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 #else
 370         if( h->param.i_scenecut_threshold >= 0 )
 371             h->param.b_pre_scenecut = 1;
 372 #endif
 373     }
 374
 375     if( h->param.b_interlaced )
 376     {
 377         if( h->param.analyse.i_me_method >= X264_ME_ESA )
 378         {
 379             x264_log( h, X264_LOG_WARNING, "interlace + me=esa is not implemented\n" );
 380             h->param.analyse.i_me_method = X264_ME_UMH;
 381         }
 382         if( h->param.analyse.i_direct_mv_pred > X264_DIRECT_PRED_SPATIAL )
 383         {
 384             x264_log( h, X264_LOG_WARNING, "interlace + direct=temporal is not implemented\n" );
 385             h->param.analyse.i_direct_mv_pred = X264_DIRECT_PRED_SPATIAL;
 386         }
 387     }
 388
 389     if( h->param.rc.i_rc_method < 0 || h->param.rc.i_rc_method > 2 )
 390     {
 391         x264_log( h, X264_LOG_ERROR, "no ratecontrol method specified\n" );
 392         return -1;
 393     }
 394     h->param.rc.f_rf_constant = x264_clip3f( h->param.rc.f_rf_constant, 0, 51 );
 395     h->param.rc.i_qp_constant = x264_clip3( h->param.rc.i_qp_constant, 0, 51 );
 396     if( h->param.rc.i_rc_method == X264_RC_CRF )
 397         h->param.rc.i_qp_constant = h->param.rc.f_rf_constant;
 398     if( (h->param.rc.i_rc_method == X264_RC_CQP || h->param.rc.i_rc_method == X264_RC_CRF)
 399         && h->param.rc.i_qp_constant == 0 )
 400     {
 401         h->mb.b_lossless = 1;
 402         h->param.i_cqm_preset = X264_CQM_FLAT;
 403         h->param.psz_cqm_file = NULL;
 404         h->param.rc.i_rc_method = X264_RC_CQP;
 405         h->param.rc.f_ip_factor = 1;
 406         h->param.rc.f_pb_factor = 1;
 407         h->param.analyse.b_psnr = 0;
 408         h->param.analyse.b_ssim = 0;
 409         h->param.analyse.i_chroma_qp_offset = 0;
 410         h->param.analyse.i_trellis = 0;
 411         h->param.analyse.b_fast_pskip = 0;
 412         h->param.analyse.i_noise_reduction = 0;
 413         h->param.analyse.f_psy_rd = 0;
 414         /* 8x8dct is not useful at all in CAVLC lossless */
 415         if( !h->param.b_cabac )
 416             h->param.analyse.b_transform_8x8 = 0;
 417     }
 418     if( h->param.rc.i_rc_method == X264_RC_CQP )
 419     {
 420         float qp_p = h->param.rc.i_qp_constant;
 421         float qp_i = qp_p - 6*log(h->param.rc.f_ip_factor)/log(2);
 422         float qp_b = qp_p + 6*log(h->param.rc.f_pb_factor)/log(2);
 423         h->param.rc.i_qp_min = x264_clip3( (int)(X264_MIN3( qp_p, qp_i, qp_b )), 0, 51 );
 424         h->param.rc.i_qp_max = x264_clip3( (int)(X264_MAX3( qp_p, qp_i, qp_b ) + .999), 0, 51 );
 425         h->param.rc.i_aq_mode = 0;
 426     }
 427     h->param.rc.i_qp_max = x264_clip3( h->param.rc.i_qp_max, 0, 51 );
 428     h->param.rc.i_qp_min = x264_clip3( h->param.rc.i_qp_min, 0, h->param.rc.i_qp_max );
 429
 430     if( ( h->param.i_width % 16 || h->param.i_height % 16 )
 431         && h->param.i_height != 1080 && !h->mb.b_lossless )
 432     {
 433         // There's nothing special about 1080 in that the warning still applies to it,
 434         // but chances are the user can't help it if his content is already 1080p,
 435         // so there's no point in warning in that case.
 436         x264_log( h, X264_LOG_WARNING,
 437                   "width or height not divisible by 16 (%dx%d), compression will suffer.\n",
 438                   h->param.i_width, h->param.i_height );
 439     }
 440
 441     h->param.i_frame_reference = x264_clip3( h->param.i_frame_reference, 1, 16 );
 442     if( h->param.i_keyint_max <= 0 )
 443         h->param.i_keyint_max = 1;
 444     h->param.i_keyint_min = x264_clip3( h->param.i_keyint_min, 1, h->param.i_keyint_max/2+1 );
 445     if( !h->param.analyse.i_subpel_refine && h->param.analyse.i_direct_mv_pred > X264_DIRECT_PRED_SPATIAL )
 446     {
 447         x264_log( h, X264_LOG_WARNING, "subme=0 + direct=temporal is not supported\n" );
 448         h->param.analyse.i_direct_mv_pred = X264_DIRECT_PRED_SPATIAL;
 449     }
 450     h->param.i_bframe = x264_clip3( h->param.i_bframe, 0, X264_BFRAME_MAX );
 451     h->param.i_bframe_bias = x264_clip3( h->param.i_bframe_bias, -90, 100 );
 452     h->param.b_bframe_pyramid = h->param.b_bframe_pyramid && h->param.i_bframe > 1;
 453     if( !h->param.i_bframe )
 454         h->param.i_bframe_adaptive = X264_B_ADAPT_NONE;
 455     h->param.analyse.b_weighted_bipred = h->param.analyse.b_weighted_bipred && h->param.i_bframe > 0;
 456     h->mb.b_direct_auto_write = h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO
 457                                 && h->param.i_bframe
 458                                 && ( h->param.rc.b_stat_write || !h->param.rc.b_stat_read );
 459     if( h->param.i_scenecut_threshold < 0 )
 460         h->param.b_pre_scenecut = 0;
 461
 462     h->param.i_deblocking_filter_alphac0 = x264_clip3( h->param.i_deblocking_filter_alphac0, -6, 6 );
 463     h->param.i_deblocking_filter_beta    = x264_clip3( h->param.i_deblocking_filter_beta, -6, 6 );
 464     h->param.analyse.i_luma_deadzone[0] = x264_clip3( h->param.analyse.i_luma_deadzone[0], 0, 32 );
 465     h->param.analyse.i_luma_deadzone[1] = x264_clip3( h->param.analyse.i_luma_deadzone[1], 0, 32 );
 466
 467     h->param.i_cabac_init_idc = x264_clip3( h->param.i_cabac_init_idc, 0, 2 );
 468
 469     if( h->param.i_cqm_preset < X264_CQM_FLAT || h->param.i_cqm_preset > X264_CQM_CUSTOM )
 470         h->param.i_cqm_preset = X264_CQM_FLAT;
 471
 472     if( h->param.analyse.i_me_method < X264_ME_DIA ||
 473         h->param.analyse.i_me_method > X264_ME_TESA )
 474         h->param.analyse.i_me_method = X264_ME_HEX;
 475     if( h->param.analyse.i_me_range < 4 )
 476         h->param.analyse.i_me_range = 4;
 477     if( h->param.analyse.i_me_range > 16 && h->param.analyse.i_me_method <= X264_ME_HEX )
 478         h->param.analyse.i_me_range = 16;
 479     if( h->param.analyse.i_me_method == X264_ME_TESA &&
 480         (h->mb.b_lossless || h->param.analyse.i_subpel_refine <= 1) )
 481         h->param.analyse.i_me_method = X264_ME_ESA;
 482     h->param.analyse.i_subpel_refine = x264_clip3( h->param.analyse.i_subpel_refine, 0, 9 );
 483     h->param.analyse.b_mixed_references = h->param.analyse.b_mixed_references && h->param.i_frame_reference > 1;
 484     h->param.analyse.inter &= X264_ANALYSE_PSUB16x16|X264_ANALYSE_PSUB8x8|X264_ANALYSE_BSUB16x16|
 485                               X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
 486     h->param.analyse.intra &= X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
 487     if( !(h->param.analyse.inter & X264_ANALYSE_PSUB16x16) )
 488         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
 489     if( !h->param.analyse.b_transform_8x8 )
 490     {
 491         h->param.analyse.inter &= ~X264_ANALYSE_I8x8;
 492         h->param.analyse.intra &= ~X264_ANALYSE_I8x8;
 493     }
 494     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -12, 12);
 495     if( !h->param.b_cabac )
 496         h->param.analyse.i_trellis = 0;
 497     h->param.analyse.i_trellis = x264_clip3( h->param.analyse.i_trellis, 0, 2 );
 498     if( !h->param.analyse.i_trellis )
 499         h->param.analyse.f_psy_trellis = 0;
 500     h->param.analyse.f_psy_rd = x264_clip3f( h->param.analyse.f_psy_rd, 0, 10 );
 501     h->param.analyse.f_psy_trellis = x264_clip3f( h->param.analyse.f_psy_trellis, 0, 10 );
 502     if( h->param.analyse.i_subpel_refine < 6 )
 503         h->param.analyse.f_psy_rd = 0;
 504     h->mb.i_psy_rd = FIX8( h->param.analyse.f_psy_rd );
 505     /* Psy RDO increases overall quantizers to improve the quality of luma--this indirectly hurts chroma quality */
 506     /* so we lower the chroma QP offset to compensate */
 507     /* This can be triggered repeatedly on multiple calls to parameter_validate, but since encoding
 508      * uses the pps chroma qp offset not the param chroma qp offset, this is not a problem. */
 509     if( h->mb.i_psy_rd )
 510         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_rd < 0.25 ? 1 : 2;
 511     h->mb.i_psy_trellis = FIX8( h->param.analyse.f_psy_trellis / 4 );
 512     /* Psy trellis has a similar effect. */
 513     if( h->mb.i_psy_trellis )
 514         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_trellis < 0.25 ? 1 : 2;
 515     else
 516         h->mb.i_psy_trellis = 0;
 517     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -12, 12);
 518     h->param.rc.i_aq_mode = x264_clip3( h->param.rc.i_aq_mode, 0, 1 );
 519     h->param.rc.f_aq_strength = x264_clip3f( h->param.rc.f_aq_strength, 0, 3 );
 520     if( h->param.rc.f_aq_strength == 0 )
 521         h->param.rc.i_aq_mode = 0;
 522     h->param.analyse.i_noise_reduction = x264_clip3( h->param.analyse.i_noise_reduction, 0, 1<<16 );
 523
 524     {
 525         const x264_level_t *l = x264_levels;
 526         if( h->param.i_level_idc < 0 )
 527         {
 528             if( h->param.rc.i_rc_method == X264_RC_ABR && h->param.rc.i_vbv_buffer_size <= 0 )
 529                 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate * 2;
 530             h->sps = h->sps_array;
 531             x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
 532             do h->param.i_level_idc = l->level_idc;
 533                 while( l[1].level_idc && x264_validate_levels( h, 0 ) && l++ );
 534             if( h->param.rc.i_vbv_buffer_size <= 0 )
 535                 h->param.rc.i_vbv_max_bitrate = 0;
 536         }
 537         else
 538         {
 539             while( l->level_idc && l->level_idc != h->param.i_level_idc )
 540                 l++;
 541             if( l->level_idc == 0 )
 542             {
 543                 x264_log( h, X264_LOG_ERROR, "invalid level_idc: %d\n", h->param.i_level_idc );
 544                 return -1;
 545             }
 546         }
 547         if( h->param.analyse.i_mv_range <= 0 )
 548             h->param.analyse.i_mv_range = l->mv_range >> h->param.b_interlaced;
 549         else
 550             h->param.analyse.i_mv_range = x264_clip3(h->param.analyse.i_mv_range, 32, 512 >> h->param.b_interlaced);
 551         if( h->param.analyse.i_direct_8x8_inference < 0 )
 552             h->param.analyse.i_direct_8x8_inference = l->direct8x8;
 553     }
 554
 555     if( h->param.i_threads > 1 )
 556     {
 557         int r = h->param.analyse.i_mv_range_thread;
 558         int r2;
 559         if( r <= 0 )
 560         {
 561             // half of the available space is reserved and divided evenly among the threads,
 562             // the rest is allocated to whichever thread is far enough ahead to use it.
 563             // reserving more space increases quality for some videos, but costs more time
 564             // in thread synchronization.
 565             int max_range = (h->param.i_height + X264_THREAD_HEIGHT) / h->param.i_threads - X264_THREAD_HEIGHT;
 566             r = max_range / 2;
 567         }
 568         r = X264_MAX( r, h->param.analyse.i_me_range );
 569         r = X264_MIN( r, h->param.analyse.i_mv_range );
 570         // round up to use the whole mb row
 571         r2 = (r & ~15) + ((-X264_THREAD_HEIGHT) & 15);
 572         if( r2 < r )
 573             r2 += 16;
 574         x264_log( h, X264_LOG_DEBUG, "using mv_range_thread = %d\n", r2 );
 575         h->param.analyse.i_mv_range_thread = r2;
 576     }
 577
 578     if( h->param.rc.f_qblur < 0 )
 579         h->param.rc.f_qblur = 0;
 580     if( h->param.rc.f_complexity_blur < 0 )
 581         h->param.rc.f_complexity_blur = 0;
 582
 583     h->param.i_sps_id &= 31;
 584
 585     if( h->param.i_log_level < X264_LOG_INFO )
 586     {
 587         h->param.analyse.b_psnr = 0;
 588         h->param.analyse.b_ssim = 0;
 589     }
 590
 591     /* ensure the booleans are 0 or 1 so they can be used in math */
 592 #define BOOLIFY(x) h->param.x = !!h->param.x
 593     BOOLIFY( b_cabac );
 594     BOOLIFY( b_deblocking_filter );
 595     BOOLIFY( b_interlaced );
 596     BOOLIFY( analyse.b_transform_8x8 );
 597     BOOLIFY( analyse.i_direct_8x8_inference );
 598     BOOLIFY( analyse.b_chroma_me );
 599     BOOLIFY( analyse.b_fast_pskip );
 600     BOOLIFY( rc.b_stat_write );
 601     BOOLIFY( rc.b_stat_read );
 602 #undef BOOLIFY
 603
 604     return 0;
 605 }
 606
 607 static void mbcmp_init( x264_t *h )
 608 {
 609     int satd = !h->mb.b_lossless && h->param.analyse.i_subpel_refine > 1;
 610     memcpy( h->pixf.mbcmp, satd ? h->pixf.satd : h->pixf.sad_aligned, sizeof(h->pixf.mbcmp) );
 611     memcpy( h->pixf.mbcmp_unaligned, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.mbcmp_unaligned) );
 612     h->pixf.intra_mbcmp_x3_16x16 = satd ? h->pixf.intra_satd_x3_16x16 : h->pixf.intra_sad_x3_16x16;
 613     satd &= h->param.analyse.i_me_method == X264_ME_TESA;
 614     memcpy( h->pixf.fpelcmp, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.fpelcmp) );
 615     memcpy( h->pixf.fpelcmp_x3, satd ? h->pixf.satd_x3 : h->pixf.sad_x3, sizeof(h->pixf.fpelcmp_x3) );
 616     memcpy( h->pixf.fpelcmp_x4, satd ? h->pixf.satd_x4 : h->pixf.sad_x4, sizeof(h->pixf.fpelcmp_x4) );
 617 }
 618
 619 /****************************************************************************
 620  * x264_encoder_open:
 621  ****************************************************************************/
 622 x264_t *x264_encoder_open   ( x264_param_t *param )
 623 {
 624     x264_t *h = x264_malloc( sizeof( x264_t ) );
 625     char buf[1000], *p;
 626     int i;
 627
 628     memset( h, 0, sizeof( x264_t ) );
 629
 630     /* Create a copy of param */
 631     memcpy( &h->param, param, sizeof( x264_param_t ) );
 632
 633     if( x264_validate_parameters( h ) < 0 )
 634     {
 635         x264_free( h );
 636         return NULL;
 637     }
 638
 639     if( h->param.psz_cqm_file )
 640         if( x264_cqm_parse_file( h, h->param.psz_cqm_file ) < 0 )
 641         {
 642             x264_free( h );
 643             return NULL;
 644         }
 645
 646     if( h->param.rc.psz_stat_out )
 647         h->param.rc.psz_stat_out = strdup( h->param.rc.psz_stat_out );
 648     if( h->param.rc.psz_stat_in )
 649         h->param.rc.psz_stat_in = strdup( h->param.rc.psz_stat_in );
 650
 651     /* VUI */
 652     if( h->param.vui.i_sar_width > 0 && h->param.vui.i_sar_height > 0 )
 653     {
 654         int i_w = param->vui.i_sar_width;
 655         int i_h = param->vui.i_sar_height;
 656
 657         x264_reduce_fraction( &i_w, &i_h );
 658
 659         while( i_w > 65535 || i_h > 65535 )
 660         {
 661             i_w /= 2;
 662             i_h /= 2;
 663         }
 664
 665         h->param.vui.i_sar_width = 0;
 666         h->param.vui.i_sar_height = 0;
 667         if( i_w == 0 || i_h == 0 )
 668         {
 669             x264_log( h, X264_LOG_WARNING, "cannot create valid sample aspect ratio\n" );
 670         }
 671         else
 672         {
 673             x264_log( h, X264_LOG_INFO, "using SAR=%d/%d\n", i_w, i_h );
 674             h->param.vui.i_sar_width = i_w;
 675             h->param.vui.i_sar_height = i_h;
 676         }
 677     }
 678
 679     x264_reduce_fraction( &h->param.i_fps_num, &h->param.i_fps_den );
 680
 681     /* Init x264_t */
 682     h->i_frame = 0;
 683     h->i_frame_num = 0;
 684     h->i_idr_pic_id = 0;
 685
 686     h->sps = &h->sps_array[0];
 687     x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
 688
 689     h->pps = &h->pps_array[0];
 690     x264_pps_init( h->pps, h->param.i_sps_id, &h->param, h->sps);
 691
 692     x264_validate_levels( h, 1 );
 693
 694     if( x264_cqm_init( h ) < 0 )
 695     {
 696         x264_free( h );
 697         return NULL;
 698     }
 699
 700     h->mb.i_mb_count = h->sps->i_mb_width * h->sps->i_mb_height;
 701
 702     /* Init frames. */
 703     if( h->param.i_bframe_adaptive == X264_B_ADAPT_TRELLIS )
 704         h->frames.i_delay = X264_MAX(h->param.i_bframe,3)*4 + h->param.i_threads - 1;
 705     else
 706         h->frames.i_delay = h->param.i_bframe + h->param.i_threads - 1;
 707     h->frames.i_max_ref0 = h->param.i_frame_reference;
 708     h->frames.i_max_ref1 = h->sps->vui.i_num_reorder_frames;
 709     h->frames.i_max_dpb  = h->sps->vui.i_max_dec_frame_buffering;
 710     h->frames.b_have_lowres = !h->param.rc.b_stat_read
 711         && ( h->param.rc.i_rc_method == X264_RC_ABR
 712           || h->param.rc.i_rc_method == X264_RC_CRF
 713           || h->param.i_bframe_adaptive
 714           || h->param.b_pre_scenecut );
 715     h->frames.b_have_lowres |= (h->param.rc.b_stat_read && h->param.rc.i_vbv_buffer_size > 0);
 716
 717     h->frames.i_last_idr = - h->param.i_keyint_max;
 718     h->frames.i_input    = 0;
 719     h->frames.last_nonb  = NULL;
 720
 721     h->i_ref0 = 0;
 722     h->i_ref1 = 0;
 723
 724     h->chroma_qp_table = i_chroma_qp_table + 12 + h->pps->i_chroma_qp_index_offset;
 725
 726     x264_rdo_init( );
 727
 728     /* init CPU functions */
 729     x264_predict_16x16_init( h->param.cpu, h->predict_16x16 );
 730     x264_predict_8x8c_init( h->param.cpu, h->predict_8x8c );
 731     x264_predict_8x8_init( h->param.cpu, h->predict_8x8 );
 732     x264_predict_4x4_init( h->param.cpu, h->predict_4x4 );
 733     if( !h->param.b_cabac );
 734         x264_init_vlc_tables();
 735     x264_pixel_init( h->param.cpu, &h->pixf );
 736     x264_dct_init( h->param.cpu, &h->dctf );
 737     x264_zigzag_init( h->param.cpu, &h->zigzagf, h->param.b_interlaced );
 738     x264_mc_init( h->param.cpu, &h->mc );
 739     x264_quant_init( h, h->param.cpu, &h->quantf );
 740     x264_deblock_init( h->param.cpu, &h->loopf );
 741     x264_dct_init_weights();
 742
 743     mbcmp_init( h );
 744
 745     p = buf + sprintf( buf, "using cpu capabilities:" );
 746     for( i=0; x264_cpu_names[i].flags; i++ )
 747     {
 748         if( !strcmp(x264_cpu_names[i].name, "SSE2")
 749             && param->cpu & (X264_CPU_SSE2_IS_FAST|X264_CPU_SSE2_IS_SLOW) )
 750             continue;
 751         if( !strcmp(x264_cpu_names[i].name, "SSE3")
 752             && (param->cpu & X264_CPU_SSSE3 || !(param->cpu & X264_CPU_CACHELINE_64)) )
 753             continue;
 754         if( !strcmp(x264_cpu_names[i].name, "SSE4.1")
 755             && (param->cpu & X264_CPU_SSE42) )
 756             continue;
 757         if( (param->cpu & x264_cpu_names[i].flags) == x264_cpu_names[i].flags
 758             && (!i || x264_cpu_names[i].flags != x264_cpu_names[i-1].flags) )
 759             p += sprintf( p, " %s", x264_cpu_names[i].name );
 760     }
 761     if( !param->cpu )
 762         p += sprintf( p, " none!" );
 763     x264_log( h, X264_LOG_INFO, "%s\n", buf );
 764
 765     h->out.i_nal = 0;
 766     h->out.i_bitstream = X264_MAX( 1000000, h->param.i_width * h->param.i_height * 4
 767         * ( h->param.rc.i_rc_method == X264_RC_ABR ? pow( 0.95, h->param.rc.i_qp_min )
 768           : pow( 0.95, h->param.rc.i_qp_constant ) * X264_MAX( 1, h->param.rc.f_ip_factor )));
 769
 770     h->thread[0] = h;
 771     h->i_thread_num = 0;
 772     for( i = 1; i < h->param.i_threads; i++ )
 773         h->thread[i] = x264_malloc( sizeof(x264_t) );
 774
 775     for( i = 0; i < h->param.i_threads; i++ )
 776     {
 777         if( i > 0 )
 778             *h->thread[i] = *h;
 779         h->thread[i]->fdec = x264_frame_pop_unused( h );
 780         h->thread[i]->out.p_bitstream = x264_malloc( h->out.i_bitstream );
 781         if( x264_macroblock_cache_init( h->thread[i] ) < 0 )
 782             return NULL;
 783     }
 784
 785     if( x264_ratecontrol_new( h ) < 0 )
 786         return NULL;
 787
 788     if( h->param.psz_dump_yuv )
 789     {
 790         /* create or truncate the reconstructed video file */
 791         FILE *f = fopen( h->param.psz_dump_yuv, "w" );
 792         if( f )
 793             fclose( f );
 794         else
 795         {
 796             x264_log( h, X264_LOG_ERROR, "can't write to fdec.yuv\n" );
 797             x264_free( h );
 798             return NULL;
 799         }
 800     }
 801
 802     x264_log( h, X264_LOG_INFO, "profile %s, level %d.%d\n",
 803         h->sps->i_profile_idc == PROFILE_BASELINE ? "Baseline" :
 804         h->sps->i_profile_idc == PROFILE_MAIN ? "Main" :
 805         h->sps->i_profile_idc == PROFILE_HIGH ? "High" :
 806         "High 4:4:4 Predictive", h->sps->i_level_idc/10, h->sps->i_level_idc%10 );
 807
 808     return h;
 809 }
 810
 811 /****************************************************************************
 812  * x264_encoder_reconfig:
 813  ****************************************************************************/
 814 int x264_encoder_reconfig( x264_t *h, x264_param_t *param )
 815 {
 816 #define COPY(var) h->param.var = param->var
 817     COPY( i_frame_reference ); // but never uses more refs than initially specified
 818     COPY( i_bframe_bias );
 819     if( h->param.i_scenecut_threshold >= 0 && param->i_scenecut_threshold >= 0 )
 820         COPY( i_scenecut_threshold ); // can't turn it on or off, only vary the threshold
 821     COPY( b_deblocking_filter );
 822     COPY( i_deblocking_filter_alphac0 );
 823     COPY( i_deblocking_filter_beta );
 824     COPY( analyse.intra );
 825     COPY( analyse.inter );
 826     COPY( analyse.i_direct_mv_pred );
 827     COPY( analyse.i_me_range );
 828     COPY( analyse.i_noise_reduction );
 829     /* We can't switch out of subme=0 during encoding. */
 830     if( h->param.analyse.i_subpel_refine )
 831         COPY( analyse.i_subpel_refine );
 832     COPY( analyse.i_trellis );
 833     COPY( analyse.b_chroma_me );
 834     COPY( analyse.b_dct_decimate );
 835     COPY( analyse.b_fast_pskip );
 836     COPY( analyse.b_mixed_references );
 837     COPY( analyse.f_psy_rd );
 838     COPY( analyse.f_psy_trellis );
 839     // can only twiddle these if they were enabled to begin with:
 840     if( h->param.analyse.i_me_method >= X264_ME_ESA || param->analyse.i_me_method < X264_ME_ESA )
 841         COPY( analyse.i_me_method );
 842     if( h->pps->b_transform_8x8_mode )
 843         COPY( analyse.b_transform_8x8 );
 844     if( h->frames.i_max_ref1 > 1 )
 845         COPY( b_bframe_pyramid );
 846 #undef COPY
 847
 848     mbcmp_init( h );
 849
 850     return x264_validate_parameters( h );
 851 }
 852
 853 /* internal usage */
 854 static void x264_nal_start( x264_t *h, int i_type, int i_ref_idc )
 855 {
 856     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
 857
 858     nal->i_ref_idc = i_ref_idc;
 859     nal->i_type    = i_type;
 860
 861     nal->i_payload= 0;
 862     nal->p_payload= &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8];
 863 }
 864 static void x264_nal_end( x264_t *h )
 865 {
 866     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
 867     nal->i_payload = &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8] - nal->p_payload;
 868     h->out.i_nal++;
 869 }
 870
 871 /****************************************************************************
 872  * x264_encoder_headers:
 873  ****************************************************************************/
 874 int x264_encoder_headers( x264_t *h, x264_nal_t **pp_nal, int *pi_nal )
 875 {
 876     /* init bitstream context */
 877     h->out.i_nal = 0;
 878     bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
 879
 880     /* Put SPS and PPS */
 881     if( h->i_frame == 0 )
 882     {
 883         /* identify ourself */
 884         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
 885         x264_sei_version_write( h, &h->out.bs );
 886         x264_nal_end( h );
 887
 888         /* generate sequence parameters */
 889         x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
 890         x264_sps_write( &h->out.bs, h->sps );
 891         x264_nal_end( h );
 892
 893         /* generate picture parameters */
 894         x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
 895         x264_pps_write( &h->out.bs, h->pps );
 896         x264_nal_end( h );
 897     }
 898     /* now set output*/
 899     *pi_nal = h->out.i_nal;
 900     *pp_nal = &h->out.nal[0];
 901     h->out.i_nal = 0;
 902
 903     return 0;
 904 }
 905
 906 static inline void x264_reference_build_list( x264_t *h, int i_poc )
 907 {
 908     int i;
 909     int b_ok;
 910
 911     /* build ref list 0/1 */
 912     h->i_ref0 = 0;
 913     h->i_ref1 = 0;
 914     for( i = 0; h->frames.reference[i]; i++ )
 915     {
 916         if( h->frames.reference[i]->i_poc < i_poc )
 917         {
 918             h->fref0[h->i_ref0++] = h->frames.reference[i];
 919         }
 920         else if( h->frames.reference[i]->i_poc > i_poc )
 921         {
 922             h->fref1[h->i_ref1++] = h->frames.reference[i];
 923         }
 924     }
 925
 926     /* Order ref0 from higher to lower poc */
 927     do
 928     {
 929         b_ok = 1;
 930         for( i = 0; i < h->i_ref0 - 1; i++ )
 931         {
 932             if( h->fref0[i]->i_poc < h->fref0[i+1]->i_poc )
 933             {
 934                 XCHG( x264_frame_t*, h->fref0[i], h->fref0[i+1] );
 935                 b_ok = 0;
 936                 break;
 937             }
 938         }
 939     } while( !b_ok );
 940     /* Order ref1 from lower to higher poc (bubble sort) for B-frame */
 941     do
 942     {
 943         b_ok = 1;
 944         for( i = 0; i < h->i_ref1 - 1; i++ )
 945         {
 946             if( h->fref1[i]->i_poc > h->fref1[i+1]->i_poc )
 947             {
 948                 XCHG( x264_frame_t*, h->fref1[i], h->fref1[i+1] );
 949                 b_ok = 0;
 950                 break;
 951             }
 952         }
 953     } while( !b_ok );
 954
 955     /* In the standard, a P-frame's ref list is sorted by frame_num.
 956      * We use POC, but check whether explicit reordering is needed */
 957     h->b_ref_reorder[0] =
 958     h->b_ref_reorder[1] = 0;
 959     if( h->sh.i_type == SLICE_TYPE_P )
 960     {
 961         for( i = 0; i < h->i_ref0 - 1; i++ )
 962             if( h->fref0[i]->i_frame_num < h->fref0[i+1]->i_frame_num )
 963             {
 964                 h->b_ref_reorder[0] = 1;
 965                 break;
 966             }
 967     }
 968
 969     h->i_ref1 = X264_MIN( h->i_ref1, h->frames.i_max_ref1 );
 970     h->i_ref0 = X264_MIN( h->i_ref0, h->frames.i_max_ref0 );
 971     h->i_ref0 = X264_MIN( h->i_ref0, h->param.i_frame_reference ); // if reconfig() has lowered the limit
 972     assert( h->i_ref0 + h->i_ref1 <= 16 );
 973     h->mb.pic.i_fref[0] = h->i_ref0;
 974     h->mb.pic.i_fref[1] = h->i_ref1;
 975 }
 976
 977 static void x264_fdec_filter_row( x264_t *h, int mb_y )
 978 {
 979     /* mb_y is the mb to be encoded next, not the mb to be filtered here */
 980     int b_hpel = h->fdec->b_kept_as_ref;
 981     int b_deblock = !h->sh.i_disable_deblocking_filter_idc;
 982     int b_end = mb_y == h->sps->i_mb_height;
 983     int min_y = mb_y - (1 << h->sh.b_mbaff);
 984     int max_y = b_end ? h->sps->i_mb_height : mb_y;
 985     b_deblock &= b_hpel || h->param.psz_dump_yuv;
 986     if( mb_y & h->sh.b_mbaff )
 987         return;
 988     if( min_y < 0 )
 989         return;
 990
 991     if( !b_end )
 992     {
 993         int i, j;
 994         for( j=0; j<=h->sh.b_mbaff; j++ )
 995             for( i=0; i<3; i++ )
 996             {
 997                 memcpy( h->mb.intra_border_backup[j][i],
 998                         h->fdec->plane[i] + ((mb_y*16 >> !!i) + j - 1 - h->sh.b_mbaff) * h->fdec->i_stride[i],
 999                         h->sps->i_mb_width*16 >> !!i );
1000             }
1001     }
1002
1003     if( b_deblock )
1004     {
1005         int y;
1006         for( y = min_y; y < max_y; y += (1 << h->sh.b_mbaff) )
1007             x264_frame_deblock_row( h, y );
1008     }
1009
1010     if( b_hpel )
1011     {
1012         x264_frame_expand_border( h, h->fdec, min_y, b_end );
1013         if( h->param.analyse.i_subpel_refine )
1014         {
1015             x264_frame_filter( h, h->fdec, min_y, b_end );
1016             x264_frame_expand_border_filtered( h, h->fdec, min_y, b_end );
1017         }
1018     }
1019
1020     if( h->param.i_threads > 1 && h->fdec->b_kept_as_ref )
1021     {
1022         x264_frame_cond_broadcast( h->fdec, mb_y*16 + (b_end ? 10000 : -(X264_THREAD_HEIGHT << h->sh.b_mbaff)) );
1023     }
1024
1025     min_y = X264_MAX( min_y*16-8, 0 );
1026     max_y = b_end ? h->param.i_height : mb_y*16-8;
1027
1028     if( h->param.analyse.b_psnr )
1029     {
1030         int i;
1031         for( i=0; i<3; i++ )
1032             h->stat.frame.i_ssd[i] +=
1033                 x264_pixel_ssd_wxh( &h->pixf,
1034                     h->fdec->plane[i] + (min_y>>!!i) * h->fdec->i_stride[i], h->fdec->i_stride[i],
1035                     h->fenc->plane[i] + (min_y>>!!i) * h->fenc->i_stride[i], h->fenc->i_stride[i],
1036                     h->param.i_width >> !!i, (max_y-min_y) >> !!i );
1037     }
1038
1039     if( h->param.analyse.b_ssim )
1040     {
1041         x264_emms();
1042         /* offset by 2 pixels to avoid alignment of ssim blocks with dct blocks,
1043          * and overlap by 4 */
1044         min_y += min_y == 0 ? 2 : -6;
1045         h->stat.frame.f_ssim +=
1046             x264_pixel_ssim_wxh( &h->pixf,
1047                 h->fdec->plane[0] + 2+min_y*h->fdec->i_stride[0], h->fdec->i_stride[0],
1048                 h->fenc->plane[0] + 2+min_y*h->fenc->i_stride[0], h->fenc->i_stride[0],
1049                 h->param.i_width-2, max_y-min_y );
1050     }
1051 }
1052
1053 static inline void x264_reference_update( x264_t *h )
1054 {
1055     int i;
1056
1057     if( h->fdec->i_frame >= 0 )
1058         h->i_frame++;
1059
1060     if( !h->fdec->b_kept_as_ref )
1061     {
1062         if( h->param.i_threads > 1 )
1063         {
1064             x264_frame_push_unused( h, h->fdec );
1065             h->fdec = x264_frame_pop_unused( h );
1066         }
1067         return;
1068     }
1069
1070     /* move lowres copy of the image to the ref frame */
1071     for( i = 0; i < 4; i++)
1072     {
1073         XCHG( uint8_t*, h->fdec->lowres[i], h->fenc->lowres[i] );
1074         XCHG( uint8_t*, h->fdec->buffer_lowres[i], h->fenc->buffer_lowres[i] );
1075     }
1076
1077     /* adaptive B decision needs a pointer, since it can't use the ref lists */
1078     if( h->sh.i_type != SLICE_TYPE_B )
1079         h->frames.last_nonb = h->fdec;
1080
1081     /* move frame in the buffer */
1082     x264_frame_push( h->frames.reference, h->fdec );
1083     if( h->frames.reference[h->frames.i_max_dpb] )
1084         x264_frame_push_unused( h, x264_frame_shift( h->frames.reference ) );
1085     h->fdec = x264_frame_pop_unused( h );
1086 }
1087
1088 static inline void x264_reference_reset( x264_t *h )
1089 {
1090     while( h->frames.reference[0] )
1091         x264_frame_push_unused( h, x264_frame_pop( h->frames.reference ) );
1092     h->fdec->i_poc =
1093     h->fenc->i_poc = 0;
1094 }
1095
1096 static inline void x264_slice_init( x264_t *h, int i_nal_type, int i_global_qp )
1097 {
1098     /* ------------------------ Create slice header  ----------------------- */
1099     if( i_nal_type == NAL_SLICE_IDR )
1100     {
1101         x264_slice_header_init( h, &h->sh, h->sps, h->pps, h->i_idr_pic_id, h->i_frame_num, i_global_qp );
1102
1103         /* increment id */
1104         h->i_idr_pic_id = ( h->i_idr_pic_id + 1 ) % 65536;
1105     }
1106     else
1107     {
1108         x264_slice_header_init( h, &h->sh, h->sps, h->pps, -1, h->i_frame_num, i_global_qp );
1109
1110         /* always set the real higher num of ref frame used */
1111         h->sh.b_num_ref_idx_override = 1;
1112         h->sh.i_num_ref_idx_l0_active = h->i_ref0 <= 0 ? 1 : h->i_ref0;
1113         h->sh.i_num_ref_idx_l1_active = h->i_ref1 <= 0 ? 1 : h->i_ref1;
1114     }
1115
1116     h->fdec->i_frame_num = h->sh.i_frame_num;
1117
1118     if( h->sps->i_poc_type == 0 )
1119     {
1120         h->sh.i_poc_lsb = h->fdec->i_poc & ( (1 << h->sps->i_log2_max_poc_lsb) - 1 );
1121         h->sh.i_delta_poc_bottom = 0;   /* XXX won't work for field */
1122     }
1123     else if( h->sps->i_poc_type == 1 )
1124     {
1125         /* FIXME TODO FIXME */
1126     }
1127     else
1128     {
1129         /* Nothing to do ? */
1130     }
1131
1132     x264_macroblock_slice_init( h );
1133 }
1134
1135 static void x264_slice_write( x264_t *h )
1136 {
1137     int i_skip;
1138     int mb_xy, i_mb_x, i_mb_y;
1139     int i, i_list, i_ref;
1140
1141     /* init stats */
1142     memset( &h->stat.frame, 0, sizeof(h->stat.frame) );
1143
1144     /* Slice */
1145     x264_nal_start( h, h->i_nal_type, h->i_nal_ref_idc );
1146
1147     /* Slice header */
1148     x264_slice_header_write( &h->out.bs, &h->sh, h->i_nal_ref_idc );
1149     if( h->param.b_cabac )
1150     {
1151         /* alignment needed */
1152         bs_align_1( &h->out.bs );
1153
1154         /* init cabac */
1155         x264_cabac_context_init( &h->cabac, h->sh.i_type, h->sh.i_qp, h->sh.i_cabac_init_idc );
1156         x264_cabac_encode_init ( &h->cabac, h->out.bs.p, h->out.bs.p_end );
1157     }
1158     h->mb.i_last_qp = h->sh.i_qp;
1159     h->mb.i_last_dqp = 0;
1160
1161     i_mb_y = h->sh.i_first_mb / h->sps->i_mb_width;
1162     i_mb_x = h->sh.i_first_mb % h->sps->i_mb_width;
1163     i_skip = 0;
1164
1165     while( (mb_xy = i_mb_x + i_mb_y * h->sps->i_mb_width) < h->sh.i_last_mb )
1166     {
1167         int mb_spos = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
1168
1169         if( i_mb_x == 0 )
1170             x264_fdec_filter_row( h, i_mb_y );
1171
1172         /* load cache */
1173         x264_macroblock_cache_load( h, i_mb_x, i_mb_y );
1174
1175         /* analyse parameters
1176          * Slice I: choose I_4x4 or I_16x16 mode
1177          * Slice P: choose between using P mode or intra (4x4 or 16x16)
1178          * */
1179         x264_macroblock_analyse( h );
1180
1181         /* encode this macroblock -> be careful it can change the mb type to P_SKIP if needed */
1182         x264_macroblock_encode( h );
1183
1184         x264_bitstream_check_buffer( h );
1185
1186         if( h->param.b_cabac )
1187         {
1188             if( mb_xy > h->sh.i_first_mb && !(h->sh.b_mbaff && (i_mb_y&1)) )
1189                 x264_cabac_encode_terminal( &h->cabac );
1190
1191             if( IS_SKIP( h->mb.i_type ) )
1192                 x264_cabac_mb_skip( h, 1 );
1193             else
1194             {
1195                 if( h->sh.i_type != SLICE_TYPE_I )
1196                     x264_cabac_mb_skip( h, 0 );
1197                 x264_macroblock_write_cabac( h, &h->cabac );
1198             }
1199         }
1200         else
1201         {
1202             if( IS_SKIP( h->mb.i_type ) )
1203                 i_skip++;
1204             else
1205             {
1206                 if( h->sh.i_type != SLICE_TYPE_I )
1207                 {
1208                     bs_write_ue( &h->out.bs, i_skip );  /* skip run */
1209                     i_skip = 0;
1210                 }
1211                 x264_macroblock_write_cavlc( h, &h->out.bs );
1212             }
1213         }
1214
1215 #if VISUALIZE
1216         if( h->param.b_visualize )
1217             x264_visualize_mb( h );
1218 #endif
1219
1220         /* save cache */
1221         x264_macroblock_cache_save( h );
1222
1223         /* accumulate mb stats */
1224         h->stat.frame.i_mb_count[h->mb.i_type]++;
1225         if( !IS_SKIP(h->mb.i_type) && !IS_INTRA(h->mb.i_type) && !IS_DIRECT(h->mb.i_type) )
1226         {
1227             if( h->mb.i_partition != D_8x8 )
1228                 h->stat.frame.i_mb_partition[h->mb.i_partition] += 4;
1229             else
1230                 for( i = 0; i < 4; i++ )
1231                     h->stat.frame.i_mb_partition[h->mb.i_sub_partition[i]] ++;
1232             if( h->param.i_frame_reference > 1 )
1233                 for( i_list = 0; i_list <= (h->sh.i_type == SLICE_TYPE_B); i_list++ )
1234                     for( i = 0; i < 4; i++ )
1235                     {
1236                         i_ref = h->mb.cache.ref[i_list][ x264_scan8[4*i] ];
1237                         if( i_ref >= 0 )
1238                             h->stat.frame.i_mb_count_ref[i_list][i_ref] ++;
1239                     }
1240         }
1241         if( h->mb.i_cbp_luma && !IS_INTRA(h->mb.i_type) )
1242         {
1243             h->stat.frame.i_mb_count_8x8dct[0] ++;
1244             h->stat.frame.i_mb_count_8x8dct[1] += h->mb.b_transform_8x8;
1245         }
1246
1247         x264_ratecontrol_mb( h, bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac) - mb_spos );
1248
1249         if( h->sh.b_mbaff )
1250         {
1251             i_mb_x += i_mb_y & 1;
1252             i_mb_y ^= i_mb_x < h->sps->i_mb_width;
1253         }
1254         else
1255             i_mb_x++;
1256         if(i_mb_x == h->sps->i_mb_width)
1257         {
1258             i_mb_y++;
1259             i_mb_x = 0;
1260         }
1261     }
1262
1263     if( h->param.b_cabac )
1264     {
1265         x264_cabac_encode_flush( h, &h->cabac );
1266         h->out.bs.p = h->cabac.p;
1267     }
1268     else
1269     {
1270         if( i_skip > 0 )
1271             bs_write_ue( &h->out.bs, i_skip );  /* last skip run */
1272         /* rbsp_slice_trailing_bits */
1273         bs_rbsp_trailing( &h->out.bs );
1274     }
1275
1276     x264_nal_end( h );
1277
1278     x264_fdec_filter_row( h, h->sps->i_mb_height );
1279
1280     /* Compute misc bits */
1281     h->stat.frame.i_misc_bits = bs_pos( &h->out.bs )
1282                               + NALU_OVERHEAD * 8
1283                               - h->stat.frame.i_tex_bits
1284                               - h->stat.frame.i_mv_bits;
1285 }
1286
1287 static void x264_thread_sync_context( x264_t *dst, x264_t *src )
1288 {
1289     x264_frame_t **f;
1290     if( dst == src )
1291         return;
1292
1293     // reference counting
1294     for( f = src->frames.reference; *f; f++ )
1295         (*f)->i_reference_count++;
1296     for( f = dst->frames.reference; *f; f++ )
1297         x264_frame_push_unused( src, *f );
1298     src->fdec->i_reference_count++;
1299     x264_frame_push_unused( src, dst->fdec );
1300
1301     // copy everything except the per-thread pointers and the constants.
1302     memcpy( &dst->i_frame, &src->i_frame, offsetof(x264_t, mb.type) - offsetof(x264_t, i_frame) );
1303     dst->stat = src->stat;
1304 }
1305
1306 static void x264_thread_sync_stat( x264_t *dst, x264_t *src )
1307 {
1308     if( dst == src )
1309         return;
1310     memcpy( &dst->stat.i_slice_count, &src->stat.i_slice_count, sizeof(dst->stat) - sizeof(dst->stat.frame) );
1311 }
1312
1313 static int x264_slices_write( x264_t *h )
1314 {
1315     int i_frame_size;
1316
1317 #ifdef HAVE_MMX
1318     /* Misalign mask has to be set separately for each thread. */
1319     if( h->param.cpu&X264_CPU_SSE_MISALIGN )
1320         x264_cpu_mask_misalign_sse();
1321 #endif
1322
1323 #if VISUALIZE
1324     if( h->param.b_visualize )
1325         x264_visualize_init( h );
1326 #endif
1327
1328     x264_stack_align( x264_slice_write, h );
1329     i_frame_size = h->out.nal[h->out.i_nal-1].i_payload;
1330
1331 #if VISUALIZE
1332     if( h->param.b_visualize )
1333     {
1334         x264_visualize_show( h );
1335         x264_visualize_close( h );
1336     }
1337 #endif
1338
1339     h->out.i_frame_size = i_frame_size;
1340     return 0;
1341 }
1342
1343 /****************************************************************************
1344  * x264_encoder_encode:
1345  *  XXX: i_poc   : is the poc of the current given picture
1346  *       i_frame : is the number of the frame being coded
1347  *  ex:  type frame poc
1348  *       I      0   2*0
1349  *       P      1   2*3
1350  *       B      2   2*1
1351  *       B      3   2*2
1352  *       P      4   2*6
1353  *       B      5   2*4
1354  *       B      6   2*5
1355  ****************************************************************************/
1356 int     x264_encoder_encode( x264_t *h,
1357                              x264_nal_t **pp_nal, int *pi_nal,
1358                              x264_picture_t *pic_in,
1359                              x264_picture_t *pic_out )
1360 {
1361     x264_t *thread_current, *thread_prev, *thread_oldest;
1362     int     i_nal_type;
1363     int     i_nal_ref_idc;
1364
1365     int   i_global_qp;
1366
1367     if( h->param.i_threads > 1)
1368     {
1369         int i = ++h->i_thread_phase;
1370         int t = h->param.i_threads;
1371         thread_current = h->thread[ i%t ];
1372         thread_prev    = h->thread[ (i-1)%t ];
1373         thread_oldest  = h->thread[ (i+1)%t ];
1374         x264_thread_sync_context( thread_current, thread_prev );
1375         x264_thread_sync_ratecontrol( thread_current, thread_prev, thread_oldest );
1376         h = thread_current;
1377 //      fprintf(stderr, "current: %p  prev: %p  oldest: %p \n", thread_current, thread_prev, thread_oldest);
1378     }
1379     else
1380     {
1381         thread_current =
1382         thread_prev    =
1383         thread_oldest  = h;
1384     }
1385
1386     // ok to call this before encoding any frames, since the initial values of fdec have b_kept_as_ref=0
1387     x264_reference_update( h );
1388     h->fdec->i_lines_completed = -1;
1389
1390     /* no data out */
1391     *pi_nal = 0;
1392     *pp_nal = NULL;
1393
1394     /* ------------------- Setup new frame from picture -------------------- */
1395     if( pic_in != NULL )
1396     {
1397         /* 1: Copy the picture to a frame and move it to a buffer */
1398         x264_frame_t *fenc = x264_frame_pop_unused( h );
1399
1400         if( x264_frame_copy_picture( h, fenc, pic_in ) < 0 )
1401             return -1;
1402
1403         if( h->param.i_width != 16 * h->sps->i_mb_width ||
1404             h->param.i_height != 16 * h->sps->i_mb_height )
1405             x264_frame_expand_border_mod16( h, fenc );
1406
1407         fenc->i_frame = h->frames.i_input++;
1408
1409         x264_frame_push( h->frames.next, fenc );
1410
1411         if( h->frames.b_have_lowres )
1412             x264_frame_init_lowres( h, fenc );
1413
1414         if( h->param.rc.i_aq_mode )
1415             x264_adaptive_quant_frame( h, fenc );
1416
1417         if( h->frames.i_input <= h->frames.i_delay + 1 - h->param.i_threads )
1418         {
1419             /* Nothing yet to encode */
1420             /* waiting for filling bframe buffer */
1421             pic_out->i_type = X264_TYPE_AUTO;
1422             return 0;
1423         }
1424     }
1425
1426     if( h->frames.current[0] == NULL )
1427     {
1428         int bframes = 0;
1429         /* 2: Select frame types */
1430         if( h->frames.next[0] == NULL )
1431         {
1432             x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
1433             return 0;
1434         }
1435
1436         x264_slicetype_decide( h );
1437
1438         /* 3: move some B-frames and 1 non-B to encode queue */
1439         while( IS_X264_TYPE_B( h->frames.next[bframes]->i_type ) )
1440             bframes++;
1441         x264_frame_push( h->frames.current, x264_frame_shift( &h->frames.next[bframes] ) );
1442         /* FIXME: when max B-frames > 3, BREF may no longer be centered after GOP closing */
1443         if( h->param.b_bframe_pyramid && bframes > 1 )
1444         {
1445             x264_frame_t *mid = x264_frame_shift( &h->frames.next[bframes/2] );
1446             mid->i_type = X264_TYPE_BREF;
1447             x264_frame_push( h->frames.current, mid );
1448             bframes--;
1449         }
1450         while( bframes-- )
1451             x264_frame_push( h->frames.current, x264_frame_shift( h->frames.next ) );
1452     }
1453
1454     /* ------------------- Get frame to be encoded ------------------------- */
1455     /* 4: get picture to encode */
1456     h->fenc = x264_frame_shift( h->frames.current );
1457     if( h->fenc == NULL )
1458     {
1459         /* Nothing yet to encode (ex: waiting for I/P with B frames) */
1460         /* waiting for filling bframe buffer */
1461         pic_out->i_type = X264_TYPE_AUTO;
1462         return 0;
1463     }
1464
1465 do_encode:
1466
1467     if( h->fenc->i_type == X264_TYPE_IDR )
1468     {
1469         h->frames.i_last_idr = h->fenc->i_frame;
1470     }
1471
1472     /* ------------------- Setup frame context ----------------------------- */
1473     /* 5: Init data dependent of frame type */
1474     if( h->fenc->i_type == X264_TYPE_IDR )
1475     {
1476         /* reset ref pictures */
1477         x264_reference_reset( h );
1478
1479         i_nal_type    = NAL_SLICE_IDR;
1480         i_nal_ref_idc = NAL_PRIORITY_HIGHEST;
1481         h->sh.i_type = SLICE_TYPE_I;
1482     }
1483     else if( h->fenc->i_type == X264_TYPE_I )
1484     {
1485         i_nal_type    = NAL_SLICE;
1486         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
1487         h->sh.i_type = SLICE_TYPE_I;
1488     }
1489     else if( h->fenc->i_type == X264_TYPE_P )
1490     {
1491         i_nal_type    = NAL_SLICE;
1492         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
1493         h->sh.i_type = SLICE_TYPE_P;
1494     }
1495     else if( h->fenc->i_type == X264_TYPE_BREF )
1496     {
1497         i_nal_type    = NAL_SLICE;
1498         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* maybe add MMCO to forget it? -> low */
1499         h->sh.i_type = SLICE_TYPE_B;
1500     }
1501     else    /* B frame */
1502     {
1503         i_nal_type    = NAL_SLICE;
1504         i_nal_ref_idc = NAL_PRIORITY_DISPOSABLE;
1505         h->sh.i_type = SLICE_TYPE_B;
1506     }
1507
1508     h->fdec->i_poc =
1509     h->fenc->i_poc = 2 * (h->fenc->i_frame - h->frames.i_last_idr);
1510     h->fdec->i_type = h->fenc->i_type;
1511     h->fdec->i_frame = h->fenc->i_frame;
1512     h->fenc->b_kept_as_ref =
1513     h->fdec->b_kept_as_ref = i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE && h->param.i_keyint_max > 1;
1514
1515
1516
1517     /* ------------------- Init                ----------------------------- */
1518     /* build ref list 0/1 */
1519     x264_reference_build_list( h, h->fdec->i_poc );
1520
1521     /* Init the rate control */
1522     x264_ratecontrol_start( h, h->fenc->i_qpplus1 );
1523     i_global_qp = x264_ratecontrol_qp( h );
1524
1525     pic_out->i_qpplus1 =
1526     h->fdec->i_qpplus1 = i_global_qp + 1;
1527
1528     if( h->sh.i_type == SLICE_TYPE_B )
1529         x264_macroblock_bipred_init( h );
1530
1531     /* ------------------------ Create slice header  ----------------------- */
1532     x264_slice_init( h, i_nal_type, i_global_qp );
1533
1534     if( i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE )
1535         h->i_frame_num++;
1536
1537     /* ---------------------- Write the bitstream -------------------------- */
1538     /* Init bitstream context */
1539     h->out.i_nal = 0;
1540     bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
1541
1542     if(h->param.b_aud){
1543         int pic_type;
1544
1545         if(h->sh.i_type == SLICE_TYPE_I)
1546             pic_type = 0;
1547         else if(h->sh.i_type == SLICE_TYPE_P)
1548             pic_type = 1;
1549         else if(h->sh.i_type == SLICE_TYPE_B)
1550             pic_type = 2;
1551         else
1552             pic_type = 7;
1553
1554         x264_nal_start(h, NAL_AUD, NAL_PRIORITY_DISPOSABLE);
1555         bs_write(&h->out.bs, 3, pic_type);
1556         bs_rbsp_trailing(&h->out.bs);
1557         x264_nal_end(h);
1558     }
1559
1560     h->i_nal_type = i_nal_type;
1561     h->i_nal_ref_idc = i_nal_ref_idc;
1562
1563     /* Write SPS and PPS */
1564     if( i_nal_type == NAL_SLICE_IDR && h->param.b_repeat_headers )
1565     {
1566         if( h->fenc->i_frame == 0 )
1567         {
1568             /* identify ourself */
1569             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
1570             x264_sei_version_write( h, &h->out.bs );
1571             x264_nal_end( h );
1572         }
1573
1574         /* generate sequence parameters */
1575         x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
1576         x264_sps_write( &h->out.bs, h->sps );
1577         x264_nal_end( h );
1578
1579         /* generate picture parameters */
1580         x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
1581         x264_pps_write( &h->out.bs, h->pps );
1582         x264_nal_end( h );
1583     }
1584
1585     /* Write frame */
1586     if( h->param.i_threads > 1 )
1587     {
1588         x264_pthread_create( &h->thread_handle, NULL, (void*)x264_slices_write, h );
1589         h->b_thread_active = 1;
1590     }
1591     else
1592         x264_slices_write( h );
1593
1594     /* restore CPU state (before using float again) */
1595     x264_emms();
1596
1597     if( h->sh.i_type == SLICE_TYPE_P && !h->param.rc.b_stat_read
1598         && h->param.i_scenecut_threshold >= 0
1599         && !h->param.b_pre_scenecut )
1600     {
1601         const int *mbs = h->stat.frame.i_mb_count;
1602         int i_mb_i = mbs[I_16x16] + mbs[I_8x8] + mbs[I_4x4];
1603         int i_mb_p = mbs[P_L0] + mbs[P_8x8];
1604         int i_mb_s = mbs[P_SKIP];
1605         int i_mb   = h->sps->i_mb_width * h->sps->i_mb_height;
1606         int64_t i_inter_cost = h->stat.frame.i_inter_cost;
1607         int64_t i_intra_cost = h->stat.frame.i_intra_cost;
1608
1609         float f_bias;
1610         int i_gop_size = h->fenc->i_frame - h->frames.i_last_idr;
1611         float f_thresh_max = h->param.i_scenecut_threshold / 100.0;
1612         /* magic numbers pulled out of thin air */
1613         float f_thresh_min = f_thresh_max * h->param.i_keyint_min
1614                              / ( h->param.i_keyint_max * 4 );
1615         if( h->param.i_keyint_min == h->param.i_keyint_max )
1616              f_thresh_min= f_thresh_max;
1617
1618         /* macroblock_analyse() doesn't further analyse skipped mbs,
1619          * so we have to guess their cost */
1620         if( h->stat.frame.i_mbs_analysed > 0 )
1621             i_intra_cost = i_intra_cost * i_mb / h->stat.frame.i_mbs_analysed;
1622
1623         if( i_gop_size < h->param.i_keyint_min / 4 )
1624             f_bias = f_thresh_min / 4;
1625         else if( i_gop_size <= h->param.i_keyint_min )
1626             f_bias = f_thresh_min * i_gop_size / h->param.i_keyint_min;
1627         else
1628         {
1629             f_bias = f_thresh_min
1630                      + ( f_thresh_max - f_thresh_min )
1631                        * ( i_gop_size - h->param.i_keyint_min )
1632                        / ( h->param.i_keyint_max - h->param.i_keyint_min );
1633         }
1634         f_bias = X264_MIN( f_bias, 1.0 );
1635
1636         /* Bad P will be reencoded as I */
1637         if( h->stat.frame.i_mbs_analysed > 0 &&
1638             i_inter_cost >= (1.0 - f_bias) * i_intra_cost )
1639         {
1640             int b;
1641
1642             x264_log( h, X264_LOG_DEBUG, "scene cut at %d Icost:%.0f Pcost:%.0f ratio:%.4f bias:%.4f gop:%d (imb:%d pmb:%d smb:%d)\n",
1643                       h->fenc->i_frame,
1644                       (double)i_intra_cost, (double)i_inter_cost,
1645                       1. - (double)i_inter_cost / i_intra_cost,
1646                       f_bias, i_gop_size,
1647                       i_mb_i, i_mb_p, i_mb_s );
1648
1649             /* Restore frame num */
1650             h->i_frame_num--;
1651
1652             for( b = 0; h->frames.current[b] && IS_X264_TYPE_B( h->frames.current[b]->i_type ); b++ );
1653             if( b > 0 )
1654             {
1655                 /* If using B-frames, force GOP to be closed.
1656                  * Even if this frame is going to be I and not IDR, forcing a
1657                  * P-frame before the scenecut will probably help compression.
1658                  *
1659                  * We don't yet know exactly which frame is the scene cut, so
1660                  * we can't assign an I-frame. Instead, change the previous
1661                  * B-frame to P, and rearrange coding order. */
1662
1663                 if( h->param.i_bframe_adaptive || b > 1 )
1664                     h->fenc->i_type = X264_TYPE_AUTO;
1665                 x264_frame_sort_pts( h->frames.current );
1666                 x264_frame_unshift( h->frames.next, h->fenc );
1667                 h->fenc = h->frames.current[b-1];
1668                 h->frames.current[b-1] = NULL;
1669                 h->fenc->i_type = X264_TYPE_P;
1670                 x264_frame_sort_dts( h->frames.current );
1671             }
1672             /* Do IDR if needed */
1673             else if( i_gop_size >= h->param.i_keyint_min )
1674             {
1675                 /* Reset */
1676                 h->i_frame_num = 0;
1677
1678                 /* Reinit field of fenc */
1679                 h->fenc->i_type = X264_TYPE_IDR;
1680                 h->fenc->i_poc = 0;
1681
1682                 /* Put enqueued frames back in the pool */
1683                 while( h->frames.current[0] )
1684                     x264_frame_push( h->frames.next, x264_frame_shift( h->frames.current ) );
1685                 x264_frame_sort_pts( h->frames.next );
1686             }
1687             else
1688             {
1689                 h->fenc->i_type = X264_TYPE_I;
1690             }
1691             goto do_encode;
1692         }
1693     }
1694
1695     x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
1696     return 0;
1697 }
1698
1699 static void x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
1700                                     x264_nal_t **pp_nal, int *pi_nal,
1701                                     x264_picture_t *pic_out )
1702 {
1703     int i, i_list;
1704     char psz_message[80];
1705
1706     if( h->b_thread_active )
1707     {
1708         x264_pthread_join( h->thread_handle, NULL );
1709         h->b_thread_active = 0;
1710     }
1711     if( !h->out.i_nal )
1712     {
1713         pic_out->i_type = X264_TYPE_AUTO;
1714         return;
1715     }
1716
1717     x264_frame_push_unused( thread_current, h->fenc );
1718
1719     /* End bitstream, set output  */
1720     *pi_nal = h->out.i_nal;
1721     *pp_nal = h->out.nal;
1722     h->out.i_nal = 0;
1723
1724     /* Set output picture properties */
1725     if( h->sh.i_type == SLICE_TYPE_I )
1726         pic_out->i_type = h->i_nal_type == NAL_SLICE_IDR ? X264_TYPE_IDR : X264_TYPE_I;
1727     else if( h->sh.i_type == SLICE_TYPE_P )
1728         pic_out->i_type = X264_TYPE_P;
1729     else
1730         pic_out->i_type = X264_TYPE_B;
1731     pic_out->i_pts = h->fenc->i_pts;
1732
1733     pic_out->img.i_plane = h->fdec->i_plane;
1734     for(i = 0; i < 4; i++){
1735         pic_out->img.i_stride[i] = h->fdec->i_stride[i];
1736         pic_out->img.plane[i] = h->fdec->plane[i];
1737     }
1738
1739     /* ---------------------- Update encoder state ------------------------- */
1740
1741     /* update rc */
1742     x264_emms();
1743     x264_ratecontrol_end( h, h->out.i_frame_size * 8 );
1744
1745     /* restore CPU state (before using float again) */
1746     x264_emms();
1747
1748     x264_noise_reduction_update( thread_current );
1749
1750     /* ---------------------- Compute/Print statistics --------------------- */
1751     x264_thread_sync_stat( h, h->thread[0] );
1752
1753     /* Slice stat */
1754     h->stat.i_slice_count[h->sh.i_type]++;
1755     h->stat.i_slice_size[h->sh.i_type] += h->out.i_frame_size + NALU_OVERHEAD;
1756     h->stat.f_slice_qp[h->sh.i_type] += h->fdec->f_qp_avg_aq;
1757
1758     for( i = 0; i < X264_MBTYPE_MAX; i++ )
1759         h->stat.i_mb_count[h->sh.i_type][i] += h->stat.frame.i_mb_count[i];
1760     for( i = 0; i < X264_PARTTYPE_MAX; i++ )
1761         h->stat.i_mb_partition[h->sh.i_type][i] += h->stat.frame.i_mb_partition[i];
1762     for( i = 0; i < 2; i++ )
1763         h->stat.i_mb_count_8x8dct[i] += h->stat.frame.i_mb_count_8x8dct[i];
1764     if( h->sh.i_type != SLICE_TYPE_I )
1765         for( i_list = 0; i_list < 2; i_list++ )
1766             for( i = 0; i < 32; i++ )
1767                 h->stat.i_mb_count_ref[h->sh.i_type][i_list][i] += h->stat.frame.i_mb_count_ref[i_list][i];
1768     if( h->sh.i_type == SLICE_TYPE_P )
1769         h->stat.i_consecutive_bframes[h->fdec->i_frame - h->fref0[0]->i_frame - 1]++;
1770     if( h->sh.i_type == SLICE_TYPE_B )
1771     {
1772         h->stat.i_direct_frames[ h->sh.b_direct_spatial_mv_pred ] ++;
1773         if( h->mb.b_direct_auto_write )
1774         {
1775             //FIXME somewhat arbitrary time constants
1776             if( h->stat.i_direct_score[0] + h->stat.i_direct_score[1] > h->mb.i_mb_count )
1777             {
1778                 for( i = 0; i < 2; i++ )
1779                     h->stat.i_direct_score[i] = h->stat.i_direct_score[i] * 9/10;
1780             }
1781             for( i = 0; i < 2; i++ )
1782                 h->stat.i_direct_score[i] += h->stat.frame.i_direct_score[i];
1783         }
1784     }
1785
1786     psz_message[0] = '\0';
1787     if( h->param.analyse.b_psnr )
1788     {
1789         int64_t ssd[3] = {
1790             h->stat.frame.i_ssd[0],
1791             h->stat.frame.i_ssd[1],
1792             h->stat.frame.i_ssd[2],
1793         };
1794
1795         h->stat.i_ssd_global[h->sh.i_type] += ssd[0] + ssd[1] + ssd[2];
1796         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 );
1797         h->stat.f_psnr_mean_y[h->sh.i_type] += x264_psnr( ssd[0], h->param.i_width * h->param.i_height );
1798         h->stat.f_psnr_mean_u[h->sh.i_type] += x264_psnr( ssd[1], h->param.i_width * h->param.i_height / 4 );
1799         h->stat.f_psnr_mean_v[h->sh.i_type] += x264_psnr( ssd[2], h->param.i_width * h->param.i_height / 4 );
1800
1801         snprintf( psz_message, 80, " PSNR Y:%5.2f U:%5.2f V:%5.2f",
1802                   x264_psnr( ssd[0], h->param.i_width * h->param.i_height ),
1803                   x264_psnr( ssd[1], h->param.i_width * h->param.i_height / 4),
1804                   x264_psnr( ssd[2], h->param.i_width * h->param.i_height / 4) );
1805     }
1806
1807     if( h->param.analyse.b_ssim )
1808     {
1809         double ssim_y = h->stat.frame.f_ssim
1810                       / (((h->param.i_width-6)>>2) * ((h->param.i_height-6)>>2));
1811         h->stat.f_ssim_mean_y[h->sh.i_type] += ssim_y;
1812         snprintf( psz_message + strlen(psz_message), 80 - strlen(psz_message),
1813                   " SSIM Y:%.5f", ssim_y );
1814     }
1815     psz_message[79] = '\0';
1816
1817     x264_log( h, X264_LOG_DEBUG,
1818                   "frame=%4d QP=%.2f NAL=%d Slice:%c Poc:%-3d I:%-4d P:%-4d SKIP:%-4d size=%d bytes%s\n",
1819               h->i_frame,
1820               h->fdec->f_qp_avg_aq,
1821               h->i_nal_ref_idc,
1822               h->sh.i_type == SLICE_TYPE_I ? 'I' : (h->sh.i_type == SLICE_TYPE_P ? 'P' : 'B' ),
1823               h->fdec->i_poc,
1824               h->stat.frame.i_mb_count_i,
1825               h->stat.frame.i_mb_count_p,
1826               h->stat.frame.i_mb_count_skip,
1827               h->out.i_frame_size,
1828               psz_message );
1829
1830     // keep stats all in one place
1831     x264_thread_sync_stat( h->thread[0], h );
1832     // for the use of the next frame
1833     x264_thread_sync_stat( thread_current, h );
1834
1835 #ifdef DEBUG_MB_TYPE
1836 {
1837     static const char mb_chars[] = { 'i', 'i', 'I', 'C', 'P', '8', 'S',
1838         'D', '<', 'X', 'B', 'X', '>', 'B', 'B', 'B', 'B', '8', 'S' };
1839     int mb_xy;
1840     for( mb_xy = 0; mb_xy < h->sps->i_mb_width * h->sps->i_mb_height; mb_xy++ )
1841     {
1842         if( h->mb.type[mb_xy] < X264_MBTYPE_MAX && h->mb.type[mb_xy] >= 0 )
1843             fprintf( stderr, "%c ", mb_chars[ h->mb.type[mb_xy] ] );
1844         else
1845             fprintf( stderr, "? " );
1846
1847         if( (mb_xy+1) % h->sps->i_mb_width == 0 )
1848             fprintf( stderr, "\n" );
1849     }
1850 }
1851 #endif
1852
1853     if( h->param.psz_dump_yuv )
1854         x264_frame_dump( h );
1855 }
1856
1857 static void x264_print_intra( int64_t *i_mb_count, double i_count, int b_print_pcm, char *intra )
1858 {
1859     intra += sprintf( intra, "I16..4%s: %4.1f%% %4.1f%% %4.1f%%",
1860         b_print_pcm ? "..PCM" : "",
1861         i_mb_count[I_16x16]/ i_count,
1862         i_mb_count[I_8x8]  / i_count,
1863         i_mb_count[I_4x4]  / i_count );
1864     if( b_print_pcm )
1865         sprintf( intra, " %4.1f%%", i_mb_count[I_PCM]  / i_count );
1866 }
1867
1868 /****************************************************************************
1869  * x264_encoder_close:
1870  ****************************************************************************/
1871 void    x264_encoder_close  ( x264_t *h )
1872 {
1873     int64_t i_yuv_size = 3 * h->param.i_width * h->param.i_height / 2;
1874     int64_t i_mb_count_size[2][7] = {{0}};
1875     char buf[200];
1876     int i, j, i_list, i_type;
1877     int b_print_pcm = h->stat.i_mb_count[SLICE_TYPE_I][I_PCM]
1878                    || h->stat.i_mb_count[SLICE_TYPE_P][I_PCM]
1879                    || h->stat.i_mb_count[SLICE_TYPE_B][I_PCM];
1880
1881     for( i=0; i<h->param.i_threads; i++ )
1882     {
1883         // don't strictly have to wait for the other threads, but it's simpler than canceling them
1884         if( h->thread[i]->b_thread_active )
1885         {
1886             x264_pthread_join( h->thread[i]->thread_handle, NULL );
1887             assert( h->thread[i]->fenc->i_reference_count == 1 );
1888             x264_frame_delete( h->thread[i]->fenc );
1889         }
1890     }
1891
1892     /* Slices used and PSNR */
1893     for( i=0; i<5; i++ )
1894     {
1895         static const int slice_order[] = { SLICE_TYPE_I, SLICE_TYPE_SI, SLICE_TYPE_P, SLICE_TYPE_SP, SLICE_TYPE_B };
1896         static const char *slice_name[] = { "P", "B", "I", "SP", "SI" };
1897         int i_slice = slice_order[i];
1898
1899         if( h->stat.i_slice_count[i_slice] > 0 )
1900         {
1901             const int i_count = h->stat.i_slice_count[i_slice];
1902             if( h->param.analyse.b_psnr )
1903             {
1904                 x264_log( h, X264_LOG_INFO,
1905                           "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",
1906                           slice_name[i_slice],
1907                           i_count,
1908                           h->stat.f_slice_qp[i_slice] / i_count,
1909                           (double)h->stat.i_slice_size[i_slice] / i_count,
1910                           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,
1911                           h->stat.f_psnr_average[i_slice] / i_count,
1912                           x264_psnr( h->stat.i_ssd_global[i_slice], i_count * i_yuv_size ) );
1913             }
1914             else
1915             {
1916                 x264_log( h, X264_LOG_INFO,
1917                           "slice %s:%-5d Avg QP:%5.2f  size:%6.0f\n",
1918                           slice_name[i_slice],
1919                           i_count,
1920                           h->stat.f_slice_qp[i_slice] / i_count,
1921                           (double)h->stat.i_slice_size[i_slice] / i_count );
1922             }
1923         }
1924     }
1925     if( h->param.i_bframe && h->stat.i_slice_count[SLICE_TYPE_P] )
1926     {
1927         char *p = buf;
1928         int den = 0;
1929         // weight by number of frames (including the P-frame) that are in a sequence of N B-frames
1930         for( i=0; i<=h->param.i_bframe; i++ )
1931             den += (i+1) * h->stat.i_consecutive_bframes[i];
1932         for( i=0; i<=h->param.i_bframe; i++ )
1933             p += sprintf( p, " %4.1f%%", 100. * (i+1) * h->stat.i_consecutive_bframes[i] / den );
1934         x264_log( h, X264_LOG_INFO, "consecutive B-frames:%s\n", buf );
1935     }
1936
1937     for( i_type = 0; i_type < 2; i_type++ )
1938         for( i = 0; i < X264_PARTTYPE_MAX; i++ )
1939         {
1940             if( i == D_DIRECT_8x8 ) continue; /* direct is counted as its own type */
1941             i_mb_count_size[i_type][x264_mb_partition_pixel_table[i]] += h->stat.i_mb_partition[i_type][i];
1942         }
1943
1944     /* MB types used */
1945     if( h->stat.i_slice_count[SLICE_TYPE_I] > 0 )
1946     {
1947         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_I];
1948         double i_count = h->stat.i_slice_count[SLICE_TYPE_I] * h->mb.i_mb_count / 100.0;
1949         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
1950         x264_log( h, X264_LOG_INFO, "mb I  %s\n", buf );
1951     }
1952     if( h->stat.i_slice_count[SLICE_TYPE_P] > 0 )
1953     {
1954         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_P];
1955         double i_count = h->stat.i_slice_count[SLICE_TYPE_P] * h->mb.i_mb_count / 100.0;
1956         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_P];
1957         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
1958         x264_log( h, X264_LOG_INFO,
1959                   "mb P  %s  P16..4: %4.1f%% %4.1f%% %4.1f%% %4.1f%% %4.1f%%    skip:%4.1f%%\n",
1960                   buf,
1961                   i_mb_size[PIXEL_16x16] / (i_count*4),
1962                   (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
1963                   i_mb_size[PIXEL_8x8] / (i_count*4),
1964                   (i_mb_size[PIXEL_8x4] + i_mb_size[PIXEL_4x8]) / (i_count*4),
1965                   i_mb_size[PIXEL_4x4] / (i_count*4),
1966                   i_mb_count[P_SKIP] / i_count );
1967     }
1968     if( h->stat.i_slice_count[SLICE_TYPE_B] > 0 )
1969     {
1970         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_B];
1971         double i_count = h->stat.i_slice_count[SLICE_TYPE_B] * h->mb.i_mb_count / 100.0;
1972         double i_mb_list_count;
1973         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_B];
1974         int64_t list_count[3] = {0}; /* 0 == L0, 1 == L1, 2 == BI */
1975         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
1976         for( i = 0; i < X264_PARTTYPE_MAX; i++ )
1977             for( j = 0; j < 2; j++ )
1978             {
1979                 int l0 = x264_mb_type_list0_table[i][j];
1980                 int l1 = x264_mb_type_list1_table[i][j];
1981                 if( l0 || l1 )
1982                     list_count[l1+l0*l1] += h->stat.i_mb_count[SLICE_TYPE_B][i] * 2;
1983             }
1984         list_count[0] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L0_8x8];
1985         list_count[1] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L1_8x8];
1986         list_count[2] += h->stat.i_mb_partition[SLICE_TYPE_B][D_BI_8x8];
1987         i_mb_count[B_DIRECT] += (h->stat.i_mb_partition[SLICE_TYPE_B][D_DIRECT_8x8]+2)/4;
1988         i_mb_list_count = (list_count[0] + list_count[1] + list_count[2]) / 100.0;
1989         x264_log( h, X264_LOG_INFO,
1990                   "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",
1991                   buf,
1992                   i_mb_size[PIXEL_16x16] / (i_count*4),
1993                   (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
1994                   i_mb_size[PIXEL_8x8] / (i_count*4),
1995                   i_mb_count[B_DIRECT] / i_count,
1996                   i_mb_count[B_SKIP]   / i_count,
1997                   list_count[0] / i_mb_list_count,
1998                   list_count[1] / i_mb_list_count,
1999                   list_count[2] / i_mb_list_count );
2000     }
2001
2002     x264_ratecontrol_summary( h );
2003
2004     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 )
2005     {
2006         const int i_count = h->stat.i_slice_count[SLICE_TYPE_I] +
2007                             h->stat.i_slice_count[SLICE_TYPE_P] +
2008                             h->stat.i_slice_count[SLICE_TYPE_B];
2009         float fps = (float) h->param.i_fps_num / h->param.i_fps_den;
2010 #define SUM3(p) (p[SLICE_TYPE_I] + p[SLICE_TYPE_P] + p[SLICE_TYPE_B])
2011 #define SUM3b(p,o) (p[SLICE_TYPE_I][o] + p[SLICE_TYPE_P][o] + p[SLICE_TYPE_B][o])
2012         float f_bitrate = fps * SUM3(h->stat.i_slice_size) / i_count / 125;
2013
2014         if( h->pps->b_transform_8x8_mode )
2015         {
2016             int64_t i_i8x8 = SUM3b( h->stat.i_mb_count, I_8x8 );
2017             int64_t i_intra = i_i8x8 + SUM3b( h->stat.i_mb_count, I_4x4 )
2018                                      + SUM3b( h->stat.i_mb_count, I_16x16 );
2019             x264_log( h, X264_LOG_INFO, "8x8 transform  intra:%.1f%%  inter:%.1f%%\n",
2020                       100. * i_i8x8 / i_intra,
2021                       100. * h->stat.i_mb_count_8x8dct[1] / h->stat.i_mb_count_8x8dct[0] );
2022         }
2023
2024         if( h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO
2025             && h->stat.i_slice_count[SLICE_TYPE_B] )
2026         {
2027             x264_log( h, X264_LOG_INFO, "direct mvs  spatial:%.1f%%  temporal:%.1f%%\n",
2028                       h->stat.i_direct_frames[1] * 100. / h->stat.i_slice_count[SLICE_TYPE_B],
2029                       h->stat.i_direct_frames[0] * 100. / h->stat.i_slice_count[SLICE_TYPE_B] );
2030         }
2031
2032         for( i_list = 0; i_list < 2; i_list++ )
2033         {
2034             int i_slice;
2035             for( i_slice = 0; i_slice < 2; i_slice++ )
2036             {
2037                 char *p = buf;
2038                 int64_t i_den = 0;
2039                 int i_max = 0;
2040                 for( i = 0; i < 32; i++ )
2041                     if( h->stat.i_mb_count_ref[i_slice][i_list][i] )
2042                     {
2043                         i_den += h->stat.i_mb_count_ref[i_slice][i_list][i];
2044                         i_max = i;
2045                     }
2046                 if( i_max == 0 )
2047                     continue;
2048                 for( i = 0; i <= i_max; i++ )
2049                     p += sprintf( p, " %4.1f%%", 100. * h->stat.i_mb_count_ref[i_slice][i_list][i] / i_den );
2050                 x264_log( h, X264_LOG_INFO, "ref %c L%d %s\n", "PB"[i_slice], i_list, buf );
2051             }
2052         }
2053
2054         if( h->param.analyse.b_ssim )
2055         {
2056             x264_log( h, X264_LOG_INFO,
2057                       "SSIM Mean Y:%.7f\n",
2058                       SUM3( h->stat.f_ssim_mean_y ) / i_count );
2059         }
2060         if( h->param.analyse.b_psnr )
2061         {
2062             x264_log( h, X264_LOG_INFO,
2063                       "PSNR Mean Y:%6.3f U:%6.3f V:%6.3f Avg:%6.3f Global:%6.3f kb/s:%.2f\n",
2064                       SUM3( h->stat.f_psnr_mean_y ) / i_count,
2065                       SUM3( h->stat.f_psnr_mean_u ) / i_count,
2066                       SUM3( h->stat.f_psnr_mean_v ) / i_count,
2067                       SUM3( h->stat.f_psnr_average ) / i_count,
2068                       x264_psnr( SUM3( h->stat.i_ssd_global ), i_count * i_yuv_size ),
2069                       f_bitrate );
2070         }
2071         else
2072             x264_log( h, X264_LOG_INFO, "kb/s:%.1f\n", f_bitrate );
2073     }
2074
2075     /* rc */
2076     x264_ratecontrol_delete( h );
2077
2078     /* param */
2079     if( h->param.rc.psz_stat_out )
2080         free( h->param.rc.psz_stat_out );
2081     if( h->param.rc.psz_stat_in )
2082         free( h->param.rc.psz_stat_in );
2083
2084     x264_cqm_delete( h );
2085
2086     if( h->param.i_threads > 1)
2087         h = h->thread[ h->i_thread_phase % h->param.i_threads ];
2088
2089     /* frames */
2090     for( i = 0; h->frames.current[i]; i++ )
2091     {
2092         assert( h->frames.current[i]->i_reference_count == 1 );
2093         x264_frame_delete( h->frames.current[i] );
2094     }
2095     for( i = 0; h->frames.next[i]; i++ )
2096     {
2097         assert( h->frames.next[i]->i_reference_count == 1 );
2098         x264_frame_delete( h->frames.next[i] );
2099     }
2100     for( i = 0; h->frames.unused[i]; i++ )
2101     {
2102         assert( h->frames.unused[i]->i_reference_count == 0 );
2103         x264_frame_delete( h->frames.unused[i] );
2104     }
2105
2106     h = h->thread[0];
2107
2108     for( i = h->param.i_threads - 1; i >= 0; i-- )
2109     {
2110         x264_frame_t **frame;
2111
2112         for( frame = h->thread[i]->frames.reference; *frame; frame++ )
2113         {
2114             assert( (*frame)->i_reference_count > 0 );
2115             (*frame)->i_reference_count--;
2116             if( (*frame)->i_reference_count == 0 )
2117                 x264_frame_delete( *frame );
2118         }
2119         frame = &h->thread[i]->fdec;
2120         assert( (*frame)->i_reference_count > 0 );
2121         (*frame)->i_reference_count--;
2122         if( (*frame)->i_reference_count == 0 )
2123             x264_frame_delete( *frame );
2124
2125         x264_macroblock_cache_end( h->thread[i] );
2126         x264_free( h->thread[i]->out.p_bitstream );
2127         x264_free( h->thread[i] );
2128     }
2129 }