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