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 void 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, 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->sps->i_mb_width * h->sps->i_mb_height;
  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 );
 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 );
 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         ( sh->pps->b_weighted_bipred == 1 && sh->i_type == SLICE_TYPE_B ) )
 262     {
 263         /* FIXME */
 264     }
 265
 266     if( i_nal_ref_idc != 0 )
 267     {
 268         if( sh->i_idr_pic_id >= 0 )
 269         {
 270             bs_write1( s, 0 );  /* no output of prior pics flag */
 271             bs_write1( s, 0 );  /* long term reference flag */
 272         }
 273         else
 274         {
 275             bs_write1( s, 0 );  /* adaptive_ref_pic_marking_mode_flag */
 276         }
 277     }
 278
 279     if( sh->pps->b_cabac && sh->i_type != SLICE_TYPE_I )
 280     {
 281         bs_write_ue( s, sh->i_cabac_init_idc );
 282     }
 283     bs_write_se( s, sh->i_qp_delta );      /* slice qp delta */
 284
 285     if( sh->pps->b_deblocking_filter_control )
 286     {
 287         bs_write_ue( s, sh->i_disable_deblocking_filter_idc );
 288         if( sh->i_disable_deblocking_filter_idc != 1 )
 289         {
 290             bs_write_se( s, sh->i_alpha_c0_offset >> 1 );
 291             bs_write_se( s, sh->i_beta_offset >> 1 );
 292         }
 293     }
 294 }
 295
 296 /* If we are within a reasonable distance of the end of the memory allocated for the bitstream, */
 297 /* reallocate, adding an arbitrary amount of space (100 kilobytes). */
 298 static void x264_bitstream_check_buffer( x264_t *h )
 299 {
 300     if( ( h->param.b_cabac && (h->cabac.p_end - h->cabac.p < 2500) )
 301      || ( h->out.bs.p_end - h->out.bs.p < 2500 ) )
 302     {
 303         uint8_t *bs_bak = h->out.p_bitstream;
 304         intptr_t delta;
 305         int i;
 306
 307         h->out.i_bitstream += 100000;
 308         h->out.p_bitstream = x264_realloc( h->out.p_bitstream, h->out.i_bitstream );
 309         delta = h->out.p_bitstream - bs_bak;
 310
 311         h->out.bs.p_start += delta;
 312         h->out.bs.p += delta;
 313         h->out.bs.p_end = h->out.p_bitstream + h->out.i_bitstream;
 314
 315         h->cabac.p_start += delta;
 316         h->cabac.p += delta;
 317         h->cabac.p_end = h->out.p_bitstream + h->out.i_bitstream;
 318
 319         for( i = 0; i <= h->out.i_nal; i++ )
 320             h->out.nal[i].p_payload += delta;
 321     }
 322 }
 323
 324 /****************************************************************************
 325  *
 326  ****************************************************************************
 327  ****************************** External API*********************************
 328  ****************************************************************************
 329  *
 330  ****************************************************************************/
 331
 332 static int x264_validate_parameters( x264_t *h )
 333 {
 334 #ifdef HAVE_MMX
 335     if( !(x264_cpu_detect() & X264_CPU_MMXEXT) )
 336     {
 337         x264_log( h, X264_LOG_ERROR, "your cpu does not support MMXEXT, but x264 was compiled with asm support\n");
 338         x264_log( h, X264_LOG_ERROR, "to run x264, recompile without asm support (configure --disable-asm)\n");
 339         return -1;
 340     }
 341 #endif
 342     if( h->param.i_width <= 0 || h->param.i_height <= 0 )
 343     {
 344         x264_log( h, X264_LOG_ERROR, "invalid width x height (%dx%d)\n",
 345                   h->param.i_width, h->param.i_height );
 346         return -1;
 347     }
 348
 349     if( h->param.i_width % 2 || h->param.i_height % 2 )
 350     {
 351         x264_log( h, X264_LOG_ERROR, "width or height not divisible by 2 (%dx%d)\n",
 352                   h->param.i_width, h->param.i_height );
 353         return -1;
 354     }
 355     if( h->param.i_csp != X264_CSP_I420 )
 356     {
 357         x264_log( h, X264_LOG_ERROR, "invalid CSP (only I420 supported)\n" );
 358         return -1;
 359     }
 360
 361     if( h->param.i_threads == 0 )
 362         h->param.i_threads = x264_cpu_num_processors() * 3/2;
 363     h->param.i_threads = x264_clip3( h->param.i_threads, 1, X264_THREAD_MAX );
 364     if( h->param.i_threads > 1 )
 365     {
 366 #ifndef HAVE_PTHREAD
 367         x264_log( h, X264_LOG_WARNING, "not compiled with pthread support!\n");
 368         h->param.i_threads = 1;
 369 #endif
 370     }
 371
 372     if( h->param.b_interlaced )
 373     {
 374         if( h->param.analyse.i_me_method >= X264_ME_ESA )
 375         {
 376             x264_log( h, X264_LOG_WARNING, "interlace + me=esa is not implemented\n" );
 377             h->param.analyse.i_me_method = X264_ME_UMH;
 378         }
 379         if( h->param.analyse.i_direct_mv_pred > X264_DIRECT_PRED_SPATIAL )
 380         {
 381             x264_log( h, X264_LOG_WARNING, "interlace + direct=temporal is not implemented\n" );
 382             h->param.analyse.i_direct_mv_pred = X264_DIRECT_PRED_SPATIAL;
 383         }
 384     }
 385
 386     /* Detect default ffmpeg settings and terminate with an error. */
 387     {
 388         int score = 0;
 389         score += h->param.analyse.i_me_range == 0;
 390         score += h->param.rc.i_qp_step == 3;
 391         score += h->param.i_keyint_max == 12;
 392         score += h->param.rc.i_qp_min == 2;
 393         score += h->param.rc.i_qp_max == 31;
 394         score += h->param.rc.f_qcompress == 0.5;
 395         score += fabs(h->param.rc.f_ip_factor - 1.25) < 0.01;
 396         score += fabs(h->param.rc.f_pb_factor - 1.25) < 0.01;
 397         score += h->param.analyse.inter == 0 && h->param.analyse.i_subpel_refine == 8;
 398         if( score >= 5 )
 399         {
 400             x264_log( h, X264_LOG_ERROR, "broken ffmpeg default settings detected\n" );
 401             x264_log( h, X264_LOG_ERROR, "use an encoding preset (vpre)\n" );
 402             return -1;
 403         }
 404     }
 405
 406     if( h->param.rc.i_rc_method < 0 || h->param.rc.i_rc_method > 2 )
 407     {
 408         x264_log( h, X264_LOG_ERROR, "no ratecontrol method specified\n" );
 409         return -1;
 410     }
 411     h->param.rc.f_rf_constant = x264_clip3f( h->param.rc.f_rf_constant, 0, 51 );
 412     h->param.rc.i_qp_constant = x264_clip3( h->param.rc.i_qp_constant, 0, 51 );
 413     if( h->param.rc.i_rc_method == X264_RC_CRF )
 414         h->param.rc.i_qp_constant = h->param.rc.f_rf_constant;
 415     if( (h->param.rc.i_rc_method == X264_RC_CQP || h->param.rc.i_rc_method == X264_RC_CRF)
 416         && h->param.rc.i_qp_constant == 0 )
 417     {
 418         h->mb.b_lossless = 1;
 419         h->param.i_cqm_preset = X264_CQM_FLAT;
 420         h->param.psz_cqm_file = NULL;
 421         h->param.rc.i_rc_method = X264_RC_CQP;
 422         h->param.rc.f_ip_factor = 1;
 423         h->param.rc.f_pb_factor = 1;
 424         h->param.analyse.b_psnr = 0;
 425         h->param.analyse.b_ssim = 0;
 426         h->param.analyse.i_chroma_qp_offset = 0;
 427         h->param.analyse.i_trellis = 0;
 428         h->param.analyse.b_fast_pskip = 0;
 429         h->param.analyse.i_noise_reduction = 0;
 430         h->param.analyse.f_psy_rd = 0;
 431         h->param.i_bframe = 0;
 432         /* 8x8dct is not useful at all in CAVLC lossless */
 433         if( !h->param.b_cabac )
 434             h->param.analyse.b_transform_8x8 = 0;
 435     }
 436     if( h->param.rc.i_rc_method == X264_RC_CQP )
 437     {
 438         float qp_p = h->param.rc.i_qp_constant;
 439         float qp_i = qp_p - 6*log(h->param.rc.f_ip_factor)/log(2);
 440         float qp_b = qp_p + 6*log(h->param.rc.f_pb_factor)/log(2);
 441         h->param.rc.i_qp_min = x264_clip3( (int)(X264_MIN3( qp_p, qp_i, qp_b )), 0, 51 );
 442         h->param.rc.i_qp_max = x264_clip3( (int)(X264_MAX3( qp_p, qp_i, qp_b ) + .999), 0, 51 );
 443         h->param.rc.i_aq_mode = 0;
 444     }
 445     h->param.rc.i_qp_max = x264_clip3( h->param.rc.i_qp_max, 0, 51 );
 446     h->param.rc.i_qp_min = x264_clip3( h->param.rc.i_qp_min, 0, h->param.rc.i_qp_max );
 447
 448     if( ( h->param.i_width % 16 || h->param.i_height % 16 )
 449         && h->param.i_height != 1080 && !h->mb.b_lossless )
 450     {
 451         // There's nothing special about 1080 in that the warning still applies to it,
 452         // but chances are the user can't help it if his content is already 1080p,
 453         // so there's no point in warning in that case.
 454         x264_log( h, X264_LOG_WARNING,
 455                   "width or height not divisible by 16 (%dx%d), compression will suffer.\n",
 456                   h->param.i_width, h->param.i_height );
 457     }
 458
 459     h->param.i_frame_reference = x264_clip3( h->param.i_frame_reference, 1, 16 );
 460     if( h->param.i_keyint_max <= 0 )
 461         h->param.i_keyint_max = 1;
 462     if( h->param.i_scenecut_threshold < 0 )
 463         h->param.i_scenecut_threshold = 0;
 464     h->param.i_keyint_min = x264_clip3( h->param.i_keyint_min, 1, h->param.i_keyint_max/2+1 );
 465     if( !h->param.analyse.i_subpel_refine && h->param.analyse.i_direct_mv_pred > X264_DIRECT_PRED_SPATIAL )
 466     {
 467         x264_log( h, X264_LOG_WARNING, "subme=0 + direct=temporal is not supported\n" );
 468         h->param.analyse.i_direct_mv_pred = X264_DIRECT_PRED_SPATIAL;
 469     }
 470     h->param.i_bframe = x264_clip3( h->param.i_bframe, 0, X264_BFRAME_MAX );
 471     h->param.i_bframe_bias = x264_clip3( h->param.i_bframe_bias, -90, 100 );
 472     h->param.b_bframe_pyramid = h->param.b_bframe_pyramid && h->param.i_bframe > 1;
 473     if( !h->param.i_bframe )
 474         h->param.i_bframe_adaptive = X264_B_ADAPT_NONE;
 475     h->param.analyse.b_weighted_bipred = h->param.analyse.b_weighted_bipred && h->param.i_bframe > 0;
 476     h->mb.b_direct_auto_write = h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO
 477                                 && h->param.i_bframe
 478                                 && ( h->param.rc.b_stat_write || !h->param.rc.b_stat_read );
 479
 480     h->param.i_deblocking_filter_alphac0 = x264_clip3( h->param.i_deblocking_filter_alphac0, -6, 6 );
 481     h->param.i_deblocking_filter_beta    = x264_clip3( h->param.i_deblocking_filter_beta, -6, 6 );
 482     h->param.analyse.i_luma_deadzone[0] = x264_clip3( h->param.analyse.i_luma_deadzone[0], 0, 32 );
 483     h->param.analyse.i_luma_deadzone[1] = x264_clip3( h->param.analyse.i_luma_deadzone[1], 0, 32 );
 484
 485     h->param.i_cabac_init_idc = x264_clip3( h->param.i_cabac_init_idc, 0, 2 );
 486
 487     if( h->param.i_cqm_preset < X264_CQM_FLAT || h->param.i_cqm_preset > X264_CQM_CUSTOM )
 488         h->param.i_cqm_preset = X264_CQM_FLAT;
 489
 490     if( h->param.analyse.i_me_method < X264_ME_DIA ||
 491         h->param.analyse.i_me_method > X264_ME_TESA )
 492         h->param.analyse.i_me_method = X264_ME_HEX;
 493     if( h->param.analyse.i_me_range < 4 )
 494         h->param.analyse.i_me_range = 4;
 495     if( h->param.analyse.i_me_range > 16 && h->param.analyse.i_me_method <= X264_ME_HEX )
 496         h->param.analyse.i_me_range = 16;
 497     if( h->param.analyse.i_me_method == X264_ME_TESA &&
 498         (h->mb.b_lossless || h->param.analyse.i_subpel_refine <= 1) )
 499         h->param.analyse.i_me_method = X264_ME_ESA;
 500     h->param.analyse.i_subpel_refine = x264_clip3( h->param.analyse.i_subpel_refine, 0, 10 );
 501     h->param.analyse.b_mixed_references = h->param.analyse.b_mixed_references && h->param.i_frame_reference > 1;
 502     h->param.analyse.inter &= X264_ANALYSE_PSUB16x16|X264_ANALYSE_PSUB8x8|X264_ANALYSE_BSUB16x16|
 503                               X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
 504     h->param.analyse.intra &= X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
 505     if( !(h->param.analyse.inter & X264_ANALYSE_PSUB16x16) )
 506         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
 507     if( !h->param.analyse.b_transform_8x8 )
 508     {
 509         h->param.analyse.inter &= ~X264_ANALYSE_I8x8;
 510         h->param.analyse.intra &= ~X264_ANALYSE_I8x8;
 511     }
 512     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -12, 12);
 513     if( !h->param.b_cabac )
 514         h->param.analyse.i_trellis = 0;
 515     h->param.analyse.i_trellis = x264_clip3( h->param.analyse.i_trellis, 0, 2 );
 516     if( !h->param.analyse.i_trellis )
 517         h->param.analyse.f_psy_trellis = 0;
 518     h->param.analyse.f_psy_rd = x264_clip3f( h->param.analyse.f_psy_rd, 0, 10 );
 519     h->param.analyse.f_psy_trellis = x264_clip3f( h->param.analyse.f_psy_trellis, 0, 10 );
 520     if( h->param.analyse.i_subpel_refine < 6 )
 521         h->param.analyse.f_psy_rd = 0;
 522     h->mb.i_psy_rd = FIX8( h->param.analyse.f_psy_rd );
 523     /* Psy RDO increases overall quantizers to improve the quality of luma--this indirectly hurts chroma quality */
 524     /* so we lower the chroma QP offset to compensate */
 525     /* This can be triggered repeatedly on multiple calls to parameter_validate, but since encoding
 526      * uses the pps chroma qp offset not the param chroma qp offset, this is not a problem. */
 527     if( h->mb.i_psy_rd )
 528         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_rd < 0.25 ? 1 : 2;
 529     h->mb.i_psy_trellis = FIX8( h->param.analyse.f_psy_trellis / 4 );
 530     /* Psy trellis has a similar effect. */
 531     if( h->mb.i_psy_trellis )
 532         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_trellis < 0.25 ? 1 : 2;
 533     else
 534         h->mb.i_psy_trellis = 0;
 535     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -12, 12);
 536     h->param.rc.i_aq_mode = x264_clip3( h->param.rc.i_aq_mode, 0, 2 );
 537     h->param.rc.f_aq_strength = x264_clip3f( h->param.rc.f_aq_strength, 0, 3 );
 538     if( h->param.rc.f_aq_strength == 0 )
 539         h->param.rc.i_aq_mode = 0;
 540     h->param.analyse.i_noise_reduction = x264_clip3( h->param.analyse.i_noise_reduction, 0, 1<<16 );
 541     if( h->param.analyse.i_subpel_refine == 10 && (h->param.analyse.i_trellis != 2 || !h->param.rc.i_aq_mode) )
 542         h->param.analyse.i_subpel_refine = 9;
 543
 544     {
 545         const x264_level_t *l = x264_levels;
 546         if( h->param.i_level_idc < 0 )
 547         {
 548             int maxrate_bak = h->param.rc.i_vbv_max_bitrate;
 549             if( h->param.rc.i_rc_method == X264_RC_ABR && h->param.rc.i_vbv_buffer_size <= 0 )
 550                 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate * 2;
 551             h->sps = h->sps_array;
 552             x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
 553             do h->param.i_level_idc = l->level_idc;
 554                 while( l[1].level_idc && x264_validate_levels( h, 0 ) && l++ );
 555             h->param.rc.i_vbv_max_bitrate = maxrate_bak;
 556         }
 557         else
 558         {
 559             while( l->level_idc && l->level_idc != h->param.i_level_idc )
 560                 l++;
 561             if( l->level_idc == 0 )
 562             {
 563                 x264_log( h, X264_LOG_ERROR, "invalid level_idc: %d\n", h->param.i_level_idc );
 564                 return -1;
 565             }
 566         }
 567         if( h->param.analyse.i_mv_range <= 0 )
 568             h->param.analyse.i_mv_range = l->mv_range >> h->param.b_interlaced;
 569         else
 570             h->param.analyse.i_mv_range = x264_clip3(h->param.analyse.i_mv_range, 32, 512 >> h->param.b_interlaced);
 571     }
 572
 573     if( h->param.i_threads > 1 )
 574     {
 575         int r = h->param.analyse.i_mv_range_thread;
 576         int r2;
 577         if( r <= 0 )
 578         {
 579             // half of the available space is reserved and divided evenly among the threads,
 580             // the rest is allocated to whichever thread is far enough ahead to use it.
 581             // reserving more space increases quality for some videos, but costs more time
 582             // in thread synchronization.
 583             int max_range = (h->param.i_height + X264_THREAD_HEIGHT) / h->param.i_threads - X264_THREAD_HEIGHT;
 584             r = max_range / 2;
 585         }
 586         r = X264_MAX( r, h->param.analyse.i_me_range );
 587         r = X264_MIN( r, h->param.analyse.i_mv_range );
 588         // round up to use the whole mb row
 589         r2 = (r & ~15) + ((-X264_THREAD_HEIGHT) & 15);
 590         if( r2 < r )
 591             r2 += 16;
 592         x264_log( h, X264_LOG_DEBUG, "using mv_range_thread = %d\n", r2 );
 593         h->param.analyse.i_mv_range_thread = r2;
 594     }
 595
 596     if( h->param.rc.f_qblur < 0 )
 597         h->param.rc.f_qblur = 0;
 598     if( h->param.rc.f_complexity_blur < 0 )
 599         h->param.rc.f_complexity_blur = 0;
 600
 601     h->param.i_sps_id &= 31;
 602
 603     if( h->param.i_log_level < X264_LOG_INFO )
 604     {
 605         h->param.analyse.b_psnr = 0;
 606         h->param.analyse.b_ssim = 0;
 607     }
 608
 609     /* ensure the booleans are 0 or 1 so they can be used in math */
 610 #define BOOLIFY(x) h->param.x = !!h->param.x
 611     BOOLIFY( b_cabac );
 612     BOOLIFY( b_deblocking_filter );
 613     BOOLIFY( b_interlaced );
 614     BOOLIFY( analyse.b_transform_8x8 );
 615     BOOLIFY( analyse.b_chroma_me );
 616     BOOLIFY( analyse.b_fast_pskip );
 617     BOOLIFY( rc.b_stat_write );
 618     BOOLIFY( rc.b_stat_read );
 619 #undef BOOLIFY
 620
 621     return 0;
 622 }
 623
 624 static void mbcmp_init( x264_t *h )
 625 {
 626     int satd = !h->mb.b_lossless && h->param.analyse.i_subpel_refine > 1;
 627     memcpy( h->pixf.mbcmp, satd ? h->pixf.satd : h->pixf.sad_aligned, sizeof(h->pixf.mbcmp) );
 628     memcpy( h->pixf.mbcmp_unaligned, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.mbcmp_unaligned) );
 629     h->pixf.intra_mbcmp_x3_16x16 = satd ? h->pixf.intra_satd_x3_16x16 : h->pixf.intra_sad_x3_16x16;
 630     h->pixf.intra_mbcmp_x3_8x8c = satd ? h->pixf.intra_satd_x3_8x8c : h->pixf.intra_sad_x3_8x8c;
 631     h->pixf.intra_mbcmp_x3_4x4 = satd ? h->pixf.intra_satd_x3_4x4 : h->pixf.intra_sad_x3_4x4;
 632     satd &= h->param.analyse.i_me_method == X264_ME_TESA;
 633     memcpy( h->pixf.fpelcmp, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.fpelcmp) );
 634     memcpy( h->pixf.fpelcmp_x3, satd ? h->pixf.satd_x3 : h->pixf.sad_x3, sizeof(h->pixf.fpelcmp_x3) );
 635     memcpy( h->pixf.fpelcmp_x4, satd ? h->pixf.satd_x4 : h->pixf.sad_x4, sizeof(h->pixf.fpelcmp_x4) );
 636 }
 637
 638 /****************************************************************************
 639  * x264_encoder_open:
 640  ****************************************************************************/
 641 x264_t *x264_encoder_open   ( x264_param_t *param )
 642 {
 643     x264_t *h = x264_malloc( sizeof( x264_t ) );
 644     char buf[1000], *p;
 645     int i;
 646
 647     memset( h, 0, sizeof( x264_t ) );
 648
 649     /* Create a copy of param */
 650     memcpy( &h->param, param, sizeof( x264_param_t ) );
 651
 652     if( x264_validate_parameters( h ) < 0 )
 653     {
 654         x264_free( h );
 655         return NULL;
 656     }
 657
 658     if( h->param.psz_cqm_file )
 659         if( x264_cqm_parse_file( h, h->param.psz_cqm_file ) < 0 )
 660         {
 661             x264_free( h );
 662             return NULL;
 663         }
 664
 665     if( h->param.rc.psz_stat_out )
 666         h->param.rc.psz_stat_out = strdup( h->param.rc.psz_stat_out );
 667     if( h->param.rc.psz_stat_in )
 668         h->param.rc.psz_stat_in = strdup( h->param.rc.psz_stat_in );
 669
 670     /* VUI */
 671     if( h->param.vui.i_sar_width > 0 && h->param.vui.i_sar_height > 0 )
 672     {
 673         int i_w = param->vui.i_sar_width;
 674         int i_h = param->vui.i_sar_height;
 675
 676         x264_reduce_fraction( &i_w, &i_h );
 677
 678         while( i_w > 65535 || i_h > 65535 )
 679         {
 680             i_w /= 2;
 681             i_h /= 2;
 682         }
 683
 684         h->param.vui.i_sar_width = 0;
 685         h->param.vui.i_sar_height = 0;
 686         if( i_w == 0 || i_h == 0 )
 687         {
 688             x264_log( h, X264_LOG_WARNING, "cannot create valid sample aspect ratio\n" );
 689         }
 690         else
 691         {
 692             x264_log( h, X264_LOG_INFO, "using SAR=%d/%d\n", i_w, i_h );
 693             h->param.vui.i_sar_width = i_w;
 694             h->param.vui.i_sar_height = i_h;
 695         }
 696     }
 697
 698     x264_reduce_fraction( &h->param.i_fps_num, &h->param.i_fps_den );
 699
 700     /* Init x264_t */
 701     h->i_frame = 0;
 702     h->i_frame_num = 0;
 703     h->i_idr_pic_id = 0;
 704
 705     h->sps = &h->sps_array[0];
 706     x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
 707
 708     h->pps = &h->pps_array[0];
 709     x264_pps_init( h->pps, h->param.i_sps_id, &h->param, h->sps);
 710
 711     x264_validate_levels( h, 1 );
 712
 713     if( x264_cqm_init( h ) < 0 )
 714     {
 715         x264_free( h );
 716         return NULL;
 717     }
 718
 719     h->mb.i_mb_count = h->sps->i_mb_width * h->sps->i_mb_height;
 720
 721     /* Init frames. */
 722     if( h->param.i_bframe_adaptive == X264_B_ADAPT_TRELLIS )
 723         h->frames.i_delay = X264_MAX(h->param.i_bframe,3)*4 + h->param.i_threads - 1;
 724     else
 725         h->frames.i_delay = h->param.i_bframe + h->param.i_threads - 1;
 726     h->frames.i_max_ref0 = h->param.i_frame_reference;
 727     h->frames.i_max_ref1 = h->sps->vui.i_num_reorder_frames;
 728     h->frames.i_max_dpb  = h->sps->vui.i_max_dec_frame_buffering;
 729     h->frames.b_have_lowres = !h->param.rc.b_stat_read
 730         && ( h->param.rc.i_rc_method == X264_RC_ABR
 731           || h->param.rc.i_rc_method == X264_RC_CRF
 732           || h->param.i_bframe_adaptive
 733           || h->param.i_scenecut_threshold );
 734     h->frames.b_have_lowres |= (h->param.rc.b_stat_read && h->param.rc.i_vbv_buffer_size > 0);
 735     h->frames.b_have_sub8x8_esa = !!(h->param.analyse.inter & X264_ANALYSE_PSUB8x8);
 736
 737     h->frames.i_last_idr = - h->param.i_keyint_max;
 738     h->frames.i_input    = 0;
 739     h->frames.last_nonb  = NULL;
 740
 741     h->i_ref0 = 0;
 742     h->i_ref1 = 0;
 743
 744     h->chroma_qp_table = i_chroma_qp_table + 12 + h->pps->i_chroma_qp_index_offset;
 745
 746     x264_rdo_init( );
 747
 748     /* init CPU functions */
 749     x264_predict_16x16_init( h->param.cpu, h->predict_16x16 );
 750     x264_predict_8x8c_init( h->param.cpu, h->predict_8x8c );
 751     x264_predict_8x8_init( h->param.cpu, h->predict_8x8, &h->predict_8x8_filter );
 752     x264_predict_4x4_init( h->param.cpu, h->predict_4x4 );
 753     if( !h->param.b_cabac )
 754         x264_init_vlc_tables();
 755     x264_pixel_init( h->param.cpu, &h->pixf );
 756     x264_dct_init( h->param.cpu, &h->dctf );
 757     x264_zigzag_init( h->param.cpu, &h->zigzagf, h->param.b_interlaced );
 758     x264_mc_init( h->param.cpu, &h->mc );
 759     x264_quant_init( h, h->param.cpu, &h->quantf );
 760     x264_deblock_init( h->param.cpu, &h->loopf );
 761     x264_dct_init_weights();
 762
 763     mbcmp_init( h );
 764
 765     p = buf + sprintf( buf, "using cpu capabilities:" );
 766     for( i=0; x264_cpu_names[i].flags; i++ )
 767     {
 768         if( !strcmp(x264_cpu_names[i].name, "SSE2")
 769             && param->cpu & (X264_CPU_SSE2_IS_FAST|X264_CPU_SSE2_IS_SLOW) )
 770             continue;
 771         if( !strcmp(x264_cpu_names[i].name, "SSE3")
 772             && (param->cpu & X264_CPU_SSSE3 || !(param->cpu & X264_CPU_CACHELINE_64)) )
 773             continue;
 774         if( !strcmp(x264_cpu_names[i].name, "SSE4.1")
 775             && (param->cpu & X264_CPU_SSE42) )
 776             continue;
 777         if( (param->cpu & x264_cpu_names[i].flags) == x264_cpu_names[i].flags
 778             && (!i || x264_cpu_names[i].flags != x264_cpu_names[i-1].flags) )
 779             p += sprintf( p, " %s", x264_cpu_names[i].name );
 780     }
 781     if( !param->cpu )
 782         p += sprintf( p, " none!" );
 783     x264_log( h, X264_LOG_INFO, "%s\n", buf );
 784
 785     h->out.i_nal = 0;
 786     h->out.i_bitstream = X264_MAX( 1000000, h->param.i_width * h->param.i_height * 4
 787         * ( h->param.rc.i_rc_method == X264_RC_ABR ? pow( 0.95, h->param.rc.i_qp_min )
 788           : pow( 0.95, h->param.rc.i_qp_constant ) * X264_MAX( 1, h->param.rc.f_ip_factor )));
 789
 790     h->thread[0] = h;
 791     h->i_thread_num = 0;
 792     for( i = 1; i < h->param.i_threads; i++ )
 793         h->thread[i] = x264_malloc( sizeof(x264_t) );
 794
 795     for( i = 0; i < h->param.i_threads; i++ )
 796     {
 797         if( i > 0 )
 798             *h->thread[i] = *h;
 799         h->thread[i]->fdec = x264_frame_pop_unused( h );
 800         h->thread[i]->out.p_bitstream = x264_malloc( h->out.i_bitstream );
 801         if( x264_macroblock_cache_init( h->thread[i] ) < 0 )
 802             return NULL;
 803     }
 804
 805     if( x264_ratecontrol_new( h ) < 0 )
 806         return NULL;
 807
 808     if( h->param.psz_dump_yuv )
 809     {
 810         /* create or truncate the reconstructed video file */
 811         FILE *f = fopen( h->param.psz_dump_yuv, "w" );
 812         if( f )
 813             fclose( f );
 814         else
 815         {
 816             x264_log( h, X264_LOG_ERROR, "can't write to fdec.yuv\n" );
 817             x264_free( h );
 818             return NULL;
 819         }
 820     }
 821
 822     x264_log( h, X264_LOG_INFO, "profile %s, level %d.%d\n",
 823         h->sps->i_profile_idc == PROFILE_BASELINE ? "Baseline" :
 824         h->sps->i_profile_idc == PROFILE_MAIN ? "Main" :
 825         h->sps->i_profile_idc == PROFILE_HIGH ? "High" :
 826         "High 4:4:4 Predictive", h->sps->i_level_idc/10, h->sps->i_level_idc%10 );
 827
 828     return h;
 829 }
 830
 831 /****************************************************************************
 832  * x264_encoder_reconfig:
 833  ****************************************************************************/
 834 int x264_encoder_reconfig( x264_t *h, x264_param_t *param )
 835 {
 836 #define COPY(var) h->param.var = param->var
 837     COPY( i_frame_reference ); // but never uses more refs than initially specified
 838     COPY( i_bframe_bias );
 839     if( h->param.i_scenecut_threshold )
 840         COPY( i_scenecut_threshold ); // can't turn it on or off, only vary the threshold
 841     COPY( b_deblocking_filter );
 842     COPY( i_deblocking_filter_alphac0 );
 843     COPY( i_deblocking_filter_beta );
 844     COPY( analyse.intra );
 845     COPY( analyse.inter );
 846     COPY( analyse.i_direct_mv_pred );
 847     /* Scratch buffer prevents me_range from being increased for esa/tesa */
 848     if( h->param.analyse.i_me_method < X264_ME_ESA || param->analyse.i_me_range < h->param.analyse.i_me_range )
 849         COPY( analyse.i_me_range );
 850     COPY( analyse.i_noise_reduction );
 851     /* We can't switch out of subme=0 during encoding. */
 852     if( h->param.analyse.i_subpel_refine )
 853         COPY( analyse.i_subpel_refine );
 854     COPY( analyse.i_trellis );
 855     COPY( analyse.b_chroma_me );
 856     COPY( analyse.b_dct_decimate );
 857     COPY( analyse.b_fast_pskip );
 858     COPY( analyse.b_mixed_references );
 859     COPY( analyse.f_psy_rd );
 860     COPY( analyse.f_psy_trellis );
 861     // can only twiddle these if they were enabled to begin with:
 862     if( h->param.analyse.i_me_method >= X264_ME_ESA || param->analyse.i_me_method < X264_ME_ESA )
 863         COPY( analyse.i_me_method );
 864     if( h->param.analyse.i_me_method >= X264_ME_ESA && !h->frames.b_have_sub8x8_esa )
 865         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
 866     if( h->pps->b_transform_8x8_mode )
 867         COPY( analyse.b_transform_8x8 );
 868     if( h->frames.i_max_ref1 > 1 )
 869         COPY( b_bframe_pyramid );
 870 #undef COPY
 871
 872     mbcmp_init( h );
 873
 874     return x264_validate_parameters( h );
 875 }
 876
 877 /* internal usage */
 878 static void x264_nal_start( x264_t *h, int i_type, int i_ref_idc )
 879 {
 880     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
 881
 882     nal->i_ref_idc = i_ref_idc;
 883     nal->i_type    = i_type;
 884
 885     nal->i_payload= 0;
 886     nal->p_payload= &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8];
 887 }
 888 static void x264_nal_end( x264_t *h )
 889 {
 890     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
 891     nal->i_payload = &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8] - nal->p_payload;
 892     h->out.i_nal++;
 893 }
 894
 895 /****************************************************************************
 896  * x264_encoder_headers:
 897  ****************************************************************************/
 898 int x264_encoder_headers( x264_t *h, x264_nal_t **pp_nal, int *pi_nal )
 899 {
 900     /* init bitstream context */
 901     h->out.i_nal = 0;
 902     bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
 903
 904     /* Put SPS and PPS */
 905     if( h->i_frame == 0 )
 906     {
 907         /* identify ourself */
 908         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
 909         x264_sei_version_write( h, &h->out.bs );
 910         x264_nal_end( h );
 911
 912         /* generate sequence parameters */
 913         x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
 914         x264_sps_write( &h->out.bs, h->sps );
 915         x264_nal_end( h );
 916
 917         /* generate picture parameters */
 918         x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
 919         x264_pps_write( &h->out.bs, h->pps );
 920         x264_nal_end( h );
 921     }
 922     /* now set output*/
 923     *pi_nal = h->out.i_nal;
 924     *pp_nal = &h->out.nal[0];
 925     h->out.i_nal = 0;
 926
 927     return 0;
 928 }
 929
 930 static inline void x264_reference_build_list( x264_t *h, int i_poc )
 931 {
 932     int i;
 933     int b_ok;
 934
 935     /* build ref list 0/1 */
 936     h->i_ref0 = 0;
 937     h->i_ref1 = 0;
 938     for( i = 0; h->frames.reference[i]; i++ )
 939     {
 940         if( h->frames.reference[i]->i_poc < i_poc )
 941         {
 942             h->fref0[h->i_ref0++] = h->frames.reference[i];
 943         }
 944         else if( h->frames.reference[i]->i_poc > i_poc )
 945         {
 946             h->fref1[h->i_ref1++] = h->frames.reference[i];
 947         }
 948     }
 949
 950     /* Order ref0 from higher to lower poc */
 951     do
 952     {
 953         b_ok = 1;
 954         for( i = 0; i < h->i_ref0 - 1; i++ )
 955         {
 956             if( h->fref0[i]->i_poc < h->fref0[i+1]->i_poc )
 957             {
 958                 XCHG( x264_frame_t*, h->fref0[i], h->fref0[i+1] );
 959                 b_ok = 0;
 960                 break;
 961             }
 962         }
 963     } while( !b_ok );
 964     /* Order ref1 from lower to higher poc (bubble sort) for B-frame */
 965     do
 966     {
 967         b_ok = 1;
 968         for( i = 0; i < h->i_ref1 - 1; i++ )
 969         {
 970             if( h->fref1[i]->i_poc > h->fref1[i+1]->i_poc )
 971             {
 972                 XCHG( x264_frame_t*, h->fref1[i], h->fref1[i+1] );
 973                 b_ok = 0;
 974                 break;
 975             }
 976         }
 977     } while( !b_ok );
 978
 979     /* In the standard, a P-frame's ref list is sorted by frame_num.
 980      * We use POC, but check whether explicit reordering is needed */
 981     h->b_ref_reorder[0] =
 982     h->b_ref_reorder[1] = 0;
 983     if( h->sh.i_type == SLICE_TYPE_P )
 984     {
 985         for( i = 0; i < h->i_ref0 - 1; i++ )
 986             if( h->fref0[i]->i_frame_num < h->fref0[i+1]->i_frame_num )
 987             {
 988                 h->b_ref_reorder[0] = 1;
 989                 break;
 990             }
 991     }
 992
 993     h->i_ref1 = X264_MIN( h->i_ref1, h->frames.i_max_ref1 );
 994     h->i_ref0 = X264_MIN( h->i_ref0, h->frames.i_max_ref0 );
 995     h->i_ref0 = X264_MIN( h->i_ref0, h->param.i_frame_reference ); // if reconfig() has lowered the limit
 996     assert( h->i_ref0 + h->i_ref1 <= 16 );
 997     h->mb.pic.i_fref[0] = h->i_ref0;
 998     h->mb.pic.i_fref[1] = h->i_ref1;
 999 }
1000
1001 static void x264_fdec_filter_row( x264_t *h, int mb_y )
1002 {
1003     /* mb_y is the mb to be encoded next, not the mb to be filtered here */
1004     int b_hpel = h->fdec->b_kept_as_ref;
1005     int b_deblock = !h->sh.i_disable_deblocking_filter_idc;
1006     int b_end = mb_y == h->sps->i_mb_height;
1007     int min_y = mb_y - (1 << h->sh.b_mbaff);
1008     int max_y = b_end ? h->sps->i_mb_height : mb_y;
1009     b_deblock &= b_hpel || h->param.psz_dump_yuv;
1010     if( mb_y & h->sh.b_mbaff )
1011         return;
1012     if( min_y < 0 )
1013         return;
1014
1015     if( !b_end )
1016     {
1017         int i, j;
1018         for( j=0; j<=h->sh.b_mbaff; j++ )
1019             for( i=0; i<3; i++ )
1020             {
1021                 memcpy( h->mb.intra_border_backup[j][i],
1022                         h->fdec->plane[i] + ((mb_y*16 >> !!i) + j - 1 - h->sh.b_mbaff) * h->fdec->i_stride[i],
1023                         h->sps->i_mb_width*16 >> !!i );
1024             }
1025     }
1026
1027     if( b_deblock )
1028     {
1029         int y;
1030         for( y = min_y; y < max_y; y += (1 << h->sh.b_mbaff) )
1031             x264_frame_deblock_row( h, y );
1032     }
1033
1034     if( b_hpel )
1035     {
1036         x264_frame_expand_border( h, h->fdec, min_y, b_end );
1037         if( h->param.analyse.i_subpel_refine )
1038         {
1039             x264_frame_filter( h, h->fdec, min_y, b_end );
1040             x264_frame_expand_border_filtered( h, h->fdec, min_y, b_end );
1041         }
1042     }
1043
1044     if( h->param.i_threads > 1 && h->fdec->b_kept_as_ref )
1045     {
1046         x264_frame_cond_broadcast( h->fdec, mb_y*16 + (b_end ? 10000 : -(X264_THREAD_HEIGHT << h->sh.b_mbaff)) );
1047     }
1048
1049     min_y = X264_MAX( min_y*16-8, 0 );
1050     max_y = b_end ? h->param.i_height : mb_y*16-8;
1051
1052     if( h->param.analyse.b_psnr )
1053     {
1054         int i;
1055         for( i=0; i<3; i++ )
1056             h->stat.frame.i_ssd[i] +=
1057                 x264_pixel_ssd_wxh( &h->pixf,
1058                     h->fdec->plane[i] + (min_y>>!!i) * h->fdec->i_stride[i], h->fdec->i_stride[i],
1059                     h->fenc->plane[i] + (min_y>>!!i) * h->fenc->i_stride[i], h->fenc->i_stride[i],
1060                     h->param.i_width >> !!i, (max_y-min_y) >> !!i );
1061     }
1062
1063     if( h->param.analyse.b_ssim )
1064     {
1065         x264_emms();
1066         /* offset by 2 pixels to avoid alignment of ssim blocks with dct blocks,
1067          * and overlap by 4 */
1068         min_y += min_y == 0 ? 2 : -6;
1069         h->stat.frame.f_ssim +=
1070             x264_pixel_ssim_wxh( &h->pixf,
1071                 h->fdec->plane[0] + 2+min_y*h->fdec->i_stride[0], h->fdec->i_stride[0],
1072                 h->fenc->plane[0] + 2+min_y*h->fenc->i_stride[0], h->fenc->i_stride[0],
1073                 h->param.i_width-2, max_y-min_y, h->scratch_buffer );
1074     }
1075 }
1076
1077 static inline void x264_reference_update( x264_t *h )
1078 {
1079     int i;
1080
1081     if( h->fdec->i_frame >= 0 )
1082         h->i_frame++;
1083
1084     if( !h->fdec->b_kept_as_ref )
1085     {
1086         if( h->param.i_threads > 1 )
1087         {
1088             x264_frame_push_unused( h, h->fdec );
1089             h->fdec = x264_frame_pop_unused( h );
1090         }
1091         return;
1092     }
1093
1094     /* move lowres copy of the image to the ref frame */
1095     for( i = 0; i < 4; i++)
1096     {
1097         XCHG( uint8_t*, h->fdec->lowres[i], h->fenc->lowres[i] );
1098         XCHG( uint8_t*, h->fdec->buffer_lowres[i], h->fenc->buffer_lowres[i] );
1099     }
1100
1101     /* adaptive B decision needs a pointer, since it can't use the ref lists */
1102     if( h->sh.i_type != SLICE_TYPE_B )
1103         h->frames.last_nonb = h->fdec;
1104
1105     /* move frame in the buffer */
1106     x264_frame_push( h->frames.reference, h->fdec );
1107     if( h->frames.reference[h->frames.i_max_dpb] )
1108         x264_frame_push_unused( h, x264_frame_shift( h->frames.reference ) );
1109     h->fdec = x264_frame_pop_unused( h );
1110 }
1111
1112 static inline void x264_reference_reset( x264_t *h )
1113 {
1114     while( h->frames.reference[0] )
1115         x264_frame_push_unused( h, x264_frame_pop( h->frames.reference ) );
1116     h->fdec->i_poc =
1117     h->fenc->i_poc = 0;
1118 }
1119
1120 static inline void x264_slice_init( x264_t *h, int i_nal_type, int i_global_qp )
1121 {
1122     /* ------------------------ Create slice header  ----------------------- */
1123     if( i_nal_type == NAL_SLICE_IDR )
1124     {
1125         x264_slice_header_init( h, &h->sh, h->sps, h->pps, h->i_idr_pic_id, h->i_frame_num, i_global_qp );
1126
1127         /* increment id */
1128         h->i_idr_pic_id = ( h->i_idr_pic_id + 1 ) % 65536;
1129     }
1130     else
1131     {
1132         x264_slice_header_init( h, &h->sh, h->sps, h->pps, -1, h->i_frame_num, i_global_qp );
1133
1134         /* always set the real higher num of ref frame used */
1135         h->sh.b_num_ref_idx_override = 1;
1136         h->sh.i_num_ref_idx_l0_active = h->i_ref0 <= 0 ? 1 : h->i_ref0;
1137         h->sh.i_num_ref_idx_l1_active = h->i_ref1 <= 0 ? 1 : h->i_ref1;
1138     }
1139
1140     h->fdec->i_frame_num = h->sh.i_frame_num;
1141
1142     if( h->sps->i_poc_type == 0 )
1143     {
1144         h->sh.i_poc_lsb = h->fdec->i_poc & ( (1 << h->sps->i_log2_max_poc_lsb) - 1 );
1145         h->sh.i_delta_poc_bottom = 0;   /* XXX won't work for field */
1146     }
1147     else if( h->sps->i_poc_type == 1 )
1148     {
1149         /* FIXME TODO FIXME */
1150     }
1151     else
1152     {
1153         /* Nothing to do ? */
1154     }
1155
1156     x264_macroblock_slice_init( h );
1157 }
1158
1159 static void x264_slice_write( x264_t *h )
1160 {
1161     int i_skip;
1162     int mb_xy, i_mb_x, i_mb_y;
1163     int i, i_list, i_ref;
1164
1165     /* init stats */
1166     memset( &h->stat.frame, 0, sizeof(h->stat.frame) );
1167
1168     /* Slice */
1169     x264_nal_start( h, h->i_nal_type, h->i_nal_ref_idc );
1170
1171     /* Slice header */
1172     x264_slice_header_write( &h->out.bs, &h->sh, h->i_nal_ref_idc );
1173     if( h->param.b_cabac )
1174     {
1175         /* alignment needed */
1176         bs_align_1( &h->out.bs );
1177
1178         /* init cabac */
1179         x264_cabac_context_init( &h->cabac, h->sh.i_type, h->sh.i_qp, h->sh.i_cabac_init_idc );
1180         x264_cabac_encode_init ( &h->cabac, h->out.bs.p, h->out.bs.p_end );
1181     }
1182     h->mb.i_last_qp = h->sh.i_qp;
1183     h->mb.i_last_dqp = 0;
1184
1185     i_mb_y = h->sh.i_first_mb / h->sps->i_mb_width;
1186     i_mb_x = h->sh.i_first_mb % h->sps->i_mb_width;
1187     i_skip = 0;
1188
1189     while( (mb_xy = i_mb_x + i_mb_y * h->sps->i_mb_width) < h->sh.i_last_mb )
1190     {
1191         int mb_spos = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
1192
1193         if( i_mb_x == 0 )
1194             x264_fdec_filter_row( h, i_mb_y );
1195
1196         /* load cache */
1197         x264_macroblock_cache_load( h, i_mb_x, i_mb_y );
1198
1199         /* analyse parameters
1200          * Slice I: choose I_4x4 or I_16x16 mode
1201          * Slice P: choose between using P mode or intra (4x4 or 16x16)
1202          * */
1203         x264_macroblock_analyse( h );
1204
1205         /* encode this macroblock -> be careful it can change the mb type to P_SKIP if needed */
1206         x264_macroblock_encode( h );
1207
1208         x264_bitstream_check_buffer( h );
1209
1210         if( h->param.b_cabac )
1211         {
1212             if( mb_xy > h->sh.i_first_mb && !(h->sh.b_mbaff && (i_mb_y&1)) )
1213                 x264_cabac_encode_terminal( &h->cabac );
1214
1215             if( IS_SKIP( h->mb.i_type ) )
1216                 x264_cabac_mb_skip( h, 1 );
1217             else
1218             {
1219                 if( h->sh.i_type != SLICE_TYPE_I )
1220                     x264_cabac_mb_skip( h, 0 );
1221                 x264_macroblock_write_cabac( h, &h->cabac );
1222             }
1223         }
1224         else
1225         {
1226             if( IS_SKIP( h->mb.i_type ) )
1227                 i_skip++;
1228             else
1229             {
1230                 if( h->sh.i_type != SLICE_TYPE_I )
1231                 {
1232                     bs_write_ue( &h->out.bs, i_skip );  /* skip run */
1233                     i_skip = 0;
1234                 }
1235                 x264_macroblock_write_cavlc( h, &h->out.bs );
1236             }
1237         }
1238
1239 #if VISUALIZE
1240         if( h->param.b_visualize )
1241             x264_visualize_mb( h );
1242 #endif
1243
1244         /* save cache */
1245         x264_macroblock_cache_save( h );
1246
1247         /* accumulate mb stats */
1248         h->stat.frame.i_mb_count[h->mb.i_type]++;
1249         if( !IS_SKIP(h->mb.i_type) && !IS_INTRA(h->mb.i_type) && !IS_DIRECT(h->mb.i_type) )
1250         {
1251             if( h->mb.i_partition != D_8x8 )
1252                 h->stat.frame.i_mb_partition[h->mb.i_partition] += 4;
1253             else
1254                 for( i = 0; i < 4; i++ )
1255                     h->stat.frame.i_mb_partition[h->mb.i_sub_partition[i]] ++;
1256             if( h->param.i_frame_reference > 1 )
1257                 for( i_list = 0; i_list <= (h->sh.i_type == SLICE_TYPE_B); i_list++ )
1258                     for( i = 0; i < 4; i++ )
1259                     {
1260                         i_ref = h->mb.cache.ref[i_list][ x264_scan8[4*i] ];
1261                         if( i_ref >= 0 )
1262                             h->stat.frame.i_mb_count_ref[i_list][i_ref] ++;
1263                     }
1264         }
1265         if( h->mb.i_cbp_luma || h->mb.i_cbp_chroma )
1266         {
1267             int cbpsum = (h->mb.i_cbp_luma&1) + ((h->mb.i_cbp_luma>>1)&1)
1268                        + ((h->mb.i_cbp_luma>>2)&1) + (h->mb.i_cbp_luma>>3);
1269             int b_intra = IS_INTRA(h->mb.i_type);
1270             h->stat.frame.i_mb_cbp[!b_intra + 0] += cbpsum;
1271             h->stat.frame.i_mb_cbp[!b_intra + 2] += h->mb.i_cbp_chroma >= 1;
1272             h->stat.frame.i_mb_cbp[!b_intra + 4] += h->mb.i_cbp_chroma == 2;
1273         }
1274         if( h->mb.i_cbp_luma && !IS_INTRA(h->mb.i_type) )
1275         {
1276             h->stat.frame.i_mb_count_8x8dct[0] ++;
1277             h->stat.frame.i_mb_count_8x8dct[1] += h->mb.b_transform_8x8;
1278         }
1279
1280         x264_ratecontrol_mb( h, bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac) - mb_spos );
1281
1282         if( h->sh.b_mbaff )
1283         {
1284             i_mb_x += i_mb_y & 1;
1285             i_mb_y ^= i_mb_x < h->sps->i_mb_width;
1286         }
1287         else
1288             i_mb_x++;
1289         if(i_mb_x == h->sps->i_mb_width)
1290         {
1291             i_mb_y++;
1292             i_mb_x = 0;
1293         }
1294     }
1295
1296     if( h->param.b_cabac )
1297     {
1298         x264_cabac_encode_flush( h, &h->cabac );
1299         h->out.bs.p = h->cabac.p;
1300     }
1301     else
1302     {
1303         if( i_skip > 0 )
1304             bs_write_ue( &h->out.bs, i_skip );  /* last skip run */
1305         /* rbsp_slice_trailing_bits */
1306         bs_rbsp_trailing( &h->out.bs );
1307     }
1308
1309     x264_nal_end( h );
1310
1311     x264_fdec_filter_row( h, h->sps->i_mb_height );
1312
1313     /* Compute misc bits */
1314     h->stat.frame.i_misc_bits = bs_pos( &h->out.bs )
1315                               + NALU_OVERHEAD * 8
1316                               - h->stat.frame.i_tex_bits
1317                               - h->stat.frame.i_mv_bits;
1318 }
1319
1320 static void x264_thread_sync_context( x264_t *dst, x264_t *src )
1321 {
1322     x264_frame_t **f;
1323     if( dst == src )
1324         return;
1325
1326     // reference counting
1327     for( f = src->frames.reference; *f; f++ )
1328         (*f)->i_reference_count++;
1329     for( f = dst->frames.reference; *f; f++ )
1330         x264_frame_push_unused( src, *f );
1331     src->fdec->i_reference_count++;
1332     x264_frame_push_unused( src, dst->fdec );
1333
1334     // copy everything except the per-thread pointers and the constants.
1335     memcpy( &dst->i_frame, &src->i_frame, offsetof(x264_t, mb.type) - offsetof(x264_t, i_frame) );
1336     dst->stat = src->stat;
1337 }
1338
1339 static void x264_thread_sync_stat( x264_t *dst, x264_t *src )
1340 {
1341     if( dst == src )
1342         return;
1343     memcpy( &dst->stat.i_slice_count, &src->stat.i_slice_count, sizeof(dst->stat) - sizeof(dst->stat.frame) );
1344 }
1345
1346 static int x264_slices_write( x264_t *h )
1347 {
1348     int i_frame_size;
1349
1350 #ifdef HAVE_MMX
1351     /* Misalign mask has to be set separately for each thread. */
1352     if( h->param.cpu&X264_CPU_SSE_MISALIGN )
1353         x264_cpu_mask_misalign_sse();
1354 #endif
1355
1356 #if VISUALIZE
1357     if( h->param.b_visualize )
1358         x264_visualize_init( h );
1359 #endif
1360
1361     x264_stack_align( x264_slice_write, h );
1362     i_frame_size = h->out.nal[h->out.i_nal-1].i_payload;
1363
1364 #if VISUALIZE
1365     if( h->param.b_visualize )
1366     {
1367         x264_visualize_show( h );
1368         x264_visualize_close( h );
1369     }
1370 #endif
1371
1372     h->out.i_frame_size = i_frame_size;
1373     return 0;
1374 }
1375
1376 /****************************************************************************
1377  * x264_encoder_encode:
1378  *  XXX: i_poc   : is the poc of the current given picture
1379  *       i_frame : is the number of the frame being coded
1380  *  ex:  type frame poc
1381  *       I      0   2*0
1382  *       P      1   2*3
1383  *       B      2   2*1
1384  *       B      3   2*2
1385  *       P      4   2*6
1386  *       B      5   2*4
1387  *       B      6   2*5
1388  ****************************************************************************/
1389 int     x264_encoder_encode( x264_t *h,
1390                              x264_nal_t **pp_nal, int *pi_nal,
1391                              x264_picture_t *pic_in,
1392                              x264_picture_t *pic_out )
1393 {
1394     x264_t *thread_current, *thread_prev, *thread_oldest;
1395     int     i_nal_type;
1396     int     i_nal_ref_idc;
1397
1398     int   i_global_qp;
1399
1400     if( h->param.i_threads > 1)
1401     {
1402         int i = ++h->i_thread_phase;
1403         int t = h->param.i_threads;
1404         thread_current = h->thread[ i%t ];
1405         thread_prev    = h->thread[ (i-1)%t ];
1406         thread_oldest  = h->thread[ (i+1)%t ];
1407         x264_thread_sync_context( thread_current, thread_prev );
1408         x264_thread_sync_ratecontrol( thread_current, thread_prev, thread_oldest );
1409         h = thread_current;
1410 //      fprintf(stderr, "current: %p  prev: %p  oldest: %p \n", thread_current, thread_prev, thread_oldest);
1411     }
1412     else
1413     {
1414         thread_current =
1415         thread_oldest  = h;
1416     }
1417
1418     // ok to call this before encoding any frames, since the initial values of fdec have b_kept_as_ref=0
1419     x264_reference_update( h );
1420     h->fdec->i_lines_completed = -1;
1421
1422     /* no data out */
1423     *pi_nal = 0;
1424     *pp_nal = NULL;
1425
1426     /* ------------------- Setup new frame from picture -------------------- */
1427     if( pic_in != NULL )
1428     {
1429         /* 1: Copy the picture to a frame and move it to a buffer */
1430         x264_frame_t *fenc = x264_frame_pop_unused( h );
1431
1432         if( x264_frame_copy_picture( h, fenc, pic_in ) < 0 )
1433             return -1;
1434
1435         if( h->param.i_width != 16 * h->sps->i_mb_width ||
1436             h->param.i_height != 16 * h->sps->i_mb_height )
1437             x264_frame_expand_border_mod16( h, fenc );
1438
1439         fenc->i_frame = h->frames.i_input++;
1440
1441         x264_frame_push( h->frames.next, fenc );
1442
1443         if( h->frames.b_have_lowres )
1444             x264_frame_init_lowres( h, fenc );
1445
1446         if( h->param.rc.i_aq_mode )
1447             x264_adaptive_quant_frame( h, fenc );
1448
1449         if( h->frames.i_input <= h->frames.i_delay + 1 - h->param.i_threads )
1450         {
1451             /* Nothing yet to encode */
1452             /* waiting for filling bframe buffer */
1453             pic_out->i_type = X264_TYPE_AUTO;
1454             return 0;
1455         }
1456     }
1457
1458     if( h->frames.current[0] == NULL )
1459     {
1460         int bframes = 0;
1461         /* 2: Select frame types */
1462         if( h->frames.next[0] == NULL )
1463         {
1464             x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
1465             return 0;
1466         }
1467
1468         x264_stack_align( x264_slicetype_decide, h );
1469
1470         /* 3: move some B-frames and 1 non-B to encode queue */
1471         while( IS_X264_TYPE_B( h->frames.next[bframes]->i_type ) )
1472             bframes++;
1473         x264_frame_push( h->frames.current, x264_frame_shift( &h->frames.next[bframes] ) );
1474         /* FIXME: when max B-frames > 3, BREF may no longer be centered after GOP closing */
1475         if( h->param.b_bframe_pyramid && bframes > 1 )
1476         {
1477             x264_frame_t *mid = x264_frame_shift( &h->frames.next[bframes/2] );
1478             mid->i_type = X264_TYPE_BREF;
1479             x264_frame_push( h->frames.current, mid );
1480             bframes--;
1481         }
1482         while( bframes-- )
1483             x264_frame_push( h->frames.current, x264_frame_shift( h->frames.next ) );
1484     }
1485
1486     /* ------------------- Get frame to be encoded ------------------------- */
1487     /* 4: get picture to encode */
1488     h->fenc = x264_frame_shift( h->frames.current );
1489     if( h->fenc == NULL )
1490     {
1491         /* Nothing yet to encode (ex: waiting for I/P with B frames) */
1492         /* waiting for filling bframe buffer */
1493         pic_out->i_type = X264_TYPE_AUTO;
1494         return 0;
1495     }
1496
1497     if( h->fenc->i_type == X264_TYPE_IDR )
1498     {
1499         h->frames.i_last_idr = h->fenc->i_frame;
1500     }
1501
1502     /* ------------------- Setup frame context ----------------------------- */
1503     /* 5: Init data dependent of frame type */
1504     if( h->fenc->i_type == X264_TYPE_IDR )
1505     {
1506         /* reset ref pictures */
1507         x264_reference_reset( h );
1508
1509         i_nal_type    = NAL_SLICE_IDR;
1510         i_nal_ref_idc = NAL_PRIORITY_HIGHEST;
1511         h->sh.i_type = SLICE_TYPE_I;
1512     }
1513     else if( h->fenc->i_type == X264_TYPE_I )
1514     {
1515         i_nal_type    = NAL_SLICE;
1516         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
1517         h->sh.i_type = SLICE_TYPE_I;
1518     }
1519     else if( h->fenc->i_type == X264_TYPE_P )
1520     {
1521         i_nal_type    = NAL_SLICE;
1522         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
1523         h->sh.i_type = SLICE_TYPE_P;
1524     }
1525     else if( h->fenc->i_type == X264_TYPE_BREF )
1526     {
1527         i_nal_type    = NAL_SLICE;
1528         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* maybe add MMCO to forget it? -> low */
1529         h->sh.i_type = SLICE_TYPE_B;
1530     }
1531     else    /* B frame */
1532     {
1533         i_nal_type    = NAL_SLICE;
1534         i_nal_ref_idc = NAL_PRIORITY_DISPOSABLE;
1535         h->sh.i_type = SLICE_TYPE_B;
1536     }
1537
1538     h->fdec->i_poc =
1539     h->fenc->i_poc = 2 * (h->fenc->i_frame - h->frames.i_last_idr);
1540     h->fdec->i_type = h->fenc->i_type;
1541     h->fdec->i_frame = h->fenc->i_frame;
1542     h->fenc->b_kept_as_ref =
1543     h->fdec->b_kept_as_ref = i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE && h->param.i_keyint_max > 1;
1544
1545
1546
1547     /* ------------------- Init                ----------------------------- */
1548     /* build ref list 0/1 */
1549     x264_reference_build_list( h, h->fdec->i_poc );
1550
1551     /* Init the rate control */
1552     x264_ratecontrol_start( h, h->fenc->i_qpplus1 );
1553     i_global_qp = x264_ratecontrol_qp( h );
1554
1555     pic_out->i_qpplus1 =
1556     h->fdec->i_qpplus1 = i_global_qp + 1;
1557
1558     if( h->sh.i_type == SLICE_TYPE_B )
1559         x264_macroblock_bipred_init( h );
1560
1561     /* ------------------------ Create slice header  ----------------------- */
1562     x264_slice_init( h, i_nal_type, i_global_qp );
1563
1564     if( i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE )
1565         h->i_frame_num++;
1566
1567     /* ---------------------- Write the bitstream -------------------------- */
1568     /* Init bitstream context */
1569     h->out.i_nal = 0;
1570     bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
1571
1572     if(h->param.b_aud){
1573         int pic_type;
1574
1575         if(h->sh.i_type == SLICE_TYPE_I)
1576             pic_type = 0;
1577         else if(h->sh.i_type == SLICE_TYPE_P)
1578             pic_type = 1;
1579         else if(h->sh.i_type == SLICE_TYPE_B)
1580             pic_type = 2;
1581         else
1582             pic_type = 7;
1583
1584         x264_nal_start(h, NAL_AUD, NAL_PRIORITY_DISPOSABLE);
1585         bs_write(&h->out.bs, 3, pic_type);
1586         bs_rbsp_trailing(&h->out.bs);
1587         x264_nal_end(h);
1588     }
1589
1590     h->i_nal_type = i_nal_type;
1591     h->i_nal_ref_idc = i_nal_ref_idc;
1592
1593     /* Write SPS and PPS */
1594     if( i_nal_type == NAL_SLICE_IDR && h->param.b_repeat_headers )
1595     {
1596         if( h->fenc->i_frame == 0 )
1597         {
1598             /* identify ourself */
1599             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
1600             x264_sei_version_write( h, &h->out.bs );
1601             x264_nal_end( h );
1602         }
1603
1604         /* generate sequence parameters */
1605         x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
1606         x264_sps_write( &h->out.bs, h->sps );
1607         x264_nal_end( h );
1608
1609         /* generate picture parameters */
1610         x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
1611         x264_pps_write( &h->out.bs, h->pps );
1612         x264_nal_end( h );
1613     }
1614
1615     /* Write frame */
1616     if( h->param.i_threads > 1 )
1617     {
1618         x264_pthread_create( &h->thread_handle, NULL, (void*)x264_slices_write, h );
1619         h->b_thread_active = 1;
1620     }
1621     else
1622         x264_slices_write( h );
1623
1624     x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
1625     return 0;
1626 }
1627
1628 static void x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
1629                                     x264_nal_t **pp_nal, int *pi_nal,
1630                                     x264_picture_t *pic_out )
1631 {
1632     int i, i_list;
1633     char psz_message[80];
1634
1635     if( h->b_thread_active )
1636     {
1637         x264_pthread_join( h->thread_handle, NULL );
1638         h->b_thread_active = 0;
1639     }
1640     if( !h->out.i_nal )
1641     {
1642         pic_out->i_type = X264_TYPE_AUTO;
1643         return;
1644     }
1645
1646     x264_frame_push_unused( thread_current, h->fenc );
1647
1648     /* End bitstream, set output  */
1649     *pi_nal = h->out.i_nal;
1650     *pp_nal = h->out.nal;
1651     h->out.i_nal = 0;
1652
1653     /* Set output picture properties */
1654     if( h->sh.i_type == SLICE_TYPE_I )
1655         pic_out->i_type = h->i_nal_type == NAL_SLICE_IDR ? X264_TYPE_IDR : X264_TYPE_I;
1656     else if( h->sh.i_type == SLICE_TYPE_P )
1657         pic_out->i_type = X264_TYPE_P;
1658     else
1659         pic_out->i_type = X264_TYPE_B;
1660     pic_out->i_pts = h->fenc->i_pts;
1661
1662     pic_out->img.i_plane = h->fdec->i_plane;
1663     for(i = 0; i < 3; i++)
1664     {
1665         pic_out->img.i_stride[i] = h->fdec->i_stride[i];
1666         pic_out->img.plane[i] = h->fdec->plane[i];
1667     }
1668
1669     /* ---------------------- Update encoder state ------------------------- */
1670
1671     /* update rc */
1672     x264_emms();
1673     x264_ratecontrol_end( h, h->out.i_frame_size * 8 );
1674
1675     /* restore CPU state (before using float again) */
1676     x264_emms();
1677
1678     x264_noise_reduction_update( thread_current );
1679
1680     /* ---------------------- Compute/Print statistics --------------------- */
1681     x264_thread_sync_stat( h, h->thread[0] );
1682
1683     /* Slice stat */
1684     h->stat.i_slice_count[h->sh.i_type]++;
1685     h->stat.i_slice_size[h->sh.i_type] += h->out.i_frame_size + NALU_OVERHEAD;
1686     h->stat.f_slice_qp[h->sh.i_type] += h->fdec->f_qp_avg_aq;
1687
1688     for( i = 0; i < X264_MBTYPE_MAX; i++ )
1689         h->stat.i_mb_count[h->sh.i_type][i] += h->stat.frame.i_mb_count[i];
1690     for( i = 0; i < X264_PARTTYPE_MAX; i++ )
1691         h->stat.i_mb_partition[h->sh.i_type][i] += h->stat.frame.i_mb_partition[i];
1692     for( i = 0; i < 2; i++ )
1693         h->stat.i_mb_count_8x8dct[i] += h->stat.frame.i_mb_count_8x8dct[i];
1694     for( i = 0; i < 6; i++ )
1695         h->stat.i_mb_cbp[i] += h->stat.frame.i_mb_cbp[i];
1696     if( h->sh.i_type != SLICE_TYPE_I )
1697         for( i_list = 0; i_list < 2; i_list++ )
1698             for( i = 0; i < 32; i++ )
1699                 h->stat.i_mb_count_ref[h->sh.i_type][i_list][i] += h->stat.frame.i_mb_count_ref[i_list][i];
1700     if( h->sh.i_type == SLICE_TYPE_P )
1701         h->stat.i_consecutive_bframes[h->fdec->i_frame - h->fref0[0]->i_frame - 1]++;
1702     if( h->sh.i_type == SLICE_TYPE_B )
1703     {
1704         h->stat.i_direct_frames[ h->sh.b_direct_spatial_mv_pred ] ++;
1705         if( h->mb.b_direct_auto_write )
1706         {
1707             //FIXME somewhat arbitrary time constants
1708             if( h->stat.i_direct_score[0] + h->stat.i_direct_score[1] > h->mb.i_mb_count )
1709             {
1710                 for( i = 0; i < 2; i++ )
1711                     h->stat.i_direct_score[i] = h->stat.i_direct_score[i] * 9/10;
1712             }
1713             for( i = 0; i < 2; i++ )
1714                 h->stat.i_direct_score[i] += h->stat.frame.i_direct_score[i];
1715         }
1716     }
1717
1718     psz_message[0] = '\0';
1719     if( h->param.analyse.b_psnr )
1720     {
1721         int64_t ssd[3] = {
1722             h->stat.frame.i_ssd[0],
1723             h->stat.frame.i_ssd[1],
1724             h->stat.frame.i_ssd[2],
1725         };
1726
1727         h->stat.i_ssd_global[h->sh.i_type] += ssd[0] + ssd[1] + ssd[2];
1728         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 );
1729         h->stat.f_psnr_mean_y[h->sh.i_type] += x264_psnr( ssd[0], h->param.i_width * h->param.i_height );
1730         h->stat.f_psnr_mean_u[h->sh.i_type] += x264_psnr( ssd[1], h->param.i_width * h->param.i_height / 4 );
1731         h->stat.f_psnr_mean_v[h->sh.i_type] += x264_psnr( ssd[2], h->param.i_width * h->param.i_height / 4 );
1732
1733         snprintf( psz_message, 80, " PSNR Y:%5.2f U:%5.2f V:%5.2f",
1734                   x264_psnr( ssd[0], h->param.i_width * h->param.i_height ),
1735                   x264_psnr( ssd[1], h->param.i_width * h->param.i_height / 4),
1736                   x264_psnr( ssd[2], h->param.i_width * h->param.i_height / 4) );
1737     }
1738
1739     if( h->param.analyse.b_ssim )
1740     {
1741         double ssim_y = h->stat.frame.f_ssim
1742                       / (((h->param.i_width-6)>>2) * ((h->param.i_height-6)>>2));
1743         h->stat.f_ssim_mean_y[h->sh.i_type] += ssim_y;
1744         snprintf( psz_message + strlen(psz_message), 80 - strlen(psz_message),
1745                   " SSIM Y:%.5f", ssim_y );
1746     }
1747     psz_message[79] = '\0';
1748
1749     x264_log( h, X264_LOG_DEBUG,
1750                   "frame=%4d QP=%.2f NAL=%d Slice:%c Poc:%-3d I:%-4d P:%-4d SKIP:%-4d size=%d bytes%s\n",
1751               h->i_frame,
1752               h->fdec->f_qp_avg_aq,
1753               h->i_nal_ref_idc,
1754               h->sh.i_type == SLICE_TYPE_I ? 'I' : (h->sh.i_type == SLICE_TYPE_P ? 'P' : 'B' ),
1755               h->fdec->i_poc,
1756               h->stat.frame.i_mb_count_i,
1757               h->stat.frame.i_mb_count_p,
1758               h->stat.frame.i_mb_count_skip,
1759               h->out.i_frame_size,
1760               psz_message );
1761
1762     // keep stats all in one place
1763     x264_thread_sync_stat( h->thread[0], h );
1764     // for the use of the next frame
1765     x264_thread_sync_stat( thread_current, h );
1766
1767 #ifdef DEBUG_MB_TYPE
1768 {
1769     static const char mb_chars[] = { 'i', 'i', 'I', 'C', 'P', '8', 'S',
1770         'D', '<', 'X', 'B', 'X', '>', 'B', 'B', 'B', 'B', '8', 'S' };
1771     int mb_xy;
1772     for( mb_xy = 0; mb_xy < h->sps->i_mb_width * h->sps->i_mb_height; mb_xy++ )
1773     {
1774         if( h->mb.type[mb_xy] < X264_MBTYPE_MAX && h->mb.type[mb_xy] >= 0 )
1775             fprintf( stderr, "%c ", mb_chars[ h->mb.type[mb_xy] ] );
1776         else
1777             fprintf( stderr, "? " );
1778
1779         if( (mb_xy+1) % h->sps->i_mb_width == 0 )
1780             fprintf( stderr, "\n" );
1781     }
1782 }
1783 #endif
1784
1785     if( h->param.psz_dump_yuv )
1786         x264_frame_dump( h );
1787 }
1788
1789 static void x264_print_intra( int64_t *i_mb_count, double i_count, int b_print_pcm, char *intra )
1790 {
1791     intra += sprintf( intra, "I16..4%s: %4.1f%% %4.1f%% %4.1f%%",
1792         b_print_pcm ? "..PCM" : "",
1793         i_mb_count[I_16x16]/ i_count,
1794         i_mb_count[I_8x8]  / i_count,
1795         i_mb_count[I_4x4]  / i_count );
1796     if( b_print_pcm )
1797         sprintf( intra, " %4.1f%%", i_mb_count[I_PCM]  / i_count );
1798 }
1799
1800 /****************************************************************************
1801  * x264_encoder_close:
1802  ****************************************************************************/
1803 void    x264_encoder_close  ( x264_t *h )
1804 {
1805     int64_t i_yuv_size = 3 * h->param.i_width * h->param.i_height / 2;
1806     int64_t i_mb_count_size[2][7] = {{0}};
1807     char buf[200];
1808     int i, j, i_list, i_type;
1809     int b_print_pcm = h->stat.i_mb_count[SLICE_TYPE_I][I_PCM]
1810                    || h->stat.i_mb_count[SLICE_TYPE_P][I_PCM]
1811                    || h->stat.i_mb_count[SLICE_TYPE_B][I_PCM];
1812
1813     for( i=0; i<h->param.i_threads; i++ )
1814     {
1815         // don't strictly have to wait for the other threads, but it's simpler than canceling them
1816         if( h->thread[i]->b_thread_active )
1817         {
1818             x264_pthread_join( h->thread[i]->thread_handle, NULL );
1819             assert( h->thread[i]->fenc->i_reference_count == 1 );
1820             x264_frame_delete( h->thread[i]->fenc );
1821         }
1822     }
1823
1824     /* Slices used and PSNR */
1825     for( i=0; i<5; i++ )
1826     {
1827         static const int slice_order[] = { SLICE_TYPE_I, SLICE_TYPE_SI, SLICE_TYPE_P, SLICE_TYPE_SP, SLICE_TYPE_B };
1828         static const char *slice_name[] = { "P", "B", "I", "SP", "SI" };
1829         int i_slice = slice_order[i];
1830
1831         if( h->stat.i_slice_count[i_slice] > 0 )
1832         {
1833             const int i_count = h->stat.i_slice_count[i_slice];
1834             if( h->param.analyse.b_psnr )
1835             {
1836                 x264_log( h, X264_LOG_INFO,
1837                           "slice %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",
1838                           slice_name[i_slice],
1839                           i_count,
1840                           h->stat.f_slice_qp[i_slice] / i_count,
1841                           (double)h->stat.i_slice_size[i_slice] / i_count,
1842                           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,
1843                           h->stat.f_psnr_average[i_slice] / i_count,
1844                           x264_psnr( h->stat.i_ssd_global[i_slice], i_count * i_yuv_size ) );
1845             }
1846             else
1847             {
1848                 x264_log( h, X264_LOG_INFO,
1849                           "slice %s:%-5d Avg QP:%5.2f  size:%6.0f\n",
1850                           slice_name[i_slice],
1851                           i_count,
1852                           h->stat.f_slice_qp[i_slice] / i_count,
1853                           (double)h->stat.i_slice_size[i_slice] / i_count );
1854             }
1855         }
1856     }
1857     if( h->param.i_bframe && h->stat.i_slice_count[SLICE_TYPE_P] )
1858     {
1859         char *p = buf;
1860         int den = 0;
1861         // weight by number of frames (including the P-frame) that are in a sequence of N B-frames
1862         for( i=0; i<=h->param.i_bframe; i++ )
1863             den += (i+1) * h->stat.i_consecutive_bframes[i];
1864         for( i=0; i<=h->param.i_bframe; i++ )
1865             p += sprintf( p, " %4.1f%%", 100. * (i+1) * h->stat.i_consecutive_bframes[i] / den );
1866         x264_log( h, X264_LOG_INFO, "consecutive B-frames:%s\n", buf );
1867     }
1868
1869     for( i_type = 0; i_type < 2; i_type++ )
1870         for( i = 0; i < X264_PARTTYPE_MAX; i++ )
1871         {
1872             if( i == D_DIRECT_8x8 ) continue; /* direct is counted as its own type */
1873             i_mb_count_size[i_type][x264_mb_partition_pixel_table[i]] += h->stat.i_mb_partition[i_type][i];
1874         }
1875
1876     /* MB types used */
1877     if( h->stat.i_slice_count[SLICE_TYPE_I] > 0 )
1878     {
1879         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_I];
1880         double i_count = h->stat.i_slice_count[SLICE_TYPE_I] * h->mb.i_mb_count / 100.0;
1881         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
1882         x264_log( h, X264_LOG_INFO, "mb I  %s\n", buf );
1883     }
1884     if( h->stat.i_slice_count[SLICE_TYPE_P] > 0 )
1885     {
1886         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_P];
1887         double i_count = h->stat.i_slice_count[SLICE_TYPE_P] * h->mb.i_mb_count / 100.0;
1888         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_P];
1889         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
1890         x264_log( h, X264_LOG_INFO,
1891                   "mb P  %s  P16..4: %4.1f%% %4.1f%% %4.1f%% %4.1f%% %4.1f%%    skip:%4.1f%%\n",
1892                   buf,
1893                   i_mb_size[PIXEL_16x16] / (i_count*4),
1894                   (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
1895                   i_mb_size[PIXEL_8x8] / (i_count*4),
1896                   (i_mb_size[PIXEL_8x4] + i_mb_size[PIXEL_4x8]) / (i_count*4),
1897                   i_mb_size[PIXEL_4x4] / (i_count*4),
1898                   i_mb_count[P_SKIP] / i_count );
1899     }
1900     if( h->stat.i_slice_count[SLICE_TYPE_B] > 0 )
1901     {
1902         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_B];
1903         double i_count = h->stat.i_slice_count[SLICE_TYPE_B] * h->mb.i_mb_count / 100.0;
1904         double i_mb_list_count;
1905         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_B];
1906         int64_t list_count[3] = {0}; /* 0 == L0, 1 == L1, 2 == BI */
1907         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
1908         for( i = 0; i < X264_PARTTYPE_MAX; i++ )
1909             for( j = 0; j < 2; j++ )
1910             {
1911                 int l0 = x264_mb_type_list_table[i][0][j];
1912                 int l1 = x264_mb_type_list_table[i][1][j];
1913                 if( l0 || l1 )
1914                     list_count[l1+l0*l1] += h->stat.i_mb_count[SLICE_TYPE_B][i] * 2;
1915             }
1916         list_count[0] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L0_8x8];
1917         list_count[1] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L1_8x8];
1918         list_count[2] += h->stat.i_mb_partition[SLICE_TYPE_B][D_BI_8x8];
1919         i_mb_count[B_DIRECT] += (h->stat.i_mb_partition[SLICE_TYPE_B][D_DIRECT_8x8]+2)/4;
1920         i_mb_list_count = (list_count[0] + list_count[1] + list_count[2]) / 100.0;
1921         x264_log( h, X264_LOG_INFO,
1922                   "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",
1923                   buf,
1924                   i_mb_size[PIXEL_16x16] / (i_count*4),
1925                   (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
1926                   i_mb_size[PIXEL_8x8] / (i_count*4),
1927                   i_mb_count[B_DIRECT] / i_count,
1928                   i_mb_count[B_SKIP]   / i_count,
1929                   list_count[0] / i_mb_list_count,
1930                   list_count[1] / i_mb_list_count,
1931                   list_count[2] / i_mb_list_count );
1932     }
1933
1934     x264_ratecontrol_summary( h );
1935
1936     if( h->stat.i_slice_count[SLICE_TYPE_I] + h->stat.i_slice_count[SLICE_TYPE_P] + h->stat.i_slice_count[SLICE_TYPE_B] > 0 )
1937     {
1938 #define SUM3(p) (p[SLICE_TYPE_I] + p[SLICE_TYPE_P] + p[SLICE_TYPE_B])
1939 #define SUM3b(p,o) (p[SLICE_TYPE_I][o] + p[SLICE_TYPE_P][o] + p[SLICE_TYPE_B][o])
1940         int64_t i_i8x8 = SUM3b( h->stat.i_mb_count, I_8x8 );
1941         int64_t i_intra = i_i8x8 + SUM3b( h->stat.i_mb_count, I_4x4 )
1942                                  + SUM3b( h->stat.i_mb_count, I_16x16 );
1943         int64_t i_all_intra = i_intra + SUM3b( h->stat.i_mb_count, I_PCM);
1944         const int i_count = h->stat.i_slice_count[SLICE_TYPE_I] +
1945                             h->stat.i_slice_count[SLICE_TYPE_P] +
1946                             h->stat.i_slice_count[SLICE_TYPE_B];
1947         int64_t i_mb_count = i_count * h->mb.i_mb_count;
1948         float fps = (float) h->param.i_fps_num / h->param.i_fps_den;
1949         float f_bitrate = fps * SUM3(h->stat.i_slice_size) / i_count / 125;
1950
1951         if( h->pps->b_transform_8x8_mode )
1952         {
1953             x264_log( h, X264_LOG_INFO, "8x8 transform  intra:%.1f%%  inter:%.1f%%\n",
1954                       100. * i_i8x8 / i_intra,
1955                       100. * h->stat.i_mb_count_8x8dct[1] / h->stat.i_mb_count_8x8dct[0] );
1956         }
1957
1958         if( h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO
1959             && h->stat.i_slice_count[SLICE_TYPE_B] )
1960         {
1961             x264_log( h, X264_LOG_INFO, "direct mvs  spatial:%.1f%%  temporal:%.1f%%\n",
1962                       h->stat.i_direct_frames[1] * 100. / h->stat.i_slice_count[SLICE_TYPE_B],
1963                       h->stat.i_direct_frames[0] * 100. / h->stat.i_slice_count[SLICE_TYPE_B] );
1964         }
1965
1966         x264_log( h, X264_LOG_INFO, "coded y,uvDC,uvAC intra:%.1f%% %.1f%% %.1f%% inter:%.1f%% %.1f%% %.1f%%\n",
1967                   h->stat.i_mb_cbp[0] * 100.0 / (i_all_intra*4),
1968                   h->stat.i_mb_cbp[2] * 100.0 / (i_all_intra  ),
1969                   h->stat.i_mb_cbp[4] * 100.0 / (i_all_intra  ),
1970                   h->stat.i_mb_cbp[1] * 100.0 / ((i_mb_count - i_all_intra)*4),
1971                   h->stat.i_mb_cbp[3] * 100.0 / ((i_mb_count - i_all_intra)  ),
1972                   h->stat.i_mb_cbp[5] * 100.0 / ((i_mb_count - i_all_intra)) );
1973
1974         for( i_list = 0; i_list < 2; i_list++ )
1975         {
1976             int i_slice;
1977             for( i_slice = 0; i_slice < 2; i_slice++ )
1978             {
1979                 char *p = buf;
1980                 int64_t i_den = 0;
1981                 int i_max = 0;
1982                 for( i = 0; i < 32; i++ )
1983                     if( h->stat.i_mb_count_ref[i_slice][i_list][i] )
1984                     {
1985                         i_den += h->stat.i_mb_count_ref[i_slice][i_list][i];
1986                         i_max = i;
1987                     }
1988                 if( i_max == 0 )
1989                     continue;
1990                 for( i = 0; i <= i_max; i++ )
1991                     p += sprintf( p, " %4.1f%%", 100. * h->stat.i_mb_count_ref[i_slice][i_list][i] / i_den );
1992                 x264_log( h, X264_LOG_INFO, "ref %c L%d %s\n", "PB"[i_slice], i_list, buf );
1993             }
1994         }
1995
1996         if( h->param.analyse.b_ssim )
1997         {
1998             x264_log( h, X264_LOG_INFO,
1999                       "SSIM Mean Y:%.7f\n",
2000                       SUM3( h->stat.f_ssim_mean_y ) / i_count );
2001         }
2002         if( h->param.analyse.b_psnr )
2003         {
2004             x264_log( h, X264_LOG_INFO,
2005                       "PSNR Mean Y:%6.3f U:%6.3f V:%6.3f Avg:%6.3f Global:%6.3f kb/s:%.2f\n",
2006                       SUM3( h->stat.f_psnr_mean_y ) / i_count,
2007                       SUM3( h->stat.f_psnr_mean_u ) / i_count,
2008                       SUM3( h->stat.f_psnr_mean_v ) / i_count,
2009                       SUM3( h->stat.f_psnr_average ) / i_count,
2010                       x264_psnr( SUM3( h->stat.i_ssd_global ), i_count * i_yuv_size ),
2011                       f_bitrate );
2012         }
2013         else
2014             x264_log( h, X264_LOG_INFO, "kb/s:%.1f\n", f_bitrate );
2015     }
2016
2017     /* rc */
2018     x264_ratecontrol_delete( h );
2019
2020     /* param */
2021     if( h->param.rc.psz_stat_out )
2022         free( h->param.rc.psz_stat_out );
2023     if( h->param.rc.psz_stat_in )
2024         free( h->param.rc.psz_stat_in );
2025
2026     x264_cqm_delete( h );
2027
2028     if( h->param.i_threads > 1)
2029         h = h->thread[ h->i_thread_phase % h->param.i_threads ];
2030
2031     /* frames */
2032     for( i = 0; h->frames.current[i]; i++ )
2033     {
2034         assert( h->frames.current[i]->i_reference_count == 1 );
2035         x264_frame_delete( h->frames.current[i] );
2036     }
2037     for( i = 0; h->frames.next[i]; i++ )
2038     {
2039         assert( h->frames.next[i]->i_reference_count == 1 );
2040         x264_frame_delete( h->frames.next[i] );
2041     }
2042     for( i = 0; h->frames.unused[i]; i++ )
2043     {
2044         assert( h->frames.unused[i]->i_reference_count == 0 );
2045         x264_frame_delete( h->frames.unused[i] );
2046     }
2047
2048     h = h->thread[0];
2049
2050     for( i = h->param.i_threads - 1; i >= 0; i-- )
2051     {
2052         x264_frame_t **frame;
2053
2054         for( frame = h->thread[i]->frames.reference; *frame; frame++ )
2055         {
2056             assert( (*frame)->i_reference_count > 0 );
2057             (*frame)->i_reference_count--;
2058             if( (*frame)->i_reference_count == 0 )
2059                 x264_frame_delete( *frame );
2060         }
2061         frame = &h->thread[i]->fdec;
2062         assert( (*frame)->i_reference_count > 0 );
2063         (*frame)->i_reference_count--;
2064         if( (*frame)->i_reference_count == 0 )
2065             x264_frame_delete( *frame );
2066
2067         x264_macroblock_cache_end( h->thread[i] );
2068         x264_free( h->thread[i]->out.p_bitstream );
2069         x264_free( h->thread[i] );
2070     }
2071 }