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

   1 /*****************************************************************************
   2  * x264: h264 encoder
   3  *****************************************************************************
   4  * Copyright (C) 2003-2008 x264 project
   5  *
   6  * Authors: Laurent Aimar <fenrir@via.ecp.fr>
   7  *          Loren Merritt <lorenm@u.washington.edu>
   8  *          Fiona Glaser <fiona@x264.com>
   9  *
  10  * This program is free software; you can redistribute it and/or modify
  11  * it under the terms of the GNU General Public License as published by
  12  * the Free Software Foundation; either version 2 of the License, or
  13  * (at your option) any later version.
  14  *
  15  * This program is distributed in the hope that it will be useful,
  16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  18  * GNU General Public License for more details.
  19  *
  20  * You should have received a copy of the GNU General Public License
  21  * along with this program; if not, write to the Free Software
  22  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02111, USA.
  23  *****************************************************************************/
  24
  25 #include <math.h>
  26
  27 #include "common/common.h"
  28 #include "common/cpu.h"
  29
  30 #include "set.h"
  31 #include "analyse.h"
  32 #include "ratecontrol.h"
  33 #include "macroblock.h"
  34
  35 #if VISUALIZE
  36 #include "common/visualize.h"
  37 #endif
  38
  39 //#define DEBUG_MB_TYPE
  40
  41 #define NALU_OVERHEAD 5 // startcode + NAL type costs 5 bytes per frame
  42
  43 #define bs_write_ue bs_write_ue_big
  44
  45 static int x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
  46                                    x264_nal_t **pp_nal, int *pi_nal,
  47                                    x264_picture_t *pic_out );
  48
  49 /****************************************************************************
  50  *
  51  ******************************* x264 libs **********************************
  52  *
  53  ****************************************************************************/
  54 static float x264_psnr( int64_t i_sqe, int64_t i_size )
  55 {
  56     double f_mse = (double)i_sqe / ((double)65025.0 * (double)i_size);
  57     if( f_mse <= 0.0000000001 ) /* Max 100dB */
  58         return 100;
  59
  60     return (float)(-10.0 * log( f_mse ) / log( 10.0 ));
  61 }
  62
  63 static void x264_frame_dump( x264_t *h )
  64 {
  65     FILE *f = fopen( h->param.psz_dump_yuv, "r+b" );
  66     int i, y;
  67     if( !f )
  68         return;
  69     /* Write the frame in display order */
  70     fseek( f, (uint64_t)h->fdec->i_frame * h->param.i_height * h->param.i_width * 3/2, SEEK_SET );
  71     for( i = 0; i < h->fdec->i_plane; i++ )
  72         for( y = 0; y < h->param.i_height >> !!i; y++ )
  73             fwrite( &h->fdec->plane[i][y*h->fdec->i_stride[i]], 1, h->param.i_width >> !!i, f );
  74     fclose( f );
  75 }
  76
  77
  78 /* Fill "default" values */
  79 static void x264_slice_header_init( x264_t *h, x264_slice_header_t *sh,
  80                                     x264_sps_t *sps, x264_pps_t *pps,
  81                                     int i_idr_pic_id, int i_frame, int i_qp )
  82 {
  83     x264_param_t *param = &h->param;
  84     int i;
  85
  86     /* First we fill all field */
  87     sh->sps = sps;
  88     sh->pps = pps;
  89
  90     sh->i_first_mb  = 0;
  91     sh->i_last_mb   = h->mb.i_mb_count - 1;
  92     sh->i_pps_id    = pps->i_id;
  93
  94     sh->i_frame_num = i_frame;
  95
  96     sh->b_mbaff = h->param.b_interlaced;
  97     sh->b_field_pic = 0;    /* no field support for now */
  98     sh->b_bottom_field = 0; /* not yet used */
  99
 100     sh->i_idr_pic_id = i_idr_pic_id;
 101
 102     /* poc stuff, fixed later */
 103     sh->i_poc_lsb = 0;
 104     sh->i_delta_poc_bottom = 0;
 105     sh->i_delta_poc[0] = 0;
 106     sh->i_delta_poc[1] = 0;
 107
 108     sh->i_redundant_pic_cnt = 0;
 109
 110     if( !h->mb.b_direct_auto_read )
 111     {
 112         if( h->mb.b_direct_auto_write )
 113             sh->b_direct_spatial_mv_pred = ( h->stat.i_direct_score[1] > h->stat.i_direct_score[0] );
 114         else
 115             sh->b_direct_spatial_mv_pred = ( param->analyse.i_direct_mv_pred == X264_DIRECT_PRED_SPATIAL );
 116     }
 117     /* else b_direct_spatial_mv_pred was read from the 2pass statsfile */
 118
 119     sh->b_num_ref_idx_override = 0;
 120     sh->i_num_ref_idx_l0_active = 1;
 121     sh->i_num_ref_idx_l1_active = 1;
 122
 123     sh->b_ref_pic_list_reordering_l0 = h->b_ref_reorder[0];
 124     sh->b_ref_pic_list_reordering_l1 = h->b_ref_reorder[1];
 125
 126     /* If the ref list isn't in the default order, construct reordering header */
 127     /* List1 reordering isn't needed yet */
 128     if( sh->b_ref_pic_list_reordering_l0 )
 129     {
 130         int pred_frame_num = i_frame;
 131         for( i = 0; i < h->i_ref0; i++ )
 132         {
 133             int diff = h->fref0[i]->i_frame_num - pred_frame_num;
 134             if( diff == 0 )
 135                 x264_log( h, X264_LOG_ERROR, "diff frame num == 0\n" );
 136             sh->ref_pic_list_order[0][i].idc = ( diff > 0 );
 137             sh->ref_pic_list_order[0][i].arg = abs( diff ) - 1;
 138             pred_frame_num = h->fref0[i]->i_frame_num;
 139         }
 140     }
 141
 142     sh->i_cabac_init_idc = param->i_cabac_init_idc;
 143
 144     sh->i_qp = i_qp;
 145     sh->i_qp_delta = i_qp - pps->i_pic_init_qp;
 146     sh->b_sp_for_swidth = 0;
 147     sh->i_qs_delta = 0;
 148
 149     /* If effective qp <= 15, deblocking would have no effect anyway */
 150     if( param->b_deblocking_filter
 151         && ( h->mb.b_variable_qp
 152         || 15 < i_qp + 2 * X264_MIN(param->i_deblocking_filter_alphac0, param->i_deblocking_filter_beta) ) )
 153     {
 154         sh->i_disable_deblocking_filter_idc = 0;
 155     }
 156     else
 157     {
 158         sh->i_disable_deblocking_filter_idc = 1;
 159     }
 160     sh->i_alpha_c0_offset = param->i_deblocking_filter_alphac0 << 1;
 161     sh->i_beta_offset = param->i_deblocking_filter_beta << 1;
 162 }
 163
 164 static void x264_slice_header_write( bs_t *s, x264_slice_header_t *sh, int i_nal_ref_idc )
 165 {
 166     int i;
 167
 168     if( sh->b_mbaff )
 169     {
 170         assert( sh->i_first_mb % (2*sh->sps->i_mb_width) == 0 );
 171         bs_write_ue( s, sh->i_first_mb >> 1 );
 172     }
 173     else
 174         bs_write_ue( s, sh->i_first_mb );
 175
 176     bs_write_ue( s, sh->i_type + 5 );   /* same type things */
 177     bs_write_ue( s, sh->i_pps_id );
 178     bs_write( s, sh->sps->i_log2_max_frame_num, sh->i_frame_num & ((1<<sh->sps->i_log2_max_frame_num)-1) );
 179
 180     if( !sh->sps->b_frame_mbs_only )
 181     {
 182         bs_write1( s, sh->b_field_pic );
 183         if( sh->b_field_pic )
 184             bs_write1( s, sh->b_bottom_field );
 185     }
 186
 187     if( sh->i_idr_pic_id >= 0 ) /* NAL IDR */
 188     {
 189         bs_write_ue( s, sh->i_idr_pic_id );
 190     }
 191
 192     if( sh->sps->i_poc_type == 0 )
 193     {
 194         bs_write( s, sh->sps->i_log2_max_poc_lsb, sh->i_poc_lsb & ((1<<sh->sps->i_log2_max_poc_lsb)-1) );
 195         if( sh->pps->b_pic_order && !sh->b_field_pic )
 196         {
 197             bs_write_se( s, sh->i_delta_poc_bottom );
 198         }
 199     }
 200     else if( sh->sps->i_poc_type == 1 && !sh->sps->b_delta_pic_order_always_zero )
 201     {
 202         bs_write_se( s, sh->i_delta_poc[0] );
 203         if( sh->pps->b_pic_order && !sh->b_field_pic )
 204         {
 205             bs_write_se( s, sh->i_delta_poc[1] );
 206         }
 207     }
 208
 209     if( sh->pps->b_redundant_pic_cnt )
 210     {
 211         bs_write_ue( s, sh->i_redundant_pic_cnt );
 212     }
 213
 214     if( sh->i_type == SLICE_TYPE_B )
 215     {
 216         bs_write1( s, sh->b_direct_spatial_mv_pred );
 217     }
 218     if( sh->i_type == SLICE_TYPE_P || sh->i_type == SLICE_TYPE_SP || sh->i_type == SLICE_TYPE_B )
 219     {
 220         bs_write1( s, sh->b_num_ref_idx_override );
 221         if( sh->b_num_ref_idx_override )
 222         {
 223             bs_write_ue( s, sh->i_num_ref_idx_l0_active - 1 );
 224             if( sh->i_type == SLICE_TYPE_B )
 225             {
 226                 bs_write_ue( s, sh->i_num_ref_idx_l1_active - 1 );
 227             }
 228         }
 229     }
 230
 231     /* ref pic list reordering */
 232     if( sh->i_type != SLICE_TYPE_I )
 233     {
 234         bs_write1( s, sh->b_ref_pic_list_reordering_l0 );
 235         if( sh->b_ref_pic_list_reordering_l0 )
 236         {
 237             for( i = 0; i < sh->i_num_ref_idx_l0_active; i++ )
 238             {
 239                 bs_write_ue( s, sh->ref_pic_list_order[0][i].idc );
 240                 bs_write_ue( s, sh->ref_pic_list_order[0][i].arg );
 241
 242             }
 243             bs_write_ue( s, 3 );
 244         }
 245     }
 246     if( sh->i_type == SLICE_TYPE_B )
 247     {
 248         bs_write1( s, sh->b_ref_pic_list_reordering_l1 );
 249         if( sh->b_ref_pic_list_reordering_l1 )
 250         {
 251             for( i = 0; i < sh->i_num_ref_idx_l1_active; i++ )
 252             {
 253                 bs_write_ue( s, sh->ref_pic_list_order[1][i].idc );
 254                 bs_write_ue( s, sh->ref_pic_list_order[1][i].arg );
 255             }
 256             bs_write_ue( s, 3 );
 257         }
 258     }
 259
 260     if( sh->pps->b_weighted_pred && ( sh->i_type == SLICE_TYPE_P || sh->i_type == SLICE_TYPE_SP ) )
 261     {
 262         /* pred_weight_table() */
 263         bs_write_ue( s, sh->weight[0][0].i_denom );
 264         bs_write_ue( s, sh->weight[0][1].i_denom );
 265         for( i = 0; i < sh->i_num_ref_idx_l0_active; i++ )
 266         {
 267             int luma_weight_l0_flag = !!sh->weight[i][0].weightfn;
 268             int chroma_weight_l0_flag = !!sh->weight[i][1].weightfn || !!sh->weight[i][2].weightfn;
 269             bs_write1( s, luma_weight_l0_flag );
 270             if( luma_weight_l0_flag )
 271             {
 272                 bs_write_se( s, sh->weight[i][0].i_scale );
 273                 bs_write_se( s, sh->weight[i][0].i_offset );
 274             }
 275             bs_write1( s, chroma_weight_l0_flag );
 276             if( chroma_weight_l0_flag )
 277             {
 278                 int j;
 279                 for( 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                 int i;
 305                 for( i = 0; i < sh->i_mmco_command_count; i++ )
 306                 {
 307                     bs_write_ue( s, 1 ); /* mark short term ref as unused */
 308                     bs_write_ue( s, sh->mmco[i].i_difference_of_pic_nums - 1 );
 309                 }
 310                 bs_write_ue( s, 0 ); /* end command list */
 311             }
 312         }
 313     }
 314
 315     if( sh->pps->b_cabac && sh->i_type != SLICE_TYPE_I )
 316     {
 317         bs_write_ue( s, sh->i_cabac_init_idc );
 318     }
 319     bs_write_se( s, sh->i_qp_delta );      /* slice qp delta */
 320
 321     if( sh->pps->b_deblocking_filter_control )
 322     {
 323         bs_write_ue( s, sh->i_disable_deblocking_filter_idc );
 324         if( sh->i_disable_deblocking_filter_idc != 1 )
 325         {
 326             bs_write_se( s, sh->i_alpha_c0_offset >> 1 );
 327             bs_write_se( s, sh->i_beta_offset >> 1 );
 328         }
 329     }
 330 }
 331
 332 /* If we are within a reasonable distance of the end of the memory allocated for the bitstream, */
 333 /* reallocate, adding an arbitrary amount of space (100 kilobytes). */
 334 static int x264_bitstream_check_buffer( x264_t *h )
 335 {
 336     uint8_t *bs_bak = h->out.p_bitstream;
 337     if( ( h->param.b_cabac && (h->cabac.p_end - h->cabac.p < 2500) )
 338      || ( h->out.bs.p_end - h->out.bs.p < 2500 ) )
 339     {
 340         intptr_t delta;
 341         int i;
 342
 343         h->out.i_bitstream += 100000;
 344         CHECKED_MALLOC( h->out.p_bitstream, h->out.i_bitstream );
 345         h->mc.memcpy_aligned( h->out.p_bitstream, bs_bak, (h->out.i_bitstream - 100000) & ~15 );
 346         delta = h->out.p_bitstream - bs_bak;
 347
 348         h->out.bs.p_start += delta;
 349         h->out.bs.p += delta;
 350         h->out.bs.p_end = h->out.p_bitstream + h->out.i_bitstream;
 351
 352         h->cabac.p_start += delta;
 353         h->cabac.p += delta;
 354         h->cabac.p_end = h->out.p_bitstream + h->out.i_bitstream;
 355
 356         for( i = 0; i <= h->out.i_nal; i++ )
 357             h->out.nal[i].p_payload += delta;
 358         x264_free( bs_bak );
 359     }
 360     return 0;
 361 fail:
 362     x264_free( bs_bak );
 363     return -1;
 364 }
 365
 366 /****************************************************************************
 367  *
 368  ****************************************************************************
 369  ****************************** External API*********************************
 370  ****************************************************************************
 371  *
 372  ****************************************************************************/
 373
 374 static int x264_validate_parameters( x264_t *h )
 375 {
 376 #ifdef HAVE_MMX
 377     if( !(x264_cpu_detect() & X264_CPU_SSE) )
 378     {
 379         x264_log( h, X264_LOG_ERROR, "your cpu does not support SSE1, but x264 was compiled with asm support\n");
 380         x264_log( h, X264_LOG_ERROR, "to run x264, recompile without asm support (configure --disable-asm)\n");
 381         return -1;
 382     }
 383 #endif
 384     if( h->param.i_width <= 0 || h->param.i_height <= 0 )
 385     {
 386         x264_log( h, X264_LOG_ERROR, "invalid width x height (%dx%d)\n",
 387                   h->param.i_width, h->param.i_height );
 388         return -1;
 389     }
 390
 391     if( h->param.i_width % 2 || h->param.i_height % 2 )
 392     {
 393         x264_log( h, X264_LOG_ERROR, "width or height not divisible by 2 (%dx%d)\n",
 394                   h->param.i_width, h->param.i_height );
 395         return -1;
 396     }
 397     if( h->param.i_csp != X264_CSP_I420 )
 398     {
 399         x264_log( h, X264_LOG_ERROR, "invalid CSP (only I420 supported)\n" );
 400         return -1;
 401     }
 402
 403     if( h->param.i_threads == X264_THREADS_AUTO )
 404         h->param.i_threads = x264_cpu_num_processors() * 3/2;
 405     h->param.i_threads = x264_clip3( h->param.i_threads, 1, X264_THREAD_MAX );
 406     if( h->param.i_threads > 1 )
 407     {
 408 #ifndef HAVE_PTHREAD
 409         x264_log( h, X264_LOG_WARNING, "not compiled with pthread support!\n");
 410         h->param.i_threads = 1;
 411 #endif
 412     }
 413
 414     if( h->param.b_interlaced )
 415     {
 416         if( h->param.analyse.i_me_method >= X264_ME_ESA )
 417         {
 418             x264_log( h, X264_LOG_WARNING, "interlace + me=esa is not implemented\n" );
 419             h->param.analyse.i_me_method = X264_ME_UMH;
 420         }
 421         if( h->param.analyse.i_direct_mv_pred > X264_DIRECT_PRED_SPATIAL )
 422         {
 423             x264_log( h, X264_LOG_WARNING, "interlace + direct=temporal is not implemented\n" );
 424             h->param.analyse.i_direct_mv_pred = X264_DIRECT_PRED_SPATIAL;
 425         }
 426         if( h->param.analyse.i_weighted_pred > 0 )
 427         {
 428             x264_log( h, X264_LOG_WARNING, "interlace + weightp is not implemented\n" );
 429             h->param.analyse.i_weighted_pred = X264_WEIGHTP_NONE;
 430         }
 431     }
 432
 433     /* Detect default ffmpeg settings and terminate with an error. */
 434     {
 435         int score = 0;
 436         score += h->param.analyse.i_me_range == 0;
 437         score += h->param.rc.i_qp_step == 3;
 438         score += h->param.i_keyint_max == 12;
 439         score += h->param.rc.i_qp_min == 2;
 440         score += h->param.rc.i_qp_max == 31;
 441         score += h->param.rc.f_qcompress == 0.5;
 442         score += fabs(h->param.rc.f_ip_factor - 1.25) < 0.01;
 443         score += fabs(h->param.rc.f_pb_factor - 1.25) < 0.01;
 444         score += h->param.analyse.inter == 0 && h->param.analyse.i_subpel_refine == 8;
 445         if( score >= 5 )
 446         {
 447             x264_log( h, X264_LOG_ERROR, "broken ffmpeg default settings detected\n" );
 448             x264_log( h, X264_LOG_ERROR, "use an encoding preset (vpre)\n" );
 449             return -1;
 450         }
 451     }
 452
 453     if( h->param.rc.i_rc_method < 0 || h->param.rc.i_rc_method > 2 )
 454     {
 455         x264_log( h, X264_LOG_ERROR, "no ratecontrol method specified\n" );
 456         return -1;
 457     }
 458     h->param.rc.f_rf_constant = x264_clip3f( h->param.rc.f_rf_constant, 0, 51 );
 459     h->param.rc.i_qp_constant = x264_clip3( h->param.rc.i_qp_constant, 0, 51 );
 460     if( h->param.rc.i_rc_method == X264_RC_CRF )
 461     {
 462         h->param.rc.i_qp_constant = h->param.rc.f_rf_constant;
 463         h->param.rc.i_bitrate = 0;
 464     }
 465     if( (h->param.rc.i_rc_method == X264_RC_CQP || h->param.rc.i_rc_method == X264_RC_CRF)
 466         && h->param.rc.i_qp_constant == 0 )
 467     {
 468         h->mb.b_lossless = 1;
 469         h->param.i_cqm_preset = X264_CQM_FLAT;
 470         h->param.psz_cqm_file = NULL;
 471         h->param.rc.i_rc_method = X264_RC_CQP;
 472         h->param.rc.f_ip_factor = 1;
 473         h->param.rc.f_pb_factor = 1;
 474         h->param.analyse.b_psnr = 0;
 475         h->param.analyse.b_ssim = 0;
 476         h->param.analyse.i_chroma_qp_offset = 0;
 477         h->param.analyse.i_trellis = 0;
 478         h->param.analyse.b_fast_pskip = 0;
 479         h->param.analyse.i_noise_reduction = 0;
 480         h->param.analyse.f_psy_rd = 0;
 481         h->param.i_bframe = 0;
 482         /* 8x8dct is not useful at all in CAVLC lossless */
 483         if( !h->param.b_cabac )
 484             h->param.analyse.b_transform_8x8 = 0;
 485     }
 486     if( h->param.rc.i_rc_method == X264_RC_CQP )
 487     {
 488         float qp_p = h->param.rc.i_qp_constant;
 489         float qp_i = qp_p - 6*log(h->param.rc.f_ip_factor)/log(2);
 490         float qp_b = qp_p + 6*log(h->param.rc.f_pb_factor)/log(2);
 491         h->param.rc.i_qp_min = x264_clip3( (int)(X264_MIN3( qp_p, qp_i, qp_b )), 0, 51 );
 492         h->param.rc.i_qp_max = x264_clip3( (int)(X264_MAX3( qp_p, qp_i, qp_b ) + .999), 0, 51 );
 493         h->param.rc.i_aq_mode = 0;
 494         h->param.rc.b_mb_tree = 0;
 495     }
 496     h->param.rc.i_qp_max = x264_clip3( h->param.rc.i_qp_max, 0, 51 );
 497     h->param.rc.i_qp_min = x264_clip3( h->param.rc.i_qp_min, 0, h->param.rc.i_qp_max );
 498
 499     int max_slices = (h->param.i_height+((16<<h->param.b_interlaced)-1))/(16<<h->param.b_interlaced);
 500     h->param.i_slice_count = x264_clip3( h->param.i_slice_count, 0, max_slices );
 501     h->param.i_slice_max_size = X264_MAX( h->param.i_slice_max_size, 0 );
 502     h->param.i_slice_max_mbs = X264_MAX( h->param.i_slice_max_mbs, 0 );
 503     if( h->param.b_interlaced && h->param.i_slice_max_size )
 504     {
 505         x264_log( h, X264_LOG_WARNING, "interlaced + slice-max-size is not implemented\n" );
 506         h->param.i_slice_max_size = 0;
 507     }
 508     if( h->param.b_interlaced && h->param.i_slice_max_mbs )
 509     {
 510         x264_log( h, X264_LOG_WARNING, "interlaced + slice-max-mbs is not implemented\n" );
 511         h->param.i_slice_max_mbs = 0;
 512     }
 513     if( h->param.i_slice_max_mbs || h->param.i_slice_max_size )
 514         h->param.i_slice_count = 0;
 515
 516     h->param.i_frame_reference = x264_clip3( h->param.i_frame_reference, 1, 16 );
 517     if( h->param.i_keyint_max <= 0 )
 518         h->param.i_keyint_max = 1;
 519     if( h->param.i_scenecut_threshold < 0 )
 520         h->param.i_scenecut_threshold = 0;
 521     h->param.i_keyint_min = x264_clip3( h->param.i_keyint_min, 1, h->param.i_keyint_max/2+1 );
 522     if( !h->param.analyse.i_subpel_refine && h->param.analyse.i_direct_mv_pred > X264_DIRECT_PRED_SPATIAL )
 523     {
 524         x264_log( h, X264_LOG_WARNING, "subme=0 + direct=temporal is not supported\n" );
 525         h->param.analyse.i_direct_mv_pred = X264_DIRECT_PRED_SPATIAL;
 526     }
 527     h->param.i_bframe = x264_clip3( h->param.i_bframe, 0, X264_BFRAME_MAX );
 528     if( h->param.i_keyint_max == 1 )
 529         h->param.i_bframe = 0;
 530     h->param.i_bframe_bias = x264_clip3( h->param.i_bframe_bias, -90, 100 );
 531     if( h->param.i_bframe <= 1 )
 532         h->param.i_bframe_pyramid = X264_B_PYRAMID_NONE;
 533     h->param.i_bframe_pyramid = x264_clip3( h->param.i_bframe_pyramid, X264_B_PYRAMID_NONE, X264_B_PYRAMID_NORMAL );
 534     if( !h->param.i_bframe )
 535     {
 536         h->param.i_bframe_adaptive = X264_B_ADAPT_NONE;
 537         h->param.analyse.i_direct_mv_pred = 0;
 538         h->param.analyse.b_weighted_bipred = 0;
 539     }
 540     h->param.rc.i_lookahead = x264_clip3( h->param.rc.i_lookahead, 0, X264_LOOKAHEAD_MAX );
 541     {
 542         int maxrate = X264_MAX( h->param.rc.i_vbv_max_bitrate, h->param.rc.i_bitrate );
 543         float bufsize = maxrate ? (float)h->param.rc.i_vbv_buffer_size / maxrate : 0;
 544         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;
 545         h->param.rc.i_lookahead = X264_MIN( h->param.rc.i_lookahead, X264_MAX( h->param.i_keyint_max, bufsize*fps ) );
 546     }
 547
 548     h->param.rc.f_qcompress = x264_clip3f( h->param.rc.f_qcompress, 0.0, 1.0 );
 549     if( !h->param.rc.i_lookahead || h->param.i_keyint_max == 1 || h->param.rc.f_qcompress == 1 )
 550         h->param.rc.b_mb_tree = 0;
 551     if( h->param.rc.b_stat_read )
 552         h->param.rc.i_lookahead = 0;
 553 #ifdef HAVE_PTHREAD
 554     if( h->param.i_sync_lookahead )
 555         h->param.i_sync_lookahead = x264_clip3( h->param.i_sync_lookahead, h->param.i_threads + h->param.i_bframe, X264_LOOKAHEAD_MAX );
 556     if( h->param.rc.b_stat_read || h->param.i_threads == 1 )
 557         h->param.i_sync_lookahead = 0;
 558 #else
 559     h->param.i_sync_lookahead = 0;
 560 #endif
 561
 562     h->mb.b_direct_auto_write = h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO
 563                                 && h->param.i_bframe
 564                                 && ( h->param.rc.b_stat_write || !h->param.rc.b_stat_read );
 565
 566     h->param.i_deblocking_filter_alphac0 = x264_clip3( h->param.i_deblocking_filter_alphac0, -6, 6 );
 567     h->param.i_deblocking_filter_beta    = x264_clip3( h->param.i_deblocking_filter_beta, -6, 6 );
 568     h->param.analyse.i_luma_deadzone[0] = x264_clip3( h->param.analyse.i_luma_deadzone[0], 0, 32 );
 569     h->param.analyse.i_luma_deadzone[1] = x264_clip3( h->param.analyse.i_luma_deadzone[1], 0, 32 );
 570
 571     h->param.i_cabac_init_idc = x264_clip3( h->param.i_cabac_init_idc, 0, 2 );
 572
 573     if( h->param.i_cqm_preset < X264_CQM_FLAT || h->param.i_cqm_preset > X264_CQM_CUSTOM )
 574         h->param.i_cqm_preset = X264_CQM_FLAT;
 575
 576     if( h->param.analyse.i_me_method < X264_ME_DIA ||
 577         h->param.analyse.i_me_method > X264_ME_TESA )
 578         h->param.analyse.i_me_method = X264_ME_HEX;
 579     if( h->param.analyse.i_me_range < 4 )
 580         h->param.analyse.i_me_range = 4;
 581     if( h->param.analyse.i_me_range > 16 && h->param.analyse.i_me_method <= X264_ME_HEX )
 582         h->param.analyse.i_me_range = 16;
 583     if( h->param.analyse.i_me_method == X264_ME_TESA &&
 584         (h->mb.b_lossless || h->param.analyse.i_subpel_refine <= 1) )
 585         h->param.analyse.i_me_method = X264_ME_ESA;
 586     h->param.analyse.i_subpel_refine = x264_clip3( h->param.analyse.i_subpel_refine, 0, 10 );
 587     h->param.analyse.b_mixed_references = h->param.analyse.b_mixed_references && h->param.i_frame_reference > 1;
 588     h->param.analyse.inter &= X264_ANALYSE_PSUB16x16|X264_ANALYSE_PSUB8x8|X264_ANALYSE_BSUB16x16|
 589                               X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
 590     h->param.analyse.intra &= X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
 591     if( !(h->param.analyse.inter & X264_ANALYSE_PSUB16x16) )
 592         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
 593     if( !h->param.analyse.b_transform_8x8 )
 594     {
 595         h->param.analyse.inter &= ~X264_ANALYSE_I8x8;
 596         h->param.analyse.intra &= ~X264_ANALYSE_I8x8;
 597     }
 598     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -12, 12);
 599     if( !h->param.b_cabac )
 600         h->param.analyse.i_trellis = 0;
 601     h->param.analyse.i_trellis = x264_clip3( h->param.analyse.i_trellis, 0, 2 );
 602     if( !h->param.analyse.b_psy )
 603     {
 604         h->param.analyse.f_psy_rd = 0;
 605         h->param.analyse.f_psy_trellis = 0;
 606     }
 607     if( !h->param.analyse.i_trellis )
 608         h->param.analyse.f_psy_trellis = 0;
 609     h->param.analyse.f_psy_rd = x264_clip3f( h->param.analyse.f_psy_rd, 0, 10 );
 610     h->param.analyse.f_psy_trellis = x264_clip3f( h->param.analyse.f_psy_trellis, 0, 10 );
 611     if( h->param.analyse.i_subpel_refine < 6 )
 612         h->param.analyse.f_psy_rd = 0;
 613     h->mb.i_psy_rd = FIX8( h->param.analyse.f_psy_rd );
 614     /* Psy RDO increases overall quantizers to improve the quality of luma--this indirectly hurts chroma quality */
 615     /* so we lower the chroma QP offset to compensate */
 616     /* This can be triggered repeatedly on multiple calls to parameter_validate, but since encoding
 617      * uses the pps chroma qp offset not the param chroma qp offset, this is not a problem. */
 618     if( h->mb.i_psy_rd )
 619         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_rd < 0.25 ? 1 : 2;
 620     h->mb.i_psy_trellis = FIX8( h->param.analyse.f_psy_trellis / 4 );
 621     /* Psy trellis has a similar effect. */
 622     if( h->mb.i_psy_trellis )
 623         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_trellis < 0.25 ? 1 : 2;
 624     else
 625         h->mb.i_psy_trellis = 0;
 626     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -12, 12);
 627     h->param.rc.i_aq_mode = x264_clip3( h->param.rc.i_aq_mode, 0, 2 );
 628     h->param.rc.f_aq_strength = x264_clip3f( h->param.rc.f_aq_strength, 0, 3 );
 629     if( h->param.rc.f_aq_strength == 0 )
 630         h->param.rc.i_aq_mode = 0;
 631     /* MB-tree requires AQ to be on, even if the strength is zero. */
 632     if( !h->param.rc.i_aq_mode && h->param.rc.b_mb_tree )
 633     {
 634         h->param.rc.i_aq_mode = 1;
 635         h->param.rc.f_aq_strength = 0;
 636     }
 637     if( h->param.rc.b_mb_tree && h->param.i_bframe_pyramid )
 638     {
 639         x264_log( h, X264_LOG_WARNING, "b-pyramid + mb-tree is not supported\n" );
 640         h->param.i_bframe_pyramid = X264_B_PYRAMID_NONE;
 641     }
 642     h->param.analyse.i_noise_reduction = x264_clip3( h->param.analyse.i_noise_reduction, 0, 1<<16 );
 643     if( h->param.analyse.i_subpel_refine == 10 && (h->param.analyse.i_trellis != 2 || !h->param.rc.i_aq_mode) )
 644         h->param.analyse.i_subpel_refine = 9;
 645
 646     {
 647         const x264_level_t *l = x264_levels;
 648         if( h->param.i_level_idc < 0 )
 649         {
 650             int maxrate_bak = h->param.rc.i_vbv_max_bitrate;
 651             if( h->param.rc.i_rc_method == X264_RC_ABR && h->param.rc.i_vbv_buffer_size <= 0 )
 652                 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate * 2;
 653             h->sps = h->sps_array;
 654             x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
 655             do h->param.i_level_idc = l->level_idc;
 656                 while( l[1].level_idc && x264_validate_levels( h, 0 ) && l++ );
 657             h->param.rc.i_vbv_max_bitrate = maxrate_bak;
 658         }
 659         else
 660         {
 661             while( l->level_idc && l->level_idc != h->param.i_level_idc )
 662                 l++;
 663             if( l->level_idc == 0 )
 664             {
 665                 x264_log( h, X264_LOG_ERROR, "invalid level_idc: %d\n", h->param.i_level_idc );
 666                 return -1;
 667             }
 668         }
 669         if( h->param.analyse.i_mv_range <= 0 )
 670             h->param.analyse.i_mv_range = l->mv_range >> h->param.b_interlaced;
 671         else
 672             h->param.analyse.i_mv_range = x264_clip3(h->param.analyse.i_mv_range, 32, 512 >> h->param.b_interlaced);
 673     }
 674
 675     h->param.analyse.i_weighted_pred = x264_clip3( h->param.analyse.i_weighted_pred, 0, X264_WEIGHTP_SMART );
 676     if( !h->param.analyse.i_weighted_pred && h->param.rc.b_mb_tree && h->param.analyse.b_psy && !h->param.b_interlaced )
 677         h->param.analyse.i_weighted_pred = X264_WEIGHTP_FAKE;
 678
 679     if( h->param.i_threads > 1 )
 680     {
 681         int r = h->param.analyse.i_mv_range_thread;
 682         int r2;
 683         if( r <= 0 )
 684         {
 685             // half of the available space is reserved and divided evenly among the threads,
 686             // the rest is allocated to whichever thread is far enough ahead to use it.
 687             // reserving more space increases quality for some videos, but costs more time
 688             // in thread synchronization.
 689             int max_range = (h->param.i_height + X264_THREAD_HEIGHT) / h->param.i_threads - X264_THREAD_HEIGHT;
 690             r = max_range / 2;
 691         }
 692         r = X264_MAX( r, h->param.analyse.i_me_range );
 693         r = X264_MIN( r, h->param.analyse.i_mv_range );
 694         // round up to use the whole mb row
 695         r2 = (r & ~15) + ((-X264_THREAD_HEIGHT) & 15);
 696         if( r2 < r )
 697             r2 += 16;
 698         x264_log( h, X264_LOG_DEBUG, "using mv_range_thread = %d\n", r2 );
 699         h->param.analyse.i_mv_range_thread = r2;
 700     }
 701
 702     if( h->param.rc.f_qblur < 0 )
 703         h->param.rc.f_qblur = 0;
 704     if( h->param.rc.f_complexity_blur < 0 )
 705         h->param.rc.f_complexity_blur = 0;
 706
 707     h->param.i_sps_id &= 31;
 708
 709     if( h->param.i_log_level < X264_LOG_INFO )
 710     {
 711         h->param.analyse.b_psnr = 0;
 712         h->param.analyse.b_ssim = 0;
 713     }
 714
 715     /* ensure the booleans are 0 or 1 so they can be used in math */
 716 #define BOOLIFY(x) h->param.x = !!h->param.x
 717     BOOLIFY( b_cabac );
 718     BOOLIFY( b_deblocking_filter );
 719     BOOLIFY( b_interlaced );
 720     BOOLIFY( analyse.b_transform_8x8 );
 721     BOOLIFY( analyse.b_chroma_me );
 722     BOOLIFY( analyse.b_fast_pskip );
 723     BOOLIFY( rc.b_stat_write );
 724     BOOLIFY( rc.b_stat_read );
 725     BOOLIFY( rc.b_mb_tree );
 726 #undef BOOLIFY
 727
 728     return 0;
 729 }
 730
 731 static void mbcmp_init( x264_t *h )
 732 {
 733     int satd = !h->mb.b_lossless && h->param.analyse.i_subpel_refine > 1;
 734     memcpy( h->pixf.mbcmp, satd ? h->pixf.satd : h->pixf.sad_aligned, sizeof(h->pixf.mbcmp) );
 735     memcpy( h->pixf.mbcmp_unaligned, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.mbcmp_unaligned) );
 736     h->pixf.intra_mbcmp_x3_16x16 = satd ? h->pixf.intra_satd_x3_16x16 : h->pixf.intra_sad_x3_16x16;
 737     h->pixf.intra_mbcmp_x3_8x8c = satd ? h->pixf.intra_satd_x3_8x8c : h->pixf.intra_sad_x3_8x8c;
 738     h->pixf.intra_mbcmp_x3_4x4 = satd ? h->pixf.intra_satd_x3_4x4 : h->pixf.intra_sad_x3_4x4;
 739     satd &= h->param.analyse.i_me_method == X264_ME_TESA;
 740     memcpy( h->pixf.fpelcmp, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.fpelcmp) );
 741     memcpy( h->pixf.fpelcmp_x3, satd ? h->pixf.satd_x3 : h->pixf.sad_x3, sizeof(h->pixf.fpelcmp_x3) );
 742     memcpy( h->pixf.fpelcmp_x4, satd ? h->pixf.satd_x4 : h->pixf.sad_x4, sizeof(h->pixf.fpelcmp_x4) );
 743 }
 744
 745 static void x264_set_aspect_ratio( x264_t *h, x264_param_t *param, int initial )
 746 {
 747     /* VUI */
 748     if( param->vui.i_sar_width > 0 && param->vui.i_sar_height > 0 )
 749     {
 750         int i_w = param->vui.i_sar_width;
 751         int i_h = param->vui.i_sar_height;
 752         int old_w = h->param.vui.i_sar_width;
 753         int old_h = h->param.vui.i_sar_height;
 754
 755         x264_reduce_fraction( &i_w, &i_h );
 756
 757         while( i_w > 65535 || i_h > 65535 )
 758         {
 759             i_w /= 2;
 760             i_h /= 2;
 761         }
 762
 763         x264_reduce_fraction( &i_w, &i_h );
 764
 765         if( i_w != old_w || i_h != old_h || initial )
 766         {
 767             h->param.vui.i_sar_width = 0;
 768             h->param.vui.i_sar_height = 0;
 769             if( i_w == 0 || i_h == 0 )
 770                 x264_log( h, X264_LOG_WARNING, "cannot create valid sample aspect ratio\n" );
 771             else
 772             {
 773                 x264_log( h, initial?X264_LOG_INFO:X264_LOG_DEBUG, "using SAR=%d/%d\n", i_w, i_h );
 774                 h->param.vui.i_sar_width = i_w;
 775                 h->param.vui.i_sar_height = i_h;
 776             }
 777         }
 778     }
 779 }
 780
 781 /****************************************************************************
 782  * x264_encoder_open:
 783  ****************************************************************************/
 784 x264_t *x264_encoder_open( x264_param_t *param )
 785 {
 786     x264_t *h;
 787     char buf[1000], *p;
 788     int i, qp, i_slicetype_length;
 789
 790     CHECKED_MALLOCZERO( h, sizeof(x264_t) );
 791
 792     /* Create a copy of param */
 793     memcpy( &h->param, param, sizeof(x264_param_t) );
 794
 795     if( param->param_free )
 796         param->param_free( param );
 797
 798     if( x264_validate_parameters( h ) < 0 )
 799         goto fail;
 800
 801     if( h->param.psz_cqm_file )
 802         if( x264_cqm_parse_file( h, h->param.psz_cqm_file ) < 0 )
 803             goto fail;
 804
 805     if( h->param.rc.psz_stat_out )
 806         h->param.rc.psz_stat_out = strdup( h->param.rc.psz_stat_out );
 807     if( h->param.rc.psz_stat_in )
 808         h->param.rc.psz_stat_in = strdup( h->param.rc.psz_stat_in );
 809
 810     x264_set_aspect_ratio( h, param, 1 );
 811
 812     x264_reduce_fraction( &h->param.i_fps_num, &h->param.i_fps_den );
 813
 814     /* Init x264_t */
 815     h->i_frame = -1;
 816     h->i_frame_num = 0;
 817     h->i_idr_pic_id = 0;
 818
 819     h->sps = &h->sps_array[0];
 820     x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
 821
 822     h->pps = &h->pps_array[0];
 823     x264_pps_init( h->pps, h->param.i_sps_id, &h->param, h->sps );
 824
 825     x264_validate_levels( h, 1 );
 826
 827     h->chroma_qp_table = i_chroma_qp_table + 12 + h->pps->i_chroma_qp_index_offset;
 828
 829     if( x264_cqm_init( h ) < 0 )
 830         goto fail;
 831
 832     h->mb.i_mb_count = h->sps->i_mb_width * h->sps->i_mb_height;
 833
 834     /* Init frames. */
 835     if( h->param.i_bframe_adaptive == X264_B_ADAPT_TRELLIS )
 836         h->frames.i_delay = X264_MAX(h->param.i_bframe,3)*4;
 837     else
 838         h->frames.i_delay = h->param.i_bframe;
 839     if( h->param.rc.b_mb_tree || h->param.rc.i_vbv_buffer_size )
 840         h->frames.i_delay = X264_MAX( h->frames.i_delay, h->param.rc.i_lookahead );
 841     i_slicetype_length = h->frames.i_delay;
 842     h->frames.i_delay += h->param.i_threads - 1;
 843     h->frames.i_delay = X264_MIN( h->frames.i_delay, X264_LOOKAHEAD_MAX );
 844     h->frames.i_delay += h->param.i_sync_lookahead;
 845
 846     h->frames.i_max_ref0 = h->param.i_frame_reference;
 847     h->frames.i_max_ref1 = h->sps->vui.i_num_reorder_frames;
 848     h->frames.i_max_dpb  = h->sps->vui.i_max_dec_frame_buffering;
 849     h->frames.b_have_lowres = !h->param.rc.b_stat_read
 850         && ( h->param.rc.i_rc_method == X264_RC_ABR
 851           || h->param.rc.i_rc_method == X264_RC_CRF
 852           || h->param.i_bframe_adaptive
 853           || h->param.i_scenecut_threshold
 854           || h->param.rc.b_mb_tree
 855           || h->param.analyse.i_weighted_pred == X264_WEIGHTP_SMART );
 856     h->frames.b_have_lowres |= h->param.rc.b_stat_read && h->param.rc.i_vbv_buffer_size > 0;
 857     h->frames.b_have_sub8x8_esa = !!(h->param.analyse.inter & X264_ANALYSE_PSUB8x8);
 858
 859     h->frames.i_last_idr = - h->param.i_keyint_max;
 860     h->frames.i_input    = 0;
 861
 862     CHECKED_MALLOCZERO( h->frames.unused[0], (h->frames.i_delay + 3) * sizeof(x264_frame_t *) );
 863     /* Allocate room for max refs plus a few extra just in case. */
 864     CHECKED_MALLOCZERO( h->frames.unused[1], (h->param.i_threads + 20) * sizeof(x264_frame_t *) );
 865     CHECKED_MALLOCZERO( h->frames.current, (h->param.i_sync_lookahead + h->param.i_bframe
 866                         + h->param.i_threads + 3) * sizeof(x264_frame_t *) );
 867     if( h->param.analyse.i_weighted_pred > 0 )
 868         CHECKED_MALLOCZERO( h->frames.blank_unused, h->param.i_threads * 4 * sizeof(x264_frame_t *) );
 869     h->i_ref0 = 0;
 870     h->i_ref1 = 0;
 871
 872     x264_rdo_init();
 873
 874     /* init CPU functions */
 875     x264_predict_16x16_init( h->param.cpu, h->predict_16x16 );
 876     x264_predict_8x8c_init( h->param.cpu, h->predict_8x8c );
 877     x264_predict_8x8_init( h->param.cpu, h->predict_8x8, &h->predict_8x8_filter );
 878     x264_predict_4x4_init( h->param.cpu, h->predict_4x4 );
 879     if( !h->param.b_cabac )
 880         x264_init_vlc_tables();
 881     x264_pixel_init( h->param.cpu, &h->pixf );
 882     x264_dct_init( h->param.cpu, &h->dctf );
 883     x264_zigzag_init( h->param.cpu, &h->zigzagf, h->param.b_interlaced );
 884     x264_mc_init( h->param.cpu, &h->mc );
 885     x264_quant_init( h, h->param.cpu, &h->quantf );
 886     x264_deblock_init( h->param.cpu, &h->loopf );
 887     x264_dct_init_weights();
 888
 889     mbcmp_init( h );
 890
 891     p = buf + sprintf( buf, "using cpu capabilities:" );
 892     for( i=0; x264_cpu_names[i].flags; i++ )
 893     {
 894         if( !strcmp(x264_cpu_names[i].name, "SSE2")
 895             && param->cpu & (X264_CPU_SSE2_IS_FAST|X264_CPU_SSE2_IS_SLOW) )
 896             continue;
 897         if( !strcmp(x264_cpu_names[i].name, "SSE3")
 898             && (param->cpu & X264_CPU_SSSE3 || !(param->cpu & X264_CPU_CACHELINE_64)) )
 899             continue;
 900         if( !strcmp(x264_cpu_names[i].name, "SSE4.1")
 901             && (param->cpu & X264_CPU_SSE42) )
 902             continue;
 903         if( (param->cpu & x264_cpu_names[i].flags) == x264_cpu_names[i].flags
 904             && (!i || x264_cpu_names[i].flags != x264_cpu_names[i-1].flags) )
 905             p += sprintf( p, " %s", x264_cpu_names[i].name );
 906     }
 907     if( !param->cpu )
 908         p += sprintf( p, " none!" );
 909     x264_log( h, X264_LOG_INFO, "%s\n", buf );
 910
 911     for( qp = h->param.rc.i_qp_min; qp <= h->param.rc.i_qp_max; qp++ )
 912         if( x264_analyse_init_costs( h, qp ) )
 913             goto fail;
 914     if( x264_analyse_init_costs( h, X264_LOOKAHEAD_QP ) )
 915         goto fail;
 916     if( h->cost_mv[1][2013] != 24 )
 917     {
 918         x264_log( h, X264_LOG_ERROR, "MV cost test failed: x264 has been miscompiled!\n" );
 919         goto fail;
 920     }
 921
 922     h->out.i_nal = 0;
 923     h->out.i_bitstream = X264_MAX( 1000000, h->param.i_width * h->param.i_height * 4
 924         * ( h->param.rc.i_rc_method == X264_RC_ABR ? pow( 0.95, h->param.rc.i_qp_min )
 925           : pow( 0.95, h->param.rc.i_qp_constant ) * X264_MAX( 1, h->param.rc.f_ip_factor )));
 926
 927     CHECKED_MALLOC( h->nal_buffer, h->out.i_bitstream * 3/2 + 4 );
 928     h->nal_buffer_size = h->out.i_bitstream * 3/2 + 4;
 929
 930     h->thread[0] = h;
 931     h->i_thread_num = 0;
 932     for( i = 1; i < h->param.i_threads + !!h->param.i_sync_lookahead; i++ )
 933         CHECKED_MALLOC( h->thread[i], sizeof(x264_t) );
 934
 935     for( i = 0; i < h->param.i_threads; i++ )
 936     {
 937         if( i > 0 )
 938             *h->thread[i] = *h;
 939         h->thread[i]->fdec = x264_frame_pop_unused( h, 1 );
 940         if( !h->thread[i]->fdec )
 941             goto fail;
 942         CHECKED_MALLOC( h->thread[i]->out.p_bitstream, h->out.i_bitstream );
 943         /* Start each thread with room for 8 NAL units; it'll realloc later if needed. */
 944         CHECKED_MALLOC( h->thread[i]->out.nal, 8*sizeof(x264_nal_t) );
 945         h->thread[i]->out.i_nals_allocated = 8;
 946         if( x264_macroblock_cache_init( h->thread[i] ) < 0 )
 947             goto fail;
 948     }
 949
 950     if( x264_lookahead_init( h, i_slicetype_length ) )
 951         goto fail;
 952
 953     if( x264_ratecontrol_new( h ) < 0 )
 954         goto fail;
 955
 956     if( h->param.psz_dump_yuv )
 957     {
 958         /* create or truncate the reconstructed video file */
 959         FILE *f = fopen( h->param.psz_dump_yuv, "w" );
 960         if( !f )
 961         {
 962             x264_log( h, X264_LOG_ERROR, "dump_yuv: can't write to %s\n", h->param.psz_dump_yuv );
 963             goto fail;
 964         }
 965         else if( !x264_is_regular_file( f ) )
 966         {
 967             x264_log( h, X264_LOG_ERROR, "dump_yuv: incompatible with non-regular file %s\n", h->param.psz_dump_yuv );
 968             goto fail;
 969         }
 970         fclose( f );
 971     }
 972
 973     x264_log( h, X264_LOG_INFO, "profile %s, level %d.%d\n",
 974         h->sps->i_profile_idc == PROFILE_BASELINE ? "Baseline" :
 975         h->sps->i_profile_idc == PROFILE_MAIN ? "Main" :
 976         h->sps->i_profile_idc == PROFILE_HIGH ? "High" :
 977         "High 4:4:4 Predictive", h->sps->i_level_idc/10, h->sps->i_level_idc%10 );
 978
 979     return h;
 980 fail:
 981     x264_free( h );
 982     return NULL;
 983 }
 984
 985 /****************************************************************************
 986  * x264_encoder_reconfig:
 987  ****************************************************************************/
 988 int x264_encoder_reconfig( x264_t *h, x264_param_t *param )
 989 {
 990     h = h->thread[h->i_thread_phase];
 991     x264_set_aspect_ratio( h, param, 0 );
 992 #define COPY(var) h->param.var = param->var
 993     COPY( i_frame_reference ); // but never uses more refs than initially specified
 994     COPY( i_bframe_bias );
 995     if( h->param.i_scenecut_threshold )
 996         COPY( i_scenecut_threshold ); // can't turn it on or off, only vary the threshold
 997     COPY( b_deblocking_filter );
 998     COPY( i_deblocking_filter_alphac0 );
 999     COPY( i_deblocking_filter_beta );
1000     COPY( analyse.intra );
1001     COPY( analyse.inter );
1002     COPY( analyse.i_direct_mv_pred );
1003     /* Scratch buffer prevents me_range from being increased for esa/tesa */
1004     if( h->param.analyse.i_me_method < X264_ME_ESA || param->analyse.i_me_range < h->param.analyse.i_me_range )
1005         COPY( analyse.i_me_range );
1006     COPY( analyse.i_noise_reduction );
1007     /* We can't switch out of subme=0 during encoding. */
1008     if( h->param.analyse.i_subpel_refine )
1009         COPY( analyse.i_subpel_refine );
1010     COPY( analyse.i_trellis );
1011     COPY( analyse.b_chroma_me );
1012     COPY( analyse.b_dct_decimate );
1013     COPY( analyse.b_fast_pskip );
1014     COPY( analyse.b_mixed_references );
1015     COPY( analyse.f_psy_rd );
1016     COPY( analyse.f_psy_trellis );
1017     // can only twiddle these if they were enabled to begin with:
1018     if( h->param.analyse.i_me_method >= X264_ME_ESA || param->analyse.i_me_method < X264_ME_ESA )
1019         COPY( analyse.i_me_method );
1020     if( h->param.analyse.i_me_method >= X264_ME_ESA && !h->frames.b_have_sub8x8_esa )
1021         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
1022     if( h->pps->b_transform_8x8_mode )
1023         COPY( analyse.b_transform_8x8 );
1024     if( h->frames.i_max_ref1 > 1 )
1025         COPY( i_bframe_pyramid );
1026     COPY( i_slice_max_size );
1027     COPY( i_slice_max_mbs );
1028     COPY( i_slice_count );
1029 #undef COPY
1030
1031     mbcmp_init( h );
1032
1033     return x264_validate_parameters( h );
1034 }
1035
1036 /* internal usage */
1037 static void x264_nal_start( x264_t *h, int i_type, int i_ref_idc )
1038 {
1039     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1040
1041     nal->i_ref_idc = i_ref_idc;
1042     nal->i_type    = i_type;
1043
1044     nal->i_payload= 0;
1045     nal->p_payload= &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8];
1046 }
1047 static int x264_nal_end( x264_t *h )
1048 {
1049     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1050     nal->i_payload = &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8] - nal->p_payload;
1051     h->out.i_nal++;
1052
1053     /* if number of allocated nals is not enough, re-allocate a larger one. */
1054     if( h->out.i_nal >= h->out.i_nals_allocated )
1055     {
1056         x264_nal_t *new_out = x264_malloc( sizeof(x264_nal_t) * (h->out.i_nals_allocated*2) );
1057         if( !new_out )
1058             return -1;
1059         memcpy( new_out, h->out.nal, sizeof(x264_nal_t) * (h->out.i_nals_allocated) );
1060         x264_free( h->out.nal );
1061         h->out.nal = new_out;
1062         h->out.i_nals_allocated *= 2;
1063     }
1064     return 0;
1065 }
1066
1067 static int x264_encoder_encapsulate_nals( x264_t *h )
1068 {
1069     int nal_size = 0, i;
1070     for( i = 0; i < h->out.i_nal; i++ )
1071         nal_size += h->out.nal[i].i_payload;
1072
1073     /* Worst-case NAL unit escaping: reallocate the buffer if it's too small. */
1074     if( h->nal_buffer_size < nal_size * 3/2 + h->out.i_nal * 4 )
1075     {
1076         uint8_t *buf = x264_malloc( nal_size * 2 + h->out.i_nal * 4 );
1077         if( !buf )
1078             return -1;
1079         x264_free( h->nal_buffer );
1080         h->nal_buffer = buf;
1081     }
1082
1083     uint8_t *nal_buffer = h->nal_buffer;
1084
1085     for( i = 0; i < h->out.i_nal; i++ )
1086     {
1087         int size = x264_nal_encode( nal_buffer, h->param.b_annexb, &h->out.nal[i] );
1088         h->out.nal[i].i_payload = size;
1089         h->out.nal[i].p_payload = nal_buffer;
1090         nal_buffer += size;
1091     }
1092
1093     return nal_buffer - h->nal_buffer;
1094 }
1095
1096 /****************************************************************************
1097  * x264_encoder_headers:
1098  ****************************************************************************/
1099 int x264_encoder_headers( x264_t *h, x264_nal_t **pp_nal, int *pi_nal )
1100 {
1101     int frame_size = 0;
1102     /* init bitstream context */
1103     h->out.i_nal = 0;
1104     bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
1105
1106     /* Write SEI, SPS and PPS. */
1107     x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
1108     if( x264_sei_version_write( h, &h->out.bs ) )
1109         return -1;
1110     if( x264_nal_end( h ) )
1111         return -1;
1112
1113     /* generate sequence parameters */
1114     x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
1115     x264_sps_write( &h->out.bs, h->sps );
1116     if( x264_nal_end( h ) )
1117         return -1;
1118
1119     /* generate picture parameters */
1120     x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
1121     x264_pps_write( &h->out.bs, h->pps );
1122     if( x264_nal_end( h ) )
1123         return -1;
1124     bs_flush( &h->out.bs );
1125
1126     frame_size = x264_encoder_encapsulate_nals( h );
1127
1128     /* now set output*/
1129     *pi_nal = h->out.i_nal;
1130     *pp_nal = &h->out.nal[0];
1131     h->out.i_nal = 0;
1132
1133     return frame_size;
1134 }
1135
1136 /* Check to see whether we have chosen a reference list ordering different
1137  * from the standard's default. */
1138 static inline void x264_reference_check_reorder( x264_t *h )
1139 {
1140     int i;
1141     for( i = 0; i < h->i_ref0 - 1; i++ )
1142         /* P and B-frames use different default orders. */
1143         if( h->sh.i_type == SLICE_TYPE_P ? h->fref0[i]->i_frame_num < h->fref0[i+1]->i_frame_num
1144                                          : h->fref0[i]->i_poc < h->fref0[i+1]->i_poc )
1145         {
1146             h->b_ref_reorder[0] = 1;
1147             break;
1148         }
1149 }
1150
1151 /* return -1 on failure, else return the index of the new reference frame */
1152 int x264_weighted_reference_duplicate( x264_t *h, int i_ref, const x264_weight_t *w )
1153 {
1154     int i = h->i_ref0;
1155     int j;
1156     x264_frame_t *newframe;
1157     if( i <= 1 ) /* empty list, definitely can't duplicate frame */
1158         return -1;
1159
1160     /* Find a place to insert the duplicate in the reference list. */
1161     for( j = 0; j < i; j++ )
1162         if( h->fref0[i_ref]->i_frame != h->fref0[j]->i_frame )
1163         {
1164             /* found a place, after j, make sure there is not already a duplicate there */
1165             if( j == i-1 || ( h->fref0[j+1] && h->fref0[i_ref]->i_frame != h->fref0[j+1]->i_frame ) )
1166                 break;
1167         }
1168
1169     if( j == i ) /* No room in the reference list for the duplicate. */
1170         return -1;
1171     j++;
1172
1173     newframe = x264_frame_pop_blank_unused( h );
1174
1175     //FIXME: probably don't need to copy everything
1176     *newframe = *h->fref0[i_ref];
1177     newframe->i_reference_count = 1;
1178     newframe->orig = h->fref0[i_ref];
1179     newframe->b_duplicate = 1;
1180     memcpy( h->fenc->weight[j], w, sizeof(h->fenc->weight[i]) );
1181
1182     /* shift the frames to make space for the dupe. */
1183     h->b_ref_reorder[0] = 1;
1184     if( h->i_ref0 < 16 )
1185         ++h->i_ref0;
1186     h->fref0[15] = NULL;
1187     x264_frame_unshift( &h->fref0[j], newframe );
1188
1189     return j;
1190 }
1191
1192 static void x264_weighted_pred_init( x264_t *h )
1193 {
1194     int i_ref;
1195     int i;
1196
1197     /* for now no analysis and set all weights to nothing */
1198     for( i_ref = 0; i_ref < h->i_ref0; i_ref++ )
1199         h->fenc->weighted[i_ref] = h->fref0[i_ref]->filtered[0];
1200
1201     // FIXME: This only supports weighting of one reference frame
1202     // and duplicates of that frame.
1203     h->fenc->i_lines_weighted = 0;
1204
1205     for( i_ref = 0; i_ref < h->i_ref0; i_ref++ )
1206         for( i = 0; i < 3; i++ )
1207             h->sh.weight[i_ref][i].weightfn = NULL;
1208
1209
1210     if( h->sh.i_type != SLICE_TYPE_P || h->param.analyse.i_weighted_pred <= 0 )
1211         return;
1212
1213     int i_padv = PADV << h->param.b_interlaced;
1214     int denom = -1;
1215     int weightluma = 0;
1216     int buffer_next = 0;
1217     int j;
1218     //FIXME: when chroma support is added, move this into loop
1219     h->sh.weight[0][1].weightfn = h->sh.weight[0][2].weightfn = NULL;
1220     h->sh.weight[0][1].i_denom = h->sh.weight[0][2].i_denom = 0;
1221     for( j = 0; j < h->i_ref0; j++ )
1222     {
1223         if( h->fenc->weight[j][0].weightfn )
1224         {
1225             h->sh.weight[j][0] = h->fenc->weight[j][0];
1226             // if weight is useless, don't write it to stream
1227             if( h->sh.weight[j][0].i_scale == 1<<h->sh.weight[j][0].i_denom && h->sh.weight[j][0].i_offset == 0 )
1228                 h->sh.weight[j][0].weightfn = NULL;
1229             else
1230             {
1231                 if( !weightluma )
1232                 {
1233                     weightluma = 1;
1234                     h->sh.weight[0][0].i_denom = denom = h->sh.weight[j][0].i_denom;
1235                     assert( x264_clip3( denom, 0, 7 ) == denom );
1236                 }
1237                 assert( h->sh.weight[j][0].i_denom == denom );
1238                 assert( x264_clip3( h->sh.weight[j][0].i_scale, 0, 127 ) == h->sh.weight[j][0].i_scale );
1239                 assert( x264_clip3( h->sh.weight[j][0].i_offset, -128, 127 ) == h->sh.weight[j][0].i_offset );
1240                 h->fenc->weighted[j] = h->mb.p_weight_buf[buffer_next++] +
1241                     h->fenc->i_stride[0] * i_padv + PADH;
1242             }
1243         }
1244
1245         //scale full resolution frame
1246         if( h->sh.weight[j][0].weightfn && h->param.i_threads == 1 )
1247         {
1248             uint8_t *src = h->fref0[j]->filtered[0] - h->fref0[j]->i_stride[0]*i_padv - PADH;
1249             uint8_t *dst = h->fenc->weighted[j] - h->fenc->i_stride[0]*i_padv - PADH;
1250             int stride = h->fenc->i_stride[0];
1251             int width = h->fenc->i_width[0] + PADH*2;
1252             int height = h->fenc->i_lines[0] + i_padv*2;
1253             x264_weight_scale_plane( h, dst, stride, src, stride, width, height, &h->sh.weight[j][0] );
1254             h->fenc->i_lines_weighted = height;
1255         }
1256     }
1257     if( !weightluma )
1258         h->sh.weight[0][0].i_denom = 0;
1259 }
1260
1261 static inline void x264_reference_build_list( x264_t *h, int i_poc )
1262 {
1263     int i;
1264     int b_ok;
1265
1266     /* build ref list 0/1 */
1267     h->i_ref0 = 0;
1268     h->i_ref1 = 0;
1269     for( i = 0; h->frames.reference[i]; i++ )
1270     {
1271         if( h->frames.reference[i]->i_poc < i_poc )
1272         {
1273             h->fref0[h->i_ref0++] = h->frames.reference[i];
1274         }
1275         else if( h->frames.reference[i]->i_poc > i_poc )
1276         {
1277             h->fref1[h->i_ref1++] = h->frames.reference[i];
1278         }
1279     }
1280
1281     /* Order ref0 from higher to lower poc */
1282     do
1283     {
1284         b_ok = 1;
1285         for( i = 0; i < h->i_ref0 - 1; i++ )
1286         {
1287             if( h->fref0[i]->i_poc < h->fref0[i+1]->i_poc )
1288             {
1289                 XCHG( x264_frame_t*, h->fref0[i], h->fref0[i+1] );
1290                 b_ok = 0;
1291                 break;
1292             }
1293         }
1294     } while( !b_ok );
1295
1296     if( h->sh.i_mmco_remove_from_end )
1297         for( i = h->i_ref0-1; i >= h->i_ref0 - h->sh.i_mmco_remove_from_end; i-- )
1298         {
1299             int diff = h->i_frame_num - h->fref0[i]->i_frame_num;
1300             h->sh.mmco[h->sh.i_mmco_command_count].i_poc = h->fref0[i]->i_poc;
1301             h->sh.mmco[h->sh.i_mmco_command_count++].i_difference_of_pic_nums = diff;
1302         }
1303
1304     /* Order ref1 from lower to higher poc (bubble sort) for B-frame */
1305     do
1306     {
1307         b_ok = 1;
1308         for( i = 0; i < h->i_ref1 - 1; i++ )
1309         {
1310             if( h->fref1[i]->i_poc > h->fref1[i+1]->i_poc )
1311             {
1312                 XCHG( x264_frame_t*, h->fref1[i], h->fref1[i+1] );
1313                 b_ok = 0;
1314                 break;
1315             }
1316         }
1317     } while( !b_ok );
1318
1319     x264_reference_check_reorder( h );
1320
1321     h->i_ref1 = X264_MIN( h->i_ref1, h->frames.i_max_ref1 );
1322     h->i_ref0 = X264_MIN( h->i_ref0, h->frames.i_max_ref0 );
1323     h->i_ref0 = X264_MIN( h->i_ref0, h->param.i_frame_reference ); // if reconfig() has lowered the limit
1324
1325     /* add duplicates */
1326     if( h->fenc->i_type == X264_TYPE_P )
1327     {
1328         if( h->param.analyse.i_weighted_pred == X264_WEIGHTP_SMART )
1329         {
1330             x264_weight_t w[3];
1331             w[1].weightfn = w[2].weightfn = NULL;
1332             if( h->param.rc.b_stat_read )
1333                 x264_ratecontrol_set_weights( h, h->fenc );
1334
1335             if( !h->fenc->weight[0][0].weightfn )
1336             {
1337                 h->fenc->weight[0][0].i_denom = 0;
1338                 SET_WEIGHT( w[0], 1, 1, 0, -1 );
1339                 x264_weighted_reference_duplicate( h, 0, w );
1340             }
1341             else
1342             {
1343                 if( h->fenc->weight[0][0].i_scale == 1<<h->fenc->weight[0][0].i_denom )
1344                 {
1345                     SET_WEIGHT( h->fenc->weight[0][0], 1, 1, 0, h->fenc->weight[0][0].i_offset );
1346                 }
1347                 x264_weighted_reference_duplicate( h, 0, weight_none );
1348                 if( h->fenc->weight[0][0].i_offset > -128 )
1349                 {
1350                     w[0] = h->fenc->weight[0][0];
1351                     w[0].i_offset--;
1352                     h->mc.weight_cache( h, &w[0] );
1353                     x264_weighted_reference_duplicate( h, 0, w );
1354                 }
1355             }
1356         }
1357         else if( h->param.analyse.i_weighted_pred == X264_WEIGHTP_BLIND )
1358         {
1359             //weighted offset=-1
1360             x264_weight_t w[3];
1361             SET_WEIGHT( w[0], 1, 1, 0, -1 );
1362             h->fenc->weight[0][0].i_denom = 0;
1363             w[1].weightfn = w[2].weightfn = NULL;
1364             x264_weighted_reference_duplicate( h, 0, w );
1365         }
1366     }
1367
1368     assert( h->i_ref0 + h->i_ref1 <= 16 );
1369     h->mb.pic.i_fref[0] = h->i_ref0;
1370     h->mb.pic.i_fref[1] = h->i_ref1;
1371 }
1372
1373 static void x264_fdec_filter_row( x264_t *h, int mb_y )
1374 {
1375     /* mb_y is the mb to be encoded next, not the mb to be filtered here */
1376     int b_hpel = h->fdec->b_kept_as_ref;
1377     int b_deblock = !h->sh.i_disable_deblocking_filter_idc;
1378     int b_end = mb_y == h->sps->i_mb_height;
1379     int min_y = mb_y - (1 << h->sh.b_mbaff);
1380     int max_y = b_end ? h->sps->i_mb_height : mb_y;
1381     b_deblock &= b_hpel || h->param.psz_dump_yuv;
1382     if( mb_y & h->sh.b_mbaff )
1383         return;
1384     if( min_y < 0 )
1385         return;
1386
1387     if( !b_end )
1388     {
1389         int i, j;
1390         for( j=0; j<=h->sh.b_mbaff; j++ )
1391             for( i=0; i<3; i++ )
1392             {
1393                 memcpy( h->mb.intra_border_backup[j][i],
1394                         h->fdec->plane[i] + ((mb_y*16 >> !!i) + j - 1 - h->sh.b_mbaff) * h->fdec->i_stride[i],
1395                         h->sps->i_mb_width*16 >> !!i );
1396             }
1397     }
1398
1399     if( b_deblock )
1400     {
1401         int y;
1402         for( y = min_y; y < max_y; y += (1 << h->sh.b_mbaff) )
1403             x264_frame_deblock_row( h, y );
1404     }
1405
1406     if( b_hpel )
1407     {
1408         x264_frame_expand_border( h, h->fdec, min_y, b_end );
1409         if( h->param.analyse.i_subpel_refine )
1410         {
1411             x264_frame_filter( h, h->fdec, min_y, b_end );
1412             x264_frame_expand_border_filtered( h, h->fdec, min_y, b_end );
1413         }
1414     }
1415
1416     if( h->param.i_threads > 1 && h->fdec->b_kept_as_ref )
1417     {
1418         x264_frame_cond_broadcast( h->fdec, mb_y*16 + (b_end ? 10000 : -(X264_THREAD_HEIGHT << h->sh.b_mbaff)) );
1419     }
1420
1421     min_y = X264_MAX( min_y*16-8, 0 );
1422     max_y = b_end ? h->param.i_height : mb_y*16-8;
1423
1424     if( h->param.analyse.b_psnr )
1425     {
1426         int i;
1427         for( i=0; i<3; i++ )
1428             h->stat.frame.i_ssd[i] +=
1429                 x264_pixel_ssd_wxh( &h->pixf,
1430                     h->fdec->plane[i] + (min_y>>!!i) * h->fdec->i_stride[i], h->fdec->i_stride[i],
1431                     h->fenc->plane[i] + (min_y>>!!i) * h->fenc->i_stride[i], h->fenc->i_stride[i],
1432                     h->param.i_width >> !!i, (max_y-min_y) >> !!i );
1433     }
1434
1435     if( h->param.analyse.b_ssim )
1436     {
1437         x264_emms();
1438         /* offset by 2 pixels to avoid alignment of ssim blocks with dct blocks,
1439          * and overlap by 4 */
1440         min_y += min_y == 0 ? 2 : -6;
1441         h->stat.frame.f_ssim +=
1442             x264_pixel_ssim_wxh( &h->pixf,
1443                 h->fdec->plane[0] + 2+min_y*h->fdec->i_stride[0], h->fdec->i_stride[0],
1444                 h->fenc->plane[0] + 2+min_y*h->fenc->i_stride[0], h->fenc->i_stride[0],
1445                 h->param.i_width-2, max_y-min_y, h->scratch_buffer );
1446     }
1447 }
1448
1449 static inline int x264_reference_update( x264_t *h )
1450 {
1451     int i, j;
1452     if( !h->fdec->b_kept_as_ref )
1453     {
1454         if( h->param.i_threads > 1 )
1455         {
1456             x264_frame_push_unused( h, h->fdec );
1457             h->fdec = x264_frame_pop_unused( h, 1 );
1458             if( !h->fdec )
1459                 return -1;
1460         }
1461         return 0;
1462     }
1463
1464     /* apply mmco from previous frame. */
1465     for( i = 0; i < h->sh.i_mmco_command_count; i++ )
1466         for( j = 0; h->frames.reference[j]; j++ )
1467             if( h->frames.reference[j]->i_poc == h->sh.mmco[i].i_poc )
1468                 x264_frame_push_unused( h, x264_frame_shift( &h->frames.reference[j] ) );
1469
1470     /* move frame in the buffer */
1471     x264_frame_push( h->frames.reference, h->fdec );
1472     if( h->frames.reference[h->sps->i_num_ref_frames] )
1473         x264_frame_push_unused( h, x264_frame_shift( h->frames.reference ) );
1474     h->fdec = x264_frame_pop_unused( h, 1 );
1475     if( !h->fdec )
1476         return -1;
1477     return 0;
1478 }
1479
1480 static inline void x264_reference_reset( x264_t *h )
1481 {
1482     while( h->frames.reference[0] )
1483         x264_frame_push_unused( h, x264_frame_pop( h->frames.reference ) );
1484     h->fdec->i_poc =
1485     h->fenc->i_poc = 0;
1486 }
1487
1488 static inline void x264_reference_hierarchy_reset( x264_t *h )
1489 {
1490     int i, ref;
1491     int b_hasdelayframe = 0;
1492     if( !h->param.i_bframe_pyramid )
1493         return;
1494
1495     /* look for delay frames -- chain must only contain frames that are disposable */
1496     for( i = 0; h->frames.current[i] && IS_DISPOSABLE( h->frames.current[i]->i_type ); i++ )
1497         b_hasdelayframe |= h->frames.current[i]->i_dts
1498                         != h->frames.current[i]->i_frame + h->sps->vui.i_num_reorder_frames;
1499
1500     if( h->param.i_bframe_pyramid != X264_B_PYRAMID_STRICT && !b_hasdelayframe )
1501         return;
1502
1503     /* Remove last BREF. There will never be old BREFs in the
1504      * dpb during a BREF decode when pyramid == STRICT */
1505     for( ref = 0; h->frames.reference[ref]; ref++ )
1506     {
1507         if( h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT
1508             && h->frames.reference[ref]->i_type == X264_TYPE_BREF )
1509         {
1510             int diff = h->i_frame_num - h->frames.reference[ref]->i_frame_num;
1511             h->sh.mmco[h->sh.i_mmco_command_count].i_difference_of_pic_nums = diff;
1512             h->sh.mmco[h->sh.i_mmco_command_count++].i_poc = h->frames.reference[ref]->i_poc;
1513             x264_frame_push_unused( h, x264_frame_pop( h->frames.reference ) );
1514             h->b_ref_reorder[0] = 1;
1515             break;
1516         }
1517     }
1518
1519     /* Prepare to room in the dpb for the delayed display time of the later b-frame's */
1520     h->sh.i_mmco_remove_from_end = X264_MAX( ref + 2 - h->frames.i_max_dpb, 0 );
1521 }
1522
1523 static inline void x264_slice_init( x264_t *h, int i_nal_type, int i_global_qp )
1524 {
1525     /* ------------------------ Create slice header  ----------------------- */
1526     if( i_nal_type == NAL_SLICE_IDR )
1527     {
1528         x264_slice_header_init( h, &h->sh, h->sps, h->pps, h->i_idr_pic_id, h->i_frame_num, i_global_qp );
1529
1530         /* increment id */
1531         h->i_idr_pic_id = ( h->i_idr_pic_id + 1 ) % 65536;
1532     }
1533     else
1534     {
1535         x264_slice_header_init( h, &h->sh, h->sps, h->pps, -1, h->i_frame_num, i_global_qp );
1536
1537         /* always set the real higher num of ref frame used */
1538         h->sh.b_num_ref_idx_override = 1;
1539         h->sh.i_num_ref_idx_l0_active = h->i_ref0 <= 0 ? 1 : h->i_ref0;
1540         h->sh.i_num_ref_idx_l1_active = h->i_ref1 <= 0 ? 1 : h->i_ref1;
1541     }
1542
1543     h->fdec->i_frame_num = h->sh.i_frame_num;
1544
1545     if( h->sps->i_poc_type == 0 )
1546     {
1547         h->sh.i_poc_lsb = h->fdec->i_poc & ( (1 << h->sps->i_log2_max_poc_lsb) - 1 );
1548         h->sh.i_delta_poc_bottom = 0;
1549     }
1550     else if( h->sps->i_poc_type == 1 )
1551     {
1552         /* FIXME TODO FIXME */
1553     }
1554     else
1555     {
1556         /* Nothing to do ? */
1557     }
1558
1559     x264_macroblock_slice_init( h );
1560 }
1561
1562 static int x264_slice_write( x264_t *h )
1563 {
1564     int i_skip;
1565     int mb_xy, i_mb_x, i_mb_y;
1566     int i, i_list, i_ref, i_skip_bak = 0; /* Shut up GCC. */
1567     bs_t bs_bak;
1568     x264_cabac_t cabac_bak;
1569     uint8_t cabac_prevbyte_bak = 0; /* Shut up GCC. */
1570     /* Assume no more than 3 bytes of NALU escaping. */
1571     int slice_max_size = h->param.i_slice_max_size > 0 ? (h->param.i_slice_max_size-3-NALU_OVERHEAD)*8 : INT_MAX;
1572     int starting_bits = bs_pos(&h->out.bs);
1573
1574     /* Slice */
1575     x264_nal_start( h, h->i_nal_type, h->i_nal_ref_idc );
1576
1577     /* Slice header */
1578     x264_slice_header_write( &h->out.bs, &h->sh, h->i_nal_ref_idc );
1579     if( h->param.b_cabac )
1580     {
1581         /* alignment needed */
1582         bs_align_1( &h->out.bs );
1583
1584         /* init cabac */
1585         x264_cabac_context_init( &h->cabac, h->sh.i_type, h->sh.i_qp, h->sh.i_cabac_init_idc );
1586         x264_cabac_encode_init ( &h->cabac, h->out.bs.p, h->out.bs.p_end );
1587     }
1588     h->mb.i_last_qp = h->sh.i_qp;
1589     h->mb.i_last_dqp = 0;
1590
1591     i_mb_y = h->sh.i_first_mb / h->sps->i_mb_width;
1592     i_mb_x = h->sh.i_first_mb % h->sps->i_mb_width;
1593     i_skip = 0;
1594
1595     while( (mb_xy = i_mb_x + i_mb_y * h->sps->i_mb_width) <= h->sh.i_last_mb )
1596     {
1597         int mb_spos = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
1598         if( h->param.i_slice_max_size > 0 )
1599         {
1600             /* We don't need the contexts because flushing the CABAC encoder has no context
1601              * dependency and macroblocks are only re-encoded in the case where a slice is
1602              * ended (and thus the content of all contexts are thrown away). */
1603             if( h->param.b_cabac )
1604             {
1605                 memcpy( &cabac_bak, &h->cabac, offsetof(x264_cabac_t, f8_bits_encoded) );
1606                 /* x264's CABAC writer modifies the previous byte during carry, so it has to be
1607                  * backed up. */
1608                 cabac_prevbyte_bak = h->cabac.p[-1];
1609             }
1610             else
1611             {
1612                 bs_bak = h->out.bs;
1613                 i_skip_bak = i_skip;
1614             }
1615         }
1616
1617         if( i_mb_x == 0 && !h->mb.b_reencode_mb )
1618             x264_fdec_filter_row( h, i_mb_y );
1619
1620         /* load cache */
1621         x264_macroblock_cache_load( h, i_mb_x, i_mb_y );
1622
1623         x264_macroblock_analyse( h );
1624
1625         /* encode this macroblock -> be careful it can change the mb type to P_SKIP if needed */
1626         x264_macroblock_encode( h );
1627
1628         if( x264_bitstream_check_buffer( h ) )
1629             return -1;
1630
1631         if( h->param.b_cabac )
1632         {
1633             if( mb_xy > h->sh.i_first_mb && !(h->sh.b_mbaff && (i_mb_y&1)) )
1634                 x264_cabac_encode_terminal( &h->cabac );
1635
1636             if( IS_SKIP( h->mb.i_type ) )
1637                 x264_cabac_mb_skip( h, 1 );
1638             else
1639             {
1640                 if( h->sh.i_type != SLICE_TYPE_I )
1641                     x264_cabac_mb_skip( h, 0 );
1642                 x264_macroblock_write_cabac( h, &h->cabac );
1643             }
1644         }
1645         else
1646         {
1647             if( IS_SKIP( h->mb.i_type ) )
1648                 i_skip++;
1649             else
1650             {
1651                 if( h->sh.i_type != SLICE_TYPE_I )
1652                 {
1653                     bs_write_ue( &h->out.bs, i_skip );  /* skip run */
1654                     i_skip = 0;
1655                 }
1656                 x264_macroblock_write_cavlc( h, &h->out.bs );
1657             }
1658         }
1659
1660         int total_bits = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
1661         int mb_size = total_bits - mb_spos;
1662
1663         /* We'll just re-encode this last macroblock if we go over the max slice size. */
1664         if( total_bits - starting_bits > slice_max_size && !h->mb.b_reencode_mb )
1665         {
1666             if( mb_xy != h->sh.i_first_mb )
1667             {
1668                 if( h->param.b_cabac )
1669                 {
1670                     memcpy( &h->cabac, &cabac_bak, offsetof(x264_cabac_t, f8_bits_encoded) );
1671                     h->cabac.p[-1] = cabac_prevbyte_bak;
1672                 }
1673                 else
1674                 {
1675                     h->out.bs = bs_bak;
1676                     i_skip = i_skip_bak;
1677                 }
1678                 h->mb.b_reencode_mb = 1;
1679                 h->sh.i_last_mb = mb_xy-1;
1680                 break;
1681             }
1682             else
1683             {
1684                 h->sh.i_last_mb = mb_xy;
1685                 h->mb.b_reencode_mb = 0;
1686             }
1687         }
1688         else
1689             h->mb.b_reencode_mb = 0;
1690
1691 #if VISUALIZE
1692         if( h->param.b_visualize )
1693             x264_visualize_mb( h );
1694 #endif
1695
1696         /* save cache */
1697         x264_macroblock_cache_save( h );
1698
1699         /* accumulate mb stats */
1700         h->stat.frame.i_mb_count[h->mb.i_type]++;
1701
1702         if( !IS_INTRA(h->mb.i_type) && !IS_SKIP(h->mb.i_type) && !IS_DIRECT(h->mb.i_type) )
1703         {
1704             if( h->mb.i_partition != D_8x8 )
1705                     h->stat.frame.i_mb_partition[h->mb.i_partition] += 4;
1706                 else
1707                     for( i = 0; i < 4; i++ )
1708                         h->stat.frame.i_mb_partition[h->mb.i_sub_partition[i]] ++;
1709             if( h->param.i_frame_reference > 1 )
1710                 for( i_list = 0; i_list <= (h->sh.i_type == SLICE_TYPE_B); i_list++ )
1711                     for( i = 0; i < 4; i++ )
1712                     {
1713                         i_ref = h->mb.cache.ref[i_list][ x264_scan8[4*i] ];
1714                         if( i_ref >= 0 )
1715                             h->stat.frame.i_mb_count_ref[i_list][i_ref] ++;
1716                     }
1717         }
1718
1719         if( h->param.i_log_level >= X264_LOG_INFO )
1720         {
1721             if( h->mb.i_cbp_luma || h->mb.i_cbp_chroma )
1722             {
1723                 int cbpsum = (h->mb.i_cbp_luma&1) + ((h->mb.i_cbp_luma>>1)&1)
1724                            + ((h->mb.i_cbp_luma>>2)&1) + (h->mb.i_cbp_luma>>3);
1725                 int b_intra = IS_INTRA(h->mb.i_type);
1726                 h->stat.frame.i_mb_cbp[!b_intra + 0] += cbpsum;
1727                 h->stat.frame.i_mb_cbp[!b_intra + 2] += h->mb.i_cbp_chroma >= 1;
1728                 h->stat.frame.i_mb_cbp[!b_intra + 4] += h->mb.i_cbp_chroma == 2;
1729             }
1730             if( h->mb.i_cbp_luma && !IS_INTRA(h->mb.i_type) )
1731             {
1732                 h->stat.frame.i_mb_count_8x8dct[0] ++;
1733                 h->stat.frame.i_mb_count_8x8dct[1] += h->mb.b_transform_8x8;
1734             }
1735             if( IS_INTRA(h->mb.i_type) && h->mb.i_type != I_PCM )
1736             {
1737                 if( h->mb.i_type == I_16x16 )
1738                     h->stat.frame.i_mb_pred_mode[0][h->mb.i_intra16x16_pred_mode]++;
1739                 else if( h->mb.i_type == I_8x8 )
1740                     for( i = 0; i < 16; i += 4 )
1741                         h->stat.frame.i_mb_pred_mode[1][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
1742                 else //if( h->mb.i_type == I_4x4 )
1743                     for( i = 0; i < 16; i++ )
1744                         h->stat.frame.i_mb_pred_mode[2][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
1745             }
1746         }
1747
1748         x264_ratecontrol_mb( h, mb_size );
1749
1750         if( h->sh.b_mbaff )
1751         {
1752             i_mb_x += i_mb_y & 1;
1753             i_mb_y ^= i_mb_x < h->sps->i_mb_width;
1754         }
1755         else
1756             i_mb_x++;
1757         if( i_mb_x == h->sps->i_mb_width )
1758         {
1759             i_mb_y++;
1760             i_mb_x = 0;
1761         }
1762     }
1763
1764     if( h->param.b_cabac )
1765     {
1766         x264_cabac_encode_flush( h, &h->cabac );
1767         h->out.bs.p = h->cabac.p;
1768     }
1769     else
1770     {
1771         if( i_skip > 0 )
1772             bs_write_ue( &h->out.bs, i_skip );  /* last skip run */
1773         /* rbsp_slice_trailing_bits */
1774         bs_rbsp_trailing( &h->out.bs );
1775         bs_flush( &h->out.bs );
1776     }
1777     if( x264_nal_end( h ) )
1778         return -1;
1779
1780     if( h->sh.i_last_mb == h->mb.i_mb_count-1 )
1781     {
1782         h->stat.frame.i_misc_bits = bs_pos( &h->out.bs )
1783                                   + (h->out.i_nal*NALU_OVERHEAD * 8)
1784                                   - h->stat.frame.i_tex_bits
1785                                   - h->stat.frame.i_mv_bits;
1786         x264_fdec_filter_row( h, h->sps->i_mb_height );
1787     }
1788
1789     return 0;
1790 }
1791
1792 static void x264_thread_sync_context( x264_t *dst, x264_t *src )
1793 {
1794     x264_frame_t **f;
1795     if( dst == src )
1796         return;
1797
1798     // reference counting
1799     for( f = src->frames.reference; *f; f++ )
1800         (*f)->i_reference_count++;
1801     for( f = dst->frames.reference; *f; f++ )
1802         x264_frame_push_unused( src, *f );
1803     src->fdec->i_reference_count++;
1804     x264_frame_push_unused( src, dst->fdec );
1805
1806     // copy everything except the per-thread pointers and the constants.
1807     memcpy( &dst->i_frame, &src->i_frame, offsetof(x264_t, mb.type) - offsetof(x264_t, i_frame) );
1808     dst->param = src->param;
1809     dst->stat = src->stat;
1810 }
1811
1812 static void x264_thread_sync_stat( x264_t *dst, x264_t *src )
1813 {
1814     if( dst == src )
1815         return;
1816     memcpy( &dst->stat.i_frame_count, &src->stat.i_frame_count, sizeof(dst->stat) - sizeof(dst->stat.frame) );
1817 }
1818
1819 static void *x264_slices_write( x264_t *h )
1820 {
1821     int i_slice_num = 0;
1822     if( h->param.i_sync_lookahead )
1823         x264_lower_thread_priority( 10 );
1824
1825 #ifdef HAVE_MMX
1826     /* Misalign mask has to be set separately for each thread. */
1827     if( h->param.cpu&X264_CPU_SSE_MISALIGN )
1828         x264_cpu_mask_misalign_sse();
1829 #endif
1830
1831 #if VISUALIZE
1832     if( h->param.b_visualize )
1833         if( x264_visualize_init( h ) )
1834             return (void *)-1;
1835 #endif
1836
1837     /* init stats */
1838     memset( &h->stat.frame, 0, sizeof(h->stat.frame) );
1839     h->mb.b_reencode_mb = 0;
1840     while( h->sh.i_first_mb < h->mb.i_mb_count )
1841     {
1842         h->sh.i_last_mb = h->mb.i_mb_count - 1;
1843         if( h->param.i_slice_max_mbs )
1844             h->sh.i_last_mb = h->sh.i_first_mb + h->param.i_slice_max_mbs - 1;
1845         else if( h->param.i_slice_count )
1846         {
1847             x264_emms();
1848             i_slice_num++;
1849             double height = h->sps->i_mb_height >> h->param.b_interlaced;
1850             int width = h->sps->i_mb_width << h->param.b_interlaced;
1851             h->sh.i_last_mb = (int)(height * i_slice_num / h->param.i_slice_count + 0.5) * width - 1;
1852         }
1853         h->sh.i_last_mb = X264_MIN( h->sh.i_last_mb, h->mb.i_mb_count - 1 );
1854         if( x264_stack_align( x264_slice_write, h ) )
1855             return (void *)-1;
1856         h->sh.i_first_mb = h->sh.i_last_mb + 1;
1857     }
1858
1859 #if VISUALIZE
1860     if( h->param.b_visualize )
1861     {
1862         x264_visualize_show( h );
1863         x264_visualize_close( h );
1864     }
1865 #endif
1866
1867     return (void *)0;
1868 }
1869
1870 /****************************************************************************
1871  * x264_encoder_encode:
1872  *  XXX: i_poc   : is the poc of the current given picture
1873  *       i_frame : is the number of the frame being coded
1874  *  ex:  type frame poc
1875  *       I      0   2*0
1876  *       P      1   2*3
1877  *       B      2   2*1
1878  *       B      3   2*2
1879  *       P      4   2*6
1880  *       B      5   2*4
1881  *       B      6   2*5
1882  ****************************************************************************/
1883 int     x264_encoder_encode( x264_t *h,
1884                              x264_nal_t **pp_nal, int *pi_nal,
1885                              x264_picture_t *pic_in,
1886                              x264_picture_t *pic_out )
1887 {
1888     x264_t *thread_current, *thread_prev, *thread_oldest;
1889     int     i_nal_type;
1890     int     i_nal_ref_idc;
1891
1892     int   i_global_qp;
1893
1894     if( h->param.i_threads > 1)
1895     {
1896         thread_prev    = h->thread[ h->i_thread_phase ];
1897         h->i_thread_phase = (h->i_thread_phase + 1) % h->param.i_threads;
1898         thread_current = h->thread[ h->i_thread_phase ];
1899         thread_oldest  = h->thread[ (h->i_thread_phase + 1) % h->param.i_threads ];
1900         x264_thread_sync_context( thread_current, thread_prev );
1901         x264_thread_sync_ratecontrol( thread_current, thread_prev, thread_oldest );
1902         h = thread_current;
1903 //      fprintf(stderr, "current: %p  prev: %p  oldest: %p \n", thread_current, thread_prev, thread_oldest);
1904     }
1905     else
1906     {
1907         thread_current =
1908         thread_oldest  = h;
1909     }
1910
1911     // ok to call this before encoding any frames, since the initial values of fdec have b_kept_as_ref=0
1912     if( x264_reference_update( h ) )
1913         return -1;
1914     h->fdec->i_lines_completed = -1;
1915
1916     /* no data out */
1917     *pi_nal = 0;
1918     *pp_nal = NULL;
1919
1920     /* ------------------- Setup new frame from picture -------------------- */
1921     if( pic_in != NULL )
1922     {
1923         /* 1: Copy the picture to a frame and move it to a buffer */
1924         x264_frame_t *fenc = x264_frame_pop_unused( h, 0 );
1925         if( !fenc )
1926             return -1;
1927
1928         if( x264_frame_copy_picture( h, fenc, pic_in ) < 0 )
1929             return -1;
1930
1931         if( h->param.i_width != 16 * h->sps->i_mb_width ||
1932             h->param.i_height != 16 * h->sps->i_mb_height )
1933             x264_frame_expand_border_mod16( h, fenc );
1934
1935         fenc->i_frame = h->frames.i_input++;
1936
1937         if( h->frames.b_have_lowres )
1938             x264_frame_init_lowres( h, fenc );
1939
1940         if( h->param.rc.b_mb_tree && h->param.rc.b_stat_read )
1941         {
1942             if( x264_macroblock_tree_read( h, fenc ) )
1943                 return -1;
1944         }
1945         else if( h->param.rc.i_aq_mode )
1946             x264_adaptive_quant_frame( h, fenc );
1947
1948         /* 2: Place the frame into the queue for its slice type decision */
1949         x264_lookahead_put_frame( h, fenc );
1950
1951         if( h->frames.i_input <= h->frames.i_delay + 1 - h->param.i_threads )
1952         {
1953             /* Nothing yet to encode, waiting for filling of buffers */
1954             pic_out->i_type = X264_TYPE_AUTO;
1955             return 0;
1956         }
1957     }
1958     else
1959     {
1960         /* signal kills for lookahead thread */
1961         x264_pthread_mutex_lock( &h->lookahead->ifbuf.mutex );
1962         h->lookahead->b_exit_thread = 1;
1963         x264_pthread_cond_broadcast( &h->lookahead->ifbuf.cv_fill );
1964         x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
1965     }
1966
1967     h->i_frame++;
1968     /* 3: The picture is analyzed in the lookahead */
1969     if( !h->frames.current[0] )
1970         x264_lookahead_get_frames( h );
1971
1972     if( !h->frames.current[0] && x264_lookahead_is_empty( h ) )
1973         return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
1974
1975     /* ------------------- Get frame to be encoded ------------------------- */
1976     /* 4: get picture to encode */
1977     h->fenc = x264_frame_shift( h->frames.current );
1978     if( h->fenc->param )
1979     {
1980         x264_encoder_reconfig( h, h->fenc->param );
1981         if( h->fenc->param->param_free )
1982             h->fenc->param->param_free( h->fenc->param );
1983     }
1984
1985     if( h->fenc->i_type == X264_TYPE_IDR )
1986     {
1987         h->frames.i_last_idr = h->fenc->i_frame;
1988         h->i_frame_num = 0;
1989     }
1990     h->sh.i_mmco_command_count = 0;
1991     h->sh.i_mmco_remove_from_end = 0;
1992     h->b_ref_reorder[0] =
1993     h->b_ref_reorder[1] = 0;
1994
1995     /* ------------------- Setup frame context ----------------------------- */
1996     /* 5: Init data dependent of frame type */
1997     if( h->fenc->i_type == X264_TYPE_IDR )
1998     {
1999         /* reset ref pictures */
2000         i_nal_type    = NAL_SLICE_IDR;
2001         i_nal_ref_idc = NAL_PRIORITY_HIGHEST;
2002         h->sh.i_type = SLICE_TYPE_I;
2003         x264_reference_reset( h );
2004     }
2005     else if( h->fenc->i_type == X264_TYPE_I )
2006     {
2007         i_nal_type    = NAL_SLICE;
2008         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
2009         h->sh.i_type = SLICE_TYPE_I;
2010         x264_reference_hierarchy_reset( h );
2011     }
2012     else if( h->fenc->i_type == X264_TYPE_P )
2013     {
2014         i_nal_type    = NAL_SLICE;
2015         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
2016         h->sh.i_type = SLICE_TYPE_P;
2017         x264_reference_hierarchy_reset( h );
2018     }
2019     else if( h->fenc->i_type == X264_TYPE_BREF )
2020     {
2021         i_nal_type    = NAL_SLICE;
2022         i_nal_ref_idc = h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT ? NAL_PRIORITY_LOW : NAL_PRIORITY_HIGH;
2023         h->sh.i_type = SLICE_TYPE_B;
2024         x264_reference_hierarchy_reset( h );
2025     }
2026     else    /* B frame */
2027     {
2028         i_nal_type    = NAL_SLICE;
2029         i_nal_ref_idc = NAL_PRIORITY_DISPOSABLE;
2030         h->sh.i_type = SLICE_TYPE_B;
2031     }
2032
2033     h->fdec->i_poc =
2034     h->fenc->i_poc = 2 * (h->fenc->i_frame - h->frames.i_last_idr);
2035     h->fdec->i_type = h->fenc->i_type;
2036     h->fdec->i_frame = h->fenc->i_frame;
2037     h->fenc->b_kept_as_ref =
2038     h->fdec->b_kept_as_ref = i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE && h->param.i_keyint_max > 1;
2039
2040
2041
2042     /* ------------------- Init                ----------------------------- */
2043     /* build ref list 0/1 */
2044     x264_reference_build_list( h, h->fdec->i_poc );
2045
2046     /* ---------------------- Write the bitstream -------------------------- */
2047     /* Init bitstream context */
2048     h->out.i_nal = 0;
2049     bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
2050
2051     if( h->param.b_aud )
2052     {
2053         int pic_type;
2054
2055         if( h->sh.i_type == SLICE_TYPE_I )
2056             pic_type = 0;
2057         else if( h->sh.i_type == SLICE_TYPE_P )
2058             pic_type = 1;
2059         else if( h->sh.i_type == SLICE_TYPE_B )
2060             pic_type = 2;
2061         else
2062             pic_type = 7;
2063
2064         x264_nal_start( h, NAL_AUD, NAL_PRIORITY_DISPOSABLE );
2065         bs_write( &h->out.bs, 3, pic_type );
2066         bs_rbsp_trailing( &h->out.bs );
2067         if( x264_nal_end( h ) )
2068             return -1;
2069     }
2070
2071     h->i_nal_type = i_nal_type;
2072     h->i_nal_ref_idc = i_nal_ref_idc;
2073
2074     int overhead = NALU_OVERHEAD;
2075
2076     /* Write SPS and PPS */
2077     if( i_nal_type == NAL_SLICE_IDR && h->param.b_repeat_headers )
2078     {
2079         if( h->fenc->i_frame == 0 )
2080         {
2081             /* identify ourself */
2082             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2083             if( x264_sei_version_write( h, &h->out.bs ) )
2084                 return -1;
2085             if( x264_nal_end( h ) )
2086                 return -1;
2087             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
2088         }
2089
2090         /* generate sequence parameters */
2091         x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
2092         x264_sps_write( &h->out.bs, h->sps );
2093         if( x264_nal_end( h ) )
2094             return -1;
2095         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
2096
2097         /* generate picture parameters */
2098         x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
2099         x264_pps_write( &h->out.bs, h->pps );
2100         if( x264_nal_end( h ) )
2101             return -1;
2102         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
2103     }
2104
2105     /* Init the rate control */
2106     /* FIXME: Include slice header bit cost. */
2107     x264_ratecontrol_start( h, h->fenc->i_qpplus1, overhead*8 );
2108     i_global_qp = x264_ratecontrol_qp( h );
2109
2110     pic_out->i_qpplus1 =
2111     h->fdec->i_qpplus1 = i_global_qp + 1;
2112
2113     if( h->param.rc.b_stat_read && h->sh.i_type != SLICE_TYPE_I )
2114     {
2115         x264_reference_build_list_optimal( h );
2116         x264_reference_check_reorder( h );
2117     }
2118
2119     if( h->sh.i_type == SLICE_TYPE_B )
2120         x264_macroblock_bipred_init( h );
2121
2122     /*------------------------- Weights -------------------------------------*/
2123     x264_weighted_pred_init( h );
2124
2125     /* ------------------------ Create slice header  ----------------------- */
2126     x264_slice_init( h, i_nal_type, i_global_qp );
2127
2128     if( i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE )
2129         h->i_frame_num++;
2130
2131     /* Write frame */
2132     if( h->param.i_threads > 1 )
2133     {
2134         if( x264_pthread_create( &h->thread_handle, NULL, (void*)x264_slices_write, h ) )
2135             return -1;
2136         h->b_thread_active = 1;
2137     }
2138     else
2139         if( (intptr_t)x264_slices_write( h ) )
2140             return -1;
2141
2142     return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
2143 }
2144
2145 static int x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
2146                                    x264_nal_t **pp_nal, int *pi_nal,
2147                                    x264_picture_t *pic_out )
2148 {
2149     int i, j, i_list, frame_size;
2150     char psz_message[80];
2151
2152     if( h->b_thread_active )
2153     {
2154         void *ret = NULL;
2155         x264_pthread_join( h->thread_handle, &ret );
2156         if( (intptr_t)ret )
2157             return (intptr_t)ret;
2158         h->b_thread_active = 0;
2159     }
2160     if( !h->out.i_nal )
2161     {
2162         pic_out->i_type = X264_TYPE_AUTO;
2163         return 0;
2164     }
2165
2166     x264_frame_push_unused( thread_current, h->fenc );
2167
2168     /* End bitstream, set output  */
2169     *pi_nal = h->out.i_nal;
2170     *pp_nal = h->out.nal;
2171
2172     frame_size = x264_encoder_encapsulate_nals( h );
2173
2174     h->out.i_nal = 0;
2175
2176     /* Set output picture properties */
2177     if( h->sh.i_type == SLICE_TYPE_I )
2178         pic_out->i_type = h->i_nal_type == NAL_SLICE_IDR ? X264_TYPE_IDR : X264_TYPE_I;
2179     else if( h->sh.i_type == SLICE_TYPE_P )
2180         pic_out->i_type = X264_TYPE_P;
2181     else
2182         pic_out->i_type = X264_TYPE_B;
2183     pic_out->i_pts = h->fenc->i_pts;
2184
2185     pic_out->img.i_plane = h->fdec->i_plane;
2186     for(i = 0; i < 3; i++)
2187     {
2188         pic_out->img.i_stride[i] = h->fdec->i_stride[i];
2189         pic_out->img.plane[i] = h->fdec->plane[i];
2190     }
2191
2192     /* ---------------------- Update encoder state ------------------------- */
2193
2194     /* update rc */
2195     x264_emms();
2196     if( x264_ratecontrol_end( h, frame_size * 8 ) < 0 )
2197         return -1;
2198
2199     x264_noise_reduction_update( thread_current );
2200
2201     /* ---------------------- Compute/Print statistics --------------------- */
2202     x264_thread_sync_stat( h, h->thread[0] );
2203
2204     /* Slice stat */
2205     h->stat.i_frame_count[h->sh.i_type]++;
2206     h->stat.i_frame_size[h->sh.i_type] += frame_size;
2207     h->stat.f_frame_qp[h->sh.i_type] += h->fdec->f_qp_avg_aq;
2208
2209     for( i = 0; i < X264_MBTYPE_MAX; i++ )
2210         h->stat.i_mb_count[h->sh.i_type][i] += h->stat.frame.i_mb_count[i];
2211     for( i = 0; i < X264_PARTTYPE_MAX; i++ )
2212         h->stat.i_mb_partition[h->sh.i_type][i] += h->stat.frame.i_mb_partition[i];
2213     for( i = 0; i < 2; i++ )
2214         h->stat.i_mb_count_8x8dct[i] += h->stat.frame.i_mb_count_8x8dct[i];
2215     for( i = 0; i < 6; i++ )
2216         h->stat.i_mb_cbp[i] += h->stat.frame.i_mb_cbp[i];
2217     for( i = 0; i < 3; i++ )
2218         for( j = 0; j < 13; j++ )
2219             h->stat.i_mb_pred_mode[i][j] += h->stat.frame.i_mb_pred_mode[i][j];
2220     if( h->sh.i_type != SLICE_TYPE_I )
2221         for( i_list = 0; i_list < 2; i_list++ )
2222             for( i = 0; i < 32; i++ )
2223                 h->stat.i_mb_count_ref[h->sh.i_type][i_list][i] += h->stat.frame.i_mb_count_ref[i_list][i];
2224     if( h->sh.i_type == SLICE_TYPE_P )
2225     {
2226         h->stat.i_consecutive_bframes[h->fdec->i_frame - h->fref0[0]->i_frame - 1]++;
2227         if( h->param.analyse.i_weighted_pred == X264_WEIGHTP_SMART )
2228         {
2229             for( i = 0; i < 3; i++ )
2230                 for( j = 0; j < h->i_ref0; j++ )
2231                     if( h->sh.weight[0][i].i_denom != 0 )
2232                     {
2233                         h->stat.i_wpred[i]++;
2234                         break;
2235                     }
2236         }
2237     }
2238     if( h->sh.i_type == SLICE_TYPE_B )
2239     {
2240         h->stat.i_direct_frames[ h->sh.b_direct_spatial_mv_pred ] ++;
2241         if( h->mb.b_direct_auto_write )
2242         {
2243             //FIXME somewhat arbitrary time constants
2244             if( h->stat.i_direct_score[0] + h->stat.i_direct_score[1] > h->mb.i_mb_count )
2245             {
2246                 for( i = 0; i < 2; i++ )
2247                     h->stat.i_direct_score[i] = h->stat.i_direct_score[i] * 9/10;
2248             }
2249             for( i = 0; i < 2; i++ )
2250                 h->stat.i_direct_score[i] += h->stat.frame.i_direct_score[i];
2251         }
2252     }
2253
2254     psz_message[0] = '\0';
2255     if( h->param.analyse.b_psnr )
2256     {
2257         int64_t ssd[3] = {
2258             h->stat.frame.i_ssd[0],
2259             h->stat.frame.i_ssd[1],
2260             h->stat.frame.i_ssd[2],
2261         };
2262
2263         h->stat.i_ssd_global[h->sh.i_type] += ssd[0] + ssd[1] + ssd[2];
2264         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 );
2265         h->stat.f_psnr_mean_y[h->sh.i_type] += x264_psnr( ssd[0], h->param.i_width * h->param.i_height );
2266         h->stat.f_psnr_mean_u[h->sh.i_type] += x264_psnr( ssd[1], h->param.i_width * h->param.i_height / 4 );
2267         h->stat.f_psnr_mean_v[h->sh.i_type] += x264_psnr( ssd[2], h->param.i_width * h->param.i_height / 4 );
2268
2269         snprintf( psz_message, 80, " PSNR Y:%5.2f U:%5.2f V:%5.2f",
2270                   x264_psnr( ssd[0], h->param.i_width * h->param.i_height ),
2271                   x264_psnr( ssd[1], h->param.i_width * h->param.i_height / 4),
2272                   x264_psnr( ssd[2], h->param.i_width * h->param.i_height / 4) );
2273     }
2274
2275     if( h->param.analyse.b_ssim )
2276     {
2277         double ssim_y = h->stat.frame.f_ssim
2278                       / (((h->param.i_width-6)>>2) * ((h->param.i_height-6)>>2));
2279         h->stat.f_ssim_mean_y[h->sh.i_type] += ssim_y;
2280         snprintf( psz_message + strlen(psz_message), 80 - strlen(psz_message),
2281                   " SSIM Y:%.5f", ssim_y );
2282     }
2283     psz_message[79] = '\0';
2284
2285     x264_log( h, X264_LOG_DEBUG,
2286                   "frame=%4d QP=%.2f NAL=%d Slice:%c Poc:%-3d I:%-4d P:%-4d SKIP:%-4d size=%d bytes%s\n",
2287               h->i_frame,
2288               h->fdec->f_qp_avg_aq,
2289               h->i_nal_ref_idc,
2290               h->sh.i_type == SLICE_TYPE_I ? 'I' : (h->sh.i_type == SLICE_TYPE_P ? 'P' : 'B' ),
2291               h->fdec->i_poc,
2292               h->stat.frame.i_mb_count_i,
2293               h->stat.frame.i_mb_count_p,
2294               h->stat.frame.i_mb_count_skip,
2295               frame_size,
2296               psz_message );
2297
2298     // keep stats all in one place
2299     x264_thread_sync_stat( h->thread[0], h );
2300     // for the use of the next frame
2301     x264_thread_sync_stat( thread_current, h );
2302
2303 #ifdef DEBUG_MB_TYPE
2304 {
2305     static const char mb_chars[] = { 'i', 'i', 'I', 'C', 'P', '8', 'S',
2306         'D', '<', 'X', 'B', 'X', '>', 'B', 'B', 'B', 'B', '8', 'S' };
2307     int mb_xy;
2308     for( mb_xy = 0; mb_xy < h->sps->i_mb_width * h->sps->i_mb_height; mb_xy++ )
2309     {
2310         if( h->mb.type[mb_xy] < X264_MBTYPE_MAX && h->mb.type[mb_xy] >= 0 )
2311             fprintf( stderr, "%c ", mb_chars[ h->mb.type[mb_xy] ] );
2312         else
2313             fprintf( stderr, "? " );
2314
2315         if( (mb_xy+1) % h->sps->i_mb_width == 0 )
2316             fprintf( stderr, "\n" );
2317     }
2318 }
2319 #endif
2320
2321     /* Remove duplicates, must be done near the end as breaks h->fref0 array
2322      * by freeing some of its pointers. */
2323      for( i = 0; i < h->i_ref0; i++ )
2324          if( h->fref0[i] && h->fref0[i]->b_duplicate )
2325          {
2326              x264_frame_push_blank_unused( h, h->fref0[i] );
2327              h->fref0[i] = 0;
2328          }
2329
2330     if( h->param.psz_dump_yuv )
2331         x264_frame_dump( h );
2332
2333     return frame_size;
2334 }
2335
2336 static void x264_print_intra( int64_t *i_mb_count, double i_count, int b_print_pcm, char *intra )
2337 {
2338     intra += sprintf( intra, "I16..4%s: %4.1f%% %4.1f%% %4.1f%%",
2339         b_print_pcm ? "..PCM" : "",
2340         i_mb_count[I_16x16]/ i_count,
2341         i_mb_count[I_8x8]  / i_count,
2342         i_mb_count[I_4x4]  / i_count );
2343     if( b_print_pcm )
2344         sprintf( intra, " %4.1f%%", i_mb_count[I_PCM]  / i_count );
2345 }
2346
2347 /****************************************************************************
2348  * x264_encoder_close:
2349  ****************************************************************************/
2350 void    x264_encoder_close  ( x264_t *h )
2351 {
2352     int64_t i_yuv_size = 3 * h->param.i_width * h->param.i_height / 2;
2353     int64_t i_mb_count_size[2][7] = {{0}};
2354     char buf[200];
2355     int i, j, i_list, i_type;
2356     int b_print_pcm = h->stat.i_mb_count[SLICE_TYPE_I][I_PCM]
2357                    || h->stat.i_mb_count[SLICE_TYPE_P][I_PCM]
2358                    || h->stat.i_mb_count[SLICE_TYPE_B][I_PCM];
2359
2360     x264_lookahead_delete( h );
2361
2362     for( i=0; i<h->param.i_threads; i++ )
2363     {
2364         // don't strictly have to wait for the other threads, but it's simpler than canceling them
2365         if( h->thread[i]->b_thread_active )
2366         {
2367             x264_pthread_join( h->thread[i]->thread_handle, NULL );
2368             assert( h->thread[i]->fenc->i_reference_count == 1 );
2369             x264_frame_delete( h->thread[i]->fenc );
2370         }
2371     }
2372
2373     if( h->param.i_threads > 1 )
2374     {
2375         x264_t *thread_prev;
2376
2377         thread_prev = h->thread[h->i_thread_phase];
2378         x264_thread_sync_ratecontrol( h, thread_prev, h );
2379         x264_thread_sync_ratecontrol( thread_prev, thread_prev, h );
2380         h->i_frame = thread_prev->i_frame + 1 - h->param.i_threads;
2381     }
2382     h->i_frame++;
2383
2384     /* Slices used and PSNR */
2385     for( i=0; i<5; i++ )
2386     {
2387         static const int slice_order[] = { SLICE_TYPE_I, SLICE_TYPE_SI, SLICE_TYPE_P, SLICE_TYPE_SP, SLICE_TYPE_B };
2388         static const char *slice_name[] = { "P", "B", "I", "SP", "SI" };
2389         int i_slice = slice_order[i];
2390
2391         if( h->stat.i_frame_count[i_slice] > 0 )
2392         {
2393             const int i_count = h->stat.i_frame_count[i_slice];
2394             if( h->param.analyse.b_psnr )
2395             {
2396                 x264_log( h, X264_LOG_INFO,
2397                           "frame %s:%-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",
2398                           slice_name[i_slice],
2399                           i_count,
2400                           h->stat.f_frame_qp[i_slice] / i_count,
2401                           (double)h->stat.i_frame_size[i_slice] / i_count,
2402                           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,
2403                           h->stat.f_psnr_average[i_slice] / i_count,
2404                           x264_psnr( h->stat.i_ssd_global[i_slice], i_count * i_yuv_size ) );
2405             }
2406             else
2407             {
2408                 x264_log( h, X264_LOG_INFO,
2409                           "frame %s:%-5d Avg QP:%5.2f  size:%6.0f\n",
2410                           slice_name[i_slice],
2411                           i_count,
2412                           h->stat.f_frame_qp[i_slice] / i_count,
2413                           (double)h->stat.i_frame_size[i_slice] / i_count );
2414             }
2415         }
2416     }
2417     if( h->param.i_bframe && h->stat.i_frame_count[SLICE_TYPE_P] )
2418     {
2419         char *p = buf;
2420         int den = 0;
2421         // weight by number of frames (including the P-frame) that are in a sequence of N B-frames
2422         for( i=0; i<=h->param.i_bframe; i++ )
2423             den += (i+1) * h->stat.i_consecutive_bframes[i];
2424         for( i=0; i<=h->param.i_bframe; i++ )
2425             p += sprintf( p, " %4.1f%%", 100. * (i+1) * h->stat.i_consecutive_bframes[i] / den );
2426         x264_log( h, X264_LOG_INFO, "consecutive B-frames:%s\n", buf );
2427     }
2428
2429     for( i_type = 0; i_type < 2; i_type++ )
2430         for( i = 0; i < X264_PARTTYPE_MAX; i++ )
2431         {
2432             if( i == D_DIRECT_8x8 ) continue; /* direct is counted as its own type */
2433             i_mb_count_size[i_type][x264_mb_partition_pixel_table[i]] += h->stat.i_mb_partition[i_type][i];
2434         }
2435
2436     /* MB types used */
2437     if( h->stat.i_frame_count[SLICE_TYPE_I] > 0 )
2438     {
2439         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_I];
2440         double i_count = h->stat.i_frame_count[SLICE_TYPE_I] * h->mb.i_mb_count / 100.0;
2441         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
2442         x264_log( h, X264_LOG_INFO, "mb I  %s\n", buf );
2443     }
2444     if( h->stat.i_frame_count[SLICE_TYPE_P] > 0 )
2445     {
2446         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_P];
2447         double i_count = h->stat.i_frame_count[SLICE_TYPE_P] * h->mb.i_mb_count / 100.0;
2448         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_P];
2449         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
2450         x264_log( h, X264_LOG_INFO,
2451                   "mb P  %s  P16..4: %4.1f%% %4.1f%% %4.1f%% %4.1f%% %4.1f%%    skip:%4.1f%%\n",
2452                   buf,
2453                   i_mb_size[PIXEL_16x16] / (i_count*4),
2454                   (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
2455                   i_mb_size[PIXEL_8x8] / (i_count*4),
2456                   (i_mb_size[PIXEL_8x4] + i_mb_size[PIXEL_4x8]) / (i_count*4),
2457                   i_mb_size[PIXEL_4x4] / (i_count*4),
2458                   i_mb_count[P_SKIP] / i_count );
2459     }
2460     if( h->stat.i_frame_count[SLICE_TYPE_B] > 0 )
2461     {
2462         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_B];
2463         double i_count = h->stat.i_frame_count[SLICE_TYPE_B] * h->mb.i_mb_count / 100.0;
2464         double i_mb_list_count;
2465         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_B];
2466         int64_t list_count[3] = {0}; /* 0 == L0, 1 == L1, 2 == BI */
2467         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
2468         for( i = 0; i < X264_PARTTYPE_MAX; i++ )
2469             for( j = 0; j < 2; j++ )
2470             {
2471                 int l0 = x264_mb_type_list_table[i][0][j];
2472                 int l1 = x264_mb_type_list_table[i][1][j];
2473                 if( l0 || l1 )
2474                     list_count[l1+l0*l1] += h->stat.i_mb_count[SLICE_TYPE_B][i] * 2;
2475             }
2476         list_count[0] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L0_8x8];
2477         list_count[1] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L1_8x8];
2478         list_count[2] += h->stat.i_mb_partition[SLICE_TYPE_B][D_BI_8x8];
2479         i_mb_count[B_DIRECT] += (h->stat.i_mb_partition[SLICE_TYPE_B][D_DIRECT_8x8]+2)/4;
2480         i_mb_list_count = (list_count[0] + list_count[1] + list_count[2]) / 100.0;
2481         x264_log( h, X264_LOG_INFO,
2482                   "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",
2483                   buf,
2484                   i_mb_size[PIXEL_16x16] / (i_count*4),
2485                   (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
2486                   i_mb_size[PIXEL_8x8] / (i_count*4),
2487                   i_mb_count[B_DIRECT] / i_count,
2488                   i_mb_count[B_SKIP]   / i_count,
2489                   list_count[0] / i_mb_list_count,
2490                   list_count[1] / i_mb_list_count,
2491                   list_count[2] / i_mb_list_count );
2492     }
2493
2494     x264_ratecontrol_summary( h );
2495
2496     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 )
2497     {
2498 #define SUM3(p) (p[SLICE_TYPE_I] + p[SLICE_TYPE_P] + p[SLICE_TYPE_B])
2499 #define SUM3b(p,o) (p[SLICE_TYPE_I][o] + p[SLICE_TYPE_P][o] + p[SLICE_TYPE_B][o])
2500         int64_t i_i8x8 = SUM3b( h->stat.i_mb_count, I_8x8 );
2501         int64_t i_intra = i_i8x8 + SUM3b( h->stat.i_mb_count, I_4x4 )
2502                                  + SUM3b( h->stat.i_mb_count, I_16x16 );
2503         int64_t i_all_intra = i_intra + SUM3b( h->stat.i_mb_count, I_PCM);
2504         const int i_count = h->stat.i_frame_count[SLICE_TYPE_I] +
2505                             h->stat.i_frame_count[SLICE_TYPE_P] +
2506                             h->stat.i_frame_count[SLICE_TYPE_B];
2507         int64_t i_mb_count = i_count * h->mb.i_mb_count;
2508         float fps = (float) h->param.i_fps_num / h->param.i_fps_den;
2509         float f_bitrate = fps * SUM3(h->stat.i_frame_size) / i_count / 125;
2510
2511         if( h->pps->b_transform_8x8_mode )
2512         {
2513             buf[0] = 0;
2514             if( h->stat.i_mb_count_8x8dct[0] )
2515                 sprintf( buf, " inter:%.1f%%", 100. * h->stat.i_mb_count_8x8dct[1] / h->stat.i_mb_count_8x8dct[0] );
2516             x264_log( h, X264_LOG_INFO, "8x8 transform intra:%.1f%%%s\n", 100. * i_i8x8 / i_intra, buf );
2517         }
2518
2519         if( h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO
2520             && h->stat.i_frame_count[SLICE_TYPE_B] )
2521         {
2522             x264_log( h, X264_LOG_INFO, "direct mvs  spatial:%.1f%% temporal:%.1f%%\n",
2523                       h->stat.i_direct_frames[1] * 100. / h->stat.i_frame_count[SLICE_TYPE_B],
2524                       h->stat.i_direct_frames[0] * 100. / h->stat.i_frame_count[SLICE_TYPE_B] );
2525         }
2526
2527         buf[0] = 0;
2528         if( i_mb_count != i_all_intra )
2529             sprintf( buf, " inter: %.1f%% %.1f%% %.1f%%",
2530                      h->stat.i_mb_cbp[1] * 100.0 / ((i_mb_count - i_all_intra)*4),
2531                      h->stat.i_mb_cbp[3] * 100.0 / ((i_mb_count - i_all_intra)  ),
2532                      h->stat.i_mb_cbp[5] * 100.0 / ((i_mb_count - i_all_intra)) );
2533         x264_log( h, X264_LOG_INFO, "coded y,uvDC,uvAC intra: %.1f%% %.1f%% %.1f%%%s\n",
2534                   h->stat.i_mb_cbp[0] * 100.0 / (i_all_intra*4),
2535                   h->stat.i_mb_cbp[2] * 100.0 / (i_all_intra  ),
2536                   h->stat.i_mb_cbp[4] * 100.0 / (i_all_intra  ), buf );
2537
2538         int64_t fixed_pred_modes[3][9] = {{0}};
2539         int64_t sum_pred_modes[3] = {0};
2540         for( i = 0; i <= I_PRED_16x16_DC_128; i++ )
2541         {
2542             fixed_pred_modes[0][x264_mb_pred_mode16x16_fix[i]] += h->stat.i_mb_pred_mode[0][i];
2543             sum_pred_modes[0] += h->stat.i_mb_pred_mode[0][i];
2544         }
2545         if( sum_pred_modes[0] )
2546             x264_log( h, X264_LOG_INFO, "i16 v,h,dc,p: %2.0f%% %2.0f%% %2.0f%% %2.0f%%\n",
2547                       fixed_pred_modes[0][0] * 100.0 / sum_pred_modes[0],
2548                       fixed_pred_modes[0][1] * 100.0 / sum_pred_modes[0],
2549                       fixed_pred_modes[0][2] * 100.0 / sum_pred_modes[0],
2550                       fixed_pred_modes[0][3] * 100.0 / sum_pred_modes[0] );
2551         for( i = 1; i <= 2; i++ )
2552         {
2553             for( j = 0; j <= I_PRED_8x8_DC_128; j++ )
2554             {
2555                 fixed_pred_modes[i][x264_mb_pred_mode4x4_fix(j)] += h->stat.i_mb_pred_mode[i][j];
2556                 sum_pred_modes[i] += h->stat.i_mb_pred_mode[i][j];
2557             }
2558             if( sum_pred_modes[i] )
2559                 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,
2560                           fixed_pred_modes[i][0] * 100.0 / sum_pred_modes[i],
2561                           fixed_pred_modes[i][1] * 100.0 / sum_pred_modes[i],
2562                           fixed_pred_modes[i][2] * 100.0 / sum_pred_modes[i],
2563                           fixed_pred_modes[i][3] * 100.0 / sum_pred_modes[i],
2564                           fixed_pred_modes[i][4] * 100.0 / sum_pred_modes[i],
2565                           fixed_pred_modes[i][5] * 100.0 / sum_pred_modes[i],
2566                           fixed_pred_modes[i][6] * 100.0 / sum_pred_modes[i],
2567                           fixed_pred_modes[i][7] * 100.0 / sum_pred_modes[i],
2568                           fixed_pred_modes[i][8] * 100.0 / sum_pred_modes[i] );
2569         }
2570
2571         if( h->param.analyse.i_weighted_pred == X264_WEIGHTP_SMART && h->stat.i_frame_count[SLICE_TYPE_P] > 0 )
2572             x264_log( h, X264_LOG_INFO, "Weighted P-Frames: Y:%.1f%%\n",
2573                       h->stat.i_wpred[0] * 100.0 / h->stat.i_frame_count[SLICE_TYPE_P] );
2574
2575         for( i_list = 0; i_list < 2; i_list++ )
2576         {
2577             int i_slice;
2578             for( i_slice = 0; i_slice < 2; i_slice++ )
2579             {
2580                 char *p = buf;
2581                 int64_t i_den = 0;
2582                 int i_max = 0;
2583                 for( i = 0; i < 32; i++ )
2584                     if( h->stat.i_mb_count_ref[i_slice][i_list][i] )
2585                     {
2586                         i_den += h->stat.i_mb_count_ref[i_slice][i_list][i];
2587                         i_max = i;
2588                     }
2589                 if( i_max == 0 )
2590                     continue;
2591                 for( i = 0; i <= i_max; i++ )
2592                     p += sprintf( p, " %4.1f%%", 100. * h->stat.i_mb_count_ref[i_slice][i_list][i] / i_den );
2593                 x264_log( h, X264_LOG_INFO, "ref %c L%d:%s\n", "PB"[i_slice], i_list, buf );
2594             }
2595         }
2596
2597         if( h->param.analyse.b_ssim )
2598         {
2599             x264_log( h, X264_LOG_INFO,
2600                       "SSIM Mean Y:%.7f\n",
2601                       SUM3( h->stat.f_ssim_mean_y ) / i_count );
2602         }
2603         if( h->param.analyse.b_psnr )
2604         {
2605             x264_log( h, X264_LOG_INFO,
2606                       "PSNR Mean Y:%6.3f U:%6.3f V:%6.3f Avg:%6.3f Global:%6.3f kb/s:%.2f\n",
2607                       SUM3( h->stat.f_psnr_mean_y ) / i_count,
2608                       SUM3( h->stat.f_psnr_mean_u ) / i_count,
2609                       SUM3( h->stat.f_psnr_mean_v ) / i_count,
2610                       SUM3( h->stat.f_psnr_average ) / i_count,
2611                       x264_psnr( SUM3( h->stat.i_ssd_global ), i_count * i_yuv_size ),
2612                       f_bitrate );
2613         }
2614         else
2615             x264_log( h, X264_LOG_INFO, "kb/s:%.2f\n", f_bitrate );
2616     }
2617
2618     /* rc */
2619     x264_ratecontrol_delete( h );
2620
2621     /* param */
2622     if( h->param.rc.psz_stat_out )
2623         free( h->param.rc.psz_stat_out );
2624     if( h->param.rc.psz_stat_in )
2625         free( h->param.rc.psz_stat_in );
2626
2627     x264_cqm_delete( h );
2628     x264_free( h->nal_buffer );
2629     x264_analyse_free_costs( h );
2630
2631     if( h->param.i_threads > 1)
2632         h = h->thread[h->i_thread_phase];
2633
2634     /* frames */
2635     x264_frame_delete_list( h->frames.unused[0] );
2636     x264_frame_delete_list( h->frames.unused[1] );
2637     x264_frame_delete_list( h->frames.current );
2638     x264_frame_delete_list( h->frames.blank_unused );
2639
2640     h = h->thread[0];
2641
2642     for( i = h->param.i_threads - 1; i >= 0; i-- )
2643     {
2644         x264_frame_t **frame;
2645
2646         for( frame = h->thread[i]->frames.reference; *frame; frame++ )
2647         {
2648             assert( (*frame)->i_reference_count > 0 );
2649             (*frame)->i_reference_count--;
2650             if( (*frame)->i_reference_count == 0 )
2651                 x264_frame_delete( *frame );
2652         }
2653         frame = &h->thread[i]->fdec;
2654         assert( (*frame)->i_reference_count > 0 );
2655         (*frame)->i_reference_count--;
2656         if( (*frame)->i_reference_count == 0 )
2657             x264_frame_delete( *frame );
2658
2659         x264_macroblock_cache_end( h->thread[i] );
2660         x264_free( h->thread[i]->out.p_bitstream );
2661         x264_free( h->thread[i]->out.nal);
2662         x264_free( h->thread[i] );
2663     }
2664 }
2665
2666 /****************************************************************************
2667  * x264_encoder_delayed_frames:
2668  ****************************************************************************/
2669 int x264_encoder_delayed_frames( x264_t *h )
2670 {
2671     int delayed_frames = 0;
2672     int i;
2673     for( i=0; i<h->param.i_threads; i++ )
2674         delayed_frames += h->thread[i]->b_thread_active;
2675     h = h->thread[h->i_thread_phase];
2676     for( i=0; h->frames.current[i]; i++ )
2677         delayed_frames++;
2678     x264_pthread_mutex_lock( &h->lookahead->ofbuf.mutex );
2679     x264_pthread_mutex_lock( &h->lookahead->ifbuf.mutex );
2680     x264_pthread_mutex_lock( &h->lookahead->next.mutex );
2681     delayed_frames += h->lookahead->ifbuf.i_size + h->lookahead->next.i_size + h->lookahead->ofbuf.i_size;
2682     x264_pthread_mutex_unlock( &h->lookahead->next.mutex );
2683     x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
2684     x264_pthread_mutex_unlock( &h->lookahead->ofbuf.mutex );
2685     return delayed_frames;
2686 }