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_MIN(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     }
 414     if( h->param.rc.i_rc_method == X264_RC_CQP )
 415     {
 416         float qp_p = h->param.rc.i_qp_constant;
 417         float qp_i = qp_p - 6*log(h->param.rc.f_ip_factor)/log(2);
 418         float qp_b = qp_p + 6*log(h->param.rc.f_pb_factor)/log(2);
 419         h->param.rc.i_qp_min = x264_clip3( (int)(X264_MIN3( qp_p, qp_i, qp_b )), 0, 51 );
 420         h->param.rc.i_qp_max = x264_clip3( (int)(X264_MAX3( qp_p, qp_i, qp_b ) + .999), 0, 51 );
 421         h->param.rc.i_aq_mode = 0;
 422     }
 423     h->param.rc.i_qp_max = x264_clip3( h->param.rc.i_qp_max, 0, 51 );
 424     h->param.rc.i_qp_min = x264_clip3( h->param.rc.i_qp_min, 0, h->param.rc.i_qp_max );
 425
 426     if( ( h->param.i_width % 16 || h->param.i_height % 16 )
 427         && h->param.i_height != 1080 && !h->mb.b_lossless )
 428     {
 429         // There's nothing special about 1080 in that the warning still applies to it,
 430         // but chances are the user can't help it if his content is already 1080p,
 431         // so there's no point in warning in that case.
 432         x264_log( h, X264_LOG_WARNING,
 433                   "width or height not divisible by 16 (%dx%d), compression will suffer.\n",
 434                   h->param.i_width, h->param.i_height );
 435     }
 436
 437     h->param.i_frame_reference = x264_clip3( h->param.i_frame_reference, 1, 16 );
 438     if( h->param.i_keyint_max <= 0 )
 439         h->param.i_keyint_max = 1;
 440     h->param.i_keyint_min = x264_clip3( h->param.i_keyint_min, 1, h->param.i_keyint_max/2+1 );
 441
 442     h->param.i_bframe = x264_clip3( h->param.i_bframe, 0, X264_BFRAME_MAX );
 443     h->param.i_bframe_bias = x264_clip3( h->param.i_bframe_bias, -90, 100 );
 444     h->param.b_bframe_pyramid = h->param.b_bframe_pyramid && h->param.i_bframe > 1;
 445     h->param.b_bframe_adaptive = h->param.b_bframe_adaptive && h->param.i_bframe > 0;
 446     h->param.analyse.b_weighted_bipred = h->param.analyse.b_weighted_bipred && h->param.i_bframe > 0;
 447     h->mb.b_direct_auto_write = h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO
 448                                 && h->param.i_bframe
 449                                 && ( h->param.rc.b_stat_write || !h->param.rc.b_stat_read );
 450     if( h->param.i_scenecut_threshold < 0 )
 451         h->param.b_pre_scenecut = 0;
 452
 453     h->param.i_deblocking_filter_alphac0 = x264_clip3( h->param.i_deblocking_filter_alphac0, -6, 6 );
 454     h->param.i_deblocking_filter_beta    = x264_clip3( h->param.i_deblocking_filter_beta, -6, 6 );
 455     h->param.analyse.i_luma_deadzone[0] = x264_clip3( h->param.analyse.i_luma_deadzone[0], 0, 32 );
 456     h->param.analyse.i_luma_deadzone[1] = x264_clip3( h->param.analyse.i_luma_deadzone[1], 0, 32 );
 457
 458     h->param.i_cabac_init_idc = x264_clip3( h->param.i_cabac_init_idc, 0, 2 );
 459
 460     if( h->param.i_cqm_preset < X264_CQM_FLAT || h->param.i_cqm_preset > X264_CQM_CUSTOM )
 461         h->param.i_cqm_preset = X264_CQM_FLAT;
 462
 463     if( h->param.analyse.i_me_method < X264_ME_DIA ||
 464         h->param.analyse.i_me_method > X264_ME_TESA )
 465         h->param.analyse.i_me_method = X264_ME_HEX;
 466     if( h->param.analyse.i_me_range < 4 )
 467         h->param.analyse.i_me_range = 4;
 468     if( h->param.analyse.i_me_range > 16 && h->param.analyse.i_me_method <= X264_ME_HEX )
 469         h->param.analyse.i_me_range = 16;
 470     if( h->param.analyse.i_me_method == X264_ME_TESA &&
 471         (h->mb.b_lossless || h->param.analyse.i_subpel_refine <= 1) )
 472         h->param.analyse.i_me_method = X264_ME_ESA;
 473     h->param.analyse.i_subpel_refine = x264_clip3( h->param.analyse.i_subpel_refine, 1, 7 );
 474     h->param.analyse.b_bframe_rdo = h->param.analyse.b_bframe_rdo && h->param.analyse.i_subpel_refine >= 6;
 475     h->param.analyse.b_mixed_references = h->param.analyse.b_mixed_references && h->param.i_frame_reference > 1;
 476     h->param.analyse.inter &= X264_ANALYSE_PSUB16x16|X264_ANALYSE_PSUB8x8|X264_ANALYSE_BSUB16x16|
 477                               X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
 478     h->param.analyse.intra &= X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
 479     if( !(h->param.analyse.inter & X264_ANALYSE_PSUB16x16) )
 480         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
 481     if( !h->param.analyse.b_transform_8x8 )
 482     {
 483         h->param.analyse.inter &= ~X264_ANALYSE_I8x8;
 484         h->param.analyse.intra &= ~X264_ANALYSE_I8x8;
 485     }
 486     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -12, 12);
 487     if( !h->param.b_cabac )
 488         h->param.analyse.i_trellis = 0;
 489     h->param.analyse.i_trellis = x264_clip3( h->param.analyse.i_trellis, 0, 2 );
 490     h->param.rc.i_aq_mode = x264_clip3( h->param.rc.i_aq_mode, 0, 2 );
 491     if( h->param.rc.f_aq_strength <= 0 )
 492         h->param.rc.i_aq_mode = 0;
 493     /* VAQ effectively replaces qcomp, so qcomp is raised towards 1 to compensate. */
 494     if( h->param.rc.i_aq_mode == X264_AQ_GLOBAL )
 495         h->param.rc.f_qcompress = x264_clip3f(h->param.rc.f_qcompress + h->param.rc.f_aq_strength / 0.7, 0, 1);
 496     h->param.analyse.i_noise_reduction = x264_clip3( h->param.analyse.i_noise_reduction, 0, 1<<16 );
 497
 498     {
 499         const x264_level_t *l = x264_levels;
 500         if( h->param.i_level_idc < 0 )
 501         {
 502             if( h->param.rc.i_rc_method == X264_RC_ABR && h->param.rc.i_vbv_buffer_size <= 0 )
 503                 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate * 2;
 504             h->sps = h->sps_array;
 505             x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
 506             do h->param.i_level_idc = l->level_idc;
 507                 while( l[1].level_idc && x264_validate_levels( h, 0 ) && l++ );
 508             if( h->param.rc.i_vbv_buffer_size <= 0 )
 509                 h->param.rc.i_vbv_max_bitrate = 0;
 510             x264_log( h, X264_LOG_DEBUG, "level_idc: %d\n", h->param.i_level_idc );
 511         }
 512         else
 513         {
 514             while( l->level_idc && l->level_idc != h->param.i_level_idc )
 515                 l++;
 516             if( l->level_idc == 0 )
 517             {
 518                 x264_log( h, X264_LOG_ERROR, "invalid level_idc: %d\n", h->param.i_level_idc );
 519                 return -1;
 520             }
 521         }
 522         if( h->param.analyse.i_mv_range <= 0 )
 523             h->param.analyse.i_mv_range = l->mv_range >> h->param.b_interlaced;
 524         else
 525             h->param.analyse.i_mv_range = x264_clip3(h->param.analyse.i_mv_range, 32, 512 >> h->param.b_interlaced);
 526         if( h->param.analyse.i_direct_8x8_inference < 0 )
 527             h->param.analyse.i_direct_8x8_inference = l->direct8x8;
 528     }
 529
 530     if( h->param.i_threads > 1 )
 531     {
 532         int r = h->param.analyse.i_mv_range_thread;
 533         int r2;
 534         if( r <= 0 )
 535         {
 536             // half of the available space is reserved and divided evenly among the threads,
 537             // the rest is allocated to whichever thread is far enough ahead to use it.
 538             // reserving more space increases quality for some videos, but costs more time
 539             // in thread synchronization.
 540             int max_range = (h->param.i_height + X264_THREAD_HEIGHT) / h->param.i_threads - X264_THREAD_HEIGHT;
 541             r = max_range / 2;
 542         }
 543         r = X264_MAX( r, h->param.analyse.i_me_range );
 544         r = X264_MIN( r, h->param.analyse.i_mv_range );
 545         // round up to use the whole mb row
 546         r2 = (r & ~15) + ((-X264_THREAD_HEIGHT) & 15);
 547         if( r2 < r )
 548             r2 += 16;
 549         x264_log( h, X264_LOG_DEBUG, "using mv_range_thread = %d\n", r2 );
 550         h->param.analyse.i_mv_range_thread = r2;
 551     }
 552
 553     if( h->param.rc.f_qblur < 0 )
 554         h->param.rc.f_qblur = 0;
 555     if( h->param.rc.f_complexity_blur < 0 )
 556         h->param.rc.f_complexity_blur = 0;
 557
 558     h->param.i_sps_id &= 31;
 559
 560     if( h->param.i_log_level < X264_LOG_INFO )
 561     {
 562         h->param.analyse.b_psnr = 0;
 563         h->param.analyse.b_ssim = 0;
 564     }
 565
 566     /* ensure the booleans are 0 or 1 so they can be used in math */
 567 #define BOOLIFY(x) h->param.x = !!h->param.x
 568     BOOLIFY( b_cabac );
 569     BOOLIFY( b_deblocking_filter );
 570     BOOLIFY( b_interlaced );
 571     BOOLIFY( analyse.b_transform_8x8 );
 572     BOOLIFY( analyse.i_direct_8x8_inference );
 573     BOOLIFY( analyse.b_bidir_me );
 574     BOOLIFY( analyse.b_chroma_me );
 575     BOOLIFY( analyse.b_fast_pskip );
 576     BOOLIFY( rc.b_stat_write );
 577     BOOLIFY( rc.b_stat_read );
 578 #undef BOOLIFY
 579
 580     return 0;
 581 }
 582
 583 static void mbcmp_init( x264_t *h )
 584 {
 585     int satd = !h->mb.b_lossless && h->param.analyse.i_subpel_refine > 1;
 586     memcpy( h->pixf.mbcmp, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.mbcmp) );
 587     satd &= h->param.analyse.i_me_method == X264_ME_TESA;
 588     memcpy( h->pixf.fpelcmp, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.fpelcmp) );
 589     memcpy( h->pixf.fpelcmp_x3, satd ? h->pixf.satd_x3 : h->pixf.sad_x3, sizeof(h->pixf.fpelcmp_x3) );
 590     memcpy( h->pixf.fpelcmp_x4, satd ? h->pixf.satd_x4 : h->pixf.sad_x4, sizeof(h->pixf.fpelcmp_x4) );
 591 }
 592
 593 /****************************************************************************
 594  * x264_encoder_open:
 595  ****************************************************************************/
 596 x264_t *x264_encoder_open   ( x264_param_t *param )
 597 {
 598     x264_t *h = x264_malloc( sizeof( x264_t ) );
 599     char buf[1000], *p;
 600     int i;
 601
 602     memset( h, 0, sizeof( x264_t ) );
 603
 604     /* Create a copy of param */
 605     memcpy( &h->param, param, sizeof( x264_param_t ) );
 606
 607     if( x264_validate_parameters( h ) < 0 )
 608     {
 609         x264_free( h );
 610         return NULL;
 611     }
 612
 613     if( h->param.psz_cqm_file )
 614         if( x264_cqm_parse_file( h, h->param.psz_cqm_file ) < 0 )
 615         {
 616             x264_free( h );
 617             return NULL;
 618         }
 619
 620     if( h->param.rc.psz_stat_out )
 621         h->param.rc.psz_stat_out = strdup( h->param.rc.psz_stat_out );
 622     if( h->param.rc.psz_stat_in )
 623         h->param.rc.psz_stat_in = strdup( h->param.rc.psz_stat_in );
 624     if( h->param.rc.psz_rc_eq )
 625         h->param.rc.psz_rc_eq = strdup( h->param.rc.psz_rc_eq );
 626
 627     /* VUI */
 628     if( h->param.vui.i_sar_width > 0 && h->param.vui.i_sar_height > 0 )
 629     {
 630         int i_w = param->vui.i_sar_width;
 631         int i_h = param->vui.i_sar_height;
 632
 633         x264_reduce_fraction( &i_w, &i_h );
 634
 635         while( i_w > 65535 || i_h > 65535 )
 636         {
 637             i_w /= 2;
 638             i_h /= 2;
 639         }
 640
 641         h->param.vui.i_sar_width = 0;
 642         h->param.vui.i_sar_height = 0;
 643         if( i_w == 0 || i_h == 0 )
 644         {
 645             x264_log( h, X264_LOG_WARNING, "cannot create valid sample aspect ratio\n" );
 646         }
 647         else
 648         {
 649             x264_log( h, X264_LOG_INFO, "using SAR=%d/%d\n", i_w, i_h );
 650             h->param.vui.i_sar_width = i_w;
 651             h->param.vui.i_sar_height = i_h;
 652         }
 653     }
 654
 655     x264_reduce_fraction( &h->param.i_fps_num, &h->param.i_fps_den );
 656
 657     /* Init x264_t */
 658     h->i_frame = 0;
 659     h->i_frame_num = 0;
 660     h->i_idr_pic_id = 0;
 661
 662     h->sps = &h->sps_array[0];
 663     x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
 664
 665     h->pps = &h->pps_array[0];
 666     x264_pps_init( h->pps, h->param.i_sps_id, &h->param, h->sps);
 667
 668     x264_validate_levels( h, 1 );
 669
 670     if( x264_cqm_init( h ) < 0 )
 671     {
 672         x264_free( h );
 673         return NULL;
 674     }
 675
 676     h->mb.i_mb_count = h->sps->i_mb_width * h->sps->i_mb_height;
 677
 678     /* Init frames. */
 679     h->frames.i_delay = h->param.i_bframe + h->param.i_threads - 1;
 680     h->frames.i_max_ref0 = h->param.i_frame_reference;
 681     h->frames.i_max_ref1 = h->sps->vui.i_num_reorder_frames;
 682     h->frames.i_max_dpb  = h->sps->vui.i_max_dec_frame_buffering;
 683     h->frames.b_have_lowres = !h->param.rc.b_stat_read
 684         && ( h->param.rc.i_rc_method == X264_RC_ABR
 685           || h->param.rc.i_rc_method == X264_RC_CRF
 686           || h->param.b_bframe_adaptive
 687           || h->param.b_pre_scenecut );
 688     h->frames.b_have_lowres |= (h->param.rc.b_stat_read && h->param.rc.i_vbv_buffer_size > 0);
 689
 690     h->frames.i_last_idr = - h->param.i_keyint_max;
 691     h->frames.i_input    = 0;
 692     h->frames.last_nonb  = NULL;
 693
 694     h->i_ref0 = 0;
 695     h->i_ref1 = 0;
 696
 697     h->chroma_qp_table = i_chroma_qp_table + 12 + h->pps->i_chroma_qp_index_offset;
 698
 699     x264_rdo_init( );
 700
 701     /* init CPU functions */
 702     x264_predict_16x16_init( h->param.cpu, h->predict_16x16 );
 703     x264_predict_8x8c_init( h->param.cpu, h->predict_8x8c );
 704     x264_predict_8x8_init( h->param.cpu, h->predict_8x8 );
 705     x264_predict_4x4_init( h->param.cpu, h->predict_4x4 );
 706
 707     x264_pixel_init( h->param.cpu, &h->pixf );
 708     x264_dct_init( h->param.cpu, &h->dctf );
 709     x264_zigzag_init( h->param.cpu, &h->zigzagf, h->param.b_interlaced );
 710     x264_mc_init( h->param.cpu, &h->mc );
 711     x264_quant_init( h, h->param.cpu, &h->quantf );
 712     x264_deblock_init( h->param.cpu, &h->loopf );
 713     x264_dct_init_weights();
 714
 715     mbcmp_init( h );
 716
 717     p = buf + sprintf( buf, "using cpu capabilities:" );
 718     for( i=0; x264_cpu_names[i].flags; i++ )
 719     {
 720         if( !strcmp(x264_cpu_names[i].name, "SSE2")
 721             && param->cpu & (X264_CPU_SSE2_IS_FAST|X264_CPU_SSE2_IS_SLOW) )
 722             continue;
 723         if( !strcmp(x264_cpu_names[i].name, "SSE3")
 724             && (param->cpu & X264_CPU_SSSE3 || !(param->cpu & X264_CPU_CACHELINE_64)) )
 725             continue;
 726         if( (param->cpu & x264_cpu_names[i].flags) == x264_cpu_names[i].flags
 727             && (!i || x264_cpu_names[i].flags != x264_cpu_names[i-1].flags) )
 728             p += sprintf( p, " %s", x264_cpu_names[i].name );
 729     }
 730     if( !param->cpu )
 731         p += sprintf( p, " none!" );
 732     x264_log( h, X264_LOG_INFO, "%s\n", buf );
 733
 734     h->out.i_nal = 0;
 735     h->out.i_bitstream = X264_MAX( 1000000, h->param.i_width * h->param.i_height * 4
 736         * ( h->param.rc.i_rc_method == X264_RC_ABR ? pow( 0.95, h->param.rc.i_qp_min )
 737           : pow( 0.95, h->param.rc.i_qp_constant ) * X264_MAX( 1, h->param.rc.f_ip_factor )));
 738
 739     h->thread[0] = h;
 740     h->i_thread_num = 0;
 741     for( i = 1; i < h->param.i_threads; i++ )
 742         h->thread[i] = x264_malloc( sizeof(x264_t) );
 743
 744     for( i = 0; i < h->param.i_threads; i++ )
 745     {
 746         if( i > 0 )
 747             *h->thread[i] = *h;
 748         h->thread[i]->fdec = x264_frame_pop_unused( h );
 749         h->thread[i]->out.p_bitstream = x264_malloc( h->out.i_bitstream );
 750         if( x264_macroblock_cache_init( h->thread[i] ) < 0 )
 751             return NULL;
 752     }
 753
 754     if( x264_ratecontrol_new( h ) < 0 )
 755         return NULL;
 756
 757     if( h->param.psz_dump_yuv )
 758     {
 759         /* create or truncate the reconstructed video file */
 760         FILE *f = fopen( h->param.psz_dump_yuv, "w" );
 761         if( f )
 762             fclose( f );
 763         else
 764         {
 765             x264_log( h, X264_LOG_ERROR, "can't write to fdec.yuv\n" );
 766             x264_free( h );
 767             return NULL;
 768         }
 769     }
 770
 771     return h;
 772 }
 773
 774 /****************************************************************************
 775  * x264_encoder_reconfig:
 776  ****************************************************************************/
 777 int x264_encoder_reconfig( x264_t *h, x264_param_t *param )
 778 {
 779 #define COPY(var) h->param.var = param->var
 780     COPY( i_frame_reference ); // but never uses more refs than initially specified
 781     COPY( i_bframe_bias );
 782     if( h->param.i_scenecut_threshold >= 0 && param->i_scenecut_threshold >= 0 )
 783         COPY( i_scenecut_threshold ); // can't turn it on or off, only vary the threshold
 784     COPY( b_deblocking_filter );
 785     COPY( i_deblocking_filter_alphac0 );
 786     COPY( i_deblocking_filter_beta );
 787     COPY( analyse.intra );
 788     COPY( analyse.inter );
 789     COPY( analyse.i_direct_mv_pred );
 790     COPY( analyse.i_me_method );
 791     COPY( analyse.i_me_range );
 792     COPY( analyse.i_noise_reduction );
 793     COPY( analyse.i_subpel_refine );
 794     COPY( analyse.i_trellis );
 795     COPY( analyse.b_bidir_me );
 796     COPY( analyse.b_bframe_rdo );
 797     COPY( analyse.b_chroma_me );
 798     COPY( analyse.b_dct_decimate );
 799     COPY( analyse.b_fast_pskip );
 800     COPY( analyse.b_mixed_references );
 801     // can only twiddle these if they were enabled to begin with:
 802     if( h->pps->b_transform_8x8_mode )
 803         COPY( analyse.b_transform_8x8 );
 804     if( h->frames.i_max_ref1 > 1 )
 805         COPY( b_bframe_pyramid );
 806 #undef COPY
 807
 808     mbcmp_init( h );
 809
 810     return x264_validate_parameters( h );
 811 }
 812
 813 /* internal usage */
 814 static void x264_nal_start( x264_t *h, int i_type, int i_ref_idc )
 815 {
 816     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
 817
 818     nal->i_ref_idc = i_ref_idc;
 819     nal->i_type    = i_type;
 820
 821     nal->i_payload= 0;
 822     nal->p_payload= &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8];
 823 }
 824 static void x264_nal_end( x264_t *h )
 825 {
 826     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
 827     nal->i_payload = &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8] - nal->p_payload;
 828     h->out.i_nal++;
 829 }
 830
 831 /****************************************************************************
 832  * x264_encoder_headers:
 833  ****************************************************************************/
 834 int x264_encoder_headers( x264_t *h, x264_nal_t **pp_nal, int *pi_nal )
 835 {
 836     /* init bitstream context */
 837     h->out.i_nal = 0;
 838     bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
 839
 840     /* Put SPS and PPS */
 841     if( h->i_frame == 0 )
 842     {
 843         /* identify ourself */
 844         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
 845         x264_sei_version_write( h, &h->out.bs );
 846         x264_nal_end( h );
 847
 848         /* generate sequence parameters */
 849         x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
 850         x264_sps_write( &h->out.bs, h->sps );
 851         x264_nal_end( h );
 852
 853         /* generate picture parameters */
 854         x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
 855         x264_pps_write( &h->out.bs, h->pps );
 856         x264_nal_end( h );
 857     }
 858     /* now set output*/
 859     *pi_nal = h->out.i_nal;
 860     *pp_nal = &h->out.nal[0];
 861     h->out.i_nal = 0;
 862
 863     return 0;
 864 }
 865
 866 static inline void x264_reference_build_list( x264_t *h, int i_poc )
 867 {
 868     int i;
 869     int b_ok;
 870
 871     /* build ref list 0/1 */
 872     h->i_ref0 = 0;
 873     h->i_ref1 = 0;
 874     for( i = 0; h->frames.reference[i]; i++ )
 875     {
 876         if( h->frames.reference[i]->i_poc < i_poc )
 877         {
 878             h->fref0[h->i_ref0++] = h->frames.reference[i];
 879         }
 880         else if( h->frames.reference[i]->i_poc > i_poc )
 881         {
 882             h->fref1[h->i_ref1++] = h->frames.reference[i];
 883         }
 884     }
 885
 886     /* Order ref0 from higher to lower poc */
 887     do
 888     {
 889         b_ok = 1;
 890         for( i = 0; i < h->i_ref0 - 1; i++ )
 891         {
 892             if( h->fref0[i]->i_poc < h->fref0[i+1]->i_poc )
 893             {
 894                 XCHG( x264_frame_t*, h->fref0[i], h->fref0[i+1] );
 895                 b_ok = 0;
 896                 break;
 897             }
 898         }
 899     } while( !b_ok );
 900     /* Order ref1 from lower to higher poc (bubble sort) for B-frame */
 901     do
 902     {
 903         b_ok = 1;
 904         for( i = 0; i < h->i_ref1 - 1; i++ )
 905         {
 906             if( h->fref1[i]->i_poc > h->fref1[i+1]->i_poc )
 907             {
 908                 XCHG( x264_frame_t*, h->fref1[i], h->fref1[i+1] );
 909                 b_ok = 0;
 910                 break;
 911             }
 912         }
 913     } while( !b_ok );
 914
 915     /* In the standard, a P-frame's ref list is sorted by frame_num.
 916      * We use POC, but check whether explicit reordering is needed */
 917     h->b_ref_reorder[0] =
 918     h->b_ref_reorder[1] = 0;
 919     if( h->sh.i_type == SLICE_TYPE_P )
 920     {
 921         for( i = 0; i < h->i_ref0 - 1; i++ )
 922             if( h->fref0[i]->i_frame_num < h->fref0[i+1]->i_frame_num )
 923             {
 924                 h->b_ref_reorder[0] = 1;
 925                 break;
 926             }
 927     }
 928
 929     h->i_ref1 = X264_MIN( h->i_ref1, h->frames.i_max_ref1 );
 930     h->i_ref0 = X264_MIN( h->i_ref0, h->frames.i_max_ref0 );
 931     h->i_ref0 = X264_MIN( h->i_ref0, h->param.i_frame_reference ); // if reconfig() has lowered the limit
 932     assert( h->i_ref0 + h->i_ref1 <= 16 );
 933     h->mb.pic.i_fref[0] = h->i_ref0;
 934     h->mb.pic.i_fref[1] = h->i_ref1;
 935 }
 936
 937 static void x264_fdec_filter_row( x264_t *h, int mb_y )
 938 {
 939     /* mb_y is the mb to be encoded next, not the mb to be filtered here */
 940     int b_hpel = h->fdec->b_kept_as_ref;
 941     int b_deblock = !h->sh.i_disable_deblocking_filter_idc;
 942     int b_end = mb_y == h->sps->i_mb_height;
 943     int min_y = mb_y - (1 << h->sh.b_mbaff);
 944     int max_y = b_end ? h->sps->i_mb_height : mb_y;
 945     b_deblock &= b_hpel || h->param.psz_dump_yuv;
 946     if( mb_y & h->sh.b_mbaff )
 947         return;
 948     if( min_y < 0 )
 949         return;
 950
 951     if( !b_end )
 952     {
 953         int i, j;
 954         for( j=0; j<=h->sh.b_mbaff; j++ )
 955             for( i=0; i<3; i++ )
 956             {
 957                 memcpy( h->mb.intra_border_backup[j][i],
 958                         h->fdec->plane[i] + ((mb_y*16 >> !!i) + j - 1 - h->sh.b_mbaff) * h->fdec->i_stride[i],
 959                         h->sps->i_mb_width*16 >> !!i );
 960             }
 961     }
 962
 963     if( b_deblock )
 964     {
 965         int y;
 966         for( y = min_y; y < max_y; y += (1 << h->sh.b_mbaff) )
 967             x264_frame_deblock_row( h, y );
 968     }
 969
 970     if( b_hpel )
 971     {
 972         x264_frame_expand_border( h, h->fdec, min_y, b_end );
 973         x264_frame_filter( h, h->fdec, min_y, b_end );
 974         x264_frame_expand_border_filtered( h, h->fdec, min_y, b_end );
 975     }
 976
 977     if( h->param.i_threads > 1 && h->fdec->b_kept_as_ref )
 978     {
 979         x264_frame_cond_broadcast( h->fdec, mb_y*16 + (b_end ? 10000 : -(X264_THREAD_HEIGHT << h->sh.b_mbaff)) );
 980     }
 981
 982     min_y = X264_MAX( min_y*16-8, 0 );
 983     max_y = b_end ? h->param.i_height : mb_y*16-8;
 984
 985     if( h->param.analyse.b_psnr )
 986     {
 987         int i;
 988         for( i=0; i<3; i++ )
 989             h->stat.frame.i_ssd[i] +=
 990                 x264_pixel_ssd_wxh( &h->pixf,
 991                     h->fdec->plane[i] + (min_y>>!!i) * h->fdec->i_stride[i], h->fdec->i_stride[i],
 992                     h->fenc->plane[i] + (min_y>>!!i) * h->fenc->i_stride[i], h->fenc->i_stride[i],
 993                     h->param.i_width >> !!i, (max_y-min_y) >> !!i );
 994     }
 995
 996     if( h->param.analyse.b_ssim )
 997     {
 998         x264_emms();
 999         /* offset by 2 pixels to avoid alignment of ssim blocks with dct blocks,
1000          * and overlap by 4 */
1001         min_y += min_y == 0 ? 2 : -6;
1002         h->stat.frame.f_ssim +=
1003             x264_pixel_ssim_wxh( &h->pixf,
1004                 h->fdec->plane[0] + 2+min_y*h->fdec->i_stride[0], h->fdec->i_stride[0],
1005                 h->fenc->plane[0] + 2+min_y*h->fenc->i_stride[0], h->fenc->i_stride[0],
1006                 h->param.i_width-2, max_y-min_y );
1007     }
1008 }
1009
1010 static inline void x264_reference_update( x264_t *h )
1011 {
1012     int i;
1013
1014     if( h->fdec->i_frame >= 0 )
1015         h->i_frame++;
1016
1017     if( !h->fdec->b_kept_as_ref )
1018     {
1019         if( h->param.i_threads > 1 )
1020         {
1021             x264_frame_push_unused( h, h->fdec );
1022             h->fdec = x264_frame_pop_unused( h );
1023         }
1024         return;
1025     }
1026
1027     /* move lowres copy of the image to the ref frame */
1028     for( i = 0; i < 4; i++)
1029     {
1030         XCHG( uint8_t*, h->fdec->lowres[i], h->fenc->lowres[i] );
1031         XCHG( uint8_t*, h->fdec->buffer_lowres[i], h->fenc->buffer_lowres[i] );
1032     }
1033
1034     /* adaptive B decision needs a pointer, since it can't use the ref lists */
1035     if( h->sh.i_type != SLICE_TYPE_B )
1036         h->frames.last_nonb = h->fdec;
1037
1038     /* move frame in the buffer */
1039     x264_frame_push( h->frames.reference, h->fdec );
1040     if( h->frames.reference[h->frames.i_max_dpb] )
1041         x264_frame_push_unused( h, x264_frame_shift( h->frames.reference ) );
1042     h->fdec = x264_frame_pop_unused( h );
1043 }
1044
1045 static inline void x264_reference_reset( x264_t *h )
1046 {
1047     while( h->frames.reference[0] )
1048         x264_frame_push_unused( h, x264_frame_pop( h->frames.reference ) );
1049     h->fdec->i_poc =
1050     h->fenc->i_poc = 0;
1051 }
1052
1053 static inline void x264_slice_init( x264_t *h, int i_nal_type, int i_global_qp )
1054 {
1055     /* ------------------------ Create slice header  ----------------------- */
1056     if( i_nal_type == NAL_SLICE_IDR )
1057     {
1058         x264_slice_header_init( h, &h->sh, h->sps, h->pps, h->i_idr_pic_id, h->i_frame_num, i_global_qp );
1059
1060         /* increment id */
1061         h->i_idr_pic_id = ( h->i_idr_pic_id + 1 ) % 65536;
1062     }
1063     else
1064     {
1065         x264_slice_header_init( h, &h->sh, h->sps, h->pps, -1, h->i_frame_num, i_global_qp );
1066
1067         /* always set the real higher num of ref frame used */
1068         h->sh.b_num_ref_idx_override = 1;
1069         h->sh.i_num_ref_idx_l0_active = h->i_ref0 <= 0 ? 1 : h->i_ref0;
1070         h->sh.i_num_ref_idx_l1_active = h->i_ref1 <= 0 ? 1 : h->i_ref1;
1071     }
1072
1073     h->fdec->i_frame_num = h->sh.i_frame_num;
1074
1075     if( h->sps->i_poc_type == 0 )
1076     {
1077         h->sh.i_poc_lsb = h->fdec->i_poc & ( (1 << h->sps->i_log2_max_poc_lsb) - 1 );
1078         h->sh.i_delta_poc_bottom = 0;   /* XXX won't work for field */
1079     }
1080     else if( h->sps->i_poc_type == 1 )
1081     {
1082         /* FIXME TODO FIXME */
1083     }
1084     else
1085     {
1086         /* Nothing to do ? */
1087     }
1088
1089     x264_macroblock_slice_init( h );
1090 }
1091
1092 static void x264_slice_write( x264_t *h )
1093 {
1094     int i_skip;
1095     int mb_xy, i_mb_x, i_mb_y;
1096     int i, i_list, i_ref;
1097
1098     /* init stats */
1099     memset( &h->stat.frame, 0, sizeof(h->stat.frame) );
1100
1101     /* Slice */
1102     x264_nal_start( h, h->i_nal_type, h->i_nal_ref_idc );
1103
1104     /* Slice header */
1105     x264_slice_header_write( &h->out.bs, &h->sh, h->i_nal_ref_idc );
1106     if( h->param.b_cabac )
1107     {
1108         /* alignment needed */
1109         bs_align_1( &h->out.bs );
1110
1111         /* init cabac */
1112         x264_cabac_context_init( &h->cabac, h->sh.i_type, h->sh.i_qp, h->sh.i_cabac_init_idc );
1113         x264_cabac_encode_init ( &h->cabac, h->out.bs.p, h->out.bs.p_end );
1114     }
1115     h->mb.i_last_qp = h->sh.i_qp;
1116     h->mb.i_last_dqp = 0;
1117
1118     i_mb_y = h->sh.i_first_mb / h->sps->i_mb_width;
1119     i_mb_x = h->sh.i_first_mb % h->sps->i_mb_width;
1120     i_skip = 0;
1121
1122     while( (mb_xy = i_mb_x + i_mb_y * h->sps->i_mb_width) < h->sh.i_last_mb )
1123     {
1124         int mb_spos = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
1125
1126         if( i_mb_x == 0 )
1127             x264_fdec_filter_row( h, i_mb_y );
1128
1129         /* load cache */
1130         x264_macroblock_cache_load( h, i_mb_x, i_mb_y );
1131
1132         /* analyse parameters
1133          * Slice I: choose I_4x4 or I_16x16 mode
1134          * Slice P: choose between using P mode or intra (4x4 or 16x16)
1135          * */
1136         x264_macroblock_analyse( h );
1137
1138         /* encode this macroblock -> be careful it can change the mb type to P_SKIP if needed */
1139         x264_macroblock_encode( h );
1140
1141         x264_bitstream_check_buffer( h );
1142
1143         if( h->param.b_cabac )
1144         {
1145             if( mb_xy > h->sh.i_first_mb && !(h->sh.b_mbaff && (i_mb_y&1)) )
1146                 x264_cabac_encode_terminal( &h->cabac );
1147
1148             if( IS_SKIP( h->mb.i_type ) )
1149                 x264_cabac_mb_skip( h, 1 );
1150             else
1151             {
1152                 if( h->sh.i_type != SLICE_TYPE_I )
1153                     x264_cabac_mb_skip( h, 0 );
1154                 x264_macroblock_write_cabac( h, &h->cabac );
1155             }
1156         }
1157         else
1158         {
1159             if( IS_SKIP( h->mb.i_type ) )
1160                 i_skip++;
1161             else
1162             {
1163                 if( h->sh.i_type != SLICE_TYPE_I )
1164                 {
1165                     bs_write_ue( &h->out.bs, i_skip );  /* skip run */
1166                     i_skip = 0;
1167                 }
1168                 x264_macroblock_write_cavlc( h, &h->out.bs );
1169             }
1170         }
1171
1172 #if VISUALIZE
1173         if( h->param.b_visualize )
1174             x264_visualize_mb( h );
1175 #endif
1176
1177         /* save cache */
1178         x264_macroblock_cache_save( h );
1179
1180         /* accumulate mb stats */
1181         h->stat.frame.i_mb_count[h->mb.i_type]++;
1182         if( !IS_SKIP(h->mb.i_type) && !IS_INTRA(h->mb.i_type) && !IS_DIRECT(h->mb.i_type) )
1183         {
1184             if( h->mb.i_partition != D_8x8 )
1185                 h->stat.frame.i_mb_partition[h->mb.i_partition] += 4;
1186             else
1187                 for( i = 0; i < 4; i++ )
1188                     h->stat.frame.i_mb_partition[h->mb.i_sub_partition[i]] ++;
1189             if( h->param.i_frame_reference > 1 )
1190                 for( i_list = 0; i_list <= (h->sh.i_type == SLICE_TYPE_B); i_list++ )
1191                     for( i = 0; i < 4; i++ )
1192                     {
1193                         i_ref = h->mb.cache.ref[i_list][ x264_scan8[4*i] ];
1194                         if( i_ref >= 0 )
1195                             h->stat.frame.i_mb_count_ref[i_list][i_ref] ++;
1196                     }
1197         }
1198         if( h->mb.i_cbp_luma && !IS_INTRA(h->mb.i_type) )
1199         {
1200             h->stat.frame.i_mb_count_8x8dct[0] ++;
1201             h->stat.frame.i_mb_count_8x8dct[1] += h->mb.b_transform_8x8;
1202         }
1203
1204         x264_ratecontrol_mb( h, bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac) - mb_spos );
1205
1206         if( h->sh.b_mbaff )
1207         {
1208             i_mb_x += i_mb_y & 1;
1209             i_mb_y ^= i_mb_x < h->sps->i_mb_width;
1210         }
1211         else
1212             i_mb_x++;
1213         if(i_mb_x == h->sps->i_mb_width)
1214         {
1215             i_mb_y++;
1216             i_mb_x = 0;
1217         }
1218     }
1219
1220     if( h->param.b_cabac )
1221     {
1222         x264_cabac_encode_flush( h, &h->cabac );
1223         h->out.bs.p = h->cabac.p;
1224     }
1225     else
1226     {
1227         if( i_skip > 0 )
1228             bs_write_ue( &h->out.bs, i_skip );  /* last skip run */
1229         /* rbsp_slice_trailing_bits */
1230         bs_rbsp_trailing( &h->out.bs );
1231     }
1232
1233     x264_nal_end( h );
1234
1235     x264_fdec_filter_row( h, h->sps->i_mb_height );
1236
1237     /* Compute misc bits */
1238     h->stat.frame.i_misc_bits = bs_pos( &h->out.bs )
1239                               + NALU_OVERHEAD * 8
1240                               - h->stat.frame.i_itex_bits
1241                               - h->stat.frame.i_ptex_bits
1242                               - h->stat.frame.i_hdr_bits;
1243 }
1244
1245 static void x264_thread_sync_context( x264_t *dst, x264_t *src )
1246 {
1247     x264_frame_t **f;
1248     if( dst == src )
1249         return;
1250
1251     // reference counting
1252     for( f = src->frames.reference; *f; f++ )
1253         (*f)->i_reference_count++;
1254     for( f = dst->frames.reference; *f; f++ )
1255         x264_frame_push_unused( src, *f );
1256     src->fdec->i_reference_count++;
1257     x264_frame_push_unused( src, dst->fdec );
1258
1259     // copy everything except the per-thread pointers and the constants.
1260     memcpy( &dst->i_frame, &src->i_frame, offsetof(x264_t, mb.type) - offsetof(x264_t, i_frame) );
1261     memcpy( &dst->mb.i_type, &src->mb.i_type, offsetof(x264_t, rc) - offsetof(x264_t, mb.i_type) );
1262     dst->stat = src->stat;
1263 }
1264
1265 static void x264_thread_sync_stat( x264_t *dst, x264_t *src )
1266 {
1267     if( dst == src )
1268         return;
1269     memcpy( &dst->stat.i_slice_count, &src->stat.i_slice_count, sizeof(dst->stat) - sizeof(dst->stat.frame) );
1270 }
1271
1272 static int x264_slices_write( x264_t *h )
1273 {
1274     int i_frame_size;
1275
1276 #if VISUALIZE
1277     if( h->param.b_visualize )
1278         x264_visualize_init( h );
1279 #endif
1280
1281     x264_stack_align( x264_slice_write, h );
1282     i_frame_size = h->out.nal[h->out.i_nal-1].i_payload;
1283
1284 #if VISUALIZE
1285     if( h->param.b_visualize )
1286     {
1287         x264_visualize_show( h );
1288         x264_visualize_close( h );
1289     }
1290 #endif
1291
1292     h->out.i_frame_size = i_frame_size;
1293     return 0;
1294 }
1295
1296 /****************************************************************************
1297  * x264_encoder_encode:
1298  *  XXX: i_poc   : is the poc of the current given picture
1299  *       i_frame : is the number of the frame being coded
1300  *  ex:  type frame poc
1301  *       I      0   2*0
1302  *       P      1   2*3
1303  *       B      2   2*1
1304  *       B      3   2*2
1305  *       P      4   2*6
1306  *       B      5   2*4
1307  *       B      6   2*5
1308  ****************************************************************************/
1309 int     x264_encoder_encode( x264_t *h,
1310                              x264_nal_t **pp_nal, int *pi_nal,
1311                              x264_picture_t *pic_in,
1312                              x264_picture_t *pic_out )
1313 {
1314     x264_t *thread_current, *thread_prev, *thread_oldest;
1315     int     i_nal_type;
1316     int     i_nal_ref_idc;
1317
1318     int   i_global_qp;
1319
1320     if( h->param.i_threads > 1)
1321     {
1322         int i = ++h->i_thread_phase;
1323         int t = h->param.i_threads;
1324         thread_current = h->thread[ i%t ];
1325         thread_prev    = h->thread[ (i-1)%t ];
1326         thread_oldest  = h->thread[ (i+1)%t ];
1327         x264_thread_sync_context( thread_current, thread_prev );
1328         x264_thread_sync_ratecontrol( thread_current, thread_prev, thread_oldest );
1329         h = thread_current;
1330 //      fprintf(stderr, "current: %p  prev: %p  oldest: %p \n", thread_current, thread_prev, thread_oldest);
1331     }
1332     else
1333     {
1334         thread_current =
1335         thread_prev    =
1336         thread_oldest  = h;
1337     }
1338
1339     // ok to call this before encoding any frames, since the initial values of fdec have b_kept_as_ref=0
1340     x264_reference_update( h );
1341     h->fdec->i_lines_completed = -1;
1342
1343     /* no data out */
1344     *pi_nal = 0;
1345     *pp_nal = NULL;
1346
1347     /* ------------------- Setup new frame from picture -------------------- */
1348     if( pic_in != NULL )
1349     {
1350         /* 1: Copy the picture to a frame and move it to a buffer */
1351         x264_frame_t *fenc = x264_frame_pop_unused( h );
1352
1353         if( x264_frame_copy_picture( h, fenc, pic_in ) < 0 )
1354             return -1;
1355
1356         if( h->param.i_width != 16 * h->sps->i_mb_width ||
1357             h->param.i_height != 16 * h->sps->i_mb_height )
1358             x264_frame_expand_border_mod16( h, fenc );
1359
1360         fenc->i_frame = h->frames.i_input++;
1361
1362         x264_frame_push( h->frames.next, fenc );
1363
1364         if( h->frames.b_have_lowres )
1365             x264_frame_init_lowres( h, fenc );
1366
1367         if( h->frames.i_input <= h->frames.i_delay + 1 - h->param.i_threads )
1368         {
1369             /* Nothing yet to encode */
1370             /* waiting for filling bframe buffer */
1371             pic_out->i_type = X264_TYPE_AUTO;
1372             return 0;
1373         }
1374     }
1375
1376     if( h->frames.current[0] == NULL )
1377     {
1378         int bframes = 0;
1379         /* 2: Select frame types */
1380         if( h->frames.next[0] == NULL )
1381         {
1382             x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
1383             return 0;
1384         }
1385
1386         x264_slicetype_decide( h );
1387
1388         /* 3: move some B-frames and 1 non-B to encode queue */
1389         while( IS_X264_TYPE_B( h->frames.next[bframes]->i_type ) )
1390             bframes++;
1391         x264_frame_push( h->frames.current, x264_frame_shift( &h->frames.next[bframes] ) );
1392         /* FIXME: when max B-frames > 3, BREF may no longer be centered after GOP closing */
1393         if( h->param.b_bframe_pyramid && bframes > 1 )
1394         {
1395             x264_frame_t *mid = x264_frame_shift( &h->frames.next[bframes/2] );
1396             mid->i_type = X264_TYPE_BREF;
1397             x264_frame_push( h->frames.current, mid );
1398             bframes--;
1399         }
1400         while( bframes-- )
1401             x264_frame_push( h->frames.current, x264_frame_shift( h->frames.next ) );
1402     }
1403
1404     /* ------------------- Get frame to be encoded ------------------------- */
1405     /* 4: get picture to encode */
1406     h->fenc = x264_frame_shift( h->frames.current );
1407     if( h->fenc == NULL )
1408     {
1409         /* Nothing yet to encode (ex: waiting for I/P with B frames) */
1410         /* waiting for filling bframe buffer */
1411         pic_out->i_type = X264_TYPE_AUTO;
1412         return 0;
1413     }
1414
1415 do_encode:
1416
1417     if( h->fenc->i_type == X264_TYPE_IDR )
1418     {
1419         h->frames.i_last_idr = h->fenc->i_frame;
1420     }
1421
1422     /* ------------------- Setup frame context ----------------------------- */
1423     /* 5: Init data dependent of frame type */
1424     if( h->fenc->i_type == X264_TYPE_IDR )
1425     {
1426         /* reset ref pictures */
1427         x264_reference_reset( h );
1428
1429         i_nal_type    = NAL_SLICE_IDR;
1430         i_nal_ref_idc = NAL_PRIORITY_HIGHEST;
1431         h->sh.i_type = SLICE_TYPE_I;
1432     }
1433     else if( h->fenc->i_type == X264_TYPE_I )
1434     {
1435         i_nal_type    = NAL_SLICE;
1436         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
1437         h->sh.i_type = SLICE_TYPE_I;
1438     }
1439     else if( h->fenc->i_type == X264_TYPE_P )
1440     {
1441         i_nal_type    = NAL_SLICE;
1442         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
1443         h->sh.i_type = SLICE_TYPE_P;
1444     }
1445     else if( h->fenc->i_type == X264_TYPE_BREF )
1446     {
1447         i_nal_type    = NAL_SLICE;
1448         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* maybe add MMCO to forget it? -> low */
1449         h->sh.i_type = SLICE_TYPE_B;
1450     }
1451     else    /* B frame */
1452     {
1453         i_nal_type    = NAL_SLICE;
1454         i_nal_ref_idc = NAL_PRIORITY_DISPOSABLE;
1455         h->sh.i_type = SLICE_TYPE_B;
1456     }
1457
1458     h->fdec->i_poc =
1459     h->fenc->i_poc = 2 * (h->fenc->i_frame - h->frames.i_last_idr);
1460     h->fdec->i_type = h->fenc->i_type;
1461     h->fdec->i_frame = h->fenc->i_frame;
1462     h->fenc->b_kept_as_ref =
1463     h->fdec->b_kept_as_ref = i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE && h->param.i_keyint_max > 1;
1464
1465
1466
1467     /* ------------------- Init                ----------------------------- */
1468     /* build ref list 0/1 */
1469     x264_reference_build_list( h, h->fdec->i_poc );
1470
1471     /* Init the rate control */
1472     x264_ratecontrol_start( h, h->fenc->i_qpplus1 );
1473     i_global_qp = x264_ratecontrol_qp( h );
1474
1475     pic_out->i_qpplus1 =
1476     h->fdec->i_qpplus1 = i_global_qp + 1;
1477
1478     if( h->sh.i_type == SLICE_TYPE_B )
1479         x264_macroblock_bipred_init( h );
1480
1481     /* ------------------------ Create slice header  ----------------------- */
1482     x264_slice_init( h, i_nal_type, i_global_qp );
1483
1484     if( i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE )
1485         h->i_frame_num++;
1486
1487     /* ---------------------- Write the bitstream -------------------------- */
1488     /* Init bitstream context */
1489     h->out.i_nal = 0;
1490     bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
1491
1492     if(h->param.b_aud){
1493         int pic_type;
1494
1495         if(h->sh.i_type == SLICE_TYPE_I)
1496             pic_type = 0;
1497         else if(h->sh.i_type == SLICE_TYPE_P)
1498             pic_type = 1;
1499         else if(h->sh.i_type == SLICE_TYPE_B)
1500             pic_type = 2;
1501         else
1502             pic_type = 7;
1503
1504         x264_nal_start(h, NAL_AUD, NAL_PRIORITY_DISPOSABLE);
1505         bs_write(&h->out.bs, 3, pic_type);
1506         bs_rbsp_trailing(&h->out.bs);
1507         x264_nal_end(h);
1508     }
1509
1510     h->i_nal_type = i_nal_type;
1511     h->i_nal_ref_idc = i_nal_ref_idc;
1512
1513     /* Write SPS and PPS */
1514     if( i_nal_type == NAL_SLICE_IDR && h->param.b_repeat_headers )
1515     {
1516         if( h->fenc->i_frame == 0 )
1517         {
1518             /* identify ourself */
1519             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
1520             x264_sei_version_write( h, &h->out.bs );
1521             x264_nal_end( h );
1522         }
1523
1524         /* generate sequence parameters */
1525         x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
1526         x264_sps_write( &h->out.bs, h->sps );
1527         x264_nal_end( h );
1528
1529         /* generate picture parameters */
1530         x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
1531         x264_pps_write( &h->out.bs, h->pps );
1532         x264_nal_end( h );
1533     }
1534
1535     /* Write frame */
1536     if( h->param.i_threads > 1 )
1537     {
1538         x264_pthread_create( &h->thread_handle, NULL, (void*)x264_slices_write, h );
1539         h->b_thread_active = 1;
1540     }
1541     else
1542         x264_slices_write( h );
1543
1544     /* restore CPU state (before using float again) */
1545     x264_emms();
1546
1547     if( h->sh.i_type == SLICE_TYPE_P && !h->param.rc.b_stat_read
1548         && h->param.i_scenecut_threshold >= 0
1549         && !h->param.b_pre_scenecut )
1550     {
1551         const int *mbs = h->stat.frame.i_mb_count;
1552         int i_mb_i = mbs[I_16x16] + mbs[I_8x8] + mbs[I_4x4];
1553         int i_mb_p = mbs[P_L0] + mbs[P_8x8];
1554         int i_mb_s = mbs[P_SKIP];
1555         int i_mb   = h->sps->i_mb_width * h->sps->i_mb_height;
1556         int64_t i_inter_cost = h->stat.frame.i_inter_cost;
1557         int64_t i_intra_cost = h->stat.frame.i_intra_cost;
1558
1559         float f_bias;
1560         int i_gop_size = h->fenc->i_frame - h->frames.i_last_idr;
1561         float f_thresh_max = h->param.i_scenecut_threshold / 100.0;
1562         /* magic numbers pulled out of thin air */
1563         float f_thresh_min = f_thresh_max * h->param.i_keyint_min
1564                              / ( h->param.i_keyint_max * 4 );
1565         if( h->param.i_keyint_min == h->param.i_keyint_max )
1566              f_thresh_min= f_thresh_max;
1567
1568         /* macroblock_analyse() doesn't further analyse skipped mbs,
1569          * so we have to guess their cost */
1570         if( h->stat.frame.i_mbs_analysed > 0 )
1571             i_intra_cost = i_intra_cost * i_mb / h->stat.frame.i_mbs_analysed;
1572
1573         if( i_gop_size < h->param.i_keyint_min / 4 )
1574             f_bias = f_thresh_min / 4;
1575         else if( i_gop_size <= h->param.i_keyint_min )
1576             f_bias = f_thresh_min * i_gop_size / h->param.i_keyint_min;
1577         else
1578         {
1579             f_bias = f_thresh_min
1580                      + ( f_thresh_max - f_thresh_min )
1581                        * ( i_gop_size - h->param.i_keyint_min )
1582                        / ( h->param.i_keyint_max - h->param.i_keyint_min );
1583         }
1584         f_bias = X264_MIN( f_bias, 1.0 );
1585
1586         /* Bad P will be reencoded as I */
1587         if( h->stat.frame.i_mbs_analysed > 0 &&
1588             i_inter_cost >= (1.0 - f_bias) * i_intra_cost )
1589         {
1590             int b;
1591
1592             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",
1593                       h->fenc->i_frame,
1594                       (double)i_intra_cost, (double)i_inter_cost,
1595                       1. - (double)i_inter_cost / i_intra_cost,
1596                       f_bias, i_gop_size,
1597                       i_mb_i, i_mb_p, i_mb_s );
1598
1599             /* Restore frame num */
1600             h->i_frame_num--;
1601
1602             for( b = 0; h->frames.current[b] && IS_X264_TYPE_B( h->frames.current[b]->i_type ); b++ );
1603             if( b > 0 )
1604             {
1605                 /* If using B-frames, force GOP to be closed.
1606                  * Even if this frame is going to be I and not IDR, forcing a
1607                  * P-frame before the scenecut will probably help compression.
1608                  *
1609                  * We don't yet know exactly which frame is the scene cut, so
1610                  * we can't assign an I-frame. Instead, change the previous
1611                  * B-frame to P, and rearrange coding order. */
1612
1613                 if( h->param.b_bframe_adaptive || b > 1 )
1614                     h->fenc->i_type = X264_TYPE_AUTO;
1615                 x264_frame_sort_pts( h->frames.current );
1616                 x264_frame_unshift( h->frames.next, h->fenc );
1617                 h->fenc = h->frames.current[b-1];
1618                 h->frames.current[b-1] = NULL;
1619                 h->fenc->i_type = X264_TYPE_P;
1620                 x264_frame_sort_dts( h->frames.current );
1621             }
1622             /* Do IDR if needed */
1623             else if( i_gop_size >= h->param.i_keyint_min )
1624             {
1625                 /* Reset */
1626                 h->i_frame_num = 0;
1627
1628                 /* Reinit field of fenc */
1629                 h->fenc->i_type = X264_TYPE_IDR;
1630                 h->fenc->i_poc = 0;
1631
1632                 /* Put enqueued frames back in the pool */
1633                 while( h->frames.current[0] )
1634                     x264_frame_push( h->frames.next, x264_frame_shift( h->frames.current ) );
1635                 x264_frame_sort_pts( h->frames.next );
1636             }
1637             else
1638             {
1639                 h->fenc->i_type = X264_TYPE_I;
1640             }
1641             goto do_encode;
1642         }
1643     }
1644
1645     x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
1646     return 0;
1647 }
1648
1649 static void x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
1650                                     x264_nal_t **pp_nal, int *pi_nal,
1651                                     x264_picture_t *pic_out )
1652 {
1653     int i, i_list;
1654     char psz_message[80];
1655
1656     if( h->b_thread_active )
1657     {
1658         x264_pthread_join( h->thread_handle, NULL );
1659         h->b_thread_active = 0;
1660     }
1661     if( !h->out.i_nal )
1662     {
1663         pic_out->i_type = X264_TYPE_AUTO;
1664         return;
1665     }
1666
1667     x264_frame_push_unused( thread_current, h->fenc );
1668
1669     /* End bitstream, set output  */
1670     *pi_nal = h->out.i_nal;
1671     *pp_nal = h->out.nal;
1672     h->out.i_nal = 0;
1673
1674     /* Set output picture properties */
1675     if( h->sh.i_type == SLICE_TYPE_I )
1676         pic_out->i_type = h->i_nal_type == NAL_SLICE_IDR ? X264_TYPE_IDR : X264_TYPE_I;
1677     else if( h->sh.i_type == SLICE_TYPE_P )
1678         pic_out->i_type = X264_TYPE_P;
1679     else
1680         pic_out->i_type = X264_TYPE_B;
1681     pic_out->i_pts = h->fenc->i_pts;
1682
1683     pic_out->img.i_plane = h->fdec->i_plane;
1684     for(i = 0; i < 4; i++){
1685         pic_out->img.i_stride[i] = h->fdec->i_stride[i];
1686         pic_out->img.plane[i] = h->fdec->plane[i];
1687     }
1688
1689     /* ---------------------- Update encoder state ------------------------- */
1690
1691     /* update rc */
1692     x264_emms();
1693     x264_ratecontrol_end( h, h->out.i_frame_size * 8 );
1694
1695     /* restore CPU state (before using float again) */
1696     x264_emms();
1697
1698     x264_noise_reduction_update( thread_current );
1699
1700     /* ---------------------- Compute/Print statistics --------------------- */
1701     x264_thread_sync_stat( h, h->thread[0] );
1702
1703     /* Slice stat */
1704     h->stat.i_slice_count[h->sh.i_type]++;
1705     h->stat.i_slice_size[h->sh.i_type] += h->out.i_frame_size + NALU_OVERHEAD;
1706     h->stat.f_slice_qp[h->sh.i_type] += h->fdec->f_qp_avg_aq;
1707
1708     for( i = 0; i < X264_MBTYPE_MAX; i++ )
1709         h->stat.i_mb_count[h->sh.i_type][i] += h->stat.frame.i_mb_count[i];
1710     for( i = 0; i < X264_PARTTYPE_MAX; i++ )
1711         h->stat.i_mb_partition[h->sh.i_type][i] += h->stat.frame.i_mb_partition[i];
1712     for( i = 0; i < 2; i++ )
1713         h->stat.i_mb_count_8x8dct[i] += h->stat.frame.i_mb_count_8x8dct[i];
1714     if( h->sh.i_type != SLICE_TYPE_I )
1715         for( i_list = 0; i_list < 2; i_list++ )
1716             for( i = 0; i < 32; i++ )
1717                 h->stat.i_mb_count_ref[h->sh.i_type][i_list][i] += h->stat.frame.i_mb_count_ref[i_list][i];
1718     if( h->sh.i_type == SLICE_TYPE_P )
1719         h->stat.i_consecutive_bframes[h->fdec->i_frame - h->fref0[0]->i_frame - 1]++;
1720     if( h->sh.i_type == SLICE_TYPE_B )
1721     {
1722         h->stat.i_direct_frames[ h->sh.b_direct_spatial_mv_pred ] ++;
1723         if( h->mb.b_direct_auto_write )
1724         {
1725             //FIXME somewhat arbitrary time constants
1726             if( h->stat.i_direct_score[0] + h->stat.i_direct_score[1] > h->mb.i_mb_count )
1727             {
1728                 for( i = 0; i < 2; i++ )
1729                     h->stat.i_direct_score[i] = h->stat.i_direct_score[i] * 9/10;
1730             }
1731             for( i = 0; i < 2; i++ )
1732                 h->stat.i_direct_score[i] += h->stat.frame.i_direct_score[i];
1733         }
1734     }
1735
1736     psz_message[0] = '\0';
1737     if( h->param.analyse.b_psnr )
1738     {
1739         int64_t sqe[3] = {
1740             h->stat.frame.i_ssd[0],
1741             h->stat.frame.i_ssd[1],
1742             h->stat.frame.i_ssd[2],
1743         };
1744
1745         h->stat.i_sqe_global[h->sh.i_type] += sqe[0] + sqe[1] + sqe[2];
1746         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 );
1747         h->stat.f_psnr_mean_y[h->sh.i_type] += x264_psnr( sqe[0], h->param.i_width * h->param.i_height );
1748         h->stat.f_psnr_mean_u[h->sh.i_type] += x264_psnr( sqe[1], h->param.i_width * h->param.i_height / 4 );
1749         h->stat.f_psnr_mean_v[h->sh.i_type] += x264_psnr( sqe[2], h->param.i_width * h->param.i_height / 4 );
1750
1751         snprintf( psz_message, 80, " PSNR Y:%5.2f U:%5.2f V:%5.2f",
1752                   x264_psnr( sqe[0], h->param.i_width * h->param.i_height ),
1753                   x264_psnr( sqe[1], h->param.i_width * h->param.i_height / 4),
1754                   x264_psnr( sqe[2], h->param.i_width * h->param.i_height / 4) );
1755     }
1756
1757     if( h->param.analyse.b_ssim )
1758     {
1759         double ssim_y = h->stat.frame.f_ssim
1760                       / (((h->param.i_width-6)>>2) * ((h->param.i_height-6)>>2));
1761         h->stat.f_ssim_mean_y[h->sh.i_type] += ssim_y;
1762         snprintf( psz_message + strlen(psz_message), 80 - strlen(psz_message),
1763                   " SSIM Y:%.5f", ssim_y );
1764     }
1765     psz_message[79] = '\0';
1766
1767     x264_log( h, X264_LOG_DEBUG,
1768                   "frame=%4d QP=%.2f NAL=%d Slice:%c Poc:%-3d I:%-4d P:%-4d SKIP:%-4d size=%d bytes%s\n",
1769               h->i_frame,
1770               h->fdec->f_qp_avg_aq,
1771               h->i_nal_ref_idc,
1772               h->sh.i_type == SLICE_TYPE_I ? 'I' : (h->sh.i_type == SLICE_TYPE_P ? 'P' : 'B' ),
1773               h->fdec->i_poc,
1774               h->stat.frame.i_mb_count_i,
1775               h->stat.frame.i_mb_count_p,
1776               h->stat.frame.i_mb_count_skip,
1777               h->out.i_frame_size,
1778               psz_message );
1779
1780     // keep stats all in one place
1781     x264_thread_sync_stat( h->thread[0], h );
1782     // for the use of the next frame
1783     x264_thread_sync_stat( thread_current, h );
1784
1785 #ifdef DEBUG_MB_TYPE
1786 {
1787     static const char mb_chars[] = { 'i', 'i', 'I', 'C', 'P', '8', 'S',
1788         'D', '<', 'X', 'B', 'X', '>', 'B', 'B', 'B', 'B', '8', 'S' };
1789     int mb_xy;
1790     for( mb_xy = 0; mb_xy < h->sps->i_mb_width * h->sps->i_mb_height; mb_xy++ )
1791     {
1792         if( h->mb.type[mb_xy] < X264_MBTYPE_MAX && h->mb.type[mb_xy] >= 0 )
1793             fprintf( stderr, "%c ", mb_chars[ h->mb.type[mb_xy] ] );
1794         else
1795             fprintf( stderr, "? " );
1796
1797         if( (mb_xy+1) % h->sps->i_mb_width == 0 )
1798             fprintf( stderr, "\n" );
1799     }
1800 }
1801 #endif
1802
1803     if( h->param.psz_dump_yuv )
1804         x264_frame_dump( h );
1805 }
1806
1807 static void x264_print_intra( int64_t *i_mb_count, double i_count, int b_print_pcm, char *intra )
1808 {
1809     intra += sprintf( intra, "I16..4%s: %4.1f%% %4.1f%% %4.1f%%",
1810         b_print_pcm ? "..PCM" : "",
1811         i_mb_count[I_16x16]/ i_count,
1812         i_mb_count[I_8x8]  / i_count,
1813         i_mb_count[I_4x4]  / i_count );
1814     if( b_print_pcm )
1815         sprintf( intra, " %4.1f%%", i_mb_count[I_PCM]  / i_count );
1816 }
1817
1818 /****************************************************************************
1819  * x264_encoder_close:
1820  ****************************************************************************/
1821 void    x264_encoder_close  ( x264_t *h )
1822 {
1823     int64_t i_yuv_size = 3 * h->param.i_width * h->param.i_height / 2;
1824     int64_t i_mb_count_size[2][7] = {{0}};
1825     char buf[200];
1826     int i, j, i_list, i_type;
1827     int b_print_pcm = h->stat.i_mb_count[SLICE_TYPE_I][I_PCM]
1828                    || h->stat.i_mb_count[SLICE_TYPE_P][I_PCM]
1829                    || h->stat.i_mb_count[SLICE_TYPE_B][I_PCM];
1830
1831     for( i=0; i<h->param.i_threads; i++ )
1832     {
1833         // don't strictly have to wait for the other threads, but it's simpler than canceling them
1834         if( h->thread[i]->b_thread_active )
1835         {
1836             x264_pthread_join( h->thread[i]->thread_handle, NULL );
1837             assert( h->thread[i]->fenc->i_reference_count == 1 );
1838             x264_frame_delete( h->thread[i]->fenc );
1839         }
1840     }
1841
1842     /* Slices used and PSNR */
1843     for( i=0; i<5; i++ )
1844     {
1845         static const int slice_order[] = { SLICE_TYPE_I, SLICE_TYPE_SI, SLICE_TYPE_P, SLICE_TYPE_SP, SLICE_TYPE_B };
1846         static const char *slice_name[] = { "P", "B", "I", "SP", "SI" };
1847         int i_slice = slice_order[i];
1848
1849         if( h->stat.i_slice_count[i_slice] > 0 )
1850         {
1851             const int i_count = h->stat.i_slice_count[i_slice];
1852             if( h->param.analyse.b_psnr )
1853             {
1854                 x264_log( h, X264_LOG_INFO,
1855                           "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",
1856                           slice_name[i_slice],
1857                           i_count,
1858                           h->stat.f_slice_qp[i_slice] / i_count,
1859                           (double)h->stat.i_slice_size[i_slice] / i_count,
1860                           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,
1861                           h->stat.f_psnr_average[i_slice] / i_count,
1862                           x264_psnr( h->stat.i_sqe_global[i_slice], i_count * i_yuv_size ) );
1863             }
1864             else
1865             {
1866                 x264_log( h, X264_LOG_INFO,
1867                           "slice %s:%-5d Avg QP:%5.2f  size:%6.0f\n",
1868                           slice_name[i_slice],
1869                           i_count,
1870                           h->stat.f_slice_qp[i_slice] / i_count,
1871                           (double)h->stat.i_slice_size[i_slice] / i_count );
1872             }
1873         }
1874     }
1875     if( h->param.i_bframe && h->stat.i_slice_count[SLICE_TYPE_P] )
1876     {
1877         char *p = buf;
1878         int den = 0;
1879         // weight by number of frames (including the P-frame) that are in a sequence of N B-frames
1880         for( i=0; i<=h->param.i_bframe; i++ )
1881             den += (i+1) * h->stat.i_consecutive_bframes[i];
1882         for( i=0; i<=h->param.i_bframe; i++ )
1883             p += sprintf( p, " %4.1f%%", 100. * (i+1) * h->stat.i_consecutive_bframes[i] / den );
1884         x264_log( h, X264_LOG_INFO, "consecutive B-frames:%s\n", buf );
1885     }
1886
1887     for( i_type = 0; i_type < 2; i_type++ )
1888         for( i = 0; i < X264_PARTTYPE_MAX; i++ )
1889         {
1890             if( i == D_DIRECT_8x8 ) continue; /* direct is counted as its own type */
1891             i_mb_count_size[i_type][x264_mb_partition_pixel_table[i]] += h->stat.i_mb_partition[i_type][i];
1892         }
1893
1894     /* MB types used */
1895     if( h->stat.i_slice_count[SLICE_TYPE_I] > 0 )
1896     {
1897         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_I];
1898         double i_count = h->stat.i_slice_count[SLICE_TYPE_I] * h->mb.i_mb_count / 100.0;
1899         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
1900         x264_log( h, X264_LOG_INFO, "mb I  %s\n", buf );
1901     }
1902     if( h->stat.i_slice_count[SLICE_TYPE_P] > 0 )
1903     {
1904         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_P];
1905         double i_count = h->stat.i_slice_count[SLICE_TYPE_P] * h->mb.i_mb_count / 100.0;
1906         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_P];
1907         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
1908         x264_log( h, X264_LOG_INFO,
1909                   "mb P  %s  P16..4: %4.1f%% %4.1f%% %4.1f%% %4.1f%% %4.1f%%    skip:%4.1f%%\n",
1910                   buf,
1911                   i_mb_size[PIXEL_16x16] / (i_count*4),
1912                   (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
1913                   i_mb_size[PIXEL_8x8] / (i_count*4),
1914                   (i_mb_size[PIXEL_8x4] + i_mb_size[PIXEL_4x8]) / (i_count*4),
1915                   i_mb_size[PIXEL_4x4] / (i_count*4),
1916                   i_mb_count[P_SKIP] / i_count );
1917     }
1918     if( h->stat.i_slice_count[SLICE_TYPE_B] > 0 )
1919     {
1920         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_B];
1921         double i_count = h->stat.i_slice_count[SLICE_TYPE_B] * h->mb.i_mb_count / 100.0;
1922         double i_mb_list_count;
1923         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_B];
1924         int64_t list_count[3] = {0}; /* 0 == L0, 1 == L1, 2 == BI */
1925         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
1926         for( i = 0; i < X264_PARTTYPE_MAX; i++ )
1927             for( j = 0; j < 2; j++ )
1928             {
1929                 int l0 = x264_mb_type_list0_table[i][j];
1930                 int l1 = x264_mb_type_list1_table[i][j];
1931                 if( l0 || l1 )
1932                     list_count[l1+l0*l1] += h->stat.i_mb_count[SLICE_TYPE_B][i] * 2;
1933             }
1934         list_count[0] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L0_8x8];
1935         list_count[1] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L1_8x8];
1936         list_count[2] += h->stat.i_mb_partition[SLICE_TYPE_B][D_BI_8x8];
1937         i_mb_count[B_DIRECT] += (h->stat.i_mb_partition[SLICE_TYPE_B][D_DIRECT_8x8]+2)/4;
1938         i_mb_list_count = (list_count[0] + list_count[1] + list_count[2]) / 100.0;
1939         x264_log( h, X264_LOG_INFO,
1940                   "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",
1941                   buf,
1942                   i_mb_size[PIXEL_16x16] / (i_count*4),
1943                   (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
1944                   i_mb_size[PIXEL_8x8] / (i_count*4),
1945                   i_mb_count[B_DIRECT] / i_count,
1946                   i_mb_count[B_SKIP]   / i_count,
1947                   list_count[0] / i_mb_list_count,
1948                   list_count[1] / i_mb_list_count,
1949                   list_count[2] / i_mb_list_count );
1950     }
1951
1952     x264_ratecontrol_summary( h );
1953
1954     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 )
1955     {
1956         const int i_count = h->stat.i_slice_count[SLICE_TYPE_I] +
1957                             h->stat.i_slice_count[SLICE_TYPE_P] +
1958                             h->stat.i_slice_count[SLICE_TYPE_B];
1959         float fps = (float) h->param.i_fps_num / h->param.i_fps_den;
1960 #define SUM3(p) (p[SLICE_TYPE_I] + p[SLICE_TYPE_P] + p[SLICE_TYPE_B])
1961 #define SUM3b(p,o) (p[SLICE_TYPE_I][o] + p[SLICE_TYPE_P][o] + p[SLICE_TYPE_B][o])
1962         float f_bitrate = fps * SUM3(h->stat.i_slice_size) / i_count / 125;
1963
1964         if( h->pps->b_transform_8x8_mode )
1965         {
1966             int64_t i_i8x8 = SUM3b( h->stat.i_mb_count, I_8x8 );
1967             int64_t i_intra = i_i8x8 + SUM3b( h->stat.i_mb_count, I_4x4 )
1968                                      + SUM3b( h->stat.i_mb_count, I_16x16 );
1969             x264_log( h, X264_LOG_INFO, "8x8 transform  intra:%.1f%%  inter:%.1f%%\n",
1970                       100. * i_i8x8 / i_intra,
1971                       100. * h->stat.i_mb_count_8x8dct[1] / h->stat.i_mb_count_8x8dct[0] );
1972         }
1973
1974         if( h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO
1975             && h->stat.i_slice_count[SLICE_TYPE_B] )
1976         {
1977             x264_log( h, X264_LOG_INFO, "direct mvs  spatial:%.1f%%  temporal:%.1f%%\n",
1978                       h->stat.i_direct_frames[1] * 100. / h->stat.i_slice_count[SLICE_TYPE_B],
1979                       h->stat.i_direct_frames[0] * 100. / h->stat.i_slice_count[SLICE_TYPE_B] );
1980         }
1981
1982         for( i_list = 0; i_list < 2; i_list++ )
1983         {
1984             int i_slice;
1985             for( i_slice = 0; i_slice < 2; i_slice++ )
1986             {
1987                 char *p = buf;
1988                 int64_t i_den = 0;
1989                 int i_max = 0;
1990                 for( i = 0; i < 32; i++ )
1991                     if( h->stat.i_mb_count_ref[i_slice][i_list][i] )
1992                     {
1993                         i_den += h->stat.i_mb_count_ref[i_slice][i_list][i];
1994                         i_max = i;
1995                     }
1996                 if( i_max == 0 )
1997                     continue;
1998                 for( i = 0; i <= i_max; i++ )
1999                     p += sprintf( p, " %4.1f%%", 100. * h->stat.i_mb_count_ref[i_slice][i_list][i] / i_den );
2000                 x264_log( h, X264_LOG_INFO, "ref %c L%d %s\n", "PB"[i_slice], i_list, buf );
2001             }
2002         }
2003
2004         if( h->param.analyse.b_ssim )
2005         {
2006             x264_log( h, X264_LOG_INFO,
2007                       "SSIM Mean Y:%.7f\n",
2008                       SUM3( h->stat.f_ssim_mean_y ) / i_count );
2009         }
2010         if( h->param.analyse.b_psnr )
2011         {
2012             x264_log( h, X264_LOG_INFO,
2013                       "PSNR Mean Y:%6.3f U:%6.3f V:%6.3f Avg:%6.3f Global:%6.3f kb/s:%.2f\n",
2014                       SUM3( h->stat.f_psnr_mean_y ) / i_count,
2015                       SUM3( h->stat.f_psnr_mean_u ) / i_count,
2016                       SUM3( h->stat.f_psnr_mean_v ) / i_count,
2017                       SUM3( h->stat.f_psnr_average ) / i_count,
2018                       x264_psnr( SUM3( h->stat.i_sqe_global ), i_count * i_yuv_size ),
2019                       f_bitrate );
2020         }
2021         else
2022             x264_log( h, X264_LOG_INFO, "kb/s:%.1f\n", f_bitrate );
2023     }
2024
2025     /* rc */
2026     x264_ratecontrol_delete( h );
2027
2028     /* param */
2029     if( h->param.rc.psz_stat_out )
2030         free( h->param.rc.psz_stat_out );
2031     if( h->param.rc.psz_stat_in )
2032         free( h->param.rc.psz_stat_in );
2033     if( h->param.rc.psz_rc_eq )
2034         free( h->param.rc.psz_rc_eq );
2035
2036     x264_cqm_delete( h );
2037
2038     if( h->param.i_threads > 1)
2039         h = h->thread[ h->i_thread_phase % h->param.i_threads ];
2040
2041     /* frames */
2042     for( i = 0; h->frames.current[i]; i++ )
2043     {
2044         assert( h->frames.current[i]->i_reference_count == 1 );
2045         x264_frame_delete( h->frames.current[i] );
2046     }
2047     for( i = 0; h->frames.next[i]; i++ )
2048     {
2049         assert( h->frames.next[i]->i_reference_count == 1 );
2050         x264_frame_delete( h->frames.next[i] );
2051     }
2052     for( i = 0; h->frames.unused[i]; i++ )
2053     {
2054         assert( h->frames.unused[i]->i_reference_count == 0 );
2055         x264_frame_delete( h->frames.unused[i] );
2056     }
2057
2058     h = h->thread[0];
2059
2060     for( i = h->param.i_threads - 1; i >= 0; i-- )
2061     {
2062         x264_frame_t **frame;
2063
2064         for( frame = h->thread[i]->frames.reference; *frame; frame++ )
2065         {
2066             assert( (*frame)->i_reference_count > 0 );
2067             (*frame)->i_reference_count--;
2068             if( (*frame)->i_reference_count == 0 )
2069                 x264_frame_delete( *frame );
2070         }
2071         frame = &h->thread[i]->fdec;
2072         assert( (*frame)->i_reference_count > 0 );
2073         (*frame)->i_reference_count--;
2074         if( (*frame)->i_reference_count == 0 )
2075             x264_frame_delete( *frame );
2076
2077         x264_macroblock_cache_end( h->thread[i] );
2078         x264_free( h->thread[i]->out.p_bitstream );
2079         x264_free( h->thread[i] );
2080     }
2081 }