /= 2;
 876         }
 877
 878         x264_reduce_fraction( &i_w, &i_h );
 879
 880         if( i_w != old_w || i_h != old_h || initial )
 881         {
 882             h->param.vui.i_sar_width = 0;
 883             h->param.vui.i_sar_height = 0;
 884             if( i_w == 0 || i_h == 0 )
 885                 x264_log( h, X264_LOG_WARNING, "cannot create valid sample aspect ratio\n" );
 886             else
 887             {
 888                 x264_log( h, initial?X264_LOG_INFO:X264_LOG_DEBUG, "using SAR=%d/%d\n", i_w, i_h );
 889                 h->param.vui.i_sar_width = i_w;
 890                 h->param.vui.i_sar_height = i_h;
 891             }
 892         }
 893     }
 894 }
 895
 896 /****************************************************************************
 897  * x264_encoder_open:
 898  ****************************************************************************/
 899 x264_t *x264_encoder_open( x264_param_t *param )
 900 {
 901     x264_t *h;
 902     char buf[1000], *p;
 903     int qp, i_slicetype_length;
 904
 905     CHECKED_MALLOCZERO( h, sizeof(x264_t) );
 906
 907     /* Create a copy of param */
 908     memcpy( &h->param, param, sizeof(x264_param_t) );
 909
 910     if( param->param_free )
 911         param->param_free( param );
 912
 913     if( x264_validate_parameters( h ) < 0 )
 914         goto fail;
 915
 916     if( h->param.psz_cqm_file )
 917         if( x264_cqm_parse_file( h, h->param.psz_cqm_file ) < 0 )
 918             goto fail;
 919
 920     if( h->param.rc.psz_stat_out )
 921         h->param.rc.psz_stat_out = strdup( h->param.rc.psz_stat_out );
 922     if( h->param.rc.psz_stat_in )
 923         h->param.rc.psz_stat_in = strdup( h->param.rc.psz_stat_in );
 924
 925     x264_set_aspect_ratio( h, &h->param, 1 );
 926
 927     x264_reduce_fraction( &h->param.i_fps_num, &h->param.i_fps_den );
 928     x264_reduce_fraction( &h->param.i_timebase_num, &h->param.i_timebase_den );
 929
 930     /* Init x264_t */
 931     h->i_frame = -1;
 932     h->i_frame_num = 0;
 933     h->i_idr_pic_id = 0;
 934     uint64_t new_timebase_den = h->param.i_timebase_den;
 935     if( h->param.b_dts_compress )
 936     {
 937         /* h->i_dts_compress_multiplier == h->frames.i_bframe_delay + 1 */
 938         h->i_dts_compress_multiplier = h->param.i_bframe ? (h->param.i_bframe_pyramid ? 3 : 2) : 1;
 939         if( h->i_dts_compress_multiplier != 1 )
 940         {
 941             new_timebase_den = h->param.i_timebase_den * h->i_dts_compress_multiplier;
 942             x264_log( h, X264_LOG_DEBUG, "DTS compression changed timebase: %u/%u -> %u/%"PRIu64"\n",
 943                       h->param.i_timebase_num, h->param.i_timebase_den,
 944                       h->param.i_timebase_num, new_timebase_den );
 945         }
 946     }
 947     else
 948         h->i_dts_compress_multiplier = 1;
 949
 950     if( new_timebase_den * 2 > UINT32_MAX )
 951     {
 952         x264_log( h, X264_LOG_ERROR, "Effective timebase denominator %"PRIu64" exceeds H.264 maximum\n", new_timebase_den );
 953         goto fail;
 954     }
 955     h->param.i_timebase_den = new_timebase_den;
 956
 957     h->sps = &h->sps_array[0];
 958     x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
 959
 960     h->pps = &h->pps_array[0];
 961     x264_pps_init( h->pps, h->param.i_sps_id, &h->param, h->sps );
 962
 963     x264_validate_levels( h, 1 );
 964
 965     h->chroma_qp_table = i_chroma_qp_table + 12 + h->pps->i_chroma_qp_index_offset;
 966
 967     if( x264_cqm_init( h ) < 0 )
 968         goto fail;
 969
 970     h->mb.i_mb_width = h->sps->i_mb_width;
 971     h->mb.i_mb_height = h->sps->i_mb_height;
 972     h->mb.i_mb_count = h->mb.i_mb_width * h->mb.i_mb_height;
 973
 974     /* Init frames. */
 975     if( h->param.i_bframe_adaptive == X264_B_ADAPT_TRELLIS && !h->param.rc.b_stat_read )
 976         h->frames.i_delay = X264_MAX(h->param.i_bframe,3)*4;
 977     else
 978         h->frames.i_delay = h->param.i_bframe;
 979     if( h->param.rc.b_mb_tree || h->param.rc.i_vbv_buffer_size )
 980         h->frames.i_delay = X264_MAX( h->frames.i_delay, h->param.rc.i_lookahead );
 981     i_slicetype_length = h->frames.i_delay;
 982     h->frames.i_delay += h->i_thread_frames - 1;
 983     h->frames.i_delay += h->param.i_sync_lookahead;
 984     h->frames.i_delay += h->param.b_vfr_input && (h->param.rc.i_rc_method == X264_RC_ABR || h->param.rc.b_stat_write
 985                                                  || h->param.rc.i_vbv_buffer_size);
 986     h->frames.i_bframe_delay = h->param.i_bframe ? (h->param.i_bframe_pyramid ? 2 : 1) : 0;
 987
 988     h->frames.i_max_ref0 = h->param.i_frame_reference;
 989     h->frames.i_max_ref1 = X264_MIN( h->sps->vui.i_num_reorder_frames, h->param.i_frame_reference );
 990     h->frames.i_max_dpb  = h->sps->vui.i_max_dec_frame_buffering;
 991     h->frames.b_have_lowres = !h->param.rc.b_stat_read
 992         && ( h->param.rc.i_rc_method == X264_RC_ABR
 993           || h->param.rc.i_rc_method == X264_RC_CRF
 994           || h->param.i_bframe_adaptive
 995           || h->param.i_scenecut_threshold
 996           || h->param.rc.b_mb_tree
 997           || h->param.analyse.i_weighted_pred == X264_WEIGHTP_SMART );
 998     h->frames.b_have_lowres |= h->param.rc.b_stat_read && h->param.rc.i_vbv_buffer_size > 0;
 999     h->frames.b_have_sub8x8_esa = !!(h->param.analyse.inter & X264_ANALYSE_PSUB8x8);
1000
1001     h->frames.i_last_idr =
1002     h->frames.i_last_keyframe = - h->param.i_keyint_max;
1003     h->frames.i_input    = 0;
1004     h->frames.i_largest_pts = h->frames.i_second_largest_pts = -1;
1005     h->frames.i_poc_last_open_gop = -1;
1006
1007     CHECKED_MALLOCZERO( h->frames.unused[0], (h->frames.i_delay + 3) * sizeof(x264_frame_t *) );
1008     /* Allocate room for max refs plus a few extra just in case. */
1009     CHECKED_MALLOCZERO( h->frames.unused[1], (h->i_thread_frames + X264_REF_MAX + 4) * sizeof(x264_frame_t *) );
1010     CHECKED_MALLOCZERO( h->frames.current, (h->param.i_sync_lookahead + h->param.i_bframe
1011                         + h->i_thread_frames + 3) * sizeof(x264_frame_t *) );
1012     if( h->param.analyse.i_weighted_pred > 0 )
1013         CHECKED_MALLOCZERO( h->frames.blank_unused, h->i_thread_frames * 4 * sizeof(x264_frame_t *) );
1014     h->i_ref0 = 0;
1015     h->i_ref1 = 0;
1016     h->i_cpb_delay = h->i_coded_fields = h->i_disp_fields = h->i_prev_duration = 0;
1017     h->i_disp_fields_last_frame = -1;
1018     x264_rdo_init();
1019
1020     /* init CPU functions */
1021     x264_predict_16x16_init( h->param.cpu, h->predict_16x16 );
1022     x264_predict_8x8c_init( h->param.cpu, h->predict_8x8c );
1023     x264_predict_8x8_init( h->param.cpu, h->predict_8x8, &h->predict_8x8_filter );
1024     x264_predict_4x4_init( h->param.cpu, h->predict_4x4 );
1025     if( !h->param.b_cabac )
1026         x264_init_vlc_tables();
1027     x264_pixel_init( h->param.cpu, &h->pixf );
1028     x264_dct_init( h->param.cpu, &h->dctf );
1029     x264_zigzag_init( h->param.cpu, &h->zigzagf, h->param.b_interlaced );
1030     x264_mc_init( h->param.cpu, &h->mc );
1031     x264_quant_init( h, h->param.cpu, &h->quantf );
1032     x264_deblock_init( h->param.cpu, &h->loopf );
1033     x264_bitstream_init( h->param.cpu, &h->bsf );
1034     x264_dct_init_weights();
1035
1036     mbcmp_init( h );
1037
1038     p = buf + sprintf( buf, "using cpu capabilities:" );
1039     for( int i = 0; x264_cpu_names[i].flags; i++ )
1040     {
1041         if( !strcmp(x264_cpu_names[i].name, "SSE2")
1042             && h->param.cpu & (X264_CPU_SSE2_IS_FAST|X264_CPU_SSE2_IS_SLOW) )
1043             continue;
1044         if( !strcmp(x264_cpu_names[i].name, "SSE3")
1045             && (h->param.cpu & X264_CPU_SSSE3 || !(h->param.cpu & X264_CPU_CACHELINE_64)) )
1046             continue;
1047         if( !strcmp(x264_cpu_names[i].name, "SSE4.1")
1048             && (h->param.cpu & X264_CPU_SSE42) )
1049             continue;
1050         if( (h->param.cpu & x264_cpu_names[i].flags) == x264_cpu_names[i].flags
1051             && (!i || x264_cpu_names[i].flags != x264_cpu_names[i-1].flags) )
1052             p += sprintf( p, " %s", x264_cpu_names[i].name );
1053     }
1054     if( !h->param.cpu )
1055         p += sprintf( p, " none!" );
1056     x264_log( h, X264_LOG_INFO, "%s\n", buf );
1057
1058     for( qp = h->param.rc.i_qp_min; qp <= h->param.rc.i_qp_max; qp++ )
1059         if( x264_analyse_init_costs( h, qp ) )
1060             goto fail;
1061     if( x264_analyse_init_costs( h, X264_LOOKAHEAD_QP ) )
1062         goto fail;
1063
1064     static const uint16_t cost_mv_correct[7] = { 24, 47, 95, 189, 379, 757, 1515 };
1065     /* Checks for known miscompilation issues. */
1066     if( h->cost_mv[x264_lambda_tab[X264_LOOKAHEAD_QP]][2013] != cost_mv_correct[BIT_DEPTH-8] )
1067     {
1068         x264_log( h, X264_LOG_ERROR, "MV cost test failed: x264 has been miscompiled!\n" );
1069         goto fail;
1070     }
1071
1072     /* Must be volatile or else GCC will optimize it out. */
1073     volatile int temp = 392;
1074     if( x264_clz( temp ) != 23 )
1075     {
1076         x264_log( h, X264_LOG_ERROR, "CLZ test failed: x264 has been miscompiled!\n" );
1077 #if ARCH_X86 || ARCH_X86_64
1078         x264_log( h, X264_LOG_ERROR, "Are you attempting to run an SSE4a-targeted build on a CPU that\n" );
1079         x264_log( h, X264_LOG_ERROR, "doesn't support it?\n" );
1080 #endif
1081         goto fail;
1082     }
1083
1084     h->out.i_nal = 0;
1085     h->out.i_bitstream = X264_MAX( 1000000, h->param.i_width * h->param.i_height * 4
1086         * ( h->param.rc.i_rc_method == X264_RC_ABR ? pow( 0.95, h->param.rc.i_qp_min )
1087           : pow( 0.95, h->param.rc.i_qp_constant ) * X264_MAX( 1, h->param.rc.f_ip_factor )));
1088
1089     CHECKED_MALLOC( h->nal_buffer, h->out.i_bitstream * 3/2 + 4 );
1090     h->nal_buffer_size = h->out.i_bitstream * 3/2 + 4;
1091
1092     if( h->param.i_threads > 1 &&
1093         x264_threadpool_init( &h->threadpool, h->param.i_threads, (void*)x264_encoder_thread_init, h ) )
1094         goto fail;
1095
1096     h->thread[0] = h;
1097     for( int i = 1; i < h->param.i_threads + !!h->param.i_sync_lookahead; i++ )
1098         CHECKED_MALLOC( h->thread[i], sizeof(x264_t) );
1099
1100     for( int i = 0; i < h->param.i_threads; i++ )
1101     {
1102         int init_nal_count = h->param.i_slice_count + 3;
1103         int allocate_threadlocal_data = !h->param.b_sliced_threads || !i;
1104         if( i > 0 )
1105             *h->thread[i] = *h;
1106
1107         if( allocate_threadlocal_data )
1108         {
1109             h->thread[i]->fdec = x264_frame_pop_unused( h, 1 );
1110             if( !h->thread[i]->fdec )
1111                 goto fail;
1112         }
1113         else
1114             h->thread[i]->fdec = h->thread[0]->fdec;
1115
1116         CHECKED_MALLOC( h->thread[i]->out.p_bitstream, h->out.i_bitstream );
1117         /* Start each thread with room for init_nal_count NAL units; it'll realloc later if needed. */
1118         CHECKED_MALLOC( h->thread[i]->out.nal, init_nal_count*sizeof(x264_nal_t) );
1119         h->thread[i]->out.i_nals_allocated = init_nal_count;
1120
1121         if( allocate_threadlocal_data && x264_macroblock_cache_allocate( h->thread[i] ) < 0 )
1122             goto fail;
1123     }
1124
1125     if( x264_lookahead_init( h, i_slicetype_length ) )
1126         goto fail;
1127
1128     for( int i = 0; i < h->param.i_threads; i++ )
1129         if( x264_macroblock_thread_allocate( h->thread[i], 0 ) < 0 )
1130             goto fail;
1131
1132     if( x264_ratecontrol_new( h ) < 0 )
1133         goto fail;
1134
1135     if( h->param.i_nal_hrd )
1136     {
1137         x264_log( h, X264_LOG_DEBUG, "HRD bitrate: %i bits/sec\n", h->sps->vui.hrd.i_bit_rate_unscaled );
1138         x264_log( h, X264_LOG_DEBUG, "CPB size: %i bits\n", h->sps->vui.hrd.i_cpb_size_unscaled );
1139     }
1140
1141     if( h->param.psz_dump_yuv )
1142     {
1143         /* create or truncate the reconstructed video file */
1144         FILE *f = fopen( h->param.psz_dump_yuv, "w" );
1145         if( !f )
1146         {
1147             x264_log( h, X264_LOG_ERROR, "dump_yuv: can't write to %s\n", h->param.psz_dump_yuv );
1148             goto fail;
1149         }
1150         else if( !x264_is_regular_file( f ) )
1151         {
1152             x264_log( h, X264_LOG_ERROR, "dump_yuv: incompatible with non-regular file %s\n", h->param.psz_dump_yuv );
1153             goto fail;
1154         }
1155         fclose( f );
1156     }
1157
1158     const char *profile = h->sps->i_profile_idc == PROFILE_BASELINE ? "Baseline" :
1159                           h->sps->i_profile_idc == PROFILE_MAIN ? "Main" :
1160                           h->sps->i_profile_idc == PROFILE_HIGH ? "High" :
1161                           h->sps->i_profile_idc == PROFILE_HIGH10 ? "High 10" :
1162                           "High 4:4:4 Predictive";
1163     char level[4];
1164     snprintf( level, sizeof(level), "%d.%d", h->sps->i_level_idc/10, h->sps->i_level_idc%10 );
1165     if( h->sps->i_level_idc == 9 || ( h->sps->i_level_idc == 11 && h->sps->b_constraint_set3 ) )
1166         strcpy( level, "1b" );
1167
1168     if( h->sps->i_profile_idc < PROFILE_HIGH10 )
1169     {
1170         x264_log( h, X264_LOG_INFO, "profile %s, level %s\n",
1171             profile, level );
1172     }
1173     else
1174     {
1175         x264_log( h, X264_LOG_INFO, "profile %s, level %s, bit depth %d\n",
1176             profile, level, BIT_DEPTH );
1177     }
1178
1179     return h;
1180 fail:
1181     x264_free( h );
1182     return NULL;
1183 }
1184
1185 /****************************************************************************
1186  * x264_encoder_reconfig:
1187  ****************************************************************************/
1188 int x264_encoder_reconfig( x264_t *h, x264_param_t *param )
1189 {
1190     int rc_reconfig = 0;
1191     h = h->thread[h->thread[0]->i_thread_phase];
1192     x264_set_aspect_ratio( h, param, 0 );
1193 #define COPY(var) h->param.var = param->var
1194     COPY( i_frame_reference ); // but never uses more refs than initially specified
1195     COPY( i_bframe_bias );
1196     if( h->param.i_scenecut_threshold )
1197         COPY( i_scenecut_threshold ); // can't turn it on or off, only vary the threshold
1198     COPY( b_deblocking_filter );
1199     COPY( i_deblocking_filter_alphac0 );
1200     COPY( i_deblocking_filter_beta );
1201     COPY( analyse.inter );
1202     COPY( analyse.intra );
1203     COPY( analyse.i_direct_mv_pred );
1204     /* Scratch buffer prevents me_range from being increased for esa/tesa */
1205     if( h->param.analyse.i_me_method < X264_ME_ESA || param->analyse.i_me_range < h->param.analyse.i_me_range )
1206         COPY( analyse.i_me_range );
1207     COPY( analyse.i_noise_reduction );
1208     /* We can't switch out of subme=0 during encoding. */
1209     if( h->param.analyse.i_subpel_refine )
1210         COPY( analyse.i_subpel_refine );
1211     COPY( analyse.i_trellis );
1212     COPY( analyse.b_chroma_me );
1213     COPY( analyse.b_dct_decimate );
1214     COPY( analyse.b_fast_pskip );
1215     COPY( analyse.b_mixed_references );
1216     COPY( analyse.f_psy_rd );
1217     COPY( analyse.f_psy_trellis );
1218     // can only twiddle these if they were enabled to begin with:
1219     if( h->param.analyse.i_me_method >= X264_ME_ESA || param->analyse.i_me_method < X264_ME_ESA )
1220         COPY( analyse.i_me_method );
1221     if( h->param.analyse.i_me_method >= X264_ME_ESA && !h->frames.b_have_sub8x8_esa )
1222         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
1223     if( h->pps->b_transform_8x8_mode )
1224         COPY( analyse.b_transform_8x8 );
1225     if( h->frames.i_max_ref1 > 1 )
1226         COPY( i_bframe_pyramid );
1227     COPY( i_slice_max_size );
1228     COPY( i_slice_max_mbs );
1229     COPY( i_slice_count );
1230
1231     /* VBV can't be turned on if it wasn't on to begin with */
1232     if( h->param.rc.i_vbv_max_bitrate > 0 && h->param.rc.i_vbv_buffer_size > 0 &&
1233           param->rc.i_vbv_max_bitrate > 0 &&   param->rc.i_vbv_buffer_size > 0 )
1234     {
1235         COPY( rc.i_vbv_max_bitrate );
1236         COPY( rc.i_vbv_buffer_size );
1237         COPY( rc.i_bitrate );
1238         rc_reconfig = 1;
1239     }
1240     if( h->param.rc.f_rf_constant != param->rc.f_rf_constant )
1241     {
1242         COPY( rc.f_rf_constant );
1243         rc_reconfig = 1;
1244     }
1245     if( h->param.rc.f_rf_constant_max != param->rc.f_rf_constant_max )
1246     {
1247         COPY( rc.f_rf_constant_max );
1248         rc_reconfig = 1;
1249     }
1250
1251 #undef COPY
1252
1253     mbcmp_init( h );
1254
1255     int ret = x264_validate_parameters( h );
1256
1257     /* Supported reconfiguration options (1-pass only):
1258      * vbv-maxrate
1259      * vbv-bufsize
1260      * crf
1261      * bitrate (CBR only) */
1262     if( !ret && rc_reconfig )
1263         x264_ratecontrol_init_reconfigurable( h, 0 );
1264
1265     return ret;
1266 }
1267
1268 /****************************************************************************
1269  * x264_encoder_parameters:
1270  ****************************************************************************/
1271 void x264_encoder_parameters( x264_t *h, x264_param_t *param )
1272 {
1273     memcpy( param, &h->thread[h->i_thread_phase]->param, sizeof(x264_param_t) );
1274 }
1275
1276 /* internal usage */
1277 static void x264_nal_start( x264_t *h, int i_type, int i_ref_idc )
1278 {
1279     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1280
1281     nal->i_ref_idc        = i_ref_idc;
1282     nal->i_type           = i_type;
1283     nal->b_long_startcode = 1;
1284
1285     nal->i_payload= 0;
1286     nal->p_payload= &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8];
1287 }
1288
1289 /* if number of allocated nals is not enough, re-allocate a larger one. */
1290 static int x264_nal_check_buffer( x264_t *h )
1291 {
1292     if( h->out.i_nal >= h->out.i_nals_allocated )
1293     {
1294         x264_nal_t *new_out = x264_malloc( sizeof(x264_nal_t) * (h->out.i_nals_allocated*2) );
1295         if( !new_out )
1296             return -1;
1297         memcpy( new_out, h->out.nal, sizeof(x264_nal_t) * (h->out.i_nals_allocated) );
1298         x264_free( h->out.nal );
1299         h->out.nal = new_out;
1300         h->out.i_nals_allocated *= 2;
1301     }
1302     return 0;
1303 }
1304
1305 static int x264_nal_end( x264_t *h )
1306 {
1307     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1308     nal->i_payload = &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8] - nal->p_payload;
1309     if( h->param.nalu_process )
1310         h->param.nalu_process( h, nal );
1311     h->out.i_nal++;
1312
1313     return x264_nal_check_buffer( h );
1314 }
1315
1316 static int x264_encoder_encapsulate_nals( x264_t *h, int start )
1317 {
1318     int nal_size = 0, previous_nal_size = 0;
1319
1320     if( h->param.nalu_process )
1321     {
1322         for( int i = start; i < h->out.i_nal; i++ )
1323             nal_size += h->out.nal[i].i_payload;
1324         return nal_size;
1325     }
1326
1327     for( int i = 0; i < start; i++ )
1328         previous_nal_size += h->out.nal[i].i_payload;
1329
1330     for( int i = start; i < h->out.i_nal; i++ )
1331         nal_size += h->out.nal[i].i_payload;
1332
1333     /* Worst-case NAL unit escaping: reallocate the buffer if it's too small. */
1334     if( h->nal_buffer_size < nal_size * 3/2 + h->out.i_nal * 4 )
1335     {
1336         uint8_t *buf = x264_malloc( nal_size * 2 + h->out.i_nal * 4 );
1337         if( !buf )
1338             return -1;
1339         if( previous_nal_size )
1340             memcpy( buf, h->nal_buffer, previous_nal_size );
1341         x264_free( h->nal_buffer );
1342         h->nal_buffer = buf;
1343     }
1344
1345     uint8_t *nal_buffer = h->nal_buffer + previous_nal_size;
1346
1347     for( int i = start; i < h->out.i_nal; i++ )
1348     {
1349         h->out.nal[i].b_long_startcode = !i || h->out.nal[i].i_type == NAL_SPS || h->out.nal[i].i_type == NAL_PPS;
1350         x264_nal_encode( h, nal_buffer, &h->out.nal[i] );
1351         nal_buffer += h->out.nal[i].i_payload;
1352     }
1353
1354     x264_emms();
1355
1356     return nal_buffer - (h->nal_buffer + previous_nal_size);
1357 }
1358
1359 /****************************************************************************
1360  * x264_encoder_headers:
1361  ****************************************************************************/
1362 int x264_encoder_headers( x264_t *h, x264_nal_t **pp_nal, int *pi_nal )
1363 {
1364     int frame_size = 0;
1365     /* init bitstream context */
1366     h->out.i_nal = 0;
1367     bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
1368
1369     /* Write SEI, SPS and PPS. */
1370
1371     /* generate sequence parameters */
1372     x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
1373     x264_sps_write( &h->out.bs, h->sps );
1374     if( x264_nal_end( h ) )
1375         return -1;
1376
1377     /* generate picture parameters */
1378     x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
1379     x264_pps_write( &h->out.bs, h->pps );
1380     if( x264_nal_end( h ) )
1381         return -1;
1382
1383     /* identify ourselves */
1384     x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
1385     if( x264_sei_version_write( h, &h->out.bs ) )
1386         return -1;
1387     if( x264_nal_end( h ) )
1388         return -1;
1389
1390     frame_size = x264_encoder_encapsulate_nals( h, 0 );
1391
1392     /* now set output*/
1393     *pi_nal = h->out.i_nal;
1394     *pp_nal = &h->out.nal[0];
1395     h->out.i_nal = 0;
1396
1397     return frame_size;
1398 }
1399
1400 /* Check to see whether we have chosen a reference list ordering different
1401  * from the standard's default. */
1402 static inline void x264_reference_check_reorder( x264_t *h )
1403 {
1404     /* The reorder check doesn't check for missing frames, so just
1405      * force a reorder if one of the reference list is corrupt. */
1406     for( int i = 0; h->frames.reference[i]; i++ )
1407         if( h->frames.reference[i]->b_corrupt )
1408         {
1409             h->b_ref_reorder[0] = 1;
1410             return;
1411         }
1412     for( int i = 0; i < h->i_ref0 - 1; i++ )
1413         /* P and B-frames use different default orders. */
1414         if( h->sh.i_type == SLICE_TYPE_P ? h->fref0[i]->i_frame_num < h->fref0[i+1]->i_frame_num
1415                                          : h->fref0[i]->i_poc < h->fref0[i+1]->i_poc )
1416         {
1417             h->b_ref_reorder[0] = 1;
1418             return;
1419         }
1420 }
1421
1422 /* return -1 on failure, else return the index of the new reference frame */
1423 int x264_weighted_reference_duplicate( x264_t *h, int i_ref, const x264_weight_t *w )
1424 {
1425     int i = h->i_ref0;
1426     int j = 1;
1427     x264_frame_t *newframe;
1428     if( i <= 1 ) /* empty list, definitely can't duplicate frame */
1429         return -1;
1430
1431     newframe = x264_frame_pop_blank_unused( h );
1432
1433     //FIXME: probably don't need to copy everything
1434     *newframe = *h->fref0[i_ref];
1435     newframe->i_reference_count = 1;
1436     newframe->orig = h->fref0[i_ref];
1437     newframe->b_duplicate = 1;
1438     memcpy( h->fenc->weight[j], w, sizeof(h->fenc->weight[i]) );
1439
1440     /* shift the frames to make space for the dupe. */
1441     h->b_ref_reorder[0] = 1;
1442     if( h->i_ref0 < X264_REF_MAX )
1443         ++h->i_ref0;
1444     h->fref0[X264_REF_MAX-1] = NULL;
1445     x264_frame_unshift( &h->fref0[j], newframe );
1446
1447     return j;
1448 }
1449
1450 static void x264_weighted_pred_init( x264_t *h )
1451 {
1452     /* for now no analysis and set all weights to nothing */
1453     for( int i_ref = 0; i_ref < h->i_ref0; i_ref++ )
1454         h->fenc->weighted[i_ref] = h->fref0[i_ref]->filtered[0];
1455
1456     // FIXME: This only supports weighting of one reference frame
1457     // and duplicates of that frame.
1458     h->fenc->i_lines_weighted = 0;
1459
1460     for( int i_ref = 0; i_ref < (h->i_ref0 << h->sh.b_mbaff); i_ref++ )
1461         for( int i = 0; i < 3; i++ )
1462             h->sh.weight[i_ref][i].weightfn = NULL;
1463
1464
1465     if( h->sh.i_type != SLICE_TYPE_P || h->param.analyse.i_weighted_pred <= 0 )
1466         return;
1467
1468     int i_padv = PADV << h->param.b_interlaced;
1469     int denom = -1;
1470     int weightluma = 0;
1471     int buffer_next = 0;
1472     //FIXME: when chroma support is added, move this into loop
1473     h->sh.weight[0][1].weightfn = h->sh.weight[0][2].weightfn = NULL;
1474     h->sh.weight[0][1].i_denom = h->sh.weight[0][2].i_denom = 0;
1475     for( int j = 0; j < h->i_ref0; j++ )
1476     {
1477         if( h->fenc->weight[j][0].weightfn )
1478         {
1479             h->sh.weight[j][0] = h->fenc->weight[j][0];
1480             // if weight is useless, don't write it to stream
1481             if( h->sh.weight[j][0].i_scale == 1<<h->sh.weight[j][0].i_denom && h->sh.weight[j][0].i_offset == 0 )
1482                 h->sh.weight[j][0].weightfn = NULL;
1483             else
1484             {
1485                 if( !weightluma )
1486                 {
1487                     weightluma = 1;
1488                     h->sh.weight[0][0].i_denom = denom = h->sh.weight[j][0].i_denom;
1489                     assert( x264_clip3( denom, 0, 7 ) == denom );
1490                 }
1491                 assert( h->sh.weight[j][0].i_denom == denom );
1492                 assert( x264_clip3( h->sh.weight[j][0].i_scale, 0, 127 ) == h->sh.weight[j][0].i_scale );
1493                 assert( x264_clip3( h->sh.weight[j][0].i_offset, -128, 127 ) == h->sh.weight[j][0].i_offset );
1494                 h->fenc->weighted[j] = h->mb.p_weight_buf[buffer_next++] +
1495                     h->fenc->i_stride[0] * i_padv + PADH;
1496             }
1497         }
1498
1499         //scale full resolution frame
1500         if( h->sh.weight[j][0].weightfn && h->param.i_threads == 1 )
1501         {
1502             pixel *src = h->fref0[j]->filtered[0] - h->fref0[j]->i_stride[0]*i_padv - PADH;
1503             pixel *dst = h->fenc->weighted[j] - h->fenc->i_stride[0]*i_padv - PADH;
1504             int stride = h->fenc->i_stride[0];
1505             int width = h->fenc->i_width[0] + PADH*2;
1506             int height = h->fenc->i_lines[0] + i_padv*2;
1507             x264_weight_scale_plane( h, dst, stride, src, stride, width, height, &h->sh.weight[j][0] );
1508             h->fenc->i_lines_weighted = height;
1509         }
1510     }
1511     if( !weightluma )
1512         h->sh.weight[0][0].i_denom = 0;
1513 }
1514
1515 static inline void x264_reference_build_list( x264_t *h, int i_poc )
1516 {
1517     int b_ok;
1518
1519     /* build ref list 0/1 */
1520     h->mb.pic.i_fref[0] = h->i_ref0 = 0;
1521     h->mb.pic.i_fref[1] = h->i_ref1 = 0;
1522     if( h->sh.i_type == SLICE_TYPE_I )
1523         return;
1524
1525     for( int i = 0; h->frames.reference[i]; i++ )
1526     {
1527         if( h->frames.reference[i]->b_corrupt )
1528             continue;
1529         if( h->frames.reference[i]->i_poc < i_poc )
1530             h->fref0[h->i_ref0++] = h->frames.reference[i];
1531         else if( h->frames.reference[i]->i_poc > i_poc )
1532             h->fref1[h->i_ref1++] = h->frames.reference[i];
1533     }
1534
1535     /* Order ref0 from higher to lower poc */
1536     do
1537     {
1538         b_ok = 1;
1539         for( int i = 0; i < h->i_ref0 - 1; i++ )
1540         {
1541             if( h->fref0[i]->i_poc < h->fref0[i+1]->i_poc )
1542             {
1543                 XCHG( x264_frame_t*, h->fref0[i], h->fref0[i+1] );
1544                 b_ok = 0;
1545                 break;
1546             }
1547         }
1548     } while( !b_ok );
1549
1550     if( h->sh.i_mmco_remove_from_end )
1551         for( int i = h->i_ref0-1; i >= h->i_ref0 - h->sh.i_mmco_remove_from_end; i-- )
1552         {
1553             int diff = h->i_frame_num - h->fref0[i]->i_frame_num;
1554             h->sh.mmco[h->sh.i_mmco_command_count].i_poc = h->fref0[i]->i_poc;
1555             h->sh.mmco[h->sh.i_mmco_command_count++].i_difference_of_pic_nums = diff;
1556         }
1557
1558     /* Order ref1 from lower to higher poc (bubble sort) for B-frame */
1559     do
1560     {
1561         b_ok = 1;
1562         for( int i = 0; i < h->i_ref1 - 1; i++ )
1563         {
1564             if( h->fref1[i]->i_poc > h->fref1[i+1]->i_poc )
1565             {
1566                 XCHG( x264_frame_t*, h->fref1[i], h->fref1[i+1] );
1567                 b_ok = 0;
1568                 break;
1569             }
1570         }
1571     } while( !b_ok );
1572
1573     x264_reference_check_reorder( h );
1574
1575     h->i_ref1 = X264_MIN( h->i_ref1, h->frames.i_max_ref1 );
1576     h->i_ref0 = X264_MIN( h->i_ref0, h->frames.i_max_ref0 );
1577     h->i_ref0 = X264_MIN( h->i_ref0, h->param.i_frame_reference ); // if reconfig() has lowered the limit
1578
1579     /* add duplicates */
1580     if( h->fenc->i_type == X264_TYPE_P )
1581     {
1582         int idx = -1;
1583         if( h->param.analyse.i_weighted_pred == X264_WEIGHTP_SMART )
1584         {
1585             x264_weight_t w[3];
1586             w[1].weightfn = w[2].weightfn = NULL;
1587             if( h->param.rc.b_stat_read )
1588                 x264_ratecontrol_set_weights( h, h->fenc );
1589
1590             if( !h->fenc->weight[0][0].weightfn )
1591             {
1592                 h->fenc->weight[0][0].i_denom = 0;
1593                 SET_WEIGHT( w[0], 1, 1, 0, -1 );
1594                 idx = x264_weighted_reference_duplicate( h, 0, w );
1595             }
1596             else
1597             {
1598                 if( h->fenc->weight[0][0].i_scale == 1<<h->fenc->weight[0][0].i_denom )
1599                 {
1600                     SET_WEIGHT( h->fenc->weight[0][0], 1, 1, 0, h->fenc->weight[0][0].i_offset );
1601                 }
1602                 x264_weighted_reference_duplicate( h, 0, weight_none );
1603                 if( h->fenc->weight[0][0].i_offset > -128 )
1604                 {
1605                     w[0] = h->fenc->weight[0][0];
1606                     w[0].i_offset--;
1607                     h->mc.weight_cache( h, &w[0] );
1608                     idx = x264_weighted_reference_duplicate( h, 0, w );
1609                 }
1610             }
1611         }
1612         else if( h->param.analyse.i_weighted_pred == X264_WEIGHTP_BLIND )
1613         {
1614             //weighted offset=-1
1615             x264_weight_t w[3];
1616             SET_WEIGHT( w[0], 1, 1, 0, -1 );
1617             h->fenc->weight[0][0].i_denom = 0;
1618             w[1].weightfn = w[2].weightfn = NULL;
1619             idx = x264_weighted_reference_duplicate( h, 0, w );
1620         }
1621         h->mb.ref_blind_dupe = idx;
1622     }
1623
1624     assert( h->i_ref0 + h->i_ref1 <= X264_REF_MAX );
1625     h->mb.pic.i_fref[0] = h->i_ref0;
1626     h->mb.pic.i_fref[1] = h->i_ref1;
1627 }
1628
1629 static void x264_fdec_filter_row( x264_t *h, int mb_y, int b_inloop )
1630 {
1631     /* mb_y is the mb to be encoded next, not the mb to be filtered here */
1632     int b_hpel = h->fdec->b_kept_as_ref;
1633     int b_deblock = h->sh.i_disable_deblocking_filter_idc != 1;
1634     int b_end = mb_y == h->i_threadslice_end;
1635     int b_measure_quality = 1;
1636     int min_y = mb_y - (1 << h->sh.b_mbaff);
1637     int b_start = min_y == h->i_threadslice_start;
1638     int max_y = b_end ? h->i_threadslice_end : mb_y;
1639     b_deblock &= b_hpel || h->param.psz_dump_yuv;
1640     if( h->param.b_sliced_threads && b_start && min_y && !b_inloop )
1641     {
1642         b_deblock = 0;         /* We already deblocked on the inloop pass. */
1643         b_measure_quality = 0; /* We already measured quality on the inloop pass. */
1644     }
1645     if( mb_y & h->sh.b_mbaff )
1646         return;
1647     if( min_y < h->i_threadslice_start )
1648         return;
1649
1650     if( b_deblock )
1651         for( int y = min_y; y < max_y; y += (1 << h->sh.b_mbaff) )
1652             x264_frame_deblock_row( h, y );
1653
1654     if( b_hpel )
1655     {
1656         int end = mb_y == h->mb.i_mb_height;
1657         x264_frame_expand_border( h, h->fdec, min_y, end );
1658         if( h->param.analyse.i_subpel_refine )
1659         {
1660             x264_frame_filter( h, h->fdec, min_y, end );
1661             x264_frame_expand_border_filtered( h, h->fdec, min_y, end );
1662         }
1663     }
1664
1665     if( h->i_thread_frames > 1 && h->fdec->b_kept_as_ref )
1666         x264_frame_cond_broadcast( h->fdec, mb_y*16 + (b_end ? 10000 : -(X264_THREAD_HEIGHT << h->sh.b_mbaff)) );
1667
1668     min_y = min_y*16 - 8 * !b_start;
1669     max_y = b_end ? X264_MIN( h->i_threadslice_end*16 , h->param.i_height ) : mb_y*16 - 8;
1670
1671     if( b_measure_quality )
1672     {
1673         if( h->param.analyse.b_psnr )
1674         {
1675             uint64_t ssd_y = x264_pixel_ssd_wxh( &h->pixf,
1676                 h->fdec->plane[0] + min_y * h->fdec->i_stride[0], h->fdec->i_stride[0],
1677                 h->fenc->plane[0] + min_y * h->fenc->i_stride[0], h->fenc->i_stride[0],
1678                 h->param.i_width, max_y-min_y );
1679             uint64_t ssd_uv = x264_pixel_ssd_nv12( &h->pixf,
1680                 h->fdec->plane[1] + (min_y>>1) * h->fdec->i_stride[1], h->fdec->i_stride[1],
1681                 h->fenc->plane[1] + (min_y>>1) * h->fenc->i_stride[1], h->fenc->i_stride[1],
1682                 h->param.i_width>>1, (max_y-min_y)>>1 );
1683             h->stat.frame.i_ssd[0] += ssd_y;
1684             h->stat.frame.i_ssd[1] += (uint32_t)ssd_uv;
1685             h->stat.frame.i_ssd[2] += ssd_uv>>32;
1686         }
1687
1688         if( h->param.analyse.b_ssim )
1689         {
1690             x264_emms();
1691             /* offset by 2 pixels to avoid alignment of ssim blocks with dct blocks,
1692              * and overlap by 4 */
1693             min_y += b_start ? 2 : -6;
1694             h->stat.frame.f_ssim +=
1695                 x264_pixel_ssim_wxh( &h->pixf,
1696                     h->fdec->plane[0] + 2+min_y*h->fdec->i_stride[0], h->fdec->i_stride[0],
1697                     h->fenc->plane[0] + 2+min_y*h->fenc->i_stride[0], h->fenc->i_stride[0],
1698                     h->param.i_width-2, max_y-min_y, h->scratch_buffer );
1699         }
1700     }
1701 }
1702
1703 static inline int x264_reference_update( x264_t *h )
1704 {
1705     if( !h->fdec->b_kept_as_ref )
1706     {
1707         if( h->i_thread_frames > 1 )
1708         {
1709             x264_frame_push_unused( h, h->fdec );
1710             h->fdec = x264_frame_pop_unused( h, 1 );
1711             if( !h->fdec )
1712                 return -1;
1713         }
1714         return 0;
1715     }
1716
1717     /* apply mmco from previous frame. */
1718     for( int i = 0; i < h->sh.i_mmco_command_count; i++ )
1719         for( int j = 0; h->frames.reference[j]; j++ )
1720             if( h->frames.reference[j]->i_poc == h->sh.mmco[i].i_poc )
1721                 x264_frame_push_unused( h, x264_frame_shift( &h->frames.reference[j] ) );
1722
1723     /* move frame in the buffer */
1724     x264_frame_push( h->frames.reference, h->fdec );
1725     if( h->frames.reference[h->sps->i_num_ref_frames] )
1726         x264_frame_push_unused( h, x264_frame_shift( h->frames.reference ) );
1727     h->fdec = x264_frame_pop_unused( h, 1 );
1728     if( !h->fdec )
1729         return -1;
1730     return 0;
1731 }
1732
1733 static inline void x264_reference_reset( x264_t *h )
1734 {
1735     while( h->frames.reference[0] )
1736         x264_frame_push_unused( h, x264_frame_pop( h->frames.reference ) );
1737     h->fdec->i_poc =
1738     h->fenc->i_poc = 0;
1739 }
1740
1741 static inline void x264_reference_hierarchy_reset( x264_t *h )
1742 {
1743     int ref;
1744     int b_hasdelayframe = 0;
1745
1746     /* look for delay frames -- chain must only contain frames that are disposable */
1747     for( int i = 0; h->frames.current[i] && IS_DISPOSABLE( h->frames.current[i]->i_type ); i++ )
1748         b_hasdelayframe |= h->frames.current[i]->i_coded
1749                         != h->frames.current[i]->i_frame + h->sps->vui.i_num_reorder_frames;
1750
1751     /* This function must handle b-pyramid and clear frames for open-gop */
1752     if( h->param.i_bframe_pyramid != X264_B_PYRAMID_STRICT && !b_hasdelayframe && h->frames.i_poc_last_open_gop == -1 )
1753         return;
1754
1755     /* Remove last BREF. There will never be old BREFs in the
1756      * dpb during a BREF decode when pyramid == STRICT */
1757     for( ref = 0; h->frames.reference[ref]; ref++ )
1758     {
1759         if( ( h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT
1760             && h->frames.reference[ref]->i_type == X264_TYPE_BREF )
1761             || ( h->frames.reference[ref]->i_poc < h->frames.i_poc_last_open_gop
1762             && h->sh.i_type != SLICE_TYPE_B ) )
1763         {
1764             int diff = h->i_frame_num - h->frames.reference[ref]->i_frame_num;
1765             h->sh.mmco[h->sh.i_mmco_command_count].i_difference_of_pic_nums = diff;
1766             h->sh.mmco[h->sh.i_mmco_command_count++].i_poc = h->frames.reference[ref]->i_poc;
1767             x264_frame_push_unused( h, x264_frame_shift( &h->frames.reference[ref] ) );
1768             h->b_ref_reorder[0] = 1;
1769             ref--;
1770         }
1771     }
1772
1773     /* Prepare room in the dpb for the delayed display time of the later b-frame's */
1774     if( h->param.i_bframe_pyramid )
1775         h->sh.i_mmco_remove_from_end = X264_MAX( ref + 2 - h->frames.i_max_dpb, 0 );
1776 }
1777
1778 static inline void x264_slice_init( x264_t *h, int i_nal_type, int i_global_qp )
1779 {
1780     /* ------------------------ Create slice header  ----------------------- */
1781     if( i_nal_type == NAL_SLICE_IDR )
1782     {
1783         x264_slice_header_init( h, &h->sh, h->sps, h->pps, h->i_idr_pic_id, h->i_frame_num, i_global_qp );
1784
1785         /* alternate id */
1786         h->i_idr_pic_id ^= 1;
1787     }
1788     else
1789     {
1790         x264_slice_header_init( h, &h->sh, h->sps, h->pps, -1, h->i_frame_num, i_global_qp );
1791
1792         h->sh.i_num_ref_idx_l0_active = h->i_ref0 <= 0 ? 1 : h->i_ref0;
1793         h->sh.i_num_ref_idx_l1_active = h->i_ref1 <= 0 ? 1 : h->i_ref1;
1794         if( h->sh.i_num_ref_idx_l0_active != h->pps->i_num_ref_idx_l0_default_active ||
1795             (h->sh.i_type == SLICE_TYPE_B && h->sh.i_num_ref_idx_l1_active != h->pps->i_num_ref_idx_l1_default_active) )
1796         {
1797             h->sh.b_num_ref_idx_override = 1;
1798         }
1799     }
1800
1801     h->fdec->i_frame_num = h->sh.i_frame_num;
1802
1803     if( h->sps->i_poc_type == 0 )
1804     {
1805         h->sh.i_poc = h->fdec->i_poc;
1806         if( h->param.b_interlaced )
1807         {
1808             h->sh.i_delta_poc_bottom = h->param.b_tff ? 1 : -1;
1809             if( h->sh.i_delta_poc_bottom == -1 )
1810                 h->sh.i_poc = h->fdec->i_poc + 1;
1811         }
1812         else
1813             h->sh.i_delta_poc_bottom = 0;
1814     }
1815     else if( h->sps->i_poc_type == 1 )
1816     {
1817         /* FIXME TODO FIXME */
1818     }
1819     else
1820     {
1821         /* Nothing to do ? */
1822     }
1823
1824     x264_macroblock_slice_init( h );
1825 }
1826
1827 static int x264_slice_write( x264_t *h )
1828 {
1829     int i_skip;
1830     int mb_xy, i_mb_x, i_mb_y;
1831     int i_skip_bak = 0; /* Shut up GCC. */
1832     bs_t bs_bak;
1833     x264_cabac_t cabac_bak;
1834     uint8_t cabac_prevbyte_bak = 0; /* Shut up GCC. */
1835     int mv_bits_bak = 0;
1836     int tex_bits_bak = 0;
1837     /* Assume no more than 3 bytes of NALU escaping.
1838      * NALUs other than the first use a 3-byte startcode. */
1839     int overhead_guess = (NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal)) + 3;
1840     int slice_max_size = h->param.i_slice_max_size > 0 ? (h->param.i_slice_max_size-overhead_guess)*8 : INT_MAX;
1841     int starting_bits = bs_pos(&h->out.bs);
1842     int b_deblock = h->sh.i_disable_deblocking_filter_idc != 1;
1843     int b_hpel = h->fdec->b_kept_as_ref;
1844     b_deblock &= b_hpel || h->param.psz_dump_yuv;
1845     bs_realign( &h->out.bs );
1846
1847     /* Slice */
1848     x264_nal_start( h, h->i_nal_type, h->i_nal_ref_idc );
1849     h->out.nal[h->out.i_nal].i_first_mb = h->sh.i_first_mb;
1850
1851     /* Slice header */
1852     x264_macroblock_thread_init( h );
1853
1854     /* If this isn't the first slice in the threadslice, set the slice QP
1855      * equal to the last QP in the previous slice for more accurate
1856      * CABAC initialization. */
1857     if( h->sh.i_first_mb != h->i_threadslice_start * h->mb.i_mb_width )
1858     {
1859         h->sh.i_qp = h->mb.i_last_qp;
1860         h->sh.i_qp_delta = h->sh.i_qp - h->pps->i_pic_init_qp;
1861     }
1862
1863     x264_slice_header_write( &h->out.bs, &h->sh, h->i_nal_ref_idc );
1864     if( h->param.b_cabac )
1865     {
1866         /* alignment needed */
1867         bs_align_1( &h->out.bs );
1868
1869         /* init cabac */
1870         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 );
1871         x264_cabac_encode_init ( &h->cabac, h->out.bs.p, h->out.bs.p_end );
1872     }
1873     h->mb.i_last_qp = h->sh.i_qp;
1874     h->mb.i_last_dqp = 0;
1875
1876     i_mb_y = h->sh.i_first_mb / h->mb.i_mb_width;
1877     i_mb_x = h->sh.i_first_mb % h->mb.i_mb_width;
1878     i_skip = 0;
1879
1880     while( (mb_xy = i_mb_x + i_mb_y * h->mb.i_mb_width) <= h->sh.i_last_mb )
1881     {
1882         int mb_spos = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
1883
1884         if( x264_bitstream_check_buffer( h ) )
1885             return -1;
1886
1887         if( h->param.i_slice_max_size > 0 )
1888         {
1889             mv_bits_bak = h->stat.frame.i_mv_bits;
1890             tex_bits_bak = h->stat.frame.i_tex_bits;
1891             /* We don't need the contexts because flushing the CABAC encoder has no context
1892              * dependency and macroblocks are only re-encoded in the case where a slice is
1893              * ended (and thus the content of all contexts are thrown away). */
1894             if( h->param.b_cabac )
1895             {
1896                 memcpy( &cabac_bak, &h->cabac, offsetof(x264_cabac_t, f8_bits_encoded) );
1897                 /* x264's CABAC writer modifies the previous byte during carry, so it has to be
1898                  * backed up. */
1899                 cabac_prevbyte_bak = h->cabac.p[-1];
1900             }
1901             else
1902             {
1903                 bs_bak = h->out.bs;
1904                 i_skip_bak = i_skip;
1905             }
1906         }
1907
1908         if( i_mb_x == 0 && !h->mb.b_reencode_mb )
1909             x264_fdec_filter_row( h, i_mb_y, 1 );
1910
1911         /* load cache */
1912         x264_macroblock_cache_load( h, i_mb_x, i_mb_y );
1913
1914         x264_macroblock_analyse( h );
1915
1916         /* encode this macroblock -> be careful it can change the mb type to P_SKIP if needed */
1917         x264_macroblock_encode( h );
1918
1919         if( h->param.b_cabac )
1920         {
1921             if( mb_xy > h->sh.i_first_mb && !(h->sh.b_mbaff && (i_mb_y&1)) )
1922                 x264_cabac_encode_terminal( &h->cabac );
1923
1924             if( IS_SKIP( h->mb.i_type ) )
1925                 x264_cabac_mb_skip( h, 1 );
1926             else
1927             {
1928                 if( h->sh.i_type != SLICE_TYPE_I )
1929                     x264_cabac_mb_skip( h, 0 );
1930                 x264_macroblock_write_cabac( h, &h->cabac );
1931             }
1932         }
1933         else
1934         {
1935             if( IS_SKIP( h->mb.i_type ) )
1936                 i_skip++;
1937             else
1938             {
1939                 if( h->sh.i_type != SLICE_TYPE_I )
1940                 {
1941                     bs_write_ue( &h->out.bs, i_skip );  /* skip run */
1942                     i_skip = 0;
1943                 }
1944                 x264_macroblock_write_cavlc( h );
1945             }
1946         }
1947
1948         int total_bits = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
1949         int mb_size = total_bits - mb_spos;
1950
1951         /* We'll just re-encode this last macroblock if we go over the max slice size. */
1952         if( total_bits - starting_bits > slice_max_size && !h->mb.b_reencode_mb )
1953         {
1954             if( mb_xy != h->sh.i_first_mb )
1955             {
1956                 h->stat.frame.i_mv_bits = mv_bits_bak;
1957                 h->stat.frame.i_tex_bits = tex_bits_bak;
1958                 if( h->param.b_cabac )
1959                 {
1960                     memcpy( &h->cabac, &cabac_bak, offsetof(x264_cabac_t, f8_bits_encoded) );
1961                     h->cabac.p[-1] = cabac_prevbyte_bak;
1962                 }
1963                 else
1964                 {
1965                     h->out.bs = bs_bak;
1966                     i_skip = i_skip_bak;
1967                 }
1968                 h->mb.b_reencode_mb = 1;
1969                 h->sh.i_last_mb = mb_xy-1;
1970                 break;
1971             }
1972             else
1973             {
1974                 h->sh.i_last_mb = mb_xy;
1975                 h->mb.b_reencode_mb = 0;
1976             }
1977         }
1978         else
1979             h->mb.b_reencode_mb = 0;
1980
1981 #if HAVE_VISUALIZE
1982         if( h->param.b_visualize )
1983             x264_visualize_mb( h );
1984 #endif
1985
1986         /* save cache */
1987         x264_macroblock_cache_save( h );
1988
1989         /* accumulate mb stats */
1990         h->stat.frame.i_mb_count[h->mb.i_type]++;
1991
1992         int b_intra = IS_INTRA( h->mb.i_type );
1993         if( h->param.i_log_level >= X264_LOG_INFO || h->param.rc.b_stat_write )
1994         {
1995             if( !b_intra && !IS_SKIP( h->mb.i_type ) && !IS_DIRECT( h->mb.i_type ) )
1996             {
1997                 if( h->mb.i_partition != D_8x8 )
1998                         h->stat.frame.i_mb_partition[h->mb.i_partition] += 4;
1999                     else
2000                         for( int i = 0; i < 4; i++ )
2001                             h->stat.frame.i_mb_partition[h->mb.i_sub_partition[i]] ++;
2002                 if( h->param.i_frame_reference > 1 )
2003                     for( int i_list = 0; i_list <= (h->sh.i_type == SLICE_TYPE_B); i_list++ )
2004                         for( int i = 0; i < 4; i++ )
2005                         {
2006                             int i_ref = h->mb.cache.ref[i_list][ x264_scan8[4*i] ];
2007                             if( i_ref >= 0 )
2008                                 h->stat.frame.i_mb_count_ref[i_list][i_ref] ++;
2009                         }
2010             }
2011         }
2012
2013         if( h->param.i_log_level >= X264_LOG_INFO )
2014         {
2015             if( h->mb.i_cbp_luma | h->mb.i_cbp_chroma )
2016             {
2017                 int cbpsum = (h->mb.i_cbp_luma&1) + ((h->mb.i_cbp_luma>>1)&1)
2018                            + ((h->mb.i_cbp_luma>>2)&1) + (h->mb.i_cbp_luma>>3);
2019                 h->stat.frame.i_mb_cbp[!b_intra + 0] += cbpsum;
2020                 h->stat.frame.i_mb_cbp[!b_intra + 2] += !!h->mb.i_cbp_chroma;
2021                 h->stat.frame.i_mb_cbp[!b_intra + 4] += h->mb.i_cbp_chroma >> 1;
2022             }
2023             if( h->mb.i_cbp_luma && !b_intra )
2024             {
2025                 h->stat.frame.i_mb_count_8x8dct[0] ++;
2026                 h->stat.frame.i_mb_count_8x8dct[1] += h->mb.b_transform_8x8;
2027             }
2028             if( b_intra && h->mb.i_type != I_PCM )
2029             {
2030                 if( h->mb.i_type == I_16x16 )
2031                     h->stat.frame.i_mb_pred_mode[0][h->mb.i_intra16x16_pred_mode]++;
2032                 else if( h->mb.i_type == I_8x8 )
2033                     for( int i = 0; i < 16; i += 4 )
2034                         h->stat.frame.i_mb_pred_mode[1][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
2035                 else //if( h->mb.i_type == I_4x4 )
2036                     for( int i = 0; i < 16; i++ )
2037                         h->stat.frame.i_mb_pred_mode[2][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
2038                 h->stat.frame.i_mb_pred_mode[3][x264_mb_pred_mode8x8c_fix[h->mb.i_chroma_pred_mode]]++;
2039             }
2040         }
2041
2042         /* calculate deblock strength values (actual deblocking is done per-row along with hpel) */
2043         if( b_deblock )
2044         {
2045             int mvy_limit = 4 >> h->sh.b_mbaff;
2046             uint8_t (*bs)[4][4] = h->deblock_strength[h->mb.i_mb_y&1][h->mb.i_mb_x];
2047             x264_macroblock_cache_load_deblock( h );
2048             if( IS_INTRA( h->mb.type[h->mb.i_mb_xy] ) )
2049                 memset( bs, 3, 2*4*4*sizeof(uint8_t) );
2050             else
2051                 h->loopf.deblock_strength( h->mb.cache.non_zero_count, h->mb.cache.ref, h->mb.cache.mv,
2052                                            bs, mvy_limit, h->sh.i_type == SLICE_TYPE_B );
2053         }
2054
2055         x264_ratecontrol_mb( h, mb_size );
2056
2057         if( h->sh.b_mbaff )
2058         {
2059             i_mb_x += i_mb_y & 1;
2060             i_mb_y ^= i_mb_x < h->mb.i_mb_width;
2061         }
2062         else
2063             i_mb_x++;
2064         if( i_mb_x == h->mb.i_mb_width )
2065         {
2066             i_mb_y++;
2067             i_mb_x = 0;
2068         }
2069     }
2070     h->out.nal[h->out.i_nal].i_last_mb = h->sh.i_last_mb;
2071
2072     if( h->param.b_cabac )
2073     {
2074         x264_cabac_encode_flush( h, &h->cabac );
2075         h->out.bs.p = h->cabac.p;
2076     }
2077     else
2078     {
2079         if( i_skip > 0 )
2080             bs_write_ue( &h->out.bs, i_skip );  /* last skip run */
2081         /* rbsp_slice_trailing_bits */
2082         bs_rbsp_trailing( &h->out.bs );
2083         bs_flush( &h->out.bs );
2084     }
2085     if( x264_nal_end( h ) )
2086         return -1;
2087
2088     if( h->sh.i_last_mb == (h->i_threadslice_end * h->mb.i_mb_width - 1) )
2089     {
2090         h->stat.frame.i_misc_bits = bs_pos( &h->out.bs )
2091                                   + (h->out.i_nal*NALU_OVERHEAD * 8)
2092                                   - h->stat.frame.i_tex_bits
2093                                   - h->stat.frame.i_mv_bits;
2094         x264_fdec_filter_row( h, h->i_threadslice_end, 1 );
2095     }
2096
2097     return 0;
2098 }
2099
2100 static void x264_thread_sync_context( x264_t *dst, x264_t *src )
2101 {
2102     if( dst == src )
2103         return;
2104
2105     // reference counting
2106     for( x264_frame_t **f = src->frames.reference; *f; f++ )
2107         (*f)->i_reference_count++;
2108     for( x264_frame_t **f = dst->frames.reference; *f; f++ )
2109         x264_frame_push_unused( src, *f );
2110     src->fdec->i_reference_count++;
2111     x264_frame_push_unused( src, dst->fdec );
2112
2113     // copy everything except the per-thread pointers and the constants.
2114     memcpy( &dst->i_frame, &src->i_frame, offsetof(x264_t, mb.type) - offsetof(x264_t, i_frame) );
2115     dst->param = src->param;
2116     dst->stat = src->stat;
2117 }
2118
2119 static void x264_thread_sync_stat( x264_t *dst, x264_t *src )
2120 {
2121     if( dst == src )
2122         return;
2123     memcpy( &dst->stat.i_frame_count, &src->stat.i_frame_count, sizeof(dst->stat) - sizeof(dst->stat.frame) );
2124 }
2125
2126 static void *x264_slices_write( x264_t *h )
2127 {
2128     int i_slice_num = 0;
2129     int last_thread_mb = h->sh.i_last_mb;
2130
2131 #if HAVE_VISUALIZE
2132     if( h->param.b_visualize )
2133         if( x264_visualize_init( h ) )
2134             return (void *)-1;
2135 #endif
2136
2137     /* init stats */
2138     memset( &h->stat.frame, 0, sizeof(h->stat.frame) );
2139     h->mb.b_reencode_mb = 0;
2140     while( h->sh.i_first_mb <= last_thread_mb )
2141     {
2142         h->sh.i_last_mb = last_thread_mb;
2143         if( h->param.i_slice_max_mbs )
2144             h->sh.i_last_mb = h->sh.i_first_mb + h->param.i_slice_max_mbs - 1;
2145         else if( h->param.i_slice_count && !h->param.b_sliced_threads )
2146         {
2147             int height = h->mb.i_mb_height >> h->param.b_interlaced;
2148             int width = h->mb.i_mb_width << h->param.b_interlaced;
2149             i_slice_num++;
2150             h->sh.i_last_mb = (height * i_slice_num + h->param.i_slice_count/2) / h->param.i_slice_count * width - 1;
2151         }
2152         h->sh.i_last_mb = X264_MIN( h->sh.i_last_mb, last_thread_mb );
2153         if( x264_stack_align( x264_slice_write, h ) )
2154             return (void *)-1;
2155         h->sh.i_first_mb = h->sh.i_last_mb + 1;
2156     }
2157
2158 #if HAVE_VISUALIZE
2159     if( h->param.b_visualize )
2160     {
2161         x264_visualize_show( h );
2162         x264_visualize_close( h );
2163     }
2164 #endif
2165
2166     return (void *)0;
2167 }
2168
2169 static int x264_threaded_slices_write( x264_t *h )
2170 {
2171     /* set first/last mb and sync contexts */
2172     for( int i = 0; i < h->param.i_threads; i++ )
2173     {
2174         x264_t *t = h->thread[i];
2175         if( i )
2176         {
2177             t->param = h->param;
2178             memcpy( &t->i_frame, &h->i_frame, offsetof(x264_t, rc) - offsetof(x264_t, i_frame) );
2179         }
2180         int height = h->mb.i_mb_height >> h->param.b_interlaced;
2181         t->i_threadslice_start = ((height *  i    + h->param.i_slice_count/2) / h->param.i_threads) << h->param.b_interlaced;
2182         t->i_threadslice_end   = ((height * (i+1) + h->param.i_slice_count/2) / h->param.i_threads) << h->param.b_interlaced;
2183         t->sh.i_first_mb = t->i_threadslice_start * h->mb.i_mb_width;
2184         t->sh.i_last_mb  =   t->i_threadslice_end * h->mb.i_mb_width - 1;
2185     }
2186
2187     x264_stack_align( x264_analyse_weight_frame, h, h->mb.i_mb_height*16 + 16 );
2188
2189     x264_threads_distribute_ratecontrol( h );
2190
2191     /* dispatch */
2192     for( int i = 0; i < h->param.i_threads; i++ )
2193     {
2194         x264_threadpool_run( h->threadpool, (void*)x264_slices_write, h->thread[i] );
2195         h->thread[i]->b_thread_active = 1;
2196     }
2197     for( int i = 0; i < h->param.i_threads; i++ )
2198     {
2199         h->thread[i]->b_thread_active = 0;
2200         if( (intptr_t)x264_threadpool_wait( h->threadpool, h->thread[i] ) )
2201             return -1;
2202     }
2203
2204     /* Go back and fix up the hpel on the borders between slices. */
2205     for( int i = 1; i < h->param.i_threads; i++ )
2206     {
2207         x264_fdec_filter_row( h->thread[i], h->thread[i]->i_threadslice_start + 1, 0 );
2208         if( h->sh.b_mbaff )
2209             x264_fdec_filter_row( h->thread[i], h->thread[i]->i_threadslice_start + 2, 0 );
2210     }
2211
2212     x264_threads_merge_ratecontrol( h );
2213
2214     for( int i = 1; i < h->param.i_threads; i++ )
2215     {
2216         x264_t *t = h->thread[i];
2217         for( int j = 0; j < t->out.i_nal; j++ )
2218         {
2219             h->out.nal[h->out.i_nal] = t->out.nal[j];
2220             h->out.i_nal++;
2221             x264_nal_check_buffer( h );
2222         }
2223         /* All entries in stat.frame are ints except for ssd/ssim. */
2224         for( int j = 0; j < (offsetof(x264_t,stat.frame.i_ssd) - offsetof(x264_t,stat.frame.i_mv_bits)) / sizeof(int); j++ )
2225             ((int*)&h->stat.frame)[j] += ((int*)&t->stat.frame)[j];
2226         for( int j = 0; j < 3; j++ )
2227             h->stat.frame.i_ssd[j] += t->stat.frame.i_ssd[j];
2228         h->stat.frame.f_ssim += t->stat.frame.f_ssim;
2229     }
2230
2231     return 0;
2232 }
2233
2234 void x264_encoder_intra_refresh( x264_t *h )
2235 {
2236     h = h->thread[h->i_thread_phase];
2237     h->b_queued_intra_refresh = 1;
2238 }
2239
2240 int x264_encoder_invalidate_reference( x264_t *h, int64_t pts )
2241 {
2242     if( h->param.i_bframe )
2243     {
2244         x264_log( h, X264_LOG_ERROR, "x264_encoder_invalidate_reference is not supported with B-frames enabled\n" );
2245         return -1;
2246     }
2247     if( h->param.b_intra_refresh )
2248     {
2249         x264_log( h, X264_LOG_ERROR, "x264_encoder_invalidate_reference is not supported with intra refresh enabled\n" );
2250         return -1;
2251     }
2252     h = h->thread[h->i_thread_phase];
2253     if( pts >= h->i_last_idr_pts )
2254     {
2255         for( int i = 0; h->frames.reference[i]; i++ )
2256             if( pts <= h->frames.reference[i]->i_pts )
2257                 h->frames.reference[i]->b_corrupt = 1;
2258         if( pts <= h->fdec->i_pts )
2259             h->fdec->b_corrupt = 1;
2260     }
2261     return 0;
2262 }
2263
2264 /****************************************************************************
2265  * x264_encoder_encode:
2266  *  XXX: i_poc   : is the poc of the current given picture
2267  *       i_frame : is the number of the frame being coded
2268  *  ex:  type frame poc
2269  *       I      0   2*0
2270  *       P      1   2*3
2271  *       B      2   2*1
2272  *       B      3   2*2
2273  *       P      4   2*6
2274  *       B      5   2*4
2275  *       B      6   2*5
2276  ****************************************************************************/
2277 int     x264_encoder_encode( x264_t *h,
2278                              x264_nal_t **pp_nal, int *pi_nal,
2279                              x264_picture_t *pic_in,
2280                              x264_picture_t *pic_out )
2281 {
2282     x264_t *thread_current, *thread_prev, *thread_oldest;
2283     int i_nal_type, i_nal_ref_idc, i_global_qp;
2284     int overhead = NALU_OVERHEAD;
2285
2286     if( h->i_thread_frames > 1 )
2287     {
2288         thread_prev    = h->thread[ h->i_thread_phase ];
2289         h->i_thread_phase = (h->i_thread_phase + 1) % h->i_thread_frames;
2290         thread_current = h->thread[ h->i_thread_phase ];
2291         thread_oldest  = h->thread[ (h->i_thread_phase + 1) % h->i_thread_frames ];
2292         x264_thread_sync_context( thread_current, thread_prev );
2293         x264_thread_sync_ratecontrol( thread_current, thread_prev, thread_oldest );
2294         h = thread_current;
2295     }
2296     else
2297     {
2298         thread_current =
2299         thread_oldest  = h;
2300     }
2301 #if HAVE_MMX
2302     if( h->i_thread_frames == 1 && h->param.cpu&X264_CPU_SSE_MISALIGN )
2303         x264_cpu_mask_misalign_sse();
2304 #endif
2305
2306     // ok to call this before encoding any frames, since the initial values of fdec have b_kept_as_ref=0
2307     if( x264_reference_update( h ) )
2308         return -1;
2309     h->fdec->i_lines_completed = -1;
2310
2311     /* no data out */
2312     *pi_nal = 0;
2313     *pp_nal = NULL;
2314
2315     /* ------------------- Setup new frame from picture -------------------- */
2316     if( pic_in != NULL )
2317     {
2318         /* 1: Copy the picture to a frame and move it to a buffer */
2319         x264_frame_t *fenc = x264_frame_pop_unused( h, 0 );
2320         if( !fenc )
2321             return -1;
2322
2323         if( x264_frame_copy_picture( h, fenc, pic_in ) < 0 )
2324             return -1;
2325
2326         if( h->param.i_width != 16 * h->mb.i_mb_width ||
2327             h->param.i_height != 16 * h->mb.i_mb_height )
2328             x264_frame_expand_border_mod16( h, fenc );
2329
2330         fenc->i_frame = h->frames.i_input++;
2331
2332         if( h->frames.i_bframe_delay && fenc->i_frame == h->frames.i_bframe_delay )
2333             h->frames.i_bframe_delay_time = fenc->i_pts;
2334
2335         if( h->param.b_vfr_input && fenc->i_pts <= h->frames.i_largest_pts )
2336             x264_log( h, X264_LOG_WARNING, "non-strictly-monotonic PTS\n" );
2337
2338         h->frames.i_second_largest_pts = h->frames.i_largest_pts;
2339         h->frames.i_largest_pts = fenc->i_pts;
2340
2341         if( (fenc->i_pic_struct < PIC_STRUCT_AUTO) || (fenc->i_pic_struct > PIC_STRUCT_TRIPLE) )
2342             fenc->i_pic_struct = PIC_STRUCT_AUTO;
2343
2344         if( fenc->i_pic_struct == PIC_STRUCT_AUTO )
2345         {
2346             int b_interlaced = fenc->param ? fenc->param->b_interlaced : h->param.b_interlaced;
2347             if( b_interlaced )
2348             {
2349                 int b_tff = fenc->param ? fenc->param->b_tff : h->param.b_tff;
2350                 fenc->i_pic_struct = b_tff ? PIC_STRUCT_TOP_BOTTOM : PIC_STRUCT_BOTTOM_TOP;
2351             }
2352             else
2353                 fenc->i_pic_struct = PIC_STRUCT_PROGRESSIVE;
2354         }
2355
2356         if( h->param.rc.b_mb_tree && h->param.rc.b_stat_read )
2357         {
2358             if( x264_macroblock_tree_read( h, fenc, pic_in->prop.quant_offsets ) )
2359                 return -1;
2360         }
2361         else
2362             x264_stack_align( x264_adaptive_quant_frame, h, fenc, pic_in->prop.quant_offsets );
2363
2364         if( pic_in->prop.quant_offsets_free )
2365             pic_in->prop.quant_offsets_free( pic_in->prop.quant_offsets );
2366
2367         if( h->frames.b_have_lowres )
2368             x264_frame_init_lowres( h, fenc );
2369
2370         /* 2: Place the frame into the queue for its slice type decision */
2371         x264_lookahead_put_frame( h, fenc );
2372
2373         if( h->frames.i_input <= h->frames.i_delay + 1 - h->i_thread_frames )
2374         {
2375             /* Nothing yet to encode, waiting for filling of buffers */
2376             pic_out->i_type = X264_TYPE_AUTO;
2377             return 0;
2378         }
2379     }
2380     else
2381     {
2382         /* signal kills for lookahead thread */
2383         x264_pthread_mutex_lock( &h->lookahead->ifbuf.mutex );
2384         h->lookahead->b_exit_thread = 1;
2385         x264_pthread_cond_broadcast( &h->lookahead->ifbuf.cv_fill );
2386         x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
2387     }
2388
2389     h->i_frame++;
2390     /* 3: The picture is analyzed in the lookahead */
2391     if( !h->frames.current[0] )
2392         x264_lookahead_get_frames( h );
2393
2394     if( !h->frames.current[0] && x264_lookahead_is_empty( h ) )
2395         return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
2396
2397     /* ------------------- Get frame to be encoded ------------------------- */
2398     /* 4: get picture to encode */
2399     h->fenc = x264_frame_shift( h->frames.current );
2400     if( h->i_frame == h->i_thread_frames - 1 )
2401         h->i_reordered_pts_delay = h->fenc->i_reordered_pts;
2402     if( h->fenc->param )
2403     {
2404         x264_encoder_reconfig( h, h->fenc->param );
2405         if( h->fenc->param->param_free )
2406             h->fenc->param->param_free( h->fenc->param );
2407     }
2408
2409     if( !IS_X264_TYPE_I( h->fenc->i_type ) )
2410     {
2411         int valid_refs_left = 0;
2412         for( int i = 0; h->frames.reference[i]; i++ )
2413             if( !h->frames.reference[i]->b_corrupt )
2414                 valid_refs_left++;
2415         /* No valid reference frames left: force an IDR. */
2416         if( !valid_refs_left )
2417         {
2418             h->fenc->b_keyframe = 1;
2419             h->fenc->i_type = X264_TYPE_IDR;
2420         }
2421     }
2422
2423     if( h->fenc->b_keyframe )
2424     {
2425         h->frames.i_last_keyframe = h->fenc->i_frame;
2426         if( h->fenc->i_type == X264_TYPE_IDR )
2427         {
2428             h->i_frame_num = 0;
2429             h->frames.i_last_idr = h->fenc->i_frame;
2430         }
2431     }
2432     h->sh.i_mmco_command_count =
2433     h->sh.i_mmco_remove_from_end = 0;
2434     h->b_ref_reorder[0] =
2435     h->b_ref_reorder[1] = 0;
2436     h->fdec->i_poc =
2437     h->fenc->i_poc = 2 * ( h->fenc->i_frame - X264_MAX( h->frames.i_last_idr, 0 ) );
2438
2439     /* ------------------- Setup frame context ----------------------------- */
2440     /* 5: Init data dependent of frame type */
2441     if( h->fenc->i_type == X264_TYPE_IDR )
2442     {
2443         /* reset ref pictures */
2444         i_nal_type    = NAL_SLICE_IDR;
2445         i_nal_ref_idc = NAL_PRIORITY_HIGHEST;
2446         h->sh.i_type = SLICE_TYPE_I;
2447         x264_reference_reset( h );
2448         h->frames.i_poc_last_open_gop = -1;
2449     }
2450     else if( h->fenc->i_type == X264_TYPE_I )
2451     {
2452         i_nal_type    = NAL_SLICE;
2453         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
2454         h->sh.i_type = SLICE_TYPE_I;
2455         x264_reference_hierarchy_reset( h );
2456         if( h->param.i_open_gop )
2457             h->frames.i_poc_last_open_gop = h->fenc->b_keyframe ? h->fenc->i_poc : -1;
2458     }
2459     else if( h->fenc->i_type == X264_TYPE_P )
2460     {
2461         i_nal_type    = NAL_SLICE;
2462         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
2463         h->sh.i_type = SLICE_TYPE_P;
2464         x264_reference_hierarchy_reset( h );
2465         h->frames.i_poc_last_open_gop = -1;
2466     }
2467     else if( h->fenc->i_type == X264_TYPE_BREF )
2468     {
2469         i_nal_type    = NAL_SLICE;
2470         i_nal_ref_idc = h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT ? NAL_PRIORITY_LOW : NAL_PRIORITY_HIGH;
2471         h->sh.i_type = SLICE_TYPE_B;
2472         x264_reference_hierarchy_reset( h );
2473     }
2474     else    /* B frame */
2475     {
2476         i_nal_type    = NAL_SLICE;
2477         i_nal_ref_idc = NAL_PRIORITY_DISPOSABLE;
2478         h->sh.i_type = SLICE_TYPE_B;
2479     }
2480
2481     h->fdec->i_type = h->fenc->i_type;
2482     h->fdec->i_frame = h->fenc->i_frame;
2483     h->fenc->b_kept_as_ref =
2484     h->fdec->b_kept_as_ref = i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE && h->param.i_keyint_max > 1;
2485
2486     h->fdec->i_pts = h->fenc->i_pts *= h->i_dts_compress_multiplier;
2487     if( h->frames.i_bframe_delay )
2488     {
2489         int64_t *prev_reordered_pts = thread_current->frames.i_prev_reordered_pts;
2490         if( h->i_frame <= h->frames.i_bframe_delay )
2491         {
2492             if( h->i_dts_compress_multiplier == 1 )
2493                 h->fdec->i_dts = h->fenc->i_reordered_pts - h->frames.i_bframe_delay_time;
2494             else
2495             {
2496                 /* DTS compression */
2497                 if( h->i_frame == 1 )
2498                     thread_current->frames.i_init_delta = h->fenc->i_reordered_pts * h->i_dts_compress_multiplier;
2499                 h->fdec->i_dts = h->i_frame * thread_current->frames.i_init_delta / h->i_dts_compress_multiplier;
2500             }
2501         }
2502         else
2503             h->fdec->i_dts = prev_reordered_pts[ (h->i_frame - h->frames.i_bframe_delay) % h->frames.i_bframe_delay ];
2504         prev_reordered_pts[ h->i_frame % h->frames.i_bframe_delay ] = h->fenc->i_reordered_pts * h->i_dts_compress_multiplier;
2505     }
2506     else
2507         h->fdec->i_dts = h->fenc->i_reordered_pts;
2508     if( h->fenc->i_type == X264_TYPE_IDR )
2509         h->i_last_idr_pts = h->fdec->i_pts;
2510
2511     /* ------------------- Init                ----------------------------- */
2512     /* build ref list 0/1 */
2513     x264_reference_build_list( h, h->fdec->i_poc );
2514
2515     /* ---------------------- Write the bitstream -------------------------- */
2516     /* Init bitstream context */
2517     if( h->param.b_sliced_threads )
2518     {
2519         for( int i = 0; i < h->param.i_threads; i++ )
2520         {
2521             bs_init( &h->thread[i]->out.bs, h->thread[i]->out.p_bitstream, h->thread[i]->out.i_bitstream );
2522             h->thread[i]->out.i_nal = 0;
2523         }
2524     }
2525     else
2526     {
2527         bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
2528         h->out.i_nal = 0;
2529     }
2530
2531     if( h->param.b_aud )
2532     {
2533         int pic_type;
2534
2535         if( h->sh.i_type == SLICE_TYPE_I )
2536             pic_type = 0;
2537         else if( h->sh.i_type == SLICE_TYPE_P )
2538             pic_type = 1;
2539         else if( h->sh.i_type == SLICE_TYPE_B )
2540             pic_type = 2;
2541         else
2542             pic_type = 7;
2543
2544         x264_nal_start( h, NAL_AUD, NAL_PRIORITY_DISPOSABLE );
2545         bs_write( &h->out.bs, 3, pic_type );
2546         bs_rbsp_trailing( &h->out.bs );
2547         if( x264_nal_end( h ) )
2548             return -1;
2549         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2550     }
2551
2552     h->i_nal_type = i_nal_type;
2553     h->i_nal_ref_idc = i_nal_ref_idc;
2554
2555     if( h->param.b_intra_refresh )
2556     {
2557         if( IS_X264_TYPE_I( h->fenc->i_type ) )
2558         {
2559             h->fdec->i_frames_since_pir = 0;
2560             h->b_queued_intra_refresh = 0;
2561             /* PIR is currently only supported with ref == 1, so any intra frame effectively refreshes
2562              * the whole frame and counts as an intra refresh. */
2563             h->fdec->f_pir_position = h->mb.i_mb_width;
2564         }
2565         else if( h->fenc->i_type == X264_TYPE_P )
2566         {
2567             int pocdiff = (h->fdec->i_poc - h->fref0[0]->i_poc)/2;
2568             float increment = X264_MAX( ((float)h->mb.i_mb_width-1) / h->param.i_keyint_max, 1 );
2569             h->fdec->f_pir_position = h->fref0[0]->f_pir_position;
2570             h->fdec->i_frames_since_pir = h->fref0[0]->i_frames_since_pir + pocdiff;
2571             if( h->fdec->i_frames_since_pir >= h->param.i_keyint_max ||
2572                 (h->b_queued_intra_refresh && h->fdec->f_pir_position + 0.5 >= h->mb.i_mb_width) )
2573             {
2574                 h->fdec->f_pir_position = 0;
2575                 h->fdec->i_frames_since_pir = 0;
2576                 h->b_queued_intra_refresh = 0;
2577                 h->fenc->b_keyframe = 1;
2578             }
2579             h->fdec->i_pir_start_col = h->fdec->f_pir_position+0.5;
2580             h->fdec->f_pir_position += increment * pocdiff;
2581             h->fdec->i_pir_end_col = h->fdec->f_pir_position+0.5;
2582             /* If our intra refresh has reached the right side of the frame, we're done. */
2583             if( h->fdec->i_pir_end_col >= h->mb.i_mb_width - 1 )
2584                 h->fdec->f_pir_position = h->mb.i_mb_width;
2585         }
2586     }
2587
2588     if( h->fenc->b_keyframe )
2589     {
2590         /* Write SPS and PPS */
2591         if( h->param.b_repeat_headers )
2592         {
2593             /* generate sequence parameters */
2594             x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
2595             x264_sps_write( &h->out.bs, h->sps );
2596             if( x264_nal_end( h ) )
2597                 return -1;
2598             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
2599
2600             /* generate picture parameters */
2601             x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
2602             x264_pps_write( &h->out.bs, h->pps );
2603             if( x264_nal_end( h ) )
2604                 return -1;
2605             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
2606         }
2607
2608         /* buffering period sei is written in x264_encoder_frame_end */
2609     }
2610
2611     /* write extra sei */
2612     for( int i = 0; i < h->fenc->extra_sei.num_payloads; i++ )
2613     {
2614         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2615         x264_sei_write( &h->out.bs, h->fenc->extra_sei.payloads[i].payload, h->fenc->extra_sei.payloads[i].payload_size,
2616                         h->fenc->extra_sei.payloads[i].payload_type );
2617         if( x264_nal_end( h ) )
2618             return -1;
2619         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2620         if( h->fenc->extra_sei.sei_free && h->fenc->extra_sei.payloads[i].payload )
2621             h->fenc->extra_sei.sei_free( h->fenc->extra_sei.payloads[i].payload );
2622     }
2623
2624     if( h->fenc->extra_sei.sei_free && h->fenc->extra_sei.payloads )
2625         h->fenc->extra_sei.sei_free( h->fenc->extra_sei.payloads );
2626
2627     if( h->fenc->b_keyframe )
2628     {
2629         if( h->param.b_repeat_headers && h->fenc->i_frame == 0 )
2630         {
2631             /* identify ourself */
2632             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2633             if( x264_sei_version_write( h, &h->out.bs ) )
2634                 return -1;
2635             if( x264_nal_end( h ) )
2636                 return -1;
2637             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2638         }
2639
2640         if( h->fenc->i_type != X264_TYPE_IDR )
2641         {
2642             int time_to_recovery = h->param.i_open_gop ? 0 : X264_MIN( h->mb.i_mb_width - 1, h->param.i_keyint_max ) + h->param.i_bframe - 1;
2643             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2644             x264_sei_recovery_point_write( h, &h->out.bs, time_to_recovery );
2645             x264_nal_end( h );
2646             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2647         }
2648     }
2649
2650     /* generate sei pic timing */
2651     if( h->sps->vui.b_pic_struct_present || h->sps->vui.b_nal_hrd_parameters_present )
2652     {
2653         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2654         x264_sei_pic_timing_write( h, &h->out.bs );
2655         if( x264_nal_end( h ) )
2656             return -1;
2657         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2658     }
2659
2660     /* Init the rate control */
2661     /* FIXME: Include slice header bit cost. */
2662     x264_ratecontrol_start( h, h->fenc->i_qpplus1, overhead*8 );
2663     i_global_qp = x264_ratecontrol_qp( h );
2664
2665     pic_out->i_qpplus1 =
2666     h->fdec->i_qpplus1 = i_global_qp + 1;
2667
2668     if( h->param.rc.b_stat_read && h->sh.i_type != SLICE_TYPE_I )
2669     {
2670         x264_reference_build_list_optimal( h );
2671         x264_reference_check_reorder( h );
2672     }
2673
2674     if( h->i_ref0 )
2675         h->fdec->i_poc_l0ref0 = h->fref0[0]->i_poc;
2676
2677     if( h->sh.i_type == SLICE_TYPE_B )
2678         x264_macroblock_bipred_init( h );
2679
2680     /*------------------------- Weights -------------------------------------*/
2681     x264_weighted_pred_init( h );
2682
2683     /* ------------------------ Create slice header  ----------------------- */
2684     x264_slice_init( h, i_nal_type, i_global_qp );
2685
2686     if( i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE )
2687         h->i_frame_num++;
2688
2689     /* Write frame */
2690     h->i_threadslice_start = 0;
2691     h->i_threadslice_end = h->mb.i_mb_height;
2692     if( h->i_thread_frames > 1 )
2693     {
2694         x264_threadpool_run( h->threadpool, (void*)x264_slices_write, h );
2695         h->b_thread_active = 1;
2696     }
2697     else if( h->param.b_sliced_threads )
2698     {
2699         if( x264_threaded_slices_write( h ) )
2700             return -1;
2701     }
2702     else
2703         if( (intptr_t)x264_slices_write( h ) )
2704             return -1;
2705
2706     return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
2707 }
2708
2709 static int x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
2710                                    x264_nal_t **pp_nal, int *pi_nal,
2711                                    x264_picture_t *pic_out )
2712 {
2713     char psz_message[80];
2714
2715     if( h->b_thread_active )
2716     {
2717         h->b_thread_active = 0;
2718         if( (intptr_t)x264_threadpool_wait( h->threadpool, h ) )
2719             return -1;
2720     }
2721     if( !h->out.i_nal )
2722     {
2723         pic_out->i_type = X264_TYPE_AUTO;
2724         return 0;
2725     }
2726
2727     x264_emms();
2728     /* generate sei buffering period and insert it into place */
2729     if( h->fenc->b_keyframe && h->sps->vui.b_nal_hrd_parameters_present )
2730     {
2731         x264_hrd_fullness( h );
2732         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2733         x264_sei_buffering_period_write( h, &h->out.bs );
2734         if( x264_nal_end( h ) )
2735            return -1;
2736         /* buffering period sei must follow AUD, SPS and PPS and precede all other SEIs */
2737         int idx = 0;
2738         while( h->out.nal[idx].i_type == NAL_AUD ||
2739                h->out.nal[idx].i_type == NAL_SPS ||
2740                h->out.nal[idx].i_type == NAL_PPS )
2741             idx++;
2742         x264_nal_t nal_tmp = h->out.nal[h->out.i_nal-1];
2743         memmove( &h->out.nal[idx+1], &h->out.nal[idx], (h->out.i_nal-idx-1)*sizeof(x264_nal_t) );
2744         h->out.nal[idx] = nal_tmp;
2745     }
2746
2747     int frame_size = x264_encoder_encapsulate_nals( h, 0 );
2748
2749     /* Set output picture properties */
2750     pic_out->i_type = h->fenc->i_type;
2751
2752     pic_out->b_keyframe = h->fenc->b_keyframe;
2753
2754     pic_out->i_pts = h->fdec->i_pts;
2755     pic_out->i_dts = h->fdec->i_dts;
2756
2757     if( pic_out->i_pts < pic_out->i_dts )
2758         x264_log( h, X264_LOG_WARNING, "invalid DTS: PTS is less than DTS\n" );
2759
2760     pic_out->img.i_csp = X264_CSP_NV12;
2761     pic_out->img.i_plane = h->fdec->i_plane;
2762     for( int i = 0; i < 2; i++ )
2763     {
2764         pic_out->img.i_stride[i] = h->fdec->i_stride[i];
2765         // FIXME This breaks the API when pixel != uint8_t.
2766         pic_out->img.plane[i] = h->fdec->plane[i];
2767     }
2768
2769     x264_frame_push_unused( thread_current, h->fenc );
2770
2771     /* ---------------------- Update encoder state ------------------------- */
2772
2773     /* update rc */
2774     int filler = 0;
2775     if( x264_ratecontrol_end( h, frame_size * 8, &filler ) < 0 )
2776         return -1;
2777
2778     pic_out->hrd_timing = h->fenc->hrd_timing;
2779
2780     while( filler > 0 )
2781     {
2782         int f, overhead;
2783         overhead = (FILLER_OVERHEAD - h->param.b_annexb);
2784         if( h->param.i_slice_max_size && filler > h->param.i_slice_max_size )
2785         {
2786             int next_size = filler - h->param.i_slice_max_size;
2787             int overflow = X264_MAX( overhead - next_size, 0 );
2788             f = h->param.i_slice_max_size - overhead - overflow;
2789         }
2790         else
2791             f = X264_MAX( 0, filler - overhead );
2792
2793         x264_nal_start( h, NAL_FILLER, NAL_PRIORITY_DISPOSABLE );
2794         x264_filler_write( h, &h->out.bs, f );
2795         if( x264_nal_end( h ) )
2796             return -1;
2797         int total_size = x264_encoder_encapsulate_nals( h, h->out.i_nal-1 );
2798         frame_size += total_size;
2799         filler -= total_size;
2800     }
2801
2802     /* End bitstream, set output  */
2803     *pi_nal = h->out.i_nal;
2804     *pp_nal = h->out.nal;
2805
2806     h->out.i_nal = 0;
2807
2808     x264_noise_reduction_update( h );
2809
2810     /* ---------------------- Compute/Print statistics --------------------- */
2811     x264_thread_sync_stat( h, h->thread[0] );
2812
2813     /* Slice stat */
2814     h->stat.i_frame_count[h->sh.i_type]++;
2815     h->stat.i_frame_size[h->sh.i_type] += frame_size;
2816     h->stat.f_frame_qp[h->sh.i_type] += h->fdec->f_qp_avg_aq;
2817
2818     for( int i = 0; i < X264_MBTYPE_MAX; i++ )
2819         h->stat.i_mb_count[h->sh.i_type][i] += h->stat.frame.i_mb_count[i];
2820     for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
2821         h->stat.i_mb_partition[h->sh.i_type][i] += h->stat.frame.i_mb_partition[i];
2822     for( int i = 0; i < 2; i++ )
2823         h->stat.i_mb_count_8x8dct[i] += h->stat.frame.i_mb_count_8x8dct[i];
2824     for( int i = 0; i < 6; i++ )
2825         h->stat.i_mb_cbp[i] += h->stat.frame.i_mb_cbp[i];
2826     for( int i = 0; i < 4; i++ )
2827         for( int j = 0; j < 13; j++ )
2828             h->stat.i_mb_pred_mode[i][j] += h->stat.frame.i_mb_pred_mode[i][j];
2829     if( h->sh.i_type != SLICE_TYPE_I )
2830         for( int i_list = 0; i_list < 2; i_list++ )
2831             for( int i = 0; i < X264_REF_MAX*2; i++ )
2832                 h->stat.i_mb_count_ref[h->sh.i_type][i_list][i] += h->stat.frame.i_mb_count_ref[i_list][i];
2833     if( h->sh.i_type == SLICE_TYPE_P )
2834     {
2835         h->stat.i_consecutive_bframes[h->fdec->i_frame - h->fref0[0]->i_frame - 1]++;
2836         if( h->param.analyse.i_weighted_pred == X264_WEIGHTP_SMART )
2837             for( int i = 0; i < 3; i++ )
2838                 for( int j = 0; j < h->i_ref0; j++ )
2839                     if( h->sh.weight[0][i].i_denom != 0 )
2840                     {
2841                         h->stat.i_wpred[i]++;
2842                         break;
2843                     }
2844     }
2845     if( h->sh.i_type == SLICE_TYPE_B )
2846     {
2847         h->stat.i_direct_frames[ h->sh.b_direct_spatial_mv_pred ] ++;
2848         if( h->mb.b_direct_auto_write )
2849         {
2850             //FIXME somewhat arbitrary time constants
2851             if( h->stat.i_direct_score[0] + h->stat.i_direct_score[1] > h->mb.i_mb_count )
2852                 for( int i = 0; i < 2; i++ )
2853                     h->stat.i_direct_score[i] = h->stat.i_direct_score[i] * 9/10;
2854             for( int i = 0; i < 2; i++ )
2855                 h->stat.i_direct_score[i] += h->stat.frame.i_direct_score[i];
2856         }
2857     }
2858
2859     psz_message[0] = '\0';
2860     if( h->param.analyse.b_psnr )
2861     {
2862         int64_t ssd[3] = {
2863             h->stat.frame.i_ssd[0],
2864             h->stat.frame.i_ssd[1],
2865             h->stat.frame.i_ssd[2],
2866         };
2867
2868         h->stat.i_ssd_global[h->sh.i_type] += ssd[0] + ssd[1] + ssd[2];
2869         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 );
2870         h->stat.f_psnr_mean_y[h->sh.i_type] += x264_psnr( ssd[0], h->param.i_width * h->param.i_height );
2871         h->stat.f_psnr_mean_u[h->sh.i_type] += x264_psnr( ssd[1], h->param.i_width * h->param.i_height / 4 );
2872         h->stat.f_psnr_mean_v[h->sh.i_type] += x264_psnr( ssd[2], h->param.i_width * h->param.i_height / 4 );
2873
2874         snprintf( psz_message, 80, " PSNR Y:%5.2f U:%5.2f V:%5.2f",
2875                   x264_psnr( ssd[0], h->param.i_width * h->param.i_height ),
2876                   x264_psnr( ssd[1], h->param.i_width * h->param.i_height / 4),
2877                   x264_psnr( ssd[2], h->param.i_width * h->param.i_height / 4) );
2878     }
2879
2880     if( h->param.analyse.b_ssim )
2881     {
2882         double ssim_y = h->stat.frame.f_ssim
2883                       / (((h->param.i_width-6)>>2) * ((h->param.i_height-6)>>2));
2884         h->stat.f_ssim_mean_y[h->sh.i_type] += ssim_y;
2885         snprintf( psz_message + strlen(psz_message), 80 - strlen(psz_message),
2886                   " SSIM Y:%.5f", ssim_y );
2887     }
2888     psz_message[79] = '\0';
2889
2890     x264_log( h, X264_LOG_DEBUG,
2891                   "frame=%4d QP=%.2f NAL=%d Slice:%c Poc:%-3d I:%-4d P:%-4d SKIP:%-4d size=%d bytes%s\n",
2892               h->i_frame,
2893               h->fdec->f_qp_avg_aq,
2894               h->i_nal_ref_idc,
2895               h->sh.i_type == SLICE_TYPE_I ? 'I' : (h->sh.i_type == SLICE_TYPE_P ? 'P' : 'B' ),
2896               h->fdec->i_poc,
2897               h->stat.frame.i_mb_count_i,
2898               h->stat.frame.i_mb_count_p,
2899               h->stat.frame.i_mb_count_skip,
2900               frame_size,
2901               psz_message );
2902
2903     // keep stats all in one place
2904     x264_thread_sync_stat( h->thread[0], h );
2905     // for the use of the next frame
2906     x264_thread_sync_stat( thread_current, h );
2907
2908 #ifdef DEBUG_MB_TYPE
2909 {
2910     static const char mb_chars[] = { 'i', 'i', 'I', 'C', 'P', '8', 'S',
2911         'D', '<', 'X', 'B', 'X', '>', 'B', 'B', 'B', 'B', '8', 'S' };
2912     for( int mb_xy = 0; mb_xy < h->mb.i_mb_width * h->mb.i_mb_height; mb_xy++ )
2913     {
2914         if( h->mb.type[mb_xy] < X264_MBTYPE_MAX && h->mb.type[mb_xy] >= 0 )
2915             fprintf( stderr, "%c ", mb_chars[ h->mb.type[mb_xy] ] );
2916         else
2917             fprintf( stderr, "? " );
2918
2919         if( (mb_xy+1) % h->mb.i_mb_width == 0 )
2920             fprintf( stderr, "\n" );
2921     }
2922 }
2923 #endif
2924
2925     /* Remove duplicates, must be done near the end as breaks h->fref0 array
2926      * by freeing some of its pointers. */
2927      for( int i = 0; i < h->i_ref0; i++ )
2928          if( h->fref0[i] && h->fref0[i]->b_duplicate )
2929          {
2930              x264_frame_push_blank_unused( h, h->fref0[i] );
2931              h->fref0[i] = 0;
2932          }
2933
2934     if( h->param.psz_dump_yuv )
2935         x264_frame_dump( h );
2936
2937     return frame_size;
2938 }
2939
2940 static void x264_print_intra( int64_t *i_mb_count, double i_count, int b_print_pcm, char *intra )
2941 {
2942     intra += sprintf( intra, "I16..4%s: %4.1f%% %4.1f%% %4.1f%%",
2943         b_print_pcm ? "..PCM" : "",
2944         i_mb_count[I_16x16]/ i_count,
2945         i_mb_count[I_8x8]  / i_count,
2946         i_mb_count[I_4x4]  / i_count );
2947     if( b_print_pcm )
2948         sprintf( intra, " %4.1f%%", i_mb_count[I_PCM]  / i_count );
2949 }
2950
2951 /****************************************************************************
2952  * x264_encoder_close:
2953  ****************************************************************************/
2954 void    x264_encoder_close  ( x264_t *h )
2955 {
2956     int64_t i_yuv_size = 3 * h->param.i_width * h->param.i_height / 2;
2957     int64_t i_mb_count_size[2][7] = {{0}};
2958     char buf[200];
2959     int b_print_pcm = h->stat.i_mb_count[SLICE_TYPE_I][I_PCM]
2960                    || h->stat.i_mb_count[SLICE_TYPE_P][I_PCM]
2961                    || h->stat.i_mb_count[SLICE_TYPE_B][I_PCM];
2962
2963     x264_lookahead_delete( h );
2964
2965     if( h->param.i_threads > 1 )
2966         x264_threadpool_delete( h->threadpool );
2967     if( h->i_thread_frames > 1 )
2968     {
2969         for( int i = 0; i < h->i_thread_frames; i++ )
2970             if( h->thread[i]->b_thread_active )
2971             {
2972                 assert( h->thread[i]->fenc->i_reference_count == 1 );
2973                 x264_frame_delete( h->thread[i]->fenc );
2974             }
2975
2976         x264_t *thread_prev = h->thread[h->i_thread_phase];
2977         x264_thread_sync_ratecontrol( h, thread_prev, h );
2978         x264_thread_sync_ratecontrol( thread_prev, thread_prev, h );
2979         h->i_frame = thread_prev->i_frame + 1 - h->i_thread_frames;
2980     }
2981     h->i_frame++;
2982
2983     /* Slices used and PSNR */
2984     for( int i = 0; i < 3; i++ )
2985     {
2986         static const uint8_t slice_order[] = { SLICE_TYPE_I, SLICE_TYPE_P, SLICE_TYPE_B };
2987         int i_slice = slice_order[i];
2988
2989         if( h->stat.i_frame_count[i_slice] > 0 )
2990         {
2991             const int i_count = h->stat.i_frame_count[i_slice];
2992             if( h->param.analyse.b_psnr )
2993             {
2994                 x264_log( h, X264_LOG_INFO,
2995                           "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",
2996                           slice_type_to_char[i_slice],
2997                           i_count,
2998                           h->stat.f_frame_qp[i_slice] / i_count,
2999                           (double)h->stat.i_frame_size[i_slice] / i_count,
3000                           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,
3001                           h->stat.f_psnr_average[i_slice] / i_count,
3002                           x264_psnr( h->stat.i_ssd_global[i_slice], i_count * i_yuv_size ) );
3003             }
3004             else
3005             {
3006                 x264_log( h, X264_LOG_INFO,
3007                           "frame %c:%-5d Avg QP:%5.2f  size:%6.0f\n",
3008                           slice_type_to_char[i_slice],
3009                           i_count,
3010                           h->stat.f_frame_qp[i_slice] / i_count,
3011                           (double)h->stat.i_frame_size[i_slice] / i_count );
3012             }
3013         }
3014     }
3015     if( h->param.i_bframe && h->stat.i_frame_count[SLICE_TYPE_P] )
3016     {
3017         char *p = buf;
3018         int den = 0;
3019         // weight by number of frames (including the P-frame) that are in a sequence of N B-frames
3020         for( int i = 0; i <= h->param.i_bframe; i++ )
3021             den += (i+1) * h->stat.i_consecutive_bframes[i];
3022         for( int i = 0; i <= h->param.i_bframe; i++ )
3023             p += sprintf( p, " %4.1f%%", 100. * (i+1) * h->stat.i_consecutive_bframes[i] / den );
3024         x264_log( h, X264_LOG_INFO, "consecutive B-frames:%s\n", buf );
3025     }
3026
3027     for( int i_type = 0; i_type < 2; i_type++ )
3028         for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
3029         {
3030             if( i == D_DIRECT_8x8 ) continue; /* direct is counted as its own type */
3031             i_mb_count_size[i_type][x264_mb_partition_pixel_table[i]] += h->stat.i_mb_partition[i_type][i];
3032         }
3033
3034     /* MB types used */
3035     if( h->stat.i_frame_count[SLICE_TYPE_I] > 0 )
3036     {
3037         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_I];
3038         double i_count = h->stat.i_frame_count[SLICE_TYPE_I] * h->mb.i_mb_count / 100.0;
3039         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
3040         x264_log( h, X264_LOG_INFO, "mb I  %s\n", buf );
3041     }
3042     if( h->stat.i_frame_count[SLICE_TYPE_P] > 0 )
3043     {
3044         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_P];
3045         double i_count = h->stat.i_frame_count[SLICE_TYPE_P] * h->mb.i_mb_count / 100.0;
3046         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_P];
3047         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
3048         x264_log( h, X264_LOG_INFO,
3049                   "mb P  %s  P16..4: %4.1f%% %4.1f%% %4.1f%% %4.1f%% %4.1f%%    skip:%4.1f%%\n",
3050                   buf,
3051                   i_mb_size[PIXEL_16x16] / (i_count*4),
3052                   (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
3053                   i_mb_size[PIXEL_8x8] / (i_count*4),
3054                   (i_mb_size[PIXEL_8x4] + i_mb_size[PIXEL_4x8]) / (i_count*4),
3055                   i_mb_size[PIXEL_4x4] / (i_count*4),
3056                   i_mb_count[P_SKIP] / i_count );
3057     }
3058     if( h->stat.i_frame_count[SLICE_TYPE_B] > 0 )
3059     {
3060         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_B];
3061         double i_count = h->stat.i_frame_count[SLICE_TYPE_B] * h->mb.i_mb_count / 100.0;
3062         double i_mb_list_count;
3063         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_B];
3064         int64_t list_count[3] = {0}; /* 0 == L0, 1 == L1, 2 == BI */
3065         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
3066         for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
3067             for( int j = 0; j < 2; j++ )
3068             {
3069                 int l0 = x264_mb_type_list_table[i][0][j];
3070                 int l1 = x264_mb_type_list_table[i][1][j];
3071                 if( l0 || l1 )
3072                     list_count[l1+l0*l1] += h->stat.i_mb_count[SLICE_TYPE_B][i] * 2;
3073             }
3074         list_count[0] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L0_8x8];
3075         list_count[1] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L1_8x8];
3076         list_count[2] += h->stat.i_mb_partition[SLICE_TYPE_B][D_BI_8x8];
3077         i_mb_count[B_DIRECT] += (h->stat.i_mb_partition[SLICE_TYPE_B][D_DIRECT_8x8]+2)/4;
3078         i_mb_list_count = (list_count[0] + list_count[1] + list_count[2]) / 100.0;
3079         x264_log( h, X264_LOG_INFO,
3080                   "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",
3081                   buf,
3082                   i_mb_size[PIXEL_16x16] / (i_count*4),
3083                   (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
3084                   i_mb_size[PIXEL_8x8] / (i_count*4),
3085                   i_mb_count[B_DIRECT] / i_count,
3086                   i_mb_count[B_SKIP]   / i_count,
3087                   list_count[0] / i_mb_list_count,
3088                   list_count[1] / i_mb_list_count,
3089                   list_count[2] / i_mb_list_count );
3090     }
3091
3092     x264_ratecontrol_summary( h );
3093
3094     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 )
3095     {
3096 #define SUM3(p) (p[SLICE_TYPE_I] + p[SLICE_TYPE_P] + p[SLICE_TYPE_B])
3097 #define SUM3b(p,o) (p[SLICE_TYPE_I][o] + p[SLICE_TYPE_P][o] + p[SLICE_TYPE_B][o])
3098         int64_t i_i8x8 = SUM3b( h->stat.i_mb_count, I_8x8 );
3099         int64_t i_intra = i_i8x8 + SUM3b( h->stat.i_mb_count, I_4x4 )
3100                                  + SUM3b( h->stat.i_mb_count, I_16x16 );
3101         int64_t i_all_intra = i_intra + SUM3b( h->stat.i_mb_count, I_PCM);
3102         const int i_count = h->stat.i_frame_count[SLICE_TYPE_I] +
3103                             h->stat.i_frame_count[SLICE_TYPE_P] +
3104                             h->stat.i_frame_count[SLICE_TYPE_B];
3105         int64_t i_mb_count = (int64_t)i_count * h->mb.i_mb_count;
3106         float fps = (float) h->param.i_fps_num / h->param.i_fps_den;
3107         float f_bitrate;
3108         /* duration algorithm fails with one frame */
3109         if( !h->param.b_vfr_input || i_count == 1 )
3110             f_bitrate = fps * SUM3(h->stat.i_frame_size) / i_count / 125;
3111         else
3112         {
3113             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;
3114             f_bitrate = SUM3(h->stat.i_frame_size) / duration / 125;
3115         }
3116
3117         if( h->pps->b_transform_8x8_mode )
3118         {
3119             buf[0] = 0;
3120             if( h->stat.i_mb_count_8x8dct[0] )
3121                 sprintf( buf, " inter:%.1f%%", 100. * h->stat.i_mb_count_8x8dct[1] / h->stat.i_mb_count_8x8dct[0] );
3122             x264_log( h, X264_LOG_INFO, "8x8 transform intra:%.1f%%%s\n", 100. * i_i8x8 / i_intra, buf );
3123         }
3124
3125         if( (h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO ||
3126             (h->stat.i_direct_frames[0] && h->stat.i_direct_frames[1]))
3127             && h->stat.i_frame_count[SLICE_TYPE_B] )
3128         {
3129             x264_log( h, X264_LOG_INFO, "direct mvs  spatial:%.1f%% temporal:%.1f%%\n",
3130                       h->stat.i_direct_frames[1] * 100. / h->stat.i_frame_count[SLICE_TYPE_B],
3131                       h->stat.i_direct_frames[0] * 100. / h->stat.i_frame_count[SLICE_TYPE_B] );
3132         }
3133
3134         buf[0] = 0;
3135         if( i_mb_count != i_all_intra )
3136             sprintf( buf, " inter: %.1f%% %.1f%% %.1f%%",
3137                      h->stat.i_mb_cbp[1] * 100.0 / ((i_mb_count - i_all_intra)*4),
3138                      h->stat.i_mb_cbp[3] * 100.0 / ((i_mb_count - i_all_intra)  ),
3139                      h->stat.i_mb_cbp[5] * 100.0 / ((i_mb_count - i_all_intra)) );
3140         x264_log( h, X264_LOG_INFO, "coded y,uvDC,uvAC intra: %.1f%% %.1f%% %.1f%%%s\n",
3141                   h->stat.i_mb_cbp[0] * 100.0 / (i_all_intra*4),
3142                   h->stat.i_mb_cbp[2] * 100.0 / (i_all_intra  ),
3143                   h->stat.i_mb_cbp[4] * 100.0 / (i_all_intra  ), buf );
3144
3145         int64_t fixed_pred_modes[4][9] = {{0}};
3146         int64_t sum_pred_modes[4] = {0};
3147         for( int i = 0; i <= I_PRED_16x16_DC_128; i++ )
3148         {
3149             fixed_pred_modes[0][x264_mb_pred_mode16x16_fix[i]] += h->stat.i_mb_pred_mode[0][i];
3150             sum_pred_modes[0] += h->stat.i_mb_pred_mode[0][i];
3151         }
3152         if( sum_pred_modes[0] )
3153             x264_log( h, X264_LOG_INFO, "i16 v,h,dc,p: %2.0f%% %2.0f%% %2.0f%% %2.0f%%\n",
3154                       fixed_pred_modes[0][0] * 100.0 / sum_pred_modes[0],
3155                       fixed_pred_modes[0][1] * 100.0 / sum_pred_modes[0],
3156                       fixed_pred_modes[0][2] * 100.0 / sum_pred_modes[0],
3157                       fixed_pred_modes[0][3] * 100.0 / sum_pred_modes[0] );
3158         for( int i = 1; i <= 2; i++ )
3159         {
3160             for( int j = 0; j <= I_PRED_8x8_DC_128; j++ )
3161             {
3162                 fixed_pred_modes[i][x264_mb_pred_mode4x4_fix(j)] += h->stat.i_mb_pred_mode[i][j];
3163                 sum_pred_modes[i] += h->stat.i_mb_pred_mode[i][j];
3164             }
3165             if( sum_pred_modes[i] )
3166                 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,
3167                           fixed_pred_modes[i][0] * 100.0 / sum_pred_modes[i],
3168                           fixed_pred_modes[i][1] * 100.0 / sum_pred_modes[i],
3169                           fixed_pred_modes[i][2] * 100.0 / sum_pred_modes[i],
3170                           fixed_pred_modes[i][3] * 100.0 / sum_pred_modes[i],
3171                           fixed_pred_modes[i][4] * 100.0 / sum_pred_modes[i],
3172                           fixed_pred_modes[i][5] * 100.0 / sum_pred_modes[i],
3173                           fixed_pred_modes[i][6] * 100.0 / sum_pred_modes[i],
3174                           fixed_pred_modes[i][7] * 100.0 / sum_pred_modes[i],
3175                           fixed_pred_modes[i][8] * 100.0 / sum_pred_modes[i] );
3176         }
3177         for( int i = 0; i <= I_PRED_CHROMA_DC_128; i++ )
3178         {
3179             fixed_pred_modes[3][x264_mb_pred_mode8x8c_fix[i]] += h->stat.i_mb_pred_mode[3][i];
3180             sum_pred_modes[3] += h->stat.i_mb_pred_mode[3][i];
3181         }
3182         if( sum_pred_modes[3] )
3183             x264_log( h, X264_LOG_INFO, "i8c dc,h,v,p: %2.0f%% %2.0f%% %2.0f%% %2.0f%%\n",
3184                       fixed_pred_modes[3][0] * 100.0 / sum_pred_modes[3],
3185                       fixed_pred_modes[3][1] * 100.0 / sum_pred_modes[3],
3186                       fixed_pred_modes[3][2] * 100.0 / sum_pred_modes[3],
3187                       fixed_pred_modes[3][3] * 100.0 / sum_pred_modes[3] );
3188
3189         if( h->param.analyse.i_weighted_pred == X264_WEIGHTP_SMART && h->stat.i_frame_count[SLICE_TYPE_P] > 0 )
3190             x264_log( h, X264_LOG_INFO, "Weighted P-Frames: Y:%.1f%%\n",
3191                       h->stat.i_wpred[0] * 100.0 / h->stat.i_frame_count[SLICE_TYPE_P] );
3192
3193         for( int i_list = 0; i_list < 2; i_list++ )
3194             for( int i_slice = 0; i_slice < 2; i_slice++ )
3195             {
3196                 char *p = buf;
3197                 int64_t i_den = 0;
3198                 int i_max = 0;
3199                 for( int i = 0; i < X264_REF_MAX*2; i++ )
3200                     if( h->stat.i_mb_count_ref[i_slice][i_list][i] )
3201                     {
3202                         i_den += h->stat.i_mb_count_ref[i_slice][i_list][i];
3203                         i_max = i;
3204                     }
3205                 if( i_max == 0 )
3206                     continue;
3207                 for( int i = 0; i <= i_max; i++ )
3208                     p += sprintf( p, " %4.1f%%", 100. * h->stat.i_mb_count_ref[i_slice][i_list][i] / i_den );
3209                 x264_log( h, X264_LOG_INFO, "ref %c L%d:%s\n", "PB"[i_slice], i_list, buf );
3210             }
3211
3212         if( h->param.analyse.b_ssim )
3213         {
3214             float ssim = SUM3( h->stat.f_ssim_mean_y ) / i_count;
3215             x264_log( h, X264_LOG_INFO, "SSIM Mean Y:%.7f (%6.3fdb)\n", ssim, x264_ssim( ssim ) );
3216         }
3217         if( h->param.analyse.b_psnr )
3218         {
3219             x264_log( h, X264_LOG_INFO,
3220                       "PSNR Mean Y:%6.3f U:%6.3f V:%6.3f Avg:%6.3f Global:%6.3f kb/s:%.2f\n",
3221                       SUM3( h->stat.f_psnr_mean_y ) / i_count,
3222                       SUM3( h->stat.f_psnr_mean_u ) / i_count,
3223                       SUM3( h->stat.f_psnr_mean_v ) / i_count,
3224                       SUM3( h->stat.f_psnr_average ) / i_count,
3225                       x264_psnr( SUM3( h->stat.i_ssd_global ), i_count * i_yuv_size ),
3226                       f_bitrate );
3227         }
3228         else
3229             x264_log( h, X264_LOG_INFO, "kb/s:%.2f\n", f_bitrate );
3230     }
3231
3232     /* rc */
3233     x264_ratecontrol_delete( h );
3234
3235     /* param */
3236     if( h->param.rc.psz_stat_out )
3237         free( h->param.rc.psz_stat_out );
3238     if( h->param.rc.psz_stat_in )
3239         free( h->param.rc.psz_stat_in );
3240
3241     x264_cqm_delete( h );
3242     x264_free( h->nal_buffer );
3243     x264_analyse_free_costs( h );
3244
3245     if( h->i_thread_frames > 1)
3246         h = h->thread[h->i_thread_phase];
3247
3248     /* frames */
3249     x264_frame_delete_list( h->frames.unused[0] );
3250     x264_frame_delete_list( h->frames.unused[1] );
3251     x264_frame_delete_list( h->frames.current );
3252     x264_frame_delete_list( h->frames.blank_unused );
3253
3254     h = h->thread[0];
3255
3256     for( int i = h->param.i_threads - 1; i >= 0; i-- )
3257     {
3258         x264_frame_t **frame;
3259
3260         if( !h->param.b_sliced_threads || i == 0 )
3261         {
3262             for( frame = h->thread[i]->frames.reference; *frame; frame++ )
3263             {
3264                 assert( (*frame)->i_reference_count > 0 );
3265                 (*frame)->i_reference_count--;
3266                 if( (*frame)->i_reference_count == 0 )
3267                     x264_frame_delete( *frame );
3268             }
3269             frame = &h->thread[i]->fdec;
3270             assert( (*frame)->i_reference_count > 0 );
3271             (*frame)->i_reference_count--;
3272             if( (*frame)->i_reference_count == 0 )
3273                 x264_frame_delete( *frame );
3274             x264_macroblock_cache_free( h->thread[i] );
3275         }
3276         x264_macroblock_thread_free( h->thread[i], 0 );
3277         x264_free( h->thread[i]->out.p_bitstream );
3278         x264_free( h->thread[i]->out.nal);
3279         x264_free( h->thread[i] );
3280     }
3281 }
3282
3283 /****************************************************************************
3284  * x264_encoder_delayed_frames:
3285  ****************************************************************************/
3286 int x264_encoder_delayed_frames( x264_t *h )
3287 {
3288     int delayed_frames = 0;
3289     if( h->i_thread_frames > 1 )
3290     {
3291         for( int i = 0; i < h->i_thread_frames; i++ )
3292             delayed_frames += h->thread[i]->b_thread_active;
3293         h = h->thread[h->i_thread_phase];
3294     }
3295     for( int i = 0; h->frames.current[i]; i++ )
3296         delayed_frames++;
3297     x264_pthread_mutex_lock( &h->lookahead->ofbuf.mutex );
3298     x264_pthread_mutex_lock( &h->lookahead->ifbuf.mutex );
3299     x264_pthread_mutex_lock( &h->lookahead->next.mutex );
3300     delayed_frames += h->lookahead->ifbuf.i_size + h->lookahead->next.i_size + h->lookahead->ofbuf.i_size;
3301     x264_pthread_mutex_unlock( &h->lookahead->next.mutex );
3302     x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
3303     x264_pthread_mutex_unlock( &h->lookahead->ofbuf.mutex );
3304     return delayed_frames;
3305 }