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  *
   9  * This program is free software; you can redistribute it and/or modify
  10  * it under the terms of the GNU General Public License as published by
  11  * the Free Software Foundation; either version 2 of the License, or
  12  * (at your option) any later version.
  13  *
  14  * This program is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  17  * GNU General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU General Public License
  20  * along with this program; if not, write to the Free Software
  21  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02111, USA.
  22  *****************************************************************************/
  23
  24 #include <math.h>
  25
  26 #include "common/common.h"
  27 #include "common/cpu.h"
  28
  29 #include "set.h"
  30 #include "analyse.h"
  31 #include "ratecontrol.h"
  32 #include "macroblock.h"
  33
  34 #if VISUALIZE
  35 #include "common/visualize.h"
  36 #endif
  37
  38 //#define DEBUG_MB_TYPE
  39
  40 #define NALU_OVERHEAD 5 // startcode + NAL type costs 5 bytes per frame
  41
  42 #define bs_write_ue bs_write_ue_big
  43
  44 static void x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
  45                                     x264_nal_t **pp_nal, int *pi_nal,
  46                                     x264_picture_t *pic_out );
  47
  48 /****************************************************************************
  49  *
  50  ******************************* x264 libs **********************************
  51  *
  52  ****************************************************************************/
  53 static float x264_psnr( int64_t i_sqe, int64_t i_size )
  54 {
  55     double f_mse = (double)i_sqe / ((double)65025.0 * (double)i_size);
  56     if( f_mse <= 0.0000000001 ) /* Max 100dB */
  57         return 100;
  58
  59     return (float)(-10.0 * log( f_mse ) / log( 10.0 ));
  60 }
  61
  62 static void x264_frame_dump( x264_t *h )
  63 {
  64     FILE *f = fopen( h->param.psz_dump_yuv, "r+b" );
  65     int i, y;
  66     if( !f )
  67         return;
  68     /* Write the frame in display order */
  69     fseek( f, h->fdec->i_frame * h->param.i_height * h->param.i_width * 3/2, SEEK_SET );
  70     for( i = 0; i < h->fdec->i_plane; i++ )
  71         for( y = 0; y < h->param.i_height >> !!i; y++ )
  72             fwrite( &h->fdec->plane[i][y*h->fdec->i_stride[i]], 1, h->param.i_width >> !!i, f );
  73     fclose( f );
  74 }
  75
  76
  77 /* Fill "default" values */
  78 static void x264_slice_header_init( x264_t *h, x264_slice_header_t *sh,
  79                                     x264_sps_t *sps, x264_pps_t *pps,
  80                                     int i_idr_pic_id, int i_frame, int i_qp )
  81 {
  82     x264_param_t *param = &h->param;
  83     int i;
  84
  85     /* First we fill all field */
  86     sh->sps = sps;
  87     sh->pps = pps;
  88
  89     sh->i_first_mb  = 0;
  90     sh->i_last_mb   = h->sps->i_mb_width * h->sps->i_mb_height;
  91     sh->i_pps_id    = pps->i_id;
  92
  93     sh->i_frame_num = i_frame;
  94
  95     sh->b_mbaff = h->param.b_interlaced;
  96     sh->b_field_pic = 0;    /* no field support for now */
  97     sh->b_bottom_field = 0; /* not yet used */
  98
  99     sh->i_idr_pic_id = i_idr_pic_id;
 100
 101     /* poc stuff, fixed later */
 102     sh->i_poc_lsb = 0;
 103     sh->i_delta_poc_bottom = 0;
 104     sh->i_delta_poc[0] = 0;
 105     sh->i_delta_poc[1] = 0;
 106
 107     sh->i_redundant_pic_cnt = 0;
 108
 109     if( !h->mb.b_direct_auto_read )
 110     {
 111         if( h->mb.b_direct_auto_write )
 112             sh->b_direct_spatial_mv_pred = ( h->stat.i_direct_score[1] > h->stat.i_direct_score[0] );
 113         else
 114             sh->b_direct_spatial_mv_pred = ( param->analyse.i_direct_mv_pred == X264_DIRECT_PRED_SPATIAL );
 115     }
 116     /* else b_direct_spatial_mv_pred was read from the 2pass statsfile */
 117
 118     sh->b_num_ref_idx_override = 0;
 119     sh->i_num_ref_idx_l0_active = 1;
 120     sh->i_num_ref_idx_l1_active = 1;
 121
 122     sh->b_ref_pic_list_reordering_l0 = h->b_ref_reorder[0];
 123     sh->b_ref_pic_list_reordering_l1 = h->b_ref_reorder[1];
 124
 125     /* If the ref list isn't in the default order, construct reordering header */
 126     /* List1 reordering isn't needed yet */
 127     if( sh->b_ref_pic_list_reordering_l0 )
 128     {
 129         int pred_frame_num = i_frame;
 130         for( i = 0; i < h->i_ref0; i++ )
 131         {
 132             int diff = h->fref0[i]->i_frame_num - pred_frame_num;
 133             if( diff == 0 )
 134                 x264_log( h, X264_LOG_ERROR, "diff frame num == 0\n" );
 135             sh->ref_pic_list_order[0][i].idc = ( diff > 0 );
 136             sh->ref_pic_list_order[0][i].arg = abs( diff ) - 1;
 137             pred_frame_num = h->fref0[i]->i_frame_num;
 138         }
 139     }
 140
 141     sh->i_cabac_init_idc = param->i_cabac_init_idc;
 142
 143     sh->i_qp = i_qp;
 144     sh->i_qp_delta = i_qp - pps->i_pic_init_qp;
 145     sh->b_sp_for_swidth = 0;
 146     sh->i_qs_delta = 0;
 147
 148     /* If effective qp <= 15, deblocking would have no effect anyway */
 149     if( param->b_deblocking_filter
 150         && ( h->mb.b_variable_qp
 151         || 15 < i_qp + 2 * X264_MAX(param->i_deblocking_filter_alphac0, param->i_deblocking_filter_beta) ) )
 152     {
 153         sh->i_disable_deblocking_filter_idc = 0;
 154     }
 155     else
 156     {
 157         sh->i_disable_deblocking_filter_idc = 1;
 158     }
 159     sh->i_alpha_c0_offset = param->i_deblocking_filter_alphac0 << 1;
 160     sh->i_beta_offset = param->i_deblocking_filter_beta << 1;
 161 }
 162
 163 static void x264_slice_header_write( bs_t *s, x264_slice_header_t *sh, int i_nal_ref_idc )
 164 {
 165     int i;
 166
 167     if( sh->b_mbaff )
 168     {
 169         assert( sh->i_first_mb % (2*sh->sps->i_mb_width) == 0 );
 170         bs_write_ue( s, sh->i_first_mb >> 1 );
 171     }
 172     else
 173         bs_write_ue( s, sh->i_first_mb );
 174
 175     bs_write_ue( s, sh->i_type + 5 );   /* same type things */
 176     bs_write_ue( s, sh->i_pps_id );
 177     bs_write( s, sh->sps->i_log2_max_frame_num, sh->i_frame_num );
 178
 179     if( !sh->sps->b_frame_mbs_only )
 180     {
 181         bs_write1( s, sh->b_field_pic );
 182         if ( sh->b_field_pic )
 183             bs_write1( s, sh->b_bottom_field );
 184     }
 185
 186     if( sh->i_idr_pic_id >= 0 ) /* NAL IDR */
 187     {
 188         bs_write_ue( s, sh->i_idr_pic_id );
 189     }
 190
 191     if( sh->sps->i_poc_type == 0 )
 192     {
 193         bs_write( s, sh->sps->i_log2_max_poc_lsb, sh->i_poc_lsb );
 194         if( sh->pps->b_pic_order && !sh->b_field_pic )
 195         {
 196             bs_write_se( s, sh->i_delta_poc_bottom );
 197         }
 198     }
 199     else if( sh->sps->i_poc_type == 1 && !sh->sps->b_delta_pic_order_always_zero )
 200     {
 201         bs_write_se( s, sh->i_delta_poc[0] );
 202         if( sh->pps->b_pic_order && !sh->b_field_pic )
 203         {
 204             bs_write_se( s, sh->i_delta_poc[1] );
 205         }
 206     }
 207
 208     if( sh->pps->b_redundant_pic_cnt )
 209     {
 210         bs_write_ue( s, sh->i_redundant_pic_cnt );
 211     }
 212
 213     if( sh->i_type == SLICE_TYPE_B )
 214     {
 215         bs_write1( s, sh->b_direct_spatial_mv_pred );
 216     }
 217     if( sh->i_type == SLICE_TYPE_P || sh->i_type == SLICE_TYPE_SP || sh->i_type == SLICE_TYPE_B )
 218     {
 219         bs_write1( s, sh->b_num_ref_idx_override );
 220         if( sh->b_num_ref_idx_override )
 221         {
 222             bs_write_ue( s, sh->i_num_ref_idx_l0_active - 1 );
 223             if( sh->i_type == SLICE_TYPE_B )
 224             {
 225                 bs_write_ue( s, sh->i_num_ref_idx_l1_active - 1 );
 226             }
 227         }
 228     }
 229
 230     /* ref pic list reordering */
 231     if( sh->i_type != SLICE_TYPE_I )
 232     {
 233         bs_write1( s, sh->b_ref_pic_list_reordering_l0 );
 234         if( sh->b_ref_pic_list_reordering_l0 )
 235         {
 236             for( i = 0; i < sh->i_num_ref_idx_l0_active; i++ )
 237             {
 238                 bs_write_ue( s, sh->ref_pic_list_order[0][i].idc );
 239                 bs_write_ue( s, sh->ref_pic_list_order[0][i].arg );
 240
 241             }
 242             bs_write_ue( s, 3 );
 243         }
 244     }
 245     if( sh->i_type == SLICE_TYPE_B )
 246     {
 247         bs_write1( s, sh->b_ref_pic_list_reordering_l1 );
 248         if( sh->b_ref_pic_list_reordering_l1 )
 249         {
 250             for( i = 0; i < sh->i_num_ref_idx_l1_active; i++ )
 251             {
 252                 bs_write_ue( s, sh->ref_pic_list_order[1][i].idc );
 253                 bs_write_ue( s, sh->ref_pic_list_order[1][i].arg );
 254             }
 255             bs_write_ue( s, 3 );
 256         }
 257     }
 258
 259     if( ( sh->pps->b_weighted_pred && ( sh->i_type == SLICE_TYPE_P || sh->i_type == SLICE_TYPE_SP ) ) ||
 260         ( sh->pps->b_weighted_bipred == 1 && sh->i_type == SLICE_TYPE_B ) )
 261     {
 262         /* FIXME */
 263     }
 264
 265     if( i_nal_ref_idc != 0 )
 266     {
 267         if( sh->i_idr_pic_id >= 0 )
 268         {
 269             bs_write1( s, 0 );  /* no output of prior pics flag */
 270             bs_write1( s, 0 );  /* long term reference flag */
 271         }
 272         else
 273         {
 274             bs_write1( s, 0 );  /* adaptive_ref_pic_marking_mode_flag */
 275         }
 276     }
 277
 278     if( sh->pps->b_cabac && sh->i_type != SLICE_TYPE_I )
 279     {
 280         bs_write_ue( s, sh->i_cabac_init_idc );
 281     }
 282     bs_write_se( s, sh->i_qp_delta );      /* slice qp delta */
 283
 284     if( sh->pps->b_deblocking_filter_control )
 285     {
 286         bs_write_ue( s, sh->i_disable_deblocking_filter_idc );
 287         if( sh->i_disable_deblocking_filter_idc != 1 )
 288         {
 289             bs_write_se( s, sh->i_alpha_c0_offset >> 1 );
 290             bs_write_se( s, sh->i_beta_offset >> 1 );
 291         }
 292     }
 293 }
 294
 295 /* If we are within a reasonable distance of the end of the memory allocated for the bitstream, */
 296 /* reallocate, adding an arbitrary amount of space (100 kilobytes). */
 297 static void x264_bitstream_check_buffer( x264_t *h )
 298 {
 299     if( ( h->param.b_cabac && (h->cabac.p_end - h->cabac.p < 2500) )
 300      || ( h->out.bs.p_end - h->out.bs.p < 2500 ) )
 301     {
 302         uint8_t *bs_bak = h->out.p_bitstream;
 303         intptr_t delta;
 304         int i;
 305
 306         h->out.i_bitstream += 100000;
 307         h->out.p_bitstream = x264_realloc( h->out.p_bitstream, h->out.i_bitstream );
 308         delta = h->out.p_bitstream - bs_bak;
 309
 310         h->out.bs.p_start += delta;
 311         h->out.bs.p += delta;
 312         h->out.bs.p_end = h->out.p_bitstream + h->out.i_bitstream;
 313
 314         h->cabac.p_start += delta;
 315         h->cabac.p += delta;
 316         h->cabac.p_end = h->out.p_bitstream + h->out.i_bitstream;
 317
 318         for( i = 0; i <= h->out.i_nal; i++ )
 319             h->out.nal[i].p_payload += delta;
 320     }
 321 }
 322
 323 /****************************************************************************
 324  *
 325  ****************************************************************************
 326  ****************************** External API*********************************
 327  ****************************************************************************
 328  *
 329  ****************************************************************************/
 330
 331 static int x264_validate_parameters( x264_t *h )
 332 {
 333 #ifdef HAVE_MMX
 334     if( !(x264_cpu_detect() & X264_CPU_MMXEXT) )
 335     {
 336         x264_log( h, X264_LOG_ERROR, "your cpu does not support MMXEXT, but x264 was compiled with asm support\n");
 337         x264_log( h, X264_LOG_ERROR, "to run x264, recompile without asm support (configure --disable-asm)\n");
 338         return -1;
 339     }
 340 #endif
 341     if( h->param.i_width <= 0 || h->param.i_height <= 0 )
 342     {
 343         x264_log( h, X264_LOG_ERROR, "invalid width x height (%dx%d)\n",
 344                   h->param.i_width, h->param.i_height );
 345         return -1;
 346     }
 347
 348     if( h->param.i_width % 2 || h->param.i_height % 2 )
 349     {
 350         x264_log( h, X264_LOG_ERROR, "width or height not divisible by 2 (%dx%d)\n",
 351                   h->param.i_width, h->param.i_height );
 352         return -1;
 353     }
 354     if( h->param.i_csp != X264_CSP_I420 )
 355     {
 356         x264_log( h, X264_LOG_ERROR, "invalid CSP (only I420 supported)\n" );
 357         return -1;
 358     }
 359
 360     if( h->param.i_threads == 0 )
 361         h->param.i_threads = x264_cpu_num_processors() * 3/2;
 362     h->param.i_threads = x264_clip3( h->param.i_threads, 1, X264_THREAD_MAX );
 363     if( h->param.i_threads > 1 )
 364     {
 365 #ifndef HAVE_PTHREAD
 366         x264_log( h, X264_LOG_WARNING, "not compiled with pthread support!\n");
 367         h->param.i_threads = 1;
 368 #else
 369         if( h->param.i_scenecut_threshold >= 0 )
 370             h->param.b_pre_scenecut = 1;
 371 #endif
 372     }
 373
 374     if( h->param.b_interlaced )
 375     {
 376         if( h->param.analyse.i_me_method >= X264_ME_ESA )
 377         {
 378             x264_log( h, X264_LOG_WARNING, "interlace + me=esa is not implemented\n" );
 379             h->param.analyse.i_me_method = X264_ME_UMH;
 380         }
 381         if( h->param.analyse.i_direct_mv_pred > X264_DIRECT_PRED_SPATIAL )
 382         {
 383             x264_log( h, X264_LOG_WARNING, "interlace + direct=temporal is not implemented\n" );
 384             h->param.analyse.i_direct_mv_pred = X264_DIRECT_PRED_SPATIAL;
 385         }
 386     }
 387
 388     if( h->param.rc.i_rc_method < 0 || h->param.rc.i_rc_method > 2 )
 389     {
 390         x264_log( h, X264_LOG_ERROR, "no ratecontrol method specified\n" );
 391         return -1;
 392     }
 393     h->param.rc.f_rf_constant = x264_clip3f( h->param.rc.f_rf_constant, 0, 51 );
 394     h->param.rc.i_qp_constant = x264_clip3( h->param.rc.i_qp_constant, 0, 51 );
 395     if( h->param.rc.i_rc_method == X264_RC_CRF )
 396         h->param.rc.i_qp_constant = h->param.rc.f_rf_constant;
 397     if( (h->param.rc.i_rc_method == X264_RC_CQP || h->param.rc.i_rc_method == X264_RC_CRF)
 398         && h->param.rc.i_qp_constant == 0 )
 399     {
 400         h->mb.b_lossless = 1;
 401         h->param.i_cqm_preset = X264_CQM_FLAT;
 402         h->param.psz_cqm_file = NULL;
 403         h->param.rc.i_rc_method = X264_RC_CQP;
 404         h->param.rc.f_ip_factor = 1;
 405         h->param.rc.f_pb_factor = 1;
 406         h->param.analyse.b_transform_8x8 = 0;
 407         h->param.analyse.b_psnr = 0;
 408         h->param.analyse.b_ssim = 0;
 409         h->param.analyse.i_chroma_qp_offset = 0;
 410         h->param.analyse.i_trellis = 0;
 411         h->param.analyse.b_fast_pskip = 0;
 412         h->param.analyse.i_noise_reduction = 0;
 413         h->param.analyse.i_subpel_refine = x264_clip3( h->param.analyse.i_subpel_refine, 1, 6 );
 414     }
 415     if( h->param.rc.i_rc_method == X264_RC_CQP )
 416     {
 417         float qp_p = h->param.rc.i_qp_constant;
 418         float qp_i = qp_p - 6*log(h->param.rc.f_ip_factor)/log(2);
 419         float qp_b = qp_p + 6*log(h->param.rc.f_pb_factor)/log(2);
 420         h->param.rc.i_qp_min = x264_clip3( (int)(X264_MIN3( qp_p, qp_i, qp_b )), 0, 51 );
 421         h->param.rc.i_qp_max = x264_clip3( (int)(X264_MAX3( qp_p, qp_i, qp_b ) + .999), 0, 51 );
 422         h->param.rc.i_aq_mode = 0;
 423     }
 424
 425     if( ( h->param.i_width % 16 || h->param.i_height % 16 )
 426         && h->param.i_height != 1080 && !h->mb.b_lossless )
 427     {
 428         // There's nothing special about 1080 in that the warning still applies to it,
 429         // but chances are the user can't help it if his content is already 1080p,
 430         // so there's no point in warning in that case.
 431         x264_log( h, X264_LOG_WARNING,
 432                   "width or height not divisible by 16 (%dx%d), compression will suffer.\n",
 433                   h->param.i_width, h->param.i_height );
 434     }
 435
 436     h->param.i_frame_reference = x264_clip3( h->param.i_frame_reference, 1, 16 );
 437     if( h->param.i_keyint_max <= 0 )
 438         h->param.i_keyint_max = 1;
 439     h->param.i_keyint_min = x264_clip3( h->param.i_keyint_min, 1, h->param.i_keyint_max/2+1 );
 440
 441     h->param.i_bframe = x264_clip3( h->param.i_bframe, 0, X264_BFRAME_MAX );
 442     h->param.i_bframe_bias = x264_clip3( h->param.i_bframe_bias, -90, 100 );
 443     h->param.b_bframe_pyramid = h->param.b_bframe_pyramid && h->param.i_bframe > 1;
 444     h->param.b_bframe_adaptive = h->param.b_bframe_adaptive && h->param.i_bframe > 0;
 445     h->param.analyse.b_weighted_bipred = h->param.analyse.b_weighted_bipred && h->param.i_bframe > 0;
 446     h->mb.b_direct_auto_write = h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO
 447                                 && h->param.i_bframe
 448                                 && ( h->param.rc.b_stat_write || !h->param.rc.b_stat_read );
 449     if( h->param.i_scenecut_threshold < 0 )
 450         h->param.b_pre_scenecut = 0;
 451
 452     h->param.i_deblocking_filter_alphac0 = x264_clip3( h->param.i_deblocking_filter_alphac0, -6, 6 );
 453     h->param.i_deblocking_filter_beta    = x264_clip3( h->param.i_deblocking_filter_beta, -6, 6 );
 454     h->param.analyse.i_luma_deadzone[0] = x264_clip3( h->param.analyse.i_luma_deadzone[0], 0, 32 );
 455     h->param.analyse.i_luma_deadzone[1] = x264_clip3( h->param.analyse.i_luma_deadzone[1], 0, 32 );
 456
 457     h->param.i_cabac_init_idc = x264_clip3( h->param.i_cabac_init_idc, 0, 2 );
 458
 459     if( h->param.i_cqm_preset < X264_CQM_FLAT || h->param.i_cqm_preset > X264_CQM_CUSTOM )
 460         h->param.i_cqm_preset = X264_CQM_FLAT;
 461
 462     if( h->param.analyse.i_me_method < X264_ME_DIA ||
 463         h->param.analyse.i_me_method > X264_ME_TESA )
 464         h->param.analyse.i_me_method = X264_ME_HEX;
 465     if( h->param.analyse.i_me_range < 4 )
 466         h->param.analyse.i_me_range = 4;
 467     if( h->param.analyse.i_me_range > 16 && h->param.analyse.i_me_method <= X264_ME_HEX )
 468         h->param.analyse.i_me_range = 16;
 469     if( h->param.analyse.i_me_method == X264_ME_TESA &&
 470         (h->mb.b_lossless || h->param.analyse.i_subpel_refine <= 1) )
 471         h->param.analyse.i_me_method = X264_ME_ESA;
 472     h->param.analyse.i_subpel_refine = x264_clip3( h->param.analyse.i_subpel_refine, 1, 7 );
 473     h->param.analyse.b_bframe_rdo = h->param.analyse.b_bframe_rdo && h->param.analyse.i_subpel_refine >= 6;
 474     h->param.analyse.b_mixed_references = h->param.analyse.b_mixed_references && h->param.i_frame_reference > 1;
 475     h->param.analyse.inter &= X264_ANALYSE_PSUB16x16|X264_ANALYSE_PSUB8x8|X264_ANALYSE_BSUB16x16|
 476                               X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
 477     h->param.analyse.intra &= X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
 478     if( !(h->param.analyse.inter & X264_ANALYSE_PSUB16x16) )
 479         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
 480     if( !h->param.analyse.b_transform_8x8 )
 481     {
 482         h->param.analyse.inter &= ~X264_ANALYSE_I8x8;
 483         h->param.analyse.intra &= ~X264_ANALYSE_I8x8;
 484     }
 485     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -12, 12);
 486     if( !h->param.b_cabac )
 487         h->param.analyse.i_trellis = 0;
 488     h->param.analyse.i_trellis = x264_clip3( h->param.analyse.i_trellis, 0, 2 );
 489     h->param.rc.i_aq_mode = x264_clip3( h->param.rc.i_aq_mode, 0, 2 );
 490     if( h->param.rc.f_aq_strength <= 0 )
 491         h->param.rc.i_aq_mode = 0;
 492     /* VAQ effectively replaces qcomp, so qcomp is raised towards 1 to compensate. */
 493     if( h->param.rc.i_aq_mode == X264_AQ_GLOBAL )
 494         h->param.rc.f_qcompress = x264_clip3f(h->param.rc.f_qcompress + h->param.rc.f_aq_strength / 0.7, 0, 1);
 495     h->param.analyse.i_noise_reduction = x264_clip3( h->param.analyse.i_noise_reduction, 0, 1<<16 );
 496
 497     {
 498         const x264_level_t *l = x264_levels;
 499         while( l->level_idc != 0 && l->level_idc != h->param.i_level_idc )
 500             l++;
 501         if( l->level_idc == 0 )
 502         {
 503             x264_log( h, X264_LOG_ERROR, "invalid level_idc: %d\n", h->param.i_level_idc );
 504             return -1;
 505         }
 506         if( h->param.analyse.i_mv_range <= 0 )
 507             h->param.analyse.i_mv_range = l->mv_range;
 508         else
 509             h->param.analyse.i_mv_range = x264_clip3(h->param.analyse.i_mv_range, 32, 512);
 510         if( h->param.analyse.i_direct_8x8_inference < 0 )
 511             h->param.analyse.i_direct_8x8_inference = l->direct8x8;
 512     }
 513
 514     if( h->param.i_threads > 1 )
 515     {
 516         int r = h->param.analyse.i_mv_range_thread;
 517         int r2;
 518         if( r <= 0 )
 519         {
 520             // half of the available space is reserved and divided evenly among the threads,
 521             // the rest is allocated to whichever thread is far enough ahead to use it.
 522             // reserving more space increases quality for some videos, but costs more time
 523             // in thread synchronization.
 524             int max_range = (h->param.i_height + X264_THREAD_HEIGHT) / h->param.i_threads - X264_THREAD_HEIGHT;
 525             r = max_range / 2;
 526         }
 527         r = X264_MAX( r, h->param.analyse.i_me_range );
 528         r = X264_MIN( r, h->param.analyse.i_mv_range );
 529         // round up to use the whole mb row
 530         r2 = (r & ~15) + ((-X264_THREAD_HEIGHT) & 15);
 531         if( r2 < r )
 532             r2 += 16;
 533         x264_log( h, X264_LOG_DEBUG, "using mv_range_thread = %d\n", r2 );
 534         h->param.analyse.i_mv_range_thread = r2;
 535     }
 536
 537     if( h->param.rc.f_qblur < 0 )
 538         h->param.rc.f_qblur = 0;
 539     if( h->param.rc.f_complexity_blur < 0 )
 540         h->param.rc.f_complexity_blur = 0;
 541
 542     h->param.i_sps_id &= 31;
 543
 544     if( h->param.i_log_level < X264_LOG_INFO )
 545     {
 546         h->param.analyse.b_psnr = 0;
 547         h->param.analyse.b_ssim = 0;
 548     }
 549
 550     /* ensure the booleans are 0 or 1 so they can be used in math */
 551 #define BOOLIFY(x) h->param.x = !!h->param.x
 552     BOOLIFY( b_cabac );
 553     BOOLIFY( b_deblocking_filter );
 554     BOOLIFY( b_interlaced );
 555     BOOLIFY( analyse.b_transform_8x8 );
 556     BOOLIFY( analyse.i_direct_8x8_inference );
 557     BOOLIFY( analyse.b_bidir_me );
 558     BOOLIFY( analyse.b_chroma_me );
 559     BOOLIFY( analyse.b_fast_pskip );
 560     BOOLIFY( rc.b_stat_write );
 561     BOOLIFY( rc.b_stat_read );
 562 #undef BOOLIFY
 563
 564     return 0;
 565 }
 566
 567 static void mbcmp_init( x264_t *h )
 568 {
 569     int satd = !h->mb.b_lossless && h->param.analyse.i_subpel_refine > 1;
 570     memcpy( h->pixf.mbcmp, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.mbcmp) );
 571     satd &= h->param.analyse.i_me_method == X264_ME_TESA;
 572     memcpy( h->pixf.fpelcmp, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.fpelcmp) );
 573     memcpy( h->pixf.fpelcmp_x3, satd ? h->pixf.satd_x3 : h->pixf.sad_x3, sizeof(h->pixf.fpelcmp_x3) );
 574     memcpy( h->pixf.fpelcmp_x4, satd ? h->pixf.satd_x4 : h->pixf.sad_x4, sizeof(h->pixf.fpelcmp_x4) );
 575 }
 576
 577 /****************************************************************************
 578  * x264_encoder_open:
 579  ****************************************************************************/
 580 x264_t *x264_encoder_open   ( x264_param_t *param )
 581 {
 582     x264_t *h = x264_malloc( sizeof( x264_t ) );
 583     char buf[1000], *p;
 584     int i;
 585
 586     memset( h, 0, sizeof( x264_t ) );
 587
 588     /* Create a copy of param */
 589     memcpy( &h->param, param, sizeof( x264_param_t ) );
 590
 591     if( x264_validate_parameters( h ) < 0 )
 592     {
 593         x264_free( h );
 594         return NULL;
 595     }
 596
 597     if( h->param.psz_cqm_file )
 598         if( x264_cqm_parse_file( h, h->param.psz_cqm_file ) < 0 )
 599         {
 600             x264_free( h );
 601             return NULL;
 602         }
 603
 604     if( h->param.rc.psz_stat_out )
 605         h->param.rc.psz_stat_out = strdup( h->param.rc.psz_stat_out );
 606     if( h->param.rc.psz_stat_in )
 607         h->param.rc.psz_stat_in = strdup( h->param.rc.psz_stat_in );
 608     if( h->param.rc.psz_rc_eq )
 609         h->param.rc.psz_rc_eq = strdup( h->param.rc.psz_rc_eq );
 610
 611     /* VUI */
 612     if( h->param.vui.i_sar_width > 0 && h->param.vui.i_sar_height > 0 )
 613     {
 614         int i_w = param->vui.i_sar_width;
 615         int i_h = param->vui.i_sar_height;
 616
 617         x264_reduce_fraction( &i_w, &i_h );
 618
 619         while( i_w > 65535 || i_h > 65535 )
 620         {
 621             i_w /= 2;
 622             i_h /= 2;
 623         }
 624
 625         h->param.vui.i_sar_width = 0;
 626         h->param.vui.i_sar_height = 0;
 627         if( i_w == 0 || i_h == 0 )
 628         {
 629             x264_log( h, X264_LOG_WARNING, "cannot create valid sample aspect ratio\n" );
 630         }
 631         else
 632         {
 633             x264_log( h, X264_LOG_INFO, "using SAR=%d/%d\n", i_w, i_h );
 634             h->param.vui.i_sar_width = i_w;
 635             h->param.vui.i_sar_height = i_h;
 636         }
 637     }
 638
 639     x264_reduce_fraction( &h->param.i_fps_num, &h->param.i_fps_den );
 640
 641     /* Init x264_t */
 642     h->i_frame = 0;
 643     h->i_frame_num = 0;
 644     h->i_idr_pic_id = 0;
 645
 646     h->sps = &h->sps_array[0];
 647     x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
 648
 649     h->pps = &h->pps_array[0];
 650     x264_pps_init( h->pps, h->param.i_sps_id, &h->param, h->sps);
 651
 652     x264_validate_levels( h );
 653
 654     if( x264_cqm_init( h ) < 0 )
 655     {
 656         x264_free( h );
 657         return NULL;
 658     }
 659
 660     h->mb.i_mb_count = h->sps->i_mb_width * h->sps->i_mb_height;
 661
 662     /* Init frames. */
 663     h->frames.i_delay = h->param.i_bframe + h->param.i_threads - 1;
 664     h->frames.i_max_ref0 = h->param.i_frame_reference;
 665     h->frames.i_max_ref1 = h->sps->vui.i_num_reorder_frames;
 666     h->frames.i_max_dpb  = h->sps->vui.i_max_dec_frame_buffering;
 667     h->frames.b_have_lowres = !h->param.rc.b_stat_read
 668         && ( h->param.rc.i_rc_method == X264_RC_ABR
 669           || h->param.rc.i_rc_method == X264_RC_CRF
 670           || h->param.b_bframe_adaptive
 671           || h->param.b_pre_scenecut );
 672     h->frames.b_have_lowres |= (h->param.rc.b_stat_read && h->param.rc.i_vbv_buffer_size > 0);
 673
 674     h->frames.i_last_idr = - h->param.i_keyint_max;
 675     h->frames.i_input    = 0;
 676     h->frames.last_nonb  = NULL;
 677
 678     h->i_ref0 = 0;
 679     h->i_ref1 = 0;
 680
 681     x264_rdo_init( );
 682
 683     /* init CPU functions */
 684     x264_predict_16x16_init( h->param.cpu, h->predict_16x16 );
 685     x264_predict_8x8c_init( h->param.cpu, h->predict_8x8c );
 686     x264_predict_8x8_init( h->param.cpu, h->predict_8x8 );
 687     x264_predict_4x4_init( h->param.cpu, h->predict_4x4 );
 688
 689     x264_pixel_init( h->param.cpu, &h->pixf );
 690     x264_dct_init( h->param.cpu, &h->dctf );
 691     x264_zigzag_init( h->param.cpu, &h->zigzagf, h->param.b_interlaced );
 692     x264_mc_init( h->param.cpu, &h->mc );
 693     x264_quant_init( h, h->param.cpu, &h->quantf );
 694     x264_deblock_init( h->param.cpu, &h->loopf );
 695     x264_dct_init_weights();
 696
 697     mbcmp_init( h );
 698
 699     p = buf + sprintf( buf, "using cpu capabilities:" );
 700     for( i=0; x264_cpu_names[i].flags; i++ )
 701     {
 702         if( !strcmp(x264_cpu_names[i].name, "SSE2")
 703             && param->cpu & (X264_CPU_SSE2_IS_FAST|X264_CPU_SSE2_IS_SLOW) )
 704             continue;
 705         if( !strcmp(x264_cpu_names[i].name, "SSE3")
 706             && (param->cpu & X264_CPU_SSSE3 || !(param->cpu & X264_CPU_CACHELINE_64)) )
 707             continue;
 708         if( (param->cpu & x264_cpu_names[i].flags) == x264_cpu_names[i].flags
 709             && (!i || x264_cpu_names[i].flags != x264_cpu_names[i-1].flags) )
 710             p += sprintf( p, " %s", x264_cpu_names[i].name );
 711     }
 712     if( !param->cpu )
 713         p += sprintf( p, " none!" );
 714     x264_log( h, X264_LOG_INFO, "%s\n", buf );
 715
 716     h->out.i_nal = 0;
 717     h->out.i_bitstream = X264_MAX( 1000000, h->param.i_width * h->param.i_height * 4
 718         * ( h->param.rc.i_rc_method == X264_RC_ABR ? pow( 0.95, h->param.rc.i_qp_min )
 719           : pow( 0.95, h->param.rc.i_qp_constant ) * X264_MAX( 1, h->param.rc.f_ip_factor )));
 720
 721     h->thread[0] = h;
 722     h->i_thread_num = 0;
 723     for( i = 1; i < h->param.i_threads; i++ )
 724         h->thread[i] = x264_malloc( sizeof(x264_t) );
 725
 726     for( i = 0; i < h->param.i_threads; i++ )
 727     {
 728         if( i > 0 )
 729             *h->thread[i] = *h;
 730         h->thread[i]->fdec = x264_frame_pop_unused( h );
 731         h->thread[i]->out.p_bitstream = x264_malloc( h->out.i_bitstream );
 732         if( x264_macroblock_cache_init( h->thread[i] ) < 0 )
 733             return NULL;
 734     }
 735
 736     if( x264_ratecontrol_new( h ) < 0 )
 737         return NULL;
 738
 739     if( h->param.psz_dump_yuv )
 740     {
 741         /* create or truncate the reconstructed video file */
 742         FILE *f = fopen( h->param.psz_dump_yuv, "w" );
 743         if( f )
 744             fclose( f );
 745         else
 746         {
 747             x264_log( h, X264_LOG_ERROR, "can't write to fdec.yuv\n" );
 748             x264_free( h );
 749             return NULL;
 750         }
 751     }
 752
 753     return h;
 754 }
 755
 756 /****************************************************************************
 757  * x264_encoder_reconfig:
 758  ****************************************************************************/
 759 int x264_encoder_reconfig( x264_t *h, x264_param_t *param )
 760 {
 761 #define COPY(var) h->param.var = param->var
 762     COPY( i_frame_reference ); // but never uses more refs than initially specified
 763     COPY( i_bframe_bias );
 764     if( h->param.i_scenecut_threshold >= 0 && param->i_scenecut_threshold >= 0 )
 765         COPY( i_scenecut_threshold ); // can't turn it on or off, only vary the threshold
 766     COPY( b_deblocking_filter );
 767     COPY( i_deblocking_filter_alphac0 );
 768     COPY( i_deblocking_filter_beta );
 769     COPY( analyse.intra );
 770     COPY( analyse.inter );
 771     COPY( analyse.i_direct_mv_pred );
 772     COPY( analyse.i_me_method );
 773     COPY( analyse.i_me_range );
 774     COPY( analyse.i_noise_reduction );
 775     COPY( analyse.i_subpel_refine );
 776     COPY( analyse.i_trellis );
 777     COPY( analyse.b_bidir_me );
 778     COPY( analyse.b_bframe_rdo );
 779     COPY( analyse.b_chroma_me );
 780     COPY( analyse.b_dct_decimate );
 781     COPY( analyse.b_fast_pskip );
 782     COPY( analyse.b_mixed_references );
 783     // can only twiddle these if they were enabled to begin with:
 784     if( h->pps->b_transform_8x8_mode )
 785         COPY( analyse.b_transform_8x8 );
 786     if( h->frames.i_max_ref1 > 1 )
 787         COPY( b_bframe_pyramid );
 788 #undef COPY
 789
 790     mbcmp_init( h );
 791
 792     return x264_validate_parameters( h );
 793 }
 794
 795 /* internal usage */
 796 static void x264_nal_start( x264_t *h, int i_type, int i_ref_idc )
 797 {
 798     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
 799
 800     nal->i_ref_idc = i_ref_idc;
 801     nal->i_type    = i_type;
 802
 803     nal->i_payload= 0;
 804     nal->p_payload= &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8];
 805 }
 806 static void x264_nal_end( x264_t *h )
 807 {
 808     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
 809     nal->i_payload = &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8] - nal->p_payload;
 810     h->out.i_nal++;
 811 }
 812
 813 /****************************************************************************
 814  * x264_encoder_headers:
 815  ****************************************************************************/
 816 int x264_encoder_headers( x264_t *h, x264_nal_t **pp_nal, int *pi_nal )
 817 {
 818     /* init bitstream context */
 819     h->out.i_nal = 0;
 820     bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
 821
 822     /* Put SPS and PPS */
 823     if( h->i_frame == 0 )
 824     {
 825         /* identify ourself */
 826         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
 827         x264_sei_version_write( h, &h->out.bs );
 828         x264_nal_end( h );
 829
 830         /* generate sequence parameters */
 831         x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
 832         x264_sps_write( &h->out.bs, h->sps );
 833         x264_nal_end( h );
 834
 835         /* generate picture parameters */
 836         x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
 837         x264_pps_write( &h->out.bs, h->pps );
 838         x264_nal_end( h );
 839     }
 840     /* now set output*/
 841     *pi_nal = h->out.i_nal;
 842     *pp_nal = &h->out.nal[0];
 843     h->out.i_nal = 0;
 844
 845     return 0;
 846 }
 847
 848 static inline void x264_reference_build_list( x264_t *h, int i_poc )
 849 {
 850     int i;
 851     int b_ok;
 852
 853     /* build ref list 0/1 */
 854     h->i_ref0 = 0;
 855     h->i_ref1 = 0;
 856     for( i = 0; h->frames.reference[i]; i++ )
 857     {
 858         if( h->frames.reference[i]->i_poc < i_poc )
 859         {
 860             h->fref0[h->i_ref0++] = h->frames.reference[i];
 861         }
 862         else if( h->frames.reference[i]->i_poc > i_poc )
 863         {
 864             h->fref1[h->i_ref1++] = h->frames.reference[i];
 865         }
 866     }
 867
 868     /* Order ref0 from higher to lower poc */
 869     do
 870     {
 871         b_ok = 1;
 872         for( i = 0; i < h->i_ref0 - 1; i++ )
 873         {
 874             if( h->fref0[i]->i_poc < h->fref0[i+1]->i_poc )
 875             {
 876                 XCHG( x264_frame_t*, h->fref0[i], h->fref0[i+1] );
 877                 b_ok = 0;
 878                 break;
 879             }
 880         }
 881     } while( !b_ok );
 882     /* Order ref1 from lower to higher poc (bubble sort) for B-frame */
 883     do
 884     {
 885         b_ok = 1;
 886         for( i = 0; i < h->i_ref1 - 1; i++ )
 887         {
 888             if( h->fref1[i]->i_poc > h->fref1[i+1]->i_poc )
 889             {
 890                 XCHG( x264_frame_t*, h->fref1[i], h->fref1[i+1] );
 891                 b_ok = 0;
 892                 break;
 893             }
 894         }
 895     } while( !b_ok );
 896
 897     /* In the standard, a P-frame's ref list is sorted by frame_num.
 898      * We use POC, but check whether explicit reordering is needed */
 899     h->b_ref_reorder[0] =
 900     h->b_ref_reorder[1] = 0;
 901     if( h->sh.i_type == SLICE_TYPE_P )
 902     {
 903         for( i = 0; i < h->i_ref0 - 1; i++ )
 904             if( h->fref0[i]->i_frame_num < h->fref0[i+1]->i_frame_num )
 905             {
 906                 h->b_ref_reorder[0] = 1;
 907                 break;
 908             }
 909     }
 910
 911     h->i_ref1 = X264_MIN( h->i_ref1, h->frames.i_max_ref1 );
 912     h->i_ref0 = X264_MIN( h->i_ref0, h->frames.i_max_ref0 );
 913     h->i_ref0 = X264_MIN( h->i_ref0, h->param.i_frame_reference ); // if reconfig() has lowered the limit
 914     assert( h->i_ref0 + h->i_ref1 <= 16 );
 915     h->mb.pic.i_fref[0] = h->i_ref0;
 916     h->mb.pic.i_fref[1] = h->i_ref1;
 917 }
 918
 919 static void x264_fdec_filter_row( x264_t *h, int mb_y )
 920 {
 921     /* mb_y is the mb to be encoded next, not the mb to be filtered here */
 922     int b_hpel = h->fdec->b_kept_as_ref;
 923     int b_deblock = !h->sh.i_disable_deblocking_filter_idc;
 924     int b_end = mb_y == h->sps->i_mb_height;
 925     int min_y = mb_y - (1 << h->sh.b_mbaff);
 926     int max_y = b_end ? h->sps->i_mb_height : mb_y;
 927     b_deblock &= b_hpel || h->param.psz_dump_yuv;
 928     if( mb_y & h->sh.b_mbaff )
 929         return;
 930     if( min_y < 0 )
 931         return;
 932
 933     if( !b_end )
 934     {
 935         int i, j;
 936         for( j=0; j<=h->sh.b_mbaff; j++ )
 937             for( i=0; i<3; i++ )
 938             {
 939                 memcpy( h->mb.intra_border_backup[j][i],
 940                         h->fdec->plane[i] + ((mb_y*16 >> !!i) + j - 1 - h->sh.b_mbaff) * h->fdec->i_stride[i],
 941                         h->sps->i_mb_width*16 >> !!i );
 942             }
 943     }
 944
 945     if( b_deblock )
 946     {
 947         int y;
 948         for( y = min_y; y < max_y; y += (1 << h->sh.b_mbaff) )
 949             x264_frame_deblock_row( h, y );
 950     }
 951
 952     if( b_hpel )
 953     {
 954         x264_frame_expand_border( h, h->fdec, min_y, b_end );
 955         x264_frame_filter( h, h->fdec, min_y, b_end );
 956         x264_frame_expand_border_filtered( h, h->fdec, min_y, b_end );
 957     }
 958
 959     if( h->param.i_threads > 1 && h->fdec->b_kept_as_ref )
 960     {
 961         x264_frame_cond_broadcast( h->fdec, mb_y*16 + (b_end ? 10000 : -(X264_THREAD_HEIGHT << h->sh.b_mbaff)) );
 962     }
 963
 964     min_y = X264_MAX( min_y*16-8, 0 );
 965     max_y = b_end ? h->param.i_height : mb_y*16-8;
 966
 967     if( h->param.analyse.b_psnr )
 968     {
 969         int i;
 970         for( i=0; i<3; i++ )
 971             h->stat.frame.i_ssd[i] +=
 972                 x264_pixel_ssd_wxh( &h->pixf,
 973                     h->fdec->plane[i] + (min_y>>!!i) * h->fdec->i_stride[i], h->fdec->i_stride[i],
 974                     h->fenc->plane[i] + (min_y>>!!i) * h->fenc->i_stride[i], h->fenc->i_stride[i],
 975                     h->param.i_width >> !!i, (max_y-min_y) >> !!i );
 976     }
 977
 978     if( h->param.analyse.b_ssim )
 979     {
 980         x264_emms();
 981         /* offset by 2 pixels to avoid alignment of ssim blocks with dct blocks,
 982          * and overlap by 4 */
 983         min_y += min_y == 0 ? 2 : -6;
 984         h->stat.frame.f_ssim +=
 985             x264_pixel_ssim_wxh( &h->pixf,
 986                 h->fdec->plane[0] + 2+min_y*h->fdec->i_stride[0], h->fdec->i_stride[0],
 987                 h->fenc->plane[0] + 2+min_y*h->fenc->i_stride[0], h->fenc->i_stride[0],
 988                 h->param.i_width-2, max_y-min_y );
 989     }
 990 }
 991
 992 static inline void x264_reference_update( x264_t *h )
 993 {
 994     int i;
 995
 996     if( h->fdec->i_frame >= 0 )
 997         h->i_frame++;
 998
 999     if( !h->fdec->b_kept_as_ref )
1000     {
1001         if( h->param.i_threads > 1 )
1002         {
1003             x264_frame_push_unused( h, h->fdec );
1004             h->fdec = x264_frame_pop_unused( h );
1005         }
1006         return;
1007     }
1008
1009     /* move lowres copy of the image to the ref frame */
1010     for( i = 0; i < 4; i++)
1011     {
1012         XCHG( uint8_t*, h->fdec->lowres[i], h->fenc->lowres[i] );
1013         XCHG( uint8_t*, h->fdec->buffer_lowres[i], h->fenc->buffer_lowres[i] );
1014     }
1015
1016     /* adaptive B decision needs a pointer, since it can't use the ref lists */
1017     if( h->sh.i_type != SLICE_TYPE_B )
1018         h->frames.last_nonb = h->fdec;
1019
1020     /* move frame in the buffer */
1021     x264_frame_push( h->frames.reference, h->fdec );
1022     if( h->frames.reference[h->frames.i_max_dpb] )
1023         x264_frame_push_unused( h, x264_frame_shift( h->frames.reference ) );
1024     h->fdec = x264_frame_pop_unused( h );
1025 }
1026
1027 static inline void x264_reference_reset( x264_t *h )
1028 {
1029     while( h->frames.reference[0] )
1030         x264_frame_push_unused( h, x264_frame_pop( h->frames.reference ) );
1031     h->fdec->i_poc =
1032     h->fenc->i_poc = 0;
1033 }
1034
1035 static inline void x264_slice_init( x264_t *h, int i_nal_type, int i_global_qp )
1036 {
1037     /* ------------------------ Create slice header  ----------------------- */
1038     if( i_nal_type == NAL_SLICE_IDR )
1039     {
1040         x264_slice_header_init( h, &h->sh, h->sps, h->pps, h->i_idr_pic_id, h->i_frame_num, i_global_qp );
1041
1042         /* increment id */
1043         h->i_idr_pic_id = ( h->i_idr_pic_id + 1 ) % 65536;
1044     }
1045     else
1046     {
1047         x264_slice_header_init( h, &h->sh, h->sps, h->pps, -1, h->i_frame_num, i_global_qp );
1048
1049         /* always set the real higher num of ref frame used */
1050         h->sh.b_num_ref_idx_override = 1;
1051         h->sh.i_num_ref_idx_l0_active = h->i_ref0 <= 0 ? 1 : h->i_ref0;
1052         h->sh.i_num_ref_idx_l1_active = h->i_ref1 <= 0 ? 1 : h->i_ref1;
1053     }
1054
1055     h->fdec->i_frame_num = h->sh.i_frame_num;
1056
1057     if( h->sps->i_poc_type == 0 )
1058     {
1059         h->sh.i_poc_lsb = h->fdec->i_poc & ( (1 << h->sps->i_log2_max_poc_lsb) - 1 );
1060         h->sh.i_delta_poc_bottom = 0;   /* XXX won't work for field */
1061     }
1062     else if( h->sps->i_poc_type == 1 )
1063     {
1064         /* FIXME TODO FIXME */
1065     }
1066     else
1067     {
1068         /* Nothing to do ? */
1069     }
1070
1071     x264_macroblock_slice_init( h );
1072 }
1073
1074 static void x264_slice_write( x264_t *h )
1075 {
1076     int i_skip;
1077     int mb_xy, i_mb_x, i_mb_y;
1078     int i, i_list, i_ref;
1079
1080     /* init stats */
1081     memset( &h->stat.frame, 0, sizeof(h->stat.frame) );
1082
1083     /* Slice */
1084     x264_nal_start( h, h->i_nal_type, h->i_nal_ref_idc );
1085
1086     /* Slice header */
1087     x264_slice_header_write( &h->out.bs, &h->sh, h->i_nal_ref_idc );
1088     if( h->param.b_cabac )
1089     {
1090         /* alignment needed */
1091         bs_align_1( &h->out.bs );
1092
1093         /* init cabac */
1094         x264_cabac_context_init( &h->cabac, h->sh.i_type, h->sh.i_qp, h->sh.i_cabac_init_idc );
1095         x264_cabac_encode_init ( &h->cabac, h->out.bs.p, h->out.bs.p_end );
1096     }
1097     h->mb.i_last_qp = h->sh.i_qp;
1098     h->mb.i_last_dqp = 0;
1099
1100     i_mb_y = h->sh.i_first_mb / h->sps->i_mb_width;
1101     i_mb_x = h->sh.i_first_mb % h->sps->i_mb_width;
1102     i_skip = 0;
1103
1104     while( (mb_xy = i_mb_x + i_mb_y * h->sps->i_mb_width) < h->sh.i_last_mb )
1105     {
1106         int mb_spos = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
1107
1108         if( i_mb_x == 0 )
1109             x264_fdec_filter_row( h, i_mb_y );
1110
1111         /* load cache */
1112         x264_macroblock_cache_load( h, i_mb_x, i_mb_y );
1113
1114         /* analyse parameters
1115          * Slice I: choose I_4x4 or I_16x16 mode
1116          * Slice P: choose between using P mode or intra (4x4 or 16x16)
1117          * */
1118         x264_macroblock_analyse( h );
1119
1120         /* encode this macroblock -> be careful it can change the mb type to P_SKIP if needed */
1121         x264_macroblock_encode( h );
1122
1123         x264_bitstream_check_buffer( h );
1124
1125         if( h->param.b_cabac )
1126         {
1127             if( mb_xy > h->sh.i_first_mb && !(h->sh.b_mbaff && (i_mb_y&1)) )
1128                 x264_cabac_encode_terminal( &h->cabac );
1129
1130             if( IS_SKIP( h->mb.i_type ) )
1131                 x264_cabac_mb_skip( h, 1 );
1132             else
1133             {
1134                 if( h->sh.i_type != SLICE_TYPE_I )
1135                     x264_cabac_mb_skip( h, 0 );
1136                 x264_macroblock_write_cabac( h, &h->cabac );
1137             }
1138         }
1139         else
1140         {
1141             if( IS_SKIP( h->mb.i_type ) )
1142                 i_skip++;
1143             else
1144             {
1145                 if( h->sh.i_type != SLICE_TYPE_I )
1146                 {
1147                     bs_write_ue( &h->out.bs, i_skip );  /* skip run */
1148                     i_skip = 0;
1149                 }
1150                 x264_macroblock_write_cavlc( h, &h->out.bs );
1151             }
1152         }
1153
1154 #if VISUALIZE
1155         if( h->param.b_visualize )
1156             x264_visualize_mb( h );
1157 #endif
1158
1159         /* save cache */
1160         x264_macroblock_cache_save( h );
1161
1162         /* accumulate mb stats */
1163         h->stat.frame.i_mb_count[h->mb.i_type]++;
1164         if( !IS_SKIP(h->mb.i_type) && !IS_INTRA(h->mb.i_type) && !IS_DIRECT(h->mb.i_type) )
1165         {
1166             if( h->mb.i_partition != D_8x8 )
1167                 h->stat.frame.i_mb_partition[h->mb.i_partition] += 4;
1168             else
1169                 for( i = 0; i < 4; i++ )
1170                     h->stat.frame.i_mb_partition[h->mb.i_sub_partition[i]] ++;
1171             if( h->param.i_frame_reference > 1 )
1172                 for( i_list = 0; i_list <= (h->sh.i_type == SLICE_TYPE_B); i_list++ )
1173                     for( i = 0; i < 4; i++ )
1174                     {
1175                         i_ref = h->mb.cache.ref[i_list][ x264_scan8[4*i] ];
1176                         if( i_ref >= 0 )
1177                             h->stat.frame.i_mb_count_ref[i_list][i_ref] ++;
1178                     }
1179         }
1180         if( h->mb.i_cbp_luma && !IS_INTRA(h->mb.i_type) )
1181         {
1182             h->stat.frame.i_mb_count_8x8dct[0] ++;
1183             h->stat.frame.i_mb_count_8x8dct[1] += h->mb.b_transform_8x8;
1184         }
1185
1186         x264_ratecontrol_mb( h, bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac) - mb_spos );
1187
1188         if( h->sh.b_mbaff )
1189         {
1190             i_mb_x += i_mb_y & 1;
1191             i_mb_y ^= i_mb_x < h->sps->i_mb_width;
1192         }
1193         else
1194             i_mb_x++;
1195         if(i_mb_x == h->sps->i_mb_width)
1196         {
1197             i_mb_y++;
1198             i_mb_x = 0;
1199         }
1200     }
1201
1202     if( h->param.b_cabac )
1203     {
1204         x264_cabac_encode_flush( h, &h->cabac );
1205         h->out.bs.p = h->cabac.p;
1206     }
1207     else
1208     {
1209         if( i_skip > 0 )
1210             bs_write_ue( &h->out.bs, i_skip );  /* last skip run */
1211         /* rbsp_slice_trailing_bits */
1212         bs_rbsp_trailing( &h->out.bs );
1213     }
1214
1215     x264_nal_end( h );
1216
1217     x264_fdec_filter_row( h, h->sps->i_mb_height );
1218
1219     /* Compute misc bits */
1220     h->stat.frame.i_misc_bits = bs_pos( &h->out.bs )
1221                               + NALU_OVERHEAD * 8
1222                               - h->stat.frame.i_itex_bits
1223                               - h->stat.frame.i_ptex_bits
1224                               - h->stat.frame.i_hdr_bits;
1225 }
1226
1227 static void x264_thread_sync_context( x264_t *dst, x264_t *src )
1228 {
1229     x264_frame_t **f;
1230     if( dst == src )
1231         return;
1232
1233     // reference counting
1234     for( f = src->frames.reference; *f; f++ )
1235         (*f)->i_reference_count++;
1236     for( f = dst->frames.reference; *f; f++ )
1237         x264_frame_push_unused( src, *f );
1238     src->fdec->i_reference_count++;
1239     x264_frame_push_unused( src, dst->fdec );
1240
1241     // copy everything except the per-thread pointers and the constants.
1242     memcpy( &dst->i_frame, &src->i_frame, offsetof(x264_t, mb.type) - offsetof(x264_t, i_frame) );
1243     memcpy( &dst->mb.i_type, &src->mb.i_type, offsetof(x264_t, rc) - offsetof(x264_t, mb.i_type) );
1244     dst->stat = src->stat;
1245 }
1246
1247 static void x264_thread_sync_stat( x264_t *dst, x264_t *src )
1248 {
1249     if( dst == src )
1250         return;
1251     memcpy( &dst->stat.i_slice_count, &src->stat.i_slice_count, sizeof(dst->stat) - sizeof(dst->stat.frame) );
1252 }
1253
1254 static int x264_slices_write( x264_t *h )
1255 {
1256     int i_frame_size;
1257
1258 #if VISUALIZE
1259     if( h->param.b_visualize )
1260         x264_visualize_init( h );
1261 #endif
1262
1263     x264_stack_align( x264_slice_write, h );
1264     i_frame_size = h->out.nal[h->out.i_nal-1].i_payload;
1265
1266 #if VISUALIZE
1267     if( h->param.b_visualize )
1268     {
1269         x264_visualize_show( h );
1270         x264_visualize_close( h );
1271     }
1272 #endif
1273
1274     h->out.i_frame_size = i_frame_size;
1275     return 0;
1276 }
1277
1278 /****************************************************************************
1279  * x264_encoder_encode:
1280  *  XXX: i_poc   : is the poc of the current given picture
1281  *       i_frame : is the number of the frame being coded
1282  *  ex:  type frame poc
1283  *       I      0   2*0
1284  *       P      1   2*3
1285  *       B      2   2*1
1286  *       B      3   2*2
1287  *       P      4   2*6
1288  *       B      5   2*4
1289  *       B      6   2*5
1290  ****************************************************************************/
1291 int     x264_encoder_encode( x264_t *h,
1292                              x264_nal_t **pp_nal, int *pi_nal,
1293                              x264_picture_t *pic_in,
1294                              x264_picture_t *pic_out )
1295 {
1296     x264_t *thread_current, *thread_prev, *thread_oldest;
1297     int     i_nal_type;
1298     int     i_nal_ref_idc;
1299
1300     int   i_global_qp;
1301
1302     if( h->param.i_threads > 1)
1303     {
1304         int i = ++h->i_thread_phase;
1305         int t = h->param.i_threads;
1306         thread_current = h->thread[ i%t ];
1307         thread_prev    = h->thread[ (i-1)%t ];
1308         thread_oldest  = h->thread[ (i+1)%t ];
1309         x264_thread_sync_context( thread_current, thread_prev );
1310         x264_thread_sync_ratecontrol( thread_current, thread_prev, thread_oldest );
1311         h = thread_current;
1312 //      fprintf(stderr, "current: %p  prev: %p  oldest: %p \n", thread_current, thread_prev, thread_oldest);
1313     }
1314     else
1315     {
1316         thread_current =
1317         thread_prev    =
1318         thread_oldest  = h;
1319     }
1320
1321     // ok to call this before encoding any frames, since the initial values of fdec have b_kept_as_ref=0
1322     x264_reference_update( h );
1323     h->fdec->i_lines_completed = -1;
1324
1325     /* no data out */
1326     *pi_nal = 0;
1327     *pp_nal = NULL;
1328
1329     /* ------------------- Setup new frame from picture -------------------- */
1330     if( pic_in != NULL )
1331     {
1332         /* 1: Copy the picture to a frame and move it to a buffer */
1333         x264_frame_t *fenc = x264_frame_pop_unused( h );
1334
1335         if( x264_frame_copy_picture( h, fenc, pic_in ) < 0 )
1336             return -1;
1337
1338         if( h->param.i_width != 16 * h->sps->i_mb_width ||
1339             h->param.i_height != 16 * h->sps->i_mb_height )
1340             x264_frame_expand_border_mod16( h, fenc );
1341
1342         fenc->i_frame = h->frames.i_input++;
1343
1344         x264_frame_push( h->frames.next, fenc );
1345
1346         if( h->frames.b_have_lowres )
1347             x264_frame_init_lowres( h, fenc );
1348
1349         if( h->frames.i_input <= h->frames.i_delay + 1 - h->param.i_threads )
1350         {
1351             /* Nothing yet to encode */
1352             /* waiting for filling bframe buffer */
1353             pic_out->i_type = X264_TYPE_AUTO;
1354             return 0;
1355         }
1356     }
1357
1358     if( h->frames.current[0] == NULL )
1359     {
1360         int bframes = 0;
1361         /* 2: Select frame types */
1362         if( h->frames.next[0] == NULL )
1363         {
1364             x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
1365             return 0;
1366         }
1367
1368         x264_slicetype_decide( h );
1369
1370         /* 3: move some B-frames and 1 non-B to encode queue */
1371         while( IS_X264_TYPE_B( h->frames.next[bframes]->i_type ) )
1372             bframes++;
1373         x264_frame_push( h->frames.current, x264_frame_shift( &h->frames.next[bframes] ) );
1374         /* FIXME: when max B-frames > 3, BREF may no longer be centered after GOP closing */
1375         if( h->param.b_bframe_pyramid && bframes > 1 )
1376         {
1377             x264_frame_t *mid = x264_frame_shift( &h->frames.next[bframes/2] );
1378             mid->i_type = X264_TYPE_BREF;
1379             x264_frame_push( h->frames.current, mid );
1380             bframes--;
1381         }
1382         while( bframes-- )
1383             x264_frame_push( h->frames.current, x264_frame_shift( h->frames.next ) );
1384     }
1385
1386     /* ------------------- Get frame to be encoded ------------------------- */
1387     /* 4: get picture to encode */
1388     h->fenc = x264_frame_shift( h->frames.current );
1389     if( h->fenc == NULL )
1390     {
1391         /* Nothing yet to encode (ex: waiting for I/P with B frames) */
1392         /* waiting for filling bframe buffer */
1393         pic_out->i_type = X264_TYPE_AUTO;
1394         return 0;
1395     }
1396
1397 do_encode:
1398
1399     if( h->fenc->i_type == X264_TYPE_IDR )
1400     {
1401         h->frames.i_last_idr = h->fenc->i_frame;
1402     }
1403
1404     /* ------------------- Setup frame context ----------------------------- */
1405     /* 5: Init data dependent of frame type */
1406     if( h->fenc->i_type == X264_TYPE_IDR )
1407     {
1408         /* reset ref pictures */
1409         x264_reference_reset( h );
1410
1411         i_nal_type    = NAL_SLICE_IDR;
1412         i_nal_ref_idc = NAL_PRIORITY_HIGHEST;
1413         h->sh.i_type = SLICE_TYPE_I;
1414     }
1415     else if( h->fenc->i_type == X264_TYPE_I )
1416     {
1417         i_nal_type    = NAL_SLICE;
1418         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
1419         h->sh.i_type = SLICE_TYPE_I;
1420     }
1421     else if( h->fenc->i_type == X264_TYPE_P )
1422     {
1423         i_nal_type    = NAL_SLICE;
1424         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
1425         h->sh.i_type = SLICE_TYPE_P;
1426     }
1427     else if( h->fenc->i_type == X264_TYPE_BREF )
1428     {
1429         i_nal_type    = NAL_SLICE;
1430         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* maybe add MMCO to forget it? -> low */
1431         h->sh.i_type = SLICE_TYPE_B;
1432     }
1433     else    /* B frame */
1434     {
1435         i_nal_type    = NAL_SLICE;
1436         i_nal_ref_idc = NAL_PRIORITY_DISPOSABLE;
1437         h->sh.i_type = SLICE_TYPE_B;
1438     }
1439
1440     h->fdec->i_poc =
1441     h->fenc->i_poc = 2 * (h->fenc->i_frame - h->frames.i_last_idr);
1442     h->fdec->i_type = h->fenc->i_type;
1443     h->fdec->i_frame = h->fenc->i_frame;
1444     h->fenc->b_kept_as_ref =
1445     h->fdec->b_kept_as_ref = i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE && h->param.i_keyint_max > 1;
1446
1447
1448
1449     /* ------------------- Init                ----------------------------- */
1450     /* build ref list 0/1 */
1451     x264_reference_build_list( h, h->fdec->i_poc );
1452
1453     /* Init the rate control */
1454     x264_ratecontrol_start( h, h->fenc->i_qpplus1 );
1455     i_global_qp = x264_ratecontrol_qp( h );
1456
1457     pic_out->i_qpplus1 =
1458     h->fdec->i_qpplus1 = i_global_qp + 1;
1459
1460     if( h->sh.i_type == SLICE_TYPE_B )
1461         x264_macroblock_bipred_init( h );
1462
1463     /* ------------------------ Create slice header  ----------------------- */
1464     x264_slice_init( h, i_nal_type, i_global_qp );
1465
1466     if( i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE )
1467         h->i_frame_num++;
1468
1469     /* ---------------------- Write the bitstream -------------------------- */
1470     /* Init bitstream context */
1471     h->out.i_nal = 0;
1472     bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
1473
1474     if(h->param.b_aud){
1475         int pic_type;
1476
1477         if(h->sh.i_type == SLICE_TYPE_I)
1478             pic_type = 0;
1479         else if(h->sh.i_type == SLICE_TYPE_P)
1480             pic_type = 1;
1481         else if(h->sh.i_type == SLICE_TYPE_B)
1482             pic_type = 2;
1483         else
1484             pic_type = 7;
1485
1486         x264_nal_start(h, NAL_AUD, NAL_PRIORITY_DISPOSABLE);
1487         bs_write(&h->out.bs, 3, pic_type);
1488         bs_rbsp_trailing(&h->out.bs);
1489         x264_nal_end(h);
1490     }
1491
1492     h->i_nal_type = i_nal_type;
1493     h->i_nal_ref_idc = i_nal_ref_idc;
1494
1495     /* Write SPS and PPS */
1496     if( i_nal_type == NAL_SLICE_IDR && h->param.b_repeat_headers )
1497     {
1498         if( h->fenc->i_frame == 0 )
1499         {
1500             /* identify ourself */
1501             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
1502             x264_sei_version_write( h, &h->out.bs );
1503             x264_nal_end( h );
1504         }
1505
1506         /* generate sequence parameters */
1507         x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
1508         x264_sps_write( &h->out.bs, h->sps );
1509         x264_nal_end( h );
1510
1511         /* generate picture parameters */
1512         x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
1513         x264_pps_write( &h->out.bs, h->pps );
1514         x264_nal_end( h );
1515     }
1516
1517     /* Write frame */
1518     if( h->param.i_threads > 1 )
1519     {
1520         x264_pthread_create( &h->thread_handle, NULL, (void*)x264_slices_write, h );
1521         h->b_thread_active = 1;
1522     }
1523     else
1524         x264_slices_write( h );
1525
1526     /* restore CPU state (before using float again) */
1527     x264_emms();
1528
1529     if( h->sh.i_type == SLICE_TYPE_P && !h->param.rc.b_stat_read
1530         && h->param.i_scenecut_threshold >= 0
1531         && !h->param.b_pre_scenecut )
1532     {
1533         const int *mbs = h->stat.frame.i_mb_count;
1534         int i_mb_i = mbs[I_16x16] + mbs[I_8x8] + mbs[I_4x4];
1535         int i_mb_p = mbs[P_L0] + mbs[P_8x8];
1536         int i_mb_s = mbs[P_SKIP];
1537         int i_mb   = h->sps->i_mb_width * h->sps->i_mb_height;
1538         int64_t i_inter_cost = h->stat.frame.i_inter_cost;
1539         int64_t i_intra_cost = h->stat.frame.i_intra_cost;
1540
1541         float f_bias;
1542         int i_gop_size = h->fenc->i_frame - h->frames.i_last_idr;
1543         float f_thresh_max = h->param.i_scenecut_threshold / 100.0;
1544         /* magic numbers pulled out of thin air */
1545         float f_thresh_min = f_thresh_max * h->param.i_keyint_min
1546                              / ( h->param.i_keyint_max * 4 );
1547         if( h->param.i_keyint_min == h->param.i_keyint_max )
1548              f_thresh_min= f_thresh_max;
1549
1550         /* macroblock_analyse() doesn't further analyse skipped mbs,
1551          * so we have to guess their cost */
1552         if( h->stat.frame.i_mbs_analysed > 0 )
1553             i_intra_cost = i_intra_cost * i_mb / h->stat.frame.i_mbs_analysed;
1554
1555         if( i_gop_size < h->param.i_keyint_min / 4 )
1556             f_bias = f_thresh_min / 4;
1557         else if( i_gop_size <= h->param.i_keyint_min )
1558             f_bias = f_thresh_min * i_gop_size / h->param.i_keyint_min;
1559         else
1560         {
1561             f_bias = f_thresh_min
1562                      + ( f_thresh_max - f_thresh_min )
1563                        * ( i_gop_size - h->param.i_keyint_min )
1564                        / ( h->param.i_keyint_max - h->param.i_keyint_min );
1565         }
1566         f_bias = X264_MIN( f_bias, 1.0 );
1567
1568         /* Bad P will be reencoded as I */
1569         if( h->stat.frame.i_mbs_analysed > 0 &&
1570             i_inter_cost >= (1.0 - f_bias) * i_intra_cost )
1571         {
1572             int b;
1573
1574             x264_log( h, X264_LOG_DEBUG, "scene cut at %d Icost:%.0f Pcost:%.0f ratio:%.4f bias:%.4f gop:%d (imb:%d pmb:%d smb:%d)\n",
1575                       h->fenc->i_frame,
1576                       (double)i_intra_cost, (double)i_inter_cost,
1577                       1. - (double)i_inter_cost / i_intra_cost,
1578                       f_bias, i_gop_size,
1579                       i_mb_i, i_mb_p, i_mb_s );
1580
1581             /* Restore frame num */
1582             h->i_frame_num--;
1583
1584             for( b = 0; h->frames.current[b] && IS_X264_TYPE_B( h->frames.current[b]->i_type ); b++ );
1585             if( b > 0 )
1586             {
1587                 /* If using B-frames, force GOP to be closed.
1588                  * Even if this frame is going to be I and not IDR, forcing a
1589                  * P-frame before the scenecut will probably help compression.
1590                  *
1591                  * We don't yet know exactly which frame is the scene cut, so
1592                  * we can't assign an I-frame. Instead, change the previous
1593                  * B-frame to P, and rearrange coding order. */
1594
1595                 if( h->param.b_bframe_adaptive || b > 1 )
1596                     h->fenc->i_type = X264_TYPE_AUTO;
1597                 x264_frame_sort_pts( h->frames.current );
1598                 x264_frame_unshift( h->frames.next, h->fenc );
1599                 h->fenc = h->frames.current[b-1];
1600                 h->frames.current[b-1] = NULL;
1601                 h->fenc->i_type = X264_TYPE_P;
1602                 x264_frame_sort_dts( h->frames.current );
1603             }
1604             /* Do IDR if needed */
1605             else if( i_gop_size >= h->param.i_keyint_min )
1606             {
1607                 /* Reset */
1608                 h->i_frame_num = 0;
1609
1610                 /* Reinit field of fenc */
1611                 h->fenc->i_type = X264_TYPE_IDR;
1612                 h->fenc->i_poc = 0;
1613
1614                 /* Put enqueued frames back in the pool */
1615                 while( h->frames.current[0] )
1616                     x264_frame_push( h->frames.next, x264_frame_shift( h->frames.current ) );
1617                 x264_frame_sort_pts( h->frames.next );
1618             }
1619             else
1620             {
1621                 h->fenc->i_type = X264_TYPE_I;
1622             }
1623             goto do_encode;
1624         }
1625     }
1626
1627     x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
1628     return 0;
1629 }
1630
1631 static void x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
1632                                     x264_nal_t **pp_nal, int *pi_nal,
1633                                     x264_picture_t *pic_out )
1634 {
1635     int i, i_list;
1636     char psz_message[80];
1637
1638     if( h->b_thread_active )
1639     {
1640         x264_pthread_join( h->thread_handle, NULL );
1641         h->b_thread_active = 0;
1642     }
1643     if( !h->out.i_nal )
1644     {
1645         pic_out->i_type = X264_TYPE_AUTO;
1646         return;
1647     }
1648
1649     x264_frame_push_unused( thread_current, h->fenc );
1650
1651     /* End bitstream, set output  */
1652     *pi_nal = h->out.i_nal;
1653     *pp_nal = h->out.nal;
1654     h->out.i_nal = 0;
1655
1656     /* Set output picture properties */
1657     if( h->sh.i_type == SLICE_TYPE_I )
1658         pic_out->i_type = h->i_nal_type == NAL_SLICE_IDR ? X264_TYPE_IDR : X264_TYPE_I;
1659     else if( h->sh.i_type == SLICE_TYPE_P )
1660         pic_out->i_type = X264_TYPE_P;
1661     else
1662         pic_out->i_type = X264_TYPE_B;
1663     pic_out->i_pts = h->fenc->i_pts;
1664
1665     pic_out->img.i_plane = h->fdec->i_plane;
1666     for(i = 0; i < 4; i++){
1667         pic_out->img.i_stride[i] = h->fdec->i_stride[i];
1668         pic_out->img.plane[i] = h->fdec->plane[i];
1669     }
1670
1671     /* ---------------------- Update encoder state ------------------------- */
1672
1673     /* update rc */
1674     x264_emms();
1675     x264_ratecontrol_end( h, h->out.i_frame_size * 8 );
1676
1677     /* restore CPU state (before using float again) */
1678     x264_emms();
1679
1680     x264_noise_reduction_update( thread_current );
1681
1682     /* ---------------------- Compute/Print statistics --------------------- */
1683     x264_thread_sync_stat( h, h->thread[0] );
1684
1685     /* Slice stat */
1686     h->stat.i_slice_count[h->sh.i_type]++;
1687     h->stat.i_slice_size[h->sh.i_type] += h->out.i_frame_size + NALU_OVERHEAD;
1688     h->stat.f_slice_qp[h->sh.i_type] += h->fdec->f_qp_avg_aq;
1689
1690     for( i = 0; i < X264_MBTYPE_MAX; i++ )
1691         h->stat.i_mb_count[h->sh.i_type][i] += h->stat.frame.i_mb_count[i];
1692     for( i = 0; i < X264_PARTTYPE_MAX; i++ )
1693         h->stat.i_mb_partition[h->sh.i_type][i] += h->stat.frame.i_mb_partition[i];
1694     for( i = 0; i < 2; i++ )
1695         h->stat.i_mb_count_8x8dct[i] += h->stat.frame.i_mb_count_8x8dct[i];
1696     if( h->sh.i_type != SLICE_TYPE_I )
1697         for( i_list = 0; i_list < 2; i_list++ )
1698             for( i = 0; i < 32; i++ )
1699                 h->stat.i_mb_count_ref[h->sh.i_type][i_list][i] += h->stat.frame.i_mb_count_ref[i_list][i];
1700     if( h->sh.i_type == SLICE_TYPE_B )
1701     {
1702         h->stat.i_direct_frames[ h->sh.b_direct_spatial_mv_pred ] ++;
1703         if( h->mb.b_direct_auto_write )
1704         {
1705             //FIXME somewhat arbitrary time constants
1706             if( h->stat.i_direct_score[0] + h->stat.i_direct_score[1] > h->mb.i_mb_count )
1707             {
1708                 for( i = 0; i < 2; i++ )
1709                     h->stat.i_direct_score[i] = h->stat.i_direct_score[i] * 9/10;
1710             }
1711             for( i = 0; i < 2; i++ )
1712                 h->stat.i_direct_score[i] += h->stat.frame.i_direct_score[i];
1713         }
1714     }
1715
1716     psz_message[0] = '\0';
1717     if( h->param.analyse.b_psnr )
1718     {
1719         int64_t sqe[3] = {
1720             h->stat.frame.i_ssd[0],
1721             h->stat.frame.i_ssd[1],
1722             h->stat.frame.i_ssd[2],
1723         };
1724
1725         h->stat.i_sqe_global[h->sh.i_type] += sqe[0] + sqe[1] + sqe[2];
1726         h->stat.f_psnr_average[h->sh.i_type] += x264_psnr( sqe[0] + sqe[1] + sqe[2], 3 * h->param.i_width * h->param.i_height / 2 );
1727         h->stat.f_psnr_mean_y[h->sh.i_type] += x264_psnr( sqe[0], h->param.i_width * h->param.i_height );
1728         h->stat.f_psnr_mean_u[h->sh.i_type] += x264_psnr( sqe[1], h->param.i_width * h->param.i_height / 4 );
1729         h->stat.f_psnr_mean_v[h->sh.i_type] += x264_psnr( sqe[2], h->param.i_width * h->param.i_height / 4 );
1730
1731         snprintf( psz_message, 80, " PSNR Y:%5.2f U:%5.2f V:%5.2f",
1732                   x264_psnr( sqe[0], h->param.i_width * h->param.i_height ),
1733                   x264_psnr( sqe[1], h->param.i_width * h->param.i_height / 4),
1734                   x264_psnr( sqe[2], h->param.i_width * h->param.i_height / 4) );
1735     }
1736
1737     if( h->param.analyse.b_ssim )
1738     {
1739         double ssim_y = h->stat.frame.f_ssim
1740                       / (((h->param.i_width-6)>>2) * ((h->param.i_height-6)>>2));
1741         h->stat.f_ssim_mean_y[h->sh.i_type] += ssim_y;
1742         snprintf( psz_message + strlen(psz_message), 80 - strlen(psz_message),
1743                   " SSIM Y:%.5f", ssim_y );
1744     }
1745     psz_message[79] = '\0';
1746
1747     x264_log( h, X264_LOG_DEBUG,
1748                   "frame=%4d QP=%.2f NAL=%d Slice:%c Poc:%-3d I:%-4d P:%-4d SKIP:%-4d size=%d bytes%s\n",
1749               h->i_frame,
1750               h->fdec->f_qp_avg_aq,
1751               h->i_nal_ref_idc,
1752               h->sh.i_type == SLICE_TYPE_I ? 'I' : (h->sh.i_type == SLICE_TYPE_P ? 'P' : 'B' ),
1753               h->fdec->i_poc,
1754               h->stat.frame.i_mb_count_i,
1755               h->stat.frame.i_mb_count_p,
1756               h->stat.frame.i_mb_count_skip,
1757               h->out.i_frame_size,
1758               psz_message );
1759
1760     // keep stats all in one place
1761     x264_thread_sync_stat( h->thread[0], h );
1762     // for the use of the next frame
1763     x264_thread_sync_stat( thread_current, h );
1764
1765 #ifdef DEBUG_MB_TYPE
1766 {
1767     static const char mb_chars[] = { 'i', 'i', 'I', 'C', 'P', '8', 'S',
1768         'D', '<', 'X', 'B', 'X', '>', 'B', 'B', 'B', 'B', '8', 'S' };
1769     int mb_xy;
1770     for( mb_xy = 0; mb_xy < h->sps->i_mb_width * h->sps->i_mb_height; mb_xy++ )
1771     {
1772         if( h->mb.type[mb_xy] < X264_MBTYPE_MAX && h->mb.type[mb_xy] >= 0 )
1773             fprintf( stderr, "%c ", mb_chars[ h->mb.type[mb_xy] ] );
1774         else
1775             fprintf( stderr, "? " );
1776
1777         if( (mb_xy+1) % h->sps->i_mb_width == 0 )
1778             fprintf( stderr, "\n" );
1779     }
1780 }
1781 #endif
1782
1783     if( h->param.psz_dump_yuv )
1784         x264_frame_dump( h );
1785 }
1786
1787 static void x264_print_intra( int64_t *i_mb_count, double i_count, int b_print_pcm, char *intra )
1788 {
1789     intra += sprintf( intra, "I16..4%s: %4.1f%% %4.1f%% %4.1f%%",
1790         b_print_pcm ? "..PCM" : "",
1791         i_mb_count[I_16x16]/ i_count,
1792         i_mb_count[I_8x8]  / i_count,
1793         i_mb_count[I_4x4]  / i_count );
1794     if( b_print_pcm )
1795         sprintf( intra, " %4.1f%%", i_mb_count[I_PCM]  / i_count );
1796 }
1797
1798 /****************************************************************************
1799  * x264_encoder_close:
1800  ****************************************************************************/
1801 void    x264_encoder_close  ( x264_t *h )
1802 {
1803     int64_t i_yuv_size = 3 * h->param.i_width * h->param.i_height / 2;
1804     int64_t i_mb_count_size[2][7] = {{0}};
1805     char intra[40];
1806     int i, j, i_list, i_type;
1807     int b_print_pcm = h->stat.i_mb_count[SLICE_TYPE_I][I_PCM]
1808                    || h->stat.i_mb_count[SLICE_TYPE_P][I_PCM]
1809                    || h->stat.i_mb_count[SLICE_TYPE_B][I_PCM];
1810
1811     for( i=0; i<h->param.i_threads; i++ )
1812     {
1813         // don't strictly have to wait for the other threads, but it's simpler than canceling them
1814         if( h->thread[i]->b_thread_active )
1815         {
1816             x264_pthread_join( h->thread[i]->thread_handle, NULL );
1817             assert( h->thread[i]->fenc->i_reference_count == 1 );
1818             x264_frame_delete( h->thread[i]->fenc );
1819         }
1820     }
1821
1822     /* Slices used and PSNR */
1823     for( i=0; i<5; i++ )
1824     {
1825         static const int slice_order[] = { SLICE_TYPE_I, SLICE_TYPE_SI, SLICE_TYPE_P, SLICE_TYPE_SP, SLICE_TYPE_B };
1826         static const char *slice_name[] = { "P", "B", "I", "SP", "SI" };
1827         int i_slice = slice_order[i];
1828
1829         if( h->stat.i_slice_count[i_slice] > 0 )
1830         {
1831             const int i_count = h->stat.i_slice_count[i_slice];
1832             if( h->param.analyse.b_psnr )
1833             {
1834                 x264_log( h, X264_LOG_INFO,
1835                           "slice %s:%-5d Avg QP:%5.2f  size:%6.0f  PSNR Mean Y:%5.2f U:%5.2f V:%5.2f Avg:%5.2f Global:%5.2f\n",
1836                           slice_name[i_slice],
1837                           i_count,
1838                           h->stat.f_slice_qp[i_slice] / i_count,
1839                           (double)h->stat.i_slice_size[i_slice] / i_count,
1840                           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,
1841                           h->stat.f_psnr_average[i_slice] / i_count,
1842                           x264_psnr( h->stat.i_sqe_global[i_slice], i_count * i_yuv_size ) );
1843             }
1844             else
1845             {
1846                 x264_log( h, X264_LOG_INFO,
1847                           "slice %s:%-5d Avg QP:%5.2f  size:%6.0f\n",
1848                           slice_name[i_slice],
1849                           i_count,
1850                           h->stat.f_slice_qp[i_slice] / i_count,
1851                           (double)h->stat.i_slice_size[i_slice] / i_count );
1852             }
1853         }
1854     }
1855
1856     for( i_type = 0; i_type < 2; i_type++ )
1857         for( i = 0; i < X264_PARTTYPE_MAX; i++ )
1858         {
1859             if( i == D_DIRECT_8x8 ) continue; /* direct is counted as its own type */
1860             i_mb_count_size[i_type][x264_mb_partition_pixel_table[i]] += h->stat.i_mb_partition[i_type][i];
1861         }
1862
1863     /* MB types used */
1864     if( h->stat.i_slice_count[SLICE_TYPE_I] > 0 )
1865     {
1866         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_I];
1867         double i_count = h->stat.i_slice_count[SLICE_TYPE_I] * h->mb.i_mb_count / 100.0;
1868         x264_print_intra( i_mb_count, i_count, b_print_pcm, intra );
1869         x264_log( h, X264_LOG_INFO, "mb I  %s\n", intra );
1870     }
1871     if( h->stat.i_slice_count[SLICE_TYPE_P] > 0 )
1872     {
1873         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_P];
1874         double i_count = h->stat.i_slice_count[SLICE_TYPE_P] * h->mb.i_mb_count / 100.0;
1875         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_P];
1876         x264_print_intra( i_mb_count, i_count, b_print_pcm, intra );
1877         x264_log( h, X264_LOG_INFO,
1878                   "mb P  %s  P16..4: %4.1f%% %4.1f%% %4.1f%% %4.1f%% %4.1f%%    skip:%4.1f%%\n",
1879                   intra,
1880                   i_mb_size[PIXEL_16x16] / (i_count*4),
1881                   (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
1882                   i_mb_size[PIXEL_8x8] / (i_count*4),
1883                   (i_mb_size[PIXEL_8x4] + i_mb_size[PIXEL_4x8]) / (i_count*4),
1884                   i_mb_size[PIXEL_4x4] / (i_count*4),
1885                   i_mb_count[P_SKIP] / i_count );
1886     }
1887     if( h->stat.i_slice_count[SLICE_TYPE_B] > 0 )
1888     {
1889         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_B];
1890         double i_count = h->stat.i_slice_count[SLICE_TYPE_B] * h->mb.i_mb_count / 100.0;
1891         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_B];
1892         int64_t list_count[3] = {0}; /* 0 == L0, 1 == L1, 2 == BI */
1893         x264_print_intra( i_mb_count, i_count, b_print_pcm, intra );
1894         for( i = 0; i < X264_PARTTYPE_MAX; i++ )
1895             for( j = 0; j < 2; j++ )
1896             {
1897                 int l0 = x264_mb_type_list0_table[i][j];
1898                 int l1 = x264_mb_type_list1_table[i][j];
1899                 if( l0 || l1 )
1900                     list_count[l1+l0*l1] += h->stat.i_mb_count[SLICE_TYPE_B][i] * 2;
1901             }
1902         list_count[0] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L0_8x8];
1903         list_count[1] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L1_8x8];
1904         list_count[2] += h->stat.i_mb_partition[SLICE_TYPE_B][D_BI_8x8];
1905         i_mb_count[B_DIRECT] += (h->stat.i_mb_partition[SLICE_TYPE_B][D_DIRECT_8x8]+2)/4;
1906         const double i_mb_list_count = (list_count[0] + list_count[1] + list_count[2]) / 100.0;
1907         x264_log( h, X264_LOG_INFO,
1908                   "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",
1909                   intra,
1910                   i_mb_size[PIXEL_16x16] / (i_count*4),
1911                   (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
1912                   i_mb_size[PIXEL_8x8] / (i_count*4),
1913                   i_mb_count[B_DIRECT] / i_count,
1914                   i_mb_count[B_SKIP]   / i_count,
1915                   list_count[0] / i_mb_list_count,
1916                   list_count[1] / i_mb_list_count,
1917                   list_count[2] / i_mb_list_count );
1918     }
1919
1920     x264_ratecontrol_summary( h );
1921
1922     if( h->stat.i_slice_count[SLICE_TYPE_I] + h->stat.i_slice_count[SLICE_TYPE_P] + h->stat.i_slice_count[SLICE_TYPE_B] > 0 )
1923     {
1924         const int i_count = h->stat.i_slice_count[SLICE_TYPE_I] +
1925                             h->stat.i_slice_count[SLICE_TYPE_P] +
1926                             h->stat.i_slice_count[SLICE_TYPE_B];
1927         float fps = (float) h->param.i_fps_num / h->param.i_fps_den;
1928 #define SUM3(p) (p[SLICE_TYPE_I] + p[SLICE_TYPE_P] + p[SLICE_TYPE_B])
1929 #define SUM3b(p,o) (p[SLICE_TYPE_I][o] + p[SLICE_TYPE_P][o] + p[SLICE_TYPE_B][o])
1930         float f_bitrate = fps * SUM3(h->stat.i_slice_size) / i_count / 125;
1931
1932         if( h->pps->b_transform_8x8_mode )
1933         {
1934             int64_t i_i8x8 = SUM3b( h->stat.i_mb_count, I_8x8 );
1935             int64_t i_intra = i_i8x8 + SUM3b( h->stat.i_mb_count, I_4x4 )
1936                                      + SUM3b( h->stat.i_mb_count, I_16x16 );
1937             x264_log( h, X264_LOG_INFO, "8x8 transform  intra:%.1f%%  inter:%.1f%%\n",
1938                       100. * i_i8x8 / i_intra,
1939                       100. * h->stat.i_mb_count_8x8dct[1] / h->stat.i_mb_count_8x8dct[0] );
1940         }
1941
1942         if( h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO
1943             && h->stat.i_slice_count[SLICE_TYPE_B] )
1944         {
1945             x264_log( h, X264_LOG_INFO, "direct mvs  spatial:%.1f%%  temporal:%.1f%%\n",
1946                       h->stat.i_direct_frames[1] * 100. / h->stat.i_slice_count[SLICE_TYPE_B],
1947                       h->stat.i_direct_frames[0] * 100. / h->stat.i_slice_count[SLICE_TYPE_B] );
1948         }
1949
1950         for( i_list = 0; i_list < 2; i_list++ )
1951         {
1952             int i_slice;
1953             for( i_slice = 0; i_slice < 2; i_slice++ )
1954             {
1955                 char buf[200];
1956                 char *p = buf;
1957                 int64_t i_den = 0;
1958                 int i_max = 0;
1959                 for( i = 0; i < 32; i++ )
1960                     if( h->stat.i_mb_count_ref[i_slice][i_list][i] )
1961                     {
1962                         i_den += h->stat.i_mb_count_ref[i_slice][i_list][i];
1963                         i_max = i;
1964                     }
1965                 if( i_max == 0 )
1966                     continue;
1967                 for( i = 0; i <= i_max; i++ )
1968                     p += sprintf( p, " %4.1f%%", 100. * h->stat.i_mb_count_ref[i_slice][i_list][i] / i_den );
1969                 x264_log( h, X264_LOG_INFO, "ref %c L%d %s\n", "PB"[i_slice], i_list, buf );
1970             }
1971         }
1972
1973         if( h->param.analyse.b_ssim )
1974         {
1975             x264_log( h, X264_LOG_INFO,
1976                       "SSIM Mean Y:%.7f\n",
1977                       SUM3( h->stat.f_ssim_mean_y ) / i_count );
1978         }
1979         if( h->param.analyse.b_psnr )
1980         {
1981             x264_log( h, X264_LOG_INFO,
1982                       "PSNR Mean Y:%6.3f U:%6.3f V:%6.3f Avg:%6.3f Global:%6.3f kb/s:%.2f\n",
1983                       SUM3( h->stat.f_psnr_mean_y ) / i_count,
1984                       SUM3( h->stat.f_psnr_mean_u ) / i_count,
1985                       SUM3( h->stat.f_psnr_mean_v ) / i_count,
1986                       SUM3( h->stat.f_psnr_average ) / i_count,
1987                       x264_psnr( SUM3( h->stat.i_sqe_global ), i_count * i_yuv_size ),
1988                       f_bitrate );
1989         }
1990         else
1991             x264_log( h, X264_LOG_INFO, "kb/s:%.1f\n", f_bitrate );
1992     }
1993
1994     /* rc */
1995     x264_ratecontrol_delete( h );
1996
1997     /* param */
1998     if( h->param.rc.psz_stat_out )
1999         free( h->param.rc.psz_stat_out );
2000     if( h->param.rc.psz_stat_in )
2001         free( h->param.rc.psz_stat_in );
2002     if( h->param.rc.psz_rc_eq )
2003         free( h->param.rc.psz_rc_eq );
2004
2005     x264_cqm_delete( h );
2006
2007     if( h->param.i_threads > 1)
2008         h = h->thread[ h->i_thread_phase % h->param.i_threads ];
2009
2010     /* frames */
2011     for( i = 0; h->frames.current[i]; i++ )
2012     {
2013         assert( h->frames.current[i]->i_reference_count == 1 );
2014         x264_frame_delete( h->frames.current[i] );
2015     }
2016     for( i = 0; h->frames.next[i]; i++ )
2017     {
2018         assert( h->frames.next[i]->i_reference_count == 1 );
2019         x264_frame_delete( h->frames.next[i] );
2020     }
2021     for( i = 0; h->frames.unused[i]; i++ )
2022     {
2023         assert( h->frames.unused[i]->i_reference_count == 0 );
2024         x264_frame_delete( h->frames.unused[i] );
2025     }
2026
2027     h = h->thread[0];
2028
2029     for( i = h->param.i_threads - 1; i >= 0; i-- )
2030     {
2031         x264_frame_t **frame;
2032
2033         for( frame = h->thread[i]->frames.reference; *frame; frame++ )
2034         {
2035             assert( (*frame)->i_reference_count > 0 );
2036             (*frame)->i_reference_count--;
2037             if( (*frame)->i_reference_count == 0 )
2038                 x264_frame_delete( *frame );
2039         }
2040         frame = &h->thread[i]->fdec;
2041         assert( (*frame)->i_reference_count > 0 );
2042         (*frame)->i_reference_count--;
2043         if( (*frame)->i_reference_count == 0 )
2044             x264_frame_delete( *frame );
2045
2046         x264_macroblock_cache_end( h->thread[i] );
2047         x264_free( h->thread[i]->out.p_bitstream );
2048         x264_free( h->thread[i] );
2049     }
2050 }