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 #include "me.h"
  35
  36 #if VISUALIZE
  37 #include "common/visualize.h"
  38 #endif
  39
  40 //#define DEBUG_MB_TYPE
  41
  42 #define NALU_OVERHEAD 5 // startcode + NAL type costs 5 bytes per frame
  43
  44 #define bs_write_ue bs_write_ue_big
  45
  46 static int x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
  47                                    x264_nal_t **pp_nal, int *pi_nal,
  48                                    x264_picture_t *pic_out );
  49
  50 /****************************************************************************
  51  *
  52  ******************************* x264 libs **********************************
  53  *
  54  ****************************************************************************/
  55 static float x264_psnr( int64_t i_sqe, int64_t i_size )
  56 {
  57     double f_mse = (double)i_sqe / ((double)65025.0 * (double)i_size);
  58     if( f_mse <= 0.0000000001 ) /* Max 100dB */
  59         return 100;
  60
  61     return (float)(-10.0 * log( f_mse ) / log( 10.0 ));
  62 }
  63
  64 static void x264_frame_dump( x264_t *h )
  65 {
  66     FILE *f = fopen( h->param.psz_dump_yuv, "r+b" );
  67     int i, y;
  68     if( !f )
  69         return;
  70     /* Write the frame in display order */
  71     fseek( f, (uint64_t)h->fdec->i_frame * h->param.i_height * h->param.i_width * 3/2, SEEK_SET );
  72     for( i = 0; i < h->fdec->i_plane; i++ )
  73         for( y = 0; y < h->param.i_height >> !!i; y++ )
  74             fwrite( &h->fdec->plane[i][y*h->fdec->i_stride[i]], 1, h->param.i_width >> !!i, f );
  75     fclose( f );
  76 }
  77
  78
  79 /* Fill "default" values */
  80 static void x264_slice_header_init( x264_t *h, x264_slice_header_t *sh,
  81                                     x264_sps_t *sps, x264_pps_t *pps,
  82                                     int i_idr_pic_id, int i_frame, int i_qp )
  83 {
  84     x264_param_t *param = &h->param;
  85     int i;
  86
  87     /* First we fill all field */
  88     sh->sps = sps;
  89     sh->pps = pps;
  90
  91     sh->i_first_mb  = 0;
  92     sh->i_last_mb   = h->mb.i_mb_count - 1;
  93     sh->i_pps_id    = pps->i_id;
  94
  95     sh->i_frame_num = i_frame;
  96
  97     sh->b_mbaff = h->param.b_interlaced;
  98     sh->b_field_pic = 0;    /* no field support for now */
  99     sh->b_bottom_field = 0; /* not yet used */
 100
 101     sh->i_idr_pic_id = i_idr_pic_id;
 102
 103     /* poc stuff, fixed later */
 104     sh->i_poc_lsb = 0;
 105     sh->i_delta_poc_bottom = 0;
 106     sh->i_delta_poc[0] = 0;
 107     sh->i_delta_poc[1] = 0;
 108
 109     sh->i_redundant_pic_cnt = 0;
 110
 111     if( !h->mb.b_direct_auto_read )
 112     {
 113         if( h->mb.b_direct_auto_write )
 114             sh->b_direct_spatial_mv_pred = ( h->stat.i_direct_score[1] > h->stat.i_direct_score[0] );
 115         else
 116             sh->b_direct_spatial_mv_pred = ( param->analyse.i_direct_mv_pred == X264_DIRECT_PRED_SPATIAL );
 117     }
 118     /* else b_direct_spatial_mv_pred was read from the 2pass statsfile */
 119
 120     sh->b_num_ref_idx_override = 0;
 121     sh->i_num_ref_idx_l0_active = 1;
 122     sh->i_num_ref_idx_l1_active = 1;
 123
 124     sh->b_ref_pic_list_reordering_l0 = h->b_ref_reorder[0];
 125     sh->b_ref_pic_list_reordering_l1 = h->b_ref_reorder[1];
 126
 127     /* If the ref list isn't in the default order, construct reordering header */
 128     /* List1 reordering isn't needed yet */
 129     if( sh->b_ref_pic_list_reordering_l0 )
 130     {
 131         int pred_frame_num = i_frame;
 132         for( i = 0; i < h->i_ref0; i++ )
 133         {
 134             int diff = h->fref0[i]->i_frame_num - pred_frame_num;
 135             if( diff == 0 )
 136                 x264_log( h, X264_LOG_ERROR, "diff frame num == 0\n" );
 137             sh->ref_pic_list_order[0][i].idc = ( diff > 0 );
 138             sh->ref_pic_list_order[0][i].arg = abs( diff ) - 1;
 139             pred_frame_num = h->fref0[i]->i_frame_num;
 140         }
 141     }
 142
 143     sh->i_cabac_init_idc = param->i_cabac_init_idc;
 144
 145     sh->i_qp = i_qp;
 146     sh->i_qp_delta = i_qp - pps->i_pic_init_qp;
 147     sh->b_sp_for_swidth = 0;
 148     sh->i_qs_delta = 0;
 149
 150     /* If effective qp <= 15, deblocking would have no effect anyway */
 151     if( param->b_deblocking_filter
 152         && ( h->mb.b_variable_qp
 153         || 15 < i_qp + 2 * X264_MIN(param->i_deblocking_filter_alphac0, param->i_deblocking_filter_beta) ) )
 154     {
 155         sh->i_disable_deblocking_filter_idc = 0;
 156     }
 157     else
 158     {
 159         sh->i_disable_deblocking_filter_idc = 1;
 160     }
 161     sh->i_alpha_c0_offset = param->i_deblocking_filter_alphac0 << 1;
 162     sh->i_beta_offset = param->i_deblocking_filter_beta << 1;
 163 }
 164
 165 static void x264_slice_header_write( bs_t *s, x264_slice_header_t *sh, int i_nal_ref_idc )
 166 {
 167     int i;
 168
 169     if( sh->b_mbaff )
 170     {
 171         assert( sh->i_first_mb % (2*sh->sps->i_mb_width) == 0 );
 172         bs_write_ue( s, sh->i_first_mb >> 1 );
 173     }
 174     else
 175         bs_write_ue( s, sh->i_first_mb );
 176
 177     bs_write_ue( s, sh->i_type + 5 );   /* same type things */
 178     bs_write_ue( s, sh->i_pps_id );
 179     bs_write( s, sh->sps->i_log2_max_frame_num, sh->i_frame_num & ((1<<sh->sps->i_log2_max_frame_num)-1) );
 180
 181     if( !sh->sps->b_frame_mbs_only )
 182     {
 183         bs_write1( s, sh->b_field_pic );
 184         if( sh->b_field_pic )
 185             bs_write1( s, sh->b_bottom_field );
 186     }
 187
 188     if( sh->i_idr_pic_id >= 0 ) /* NAL IDR */
 189     {
 190         bs_write_ue( s, sh->i_idr_pic_id );
 191     }
 192
 193     if( sh->sps->i_poc_type == 0 )
 194     {
 195         bs_write( s, sh->sps->i_log2_max_poc_lsb, sh->i_poc_lsb & ((1<<sh->sps->i_log2_max_poc_lsb)-1) );
 196         if( sh->pps->b_pic_order && !sh->b_field_pic )
 197         {
 198             bs_write_se( s, sh->i_delta_poc_bottom );
 199         }
 200     }
 201     else if( sh->sps->i_poc_type == 1 && !sh->sps->b_delta_pic_order_always_zero )
 202     {
 203         bs_write_se( s, sh->i_delta_poc[0] );
 204         if( sh->pps->b_pic_order && !sh->b_field_pic )
 205         {
 206             bs_write_se( s, sh->i_delta_poc[1] );
 207         }
 208     }
 209
 210     if( sh->pps->b_redundant_pic_cnt )
 211     {
 212         bs_write_ue( s, sh->i_redundant_pic_cnt );
 213     }
 214
 215     if( sh->i_type == SLICE_TYPE_B )
 216     {
 217         bs_write1( s, sh->b_direct_spatial_mv_pred );
 218     }
 219     if( sh->i_type == SLICE_TYPE_P || sh->i_type == SLICE_TYPE_SP || sh->i_type == SLICE_TYPE_B )
 220     {
 221         bs_write1( s, sh->b_num_ref_idx_override );
 222         if( sh->b_num_ref_idx_override )
 223         {
 224             bs_write_ue( s, sh->i_num_ref_idx_l0_active - 1 );
 225             if( sh->i_type == SLICE_TYPE_B )
 226             {
 227                 bs_write_ue( s, sh->i_num_ref_idx_l1_active - 1 );
 228             }
 229         }
 230     }
 231
 232     /* ref pic list reordering */
 233     if( sh->i_type != SLICE_TYPE_I )
 234     {
 235         bs_write1( s, sh->b_ref_pic_list_reordering_l0 );
 236         if( sh->b_ref_pic_list_reordering_l0 )
 237         {
 238             for( i = 0; i < sh->i_num_ref_idx_l0_active; i++ )
 239             {
 240                 bs_write_ue( s, sh->ref_pic_list_order[0][i].idc );
 241                 bs_write_ue( s, sh->ref_pic_list_order[0][i].arg );
 242
 243             }
 244             bs_write_ue( s, 3 );
 245         }
 246     }
 247     if( sh->i_type == SLICE_TYPE_B )
 248     {
 249         bs_write1( s, sh->b_ref_pic_list_reordering_l1 );
 250         if( sh->b_ref_pic_list_reordering_l1 )
 251         {
 252             for( i = 0; i < sh->i_num_ref_idx_l1_active; i++ )
 253             {
 254                 bs_write_ue( s, sh->ref_pic_list_order[1][i].idc );
 255                 bs_write_ue( s, sh->ref_pic_list_order[1][i].arg );
 256             }
 257             bs_write_ue( s, 3 );
 258         }
 259     }
 260
 261     if( sh->pps->b_weighted_pred && ( sh->i_type == SLICE_TYPE_P || sh->i_type == SLICE_TYPE_SP ) )
 262     {
 263         /* pred_weight_table() */
 264         bs_write_ue( s, sh->weight[0][0].i_denom );
 265         bs_write_ue( s, sh->weight[0][1].i_denom );
 266         for( i = 0; i < sh->i_num_ref_idx_l0_active; i++ )
 267         {
 268             int luma_weight_l0_flag = !!sh->weight[i][0].weightfn;
 269             int chroma_weight_l0_flag = !!sh->weight[i][1].weightfn || !!sh->weight[i][2].weightfn;
 270             bs_write1( s, luma_weight_l0_flag );
 271             if( luma_weight_l0_flag )
 272             {
 273                 bs_write_se( s, sh->weight[i][0].i_scale );
 274                 bs_write_se( s, sh->weight[i][0].i_offset );
 275             }
 276             bs_write1( s, chroma_weight_l0_flag );
 277             if( chroma_weight_l0_flag )
 278             {
 279                 int j;
 280                 for( j = 1; j < 3; j++ )
 281                 {
 282                     bs_write_se( s, sh->weight[i][j].i_scale );
 283                     bs_write_se( s, sh->weight[i][j].i_offset );
 284                 }
 285             }
 286         }
 287     }
 288     else if( sh->pps->b_weighted_bipred == 1 && sh->i_type == SLICE_TYPE_B )
 289     {
 290       /* TODO */
 291     }
 292
 293     if( i_nal_ref_idc != 0 )
 294     {
 295         if( sh->i_idr_pic_id >= 0 )
 296         {
 297             bs_write1( s, 0 );  /* no output of prior pics flag */
 298             bs_write1( s, 0 );  /* long term reference flag */
 299         }
 300         else
 301         {
 302             bs_write1( s, sh->i_mmco_command_count > 0 ); /* adaptive_ref_pic_marking_mode_flag */
 303             if( sh->i_mmco_command_count > 0 )
 304             {
 305                 int i;
 306                 for( i = 0; i < sh->i_mmco_command_count; i++ )
 307                 {
 308                     bs_write_ue( s, 1 ); /* mark short term ref as unused */
 309                     bs_write_ue( s, sh->mmco[i].i_difference_of_pic_nums - 1 );
 310                 }
 311                 bs_write_ue( s, 0 ); /* end command list */
 312             }
 313         }
 314     }
 315
 316     if( sh->pps->b_cabac && sh->i_type != SLICE_TYPE_I )
 317     {
 318         bs_write_ue( s, sh->i_cabac_init_idc );
 319     }
 320     bs_write_se( s, sh->i_qp_delta );      /* slice qp delta */
 321
 322     if( sh->pps->b_deblocking_filter_control )
 323     {
 324         bs_write_ue( s, sh->i_disable_deblocking_filter_idc );
 325         if( sh->i_disable_deblocking_filter_idc != 1 )
 326         {
 327             bs_write_se( s, sh->i_alpha_c0_offset >> 1 );
 328             bs_write_se( s, sh->i_beta_offset >> 1 );
 329         }
 330     }
 331 }
 332
 333 /* If we are within a reasonable distance of the end of the memory allocated for the bitstream, */
 334 /* reallocate, adding an arbitrary amount of space (100 kilobytes). */
 335 static int x264_bitstream_check_buffer( x264_t *h )
 336 {
 337     uint8_t *bs_bak = h->out.p_bitstream;
 338     if( ( h->param.b_cabac && (h->cabac.p_end - h->cabac.p < 2500) )
 339      || ( h->out.bs.p_end - h->out.bs.p < 2500 ) )
 340     {
 341         intptr_t delta;
 342         int i;
 343
 344         h->out.i_bitstream += 100000;
 345         CHECKED_MALLOC( h->out.p_bitstream, h->out.i_bitstream );
 346         h->mc.memcpy_aligned( h->out.p_bitstream, bs_bak, (h->out.i_bitstream - 100000) & ~15 );
 347         delta = h->out.p_bitstream - bs_bak;
 348
 349         h->out.bs.p_start += delta;
 350         h->out.bs.p += delta;
 351         h->out.bs.p_end = h->out.p_bitstream + h->out.i_bitstream;
 352
 353         h->cabac.p_start += delta;
 354         h->cabac.p += delta;
 355         h->cabac.p_end = h->out.p_bitstream + h->out.i_bitstream;
 356
 357         for( i = 0; i <= h->out.i_nal; i++ )
 358             h->out.nal[i].p_payload += delta;
 359         x264_free( bs_bak );
 360     }
 361     return 0;
 362 fail:
 363     x264_free( bs_bak );
 364     return -1;
 365 }
 366
 367 /****************************************************************************
 368  *
 369  ****************************************************************************
 370  ****************************** External API*********************************
 371  ****************************************************************************
 372  *
 373  ****************************************************************************/
 374
 375 static int x264_validate_parameters( x264_t *h )
 376 {
 377 #ifdef HAVE_MMX
 378     if( !(x264_cpu_detect() & X264_CPU_SSE) )
 379     {
 380         x264_log( h, X264_LOG_ERROR, "your cpu does not support SSE1, but x264 was compiled with asm support\n");
 381         x264_log( h, X264_LOG_ERROR, "to run x264, recompile without asm support (configure --disable-asm)\n");
 382         return -1;
 383     }
 384 #endif
 385     if( h->param.i_width <= 0 || h->param.i_height <= 0 )
 386     {
 387         x264_log( h, X264_LOG_ERROR, "invalid width x height (%dx%d)\n",
 388                   h->param.i_width, h->param.i_height );
 389         return -1;
 390     }
 391
 392     if( h->param.i_width % 2 || h->param.i_height % 2 )
 393     {
 394         x264_log( h, X264_LOG_ERROR, "width or height not divisible by 2 (%dx%d)\n",
 395                   h->param.i_width, h->param.i_height );
 396         return -1;
 397     }
 398     int i_csp = h->param.i_csp & X264_CSP_MASK;
 399     if( i_csp != X264_CSP_I420 && i_csp != X264_CSP_YV12 )
 400     {
 401         x264_log( h, X264_LOG_ERROR, "invalid CSP (only I420/YV12 supported)\n" );
 402         return -1;
 403     }
 404
 405     if( h->param.i_threads == X264_THREADS_AUTO )
 406         h->param.i_threads = x264_cpu_num_processors() * (h->param.b_sliced_threads?2:3)/2;
 407     h->param.i_threads = x264_clip3( h->param.i_threads, 1, X264_THREAD_MAX );
 408     if( h->param.i_threads > 1 )
 409     {
 410 #ifndef HAVE_PTHREAD
 411         x264_log( h, X264_LOG_WARNING, "not compiled with pthread support!\n");
 412         h->param.i_threads = 1;
 413 #endif
 414         /* Avoid absurdly small thread slices as they can reduce performance
 415          * and VBV compliance.  Capped at an arbitrary 4 rows per thread. */
 416         if( h->param.b_sliced_threads )
 417         {
 418             int max_threads = (h->param.i_height+15)/16 / 4;
 419             h->param.i_threads = X264_MIN( h->param.i_threads, max_threads );
 420         }
 421     }
 422     else
 423         h->param.b_sliced_threads = 0;
 424
 425     if( h->param.b_interlaced )
 426     {
 427         if( h->param.analyse.i_me_method >= X264_ME_ESA )
 428         {
 429             x264_log( h, X264_LOG_WARNING, "interlace + me=esa is not implemented\n" );
 430             h->param.analyse.i_me_method = X264_ME_UMH;
 431         }
 432         if( h->param.analyse.i_direct_mv_pred > X264_DIRECT_PRED_SPATIAL )
 433         {
 434             x264_log( h, X264_LOG_WARNING, "interlace + direct=temporal is not implemented\n" );
 435             h->param.analyse.i_direct_mv_pred = X264_DIRECT_PRED_SPATIAL;
 436         }
 437         if( h->param.analyse.i_weighted_pred > 0 )
 438         {
 439             x264_log( h, X264_LOG_WARNING, "interlace + weightp is not implemented\n" );
 440             h->param.analyse.i_weighted_pred = X264_WEIGHTP_NONE;
 441         }
 442     }
 443
 444     /* Detect default ffmpeg settings and terminate with an error. */
 445     {
 446         int score = 0;
 447         score += h->param.analyse.i_me_range == 0;
 448         score += h->param.rc.i_qp_step == 3;
 449         score += h->param.i_keyint_max == 12;
 450         score += h->param.rc.i_qp_min == 2;
 451         score += h->param.rc.i_qp_max == 31;
 452         score += h->param.rc.f_qcompress == 0.5;
 453         score += fabs(h->param.rc.f_ip_factor - 1.25) < 0.01;
 454         score += fabs(h->param.rc.f_pb_factor - 1.25) < 0.01;
 455         score += h->param.analyse.inter == 0 && h->param.analyse.i_subpel_refine == 8;
 456         if( score >= 5 )
 457         {
 458             x264_log( h, X264_LOG_ERROR, "broken ffmpeg default settings detected\n" );
 459             x264_log( h, X264_LOG_ERROR, "use an encoding preset (vpre)\n" );
 460             return -1;
 461         }
 462     }
 463
 464     if( h->param.rc.i_rc_method < 0 || h->param.rc.i_rc_method > 2 )
 465     {
 466         x264_log( h, X264_LOG_ERROR, "no ratecontrol method specified\n" );
 467         return -1;
 468     }
 469     h->param.rc.f_rf_constant = x264_clip3f( h->param.rc.f_rf_constant, 0, 51 );
 470     h->param.rc.i_qp_constant = x264_clip3( h->param.rc.i_qp_constant, 0, 51 );
 471     if( h->param.rc.i_rc_method == X264_RC_CRF )
 472     {
 473         h->param.rc.i_qp_constant = h->param.rc.f_rf_constant;
 474         h->param.rc.i_bitrate = 0;
 475     }
 476     if( (h->param.rc.i_rc_method == X264_RC_CQP || h->param.rc.i_rc_method == X264_RC_CRF)
 477         && h->param.rc.i_qp_constant == 0 )
 478     {
 479         h->mb.b_lossless = 1;
 480         h->param.i_cqm_preset = X264_CQM_FLAT;
 481         h->param.psz_cqm_file = NULL;
 482         h->param.rc.i_rc_method = X264_RC_CQP;
 483         h->param.rc.f_ip_factor = 1;
 484         h->param.rc.f_pb_factor = 1;
 485         h->param.analyse.b_psnr = 0;
 486         h->param.analyse.b_ssim = 0;
 487         h->param.analyse.i_chroma_qp_offset = 0;
 488         h->param.analyse.i_trellis = 0;
 489         h->param.analyse.b_fast_pskip = 0;
 490         h->param.analyse.i_noise_reduction = 0;
 491         h->param.analyse.f_psy_rd = 0;
 492         h->param.i_bframe = 0;
 493         /* 8x8dct is not useful at all in CAVLC lossless */
 494         if( !h->param.b_cabac )
 495             h->param.analyse.b_transform_8x8 = 0;
 496     }
 497     if( h->param.rc.i_rc_method == X264_RC_CQP )
 498     {
 499         float qp_p = h->param.rc.i_qp_constant;
 500         float qp_i = qp_p - 6*log(h->param.rc.f_ip_factor)/log(2);
 501         float qp_b = qp_p + 6*log(h->param.rc.f_pb_factor)/log(2);
 502         h->param.rc.i_qp_min = x264_clip3( (int)(X264_MIN3( qp_p, qp_i, qp_b )), 0, 51 );
 503         h->param.rc.i_qp_max = x264_clip3( (int)(X264_MAX3( qp_p, qp_i, qp_b ) + .999), 0, 51 );
 504         h->param.rc.i_aq_mode = 0;
 505         h->param.rc.b_mb_tree = 0;
 506     }
 507     h->param.rc.i_qp_max = x264_clip3( h->param.rc.i_qp_max, 0, 51 );
 508     h->param.rc.i_qp_min = x264_clip3( h->param.rc.i_qp_min, 0, h->param.rc.i_qp_max );
 509
 510     int max_slices = (h->param.i_height+((16<<h->param.b_interlaced)-1))/(16<<h->param.b_interlaced);
 511     if( h->param.b_sliced_threads )
 512         h->param.i_slice_count = x264_clip3( h->param.i_threads, 0, max_slices );
 513     else
 514     {
 515         h->param.i_slice_count = x264_clip3( h->param.i_slice_count, 0, max_slices );
 516         h->param.i_slice_max_size = X264_MAX( h->param.i_slice_max_size, 0 );
 517         h->param.i_slice_max_mbs = X264_MAX( h->param.i_slice_max_mbs, 0 );
 518         if( h->param.b_interlaced && h->param.i_slice_max_size )
 519         {
 520             x264_log( h, X264_LOG_WARNING, "interlaced + slice-max-size is not implemented\n" );
 521             h->param.i_slice_max_size = 0;
 522         }
 523         if( h->param.b_interlaced && h->param.i_slice_max_mbs )
 524         {
 525             x264_log( h, X264_LOG_WARNING, "interlaced + slice-max-mbs is not implemented\n" );
 526             h->param.i_slice_max_mbs = 0;
 527         }
 528         if( h->param.i_slice_max_mbs || h->param.i_slice_max_size )
 529             h->param.i_slice_count = 0;
 530     }
 531
 532     h->param.i_frame_reference = x264_clip3( h->param.i_frame_reference, 1, 16 );
 533     if( h->param.i_keyint_max <= 0 )
 534         h->param.i_keyint_max = 1;
 535     if( h->param.i_scenecut_threshold < 0 )
 536         h->param.i_scenecut_threshold = 0;
 537     if( !h->param.analyse.i_subpel_refine && h->param.analyse.i_direct_mv_pred > X264_DIRECT_PRED_SPATIAL )
 538     {
 539         x264_log( h, X264_LOG_WARNING, "subme=0 + direct=temporal is not supported\n" );
 540         h->param.analyse.i_direct_mv_pred = X264_DIRECT_PRED_SPATIAL;
 541     }
 542     h->param.i_bframe = x264_clip3( h->param.i_bframe, 0, X264_BFRAME_MAX );
 543     if( h->param.i_keyint_max == 1 )
 544     {
 545         h->param.i_bframe = 0;
 546         h->param.b_intra_refresh = 0;
 547     }
 548     h->param.i_bframe_bias = x264_clip3( h->param.i_bframe_bias, -90, 100 );
 549     if( h->param.i_bframe <= 1 )
 550         h->param.i_bframe_pyramid = X264_B_PYRAMID_NONE;
 551     h->param.i_bframe_pyramid = x264_clip3( h->param.i_bframe_pyramid, X264_B_PYRAMID_NONE, X264_B_PYRAMID_NORMAL );
 552     if( !h->param.i_bframe )
 553     {
 554         h->param.i_bframe_adaptive = X264_B_ADAPT_NONE;
 555         h->param.analyse.i_direct_mv_pred = 0;
 556         h->param.analyse.b_weighted_bipred = 0;
 557     }
 558     if( h->param.b_intra_refresh && h->param.i_bframe_pyramid == X264_B_PYRAMID_NORMAL )
 559     {
 560         x264_log( h, X264_LOG_WARNING, "b-pyramid normal + intra-refresh is not supported\n" );
 561         h->param.i_bframe_pyramid = X264_B_PYRAMID_STRICT;
 562     }
 563     if( h->param.b_intra_refresh && h->param.i_frame_reference > 1 )
 564     {
 565         x264_log( h, X264_LOG_WARNING, "ref > 1 + intra-refresh is not supported\n" );
 566         h->param.i_frame_reference = 1;
 567     }
 568     if( h->param.b_intra_refresh )
 569         h->param.i_keyint_max = X264_MIN( h->param.i_keyint_max, (h->param.i_width+15)/16 - 1 );
 570     h->param.i_keyint_min = x264_clip3( h->param.i_keyint_min, 1, h->param.i_keyint_max/2+1 );
 571     h->param.rc.i_lookahead = x264_clip3( h->param.rc.i_lookahead, 0, X264_LOOKAHEAD_MAX );
 572     {
 573         int maxrate = X264_MAX( h->param.rc.i_vbv_max_bitrate, h->param.rc.i_bitrate );
 574         float bufsize = maxrate ? (float)h->param.rc.i_vbv_buffer_size / maxrate : 0;
 575         float fps = h->param.i_fps_num > 0 && h->param.i_fps_den > 0 ? (float) h->param.i_fps_num / h->param.i_fps_den : 25.0;
 576         h->param.rc.i_lookahead = X264_MIN( h->param.rc.i_lookahead, X264_MAX( h->param.i_keyint_max, bufsize*fps ) );
 577     }
 578
 579     if( !h->param.i_timebase_num || !h->param.i_timebase_den )
 580     {
 581         h->param.i_timebase_num = h->param.i_fps_den;
 582         h->param.i_timebase_den = h->param.i_fps_num;
 583     }
 584
 585     h->param.rc.f_qcompress = x264_clip3f( h->param.rc.f_qcompress, 0.0, 1.0 );
 586     if( !h->param.rc.i_lookahead || h->param.i_keyint_max == 1 || h->param.rc.f_qcompress == 1 )
 587         h->param.rc.b_mb_tree = 0;
 588     if( h->param.rc.b_stat_read )
 589         h->param.rc.i_lookahead = 0;
 590 #ifdef HAVE_PTHREAD
 591     if( h->param.i_sync_lookahead )
 592         h->param.i_sync_lookahead = x264_clip3( h->param.i_sync_lookahead, h->param.i_threads + h->param.i_bframe, X264_LOOKAHEAD_MAX );
 593     if( h->param.rc.b_stat_read || h->param.i_threads == 1 || h->param.b_sliced_threads )
 594         h->param.i_sync_lookahead = 0;
 595 #else
 596     h->param.i_sync_lookahead = 0;
 597 #endif
 598
 599     h->mb.b_direct_auto_write = h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO
 600                                 && h->param.i_bframe
 601                                 && ( h->param.rc.b_stat_write || !h->param.rc.b_stat_read );
 602
 603     h->param.i_deblocking_filter_alphac0 = x264_clip3( h->param.i_deblocking_filter_alphac0, -6, 6 );
 604     h->param.i_deblocking_filter_beta    = x264_clip3( h->param.i_deblocking_filter_beta, -6, 6 );
 605     h->param.analyse.i_luma_deadzone[0] = x264_clip3( h->param.analyse.i_luma_deadzone[0], 0, 32 );
 606     h->param.analyse.i_luma_deadzone[1] = x264_clip3( h->param.analyse.i_luma_deadzone[1], 0, 32 );
 607
 608     h->param.i_cabac_init_idc = x264_clip3( h->param.i_cabac_init_idc, 0, 2 );
 609
 610     if( h->param.i_cqm_preset < X264_CQM_FLAT || h->param.i_cqm_preset > X264_CQM_CUSTOM )
 611         h->param.i_cqm_preset = X264_CQM_FLAT;
 612
 613     if( h->param.analyse.i_me_method < X264_ME_DIA ||
 614         h->param.analyse.i_me_method > X264_ME_TESA )
 615         h->param.analyse.i_me_method = X264_ME_HEX;
 616     if( h->param.analyse.i_me_range < 4 )
 617         h->param.analyse.i_me_range = 4;
 618     if( h->param.analyse.i_me_range > 16 && h->param.analyse.i_me_method <= X264_ME_HEX )
 619         h->param.analyse.i_me_range = 16;
 620     if( h->param.analyse.i_me_method == X264_ME_TESA &&
 621         (h->mb.b_lossless || h->param.analyse.i_subpel_refine <= 1) )
 622         h->param.analyse.i_me_method = X264_ME_ESA;
 623     h->param.analyse.i_subpel_refine = x264_clip3( h->param.analyse.i_subpel_refine, 0, 10 );
 624     h->param.analyse.b_mixed_references = h->param.analyse.b_mixed_references && h->param.i_frame_reference > 1;
 625     h->param.analyse.inter &= X264_ANALYSE_PSUB16x16|X264_ANALYSE_PSUB8x8|X264_ANALYSE_BSUB16x16|
 626                               X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
 627     h->param.analyse.intra &= X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
 628     if( !(h->param.analyse.inter & X264_ANALYSE_PSUB16x16) )
 629         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
 630     if( !h->param.analyse.b_transform_8x8 )
 631     {
 632         h->param.analyse.inter &= ~X264_ANALYSE_I8x8;
 633         h->param.analyse.intra &= ~X264_ANALYSE_I8x8;
 634     }
 635     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -12, 12);
 636     if( !h->param.b_cabac )
 637         h->param.analyse.i_trellis = 0;
 638     h->param.analyse.i_trellis = x264_clip3( h->param.analyse.i_trellis, 0, 2 );
 639     if( !h->param.analyse.b_psy )
 640     {
 641         h->param.analyse.f_psy_rd = 0;
 642         h->param.analyse.f_psy_trellis = 0;
 643     }
 644     if( !h->param.analyse.i_trellis )
 645         h->param.analyse.f_psy_trellis = 0;
 646     h->param.analyse.f_psy_rd = x264_clip3f( h->param.analyse.f_psy_rd, 0, 10 );
 647     h->param.analyse.f_psy_trellis = x264_clip3f( h->param.analyse.f_psy_trellis, 0, 10 );
 648     if( h->param.analyse.i_subpel_refine < 6 )
 649         h->param.analyse.f_psy_rd = 0;
 650     h->mb.i_psy_rd = FIX8( h->param.analyse.f_psy_rd );
 651     /* Psy RDO increases overall quantizers to improve the quality of luma--this indirectly hurts chroma quality */
 652     /* so we lower the chroma QP offset to compensate */
 653     /* This can be triggered repeatedly on multiple calls to parameter_validate, but since encoding
 654      * uses the pps chroma qp offset not the param chroma qp offset, this is not a problem. */
 655     if( h->mb.i_psy_rd )
 656         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_rd < 0.25 ? 1 : 2;
 657     h->mb.i_psy_trellis = FIX8( h->param.analyse.f_psy_trellis / 4 );
 658     /* Psy trellis has a similar effect. */
 659     if( h->mb.i_psy_trellis )
 660         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_trellis < 0.25 ? 1 : 2;
 661     else
 662         h->mb.i_psy_trellis = 0;
 663     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -12, 12);
 664     h->param.rc.i_aq_mode = x264_clip3( h->param.rc.i_aq_mode, 0, 2 );
 665     h->param.rc.f_aq_strength = x264_clip3f( h->param.rc.f_aq_strength, 0, 3 );
 666     if( h->param.rc.f_aq_strength == 0 )
 667         h->param.rc.i_aq_mode = 0;
 668     /* MB-tree requires AQ to be on, even if the strength is zero. */
 669     if( !h->param.rc.i_aq_mode && h->param.rc.b_mb_tree )
 670     {
 671         h->param.rc.i_aq_mode = 1;
 672         h->param.rc.f_aq_strength = 0;
 673     }
 674     h->param.analyse.i_noise_reduction = x264_clip3( h->param.analyse.i_noise_reduction, 0, 1<<16 );
 675     if( h->param.analyse.i_subpel_refine == 10 && (h->param.analyse.i_trellis != 2 || !h->param.rc.i_aq_mode) )
 676         h->param.analyse.i_subpel_refine = 9;
 677
 678     {
 679         const x264_level_t *l = x264_levels;
 680         if( h->param.i_level_idc < 0 )
 681         {
 682             int maxrate_bak = h->param.rc.i_vbv_max_bitrate;
 683             if( h->param.rc.i_rc_method == X264_RC_ABR && h->param.rc.i_vbv_buffer_size <= 0 )
 684                 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate * 2;
 685             h->sps = h->sps_array;
 686             x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
 687             do h->param.i_level_idc = l->level_idc;
 688                 while( l[1].level_idc && x264_validate_levels( h, 0 ) && l++ );
 689             h->param.rc.i_vbv_max_bitrate = maxrate_bak;
 690         }
 691         else
 692         {
 693             while( l->level_idc && l->level_idc != h->param.i_level_idc )
 694                 l++;
 695             if( l->level_idc == 0 )
 696             {
 697                 x264_log( h, X264_LOG_ERROR, "invalid level_idc: %d\n", h->param.i_level_idc );
 698                 return -1;
 699             }
 700         }
 701         if( h->param.analyse.i_mv_range <= 0 )
 702             h->param.analyse.i_mv_range = l->mv_range >> h->param.b_interlaced;
 703         else
 704             h->param.analyse.i_mv_range = x264_clip3(h->param.analyse.i_mv_range, 32, 512 >> h->param.b_interlaced);
 705     }
 706
 707     h->param.analyse.i_weighted_pred = x264_clip3( h->param.analyse.i_weighted_pred, 0, X264_WEIGHTP_SMART );
 708     if( !h->param.analyse.i_weighted_pred && h->param.rc.b_mb_tree && h->param.analyse.b_psy && !h->param.b_interlaced )
 709         h->param.analyse.i_weighted_pred = X264_WEIGHTP_FAKE;
 710
 711     if( h->param.i_threads > 1 && !h->param.b_sliced_threads )
 712     {
 713         int r = h->param.analyse.i_mv_range_thread;
 714         int r2;
 715         if( r <= 0 )
 716         {
 717             // half of the available space is reserved and divided evenly among the threads,
 718             // the rest is allocated to whichever thread is far enough ahead to use it.
 719             // reserving more space increases quality for some videos, but costs more time
 720             // in thread synchronization.
 721             int max_range = (h->param.i_height + X264_THREAD_HEIGHT) / h->param.i_threads - X264_THREAD_HEIGHT;
 722             r = max_range / 2;
 723         }
 724         r = X264_MAX( r, h->param.analyse.i_me_range );
 725         r = X264_MIN( r, h->param.analyse.i_mv_range );
 726         // round up to use the whole mb row
 727         r2 = (r & ~15) + ((-X264_THREAD_HEIGHT) & 15);
 728         if( r2 < r )
 729             r2 += 16;
 730         x264_log( h, X264_LOG_DEBUG, "using mv_range_thread = %d\n", r2 );
 731         h->param.analyse.i_mv_range_thread = r2;
 732     }
 733
 734     if( h->param.rc.f_qblur < 0 )
 735         h->param.rc.f_qblur = 0;
 736     if( h->param.rc.f_complexity_blur < 0 )
 737         h->param.rc.f_complexity_blur = 0;
 738
 739     h->param.i_sps_id &= 31;
 740
 741     if( h->param.i_log_level < X264_LOG_INFO )
 742     {
 743         h->param.analyse.b_psnr = 0;
 744         h->param.analyse.b_ssim = 0;
 745     }
 746
 747     /* ensure the booleans are 0 or 1 so they can be used in math */
 748 #define BOOLIFY(x) h->param.x = !!h->param.x
 749     BOOLIFY( b_cabac );
 750     BOOLIFY( b_constrained_intra );
 751     BOOLIFY( b_deblocking_filter );
 752     BOOLIFY( b_deterministic );
 753     BOOLIFY( b_sliced_threads );
 754     BOOLIFY( b_interlaced );
 755     BOOLIFY( b_intra_refresh );
 756     BOOLIFY( b_visualize );
 757     BOOLIFY( b_aud );
 758     BOOLIFY( b_repeat_headers );
 759     BOOLIFY( b_annexb );
 760     BOOLIFY( analyse.b_transform_8x8 );
 761     BOOLIFY( analyse.b_weighted_bipred );
 762     BOOLIFY( analyse.b_chroma_me );
 763     BOOLIFY( analyse.b_mixed_references );
 764     BOOLIFY( analyse.b_fast_pskip );
 765     BOOLIFY( analyse.b_dct_decimate );
 766     BOOLIFY( analyse.b_psy );
 767     BOOLIFY( analyse.b_psnr );
 768     BOOLIFY( analyse.b_ssim );
 769     BOOLIFY( rc.b_stat_write );
 770     BOOLIFY( rc.b_stat_read );
 771     BOOLIFY( rc.b_mb_tree );
 772 #undef BOOLIFY
 773
 774     return 0;
 775 }
 776
 777 static void mbcmp_init( x264_t *h )
 778 {
 779     int satd = !h->mb.b_lossless && h->param.analyse.i_subpel_refine > 1;
 780     memcpy( h->pixf.mbcmp, satd ? h->pixf.satd : h->pixf.sad_aligned, sizeof(h->pixf.mbcmp) );
 781     memcpy( h->pixf.mbcmp_unaligned, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.mbcmp_unaligned) );
 782     h->pixf.intra_mbcmp_x3_16x16 = satd ? h->pixf.intra_satd_x3_16x16 : h->pixf.intra_sad_x3_16x16;
 783     h->pixf.intra_mbcmp_x3_8x8c = satd ? h->pixf.intra_satd_x3_8x8c : h->pixf.intra_sad_x3_8x8c;
 784     h->pixf.intra_mbcmp_x3_4x4 = satd ? h->pixf.intra_satd_x3_4x4 : h->pixf.intra_sad_x3_4x4;
 785     satd &= h->param.analyse.i_me_method == X264_ME_TESA;
 786     memcpy( h->pixf.fpelcmp, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.fpelcmp) );
 787     memcpy( h->pixf.fpelcmp_x3, satd ? h->pixf.satd_x3 : h->pixf.sad_x3, sizeof(h->pixf.fpelcmp_x3) );
 788     memcpy( h->pixf.fpelcmp_x4, satd ? h->pixf.satd_x4 : h->pixf.sad_x4, sizeof(h->pixf.fpelcmp_x4) );
 789 }
 790
 791 static void x264_set_aspect_ratio( x264_t *h, x264_param_t *param, int initial )
 792 {
 793     /* VUI */
 794     if( param->vui.i_sar_width > 0 && param->vui.i_sar_height > 0 )
 795     {
 796         int i_w = param->vui.i_sar_width;
 797         int i_h = param->vui.i_sar_height;
 798         int old_w = h->param.vui.i_sar_width;
 799         int old_h = h->param.vui.i_sar_height;
 800
 801         x264_reduce_fraction( &i_w, &i_h );
 802
 803         while( i_w > 65535 || i_h > 65535 )
 804         {
 805             i_w /= 2;
 806             i_h /= 2;
 807         }
 808
 809         x264_reduce_fraction( &i_w, &i_h );
 810
 811         if( i_w != old_w || i_h != old_h || initial )
 812         {
 813             h->param.vui.i_sar_width = 0;
 814             h->param.vui.i_sar_height = 0;
 815             if( i_w == 0 || i_h == 0 )
 816                 x264_log( h, X264_LOG_WARNING, "cannot create valid sample aspect ratio\n" );
 817             else
 818             {
 819                 x264_log( h, initial?X264_LOG_INFO:X264_LOG_DEBUG, "using SAR=%d/%d\n", i_w, i_h );
 820                 h->param.vui.i_sar_width = i_w;
 821                 h->param.vui.i_sar_height = i_h;
 822             }
 823         }
 824     }
 825 }
 826
 827 /****************************************************************************
 828  * x264_encoder_open:
 829  ****************************************************************************/
 830 x264_t *x264_encoder_open( x264_param_t *param )
 831 {
 832     x264_t *h;
 833     char buf[1000], *p;
 834     int i, qp, i_slicetype_length;
 835
 836     CHECKED_MALLOCZERO( h, sizeof(x264_t) );
 837
 838     /* Create a copy of param */
 839     memcpy( &h->param, param, sizeof(x264_param_t) );
 840
 841     if( param->param_free )
 842         param->param_free( param );
 843
 844     if( x264_validate_parameters( h ) < 0 )
 845         goto fail;
 846
 847     if( h->param.psz_cqm_file )
 848         if( x264_cqm_parse_file( h, h->param.psz_cqm_file ) < 0 )
 849             goto fail;
 850
 851     if( h->param.rc.psz_stat_out )
 852         h->param.rc.psz_stat_out = strdup( h->param.rc.psz_stat_out );
 853     if( h->param.rc.psz_stat_in )
 854         h->param.rc.psz_stat_in = strdup( h->param.rc.psz_stat_in );
 855
 856     x264_set_aspect_ratio( h, &h->param, 1 );
 857
 858     x264_reduce_fraction( &h->param.i_fps_num, &h->param.i_fps_den );
 859     x264_reduce_fraction( &h->param.i_timebase_num, &h->param.i_timebase_den );
 860
 861     /* Init x264_t */
 862     h->i_frame = -1;
 863     h->i_frame_num = 0;
 864     h->i_idr_pic_id = 0;
 865
 866     h->sps = &h->sps_array[0];
 867     x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
 868
 869     h->pps = &h->pps_array[0];
 870     x264_pps_init( h->pps, h->param.i_sps_id, &h->param, h->sps );
 871
 872     x264_validate_levels( h, 1 );
 873
 874     h->chroma_qp_table = i_chroma_qp_table + 12 + h->pps->i_chroma_qp_index_offset;
 875
 876     if( x264_cqm_init( h ) < 0 )
 877         goto fail;
 878
 879     h->mb.i_mb_count = h->sps->i_mb_width * h->sps->i_mb_height;
 880
 881     /* Init frames. */
 882     if( h->param.i_bframe_adaptive == X264_B_ADAPT_TRELLIS )
 883         h->frames.i_delay = X264_MAX(h->param.i_bframe,3)*4;
 884     else
 885         h->frames.i_delay = h->param.i_bframe;
 886     if( h->param.rc.b_mb_tree || h->param.rc.i_vbv_buffer_size )
 887         h->frames.i_delay = X264_MAX( h->frames.i_delay, h->param.rc.i_lookahead );
 888     i_slicetype_length = h->frames.i_delay;
 889     if( !h->param.b_sliced_threads )
 890         h->frames.i_delay += h->param.i_threads - 1;
 891     h->frames.i_delay = X264_MIN( h->frames.i_delay, X264_LOOKAHEAD_MAX );
 892     h->frames.i_delay += h->param.i_sync_lookahead;
 893     h->frames.i_bframe_delay = h->param.i_bframe ? (h->param.i_bframe_pyramid ? 2 : 1) : 0;
 894
 895     h->frames.i_max_ref0 = h->param.i_frame_reference;
 896     h->frames.i_max_ref1 = h->sps->vui.i_num_reorder_frames;
 897     h->frames.i_max_dpb  = h->sps->vui.i_max_dec_frame_buffering;
 898     h->frames.b_have_lowres = !h->param.rc.b_stat_read
 899         && ( h->param.rc.i_rc_method == X264_RC_ABR
 900           || h->param.rc.i_rc_method == X264_RC_CRF
 901           || h->param.i_bframe_adaptive
 902           || h->param.i_scenecut_threshold
 903           || h->param.rc.b_mb_tree
 904           || h->param.analyse.i_weighted_pred == X264_WEIGHTP_SMART );
 905     h->frames.b_have_lowres |= h->param.rc.b_stat_read && h->param.rc.i_vbv_buffer_size > 0;
 906     h->frames.b_have_sub8x8_esa = !!(h->param.analyse.inter & X264_ANALYSE_PSUB8x8);
 907
 908     h->frames.i_last_keyframe = - h->param.i_keyint_max;
 909     h->frames.i_input    = 0;
 910
 911     CHECKED_MALLOCZERO( h->frames.unused[0], (h->frames.i_delay + 3) * sizeof(x264_frame_t *) );
 912     /* Allocate room for max refs plus a few extra just in case. */
 913     CHECKED_MALLOCZERO( h->frames.unused[1], (h->param.i_threads + 20) * sizeof(x264_frame_t *) );
 914     CHECKED_MALLOCZERO( h->frames.current, (h->param.i_sync_lookahead + h->param.i_bframe
 915                         + h->param.i_threads + 3) * sizeof(x264_frame_t *) );
 916     if( h->param.analyse.i_weighted_pred > 0 )
 917         CHECKED_MALLOCZERO( h->frames.blank_unused, h->param.i_threads * 4 * sizeof(x264_frame_t *) );
 918     h->i_ref0 = 0;
 919     h->i_ref1 = 0;
 920
 921     x264_rdo_init();
 922
 923     /* init CPU functions */
 924     x264_predict_16x16_init( h->param.cpu, h->predict_16x16 );
 925     x264_predict_8x8c_init( h->param.cpu, h->predict_8x8c );
 926     x264_predict_8x8_init( h->param.cpu, h->predict_8x8, &h->predict_8x8_filter );
 927     x264_predict_4x4_init( h->param.cpu, h->predict_4x4 );
 928     if( !h->param.b_cabac )
 929         x264_init_vlc_tables();
 930     x264_pixel_init( h->param.cpu, &h->pixf );
 931     x264_dct_init( h->param.cpu, &h->dctf );
 932     x264_zigzag_init( h->param.cpu, &h->zigzagf, h->param.b_interlaced );
 933     x264_mc_init( h->param.cpu, &h->mc );
 934     x264_quant_init( h, h->param.cpu, &h->quantf );
 935     x264_deblock_init( h->param.cpu, &h->loopf );
 936     x264_dct_init_weights();
 937
 938     mbcmp_init( h );
 939
 940     p = buf + sprintf( buf, "using cpu capabilities:" );
 941     for( i=0; x264_cpu_names[i].flags; i++ )
 942     {
 943         if( !strcmp(x264_cpu_names[i].name, "SSE2")
 944             && h->param.cpu & (X264_CPU_SSE2_IS_FAST|X264_CPU_SSE2_IS_SLOW) )
 945             continue;
 946         if( !strcmp(x264_cpu_names[i].name, "SSE3")
 947             && (h->param.cpu & X264_CPU_SSSE3 || !(h->param.cpu & X264_CPU_CACHELINE_64)) )
 948             continue;
 949         if( !strcmp(x264_cpu_names[i].name, "SSE4.1")
 950             && (h->param.cpu & X264_CPU_SSE42) )
 951             continue;
 952         if( (h->param.cpu & x264_cpu_names[i].flags) == x264_cpu_names[i].flags
 953             && (!i || x264_cpu_names[i].flags != x264_cpu_names[i-1].flags) )
 954             p += sprintf( p, " %s", x264_cpu_names[i].name );
 955     }
 956     if( !h->param.cpu )
 957         p += sprintf( p, " none!" );
 958     x264_log( h, X264_LOG_INFO, "%s\n", buf );
 959
 960     for( qp = h->param.rc.i_qp_min; qp <= h->param.rc.i_qp_max; qp++ )
 961         if( x264_analyse_init_costs( h, qp ) )
 962             goto fail;
 963     if( x264_analyse_init_costs( h, X264_LOOKAHEAD_QP ) )
 964         goto fail;
 965     if( h->cost_mv[1][2013] != 24 )
 966     {
 967         x264_log( h, X264_LOG_ERROR, "MV cost test failed: x264 has been miscompiled!\n" );
 968         goto fail;
 969     }
 970
 971     h->out.i_nal = 0;
 972     h->out.i_bitstream = X264_MAX( 1000000, h->param.i_width * h->param.i_height * 4
 973         * ( h->param.rc.i_rc_method == X264_RC_ABR ? pow( 0.95, h->param.rc.i_qp_min )
 974           : pow( 0.95, h->param.rc.i_qp_constant ) * X264_MAX( 1, h->param.rc.f_ip_factor )));
 975
 976     CHECKED_MALLOC( h->nal_buffer, h->out.i_bitstream * 3/2 + 4 );
 977     h->nal_buffer_size = h->out.i_bitstream * 3/2 + 4;
 978
 979     h->thread[0] = h;
 980     h->i_thread_num = 0;
 981     for( i = 1; i < h->param.i_threads + !!h->param.i_sync_lookahead; i++ )
 982         CHECKED_MALLOC( h->thread[i], sizeof(x264_t) );
 983
 984     if( x264_lookahead_init( h, i_slicetype_length ) )
 985         goto fail;
 986
 987     for( i = 0; i < h->param.i_threads; i++ )
 988     {
 989         int init_nal_count = h->param.i_slice_count + 3;
 990         int allocate_threadlocal_data = !h->param.b_sliced_threads || !i;
 991         if( i > 0 )
 992             *h->thread[i] = *h;
 993
 994         if( allocate_threadlocal_data )
 995         {
 996             h->thread[i]->fdec = x264_frame_pop_unused( h, 1 );
 997             if( !h->thread[i]->fdec )
 998                 goto fail;
 999         }
1000         else
1001             h->thread[i]->fdec = h->thread[0]->fdec;
1002
1003         CHECKED_MALLOC( h->thread[i]->out.p_bitstream, h->out.i_bitstream );
1004         /* Start each thread with room for init_nal_count NAL units; it'll realloc later if needed. */
1005         CHECKED_MALLOC( h->thread[i]->out.nal, init_nal_count*sizeof(x264_nal_t) );
1006         h->thread[i]->out.i_nals_allocated = init_nal_count;
1007
1008         if( allocate_threadlocal_data && x264_macroblock_cache_init( h->thread[i] ) < 0 )
1009             goto fail;
1010     }
1011
1012     /* Allocate scratch buffer */
1013     for( i = 0; i < h->param.i_threads + !!h->param.i_sync_lookahead; i++ )
1014     {
1015         int buf_hpel = (h->param.i_width+48) * sizeof(int16_t);
1016         int buf_ssim = h->param.analyse.b_ssim * 8 * (h->param.i_width/4+3) * sizeof(int);
1017         int me_range = X264_MIN(h->param.analyse.i_me_range, h->param.analyse.i_mv_range);
1018         int buf_tesa = (h->param.analyse.i_me_method >= X264_ME_ESA) *
1019             ((me_range*2+18) * sizeof(int16_t) + (me_range+4) * (me_range+1) * 4 * sizeof(mvsad_t));
1020         int buf_mbtree = h->param.rc.b_mb_tree * ((h->sps->i_mb_width+3)&~3) * sizeof(int);
1021         int buf_nnz = !h->param.b_cabac * h->pps->b_transform_8x8_mode * (h->sps->i_mb_width * 4 * 16 * sizeof(uint8_t));
1022         int scratch_size = X264_MAX4( buf_hpel, buf_ssim, buf_tesa, X264_MAX( buf_mbtree, buf_nnz ) );
1023         CHECKED_MALLOC( h->thread[i]->scratch_buffer, scratch_size );
1024     }
1025
1026     if( x264_ratecontrol_new( h ) < 0 )
1027         goto fail;
1028
1029     if( h->param.psz_dump_yuv )
1030     {
1031         /* create or truncate the reconstructed video file */
1032         FILE *f = fopen( h->param.psz_dump_yuv, "w" );
1033         if( !f )
1034         {
1035             x264_log( h, X264_LOG_ERROR, "dump_yuv: can't write to %s\n", h->param.psz_dump_yuv );
1036             goto fail;
1037         }
1038         else if( !x264_is_regular_file( f ) )
1039         {
1040             x264_log( h, X264_LOG_ERROR, "dump_yuv: incompatible with non-regular file %s\n", h->param.psz_dump_yuv );
1041             goto fail;
1042         }
1043         fclose( f );
1044     }
1045
1046     x264_log( h, X264_LOG_INFO, "profile %s, level %d.%d\n",
1047         h->sps->i_profile_idc == PROFILE_BASELINE ? "Baseline" :
1048         h->sps->i_profile_idc == PROFILE_MAIN ? "Main" :
1049         h->sps->i_profile_idc == PROFILE_HIGH ? "High" :
1050         "High 4:4:4 Predictive", h->sps->i_level_idc/10, h->sps->i_level_idc%10 );
1051
1052     return h;
1053 fail:
1054     x264_free( h );
1055     return NULL;
1056 }
1057
1058 /****************************************************************************
1059  * x264_encoder_reconfig:
1060  ****************************************************************************/
1061 int x264_encoder_reconfig( x264_t *h, x264_param_t *param )
1062 {
1063     h = h->thread[h->i_thread_phase];
1064     x264_set_aspect_ratio( h, param, 0 );
1065 #define COPY(var) h->param.var = param->var
1066     COPY( i_frame_reference ); // but never uses more refs than initially specified
1067     COPY( i_bframe_bias );
1068     if( h->param.i_scenecut_threshold )
1069         COPY( i_scenecut_threshold ); // can't turn it on or off, only vary the threshold
1070     COPY( b_deblocking_filter );
1071     COPY( i_deblocking_filter_alphac0 );
1072     COPY( i_deblocking_filter_beta );
1073     COPY( analyse.inter );
1074     COPY( analyse.intra );
1075     COPY( analyse.i_direct_mv_pred );
1076     /* Scratch buffer prevents me_range from being increased for esa/tesa */
1077     if( h->param.analyse.i_me_method < X264_ME_ESA || param->analyse.i_me_range < h->param.analyse.i_me_range )
1078         COPY( analyse.i_me_range );
1079     COPY( analyse.i_noise_reduction );
1080     /* We can't switch out of subme=0 during encoding. */
1081     if( h->param.analyse.i_subpel_refine )
1082         COPY( analyse.i_subpel_refine );
1083     COPY( analyse.i_trellis );
1084     COPY( analyse.b_chroma_me );
1085     COPY( analyse.b_dct_decimate );
1086     COPY( analyse.b_fast_pskip );
1087     COPY( analyse.b_mixed_references );
1088     COPY( analyse.f_psy_rd );
1089     COPY( analyse.f_psy_trellis );
1090     // can only twiddle these if they were enabled to begin with:
1091     if( h->param.analyse.i_me_method >= X264_ME_ESA || param->analyse.i_me_method < X264_ME_ESA )
1092         COPY( analyse.i_me_method );
1093     if( h->param.analyse.i_me_method >= X264_ME_ESA && !h->frames.b_have_sub8x8_esa )
1094         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
1095     if( h->pps->b_transform_8x8_mode )
1096         COPY( analyse.b_transform_8x8 );
1097     if( h->frames.i_max_ref1 > 1 )
1098         COPY( i_bframe_pyramid );
1099     COPY( i_slice_max_size );
1100     COPY( i_slice_max_mbs );
1101     COPY( i_slice_count );
1102 #undef COPY
1103
1104     mbcmp_init( h );
1105
1106     return x264_validate_parameters( h );
1107 }
1108
1109 /****************************************************************************
1110  * x264_encoder_parameters:
1111  ****************************************************************************/
1112 void x264_encoder_parameters( x264_t *h, x264_param_t *param )
1113 {
1114     memcpy( param, &h->thread[h->i_thread_phase]->param, sizeof(x264_param_t) );
1115 }
1116
1117 /* internal usage */
1118 static void x264_nal_start( x264_t *h, int i_type, int i_ref_idc )
1119 {
1120     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1121
1122     nal->i_ref_idc = i_ref_idc;
1123     nal->i_type    = i_type;
1124
1125     nal->i_payload= 0;
1126     nal->p_payload= &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8];
1127 }
1128 /* if number of allocated nals is not enough, re-allocate a larger one. */
1129 static int x264_nal_check_buffer( x264_t *h )
1130 {
1131     if( h->out.i_nal >= h->out.i_nals_allocated )
1132     {
1133         x264_nal_t *new_out = x264_malloc( sizeof(x264_nal_t) * (h->out.i_nals_allocated*2) );
1134         if( !new_out )
1135             return -1;
1136         memcpy( new_out, h->out.nal, sizeof(x264_nal_t) * (h->out.i_nals_allocated) );
1137         x264_free( h->out.nal );
1138         h->out.nal = new_out;
1139         h->out.i_nals_allocated *= 2;
1140     }
1141     return 0;
1142 }
1143 static int x264_nal_end( x264_t *h )
1144 {
1145     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1146     nal->i_payload = &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8] - nal->p_payload;
1147     h->out.i_nal++;
1148
1149     return x264_nal_check_buffer( h );
1150 }
1151
1152 static int x264_encoder_encapsulate_nals( x264_t *h )
1153 {
1154     int nal_size = 0, i;
1155     for( i = 0; i < h->out.i_nal; i++ )
1156         nal_size += h->out.nal[i].i_payload;
1157
1158     /* Worst-case NAL unit escaping: reallocate the buffer if it's too small. */
1159     if( h->nal_buffer_size < nal_size * 3/2 + h->out.i_nal * 4 )
1160     {
1161         uint8_t *buf = x264_malloc( nal_size * 2 + h->out.i_nal * 4 );
1162         if( !buf )
1163             return -1;
1164         x264_free( h->nal_buffer );
1165         h->nal_buffer = buf;
1166     }
1167
1168     uint8_t *nal_buffer = h->nal_buffer;
1169
1170     for( i = 0; i < h->out.i_nal; i++ )
1171     {
1172         int size = x264_nal_encode( nal_buffer, h->param.b_annexb, &h->out.nal[i] );
1173         h->out.nal[i].i_payload = size;
1174         h->out.nal[i].p_payload = nal_buffer;
1175         nal_buffer += size;
1176     }
1177
1178     return nal_buffer - h->nal_buffer;
1179 }
1180
1181 /****************************************************************************
1182  * x264_encoder_headers:
1183  ****************************************************************************/
1184 int x264_encoder_headers( x264_t *h, x264_nal_t **pp_nal, int *pi_nal )
1185 {
1186     int frame_size = 0;
1187     /* init bitstream context */
1188     h->out.i_nal = 0;
1189     bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
1190
1191     /* Write SEI, SPS and PPS. */
1192     x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
1193     if( x264_sei_version_write( h, &h->out.bs ) )
1194         return -1;
1195     if( x264_nal_end( h ) )
1196         return -1;
1197
1198     /* generate sequence parameters */
1199     x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
1200     x264_sps_write( &h->out.bs, h->sps );
1201     if( x264_nal_end( h ) )
1202         return -1;
1203
1204     /* generate picture parameters */
1205     x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
1206     x264_pps_write( &h->out.bs, h->pps );
1207     if( x264_nal_end( h ) )
1208         return -1;
1209
1210     frame_size = x264_encoder_encapsulate_nals( h );
1211
1212     /* now set output*/
1213     *pi_nal = h->out.i_nal;
1214     *pp_nal = &h->out.nal[0];
1215     h->out.i_nal = 0;
1216
1217     return frame_size;
1218 }
1219
1220 /* Check to see whether we have chosen a reference list ordering different
1221  * from the standard's default. */
1222 static inline void x264_reference_check_reorder( x264_t *h )
1223 {
1224     int i;
1225     for( i = 0; i < h->i_ref0 - 1; i++ )
1226         /* P and B-frames use different default orders. */
1227         if( h->sh.i_type == SLICE_TYPE_P ? h->fref0[i]->i_frame_num < h->fref0[i+1]->i_frame_num
1228                                          : h->fref0[i]->i_poc < h->fref0[i+1]->i_poc )
1229         {
1230             h->b_ref_reorder[0] = 1;
1231             break;
1232         }
1233 }
1234
1235 /* return -1 on failure, else return the index of the new reference frame */
1236 int x264_weighted_reference_duplicate( x264_t *h, int i_ref, const x264_weight_t *w )
1237 {
1238     int i = h->i_ref0;
1239     int j;
1240     x264_frame_t *newframe;
1241     if( i <= 1 ) /* empty list, definitely can't duplicate frame */
1242         return -1;
1243
1244     /* Find a place to insert the duplicate in the reference list. */
1245     for( j = 0; j < i; j++ )
1246         if( h->fref0[i_ref]->i_frame != h->fref0[j]->i_frame )
1247         {
1248             /* found a place, after j, make sure there is not already a duplicate there */
1249             if( j == i-1 || ( h->fref0[j+1] && h->fref0[i_ref]->i_frame != h->fref0[j+1]->i_frame ) )
1250                 break;
1251         }
1252
1253     if( j == i ) /* No room in the reference list for the duplicate. */
1254         return -1;
1255     j++;
1256
1257     newframe = x264_frame_pop_blank_unused( h );
1258
1259     //FIXME: probably don't need to copy everything
1260     *newframe = *h->fref0[i_ref];
1261     newframe->i_reference_count = 1;
1262     newframe->orig = h->fref0[i_ref];
1263     newframe->b_duplicate = 1;
1264     memcpy( h->fenc->weight[j], w, sizeof(h->fenc->weight[i]) );
1265
1266     /* shift the frames to make space for the dupe. */
1267     h->b_ref_reorder[0] = 1;
1268     if( h->i_ref0 < 16 )
1269         ++h->i_ref0;
1270     h->fref0[15] = NULL;
1271     x264_frame_unshift( &h->fref0[j], newframe );
1272
1273     return j;
1274 }
1275
1276 static void x264_weighted_pred_init( x264_t *h )
1277 {
1278     int i_ref;
1279     int i;
1280
1281     /* for now no analysis and set all weights to nothing */
1282     for( i_ref = 0; i_ref < h->i_ref0; i_ref++ )
1283         h->fenc->weighted[i_ref] = h->fref0[i_ref]->filtered[0];
1284
1285     // FIXME: This only supports weighting of one reference frame
1286     // and duplicates of that frame.
1287     h->fenc->i_lines_weighted = 0;
1288
1289     for( i_ref = 0; i_ref < (h->i_ref0 << h->sh.b_mbaff); i_ref++ )
1290         for( i = 0; i < 3; i++ )
1291             h->sh.weight[i_ref][i].weightfn = NULL;
1292
1293
1294     if( h->sh.i_type != SLICE_TYPE_P || h->param.analyse.i_weighted_pred <= 0 )
1295         return;
1296
1297     int i_padv = PADV << h->param.b_interlaced;
1298     int denom = -1;
1299     int weightluma = 0;
1300     int buffer_next = 0;
1301     int j;
1302     //FIXME: when chroma support is added, move this into loop
1303     h->sh.weight[0][1].weightfn = h->sh.weight[0][2].weightfn = NULL;
1304     h->sh.weight[0][1].i_denom = h->sh.weight[0][2].i_denom = 0;
1305     for( j = 0; j < h->i_ref0; j++ )
1306     {
1307         if( h->fenc->weight[j][0].weightfn )
1308         {
1309             h->sh.weight[j][0] = h->fenc->weight[j][0];
1310             // if weight is useless, don't write it to stream
1311             if( h->sh.weight[j][0].i_scale == 1<<h->sh.weight[j][0].i_denom && h->sh.weight[j][0].i_offset == 0 )
1312                 h->sh.weight[j][0].weightfn = NULL;
1313             else
1314             {
1315                 if( !weightluma )
1316                 {
1317                     weightluma = 1;
1318                     h->sh.weight[0][0].i_denom = denom = h->sh.weight[j][0].i_denom;
1319                     assert( x264_clip3( denom, 0, 7 ) == denom );
1320                 }
1321                 assert( h->sh.weight[j][0].i_denom == denom );
1322                 assert( x264_clip3( h->sh.weight[j][0].i_scale, 0, 127 ) == h->sh.weight[j][0].i_scale );
1323                 assert( x264_clip3( h->sh.weight[j][0].i_offset, -128, 127 ) == h->sh.weight[j][0].i_offset );
1324                 h->fenc->weighted[j] = h->mb.p_weight_buf[buffer_next++] +
1325                     h->fenc->i_stride[0] * i_padv + PADH;
1326             }
1327         }
1328
1329         //scale full resolution frame
1330         if( h->sh.weight[j][0].weightfn && h->param.i_threads == 1 )
1331         {
1332             uint8_t *src = h->fref0[j]->filtered[0] - h->fref0[j]->i_stride[0]*i_padv - PADH;
1333             uint8_t *dst = h->fenc->weighted[j] - h->fenc->i_stride[0]*i_padv - PADH;
1334             int stride = h->fenc->i_stride[0];
1335             int width = h->fenc->i_width[0] + PADH*2;
1336             int height = h->fenc->i_lines[0] + i_padv*2;
1337             x264_weight_scale_plane( h, dst, stride, src, stride, width, height, &h->sh.weight[j][0] );
1338             h->fenc->i_lines_weighted = height;
1339         }
1340     }
1341     if( !weightluma )
1342         h->sh.weight[0][0].i_denom = 0;
1343 }
1344
1345 static inline void x264_reference_build_list( x264_t *h, int i_poc )
1346 {
1347     int i, b_ok;
1348
1349     /* build ref list 0/1 */
1350     h->mb.pic.i_fref[0] = h->i_ref0 = 0;
1351     h->mb.pic.i_fref[1] = h->i_ref1 = 0;
1352     if( h->sh.i_type == SLICE_TYPE_I )
1353         return;
1354
1355     for( i = 0; h->frames.reference[i]; i++ )
1356     {
1357         if( h->frames.reference[i]->i_poc < i_poc )
1358         {
1359             h->fref0[h->i_ref0++] = h->frames.reference[i];
1360         }
1361         else if( h->frames.reference[i]->i_poc > i_poc )
1362         {
1363             h->fref1[h->i_ref1++] = h->frames.reference[i];
1364         }
1365     }
1366
1367     /* Order ref0 from higher to lower poc */
1368     do
1369     {
1370         b_ok = 1;
1371         for( i = 0; i < h->i_ref0 - 1; i++ )
1372         {
1373             if( h->fref0[i]->i_poc < h->fref0[i+1]->i_poc )
1374             {
1375                 XCHG( x264_frame_t*, h->fref0[i], h->fref0[i+1] );
1376                 b_ok = 0;
1377                 break;
1378             }
1379         }
1380     } while( !b_ok );
1381
1382     if( h->sh.i_mmco_remove_from_end )
1383         for( i = h->i_ref0-1; i >= h->i_ref0 - h->sh.i_mmco_remove_from_end; i-- )
1384         {
1385             int diff = h->i_frame_num - h->fref0[i]->i_frame_num;
1386             h->sh.mmco[h->sh.i_mmco_command_count].i_poc = h->fref0[i]->i_poc;
1387             h->sh.mmco[h->sh.i_mmco_command_count++].i_difference_of_pic_nums = diff;
1388         }
1389
1390     /* Order ref1 from lower to higher poc (bubble sort) for B-frame */
1391     do
1392     {
1393         b_ok = 1;
1394         for( i = 0; i < h->i_ref1 - 1; i++ )
1395         {
1396             if( h->fref1[i]->i_poc > h->fref1[i+1]->i_poc )
1397             {
1398                 XCHG( x264_frame_t*, h->fref1[i], h->fref1[i+1] );
1399                 b_ok = 0;
1400                 break;
1401             }
1402         }
1403     } while( !b_ok );
1404
1405     x264_reference_check_reorder( h );
1406
1407     h->i_ref1 = X264_MIN( h->i_ref1, h->frames.i_max_ref1 );
1408     h->i_ref0 = X264_MIN( h->i_ref0, h->frames.i_max_ref0 );
1409     h->i_ref0 = X264_MIN( h->i_ref0, h->param.i_frame_reference ); // if reconfig() has lowered the limit
1410
1411     /* add duplicates */
1412     if( h->fenc->i_type == X264_TYPE_P )
1413     {
1414         int idx = -1;
1415         if( h->param.analyse.i_weighted_pred == X264_WEIGHTP_SMART )
1416         {
1417             x264_weight_t w[3];
1418             w[1].weightfn = w[2].weightfn = NULL;
1419             if( h->param.rc.b_stat_read )
1420                 x264_ratecontrol_set_weights( h, h->fenc );
1421
1422             if( !h->fenc->weight[0][0].weightfn )
1423             {
1424                 h->fenc->weight[0][0].i_denom = 0;
1425                 SET_WEIGHT( w[0], 1, 1, 0, -1 );
1426                 idx = x264_weighted_reference_duplicate( h, 0, w );
1427             }
1428             else
1429             {
1430                 if( h->fenc->weight[0][0].i_scale == 1<<h->fenc->weight[0][0].i_denom )
1431                 {
1432                     SET_WEIGHT( h->fenc->weight[0][0], 1, 1, 0, h->fenc->weight[0][0].i_offset );
1433                 }
1434                 x264_weighted_reference_duplicate( h, 0, weight_none );
1435                 if( h->fenc->weight[0][0].i_offset > -128 )
1436                 {
1437                     w[0] = h->fenc->weight[0][0];
1438                     w[0].i_offset--;
1439                     h->mc.weight_cache( h, &w[0] );
1440                     idx = x264_weighted_reference_duplicate( h, 0, w );
1441                 }
1442             }
1443         }
1444         else if( h->param.analyse.i_weighted_pred == X264_WEIGHTP_BLIND )
1445         {
1446             //weighted offset=-1
1447             x264_weight_t w[3];
1448             SET_WEIGHT( w[0], 1, 1, 0, -1 );
1449             h->fenc->weight[0][0].i_denom = 0;
1450             w[1].weightfn = w[2].weightfn = NULL;
1451             idx = x264_weighted_reference_duplicate( h, 0, w );
1452         }
1453         h->mb.ref_blind_dupe = idx;
1454     }
1455
1456     assert( h->i_ref0 + h->i_ref1 <= 16 );
1457     h->mb.pic.i_fref[0] = h->i_ref0;
1458     h->mb.pic.i_fref[1] = h->i_ref1;
1459 }
1460
1461 static void x264_fdec_filter_row( x264_t *h, int mb_y )
1462 {
1463     /* mb_y is the mb to be encoded next, not the mb to be filtered here */
1464     int b_hpel = h->fdec->b_kept_as_ref;
1465     int b_deblock = !h->sh.i_disable_deblocking_filter_idc;
1466     int b_end = mb_y == h->sps->i_mb_height;
1467     int min_y = mb_y - (1 << h->sh.b_mbaff);
1468     int max_y = b_end ? h->sps->i_mb_height : mb_y;
1469     b_deblock &= b_hpel || h->param.psz_dump_yuv;
1470     if( mb_y & h->sh.b_mbaff )
1471         return;
1472     if( min_y < 0 )
1473         return;
1474
1475     if( !b_end && !h->param.b_sliced_threads )
1476     {
1477         int i, j;
1478         for( j=0; j<=h->sh.b_mbaff; j++ )
1479             for( i=0; i<3; i++ )
1480             {
1481                 memcpy( h->mb.intra_border_backup[j][i],
1482                         h->fdec->plane[i] + ((mb_y*16 >> !!i) + j - 1 - h->sh.b_mbaff) * h->fdec->i_stride[i],
1483                         h->sps->i_mb_width*16 >> !!i );
1484             }
1485     }
1486
1487     if( b_deblock )
1488     {
1489         int y;
1490         for( y = min_y; y < max_y; y += (1 << h->sh.b_mbaff) )
1491             x264_frame_deblock_row( h, y );
1492     }
1493
1494     if( b_hpel )
1495     {
1496         x264_frame_expand_border( h, h->fdec, min_y, b_end );
1497         if( h->param.analyse.i_subpel_refine )
1498         {
1499             x264_frame_filter( h, h->fdec, min_y, b_end );
1500             x264_frame_expand_border_filtered( h, h->fdec, min_y, b_end );
1501         }
1502     }
1503
1504     if( h->param.i_threads > 1 && h->fdec->b_kept_as_ref && !h->param.b_sliced_threads )
1505         x264_frame_cond_broadcast( h->fdec, mb_y*16 + (b_end ? 10000 : -(X264_THREAD_HEIGHT << h->sh.b_mbaff)) );
1506
1507     min_y = X264_MAX( min_y*16-8, 0 );
1508     max_y = b_end ? h->param.i_height : mb_y*16-8;
1509
1510     if( h->param.analyse.b_psnr )
1511     {
1512         int i;
1513         for( i=0; i<3; i++ )
1514             h->stat.frame.i_ssd[i] +=
1515                 x264_pixel_ssd_wxh( &h->pixf,
1516                     h->fdec->plane[i] + (min_y>>!!i) * h->fdec->i_stride[i], h->fdec->i_stride[i],
1517                     h->fenc->plane[i] + (min_y>>!!i) * h->fenc->i_stride[i], h->fenc->i_stride[i],
1518                     h->param.i_width >> !!i, (max_y-min_y) >> !!i );
1519     }
1520
1521     if( h->param.analyse.b_ssim )
1522     {
1523         x264_emms();
1524         /* offset by 2 pixels to avoid alignment of ssim blocks with dct blocks,
1525          * and overlap by 4 */
1526         min_y += min_y == 0 ? 2 : -6;
1527         h->stat.frame.f_ssim +=
1528             x264_pixel_ssim_wxh( &h->pixf,
1529                 h->fdec->plane[0] + 2+min_y*h->fdec->i_stride[0], h->fdec->i_stride[0],
1530                 h->fenc->plane[0] + 2+min_y*h->fenc->i_stride[0], h->fenc->i_stride[0],
1531                 h->param.i_width-2, max_y-min_y, h->scratch_buffer );
1532     }
1533 }
1534
1535 static inline int x264_reference_update( x264_t *h )
1536 {
1537     int i, j;
1538     if( !h->fdec->b_kept_as_ref )
1539     {
1540         if( h->param.i_threads > 1 && !h->param.b_sliced_threads )
1541         {
1542             x264_frame_push_unused( h, h->fdec );
1543             h->fdec = x264_frame_pop_unused( h, 1 );
1544             if( !h->fdec )
1545                 return -1;
1546         }
1547         return 0;
1548     }
1549
1550     /* apply mmco from previous frame. */
1551     for( i = 0; i < h->sh.i_mmco_command_count; i++ )
1552         for( j = 0; h->frames.reference[j]; j++ )
1553             if( h->frames.reference[j]->i_poc == h->sh.mmco[i].i_poc )
1554                 x264_frame_push_unused( h, x264_frame_shift( &h->frames.reference[j] ) );
1555
1556     /* move frame in the buffer */
1557     x264_frame_push( h->frames.reference, h->fdec );
1558     if( h->frames.reference[h->sps->i_num_ref_frames] )
1559         x264_frame_push_unused( h, x264_frame_shift( h->frames.reference ) );
1560     h->fdec = x264_frame_pop_unused( h, 1 );
1561     if( !h->fdec )
1562         return -1;
1563     return 0;
1564 }
1565
1566 static inline void x264_reference_reset( x264_t *h )
1567 {
1568     while( h->frames.reference[0] )
1569         x264_frame_push_unused( h, x264_frame_pop( h->frames.reference ) );
1570     h->fdec->i_poc =
1571     h->fenc->i_poc = 0;
1572 }
1573
1574 static inline void x264_reference_hierarchy_reset( x264_t *h )
1575 {
1576     int i, ref;
1577     int b_hasdelayframe = 0;
1578     if( !h->param.i_bframe_pyramid )
1579         return;
1580
1581     /* look for delay frames -- chain must only contain frames that are disposable */
1582     for( i = 0; h->frames.current[i] && IS_DISPOSABLE( h->frames.current[i]->i_type ); i++ )
1583         b_hasdelayframe |= h->frames.current[i]->i_coded
1584                         != h->frames.current[i]->i_frame + h->sps->vui.i_num_reorder_frames;
1585
1586     if( h->param.i_bframe_pyramid != X264_B_PYRAMID_STRICT && !b_hasdelayframe )
1587         return;
1588
1589     /* Remove last BREF. There will never be old BREFs in the
1590      * dpb during a BREF decode when pyramid == STRICT */
1591     for( ref = 0; h->frames.reference[ref]; ref++ )
1592     {
1593         if( h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT
1594             && h->frames.reference[ref]->i_type == X264_TYPE_BREF )
1595         {
1596             int diff = h->i_frame_num - h->frames.reference[ref]->i_frame_num;
1597             h->sh.mmco[h->sh.i_mmco_command_count].i_difference_of_pic_nums = diff;
1598             h->sh.mmco[h->sh.i_mmco_command_count++].i_poc = h->frames.reference[ref]->i_poc;
1599             x264_frame_push_unused( h, x264_frame_pop( h->frames.reference ) );
1600             h->b_ref_reorder[0] = 1;
1601             break;
1602         }
1603     }
1604
1605     /* Prepare to room in the dpb for the delayed display time of the later b-frame's */
1606     h->sh.i_mmco_remove_from_end = X264_MAX( ref + 2 - h->frames.i_max_dpb, 0 );
1607 }
1608
1609 static inline void x264_slice_init( x264_t *h, int i_nal_type, int i_global_qp )
1610 {
1611     /* ------------------------ Create slice header  ----------------------- */
1612     if( i_nal_type == NAL_SLICE_IDR )
1613     {
1614         x264_slice_header_init( h, &h->sh, h->sps, h->pps, h->i_idr_pic_id, h->i_frame_num, i_global_qp );
1615
1616         /* increment id */
1617         h->i_idr_pic_id = ( h->i_idr_pic_id + 1 ) % 65536;
1618     }
1619     else
1620     {
1621         x264_slice_header_init( h, &h->sh, h->sps, h->pps, -1, h->i_frame_num, i_global_qp );
1622
1623         /* always set the real higher num of ref frame used */
1624         h->sh.b_num_ref_idx_override = 1;
1625         h->sh.i_num_ref_idx_l0_active = h->i_ref0 <= 0 ? 1 : h->i_ref0;
1626         h->sh.i_num_ref_idx_l1_active = h->i_ref1 <= 0 ? 1 : h->i_ref1;
1627     }
1628
1629     h->fdec->i_frame_num = h->sh.i_frame_num;
1630
1631     if( h->sps->i_poc_type == 0 )
1632     {
1633         h->sh.i_poc_lsb = h->fdec->i_poc & ( (1 << h->sps->i_log2_max_poc_lsb) - 1 );
1634         h->sh.i_delta_poc_bottom = 0;
1635     }
1636     else if( h->sps->i_poc_type == 1 )
1637     {
1638         /* FIXME TODO FIXME */
1639     }
1640     else
1641     {
1642         /* Nothing to do ? */
1643     }
1644
1645     x264_macroblock_slice_init( h );
1646 }
1647
1648 static int x264_slice_write( x264_t *h )
1649 {
1650     int i_skip;
1651     int mb_xy, i_mb_x, i_mb_y;
1652     int i, i_list, i_ref, i_skip_bak = 0; /* Shut up GCC. */
1653     bs_t bs_bak;
1654     x264_cabac_t cabac_bak;
1655     uint8_t cabac_prevbyte_bak = 0; /* Shut up GCC. */
1656     /* Assume no more than 3 bytes of NALU escaping. */
1657     int slice_max_size = h->param.i_slice_max_size > 0 ? (h->param.i_slice_max_size-3-NALU_OVERHEAD)*8 : INT_MAX;
1658     int starting_bits = bs_pos(&h->out.bs);
1659     bs_realign( &h->out.bs );
1660
1661     /* Slice */
1662     x264_nal_start( h, h->i_nal_type, h->i_nal_ref_idc );
1663
1664     /* Slice header */
1665     x264_macroblock_thread_init( h );
1666     x264_slice_header_write( &h->out.bs, &h->sh, h->i_nal_ref_idc );
1667     if( h->param.b_cabac )
1668     {
1669         /* alignment needed */
1670         bs_align_1( &h->out.bs );
1671
1672         /* init cabac */
1673         x264_cabac_context_init( &h->cabac, h->sh.i_type, h->sh.i_qp, h->sh.i_cabac_init_idc );
1674         x264_cabac_encode_init ( &h->cabac, h->out.bs.p, h->out.bs.p_end );
1675     }
1676     h->mb.i_last_qp = h->sh.i_qp;
1677     h->mb.i_last_dqp = 0;
1678
1679     i_mb_y = h->sh.i_first_mb / h->sps->i_mb_width;
1680     i_mb_x = h->sh.i_first_mb % h->sps->i_mb_width;
1681     i_skip = 0;
1682
1683     while( (mb_xy = i_mb_x + i_mb_y * h->sps->i_mb_width) <= h->sh.i_last_mb )
1684     {
1685         int mb_spos = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
1686         if( h->param.i_slice_max_size > 0 )
1687         {
1688             /* We don't need the contexts because flushing the CABAC encoder has no context
1689              * dependency and macroblocks are only re-encoded in the case where a slice is
1690              * ended (and thus the content of all contexts are thrown away). */
1691             if( h->param.b_cabac )
1692             {
1693                 memcpy( &cabac_bak, &h->cabac, offsetof(x264_cabac_t, f8_bits_encoded) );
1694                 /* x264's CABAC writer modifies the previous byte during carry, so it has to be
1695                  * backed up. */
1696                 cabac_prevbyte_bak = h->cabac.p[-1];
1697             }
1698             else
1699             {
1700                 bs_bak = h->out.bs;
1701                 i_skip_bak = i_skip;
1702             }
1703         }
1704
1705         if( i_mb_x == 0 && !h->mb.b_reencode_mb && !h->param.b_sliced_threads )
1706             x264_fdec_filter_row( h, i_mb_y );
1707
1708         /* load cache */
1709         x264_macroblock_cache_load( h, i_mb_x, i_mb_y );
1710
1711         x264_macroblock_analyse( h );
1712
1713         /* encode this macroblock -> be careful it can change the mb type to P_SKIP if needed */
1714         x264_macroblock_encode( h );
1715
1716         if( x264_bitstream_check_buffer( h ) )
1717             return -1;
1718
1719         if( h->param.b_cabac )
1720         {
1721             if( mb_xy > h->sh.i_first_mb && !(h->sh.b_mbaff && (i_mb_y&1)) )
1722                 x264_cabac_encode_terminal( &h->cabac );
1723
1724             if( IS_SKIP( h->mb.i_type ) )
1725                 x264_cabac_mb_skip( h, 1 );
1726             else
1727             {
1728                 if( h->sh.i_type != SLICE_TYPE_I )
1729                     x264_cabac_mb_skip( h, 0 );
1730                 x264_macroblock_write_cabac( h, &h->cabac );
1731             }
1732         }
1733         else
1734         {
1735             if( IS_SKIP( h->mb.i_type ) )
1736                 i_skip++;
1737             else
1738             {
1739                 if( h->sh.i_type != SLICE_TYPE_I )
1740                 {
1741                     bs_write_ue( &h->out.bs, i_skip );  /* skip run */
1742                     i_skip = 0;
1743                 }
1744                 x264_macroblock_write_cavlc( h );
1745             }
1746         }
1747
1748         int total_bits = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
1749         int mb_size = total_bits - mb_spos;
1750
1751         /* We'll just re-encode this last macroblock if we go over the max slice size. */
1752         if( total_bits - starting_bits > slice_max_size && !h->mb.b_reencode_mb )
1753         {
1754             if( mb_xy != h->sh.i_first_mb )
1755             {
1756                 if( h->param.b_cabac )
1757                 {
1758                     memcpy( &h->cabac, &cabac_bak, offsetof(x264_cabac_t, f8_bits_encoded) );
1759                     h->cabac.p[-1] = cabac_prevbyte_bak;
1760                 }
1761                 else
1762                 {
1763                     h->out.bs = bs_bak;
1764                     i_skip = i_skip_bak;
1765                 }
1766                 h->mb.b_reencode_mb = 1;
1767                 h->sh.i_last_mb = mb_xy-1;
1768                 break;
1769             }
1770             else
1771             {
1772                 h->sh.i_last_mb = mb_xy;
1773                 h->mb.b_reencode_mb = 0;
1774             }
1775         }
1776         else
1777             h->mb.b_reencode_mb = 0;
1778
1779 #if VISUALIZE
1780         if( h->param.b_visualize )
1781             x264_visualize_mb( h );
1782 #endif
1783
1784         /* save cache */
1785         x264_macroblock_cache_save( h );
1786
1787         /* accumulate mb stats */
1788         h->stat.frame.i_mb_count[h->mb.i_type]++;
1789
1790         if( !IS_INTRA(h->mb.i_type) && !IS_SKIP(h->mb.i_type) && !IS_DIRECT(h->mb.i_type) )
1791         {
1792             if( h->mb.i_partition != D_8x8 )
1793                     h->stat.frame.i_mb_partition[h->mb.i_partition] += 4;
1794                 else
1795                     for( i = 0; i < 4; i++ )
1796                         h->stat.frame.i_mb_partition[h->mb.i_sub_partition[i]] ++;
1797             if( h->param.i_frame_reference > 1 )
1798                 for( i_list = 0; i_list <= (h->sh.i_type == SLICE_TYPE_B); i_list++ )
1799                     for( i = 0; i < 4; i++ )
1800                     {
1801                         i_ref = h->mb.cache.ref[i_list][ x264_scan8[4*i] ];
1802                         if( i_ref >= 0 )
1803                             h->stat.frame.i_mb_count_ref[i_list][i_ref] ++;
1804                     }
1805         }
1806
1807         if( h->param.i_log_level >= X264_LOG_INFO )
1808         {
1809             if( h->mb.i_cbp_luma || h->mb.i_cbp_chroma )
1810             {
1811                 int cbpsum = (h->mb.i_cbp_luma&1) + ((h->mb.i_cbp_luma>>1)&1)
1812                            + ((h->mb.i_cbp_luma>>2)&1) + (h->mb.i_cbp_luma>>3);
1813                 int b_intra = IS_INTRA(h->mb.i_type);
1814                 h->stat.frame.i_mb_cbp[!b_intra + 0] += cbpsum;
1815                 h->stat.frame.i_mb_cbp[!b_intra + 2] += h->mb.i_cbp_chroma >= 1;
1816                 h->stat.frame.i_mb_cbp[!b_intra + 4] += h->mb.i_cbp_chroma == 2;
1817             }
1818             if( h->mb.i_cbp_luma && !IS_INTRA(h->mb.i_type) )
1819             {
1820                 h->stat.frame.i_mb_count_8x8dct[0] ++;
1821                 h->stat.frame.i_mb_count_8x8dct[1] += h->mb.b_transform_8x8;
1822             }
1823             if( IS_INTRA(h->mb.i_type) && h->mb.i_type != I_PCM )
1824             {
1825                 if( h->mb.i_type == I_16x16 )
1826                     h->stat.frame.i_mb_pred_mode[0][h->mb.i_intra16x16_pred_mode]++;
1827                 else if( h->mb.i_type == I_8x8 )
1828                     for( i = 0; i < 16; i += 4 )
1829                         h->stat.frame.i_mb_pred_mode[1][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
1830                 else //if( h->mb.i_type == I_4x4 )
1831                     for( i = 0; i < 16; i++ )
1832                         h->stat.frame.i_mb_pred_mode[2][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
1833             }
1834         }
1835
1836         x264_ratecontrol_mb( h, mb_size );
1837
1838         if( h->sh.b_mbaff )
1839         {
1840             i_mb_x += i_mb_y & 1;
1841             i_mb_y ^= i_mb_x < h->sps->i_mb_width;
1842         }
1843         else
1844             i_mb_x++;
1845         if( i_mb_x == h->sps->i_mb_width )
1846         {
1847             i_mb_y++;
1848             i_mb_x = 0;
1849         }
1850     }
1851
1852     if( h->param.b_cabac )
1853     {
1854         x264_cabac_encode_flush( h, &h->cabac );
1855         h->out.bs.p = h->cabac.p;
1856     }
1857     else
1858     {
1859         if( i_skip > 0 )
1860             bs_write_ue( &h->out.bs, i_skip );  /* last skip run */
1861         /* rbsp_slice_trailing_bits */
1862         bs_rbsp_trailing( &h->out.bs );
1863         bs_flush( &h->out.bs );
1864     }
1865     if( x264_nal_end( h ) )
1866         return -1;
1867
1868     if( h->sh.i_last_mb == h->mb.i_mb_count-1 )
1869     {
1870         h->stat.frame.i_misc_bits = bs_pos( &h->out.bs )
1871                                   + (h->out.i_nal*NALU_OVERHEAD * 8)
1872                                   - h->stat.frame.i_tex_bits
1873                                   - h->stat.frame.i_mv_bits;
1874         if( !h->param.b_sliced_threads )
1875             x264_fdec_filter_row( h, h->sps->i_mb_height );
1876     }
1877
1878     return 0;
1879 }
1880
1881 static void x264_thread_sync_context( x264_t *dst, x264_t *src )
1882 {
1883     if( dst == src )
1884         return;
1885
1886     // reference counting
1887     x264_frame_t **f;
1888     for( f = src->frames.reference; *f; f++ )
1889         (*f)->i_reference_count++;
1890     for( f = dst->frames.reference; *f; f++ )
1891         x264_frame_push_unused( src, *f );
1892     src->fdec->i_reference_count++;
1893     x264_frame_push_unused( src, dst->fdec );
1894
1895     // copy everything except the per-thread pointers and the constants.
1896     memcpy( &dst->i_frame, &src->i_frame, offsetof(x264_t, mb.type) - offsetof(x264_t, i_frame) );
1897     dst->param = src->param;
1898     dst->stat = src->stat;
1899 }
1900
1901 static void x264_thread_sync_stat( x264_t *dst, x264_t *src )
1902 {
1903     if( dst == src )
1904         return;
1905     memcpy( &dst->stat.i_frame_count, &src->stat.i_frame_count, sizeof(dst->stat) - sizeof(dst->stat.frame) );
1906 }
1907
1908 static void *x264_slices_write( x264_t *h )
1909 {
1910     int i_slice_num = 0;
1911     int last_thread_mb = h->sh.i_last_mb;
1912     if( h->param.i_sync_lookahead )
1913         x264_lower_thread_priority( 10 );
1914
1915 #ifdef HAVE_MMX
1916     /* Misalign mask has to be set separately for each thread. */
1917     if( h->param.cpu&X264_CPU_SSE_MISALIGN )
1918         x264_cpu_mask_misalign_sse();
1919 #endif
1920
1921 #if VISUALIZE
1922     if( h->param.b_visualize )
1923         if( x264_visualize_init( h ) )
1924             return (void *)-1;
1925 #endif
1926
1927     /* init stats */
1928     memset( &h->stat.frame, 0, sizeof(h->stat.frame) );
1929     h->mb.b_reencode_mb = 0;
1930     while( h->sh.i_first_mb <= last_thread_mb )
1931     {
1932         h->sh.i_last_mb = last_thread_mb;
1933         if( h->param.i_slice_max_mbs )
1934             h->sh.i_last_mb = h->sh.i_first_mb + h->param.i_slice_max_mbs - 1;
1935         else if( h->param.i_slice_count && !h->param.b_sliced_threads )
1936         {
1937             int height = h->sps->i_mb_height >> h->param.b_interlaced;
1938             int width = h->sps->i_mb_width << h->param.b_interlaced;
1939             i_slice_num++;
1940             h->sh.i_last_mb = (height * i_slice_num + h->param.i_slice_count/2) / h->param.i_slice_count * width - 1;
1941         }
1942         h->sh.i_last_mb = X264_MIN( h->sh.i_last_mb, last_thread_mb );
1943         if( x264_stack_align( x264_slice_write, h ) )
1944             return (void *)-1;
1945         h->sh.i_first_mb = h->sh.i_last_mb + 1;
1946     }
1947
1948 #if VISUALIZE
1949     if( h->param.b_visualize )
1950     {
1951         x264_visualize_show( h );
1952         x264_visualize_close( h );
1953     }
1954 #endif
1955
1956     return (void *)0;
1957 }
1958
1959 static int x264_threaded_slices_write( x264_t *h )
1960 {
1961     int i, j;
1962     void *ret = NULL;
1963     /* set first/last mb and sync contexts */
1964     for( i = 0; i < h->param.i_threads; i++ )
1965     {
1966         x264_t *t = h->thread[i];
1967         if( i )
1968         {
1969             t->param = h->param;
1970             memcpy( &t->i_frame, &h->i_frame, offsetof(x264_t, rc) - offsetof(x264_t, i_frame) );
1971         }
1972         int height = h->sps->i_mb_height >> h->param.b_interlaced;
1973         t->i_threadslice_start = ((height *  i    + h->param.i_slice_count/2) / h->param.i_threads) << h->param.b_interlaced;
1974         t->i_threadslice_end   = ((height * (i+1) + h->param.i_slice_count/2) / h->param.i_threads) << h->param.b_interlaced;
1975         t->sh.i_first_mb = t->i_threadslice_start * h->sps->i_mb_width;
1976         t->sh.i_last_mb  =   t->i_threadslice_end * h->sps->i_mb_width - 1;
1977     }
1978
1979     x264_analyse_weight_frame( h, h->sps->i_mb_height*16 + 16 );
1980
1981     x264_threads_distribute_ratecontrol( h );
1982
1983     /* dispatch */
1984     for( i = 0; i < h->param.i_threads; i++ )
1985         if( x264_pthread_create( &h->thread[i]->thread_handle, NULL, (void*)x264_slices_write, (void*)h->thread[i] ) )
1986             return -1;
1987     for( i = 0; i < h->param.i_threads; i++ )
1988     {
1989         x264_pthread_join( h->thread[i]->thread_handle, &ret );
1990         if( (intptr_t)ret )
1991             return (intptr_t)ret;
1992     }
1993
1994     /* deblocking and hpel filtering */
1995     for( i = 0; i <= h->sps->i_mb_height; i++ )
1996         x264_fdec_filter_row( h, i );
1997
1998     for( i = 1; i < h->param.i_threads; i++ )
1999     {
2000         x264_t *t = h->thread[i];
2001         for( j = 0; j < t->out.i_nal; j++ )
2002         {
2003             h->out.nal[h->out.i_nal] = t->out.nal[j];
2004             h->out.i_nal++;
2005             x264_nal_check_buffer( h );
2006         }
2007         /* All entries in stat.frame are ints except for ssd/ssim,
2008          * which are only calculated in the main thread. */
2009         for( j = 0; j < (offsetof(x264_t,stat.frame.i_ssd) - offsetof(x264_t,stat.frame.i_mv_bits)) / sizeof(int); j++ )
2010             ((int*)&h->stat.frame)[j] += ((int*)&t->stat.frame)[j];
2011     }
2012
2013     x264_threads_merge_ratecontrol( h );
2014
2015     return 0;
2016 }
2017
2018 /****************************************************************************
2019  * x264_encoder_encode:
2020  *  XXX: i_poc   : is the poc of the current given picture
2021  *       i_frame : is the number of the frame being coded
2022  *  ex:  type frame poc
2023  *       I      0   2*0
2024  *       P      1   2*3
2025  *       B      2   2*1
2026  *       B      3   2*2
2027  *       P      4   2*6
2028  *       B      5   2*4
2029  *       B      6   2*5
2030  ****************************************************************************/
2031 int     x264_encoder_encode( x264_t *h,
2032                              x264_nal_t **pp_nal, int *pi_nal,
2033                              x264_picture_t *pic_in,
2034                              x264_picture_t *pic_out )
2035 {
2036     x264_t *thread_current, *thread_prev, *thread_oldest;
2037     int i_nal_type, i_nal_ref_idc, i_global_qp, i;
2038
2039     if( h->param.i_threads > 1 && !h->param.b_sliced_threads )
2040     {
2041         thread_prev    = h->thread[ h->i_thread_phase ];
2042         h->i_thread_phase = (h->i_thread_phase + 1) % h->param.i_threads;
2043         thread_current = h->thread[ h->i_thread_phase ];
2044         thread_oldest  = h->thread[ (h->i_thread_phase + 1) % h->param.i_threads ];
2045         x264_thread_sync_context( thread_current, thread_prev );
2046         x264_thread_sync_ratecontrol( thread_current, thread_prev, thread_oldest );
2047         h = thread_current;
2048 //      fprintf(stderr, "current: %p  prev: %p  oldest: %p \n", thread_current, thread_prev, thread_oldest);
2049     }
2050     else
2051     {
2052         thread_current =
2053         thread_oldest  = h;
2054     }
2055
2056     // ok to call this before encoding any frames, since the initial values of fdec have b_kept_as_ref=0
2057     if( x264_reference_update( h ) )
2058         return -1;
2059     h->fdec->i_lines_completed = -1;
2060
2061     /* no data out */
2062     *pi_nal = 0;
2063     *pp_nal = NULL;
2064
2065     /* ------------------- Setup new frame from picture -------------------- */
2066     if( pic_in != NULL )
2067     {
2068         /* 1: Copy the picture to a frame and move it to a buffer */
2069         x264_frame_t *fenc = x264_frame_pop_unused( h, 0 );
2070         if( !fenc )
2071             return -1;
2072
2073         if( x264_frame_copy_picture( h, fenc, pic_in ) < 0 )
2074             return -1;
2075
2076         if( h->param.i_width != 16 * h->sps->i_mb_width ||
2077             h->param.i_height != 16 * h->sps->i_mb_height )
2078             x264_frame_expand_border_mod16( h, fenc );
2079
2080         fenc->i_frame = h->frames.i_input++;
2081
2082         if( h->frames.i_bframe_delay && fenc->i_frame == h->frames.i_bframe_delay )
2083             h->frames.i_bframe_delay_time = fenc->i_pts;
2084
2085         if( h->frames.b_have_lowres )
2086         {
2087             if( h->param.analyse.i_weighted_pred == X264_WEIGHTP_FAKE || h->param.analyse.i_weighted_pred == X264_WEIGHTP_SMART )
2088                 x264_weight_plane_analyse( h, fenc );
2089             x264_frame_init_lowres( h, fenc );
2090         }
2091
2092         if( h->param.rc.b_mb_tree && h->param.rc.b_stat_read )
2093         {
2094             if( x264_macroblock_tree_read( h, fenc ) )
2095                 return -1;
2096         }
2097         else if( h->param.rc.i_aq_mode )
2098             x264_adaptive_quant_frame( h, fenc );
2099
2100         /* 2: Place the frame into the queue for its slice type decision */
2101         x264_lookahead_put_frame( h, fenc );
2102
2103         if( h->frames.i_input <= h->frames.i_delay + (h->param.b_sliced_threads ? 0 : 1 - h->param.i_threads) )
2104         {
2105             /* Nothing yet to encode, waiting for filling of buffers */
2106             pic_out->i_type = X264_TYPE_AUTO;
2107             return 0;
2108         }
2109     }
2110     else
2111     {
2112         /* signal kills for lookahead thread */
2113         x264_pthread_mutex_lock( &h->lookahead->ifbuf.mutex );
2114         h->lookahead->b_exit_thread = 1;
2115         x264_pthread_cond_broadcast( &h->lookahead->ifbuf.cv_fill );
2116         x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
2117     }
2118
2119     h->i_frame++;
2120     /* 3: The picture is analyzed in the lookahead */
2121     if( !h->frames.current[0] )
2122         x264_lookahead_get_frames( h );
2123
2124     if( !h->frames.current[0] && x264_lookahead_is_empty( h ) )
2125         return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
2126
2127     /* ------------------- Get frame to be encoded ------------------------- */
2128     /* 4: get picture to encode */
2129     h->fenc = x264_frame_shift( h->frames.current );
2130     if( h->fenc->param )
2131     {
2132         x264_encoder_reconfig( h, h->fenc->param );
2133         if( h->fenc->param->param_free )
2134             h->fenc->param->param_free( h->fenc->param );
2135     }
2136
2137     if( h->fenc->b_keyframe )
2138     {
2139         h->frames.i_last_keyframe = h->fenc->i_frame;
2140         if( h->fenc->i_type == X264_TYPE_IDR )
2141             h->i_frame_num = 0;
2142     }
2143     h->sh.i_mmco_command_count =
2144     h->sh.i_mmco_remove_from_end = 0;
2145     h->b_ref_reorder[0] =
2146     h->b_ref_reorder[1] = 0;
2147
2148     /* ------------------- Setup frame context ----------------------------- */
2149     /* 5: Init data dependent of frame type */
2150     if( h->fenc->i_type == X264_TYPE_IDR )
2151     {
2152         /* reset ref pictures */
2153         i_nal_type    = NAL_SLICE_IDR;
2154         i_nal_ref_idc = NAL_PRIORITY_HIGHEST;
2155         h->sh.i_type = SLICE_TYPE_I;
2156         x264_reference_reset( h );
2157     }
2158     else if( h->fenc->i_type == X264_TYPE_I )
2159     {
2160         i_nal_type    = NAL_SLICE;
2161         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
2162         h->sh.i_type = SLICE_TYPE_I;
2163         x264_reference_hierarchy_reset( h );
2164     }
2165     else if( h->fenc->i_type == X264_TYPE_P )
2166     {
2167         i_nal_type    = NAL_SLICE;
2168         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
2169         h->sh.i_type = SLICE_TYPE_P;
2170         x264_reference_hierarchy_reset( h );
2171     }
2172     else if( h->fenc->i_type == X264_TYPE_BREF )
2173     {
2174         i_nal_type    = NAL_SLICE;
2175         i_nal_ref_idc = h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT ? NAL_PRIORITY_LOW : NAL_PRIORITY_HIGH;
2176         h->sh.i_type = SLICE_TYPE_B;
2177         x264_reference_hierarchy_reset( h );
2178     }
2179     else    /* B frame */
2180     {
2181         i_nal_type    = NAL_SLICE;
2182         i_nal_ref_idc = NAL_PRIORITY_DISPOSABLE;
2183         h->sh.i_type = SLICE_TYPE_B;
2184     }
2185
2186     h->fdec->i_poc =
2187     h->fenc->i_poc = 2 * (h->fenc->i_frame - h->frames.i_last_keyframe);
2188     h->fdec->i_type = h->fenc->i_type;
2189     h->fdec->i_frame = h->fenc->i_frame;
2190     h->fenc->b_kept_as_ref =
2191     h->fdec->b_kept_as_ref = i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE && h->param.i_keyint_max > 1;
2192
2193
2194
2195     /* ------------------- Init                ----------------------------- */
2196     /* build ref list 0/1 */
2197     x264_reference_build_list( h, h->fdec->i_poc );
2198
2199     /* ---------------------- Write the bitstream -------------------------- */
2200     /* Init bitstream context */
2201     if( h->param.b_sliced_threads )
2202     {
2203         for( i = 0; i < h->param.i_threads; i++ )
2204         {
2205             bs_init( &h->thread[i]->out.bs, h->thread[i]->out.p_bitstream, h->thread[i]->out.i_bitstream );
2206             h->thread[i]->out.i_nal = 0;
2207         }
2208     }
2209     else
2210     {
2211         bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
2212         h->out.i_nal = 0;
2213     }
2214
2215     if( h->param.b_aud )
2216     {
2217         int pic_type;
2218
2219         if( h->sh.i_type == SLICE_TYPE_I )
2220             pic_type = 0;
2221         else if( h->sh.i_type == SLICE_TYPE_P )
2222             pic_type = 1;
2223         else if( h->sh.i_type == SLICE_TYPE_B )
2224             pic_type = 2;
2225         else
2226             pic_type = 7;
2227
2228         x264_nal_start( h, NAL_AUD, NAL_PRIORITY_DISPOSABLE );
2229         bs_write( &h->out.bs, 3, pic_type );
2230         bs_rbsp_trailing( &h->out.bs );
2231         if( x264_nal_end( h ) )
2232             return -1;
2233     }
2234
2235     h->i_nal_type = i_nal_type;
2236     h->i_nal_ref_idc = i_nal_ref_idc;
2237
2238     int overhead = NALU_OVERHEAD;
2239
2240     if( h->param.b_intra_refresh && h->fenc->i_type == X264_TYPE_P )
2241     {
2242         int pocdiff = (h->fdec->i_poc - h->fref0[0]->i_poc)/2;
2243         float increment = ((float)h->sps->i_mb_width-1) / h->param.i_keyint_max;
2244         if( IS_X264_TYPE_I( h->fref0[0]->i_type ) )
2245             h->fdec->f_pir_position = 0;
2246         else
2247         {
2248             if( h->fref0[0]->i_pir_end_col == h->sps->i_mb_width - 1 )
2249             {
2250                 h->fdec->f_pir_position = 0;
2251                 h->fenc->b_keyframe = 1;
2252             }
2253             else
2254                 h->fdec->f_pir_position = h->fref0[0]->f_pir_position;
2255         }
2256         h->fdec->i_pir_start_col = h->fdec->f_pir_position+0.5;
2257         h->fdec->f_pir_position += increment * pocdiff;
2258         h->fdec->i_pir_end_col = X264_MIN( h->fdec->f_pir_position+0.5, h->sps->i_mb_width-1 );
2259     }
2260
2261     /* Write SPS and PPS */
2262     if( h->fenc->b_keyframe )
2263     {
2264         if( h->param.b_repeat_headers )
2265         {
2266             if( h->fenc->i_frame == 0 )
2267             {
2268                 /* identify ourself */
2269                 x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2270                 if( x264_sei_version_write( h, &h->out.bs ) )
2271                     return -1;
2272                 if( x264_nal_end( h ) )
2273                     return -1;
2274                 overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
2275             }
2276
2277             /* generate sequence parameters */
2278             x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
2279             x264_sps_write( &h->out.bs, h->sps );
2280             if( x264_nal_end( h ) )
2281                 return -1;
2282             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
2283
2284             /* generate picture parameters */
2285             x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
2286             x264_pps_write( &h->out.bs, h->pps );
2287             if( x264_nal_end( h ) )
2288                 return -1;
2289             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
2290         }
2291
2292         if( h->fenc->i_type != X264_TYPE_IDR )
2293         {
2294             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2295             x264_sei_recovery_point_write( h, &h->out.bs, h->param.i_keyint_max );
2296             x264_nal_end( h );
2297             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
2298         }
2299     }
2300
2301     /* Init the rate control */
2302     /* FIXME: Include slice header bit cost. */
2303     x264_ratecontrol_start( h, h->fenc->i_qpplus1, overhead*8 );
2304     i_global_qp = x264_ratecontrol_qp( h );
2305
2306     pic_out->i_qpplus1 =
2307     h->fdec->i_qpplus1 = i_global_qp + 1;
2308
2309     if( h->param.rc.b_stat_read && h->sh.i_type != SLICE_TYPE_I )
2310     {
2311         x264_reference_build_list_optimal( h );
2312         x264_reference_check_reorder( h );
2313     }
2314
2315     if( h->sh.i_type == SLICE_TYPE_B )
2316         x264_macroblock_bipred_init( h );
2317
2318     /*------------------------- Weights -------------------------------------*/
2319     x264_weighted_pred_init( h );
2320
2321     /* ------------------------ Create slice header  ----------------------- */
2322     x264_slice_init( h, i_nal_type, i_global_qp );
2323
2324     if( i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE )
2325         h->i_frame_num++;
2326
2327     /* Write frame */
2328     h->i_threadslice_start = 0;
2329     h->i_threadslice_end = h->sps->i_mb_height;
2330     if( !h->param.b_sliced_threads && h->param.i_threads > 1 )
2331     {
2332         if( x264_pthread_create( &h->thread_handle, NULL, (void*)x264_slices_write, h ) )
2333             return -1;
2334         h->b_thread_active = 1;
2335     }
2336     else if( h->param.b_sliced_threads )
2337     {
2338         if( x264_threaded_slices_write( h ) )
2339             return -1;
2340     }
2341     else
2342         if( (intptr_t)x264_slices_write( h ) )
2343             return -1;
2344
2345     return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
2346 }
2347
2348 static int x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
2349                                    x264_nal_t **pp_nal, int *pi_nal,
2350                                    x264_picture_t *pic_out )
2351 {
2352     int i, j, i_list, frame_size;
2353     char psz_message[80];
2354
2355     if( h->b_thread_active )
2356     {
2357         void *ret = NULL;
2358         x264_pthread_join( h->thread_handle, &ret );
2359         if( (intptr_t)ret )
2360             return (intptr_t)ret;
2361         h->b_thread_active = 0;
2362     }
2363     if( !h->out.i_nal )
2364     {
2365         pic_out->i_type = X264_TYPE_AUTO;
2366         return 0;
2367     }
2368
2369     x264_frame_push_unused( thread_current, h->fenc );
2370
2371     /* End bitstream, set output  */
2372     *pi_nal = h->out.i_nal;
2373     *pp_nal = h->out.nal;
2374
2375     frame_size = x264_encoder_encapsulate_nals( h );
2376
2377     h->out.i_nal = 0;
2378
2379     /* Set output picture properties */
2380     if( h->sh.i_type == SLICE_TYPE_I )
2381         pic_out->i_type = h->i_nal_type == NAL_SLICE_IDR ? X264_TYPE_IDR : X264_TYPE_I;
2382     else if( h->sh.i_type == SLICE_TYPE_P )
2383         pic_out->i_type = X264_TYPE_P;
2384     else
2385         pic_out->i_type = X264_TYPE_B;
2386
2387     pic_out->b_keyframe = h->fenc->b_keyframe;
2388     pic_out->i_pts = h->fenc->i_pts;
2389     pic_out->i_dts = h->fenc->i_dts - h->frames.i_bframe_delay_time;
2390     pic_out->img.i_plane = h->fdec->i_plane;
2391     for(i = 0; i < 3; i++)
2392     {
2393         pic_out->img.i_stride[i] = h->fdec->i_stride[i];
2394         pic_out->img.plane[i] = h->fdec->plane[i];
2395     }
2396
2397     /* ---------------------- Update encoder state ------------------------- */
2398
2399     /* update rc */
2400     x264_emms();
2401     if( x264_ratecontrol_end( h, frame_size * 8 ) < 0 )
2402         return -1;
2403
2404     x264_noise_reduction_update( thread_current );
2405
2406     /* ---------------------- Compute/Print statistics --------------------- */
2407     x264_thread_sync_stat( h, h->thread[0] );
2408
2409     /* Slice stat */
2410     h->stat.i_frame_count[h->sh.i_type]++;
2411     h->stat.i_frame_size[h->sh.i_type] += frame_size;
2412     h->stat.f_frame_qp[h->sh.i_type] += h->fdec->f_qp_avg_aq;
2413
2414     for( i = 0; i < X264_MBTYPE_MAX; i++ )
2415         h->stat.i_mb_count[h->sh.i_type][i] += h->stat.frame.i_mb_count[i];
2416     for( i = 0; i < X264_PARTTYPE_MAX; i++ )
2417         h->stat.i_mb_partition[h->sh.i_type][i] += h->stat.frame.i_mb_partition[i];
2418     for( i = 0; i < 2; i++ )
2419         h->stat.i_mb_count_8x8dct[i] += h->stat.frame.i_mb_count_8x8dct[i];
2420     for( i = 0; i < 6; i++ )
2421         h->stat.i_mb_cbp[i] += h->stat.frame.i_mb_cbp[i];
2422     for( i = 0; i < 3; i++ )
2423         for( j = 0; j < 13; j++ )
2424             h->stat.i_mb_pred_mode[i][j] += h->stat.frame.i_mb_pred_mode[i][j];
2425     if( h->sh.i_type != SLICE_TYPE_I )
2426         for( i_list = 0; i_list < 2; i_list++ )
2427             for( i = 0; i < 32; i++ )
2428                 h->stat.i_mb_count_ref[h->sh.i_type][i_list][i] += h->stat.frame.i_mb_count_ref[i_list][i];
2429     if( h->sh.i_type == SLICE_TYPE_P )
2430     {
2431         h->stat.i_consecutive_bframes[h->fdec->i_frame - h->fref0[0]->i_frame - 1]++;
2432         if( h->param.analyse.i_weighted_pred == X264_WEIGHTP_SMART )
2433         {
2434             for( i = 0; i < 3; i++ )
2435                 for( j = 0; j < h->i_ref0; j++ )
2436                     if( h->sh.weight[0][i].i_denom != 0 )
2437                     {
2438                         h->stat.i_wpred[i]++;
2439                         break;
2440                     }
2441         }
2442     }
2443     if( h->sh.i_type == SLICE_TYPE_B )
2444     {
2445         h->stat.i_direct_frames[ h->sh.b_direct_spatial_mv_pred ] ++;
2446         if( h->mb.b_direct_auto_write )
2447         {
2448             //FIXME somewhat arbitrary time constants
2449             if( h->stat.i_direct_score[0] + h->stat.i_direct_score[1] > h->mb.i_mb_count )
2450             {
2451                 for( i = 0; i < 2; i++ )
2452                     h->stat.i_direct_score[i] = h->stat.i_direct_score[i] * 9/10;
2453             }
2454             for( i = 0; i < 2; i++ )
2455                 h->stat.i_direct_score[i] += h->stat.frame.i_direct_score[i];
2456         }
2457     }
2458
2459     psz_message[0] = '\0';
2460     if( h->param.analyse.b_psnr )
2461     {
2462         int64_t ssd[3] = {
2463             h->stat.frame.i_ssd[0],
2464             h->stat.frame.i_ssd[1],
2465             h->stat.frame.i_ssd[2],
2466         };
2467
2468         h->stat.i_ssd_global[h->sh.i_type] += ssd[0] + ssd[1] + ssd[2];
2469         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 );
2470         h->stat.f_psnr_mean_y[h->sh.i_type] += x264_psnr( ssd[0], h->param.i_width * h->param.i_height );
2471         h->stat.f_psnr_mean_u[h->sh.i_type] += x264_psnr( ssd[1], h->param.i_width * h->param.i_height / 4 );
2472         h->stat.f_psnr_mean_v[h->sh.i_type] += x264_psnr( ssd[2], h->param.i_width * h->param.i_height / 4 );
2473
2474         snprintf( psz_message, 80, " PSNR Y:%5.2f U:%5.2f V:%5.2f",
2475                   x264_psnr( ssd[0], h->param.i_width * h->param.i_height ),
2476                   x264_psnr( ssd[1], h->param.i_width * h->param.i_height / 4),
2477                   x264_psnr( ssd[2], h->param.i_width * h->param.i_height / 4) );
2478     }
2479
2480     if( h->param.analyse.b_ssim )
2481     {
2482         double ssim_y = h->stat.frame.f_ssim
2483                       / (((h->param.i_width-6)>>2) * ((h->param.i_height-6)>>2));
2484         h->stat.f_ssim_mean_y[h->sh.i_type] += ssim_y;
2485         snprintf( psz_message + strlen(psz_message), 80 - strlen(psz_message),
2486                   " SSIM Y:%.5f", ssim_y );
2487     }
2488     psz_message[79] = '\0';
2489
2490     x264_log( h, X264_LOG_DEBUG,
2491                   "frame=%4d QP=%.2f NAL=%d Slice:%c Poc:%-3d I:%-4d P:%-4d SKIP:%-4d size=%d bytes%s\n",
2492               h->i_frame,
2493               h->fdec->f_qp_avg_aq,
2494               h->i_nal_ref_idc,
2495               h->sh.i_type == SLICE_TYPE_I ? 'I' : (h->sh.i_type == SLICE_TYPE_P ? 'P' : 'B' ),
2496               h->fdec->i_poc,
2497               h->stat.frame.i_mb_count_i,
2498               h->stat.frame.i_mb_count_p,
2499               h->stat.frame.i_mb_count_skip,
2500               frame_size,
2501               psz_message );
2502
2503     // keep stats all in one place
2504     x264_thread_sync_stat( h->thread[0], h );
2505     // for the use of the next frame
2506     x264_thread_sync_stat( thread_current, h );
2507
2508 #ifdef DEBUG_MB_TYPE
2509 {
2510     static const char mb_chars[] = { 'i', 'i', 'I', 'C', 'P', '8', 'S',
2511         'D', '<', 'X', 'B', 'X', '>', 'B', 'B', 'B', 'B', '8', 'S' };
2512     int mb_xy;
2513     for( mb_xy = 0; mb_xy < h->sps->i_mb_width * h->sps->i_mb_height; mb_xy++ )
2514     {
2515         if( h->mb.type[mb_xy] < X264_MBTYPE_MAX && h->mb.type[mb_xy] >= 0 )
2516             fprintf( stderr, "%c ", mb_chars[ h->mb.type[mb_xy] ] );
2517         else
2518             fprintf( stderr, "? " );
2519
2520         if( (mb_xy+1) % h->sps->i_mb_width == 0 )
2521             fprintf( stderr, "\n" );
2522     }
2523 }
2524 #endif
2525
2526     /* Remove duplicates, must be done near the end as breaks h->fref0 array
2527      * by freeing some of its pointers. */
2528      for( i = 0; i < h->i_ref0; i++ )
2529          if( h->fref0[i] && h->fref0[i]->b_duplicate )
2530          {
2531              x264_frame_push_blank_unused( h, h->fref0[i] );
2532              h->fref0[i] = 0;
2533          }
2534
2535     if( h->param.psz_dump_yuv )
2536         x264_frame_dump( h );
2537
2538     return frame_size;
2539 }
2540
2541 static void x264_print_intra( int64_t *i_mb_count, double i_count, int b_print_pcm, char *intra )
2542 {
2543     intra += sprintf( intra, "I16..4%s: %4.1f%% %4.1f%% %4.1f%%",
2544         b_print_pcm ? "..PCM" : "",
2545         i_mb_count[I_16x16]/ i_count,
2546         i_mb_count[I_8x8]  / i_count,
2547         i_mb_count[I_4x4]  / i_count );
2548     if( b_print_pcm )
2549         sprintf( intra, " %4.1f%%", i_mb_count[I_PCM]  / i_count );
2550 }
2551
2552 /****************************************************************************
2553  * x264_encoder_close:
2554  ****************************************************************************/
2555 void    x264_encoder_close  ( x264_t *h )
2556 {
2557     int64_t i_yuv_size = 3 * h->param.i_width * h->param.i_height / 2;
2558     int64_t i_mb_count_size[2][7] = {{0}};
2559     char buf[200];
2560     int i, j, i_list, i_type;
2561     int b_print_pcm = h->stat.i_mb_count[SLICE_TYPE_I][I_PCM]
2562                    || h->stat.i_mb_count[SLICE_TYPE_P][I_PCM]
2563                    || h->stat.i_mb_count[SLICE_TYPE_B][I_PCM];
2564
2565     x264_lookahead_delete( h );
2566
2567     for( i = 0; i < h->param.i_threads; i++ )
2568     {
2569         // don't strictly have to wait for the other threads, but it's simpler than canceling them
2570         if( h->thread[i]->b_thread_active )
2571         {
2572             x264_pthread_join( h->thread[i]->thread_handle, NULL );
2573             assert( h->thread[i]->fenc->i_reference_count == 1 );
2574             x264_frame_delete( h->thread[i]->fenc );
2575         }
2576     }
2577
2578     if( h->param.i_threads > 1 && !h->param.b_sliced_threads )
2579     {
2580         x264_t *thread_prev;
2581
2582         thread_prev = h->thread[h->i_thread_phase];
2583         x264_thread_sync_ratecontrol( h, thread_prev, h );
2584         x264_thread_sync_ratecontrol( thread_prev, thread_prev, h );
2585         h->i_frame = thread_prev->i_frame + 1 - h->param.i_threads;
2586     }
2587     h->i_frame++;
2588
2589     /* Slices used and PSNR */
2590     for( i=0; i<5; i++ )
2591     {
2592         static const int slice_order[] = { SLICE_TYPE_I, SLICE_TYPE_SI, SLICE_TYPE_P, SLICE_TYPE_SP, SLICE_TYPE_B };
2593         static const char *slice_name[] = { "P", "B", "I", "SP", "SI" };
2594         int i_slice = slice_order[i];
2595
2596         if( h->stat.i_frame_count[i_slice] > 0 )
2597         {
2598             const int i_count = h->stat.i_frame_count[i_slice];
2599             if( h->param.analyse.b_psnr )
2600             {
2601                 x264_log( h, X264_LOG_INFO,
2602                           "frame %s:%-5d Avg QP:%5.2f  size:%6.0f  PSNR Mean Y:%5.2f U:%5.2f V:%5.2f Avg:%5.2f Global:%5.2f\n",
2603                           slice_name[i_slice],
2604                           i_count,
2605                           h->stat.f_frame_qp[i_slice] / i_count,
2606                           (double)h->stat.i_frame_size[i_slice] / i_count,
2607                           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,
2608                           h->stat.f_psnr_average[i_slice] / i_count,
2609                           x264_psnr( h->stat.i_ssd_global[i_slice], i_count * i_yuv_size ) );
2610             }
2611             else
2612             {
2613                 x264_log( h, X264_LOG_INFO,
2614                           "frame %s:%-5d Avg QP:%5.2f  size:%6.0f\n",
2615                           slice_name[i_slice],
2616                           i_count,
2617                           h->stat.f_frame_qp[i_slice] / i_count,
2618                           (double)h->stat.i_frame_size[i_slice] / i_count );
2619             }
2620         }
2621     }
2622     if( h->param.i_bframe && h->stat.i_frame_count[SLICE_TYPE_P] )
2623     {
2624         char *p = buf;
2625         int den = 0;
2626         // weight by number of frames (including the P-frame) that are in a sequence of N B-frames
2627         for( i=0; i<=h->param.i_bframe; i++ )
2628             den += (i+1) * h->stat.i_consecutive_bframes[i];
2629         for( i=0; i<=h->param.i_bframe; i++ )
2630             p += sprintf( p, " %4.1f%%", 100. * (i+1) * h->stat.i_consecutive_bframes[i] / den );
2631         x264_log( h, X264_LOG_INFO, "consecutive B-frames:%s\n", buf );
2632     }
2633
2634     for( i_type = 0; i_type < 2; i_type++ )
2635         for( i = 0; i < X264_PARTTYPE_MAX; i++ )
2636         {
2637             if( i == D_DIRECT_8x8 ) continue; /* direct is counted as its own type */
2638             i_mb_count_size[i_type][x264_mb_partition_pixel_table[i]] += h->stat.i_mb_partition[i_type][i];
2639         }
2640
2641     /* MB types used */
2642     if( h->stat.i_frame_count[SLICE_TYPE_I] > 0 )
2643     {
2644         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_I];
2645         double i_count = h->stat.i_frame_count[SLICE_TYPE_I] * h->mb.i_mb_count / 100.0;
2646         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
2647         x264_log( h, X264_LOG_INFO, "mb I  %s\n", buf );
2648     }
2649     if( h->stat.i_frame_count[SLICE_TYPE_P] > 0 )
2650     {
2651         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_P];
2652         double i_count = h->stat.i_frame_count[SLICE_TYPE_P] * h->mb.i_mb_count / 100.0;
2653         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_P];
2654         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
2655         x264_log( h, X264_LOG_INFO,
2656                   "mb P  %s  P16..4: %4.1f%% %4.1f%% %4.1f%% %4.1f%% %4.1f%%    skip:%4.1f%%\n",
2657                   buf,
2658                   i_mb_size[PIXEL_16x16] / (i_count*4),
2659                   (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
2660                   i_mb_size[PIXEL_8x8] / (i_count*4),
2661                   (i_mb_size[PIXEL_8x4] + i_mb_size[PIXEL_4x8]) / (i_count*4),
2662                   i_mb_size[PIXEL_4x4] / (i_count*4),
2663                   i_mb_count[P_SKIP] / i_count );
2664     }
2665     if( h->stat.i_frame_count[SLICE_TYPE_B] > 0 )
2666     {
2667         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_B];
2668         double i_count = h->stat.i_frame_count[SLICE_TYPE_B] * h->mb.i_mb_count / 100.0;
2669         double i_mb_list_count;
2670         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_B];
2671         int64_t list_count[3] = {0}; /* 0 == L0, 1 == L1, 2 == BI */
2672         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
2673         for( i = 0; i < X264_PARTTYPE_MAX; i++ )
2674             for( j = 0; j < 2; j++ )
2675             {
2676                 int l0 = x264_mb_type_list_table[i][0][j];
2677                 int l1 = x264_mb_type_list_table[i][1][j];
2678                 if( l0 || l1 )
2679                     list_count[l1+l0*l1] += h->stat.i_mb_count[SLICE_TYPE_B][i] * 2;
2680             }
2681         list_count[0] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L0_8x8];
2682         list_count[1] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L1_8x8];
2683         list_count[2] += h->stat.i_mb_partition[SLICE_TYPE_B][D_BI_8x8];
2684         i_mb_count[B_DIRECT] += (h->stat.i_mb_partition[SLICE_TYPE_B][D_DIRECT_8x8]+2)/4;
2685         i_mb_list_count = (list_count[0] + list_count[1] + list_count[2]) / 100.0;
2686         x264_log( h, X264_LOG_INFO,
2687                   "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",
2688                   buf,
2689                   i_mb_size[PIXEL_16x16] / (i_count*4),
2690                   (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
2691                   i_mb_size[PIXEL_8x8] / (i_count*4),
2692                   i_mb_count[B_DIRECT] / i_count,
2693                   i_mb_count[B_SKIP]   / i_count,
2694                   list_count[0] / i_mb_list_count,
2695                   list_count[1] / i_mb_list_count,
2696                   list_count[2] / i_mb_list_count );
2697     }
2698
2699     x264_ratecontrol_summary( h );
2700
2701     if( h->stat.i_frame_count[SLICE_TYPE_I] + h->stat.i_frame_count[SLICE_TYPE_P] + h->stat.i_frame_count[SLICE_TYPE_B] > 0 )
2702     {
2703 #define SUM3(p) (p[SLICE_TYPE_I] + p[SLICE_TYPE_P] + p[SLICE_TYPE_B])
2704 #define SUM3b(p,o) (p[SLICE_TYPE_I][o] + p[SLICE_TYPE_P][o] + p[SLICE_TYPE_B][o])
2705         int64_t i_i8x8 = SUM3b( h->stat.i_mb_count, I_8x8 );
2706         int64_t i_intra = i_i8x8 + SUM3b( h->stat.i_mb_count, I_4x4 )
2707                                  + SUM3b( h->stat.i_mb_count, I_16x16 );
2708         int64_t i_all_intra = i_intra + SUM3b( h->stat.i_mb_count, I_PCM);
2709         const int i_count = h->stat.i_frame_count[SLICE_TYPE_I] +
2710                             h->stat.i_frame_count[SLICE_TYPE_P] +
2711                             h->stat.i_frame_count[SLICE_TYPE_B];
2712         int64_t i_mb_count = i_count * h->mb.i_mb_count;
2713         float fps = (float) h->param.i_fps_num / h->param.i_fps_den;
2714         float f_bitrate = fps * SUM3(h->stat.i_frame_size) / i_count / 125;
2715
2716         if( h->pps->b_transform_8x8_mode )
2717         {
2718             buf[0] = 0;
2719             if( h->stat.i_mb_count_8x8dct[0] )
2720                 sprintf( buf, " inter:%.1f%%", 100. * h->stat.i_mb_count_8x8dct[1] / h->stat.i_mb_count_8x8dct[0] );
2721             x264_log( h, X264_LOG_INFO, "8x8 transform intra:%.1f%%%s\n", 100. * i_i8x8 / i_intra, buf );
2722         }
2723
2724         if( h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO
2725             && h->stat.i_frame_count[SLICE_TYPE_B] )
2726         {
2727             x264_log( h, X264_LOG_INFO, "direct mvs  spatial:%.1f%% temporal:%.1f%%\n",
2728                       h->stat.i_direct_frames[1] * 100. / h->stat.i_frame_count[SLICE_TYPE_B],
2729                       h->stat.i_direct_frames[0] * 100. / h->stat.i_frame_count[SLICE_TYPE_B] );
2730         }
2731
2732         buf[0] = 0;
2733         if( i_mb_count != i_all_intra )
2734             sprintf( buf, " inter: %.1f%% %.1f%% %.1f%%",
2735                      h->stat.i_mb_cbp[1] * 100.0 / ((i_mb_count - i_all_intra)*4),
2736                      h->stat.i_mb_cbp[3] * 100.0 / ((i_mb_count - i_all_intra)  ),
2737                      h->stat.i_mb_cbp[5] * 100.0 / ((i_mb_count - i_all_intra)) );
2738         x264_log( h, X264_LOG_INFO, "coded y,uvDC,uvAC intra: %.1f%% %.1f%% %.1f%%%s\n",
2739                   h->stat.i_mb_cbp[0] * 100.0 / (i_all_intra*4),
2740                   h->stat.i_mb_cbp[2] * 100.0 / (i_all_intra  ),
2741                   h->stat.i_mb_cbp[4] * 100.0 / (i_all_intra  ), buf );
2742
2743         int64_t fixed_pred_modes[3][9] = {{0}};
2744         int64_t sum_pred_modes[3] = {0};
2745         for( i = 0; i <= I_PRED_16x16_DC_128; i++ )
2746         {
2747             fixed_pred_modes[0][x264_mb_pred_mode16x16_fix[i]] += h->stat.i_mb_pred_mode[0][i];
2748             sum_pred_modes[0] += h->stat.i_mb_pred_mode[0][i];
2749         }
2750         if( sum_pred_modes[0] )
2751             x264_log( h, X264_LOG_INFO, "i16 v,h,dc,p: %2.0f%% %2.0f%% %2.0f%% %2.0f%%\n",
2752                       fixed_pred_modes[0][0] * 100.0 / sum_pred_modes[0],
2753                       fixed_pred_modes[0][1] * 100.0 / sum_pred_modes[0],
2754                       fixed_pred_modes[0][2] * 100.0 / sum_pred_modes[0],
2755                       fixed_pred_modes[0][3] * 100.0 / sum_pred_modes[0] );
2756         for( i = 1; i <= 2; i++ )
2757         {
2758             for( j = 0; j <= I_PRED_8x8_DC_128; j++ )
2759             {
2760                 fixed_pred_modes[i][x264_mb_pred_mode4x4_fix(j)] += h->stat.i_mb_pred_mode[i][j];
2761                 sum_pred_modes[i] += h->stat.i_mb_pred_mode[i][j];
2762             }
2763             if( sum_pred_modes[i] )
2764                 x264_log( h, X264_LOG_INFO, "i%d v,h,dc,ddl,ddr,vr,hd,vl,hu: %2.0f%% %2.0f%% %2.0f%% %2.0f%% %2.0f%% %2.0f%% %2.0f%% %2.0f%% %2.0f%%\n", (3-i)*4,
2765                           fixed_pred_modes[i][0] * 100.0 / sum_pred_modes[i],
2766                           fixed_pred_modes[i][1] * 100.0 / sum_pred_modes[i],
2767                           fixed_pred_modes[i][2] * 100.0 / sum_pred_modes[i],
2768                           fixed_pred_modes[i][3] * 100.0 / sum_pred_modes[i],
2769                           fixed_pred_modes[i][4] * 100.0 / sum_pred_modes[i],
2770                           fixed_pred_modes[i][5] * 100.0 / sum_pred_modes[i],
2771                           fixed_pred_modes[i][6] * 100.0 / sum_pred_modes[i],
2772                           fixed_pred_modes[i][7] * 100.0 / sum_pred_modes[i],
2773                           fixed_pred_modes[i][8] * 100.0 / sum_pred_modes[i] );
2774         }
2775
2776         if( h->param.analyse.i_weighted_pred == X264_WEIGHTP_SMART && h->stat.i_frame_count[SLICE_TYPE_P] > 0 )
2777             x264_log( h, X264_LOG_INFO, "Weighted P-Frames: Y:%.1f%%\n",
2778                       h->stat.i_wpred[0] * 100.0 / h->stat.i_frame_count[SLICE_TYPE_P] );
2779
2780         for( i_list = 0; i_list < 2; i_list++ )
2781         {
2782             int i_slice;
2783             for( i_slice = 0; i_slice < 2; i_slice++ )
2784             {
2785                 char *p = buf;
2786                 int64_t i_den = 0;
2787                 int i_max = 0;
2788                 for( i = 0; i < 32; i++ )
2789                     if( h->stat.i_mb_count_ref[i_slice][i_list][i] )
2790                     {
2791                         i_den += h->stat.i_mb_count_ref[i_slice][i_list][i];
2792                         i_max = i;
2793                     }
2794                 if( i_max == 0 )
2795                     continue;
2796                 for( i = 0; i <= i_max; i++ )
2797                     p += sprintf( p, " %4.1f%%", 100. * h->stat.i_mb_count_ref[i_slice][i_list][i] / i_den );
2798                 x264_log( h, X264_LOG_INFO, "ref %c L%d:%s\n", "PB"[i_slice], i_list, buf );
2799             }
2800         }
2801
2802         if( h->param.analyse.b_ssim )
2803         {
2804             x264_log( h, X264_LOG_INFO,
2805                       "SSIM Mean Y:%.7f\n",
2806                       SUM3( h->stat.f_ssim_mean_y ) / i_count );
2807         }
2808         if( h->param.analyse.b_psnr )
2809         {
2810             x264_log( h, X264_LOG_INFO,
2811                       "PSNR Mean Y:%6.3f U:%6.3f V:%6.3f Avg:%6.3f Global:%6.3f kb/s:%.2f\n",
2812                       SUM3( h->stat.f_psnr_mean_y ) / i_count,
2813                       SUM3( h->stat.f_psnr_mean_u ) / i_count,
2814                       SUM3( h->stat.f_psnr_mean_v ) / i_count,
2815                       SUM3( h->stat.f_psnr_average ) / i_count,
2816                       x264_psnr( SUM3( h->stat.i_ssd_global ), i_count * i_yuv_size ),
2817                       f_bitrate );
2818         }
2819         else
2820             x264_log( h, X264_LOG_INFO, "kb/s:%.2f\n", f_bitrate );
2821     }
2822
2823     /* rc */
2824     x264_ratecontrol_delete( h );
2825
2826     /* param */
2827     if( h->param.rc.psz_stat_out )
2828         free( h->param.rc.psz_stat_out );
2829     if( h->param.rc.psz_stat_in )
2830         free( h->param.rc.psz_stat_in );
2831
2832     x264_cqm_delete( h );
2833     x264_free( h->nal_buffer );
2834     x264_analyse_free_costs( h );
2835
2836     if( h->param.i_threads > 1)
2837         h = h->thread[h->i_thread_phase];
2838
2839     /* frames */
2840     x264_frame_delete_list( h->frames.unused[0] );
2841     x264_frame_delete_list( h->frames.unused[1] );
2842     x264_frame_delete_list( h->frames.current );
2843     x264_frame_delete_list( h->frames.blank_unused );
2844
2845     h = h->thread[0];
2846
2847     for( i = h->param.i_threads - 1; i >= 0; i-- )
2848     {
2849         x264_frame_t **frame;
2850
2851         if( !h->param.b_sliced_threads || i == 0 )
2852         {
2853             for( frame = h->thread[i]->frames.reference; *frame; frame++ )
2854             {
2855                 assert( (*frame)->i_reference_count > 0 );
2856                 (*frame)->i_reference_count--;
2857                 if( (*frame)->i_reference_count == 0 )
2858                     x264_frame_delete( *frame );
2859             }
2860             frame = &h->thread[i]->fdec;
2861             assert( (*frame)->i_reference_count > 0 );
2862             (*frame)->i_reference_count--;
2863             if( (*frame)->i_reference_count == 0 )
2864                 x264_frame_delete( *frame );
2865             x264_macroblock_cache_end( h->thread[i] );
2866         }
2867         x264_free( h->thread[i]->scratch_buffer );
2868         x264_free( h->thread[i]->out.p_bitstream );
2869         x264_free( h->thread[i]->out.nal);
2870         x264_free( h->thread[i] );
2871     }
2872 }
2873
2874 /****************************************************************************
2875  * x264_encoder_delayed_frames:
2876  ****************************************************************************/
2877 int x264_encoder_delayed_frames( x264_t *h )
2878 {
2879     int delayed_frames = 0;
2880     int i;
2881     for( i=0; i<h->param.i_threads; i++ )
2882         delayed_frames += h->thread[i]->b_thread_active;
2883     h = h->thread[h->i_thread_phase];
2884     for( i=0; h->frames.current[i]; i++ )
2885         delayed_frames++;
2886     x264_pthread_mutex_lock( &h->lookahead->ofbuf.mutex );
2887     x264_pthread_mutex_lock( &h->lookahead->ifbuf.mutex );
2888     x264_pthread_mutex_lock( &h->lookahead->next.mutex );
2889     delayed_frames += h->lookahead->ifbuf.i_size + h->lookahead->next.i_size + h->lookahead->ofbuf.i_size;
2890     x264_pthread_mutex_unlock( &h->lookahead->next.mutex );
2891     x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
2892     x264_pthread_mutex_unlock( &h->lookahead->ofbuf.mutex );
2893     return delayed_frames;
2894 }