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     h->i_thread_frames = h->param.b_sliced_threads ? 1 : h->param.i_threads;
 425
 426     if( h->param.b_interlaced )
 427     {
 428         if( h->param.analyse.i_me_method >= X264_ME_ESA )
 429         {
 430             x264_log( h, X264_LOG_WARNING, "interlace + me=esa is not implemented\n" );
 431             h->param.analyse.i_me_method = X264_ME_UMH;
 432         }
 433         if( h->param.analyse.i_direct_mv_pred > X264_DIRECT_PRED_SPATIAL )
 434         {
 435             x264_log( h, X264_LOG_WARNING, "interlace + direct=temporal is not implemented\n" );
 436             h->param.analyse.i_direct_mv_pred = X264_DIRECT_PRED_SPATIAL;
 437         }
 438         if( h->param.analyse.i_weighted_pred > 0 )
 439         {
 440             x264_log( h, X264_LOG_WARNING, "interlace + weightp is not implemented\n" );
 441             h->param.analyse.i_weighted_pred = X264_WEIGHTP_NONE;
 442         }
 443     }
 444
 445     /* Detect default ffmpeg settings and terminate with an error. */
 446     {
 447         int score = 0;
 448         score += h->param.analyse.i_me_range == 0;
 449         score += h->param.rc.i_qp_step == 3;
 450         score += h->param.i_keyint_max == 12;
 451         score += h->param.rc.i_qp_min == 2;
 452         score += h->param.rc.i_qp_max == 31;
 453         score += h->param.rc.f_qcompress == 0.5;
 454         score += fabs(h->param.rc.f_ip_factor - 1.25) < 0.01;
 455         score += fabs(h->param.rc.f_pb_factor - 1.25) < 0.01;
 456         score += h->param.analyse.inter == 0 && h->param.analyse.i_subpel_refine == 8;
 457         if( score >= 5 )
 458         {
 459             x264_log( h, X264_LOG_ERROR, "broken ffmpeg default settings detected\n" );
 460             x264_log( h, X264_LOG_ERROR, "use an encoding preset (vpre)\n" );
 461             return -1;
 462         }
 463     }
 464
 465     if( h->param.rc.i_rc_method < 0 || h->param.rc.i_rc_method > 2 )
 466     {
 467         x264_log( h, X264_LOG_ERROR, "no ratecontrol method specified\n" );
 468         return -1;
 469     }
 470     h->param.rc.f_rf_constant = x264_clip3f( h->param.rc.f_rf_constant, 0, 51 );
 471     h->param.rc.i_qp_constant = x264_clip3( h->param.rc.i_qp_constant, 0, 51 );
 472     if( h->param.rc.i_rc_method == X264_RC_CRF )
 473     {
 474         h->param.rc.i_qp_constant = h->param.rc.f_rf_constant;
 475         h->param.rc.i_bitrate = 0;
 476     }
 477     if( (h->param.rc.i_rc_method == X264_RC_CQP || h->param.rc.i_rc_method == X264_RC_CRF)
 478         && h->param.rc.i_qp_constant == 0 )
 479     {
 480         h->mb.b_lossless = 1;
 481         h->param.i_cqm_preset = X264_CQM_FLAT;
 482         h->param.psz_cqm_file = NULL;
 483         h->param.rc.i_rc_method = X264_RC_CQP;
 484         h->param.rc.f_ip_factor = 1;
 485         h->param.rc.f_pb_factor = 1;
 486         h->param.analyse.b_psnr = 0;
 487         h->param.analyse.b_ssim = 0;
 488         h->param.analyse.i_chroma_qp_offset = 0;
 489         h->param.analyse.i_trellis = 0;
 490         h->param.analyse.b_fast_pskip = 0;
 491         h->param.analyse.i_noise_reduction = 0;
 492         h->param.analyse.f_psy_rd = 0;
 493         h->param.i_bframe = 0;
 494         /* 8x8dct is not useful at all in CAVLC lossless */
 495         if( !h->param.b_cabac )
 496             h->param.analyse.b_transform_8x8 = 0;
 497     }
 498     if( h->param.rc.i_rc_method == X264_RC_CQP )
 499     {
 500         float qp_p = h->param.rc.i_qp_constant;
 501         float qp_i = qp_p - 6*log(h->param.rc.f_ip_factor)/log(2);
 502         float qp_b = qp_p + 6*log(h->param.rc.f_pb_factor)/log(2);
 503         h->param.rc.i_qp_min = x264_clip3( (int)(X264_MIN3( qp_p, qp_i, qp_b )), 0, 51 );
 504         h->param.rc.i_qp_max = x264_clip3( (int)(X264_MAX3( qp_p, qp_i, qp_b ) + .999), 0, 51 );
 505         h->param.rc.i_aq_mode = 0;
 506         h->param.rc.b_mb_tree = 0;
 507     }
 508     h->param.rc.i_qp_max = x264_clip3( h->param.rc.i_qp_max, 0, 51 );
 509     h->param.rc.i_qp_min = x264_clip3( h->param.rc.i_qp_min, 0, h->param.rc.i_qp_max );
 510
 511     int max_slices = (h->param.i_height+((16<<h->param.b_interlaced)-1))/(16<<h->param.b_interlaced);
 512     if( h->param.b_sliced_threads )
 513         h->param.i_slice_count = x264_clip3( h->param.i_threads, 0, max_slices );
 514     else
 515     {
 516         h->param.i_slice_count = x264_clip3( h->param.i_slice_count, 0, max_slices );
 517         h->param.i_slice_max_size = X264_MAX( h->param.i_slice_max_size, 0 );
 518         h->param.i_slice_max_mbs = X264_MAX( h->param.i_slice_max_mbs, 0 );
 519         if( h->param.b_interlaced && h->param.i_slice_max_size )
 520         {
 521             x264_log( h, X264_LOG_WARNING, "interlaced + slice-max-size is not implemented\n" );
 522             h->param.i_slice_max_size = 0;
 523         }
 524         if( h->param.b_interlaced && h->param.i_slice_max_mbs )
 525         {
 526             x264_log( h, X264_LOG_WARNING, "interlaced + slice-max-mbs is not implemented\n" );
 527             h->param.i_slice_max_mbs = 0;
 528         }
 529         if( h->param.i_slice_max_mbs || h->param.i_slice_max_size )
 530             h->param.i_slice_count = 0;
 531     }
 532
 533     h->param.i_frame_reference = x264_clip3( h->param.i_frame_reference, 1, 16 );
 534     if( h->param.i_keyint_max <= 0 )
 535         h->param.i_keyint_max = 1;
 536     if( h->param.i_scenecut_threshold < 0 )
 537         h->param.i_scenecut_threshold = 0;
 538     if( !h->param.analyse.i_subpel_refine && h->param.analyse.i_direct_mv_pred > X264_DIRECT_PRED_SPATIAL )
 539     {
 540         x264_log( h, X264_LOG_WARNING, "subme=0 + direct=temporal is not supported\n" );
 541         h->param.analyse.i_direct_mv_pred = X264_DIRECT_PRED_SPATIAL;
 542     }
 543     h->param.i_bframe = x264_clip3( h->param.i_bframe, 0, X264_BFRAME_MAX );
 544     if( h->param.i_keyint_max == 1 )
 545     {
 546         h->param.i_bframe = 0;
 547         h->param.b_intra_refresh = 0;
 548     }
 549     h->param.i_bframe_bias = x264_clip3( h->param.i_bframe_bias, -90, 100 );
 550     if( h->param.i_bframe <= 1 )
 551         h->param.i_bframe_pyramid = X264_B_PYRAMID_NONE;
 552     h->param.i_bframe_pyramid = x264_clip3( h->param.i_bframe_pyramid, X264_B_PYRAMID_NONE, X264_B_PYRAMID_NORMAL );
 553     if( !h->param.i_bframe )
 554     {
 555         h->param.i_bframe_adaptive = X264_B_ADAPT_NONE;
 556         h->param.analyse.i_direct_mv_pred = 0;
 557         h->param.analyse.b_weighted_bipred = 0;
 558     }
 559     if( h->param.b_intra_refresh && h->param.i_bframe_pyramid == X264_B_PYRAMID_NORMAL )
 560     {
 561         x264_log( h, X264_LOG_WARNING, "b-pyramid normal + intra-refresh is not supported\n" );
 562         h->param.i_bframe_pyramid = X264_B_PYRAMID_STRICT;
 563     }
 564     if( h->param.b_intra_refresh && h->param.i_frame_reference > 1 )
 565     {
 566         x264_log( h, X264_LOG_WARNING, "ref > 1 + intra-refresh is not supported\n" );
 567         h->param.i_frame_reference = 1;
 568     }
 569     if( h->param.b_intra_refresh )
 570         h->param.i_keyint_max = X264_MIN( h->param.i_keyint_max, (h->param.i_width+15)/16 - 1 );
 571     h->param.i_keyint_min = x264_clip3( h->param.i_keyint_min, 1, h->param.i_keyint_max/2+1 );
 572     h->param.rc.i_lookahead = x264_clip3( h->param.rc.i_lookahead, 0, X264_LOOKAHEAD_MAX );
 573     {
 574         int maxrate = X264_MAX( h->param.rc.i_vbv_max_bitrate, h->param.rc.i_bitrate );
 575         float bufsize = maxrate ? (float)h->param.rc.i_vbv_buffer_size / maxrate : 0;
 576         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;
 577         h->param.rc.i_lookahead = X264_MIN( h->param.rc.i_lookahead, X264_MAX( h->param.i_keyint_max, bufsize*fps ) );
 578     }
 579
 580     if( !h->param.i_timebase_num || !h->param.i_timebase_den )
 581     {
 582         h->param.i_timebase_num = h->param.i_fps_den;
 583         h->param.i_timebase_den = h->param.i_fps_num;
 584     }
 585
 586     h->param.rc.f_qcompress = x264_clip3f( h->param.rc.f_qcompress, 0.0, 1.0 );
 587     if( !h->param.rc.i_lookahead || h->param.i_keyint_max == 1 || h->param.rc.f_qcompress == 1 )
 588         h->param.rc.b_mb_tree = 0;
 589     if( h->param.rc.b_stat_read )
 590         h->param.rc.i_lookahead = 0;
 591 #ifdef HAVE_PTHREAD
 592     if( h->param.i_sync_lookahead )
 593         h->param.i_sync_lookahead = x264_clip3( h->param.i_sync_lookahead, h->i_thread_frames + h->param.i_bframe, X264_LOOKAHEAD_MAX );
 594     if( h->param.rc.b_stat_read || h->i_thread_frames == 1 )
 595         h->param.i_sync_lookahead = 0;
 596 #else
 597     h->param.i_sync_lookahead = 0;
 598 #endif
 599
 600     h->mb.b_direct_auto_write = h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO
 601                                 && h->param.i_bframe
 602                                 && ( h->param.rc.b_stat_write || !h->param.rc.b_stat_read );
 603
 604     h->param.i_deblocking_filter_alphac0 = x264_clip3( h->param.i_deblocking_filter_alphac0, -6, 6 );
 605     h->param.i_deblocking_filter_beta    = x264_clip3( h->param.i_deblocking_filter_beta, -6, 6 );
 606     h->param.analyse.i_luma_deadzone[0] = x264_clip3( h->param.analyse.i_luma_deadzone[0], 0, 32 );
 607     h->param.analyse.i_luma_deadzone[1] = x264_clip3( h->param.analyse.i_luma_deadzone[1], 0, 32 );
 608
 609     h->param.i_cabac_init_idc = x264_clip3( h->param.i_cabac_init_idc, 0, 2 );
 610
 611     if( h->param.i_cqm_preset < X264_CQM_FLAT || h->param.i_cqm_preset > X264_CQM_CUSTOM )
 612         h->param.i_cqm_preset = X264_CQM_FLAT;
 613
 614     if( h->param.analyse.i_me_method < X264_ME_DIA ||
 615         h->param.analyse.i_me_method > X264_ME_TESA )
 616         h->param.analyse.i_me_method = X264_ME_HEX;
 617     if( h->param.analyse.i_me_range < 4 )
 618         h->param.analyse.i_me_range = 4;
 619     if( h->param.analyse.i_me_range > 16 && h->param.analyse.i_me_method <= X264_ME_HEX )
 620         h->param.analyse.i_me_range = 16;
 621     if( h->param.analyse.i_me_method == X264_ME_TESA &&
 622         (h->mb.b_lossless || h->param.analyse.i_subpel_refine <= 1) )
 623         h->param.analyse.i_me_method = X264_ME_ESA;
 624     h->param.analyse.i_subpel_refine = x264_clip3( h->param.analyse.i_subpel_refine, 0, 10 );
 625     h->param.analyse.b_mixed_references = h->param.analyse.b_mixed_references && h->param.i_frame_reference > 1;
 626     h->param.analyse.inter &= X264_ANALYSE_PSUB16x16|X264_ANALYSE_PSUB8x8|X264_ANALYSE_BSUB16x16|
 627                               X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
 628     h->param.analyse.intra &= X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
 629     if( !(h->param.analyse.inter & X264_ANALYSE_PSUB16x16) )
 630         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
 631     if( !h->param.analyse.b_transform_8x8 )
 632     {
 633         h->param.analyse.inter &= ~X264_ANALYSE_I8x8;
 634         h->param.analyse.intra &= ~X264_ANALYSE_I8x8;
 635     }
 636     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -12, 12);
 637     if( !h->param.b_cabac )
 638         h->param.analyse.i_trellis = 0;
 639     h->param.analyse.i_trellis = x264_clip3( h->param.analyse.i_trellis, 0, 2 );
 640     if( !h->param.analyse.b_psy )
 641     {
 642         h->param.analyse.f_psy_rd = 0;
 643         h->param.analyse.f_psy_trellis = 0;
 644     }
 645     if( !h->param.analyse.i_trellis )
 646         h->param.analyse.f_psy_trellis = 0;
 647     h->param.analyse.f_psy_rd = x264_clip3f( h->param.analyse.f_psy_rd, 0, 10 );
 648     h->param.analyse.f_psy_trellis = x264_clip3f( h->param.analyse.f_psy_trellis, 0, 10 );
 649     if( h->param.analyse.i_subpel_refine < 6 )
 650         h->param.analyse.f_psy_rd = 0;
 651     h->mb.i_psy_rd = FIX8( h->param.analyse.f_psy_rd );
 652     /* Psy RDO increases overall quantizers to improve the quality of luma--this indirectly hurts chroma quality */
 653     /* so we lower the chroma QP offset to compensate */
 654     /* This can be triggered repeatedly on multiple calls to parameter_validate, but since encoding
 655      * uses the pps chroma qp offset not the param chroma qp offset, this is not a problem. */
 656     if( h->mb.i_psy_rd )
 657         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_rd < 0.25 ? 1 : 2;
 658     h->mb.i_psy_trellis = FIX8( h->param.analyse.f_psy_trellis / 4 );
 659     /* Psy trellis has a similar effect. */
 660     if( h->mb.i_psy_trellis )
 661         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_trellis < 0.25 ? 1 : 2;
 662     else
 663         h->mb.i_psy_trellis = 0;
 664     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -12, 12);
 665     h->param.rc.i_aq_mode = x264_clip3( h->param.rc.i_aq_mode, 0, 2 );
 666     h->param.rc.f_aq_strength = x264_clip3f( h->param.rc.f_aq_strength, 0, 3 );
 667     if( h->param.rc.f_aq_strength == 0 )
 668         h->param.rc.i_aq_mode = 0;
 669     /* MB-tree requires AQ to be on, even if the strength is zero. */
 670     if( !h->param.rc.i_aq_mode && h->param.rc.b_mb_tree )
 671     {
 672         h->param.rc.i_aq_mode = 1;
 673         h->param.rc.f_aq_strength = 0;
 674     }
 675     h->param.analyse.i_noise_reduction = x264_clip3( h->param.analyse.i_noise_reduction, 0, 1<<16 );
 676     if( h->param.analyse.i_subpel_refine == 10 && (h->param.analyse.i_trellis != 2 || !h->param.rc.i_aq_mode) )
 677         h->param.analyse.i_subpel_refine = 9;
 678
 679     {
 680         const x264_level_t *l = x264_levels;
 681         if( h->param.i_level_idc < 0 )
 682         {
 683             int maxrate_bak = h->param.rc.i_vbv_max_bitrate;
 684             if( h->param.rc.i_rc_method == X264_RC_ABR && h->param.rc.i_vbv_buffer_size <= 0 )
 685                 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate * 2;
 686             h->sps = h->sps_array;
 687             x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
 688             do h->param.i_level_idc = l->level_idc;
 689                 while( l[1].level_idc && x264_validate_levels( h, 0 ) && l++ );
 690             h->param.rc.i_vbv_max_bitrate = maxrate_bak;
 691         }
 692         else
 693         {
 694             while( l->level_idc && l->level_idc != h->param.i_level_idc )
 695                 l++;
 696             if( l->level_idc == 0 )
 697             {
 698                 x264_log( h, X264_LOG_ERROR, "invalid level_idc: %d\n", h->param.i_level_idc );
 699                 return -1;
 700             }
 701         }
 702         if( h->param.analyse.i_mv_range <= 0 )
 703             h->param.analyse.i_mv_range = l->mv_range >> h->param.b_interlaced;
 704         else
 705             h->param.analyse.i_mv_range = x264_clip3(h->param.analyse.i_mv_range, 32, 512 >> h->param.b_interlaced);
 706     }
 707
 708     h->param.analyse.i_weighted_pred = x264_clip3( h->param.analyse.i_weighted_pred, 0, X264_WEIGHTP_SMART );
 709     if( !h->param.analyse.i_weighted_pred && h->param.rc.b_mb_tree && h->param.analyse.b_psy && !h->param.b_interlaced )
 710         h->param.analyse.i_weighted_pred = X264_WEIGHTP_FAKE;
 711
 712     if( h->i_thread_frames > 1 )
 713     {
 714         int r = h->param.analyse.i_mv_range_thread;
 715         int r2;
 716         if( r <= 0 )
 717         {
 718             // half of the available space is reserved and divided evenly among the threads,
 719             // the rest is allocated to whichever thread is far enough ahead to use it.
 720             // reserving more space increases quality for some videos, but costs more time
 721             // in thread synchronization.
 722             int max_range = (h->param.i_height + X264_THREAD_HEIGHT) / h->i_thread_frames - X264_THREAD_HEIGHT;
 723             r = max_range / 2;
 724         }
 725         r = X264_MAX( r, h->param.analyse.i_me_range );
 726         r = X264_MIN( r, h->param.analyse.i_mv_range );
 727         // round up to use the whole mb row
 728         r2 = (r & ~15) + ((-X264_THREAD_HEIGHT) & 15);
 729         if( r2 < r )
 730             r2 += 16;
 731         x264_log( h, X264_LOG_DEBUG, "using mv_range_thread = %d\n", r2 );
 732         h->param.analyse.i_mv_range_thread = r2;
 733     }
 734
 735     if( h->param.rc.f_qblur < 0 )
 736         h->param.rc.f_qblur = 0;
 737     if( h->param.rc.f_complexity_blur < 0 )
 738         h->param.rc.f_complexity_blur = 0;
 739
 740     h->param.i_sps_id &= 31;
 741
 742     if( h->param.i_log_level < X264_LOG_INFO )
 743     {
 744         h->param.analyse.b_psnr = 0;
 745         h->param.analyse.b_ssim = 0;
 746     }
 747
 748     /* ensure the booleans are 0 or 1 so they can be used in math */
 749 #define BOOLIFY(x) h->param.x = !!h->param.x
 750     BOOLIFY( b_cabac );
 751     BOOLIFY( b_constrained_intra );
 752     BOOLIFY( b_deblocking_filter );
 753     BOOLIFY( b_deterministic );
 754     BOOLIFY( b_sliced_threads );
 755     BOOLIFY( b_interlaced );
 756     BOOLIFY( b_intra_refresh );
 757     BOOLIFY( b_visualize );
 758     BOOLIFY( b_aud );
 759     BOOLIFY( b_repeat_headers );
 760     BOOLIFY( b_annexb );
 761     BOOLIFY( analyse.b_transform_8x8 );
 762     BOOLIFY( analyse.b_weighted_bipred );
 763     BOOLIFY( analyse.b_chroma_me );
 764     BOOLIFY( analyse.b_mixed_references );
 765     BOOLIFY( analyse.b_fast_pskip );
 766     BOOLIFY( analyse.b_dct_decimate );
 767     BOOLIFY( analyse.b_psy );
 768     BOOLIFY( analyse.b_psnr );
 769     BOOLIFY( analyse.b_ssim );
 770     BOOLIFY( rc.b_stat_write );
 771     BOOLIFY( rc.b_stat_read );
 772     BOOLIFY( rc.b_mb_tree );
 773 #undef BOOLIFY
 774
 775     return 0;
 776 }
 777
 778 static void mbcmp_init( x264_t *h )
 779 {
 780     int satd = !h->mb.b_lossless && h->param.analyse.i_subpel_refine > 1;
 781     memcpy( h->pixf.mbcmp, satd ? h->pixf.satd : h->pixf.sad_aligned, sizeof(h->pixf.mbcmp) );
 782     memcpy( h->pixf.mbcmp_unaligned, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.mbcmp_unaligned) );
 783     h->pixf.intra_mbcmp_x3_16x16 = satd ? h->pixf.intra_satd_x3_16x16 : h->pixf.intra_sad_x3_16x16;
 784     h->pixf.intra_mbcmp_x3_8x8c = satd ? h->pixf.intra_satd_x3_8x8c : h->pixf.intra_sad_x3_8x8c;
 785     h->pixf.intra_mbcmp_x3_4x4 = satd ? h->pixf.intra_satd_x3_4x4 : h->pixf.intra_sad_x3_4x4;
 786     satd &= h->param.analyse.i_me_method == X264_ME_TESA;
 787     memcpy( h->pixf.fpelcmp, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.fpelcmp) );
 788     memcpy( h->pixf.fpelcmp_x3, satd ? h->pixf.satd_x3 : h->pixf.sad_x3, sizeof(h->pixf.fpelcmp_x3) );
 789     memcpy( h->pixf.fpelcmp_x4, satd ? h->pixf.satd_x4 : h->pixf.sad_x4, sizeof(h->pixf.fpelcmp_x4) );
 790 }
 791
 792 static void x264_set_aspect_ratio( x264_t *h, x264_param_t *param, int initial )
 793 {
 794     /* VUI */
 795     if( param->vui.i_sar_width > 0 && param->vui.i_sar_height > 0 )
 796     {
 797         int i_w = param->vui.i_sar_width;
 798         int i_h = param->vui.i_sar_height;
 799         int old_w = h->param.vui.i_sar_width;
 800         int old_h = h->param.vui.i_sar_height;
 801
 802         x264_reduce_fraction( &i_w, &i_h );
 803
 804         while( i_w > 65535 || i_h > 65535 )
 805         {
 806             i_w /= 2;
 807             i_h /= 2;
 808         }
 809
 810         x264_reduce_fraction( &i_w, &i_h );
 811
 812         if( i_w != old_w || i_h != old_h || initial )
 813         {
 814             h->param.vui.i_sar_width = 0;
 815             h->param.vui.i_sar_height = 0;
 816             if( i_w == 0 || i_h == 0 )
 817                 x264_log( h, X264_LOG_WARNING, "cannot create valid sample aspect ratio\n" );
 818             else
 819             {
 820                 x264_log( h, initial?X264_LOG_INFO:X264_LOG_DEBUG, "using SAR=%d/%d\n", i_w, i_h );
 821                 h->param.vui.i_sar_width = i_w;
 822                 h->param.vui.i_sar_height = i_h;
 823             }
 824         }
 825     }
 826 }
 827
 828 /****************************************************************************
 829  * x264_encoder_open:
 830  ****************************************************************************/
 831 x264_t *x264_encoder_open( x264_param_t *param )
 832 {
 833     x264_t *h;
 834     char buf[1000], *p;
 835     int i, qp, i_slicetype_length;
 836
 837     CHECKED_MALLOCZERO( h, sizeof(x264_t) );
 838
 839     /* Create a copy of param */
 840     memcpy( &h->param, param, sizeof(x264_param_t) );
 841
 842     if( param->param_free )
 843         param->param_free( param );
 844
 845     if( x264_validate_parameters( h ) < 0 )
 846         goto fail;
 847
 848     if( h->param.psz_cqm_file )
 849         if( x264_cqm_parse_file( h, h->param.psz_cqm_file ) < 0 )
 850             goto fail;
 851
 852     if( h->param.rc.psz_stat_out )
 853         h->param.rc.psz_stat_out = strdup( h->param.rc.psz_stat_out );
 854     if( h->param.rc.psz_stat_in )
 855         h->param.rc.psz_stat_in = strdup( h->param.rc.psz_stat_in );
 856
 857     x264_set_aspect_ratio( h, &h->param, 1 );
 858
 859     x264_reduce_fraction( &h->param.i_fps_num, &h->param.i_fps_den );
 860     x264_reduce_fraction( &h->param.i_timebase_num, &h->param.i_timebase_den );
 861
 862     /* Init x264_t */
 863     h->i_frame = -1;
 864     h->i_frame_num = 0;
 865     h->i_idr_pic_id = 0;
 866     if( h->param.b_dts_compress )
 867     {
 868         /* h->i_dts_compress_multiplier == h->frames.i_bframe_delay + 1 */
 869         h->i_dts_compress_multiplier = h->param.i_bframe ? (h->param.i_bframe_pyramid ? 3 : 2) : 1;
 870         if( h->i_dts_compress_multiplier != 1 )
 871             x264_log( h, X264_LOG_DEBUG, "DTS compresion changed timebase: %d/%d -> %d/%d\n",
 872                       h->param.i_timebase_num, h->param.i_timebase_den,
 873                       h->param.i_timebase_num, h->param.i_timebase_den * h->i_dts_compress_multiplier );
 874         h->param.i_timebase_den *= h->i_dts_compress_multiplier;
 875     }
 876     else
 877         h->i_dts_compress_multiplier = 1;
 878
 879     h->sps = &h->sps_array[0];
 880     x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
 881
 882     h->pps = &h->pps_array[0];
 883     x264_pps_init( h->pps, h->param.i_sps_id, &h->param, h->sps );
 884
 885     x264_validate_levels( h, 1 );
 886
 887     h->chroma_qp_table = i_chroma_qp_table + 12 + h->pps->i_chroma_qp_index_offset;
 888
 889     if( x264_cqm_init( h ) < 0 )
 890         goto fail;
 891
 892     h->mb.i_mb_count = h->sps->i_mb_width * h->sps->i_mb_height;
 893
 894     /* Init frames. */
 895     if( h->param.i_bframe_adaptive == X264_B_ADAPT_TRELLIS )
 896         h->frames.i_delay = X264_MAX(h->param.i_bframe,3)*4;
 897     else
 898         h->frames.i_delay = h->param.i_bframe;
 899     if( h->param.rc.b_mb_tree || h->param.rc.i_vbv_buffer_size )
 900         h->frames.i_delay = X264_MAX( h->frames.i_delay, h->param.rc.i_lookahead );
 901     i_slicetype_length = h->frames.i_delay;
 902     h->frames.i_delay += h->i_thread_frames - 1;
 903     h->frames.i_delay = X264_MIN( h->frames.i_delay, X264_LOOKAHEAD_MAX );
 904     h->frames.i_delay += h->param.i_sync_lookahead;
 905     h->frames.i_bframe_delay = h->param.i_bframe ? (h->param.i_bframe_pyramid ? 2 : 1) : 0;
 906
 907     h->frames.i_max_ref0 = h->param.i_frame_reference;
 908     h->frames.i_max_ref1 = h->sps->vui.i_num_reorder_frames;
 909     h->frames.i_max_dpb  = h->sps->vui.i_max_dec_frame_buffering;
 910     h->frames.b_have_lowres = !h->param.rc.b_stat_read
 911         && ( h->param.rc.i_rc_method == X264_RC_ABR
 912           || h->param.rc.i_rc_method == X264_RC_CRF
 913           || h->param.i_bframe_adaptive
 914           || h->param.i_scenecut_threshold
 915           || h->param.rc.b_mb_tree
 916           || h->param.analyse.i_weighted_pred == X264_WEIGHTP_SMART );
 917     h->frames.b_have_lowres |= h->param.rc.b_stat_read && h->param.rc.i_vbv_buffer_size > 0;
 918     h->frames.b_have_sub8x8_esa = !!(h->param.analyse.inter & X264_ANALYSE_PSUB8x8);
 919
 920     h->frames.i_last_keyframe = - h->param.i_keyint_max;
 921     h->frames.i_input    = 0;
 922
 923     CHECKED_MALLOCZERO( h->frames.unused[0], (h->frames.i_delay + 3) * sizeof(x264_frame_t *) );
 924     /* Allocate room for max refs plus a few extra just in case. */
 925     CHECKED_MALLOCZERO( h->frames.unused[1], (h->i_thread_frames + 20) * sizeof(x264_frame_t *) );
 926     CHECKED_MALLOCZERO( h->frames.current, (h->param.i_sync_lookahead + h->param.i_bframe
 927                         + h->i_thread_frames + 3) * sizeof(x264_frame_t *) );
 928     if( h->param.analyse.i_weighted_pred > 0 )
 929         CHECKED_MALLOCZERO( h->frames.blank_unused, h->i_thread_frames * 4 * sizeof(x264_frame_t *) );
 930     h->i_ref0 = 0;
 931     h->i_ref1 = 0;
 932
 933     x264_rdo_init();
 934
 935     /* init CPU functions */
 936     x264_predict_16x16_init( h->param.cpu, h->predict_16x16 );
 937     x264_predict_8x8c_init( h->param.cpu, h->predict_8x8c );
 938     x264_predict_8x8_init( h->param.cpu, h->predict_8x8, &h->predict_8x8_filter );
 939     x264_predict_4x4_init( h->param.cpu, h->predict_4x4 );
 940     if( !h->param.b_cabac )
 941         x264_init_vlc_tables();
 942     x264_pixel_init( h->param.cpu, &h->pixf );
 943     x264_dct_init( h->param.cpu, &h->dctf );
 944     x264_zigzag_init( h->param.cpu, &h->zigzagf, h->param.b_interlaced );
 945     x264_mc_init( h->param.cpu, &h->mc );
 946     x264_quant_init( h, h->param.cpu, &h->quantf );
 947     x264_deblock_init( h->param.cpu, &h->loopf );
 948     x264_dct_init_weights();
 949
 950     mbcmp_init( h );
 951
 952     p = buf + sprintf( buf, "using cpu capabilities:" );
 953     for( i=0; x264_cpu_names[i].flags; i++ )
 954     {
 955         if( !strcmp(x264_cpu_names[i].name, "SSE2")
 956             && h->param.cpu & (X264_CPU_SSE2_IS_FAST|X264_CPU_SSE2_IS_SLOW) )
 957             continue;
 958         if( !strcmp(x264_cpu_names[i].name, "SSE3")
 959             && (h->param.cpu & X264_CPU_SSSE3 || !(h->param.cpu & X264_CPU_CACHELINE_64)) )
 960             continue;
 961         if( !strcmp(x264_cpu_names[i].name, "SSE4.1")
 962             && (h->param.cpu & X264_CPU_SSE42) )
 963             continue;
 964         if( (h->param.cpu & x264_cpu_names[i].flags) == x264_cpu_names[i].flags
 965             && (!i || x264_cpu_names[i].flags != x264_cpu_names[i-1].flags) )
 966             p += sprintf( p, " %s", x264_cpu_names[i].name );
 967     }
 968     if( !h->param.cpu )
 969         p += sprintf( p, " none!" );
 970     x264_log( h, X264_LOG_INFO, "%s\n", buf );
 971
 972     for( qp = h->param.rc.i_qp_min; qp <= h->param.rc.i_qp_max; qp++ )
 973         if( x264_analyse_init_costs( h, qp ) )
 974             goto fail;
 975     if( x264_analyse_init_costs( h, X264_LOOKAHEAD_QP ) )
 976         goto fail;
 977     if( h->cost_mv[1][2013] != 24 )
 978     {
 979         x264_log( h, X264_LOG_ERROR, "MV cost test failed: x264 has been miscompiled!\n" );
 980         goto fail;
 981     }
 982
 983     h->out.i_nal = 0;
 984     h->out.i_bitstream = X264_MAX( 1000000, h->param.i_width * h->param.i_height * 4
 985         * ( h->param.rc.i_rc_method == X264_RC_ABR ? pow( 0.95, h->param.rc.i_qp_min )
 986           : pow( 0.95, h->param.rc.i_qp_constant ) * X264_MAX( 1, h->param.rc.f_ip_factor )));
 987
 988     CHECKED_MALLOC( h->nal_buffer, h->out.i_bitstream * 3/2 + 4 );
 989     h->nal_buffer_size = h->out.i_bitstream * 3/2 + 4;
 990
 991     h->thread[0] = h;
 992     for( i = 1; i < h->param.i_threads + !!h->param.i_sync_lookahead; i++ )
 993         CHECKED_MALLOC( h->thread[i], sizeof(x264_t) );
 994
 995     if( x264_lookahead_init( h, i_slicetype_length ) )
 996         goto fail;
 997
 998     for( i = 0; i < h->param.i_threads; i++ )
 999     {
1000         int init_nal_count = h->param.i_slice_count + 3;
1001         int allocate_threadlocal_data = !h->param.b_sliced_threads || !i;
1002         if( i > 0 )
1003             *h->thread[i] = *h;
1004
1005         if( allocate_threadlocal_data )
1006         {
1007             h->thread[i]->fdec = x264_frame_pop_unused( h, 1 );
1008             if( !h->thread[i]->fdec )
1009                 goto fail;
1010         }
1011         else
1012             h->thread[i]->fdec = h->thread[0]->fdec;
1013
1014         CHECKED_MALLOC( h->thread[i]->out.p_bitstream, h->out.i_bitstream );
1015         /* Start each thread with room for init_nal_count NAL units; it'll realloc later if needed. */
1016         CHECKED_MALLOC( h->thread[i]->out.nal, init_nal_count*sizeof(x264_nal_t) );
1017         h->thread[i]->out.i_nals_allocated = init_nal_count;
1018
1019         if( allocate_threadlocal_data && x264_macroblock_cache_init( h->thread[i] ) < 0 )
1020             goto fail;
1021     }
1022
1023     /* Allocate scratch buffer */
1024     for( i = 0; i < h->param.i_threads + !!h->param.i_sync_lookahead; i++ )
1025     {
1026         int buf_hpel = (h->param.i_width+48) * sizeof(int16_t);
1027         int buf_ssim = h->param.analyse.b_ssim * 8 * (h->param.i_width/4+3) * sizeof(int);
1028         int me_range = X264_MIN(h->param.analyse.i_me_range, h->param.analyse.i_mv_range);
1029         int buf_tesa = (h->param.analyse.i_me_method >= X264_ME_ESA) *
1030             ((me_range*2+18) * sizeof(int16_t) + (me_range+4) * (me_range+1) * 4 * sizeof(mvsad_t));
1031         int buf_mbtree = h->param.rc.b_mb_tree * ((h->sps->i_mb_width+3)&~3) * sizeof(int);
1032         int buf_nnz = !h->param.b_cabac * h->pps->b_transform_8x8_mode * (h->sps->i_mb_width * 4 * 16 * sizeof(uint8_t));
1033         int scratch_size = X264_MAX4( buf_hpel, buf_ssim, buf_tesa, X264_MAX( buf_mbtree, buf_nnz ) );
1034         CHECKED_MALLOC( h->thread[i]->scratch_buffer, scratch_size );
1035     }
1036
1037     if( x264_ratecontrol_new( h ) < 0 )
1038         goto fail;
1039
1040     if( h->param.psz_dump_yuv )
1041     {
1042         /* create or truncate the reconstructed video file */
1043         FILE *f = fopen( h->param.psz_dump_yuv, "w" );
1044         if( !f )
1045         {
1046             x264_log( h, X264_LOG_ERROR, "dump_yuv: can't write to %s\n", h->param.psz_dump_yuv );
1047             goto fail;
1048         }
1049         else if( !x264_is_regular_file( f ) )
1050         {
1051             x264_log( h, X264_LOG_ERROR, "dump_yuv: incompatible with non-regular file %s\n", h->param.psz_dump_yuv );
1052             goto fail;
1053         }
1054         fclose( f );
1055     }
1056
1057     x264_log( h, X264_LOG_INFO, "profile %s, level %d.%d\n",
1058         h->sps->i_profile_idc == PROFILE_BASELINE ? "Baseline" :
1059         h->sps->i_profile_idc == PROFILE_MAIN ? "Main" :
1060         h->sps->i_profile_idc == PROFILE_HIGH ? "High" :
1061         "High 4:4:4 Predictive", h->sps->i_level_idc/10, h->sps->i_level_idc%10 );
1062
1063     return h;
1064 fail:
1065     x264_free( h );
1066     return NULL;
1067 }
1068
1069 /****************************************************************************
1070  * x264_encoder_reconfig:
1071  ****************************************************************************/
1072 int x264_encoder_reconfig( x264_t *h, x264_param_t *param )
1073 {
1074     h = h->thread[h->i_thread_phase];
1075     x264_set_aspect_ratio( h, param, 0 );
1076 #define COPY(var) h->param.var = param->var
1077     COPY( i_frame_reference ); // but never uses more refs than initially specified
1078     COPY( i_bframe_bias );
1079     if( h->param.i_scenecut_threshold )
1080         COPY( i_scenecut_threshold ); // can't turn it on or off, only vary the threshold
1081     COPY( b_deblocking_filter );
1082     COPY( i_deblocking_filter_alphac0 );
1083     COPY( i_deblocking_filter_beta );
1084     COPY( analyse.inter );
1085     COPY( analyse.intra );
1086     COPY( analyse.i_direct_mv_pred );
1087     /* Scratch buffer prevents me_range from being increased for esa/tesa */
1088     if( h->param.analyse.i_me_method < X264_ME_ESA || param->analyse.i_me_range < h->param.analyse.i_me_range )
1089         COPY( analyse.i_me_range );
1090     COPY( analyse.i_noise_reduction );
1091     /* We can't switch out of subme=0 during encoding. */
1092     if( h->param.analyse.i_subpel_refine )
1093         COPY( analyse.i_subpel_refine );
1094     COPY( analyse.i_trellis );
1095     COPY( analyse.b_chroma_me );
1096     COPY( analyse.b_dct_decimate );
1097     COPY( analyse.b_fast_pskip );
1098     COPY( analyse.b_mixed_references );
1099     COPY( analyse.f_psy_rd );
1100     COPY( analyse.f_psy_trellis );
1101     // can only twiddle these if they were enabled to begin with:
1102     if( h->param.analyse.i_me_method >= X264_ME_ESA || param->analyse.i_me_method < X264_ME_ESA )
1103         COPY( analyse.i_me_method );
1104     if( h->param.analyse.i_me_method >= X264_ME_ESA && !h->frames.b_have_sub8x8_esa )
1105         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
1106     if( h->pps->b_transform_8x8_mode )
1107         COPY( analyse.b_transform_8x8 );
1108     if( h->frames.i_max_ref1 > 1 )
1109         COPY( i_bframe_pyramid );
1110     COPY( i_slice_max_size );
1111     COPY( i_slice_max_mbs );
1112     COPY( i_slice_count );
1113 #undef COPY
1114
1115     mbcmp_init( h );
1116
1117     return x264_validate_parameters( h );
1118 }
1119
1120 /****************************************************************************
1121  * x264_encoder_parameters:
1122  ****************************************************************************/
1123 void x264_encoder_parameters( x264_t *h, x264_param_t *param )
1124 {
1125     memcpy( param, &h->thread[h->i_thread_phase]->param, sizeof(x264_param_t) );
1126 }
1127
1128 /* internal usage */
1129 static void x264_nal_start( x264_t *h, int i_type, int i_ref_idc )
1130 {
1131     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1132
1133     nal->i_ref_idc = i_ref_idc;
1134     nal->i_type    = i_type;
1135
1136     nal->i_payload= 0;
1137     nal->p_payload= &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8];
1138 }
1139 /* if number of allocated nals is not enough, re-allocate a larger one. */
1140 static int x264_nal_check_buffer( x264_t *h )
1141 {
1142     if( h->out.i_nal >= h->out.i_nals_allocated )
1143     {
1144         x264_nal_t *new_out = x264_malloc( sizeof(x264_nal_t) * (h->out.i_nals_allocated*2) );
1145         if( !new_out )
1146             return -1;
1147         memcpy( new_out, h->out.nal, sizeof(x264_nal_t) * (h->out.i_nals_allocated) );
1148         x264_free( h->out.nal );
1149         h->out.nal = new_out;
1150         h->out.i_nals_allocated *= 2;
1151     }
1152     return 0;
1153 }
1154 static int x264_nal_end( x264_t *h )
1155 {
1156     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1157     nal->i_payload = &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8] - nal->p_payload;
1158     h->out.i_nal++;
1159
1160     return x264_nal_check_buffer( h );
1161 }
1162
1163 static int x264_encoder_encapsulate_nals( x264_t *h )
1164 {
1165     int nal_size = 0, i;
1166     for( i = 0; i < h->out.i_nal; i++ )
1167         nal_size += h->out.nal[i].i_payload;
1168
1169     /* Worst-case NAL unit escaping: reallocate the buffer if it's too small. */
1170     if( h->nal_buffer_size < nal_size * 3/2 + h->out.i_nal * 4 )
1171     {
1172         uint8_t *buf = x264_malloc( nal_size * 2 + h->out.i_nal * 4 );
1173         if( !buf )
1174             return -1;
1175         x264_free( h->nal_buffer );
1176         h->nal_buffer = buf;
1177     }
1178
1179     uint8_t *nal_buffer = h->nal_buffer;
1180
1181     for( i = 0; i < h->out.i_nal; i++ )
1182     {
1183         int size = x264_nal_encode( nal_buffer, h->param.b_annexb, &h->out.nal[i] );
1184         h->out.nal[i].i_payload = size;
1185         h->out.nal[i].p_payload = nal_buffer;
1186         nal_buffer += size;
1187     }
1188
1189     return nal_buffer - h->nal_buffer;
1190 }
1191
1192 /****************************************************************************
1193  * x264_encoder_headers:
1194  ****************************************************************************/
1195 int x264_encoder_headers( x264_t *h, x264_nal_t **pp_nal, int *pi_nal )
1196 {
1197     int frame_size = 0;
1198     /* init bitstream context */
1199     h->out.i_nal = 0;
1200     bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
1201
1202     /* Write SEI, SPS and PPS. */
1203     x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
1204     if( x264_sei_version_write( h, &h->out.bs ) )
1205         return -1;
1206     if( x264_nal_end( h ) )
1207         return -1;
1208
1209     /* generate sequence parameters */
1210     x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
1211     x264_sps_write( &h->out.bs, h->sps );
1212     if( x264_nal_end( h ) )
1213         return -1;
1214
1215     /* generate picture parameters */
1216     x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
1217     x264_pps_write( &h->out.bs, h->pps );
1218     if( x264_nal_end( h ) )
1219         return -1;
1220
1221     frame_size = x264_encoder_encapsulate_nals( h );
1222
1223     /* now set output*/
1224     *pi_nal = h->out.i_nal;
1225     *pp_nal = &h->out.nal[0];
1226     h->out.i_nal = 0;
1227
1228     return frame_size;
1229 }
1230
1231 /* Check to see whether we have chosen a reference list ordering different
1232  * from the standard's default. */
1233 static inline void x264_reference_check_reorder( x264_t *h )
1234 {
1235     int i;
1236     for( i = 0; i < h->i_ref0 - 1; i++ )
1237         /* P and B-frames use different default orders. */
1238         if( h->sh.i_type == SLICE_TYPE_P ? h->fref0[i]->i_frame_num < h->fref0[i+1]->i_frame_num
1239                                          : h->fref0[i]->i_poc < h->fref0[i+1]->i_poc )
1240         {
1241             h->b_ref_reorder[0] = 1;
1242             break;
1243         }
1244 }
1245
1246 /* return -1 on failure, else return the index of the new reference frame */
1247 int x264_weighted_reference_duplicate( x264_t *h, int i_ref, const x264_weight_t *w )
1248 {
1249     int i = h->i_ref0;
1250     int j;
1251     x264_frame_t *newframe;
1252     if( i <= 1 ) /* empty list, definitely can't duplicate frame */
1253         return -1;
1254
1255     /* Find a place to insert the duplicate in the reference list. */
1256     for( j = 0; j < i; j++ )
1257         if( h->fref0[i_ref]->i_frame != h->fref0[j]->i_frame )
1258         {
1259             /* found a place, after j, make sure there is not already a duplicate there */
1260             if( j == i-1 || ( h->fref0[j+1] && h->fref0[i_ref]->i_frame != h->fref0[j+1]->i_frame ) )
1261                 break;
1262         }
1263
1264     if( j == i ) /* No room in the reference list for the duplicate. */
1265         return -1;
1266     j++;
1267
1268     newframe = x264_frame_pop_blank_unused( h );
1269
1270     //FIXME: probably don't need to copy everything
1271     *newframe = *h->fref0[i_ref];
1272     newframe->i_reference_count = 1;
1273     newframe->orig = h->fref0[i_ref];
1274     newframe->b_duplicate = 1;
1275     memcpy( h->fenc->weight[j], w, sizeof(h->fenc->weight[i]) );
1276
1277     /* shift the frames to make space for the dupe. */
1278     h->b_ref_reorder[0] = 1;
1279     if( h->i_ref0 < 16 )
1280         ++h->i_ref0;
1281     h->fref0[15] = NULL;
1282     x264_frame_unshift( &h->fref0[j], newframe );
1283
1284     return j;
1285 }
1286
1287 static void x264_weighted_pred_init( x264_t *h )
1288 {
1289     int i_ref;
1290     int i;
1291
1292     /* for now no analysis and set all weights to nothing */
1293     for( i_ref = 0; i_ref < h->i_ref0; i_ref++ )
1294         h->fenc->weighted[i_ref] = h->fref0[i_ref]->filtered[0];
1295
1296     // FIXME: This only supports weighting of one reference frame
1297     // and duplicates of that frame.
1298     h->fenc->i_lines_weighted = 0;
1299
1300     for( i_ref = 0; i_ref < (h->i_ref0 << h->sh.b_mbaff); i_ref++ )
1301         for( i = 0; i < 3; i++ )
1302             h->sh.weight[i_ref][i].weightfn = NULL;
1303
1304
1305     if( h->sh.i_type != SLICE_TYPE_P || h->param.analyse.i_weighted_pred <= 0 )
1306         return;
1307
1308     int i_padv = PADV << h->param.b_interlaced;
1309     int denom = -1;
1310     int weightluma = 0;
1311     int buffer_next = 0;
1312     int j;
1313     //FIXME: when chroma support is added, move this into loop
1314     h->sh.weight[0][1].weightfn = h->sh.weight[0][2].weightfn = NULL;
1315     h->sh.weight[0][1].i_denom = h->sh.weight[0][2].i_denom = 0;
1316     for( j = 0; j < h->i_ref0; j++ )
1317     {
1318         if( h->fenc->weight[j][0].weightfn )
1319         {
1320             h->sh.weight[j][0] = h->fenc->weight[j][0];
1321             // if weight is useless, don't write it to stream
1322             if( h->sh.weight[j][0].i_scale == 1<<h->sh.weight[j][0].i_denom && h->sh.weight[j][0].i_offset == 0 )
1323                 h->sh.weight[j][0].weightfn = NULL;
1324             else
1325             {
1326                 if( !weightluma )
1327                 {
1328                     weightluma = 1;
1329                     h->sh.weight[0][0].i_denom = denom = h->sh.weight[j][0].i_denom;
1330                     assert( x264_clip3( denom, 0, 7 ) == denom );
1331                 }
1332                 assert( h->sh.weight[j][0].i_denom == denom );
1333                 assert( x264_clip3( h->sh.weight[j][0].i_scale, 0, 127 ) == h->sh.weight[j][0].i_scale );
1334                 assert( x264_clip3( h->sh.weight[j][0].i_offset, -128, 127 ) == h->sh.weight[j][0].i_offset );
1335                 h->fenc->weighted[j] = h->mb.p_weight_buf[buffer_next++] +
1336                     h->fenc->i_stride[0] * i_padv + PADH;
1337             }
1338         }
1339
1340         //scale full resolution frame
1341         if( h->sh.weight[j][0].weightfn && h->param.i_threads == 1 )
1342         {
1343             uint8_t *src = h->fref0[j]->filtered[0] - h->fref0[j]->i_stride[0]*i_padv - PADH;
1344             uint8_t *dst = h->fenc->weighted[j] - h->fenc->i_stride[0]*i_padv - PADH;
1345             int stride = h->fenc->i_stride[0];
1346             int width = h->fenc->i_width[0] + PADH*2;
1347             int height = h->fenc->i_lines[0] + i_padv*2;
1348             x264_weight_scale_plane( h, dst, stride, src, stride, width, height, &h->sh.weight[j][0] );
1349             h->fenc->i_lines_weighted = height;
1350         }
1351     }
1352     if( !weightluma )
1353         h->sh.weight[0][0].i_denom = 0;
1354 }
1355
1356 static inline void x264_reference_build_list( x264_t *h, int i_poc )
1357 {
1358     int i, b_ok;
1359
1360     /* build ref list 0/1 */
1361     h->mb.pic.i_fref[0] = h->i_ref0 = 0;
1362     h->mb.pic.i_fref[1] = h->i_ref1 = 0;
1363     if( h->sh.i_type == SLICE_TYPE_I )
1364         return;
1365
1366     for( i = 0; h->frames.reference[i]; i++ )
1367     {
1368         if( h->frames.reference[i]->i_poc < i_poc )
1369         {
1370             h->fref0[h->i_ref0++] = h->frames.reference[i];
1371         }
1372         else if( h->frames.reference[i]->i_poc > i_poc )
1373         {
1374             h->fref1[h->i_ref1++] = h->frames.reference[i];
1375         }
1376     }
1377
1378     /* Order ref0 from higher to lower poc */
1379     do
1380     {
1381         b_ok = 1;
1382         for( i = 0; i < h->i_ref0 - 1; i++ )
1383         {
1384             if( h->fref0[i]->i_poc < h->fref0[i+1]->i_poc )
1385             {
1386                 XCHG( x264_frame_t*, h->fref0[i], h->fref0[i+1] );
1387                 b_ok = 0;
1388                 break;
1389             }
1390         }
1391     } while( !b_ok );
1392
1393     if( h->sh.i_mmco_remove_from_end )
1394         for( i = h->i_ref0-1; i >= h->i_ref0 - h->sh.i_mmco_remove_from_end; i-- )
1395         {
1396             int diff = h->i_frame_num - h->fref0[i]->i_frame_num;
1397             h->sh.mmco[h->sh.i_mmco_command_count].i_poc = h->fref0[i]->i_poc;
1398             h->sh.mmco[h->sh.i_mmco_command_count++].i_difference_of_pic_nums = diff;
1399         }
1400
1401     /* Order ref1 from lower to higher poc (bubble sort) for B-frame */
1402     do
1403     {
1404         b_ok = 1;
1405         for( i = 0; i < h->i_ref1 - 1; i++ )
1406         {
1407             if( h->fref1[i]->i_poc > h->fref1[i+1]->i_poc )
1408             {
1409                 XCHG( x264_frame_t*, h->fref1[i], h->fref1[i+1] );
1410                 b_ok = 0;
1411                 break;
1412             }
1413         }
1414     } while( !b_ok );
1415
1416     x264_reference_check_reorder( h );
1417
1418     h->i_ref1 = X264_MIN( h->i_ref1, h->frames.i_max_ref1 );
1419     h->i_ref0 = X264_MIN( h->i_ref0, h->frames.i_max_ref0 );
1420     h->i_ref0 = X264_MIN( h->i_ref0, h->param.i_frame_reference ); // if reconfig() has lowered the limit
1421
1422     /* add duplicates */
1423     if( h->fenc->i_type == X264_TYPE_P )
1424     {
1425         int idx = -1;
1426         if( h->param.analyse.i_weighted_pred == X264_WEIGHTP_SMART )
1427         {
1428             x264_weight_t w[3];
1429             w[1].weightfn = w[2].weightfn = NULL;
1430             if( h->param.rc.b_stat_read )
1431                 x264_ratecontrol_set_weights( h, h->fenc );
1432
1433             if( !h->fenc->weight[0][0].weightfn )
1434             {
1435                 h->fenc->weight[0][0].i_denom = 0;
1436                 SET_WEIGHT( w[0], 1, 1, 0, -1 );
1437                 idx = x264_weighted_reference_duplicate( h, 0, w );
1438             }
1439             else
1440             {
1441                 if( h->fenc->weight[0][0].i_scale == 1<<h->fenc->weight[0][0].i_denom )
1442                 {
1443                     SET_WEIGHT( h->fenc->weight[0][0], 1, 1, 0, h->fenc->weight[0][0].i_offset );
1444                 }
1445                 x264_weighted_reference_duplicate( h, 0, weight_none );
1446                 if( h->fenc->weight[0][0].i_offset > -128 )
1447                 {
1448                     w[0] = h->fenc->weight[0][0];
1449                     w[0].i_offset--;
1450                     h->mc.weight_cache( h, &w[0] );
1451                     idx = x264_weighted_reference_duplicate( h, 0, w );
1452                 }
1453             }
1454         }
1455         else if( h->param.analyse.i_weighted_pred == X264_WEIGHTP_BLIND )
1456         {
1457             //weighted offset=-1
1458             x264_weight_t w[3];
1459             SET_WEIGHT( w[0], 1, 1, 0, -1 );
1460             h->fenc->weight[0][0].i_denom = 0;
1461             w[1].weightfn = w[2].weightfn = NULL;
1462             idx = x264_weighted_reference_duplicate( h, 0, w );
1463         }
1464         h->mb.ref_blind_dupe = idx;
1465     }
1466
1467     assert( h->i_ref0 + h->i_ref1 <= 16 );
1468     h->mb.pic.i_fref[0] = h->i_ref0;
1469     h->mb.pic.i_fref[1] = h->i_ref1;
1470 }
1471
1472 static void x264_fdec_filter_row( x264_t *h, int mb_y )
1473 {
1474     /* mb_y is the mb to be encoded next, not the mb to be filtered here */
1475     int b_hpel = h->fdec->b_kept_as_ref;
1476     int b_deblock = !h->sh.i_disable_deblocking_filter_idc;
1477     int b_end = mb_y == h->sps->i_mb_height;
1478     int min_y = mb_y - (1 << h->sh.b_mbaff);
1479     int max_y = b_end ? h->sps->i_mb_height : mb_y;
1480     b_deblock &= b_hpel || h->param.psz_dump_yuv;
1481     if( mb_y & h->sh.b_mbaff )
1482         return;
1483     if( min_y < 0 )
1484         return;
1485
1486     if( !b_end && !h->param.b_sliced_threads )
1487     {
1488         int i, j;
1489         for( j=0; j<=h->sh.b_mbaff; j++ )
1490             for( i=0; i<3; i++ )
1491             {
1492                 memcpy( h->mb.intra_border_backup[j][i],
1493                         h->fdec->plane[i] + ((mb_y*16 >> !!i) + j - 1 - h->sh.b_mbaff) * h->fdec->i_stride[i],
1494                         h->sps->i_mb_width*16 >> !!i );
1495             }
1496     }
1497
1498     if( b_deblock )
1499     {
1500         int y;
1501         for( y = min_y; y < max_y; y += (1 << h->sh.b_mbaff) )
1502             x264_frame_deblock_row( h, y );
1503     }
1504
1505     if( b_hpel )
1506     {
1507         x264_frame_expand_border( h, h->fdec, min_y, b_end );
1508         if( h->param.analyse.i_subpel_refine )
1509         {
1510             x264_frame_filter( h, h->fdec, min_y, b_end );
1511             x264_frame_expand_border_filtered( h, h->fdec, min_y, b_end );
1512         }
1513     }
1514
1515     if( h->i_thread_frames > 1 && h->fdec->b_kept_as_ref )
1516         x264_frame_cond_broadcast( h->fdec, mb_y*16 + (b_end ? 10000 : -(X264_THREAD_HEIGHT << h->sh.b_mbaff)) );
1517
1518     min_y = X264_MAX( min_y*16-8, 0 );
1519     max_y = b_end ? h->param.i_height : mb_y*16-8;
1520
1521     if( h->param.analyse.b_psnr )
1522     {
1523         int i;
1524         for( i=0; i<3; i++ )
1525             h->stat.frame.i_ssd[i] +=
1526                 x264_pixel_ssd_wxh( &h->pixf,
1527                     h->fdec->plane[i] + (min_y>>!!i) * h->fdec->i_stride[i], h->fdec->i_stride[i],
1528                     h->fenc->plane[i] + (min_y>>!!i) * h->fenc->i_stride[i], h->fenc->i_stride[i],
1529                     h->param.i_width >> !!i, (max_y-min_y) >> !!i );
1530     }
1531
1532     if( h->param.analyse.b_ssim )
1533     {
1534         x264_emms();
1535         /* offset by 2 pixels to avoid alignment of ssim blocks with dct blocks,
1536          * and overlap by 4 */
1537         min_y += min_y == 0 ? 2 : -6;
1538         h->stat.frame.f_ssim +=
1539             x264_pixel_ssim_wxh( &h->pixf,
1540                 h->fdec->plane[0] + 2+min_y*h->fdec->i_stride[0], h->fdec->i_stride[0],
1541                 h->fenc->plane[0] + 2+min_y*h->fenc->i_stride[0], h->fenc->i_stride[0],
1542                 h->param.i_width-2, max_y-min_y, h->scratch_buffer );
1543     }
1544 }
1545
1546 static inline int x264_reference_update( x264_t *h )
1547 {
1548     int i, j;
1549     if( !h->fdec->b_kept_as_ref )
1550     {
1551         if( h->i_thread_frames > 1 )
1552         {
1553             x264_frame_push_unused( h, h->fdec );
1554             h->fdec = x264_frame_pop_unused( h, 1 );
1555             if( !h->fdec )
1556                 return -1;
1557         }
1558         return 0;
1559     }
1560
1561     /* apply mmco from previous frame. */
1562     for( i = 0; i < h->sh.i_mmco_command_count; i++ )
1563         for( j = 0; h->frames.reference[j]; j++ )
1564             if( h->frames.reference[j]->i_poc == h->sh.mmco[i].i_poc )
1565                 x264_frame_push_unused( h, x264_frame_shift( &h->frames.reference[j] ) );
1566
1567     /* move frame in the buffer */
1568     x264_frame_push( h->frames.reference, h->fdec );
1569     if( h->frames.reference[h->sps->i_num_ref_frames] )
1570         x264_frame_push_unused( h, x264_frame_shift( h->frames.reference ) );
1571     h->fdec = x264_frame_pop_unused( h, 1 );
1572     if( !h->fdec )
1573         return -1;
1574     return 0;
1575 }
1576
1577 static inline void x264_reference_reset( x264_t *h )
1578 {
1579     while( h->frames.reference[0] )
1580         x264_frame_push_unused( h, x264_frame_pop( h->frames.reference ) );
1581     h->fdec->i_poc =
1582     h->fenc->i_poc = 0;
1583 }
1584
1585 static inline void x264_reference_hierarchy_reset( x264_t *h )
1586 {
1587     int i, ref;
1588     int b_hasdelayframe = 0;
1589     if( !h->param.i_bframe_pyramid )
1590         return;
1591
1592     /* look for delay frames -- chain must only contain frames that are disposable */
1593     for( i = 0; h->frames.current[i] && IS_DISPOSABLE( h->frames.current[i]->i_type ); i++ )
1594         b_hasdelayframe |= h->frames.current[i]->i_coded
1595                         != h->frames.current[i]->i_frame + h->sps->vui.i_num_reorder_frames;
1596
1597     if( h->param.i_bframe_pyramid != X264_B_PYRAMID_STRICT && !b_hasdelayframe )
1598         return;
1599
1600     /* Remove last BREF. There will never be old BREFs in the
1601      * dpb during a BREF decode when pyramid == STRICT */
1602     for( ref = 0; h->frames.reference[ref]; ref++ )
1603     {
1604         if( h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT
1605             && h->frames.reference[ref]->i_type == X264_TYPE_BREF )
1606         {
1607             int diff = h->i_frame_num - h->frames.reference[ref]->i_frame_num;
1608             h->sh.mmco[h->sh.i_mmco_command_count].i_difference_of_pic_nums = diff;
1609             h->sh.mmco[h->sh.i_mmco_command_count++].i_poc = h->frames.reference[ref]->i_poc;
1610             x264_frame_push_unused( h, x264_frame_pop( h->frames.reference ) );
1611             h->b_ref_reorder[0] = 1;
1612             break;
1613         }
1614     }
1615
1616     /* Prepare to room in the dpb for the delayed display time of the later b-frame's */
1617     h->sh.i_mmco_remove_from_end = X264_MAX( ref + 2 - h->frames.i_max_dpb, 0 );
1618 }
1619
1620 static inline void x264_slice_init( x264_t *h, int i_nal_type, int i_global_qp )
1621 {
1622     /* ------------------------ Create slice header  ----------------------- */
1623     if( i_nal_type == NAL_SLICE_IDR )
1624     {
1625         x264_slice_header_init( h, &h->sh, h->sps, h->pps, h->i_idr_pic_id, h->i_frame_num, i_global_qp );
1626
1627         /* increment id */
1628         h->i_idr_pic_id = ( h->i_idr_pic_id + 1 ) % 65536;
1629     }
1630     else
1631     {
1632         x264_slice_header_init( h, &h->sh, h->sps, h->pps, -1, h->i_frame_num, i_global_qp );
1633
1634         /* always set the real higher num of ref frame used */
1635         h->sh.b_num_ref_idx_override = 1;
1636         h->sh.i_num_ref_idx_l0_active = h->i_ref0 <= 0 ? 1 : h->i_ref0;
1637         h->sh.i_num_ref_idx_l1_active = h->i_ref1 <= 0 ? 1 : h->i_ref1;
1638     }
1639
1640     h->fdec->i_frame_num = h->sh.i_frame_num;
1641
1642     if( h->sps->i_poc_type == 0 )
1643     {
1644         h->sh.i_poc_lsb = h->fdec->i_poc & ( (1 << h->sps->i_log2_max_poc_lsb) - 1 );
1645         h->sh.i_delta_poc_bottom = 0;
1646     }
1647     else if( h->sps->i_poc_type == 1 )
1648     {
1649         /* FIXME TODO FIXME */
1650     }
1651     else
1652     {
1653         /* Nothing to do ? */
1654     }
1655
1656     x264_macroblock_slice_init( h );
1657 }
1658
1659 static int x264_slice_write( x264_t *h )
1660 {
1661     int i_skip;
1662     int mb_xy, i_mb_x, i_mb_y;
1663     int i, i_list, i_ref, i_skip_bak = 0; /* Shut up GCC. */
1664     bs_t bs_bak;
1665     x264_cabac_t cabac_bak;
1666     uint8_t cabac_prevbyte_bak = 0; /* Shut up GCC. */
1667     /* Assume no more than 3 bytes of NALU escaping. */
1668     int slice_max_size = h->param.i_slice_max_size > 0 ? (h->param.i_slice_max_size-3-NALU_OVERHEAD)*8 : INT_MAX;
1669     int starting_bits = bs_pos(&h->out.bs);
1670     bs_realign( &h->out.bs );
1671
1672     /* Slice */
1673     x264_nal_start( h, h->i_nal_type, h->i_nal_ref_idc );
1674
1675     /* Slice header */
1676     x264_macroblock_thread_init( h );
1677     x264_slice_header_write( &h->out.bs, &h->sh, h->i_nal_ref_idc );
1678     if( h->param.b_cabac )
1679     {
1680         /* alignment needed */
1681         bs_align_1( &h->out.bs );
1682
1683         /* init cabac */
1684         x264_cabac_context_init( &h->cabac, h->sh.i_type, h->sh.i_qp, h->sh.i_cabac_init_idc );
1685         x264_cabac_encode_init ( &h->cabac, h->out.bs.p, h->out.bs.p_end );
1686     }
1687     h->mb.i_last_qp = h->sh.i_qp;
1688     h->mb.i_last_dqp = 0;
1689
1690     i_mb_y = h->sh.i_first_mb / h->sps->i_mb_width;
1691     i_mb_x = h->sh.i_first_mb % h->sps->i_mb_width;
1692     i_skip = 0;
1693
1694     while( (mb_xy = i_mb_x + i_mb_y * h->sps->i_mb_width) <= h->sh.i_last_mb )
1695     {
1696         int mb_spos = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
1697         if( h->param.i_slice_max_size > 0 )
1698         {
1699             /* We don't need the contexts because flushing the CABAC encoder has no context
1700              * dependency and macroblocks are only re-encoded in the case where a slice is
1701              * ended (and thus the content of all contexts are thrown away). */
1702             if( h->param.b_cabac )
1703             {
1704                 memcpy( &cabac_bak, &h->cabac, offsetof(x264_cabac_t, f8_bits_encoded) );
1705                 /* x264's CABAC writer modifies the previous byte during carry, so it has to be
1706                  * backed up. */
1707                 cabac_prevbyte_bak = h->cabac.p[-1];
1708             }
1709             else
1710             {
1711                 bs_bak = h->out.bs;
1712                 i_skip_bak = i_skip;
1713             }
1714         }
1715
1716         if( i_mb_x == 0 && !h->mb.b_reencode_mb && !h->param.b_sliced_threads )
1717             x264_fdec_filter_row( h, i_mb_y );
1718
1719         /* load cache */
1720         x264_macroblock_cache_load( h, i_mb_x, i_mb_y );
1721
1722         x264_macroblock_analyse( h );
1723
1724         /* encode this macroblock -> be careful it can change the mb type to P_SKIP if needed */
1725         x264_macroblock_encode( h );
1726
1727         if( x264_bitstream_check_buffer( h ) )
1728             return -1;
1729
1730         if( h->param.b_cabac )
1731         {
1732             if( mb_xy > h->sh.i_first_mb && !(h->sh.b_mbaff && (i_mb_y&1)) )
1733                 x264_cabac_encode_terminal( &h->cabac );
1734
1735             if( IS_SKIP( h->mb.i_type ) )
1736                 x264_cabac_mb_skip( h, 1 );
1737             else
1738             {
1739                 if( h->sh.i_type != SLICE_TYPE_I )
1740                     x264_cabac_mb_skip( h, 0 );
1741                 x264_macroblock_write_cabac( h, &h->cabac );
1742             }
1743         }
1744         else
1745         {
1746             if( IS_SKIP( h->mb.i_type ) )
1747                 i_skip++;
1748             else
1749             {
1750                 if( h->sh.i_type != SLICE_TYPE_I )
1751                 {
1752                     bs_write_ue( &h->out.bs, i_skip );  /* skip run */
1753                     i_skip = 0;
1754                 }
1755                 x264_macroblock_write_cavlc( h );
1756             }
1757         }
1758
1759         int total_bits = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
1760         int mb_size = total_bits - mb_spos;
1761
1762         /* We'll just re-encode this last macroblock if we go over the max slice size. */
1763         if( total_bits - starting_bits > slice_max_size && !h->mb.b_reencode_mb )
1764         {
1765             if( mb_xy != h->sh.i_first_mb )
1766             {
1767                 if( h->param.b_cabac )
1768                 {
1769                     memcpy( &h->cabac, &cabac_bak, offsetof(x264_cabac_t, f8_bits_encoded) );
1770                     h->cabac.p[-1] = cabac_prevbyte_bak;
1771                 }
1772                 else
1773                 {
1774                     h->out.bs = bs_bak;
1775                     i_skip = i_skip_bak;
1776                 }
1777                 h->mb.b_reencode_mb = 1;
1778                 h->sh.i_last_mb = mb_xy-1;
1779                 break;
1780             }
1781             else
1782             {
1783                 h->sh.i_last_mb = mb_xy;
1784                 h->mb.b_reencode_mb = 0;
1785             }
1786         }
1787         else
1788             h->mb.b_reencode_mb = 0;
1789
1790 #if VISUALIZE
1791         if( h->param.b_visualize )
1792             x264_visualize_mb( h );
1793 #endif
1794
1795         /* save cache */
1796         x264_macroblock_cache_save( h );
1797
1798         /* accumulate mb stats */
1799         h->stat.frame.i_mb_count[h->mb.i_type]++;
1800
1801         if( !IS_INTRA(h->mb.i_type) && !IS_SKIP(h->mb.i_type) && !IS_DIRECT(h->mb.i_type) )
1802         {
1803             if( h->mb.i_partition != D_8x8 )
1804                     h->stat.frame.i_mb_partition[h->mb.i_partition] += 4;
1805                 else
1806                     for( i = 0; i < 4; i++ )
1807                         h->stat.frame.i_mb_partition[h->mb.i_sub_partition[i]] ++;
1808             if( h->param.i_frame_reference > 1 )
1809                 for( i_list = 0; i_list <= (h->sh.i_type == SLICE_TYPE_B); i_list++ )
1810                     for( i = 0; i < 4; i++ )
1811                     {
1812                         i_ref = h->mb.cache.ref[i_list][ x264_scan8[4*i] ];
1813                         if( i_ref >= 0 )
1814                             h->stat.frame.i_mb_count_ref[i_list][i_ref] ++;
1815                     }
1816         }
1817
1818         if( h->param.i_log_level >= X264_LOG_INFO )
1819         {
1820             if( h->mb.i_cbp_luma || h->mb.i_cbp_chroma )
1821             {
1822                 int cbpsum = (h->mb.i_cbp_luma&1) + ((h->mb.i_cbp_luma>>1)&1)
1823                            + ((h->mb.i_cbp_luma>>2)&1) + (h->mb.i_cbp_luma>>3);
1824                 int b_intra = IS_INTRA(h->mb.i_type);
1825                 h->stat.frame.i_mb_cbp[!b_intra + 0] += cbpsum;
1826                 h->stat.frame.i_mb_cbp[!b_intra + 2] += h->mb.i_cbp_chroma >= 1;
1827                 h->stat.frame.i_mb_cbp[!b_intra + 4] += h->mb.i_cbp_chroma == 2;
1828             }
1829             if( h->mb.i_cbp_luma && !IS_INTRA(h->mb.i_type) )
1830             {
1831                 h->stat.frame.i_mb_count_8x8dct[0] ++;
1832                 h->stat.frame.i_mb_count_8x8dct[1] += h->mb.b_transform_8x8;
1833             }
1834             if( IS_INTRA(h->mb.i_type) && h->mb.i_type != I_PCM )
1835             {
1836                 if( h->mb.i_type == I_16x16 )
1837                     h->stat.frame.i_mb_pred_mode[0][h->mb.i_intra16x16_pred_mode]++;
1838                 else if( h->mb.i_type == I_8x8 )
1839                     for( i = 0; i < 16; i += 4 )
1840                         h->stat.frame.i_mb_pred_mode[1][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
1841                 else //if( h->mb.i_type == I_4x4 )
1842                     for( i = 0; i < 16; i++ )
1843                         h->stat.frame.i_mb_pred_mode[2][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
1844             }
1845         }
1846
1847         x264_ratecontrol_mb( h, mb_size );
1848
1849         if( h->sh.b_mbaff )
1850         {
1851             i_mb_x += i_mb_y & 1;
1852             i_mb_y ^= i_mb_x < h->sps->i_mb_width;
1853         }
1854         else
1855             i_mb_x++;
1856         if( i_mb_x == h->sps->i_mb_width )
1857         {
1858             i_mb_y++;
1859             i_mb_x = 0;
1860         }
1861     }
1862
1863     if( h->param.b_cabac )
1864     {
1865         x264_cabac_encode_flush( h, &h->cabac );
1866         h->out.bs.p = h->cabac.p;
1867     }
1868     else
1869     {
1870         if( i_skip > 0 )
1871             bs_write_ue( &h->out.bs, i_skip );  /* last skip run */
1872         /* rbsp_slice_trailing_bits */
1873         bs_rbsp_trailing( &h->out.bs );
1874         bs_flush( &h->out.bs );
1875     }
1876     if( x264_nal_end( h ) )
1877         return -1;
1878
1879     if( h->sh.i_last_mb == h->mb.i_mb_count-1 )
1880     {
1881         h->stat.frame.i_misc_bits = bs_pos( &h->out.bs )
1882                                   + (h->out.i_nal*NALU_OVERHEAD * 8)
1883                                   - h->stat.frame.i_tex_bits
1884                                   - h->stat.frame.i_mv_bits;
1885         if( !h->param.b_sliced_threads )
1886             x264_fdec_filter_row( h, h->sps->i_mb_height );
1887     }
1888
1889     return 0;
1890 }
1891
1892 static void x264_thread_sync_context( x264_t *dst, x264_t *src )
1893 {
1894     if( dst == src )
1895         return;
1896
1897     // reference counting
1898     x264_frame_t **f;
1899     for( f = src->frames.reference; *f; f++ )
1900         (*f)->i_reference_count++;
1901     for( f = dst->frames.reference; *f; f++ )
1902         x264_frame_push_unused( src, *f );
1903     src->fdec->i_reference_count++;
1904     x264_frame_push_unused( src, dst->fdec );
1905
1906     // copy everything except the per-thread pointers and the constants.
1907     memcpy( &dst->i_frame, &src->i_frame, offsetof(x264_t, mb.type) - offsetof(x264_t, i_frame) );
1908     dst->param = src->param;
1909     dst->stat = src->stat;
1910 }
1911
1912 static void x264_thread_sync_stat( x264_t *dst, x264_t *src )
1913 {
1914     if( dst == src )
1915         return;
1916     memcpy( &dst->stat.i_frame_count, &src->stat.i_frame_count, sizeof(dst->stat) - sizeof(dst->stat.frame) );
1917 }
1918
1919 static void *x264_slices_write( x264_t *h )
1920 {
1921     int i_slice_num = 0;
1922     int last_thread_mb = h->sh.i_last_mb;
1923     if( h->param.i_sync_lookahead )
1924         x264_lower_thread_priority( 10 );
1925
1926 #ifdef HAVE_MMX
1927     /* Misalign mask has to be set separately for each thread. */
1928     if( h->param.cpu&X264_CPU_SSE_MISALIGN )
1929         x264_cpu_mask_misalign_sse();
1930 #endif
1931
1932 #if VISUALIZE
1933     if( h->param.b_visualize )
1934         if( x264_visualize_init( h ) )
1935             return (void *)-1;
1936 #endif
1937
1938     /* init stats */
1939     memset( &h->stat.frame, 0, sizeof(h->stat.frame) );
1940     h->mb.b_reencode_mb = 0;
1941     while( h->sh.i_first_mb <= last_thread_mb )
1942     {
1943         h->sh.i_last_mb = last_thread_mb;
1944         if( h->param.i_slice_max_mbs )
1945             h->sh.i_last_mb = h->sh.i_first_mb + h->param.i_slice_max_mbs - 1;
1946         else if( h->param.i_slice_count && !h->param.b_sliced_threads )
1947         {
1948             int height = h->sps->i_mb_height >> h->param.b_interlaced;
1949             int width = h->sps->i_mb_width << h->param.b_interlaced;
1950             i_slice_num++;
1951             h->sh.i_last_mb = (height * i_slice_num + h->param.i_slice_count/2) / h->param.i_slice_count * width - 1;
1952         }
1953         h->sh.i_last_mb = X264_MIN( h->sh.i_last_mb, last_thread_mb );
1954         if( x264_stack_align( x264_slice_write, h ) )
1955             return (void *)-1;
1956         h->sh.i_first_mb = h->sh.i_last_mb + 1;
1957     }
1958
1959 #if VISUALIZE
1960     if( h->param.b_visualize )
1961     {
1962         x264_visualize_show( h );
1963         x264_visualize_close( h );
1964     }
1965 #endif
1966
1967     return (void *)0;
1968 }
1969
1970 static int x264_threaded_slices_write( x264_t *h )
1971 {
1972     int i, j;
1973     void *ret = NULL;
1974     /* set first/last mb and sync contexts */
1975     for( i = 0; i < h->param.i_threads; i++ )
1976     {
1977         x264_t *t = h->thread[i];
1978         if( i )
1979         {
1980             t->param = h->param;
1981             memcpy( &t->i_frame, &h->i_frame, offsetof(x264_t, rc) - offsetof(x264_t, i_frame) );
1982         }
1983         int height = h->sps->i_mb_height >> h->param.b_interlaced;
1984         t->i_threadslice_start = ((height *  i    + h->param.i_slice_count/2) / h->param.i_threads) << h->param.b_interlaced;
1985         t->i_threadslice_end   = ((height * (i+1) + h->param.i_slice_count/2) / h->param.i_threads) << h->param.b_interlaced;
1986         t->sh.i_first_mb = t->i_threadslice_start * h->sps->i_mb_width;
1987         t->sh.i_last_mb  =   t->i_threadslice_end * h->sps->i_mb_width - 1;
1988     }
1989
1990     x264_analyse_weight_frame( h, h->sps->i_mb_height*16 + 16 );
1991
1992     x264_threads_distribute_ratecontrol( h );
1993
1994     /* dispatch */
1995     for( i = 0; i < h->param.i_threads; i++ )
1996     {
1997         if( x264_pthread_create( &h->thread[i]->thread_handle, NULL, (void*)x264_slices_write, (void*)h->thread[i] ) )
1998             return -1;
1999         h->thread[i]->b_thread_active = 1;
2000     }
2001     for( i = 0; i < h->param.i_threads; i++ )
2002     {
2003         x264_pthread_join( h->thread[i]->thread_handle, &ret );
2004         h->thread[i]->b_thread_active = 0;
2005         if( (intptr_t)ret )
2006             return (intptr_t)ret;
2007     }
2008
2009     /* deblocking and hpel filtering */
2010     for( i = 0; i <= h->sps->i_mb_height; i++ )
2011         x264_fdec_filter_row( h, i );
2012
2013     for( i = 1; i < h->param.i_threads; i++ )
2014     {
2015         x264_t *t = h->thread[i];
2016         for( j = 0; j < t->out.i_nal; j++ )
2017         {
2018             h->out.nal[h->out.i_nal] = t->out.nal[j];
2019             h->out.i_nal++;
2020             x264_nal_check_buffer( h );
2021         }
2022         /* All entries in stat.frame are ints except for ssd/ssim,
2023          * which are only calculated in the main thread. */
2024         for( j = 0; j < (offsetof(x264_t,stat.frame.i_ssd) - offsetof(x264_t,stat.frame.i_mv_bits)) / sizeof(int); j++ )
2025             ((int*)&h->stat.frame)[j] += ((int*)&t->stat.frame)[j];
2026     }
2027
2028     x264_threads_merge_ratecontrol( h );
2029
2030     return 0;
2031 }
2032
2033 /****************************************************************************
2034  * x264_encoder_encode:
2035  *  XXX: i_poc   : is the poc of the current given picture
2036  *       i_frame : is the number of the frame being coded
2037  *  ex:  type frame poc
2038  *       I      0   2*0
2039  *       P      1   2*3
2040  *       B      2   2*1
2041  *       B      3   2*2
2042  *       P      4   2*6
2043  *       B      5   2*4
2044  *       B      6   2*5
2045  ****************************************************************************/
2046 int     x264_encoder_encode( x264_t *h,
2047                              x264_nal_t **pp_nal, int *pi_nal,
2048                              x264_picture_t *pic_in,
2049                              x264_picture_t *pic_out )
2050 {
2051     x264_t *thread_current, *thread_prev, *thread_oldest;
2052     int i_nal_type, i_nal_ref_idc, i_global_qp, i;
2053
2054     if( h->i_thread_frames > 1 )
2055     {
2056         thread_prev    = h->thread[ h->i_thread_phase ];
2057         h->i_thread_phase = (h->i_thread_phase + 1) % h->i_thread_frames;
2058         thread_current = h->thread[ h->i_thread_phase ];
2059         thread_oldest  = h->thread[ (h->i_thread_phase + 1) % h->i_thread_frames ];
2060         x264_thread_sync_context( thread_current, thread_prev );
2061         x264_thread_sync_ratecontrol( thread_current, thread_prev, thread_oldest );
2062         h = thread_current;
2063 //      fprintf(stderr, "current: %p  prev: %p  oldest: %p \n", thread_current, thread_prev, thread_oldest);
2064     }
2065     else
2066     {
2067         thread_current =
2068         thread_oldest  = h;
2069     }
2070
2071     // ok to call this before encoding any frames, since the initial values of fdec have b_kept_as_ref=0
2072     if( x264_reference_update( h ) )
2073         return -1;
2074     h->fdec->i_lines_completed = -1;
2075
2076     /* no data out */
2077     *pi_nal = 0;
2078     *pp_nal = NULL;
2079
2080     /* ------------------- Setup new frame from picture -------------------- */
2081     if( pic_in != NULL )
2082     {
2083         /* 1: Copy the picture to a frame and move it to a buffer */
2084         x264_frame_t *fenc = x264_frame_pop_unused( h, 0 );
2085         if( !fenc )
2086             return -1;
2087
2088         if( x264_frame_copy_picture( h, fenc, pic_in ) < 0 )
2089             return -1;
2090
2091         if( h->param.i_width != 16 * h->sps->i_mb_width ||
2092             h->param.i_height != 16 * h->sps->i_mb_height )
2093             x264_frame_expand_border_mod16( h, fenc );
2094
2095         fenc->i_frame = h->frames.i_input++;
2096
2097         if( h->frames.i_bframe_delay && fenc->i_frame == h->frames.i_bframe_delay )
2098             h->frames.i_bframe_delay_time = fenc->i_pts;
2099
2100         if( h->frames.b_have_lowres )
2101         {
2102             if( h->param.analyse.i_weighted_pred == X264_WEIGHTP_FAKE || h->param.analyse.i_weighted_pred == X264_WEIGHTP_SMART )
2103                 x264_weight_plane_analyse( h, fenc );
2104             x264_frame_init_lowres( h, fenc );
2105         }
2106
2107         if( h->param.rc.b_mb_tree && h->param.rc.b_stat_read )
2108         {
2109             if( x264_macroblock_tree_read( h, fenc ) )
2110                 return -1;
2111         }
2112         else if( h->param.rc.i_aq_mode )
2113             x264_adaptive_quant_frame( h, fenc );
2114
2115         /* 2: Place the frame into the queue for its slice type decision */
2116         x264_lookahead_put_frame( h, fenc );
2117
2118         if( h->frames.i_input <= h->frames.i_delay + 1 - h->i_thread_frames )
2119         {
2120             /* Nothing yet to encode, waiting for filling of buffers */
2121             pic_out->i_type = X264_TYPE_AUTO;
2122             return 0;
2123         }
2124     }
2125     else
2126     {
2127         /* signal kills for lookahead thread */
2128         x264_pthread_mutex_lock( &h->lookahead->ifbuf.mutex );
2129         h->lookahead->b_exit_thread = 1;
2130         x264_pthread_cond_broadcast( &h->lookahead->ifbuf.cv_fill );
2131         x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
2132     }
2133
2134     h->i_frame++;
2135     /* 3: The picture is analyzed in the lookahead */
2136     if( !h->frames.current[0] )
2137         x264_lookahead_get_frames( h );
2138
2139     if( !h->frames.current[0] && x264_lookahead_is_empty( h ) )
2140         return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
2141
2142     /* ------------------- Get frame to be encoded ------------------------- */
2143     /* 4: get picture to encode */
2144     h->fenc = x264_frame_shift( h->frames.current );
2145     if( h->fenc->param )
2146     {
2147         x264_encoder_reconfig( h, h->fenc->param );
2148         if( h->fenc->param->param_free )
2149             h->fenc->param->param_free( h->fenc->param );
2150     }
2151
2152     if( h->fenc->b_keyframe )
2153     {
2154         h->frames.i_last_keyframe = h->fenc->i_frame;
2155         if( h->fenc->i_type == X264_TYPE_IDR )
2156             h->i_frame_num = 0;
2157     }
2158     h->sh.i_mmco_command_count =
2159     h->sh.i_mmco_remove_from_end = 0;
2160     h->b_ref_reorder[0] =
2161     h->b_ref_reorder[1] = 0;
2162
2163     /* ------------------- Setup frame context ----------------------------- */
2164     /* 5: Init data dependent of frame type */
2165     if( h->fenc->i_type == X264_TYPE_IDR )
2166     {
2167         /* reset ref pictures */
2168         i_nal_type    = NAL_SLICE_IDR;
2169         i_nal_ref_idc = NAL_PRIORITY_HIGHEST;
2170         h->sh.i_type = SLICE_TYPE_I;
2171         x264_reference_reset( h );
2172     }
2173     else if( h->fenc->i_type == X264_TYPE_I )
2174     {
2175         i_nal_type    = NAL_SLICE;
2176         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
2177         h->sh.i_type = SLICE_TYPE_I;
2178         x264_reference_hierarchy_reset( h );
2179     }
2180     else if( h->fenc->i_type == X264_TYPE_P )
2181     {
2182         i_nal_type    = NAL_SLICE;
2183         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
2184         h->sh.i_type = SLICE_TYPE_P;
2185         x264_reference_hierarchy_reset( h );
2186     }
2187     else if( h->fenc->i_type == X264_TYPE_BREF )
2188     {
2189         i_nal_type    = NAL_SLICE;
2190         i_nal_ref_idc = h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT ? NAL_PRIORITY_LOW : NAL_PRIORITY_HIGH;
2191         h->sh.i_type = SLICE_TYPE_B;
2192         x264_reference_hierarchy_reset( h );
2193     }
2194     else    /* B frame */
2195     {
2196         i_nal_type    = NAL_SLICE;
2197         i_nal_ref_idc = NAL_PRIORITY_DISPOSABLE;
2198         h->sh.i_type = SLICE_TYPE_B;
2199     }
2200
2201     h->fdec->i_poc =
2202     h->fenc->i_poc = 2 * (h->fenc->i_frame - h->frames.i_last_keyframe);
2203     h->fdec->i_type = h->fenc->i_type;
2204     h->fdec->i_frame = h->fenc->i_frame;
2205     h->fenc->b_kept_as_ref =
2206     h->fdec->b_kept_as_ref = i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE && h->param.i_keyint_max > 1;
2207
2208
2209
2210     /* ------------------- Init                ----------------------------- */
2211     /* build ref list 0/1 */
2212     x264_reference_build_list( h, h->fdec->i_poc );
2213
2214     /* ---------------------- Write the bitstream -------------------------- */
2215     /* Init bitstream context */
2216     if( h->param.b_sliced_threads )
2217     {
2218         for( i = 0; i < h->param.i_threads; i++ )
2219         {
2220             bs_init( &h->thread[i]->out.bs, h->thread[i]->out.p_bitstream, h->thread[i]->out.i_bitstream );
2221             h->thread[i]->out.i_nal = 0;
2222         }
2223     }
2224     else
2225     {
2226         bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
2227         h->out.i_nal = 0;
2228     }
2229
2230     if( h->param.b_aud )
2231     {
2232         int pic_type;
2233
2234         if( h->sh.i_type == SLICE_TYPE_I )
2235             pic_type = 0;
2236         else if( h->sh.i_type == SLICE_TYPE_P )
2237             pic_type = 1;
2238         else if( h->sh.i_type == SLICE_TYPE_B )
2239             pic_type = 2;
2240         else
2241             pic_type = 7;
2242
2243         x264_nal_start( h, NAL_AUD, NAL_PRIORITY_DISPOSABLE );
2244         bs_write( &h->out.bs, 3, pic_type );
2245         bs_rbsp_trailing( &h->out.bs );
2246         if( x264_nal_end( h ) )
2247             return -1;
2248     }
2249
2250     h->i_nal_type = i_nal_type;
2251     h->i_nal_ref_idc = i_nal_ref_idc;
2252
2253     int overhead = NALU_OVERHEAD;
2254
2255     if( h->param.b_intra_refresh && h->fenc->i_type == X264_TYPE_P )
2256     {
2257         int pocdiff = (h->fdec->i_poc - h->fref0[0]->i_poc)/2;
2258         float increment = ((float)h->sps->i_mb_width-1) / h->param.i_keyint_max;
2259         if( IS_X264_TYPE_I( h->fref0[0]->i_type ) )
2260             h->fdec->f_pir_position = 0;
2261         else
2262         {
2263             if( h->fref0[0]->i_pir_end_col == h->sps->i_mb_width - 1 )
2264             {
2265                 h->fdec->f_pir_position = 0;
2266                 h->fenc->b_keyframe = 1;
2267             }
2268             else
2269                 h->fdec->f_pir_position = h->fref0[0]->f_pir_position;
2270         }
2271         h->fdec->i_pir_start_col = h->fdec->f_pir_position+0.5;
2272         h->fdec->f_pir_position += increment * pocdiff;
2273         h->fdec->i_pir_end_col = X264_MIN( h->fdec->f_pir_position+0.5, h->sps->i_mb_width-1 );
2274     }
2275
2276     /* Write SPS and PPS */
2277     if( h->fenc->b_keyframe )
2278     {
2279         if( h->param.b_repeat_headers )
2280         {
2281             if( h->fenc->i_frame == 0 )
2282             {
2283                 /* identify ourself */
2284                 x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2285                 if( x264_sei_version_write( h, &h->out.bs ) )
2286                     return -1;
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             /* generate sequence parameters */
2293             x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
2294             x264_sps_write( &h->out.bs, h->sps );
2295             if( x264_nal_end( h ) )
2296                 return -1;
2297             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
2298
2299             /* generate picture parameters */
2300             x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
2301             x264_pps_write( &h->out.bs, h->pps );
2302             if( x264_nal_end( h ) )
2303                 return -1;
2304             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
2305         }
2306
2307         if( h->fenc->i_type != X264_TYPE_IDR )
2308         {
2309             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2310             x264_sei_recovery_point_write( h, &h->out.bs, h->param.i_keyint_max );
2311             x264_nal_end( h );
2312             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
2313         }
2314     }
2315
2316     /* Init the rate control */
2317     /* FIXME: Include slice header bit cost. */
2318     x264_ratecontrol_start( h, h->fenc->i_qpplus1, overhead*8 );
2319     i_global_qp = x264_ratecontrol_qp( h );
2320
2321     pic_out->i_qpplus1 =
2322     h->fdec->i_qpplus1 = i_global_qp + 1;
2323
2324     if( h->param.rc.b_stat_read && h->sh.i_type != SLICE_TYPE_I )
2325     {
2326         x264_reference_build_list_optimal( h );
2327         x264_reference_check_reorder( h );
2328     }
2329
2330     if( h->sh.i_type == SLICE_TYPE_B )
2331         x264_macroblock_bipred_init( h );
2332
2333     /*------------------------- Weights -------------------------------------*/
2334     x264_weighted_pred_init( h );
2335
2336     /* ------------------------ Create slice header  ----------------------- */
2337     x264_slice_init( h, i_nal_type, i_global_qp );
2338
2339     if( i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE )
2340         h->i_frame_num++;
2341
2342     /* Write frame */
2343     h->i_threadslice_start = 0;
2344     h->i_threadslice_end = h->sps->i_mb_height;
2345     if( h->i_thread_frames > 1 )
2346     {
2347         if( x264_pthread_create( &h->thread_handle, NULL, (void*)x264_slices_write, h ) )
2348             return -1;
2349         h->b_thread_active = 1;
2350     }
2351     else if( h->param.b_sliced_threads )
2352     {
2353         if( x264_threaded_slices_write( h ) )
2354             return -1;
2355     }
2356     else
2357         if( (intptr_t)x264_slices_write( h ) )
2358             return -1;
2359
2360     return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
2361 }
2362
2363 static int x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
2364                                    x264_nal_t **pp_nal, int *pi_nal,
2365                                    x264_picture_t *pic_out )
2366 {
2367     int i, j, i_list, frame_size;
2368     char psz_message[80];
2369
2370     if( h->b_thread_active )
2371     {
2372         void *ret = NULL;
2373         x264_pthread_join( h->thread_handle, &ret );
2374         h->b_thread_active = 0;
2375         if( (intptr_t)ret )
2376             return (intptr_t)ret;
2377     }
2378     if( !h->out.i_nal )
2379     {
2380         pic_out->i_type = X264_TYPE_AUTO;
2381         return 0;
2382     }
2383
2384     x264_frame_push_unused( thread_current, h->fenc );
2385
2386     /* End bitstream, set output  */
2387     *pi_nal = h->out.i_nal;
2388     *pp_nal = h->out.nal;
2389
2390     frame_size = x264_encoder_encapsulate_nals( h );
2391
2392     h->out.i_nal = 0;
2393
2394     /* Set output picture properties */
2395     if( h->sh.i_type == SLICE_TYPE_I )
2396         pic_out->i_type = h->i_nal_type == NAL_SLICE_IDR ? X264_TYPE_IDR : X264_TYPE_I;
2397     else if( h->sh.i_type == SLICE_TYPE_P )
2398         pic_out->i_type = X264_TYPE_P;
2399     else
2400         pic_out->i_type = X264_TYPE_B;
2401
2402     pic_out->b_keyframe = h->fenc->b_keyframe;
2403
2404     pic_out->i_pts = h->fenc->i_pts *= h->i_dts_compress_multiplier;
2405     if( h->frames.i_bframe_delay )
2406     {
2407         int64_t *i_prev_dts = thread_current->frames.i_prev_dts;
2408         if( h->i_frame <= h->frames.i_bframe_delay )
2409         {
2410             if( h->i_dts_compress_multiplier == 1 )
2411                 pic_out->i_dts = h->fenc->i_reordered_pts - h->frames.i_bframe_delay_time;
2412             else
2413             {
2414                 /* DTS compression */
2415                 if( h->i_frame == 1 )
2416                     thread_current->frames.i_init_delta = h->fenc->i_reordered_pts * h->i_dts_compress_multiplier;
2417                 pic_out->i_dts = h->i_frame * thread_current->frames.i_init_delta / h->i_dts_compress_multiplier;
2418             }
2419         }
2420         else
2421             pic_out->i_dts = i_prev_dts[ (h->i_frame - h->frames.i_bframe_delay) % h->frames.i_bframe_delay ];
2422         i_prev_dts[ h->i_frame % h->frames.i_bframe_delay ] = h->fenc->i_reordered_pts * h->i_dts_compress_multiplier;
2423     }
2424     else
2425         pic_out->i_dts = h->fenc->i_reordered_pts;
2426     assert( pic_out->i_pts >= pic_out->i_dts );
2427
2428     pic_out->img.i_plane = h->fdec->i_plane;
2429     for(i = 0; i < 3; i++)
2430     {
2431         pic_out->img.i_stride[i] = h->fdec->i_stride[i];
2432         pic_out->img.plane[i] = h->fdec->plane[i];
2433     }
2434
2435     /* ---------------------- Update encoder state ------------------------- */
2436
2437     /* update rc */
2438     x264_emms();
2439     if( x264_ratecontrol_end( h, frame_size * 8 ) < 0 )
2440         return -1;
2441
2442     x264_noise_reduction_update( thread_current );
2443
2444     /* ---------------------- Compute/Print statistics --------------------- */
2445     x264_thread_sync_stat( h, h->thread[0] );
2446
2447     /* Slice stat */
2448     h->stat.i_frame_count[h->sh.i_type]++;
2449     h->stat.i_frame_size[h->sh.i_type] += frame_size;
2450     h->stat.f_frame_qp[h->sh.i_type] += h->fdec->f_qp_avg_aq;
2451
2452     for( i = 0; i < X264_MBTYPE_MAX; i++ )
2453         h->stat.i_mb_count[h->sh.i_type][i] += h->stat.frame.i_mb_count[i];
2454     for( i = 0; i < X264_PARTTYPE_MAX; i++ )
2455         h->stat.i_mb_partition[h->sh.i_type][i] += h->stat.frame.i_mb_partition[i];
2456     for( i = 0; i < 2; i++ )
2457         h->stat.i_mb_count_8x8dct[i] += h->stat.frame.i_mb_count_8x8dct[i];
2458     for( i = 0; i < 6; i++ )
2459         h->stat.i_mb_cbp[i] += h->stat.frame.i_mb_cbp[i];
2460     for( i = 0; i < 3; i++ )
2461         for( j = 0; j < 13; j++ )
2462             h->stat.i_mb_pred_mode[i][j] += h->stat.frame.i_mb_pred_mode[i][j];
2463     if( h->sh.i_type != SLICE_TYPE_I )
2464         for( i_list = 0; i_list < 2; i_list++ )
2465             for( i = 0; i < 32; i++ )
2466                 h->stat.i_mb_count_ref[h->sh.i_type][i_list][i] += h->stat.frame.i_mb_count_ref[i_list][i];
2467     if( h->sh.i_type == SLICE_TYPE_P )
2468     {
2469         h->stat.i_consecutive_bframes[h->fdec->i_frame - h->fref0[0]->i_frame - 1]++;
2470         if( h->param.analyse.i_weighted_pred == X264_WEIGHTP_SMART )
2471         {
2472             for( i = 0; i < 3; i++ )
2473                 for( j = 0; j < h->i_ref0; j++ )
2474                     if( h->sh.weight[0][i].i_denom != 0 )
2475                     {
2476                         h->stat.i_wpred[i]++;
2477                         break;
2478                     }
2479         }
2480     }
2481     if( h->sh.i_type == SLICE_TYPE_B )
2482     {
2483         h->stat.i_direct_frames[ h->sh.b_direct_spatial_mv_pred ] ++;
2484         if( h->mb.b_direct_auto_write )
2485         {
2486             //FIXME somewhat arbitrary time constants
2487             if( h->stat.i_direct_score[0] + h->stat.i_direct_score[1] > h->mb.i_mb_count )
2488             {
2489                 for( i = 0; i < 2; i++ )
2490                     h->stat.i_direct_score[i] = h->stat.i_direct_score[i] * 9/10;
2491             }
2492             for( i = 0; i < 2; i++ )
2493                 h->stat.i_direct_score[i] += h->stat.frame.i_direct_score[i];
2494         }
2495     }
2496
2497     psz_message[0] = '\0';
2498     if( h->param.analyse.b_psnr )
2499     {
2500         int64_t ssd[3] = {
2501             h->stat.frame.i_ssd[0],
2502             h->stat.frame.i_ssd[1],
2503             h->stat.frame.i_ssd[2],
2504         };
2505
2506         h->stat.i_ssd_global[h->sh.i_type] += ssd[0] + ssd[1] + ssd[2];
2507         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 );
2508         h->stat.f_psnr_mean_y[h->sh.i_type] += x264_psnr( ssd[0], h->param.i_width * h->param.i_height );
2509         h->stat.f_psnr_mean_u[h->sh.i_type] += x264_psnr( ssd[1], h->param.i_width * h->param.i_height / 4 );
2510         h->stat.f_psnr_mean_v[h->sh.i_type] += x264_psnr( ssd[2], h->param.i_width * h->param.i_height / 4 );
2511
2512         snprintf( psz_message, 80, " PSNR Y:%5.2f U:%5.2f V:%5.2f",
2513                   x264_psnr( ssd[0], h->param.i_width * h->param.i_height ),
2514                   x264_psnr( ssd[1], h->param.i_width * h->param.i_height / 4),
2515                   x264_psnr( ssd[2], h->param.i_width * h->param.i_height / 4) );
2516     }
2517
2518     if( h->param.analyse.b_ssim )
2519     {
2520         double ssim_y = h->stat.frame.f_ssim
2521                       / (((h->param.i_width-6)>>2) * ((h->param.i_height-6)>>2));
2522         h->stat.f_ssim_mean_y[h->sh.i_type] += ssim_y;
2523         snprintf( psz_message + strlen(psz_message), 80 - strlen(psz_message),
2524                   " SSIM Y:%.5f", ssim_y );
2525     }
2526     psz_message[79] = '\0';
2527
2528     x264_log( h, X264_LOG_DEBUG,
2529                   "frame=%4d QP=%.2f NAL=%d Slice:%c Poc:%-3d I:%-4d P:%-4d SKIP:%-4d size=%d bytes%s\n",
2530               h->i_frame,
2531               h->fdec->f_qp_avg_aq,
2532               h->i_nal_ref_idc,
2533               h->sh.i_type == SLICE_TYPE_I ? 'I' : (h->sh.i_type == SLICE_TYPE_P ? 'P' : 'B' ),
2534               h->fdec->i_poc,
2535               h->stat.frame.i_mb_count_i,
2536               h->stat.frame.i_mb_count_p,
2537               h->stat.frame.i_mb_count_skip,
2538               frame_size,
2539               psz_message );
2540
2541     // keep stats all in one place
2542     x264_thread_sync_stat( h->thread[0], h );
2543     // for the use of the next frame
2544     x264_thread_sync_stat( thread_current, h );
2545
2546 #ifdef DEBUG_MB_TYPE
2547 {
2548     static const char mb_chars[] = { 'i', 'i', 'I', 'C', 'P', '8', 'S',
2549         'D', '<', 'X', 'B', 'X', '>', 'B', 'B', 'B', 'B', '8', 'S' };
2550     int mb_xy;
2551     for( mb_xy = 0; mb_xy < h->sps->i_mb_width * h->sps->i_mb_height; mb_xy++ )
2552     {
2553         if( h->mb.type[mb_xy] < X264_MBTYPE_MAX && h->mb.type[mb_xy] >= 0 )
2554             fprintf( stderr, "%c ", mb_chars[ h->mb.type[mb_xy] ] );
2555         else
2556             fprintf( stderr, "? " );
2557
2558         if( (mb_xy+1) % h->sps->i_mb_width == 0 )
2559             fprintf( stderr, "\n" );
2560     }
2561 }
2562 #endif
2563
2564     /* Remove duplicates, must be done near the end as breaks h->fref0 array
2565      * by freeing some of its pointers. */
2566      for( i = 0; i < h->i_ref0; i++ )
2567          if( h->fref0[i] && h->fref0[i]->b_duplicate )
2568          {
2569              x264_frame_push_blank_unused( h, h->fref0[i] );
2570              h->fref0[i] = 0;
2571          }
2572
2573     if( h->param.psz_dump_yuv )
2574         x264_frame_dump( h );
2575
2576     return frame_size;
2577 }
2578
2579 static void x264_print_intra( int64_t *i_mb_count, double i_count, int b_print_pcm, char *intra )
2580 {
2581     intra += sprintf( intra, "I16..4%s: %4.1f%% %4.1f%% %4.1f%%",
2582         b_print_pcm ? "..PCM" : "",
2583         i_mb_count[I_16x16]/ i_count,
2584         i_mb_count[I_8x8]  / i_count,
2585         i_mb_count[I_4x4]  / i_count );
2586     if( b_print_pcm )
2587         sprintf( intra, " %4.1f%%", i_mb_count[I_PCM]  / i_count );
2588 }
2589
2590 /****************************************************************************
2591  * x264_encoder_close:
2592  ****************************************************************************/
2593 void    x264_encoder_close  ( x264_t *h )
2594 {
2595     int64_t i_yuv_size = 3 * h->param.i_width * h->param.i_height / 2;
2596     int64_t i_mb_count_size[2][7] = {{0}};
2597     char buf[200];
2598     int i, j, i_list, i_type;
2599     int b_print_pcm = h->stat.i_mb_count[SLICE_TYPE_I][I_PCM]
2600                    || h->stat.i_mb_count[SLICE_TYPE_P][I_PCM]
2601                    || h->stat.i_mb_count[SLICE_TYPE_B][I_PCM];
2602
2603     x264_lookahead_delete( h );
2604
2605     if( h->param.i_threads > 1 )
2606     {
2607         // don't strictly have to wait for the other threads, but it's simpler than canceling them
2608         for( i = 0; i < h->param.i_threads; i++ )
2609             if( h->thread[i]->b_thread_active )
2610                 x264_pthread_join( h->thread[i]->thread_handle, NULL );
2611         if( h->i_thread_frames > 1 )
2612         {
2613             for( i = 0; i < h->i_thread_frames; i++ )
2614             {
2615                 if( h->thread[i]->b_thread_active )
2616                 {
2617                     assert( h->thread[i]->fenc->i_reference_count == 1 );
2618                     x264_frame_delete( h->thread[i]->fenc );
2619                 }
2620             }
2621
2622             x264_t *thread_prev = h->thread[h->i_thread_phase];
2623             x264_thread_sync_ratecontrol( h, thread_prev, h );
2624             x264_thread_sync_ratecontrol( thread_prev, thread_prev, h );
2625             h->i_frame = thread_prev->i_frame + 1 - h->i_thread_frames;
2626         }
2627     }
2628     h->i_frame++;
2629
2630     /* Slices used and PSNR */
2631     for( i=0; i<5; i++ )
2632     {
2633         static const int slice_order[] = { SLICE_TYPE_I, SLICE_TYPE_SI, SLICE_TYPE_P, SLICE_TYPE_SP, SLICE_TYPE_B };
2634         static const char *slice_name[] = { "P", "B", "I", "SP", "SI" };
2635         int i_slice = slice_order[i];
2636
2637         if( h->stat.i_frame_count[i_slice] > 0 )
2638         {
2639             const int i_count = h->stat.i_frame_count[i_slice];
2640             if( h->param.analyse.b_psnr )
2641             {
2642                 x264_log( h, X264_LOG_INFO,
2643                           "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",
2644                           slice_name[i_slice],
2645                           i_count,
2646                           h->stat.f_frame_qp[i_slice] / i_count,
2647                           (double)h->stat.i_frame_size[i_slice] / i_count,
2648                           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,
2649                           h->stat.f_psnr_average[i_slice] / i_count,
2650                           x264_psnr( h->stat.i_ssd_global[i_slice], i_count * i_yuv_size ) );
2651             }
2652             else
2653             {
2654                 x264_log( h, X264_LOG_INFO,
2655                           "frame %s:%-5d Avg QP:%5.2f  size:%6.0f\n",
2656                           slice_name[i_slice],
2657                           i_count,
2658                           h->stat.f_frame_qp[i_slice] / i_count,
2659                           (double)h->stat.i_frame_size[i_slice] / i_count );
2660             }
2661         }
2662     }
2663     if( h->param.i_bframe && h->stat.i_frame_count[SLICE_TYPE_P] )
2664     {
2665         char *p = buf;
2666         int den = 0;
2667         // weight by number of frames (including the P-frame) that are in a sequence of N B-frames
2668         for( i=0; i<=h->param.i_bframe; i++ )
2669             den += (i+1) * h->stat.i_consecutive_bframes[i];
2670         for( i=0; i<=h->param.i_bframe; i++ )
2671             p += sprintf( p, " %4.1f%%", 100. * (i+1) * h->stat.i_consecutive_bframes[i] / den );
2672         x264_log( h, X264_LOG_INFO, "consecutive B-frames:%s\n", buf );
2673     }
2674
2675     for( i_type = 0; i_type < 2; i_type++ )
2676         for( i = 0; i < X264_PARTTYPE_MAX; i++ )
2677         {
2678             if( i == D_DIRECT_8x8 ) continue; /* direct is counted as its own type */
2679             i_mb_count_size[i_type][x264_mb_partition_pixel_table[i]] += h->stat.i_mb_partition[i_type][i];
2680         }
2681
2682     /* MB types used */
2683     if( h->stat.i_frame_count[SLICE_TYPE_I] > 0 )
2684     {
2685         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_I];
2686         double i_count = h->stat.i_frame_count[SLICE_TYPE_I] * h->mb.i_mb_count / 100.0;
2687         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
2688         x264_log( h, X264_LOG_INFO, "mb I  %s\n", buf );
2689     }
2690     if( h->stat.i_frame_count[SLICE_TYPE_P] > 0 )
2691     {
2692         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_P];
2693         double i_count = h->stat.i_frame_count[SLICE_TYPE_P] * h->mb.i_mb_count / 100.0;
2694         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_P];
2695         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
2696         x264_log( h, X264_LOG_INFO,
2697                   "mb P  %s  P16..4: %4.1f%% %4.1f%% %4.1f%% %4.1f%% %4.1f%%    skip:%4.1f%%\n",
2698                   buf,
2699                   i_mb_size[PIXEL_16x16] / (i_count*4),
2700                   (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
2701                   i_mb_size[PIXEL_8x8] / (i_count*4),
2702                   (i_mb_size[PIXEL_8x4] + i_mb_size[PIXEL_4x8]) / (i_count*4),
2703                   i_mb_size[PIXEL_4x4] / (i_count*4),
2704                   i_mb_count[P_SKIP] / i_count );
2705     }
2706     if( h->stat.i_frame_count[SLICE_TYPE_B] > 0 )
2707     {
2708         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_B];
2709         double i_count = h->stat.i_frame_count[SLICE_TYPE_B] * h->mb.i_mb_count / 100.0;
2710         double i_mb_list_count;
2711         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_B];
2712         int64_t list_count[3] = {0}; /* 0 == L0, 1 == L1, 2 == BI */
2713         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
2714         for( i = 0; i < X264_PARTTYPE_MAX; i++ )
2715             for( j = 0; j < 2; j++ )
2716             {
2717                 int l0 = x264_mb_type_list_table[i][0][j];
2718                 int l1 = x264_mb_type_list_table[i][1][j];
2719                 if( l0 || l1 )
2720                     list_count[l1+l0*l1] += h->stat.i_mb_count[SLICE_TYPE_B][i] * 2;
2721             }
2722         list_count[0] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L0_8x8];
2723         list_count[1] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L1_8x8];
2724         list_count[2] += h->stat.i_mb_partition[SLICE_TYPE_B][D_BI_8x8];
2725         i_mb_count[B_DIRECT] += (h->stat.i_mb_partition[SLICE_TYPE_B][D_DIRECT_8x8]+2)/4;
2726         i_mb_list_count = (list_count[0] + list_count[1] + list_count[2]) / 100.0;
2727         x264_log( h, X264_LOG_INFO,
2728                   "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",
2729                   buf,
2730                   i_mb_size[PIXEL_16x16] / (i_count*4),
2731                   (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
2732                   i_mb_size[PIXEL_8x8] / (i_count*4),
2733                   i_mb_count[B_DIRECT] / i_count,
2734                   i_mb_count[B_SKIP]   / i_count,
2735                   list_count[0] / i_mb_list_count,
2736                   list_count[1] / i_mb_list_count,
2737                   list_count[2] / i_mb_list_count );
2738     }
2739
2740     x264_ratecontrol_summary( h );
2741
2742     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 )
2743     {
2744 #define SUM3(p) (p[SLICE_TYPE_I] + p[SLICE_TYPE_P] + p[SLICE_TYPE_B])
2745 #define SUM3b(p,o) (p[SLICE_TYPE_I][o] + p[SLICE_TYPE_P][o] + p[SLICE_TYPE_B][o])
2746         int64_t i_i8x8 = SUM3b( h->stat.i_mb_count, I_8x8 );
2747         int64_t i_intra = i_i8x8 + SUM3b( h->stat.i_mb_count, I_4x4 )
2748                                  + SUM3b( h->stat.i_mb_count, I_16x16 );
2749         int64_t i_all_intra = i_intra + SUM3b( h->stat.i_mb_count, I_PCM);
2750         const int i_count = h->stat.i_frame_count[SLICE_TYPE_I] +
2751                             h->stat.i_frame_count[SLICE_TYPE_P] +
2752                             h->stat.i_frame_count[SLICE_TYPE_B];
2753         int64_t i_mb_count = i_count * h->mb.i_mb_count;
2754         float fps = (float) h->param.i_fps_num / h->param.i_fps_den;
2755         float f_bitrate = fps * SUM3(h->stat.i_frame_size) / i_count / 125;
2756
2757         if( h->pps->b_transform_8x8_mode )
2758         {
2759             buf[0] = 0;
2760             if( h->stat.i_mb_count_8x8dct[0] )
2761                 sprintf( buf, " inter:%.1f%%", 100. * h->stat.i_mb_count_8x8dct[1] / h->stat.i_mb_count_8x8dct[0] );
2762             x264_log( h, X264_LOG_INFO, "8x8 transform intra:%.1f%%%s\n", 100. * i_i8x8 / i_intra, buf );
2763         }
2764
2765         if( h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO
2766             && h->stat.i_frame_count[SLICE_TYPE_B] )
2767         {
2768             x264_log( h, X264_LOG_INFO, "direct mvs  spatial:%.1f%% temporal:%.1f%%\n",
2769                       h->stat.i_direct_frames[1] * 100. / h->stat.i_frame_count[SLICE_TYPE_B],
2770                       h->stat.i_direct_frames[0] * 100. / h->stat.i_frame_count[SLICE_TYPE_B] );
2771         }
2772
2773         buf[0] = 0;
2774         if( i_mb_count != i_all_intra )
2775             sprintf( buf, " inter: %.1f%% %.1f%% %.1f%%",
2776                      h->stat.i_mb_cbp[1] * 100.0 / ((i_mb_count - i_all_intra)*4),
2777                      h->stat.i_mb_cbp[3] * 100.0 / ((i_mb_count - i_all_intra)  ),
2778                      h->stat.i_mb_cbp[5] * 100.0 / ((i_mb_count - i_all_intra)) );
2779         x264_log( h, X264_LOG_INFO, "coded y,uvDC,uvAC intra: %.1f%% %.1f%% %.1f%%%s\n",
2780                   h->stat.i_mb_cbp[0] * 100.0 / (i_all_intra*4),
2781                   h->stat.i_mb_cbp[2] * 100.0 / (i_all_intra  ),
2782                   h->stat.i_mb_cbp[4] * 100.0 / (i_all_intra  ), buf );
2783
2784         int64_t fixed_pred_modes[3][9] = {{0}};
2785         int64_t sum_pred_modes[3] = {0};
2786         for( i = 0; i <= I_PRED_16x16_DC_128; i++ )
2787         {
2788             fixed_pred_modes[0][x264_mb_pred_mode16x16_fix[i]] += h->stat.i_mb_pred_mode[0][i];
2789             sum_pred_modes[0] += h->stat.i_mb_pred_mode[0][i];
2790         }
2791         if( sum_pred_modes[0] )
2792             x264_log( h, X264_LOG_INFO, "i16 v,h,dc,p: %2.0f%% %2.0f%% %2.0f%% %2.0f%%\n",
2793                       fixed_pred_modes[0][0] * 100.0 / sum_pred_modes[0],
2794                       fixed_pred_modes[0][1] * 100.0 / sum_pred_modes[0],
2795                       fixed_pred_modes[0][2] * 100.0 / sum_pred_modes[0],
2796                       fixed_pred_modes[0][3] * 100.0 / sum_pred_modes[0] );
2797         for( i = 1; i <= 2; i++ )
2798         {
2799             for( j = 0; j <= I_PRED_8x8_DC_128; j++ )
2800             {
2801                 fixed_pred_modes[i][x264_mb_pred_mode4x4_fix(j)] += h->stat.i_mb_pred_mode[i][j];
2802                 sum_pred_modes[i] += h->stat.i_mb_pred_mode[i][j];
2803             }
2804             if( sum_pred_modes[i] )
2805                 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,
2806                           fixed_pred_modes[i][0] * 100.0 / sum_pred_modes[i],
2807                           fixed_pred_modes[i][1] * 100.0 / sum_pred_modes[i],
2808                           fixed_pred_modes[i][2] * 100.0 / sum_pred_modes[i],
2809                           fixed_pred_modes[i][3] * 100.0 / sum_pred_modes[i],
2810                           fixed_pred_modes[i][4] * 100.0 / sum_pred_modes[i],
2811                           fixed_pred_modes[i][5] * 100.0 / sum_pred_modes[i],
2812                           fixed_pred_modes[i][6] * 100.0 / sum_pred_modes[i],
2813                           fixed_pred_modes[i][7] * 100.0 / sum_pred_modes[i],
2814                           fixed_pred_modes[i][8] * 100.0 / sum_pred_modes[i] );
2815         }
2816
2817         if( h->param.analyse.i_weighted_pred == X264_WEIGHTP_SMART && h->stat.i_frame_count[SLICE_TYPE_P] > 0 )
2818             x264_log( h, X264_LOG_INFO, "Weighted P-Frames: Y:%.1f%%\n",
2819                       h->stat.i_wpred[0] * 100.0 / h->stat.i_frame_count[SLICE_TYPE_P] );
2820
2821         for( i_list = 0; i_list < 2; i_list++ )
2822         {
2823             int i_slice;
2824             for( i_slice = 0; i_slice < 2; i_slice++ )
2825             {
2826                 char *p = buf;
2827                 int64_t i_den = 0;
2828                 int i_max = 0;
2829                 for( i = 0; i < 32; i++ )
2830                     if( h->stat.i_mb_count_ref[i_slice][i_list][i] )
2831                     {
2832                         i_den += h->stat.i_mb_count_ref[i_slice][i_list][i];
2833                         i_max = i;
2834                     }
2835                 if( i_max == 0 )
2836                     continue;
2837                 for( i = 0; i <= i_max; i++ )
2838                     p += sprintf( p, " %4.1f%%", 100. * h->stat.i_mb_count_ref[i_slice][i_list][i] / i_den );
2839                 x264_log( h, X264_LOG_INFO, "ref %c L%d:%s\n", "PB"[i_slice], i_list, buf );
2840             }
2841         }
2842
2843         if( h->param.analyse.b_ssim )
2844         {
2845             x264_log( h, X264_LOG_INFO,
2846                       "SSIM Mean Y:%.7f\n",
2847                       SUM3( h->stat.f_ssim_mean_y ) / i_count );
2848         }
2849         if( h->param.analyse.b_psnr )
2850         {
2851             x264_log( h, X264_LOG_INFO,
2852                       "PSNR Mean Y:%6.3f U:%6.3f V:%6.3f Avg:%6.3f Global:%6.3f kb/s:%.2f\n",
2853                       SUM3( h->stat.f_psnr_mean_y ) / i_count,
2854                       SUM3( h->stat.f_psnr_mean_u ) / i_count,
2855                       SUM3( h->stat.f_psnr_mean_v ) / i_count,
2856                       SUM3( h->stat.f_psnr_average ) / i_count,
2857                       x264_psnr( SUM3( h->stat.i_ssd_global ), i_count * i_yuv_size ),
2858                       f_bitrate );
2859         }
2860         else
2861             x264_log( h, X264_LOG_INFO, "kb/s:%.2f\n", f_bitrate );
2862     }
2863
2864     /* rc */
2865     x264_ratecontrol_delete( h );
2866
2867     /* param */
2868     if( h->param.rc.psz_stat_out )
2869         free( h->param.rc.psz_stat_out );
2870     if( h->param.rc.psz_stat_in )
2871         free( h->param.rc.psz_stat_in );
2872
2873     x264_cqm_delete( h );
2874     x264_free( h->nal_buffer );
2875     x264_analyse_free_costs( h );
2876
2877     if( h->i_thread_frames > 1)
2878         h = h->thread[h->i_thread_phase];
2879
2880     /* frames */
2881     x264_frame_delete_list( h->frames.unused[0] );
2882     x264_frame_delete_list( h->frames.unused[1] );
2883     x264_frame_delete_list( h->frames.current );
2884     x264_frame_delete_list( h->frames.blank_unused );
2885
2886     h = h->thread[0];
2887
2888     for( i = h->param.i_threads - 1; i >= 0; i-- )
2889     {
2890         x264_frame_t **frame;
2891
2892         if( !h->param.b_sliced_threads || i == 0 )
2893         {
2894             for( frame = h->thread[i]->frames.reference; *frame; frame++ )
2895             {
2896                 assert( (*frame)->i_reference_count > 0 );
2897                 (*frame)->i_reference_count--;
2898                 if( (*frame)->i_reference_count == 0 )
2899                     x264_frame_delete( *frame );
2900             }
2901             frame = &h->thread[i]->fdec;
2902             assert( (*frame)->i_reference_count > 0 );
2903             (*frame)->i_reference_count--;
2904             if( (*frame)->i_reference_count == 0 )
2905                 x264_frame_delete( *frame );
2906             x264_macroblock_cache_end( h->thread[i] );
2907         }
2908         x264_free( h->thread[i]->scratch_buffer );
2909         x264_free( h->thread[i]->out.p_bitstream );
2910         x264_free( h->thread[i]->out.nal);
2911         x264_free( h->thread[i] );
2912     }
2913 }
2914
2915 /****************************************************************************
2916  * x264_encoder_delayed_frames:
2917  ****************************************************************************/
2918 int x264_encoder_delayed_frames( x264_t *h )
2919 {
2920     int delayed_frames = 0;
2921     int i;
2922     if( h->i_thread_frames > 1 )
2923     {
2924         for( i=0; i<h->i_thread_frames; i++ )
2925             delayed_frames += h->thread[i]->b_thread_active;
2926         h = h->thread[h->i_thread_phase];
2927     }
2928     for( i=0; h->frames.current[i]; i++ )
2929         delayed_frames++;
2930     x264_pthread_mutex_lock( &h->lookahead->ofbuf.mutex );
2931     x264_pthread_mutex_lock( &h->lookahead->ifbuf.mutex );
2932     x264_pthread_mutex_lock( &h->lookahead->next.mutex );
2933     delayed_frames += h->lookahead->ifbuf.i_size + h->lookahead->next.i_size + h->lookahead->ofbuf.i_size;
2934     x264_pthread_mutex_unlock( &h->lookahead->next.mutex );
2935     x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
2936     x264_pthread_mutex_unlock( &h->lookahead->ofbuf.mutex );
2937     return delayed_frames;
2938 }