git.sesse.net Git - x264/blob - encoder/encoder.c

   1 /*****************************************************************************
   2  * x264: h264 encoder
   3  *****************************************************************************
   4  * Copyright (C) 2003-2008 x264 project
   5  *
   6  * Authors: Laurent Aimar <fenrir@via.ecp.fr>
   7  *          Loren Merritt <lorenm@u.washington.edu>
   8  *          Fiona Glaser <fiona@x264.com>
   9  *
  10  * This program is free software; you can redistribute it and/or modify
  11  * it under the terms of the GNU General Public License as published by
  12  * the Free Software Foundation; either version 2 of the License, or
  13  * (at your option) any later version.
  14  *
  15  * This program is distributed in the hope that it will be useful,
  16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  18  * GNU General Public License for more details.
  19  *
  20  * You should have received a copy of the GNU General Public License
  21  * along with this program; if not, write to the Free Software
  22  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02111, USA.
  23  *****************************************************************************/
  24
  25 #include <math.h>
  26
  27 #include "common/common.h"
  28 #include "common/cpu.h"
  29
  30 #include "set.h"
  31 #include "analyse.h"
  32 #include "ratecontrol.h"
  33 #include "macroblock.h"
  34
  35 #if VISUALIZE
  36 #include "common/visualize.h"
  37 #endif
  38
  39 //#define DEBUG_MB_TYPE
  40
  41 #define NALU_OVERHEAD 5 // startcode + NAL type costs 5 bytes per frame
  42
  43 #define bs_write_ue bs_write_ue_big
  44
  45 static void x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
  46                                     x264_nal_t **pp_nal, int *pi_nal,
  47                                     x264_picture_t *pic_out );
  48
  49 /****************************************************************************
  50  *
  51  ******************************* x264 libs **********************************
  52  *
  53  ****************************************************************************/
  54 static float x264_psnr( int64_t i_sqe, int64_t i_size )
  55 {
  56     double f_mse = (double)i_sqe / ((double)65025.0 * (double)i_size);
  57     if( f_mse <= 0.0000000001 ) /* Max 100dB */
  58         return 100;
  59
  60     return (float)(-10.0 * log( f_mse ) / log( 10.0 ));
  61 }
  62
  63 static void x264_frame_dump( x264_t *h )
  64 {
  65     FILE *f = fopen( h->param.psz_dump_yuv, "r+b" );
  66     int i, y;
  67     if( !f )
  68         return;
  69     /* Write the frame in display order */
  70     fseek( f, h->fdec->i_frame * h->param.i_height * h->param.i_width * 3/2, SEEK_SET );
  71     for( i = 0; i < h->fdec->i_plane; i++ )
  72         for( y = 0; y < h->param.i_height >> !!i; y++ )
  73             fwrite( &h->fdec->plane[i][y*h->fdec->i_stride[i]], 1, h->param.i_width >> !!i, f );
  74     fclose( f );
  75 }
  76
  77
  78 /* Fill "default" values */
  79 static void x264_slice_header_init( x264_t *h, x264_slice_header_t *sh,
  80                                     x264_sps_t *sps, x264_pps_t *pps,
  81                                     int i_idr_pic_id, int i_frame, int i_qp )
  82 {
  83     x264_param_t *param = &h->param;
  84     int i;
  85
  86     /* First we fill all field */
  87     sh->sps = sps;
  88     sh->pps = pps;
  89
  90     sh->i_first_mb  = 0;
  91     sh->i_last_mb   = h->sps->i_mb_width * h->sps->i_mb_height;
  92     sh->i_pps_id    = pps->i_id;
  93
  94     sh->i_frame_num = i_frame;
  95
  96     sh->b_mbaff = h->param.b_interlaced;
  97     sh->b_field_pic = 0;    /* no field support for now */
  98     sh->b_bottom_field = 0; /* not yet used */
  99
 100     sh->i_idr_pic_id = i_idr_pic_id;
 101
 102     /* poc stuff, fixed later */
 103     sh->i_poc_lsb = 0;
 104     sh->i_delta_poc_bottom = 0;
 105     sh->i_delta_poc[0] = 0;
 106     sh->i_delta_poc[1] = 0;
 107
 108     sh->i_redundant_pic_cnt = 0;
 109
 110     if( !h->mb.b_direct_auto_read )
 111     {
 112         if( h->mb.b_direct_auto_write )
 113             sh->b_direct_spatial_mv_pred = ( h->stat.i_direct_score[1] > h->stat.i_direct_score[0] );
 114         else
 115             sh->b_direct_spatial_mv_pred = ( param->analyse.i_direct_mv_pred == X264_DIRECT_PRED_SPATIAL );
 116     }
 117     /* else b_direct_spatial_mv_pred was read from the 2pass statsfile */
 118
 119     sh->b_num_ref_idx_override = 0;
 120     sh->i_num_ref_idx_l0_active = 1;
 121     sh->i_num_ref_idx_l1_active = 1;
 122
 123     sh->b_ref_pic_list_reordering_l0 = h->b_ref_reorder[0];
 124     sh->b_ref_pic_list_reordering_l1 = h->b_ref_reorder[1];
 125
 126     /* If the ref list isn't in the default order, construct reordering header */
 127     /* List1 reordering isn't needed yet */
 128     if( sh->b_ref_pic_list_reordering_l0 )
 129     {
 130         int pred_frame_num = i_frame;
 131         for( i = 0; i < h->i_ref0; i++ )
 132         {
 133             int diff = h->fref0[i]->i_frame_num - pred_frame_num;
 134             if( diff == 0 )
 135                 x264_log( h, X264_LOG_ERROR, "diff frame num == 0\n" );
 136             sh->ref_pic_list_order[0][i].idc = ( diff > 0 );
 137             sh->ref_pic_list_order[0][i].arg = abs( diff ) - 1;
 138             pred_frame_num = h->fref0[i]->i_frame_num;
 139         }
 140     }
 141
 142     sh->i_cabac_init_idc = param->i_cabac_init_idc;
 143
 144     sh->i_qp = i_qp;
 145     sh->i_qp_delta = i_qp - pps->i_pic_init_qp;
 146     sh->b_sp_for_swidth = 0;
 147     sh->i_qs_delta = 0;
 148
 149     /* If effective qp <= 15, deblocking would have no effect anyway */
 150     if( param->b_deblocking_filter
 151         && ( h->mb.b_variable_qp
 152         || 15 < i_qp + 2 * X264_MIN(param->i_deblocking_filter_alphac0, param->i_deblocking_filter_beta) ) )
 153     {
 154         sh->i_disable_deblocking_filter_idc = 0;
 155     }
 156     else
 157     {
 158         sh->i_disable_deblocking_filter_idc = 1;
 159     }
 160     sh->i_alpha_c0_offset = param->i_deblocking_filter_alphac0 << 1;
 161     sh->i_beta_offset = param->i_deblocking_filter_beta << 1;
 162 }
 163
 164 static void x264_slice_header_write( bs_t *s, x264_slice_header_t *sh, int i_nal_ref_idc )
 165 {
 166     int i;
 167
 168     if( sh->b_mbaff )
 169     {
 170         assert( sh->i_first_mb % (2*sh->sps->i_mb_width) == 0 );
 171         bs_write_ue( s, sh->i_first_mb >> 1 );
 172     }
 173     else
 174         bs_write_ue( s, sh->i_first_mb );
 175
 176     bs_write_ue( s, sh->i_type + 5 );   /* same type things */
 177     bs_write_ue( s, sh->i_pps_id );
 178     bs_write( s, sh->sps->i_log2_max_frame_num, sh->i_frame_num );
 179
 180     if( !sh->sps->b_frame_mbs_only )
 181     {
 182         bs_write1( s, sh->b_field_pic );
 183         if ( sh->b_field_pic )
 184             bs_write1( s, sh->b_bottom_field );
 185     }
 186
 187     if( sh->i_idr_pic_id >= 0 ) /* NAL IDR */
 188     {
 189         bs_write_ue( s, sh->i_idr_pic_id );
 190     }
 191
 192     if( sh->sps->i_poc_type == 0 )
 193     {
 194         bs_write( s, sh->sps->i_log2_max_poc_lsb, sh->i_poc_lsb );
 195         if( sh->pps->b_pic_order && !sh->b_field_pic )
 196         {
 197             bs_write_se( s, sh->i_delta_poc_bottom );
 198         }
 199     }
 200     else if( sh->sps->i_poc_type == 1 && !sh->sps->b_delta_pic_order_always_zero )
 201     {
 202         bs_write_se( s, sh->i_delta_poc[0] );
 203         if( sh->pps->b_pic_order && !sh->b_field_pic )
 204         {
 205             bs_write_se( s, sh->i_delta_poc[1] );
 206         }
 207     }
 208
 209     if( sh->pps->b_redundant_pic_cnt )
 210     {
 211         bs_write_ue( s, sh->i_redundant_pic_cnt );
 212     }
 213
 214     if( sh->i_type == SLICE_TYPE_B )
 215     {
 216         bs_write1( s, sh->b_direct_spatial_mv_pred );
 217     }
 218     if( sh->i_type == SLICE_TYPE_P || sh->i_type == SLICE_TYPE_SP || sh->i_type == SLICE_TYPE_B )
 219     {
 220         bs_write1( s, sh->b_num_ref_idx_override );
 221         if( sh->b_num_ref_idx_override )
 222         {
 223             bs_write_ue( s, sh->i_num_ref_idx_l0_active - 1 );
 224             if( sh->i_type == SLICE_TYPE_B )
 225             {
 226                 bs_write_ue( s, sh->i_num_ref_idx_l1_active - 1 );
 227             }
 228         }
 229     }
 230
 231     /* ref pic list reordering */
 232     if( sh->i_type != SLICE_TYPE_I )
 233     {
 234         bs_write1( s, sh->b_ref_pic_list_reordering_l0 );
 235         if( sh->b_ref_pic_list_reordering_l0 )
 236         {
 237             for( i = 0; i < sh->i_num_ref_idx_l0_active; i++ )
 238             {
 239                 bs_write_ue( s, sh->ref_pic_list_order[0][i].idc );
 240                 bs_write_ue( s, sh->ref_pic_list_order[0][i].arg );
 241
 242             }
 243             bs_write_ue( s, 3 );
 244         }
 245     }
 246     if( sh->i_type == SLICE_TYPE_B )
 247     {
 248         bs_write1( s, sh->b_ref_pic_list_reordering_l1 );
 249         if( sh->b_ref_pic_list_reordering_l1 )
 250         {
 251             for( i = 0; i < sh->i_num_ref_idx_l1_active; i++ )
 252             {
 253                 bs_write_ue( s, sh->ref_pic_list_order[1][i].idc );
 254                 bs_write_ue( s, sh->ref_pic_list_order[1][i].arg );
 255             }
 256             bs_write_ue( s, 3 );
 257         }
 258     }
 259
 260     if( ( sh->pps->b_weighted_pred && ( sh->i_type == SLICE_TYPE_P || sh->i_type == SLICE_TYPE_SP ) ) ||
 261         ( sh->pps->b_weighted_bipred == 1 && sh->i_type == SLICE_TYPE_B ) )
 262     {
 263         /* FIXME */
 264     }
 265
 266     if( i_nal_ref_idc != 0 )
 267     {
 268         if( sh->i_idr_pic_id >= 0 )
 269         {
 270             bs_write1( s, 0 );  /* no output of prior pics flag */
 271             bs_write1( s, 0 );  /* long term reference flag */
 272         }
 273         else
 274         {
 275             bs_write1( s, 0 );  /* adaptive_ref_pic_marking_mode_flag */
 276         }
 277     }
 278
 279     if( sh->pps->b_cabac && sh->i_type != SLICE_TYPE_I )
 280     {
 281         bs_write_ue( s, sh->i_cabac_init_idc );
 282     }
 283     bs_write_se( s, sh->i_qp_delta );      /* slice qp delta */
 284
 285     if( sh->pps->b_deblocking_filter_control )
 286     {
 287         bs_write_ue( s, sh->i_disable_deblocking_filter_idc );
 288         if( sh->i_disable_deblocking_filter_idc != 1 )
 289         {
 290             bs_write_se( s, sh->i_alpha_c0_offset >> 1 );
 291             bs_write_se( s, sh->i_beta_offset >> 1 );
 292         }
 293     }
 294 }
 295
 296 /* If we are within a reasonable distance of the end of the memory allocated for the bitstream, */
 297 /* reallocate, adding an arbitrary amount of space (100 kilobytes). */
 298 static void x264_bitstream_check_buffer( x264_t *h )
 299 {
 300     if( ( h->param.b_cabac && (h->cabac.p_end - h->cabac.p < 2500) )
 301      || ( h->out.bs.p_end - h->out.bs.p < 2500 ) )
 302     {
 303         uint8_t *bs_bak = h->out.p_bitstream;
 304         intptr_t delta;
 305         int i;
 306
 307         h->out.i_bitstream += 100000;
 308         h->out.p_bitstream = x264_realloc( h->out.p_bitstream, h->out.i_bitstream );
 309         delta = h->out.p_bitstream - bs_bak;
 310
 311         h->out.bs.p_start += delta;
 312         h->out.bs.p += delta;
 313         h->out.bs.p_end = h->out.p_bitstream + h->out.i_bitstream;
 314
 315         h->cabac.p_start += delta;
 316         h->cabac.p += delta;
 317         h->cabac.p_end = h->out.p_bitstream + h->out.i_bitstream;
 318
 319         for( i = 0; i <= h->out.i_nal; i++ )
 320             h->out.nal[i].p_payload += delta;
 321     }
 322 }
 323
 324 /****************************************************************************
 325  *
 326  ****************************************************************************
 327  ****************************** External API*********************************
 328  ****************************************************************************
 329  *
 330  ****************************************************************************/
 331
 332 static int x264_validate_parameters( x264_t *h )
 333 {
 334 #ifdef HAVE_MMX
 335     if( !(x264_cpu_detect() & X264_CPU_MMXEXT) )
 336     {
 337         x264_log( h, X264_LOG_ERROR, "your cpu does not support MMXEXT, but x264 was compiled with asm support\n");
 338         x264_log( h, X264_LOG_ERROR, "to run x264, recompile without asm support (configure --disable-asm)\n");
 339         return -1;
 340     }
 341 #endif
 342     if( h->param.i_width <= 0 || h->param.i_height <= 0 )
 343     {
 344         x264_log( h, X264_LOG_ERROR, "invalid width x height (%dx%d)\n",
 345                   h->param.i_width, h->param.i_height );
 346         return -1;
 347     }
 348
 349     if( h->param.i_width % 2 || h->param.i_height % 2 )
 350     {
 351         x264_log( h, X264_LOG_ERROR, "width or height not divisible by 2 (%dx%d)\n",
 352                   h->param.i_width, h->param.i_height );
 353         return -1;
 354     }
 355     if( h->param.i_csp != X264_CSP_I420 )
 356     {
 357         x264_log( h, X264_LOG_ERROR, "invalid CSP (only I420 supported)\n" );
 358         return -1;
 359     }
 360
 361     if( h->param.i_threads == 0 )
 362         h->param.i_threads = x264_cpu_num_processors() * 3/2;
 363     h->param.i_threads = x264_clip3( h->param.i_threads, 1, X264_THREAD_MAX );
 364     if( h->param.i_threads > 1 )
 365     {
 366 #ifndef HAVE_PTHREAD
 367         x264_log( h, X264_LOG_WARNING, "not compiled with pthread support!\n");
 368         h->param.i_threads = 1;
 369 #endif
 370     }
 371
 372     if( h->param.b_interlaced )
 373     {
 374         if( h->param.analyse.i_me_method >= X264_ME_ESA )
 375         {
 376             x264_log( h, X264_LOG_WARNING, "interlace + me=esa is not implemented\n" );
 377             h->param.analyse.i_me_method = X264_ME_UMH;
 378         }
 379         if( h->param.analyse.i_direct_mv_pred > X264_DIRECT_PRED_SPATIAL )
 380         {
 381             x264_log( h, X264_LOG_WARNING, "interlace + direct=temporal is not implemented\n" );
 382             h->param.analyse.i_direct_mv_pred = X264_DIRECT_PRED_SPATIAL;
 383         }
 384     }
 385
 386     if( h->param.rc.i_rc_method < 0 || h->param.rc.i_rc_method > 2 )
 387     {
 388         x264_log( h, X264_LOG_ERROR, "no ratecontrol method specified\n" );
 389         return -1;
 390     }
 391     h->param.rc.f_rf_constant = x264_clip3f( h->param.rc.f_rf_constant, 0, 51 );
 392     h->param.rc.i_qp_constant = x264_clip3( h->param.rc.i_qp_constant, 0, 51 );
 393     if( h->param.rc.i_rc_method == X264_RC_CRF )
 394         h->param.rc.i_qp_constant = h->param.rc.f_rf_constant;
 395     if( (h->param.rc.i_rc_method == X264_RC_CQP || h->param.rc.i_rc_method == X264_RC_CRF)
 396         && h->param.rc.i_qp_constant == 0 )
 397     {
 398         h->mb.b_lossless = 1;
 399         h->param.i_cqm_preset = X264_CQM_FLAT;
 400         h->param.psz_cqm_file = NULL;
 401         h->param.rc.i_rc_method = X264_RC_CQP;
 402         h->param.rc.f_ip_factor = 1;
 403         h->param.rc.f_pb_factor = 1;
 404         h->param.analyse.b_psnr = 0;
 405         h->param.analyse.b_ssim = 0;
 406         h->param.analyse.i_chroma_qp_offset = 0;
 407         h->param.analyse.i_trellis = 0;
 408         h->param.analyse.b_fast_pskip = 0;
 409         h->param.analyse.i_noise_reduction = 0;
 410         h->param.analyse.f_psy_rd = 0;
 411         h->param.i_bframe = 0;
 412         /* 8x8dct is not useful at all in CAVLC lossless */
 413         if( !h->param.b_cabac )
 414             h->param.analyse.b_transform_8x8 = 0;
 415     }
 416     if( h->param.rc.i_rc_method == X264_RC_CQP )
 417     {
 418         float qp_p = h->param.rc.i_qp_constant;
 419         float qp_i = qp_p - 6*log(h->param.rc.f_ip_factor)/log(2);
 420         float qp_b = qp_p + 6*log(h->param.rc.f_pb_factor)/log(2);
 421         h->param.rc.i_qp_min = x264_clip3( (int)(X264_MIN3( qp_p, qp_i, qp_b )), 0, 51 );
 422         h->param.rc.i_qp_max = x264_clip3( (int)(X264_MAX3( qp_p, qp_i, qp_b ) + .999), 0, 51 );
 423         h->param.rc.i_aq_mode = 0;
 424     }
 425     h->param.rc.i_qp_max = x264_clip3( h->param.rc.i_qp_max, 0, 51 );
 426     h->param.rc.i_qp_min = x264_clip3( h->param.rc.i_qp_min, 0, h->param.rc.i_qp_max );
 427
 428     if( ( h->param.i_width % 16 || h->param.i_height % 16 )
 429         && h->param.i_height != 1080 && !h->mb.b_lossless )
 430     {
 431         // There's nothing special about 1080 in that the warning still applies to it,
 432         // but chances are the user can't help it if his content is already 1080p,
 433         // so there's no point in warning in that case.
 434         x264_log( h, X264_LOG_WARNING,
 435                   "width or height not divisible by 16 (%dx%d), compression will suffer.\n",
 436                   h->param.i_width, h->param.i_height );
 437     }
 438
 439     h->param.i_frame_reference = x264_clip3( h->param.i_frame_reference, 1, 16 );
 440     if( h->param.i_keyint_max <= 0 )
 441         h->param.i_keyint_max = 1;
 442     if( h->param.i_scenecut_threshold < 0 )
 443         h->param.i_scenecut_threshold = 0;
 444     h->param.i_keyint_min = x264_clip3( h->param.i_keyint_min, 1, h->param.i_keyint_max/2+1 );
 445     if( !h->param.analyse.i_subpel_refine && h->param.analyse.i_direct_mv_pred > X264_DIRECT_PRED_SPATIAL )
 446     {
 447         x264_log( h, X264_LOG_WARNING, "subme=0 + direct=temporal is not supported\n" );
 448         h->param.analyse.i_direct_mv_pred = X264_DIRECT_PRED_SPATIAL;
 449     }
 450     h->param.i_bframe = x264_clip3( h->param.i_bframe, 0, X264_BFRAME_MAX );
 451     h->param.i_bframe_bias = x264_clip3( h->param.i_bframe_bias, -90, 100 );
 452     h->param.b_bframe_pyramid = h->param.b_bframe_pyramid && h->param.i_bframe > 1;
 453     if( !h->param.i_bframe )
 454         h->param.i_bframe_adaptive = X264_B_ADAPT_NONE;
 455     h->param.analyse.b_weighted_bipred = h->param.analyse.b_weighted_bipred && h->param.i_bframe > 0;
 456     h->mb.b_direct_auto_write = h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO
 457                                 && h->param.i_bframe
 458                                 && ( h->param.rc.b_stat_write || !h->param.rc.b_stat_read );
 459
 460     h->param.i_deblocking_filter_alphac0 = x264_clip3( h->param.i_deblocking_filter_alphac0, -6, 6 );
 461     h->param.i_deblocking_filter_beta    = x264_clip3( h->param.i_deblocking_filter_beta, -6, 6 );
 462     h->param.analyse.i_luma_deadzone[0] = x264_clip3( h->param.analyse.i_luma_deadzone[0], 0, 32 );
 463     h->param.analyse.i_luma_deadzone[1] = x264_clip3( h->param.analyse.i_luma_deadzone[1], 0, 32 );
 464
 465     h->param.i_cabac_init_idc = x264_clip3( h->param.i_cabac_init_idc, 0, 2 );
 466
 467     if( h->param.i_cqm_preset < X264_CQM_FLAT || h->param.i_cqm_preset > X264_CQM_CUSTOM )
 468         h->param.i_cqm_preset = X264_CQM_FLAT;
 469
 470     if( h->param.analyse.i_me_method < X264_ME_DIA ||
 471         h->param.analyse.i_me_method > X264_ME_TESA )
 472         h->param.analyse.i_me_method = X264_ME_HEX;
 473     if( h->param.analyse.i_me_range < 4 )
 474         h->param.analyse.i_me_range = 4;
 475     if( h->param.analyse.i_me_range > 16 && h->param.analyse.i_me_method <= X264_ME_HEX )
 476         h->param.analyse.i_me_range = 16;
 477     if( h->param.analyse.i_me_method == X264_ME_TESA &&
 478         (h->mb.b_lossless || h->param.analyse.i_subpel_refine <= 1) )
 479         h->param.analyse.i_me_method = X264_ME_ESA;
 480     h->param.analyse.i_subpel_refine = x264_clip3( h->param.analyse.i_subpel_refine, 0, 9 );
 481     h->param.analyse.b_mixed_references = h->param.analyse.b_mixed_references && h->param.i_frame_reference > 1;
 482     h->param.analyse.inter &= X264_ANALYSE_PSUB16x16|X264_ANALYSE_PSUB8x8|X264_ANALYSE_BSUB16x16|
 483                               X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
 484     h->param.analyse.intra &= X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
 485     if( !(h->param.analyse.inter & X264_ANALYSE_PSUB16x16) )
 486         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
 487     if( !h->param.analyse.b_transform_8x8 )
 488     {
 489         h->param.analyse.inter &= ~X264_ANALYSE_I8x8;
 490         h->param.analyse.intra &= ~X264_ANALYSE_I8x8;
 491     }
 492     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -12, 12);
 493     if( !h->param.b_cabac )
 494         h->param.analyse.i_trellis = 0;
 495     h->param.analyse.i_trellis = x264_clip3( h->param.analyse.i_trellis, 0, 2 );
 496     if( !h->param.analyse.i_trellis )
 497         h->param.analyse.f_psy_trellis = 0;
 498     h->param.analyse.f_psy_rd = x264_clip3f( h->param.analyse.f_psy_rd, 0, 10 );
 499     h->param.analyse.f_psy_trellis = x264_clip3f( h->param.analyse.f_psy_trellis, 0, 10 );
 500     if( h->param.analyse.i_subpel_refine < 6 )
 501         h->param.analyse.f_psy_rd = 0;
 502     h->mb.i_psy_rd = FIX8( h->param.analyse.f_psy_rd );
 503     /* Psy RDO increases overall quantizers to improve the quality of luma--this indirectly hurts chroma quality */
 504     /* so we lower the chroma QP offset to compensate */
 505     /* This can be triggered repeatedly on multiple calls to parameter_validate, but since encoding
 506      * uses the pps chroma qp offset not the param chroma qp offset, this is not a problem. */
 507     if( h->mb.i_psy_rd )
 508         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_rd < 0.25 ? 1 : 2;
 509     h->mb.i_psy_trellis = FIX8( h->param.analyse.f_psy_trellis / 4 );
 510     /* Psy trellis has a similar effect. */
 511     if( h->mb.i_psy_trellis )
 512         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_trellis < 0.25 ? 1 : 2;
 513     else
 514         h->mb.i_psy_trellis = 0;
 515     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -12, 12);
 516     h->param.rc.i_aq_mode = x264_clip3( h->param.rc.i_aq_mode, 0, 1 );
 517     h->param.rc.f_aq_strength = x264_clip3f( h->param.rc.f_aq_strength, 0, 3 );
 518     if( h->param.rc.f_aq_strength == 0 )
 519         h->param.rc.i_aq_mode = 0;
 520     h->param.analyse.i_noise_reduction = x264_clip3( h->param.analyse.i_noise_reduction, 0, 1<<16 );
 521
 522     {
 523         const x264_level_t *l = x264_levels;
 524         if( h->param.i_level_idc < 0 )
 525         {
 526             if( h->param.rc.i_rc_method == X264_RC_ABR && h->param.rc.i_vbv_buffer_size <= 0 )
 527                 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate * 2;
 528             h->sps = h->sps_array;
 529             x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
 530             do h->param.i_level_idc = l->level_idc;
 531                 while( l[1].level_idc && x264_validate_levels( h, 0 ) && l++ );
 532             if( h->param.rc.i_vbv_buffer_size <= 0 )
 533                 h->param.rc.i_vbv_max_bitrate = 0;
 534         }
 535         else
 536         {
 537             while( l->level_idc && l->level_idc != h->param.i_level_idc )
 538                 l++;
 539             if( l->level_idc == 0 )
 540             {
 541                 x264_log( h, X264_LOG_ERROR, "invalid level_idc: %d\n", h->param.i_level_idc );
 542                 return -1;
 543             }
 544         }
 545         if( h->param.analyse.i_mv_range <= 0 )
 546             h->param.analyse.i_mv_range = l->mv_range >> h->param.b_interlaced;
 547         else
 548             h->param.analyse.i_mv_range = x264_clip3(h->param.analyse.i_mv_range, 32, 512 >> h->param.b_interlaced);
 549     }
 550
 551     if( h->param.i_threads > 1 )
 552     {
 553         int r = h->param.analyse.i_mv_range_thread;
 554         int r2;
 555         if( r <= 0 )
 556         {
 557             // half of the available space is reserved and divided evenly among the threads,
 558             // the rest is allocated to whichever thread is far enough ahead to use it.
 559             // reserving more space increases quality for some videos, but costs more time
 560             // in thread synchronization.
 561             int max_range = (h->param.i_height + X264_THREAD_HEIGHT) / h->param.i_threads - X264_THREAD_HEIGHT;
 562             r = max_range / 2;
 563         }
 564         r = X264_MAX( r, h->param.analyse.i_me_range );
 565         r = X264_MIN( r, h->param.analyse.i_mv_range );
 566         // round up to use the whole mb row
 567         r2 = (r & ~15) + ((-X264_THREAD_HEIGHT) & 15);
 568         if( r2 < r )
 569             r2 += 16;
 570         x264_log( h, X264_LOG_DEBUG, "using mv_range_thread = %d\n", r2 );
 571         h->param.analyse.i_mv_range_thread = r2;
 572     }
 573
 574     if( h->param.rc.f_qblur < 0 )
 575         h->param.rc.f_qblur = 0;
 576     if( h->param.rc.f_complexity_blur < 0 )
 577         h->param.rc.f_complexity_blur = 0;
 578
 579     h->param.i_sps_id &= 31;
 580
 581     if( h->param.i_log_level < X264_LOG_INFO )
 582     {
 583         h->param.analyse.b_psnr = 0;
 584         h->param.analyse.b_ssim = 0;
 585     }
 586
 587     /* ensure the booleans are 0 or 1 so they can be used in math */
 588 #define BOOLIFY(x) h->param.x = !!h->param.x
 589     BOOLIFY( b_cabac );
 590     BOOLIFY( b_deblocking_filter );
 591     BOOLIFY( b_interlaced );
 592     BOOLIFY( analyse.b_transform_8x8 );
 593     BOOLIFY( analyse.b_chroma_me );
 594     BOOLIFY( analyse.b_fast_pskip );
 595     BOOLIFY( rc.b_stat_write );
 596     BOOLIFY( rc.b_stat_read );
 597 #undef BOOLIFY
 598
 599     return 0;
 600 }
 601
 602 static void mbcmp_init( x264_t *h )
 603 {
 604     int satd = !h->mb.b_lossless && h->param.analyse.i_subpel_refine > 1;
 605     memcpy( h->pixf.mbcmp, satd ? h->pixf.satd : h->pixf.sad_aligned, sizeof(h->pixf.mbcmp) );
 606     memcpy( h->pixf.mbcmp_unaligned, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.mbcmp_unaligned) );
 607     h->pixf.intra_mbcmp_x3_16x16 = satd ? h->pixf.intra_satd_x3_16x16 : h->pixf.intra_sad_x3_16x16;
 608     h->pixf.intra_mbcmp_x3_8x8c = satd ? h->pixf.intra_satd_x3_8x8c : h->pixf.intra_sad_x3_8x8c;
 609     satd &= h->param.analyse.i_me_method == X264_ME_TESA;
 610     memcpy( h->pixf.fpelcmp, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.fpelcmp) );
 611     memcpy( h->pixf.fpelcmp_x3, satd ? h->pixf.satd_x3 : h->pixf.sad_x3, sizeof(h->pixf.fpelcmp_x3) );
 612     memcpy( h->pixf.fpelcmp_x4, satd ? h->pixf.satd_x4 : h->pixf.sad_x4, sizeof(h->pixf.fpelcmp_x4) );
 613 }
 614
 615 /****************************************************************************
 616  * x264_encoder_open:
 617  ****************************************************************************/
 618 x264_t *x264_encoder_open   ( x264_param_t *param )
 619 {
 620     x264_t *h = x264_malloc( sizeof( x264_t ) );
 621     char buf[1000], *p;
 622     int i;
 623
 624     memset( h, 0, sizeof( x264_t ) );
 625
 626     /* Create a copy of param */
 627     memcpy( &h->param, param, sizeof( x264_param_t ) );
 628
 629     if( x264_validate_parameters( h ) < 0 )
 630     {
 631         x264_free( h );
 632         return NULL;
 633     }
 634
 635     if( h->param.psz_cqm_file )
 636         if( x264_cqm_parse_file( h, h->param.psz_cqm_file ) < 0 )
 637         {
 638             x264_free( h );
 639             return NULL;
 640         }
 641
 642     if( h->param.rc.psz_stat_out )
 643         h->param.rc.psz_stat_out = strdup( h->param.rc.psz_stat_out );
 644     if( h->param.rc.psz_stat_in )
 645         h->param.rc.psz_stat_in = strdup( h->param.rc.psz_stat_in );
 646
 647     /* VUI */
 648     if( h->param.vui.i_sar_width > 0 && h->param.vui.i_sar_height > 0 )
 649     {
 650         int i_w = param->vui.i_sar_width;
 651         int i_h = param->vui.i_sar_height;
 652
 653         x264_reduce_fraction( &i_w, &i_h );
 654
 655         while( i_w > 65535 || i_h > 65535 )
 656         {
 657             i_w /= 2;
 658             i_h /= 2;
 659         }
 660
 661         h->param.vui.i_sar_width = 0;
 662         h->param.vui.i_sar_height = 0;
 663         if( i_w == 0 || i_h == 0 )
 664         {
 665             x264_log( h, X264_LOG_WARNING, "cannot create valid sample aspect ratio\n" );
 666         }
 667         else
 668         {
 669             x264_log( h, X264_LOG_INFO, "using SAR=%d/%d\n", i_w, i_h );
 670             h->param.vui.i_sar_width = i_w;
 671             h->param.vui.i_sar_height = i_h;
 672         }
 673     }
 674
 675     x264_reduce_fraction( &h->param.i_fps_num, &h->param.i_fps_den );
 676
 677     /* Init x264_t */
 678     h->i_frame = 0;
 679     h->i_frame_num = 0;
 680     h->i_idr_pic_id = 0;
 681
 682     h->sps = &h->sps_array[0];
 683     x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
 684
 685     h->pps = &h->pps_array[0];
 686     x264_pps_init( h->pps, h->param.i_sps_id, &h->param, h->sps);
 687
 688     x264_validate_levels( h, 1 );
 689
 690     if( x264_cqm_init( h ) < 0 )
 691     {
 692         x264_free( h );
 693         return NULL;
 694     }
 695
 696     h->mb.i_mb_count = h->sps->i_mb_width * h->sps->i_mb_height;
 697
 698     /* Init frames. */
 699     if( h->param.i_bframe_adaptive == X264_B_ADAPT_TRELLIS )
 700         h->frames.i_delay = X264_MAX(h->param.i_bframe,3)*4 + h->param.i_threads - 1;
 701     else
 702         h->frames.i_delay = h->param.i_bframe + h->param.i_threads - 1;
 703     h->frames.i_max_ref0 = h->param.i_frame_reference;
 704     h->frames.i_max_ref1 = h->sps->vui.i_num_reorder_frames;
 705     h->frames.i_max_dpb  = h->sps->vui.i_max_dec_frame_buffering;
 706     h->frames.b_have_lowres = !h->param.rc.b_stat_read
 707         && ( h->param.rc.i_rc_method == X264_RC_ABR
 708           || h->param.rc.i_rc_method == X264_RC_CRF
 709           || h->param.i_bframe_adaptive
 710           || h->param.i_scenecut_threshold );
 711     h->frames.b_have_lowres |= (h->param.rc.b_stat_read && h->param.rc.i_vbv_buffer_size > 0);
 712     h->frames.b_have_sub8x8_esa = !!(h->param.analyse.inter & X264_ANALYSE_PSUB8x8);
 713
 714     h->frames.i_last_idr = - h->param.i_keyint_max;
 715     h->frames.i_input    = 0;
 716     h->frames.last_nonb  = NULL;
 717
 718     h->i_ref0 = 0;
 719     h->i_ref1 = 0;
 720
 721     h->chroma_qp_table = i_chroma_qp_table + 12 + h->pps->i_chroma_qp_index_offset;
 722
 723     x264_rdo_init( );
 724
 725     /* init CPU functions */
 726     x264_predict_16x16_init( h->param.cpu, h->predict_16x16 );
 727     x264_predict_8x8c_init( h->param.cpu, h->predict_8x8c );
 728     x264_predict_8x8_init( h->param.cpu, h->predict_8x8, &h->predict_8x8_filter );
 729     x264_predict_4x4_init( h->param.cpu, h->predict_4x4 );
 730     if( !h->param.b_cabac );
 731         x264_init_vlc_tables();
 732     x264_pixel_init( h->param.cpu, &h->pixf );
 733     x264_dct_init( h->param.cpu, &h->dctf );
 734     x264_zigzag_init( h->param.cpu, &h->zigzagf, h->param.b_interlaced );
 735     x264_mc_init( h->param.cpu, &h->mc );
 736     x264_quant_init( h, h->param.cpu, &h->quantf );
 737     x264_deblock_init( h->param.cpu, &h->loopf );
 738     x264_dct_init_weights();
 739
 740     mbcmp_init( h );
 741
 742     p = buf + sprintf( buf, "using cpu capabilities:" );
 743     for( i=0; x264_cpu_names[i].flags; i++ )
 744     {
 745         if( !strcmp(x264_cpu_names[i].name, "SSE2")
 746             && param->cpu & (X264_CPU_SSE2_IS_FAST|X264_CPU_SSE2_IS_SLOW) )
 747             continue;
 748         if( !strcmp(x264_cpu_names[i].name, "SSE3")
 749             && (param->cpu & X264_CPU_SSSE3 || !(param->cpu & X264_CPU_CACHELINE_64)) )
 750             continue;
 751         if( !strcmp(x264_cpu_names[i].name, "SSE4.1")
 752             && (param->cpu & X264_CPU_SSE42) )
 753             continue;
 754         if( (param->cpu & x264_cpu_names[i].flags) == x264_cpu_names[i].flags
 755             && (!i || x264_cpu_names[i].flags != x264_cpu_names[i-1].flags) )
 756             p += sprintf( p, " %s", x264_cpu_names[i].name );
 757     }
 758     if( !param->cpu )
 759         p += sprintf( p, " none!" );
 760     x264_log( h, X264_LOG_INFO, "%s\n", buf );
 761
 762     h->out.i_nal = 0;
 763     h->out.i_bitstream = X264_MAX( 1000000, h->param.i_width * h->param.i_height * 4
 764         * ( h->param.rc.i_rc_method == X264_RC_ABR ? pow( 0.95, h->param.rc.i_qp_min )
 765           : pow( 0.95, h->param.rc.i_qp_constant ) * X264_MAX( 1, h->param.rc.f_ip_factor )));
 766
 767     h->thread[0] = h;
 768     h->i_thread_num = 0;
 769     for( i = 1; i < h->param.i_threads; i++ )
 770         h->thread[i] = x264_malloc( sizeof(x264_t) );
 771
 772     for( i = 0; i < h->param.i_threads; i++ )
 773     {
 774         if( i > 0 )
 775             *h->thread[i] = *h;
 776         h->thread[i]->fdec = x264_frame_pop_unused( h );
 777         h->thread[i]->out.p_bitstream = x264_malloc( h->out.i_bitstream );
 778         if( x264_macroblock_cache_init( h->thread[i] ) < 0 )
 779             return NULL;
 780     }
 781
 782     if( x264_ratecontrol_new( h ) < 0 )
 783         return NULL;
 784
 785     if( h->param.psz_dump_yuv )
 786     {
 787         /* create or truncate the reconstructed video file */
 788         FILE *f = fopen( h->param.psz_dump_yuv, "w" );
 789         if( f )
 790             fclose( f );
 791         else
 792         {
 793             x264_log( h, X264_LOG_ERROR, "can't write to fdec.yuv\n" );
 794             x264_free( h );
 795             return NULL;
 796         }
 797     }
 798
 799     x264_log( h, X264_LOG_INFO, "profile %s, level %d.%d\n",
 800         h->sps->i_profile_idc == PROFILE_BASELINE ? "Baseline" :
 801         h->sps->i_profile_idc == PROFILE_MAIN ? "Main" :
 802         h->sps->i_profile_idc == PROFILE_HIGH ? "High" :
 803         "High 4:4:4 Predictive", h->sps->i_level_idc/10, h->sps->i_level_idc%10 );
 804
 805     return h;
 806 }
 807
 808 /****************************************************************************
 809  * x264_encoder_reconfig:
 810  ****************************************************************************/
 811 int x264_encoder_reconfig( x264_t *h, x264_param_t *param )
 812 {
 813 #define COPY(var) h->param.var = param->var
 814     COPY( i_frame_reference ); // but never uses more refs than initially specified
 815     COPY( i_bframe_bias );
 816     if( h->param.i_scenecut_threshold )
 817         COPY( i_scenecut_threshold ); // can't turn it on or off, only vary the threshold
 818     COPY( b_deblocking_filter );
 819     COPY( i_deblocking_filter_alphac0 );
 820     COPY( i_deblocking_filter_beta );
 821     COPY( analyse.intra );
 822     COPY( analyse.inter );
 823     COPY( analyse.i_direct_mv_pred );
 824     /* Scratch buffer prevents me_range from being increased for esa/tesa */
 825     if( h->param.analyse.i_me_method < X264_ME_ESA || param->analyse.i_me_range < h->param.analyse.i_me_range )
 826         COPY( analyse.i_me_range );
 827     COPY( analyse.i_noise_reduction );
 828     /* We can't switch out of subme=0 during encoding. */
 829     if( h->param.analyse.i_subpel_refine )
 830         COPY( analyse.i_subpel_refine );
 831     COPY( analyse.i_trellis );
 832     COPY( analyse.b_chroma_me );
 833     COPY( analyse.b_dct_decimate );
 834     COPY( analyse.b_fast_pskip );
 835     COPY( analyse.b_mixed_references );
 836     COPY( analyse.f_psy_rd );
 837     COPY( analyse.f_psy_trellis );
 838     // can only twiddle these if they were enabled to begin with:
 839     if( h->param.analyse.i_me_method >= X264_ME_ESA || param->analyse.i_me_method < X264_ME_ESA )
 840         COPY( analyse.i_me_method );
 841     if( h->param.analyse.i_me_method >= X264_ME_ESA && !h->frames.b_have_sub8x8_esa )
 842         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
 843     if( h->pps->b_transform_8x8_mode )
 844         COPY( analyse.b_transform_8x8 );
 845     if( h->frames.i_max_ref1 > 1 )
 846         COPY( b_bframe_pyramid );
 847 #undef COPY
 848
 849     mbcmp_init( h );
 850
 851     return x264_validate_parameters( h );
 852 }
 853
 854 /* internal usage */
 855 static void x264_nal_start( x264_t *h, int i_type, int i_ref_idc )
 856 {
 857     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
 858
 859     nal->i_ref_idc = i_ref_idc;
 860     nal->i_type    = i_type;
 861
 862     nal->i_payload= 0;
 863     nal->p_payload= &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8];
 864 }
 865 static void x264_nal_end( x264_t *h )
 866 {
 867     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
 868     nal->i_payload = &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8] - nal->p_payload;
 869     h->out.i_nal++;
 870 }
 871
 872 /****************************************************************************
 873  * x264_encoder_headers:
 874  ****************************************************************************/
 875 int x264_encoder_headers( x264_t *h, x264_nal_t **pp_nal, int *pi_nal )
 876 {
 877     /* init bitstream context */
 878     h->out.i_nal = 0;
 879     bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
 880
 881     /* Put SPS and PPS */
 882     if( h->i_frame == 0 )
 883     {
 884         /* identify ourself */
 885         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
 886         x264_sei_version_write( h, &h->out.bs );
 887         x264_nal_end( h );
 888
 889         /* generate sequence parameters */
 890         x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
 891         x264_sps_write( &h->out.bs, h->sps );
 892         x264_nal_end( h );
 893
 894         /* generate picture parameters */
 895         x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
 896         x264_pps_write( &h->out.bs, h->pps );
 897         x264_nal_end( h );
 898     }
 899     /* now set output*/
 900     *pi_nal = h->out.i_nal;
 901     *pp_nal = &h->out.nal[0];
 902     h->out.i_nal = 0;
 903
 904     return 0;
 905 }
 906
 907 static inline void x264_reference_build_list( x264_t *h, int i_poc )
 908 {
 909     int i;
 910     int b_ok;
 911
 912     /* build ref list 0/1 */
 913     h->i_ref0 = 0;
 914     h->i_ref1 = 0;
 915     for( i = 0; h->frames.reference[i]; i++ )
 916     {
 917         if( h->frames.reference[i]->i_poc < i_poc )
 918         {
 919             h->fref0[h->i_ref0++] = h->frames.reference[i];
 920         }
 921         else if( h->frames.reference[i]->i_poc > i_poc )
 922         {
 923             h->fref1[h->i_ref1++] = h->frames.reference[i];
 924         }
 925     }
 926
 927     /* Order ref0 from higher to lower poc */
 928     do
 929     {
 930         b_ok = 1;
 931         for( i = 0; i < h->i_ref0 - 1; i++ )
 932         {
 933             if( h->fref0[i]->i_poc < h->fref0[i+1]->i_poc )
 934             {
 935                 XCHG( x264_frame_t*, h->fref0[i], h->fref0[i+1] );
 936                 b_ok = 0;
 937                 break;
 938             }
 939         }
 940     } while( !b_ok );
 941     /* Order ref1 from lower to higher poc (bubble sort) for B-frame */
 942     do
 943     {
 944         b_ok = 1;
 945         for( i = 0; i < h->i_ref1 - 1; i++ )
 946         {
 947             if( h->fref1[i]->i_poc > h->fref1[i+1]->i_poc )
 948             {
 949                 XCHG( x264_frame_t*, h->fref1[i], h->fref1[i+1] );
 950                 b_ok = 0;
 951                 break;
 952             }
 953         }
 954     } while( !b_ok );
 955
 956     /* In the standard, a P-frame's ref list is sorted by frame_num.
 957      * We use POC, but check whether explicit reordering is needed */
 958     h->b_ref_reorder[0] =
 959     h->b_ref_reorder[1] = 0;
 960     if( h->sh.i_type == SLICE_TYPE_P )
 961     {
 962         for( i = 0; i < h->i_ref0 - 1; i++ )
 963             if( h->fref0[i]->i_frame_num < h->fref0[i+1]->i_frame_num )
 964             {
 965                 h->b_ref_reorder[0] = 1;
 966                 break;
 967             }
 968     }
 969
 970     h->i_ref1 = X264_MIN( h->i_ref1, h->frames.i_max_ref1 );
 971     h->i_ref0 = X264_MIN( h->i_ref0, h->frames.i_max_ref0 );
 972     h->i_ref0 = X264_MIN( h->i_ref0, h->param.i_frame_reference ); // if reconfig() has lowered the limit
 973     assert( h->i_ref0 + h->i_ref1 <= 16 );
 974     h->mb.pic.i_fref[0] = h->i_ref0;
 975     h->mb.pic.i_fref[1] = h->i_ref1;
 976 }
 977
 978 static void x264_fdec_filter_row( x264_t *h, int mb_y )
 979 {
 980     /* mb_y is the mb to be encoded next, not the mb to be filtered here */
 981     int b_hpel = h->fdec->b_kept_as_ref;
 982     int b_deblock = !h->sh.i_disable_deblocking_filter_idc;
 983     int b_end = mb_y == h->sps->i_mb_height;
 984     int min_y = mb_y - (1 << h->sh.b_mbaff);
 985     int max_y = b_end ? h->sps->i_mb_height : mb_y;
 986     b_deblock &= b_hpel || h->param.psz_dump_yuv;
 987     if( mb_y & h->sh.b_mbaff )
 988         return;
 989     if( min_y < 0 )
 990         return;
 991
 992     if( !b_end )
 993     {
 994         int i, j;
 995         for( j=0; j<=h->sh.b_mbaff; j++ )
 996             for( i=0; i<3; i++ )
 997             {
 998                 memcpy( h->mb.intra_border_backup[j][i],
 999                         h->fdec->plane[i] + ((mb_y*16 >> !!i) + j - 1 - h->sh.b_mbaff) * h->fdec->i_stride[i],
1000                         h->sps->i_mb_width*16 >> !!i );
1001             }
1002     }
1003
1004     if( b_deblock )
1005     {
1006         int y;
1007         for( y = min_y; y < max_y; y += (1 << h->sh.b_mbaff) )
1008             x264_frame_deblock_row( h, y );
1009     }
1010
1011     if( b_hpel )
1012     {
1013         x264_frame_expand_border( h, h->fdec, min_y, b_end );
1014         if( h->param.analyse.i_subpel_refine )
1015         {
1016             x264_frame_filter( h, h->fdec, min_y, b_end );
1017             x264_frame_expand_border_filtered( h, h->fdec, min_y, b_end );
1018         }
1019     }
1020
1021     if( h->param.i_threads > 1 && h->fdec->b_kept_as_ref )
1022     {
1023         x264_frame_cond_broadcast( h->fdec, mb_y*16 + (b_end ? 10000 : -(X264_THREAD_HEIGHT << h->sh.b_mbaff)) );
1024     }
1025
1026     min_y = X264_MAX( min_y*16-8, 0 );
1027     max_y = b_end ? h->param.i_height : mb_y*16-8;
1028
1029     if( h->param.analyse.b_psnr )
1030     {
1031         int i;
1032         for( i=0; i<3; i++ )
1033             h->stat.frame.i_ssd[i] +=
1034                 x264_pixel_ssd_wxh( &h->pixf,
1035                     h->fdec->plane[i] + (min_y>>!!i) * h->fdec->i_stride[i], h->fdec->i_stride[i],
1036                     h->fenc->plane[i] + (min_y>>!!i) * h->fenc->i_stride[i], h->fenc->i_stride[i],
1037                     h->param.i_width >> !!i, (max_y-min_y) >> !!i );
1038     }
1039
1040     if( h->param.analyse.b_ssim )
1041     {
1042         x264_emms();
1043         /* offset by 2 pixels to avoid alignment of ssim blocks with dct blocks,
1044          * and overlap by 4 */
1045         min_y += min_y == 0 ? 2 : -6;
1046         h->stat.frame.f_ssim +=
1047             x264_pixel_ssim_wxh( &h->pixf,
1048                 h->fdec->plane[0] + 2+min_y*h->fdec->i_stride[0], h->fdec->i_stride[0],
1049                 h->fenc->plane[0] + 2+min_y*h->fenc->i_stride[0], h->fenc->i_stride[0],
1050                 h->param.i_width-2, max_y-min_y, h->scratch_buffer );
1051     }
1052 }
1053
1054 static inline void x264_reference_update( x264_t *h )
1055 {
1056     int i;
1057
1058     if( h->fdec->i_frame >= 0 )
1059         h->i_frame++;
1060
1061     if( !h->fdec->b_kept_as_ref )
1062     {
1063         if( h->param.i_threads > 1 )
1064         {
1065             x264_frame_push_unused( h, h->fdec );
1066             h->fdec = x264_frame_pop_unused( h );
1067         }
1068         return;
1069     }
1070
1071     /* move lowres copy of the image to the ref frame */
1072     for( i = 0; i < 4; i++)
1073     {
1074         XCHG( uint8_t*, h->fdec->lowres[i], h->fenc->lowres[i] );
1075         XCHG( uint8_t*, h->fdec->buffer_lowres[i], h->fenc->buffer_lowres[i] );
1076     }
1077
1078     /* adaptive B decision needs a pointer, since it can't use the ref lists */
1079     if( h->sh.i_type != SLICE_TYPE_B )
1080         h->frames.last_nonb = h->fdec;
1081
1082     /* move frame in the buffer */
1083     x264_frame_push( h->frames.reference, h->fdec );
1084     if( h->frames.reference[h->frames.i_max_dpb] )
1085         x264_frame_push_unused( h, x264_frame_shift( h->frames.reference ) );
1086     h->fdec = x264_frame_pop_unused( h );
1087 }
1088
1089 static inline void x264_reference_reset( x264_t *h )
1090 {
1091     while( h->frames.reference[0] )
1092         x264_frame_push_unused( h, x264_frame_pop( h->frames.reference ) );
1093     h->fdec->i_poc =
1094     h->fenc->i_poc = 0;
1095 }
1096
1097 static inline void x264_slice_init( x264_t *h, int i_nal_type, int i_global_qp )
1098 {
1099     /* ------------------------ Create slice header  ----------------------- */
1100     if( i_nal_type == NAL_SLICE_IDR )
1101     {
1102         x264_slice_header_init( h, &h->sh, h->sps, h->pps, h->i_idr_pic_id, h->i_frame_num, i_global_qp );
1103
1104         /* increment id */
1105         h->i_idr_pic_id = ( h->i_idr_pic_id + 1 ) % 65536;
1106     }
1107     else
1108     {
1109         x264_slice_header_init( h, &h->sh, h->sps, h->pps, -1, h->i_frame_num, i_global_qp );
1110
1111         /* always set the real higher num of ref frame used */
1112         h->sh.b_num_ref_idx_override = 1;
1113         h->sh.i_num_ref_idx_l0_active = h->i_ref0 <= 0 ? 1 : h->i_ref0;
1114         h->sh.i_num_ref_idx_l1_active = h->i_ref1 <= 0 ? 1 : h->i_ref1;
1115     }
1116
1117     h->fdec->i_frame_num = h->sh.i_frame_num;
1118
1119     if( h->sps->i_poc_type == 0 )
1120     {
1121         h->sh.i_poc_lsb = h->fdec->i_poc & ( (1 << h->sps->i_log2_max_poc_lsb) - 1 );
1122         h->sh.i_delta_poc_bottom = 0;   /* XXX won't work for field */
1123     }
1124     else if( h->sps->i_poc_type == 1 )
1125     {
1126         /* FIXME TODO FIXME */
1127     }
1128     else
1129     {
1130         /* Nothing to do ? */
1131     }
1132
1133     x264_macroblock_slice_init( h );
1134 }
1135
1136 static void x264_slice_write( x264_t *h )
1137 {
1138     int i_skip;
1139     int mb_xy, i_mb_x, i_mb_y;
1140     int i, i_list, i_ref;
1141
1142     /* init stats */
1143     memset( &h->stat.frame, 0, sizeof(h->stat.frame) );
1144
1145     /* Slice */
1146     x264_nal_start( h, h->i_nal_type, h->i_nal_ref_idc );
1147
1148     /* Slice header */
1149     x264_slice_header_write( &h->out.bs, &h->sh, h->i_nal_ref_idc );
1150     if( h->param.b_cabac )
1151     {
1152         /* alignment needed */
1153         bs_align_1( &h->out.bs );
1154
1155         /* init cabac */
1156         x264_cabac_context_init( &h->cabac, h->sh.i_type, h->sh.i_qp, h->sh.i_cabac_init_idc );
1157         x264_cabac_encode_init ( &h->cabac, h->out.bs.p, h->out.bs.p_end );
1158     }
1159     h->mb.i_last_qp = h->sh.i_qp;
1160     h->mb.i_last_dqp = 0;
1161
1162     i_mb_y = h->sh.i_first_mb / h->sps->i_mb_width;
1163     i_mb_x = h->sh.i_first_mb % h->sps->i_mb_width;
1164     i_skip = 0;
1165
1166     while( (mb_xy = i_mb_x + i_mb_y * h->sps->i_mb_width) < h->sh.i_last_mb )
1167     {
1168         int mb_spos = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
1169
1170         if( i_mb_x == 0 )
1171             x264_fdec_filter_row( h, i_mb_y );
1172
1173         /* load cache */
1174         x264_macroblock_cache_load( h, i_mb_x, i_mb_y );
1175
1176         /* analyse parameters
1177          * Slice I: choose I_4x4 or I_16x16 mode
1178          * Slice P: choose between using P mode or intra (4x4 or 16x16)
1179          * */
1180         x264_macroblock_analyse( h );
1181
1182         /* encode this macroblock -> be careful it can change the mb type to P_SKIP if needed */
1183         x264_macroblock_encode( h );
1184
1185         x264_bitstream_check_buffer( h );
1186
1187         if( h->param.b_cabac )
1188         {
1189             if( mb_xy > h->sh.i_first_mb && !(h->sh.b_mbaff && (i_mb_y&1)) )
1190                 x264_cabac_encode_terminal( &h->cabac );
1191
1192             if( IS_SKIP( h->mb.i_type ) )
1193                 x264_cabac_mb_skip( h, 1 );
1194             else
1195             {
1196                 if( h->sh.i_type != SLICE_TYPE_I )
1197                     x264_cabac_mb_skip( h, 0 );
1198                 x264_macroblock_write_cabac( h, &h->cabac );
1199             }
1200         }
1201         else
1202         {
1203             if( IS_SKIP( h->mb.i_type ) )
1204                 i_skip++;
1205             else
1206             {
1207                 if( h->sh.i_type != SLICE_TYPE_I )
1208                 {
1209                     bs_write_ue( &h->out.bs, i_skip );  /* skip run */
1210                     i_skip = 0;
1211                 }
1212                 x264_macroblock_write_cavlc( h, &h->out.bs );
1213             }
1214         }
1215
1216 #if VISUALIZE
1217         if( h->param.b_visualize )
1218             x264_visualize_mb( h );
1219 #endif
1220
1221         /* save cache */
1222         x264_macroblock_cache_save( h );
1223
1224         /* accumulate mb stats */
1225         h->stat.frame.i_mb_count[h->mb.i_type]++;
1226         if( !IS_SKIP(h->mb.i_type) && !IS_INTRA(h->mb.i_type) && !IS_DIRECT(h->mb.i_type) )
1227         {
1228             if( h->mb.i_partition != D_8x8 )
1229                 h->stat.frame.i_mb_partition[h->mb.i_partition] += 4;
1230             else
1231                 for( i = 0; i < 4; i++ )
1232                     h->stat.frame.i_mb_partition[h->mb.i_sub_partition[i]] ++;
1233             if( h->param.i_frame_reference > 1 )
1234                 for( i_list = 0; i_list <= (h->sh.i_type == SLICE_TYPE_B); i_list++ )
1235                     for( i = 0; i < 4; i++ )
1236                     {
1237                         i_ref = h->mb.cache.ref[i_list][ x264_scan8[4*i] ];
1238                         if( i_ref >= 0 )
1239                             h->stat.frame.i_mb_count_ref[i_list][i_ref] ++;
1240                     }
1241         }
1242         if( h->mb.i_cbp_luma && !IS_INTRA(h->mb.i_type) )
1243         {
1244             h->stat.frame.i_mb_count_8x8dct[0] ++;
1245             h->stat.frame.i_mb_count_8x8dct[1] += h->mb.b_transform_8x8;
1246         }
1247
1248         x264_ratecontrol_mb( h, bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac) - mb_spos );
1249
1250         if( h->sh.b_mbaff )
1251         {
1252             i_mb_x += i_mb_y & 1;
1253             i_mb_y ^= i_mb_x < h->sps->i_mb_width;
1254         }
1255         else
1256             i_mb_x++;
1257         if(i_mb_x == h->sps->i_mb_width)
1258         {
1259             i_mb_y++;
1260             i_mb_x = 0;
1261         }
1262     }
1263
1264     if( h->param.b_cabac )
1265     {
1266         x264_cabac_encode_flush( h, &h->cabac );
1267         h->out.bs.p = h->cabac.p;
1268     }
1269     else
1270     {
1271         if( i_skip > 0 )
1272             bs_write_ue( &h->out.bs, i_skip );  /* last skip run */
1273         /* rbsp_slice_trailing_bits */
1274         bs_rbsp_trailing( &h->out.bs );
1275     }
1276
1277     x264_nal_end( h );
1278
1279     x264_fdec_filter_row( h, h->sps->i_mb_height );
1280
1281     /* Compute misc bits */
1282     h->stat.frame.i_misc_bits = bs_pos( &h->out.bs )
1283                               + NALU_OVERHEAD * 8
1284                               - h->stat.frame.i_tex_bits
1285                               - h->stat.frame.i_mv_bits;
1286 }
1287
1288 static void x264_thread_sync_context( x264_t *dst, x264_t *src )
1289 {
1290     x264_frame_t **f;
1291     if( dst == src )
1292         return;
1293
1294     // reference counting
1295     for( f = src->frames.reference; *f; f++ )
1296         (*f)->i_reference_count++;
1297     for( f = dst->frames.reference; *f; f++ )
1298         x264_frame_push_unused( src, *f );
1299     src->fdec->i_reference_count++;
1300     x264_frame_push_unused( src, dst->fdec );
1301
1302     // copy everything except the per-thread pointers and the constants.
1303     memcpy( &dst->i_frame, &src->i_frame, offsetof(x264_t, mb.type) - offsetof(x264_t, i_frame) );
1304     dst->stat = src->stat;
1305 }
1306
1307 static void x264_thread_sync_stat( x264_t *dst, x264_t *src )
1308 {
1309     if( dst == src )
1310         return;
1311     memcpy( &dst->stat.i_slice_count, &src->stat.i_slice_count, sizeof(dst->stat) - sizeof(dst->stat.frame) );
1312 }
1313
1314 static int x264_slices_write( x264_t *h )
1315 {
1316     int i_frame_size;
1317
1318 #ifdef HAVE_MMX
1319     /* Misalign mask has to be set separately for each thread. */
1320     if( h->param.cpu&X264_CPU_SSE_MISALIGN )
1321         x264_cpu_mask_misalign_sse();
1322 #endif
1323
1324 #if VISUALIZE
1325     if( h->param.b_visualize )
1326         x264_visualize_init( h );
1327 #endif
1328
1329     x264_stack_align( x264_slice_write, h );
1330     i_frame_size = h->out.nal[h->out.i_nal-1].i_payload;
1331
1332 #if VISUALIZE
1333     if( h->param.b_visualize )
1334     {
1335         x264_visualize_show( h );
1336         x264_visualize_close( h );
1337     }
1338 #endif
1339
1340     h->out.i_frame_size = i_frame_size;
1341     return 0;
1342 }
1343
1344 /****************************************************************************
1345  * x264_encoder_encode:
1346  *  XXX: i_poc   : is the poc of the current given picture
1347  *       i_frame : is the number of the frame being coded
1348  *  ex:  type frame poc
1349  *       I      0   2*0
1350  *       P      1   2*3
1351  *       B      2   2*1
1352  *       B      3   2*2
1353  *       P      4   2*6
1354  *       B      5   2*4
1355  *       B      6   2*5
1356  ****************************************************************************/
1357 int     x264_encoder_encode( x264_t *h,
1358                              x264_nal_t **pp_nal, int *pi_nal,
1359                              x264_picture_t *pic_in,
1360                              x264_picture_t *pic_out )
1361 {
1362     x264_t *thread_current, *thread_prev, *thread_oldest;
1363     int     i_nal_type;
1364     int     i_nal_ref_idc;
1365
1366     int   i_global_qp;
1367
1368     if( h->param.i_threads > 1)
1369     {
1370         int i = ++h->i_thread_phase;
1371         int t = h->param.i_threads;
1372         thread_current = h->thread[ i%t ];
1373         thread_prev    = h->thread[ (i-1)%t ];
1374         thread_oldest  = h->thread[ (i+1)%t ];
1375         x264_thread_sync_context( thread_current, thread_prev );
1376         x264_thread_sync_ratecontrol( thread_current, thread_prev, thread_oldest );
1377         h = thread_current;
1378 //      fprintf(stderr, "current: %p  prev: %p  oldest: %p \n", thread_current, thread_prev, thread_oldest);
1379     }
1380     else
1381     {
1382         thread_current =
1383         thread_prev    =
1384         thread_oldest  = h;
1385     }
1386
1387     // ok to call this before encoding any frames, since the initial values of fdec have b_kept_as_ref=0
1388     x264_reference_update( h );
1389     h->fdec->i_lines_completed = -1;
1390
1391     /* no data out */
1392     *pi_nal = 0;
1393     *pp_nal = NULL;
1394
1395     /* ------------------- Setup new frame from picture -------------------- */
1396     if( pic_in != NULL )
1397     {
1398         /* 1: Copy the picture to a frame and move it to a buffer */
1399         x264_frame_t *fenc = x264_frame_pop_unused( h );
1400
1401         if( x264_frame_copy_picture( h, fenc, pic_in ) < 0 )
1402             return -1;
1403
1404         if( h->param.i_width != 16 * h->sps->i_mb_width ||
1405             h->param.i_height != 16 * h->sps->i_mb_height )
1406             x264_frame_expand_border_mod16( h, fenc );
1407
1408         fenc->i_frame = h->frames.i_input++;
1409
1410         x264_frame_push( h->frames.next, fenc );
1411
1412         if( h->frames.b_have_lowres )
1413             x264_frame_init_lowres( h, fenc );
1414
1415         if( h->param.rc.i_aq_mode )
1416             x264_adaptive_quant_frame( h, fenc );
1417
1418         if( h->frames.i_input <= h->frames.i_delay + 1 - h->param.i_threads )
1419         {
1420             /* Nothing yet to encode */
1421             /* waiting for filling bframe buffer */
1422             pic_out->i_type = X264_TYPE_AUTO;
1423             return 0;
1424         }
1425     }
1426
1427     if( h->frames.current[0] == NULL )
1428     {
1429         int bframes = 0;
1430         /* 2: Select frame types */
1431         if( h->frames.next[0] == NULL )
1432         {
1433             x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
1434             return 0;
1435         }
1436
1437         x264_stack_align( x264_slicetype_decide, h );
1438
1439         /* 3: move some B-frames and 1 non-B to encode queue */
1440         while( IS_X264_TYPE_B( h->frames.next[bframes]->i_type ) )
1441             bframes++;
1442         x264_frame_push( h->frames.current, x264_frame_shift( &h->frames.next[bframes] ) );
1443         /* FIXME: when max B-frames > 3, BREF may no longer be centered after GOP closing */
1444         if( h->param.b_bframe_pyramid && bframes > 1 )
1445         {
1446             x264_frame_t *mid = x264_frame_shift( &h->frames.next[bframes/2] );
1447             mid->i_type = X264_TYPE_BREF;
1448             x264_frame_push( h->frames.current, mid );
1449             bframes--;
1450         }
1451         while( bframes-- )
1452             x264_frame_push( h->frames.current, x264_frame_shift( h->frames.next ) );
1453     }
1454
1455     /* ------------------- Get frame to be encoded ------------------------- */
1456     /* 4: get picture to encode */
1457     h->fenc = x264_frame_shift( h->frames.current );
1458     if( h->fenc == NULL )
1459     {
1460         /* Nothing yet to encode (ex: waiting for I/P with B frames) */
1461         /* waiting for filling bframe buffer */
1462         pic_out->i_type = X264_TYPE_AUTO;
1463         return 0;
1464     }
1465
1466     if( h->fenc->i_type == X264_TYPE_IDR )
1467     {
1468         h->frames.i_last_idr = h->fenc->i_frame;
1469     }
1470
1471     /* ------------------- Setup frame context ----------------------------- */
1472     /* 5: Init data dependent of frame type */
1473     if( h->fenc->i_type == X264_TYPE_IDR )
1474     {
1475         /* reset ref pictures */
1476         x264_reference_reset( h );
1477
1478         i_nal_type    = NAL_SLICE_IDR;
1479         i_nal_ref_idc = NAL_PRIORITY_HIGHEST;
1480         h->sh.i_type = SLICE_TYPE_I;
1481     }
1482     else if( h->fenc->i_type == X264_TYPE_I )
1483     {
1484         i_nal_type    = NAL_SLICE;
1485         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
1486         h->sh.i_type = SLICE_TYPE_I;
1487     }
1488     else if( h->fenc->i_type == X264_TYPE_P )
1489     {
1490         i_nal_type    = NAL_SLICE;
1491         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
1492         h->sh.i_type = SLICE_TYPE_P;
1493     }
1494     else if( h->fenc->i_type == X264_TYPE_BREF )
1495     {
1496         i_nal_type    = NAL_SLICE;
1497         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* maybe add MMCO to forget it? -> low */
1498         h->sh.i_type = SLICE_TYPE_B;
1499     }
1500     else    /* B frame */
1501     {
1502         i_nal_type    = NAL_SLICE;
1503         i_nal_ref_idc = NAL_PRIORITY_DISPOSABLE;
1504         h->sh.i_type = SLICE_TYPE_B;
1505     }
1506
1507     h->fdec->i_poc =
1508     h->fenc->i_poc = 2 * (h->fenc->i_frame - h->frames.i_last_idr);
1509     h->fdec->i_type = h->fenc->i_type;
1510     h->fdec->i_frame = h->fenc->i_frame;
1511     h->fenc->b_kept_as_ref =
1512     h->fdec->b_kept_as_ref = i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE && h->param.i_keyint_max > 1;
1513
1514
1515
1516     /* ------------------- Init                ----------------------------- */
1517     /* build ref list 0/1 */
1518     x264_reference_build_list( h, h->fdec->i_poc );
1519
1520     /* Init the rate control */
1521     x264_ratecontrol_start( h, h->fenc->i_qpplus1 );
1522     i_global_qp = x264_ratecontrol_qp( h );
1523
1524     pic_out->i_qpplus1 =
1525     h->fdec->i_qpplus1 = i_global_qp + 1;
1526
1527     if( h->sh.i_type == SLICE_TYPE_B )
1528         x264_macroblock_bipred_init( h );
1529
1530     /* ------------------------ Create slice header  ----------------------- */
1531     x264_slice_init( h, i_nal_type, i_global_qp );
1532
1533     if( i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE )
1534         h->i_frame_num++;
1535
1536     /* ---------------------- Write the bitstream -------------------------- */
1537     /* Init bitstream context */
1538     h->out.i_nal = 0;
1539     bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
1540
1541     if(h->param.b_aud){
1542         int pic_type;
1543
1544         if(h->sh.i_type == SLICE_TYPE_I)
1545             pic_type = 0;
1546         else if(h->sh.i_type == SLICE_TYPE_P)
1547             pic_type = 1;
1548         else if(h->sh.i_type == SLICE_TYPE_B)
1549             pic_type = 2;
1550         else
1551             pic_type = 7;
1552
1553         x264_nal_start(h, NAL_AUD, NAL_PRIORITY_DISPOSABLE);
1554         bs_write(&h->out.bs, 3, pic_type);
1555         bs_rbsp_trailing(&h->out.bs);
1556         x264_nal_end(h);
1557     }
1558
1559     h->i_nal_type = i_nal_type;
1560     h->i_nal_ref_idc = i_nal_ref_idc;
1561
1562     /* Write SPS and PPS */
1563     if( i_nal_type == NAL_SLICE_IDR && h->param.b_repeat_headers )
1564     {
1565         if( h->fenc->i_frame == 0 )
1566         {
1567             /* identify ourself */
1568             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
1569             x264_sei_version_write( h, &h->out.bs );
1570             x264_nal_end( h );
1571         }
1572
1573         /* generate sequence parameters */
1574         x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
1575         x264_sps_write( &h->out.bs, h->sps );
1576         x264_nal_end( h );
1577
1578         /* generate picture parameters */
1579         x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
1580         x264_pps_write( &h->out.bs, h->pps );
1581         x264_nal_end( h );
1582     }
1583
1584     /* Write frame */
1585     if( h->param.i_threads > 1 )
1586     {
1587         x264_pthread_create( &h->thread_handle, NULL, (void*)x264_slices_write, h );
1588         h->b_thread_active = 1;
1589     }
1590     else
1591         x264_slices_write( h );
1592
1593     x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
1594     return 0;
1595 }
1596
1597 static void x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
1598                                     x264_nal_t **pp_nal, int *pi_nal,
1599                                     x264_picture_t *pic_out )
1600 {
1601     int i, i_list;
1602     char psz_message[80];
1603
1604     if( h->b_thread_active )
1605     {
1606         x264_pthread_join( h->thread_handle, NULL );
1607         h->b_thread_active = 0;
1608     }
1609     if( !h->out.i_nal )
1610     {
1611         pic_out->i_type = X264_TYPE_AUTO;
1612         return;
1613     }
1614
1615     x264_frame_push_unused( thread_current, h->fenc );
1616
1617     /* End bitstream, set output  */
1618     *pi_nal = h->out.i_nal;
1619     *pp_nal = h->out.nal;
1620     h->out.i_nal = 0;
1621
1622     /* Set output picture properties */
1623     if( h->sh.i_type == SLICE_TYPE_I )
1624         pic_out->i_type = h->i_nal_type == NAL_SLICE_IDR ? X264_TYPE_IDR : X264_TYPE_I;
1625     else if( h->sh.i_type == SLICE_TYPE_P )
1626         pic_out->i_type = X264_TYPE_P;
1627     else
1628         pic_out->i_type = X264_TYPE_B;
1629     pic_out->i_pts = h->fenc->i_pts;
1630
1631     pic_out->img.i_plane = h->fdec->i_plane;
1632     for(i = 0; i < 3; i++)
1633     {
1634         pic_out->img.i_stride[i] = h->fdec->i_stride[i];
1635         pic_out->img.plane[i] = h->fdec->plane[i];
1636     }
1637
1638     /* ---------------------- Update encoder state ------------------------- */
1639
1640     /* update rc */
1641     x264_emms();
1642     x264_ratecontrol_end( h, h->out.i_frame_size * 8 );
1643
1644     /* restore CPU state (before using float again) */
1645     x264_emms();
1646
1647     x264_noise_reduction_update( thread_current );
1648
1649     /* ---------------------- Compute/Print statistics --------------------- */
1650     x264_thread_sync_stat( h, h->thread[0] );
1651
1652     /* Slice stat */
1653     h->stat.i_slice_count[h->sh.i_type]++;
1654     h->stat.i_slice_size[h->sh.i_type] += h->out.i_frame_size + NALU_OVERHEAD;
1655     h->stat.f_slice_qp[h->sh.i_type] += h->fdec->f_qp_avg_aq;
1656
1657     for( i = 0; i < X264_MBTYPE_MAX; i++ )
1658         h->stat.i_mb_count[h->sh.i_type][i] += h->stat.frame.i_mb_count[i];
1659     for( i = 0; i < X264_PARTTYPE_MAX; i++ )
1660         h->stat.i_mb_partition[h->sh.i_type][i] += h->stat.frame.i_mb_partition[i];
1661     for( i = 0; i < 2; i++ )
1662         h->stat.i_mb_count_8x8dct[i] += h->stat.frame.i_mb_count_8x8dct[i];
1663     if( h->sh.i_type != SLICE_TYPE_I )
1664         for( i_list = 0; i_list < 2; i_list++ )
1665             for( i = 0; i < 32; i++ )
1666                 h->stat.i_mb_count_ref[h->sh.i_type][i_list][i] += h->stat.frame.i_mb_count_ref[i_list][i];
1667     if( h->sh.i_type == SLICE_TYPE_P )
1668         h->stat.i_consecutive_bframes[h->fdec->i_frame - h->fref0[0]->i_frame - 1]++;
1669     if( h->sh.i_type == SLICE_TYPE_B )
1670     {
1671         h->stat.i_direct_frames[ h->sh.b_direct_spatial_mv_pred ] ++;
1672         if( h->mb.b_direct_auto_write )
1673         {
1674             //FIXME somewhat arbitrary time constants
1675             if( h->stat.i_direct_score[0] + h->stat.i_direct_score[1] > h->mb.i_mb_count )
1676             {
1677                 for( i = 0; i < 2; i++ )
1678                     h->stat.i_direct_score[i] = h->stat.i_direct_score[i] * 9/10;
1679             }
1680             for( i = 0; i < 2; i++ )
1681                 h->stat.i_direct_score[i] += h->stat.frame.i_direct_score[i];
1682         }
1683     }
1684
1685     psz_message[0] = '\0';
1686     if( h->param.analyse.b_psnr )
1687     {
1688         int64_t ssd[3] = {
1689             h->stat.frame.i_ssd[0],
1690             h->stat.frame.i_ssd[1],
1691             h->stat.frame.i_ssd[2],
1692         };
1693
1694         h->stat.i_ssd_global[h->sh.i_type] += ssd[0] + ssd[1] + ssd[2];
1695         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 );
1696         h->stat.f_psnr_mean_y[h->sh.i_type] += x264_psnr( ssd[0], h->param.i_width * h->param.i_height );
1697         h->stat.f_psnr_mean_u[h->sh.i_type] += x264_psnr( ssd[1], h->param.i_width * h->param.i_height / 4 );
1698         h->stat.f_psnr_mean_v[h->sh.i_type] += x264_psnr( ssd[2], h->param.i_width * h->param.i_height / 4 );
1699
1700         snprintf( psz_message, 80, " PSNR Y:%5.2f U:%5.2f V:%5.2f",
1701                   x264_psnr( ssd[0], h->param.i_width * h->param.i_height ),
1702                   x264_psnr( ssd[1], h->param.i_width * h->param.i_height / 4),
1703                   x264_psnr( ssd[2], h->param.i_width * h->param.i_height / 4) );
1704     }
1705
1706     if( h->param.analyse.b_ssim )
1707     {
1708         double ssim_y = h->stat.frame.f_ssim
1709                       / (((h->param.i_width-6)>>2) * ((h->param.i_height-6)>>2));
1710         h->stat.f_ssim_mean_y[h->sh.i_type] += ssim_y;
1711         snprintf( psz_message + strlen(psz_message), 80 - strlen(psz_message),
1712                   " SSIM Y:%.5f", ssim_y );
1713     }
1714     psz_message[79] = '\0';
1715
1716     x264_log( h, X264_LOG_DEBUG,
1717                   "frame=%4d QP=%.2f NAL=%d Slice:%c Poc:%-3d I:%-4d P:%-4d SKIP:%-4d size=%d bytes%s\n",
1718               h->i_frame,
1719               h->fdec->f_qp_avg_aq,
1720               h->i_nal_ref_idc,
1721               h->sh.i_type == SLICE_TYPE_I ? 'I' : (h->sh.i_type == SLICE_TYPE_P ? 'P' : 'B' ),
1722               h->fdec->i_poc,
1723               h->stat.frame.i_mb_count_i,
1724               h->stat.frame.i_mb_count_p,
1725               h->stat.frame.i_mb_count_skip,
1726               h->out.i_frame_size,
1727               psz_message );
1728
1729     // keep stats all in one place
1730     x264_thread_sync_stat( h->thread[0], h );
1731     // for the use of the next frame
1732     x264_thread_sync_stat( thread_current, h );
1733
1734 #ifdef DEBUG_MB_TYPE
1735 {
1736     static const char mb_chars[] = { 'i', 'i', 'I', 'C', 'P', '8', 'S',
1737         'D', '<', 'X', 'B', 'X', '>', 'B', 'B', 'B', 'B', '8', 'S' };
1738     int mb_xy;
1739     for( mb_xy = 0; mb_xy < h->sps->i_mb_width * h->sps->i_mb_height; mb_xy++ )
1740     {
1741         if( h->mb.type[mb_xy] < X264_MBTYPE_MAX && h->mb.type[mb_xy] >= 0 )
1742             fprintf( stderr, "%c ", mb_chars[ h->mb.type[mb_xy] ] );
1743         else
1744             fprintf( stderr, "? " );
1745
1746         if( (mb_xy+1) % h->sps->i_mb_width == 0 )
1747             fprintf( stderr, "\n" );
1748     }
1749 }
1750 #endif
1751
1752     if( h->param.psz_dump_yuv )
1753         x264_frame_dump( h );
1754 }
1755
1756 static void x264_print_intra( int64_t *i_mb_count, double i_count, int b_print_pcm, char *intra )
1757 {
1758     intra += sprintf( intra, "I16..4%s: %4.1f%% %4.1f%% %4.1f%%",
1759         b_print_pcm ? "..PCM" : "",
1760         i_mb_count[I_16x16]/ i_count,
1761         i_mb_count[I_8x8]  / i_count,
1762         i_mb_count[I_4x4]  / i_count );
1763     if( b_print_pcm )
1764         sprintf( intra, " %4.1f%%", i_mb_count[I_PCM]  / i_count );
1765 }
1766
1767 /****************************************************************************
1768  * x264_encoder_close:
1769  ****************************************************************************/
1770 void    x264_encoder_close  ( x264_t *h )
1771 {
1772     int64_t i_yuv_size = 3 * h->param.i_width * h->param.i_height / 2;
1773     int64_t i_mb_count_size[2][7] = {{0}};
1774     char buf[200];
1775     int i, j, i_list, i_type;
1776     int b_print_pcm = h->stat.i_mb_count[SLICE_TYPE_I][I_PCM]
1777                    || h->stat.i_mb_count[SLICE_TYPE_P][I_PCM]
1778                    || h->stat.i_mb_count[SLICE_TYPE_B][I_PCM];
1779
1780     for( i=0; i<h->param.i_threads; i++ )
1781     {
1782         // don't strictly have to wait for the other threads, but it's simpler than canceling them
1783         if( h->thread[i]->b_thread_active )
1784         {
1785             x264_pthread_join( h->thread[i]->thread_handle, NULL );
1786             assert( h->thread[i]->fenc->i_reference_count == 1 );
1787             x264_frame_delete( h->thread[i]->fenc );
1788         }
1789     }
1790
1791     /* Slices used and PSNR */
1792     for( i=0; i<5; i++ )
1793     {
1794         static const int slice_order[] = { SLICE_TYPE_I, SLICE_TYPE_SI, SLICE_TYPE_P, SLICE_TYPE_SP, SLICE_TYPE_B };
1795         static const char *slice_name[] = { "P", "B", "I", "SP", "SI" };
1796         int i_slice = slice_order[i];
1797
1798         if( h->stat.i_slice_count[i_slice] > 0 )
1799         {
1800             const int i_count = h->stat.i_slice_count[i_slice];
1801             if( h->param.analyse.b_psnr )
1802             {
1803                 x264_log( h, X264_LOG_INFO,
1804                           "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",
1805                           slice_name[i_slice],
1806                           i_count,
1807                           h->stat.f_slice_qp[i_slice] / i_count,
1808                           (double)h->stat.i_slice_size[i_slice] / i_count,
1809                           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,
1810                           h->stat.f_psnr_average[i_slice] / i_count,
1811                           x264_psnr( h->stat.i_ssd_global[i_slice], i_count * i_yuv_size ) );
1812             }
1813             else
1814             {
1815                 x264_log( h, X264_LOG_INFO,
1816                           "slice %s:%-5d Avg QP:%5.2f  size:%6.0f\n",
1817                           slice_name[i_slice],
1818                           i_count,
1819                           h->stat.f_slice_qp[i_slice] / i_count,
1820                           (double)h->stat.i_slice_size[i_slice] / i_count );
1821             }
1822         }
1823     }
1824     if( h->param.i_bframe && h->stat.i_slice_count[SLICE_TYPE_P] )
1825     {
1826         char *p = buf;
1827         int den = 0;
1828         // weight by number of frames (including the P-frame) that are in a sequence of N B-frames
1829         for( i=0; i<=h->param.i_bframe; i++ )
1830             den += (i+1) * h->stat.i_consecutive_bframes[i];
1831         for( i=0; i<=h->param.i_bframe; i++ )
1832             p += sprintf( p, " %4.1f%%", 100. * (i+1) * h->stat.i_consecutive_bframes[i] / den );
1833         x264_log( h, X264_LOG_INFO, "consecutive B-frames:%s\n", buf );
1834     }
1835
1836     for( i_type = 0; i_type < 2; i_type++ )
1837         for( i = 0; i < X264_PARTTYPE_MAX; i++ )
1838         {
1839             if( i == D_DIRECT_8x8 ) continue; /* direct is counted as its own type */
1840             i_mb_count_size[i_type][x264_mb_partition_pixel_table[i]] += h->stat.i_mb_partition[i_type][i];
1841         }
1842
1843     /* MB types used */
1844     if( h->stat.i_slice_count[SLICE_TYPE_I] > 0 )
1845     {
1846         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_I];
1847         double i_count = h->stat.i_slice_count[SLICE_TYPE_I] * h->mb.i_mb_count / 100.0;
1848         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
1849         x264_log( h, X264_LOG_INFO, "mb I  %s\n", buf );
1850     }
1851     if( h->stat.i_slice_count[SLICE_TYPE_P] > 0 )
1852     {
1853         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_P];
1854         double i_count = h->stat.i_slice_count[SLICE_TYPE_P] * h->mb.i_mb_count / 100.0;
1855         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_P];
1856         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
1857         x264_log( h, X264_LOG_INFO,
1858                   "mb P  %s  P16..4: %4.1f%% %4.1f%% %4.1f%% %4.1f%% %4.1f%%    skip:%4.1f%%\n",
1859                   buf,
1860                   i_mb_size[PIXEL_16x16] / (i_count*4),
1861                   (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
1862                   i_mb_size[PIXEL_8x8] / (i_count*4),
1863                   (i_mb_size[PIXEL_8x4] + i_mb_size[PIXEL_4x8]) / (i_count*4),
1864                   i_mb_size[PIXEL_4x4] / (i_count*4),
1865                   i_mb_count[P_SKIP] / i_count );
1866     }
1867     if( h->stat.i_slice_count[SLICE_TYPE_B] > 0 )
1868     {
1869         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_B];
1870         double i_count = h->stat.i_slice_count[SLICE_TYPE_B] * h->mb.i_mb_count / 100.0;
1871         double i_mb_list_count;
1872         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_B];
1873         int64_t list_count[3] = {0}; /* 0 == L0, 1 == L1, 2 == BI */
1874         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
1875         for( i = 0; i < X264_PARTTYPE_MAX; i++ )
1876             for( j = 0; j < 2; j++ )
1877             {
1878                 int l0 = x264_mb_type_list_table[i][0][j];
1879                 int l1 = x264_mb_type_list_table[i][1][j];
1880                 if( l0 || l1 )
1881                     list_count[l1+l0*l1] += h->stat.i_mb_count[SLICE_TYPE_B][i] * 2;
1882             }
1883         list_count[0] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L0_8x8];
1884         list_count[1] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L1_8x8];
1885         list_count[2] += h->stat.i_mb_partition[SLICE_TYPE_B][D_BI_8x8];
1886         i_mb_count[B_DIRECT] += (h->stat.i_mb_partition[SLICE_TYPE_B][D_DIRECT_8x8]+2)/4;
1887         i_mb_list_count = (list_count[0] + list_count[1] + list_count[2]) / 100.0;
1888         x264_log( h, X264_LOG_INFO,
1889                   "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",
1890                   buf,
1891                   i_mb_size[PIXEL_16x16] / (i_count*4),
1892                   (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
1893                   i_mb_size[PIXEL_8x8] / (i_count*4),
1894                   i_mb_count[B_DIRECT] / i_count,
1895                   i_mb_count[B_SKIP]   / i_count,
1896                   list_count[0] / i_mb_list_count,
1897                   list_count[1] / i_mb_list_count,
1898                   list_count[2] / i_mb_list_count );
1899     }
1900
1901     x264_ratecontrol_summary( h );
1902
1903     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 )
1904     {
1905         const int i_count = h->stat.i_slice_count[SLICE_TYPE_I] +
1906                             h->stat.i_slice_count[SLICE_TYPE_P] +
1907                             h->stat.i_slice_count[SLICE_TYPE_B];
1908         float fps = (float) h->param.i_fps_num / h->param.i_fps_den;
1909 #define SUM3(p) (p[SLICE_TYPE_I] + p[SLICE_TYPE_P] + p[SLICE_TYPE_B])
1910 #define SUM3b(p,o) (p[SLICE_TYPE_I][o] + p[SLICE_TYPE_P][o] + p[SLICE_TYPE_B][o])
1911         float f_bitrate = fps * SUM3(h->stat.i_slice_size) / i_count / 125;
1912
1913         if( h->pps->b_transform_8x8_mode )
1914         {
1915             int64_t i_i8x8 = SUM3b( h->stat.i_mb_count, I_8x8 );
1916             int64_t i_intra = i_i8x8 + SUM3b( h->stat.i_mb_count, I_4x4 )
1917                                      + SUM3b( h->stat.i_mb_count, I_16x16 );
1918             x264_log( h, X264_LOG_INFO, "8x8 transform  intra:%.1f%%  inter:%.1f%%\n",
1919                       100. * i_i8x8 / i_intra,
1920                       100. * h->stat.i_mb_count_8x8dct[1] / h->stat.i_mb_count_8x8dct[0] );
1921         }
1922
1923         if( h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO
1924             && h->stat.i_slice_count[SLICE_TYPE_B] )
1925         {
1926             x264_log( h, X264_LOG_INFO, "direct mvs  spatial:%.1f%%  temporal:%.1f%%\n",
1927                       h->stat.i_direct_frames[1] * 100. / h->stat.i_slice_count[SLICE_TYPE_B],
1928                       h->stat.i_direct_frames[0] * 100. / h->stat.i_slice_count[SLICE_TYPE_B] );
1929         }
1930
1931         for( i_list = 0; i_list < 2; i_list++ )
1932         {
1933             int i_slice;
1934             for( i_slice = 0; i_slice < 2; i_slice++ )
1935             {
1936                 char *p = buf;
1937                 int64_t i_den = 0;
1938                 int i_max = 0;
1939                 for( i = 0; i < 32; i++ )
1940                     if( h->stat.i_mb_count_ref[i_slice][i_list][i] )
1941                     {
1942                         i_den += h->stat.i_mb_count_ref[i_slice][i_list][i];
1943                         i_max = i;
1944                     }
1945                 if( i_max == 0 )
1946                     continue;
1947                 for( i = 0; i <= i_max; i++ )
1948                     p += sprintf( p, " %4.1f%%", 100. * h->stat.i_mb_count_ref[i_slice][i_list][i] / i_den );
1949                 x264_log( h, X264_LOG_INFO, "ref %c L%d %s\n", "PB"[i_slice], i_list, buf );
1950             }
1951         }
1952
1953         if( h->param.analyse.b_ssim )
1954         {
1955             x264_log( h, X264_LOG_INFO,
1956                       "SSIM Mean Y:%.7f\n",
1957                       SUM3( h->stat.f_ssim_mean_y ) / i_count );
1958         }
1959         if( h->param.analyse.b_psnr )
1960         {
1961             x264_log( h, X264_LOG_INFO,
1962                       "PSNR Mean Y:%6.3f U:%6.3f V:%6.3f Avg:%6.3f Global:%6.3f kb/s:%.2f\n",
1963                       SUM3( h->stat.f_psnr_mean_y ) / i_count,
1964                       SUM3( h->stat.f_psnr_mean_u ) / i_count,
1965                       SUM3( h->stat.f_psnr_mean_v ) / i_count,
1966                       SUM3( h->stat.f_psnr_average ) / i_count,
1967                       x264_psnr( SUM3( h->stat.i_ssd_global ), i_count * i_yuv_size ),
1968                       f_bitrate );
1969         }
1970         else
1971             x264_log( h, X264_LOG_INFO, "kb/s:%.1f\n", f_bitrate );
1972     }
1973
1974     /* rc */
1975     x264_ratecontrol_delete( h );
1976
1977     /* param */
1978     if( h->param.rc.psz_stat_out )
1979         free( h->param.rc.psz_stat_out );
1980     if( h->param.rc.psz_stat_in )
1981         free( h->param.rc.psz_stat_in );
1982
1983     x264_cqm_delete( h );
1984
1985     if( h->param.i_threads > 1)
1986         h = h->thread[ h->i_thread_phase % h->param.i_threads ];
1987
1988     /* frames */
1989     for( i = 0; h->frames.current[i]; i++ )
1990     {
1991         assert( h->frames.current[i]->i_reference_count == 1 );
1992         x264_frame_delete( h->frames.current[i] );
1993     }
1994     for( i = 0; h->frames.next[i]; i++ )
1995     {
1996         assert( h->frames.next[i]->i_reference_count == 1 );
1997         x264_frame_delete( h->frames.next[i] );
1998     }
1999     for( i = 0; h->frames.unused[i]; i++ )
2000     {
2001         assert( h->frames.unused[i]->i_reference_count == 0 );
2002         x264_frame_delete( h->frames.unused[i] );
2003     }
2004
2005     h = h->thread[0];
2006
2007     for( i = h->param.i_threads - 1; i >= 0; i-- )
2008     {
2009         x264_frame_t **frame;
2010
2011         for( frame = h->thread[i]->frames.reference; *frame; frame++ )
2012         {
2013             assert( (*frame)->i_reference_count > 0 );
2014             (*frame)->i_reference_count--;
2015             if( (*frame)->i_reference_count == 0 )
2016                 x264_frame_delete( *frame );
2017         }
2018         frame = &h->thread[i]->fdec;
2019         assert( (*frame)->i_reference_count > 0 );
2020         (*frame)->i_reference_count--;
2021         if( (*frame)->i_reference_count == 0 )
2022             x264_frame_delete( *frame );
2023
2024         x264_macroblock_cache_end( h->thread[i] );
2025         x264_free( h->thread[i]->out.p_bitstream );
2026         x264_free( h->thread[i] );
2027     }
2028 }