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

   1 /*****************************************************************************
   2  * encoder.c: top-level encoder functions
   3  *****************************************************************************
   4  * Copyright (C) 2003-2014 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  * This program is also available under a commercial proprietary license.
  25  * For more information, contact us at licensing@x264.com.
  26  *****************************************************************************/
  27
  28 #include "common/common.h"
  29
  30 #include "set.h"
  31 #include "analyse.h"
  32 #include "ratecontrol.h"
  33 #include "macroblock.h"
  34 #include "me.h"
  35
  36 //#define DEBUG_MB_TYPE
  37
  38 #define bs_write_ue bs_write_ue_big
  39
  40 static int x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
  41                                    x264_nal_t **pp_nal, int *pi_nal,
  42                                    x264_picture_t *pic_out );
  43
  44 /****************************************************************************
  45  *
  46  ******************************* x264 libs **********************************
  47  *
  48  ****************************************************************************/
  49 static double x264_psnr( double sqe, double size )
  50 {
  51     double mse = sqe / (PIXEL_MAX*PIXEL_MAX * size);
  52     if( mse <= 0.0000000001 ) /* Max 100dB */
  53         return 100;
  54
  55     return -10.0 * log10( mse );
  56 }
  57
  58 static double x264_ssim( double ssim )
  59 {
  60     double inv_ssim = 1 - ssim;
  61     if( inv_ssim <= 0.0000000001 ) /* Max 100dB */
  62         return 100;
  63
  64     return -10.0 * log10( inv_ssim );
  65 }
  66
  67 static int x264_threadpool_wait_all( x264_t *h )
  68 {
  69     for( int i = 0; i < h->param.i_threads; i++ )
  70         if( h->thread[i]->b_thread_active )
  71         {
  72             h->thread[i]->b_thread_active = 0;
  73             if( (intptr_t)x264_threadpool_wait( h->threadpool, h->thread[i] ) < 0 )
  74                 return -1;
  75         }
  76     return 0;
  77 }
  78
  79 static void x264_frame_dump( x264_t *h )
  80 {
  81     FILE *f = x264_fopen( h->param.psz_dump_yuv, "r+b" );
  82     if( !f )
  83         return;
  84
  85     /* Wait for the threads to finish deblocking */
  86     if( h->param.b_sliced_threads )
  87         x264_threadpool_wait_all( h );
  88
  89     /* Write the frame in display order */
  90     int frame_size = FRAME_SIZE( h->param.i_height * h->param.i_width * sizeof(pixel) );
  91     fseek( f, (uint64_t)h->fdec->i_frame * frame_size, SEEK_SET );
  92     for( int p = 0; p < (CHROMA444 ? 3 : 1); p++ )
  93         for( int y = 0; y < h->param.i_height; y++ )
  94             fwrite( &h->fdec->plane[p][y*h->fdec->i_stride[p]], sizeof(pixel), h->param.i_width, f );
  95     if( !CHROMA444 )
  96     {
  97         int cw = h->param.i_width>>1;
  98         int ch = h->param.i_height>>CHROMA_V_SHIFT;
  99         pixel *planeu = x264_malloc( (cw*ch*2+32)*sizeof(pixel) );
 100         if( planeu )
 101         {
 102             pixel *planev = planeu + cw*ch + 16;
 103             h->mc.plane_copy_deinterleave( planeu, cw, planev, cw, h->fdec->plane[1], h->fdec->i_stride[1], cw, ch );
 104             fwrite( planeu, 1, cw*ch*sizeof(pixel), f );
 105             fwrite( planev, 1, cw*ch*sizeof(pixel), f );
 106             x264_free( planeu );
 107         }
 108     }
 109     fclose( f );
 110 }
 111
 112 /* Fill "default" values */
 113 static void x264_slice_header_init( x264_t *h, x264_slice_header_t *sh,
 114                                     x264_sps_t *sps, x264_pps_t *pps,
 115                                     int i_idr_pic_id, int i_frame, int i_qp )
 116 {
 117     x264_param_t *param = &h->param;
 118
 119     /* First we fill all fields */
 120     sh->sps = sps;
 121     sh->pps = pps;
 122
 123     sh->i_first_mb  = 0;
 124     sh->i_last_mb   = h->mb.i_mb_count - 1;
 125     sh->i_pps_id    = pps->i_id;
 126
 127     sh->i_frame_num = i_frame;
 128
 129     sh->b_mbaff = PARAM_INTERLACED;
 130     sh->b_field_pic = 0;    /* no field support for now */
 131     sh->b_bottom_field = 0; /* not yet used */
 132
 133     sh->i_idr_pic_id = i_idr_pic_id;
 134
 135     /* poc stuff, fixed later */
 136     sh->i_poc = 0;
 137     sh->i_delta_poc_bottom = 0;
 138     sh->i_delta_poc[0] = 0;
 139     sh->i_delta_poc[1] = 0;
 140
 141     sh->i_redundant_pic_cnt = 0;
 142
 143     h->mb.b_direct_auto_write = h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO
 144                                 && h->param.i_bframe
 145                                 && ( h->param.rc.b_stat_write || !h->param.rc.b_stat_read );
 146
 147     if( !h->mb.b_direct_auto_read && sh->i_type == SLICE_TYPE_B )
 148     {
 149         if( h->fref[1][0]->i_poc_l0ref0 == h->fref[0][0]->i_poc )
 150         {
 151             if( h->mb.b_direct_auto_write )
 152                 sh->b_direct_spatial_mv_pred = ( h->stat.i_direct_score[1] > h->stat.i_direct_score[0] );
 153             else
 154                 sh->b_direct_spatial_mv_pred = ( param->analyse.i_direct_mv_pred == X264_DIRECT_PRED_SPATIAL );
 155         }
 156         else
 157         {
 158             h->mb.b_direct_auto_write = 0;
 159             sh->b_direct_spatial_mv_pred = 1;
 160         }
 161     }
 162     /* else b_direct_spatial_mv_pred was read from the 2pass statsfile */
 163
 164     sh->b_num_ref_idx_override = 0;
 165     sh->i_num_ref_idx_l0_active = 1;
 166     sh->i_num_ref_idx_l1_active = 1;
 167
 168     sh->b_ref_pic_list_reordering[0] = h->b_ref_reorder[0];
 169     sh->b_ref_pic_list_reordering[1] = h->b_ref_reorder[1];
 170
 171     /* If the ref list isn't in the default order, construct reordering header */
 172     for( int list = 0; list < 2; list++ )
 173     {
 174         if( sh->b_ref_pic_list_reordering[list] )
 175         {
 176             int pred_frame_num = i_frame;
 177             for( int i = 0; i < h->i_ref[list]; i++ )
 178             {
 179                 int diff = h->fref[list][i]->i_frame_num - pred_frame_num;
 180                 sh->ref_pic_list_order[list][i].idc = ( diff > 0 );
 181                 sh->ref_pic_list_order[list][i].arg = (abs(diff) - 1) & ((1 << sps->i_log2_max_frame_num) - 1);
 182                 pred_frame_num = h->fref[list][i]->i_frame_num;
 183             }
 184         }
 185     }
 186
 187     sh->i_cabac_init_idc = param->i_cabac_init_idc;
 188
 189     sh->i_qp = SPEC_QP(i_qp);
 190     sh->i_qp_delta = sh->i_qp - pps->i_pic_init_qp;
 191     sh->b_sp_for_swidth = 0;
 192     sh->i_qs_delta = 0;
 193
 194     int deblock_thresh = i_qp + 2 * X264_MIN(param->i_deblocking_filter_alphac0, param->i_deblocking_filter_beta);
 195     /* If effective qp <= 15, deblocking would have no effect anyway */
 196     if( param->b_deblocking_filter && (h->mb.b_variable_qp || 15 < deblock_thresh ) )
 197         sh->i_disable_deblocking_filter_idc = param->b_sliced_threads ? 2 : 0;
 198     else
 199         sh->i_disable_deblocking_filter_idc = 1;
 200     sh->i_alpha_c0_offset = param->i_deblocking_filter_alphac0 << 1;
 201     sh->i_beta_offset = param->i_deblocking_filter_beta << 1;
 202 }
 203
 204 static void x264_slice_header_write( bs_t *s, x264_slice_header_t *sh, int i_nal_ref_idc )
 205 {
 206     if( sh->b_mbaff )
 207     {
 208         int first_x = sh->i_first_mb % sh->sps->i_mb_width;
 209         int first_y = sh->i_first_mb / sh->sps->i_mb_width;
 210         assert( (first_y&1) == 0 );
 211         bs_write_ue( s, (2*first_x + sh->sps->i_mb_width*(first_y&~1) + (first_y&1)) >> 1 );
 212     }
 213     else
 214         bs_write_ue( s, sh->i_first_mb );
 215
 216     bs_write_ue( s, sh->i_type + 5 );   /* same type things */
 217     bs_write_ue( s, sh->i_pps_id );
 218     bs_write( s, sh->sps->i_log2_max_frame_num, sh->i_frame_num & ((1<<sh->sps->i_log2_max_frame_num)-1) );
 219
 220     if( !sh->sps->b_frame_mbs_only )
 221     {
 222         bs_write1( s, sh->b_field_pic );
 223         if( sh->b_field_pic )
 224             bs_write1( s, sh->b_bottom_field );
 225     }
 226
 227     if( sh->i_idr_pic_id >= 0 ) /* NAL IDR */
 228         bs_write_ue( s, sh->i_idr_pic_id );
 229
 230     if( sh->sps->i_poc_type == 0 )
 231     {
 232         bs_write( s, sh->sps->i_log2_max_poc_lsb, sh->i_poc & ((1<<sh->sps->i_log2_max_poc_lsb)-1) );
 233         if( sh->pps->b_pic_order && !sh->b_field_pic )
 234             bs_write_se( s, sh->i_delta_poc_bottom );
 235     }
 236
 237     if( sh->pps->b_redundant_pic_cnt )
 238         bs_write_ue( s, sh->i_redundant_pic_cnt );
 239
 240     if( sh->i_type == SLICE_TYPE_B )
 241         bs_write1( s, sh->b_direct_spatial_mv_pred );
 242
 243     if( sh->i_type == SLICE_TYPE_P || sh->i_type == SLICE_TYPE_B )
 244     {
 245         bs_write1( s, sh->b_num_ref_idx_override );
 246         if( sh->b_num_ref_idx_override )
 247         {
 248             bs_write_ue( s, sh->i_num_ref_idx_l0_active - 1 );
 249             if( sh->i_type == SLICE_TYPE_B )
 250                 bs_write_ue( s, sh->i_num_ref_idx_l1_active - 1 );
 251         }
 252     }
 253
 254     /* ref pic list reordering */
 255     if( sh->i_type != SLICE_TYPE_I )
 256     {
 257         bs_write1( s, sh->b_ref_pic_list_reordering[0] );
 258         if( sh->b_ref_pic_list_reordering[0] )
 259         {
 260             for( int i = 0; i < sh->i_num_ref_idx_l0_active; i++ )
 261             {
 262                 bs_write_ue( s, sh->ref_pic_list_order[0][i].idc );
 263                 bs_write_ue( s, sh->ref_pic_list_order[0][i].arg );
 264             }
 265             bs_write_ue( s, 3 );
 266         }
 267     }
 268     if( sh->i_type == SLICE_TYPE_B )
 269     {
 270         bs_write1( s, sh->b_ref_pic_list_reordering[1] );
 271         if( sh->b_ref_pic_list_reordering[1] )
 272         {
 273             for( int i = 0; i < sh->i_num_ref_idx_l1_active; i++ )
 274             {
 275                 bs_write_ue( s, sh->ref_pic_list_order[1][i].idc );
 276                 bs_write_ue( s, sh->ref_pic_list_order[1][i].arg );
 277             }
 278             bs_write_ue( s, 3 );
 279         }
 280     }
 281
 282     sh->b_weighted_pred = 0;
 283     if( sh->pps->b_weighted_pred && sh->i_type == SLICE_TYPE_P )
 284     {
 285         sh->b_weighted_pred = sh->weight[0][0].weightfn || sh->weight[0][1].weightfn || sh->weight[0][2].weightfn;
 286         /* pred_weight_table() */
 287         bs_write_ue( s, sh->weight[0][0].i_denom );
 288         bs_write_ue( s, sh->weight[0][1].i_denom );
 289         for( int i = 0; i < sh->i_num_ref_idx_l0_active; i++ )
 290         {
 291             int luma_weight_l0_flag = !!sh->weight[i][0].weightfn;
 292             int chroma_weight_l0_flag = !!sh->weight[i][1].weightfn || !!sh->weight[i][2].weightfn;
 293             bs_write1( s, luma_weight_l0_flag );
 294             if( luma_weight_l0_flag )
 295             {
 296                 bs_write_se( s, sh->weight[i][0].i_scale );
 297                 bs_write_se( s, sh->weight[i][0].i_offset );
 298             }
 299             bs_write1( s, chroma_weight_l0_flag );
 300             if( chroma_weight_l0_flag )
 301             {
 302                 for( int j = 1; j < 3; j++ )
 303                 {
 304                     bs_write_se( s, sh->weight[i][j].i_scale );
 305                     bs_write_se( s, sh->weight[i][j].i_offset );
 306                 }
 307             }
 308         }
 309     }
 310     else if( sh->pps->b_weighted_bipred == 1 && sh->i_type == SLICE_TYPE_B )
 311     {
 312       /* TODO */
 313     }
 314
 315     if( i_nal_ref_idc != 0 )
 316     {
 317         if( sh->i_idr_pic_id >= 0 )
 318         {
 319             bs_write1( s, 0 );  /* no output of prior pics flag */
 320             bs_write1( s, 0 );  /* long term reference flag */
 321         }
 322         else
 323         {
 324             bs_write1( s, sh->i_mmco_command_count > 0 ); /* adaptive_ref_pic_marking_mode_flag */
 325             if( sh->i_mmco_command_count > 0 )
 326             {
 327                 for( int i = 0; i < sh->i_mmco_command_count; i++ )
 328                 {
 329                     bs_write_ue( s, 1 ); /* mark short term ref as unused */
 330                     bs_write_ue( s, sh->mmco[i].i_difference_of_pic_nums - 1 );
 331                 }
 332                 bs_write_ue( s, 0 ); /* end command list */
 333             }
 334         }
 335     }
 336
 337     if( sh->pps->b_cabac && sh->i_type != SLICE_TYPE_I )
 338         bs_write_ue( s, sh->i_cabac_init_idc );
 339
 340     bs_write_se( s, sh->i_qp_delta );      /* slice qp delta */
 341
 342     if( sh->pps->b_deblocking_filter_control )
 343     {
 344         bs_write_ue( s, sh->i_disable_deblocking_filter_idc );
 345         if( sh->i_disable_deblocking_filter_idc != 1 )
 346         {
 347             bs_write_se( s, sh->i_alpha_c0_offset >> 1 );
 348             bs_write_se( s, sh->i_beta_offset >> 1 );
 349         }
 350     }
 351 }
 352
 353 /* If we are within a reasonable distance of the end of the memory allocated for the bitstream, */
 354 /* reallocate, adding an arbitrary amount of space. */
 355 static int x264_bitstream_check_buffer_internal( x264_t *h, int size, int b_cabac, int i_nal )
 356 {
 357     if( (b_cabac && (h->cabac.p_end - h->cabac.p < size)) ||
 358         (h->out.bs.p_end - h->out.bs.p < size) )
 359     {
 360         int buf_size = h->out.i_bitstream + size;
 361         uint8_t *buf = x264_malloc( buf_size );
 362         if( !buf )
 363             return -1;
 364         int aligned_size = h->out.i_bitstream & ~15;
 365         h->mc.memcpy_aligned( buf, h->out.p_bitstream, aligned_size );
 366         memcpy( buf + aligned_size, h->out.p_bitstream + aligned_size, h->out.i_bitstream - aligned_size );
 367
 368         intptr_t delta = buf - h->out.p_bitstream;
 369
 370         h->out.bs.p_start += delta;
 371         h->out.bs.p += delta;
 372         h->out.bs.p_end = buf + buf_size;
 373
 374         h->cabac.p_start += delta;
 375         h->cabac.p += delta;
 376         h->cabac.p_end = buf + buf_size;
 377
 378         for( int i = 0; i <= i_nal; i++ )
 379             h->out.nal[i].p_payload += delta;
 380
 381         x264_free( h->out.p_bitstream );
 382         h->out.p_bitstream = buf;
 383         h->out.i_bitstream = buf_size;
 384     }
 385     return 0;
 386 }
 387
 388 static int x264_bitstream_check_buffer( x264_t *h )
 389 {
 390     int max_row_size = (2500 << SLICE_MBAFF) * h->mb.i_mb_width;
 391     return x264_bitstream_check_buffer_internal( h, max_row_size, h->param.b_cabac, h->out.i_nal );
 392 }
 393
 394 static int x264_bitstream_check_buffer_filler( x264_t *h, int filler )
 395 {
 396     filler += 32; // add padding for safety
 397     return x264_bitstream_check_buffer_internal( h, filler, 0, -1 );
 398 }
 399
 400 #if HAVE_THREAD
 401 static void x264_encoder_thread_init( x264_t *h )
 402 {
 403     if( h->param.i_sync_lookahead )
 404         x264_lower_thread_priority( 10 );
 405 }
 406 #endif
 407
 408 /****************************************************************************
 409  *
 410  ****************************************************************************
 411  ****************************** External API*********************************
 412  ****************************************************************************
 413  *
 414  ****************************************************************************/
 415
 416 static int x264_validate_parameters( x264_t *h, int b_open )
 417 {
 418     if( !h->param.pf_log )
 419     {
 420         x264_log( NULL, X264_LOG_ERROR, "pf_log not set! did you forget to call x264_param_default?\n" );
 421         return -1;
 422     }
 423
 424 #if HAVE_MMX
 425     if( b_open )
 426     {
 427         int cpuflags = x264_cpu_detect();
 428         int fail = 0;
 429 #ifdef __SSE__
 430         if( !(cpuflags & X264_CPU_SSE) )
 431         {
 432             x264_log( h, X264_LOG_ERROR, "your cpu does not support SSE1, but x264 was compiled with asm\n");
 433             fail = 1;
 434         }
 435 #else
 436         if( !(cpuflags & X264_CPU_MMX2) )
 437         {
 438             x264_log( h, X264_LOG_ERROR, "your cpu does not support MMXEXT, but x264 was compiled with asm\n");
 439             fail = 1;
 440         }
 441 #endif
 442         if( !fail && !(cpuflags & X264_CPU_CMOV) )
 443         {
 444             x264_log( h, X264_LOG_ERROR, "your cpu does not support CMOV, but x264 was compiled with asm\n");
 445             fail = 1;
 446         }
 447         if( fail )
 448         {
 449             x264_log( h, X264_LOG_ERROR, "to run x264, recompile without asm (configure --disable-asm)\n");
 450             return -1;
 451         }
 452     }
 453 #endif
 454
 455 #if HAVE_INTERLACED
 456     h->param.b_interlaced = !!PARAM_INTERLACED;
 457 #else
 458     if( h->param.b_interlaced )
 459     {
 460         x264_log( h, X264_LOG_ERROR, "not compiled with interlaced support\n" );
 461         return -1;
 462     }
 463 #endif
 464
 465     if( h->param.i_width <= 0 || h->param.i_height <= 0 )
 466     {
 467         x264_log( h, X264_LOG_ERROR, "invalid width x height (%dx%d)\n",
 468                   h->param.i_width, h->param.i_height );
 469         return -1;
 470     }
 471
 472     int i_csp = h->param.i_csp & X264_CSP_MASK;
 473 #if X264_CHROMA_FORMAT
 474     if( CHROMA_FORMAT != CHROMA_420 && i_csp >= X264_CSP_I420 && i_csp <= X264_CSP_NV12 )
 475     {
 476         x264_log( h, X264_LOG_ERROR, "not compiled with 4:2:0 support\n" );
 477         return -1;
 478     }
 479     else if( CHROMA_FORMAT != CHROMA_422 && i_csp >= X264_CSP_I422 && i_csp <= X264_CSP_V210 )
 480     {
 481         x264_log( h, X264_LOG_ERROR, "not compiled with 4:2:2 support\n" );
 482         return -1;
 483     }
 484     else if( CHROMA_FORMAT != CHROMA_444 && i_csp >= X264_CSP_I444 && i_csp <= X264_CSP_RGB )
 485     {
 486         x264_log( h, X264_LOG_ERROR, "not compiled with 4:4:4 support\n" );
 487         return -1;
 488     }
 489 #endif
 490     if( i_csp <= X264_CSP_NONE || i_csp >= X264_CSP_MAX )
 491     {
 492         x264_log( h, X264_LOG_ERROR, "invalid CSP (only I420/YV12/NV12/I422/YV16/NV16/I444/YV24/BGR/BGRA/RGB supported)\n" );
 493         return -1;
 494     }
 495
 496     if( i_csp < X264_CSP_I444 && h->param.i_width % 2 )
 497     {
 498         x264_log( h, X264_LOG_ERROR, "width not divisible by 2 (%dx%d)\n",
 499                   h->param.i_width, h->param.i_height );
 500         return -1;
 501     }
 502
 503     if( i_csp < X264_CSP_I422 && PARAM_INTERLACED && h->param.i_height % 4 )
 504     {
 505         x264_log( h, X264_LOG_ERROR, "height not divisible by 4 (%dx%d)\n",
 506                   h->param.i_width, h->param.i_height );
 507         return -1;
 508     }
 509
 510     if( (i_csp < X264_CSP_I422 || PARAM_INTERLACED) && h->param.i_height % 2 )
 511     {
 512         x264_log( h, X264_LOG_ERROR, "height not divisible by 2 (%dx%d)\n",
 513                   h->param.i_width, h->param.i_height );
 514         return -1;
 515     }
 516
 517     if( (h->param.crop_rect.i_left + h->param.crop_rect.i_right ) >= h->param.i_width ||
 518         (h->param.crop_rect.i_top  + h->param.crop_rect.i_bottom) >= h->param.i_height )
 519     {
 520         x264_log( h, X264_LOG_ERROR, "invalid crop-rect %u,%u,%u,%u\n", h->param.crop_rect.i_left,
 521                   h->param.crop_rect.i_top, h->param.crop_rect.i_right,  h->param.crop_rect.i_bottom );
 522         return -1;
 523     }
 524
 525     if( h->param.vui.i_sar_width <= 0 || h->param.vui.i_sar_height <= 0 )
 526     {
 527         h->param.vui.i_sar_width = 0;
 528         h->param.vui.i_sar_height = 0;
 529     }
 530
 531     if( h->param.i_threads == X264_THREADS_AUTO )
 532         h->param.i_threads = x264_cpu_num_processors() * (h->param.b_sliced_threads?2:3)/2;
 533     int max_sliced_threads = X264_MAX( 1, (h->param.i_height+15)/16 / 4 );
 534     if( h->param.i_threads > 1 )
 535     {
 536 #if !HAVE_THREAD
 537         x264_log( h, X264_LOG_WARNING, "not compiled with thread support!\n");
 538         h->param.i_threads = 1;
 539 #endif
 540         /* Avoid absurdly small thread slices as they can reduce performance
 541          * and VBV compliance.  Capped at an arbitrary 4 rows per thread. */
 542         if( h->param.b_sliced_threads )
 543             h->param.i_threads = X264_MIN( h->param.i_threads, max_sliced_threads );
 544     }
 545     h->param.i_threads = x264_clip3( h->param.i_threads, 1, X264_THREAD_MAX );
 546     if( h->param.i_threads == 1 )
 547     {
 548         h->param.b_sliced_threads = 0;
 549         h->param.i_lookahead_threads = 1;
 550     }
 551     h->i_thread_frames = h->param.b_sliced_threads ? 1 : h->param.i_threads;
 552     if( h->i_thread_frames > 1 )
 553         h->param.nalu_process = NULL;
 554
 555     if( h->param.b_opencl )
 556     {
 557 #if !HAVE_OPENCL
 558         x264_log( h, X264_LOG_WARNING, "OpenCL: not compiled with OpenCL support, disabling\n" );
 559         h->param.b_opencl = 0;
 560 #elif BIT_DEPTH > 8
 561         x264_log( h, X264_LOG_WARNING, "OpenCL lookahead does not support high bit depth, disabling opencl\n" );
 562         h->param.b_opencl = 0;
 563 #else
 564         if( h->param.i_width < 32 || h->param.i_height < 32 )
 565         {
 566             x264_log( h, X264_LOG_WARNING, "OpenCL: frame size is too small, disabling opencl\n" );
 567             h->param.b_opencl = 0;
 568         }
 569 #endif
 570         if( h->param.opencl_device_id && h->param.i_opencl_device )
 571         {
 572             x264_log( h, X264_LOG_WARNING, "OpenCL: device id and device skip count configured; dropping skip\n" );
 573             h->param.i_opencl_device = 0;
 574         }
 575     }
 576
 577     h->param.i_keyint_max = x264_clip3( h->param.i_keyint_max, 1, X264_KEYINT_MAX_INFINITE );
 578     if( h->param.i_keyint_max == 1 )
 579     {
 580         h->param.b_intra_refresh = 0;
 581         h->param.analyse.i_weighted_pred = 0;
 582         h->param.i_frame_reference = 1;
 583         h->param.i_dpb_size = 1;
 584     }
 585
 586     h->param.i_frame_packing = x264_clip3( h->param.i_frame_packing, -1, 5 );
 587
 588     /* Detect default ffmpeg settings and terminate with an error. */
 589     if( b_open )
 590     {
 591         int score = 0;
 592         score += h->param.analyse.i_me_range == 0;
 593         score += h->param.rc.i_qp_step == 3;
 594         score += h->param.i_keyint_max == 12;
 595         score += h->param.rc.i_qp_min == 2;
 596         score += h->param.rc.i_qp_max == 31;
 597         score += h->param.rc.f_qcompress == 0.5;
 598         score += fabs(h->param.rc.f_ip_factor - 1.25) < 0.01;
 599         score += fabs(h->param.rc.f_pb_factor - 1.25) < 0.01;
 600         score += h->param.analyse.inter == 0 && h->param.analyse.i_subpel_refine == 8;
 601         if( score >= 5 )
 602         {
 603             x264_log( h, X264_LOG_ERROR, "broken ffmpeg default settings detected\n" );
 604             x264_log( h, X264_LOG_ERROR, "use an encoding preset (e.g. -vpre medium)\n" );
 605             x264_log( h, X264_LOG_ERROR, "preset usage: -vpre <speed> -vpre <profile>\n" );
 606             x264_log( h, X264_LOG_ERROR, "speed presets are listed in x264 --help\n" );
 607             x264_log( h, X264_LOG_ERROR, "profile is optional; x264 defaults to high\n" );
 608             return -1;
 609         }
 610     }
 611
 612     if( h->param.rc.i_rc_method < 0 || h->param.rc.i_rc_method > 2 )
 613     {
 614         x264_log( h, X264_LOG_ERROR, "no ratecontrol method specified\n" );
 615         return -1;
 616     }
 617
 618     if( PARAM_INTERLACED )
 619         h->param.b_pic_struct = 1;
 620
 621     if( h->param.i_avcintra_class )
 622     {
 623         if( BIT_DEPTH != 10 )
 624         {
 625             x264_log( h, X264_LOG_ERROR, "%2d-bit AVC-Intra is not widely compatible\n", BIT_DEPTH );
 626             x264_log( h, X264_LOG_ERROR, "10-bit x264 is required to encode AVC-Intra\n" );
 627             return -1;
 628         }
 629
 630         int type = h->param.i_avcintra_class == 200 ? 2 :
 631                    h->param.i_avcintra_class == 100 ? 1 :
 632                    h->param.i_avcintra_class == 50 ? 0 : -1;
 633         if( type < 0 )
 634         {
 635             x264_log( h, X264_LOG_ERROR, "Invalid AVC-Intra class\n" );
 636             return -1;
 637         }
 638
 639         /* [50/100/200][res][fps] */
 640         static const struct
 641         {
 642             uint16_t fps_num;
 643             uint16_t fps_den;
 644             uint8_t interlaced;
 645             uint16_t frame_size;
 646             const uint8_t *cqm_4ic;
 647             const uint8_t *cqm_8iy;
 648         } avcintra_lut[3][2][7] =
 649         {
 650             {{{ 60000, 1001, 0,  912, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy },
 651               {    50,    1, 0, 1100, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy },
 652               { 30000, 1001, 0,  912, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy },
 653               {    25,    1, 0, 1100, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy },
 654               { 24000, 1001, 0,  912, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy }},
 655              {{ 30000, 1001, 1, 1820, x264_cqm_avci50_4ic, x264_cqm_avci50_1080i_8iy },
 656               {    25,    1, 1, 2196, x264_cqm_avci50_4ic, x264_cqm_avci50_1080i_8iy },
 657               { 60000, 1001, 0, 1820, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy },
 658               { 30000, 1001, 0, 1820, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy },
 659               {    50,    1, 0, 2196, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy },
 660               {    25,    1, 0, 2196, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy },
 661               { 24000, 1001, 0, 1820, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy }}},
 662             {{{ 60000, 1001, 0, 1848, x264_cqm_avci100_720p_4ic, x264_cqm_avci100_720p_8iy },
 663               {    50,    1, 0, 2224, x264_cqm_avci100_720p_4ic, x264_cqm_avci100_720p_8iy },
 664               { 30000, 1001, 0, 1848, x264_cqm_avci100_720p_4ic, x264_cqm_avci100_720p_8iy },
 665               {    25,    1, 0, 2224, x264_cqm_avci100_720p_4ic, x264_cqm_avci100_720p_8iy },
 666               { 24000, 1001, 0, 1848, x264_cqm_avci100_720p_4ic, x264_cqm_avci100_720p_8iy }},
 667              {{ 30000, 1001, 1, 3692, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080i_8iy },
 668               {    25,    1, 1, 4444, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080i_8iy },
 669               { 60000, 1001, 0, 3692, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080p_8iy },
 670               { 30000, 1001, 0, 3692, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080p_8iy },
 671               {    50,    1, 0, 4444, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080p_8iy },
 672               {    25,    1, 0, 4444, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080p_8iy },
 673               { 24000, 1001, 0, 3692, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080p_8iy }}},
 674             {{{ 60000, 1001, 0, 3724, x264_cqm_avci100_720p_4ic, x264_cqm_avci100_720p_8iy },
 675               {    50,    1, 0, 4472, x264_cqm_avci100_720p_4ic, x264_cqm_avci100_720p_8iy }},
 676              {{ 30000, 1001, 1, 7444, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080i_8iy },
 677               {    25,    1, 1, 8940, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080i_8iy },
 678               { 60000, 1001, 0, 7444, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080p_8iy },
 679               { 30000, 1001, 0, 7444, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080p_8iy },
 680               {    50,    1, 0, 8940, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080p_8iy },
 681               {    25,    1, 0, 8940, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080p_8iy },
 682               { 24000, 1001, 0, 7444, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080p_8iy }}}
 683         };
 684
 685         int res = -1;
 686         if( i_csp >= X264_CSP_I420 && i_csp < X264_CSP_I422 && !type )
 687         {
 688             if(      h->param.i_width == 1440 && h->param.i_height == 1080 ) res =  1;
 689             else if( h->param.i_width ==  960 && h->param.i_height ==  720 ) res =  0;
 690         }
 691         else if( i_csp >= X264_CSP_I422 && i_csp < X264_CSP_I444 && type )
 692         {
 693             if(      h->param.i_width == 1920 && h->param.i_height == 1080 ) res =  1;
 694             else if( h->param.i_width == 1280 && h->param.i_height ==  720 ) res =  0;
 695         }
 696         else
 697         {
 698             x264_log( h, X264_LOG_ERROR, "Invalid colorspace for AVC-Intra %d\n", h->param.i_avcintra_class );
 699             return -1;
 700         }
 701
 702         if( res < 0 )
 703         {
 704             x264_log( h, X264_LOG_ERROR, "Resolution %dx%d invalid for AVC-Intra %d\n",
 705                       h->param.i_width, h->param.i_height, h->param.i_avcintra_class );
 706             return -1;
 707         }
 708
 709         if( h->param.nalu_process )
 710         {
 711             x264_log( h, X264_LOG_ERROR, "nalu_process is not supported in AVC-Intra mode\n" );
 712             return -1;
 713         }
 714
 715         if( !h->param.b_repeat_headers )
 716         {
 717             x264_log( h, X264_LOG_ERROR, "Separate headers not supported in AVC-Intra mode\n" );
 718             return -1;
 719         }
 720
 721         int i;
 722         uint32_t fps_num = h->param.i_fps_num, fps_den = h->param.i_fps_den;
 723         x264_reduce_fraction( &fps_num, &fps_den );
 724         for( i = 0; i < 7; i++ )
 725         {
 726             if( avcintra_lut[type][res][i].fps_num == fps_num &&
 727                 avcintra_lut[type][res][i].fps_den == fps_den &&
 728                 avcintra_lut[type][res][i].interlaced == PARAM_INTERLACED )
 729             {
 730                 break;
 731             }
 732         }
 733         if( i == 7 )
 734         {
 735             x264_log( h, X264_LOG_ERROR, "FPS %d/%d%c not compatible with AVC-Intra\n",
 736                       h->param.i_fps_num, h->param.i_fps_den, PARAM_INTERLACED ? 'i' : 'p' );
 737             return -1;
 738         }
 739
 740         h->param.i_keyint_max = 1;
 741         h->param.b_intra_refresh = 0;
 742         h->param.analyse.i_weighted_pred = 0;
 743         h->param.i_frame_reference = 1;
 744         h->param.i_dpb_size = 1;
 745
 746         h->param.b_bluray_compat = 0;
 747         h->param.b_vfr_input = 0;
 748         h->param.b_aud = 1;
 749         h->param.vui.i_chroma_loc = 0;
 750         h->param.i_nal_hrd = X264_NAL_HRD_NONE;
 751         h->param.b_deblocking_filter = 0;
 752         h->param.b_stitchable = 1;
 753         h->param.b_pic_struct = 0;
 754         h->param.analyse.b_transform_8x8 = 1;
 755         h->param.analyse.intra = X264_ANALYSE_I8x8;
 756         h->param.analyse.i_chroma_qp_offset = res && type ? 3 : 4;
 757         h->param.b_cabac = !type;
 758         h->param.rc.i_vbv_buffer_size = avcintra_lut[type][res][i].frame_size;
 759         h->param.rc.i_vbv_max_bitrate =
 760         h->param.rc.i_bitrate = h->param.rc.i_vbv_buffer_size * fps_num / fps_den;
 761         h->param.rc.i_rc_method = X264_RC_ABR;
 762         h->param.rc.f_vbv_buffer_init = 1.0;
 763         h->param.rc.b_filler = 1;
 764         h->param.i_cqm_preset = X264_CQM_CUSTOM;
 765         memcpy( h->param.cqm_4iy, x264_cqm_jvt4i, sizeof(h->param.cqm_4iy) );
 766         memcpy( h->param.cqm_4ic, avcintra_lut[type][res][i].cqm_4ic, sizeof(h->param.cqm_4ic) );
 767         memcpy( h->param.cqm_8iy, avcintra_lut[type][res][i].cqm_8iy, sizeof(h->param.cqm_8iy) );
 768
 769         /* Need exactly 10 slices of equal MB count... why?  $deity knows... */
 770         h->param.i_slice_max_mbs = ((h->param.i_width + 15) / 16) * ((h->param.i_height + 15) / 16) / 10;
 771         h->param.i_slice_max_size = 0;
 772         /* The slice structure only allows a maximum of 2 threads for 1080i/p
 773          * and 1 or 5 threads for 720p */
 774         if( h->param.b_sliced_threads )
 775         {
 776             if( res )
 777                 h->param.i_threads = X264_MIN( 2, h->param.i_threads );
 778             else
 779             {
 780                 h->param.i_threads = X264_MIN( 5, h->param.i_threads );
 781                 if( h->param.i_threads < 5 )
 782                     h->param.i_threads = 1;
 783             }
 784         }
 785
 786         if( type )
 787             h->param.vui.i_sar_width = h->param.vui.i_sar_height = 1;
 788         else
 789         {
 790             h->param.vui.i_sar_width  = 4;
 791             h->param.vui.i_sar_height = 3;
 792         }
 793
 794         /* Official encoder doesn't appear to go under 13
 795          * and Avid cannot handle negative QPs */
 796         h->param.rc.i_qp_min = X264_MAX( h->param.rc.i_qp_min, QP_BD_OFFSET + 1 );
 797     }
 798
 799     h->param.rc.f_rf_constant = x264_clip3f( h->param.rc.f_rf_constant, -QP_BD_OFFSET, 51 );
 800     h->param.rc.f_rf_constant_max = x264_clip3f( h->param.rc.f_rf_constant_max, -QP_BD_OFFSET, 51 );
 801     h->param.rc.i_qp_constant = x264_clip3( h->param.rc.i_qp_constant, 0, QP_MAX );
 802     h->param.analyse.i_subpel_refine = x264_clip3( h->param.analyse.i_subpel_refine, 0, 11 );
 803     h->param.rc.f_ip_factor = X264_MAX( h->param.rc.f_ip_factor, 0.01f );
 804     h->param.rc.f_pb_factor = X264_MAX( h->param.rc.f_pb_factor, 0.01f );
 805     if( h->param.rc.i_rc_method == X264_RC_CRF )
 806     {
 807         h->param.rc.i_qp_constant = h->param.rc.f_rf_constant + QP_BD_OFFSET;
 808         h->param.rc.i_bitrate = 0;
 809     }
 810     if( b_open && (h->param.rc.i_rc_method == X264_RC_CQP || h->param.rc.i_rc_method == X264_RC_CRF)
 811         && h->param.rc.i_qp_constant == 0 )
 812     {
 813         h->mb.b_lossless = 1;
 814         h->param.i_cqm_preset = X264_CQM_FLAT;
 815         h->param.psz_cqm_file = NULL;
 816         h->param.rc.i_rc_method = X264_RC_CQP;
 817         h->param.rc.f_ip_factor = 1;
 818         h->param.rc.f_pb_factor = 1;
 819         h->param.analyse.b_psnr = 0;
 820         h->param.analyse.b_ssim = 0;
 821         h->param.analyse.i_chroma_qp_offset = 0;
 822         h->param.analyse.i_trellis = 0;
 823         h->param.analyse.b_fast_pskip = 0;
 824         h->param.analyse.i_noise_reduction = 0;
 825         h->param.analyse.b_psy = 0;
 826         h->param.i_bframe = 0;
 827         /* 8x8dct is not useful without RD in CAVLC lossless */
 828         if( !h->param.b_cabac && h->param.analyse.i_subpel_refine < 6 )
 829             h->param.analyse.b_transform_8x8 = 0;
 830         h->param.analyse.inter &= ~X264_ANALYSE_I8x8;
 831         h->param.analyse.intra &= ~X264_ANALYSE_I8x8;
 832     }
 833     if( h->param.rc.i_rc_method == X264_RC_CQP )
 834     {
 835         float qp_p = h->param.rc.i_qp_constant;
 836         float qp_i = qp_p - 6*log2f( h->param.rc.f_ip_factor );
 837         float qp_b = qp_p + 6*log2f( h->param.rc.f_pb_factor );
 838         h->param.rc.i_qp_min = x264_clip3( (int)(X264_MIN3( qp_p, qp_i, qp_b )), 0, QP_MAX );
 839         h->param.rc.i_qp_max = x264_clip3( (int)(X264_MAX3( qp_p, qp_i, qp_b ) + .999), 0, QP_MAX );
 840         h->param.rc.i_aq_mode = 0;
 841         h->param.rc.b_mb_tree = 0;
 842         h->param.rc.i_bitrate = 0;
 843     }
 844     h->param.rc.i_qp_max = x264_clip3( h->param.rc.i_qp_max, 0, QP_MAX );
 845     h->param.rc.i_qp_min = x264_clip3( h->param.rc.i_qp_min, 0, h->param.rc.i_qp_max );
 846     h->param.rc.i_qp_step = x264_clip3( h->param.rc.i_qp_step, 2, QP_MAX );
 847     h->param.rc.i_bitrate = x264_clip3( h->param.rc.i_bitrate, 0, 2000000 );
 848     if( h->param.rc.i_rc_method == X264_RC_ABR && !h->param.rc.i_bitrate )
 849     {
 850         x264_log( h, X264_LOG_ERROR, "bitrate not specified\n" );
 851         return -1;
 852     }
 853     h->param.rc.i_vbv_buffer_size = x264_clip3( h->param.rc.i_vbv_buffer_size, 0, 2000000 );
 854     h->param.rc.i_vbv_max_bitrate = x264_clip3( h->param.rc.i_vbv_max_bitrate, 0, 2000000 );
 855     h->param.rc.f_vbv_buffer_init = x264_clip3f( h->param.rc.f_vbv_buffer_init, 0, 2000000 );
 856     if( h->param.rc.i_vbv_buffer_size )
 857     {
 858         if( h->param.rc.i_rc_method == X264_RC_CQP )
 859         {
 860             x264_log( h, X264_LOG_WARNING, "VBV is incompatible with constant QP, ignored.\n" );
 861             h->param.rc.i_vbv_max_bitrate = 0;
 862             h->param.rc.i_vbv_buffer_size = 0;
 863         }
 864         else if( h->param.rc.i_vbv_max_bitrate == 0 )
 865         {
 866             if( h->param.rc.i_rc_method == X264_RC_ABR )
 867             {
 868                 x264_log( h, X264_LOG_WARNING, "VBV maxrate unspecified, assuming CBR\n" );
 869                 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate;
 870             }
 871             else
 872             {
 873                 x264_log( h, X264_LOG_WARNING, "VBV bufsize set but maxrate unspecified, ignored\n" );
 874                 h->param.rc.i_vbv_buffer_size = 0;
 875             }
 876         }
 877         else if( h->param.rc.i_vbv_max_bitrate < h->param.rc.i_bitrate &&
 878                  h->param.rc.i_rc_method == X264_RC_ABR )
 879         {
 880             x264_log( h, X264_LOG_WARNING, "max bitrate less than average bitrate, assuming CBR\n" );
 881             h->param.rc.i_bitrate = h->param.rc.i_vbv_max_bitrate;
 882         }
 883     }
 884     else if( h->param.rc.i_vbv_max_bitrate )
 885     {
 886         x264_log( h, X264_LOG_WARNING, "VBV maxrate specified, but no bufsize, ignored\n" );
 887         h->param.rc.i_vbv_max_bitrate = 0;
 888     }
 889
 890     h->param.i_slice_max_size = X264_MAX( h->param.i_slice_max_size, 0 );
 891     h->param.i_slice_max_mbs = X264_MAX( h->param.i_slice_max_mbs, 0 );
 892     h->param.i_slice_min_mbs = X264_MAX( h->param.i_slice_min_mbs, 0 );
 893     if( h->param.i_slice_max_mbs )
 894         h->param.i_slice_min_mbs = X264_MIN( h->param.i_slice_min_mbs, h->param.i_slice_max_mbs/2 );
 895     else if( !h->param.i_slice_max_size )
 896         h->param.i_slice_min_mbs = 0;
 897     if( PARAM_INTERLACED && h->param.i_slice_min_mbs )
 898     {
 899         x264_log( h, X264_LOG_WARNING, "interlace + slice-min-mbs is not implemented\n" );
 900         h->param.i_slice_min_mbs = 0;
 901     }
 902     int mb_width = (h->param.i_width+15)/16;
 903     if( h->param.i_slice_min_mbs > mb_width )
 904     {
 905         x264_log( h, X264_LOG_WARNING, "slice-min-mbs > row mb size (%d) not implemented\n", mb_width );
 906         h->param.i_slice_min_mbs = mb_width;
 907     }
 908
 909     int max_slices = (h->param.i_height+((16<<PARAM_INTERLACED)-1))/(16<<PARAM_INTERLACED);
 910     if( h->param.b_sliced_threads )
 911         h->param.i_slice_count = x264_clip3( h->param.i_threads, 0, max_slices );
 912     else
 913     {
 914         h->param.i_slice_count = x264_clip3( h->param.i_slice_count, 0, max_slices );
 915         if( h->param.i_slice_max_mbs || h->param.i_slice_max_size )
 916             h->param.i_slice_count = 0;
 917     }
 918     if( h->param.i_slice_count_max > 0 )
 919         h->param.i_slice_count_max = X264_MAX( h->param.i_slice_count, h->param.i_slice_count_max );
 920
 921     if( h->param.b_bluray_compat )
 922     {
 923         h->param.i_bframe_pyramid = X264_MIN( X264_B_PYRAMID_STRICT, h->param.i_bframe_pyramid );
 924         h->param.i_bframe = X264_MIN( h->param.i_bframe, 3 );
 925         h->param.b_aud = 1;
 926         h->param.i_nal_hrd = X264_MAX( h->param.i_nal_hrd, X264_NAL_HRD_VBR );
 927         h->param.i_slice_max_size = 0;
 928         h->param.i_slice_max_mbs = 0;
 929         h->param.b_intra_refresh = 0;
 930         h->param.i_frame_reference = X264_MIN( h->param.i_frame_reference, 6 );
 931         h->param.i_dpb_size = X264_MIN( h->param.i_dpb_size, 6 );
 932         /* Don't use I-frames, because Blu-ray treats them the same as IDR. */
 933         h->param.i_keyint_min = 1;
 934         /* Due to the proliferation of broken players that don't handle dupes properly. */
 935         h->param.analyse.i_weighted_pred = X264_MIN( h->param.analyse.i_weighted_pred, X264_WEIGHTP_SIMPLE );
 936         if( h->param.b_fake_interlaced )
 937             h->param.b_pic_struct = 1;
 938     }
 939
 940     h->param.i_frame_reference = x264_clip3( h->param.i_frame_reference, 1, X264_REF_MAX );
 941     h->param.i_dpb_size = x264_clip3( h->param.i_dpb_size, 1, X264_REF_MAX );
 942     if( h->param.i_scenecut_threshold < 0 )
 943         h->param.i_scenecut_threshold = 0;
 944     h->param.analyse.i_direct_mv_pred = x264_clip3( h->param.analyse.i_direct_mv_pred, X264_DIRECT_PRED_NONE, X264_DIRECT_PRED_AUTO );
 945     if( !h->param.analyse.i_subpel_refine && h->param.analyse.i_direct_mv_pred > X264_DIRECT_PRED_SPATIAL )
 946     {
 947         x264_log( h, X264_LOG_WARNING, "subme=0 + direct=temporal is not supported\n" );
 948         h->param.analyse.i_direct_mv_pred = X264_DIRECT_PRED_SPATIAL;
 949     }
 950     h->param.i_bframe = x264_clip3( h->param.i_bframe, 0, X264_MIN( X264_BFRAME_MAX, h->param.i_keyint_max-1 ) );
 951     h->param.i_bframe_bias = x264_clip3( h->param.i_bframe_bias, -90, 100 );
 952     if( h->param.i_bframe <= 1 )
 953         h->param.i_bframe_pyramid = X264_B_PYRAMID_NONE;
 954     h->param.i_bframe_pyramid = x264_clip3( h->param.i_bframe_pyramid, X264_B_PYRAMID_NONE, X264_B_PYRAMID_NORMAL );
 955     h->param.i_bframe_adaptive = x264_clip3( h->param.i_bframe_adaptive, X264_B_ADAPT_NONE, X264_B_ADAPT_TRELLIS );
 956     if( !h->param.i_bframe )
 957     {
 958         h->param.i_bframe_adaptive = X264_B_ADAPT_NONE;
 959         h->param.analyse.i_direct_mv_pred = 0;
 960         h->param.analyse.b_weighted_bipred = 0;
 961         h->param.b_open_gop = 0;
 962     }
 963     if( h->param.b_intra_refresh && h->param.i_bframe_pyramid == X264_B_PYRAMID_NORMAL )
 964     {
 965         x264_log( h, X264_LOG_WARNING, "b-pyramid normal + intra-refresh is not supported\n" );
 966         h->param.i_bframe_pyramid = X264_B_PYRAMID_STRICT;
 967     }
 968     if( h->param.b_intra_refresh && (h->param.i_frame_reference > 1 || h->param.i_dpb_size > 1) )
 969     {
 970         x264_log( h, X264_LOG_WARNING, "ref > 1 + intra-refresh is not supported\n" );
 971         h->param.i_frame_reference = 1;
 972         h->param.i_dpb_size = 1;
 973     }
 974     if( h->param.b_intra_refresh && h->param.b_open_gop )
 975     {
 976         x264_log( h, X264_LOG_WARNING, "intra-refresh is not compatible with open-gop\n" );
 977         h->param.b_open_gop = 0;
 978     }
 979     if( !h->param.i_fps_num || !h->param.i_fps_den )
 980     {
 981         h->param.i_fps_num = 25;
 982         h->param.i_fps_den = 1;
 983     }
 984     float fps = (float) h->param.i_fps_num / h->param.i_fps_den;
 985     if( h->param.i_keyint_min == X264_KEYINT_MIN_AUTO )
 986         h->param.i_keyint_min = X264_MIN( h->param.i_keyint_max / 10, fps );
 987     h->param.i_keyint_min = x264_clip3( h->param.i_keyint_min, 1, h->param.i_keyint_max/2+1 );
 988     h->param.rc.i_lookahead = x264_clip3( h->param.rc.i_lookahead, 0, X264_LOOKAHEAD_MAX );
 989     {
 990         int maxrate = X264_MAX( h->param.rc.i_vbv_max_bitrate, h->param.rc.i_bitrate );
 991         float bufsize = maxrate ? (float)h->param.rc.i_vbv_buffer_size / maxrate : 0;
 992         h->param.rc.i_lookahead = X264_MIN( h->param.rc.i_lookahead, X264_MAX( h->param.i_keyint_max, bufsize*fps ) );
 993     }
 994
 995     if( !h->param.i_timebase_num || !h->param.i_timebase_den || !(h->param.b_vfr_input || h->param.b_pulldown) )
 996     {
 997         h->param.i_timebase_num = h->param.i_fps_den;
 998         h->param.i_timebase_den = h->param.i_fps_num;
 999     }
1000
1001     h->param.rc.f_qcompress = x264_clip3f( h->param.rc.f_qcompress, 0.0, 1.0 );
1002     if( h->param.i_keyint_max == 1 || h->param.rc.f_qcompress == 1 )
1003         h->param.rc.b_mb_tree = 0;
1004     if( (!h->param.b_intra_refresh && h->param.i_keyint_max != X264_KEYINT_MAX_INFINITE) &&
1005         !h->param.rc.i_lookahead && h->param.rc.b_mb_tree )
1006     {
1007         x264_log( h, X264_LOG_WARNING, "lookaheadless mb-tree requires intra refresh or infinite keyint\n" );
1008         h->param.rc.b_mb_tree = 0;
1009     }
1010     if( b_open && h->param.rc.b_stat_read )
1011         h->param.rc.i_lookahead = 0;
1012 #if HAVE_THREAD
1013     if( h->param.i_sync_lookahead < 0 )
1014         h->param.i_sync_lookahead = h->param.i_bframe + 1;
1015     h->param.i_sync_lookahead = X264_MIN( h->param.i_sync_lookahead, X264_LOOKAHEAD_MAX );
1016     if( h->param.rc.b_stat_read || h->i_thread_frames == 1 )
1017         h->param.i_sync_lookahead = 0;
1018 #else
1019     h->param.i_sync_lookahead = 0;
1020 #endif
1021
1022     h->param.i_deblocking_filter_alphac0 = x264_clip3( h->param.i_deblocking_filter_alphac0, -6, 6 );
1023     h->param.i_deblocking_filter_beta    = x264_clip3( h->param.i_deblocking_filter_beta, -6, 6 );
1024     h->param.analyse.i_luma_deadzone[0] = x264_clip3( h->param.analyse.i_luma_deadzone[0], 0, 32 );
1025     h->param.analyse.i_luma_deadzone[1] = x264_clip3( h->param.analyse.i_luma_deadzone[1], 0, 32 );
1026
1027     h->param.i_cabac_init_idc = x264_clip3( h->param.i_cabac_init_idc, 0, 2 );
1028
1029     if( h->param.i_cqm_preset < X264_CQM_FLAT || h->param.i_cqm_preset > X264_CQM_CUSTOM )
1030         h->param.i_cqm_preset = X264_CQM_FLAT;
1031
1032     if( h->param.analyse.i_me_method < X264_ME_DIA ||
1033         h->param.analyse.i_me_method > X264_ME_TESA )
1034         h->param.analyse.i_me_method = X264_ME_HEX;
1035     h->param.analyse.i_me_range = x264_clip3( h->param.analyse.i_me_range, 4, 1024 );
1036     if( h->param.analyse.i_me_range > 16 && h->param.analyse.i_me_method <= X264_ME_HEX )
1037         h->param.analyse.i_me_range = 16;
1038     if( h->param.analyse.i_me_method == X264_ME_TESA &&
1039         (h->mb.b_lossless || h->param.analyse.i_subpel_refine <= 1) )
1040         h->param.analyse.i_me_method = X264_ME_ESA;
1041     h->param.analyse.b_mixed_references = h->param.analyse.b_mixed_references && h->param.i_frame_reference > 1;
1042     h->param.analyse.inter &= X264_ANALYSE_PSUB16x16|X264_ANALYSE_PSUB8x8|X264_ANALYSE_BSUB16x16|
1043                               X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
1044     h->param.analyse.intra &= X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
1045     if( !(h->param.analyse.inter & X264_ANALYSE_PSUB16x16) )
1046         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
1047     if( !h->param.analyse.b_transform_8x8 )
1048     {
1049         h->param.analyse.inter &= ~X264_ANALYSE_I8x8;
1050         h->param.analyse.intra &= ~X264_ANALYSE_I8x8;
1051     }
1052     h->param.analyse.i_trellis = x264_clip3( h->param.analyse.i_trellis, 0, 2 );
1053     h->param.rc.i_aq_mode = x264_clip3( h->param.rc.i_aq_mode, 0, 2 );
1054     h->param.rc.f_aq_strength = x264_clip3f( h->param.rc.f_aq_strength, 0, 3 );
1055     if( h->param.rc.f_aq_strength == 0 )
1056         h->param.rc.i_aq_mode = 0;
1057
1058     if( h->param.i_log_level < X264_LOG_INFO )
1059     {
1060         h->param.analyse.b_psnr = 0;
1061         h->param.analyse.b_ssim = 0;
1062     }
1063     /* Warn users trying to measure PSNR/SSIM with psy opts on. */
1064     if( b_open && (h->param.analyse.b_psnr || h->param.analyse.b_ssim) )
1065     {
1066         char *s = NULL;
1067
1068         if( h->param.analyse.b_psy )
1069         {
1070             s = h->param.analyse.b_psnr ? "psnr" : "ssim";
1071             x264_log( h, X264_LOG_WARNING, "--%s used with psy on: results will be invalid!\n", s );
1072         }
1073         else if( !h->param.rc.i_aq_mode && h->param.analyse.b_ssim )
1074         {
1075             x264_log( h, X264_LOG_WARNING, "--ssim used with AQ off: results will be invalid!\n" );
1076             s = "ssim";
1077         }
1078         else if(  h->param.rc.i_aq_mode && h->param.analyse.b_psnr )
1079         {
1080             x264_log( h, X264_LOG_WARNING, "--psnr used with AQ on: results will be invalid!\n" );
1081             s = "psnr";
1082         }
1083         if( s )
1084             x264_log( h, X264_LOG_WARNING, "--tune %s should be used if attempting to benchmark %s!\n", s, s );
1085     }
1086
1087     if( !h->param.analyse.b_psy )
1088     {
1089         h->param.analyse.f_psy_rd = 0;
1090         h->param.analyse.f_psy_trellis = 0;
1091     }
1092     h->param.analyse.f_psy_rd = x264_clip3f( h->param.analyse.f_psy_rd, 0, 10 );
1093     h->param.analyse.f_psy_trellis = x264_clip3f( h->param.analyse.f_psy_trellis, 0, 10 );
1094     h->mb.i_psy_rd = h->param.analyse.i_subpel_refine >= 6 ? FIX8( h->param.analyse.f_psy_rd ) : 0;
1095     h->mb.i_psy_trellis = h->param.analyse.i_trellis ? FIX8( h->param.analyse.f_psy_trellis / 4 ) : 0;
1096     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -32, 32);
1097     /* In 4:4:4 mode, chroma gets twice as much resolution, so we can halve its quality. */
1098     if( b_open && i_csp >= X264_CSP_I444 && i_csp < X264_CSP_BGR && h->param.analyse.b_psy )
1099         h->param.analyse.i_chroma_qp_offset += 6;
1100     /* Psy RDO increases overall quantizers to improve the quality of luma--this indirectly hurts chroma quality */
1101     /* so we lower the chroma QP offset to compensate */
1102     if( b_open && h->mb.i_psy_rd && !h->param.i_avcintra_class )
1103         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_rd < 0.25 ? 1 : 2;
1104     /* Psy trellis has a similar effect. */
1105     if( b_open && h->mb.i_psy_trellis && !h->param.i_avcintra_class )
1106         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_trellis < 0.25 ? 1 : 2;
1107     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -12, 12);
1108     /* MB-tree requires AQ to be on, even if the strength is zero. */
1109     if( !h->param.rc.i_aq_mode && h->param.rc.b_mb_tree )
1110     {
1111         h->param.rc.i_aq_mode = 1;
1112         h->param.rc.f_aq_strength = 0;
1113     }
1114     h->param.analyse.i_noise_reduction = x264_clip3( h->param.analyse.i_noise_reduction, 0, 1<<16 );
1115     if( h->param.analyse.i_subpel_refine >= 10 && (h->param.analyse.i_trellis != 2 || !h->param.rc.i_aq_mode) )
1116         h->param.analyse.i_subpel_refine = 9;
1117
1118     {
1119         const x264_level_t *l = x264_levels;
1120         if( h->param.i_level_idc < 0 )
1121         {
1122             int maxrate_bak = h->param.rc.i_vbv_max_bitrate;
1123             if( h->param.rc.i_rc_method == X264_RC_ABR && h->param.rc.i_vbv_buffer_size <= 0 )
1124                 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate * 2;
1125             x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
1126             do h->param.i_level_idc = l->level_idc;
1127                 while( l[1].level_idc && x264_validate_levels( h, 0 ) && l++ );
1128             h->param.rc.i_vbv_max_bitrate = maxrate_bak;
1129         }
1130         else
1131         {
1132             while( l->level_idc && l->level_idc != h->param.i_level_idc )
1133                 l++;
1134             if( l->level_idc == 0 )
1135             {
1136                 x264_log( h, X264_LOG_ERROR, "invalid level_idc: %d\n", h->param.i_level_idc );
1137                 return -1;
1138             }
1139         }
1140         if( h->param.analyse.i_mv_range <= 0 )
1141             h->param.analyse.i_mv_range = l->mv_range >> PARAM_INTERLACED;
1142         else
1143             h->param.analyse.i_mv_range = x264_clip3(h->param.analyse.i_mv_range, 32, 512 >> PARAM_INTERLACED);
1144     }
1145
1146     h->param.analyse.i_weighted_pred = x264_clip3( h->param.analyse.i_weighted_pred, X264_WEIGHTP_NONE, X264_WEIGHTP_SMART );
1147
1148     if( h->param.i_lookahead_threads == X264_THREADS_AUTO )
1149     {
1150         if( h->param.b_sliced_threads )
1151             h->param.i_lookahead_threads = h->param.i_threads;
1152         else
1153         {
1154             /* If we're using much slower lookahead settings than encoding settings, it helps a lot to use
1155              * more lookahead threads.  This typically happens in the first pass of a two-pass encode, so
1156              * try to guess at this sort of case.
1157              *
1158              * Tuned by a little bit of real encoding with the various presets. */
1159             int badapt = h->param.i_bframe_adaptive == X264_B_ADAPT_TRELLIS;
1160             int subme = X264_MIN( h->param.analyse.i_subpel_refine / 3, 3 ) + (h->param.analyse.i_subpel_refine > 1);
1161             int bframes = X264_MIN( (h->param.i_bframe - 1) / 3, 3 );
1162
1163             /* [b-adapt 0/1 vs 2][quantized subme][quantized bframes] */
1164             static const uint8_t lookahead_thread_div[2][5][4] =
1165             {{{6,6,6,6}, {3,3,3,3}, {4,4,4,4}, {6,6,6,6}, {12,12,12,12}},
1166              {{3,2,1,1}, {2,1,1,1}, {4,3,2,1}, {6,4,3,2}, {12, 9, 6, 4}}};
1167
1168             h->param.i_lookahead_threads = h->param.i_threads / lookahead_thread_div[badapt][subme][bframes];
1169             /* Since too many lookahead threads significantly degrades lookahead accuracy, limit auto
1170              * lookahead threads to about 8 macroblock rows high each at worst.  This number is chosen
1171              * pretty much arbitrarily. */
1172             h->param.i_lookahead_threads = X264_MIN( h->param.i_lookahead_threads, h->param.i_height / 128 );
1173         }
1174     }
1175     h->param.i_lookahead_threads = x264_clip3( h->param.i_lookahead_threads, 1, X264_MIN( max_sliced_threads, X264_LOOKAHEAD_THREAD_MAX ) );
1176
1177     if( PARAM_INTERLACED )
1178     {
1179         if( h->param.analyse.i_me_method >= X264_ME_ESA )
1180         {
1181             x264_log( h, X264_LOG_WARNING, "interlace + me=esa is not implemented\n" );
1182             h->param.analyse.i_me_method = X264_ME_UMH;
1183         }
1184         if( h->param.analyse.i_weighted_pred > 0 )
1185         {
1186             x264_log( h, X264_LOG_WARNING, "interlace + weightp is not implemented\n" );
1187             h->param.analyse.i_weighted_pred = X264_WEIGHTP_NONE;
1188         }
1189     }
1190
1191     if( !h->param.analyse.i_weighted_pred && h->param.rc.b_mb_tree && h->param.analyse.b_psy )
1192         h->param.analyse.i_weighted_pred = X264_WEIGHTP_FAKE;
1193
1194     if( h->i_thread_frames > 1 )
1195     {
1196         int r = h->param.analyse.i_mv_range_thread;
1197         int r2;
1198         if( r <= 0 )
1199         {
1200             // half of the available space is reserved and divided evenly among the threads,
1201             // the rest is allocated to whichever thread is far enough ahead to use it.
1202             // reserving more space increases quality for some videos, but costs more time
1203             // in thread synchronization.
1204             int max_range = (h->param.i_height + X264_THREAD_HEIGHT) / h->i_thread_frames - X264_THREAD_HEIGHT;
1205             r = max_range / 2;
1206         }
1207         r = X264_MAX( r, h->param.analyse.i_me_range );
1208         r = X264_MIN( r, h->param.analyse.i_mv_range );
1209         // round up to use the whole mb row
1210         r2 = (r & ~15) + ((-X264_THREAD_HEIGHT) & 15);
1211         if( r2 < r )
1212             r2 += 16;
1213         x264_log( h, X264_LOG_DEBUG, "using mv_range_thread = %d\n", r2 );
1214         h->param.analyse.i_mv_range_thread = r2;
1215     }
1216
1217     if( h->param.rc.f_rate_tolerance < 0 )
1218         h->param.rc.f_rate_tolerance = 0;
1219     if( h->param.rc.f_qblur < 0 )
1220         h->param.rc.f_qblur = 0;
1221     if( h->param.rc.f_complexity_blur < 0 )
1222         h->param.rc.f_complexity_blur = 0;
1223
1224     h->param.i_sps_id &= 31;
1225
1226     h->param.i_nal_hrd = x264_clip3( h->param.i_nal_hrd, X264_NAL_HRD_NONE, X264_NAL_HRD_CBR );
1227
1228     if( h->param.i_nal_hrd && !h->param.rc.i_vbv_buffer_size )
1229     {
1230         x264_log( h, X264_LOG_WARNING, "NAL HRD parameters require VBV parameters\n" );
1231         h->param.i_nal_hrd = X264_NAL_HRD_NONE;
1232     }
1233
1234     if( h->param.i_nal_hrd == X264_NAL_HRD_CBR &&
1235        (h->param.rc.i_bitrate != h->param.rc.i_vbv_max_bitrate || !h->param.rc.i_vbv_max_bitrate) )
1236     {
1237         x264_log( h, X264_LOG_WARNING, "CBR HRD requires constant bitrate\n" );
1238         h->param.i_nal_hrd = X264_NAL_HRD_VBR;
1239     }
1240
1241     if( h->param.i_nal_hrd == X264_NAL_HRD_CBR )
1242         h->param.rc.b_filler = 1;
1243
1244     /* ensure the booleans are 0 or 1 so they can be used in math */
1245 #define BOOLIFY(x) h->param.x = !!h->param.x
1246     BOOLIFY( b_cabac );
1247     BOOLIFY( b_constrained_intra );
1248     BOOLIFY( b_deblocking_filter );
1249     BOOLIFY( b_deterministic );
1250     BOOLIFY( b_sliced_threads );
1251     BOOLIFY( b_interlaced );
1252     BOOLIFY( b_intra_refresh );
1253     BOOLIFY( b_aud );
1254     BOOLIFY( b_repeat_headers );
1255     BOOLIFY( b_annexb );
1256     BOOLIFY( b_vfr_input );
1257     BOOLIFY( b_pulldown );
1258     BOOLIFY( b_tff );
1259     BOOLIFY( b_pic_struct );
1260     BOOLIFY( b_fake_interlaced );
1261     BOOLIFY( b_open_gop );
1262     BOOLIFY( b_bluray_compat );
1263     BOOLIFY( b_stitchable );
1264     BOOLIFY( b_full_recon );
1265     BOOLIFY( b_opencl );
1266     BOOLIFY( analyse.b_transform_8x8 );
1267     BOOLIFY( analyse.b_weighted_bipred );
1268     BOOLIFY( analyse.b_chroma_me );
1269     BOOLIFY( analyse.b_mixed_references );
1270     BOOLIFY( analyse.b_fast_pskip );
1271     BOOLIFY( analyse.b_dct_decimate );
1272     BOOLIFY( analyse.b_psy );
1273     BOOLIFY( analyse.b_psnr );
1274     BOOLIFY( analyse.b_ssim );
1275     BOOLIFY( rc.b_stat_write );
1276     BOOLIFY( rc.b_stat_read );
1277     BOOLIFY( rc.b_mb_tree );
1278     BOOLIFY( rc.b_filler );
1279 #undef BOOLIFY
1280
1281     return 0;
1282 }
1283
1284 static void mbcmp_init( x264_t *h )
1285 {
1286     int satd = !h->mb.b_lossless && h->param.analyse.i_subpel_refine > 1;
1287     memcpy( h->pixf.mbcmp, satd ? h->pixf.satd : h->pixf.sad_aligned, sizeof(h->pixf.mbcmp) );
1288     memcpy( h->pixf.mbcmp_unaligned, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.mbcmp_unaligned) );
1289     h->pixf.intra_mbcmp_x3_16x16 = satd ? h->pixf.intra_satd_x3_16x16 : h->pixf.intra_sad_x3_16x16;
1290     h->pixf.intra_mbcmp_x3_8x16c = satd ? h->pixf.intra_satd_x3_8x16c : h->pixf.intra_sad_x3_8x16c;
1291     h->pixf.intra_mbcmp_x3_8x8c  = satd ? h->pixf.intra_satd_x3_8x8c  : h->pixf.intra_sad_x3_8x8c;
1292     h->pixf.intra_mbcmp_x3_8x8 = satd ? h->pixf.intra_sa8d_x3_8x8 : h->pixf.intra_sad_x3_8x8;
1293     h->pixf.intra_mbcmp_x3_4x4 = satd ? h->pixf.intra_satd_x3_4x4 : h->pixf.intra_sad_x3_4x4;
1294     h->pixf.intra_mbcmp_x9_4x4 = h->param.b_cpu_independent || h->mb.b_lossless ? NULL
1295                                : satd ? h->pixf.intra_satd_x9_4x4 : h->pixf.intra_sad_x9_4x4;
1296     h->pixf.intra_mbcmp_x9_8x8 = h->param.b_cpu_independent || h->mb.b_lossless ? NULL
1297                                : satd ? h->pixf.intra_sa8d_x9_8x8 : h->pixf.intra_sad_x9_8x8;
1298     satd &= h->param.analyse.i_me_method == X264_ME_TESA;
1299     memcpy( h->pixf.fpelcmp, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.fpelcmp) );
1300     memcpy( h->pixf.fpelcmp_x3, satd ? h->pixf.satd_x3 : h->pixf.sad_x3, sizeof(h->pixf.fpelcmp_x3) );
1301     memcpy( h->pixf.fpelcmp_x4, satd ? h->pixf.satd_x4 : h->pixf.sad_x4, sizeof(h->pixf.fpelcmp_x4) );
1302 }
1303
1304 static void chroma_dsp_init( x264_t *h )
1305 {
1306     memcpy( h->luma2chroma_pixel, x264_luma2chroma_pixel[CHROMA_FORMAT], sizeof(h->luma2chroma_pixel) );
1307
1308     switch( CHROMA_FORMAT )
1309     {
1310         case CHROMA_420:
1311             memcpy( h->predict_chroma, h->predict_8x8c, sizeof(h->predict_chroma) );
1312             h->mc.prefetch_fenc = h->mc.prefetch_fenc_420;
1313             h->loopf.deblock_chroma[0] = h->loopf.deblock_h_chroma_420;
1314             h->loopf.deblock_chroma_intra[0] = h->loopf.deblock_h_chroma_420_intra;
1315             h->loopf.deblock_chroma_mbaff = h->loopf.deblock_chroma_420_mbaff;
1316             h->loopf.deblock_chroma_intra_mbaff = h->loopf.deblock_chroma_420_intra_mbaff;
1317             h->pixf.intra_mbcmp_x3_chroma = h->pixf.intra_mbcmp_x3_8x8c;
1318             h->quantf.coeff_last[DCT_CHROMA_DC] = h->quantf.coeff_last4;
1319             h->quantf.coeff_level_run[DCT_CHROMA_DC] = h->quantf.coeff_level_run4;
1320             break;
1321         case CHROMA_422:
1322             memcpy( h->predict_chroma, h->predict_8x16c, sizeof(h->predict_chroma) );
1323             h->mc.prefetch_fenc = h->mc.prefetch_fenc_422;
1324             h->loopf.deblock_chroma[0] = h->loopf.deblock_h_chroma_422;
1325             h->loopf.deblock_chroma_intra[0] = h->loopf.deblock_h_chroma_422_intra;
1326             h->loopf.deblock_chroma_mbaff = h->loopf.deblock_chroma_422_mbaff;
1327             h->loopf.deblock_chroma_intra_mbaff = h->loopf.deblock_chroma_422_intra_mbaff;
1328             h->pixf.intra_mbcmp_x3_chroma = h->pixf.intra_mbcmp_x3_8x16c;
1329             h->quantf.coeff_last[DCT_CHROMA_DC] = h->quantf.coeff_last8;
1330             h->quantf.coeff_level_run[DCT_CHROMA_DC] = h->quantf.coeff_level_run8;
1331             break;
1332         case CHROMA_444:
1333             h->mc.prefetch_fenc = h->mc.prefetch_fenc_422; /* FIXME: doesn't cover V plane */
1334             h->loopf.deblock_chroma_mbaff = h->loopf.deblock_luma_mbaff;
1335             h->loopf.deblock_chroma_intra_mbaff = h->loopf.deblock_luma_intra_mbaff;
1336             break;
1337     }
1338 }
1339
1340 static void x264_set_aspect_ratio( x264_t *h, x264_param_t *param, int initial )
1341 {
1342     /* VUI */
1343     if( param->vui.i_sar_width > 0 && param->vui.i_sar_height > 0 )
1344     {
1345         uint32_t i_w = param->vui.i_sar_width;
1346         uint32_t i_h = param->vui.i_sar_height;
1347         uint32_t old_w = h->param.vui.i_sar_width;
1348         uint32_t old_h = h->param.vui.i_sar_height;
1349
1350         x264_reduce_fraction( &i_w, &i_h );
1351
1352         while( i_w > 65535 || i_h > 65535 )
1353         {
1354             i_w /= 2;
1355             i_h /= 2;
1356         }
1357
1358         x264_reduce_fraction( &i_w, &i_h );
1359
1360         if( i_w != old_w || i_h != old_h || initial )
1361         {
1362             h->param.vui.i_sar_width = 0;
1363             h->param.vui.i_sar_height = 0;
1364             if( i_w == 0 || i_h == 0 )
1365                 x264_log( h, X264_LOG_WARNING, "cannot create valid sample aspect ratio\n" );
1366             else
1367             {
1368                 x264_log( h, initial?X264_LOG_INFO:X264_LOG_DEBUG, "using SAR=%d/%d\n", i_w, i_h );
1369                 h->param.vui.i_sar_width = i_w;
1370                 h->param.vui.i_sar_height = i_h;
1371             }
1372         }
1373     }
1374 }
1375
1376 /****************************************************************************
1377  * x264_encoder_open:
1378  ****************************************************************************/
1379 x264_t *x264_encoder_open( x264_param_t *param )
1380 {
1381     x264_t *h;
1382     char buf[1000], *p;
1383     int qp, i_slicetype_length;
1384
1385     CHECKED_MALLOCZERO( h, sizeof(x264_t) );
1386
1387     /* Create a copy of param */
1388     memcpy( &h->param, param, sizeof(x264_param_t) );
1389
1390     if( param->param_free )
1391         param->param_free( param );
1392
1393     if( x264_threading_init() )
1394     {
1395         x264_log( h, X264_LOG_ERROR, "unable to initialize threading\n" );
1396         goto fail;
1397     }
1398
1399     if( x264_validate_parameters( h, 1 ) < 0 )
1400         goto fail;
1401
1402     if( h->param.psz_cqm_file )
1403         if( x264_cqm_parse_file( h, h->param.psz_cqm_file ) < 0 )
1404             goto fail;
1405
1406     if( h->param.rc.psz_stat_out )
1407         h->param.rc.psz_stat_out = strdup( h->param.rc.psz_stat_out );
1408     if( h->param.rc.psz_stat_in )
1409         h->param.rc.psz_stat_in = strdup( h->param.rc.psz_stat_in );
1410
1411     x264_reduce_fraction( &h->param.i_fps_num, &h->param.i_fps_den );
1412     x264_reduce_fraction( &h->param.i_timebase_num, &h->param.i_timebase_den );
1413
1414     /* Init x264_t */
1415     h->i_frame = -1;
1416     h->i_frame_num = 0;
1417
1418     if( h->param.i_avcintra_class )
1419         h->i_idr_pic_id = 5;
1420     else
1421         h->i_idr_pic_id = 0;
1422
1423     if( (uint64_t)h->param.i_timebase_den * 2 > UINT32_MAX )
1424     {
1425         x264_log( h, X264_LOG_ERROR, "Effective timebase denominator %u exceeds H.264 maximum\n", h->param.i_timebase_den );
1426         goto fail;
1427     }
1428
1429     x264_set_aspect_ratio( h, &h->param, 1 );
1430
1431     x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
1432     x264_pps_init( h->pps, h->param.i_sps_id, &h->param, h->sps );
1433
1434     x264_validate_levels( h, 1 );
1435
1436     h->chroma_qp_table = i_chroma_qp_table + 12 + h->pps->i_chroma_qp_index_offset;
1437
1438     if( x264_cqm_init( h ) < 0 )
1439         goto fail;
1440
1441     h->mb.i_mb_width = h->sps->i_mb_width;
1442     h->mb.i_mb_height = h->sps->i_mb_height;
1443     h->mb.i_mb_count = h->mb.i_mb_width * h->mb.i_mb_height;
1444
1445     h->mb.chroma_h_shift = CHROMA_FORMAT == CHROMA_420 || CHROMA_FORMAT == CHROMA_422;
1446     h->mb.chroma_v_shift = CHROMA_FORMAT == CHROMA_420;
1447
1448     /* Adaptive MBAFF and subme 0 are not supported as we require halving motion
1449      * vectors during prediction, resulting in hpel mvs.
1450      * The chosen solution is to make MBAFF non-adaptive in this case. */
1451     h->mb.b_adaptive_mbaff = PARAM_INTERLACED && h->param.analyse.i_subpel_refine;
1452
1453     /* Init frames. */
1454     if( h->param.i_bframe_adaptive == X264_B_ADAPT_TRELLIS && !h->param.rc.b_stat_read )
1455         h->frames.i_delay = X264_MAX(h->param.i_bframe,3)*4;
1456     else
1457         h->frames.i_delay = h->param.i_bframe;
1458     if( h->param.rc.b_mb_tree || h->param.rc.i_vbv_buffer_size )
1459         h->frames.i_delay = X264_MAX( h->frames.i_delay, h->param.rc.i_lookahead );
1460     i_slicetype_length = h->frames.i_delay;
1461     h->frames.i_delay += h->i_thread_frames - 1;
1462     h->frames.i_delay += h->param.i_sync_lookahead;
1463     h->frames.i_delay += h->param.b_vfr_input;
1464     h->frames.i_bframe_delay = h->param.i_bframe ? (h->param.i_bframe_pyramid ? 2 : 1) : 0;
1465
1466     h->frames.i_max_ref0 = h->param.i_frame_reference;
1467     h->frames.i_max_ref1 = X264_MIN( h->sps->vui.i_num_reorder_frames, h->param.i_frame_reference );
1468     h->frames.i_max_dpb  = h->sps->vui.i_max_dec_frame_buffering;
1469     h->frames.b_have_lowres = !h->param.rc.b_stat_read
1470         && ( h->param.rc.i_rc_method == X264_RC_ABR
1471           || h->param.rc.i_rc_method == X264_RC_CRF
1472           || h->param.i_bframe_adaptive
1473           || h->param.i_scenecut_threshold
1474           || h->param.rc.b_mb_tree
1475           || h->param.analyse.i_weighted_pred );
1476     h->frames.b_have_lowres |= h->param.rc.b_stat_read && h->param.rc.i_vbv_buffer_size > 0;
1477     h->frames.b_have_sub8x8_esa = !!(h->param.analyse.inter & X264_ANALYSE_PSUB8x8);
1478
1479     h->frames.i_last_idr =
1480     h->frames.i_last_keyframe = - h->param.i_keyint_max;
1481     h->frames.i_input    = 0;
1482     h->frames.i_largest_pts = h->frames.i_second_largest_pts = -1;
1483     h->frames.i_poc_last_open_gop = -1;
1484
1485     CHECKED_MALLOCZERO( h->frames.unused[0], (h->frames.i_delay + 3) * sizeof(x264_frame_t *) );
1486     /* Allocate room for max refs plus a few extra just in case. */
1487     CHECKED_MALLOCZERO( h->frames.unused[1], (h->i_thread_frames + X264_REF_MAX + 4) * sizeof(x264_frame_t *) );
1488     CHECKED_MALLOCZERO( h->frames.current, (h->param.i_sync_lookahead + h->param.i_bframe
1489                         + h->i_thread_frames + 3) * sizeof(x264_frame_t *) );
1490     if( h->param.analyse.i_weighted_pred > 0 )
1491         CHECKED_MALLOCZERO( h->frames.blank_unused, h->i_thread_frames * 4 * sizeof(x264_frame_t *) );
1492     h->i_ref[0] = h->i_ref[1] = 0;
1493     h->i_cpb_delay = h->i_coded_fields = h->i_disp_fields = 0;
1494     h->i_prev_duration = ((uint64_t)h->param.i_fps_den * h->sps->vui.i_time_scale) / ((uint64_t)h->param.i_fps_num * h->sps->vui.i_num_units_in_tick);
1495     h->i_disp_fields_last_frame = -1;
1496     x264_rdo_init();
1497
1498     /* init CPU functions */
1499     x264_predict_16x16_init( h->param.cpu, h->predict_16x16 );
1500     x264_predict_8x8c_init( h->param.cpu, h->predict_8x8c );
1501     x264_predict_8x16c_init( h->param.cpu, h->predict_8x16c );
1502     x264_predict_8x8_init( h->param.cpu, h->predict_8x8, &h->predict_8x8_filter );
1503     x264_predict_4x4_init( h->param.cpu, h->predict_4x4 );
1504     x264_pixel_init( h->param.cpu, &h->pixf );
1505     x264_dct_init( h->param.cpu, &h->dctf );
1506     x264_zigzag_init( h->param.cpu, &h->zigzagf_progressive, &h->zigzagf_interlaced );
1507     memcpy( &h->zigzagf, PARAM_INTERLACED ? &h->zigzagf_interlaced : &h->zigzagf_progressive, sizeof(h->zigzagf) );
1508     x264_mc_init( h->param.cpu, &h->mc, h->param.b_cpu_independent );
1509     x264_quant_init( h, h->param.cpu, &h->quantf );
1510     x264_deblock_init( h->param.cpu, &h->loopf, PARAM_INTERLACED );
1511     x264_bitstream_init( h->param.cpu, &h->bsf );
1512     if( h->param.b_cabac )
1513         x264_cabac_init( h );
1514     else
1515         x264_stack_align( x264_cavlc_init, h );
1516
1517     mbcmp_init( h );
1518     chroma_dsp_init( h );
1519
1520     p = buf + sprintf( buf, "using cpu capabilities:" );
1521     for( int i = 0; x264_cpu_names[i].flags; i++ )
1522     {
1523         if( !strcmp(x264_cpu_names[i].name, "SSE")
1524             && h->param.cpu & (X264_CPU_SSE2) )
1525             continue;
1526         if( !strcmp(x264_cpu_names[i].name, "SSE2")
1527             && h->param.cpu & (X264_CPU_SSE2_IS_FAST|X264_CPU_SSE2_IS_SLOW) )
1528             continue;
1529         if( !strcmp(x264_cpu_names[i].name, "SSE3")
1530             && (h->param.cpu & X264_CPU_SSSE3 || !(h->param.cpu & X264_CPU_CACHELINE_64)) )
1531             continue;
1532         if( !strcmp(x264_cpu_names[i].name, "SSE4.1")
1533             && (h->param.cpu & X264_CPU_SSE42) )
1534             continue;
1535         if( !strcmp(x264_cpu_names[i].name, "BMI1")
1536             && (h->param.cpu & X264_CPU_BMI2) )
1537             continue;
1538         if( (h->param.cpu & x264_cpu_names[i].flags) == x264_cpu_names[i].flags
1539             && (!i || x264_cpu_names[i].flags != x264_cpu_names[i-1].flags) )
1540             p += sprintf( p, " %s", x264_cpu_names[i].name );
1541     }
1542     if( !h->param.cpu )
1543         p += sprintf( p, " none!" );
1544     x264_log( h, X264_LOG_INFO, "%s\n", buf );
1545
1546     float *logs = x264_analyse_prepare_costs( h );
1547     if( !logs )
1548         goto fail;
1549     for( qp = X264_MIN( h->param.rc.i_qp_min, QP_MAX_SPEC ); qp <= h->param.rc.i_qp_max; qp++ )
1550         if( x264_analyse_init_costs( h, logs, qp ) )
1551             goto fail;
1552     if( x264_analyse_init_costs( h, logs, X264_LOOKAHEAD_QP ) )
1553         goto fail;
1554     x264_free( logs );
1555
1556     static const uint16_t cost_mv_correct[7] = { 24, 47, 95, 189, 379, 757, 1515 };
1557     /* Checks for known miscompilation issues. */
1558     if( h->cost_mv[X264_LOOKAHEAD_QP][2013] != cost_mv_correct[BIT_DEPTH-8] )
1559     {
1560         x264_log( h, X264_LOG_ERROR, "MV cost test failed: x264 has been miscompiled!\n" );
1561         goto fail;
1562     }
1563
1564     /* Must be volatile or else GCC will optimize it out. */
1565     volatile int temp = 392;
1566     if( x264_clz( temp ) != 23 )
1567     {
1568         x264_log( h, X264_LOG_ERROR, "CLZ test failed: x264 has been miscompiled!\n" );
1569 #if ARCH_X86 || ARCH_X86_64
1570         x264_log( h, X264_LOG_ERROR, "Are you attempting to run an SSE4a/LZCNT-targeted build on a CPU that\n" );
1571         x264_log( h, X264_LOG_ERROR, "doesn't support it?\n" );
1572 #endif
1573         goto fail;
1574     }
1575
1576     h->out.i_nal = 0;
1577     h->out.i_bitstream = X264_MAX( 1000000, h->param.i_width * h->param.i_height * 4
1578         * ( h->param.rc.i_rc_method == X264_RC_ABR ? pow( 0.95, h->param.rc.i_qp_min )
1579           : pow( 0.95, h->param.rc.i_qp_constant ) * X264_MAX( 1, h->param.rc.f_ip_factor )));
1580
1581     h->nal_buffer_size = h->out.i_bitstream * 3/2 + 4 + 64; /* +4 for startcode, +64 for nal_escape assembly padding */
1582     CHECKED_MALLOC( h->nal_buffer, h->nal_buffer_size );
1583
1584     CHECKED_MALLOC( h->reconfig_h, sizeof(x264_t) );
1585
1586     if( h->param.i_threads > 1 &&
1587         x264_threadpool_init( &h->threadpool, h->param.i_threads, (void*)x264_encoder_thread_init, h ) )
1588         goto fail;
1589     if( h->param.i_lookahead_threads > 1 &&
1590         x264_threadpool_init( &h->lookaheadpool, h->param.i_lookahead_threads, NULL, NULL ) )
1591         goto fail;
1592
1593 #if HAVE_OPENCL
1594     if( h->param.b_opencl )
1595     {
1596         h->opencl.ocl = x264_opencl_load_library();
1597         if( !h->opencl.ocl )
1598         {
1599             x264_log( h, X264_LOG_WARNING, "failed to load OpenCL\n" );
1600             h->param.b_opencl = 0;
1601         }
1602     }
1603 #endif
1604
1605     h->thread[0] = h;
1606     for( int i = 1; i < h->param.i_threads + !!h->param.i_sync_lookahead; i++ )
1607         CHECKED_MALLOC( h->thread[i], sizeof(x264_t) );
1608     if( h->param.i_lookahead_threads > 1 )
1609         for( int i = 0; i < h->param.i_lookahead_threads; i++ )
1610         {
1611             CHECKED_MALLOC( h->lookahead_thread[i], sizeof(x264_t) );
1612             *h->lookahead_thread[i] = *h;
1613         }
1614     *h->reconfig_h = *h;
1615
1616     for( int i = 0; i < h->param.i_threads; i++ )
1617     {
1618         int init_nal_count = h->param.i_slice_count + 3;
1619         int allocate_threadlocal_data = !h->param.b_sliced_threads || !i;
1620         if( i > 0 )
1621             *h->thread[i] = *h;
1622
1623         if( x264_pthread_mutex_init( &h->thread[i]->mutex, NULL ) )
1624             goto fail;
1625         if( x264_pthread_cond_init( &h->thread[i]->cv, NULL ) )
1626             goto fail;
1627
1628         if( allocate_threadlocal_data )
1629         {
1630             h->thread[i]->fdec = x264_frame_pop_unused( h, 1 );
1631             if( !h->thread[i]->fdec )
1632                 goto fail;
1633         }
1634         else
1635             h->thread[i]->fdec = h->thread[0]->fdec;
1636
1637         CHECKED_MALLOC( h->thread[i]->out.p_bitstream, h->out.i_bitstream );
1638         /* Start each thread with room for init_nal_count NAL units; it'll realloc later if needed. */
1639         CHECKED_MALLOC( h->thread[i]->out.nal, init_nal_count*sizeof(x264_nal_t) );
1640         h->thread[i]->out.i_nals_allocated = init_nal_count;
1641
1642         if( allocate_threadlocal_data && x264_macroblock_cache_allocate( h->thread[i] ) < 0 )
1643             goto fail;
1644     }
1645
1646 #if HAVE_OPENCL
1647     if( h->param.b_opencl && x264_opencl_lookahead_init( h ) < 0 )
1648         h->param.b_opencl = 0;
1649 #endif
1650
1651     if( x264_lookahead_init( h, i_slicetype_length ) )
1652         goto fail;
1653
1654     for( int i = 0; i < h->param.i_threads; i++ )
1655         if( x264_macroblock_thread_allocate( h->thread[i], 0 ) < 0 )
1656             goto fail;
1657
1658     if( x264_ratecontrol_new( h ) < 0 )
1659         goto fail;
1660
1661     if( h->param.i_nal_hrd )
1662     {
1663         x264_log( h, X264_LOG_DEBUG, "HRD bitrate: %i bits/sec\n", h->sps->vui.hrd.i_bit_rate_unscaled );
1664         x264_log( h, X264_LOG_DEBUG, "CPB size: %i bits\n", h->sps->vui.hrd.i_cpb_size_unscaled );
1665     }
1666
1667     if( h->param.psz_dump_yuv )
1668     {
1669         /* create or truncate the reconstructed video file */
1670         FILE *f = x264_fopen( h->param.psz_dump_yuv, "w" );
1671         if( !f )
1672         {
1673             x264_log( h, X264_LOG_ERROR, "dump_yuv: can't write to %s\n", h->param.psz_dump_yuv );
1674             goto fail;
1675         }
1676         else if( !x264_is_regular_file( f ) )
1677         {
1678             x264_log( h, X264_LOG_ERROR, "dump_yuv: incompatible with non-regular file %s\n", h->param.psz_dump_yuv );
1679             goto fail;
1680         }
1681         fclose( f );
1682     }
1683
1684     const char *profile = h->sps->i_profile_idc == PROFILE_BASELINE ? "Constrained Baseline" :
1685                           h->sps->i_profile_idc == PROFILE_MAIN ? "Main" :
1686                           h->sps->i_profile_idc == PROFILE_HIGH ? "High" :
1687                           h->sps->i_profile_idc == PROFILE_HIGH10 ? (h->sps->b_constraint_set3 == 1 ? "High 10 Intra" : "High 10") :
1688                           h->sps->i_profile_idc == PROFILE_HIGH422 ? (h->sps->b_constraint_set3 == 1 ? "High 4:2:2 Intra" : "High 4:2:2") :
1689                           h->sps->b_constraint_set3 == 1 ? "High 4:4:4 Intra" : "High 4:4:4 Predictive";
1690     char level[4];
1691     snprintf( level, sizeof(level), "%d.%d", h->sps->i_level_idc/10, h->sps->i_level_idc%10 );
1692     if( h->sps->i_level_idc == 9 || ( h->sps->i_level_idc == 11 && h->sps->b_constraint_set3 &&
1693         (h->sps->i_profile_idc == PROFILE_BASELINE || h->sps->i_profile_idc == PROFILE_MAIN) ) )
1694         strcpy( level, "1b" );
1695
1696     if( h->sps->i_profile_idc < PROFILE_HIGH10 )
1697     {
1698         x264_log( h, X264_LOG_INFO, "profile %s, level %s\n",
1699             profile, level );
1700     }
1701     else
1702     {
1703         static const char * const subsampling[4] = { "4:0:0", "4:2:0", "4:2:2", "4:4:4" };
1704         x264_log( h, X264_LOG_INFO, "profile %s, level %s, %s %d-bit\n",
1705             profile, level, subsampling[CHROMA_FORMAT], BIT_DEPTH );
1706     }
1707
1708     return h;
1709 fail:
1710     x264_free( h );
1711     return NULL;
1712 }
1713
1714 /****************************************************************************/
1715 static int x264_encoder_try_reconfig( x264_t *h, x264_param_t *param, int *rc_reconfig )
1716 {
1717     *rc_reconfig = 0;
1718     x264_set_aspect_ratio( h, param, 0 );
1719 #define COPY(var) h->param.var = param->var
1720     COPY( i_frame_reference ); // but never uses more refs than initially specified
1721     COPY( i_bframe_bias );
1722     if( h->param.i_scenecut_threshold )
1723         COPY( i_scenecut_threshold ); // can't turn it on or off, only vary the threshold
1724     COPY( b_deblocking_filter );
1725     COPY( i_deblocking_filter_alphac0 );
1726     COPY( i_deblocking_filter_beta );
1727     COPY( i_frame_packing );
1728     COPY( analyse.inter );
1729     COPY( analyse.intra );
1730     COPY( analyse.i_direct_mv_pred );
1731     /* Scratch buffer prevents me_range from being increased for esa/tesa */
1732     if( h->param.analyse.i_me_method < X264_ME_ESA || param->analyse.i_me_range < h->param.analyse.i_me_range )
1733         COPY( analyse.i_me_range );
1734     COPY( analyse.i_noise_reduction );
1735     /* We can't switch out of subme=0 during encoding. */
1736     if( h->param.analyse.i_subpel_refine )
1737         COPY( analyse.i_subpel_refine );
1738     COPY( analyse.i_trellis );
1739     COPY( analyse.b_chroma_me );
1740     COPY( analyse.b_dct_decimate );
1741     COPY( analyse.b_fast_pskip );
1742     COPY( analyse.b_mixed_references );
1743     COPY( analyse.f_psy_rd );
1744     COPY( analyse.f_psy_trellis );
1745     COPY( crop_rect );
1746     // can only twiddle these if they were enabled to begin with:
1747     if( h->param.analyse.i_me_method >= X264_ME_ESA || param->analyse.i_me_method < X264_ME_ESA )
1748         COPY( analyse.i_me_method );
1749     if( h->param.analyse.i_me_method >= X264_ME_ESA && !h->frames.b_have_sub8x8_esa )
1750         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
1751     if( h->pps->b_transform_8x8_mode )
1752         COPY( analyse.b_transform_8x8 );
1753     if( h->frames.i_max_ref1 > 1 )
1754         COPY( i_bframe_pyramid );
1755     COPY( i_slice_max_size );
1756     COPY( i_slice_max_mbs );
1757     COPY( i_slice_min_mbs );
1758     COPY( i_slice_count );
1759     COPY( i_slice_count_max );
1760     COPY( b_tff );
1761
1762     /* VBV can't be turned on if it wasn't on to begin with */
1763     if( h->param.rc.i_vbv_max_bitrate > 0 && h->param.rc.i_vbv_buffer_size > 0 &&
1764           param->rc.i_vbv_max_bitrate > 0 &&   param->rc.i_vbv_buffer_size > 0 )
1765     {
1766         *rc_reconfig |= h->param.rc.i_vbv_max_bitrate != param->rc.i_vbv_max_bitrate;
1767         *rc_reconfig |= h->param.rc.i_vbv_buffer_size != param->rc.i_vbv_buffer_size;
1768         *rc_reconfig |= h->param.rc.i_bitrate != param->rc.i_bitrate;
1769         COPY( rc.i_vbv_max_bitrate );
1770         COPY( rc.i_vbv_buffer_size );
1771         COPY( rc.i_bitrate );
1772     }
1773     *rc_reconfig |= h->param.rc.f_rf_constant != param->rc.f_rf_constant;
1774     *rc_reconfig |= h->param.rc.f_rf_constant_max != param->rc.f_rf_constant_max;
1775     COPY( rc.f_rf_constant );
1776     COPY( rc.f_rf_constant_max );
1777 #undef COPY
1778
1779     return x264_validate_parameters( h, 0 );
1780 }
1781
1782 int x264_encoder_reconfig_apply( x264_t *h, x264_param_t *param )
1783 {
1784     int rc_reconfig;
1785     int ret = x264_encoder_try_reconfig( h, param, &rc_reconfig );
1786
1787     mbcmp_init( h );
1788     if( !ret )
1789         x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
1790
1791     /* Supported reconfiguration options (1-pass only):
1792      * vbv-maxrate
1793      * vbv-bufsize
1794      * crf
1795      * bitrate (CBR only) */
1796     if( !ret && rc_reconfig )
1797         x264_ratecontrol_init_reconfigurable( h, 0 );
1798
1799     return ret;
1800 }
1801
1802 /****************************************************************************
1803  * x264_encoder_reconfig:
1804  ****************************************************************************/
1805 int x264_encoder_reconfig( x264_t *h, x264_param_t *param )
1806 {
1807     h = h->thread[h->thread[0]->i_thread_phase];
1808     x264_param_t param_save = h->reconfig_h->param;
1809     h->reconfig_h->param = h->param;
1810
1811     int rc_reconfig;
1812     int ret = x264_encoder_try_reconfig( h->reconfig_h, param, &rc_reconfig );
1813     if( !ret )
1814         h->reconfig = 1;
1815     else
1816         h->reconfig_h->param = param_save;
1817
1818     return ret;
1819 }
1820
1821 /****************************************************************************
1822  * x264_encoder_parameters:
1823  ****************************************************************************/
1824 void x264_encoder_parameters( x264_t *h, x264_param_t *param )
1825 {
1826     memcpy( param, &h->thread[h->i_thread_phase]->param, sizeof(x264_param_t) );
1827 }
1828
1829 /* internal usage */
1830 static void x264_nal_start( x264_t *h, int i_type, int i_ref_idc )
1831 {
1832     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1833
1834     nal->i_ref_idc        = i_ref_idc;
1835     nal->i_type           = i_type;
1836     nal->b_long_startcode = 1;
1837
1838     nal->i_payload= 0;
1839     nal->p_payload= &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8];
1840     nal->i_padding= 0;
1841 }
1842
1843 /* if number of allocated nals is not enough, re-allocate a larger one. */
1844 static int x264_nal_check_buffer( x264_t *h )
1845 {
1846     if( h->out.i_nal >= h->out.i_nals_allocated )
1847     {
1848         x264_nal_t *new_out = x264_malloc( sizeof(x264_nal_t) * (h->out.i_nals_allocated*2) );
1849         if( !new_out )
1850             return -1;
1851         memcpy( new_out, h->out.nal, sizeof(x264_nal_t) * (h->out.i_nals_allocated) );
1852         x264_free( h->out.nal );
1853         h->out.nal = new_out;
1854         h->out.i_nals_allocated *= 2;
1855     }
1856     return 0;
1857 }
1858
1859 static int x264_nal_end( x264_t *h )
1860 {
1861     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1862     uint8_t *end = &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8];
1863     nal->i_payload = end - nal->p_payload;
1864     /* Assembly implementation of nal_escape reads past the end of the input.
1865      * While undefined padding wouldn't actually affect the output, it makes valgrind unhappy. */
1866     memset( end, 0xff, 64 );
1867     if( h->param.nalu_process )
1868         h->param.nalu_process( h, nal, h->fenc->opaque );
1869     h->out.i_nal++;
1870
1871     return x264_nal_check_buffer( h );
1872 }
1873
1874 static int x264_check_encapsulated_buffer( x264_t *h, x264_t *h0, int start,
1875                                            int previous_nal_size, int necessary_size )
1876 {
1877     if( h0->nal_buffer_size < necessary_size )
1878     {
1879         necessary_size *= 2;
1880         uint8_t *buf = x264_malloc( necessary_size );
1881         if( !buf )
1882             return -1;
1883         if( previous_nal_size )
1884             memcpy( buf, h0->nal_buffer, previous_nal_size );
1885
1886         intptr_t delta = buf - h0->nal_buffer;
1887         for( int i = 0; i < start; i++ )
1888             h->out.nal[i].p_payload += delta;
1889
1890         x264_free( h0->nal_buffer );
1891         h0->nal_buffer = buf;
1892         h0->nal_buffer_size = necessary_size;
1893     }
1894
1895     return 0;
1896 }
1897
1898 static int x264_encoder_encapsulate_nals( x264_t *h, int start )
1899 {
1900     x264_t *h0 = h->thread[0];
1901     int nal_size = 0, previous_nal_size = 0;
1902
1903     if( h->param.nalu_process )
1904     {
1905         for( int i = start; i < h->out.i_nal; i++ )
1906             nal_size += h->out.nal[i].i_payload;
1907         return nal_size;
1908     }
1909
1910     for( int i = 0; i < start; i++ )
1911         previous_nal_size += h->out.nal[i].i_payload;
1912
1913     for( int i = start; i < h->out.i_nal; i++ )
1914         nal_size += h->out.nal[i].i_payload;
1915
1916     /* Worst-case NAL unit escaping: reallocate the buffer if it's too small. */
1917     int necessary_size = previous_nal_size + nal_size * 3/2 + h->out.i_nal * 4 + 4 + 64;
1918     for( int i = start; i < h->out.i_nal; i++ )
1919         necessary_size += h->out.nal[i].i_padding;
1920     if( x264_check_encapsulated_buffer( h, h0, start, previous_nal_size, necessary_size ) )
1921         return -1;
1922
1923     uint8_t *nal_buffer = h0->nal_buffer + previous_nal_size;
1924
1925     for( int i = start; i < h->out.i_nal; i++ )
1926     {
1927         int old_payload_len = h->out.nal[i].i_payload;
1928         h->out.nal[i].b_long_startcode = !i || h->out.nal[i].i_type == NAL_SPS || h->out.nal[i].i_type == NAL_PPS ||
1929                                          h->param.i_avcintra_class;
1930         x264_nal_encode( h, nal_buffer, &h->out.nal[i] );
1931         nal_buffer += h->out.nal[i].i_payload;
1932         if( h->param.i_avcintra_class )
1933         {
1934             h->out.nal[i].i_padding -= h->out.nal[i].i_payload - (old_payload_len + NALU_OVERHEAD);
1935             if( h->out.nal[i].i_padding > 0 )
1936             {
1937                 memset( nal_buffer, 0, h->out.nal[i].i_padding );
1938                 nal_buffer += h->out.nal[i].i_padding;
1939                 h->out.nal[i].i_payload += h->out.nal[i].i_padding;
1940             }
1941             h->out.nal[i].i_padding = X264_MAX( h->out.nal[i].i_padding, 0 );
1942         }
1943     }
1944
1945     x264_emms();
1946
1947     return nal_buffer - (h0->nal_buffer + previous_nal_size);
1948 }
1949
1950 /****************************************************************************
1951  * x264_encoder_headers:
1952  ****************************************************************************/
1953 int x264_encoder_headers( x264_t *h, x264_nal_t **pp_nal, int *pi_nal )
1954 {
1955     int frame_size = 0;
1956     /* init bitstream context */
1957     h->out.i_nal = 0;
1958     bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
1959
1960     /* Write SEI, SPS and PPS. */
1961
1962     /* generate sequence parameters */
1963     x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
1964     x264_sps_write( &h->out.bs, h->sps );
1965     if( x264_nal_end( h ) )
1966         return -1;
1967
1968     /* generate picture parameters */
1969     x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
1970     x264_pps_write( &h->out.bs, h->sps, h->pps );
1971     if( x264_nal_end( h ) )
1972         return -1;
1973
1974     /* identify ourselves */
1975     x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
1976     if( x264_sei_version_write( h, &h->out.bs ) )
1977         return -1;
1978     if( x264_nal_end( h ) )
1979         return -1;
1980
1981     frame_size = x264_encoder_encapsulate_nals( h, 0 );
1982     if( frame_size < 0 )
1983         return -1;
1984
1985     /* now set output*/
1986     *pi_nal = h->out.i_nal;
1987     *pp_nal = &h->out.nal[0];
1988     h->out.i_nal = 0;
1989
1990     return frame_size;
1991 }
1992
1993 /* Check to see whether we have chosen a reference list ordering different
1994  * from the standard's default. */
1995 static inline void x264_reference_check_reorder( x264_t *h )
1996 {
1997     /* The reorder check doesn't check for missing frames, so just
1998      * force a reorder if one of the reference list is corrupt. */
1999     for( int i = 0; h->frames.reference[i]; i++ )
2000         if( h->frames.reference[i]->b_corrupt )
2001         {
2002             h->b_ref_reorder[0] = 1;
2003             return;
2004         }
2005     for( int list = 0; list <= (h->sh.i_type == SLICE_TYPE_B); list++ )
2006         for( int i = 0; i < h->i_ref[list] - 1; i++ )
2007         {
2008             int framenum_diff = h->fref[list][i+1]->i_frame_num - h->fref[list][i]->i_frame_num;
2009             int poc_diff = h->fref[list][i+1]->i_poc - h->fref[list][i]->i_poc;
2010             /* P and B-frames use different default orders. */
2011             if( h->sh.i_type == SLICE_TYPE_P ? framenum_diff > 0 : list == 1 ? poc_diff < 0 : poc_diff > 0 )
2012             {
2013                 h->b_ref_reorder[list] = 1;
2014                 return;
2015             }
2016         }
2017 }
2018
2019 /* return -1 on failure, else return the index of the new reference frame */
2020 int x264_weighted_reference_duplicate( x264_t *h, int i_ref, const x264_weight_t *w )
2021 {
2022     int i = h->i_ref[0];
2023     int j = 1;
2024     x264_frame_t *newframe;
2025     if( i <= 1 ) /* empty list, definitely can't duplicate frame */
2026         return -1;
2027
2028     //Duplication is only used in X264_WEIGHTP_SMART
2029     if( h->param.analyse.i_weighted_pred != X264_WEIGHTP_SMART )
2030         return -1;
2031
2032     /* Duplication is a hack to compensate for crappy rounding in motion compensation.
2033      * With high bit depth, it's not worth doing, so turn it off except in the case of
2034      * unweighted dupes. */
2035     if( BIT_DEPTH > 8 && w != x264_weight_none )
2036         return -1;
2037
2038     newframe = x264_frame_pop_blank_unused( h );
2039     if( !newframe )
2040         return -1;
2041
2042     //FIXME: probably don't need to copy everything
2043     *newframe = *h->fref[0][i_ref];
2044     newframe->i_reference_count = 1;
2045     newframe->orig = h->fref[0][i_ref];
2046     newframe->b_duplicate = 1;
2047     memcpy( h->fenc->weight[j], w, sizeof(h->fenc->weight[i]) );
2048
2049     /* shift the frames to make space for the dupe. */
2050     h->b_ref_reorder[0] = 1;
2051     if( h->i_ref[0] < X264_REF_MAX )
2052         ++h->i_ref[0];
2053     h->fref[0][X264_REF_MAX-1] = NULL;
2054     x264_frame_unshift( &h->fref[0][j], newframe );
2055
2056     return j;
2057 }
2058
2059 static void x264_weighted_pred_init( x264_t *h )
2060 {
2061     /* for now no analysis and set all weights to nothing */
2062     for( int i_ref = 0; i_ref < h->i_ref[0]; i_ref++ )
2063         h->fenc->weighted[i_ref] = h->fref[0][i_ref]->filtered[0][0];
2064
2065     // FIXME: This only supports weighting of one reference frame
2066     // and duplicates of that frame.
2067     h->fenc->i_lines_weighted = 0;
2068
2069     for( int i_ref = 0; i_ref < (h->i_ref[0] << SLICE_MBAFF); i_ref++ )
2070         for( int i = 0; i < 3; i++ )
2071             h->sh.weight[i_ref][i].weightfn = NULL;
2072
2073
2074     if( h->sh.i_type != SLICE_TYPE_P || h->param.analyse.i_weighted_pred <= 0 )
2075         return;
2076
2077     int i_padv = PADV << PARAM_INTERLACED;
2078     int denom = -1;
2079     int weightplane[2] = { 0, 0 };
2080     int buffer_next = 0;
2081     for( int i = 0; i < 3; i++ )
2082     {
2083         for( int j = 0; j < h->i_ref[0]; j++ )
2084         {
2085             if( h->fenc->weight[j][i].weightfn )
2086             {
2087                 h->sh.weight[j][i] = h->fenc->weight[j][i];
2088                 // if weight is useless, don't write it to stream
2089                 if( h->sh.weight[j][i].i_scale == 1<<h->sh.weight[j][i].i_denom && h->sh.weight[j][i].i_offset == 0 )
2090                     h->sh.weight[j][i].weightfn = NULL;
2091                 else
2092                 {
2093                     if( !weightplane[!!i] )
2094                     {
2095                         weightplane[!!i] = 1;
2096                         h->sh.weight[0][!!i].i_denom = denom = h->sh.weight[j][i].i_denom;
2097                         assert( x264_clip3( denom, 0, 7 ) == denom );
2098                     }
2099
2100                     assert( h->sh.weight[j][i].i_denom == denom );
2101                     if( !i )
2102                     {
2103                         h->fenc->weighted[j] = h->mb.p_weight_buf[buffer_next++] + h->fenc->i_stride[0] * i_padv + PADH;
2104                         //scale full resolution frame
2105                         if( h->param.i_threads == 1 )
2106                         {
2107                             pixel *src = h->fref[0][j]->filtered[0][0] - h->fref[0][j]->i_stride[0]*i_padv - PADH;
2108                             pixel *dst = h->fenc->weighted[j] - h->fenc->i_stride[0]*i_padv - PADH;
2109                             int stride = h->fenc->i_stride[0];
2110                             int width = h->fenc->i_width[0] + PADH*2;
2111                             int height = h->fenc->i_lines[0] + i_padv*2;
2112                             x264_weight_scale_plane( h, dst, stride, src, stride, width, height, &h->sh.weight[j][0] );
2113                             h->fenc->i_lines_weighted = height;
2114                         }
2115                     }
2116                 }
2117             }
2118         }
2119     }
2120
2121     if( weightplane[1] )
2122         for( int i = 0; i < h->i_ref[0]; i++ )
2123         {
2124             if( h->sh.weight[i][1].weightfn && !h->sh.weight[i][2].weightfn )
2125             {
2126                 h->sh.weight[i][2].i_scale = 1 << h->sh.weight[0][1].i_denom;
2127                 h->sh.weight[i][2].i_offset = 0;
2128             }
2129             else if( h->sh.weight[i][2].weightfn && !h->sh.weight[i][1].weightfn )
2130             {
2131                 h->sh.weight[i][1].i_scale = 1 << h->sh.weight[0][1].i_denom;
2132                 h->sh.weight[i][1].i_offset = 0;
2133             }
2134         }
2135
2136     if( !weightplane[0] )
2137         h->sh.weight[0][0].i_denom = 0;
2138     if( !weightplane[1] )
2139         h->sh.weight[0][1].i_denom = 0;
2140     h->sh.weight[0][2].i_denom = h->sh.weight[0][1].i_denom;
2141 }
2142
2143 static inline int x264_reference_distance( x264_t *h, x264_frame_t *frame )
2144 {
2145     if( h->param.i_frame_packing == 5 )
2146         return abs((h->fenc->i_frame&~1) - (frame->i_frame&~1)) +
2147                   ((h->fenc->i_frame&1) != (frame->i_frame&1));
2148     else
2149         return abs(h->fenc->i_frame - frame->i_frame);
2150 }
2151
2152 static inline void x264_reference_build_list( x264_t *h, int i_poc )
2153 {
2154     int b_ok;
2155
2156     /* build ref list 0/1 */
2157     h->mb.pic.i_fref[0] = h->i_ref[0] = 0;
2158     h->mb.pic.i_fref[1] = h->i_ref[1] = 0;
2159     if( h->sh.i_type == SLICE_TYPE_I )
2160         return;
2161
2162     for( int i = 0; h->frames.reference[i]; i++ )
2163     {
2164         if( h->frames.reference[i]->b_corrupt )
2165             continue;
2166         if( h->frames.reference[i]->i_poc < i_poc )
2167             h->fref[0][h->i_ref[0]++] = h->frames.reference[i];
2168         else if( h->frames.reference[i]->i_poc > i_poc )
2169             h->fref[1][h->i_ref[1]++] = h->frames.reference[i];
2170     }
2171
2172     if( h->sh.i_mmco_remove_from_end )
2173     {
2174         /* Order ref0 for MMCO remove */
2175         do
2176         {
2177             b_ok = 1;
2178             for( int i = 0; i < h->i_ref[0] - 1; i++ )
2179             {
2180                 if( h->fref[0][i]->i_frame < h->fref[0][i+1]->i_frame )
2181                 {
2182                     XCHG( x264_frame_t*, h->fref[0][i], h->fref[0][i+1] );
2183                     b_ok = 0;
2184                     break;
2185                 }
2186             }
2187         } while( !b_ok );
2188
2189         for( int i = h->i_ref[0]-1; i >= h->i_ref[0] - h->sh.i_mmco_remove_from_end; i-- )
2190         {
2191             int diff = h->i_frame_num - h->fref[0][i]->i_frame_num;
2192             h->sh.mmco[h->sh.i_mmco_command_count].i_poc = h->fref[0][i]->i_poc;
2193             h->sh.mmco[h->sh.i_mmco_command_count++].i_difference_of_pic_nums = diff;
2194         }
2195     }
2196
2197     /* Order reference lists by distance from the current frame. */
2198     for( int list = 0; list < 2; list++ )
2199     {
2200         h->fref_nearest[list] = h->fref[list][0];
2201         do
2202         {
2203             b_ok = 1;
2204             for( int i = 0; i < h->i_ref[list] - 1; i++ )
2205             {
2206                 if( list ? h->fref[list][i+1]->i_poc < h->fref_nearest[list]->i_poc
2207                          : h->fref[list][i+1]->i_poc > h->fref_nearest[list]->i_poc )
2208                     h->fref_nearest[list] = h->fref[list][i+1];
2209                 if( x264_reference_distance( h, h->fref[list][i] ) > x264_reference_distance( h, h->fref[list][i+1] ) )
2210                 {
2211                     XCHG( x264_frame_t*, h->fref[list][i], h->fref[list][i+1] );
2212                     b_ok = 0;
2213                     break;
2214                 }
2215             }
2216         } while( !b_ok );
2217     }
2218
2219     x264_reference_check_reorder( h );
2220
2221     h->i_ref[1] = X264_MIN( h->i_ref[1], h->frames.i_max_ref1 );
2222     h->i_ref[0] = X264_MIN( h->i_ref[0], h->frames.i_max_ref0 );
2223     h->i_ref[0] = X264_MIN( h->i_ref[0], h->param.i_frame_reference ); // if reconfig() has lowered the limit
2224
2225     /* For Blu-ray compliance, don't reference frames outside of the minigop. */
2226     if( IS_X264_TYPE_B( h->fenc->i_type ) && h->param.b_bluray_compat )
2227         h->i_ref[0] = X264_MIN( h->i_ref[0], IS_X264_TYPE_B( h->fref[0][0]->i_type ) + 1 );
2228
2229     /* add duplicates */
2230     if( h->fenc->i_type == X264_TYPE_P )
2231     {
2232         int idx = -1;
2233         if( h->param.analyse.i_weighted_pred >= X264_WEIGHTP_SIMPLE )
2234         {
2235             x264_weight_t w[3];
2236             w[1].weightfn = w[2].weightfn = NULL;
2237             if( h->param.rc.b_stat_read )
2238                 x264_ratecontrol_set_weights( h, h->fenc );
2239
2240             if( !h->fenc->weight[0][0].weightfn )
2241             {
2242                 h->fenc->weight[0][0].i_denom = 0;
2243                 SET_WEIGHT( w[0], 1, 1, 0, -1 );
2244                 idx = x264_weighted_reference_duplicate( h, 0, w );
2245             }
2246             else
2247             {
2248                 if( h->fenc->weight[0][0].i_scale == 1<<h->fenc->weight[0][0].i_denom )
2249                 {
2250                     SET_WEIGHT( h->fenc->weight[0][0], 1, 1, 0, h->fenc->weight[0][0].i_offset );
2251                 }
2252                 x264_weighted_reference_duplicate( h, 0, x264_weight_none );
2253                 if( h->fenc->weight[0][0].i_offset > -128 )
2254                 {
2255                     w[0] = h->fenc->weight[0][0];
2256                     w[0].i_offset--;
2257                     h->mc.weight_cache( h, &w[0] );
2258                     idx = x264_weighted_reference_duplicate( h, 0, w );
2259                 }
2260             }
2261         }
2262         h->mb.ref_blind_dupe = idx;
2263     }
2264
2265     assert( h->i_ref[0] + h->i_ref[1] <= X264_REF_MAX );
2266     h->mb.pic.i_fref[0] = h->i_ref[0];
2267     h->mb.pic.i_fref[1] = h->i_ref[1];
2268 }
2269
2270 static void x264_fdec_filter_row( x264_t *h, int mb_y, int pass )
2271 {
2272     /* mb_y is the mb to be encoded next, not the mb to be filtered here */
2273     int b_hpel = h->fdec->b_kept_as_ref;
2274     int b_deblock = h->sh.i_disable_deblocking_filter_idc != 1;
2275     int b_end = mb_y == h->i_threadslice_end;
2276     int b_measure_quality = 1;
2277     int min_y = mb_y - (1 << SLICE_MBAFF);
2278     int b_start = min_y == h->i_threadslice_start;
2279     /* Even in interlaced mode, deblocking never modifies more than 4 pixels
2280      * above each MB, as bS=4 doesn't happen for the top of interlaced mbpairs. */
2281     int minpix_y = min_y*16 - 4 * !b_start;
2282     int maxpix_y = mb_y*16 - 4 * !b_end;
2283     b_deblock &= b_hpel || h->param.b_full_recon || h->param.psz_dump_yuv;
2284     if( h->param.b_sliced_threads )
2285     {
2286         switch( pass )
2287         {
2288             /* During encode: only do deblock if asked for */
2289             default:
2290             case 0:
2291                 b_deblock &= h->param.b_full_recon;
2292                 b_hpel = 0;
2293                 break;
2294             /* During post-encode pass: do deblock if not done yet, do hpel for all
2295              * rows except those between slices. */
2296             case 1:
2297                 b_deblock &= !h->param.b_full_recon;
2298                 b_hpel &= !(b_start && min_y > 0);
2299                 b_measure_quality = 0;
2300                 break;
2301             /* Final pass: do the rows between slices in sequence. */
2302             case 2:
2303                 b_deblock = 0;
2304                 b_measure_quality = 0;
2305                 break;
2306         }
2307     }
2308     if( mb_y & SLICE_MBAFF )
2309         return;
2310     if( min_y < h->i_threadslice_start )
2311         return;
2312
2313     if( b_deblock )
2314         for( int y = min_y; y < mb_y; y += (1 << SLICE_MBAFF) )
2315             x264_frame_deblock_row( h, y );
2316
2317     /* FIXME: Prediction requires different borders for interlaced/progressive mc,
2318      * but the actual image data is equivalent. For now, maintain this
2319      * consistency by copying deblocked pixels between planes. */
2320     if( PARAM_INTERLACED && (!h->param.b_sliced_threads || pass == 1) )
2321         for( int p = 0; p < h->fdec->i_plane; p++ )
2322             for( int i = minpix_y>>(CHROMA_V_SHIFT && p); i < maxpix_y>>(CHROMA_V_SHIFT && p); i++ )
2323                 memcpy( h->fdec->plane_fld[p] + i*h->fdec->i_stride[p],
2324                         h->fdec->plane[p]     + i*h->fdec->i_stride[p],
2325                         h->mb.i_mb_width*16*sizeof(pixel) );
2326
2327     if( h->fdec->b_kept_as_ref && (!h->param.b_sliced_threads || pass == 1) )
2328         x264_frame_expand_border( h, h->fdec, min_y );
2329     if( b_hpel )
2330     {
2331         int end = mb_y == h->mb.i_mb_height;
2332         /* Can't do hpel until the previous slice is done encoding. */
2333         if( h->param.analyse.i_subpel_refine )
2334         {
2335             x264_frame_filter( h, h->fdec, min_y, end );
2336             x264_frame_expand_border_filtered( h, h->fdec, min_y, end );
2337         }
2338     }
2339
2340     if( SLICE_MBAFF && pass == 0 )
2341         for( int i = 0; i < 3; i++ )
2342         {
2343             XCHG( pixel *, h->intra_border_backup[0][i], h->intra_border_backup[3][i] );
2344             XCHG( pixel *, h->intra_border_backup[1][i], h->intra_border_backup[4][i] );
2345         }
2346
2347     if( h->i_thread_frames > 1 && h->fdec->b_kept_as_ref )
2348         x264_frame_cond_broadcast( h->fdec, mb_y*16 + (b_end ? 10000 : -(X264_THREAD_HEIGHT << SLICE_MBAFF)) );
2349
2350     if( b_measure_quality )
2351     {
2352         maxpix_y = X264_MIN( maxpix_y, h->param.i_height );
2353         if( h->param.analyse.b_psnr )
2354         {
2355             for( int p = 0; p < (CHROMA444 ? 3 : 1); p++ )
2356                 h->stat.frame.i_ssd[p] += x264_pixel_ssd_wxh( &h->pixf,
2357                     h->fdec->plane[p] + minpix_y * h->fdec->i_stride[p], h->fdec->i_stride[p],
2358                     h->fenc->plane[p] + minpix_y * h->fenc->i_stride[p], h->fenc->i_stride[p],
2359                     h->param.i_width, maxpix_y-minpix_y );
2360             if( !CHROMA444 )
2361             {
2362                 uint64_t ssd_u, ssd_v;
2363                 int v_shift = CHROMA_V_SHIFT;
2364                 x264_pixel_ssd_nv12( &h->pixf,
2365                     h->fdec->plane[1] + (minpix_y>>v_shift) * h->fdec->i_stride[1], h->fdec->i_stride[1],
2366                     h->fenc->plane[1] + (minpix_y>>v_shift) * h->fenc->i_stride[1], h->fenc->i_stride[1],
2367                     h->param.i_width>>1, (maxpix_y-minpix_y)>>v_shift, &ssd_u, &ssd_v );
2368                 h->stat.frame.i_ssd[1] += ssd_u;
2369                 h->stat.frame.i_ssd[2] += ssd_v;
2370             }
2371         }
2372
2373         if( h->param.analyse.b_ssim )
2374         {
2375             int ssim_cnt;
2376             x264_emms();
2377             /* offset by 2 pixels to avoid alignment of ssim blocks with dct blocks,
2378              * and overlap by 4 */
2379             minpix_y += b_start ? 2 : -6;
2380             h->stat.frame.f_ssim +=
2381                 x264_pixel_ssim_wxh( &h->pixf,
2382                     h->fdec->plane[0] + 2+minpix_y*h->fdec->i_stride[0], h->fdec->i_stride[0],
2383                     h->fenc->plane[0] + 2+minpix_y*h->fenc->i_stride[0], h->fenc->i_stride[0],
2384                     h->param.i_width-2, maxpix_y-minpix_y, h->scratch_buffer, &ssim_cnt );
2385             h->stat.frame.i_ssim_cnt += ssim_cnt;
2386         }
2387     }
2388 }
2389
2390 static inline int x264_reference_update( x264_t *h )
2391 {
2392     if( !h->fdec->b_kept_as_ref )
2393     {
2394         if( h->i_thread_frames > 1 )
2395         {
2396             x264_frame_push_unused( h, h->fdec );
2397             h->fdec = x264_frame_pop_unused( h, 1 );
2398             if( !h->fdec )
2399                 return -1;
2400         }
2401         return 0;
2402     }
2403
2404     /* apply mmco from previous frame. */
2405     for( int i = 0; i < h->sh.i_mmco_command_count; i++ )
2406         for( int j = 0; h->frames.reference[j]; j++ )
2407             if( h->frames.reference[j]->i_poc == h->sh.mmco[i].i_poc )
2408                 x264_frame_push_unused( h, x264_frame_shift( &h->frames.reference[j] ) );
2409
2410     /* move frame in the buffer */
2411     x264_frame_push( h->frames.reference, h->fdec );
2412     if( h->frames.reference[h->sps->i_num_ref_frames] )
2413         x264_frame_push_unused( h, x264_frame_shift( h->frames.reference ) );
2414     h->fdec = x264_frame_pop_unused( h, 1 );
2415     if( !h->fdec )
2416         return -1;
2417     return 0;
2418 }
2419
2420 static inline void x264_reference_reset( x264_t *h )
2421 {
2422     while( h->frames.reference[0] )
2423         x264_frame_push_unused( h, x264_frame_pop( h->frames.reference ) );
2424     h->fdec->i_poc =
2425     h->fenc->i_poc = 0;
2426 }
2427
2428 static inline void x264_reference_hierarchy_reset( x264_t *h )
2429 {
2430     int ref;
2431     int b_hasdelayframe = 0;
2432
2433     /* look for delay frames -- chain must only contain frames that are disposable */
2434     for( int i = 0; h->frames.current[i] && IS_DISPOSABLE( h->frames.current[i]->i_type ); i++ )
2435         b_hasdelayframe |= h->frames.current[i]->i_coded
2436                         != h->frames.current[i]->i_frame + h->sps->vui.i_num_reorder_frames;
2437
2438     /* This function must handle b-pyramid and clear frames for open-gop */
2439     if( h->param.i_bframe_pyramid != X264_B_PYRAMID_STRICT && !b_hasdelayframe && h->frames.i_poc_last_open_gop == -1 )
2440         return;
2441
2442     /* Remove last BREF. There will never be old BREFs in the
2443      * dpb during a BREF decode when pyramid == STRICT */
2444     for( ref = 0; h->frames.reference[ref]; ref++ )
2445     {
2446         if( ( h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT
2447             && h->frames.reference[ref]->i_type == X264_TYPE_BREF )
2448             || ( h->frames.reference[ref]->i_poc < h->frames.i_poc_last_open_gop
2449             && h->sh.i_type != SLICE_TYPE_B ) )
2450         {
2451             int diff = h->i_frame_num - h->frames.reference[ref]->i_frame_num;
2452             h->sh.mmco[h->sh.i_mmco_command_count].i_difference_of_pic_nums = diff;
2453             h->sh.mmco[h->sh.i_mmco_command_count++].i_poc = h->frames.reference[ref]->i_poc;
2454             x264_frame_push_unused( h, x264_frame_shift( &h->frames.reference[ref] ) );
2455             h->b_ref_reorder[0] = 1;
2456             ref--;
2457         }
2458     }
2459
2460     /* Prepare room in the dpb for the delayed display time of the later b-frame's */
2461     if( h->param.i_bframe_pyramid )
2462         h->sh.i_mmco_remove_from_end = X264_MAX( ref + 2 - h->frames.i_max_dpb, 0 );
2463 }
2464
2465 static inline void x264_slice_init( x264_t *h, int i_nal_type, int i_global_qp )
2466 {
2467     /* ------------------------ Create slice header  ----------------------- */
2468     if( i_nal_type == NAL_SLICE_IDR )
2469     {
2470         x264_slice_header_init( h, &h->sh, h->sps, h->pps, h->i_idr_pic_id, h->i_frame_num, i_global_qp );
2471
2472         /* alternate id */
2473         if( h->param.i_avcintra_class )
2474         {
2475             switch( h->i_idr_pic_id )
2476             {
2477                 case 5:
2478                     h->i_idr_pic_id = 3;
2479                     break;
2480                 case 3:
2481                     h->i_idr_pic_id = 4;
2482                     break;
2483                 case 4:
2484                 default:
2485                     h->i_idr_pic_id = 5;
2486                     break;
2487             }
2488         }
2489         else
2490             h->i_idr_pic_id ^= 1;
2491     }
2492     else
2493     {
2494         x264_slice_header_init( h, &h->sh, h->sps, h->pps, -1, h->i_frame_num, i_global_qp );
2495
2496         h->sh.i_num_ref_idx_l0_active = h->i_ref[0] <= 0 ? 1 : h->i_ref[0];
2497         h->sh.i_num_ref_idx_l1_active = h->i_ref[1] <= 0 ? 1 : h->i_ref[1];
2498         if( h->sh.i_num_ref_idx_l0_active != h->pps->i_num_ref_idx_l0_default_active ||
2499             (h->sh.i_type == SLICE_TYPE_B && h->sh.i_num_ref_idx_l1_active != h->pps->i_num_ref_idx_l1_default_active) )
2500         {
2501             h->sh.b_num_ref_idx_override = 1;
2502         }
2503     }
2504
2505     if( h->fenc->i_type == X264_TYPE_BREF && h->param.b_bluray_compat && h->sh.i_mmco_command_count )
2506     {
2507         h->b_sh_backup = 1;
2508         h->sh_backup = h->sh;
2509     }
2510
2511     h->fdec->i_frame_num = h->sh.i_frame_num;
2512
2513     if( h->sps->i_poc_type == 0 )
2514     {
2515         h->sh.i_poc = h->fdec->i_poc;
2516         if( PARAM_INTERLACED )
2517         {
2518             h->sh.i_delta_poc_bottom = h->param.b_tff ? 1 : -1;
2519             h->sh.i_poc += h->sh.i_delta_poc_bottom == -1;
2520         }
2521         else
2522             h->sh.i_delta_poc_bottom = 0;
2523         h->fdec->i_delta_poc[0] = h->sh.i_delta_poc_bottom == -1;
2524         h->fdec->i_delta_poc[1] = h->sh.i_delta_poc_bottom ==  1;
2525     }
2526     else
2527     {
2528         /* Nothing to do ? */
2529     }
2530
2531     x264_macroblock_slice_init( h );
2532 }
2533
2534 typedef struct
2535 {
2536     int skip;
2537     uint8_t cabac_prevbyte;
2538     bs_t bs;
2539     x264_cabac_t cabac;
2540     x264_frame_stat_t stat;
2541     int last_qp;
2542     int last_dqp;
2543     int field_decoding_flag;
2544 } x264_bs_bak_t;
2545
2546 static ALWAYS_INLINE void x264_bitstream_backup( x264_t *h, x264_bs_bak_t *bak, int i_skip, int full )
2547 {
2548     if( full )
2549     {
2550         bak->stat = h->stat.frame;
2551         bak->last_qp = h->mb.i_last_qp;
2552         bak->last_dqp = h->mb.i_last_dqp;
2553         bak->field_decoding_flag = h->mb.field_decoding_flag;
2554     }
2555     else
2556     {
2557         bak->stat.i_mv_bits = h->stat.frame.i_mv_bits;
2558         bak->stat.i_tex_bits = h->stat.frame.i_tex_bits;
2559     }
2560     /* In the per-MB backup, we don't need the contexts because flushing the CABAC
2561      * encoder has no context dependency and in this case, a slice is ended (and
2562      * thus the content of all contexts are thrown away). */
2563     if( h->param.b_cabac )
2564     {
2565         if( full )
2566             memcpy( &bak->cabac, &h->cabac, sizeof(x264_cabac_t) );
2567         else
2568             memcpy( &bak->cabac, &h->cabac, offsetof(x264_cabac_t, f8_bits_encoded) );
2569         /* x264's CABAC writer modifies the previous byte during carry, so it has to be
2570          * backed up. */
2571         bak->cabac_prevbyte = h->cabac.p[-1];
2572     }
2573     else
2574     {
2575         bak->bs = h->out.bs;
2576         bak->skip = i_skip;
2577     }
2578 }
2579
2580 static ALWAYS_INLINE void x264_bitstream_restore( x264_t *h, x264_bs_bak_t *bak, int *skip, int full )
2581 {
2582     if( full )
2583     {
2584         h->stat.frame = bak->stat;
2585         h->mb.i_last_qp = bak->last_qp;
2586         h->mb.i_last_dqp = bak->last_dqp;
2587         h->mb.field_decoding_flag = bak->field_decoding_flag;
2588     }
2589     else
2590     {
2591         h->stat.frame.i_mv_bits = bak->stat.i_mv_bits;
2592         h->stat.frame.i_tex_bits = bak->stat.i_tex_bits;
2593     }
2594     if( h->param.b_cabac )
2595     {
2596         if( full )
2597             memcpy( &h->cabac, &bak->cabac, sizeof(x264_cabac_t) );
2598         else
2599             memcpy( &h->cabac, &bak->cabac, offsetof(x264_cabac_t, f8_bits_encoded) );
2600         h->cabac.p[-1] = bak->cabac_prevbyte;
2601     }
2602     else
2603     {
2604         h->out.bs = bak->bs;
2605         *skip = bak->skip;
2606     }
2607 }
2608
2609 static intptr_t x264_slice_write( x264_t *h )
2610 {
2611     int i_skip;
2612     int mb_xy, i_mb_x, i_mb_y;
2613     /* NALUs other than the first use a 3-byte startcode.
2614      * Add one extra byte for the rbsp, and one more for the final CABAC putbyte.
2615      * Then add an extra 5 bytes just in case, to account for random NAL escapes and
2616      * other inaccuracies. */
2617     int overhead_guess = (NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal)) + 1 + h->param.b_cabac + 5;
2618     int slice_max_size = h->param.i_slice_max_size > 0 ? (h->param.i_slice_max_size-overhead_guess)*8 : 0;
2619     int back_up_bitstream_cavlc = !h->param.b_cabac && h->sps->i_profile_idc < PROFILE_HIGH;
2620     int back_up_bitstream = slice_max_size || back_up_bitstream_cavlc;
2621     int starting_bits = bs_pos(&h->out.bs);
2622     int b_deblock = h->sh.i_disable_deblocking_filter_idc != 1;
2623     int b_hpel = h->fdec->b_kept_as_ref;
2624     int orig_last_mb = h->sh.i_last_mb;
2625     int thread_last_mb = h->i_threadslice_end * h->mb.i_mb_width - 1;
2626     uint8_t *last_emu_check;
2627 #define BS_BAK_SLICE_MAX_SIZE 0
2628 #define BS_BAK_CAVLC_OVERFLOW 1
2629 #define BS_BAK_SLICE_MIN_MBS  2
2630 #define BS_BAK_ROW_VBV        3
2631     x264_bs_bak_t bs_bak[4];
2632     b_deblock &= b_hpel || h->param.b_full_recon || h->param.psz_dump_yuv;
2633     bs_realign( &h->out.bs );
2634
2635     /* Slice */
2636     x264_nal_start( h, h->i_nal_type, h->i_nal_ref_idc );
2637     h->out.nal[h->out.i_nal].i_first_mb = h->sh.i_first_mb;
2638
2639     /* Slice header */
2640     x264_macroblock_thread_init( h );
2641
2642     /* Set the QP equal to the first QP in the slice for more accurate CABAC initialization. */
2643     h->mb.i_mb_xy = h->sh.i_first_mb;
2644     h->sh.i_qp = x264_ratecontrol_mb_qp( h );
2645     h->sh.i_qp = SPEC_QP( h->sh.i_qp );
2646     h->sh.i_qp_delta = h->sh.i_qp - h->pps->i_pic_init_qp;
2647
2648     x264_slice_header_write( &h->out.bs, &h->sh, h->i_nal_ref_idc );
2649     if( h->param.b_cabac )
2650     {
2651         /* alignment needed */
2652         bs_align_1( &h->out.bs );
2653
2654         /* init cabac */
2655         x264_cabac_context_init( h, &h->cabac, h->sh.i_type, x264_clip3( h->sh.i_qp-QP_BD_OFFSET, 0, 51 ), h->sh.i_cabac_init_idc );
2656         x264_cabac_encode_init ( &h->cabac, h->out.bs.p, h->out.bs.p_end );
2657         last_emu_check = h->cabac.p;
2658     }
2659     else
2660         last_emu_check = h->out.bs.p;
2661     h->mb.i_last_qp = h->sh.i_qp;
2662     h->mb.i_last_dqp = 0;
2663     h->mb.field_decoding_flag = 0;
2664
2665     i_mb_y = h->sh.i_first_mb / h->mb.i_mb_width;
2666     i_mb_x = h->sh.i_first_mb % h->mb.i_mb_width;
2667     i_skip = 0;
2668
2669     while( 1 )
2670     {
2671         mb_xy = i_mb_x + i_mb_y * h->mb.i_mb_width;
2672         int mb_spos = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
2673
2674         if( i_mb_x == 0 )
2675         {
2676             if( x264_bitstream_check_buffer( h ) )
2677                 return -1;
2678             if( !(i_mb_y & SLICE_MBAFF) && h->param.rc.i_vbv_buffer_size )
2679                 x264_bitstream_backup( h, &bs_bak[BS_BAK_ROW_VBV], i_skip, 1 );
2680             if( !h->mb.b_reencode_mb )
2681                 x264_fdec_filter_row( h, i_mb_y, 0 );
2682         }
2683
2684         if( back_up_bitstream )
2685         {
2686             if( back_up_bitstream_cavlc )
2687                 x264_bitstream_backup( h, &bs_bak[BS_BAK_CAVLC_OVERFLOW], i_skip, 0 );
2688             if( slice_max_size && !(i_mb_y & SLICE_MBAFF) )
2689             {
2690                 x264_bitstream_backup( h, &bs_bak[BS_BAK_SLICE_MAX_SIZE], i_skip, 0 );
2691                 if( (thread_last_mb+1-mb_xy) == h->param.i_slice_min_mbs )
2692                     x264_bitstream_backup( h, &bs_bak[BS_BAK_SLICE_MIN_MBS], i_skip, 0 );
2693             }
2694         }
2695
2696         if( PARAM_INTERLACED )
2697         {
2698             if( h->mb.b_adaptive_mbaff )
2699             {
2700                 if( !(i_mb_y&1) )
2701                 {
2702                     /* FIXME: VSAD is fast but fairly poor at choosing the best interlace type. */
2703                     h->mb.b_interlaced = x264_field_vsad( h, i_mb_x, i_mb_y );
2704                     memcpy( &h->zigzagf, MB_INTERLACED ? &h->zigzagf_interlaced : &h->zigzagf_progressive, sizeof(h->zigzagf) );
2705                     if( !MB_INTERLACED && (i_mb_y+2) == h->mb.i_mb_height )
2706                         x264_expand_border_mbpair( h, i_mb_x, i_mb_y );
2707                 }
2708             }
2709             h->mb.field[mb_xy] = MB_INTERLACED;
2710         }
2711
2712         /* load cache */
2713         if( SLICE_MBAFF )
2714             x264_macroblock_cache_load_interlaced( h, i_mb_x, i_mb_y );
2715         else
2716             x264_macroblock_cache_load_progressive( h, i_mb_x, i_mb_y );
2717
2718         x264_macroblock_analyse( h );
2719
2720         /* encode this macroblock -> be careful it can change the mb type to P_SKIP if needed */
2721 reencode:
2722         x264_macroblock_encode( h );
2723
2724         if( h->param.b_cabac )
2725         {
2726             if( mb_xy > h->sh.i_first_mb && !(SLICE_MBAFF && (i_mb_y&1)) )
2727                 x264_cabac_encode_terminal( &h->cabac );
2728
2729             if( IS_SKIP( h->mb.i_type ) )
2730                 x264_cabac_mb_skip( h, 1 );
2731             else
2732             {
2733                 if( h->sh.i_type != SLICE_TYPE_I )
2734                     x264_cabac_mb_skip( h, 0 );
2735                 x264_macroblock_write_cabac( h, &h->cabac );
2736             }
2737         }
2738         else
2739         {
2740             if( IS_SKIP( h->mb.i_type ) )
2741                 i_skip++;
2742             else
2743             {
2744                 if( h->sh.i_type != SLICE_TYPE_I )
2745                 {
2746                     bs_write_ue( &h->out.bs, i_skip );  /* skip run */
2747                     i_skip = 0;
2748                 }
2749                 x264_macroblock_write_cavlc( h );
2750                 /* If there was a CAVLC level code overflow, try again at a higher QP. */
2751                 if( h->mb.b_overflow )
2752                 {
2753                     h->mb.i_chroma_qp = h->chroma_qp_table[++h->mb.i_qp];
2754                     h->mb.i_skip_intra = 0;
2755                     h->mb.b_skip_mc = 0;
2756                     h->mb.b_overflow = 0;
2757                     x264_bitstream_restore( h, &bs_bak[BS_BAK_CAVLC_OVERFLOW], &i_skip, 0 );
2758                     goto reencode;
2759                 }
2760             }
2761         }
2762
2763         int total_bits = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
2764         int mb_size = total_bits - mb_spos;
2765
2766         if( slice_max_size && (!SLICE_MBAFF || (i_mb_y&1)) )
2767         {
2768             /* Count the skip run, just in case. */
2769             if( !h->param.b_cabac )
2770                 total_bits += bs_size_ue_big( i_skip );
2771             /* Check for escape bytes. */
2772             uint8_t *end = h->param.b_cabac ? h->cabac.p : h->out.bs.p;
2773             for( ; last_emu_check < end - 2; last_emu_check++ )
2774                 if( last_emu_check[0] == 0 && last_emu_check[1] == 0 && last_emu_check[2] <= 3 )
2775                 {
2776                     slice_max_size -= 8;
2777                     last_emu_check++;
2778                 }
2779             /* We'll just re-encode this last macroblock if we go over the max slice size. */
2780             if( total_bits - starting_bits > slice_max_size && !h->mb.b_reencode_mb )
2781             {
2782                 if( !x264_frame_new_slice( h, h->fdec ) )
2783                 {
2784                     /* Handle the most obnoxious slice-min-mbs edge case: we need to end the slice
2785                      * because it's gone over the maximum size, but doing so would violate slice-min-mbs.
2786                      * If possible, roll back to the last checkpoint and try again.
2787                      * We could try raising QP, but that would break in the case where a slice spans multiple
2788                      * rows, which the re-encoding infrastructure can't currently handle. */
2789                     if( mb_xy <= thread_last_mb && (thread_last_mb+1-mb_xy) < h->param.i_slice_min_mbs )
2790                     {
2791                         if( thread_last_mb-h->param.i_slice_min_mbs < h->sh.i_first_mb+h->param.i_slice_min_mbs )
2792                         {
2793                             x264_log( h, X264_LOG_WARNING, "slice-max-size violated (frame %d, cause: slice-min-mbs)\n", h->i_frame );
2794                             slice_max_size = 0;
2795                             goto cont;
2796                         }
2797                         x264_bitstream_restore( h, &bs_bak[BS_BAK_SLICE_MIN_MBS], &i_skip, 0 );
2798                         h->mb.b_reencode_mb = 1;
2799                         h->sh.i_last_mb = thread_last_mb-h->param.i_slice_min_mbs;
2800                         break;
2801                     }
2802                     if( mb_xy-SLICE_MBAFF*h->mb.i_mb_stride != h->sh.i_first_mb )
2803                     {
2804                         x264_bitstream_restore( h, &bs_bak[BS_BAK_SLICE_MAX_SIZE], &i_skip, 0 );
2805                         h->mb.b_reencode_mb = 1;
2806                         if( SLICE_MBAFF )
2807                         {
2808                             // set to bottom of previous mbpair
2809                             if( i_mb_x )
2810                                 h->sh.i_last_mb = mb_xy-1+h->mb.i_mb_stride*(!(i_mb_y&1));
2811                             else
2812                                 h->sh.i_last_mb = (i_mb_y-2+!(i_mb_y&1))*h->mb.i_mb_stride + h->mb.i_mb_width - 1;
2813                         }
2814                         else
2815                             h->sh.i_last_mb = mb_xy-1;
2816                         break;
2817                     }
2818                     else
2819                         h->sh.i_last_mb = mb_xy;
2820                 }
2821                 else
2822                     slice_max_size = 0;
2823             }
2824         }
2825 cont:
2826         h->mb.b_reencode_mb = 0;
2827
2828         /* save cache */
2829         x264_macroblock_cache_save( h );
2830
2831         if( x264_ratecontrol_mb( h, mb_size ) < 0 )
2832         {
2833             x264_bitstream_restore( h, &bs_bak[BS_BAK_ROW_VBV], &i_skip, 1 );
2834             h->mb.b_reencode_mb = 1;
2835             i_mb_x = 0;
2836             i_mb_y = i_mb_y - SLICE_MBAFF;
2837             h->mb.i_mb_prev_xy = i_mb_y * h->mb.i_mb_stride - 1;
2838             h->sh.i_last_mb = orig_last_mb;
2839             continue;
2840         }
2841
2842         /* accumulate mb stats */
2843         h->stat.frame.i_mb_count[h->mb.i_type]++;
2844
2845         int b_intra = IS_INTRA( h->mb.i_type );
2846         int b_skip = IS_SKIP( h->mb.i_type );
2847         if( h->param.i_log_level >= X264_LOG_INFO || h->param.rc.b_stat_write )
2848         {
2849             if( !b_intra && !b_skip && !IS_DIRECT( h->mb.i_type ) )
2850             {
2851                 if( h->mb.i_partition != D_8x8 )
2852                         h->stat.frame.i_mb_partition[h->mb.i_partition] += 4;
2853                     else
2854                         for( int i = 0; i < 4; i++ )
2855                             h->stat.frame.i_mb_partition[h->mb.i_sub_partition[i]] ++;
2856                 if( h->param.i_frame_reference > 1 )
2857                     for( int i_list = 0; i_list <= (h->sh.i_type == SLICE_TYPE_B); i_list++ )
2858                         for( int i = 0; i < 4; i++ )
2859                         {
2860                             int i_ref = h->mb.cache.ref[i_list][ x264_scan8[4*i] ];
2861                             if( i_ref >= 0 )
2862                                 h->stat.frame.i_mb_count_ref[i_list][i_ref] ++;
2863                         }
2864             }
2865         }
2866
2867         if( h->param.i_log_level >= X264_LOG_INFO )
2868         {
2869             if( h->mb.i_cbp_luma | h->mb.i_cbp_chroma )
2870             {
2871                 if( CHROMA444 )
2872                 {
2873                     for( int i = 0; i < 4; i++ )
2874                         if( h->mb.i_cbp_luma & (1 << i) )
2875                             for( int p = 0; p < 3; p++ )
2876                             {
2877                                 int s8 = i*4+p*16;
2878                                 int nnz8x8 = M16( &h->mb.cache.non_zero_count[x264_scan8[s8]+0] )
2879                                            | M16( &h->mb.cache.non_zero_count[x264_scan8[s8]+8] );
2880                                 h->stat.frame.i_mb_cbp[!b_intra + p*2] += !!nnz8x8;
2881                             }
2882                 }
2883                 else
2884                 {
2885                     int cbpsum = (h->mb.i_cbp_luma&1) + ((h->mb.i_cbp_luma>>1)&1)
2886                                + ((h->mb.i_cbp_luma>>2)&1) + (h->mb.i_cbp_luma>>3);
2887                     h->stat.frame.i_mb_cbp[!b_intra + 0] += cbpsum;
2888                     h->stat.frame.i_mb_cbp[!b_intra + 2] += !!h->mb.i_cbp_chroma;
2889                     h->stat.frame.i_mb_cbp[!b_intra + 4] += h->mb.i_cbp_chroma >> 1;
2890                 }
2891             }
2892             if( h->mb.i_cbp_luma && !b_intra )
2893             {
2894                 h->stat.frame.i_mb_count_8x8dct[0] ++;
2895                 h->stat.frame.i_mb_count_8x8dct[1] += h->mb.b_transform_8x8;
2896             }
2897             if( b_intra && h->mb.i_type != I_PCM )
2898             {
2899                 if( h->mb.i_type == I_16x16 )
2900                     h->stat.frame.i_mb_pred_mode[0][h->mb.i_intra16x16_pred_mode]++;
2901                 else if( h->mb.i_type == I_8x8 )
2902                     for( int i = 0; i < 16; i += 4 )
2903                         h->stat.frame.i_mb_pred_mode[1][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
2904                 else //if( h->mb.i_type == I_4x4 )
2905                     for( int i = 0; i < 16; i++ )
2906                         h->stat.frame.i_mb_pred_mode[2][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
2907                 h->stat.frame.i_mb_pred_mode[3][x264_mb_chroma_pred_mode_fix[h->mb.i_chroma_pred_mode]]++;
2908             }
2909             h->stat.frame.i_mb_field[b_intra?0:b_skip?2:1] += MB_INTERLACED;
2910         }
2911
2912         /* calculate deblock strength values (actual deblocking is done per-row along with hpel) */
2913         if( b_deblock )
2914             x264_macroblock_deblock_strength( h );
2915
2916         if( mb_xy == h->sh.i_last_mb )
2917             break;
2918
2919         if( SLICE_MBAFF )
2920         {
2921             i_mb_x += i_mb_y & 1;
2922             i_mb_y ^= i_mb_x < h->mb.i_mb_width;
2923         }
2924         else
2925             i_mb_x++;
2926         if( i_mb_x == h->mb.i_mb_width )
2927         {
2928             i_mb_y++;
2929             i_mb_x = 0;
2930         }
2931     }
2932     if( h->sh.i_last_mb < h->sh.i_first_mb )
2933         return 0;
2934
2935     h->out.nal[h->out.i_nal].i_last_mb = h->sh.i_last_mb;
2936
2937     if( h->param.b_cabac )
2938     {
2939         x264_cabac_encode_flush( h, &h->cabac );
2940         h->out.bs.p = h->cabac.p;
2941     }
2942     else
2943     {
2944         if( i_skip > 0 )
2945             bs_write_ue( &h->out.bs, i_skip );  /* last skip run */
2946         /* rbsp_slice_trailing_bits */
2947         bs_rbsp_trailing( &h->out.bs );
2948         bs_flush( &h->out.bs );
2949     }
2950     if( x264_nal_end( h ) )
2951         return -1;
2952
2953     if( h->sh.i_last_mb == (h->i_threadslice_end * h->mb.i_mb_width - 1) )
2954     {
2955         h->stat.frame.i_misc_bits = bs_pos( &h->out.bs )
2956                                   + (h->out.i_nal*NALU_OVERHEAD * 8)
2957                                   - h->stat.frame.i_tex_bits
2958                                   - h->stat.frame.i_mv_bits;
2959         x264_fdec_filter_row( h, h->i_threadslice_end, 0 );
2960
2961         if( h->param.b_sliced_threads )
2962         {
2963             /* Tell the main thread we're done. */
2964             x264_threadslice_cond_broadcast( h, 1 );
2965             /* Do hpel now */
2966             for( int mb_y = h->i_threadslice_start; mb_y <= h->i_threadslice_end; mb_y++ )
2967                 x264_fdec_filter_row( h, mb_y, 1 );
2968             x264_threadslice_cond_broadcast( h, 2 );
2969             /* Do the first row of hpel, now that the previous slice is done */
2970             if( h->i_thread_idx > 0 )
2971             {
2972                 x264_threadslice_cond_wait( h->thread[h->i_thread_idx-1], 2 );
2973                 x264_fdec_filter_row( h, h->i_threadslice_start + (1 << SLICE_MBAFF), 2 );
2974             }
2975         }
2976
2977         /* Free mb info after the last thread's done using it */
2978         if( h->fdec->mb_info_free && (!h->param.b_sliced_threads || h->i_thread_idx == (h->param.i_threads-1)) )
2979         {
2980             h->fdec->mb_info_free( h->fdec->mb_info );
2981             h->fdec->mb_info = NULL;
2982             h->fdec->mb_info_free = NULL;
2983         }
2984     }
2985
2986     return 0;
2987 }
2988
2989 static void x264_thread_sync_context( x264_t *dst, x264_t *src )
2990 {
2991     if( dst == src )
2992         return;
2993
2994     // reference counting
2995     for( x264_frame_t **f = src->frames.reference; *f; f++ )
2996         (*f)->i_reference_count++;
2997     for( x264_frame_t **f = dst->frames.reference; *f; f++ )
2998         x264_frame_push_unused( src, *f );
2999     src->fdec->i_reference_count++;
3000     x264_frame_push_unused( src, dst->fdec );
3001
3002     // copy everything except the per-thread pointers and the constants.
3003     memcpy( &dst->i_frame, &src->i_frame, offsetof(x264_t, mb.base) - offsetof(x264_t, i_frame) );
3004     dst->param = src->param;
3005     dst->stat = src->stat;
3006     dst->pixf = src->pixf;
3007     dst->reconfig = src->reconfig;
3008 }
3009
3010 static void x264_thread_sync_stat( x264_t *dst, x264_t *src )
3011 {
3012     if( dst == src )
3013         return;
3014     memcpy( &dst->stat.i_frame_count, &src->stat.i_frame_count, sizeof(dst->stat) - sizeof(dst->stat.frame) );
3015 }
3016
3017 static void *x264_slices_write( x264_t *h )
3018 {
3019     int i_slice_num = 0;
3020     int last_thread_mb = h->sh.i_last_mb;
3021
3022     /* init stats */
3023     memset( &h->stat.frame, 0, sizeof(h->stat.frame) );
3024     h->mb.b_reencode_mb = 0;
3025     while( h->sh.i_first_mb + SLICE_MBAFF*h->mb.i_mb_stride <= last_thread_mb )
3026     {
3027         h->sh.i_last_mb = last_thread_mb;
3028         if( !i_slice_num || !x264_frame_new_slice( h, h->fdec ) )
3029         {
3030             if( h->param.i_slice_max_mbs )
3031             {
3032                 if( SLICE_MBAFF )
3033                 {
3034                     // convert first to mbaff form, add slice-max-mbs, then convert back to normal form
3035                     int last_mbaff = 2*(h->sh.i_first_mb % h->mb.i_mb_width)
3036                         + h->mb.i_mb_width*(h->sh.i_first_mb / h->mb.i_mb_width)
3037                         + h->param.i_slice_max_mbs - 1;
3038                     int last_x = (last_mbaff % (2*h->mb.i_mb_width))/2;
3039                     int last_y = (last_mbaff / (2*h->mb.i_mb_width))*2 + 1;
3040                     h->sh.i_last_mb = last_x + h->mb.i_mb_stride*last_y;
3041                 }
3042                 else
3043                 {
3044                     h->sh.i_last_mb = h->sh.i_first_mb + h->param.i_slice_max_mbs - 1;
3045                     if( h->sh.i_last_mb < last_thread_mb && last_thread_mb - h->sh.i_last_mb < h->param.i_slice_min_mbs )
3046                         h->sh.i_last_mb = last_thread_mb - h->param.i_slice_min_mbs;
3047                 }
3048                 i_slice_num++;
3049             }
3050             else if( h->param.i_slice_count && !h->param.b_sliced_threads )
3051             {
3052                 int height = h->mb.i_mb_height >> PARAM_INTERLACED;
3053                 int width = h->mb.i_mb_width << PARAM_INTERLACED;
3054                 i_slice_num++;
3055                 h->sh.i_last_mb = (height * i_slice_num + h->param.i_slice_count/2) / h->param.i_slice_count * width - 1;
3056             }
3057         }
3058         h->sh.i_last_mb = X264_MIN( h->sh.i_last_mb, last_thread_mb );
3059         if( x264_stack_align( x264_slice_write, h ) )
3060             goto fail;
3061         h->sh.i_first_mb = h->sh.i_last_mb + 1;
3062         // if i_first_mb is not the last mb in a row then go to the next mb in MBAFF order
3063         if( SLICE_MBAFF && h->sh.i_first_mb % h->mb.i_mb_width )
3064             h->sh.i_first_mb -= h->mb.i_mb_stride;
3065     }
3066
3067     return (void *)0;
3068
3069 fail:
3070     /* Tell other threads we're done, so they wouldn't wait for it */
3071     if( h->param.b_sliced_threads )
3072         x264_threadslice_cond_broadcast( h, 2 );
3073     return (void *)-1;
3074 }
3075
3076 static int x264_threaded_slices_write( x264_t *h )
3077 {
3078     /* set first/last mb and sync contexts */
3079     for( int i = 0; i < h->param.i_threads; i++ )
3080     {
3081         x264_t *t = h->thread[i];
3082         if( i )
3083         {
3084             t->param = h->param;
3085             memcpy( &t->i_frame, &h->i_frame, offsetof(x264_t, rc) - offsetof(x264_t, i_frame) );
3086         }
3087         int height = h->mb.i_mb_height >> PARAM_INTERLACED;
3088         t->i_threadslice_start = ((height *  i    + h->param.i_slice_count/2) / h->param.i_threads) << PARAM_INTERLACED;
3089         t->i_threadslice_end   = ((height * (i+1) + h->param.i_slice_count/2) / h->param.i_threads) << PARAM_INTERLACED;
3090         t->sh.i_first_mb = t->i_threadslice_start * h->mb.i_mb_width;
3091         t->sh.i_last_mb  =   t->i_threadslice_end * h->mb.i_mb_width - 1;
3092     }
3093
3094     x264_stack_align( x264_analyse_weight_frame, h, h->mb.i_mb_height*16 + 16 );
3095
3096     x264_threads_distribute_ratecontrol( h );
3097
3098     /* setup */
3099     for( int i = 0; i < h->param.i_threads; i++ )
3100     {
3101         h->thread[i]->i_thread_idx = i;
3102         h->thread[i]->b_thread_active = 1;
3103         x264_threadslice_cond_broadcast( h->thread[i], 0 );
3104     }
3105     /* dispatch */
3106     for( int i = 0; i < h->param.i_threads; i++ )
3107         x264_threadpool_run( h->threadpool, (void*)x264_slices_write, h->thread[i] );
3108     /* wait */
3109     for( int i = 0; i < h->param.i_threads; i++ )
3110         x264_threadslice_cond_wait( h->thread[i], 1 );
3111
3112     x264_threads_merge_ratecontrol( h );
3113
3114     for( int i = 1; i < h->param.i_threads; i++ )
3115     {
3116         x264_t *t = h->thread[i];
3117         for( int j = 0; j < t->out.i_nal; j++ )
3118         {
3119             h->out.nal[h->out.i_nal] = t->out.nal[j];
3120             h->out.i_nal++;
3121             x264_nal_check_buffer( h );
3122         }
3123         /* All entries in stat.frame are ints except for ssd/ssim. */
3124         for( int j = 0; j < (offsetof(x264_t,stat.frame.i_ssd) - offsetof(x264_t,stat.frame.i_mv_bits)) / sizeof(int); j++ )
3125             ((int*)&h->stat.frame)[j] += ((int*)&t->stat.frame)[j];
3126         for( int j = 0; j < 3; j++ )
3127             h->stat.frame.i_ssd[j] += t->stat.frame.i_ssd[j];
3128         h->stat.frame.f_ssim += t->stat.frame.f_ssim;
3129         h->stat.frame.i_ssim_cnt += t->stat.frame.i_ssim_cnt;
3130     }
3131
3132     return 0;
3133 }
3134
3135 void x264_encoder_intra_refresh( x264_t *h )
3136 {
3137     h = h->thread[h->i_thread_phase];
3138     h->b_queued_intra_refresh = 1;
3139 }
3140
3141 int x264_encoder_invalidate_reference( x264_t *h, int64_t pts )
3142 {
3143     if( h->param.i_bframe )
3144     {
3145         x264_log( h, X264_LOG_ERROR, "x264_encoder_invalidate_reference is not supported with B-frames enabled\n" );
3146         return -1;
3147     }
3148     if( h->param.b_intra_refresh )
3149     {
3150         x264_log( h, X264_LOG_ERROR, "x264_encoder_invalidate_reference is not supported with intra refresh enabled\n" );
3151         return -1;
3152     }
3153     h = h->thread[h->i_thread_phase];
3154     if( pts >= h->i_last_idr_pts )
3155     {
3156         for( int i = 0; h->frames.reference[i]; i++ )
3157             if( pts <= h->frames.reference[i]->i_pts )
3158                 h->frames.reference[i]->b_corrupt = 1;
3159         if( pts <= h->fdec->i_pts )
3160             h->fdec->b_corrupt = 1;
3161     }
3162     return 0;
3163 }
3164
3165 /****************************************************************************
3166  * x264_encoder_encode:
3167  *  XXX: i_poc   : is the poc of the current given picture
3168  *       i_frame : is the number of the frame being coded
3169  *  ex:  type frame poc
3170  *       I      0   2*0
3171  *       P      1   2*3
3172  *       B      2   2*1
3173  *       B      3   2*2
3174  *       P      4   2*6
3175  *       B      5   2*4
3176  *       B      6   2*5
3177  ****************************************************************************/
3178 int     x264_encoder_encode( x264_t *h,
3179                              x264_nal_t **pp_nal, int *pi_nal,
3180                              x264_picture_t *pic_in,
3181                              x264_picture_t *pic_out )
3182 {
3183     x264_t *thread_current, *thread_prev, *thread_oldest;
3184     int i_nal_type, i_nal_ref_idc, i_global_qp;
3185     int overhead = NALU_OVERHEAD;
3186
3187 #if HAVE_OPENCL
3188     if( h->opencl.b_fatal_error )
3189         return -1;
3190 #endif
3191
3192     if( h->i_thread_frames > 1 )
3193     {
3194         thread_prev    = h->thread[ h->i_thread_phase ];
3195         h->i_thread_phase = (h->i_thread_phase + 1) % h->i_thread_frames;
3196         thread_current = h->thread[ h->i_thread_phase ];
3197         thread_oldest  = h->thread[ (h->i_thread_phase + 1) % h->i_thread_frames ];
3198         x264_thread_sync_context( thread_current, thread_prev );
3199         x264_thread_sync_ratecontrol( thread_current, thread_prev, thread_oldest );
3200         h = thread_current;
3201     }
3202     else
3203     {
3204         thread_current =
3205         thread_oldest  = h;
3206     }
3207     h->i_cpb_delay_pir_offset = h->i_cpb_delay_pir_offset_next;
3208
3209     /* no data out */
3210     *pi_nal = 0;
3211     *pp_nal = NULL;
3212
3213     /* ------------------- Setup new frame from picture -------------------- */
3214     if( pic_in != NULL )
3215     {
3216         /* 1: Copy the picture to a frame and move it to a buffer */
3217         x264_frame_t *fenc = x264_frame_pop_unused( h, 0 );
3218         if( !fenc )
3219             return -1;
3220
3221         if( x264_frame_copy_picture( h, fenc, pic_in ) < 0 )
3222             return -1;
3223
3224         if( h->param.i_width != 16 * h->mb.i_mb_width ||
3225             h->param.i_height != 16 * h->mb.i_mb_height )
3226             x264_frame_expand_border_mod16( h, fenc );
3227
3228         fenc->i_frame = h->frames.i_input++;
3229
3230         if( fenc->i_frame == 0 )
3231             h->frames.i_first_pts = fenc->i_pts;
3232         if( h->frames.i_bframe_delay && fenc->i_frame == h->frames.i_bframe_delay )
3233             h->frames.i_bframe_delay_time = fenc->i_pts - h->frames.i_first_pts;
3234
3235         if( h->param.b_vfr_input && fenc->i_pts <= h->frames.i_largest_pts )
3236             x264_log( h, X264_LOG_WARNING, "non-strictly-monotonic PTS\n" );
3237
3238         h->frames.i_second_largest_pts = h->frames.i_largest_pts;
3239         h->frames.i_largest_pts = fenc->i_pts;
3240
3241         if( (fenc->i_pic_struct < PIC_STRUCT_AUTO) || (fenc->i_pic_struct > PIC_STRUCT_TRIPLE) )
3242             fenc->i_pic_struct = PIC_STRUCT_AUTO;
3243
3244         if( fenc->i_pic_struct == PIC_STRUCT_AUTO )
3245         {
3246 #if HAVE_INTERLACED
3247             int b_interlaced = fenc->param ? fenc->param->b_interlaced : h->param.b_interlaced;
3248 #else
3249             int b_interlaced = 0;
3250 #endif
3251             if( b_interlaced )
3252             {
3253                 int b_tff = fenc->param ? fenc->param->b_tff : h->param.b_tff;
3254                 fenc->i_pic_struct = b_tff ? PIC_STRUCT_TOP_BOTTOM : PIC_STRUCT_BOTTOM_TOP;
3255             }
3256             else
3257                 fenc->i_pic_struct = PIC_STRUCT_PROGRESSIVE;
3258         }
3259
3260         if( h->param.rc.b_mb_tree && h->param.rc.b_stat_read )
3261         {
3262             if( x264_macroblock_tree_read( h, fenc, pic_in->prop.quant_offsets ) )
3263                 return -1;
3264         }
3265         else
3266             x264_stack_align( x264_adaptive_quant_frame, h, fenc, pic_in->prop.quant_offsets );
3267
3268         if( pic_in->prop.quant_offsets_free )
3269             pic_in->prop.quant_offsets_free( pic_in->prop.quant_offsets );
3270
3271         if( h->frames.b_have_lowres )
3272             x264_frame_init_lowres( h, fenc );
3273
3274         /* 2: Place the frame into the queue for its slice type decision */
3275         x264_lookahead_put_frame( h, fenc );
3276
3277         if( h->frames.i_input <= h->frames.i_delay + 1 - h->i_thread_frames )
3278         {
3279             /* Nothing yet to encode, waiting for filling of buffers */
3280             pic_out->i_type = X264_TYPE_AUTO;
3281             return 0;
3282         }
3283     }
3284     else
3285     {
3286         /* signal kills for lookahead thread */
3287         x264_pthread_mutex_lock( &h->lookahead->ifbuf.mutex );
3288         h->lookahead->b_exit_thread = 1;
3289         x264_pthread_cond_broadcast( &h->lookahead->ifbuf.cv_fill );
3290         x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
3291     }
3292
3293     h->i_frame++;
3294     /* 3: The picture is analyzed in the lookahead */
3295     if( !h->frames.current[0] )
3296         x264_lookahead_get_frames( h );
3297
3298     if( !h->frames.current[0] && x264_lookahead_is_empty( h ) )
3299         return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
3300
3301     /* ------------------- Get frame to be encoded ------------------------- */
3302     /* 4: get picture to encode */
3303     h->fenc = x264_frame_shift( h->frames.current );
3304
3305     /* If applicable, wait for previous frame reconstruction to finish */
3306     if( h->param.b_sliced_threads )
3307         if( x264_threadpool_wait_all( h ) < 0 )
3308             return -1;
3309
3310     if( h->i_frame == h->i_thread_frames - 1 )
3311         h->i_reordered_pts_delay = h->fenc->i_reordered_pts;
3312     if( h->reconfig )
3313     {
3314         x264_encoder_reconfig_apply( h, &h->reconfig_h->param );
3315         h->reconfig = 0;
3316     }
3317     if( h->fenc->param )
3318     {
3319         x264_encoder_reconfig_apply( h, h->fenc->param );
3320         if( h->fenc->param->param_free )
3321         {
3322             h->fenc->param->param_free( h->fenc->param );
3323             h->fenc->param = NULL;
3324         }
3325     }
3326
3327     // ok to call this before encoding any frames, since the initial values of fdec have b_kept_as_ref=0
3328     if( x264_reference_update( h ) )
3329         return -1;
3330     h->fdec->i_lines_completed = -1;
3331
3332     if( !IS_X264_TYPE_I( h->fenc->i_type ) )
3333     {
3334         int valid_refs_left = 0;
3335         for( int i = 0; h->frames.reference[i]; i++ )
3336             if( !h->frames.reference[i]->b_corrupt )
3337                 valid_refs_left++;
3338         /* No valid reference frames left: force an IDR. */
3339         if( !valid_refs_left )
3340         {
3341             h->fenc->b_keyframe = 1;
3342             h->fenc->i_type = X264_TYPE_IDR;
3343         }
3344     }
3345
3346     if( h->fenc->b_keyframe )
3347     {
3348         h->frames.i_last_keyframe = h->fenc->i_frame;
3349         if( h->fenc->i_type == X264_TYPE_IDR )
3350         {
3351             h->i_frame_num = 0;
3352             h->frames.i_last_idr = h->fenc->i_frame;
3353         }
3354     }
3355     h->sh.i_mmco_command_count =
3356     h->sh.i_mmco_remove_from_end = 0;
3357     h->b_ref_reorder[0] =
3358     h->b_ref_reorder[1] = 0;
3359     h->fdec->i_poc =
3360     h->fenc->i_poc = 2 * ( h->fenc->i_frame - X264_MAX( h->frames.i_last_idr, 0 ) );
3361
3362     /* ------------------- Setup frame context ----------------------------- */
3363     /* 5: Init data dependent of frame type */
3364     if( h->fenc->i_type == X264_TYPE_IDR )
3365     {
3366         /* reset ref pictures */
3367         i_nal_type    = NAL_SLICE_IDR;
3368         i_nal_ref_idc = NAL_PRIORITY_HIGHEST;
3369         h->sh.i_type = SLICE_TYPE_I;
3370         x264_reference_reset( h );
3371         h->frames.i_poc_last_open_gop = -1;
3372     }
3373     else if( h->fenc->i_type == X264_TYPE_I )
3374     {
3375         i_nal_type    = NAL_SLICE;
3376         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
3377         h->sh.i_type = SLICE_TYPE_I;
3378         x264_reference_hierarchy_reset( h );
3379         if( h->param.b_open_gop )
3380             h->frames.i_poc_last_open_gop = h->fenc->b_keyframe ? h->fenc->i_poc : -1;
3381     }
3382     else if( h->fenc->i_type == X264_TYPE_P )
3383     {
3384         i_nal_type    = NAL_SLICE;
3385         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
3386         h->sh.i_type = SLICE_TYPE_P;
3387         x264_reference_hierarchy_reset( h );
3388         h->frames.i_poc_last_open_gop = -1;
3389     }
3390     else if( h->fenc->i_type == X264_TYPE_BREF )
3391     {
3392         i_nal_type    = NAL_SLICE;
3393         i_nal_ref_idc = h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT ? NAL_PRIORITY_LOW : NAL_PRIORITY_HIGH;
3394         h->sh.i_type = SLICE_TYPE_B;
3395         x264_reference_hierarchy_reset( h );
3396     }
3397     else    /* B frame */
3398     {
3399         i_nal_type    = NAL_SLICE;
3400         i_nal_ref_idc = NAL_PRIORITY_DISPOSABLE;
3401         h->sh.i_type = SLICE_TYPE_B;
3402     }
3403
3404     h->fdec->i_type = h->fenc->i_type;
3405     h->fdec->i_frame = h->fenc->i_frame;
3406     h->fenc->b_kept_as_ref =
3407     h->fdec->b_kept_as_ref = i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE && h->param.i_keyint_max > 1;
3408
3409     h->fdec->mb_info = h->fenc->mb_info;
3410     h->fdec->mb_info_free = h->fenc->mb_info_free;
3411     h->fenc->mb_info = NULL;
3412     h->fenc->mb_info_free = NULL;
3413
3414     h->fdec->i_pts = h->fenc->i_pts;
3415     if( h->frames.i_bframe_delay )
3416     {
3417         int64_t *prev_reordered_pts = thread_current->frames.i_prev_reordered_pts;
3418         h->fdec->i_dts = h->i_frame > h->frames.i_bframe_delay
3419                        ? prev_reordered_pts[ (h->i_frame - h->frames.i_bframe_delay) % h->frames.i_bframe_delay ]
3420                        : h->fenc->i_reordered_pts - h->frames.i_bframe_delay_time;
3421         prev_reordered_pts[ h->i_frame % h->frames.i_bframe_delay ] = h->fenc->i_reordered_pts;
3422     }
3423     else
3424         h->fdec->i_dts = h->fenc->i_reordered_pts;
3425     if( h->fenc->i_type == X264_TYPE_IDR )
3426         h->i_last_idr_pts = h->fdec->i_pts;
3427
3428     /* ------------------- Init                ----------------------------- */
3429     /* build ref list 0/1 */
3430     x264_reference_build_list( h, h->fdec->i_poc );
3431
3432     /* ---------------------- Write the bitstream -------------------------- */
3433     /* Init bitstream context */
3434     if( h->param.b_sliced_threads )
3435     {
3436         for( int i = 0; i < h->param.i_threads; i++ )
3437         {
3438             bs_init( &h->thread[i]->out.bs, h->thread[i]->out.p_bitstream, h->thread[i]->out.i_bitstream );
3439             h->thread[i]->out.i_nal = 0;
3440         }
3441     }
3442     else
3443     {
3444         bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
3445         h->out.i_nal = 0;
3446     }
3447
3448     if( h->param.b_aud )
3449     {
3450         int pic_type;
3451
3452         if( h->sh.i_type == SLICE_TYPE_I )
3453             pic_type = 0;
3454         else if( h->sh.i_type == SLICE_TYPE_P )
3455             pic_type = 1;
3456         else if( h->sh.i_type == SLICE_TYPE_B )
3457             pic_type = 2;
3458         else
3459             pic_type = 7;
3460
3461         x264_nal_start( h, NAL_AUD, NAL_PRIORITY_DISPOSABLE );
3462         bs_write( &h->out.bs, 3, pic_type );
3463         bs_rbsp_trailing( &h->out.bs );
3464         if( x264_nal_end( h ) )
3465             return -1;
3466         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
3467     }
3468
3469     h->i_nal_type = i_nal_type;
3470     h->i_nal_ref_idc = i_nal_ref_idc;
3471
3472     if( h->param.b_intra_refresh )
3473     {
3474         if( IS_X264_TYPE_I( h->fenc->i_type ) )
3475         {
3476             h->fdec->i_frames_since_pir = 0;
3477             h->b_queued_intra_refresh = 0;
3478             /* PIR is currently only supported with ref == 1, so any intra frame effectively refreshes
3479              * the whole frame and counts as an intra refresh. */
3480             h->fdec->f_pir_position = h->mb.i_mb_width;
3481         }
3482         else if( h->fenc->i_type == X264_TYPE_P )
3483         {
3484             int pocdiff = (h->fdec->i_poc - h->fref[0][0]->i_poc)/2;
3485             float increment = X264_MAX( ((float)h->mb.i_mb_width-1) / h->param.i_keyint_max, 1 );
3486             h->fdec->f_pir_position = h->fref[0][0]->f_pir_position;
3487             h->fdec->i_frames_since_pir = h->fref[0][0]->i_frames_since_pir + pocdiff;
3488             if( h->fdec->i_frames_since_pir >= h->param.i_keyint_max ||
3489                 (h->b_queued_intra_refresh && h->fdec->f_pir_position + 0.5 >= h->mb.i_mb_width) )
3490             {
3491                 h->fdec->f_pir_position = 0;
3492                 h->fdec->i_frames_since_pir = 0;
3493                 h->b_queued_intra_refresh = 0;
3494                 h->fenc->b_keyframe = 1;
3495             }
3496             h->fdec->i_pir_start_col = h->fdec->f_pir_position+0.5;
3497             h->fdec->f_pir_position += increment * pocdiff;
3498             h->fdec->i_pir_end_col = h->fdec->f_pir_position+0.5;
3499             /* If our intra refresh has reached the right side of the frame, we're done. */
3500             if( h->fdec->i_pir_end_col >= h->mb.i_mb_width - 1 )
3501             {
3502                 h->fdec->f_pir_position = h->mb.i_mb_width;
3503                 h->fdec->i_pir_end_col = h->mb.i_mb_width - 1;
3504             }
3505         }
3506     }
3507
3508     if( h->fenc->b_keyframe )
3509     {
3510         /* Write SPS and PPS */
3511         if( h->param.b_repeat_headers )
3512         {
3513             /* generate sequence parameters */
3514             x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
3515             x264_sps_write( &h->out.bs, h->sps );
3516             if( x264_nal_end( h ) )
3517                 return -1;
3518             /* Pad AUD/SPS to 256 bytes like Panasonic */
3519             if( h->param.i_avcintra_class )
3520                 h->out.nal[h->out.i_nal-1].i_padding = 256 - bs_pos( &h->out.bs ) / 8 - 2*NALU_OVERHEAD;
3521             overhead += h->out.nal[h->out.i_nal-1].i_payload + h->out.nal[h->out.i_nal-1].i_padding + NALU_OVERHEAD;
3522
3523             /* generate picture parameters */
3524             x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
3525             x264_pps_write( &h->out.bs, h->sps, h->pps );
3526             if( x264_nal_end( h ) )
3527                 return -1;
3528             if( h->param.i_avcintra_class )
3529                 h->out.nal[h->out.i_nal-1].i_padding = 256 - h->out.nal[h->out.i_nal-1].i_payload - NALU_OVERHEAD;
3530             overhead += h->out.nal[h->out.i_nal-1].i_payload + h->out.nal[h->out.i_nal-1].i_padding + NALU_OVERHEAD;
3531         }
3532
3533         /* when frame threading is used, buffering period sei is written in x264_encoder_frame_end */
3534         if( h->i_thread_frames == 1 && h->sps->vui.b_nal_hrd_parameters_present )
3535         {
3536             x264_hrd_fullness( h );
3537             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3538             x264_sei_buffering_period_write( h, &h->out.bs );
3539             if( x264_nal_end( h ) )
3540                return -1;
3541             overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3542         }
3543     }
3544
3545     /* write extra sei */
3546     for( int i = 0; i < h->fenc->extra_sei.num_payloads; i++ )
3547     {
3548         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3549         x264_sei_write( &h->out.bs, h->fenc->extra_sei.payloads[i].payload, h->fenc->extra_sei.payloads[i].payload_size,
3550                         h->fenc->extra_sei.payloads[i].payload_type );
3551         if( x264_nal_end( h ) )
3552             return -1;
3553         overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3554         if( h->fenc->extra_sei.sei_free )
3555         {
3556             h->fenc->extra_sei.sei_free( h->fenc->extra_sei.payloads[i].payload );
3557             h->fenc->extra_sei.payloads[i].payload = NULL;
3558         }
3559     }
3560
3561     if( h->fenc->extra_sei.sei_free )
3562     {
3563         h->fenc->extra_sei.sei_free( h->fenc->extra_sei.payloads );
3564         h->fenc->extra_sei.payloads = NULL;
3565         h->fenc->extra_sei.sei_free = NULL;
3566     }
3567
3568     if( h->fenc->b_keyframe )
3569     {
3570         /* Avid's decoder strictly wants two SEIs for AVC-Intra so we can't insert the x264 SEI */
3571         if( h->param.b_repeat_headers && h->fenc->i_frame == 0 && !h->param.i_avcintra_class )
3572         {
3573             /* identify ourself */
3574             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3575             if( x264_sei_version_write( h, &h->out.bs ) )
3576                 return -1;
3577             if( x264_nal_end( h ) )
3578                 return -1;
3579             overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3580         }
3581
3582         if( h->fenc->i_type != X264_TYPE_IDR )
3583         {
3584             int time_to_recovery = h->param.b_open_gop ? 0 : X264_MIN( h->mb.i_mb_width - 1, h->param.i_keyint_max ) + h->param.i_bframe - 1;
3585             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3586             x264_sei_recovery_point_write( h, &h->out.bs, time_to_recovery );
3587             if( x264_nal_end( h ) )
3588                 return -1;
3589             overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3590         }
3591     }
3592
3593     if( h->param.i_frame_packing >= 0 && (h->fenc->b_keyframe || h->param.i_frame_packing == 5) )
3594     {
3595         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3596         x264_sei_frame_packing_write( h, &h->out.bs );
3597         if( x264_nal_end( h ) )
3598             return -1;
3599         overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3600     }
3601
3602     /* generate sei pic timing */
3603     if( h->sps->vui.b_pic_struct_present || h->sps->vui.b_nal_hrd_parameters_present )
3604     {
3605         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3606         x264_sei_pic_timing_write( h, &h->out.bs );
3607         if( x264_nal_end( h ) )
3608             return -1;
3609         overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3610     }
3611
3612     /* As required by Blu-ray. */
3613     if( !IS_X264_TYPE_B( h->fenc->i_type ) && h->b_sh_backup )
3614     {
3615         h->b_sh_backup = 0;
3616         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3617         x264_sei_dec_ref_pic_marking_write( h, &h->out.bs );
3618         if( x264_nal_end( h ) )
3619             return -1;
3620         overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3621     }
3622
3623     if( h->fenc->b_keyframe && h->param.b_intra_refresh )
3624         h->i_cpb_delay_pir_offset_next = h->fenc->i_cpb_delay;
3625
3626     /* Filler space: 10 or 18 SEIs' worth of space, depending on resolution */
3627     if( h->param.i_avcintra_class )
3628     {
3629         /* Write an empty filler NAL to mimic the AUD in the P2 format*/
3630         x264_nal_start( h, NAL_FILLER, NAL_PRIORITY_DISPOSABLE );
3631         x264_filler_write( h, &h->out.bs, 0 );
3632         if( x264_nal_end( h ) )
3633             return -1;
3634         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
3635
3636         /* All lengths are magic lengths that decoders expect to see */
3637         /* "UMID" SEI */
3638         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3639         if( x264_sei_avcintra_umid_write( h, &h->out.bs ) < 0 )
3640             return -1;
3641         if( x264_nal_end( h ) )
3642             return -1;
3643         overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3644
3645         int unpadded_len;
3646         int total_len;
3647         if( h->param.i_height == 1080 )
3648         {
3649             unpadded_len = 5780;
3650             total_len = 17*512;
3651         }
3652         else
3653         {
3654             unpadded_len = 2900;
3655             total_len = 9*512;
3656         }
3657         /* "VANC" SEI */
3658         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3659         if( x264_sei_avcintra_vanc_write( h, &h->out.bs, unpadded_len ) < 0 )
3660             return -1;
3661         if( x264_nal_end( h ) )
3662             return -1;
3663
3664         h->out.nal[h->out.i_nal-1].i_padding = total_len - h->out.nal[h->out.i_nal-1].i_payload - SEI_OVERHEAD;
3665         overhead += h->out.nal[h->out.i_nal-1].i_payload + h->out.nal[h->out.i_nal-1].i_padding + SEI_OVERHEAD;
3666     }
3667
3668     /* Init the rate control */
3669     /* FIXME: Include slice header bit cost. */
3670     x264_ratecontrol_start( h, h->fenc->i_qpplus1, overhead*8 );
3671     i_global_qp = x264_ratecontrol_qp( h );
3672
3673     pic_out->i_qpplus1 =
3674     h->fdec->i_qpplus1 = i_global_qp + 1;
3675
3676     if( h->param.rc.b_stat_read && h->sh.i_type != SLICE_TYPE_I )
3677     {
3678         x264_reference_build_list_optimal( h );
3679         x264_reference_check_reorder( h );
3680     }
3681
3682     if( h->i_ref[0] )
3683         h->fdec->i_poc_l0ref0 = h->fref[0][0]->i_poc;
3684
3685     /* ------------------------ Create slice header  ----------------------- */
3686     x264_slice_init( h, i_nal_type, i_global_qp );
3687
3688     /*------------------------- Weights -------------------------------------*/
3689     if( h->sh.i_type == SLICE_TYPE_B )
3690         x264_macroblock_bipred_init( h );
3691
3692     x264_weighted_pred_init( h );
3693
3694     if( i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE )
3695         h->i_frame_num++;
3696
3697     /* Write frame */
3698     h->i_threadslice_start = 0;
3699     h->i_threadslice_end = h->mb.i_mb_height;
3700     if( h->i_thread_frames > 1 )
3701     {
3702         x264_threadpool_run( h->threadpool, (void*)x264_slices_write, h );
3703         h->b_thread_active = 1;
3704     }
3705     else if( h->param.b_sliced_threads )
3706     {
3707         if( x264_threaded_slices_write( h ) )
3708             return -1;
3709     }
3710     else
3711         if( (intptr_t)x264_slices_write( h ) )
3712             return -1;
3713
3714     return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
3715 }
3716
3717 static int x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
3718                                    x264_nal_t **pp_nal, int *pi_nal,
3719                                    x264_picture_t *pic_out )
3720 {
3721     char psz_message[80];
3722
3723     if( !h->param.b_sliced_threads && h->b_thread_active )
3724     {
3725         h->b_thread_active = 0;
3726         if( (intptr_t)x264_threadpool_wait( h->threadpool, h ) )
3727             return -1;
3728     }
3729     if( !h->out.i_nal )
3730     {
3731         pic_out->i_type = X264_TYPE_AUTO;
3732         return 0;
3733     }
3734
3735     x264_emms();
3736
3737     /* generate buffering period sei and insert it into place */
3738     if( h->i_thread_frames > 1 && h->fenc->b_keyframe && h->sps->vui.b_nal_hrd_parameters_present )
3739     {
3740         x264_hrd_fullness( h );
3741         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3742         x264_sei_buffering_period_write( h, &h->out.bs );
3743         if( x264_nal_end( h ) )
3744            return -1;
3745         /* buffering period sei must follow AUD, SPS and PPS and precede all other SEIs */
3746         int idx = 0;
3747         while( h->out.nal[idx].i_type == NAL_AUD ||
3748                h->out.nal[idx].i_type == NAL_SPS ||
3749                h->out.nal[idx].i_type == NAL_PPS )
3750             idx++;
3751         x264_nal_t nal_tmp = h->out.nal[h->out.i_nal-1];
3752         memmove( &h->out.nal[idx+1], &h->out.nal[idx], (h->out.i_nal-idx-1)*sizeof(x264_nal_t) );
3753         h->out.nal[idx] = nal_tmp;
3754     }
3755
3756     int frame_size = x264_encoder_encapsulate_nals( h, 0 );
3757     if( frame_size < 0 )
3758         return -1;
3759
3760     /* Set output picture properties */
3761     pic_out->i_type = h->fenc->i_type;
3762
3763     pic_out->b_keyframe = h->fenc->b_keyframe;
3764     pic_out->i_pic_struct = h->fenc->i_pic_struct;
3765
3766     pic_out->i_pts = h->fdec->i_pts;
3767     pic_out->i_dts = h->fdec->i_dts;
3768
3769     if( pic_out->i_pts < pic_out->i_dts )
3770         x264_log( h, X264_LOG_WARNING, "invalid DTS: PTS is less than DTS\n" );
3771
3772     pic_out->opaque = h->fenc->opaque;
3773
3774     pic_out->img.i_csp = h->fdec->i_csp;
3775 #if HIGH_BIT_DEPTH
3776     pic_out->img.i_csp |= X264_CSP_HIGH_DEPTH;
3777 #endif
3778     pic_out->img.i_plane = h->fdec->i_plane;
3779     for( int i = 0; i < pic_out->img.i_plane; i++ )
3780     {
3781         pic_out->img.i_stride[i] = h->fdec->i_stride[i] * sizeof(pixel);
3782         pic_out->img.plane[i] = (uint8_t*)h->fdec->plane[i];
3783     }
3784
3785     x264_frame_push_unused( thread_current, h->fenc );
3786
3787     /* ---------------------- Update encoder state ------------------------- */
3788
3789     /* update rc */
3790     int filler = 0;
3791     if( x264_ratecontrol_end( h, frame_size * 8, &filler ) < 0 )
3792         return -1;
3793
3794     pic_out->hrd_timing = h->fenc->hrd_timing;
3795     pic_out->prop.f_crf_avg = h->fdec->f_crf_avg;
3796
3797     /* Filler in AVC-Intra mode is written as zero bytes to the last slice
3798      * We don't know the size of the last slice until encapsulation so we add filler to the encapsulated NAL */
3799     if( h->param.i_avcintra_class )
3800     {
3801         x264_t *h0 = h->thread[0];
3802         int ret = x264_check_encapsulated_buffer( h, h0, h->out.i_nal, frame_size, frame_size + filler );
3803         if( ret < 0 )
3804             return -1;
3805         memset( h->out.nal[0].p_payload + frame_size, 0, filler );
3806         h->out.nal[h->out.i_nal-1].i_payload += filler;
3807         h->out.nal[h->out.i_nal-1].i_padding = filler;
3808         frame_size += filler;
3809     }
3810     else
3811     {
3812         while( filler > 0 )
3813         {
3814             int f, overhead;
3815             overhead = (FILLER_OVERHEAD - h->param.b_annexb);
3816             if( h->param.i_slice_max_size && filler > h->param.i_slice_max_size )
3817             {
3818                 int next_size = filler - h->param.i_slice_max_size;
3819                 int overflow = X264_MAX( overhead - next_size, 0 );
3820                 f = h->param.i_slice_max_size - overhead - overflow;
3821             }
3822             else
3823                 f = X264_MAX( 0, filler - overhead );
3824
3825             if( x264_bitstream_check_buffer_filler( h, f ) )
3826                 return -1;
3827             x264_nal_start( h, NAL_FILLER, NAL_PRIORITY_DISPOSABLE );
3828             x264_filler_write( h, &h->out.bs, f );
3829             if( x264_nal_end( h ) )
3830                 return -1;
3831             int total_size = x264_encoder_encapsulate_nals( h, h->out.i_nal-1 );
3832             if( total_size < 0 )
3833                 return -1;
3834             frame_size += total_size;
3835             filler -= total_size;
3836         }
3837     }
3838
3839     /* End bitstream, set output  */
3840     *pi_nal = h->out.i_nal;
3841     *pp_nal = h->out.nal;
3842
3843     h->out.i_nal = 0;
3844
3845     x264_noise_reduction_update( h );
3846
3847     /* ---------------------- Compute/Print statistics --------------------- */
3848     x264_thread_sync_stat( h, h->thread[0] );
3849
3850     /* Slice stat */
3851     h->stat.i_frame_count[h->sh.i_type]++;
3852     h->stat.i_frame_size[h->sh.i_type] += frame_size;
3853     h->stat.f_frame_qp[h->sh.i_type] += h->fdec->f_qp_avg_aq;
3854
3855     for( int i = 0; i < X264_MBTYPE_MAX; i++ )
3856         h->stat.i_mb_count[h->sh.i_type][i] += h->stat.frame.i_mb_count[i];
3857     for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
3858         h->stat.i_mb_partition[h->sh.i_type][i] += h->stat.frame.i_mb_partition[i];
3859     for( int i = 0; i < 2; i++ )
3860         h->stat.i_mb_count_8x8dct[i] += h->stat.frame.i_mb_count_8x8dct[i];
3861     for( int i = 0; i < 6; i++ )
3862         h->stat.i_mb_cbp[i] += h->stat.frame.i_mb_cbp[i];
3863     for( int i = 0; i < 4; i++ )
3864         for( int j = 0; j < 13; j++ )
3865             h->stat.i_mb_pred_mode[i][j] += h->stat.frame.i_mb_pred_mode[i][j];
3866     if( h->sh.i_type != SLICE_TYPE_I )
3867         for( int i_list = 0; i_list < 2; i_list++ )
3868             for( int i = 0; i < X264_REF_MAX*2; i++ )
3869                 h->stat.i_mb_count_ref[h->sh.i_type][i_list][i] += h->stat.frame.i_mb_count_ref[i_list][i];
3870     for( int i = 0; i < 3; i++ )
3871         h->stat.i_mb_field[i] += h->stat.frame.i_mb_field[i];
3872     if( h->sh.i_type == SLICE_TYPE_P && h->param.analyse.i_weighted_pred >= X264_WEIGHTP_SIMPLE )
3873     {
3874         h->stat.i_wpred[0] += !!h->sh.weight[0][0].weightfn;
3875         h->stat.i_wpred[1] += !!h->sh.weight[0][1].weightfn || !!h->sh.weight[0][2].weightfn;
3876     }
3877     if( h->sh.i_type == SLICE_TYPE_B )
3878     {
3879         h->stat.i_direct_frames[ h->sh.b_direct_spatial_mv_pred ] ++;
3880         if( h->mb.b_direct_auto_write )
3881         {
3882             //FIXME somewhat arbitrary time constants
3883             if( h->stat.i_direct_score[0] + h->stat.i_direct_score[1] > h->mb.i_mb_count )
3884                 for( int i = 0; i < 2; i++ )
3885                     h->stat.i_direct_score[i] = h->stat.i_direct_score[i] * 9/10;
3886             for( int i = 0; i < 2; i++ )
3887                 h->stat.i_direct_score[i] += h->stat.frame.i_direct_score[i];
3888         }
3889     }
3890     else
3891         h->stat.i_consecutive_bframes[h->fenc->i_bframes]++;
3892
3893     psz_message[0] = '\0';
3894     double dur = h->fenc->f_duration;
3895     h->stat.f_frame_duration[h->sh.i_type] += dur;
3896     if( h->param.analyse.b_psnr )
3897     {
3898         int64_t ssd[3] =
3899         {
3900             h->stat.frame.i_ssd[0],
3901             h->stat.frame.i_ssd[1],
3902             h->stat.frame.i_ssd[2],
3903         };
3904         int luma_size = h->param.i_width * h->param.i_height;
3905         int chroma_size = CHROMA_SIZE( luma_size );
3906         pic_out->prop.f_psnr[0] = x264_psnr( ssd[0], luma_size );
3907         pic_out->prop.f_psnr[1] = x264_psnr( ssd[1], chroma_size );
3908         pic_out->prop.f_psnr[2] = x264_psnr( ssd[2], chroma_size );
3909         pic_out->prop.f_psnr_avg = x264_psnr( ssd[0] + ssd[1] + ssd[2], luma_size + chroma_size*2 );
3910
3911         h->stat.f_ssd_global[h->sh.i_type]   += dur * (ssd[0] + ssd[1] + ssd[2]);
3912         h->stat.f_psnr_average[h->sh.i_type] += dur * pic_out->prop.f_psnr_avg;
3913         h->stat.f_psnr_mean_y[h->sh.i_type]  += dur * pic_out->prop.f_psnr[0];
3914         h->stat.f_psnr_mean_u[h->sh.i_type]  += dur * pic_out->prop.f_psnr[1];
3915         h->stat.f_psnr_mean_v[h->sh.i_type]  += dur * pic_out->prop.f_psnr[2];
3916
3917         snprintf( psz_message, 80, " PSNR Y:%5.2f U:%5.2f V:%5.2f", pic_out->prop.f_psnr[0],
3918                                                                     pic_out->prop.f_psnr[1],
3919                                                                     pic_out->prop.f_psnr[2] );
3920     }
3921
3922     if( h->param.analyse.b_ssim )
3923     {
3924         pic_out->prop.f_ssim = h->stat.frame.f_ssim / h->stat.frame.i_ssim_cnt;
3925         h->stat.f_ssim_mean_y[h->sh.i_type] += pic_out->prop.f_ssim * dur;
3926         snprintf( psz_message + strlen(psz_message), 80 - strlen(psz_message),
3927                   " SSIM Y:%.5f", pic_out->prop.f_ssim );
3928     }
3929     psz_message[79] = '\0';
3930
3931     x264_log( h, X264_LOG_DEBUG,
3932                   "frame=%4d QP=%.2f NAL=%d Slice:%c Poc:%-3d I:%-4d P:%-4d SKIP:%-4d size=%d bytes%s\n",
3933               h->i_frame,
3934               h->fdec->f_qp_avg_aq,
3935               h->i_nal_ref_idc,
3936               h->sh.i_type == SLICE_TYPE_I ? 'I' : (h->sh.i_type == SLICE_TYPE_P ? 'P' : 'B' ),
3937               h->fdec->i_poc,
3938               h->stat.frame.i_mb_count_i,
3939               h->stat.frame.i_mb_count_p,
3940               h->stat.frame.i_mb_count_skip,
3941               frame_size,
3942               psz_message );
3943
3944     // keep stats all in one place
3945     x264_thread_sync_stat( h->thread[0], h );
3946     // for the use of the next frame
3947     x264_thread_sync_stat( thread_current, h );
3948
3949 #ifdef DEBUG_MB_TYPE
3950 {
3951     static const char mb_chars[] = { 'i', 'i', 'I', 'C', 'P', '8', 'S',
3952         'D', '<', 'X', 'B', 'X', '>', 'B', 'B', 'B', 'B', '8', 'S' };
3953     for( int mb_xy = 0; mb_xy < h->mb.i_mb_width * h->mb.i_mb_height; mb_xy++ )
3954     {
3955         if( h->mb.type[mb_xy] < X264_MBTYPE_MAX && h->mb.type[mb_xy] >= 0 )
3956             fprintf( stderr, "%c ", mb_chars[ h->mb.type[mb_xy] ] );
3957         else
3958             fprintf( stderr, "? " );
3959
3960         if( (mb_xy+1) % h->mb.i_mb_width == 0 )
3961             fprintf( stderr, "\n" );
3962     }
3963 }
3964 #endif
3965
3966     /* Remove duplicates, must be done near the end as breaks h->fref0 array
3967      * by freeing some of its pointers. */
3968     for( int i = 0; i < h->i_ref[0]; i++ )
3969         if( h->fref[0][i] && h->fref[0][i]->b_duplicate )
3970         {
3971             x264_frame_push_blank_unused( h, h->fref[0][i] );
3972             h->fref[0][i] = 0;
3973         }
3974
3975     if( h->param.psz_dump_yuv )
3976         x264_frame_dump( h );
3977     x264_emms();
3978
3979     return frame_size;
3980 }
3981
3982 static void x264_print_intra( int64_t *i_mb_count, double i_count, int b_print_pcm, char *intra )
3983 {
3984     intra += sprintf( intra, "I16..4%s: %4.1f%% %4.1f%% %4.1f%%",
3985         b_print_pcm ? "..PCM" : "",
3986         i_mb_count[I_16x16]/ i_count,
3987         i_mb_count[I_8x8]  / i_count,
3988         i_mb_count[I_4x4]  / i_count );
3989     if( b_print_pcm )
3990         sprintf( intra, " %4.1f%%", i_mb_count[I_PCM]  / i_count );
3991 }
3992
3993 /****************************************************************************
3994  * x264_encoder_close:
3995  ****************************************************************************/
3996 void    x264_encoder_close  ( x264_t *h )
3997 {
3998     int64_t i_yuv_size = FRAME_SIZE( h->param.i_width * h->param.i_height );
3999     int64_t i_mb_count_size[2][7] = {{0}};
4000     char buf[200];
4001     int b_print_pcm = h->stat.i_mb_count[SLICE_TYPE_I][I_PCM]
4002                    || h->stat.i_mb_count[SLICE_TYPE_P][I_PCM]
4003                    || h->stat.i_mb_count[SLICE_TYPE_B][I_PCM];
4004
4005     x264_lookahead_delete( h );
4006
4007 #if HAVE_OPENCL
4008     x264_opencl_lookahead_delete( h );
4009     x264_opencl_function_t *ocl = h->opencl.ocl;
4010 #endif
4011
4012     if( h->param.b_sliced_threads )
4013         x264_threadpool_wait_all( h );
4014     if( h->param.i_threads > 1 )
4015         x264_threadpool_delete( h->threadpool );
4016     if( h->param.i_lookahead_threads > 1 )
4017         x264_threadpool_delete( h->lookaheadpool );
4018     if( h->i_thread_frames > 1 )
4019     {
4020         for( int i = 0; i < h->i_thread_frames; i++ )
4021             if( h->thread[i]->b_thread_active )
4022             {
4023                 assert( h->thread[i]->fenc->i_reference_count == 1 );
4024                 x264_frame_delete( h->thread[i]->fenc );
4025             }
4026
4027         x264_t *thread_prev = h->thread[h->i_thread_phase];
4028         x264_thread_sync_ratecontrol( h, thread_prev, h );
4029         x264_thread_sync_ratecontrol( thread_prev, thread_prev, h );
4030         h->i_frame = thread_prev->i_frame + 1 - h->i_thread_frames;
4031     }
4032     h->i_frame++;
4033
4034     /* Slices used and PSNR */
4035     for( int i = 0; i < 3; i++ )
4036     {
4037         static const uint8_t slice_order[] = { SLICE_TYPE_I, SLICE_TYPE_P, SLICE_TYPE_B };
4038         int i_slice = slice_order[i];
4039
4040         if( h->stat.i_frame_count[i_slice] > 0 )
4041         {
4042             int i_count = h->stat.i_frame_count[i_slice];
4043             double dur =  h->stat.f_frame_duration[i_slice];
4044             if( h->param.analyse.b_psnr )
4045             {
4046                 x264_log( h, X264_LOG_INFO,
4047                           "frame %c:%-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",
4048                           slice_type_to_char[i_slice],
4049                           i_count,
4050                           h->stat.f_frame_qp[i_slice] / i_count,
4051                           (double)h->stat.i_frame_size[i_slice] / i_count,
4052                           h->stat.f_psnr_mean_y[i_slice] / dur, h->stat.f_psnr_mean_u[i_slice] / dur, h->stat.f_psnr_mean_v[i_slice] / dur,
4053                           h->stat.f_psnr_average[i_slice] / dur,
4054                           x264_psnr( h->stat.f_ssd_global[i_slice], dur * i_yuv_size ) );
4055             }
4056             else
4057             {
4058                 x264_log( h, X264_LOG_INFO,
4059                           "frame %c:%-5d Avg QP:%5.2f  size:%6.0f\n",
4060                           slice_type_to_char[i_slice],
4061                           i_count,
4062                           h->stat.f_frame_qp[i_slice] / i_count,
4063                           (double)h->stat.i_frame_size[i_slice] / i_count );
4064             }
4065         }
4066     }
4067     if( h->param.i_bframe && h->stat.i_frame_count[SLICE_TYPE_B] )
4068     {
4069         char *p = buf;
4070         int den = 0;
4071         // weight by number of frames (including the I/P-frames) that are in a sequence of N B-frames
4072         for( int i = 0; i <= h->param.i_bframe; i++ )
4073             den += (i+1) * h->stat.i_consecutive_bframes[i];
4074         for( int i = 0; i <= h->param.i_bframe; i++ )
4075             p += sprintf( p, " %4.1f%%", 100. * (i+1) * h->stat.i_consecutive_bframes[i] / den );
4076         x264_log( h, X264_LOG_INFO, "consecutive B-frames:%s\n", buf );
4077     }
4078
4079     for( int i_type = 0; i_type < 2; i_type++ )
4080         for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
4081         {
4082             if( i == D_DIRECT_8x8 ) continue; /* direct is counted as its own type */
4083             i_mb_count_size[i_type][x264_mb_partition_pixel_table[i]] += h->stat.i_mb_partition[i_type][i];
4084         }
4085
4086     /* MB types used */
4087     if( h->stat.i_frame_count[SLICE_TYPE_I] > 0 )
4088     {
4089         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_I];
4090         double i_count = h->stat.i_frame_count[SLICE_TYPE_I] * h->mb.i_mb_count / 100.0;
4091         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
4092         x264_log( h, X264_LOG_INFO, "mb I  %s\n", buf );
4093     }
4094     if( h->stat.i_frame_count[SLICE_TYPE_P] > 0 )
4095     {
4096         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_P];
4097         double i_count = h->stat.i_frame_count[SLICE_TYPE_P] * h->mb.i_mb_count / 100.0;
4098         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_P];
4099         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
4100         x264_log( h, X264_LOG_INFO,
4101                   "mb P  %s  P16..4: %4.1f%% %4.1f%% %4.1f%% %4.1f%% %4.1f%%    skip:%4.1f%%\n",
4102                   buf,
4103                   i_mb_size[PIXEL_16x16] / (i_count*4),
4104                   (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
4105                   i_mb_size[PIXEL_8x8] / (i_count*4),
4106                   (i_mb_size[PIXEL_8x4] + i_mb_size[PIXEL_4x8]) / (i_count*4),
4107                   i_mb_size[PIXEL_4x4] / (i_count*4),
4108                   i_mb_count[P_SKIP] / i_count );
4109     }
4110     if( h->stat.i_frame_count[SLICE_TYPE_B] > 0 )
4111     {
4112         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_B];
4113         double i_count = h->stat.i_frame_count[SLICE_TYPE_B] * h->mb.i_mb_count / 100.0;
4114         double i_mb_list_count;
4115         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_B];
4116         int64_t list_count[3] = {0}; /* 0 == L0, 1 == L1, 2 == BI */
4117         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
4118         for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
4119             for( int j = 0; j < 2; j++ )
4120             {
4121                 int l0 = x264_mb_type_list_table[i][0][j];
4122                 int l1 = x264_mb_type_list_table[i][1][j];
4123                 if( l0 || l1 )
4124                     list_count[l1+l0*l1] += h->stat.i_mb_count[SLICE_TYPE_B][i] * 2;
4125             }
4126         list_count[0] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L0_8x8];
4127         list_count[1] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L1_8x8];
4128         list_count[2] += h->stat.i_mb_partition[SLICE_TYPE_B][D_BI_8x8];
4129         i_mb_count[B_DIRECT] += (h->stat.i_mb_partition[SLICE_TYPE_B][D_DIRECT_8x8]+2)/4;
4130         i_mb_list_count = (list_count[0] + list_count[1] + list_count[2]) / 100.0;
4131         sprintf( buf + strlen(buf), "  B16..8: %4.1f%% %4.1f%% %4.1f%%  direct:%4.1f%%  skip:%4.1f%%",
4132                  i_mb_size[PIXEL_16x16] / (i_count*4),
4133                  (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
4134                  i_mb_size[PIXEL_8x8] / (i_count*4),
4135                  i_mb_count[B_DIRECT] / i_count,
4136                  i_mb_count[B_SKIP]   / i_count );
4137         if( i_mb_list_count != 0 )
4138             sprintf( buf + strlen(buf), "  L0:%4.1f%% L1:%4.1f%% BI:%4.1f%%",
4139                      list_count[0] / i_mb_list_count,
4140                      list_count[1] / i_mb_list_count,
4141                      list_count[2] / i_mb_list_count );
4142         x264_log( h, X264_LOG_INFO, "mb B  %s\n", buf );
4143     }
4144
4145     x264_ratecontrol_summary( h );
4146
4147     if( h->stat.i_frame_count[SLICE_TYPE_I] + h->stat.i_frame_count[SLICE_TYPE_P] + h->stat.i_frame_count[SLICE_TYPE_B] > 0 )
4148     {
4149 #define SUM3(p) (p[SLICE_TYPE_I] + p[SLICE_TYPE_P] + p[SLICE_TYPE_B])
4150 #define SUM3b(p,o) (p[SLICE_TYPE_I][o] + p[SLICE_TYPE_P][o] + p[SLICE_TYPE_B][o])
4151         int64_t i_i8x8 = SUM3b( h->stat.i_mb_count, I_8x8 );
4152         int64_t i_intra = i_i8x8 + SUM3b( h->stat.i_mb_count, I_4x4 )
4153                                  + SUM3b( h->stat.i_mb_count, I_16x16 );
4154         int64_t i_all_intra = i_intra + SUM3b( h->stat.i_mb_count, I_PCM);
4155         int64_t i_skip = SUM3b( h->stat.i_mb_count, P_SKIP )
4156                        + SUM3b( h->stat.i_mb_count, B_SKIP );
4157         const int i_count = h->stat.i_frame_count[SLICE_TYPE_I] +
4158                             h->stat.i_frame_count[SLICE_TYPE_P] +
4159                             h->stat.i_frame_count[SLICE_TYPE_B];
4160         int64_t i_mb_count = (int64_t)i_count * h->mb.i_mb_count;
4161         int64_t i_inter = i_mb_count - i_skip - i_intra;
4162         const double duration = h->stat.f_frame_duration[SLICE_TYPE_I] +
4163                                 h->stat.f_frame_duration[SLICE_TYPE_P] +
4164                                 h->stat.f_frame_duration[SLICE_TYPE_B];
4165         float f_bitrate = SUM3(h->stat.i_frame_size) / duration / 125;
4166
4167         if( PARAM_INTERLACED )
4168         {
4169             char *fieldstats = buf;
4170             fieldstats[0] = 0;
4171             if( i_inter )
4172                 fieldstats += sprintf( fieldstats, " inter:%.1f%%", h->stat.i_mb_field[1] * 100.0 / i_inter );
4173             if( i_skip )
4174                 fieldstats += sprintf( fieldstats, " skip:%.1f%%", h->stat.i_mb_field[2] * 100.0 / i_skip );
4175             x264_log( h, X264_LOG_INFO, "field mbs: intra: %.1f%%%s\n",
4176                       h->stat.i_mb_field[0] * 100.0 / i_intra, buf );
4177         }
4178
4179         if( h->pps->b_transform_8x8_mode )
4180         {
4181             buf[0] = 0;
4182             if( h->stat.i_mb_count_8x8dct[0] )
4183                 sprintf( buf, " inter:%.1f%%", 100. * h->stat.i_mb_count_8x8dct[1] / h->stat.i_mb_count_8x8dct[0] );
4184             x264_log( h, X264_LOG_INFO, "8x8 transform intra:%.1f%%%s\n", 100. * i_i8x8 / i_intra, buf );
4185         }
4186
4187         if( (h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO ||
4188             (h->stat.i_direct_frames[0] && h->stat.i_direct_frames[1]))
4189             && h->stat.i_frame_count[SLICE_TYPE_B] )
4190         {
4191             x264_log( h, X264_LOG_INFO, "direct mvs  spatial:%.1f%% temporal:%.1f%%\n",
4192                       h->stat.i_direct_frames[1] * 100. / h->stat.i_frame_count[SLICE_TYPE_B],
4193                       h->stat.i_direct_frames[0] * 100. / h->stat.i_frame_count[SLICE_TYPE_B] );
4194         }
4195
4196         buf[0] = 0;
4197         int csize = CHROMA444 ? 4 : 1;
4198         if( i_mb_count != i_all_intra )
4199             sprintf( buf, " inter: %.1f%% %.1f%% %.1f%%",
4200                      h->stat.i_mb_cbp[1] * 100.0 / ((i_mb_count - i_all_intra)*4),
4201                      h->stat.i_mb_cbp[3] * 100.0 / ((i_mb_count - i_all_intra)*csize),
4202                      h->stat.i_mb_cbp[5] * 100.0 / ((i_mb_count - i_all_intra)*csize) );
4203         x264_log( h, X264_LOG_INFO, "coded y,%s,%s intra: %.1f%% %.1f%% %.1f%%%s\n",
4204                   CHROMA444?"u":"uvDC", CHROMA444?"v":"uvAC",
4205                   h->stat.i_mb_cbp[0] * 100.0 / (i_all_intra*4),
4206                   h->stat.i_mb_cbp[2] * 100.0 / (i_all_intra*csize),
4207                   h->stat.i_mb_cbp[4] * 100.0 / (i_all_intra*csize), buf );
4208
4209         int64_t fixed_pred_modes[4][9] = {{0}};
4210         int64_t sum_pred_modes[4] = {0};
4211         for( int i = 0; i <= I_PRED_16x16_DC_128; i++ )
4212         {
4213             fixed_pred_modes[0][x264_mb_pred_mode16x16_fix[i]] += h->stat.i_mb_pred_mode[0][i];
4214             sum_pred_modes[0] += h->stat.i_mb_pred_mode[0][i];
4215         }
4216         if( sum_pred_modes[0] )
4217             x264_log( h, X264_LOG_INFO, "i16 v,h,dc,p: %2.0f%% %2.0f%% %2.0f%% %2.0f%%\n",
4218                       fixed_pred_modes[0][0] * 100.0 / sum_pred_modes[0],
4219                       fixed_pred_modes[0][1] * 100.0 / sum_pred_modes[0],
4220                       fixed_pred_modes[0][2] * 100.0 / sum_pred_modes[0],
4221                       fixed_pred_modes[0][3] * 100.0 / sum_pred_modes[0] );
4222         for( int i = 1; i <= 2; i++ )
4223         {
4224             for( int j = 0; j <= I_PRED_8x8_DC_128; j++ )
4225             {
4226                 fixed_pred_modes[i][x264_mb_pred_mode4x4_fix(j)] += h->stat.i_mb_pred_mode[i][j];
4227                 sum_pred_modes[i] += h->stat.i_mb_pred_mode[i][j];
4228             }
4229             if( sum_pred_modes[i] )
4230                 x264_log( h, X264_LOG_INFO, "i%d v,h,dc,ddl,ddr,vr,hd,vl,hu: %2.0f%% %2.0f%% %2.0f%% %2.0f%% %2.0f%% %2.0f%% %2.0f%% %2.0f%% %2.0f%%\n", (3-i)*4,
4231                           fixed_pred_modes[i][0] * 100.0 / sum_pred_modes[i],
4232                           fixed_pred_modes[i][1] * 100.0 / sum_pred_modes[i],
4233                           fixed_pred_modes[i][2] * 100.0 / sum_pred_modes[i],
4234                           fixed_pred_modes[i][3] * 100.0 / sum_pred_modes[i],
4235                           fixed_pred_modes[i][4] * 100.0 / sum_pred_modes[i],
4236                           fixed_pred_modes[i][5] * 100.0 / sum_pred_modes[i],
4237                           fixed_pred_modes[i][6] * 100.0 / sum_pred_modes[i],
4238                           fixed_pred_modes[i][7] * 100.0 / sum_pred_modes[i],
4239                           fixed_pred_modes[i][8] * 100.0 / sum_pred_modes[i] );
4240         }
4241         for( int i = 0; i <= I_PRED_CHROMA_DC_128; i++ )
4242         {
4243             fixed_pred_modes[3][x264_mb_chroma_pred_mode_fix[i]] += h->stat.i_mb_pred_mode[3][i];
4244             sum_pred_modes[3] += h->stat.i_mb_pred_mode[3][i];
4245         }
4246         if( sum_pred_modes[3] && !CHROMA444 )
4247             x264_log( h, X264_LOG_INFO, "i8c dc,h,v,p: %2.0f%% %2.0f%% %2.0f%% %2.0f%%\n",
4248                       fixed_pred_modes[3][0] * 100.0 / sum_pred_modes[3],
4249                       fixed_pred_modes[3][1] * 100.0 / sum_pred_modes[3],
4250                       fixed_pred_modes[3][2] * 100.0 / sum_pred_modes[3],
4251                       fixed_pred_modes[3][3] * 100.0 / sum_pred_modes[3] );
4252
4253         if( h->param.analyse.i_weighted_pred >= X264_WEIGHTP_SIMPLE && h->stat.i_frame_count[SLICE_TYPE_P] > 0 )
4254             x264_log( h, X264_LOG_INFO, "Weighted P-Frames: Y:%.1f%% UV:%.1f%%\n",
4255                       h->stat.i_wpred[0] * 100.0 / h->stat.i_frame_count[SLICE_TYPE_P],
4256                       h->stat.i_wpred[1] * 100.0 / h->stat.i_frame_count[SLICE_TYPE_P] );
4257
4258         for( int i_list = 0; i_list < 2; i_list++ )
4259             for( int i_slice = 0; i_slice < 2; i_slice++ )
4260             {
4261                 char *p = buf;
4262                 int64_t i_den = 0;
4263                 int i_max = 0;
4264                 for( int i = 0; i < X264_REF_MAX*2; i++ )
4265                     if( h->stat.i_mb_count_ref[i_slice][i_list][i] )
4266                     {
4267                         i_den += h->stat.i_mb_count_ref[i_slice][i_list][i];
4268                         i_max = i;
4269                     }
4270                 if( i_max == 0 )
4271                     continue;
4272                 for( int i = 0; i <= i_max; i++ )
4273                     p += sprintf( p, " %4.1f%%", 100. * h->stat.i_mb_count_ref[i_slice][i_list][i] / i_den );
4274                 x264_log( h, X264_LOG_INFO, "ref %c L%d:%s\n", "PB"[i_slice], i_list, buf );
4275             }
4276
4277         if( h->param.analyse.b_ssim )
4278         {
4279             float ssim = SUM3( h->stat.f_ssim_mean_y ) / duration;
4280             x264_log( h, X264_LOG_INFO, "SSIM Mean Y:%.7f (%6.3fdb)\n", ssim, x264_ssim( ssim ) );
4281         }
4282         if( h->param.analyse.b_psnr )
4283         {
4284             x264_log( h, X264_LOG_INFO,
4285                       "PSNR Mean Y:%6.3f U:%6.3f V:%6.3f Avg:%6.3f Global:%6.3f kb/s:%.2f\n",
4286                       SUM3( h->stat.f_psnr_mean_y ) / duration,
4287                       SUM3( h->stat.f_psnr_mean_u ) / duration,
4288                       SUM3( h->stat.f_psnr_mean_v ) / duration,
4289                       SUM3( h->stat.f_psnr_average ) / duration,
4290                       x264_psnr( SUM3( h->stat.f_ssd_global ), duration * i_yuv_size ),
4291                       f_bitrate );
4292         }
4293         else
4294             x264_log( h, X264_LOG_INFO, "kb/s:%.2f\n", f_bitrate );
4295     }
4296
4297     /* rc */
4298     x264_ratecontrol_delete( h );
4299
4300     /* param */
4301     if( h->param.rc.psz_stat_out )
4302         free( h->param.rc.psz_stat_out );
4303     if( h->param.rc.psz_stat_in )
4304         free( h->param.rc.psz_stat_in );
4305
4306     x264_cqm_delete( h );
4307     x264_free( h->nal_buffer );
4308     x264_free( h->reconfig_h );
4309     x264_analyse_free_costs( h );
4310
4311     if( h->i_thread_frames > 1 )
4312         h = h->thread[h->i_thread_phase];
4313
4314     /* frames */
4315     x264_frame_delete_list( h->frames.unused[0] );
4316     x264_frame_delete_list( h->frames.unused[1] );
4317     x264_frame_delete_list( h->frames.current );
4318     x264_frame_delete_list( h->frames.blank_unused );
4319
4320     h = h->thread[0];
4321
4322     for( int i = 0; i < h->i_thread_frames; i++ )
4323         if( h->thread[i]->b_thread_active )
4324             for( int j = 0; j < h->thread[i]->i_ref[0]; j++ )
4325                 if( h->thread[i]->fref[0][j] && h->thread[i]->fref[0][j]->b_duplicate )
4326                     x264_frame_delete( h->thread[i]->fref[0][j] );
4327
4328     if( h->param.i_lookahead_threads > 1 )
4329         for( int i = 0; i < h->param.i_lookahead_threads; i++ )
4330             x264_free( h->lookahead_thread[i] );
4331
4332     for( int i = h->param.i_threads - 1; i >= 0; i-- )
4333     {
4334         x264_frame_t **frame;
4335
4336         if( !h->param.b_sliced_threads || i == 0 )
4337         {
4338             for( frame = h->thread[i]->frames.reference; *frame; frame++ )
4339             {
4340                 assert( (*frame)->i_reference_count > 0 );
4341                 (*frame)->i_reference_count--;
4342                 if( (*frame)->i_reference_count == 0 )
4343                     x264_frame_delete( *frame );
4344             }
4345             frame = &h->thread[i]->fdec;
4346             if( *frame )
4347             {
4348                 assert( (*frame)->i_reference_count > 0 );
4349                 (*frame)->i_reference_count--;
4350                 if( (*frame)->i_reference_count == 0 )
4351                     x264_frame_delete( *frame );
4352             }
4353             x264_macroblock_cache_free( h->thread[i] );
4354         }
4355         x264_macroblock_thread_free( h->thread[i], 0 );
4356         x264_free( h->thread[i]->out.p_bitstream );
4357         x264_free( h->thread[i]->out.nal );
4358         x264_pthread_mutex_destroy( &h->thread[i]->mutex );
4359         x264_pthread_cond_destroy( &h->thread[i]->cv );
4360         x264_free( h->thread[i] );
4361     }
4362 #if HAVE_OPENCL
4363     x264_opencl_close_library( ocl );
4364 #endif
4365 }
4366
4367 int x264_encoder_delayed_frames( x264_t *h )
4368 {
4369     int delayed_frames = 0;
4370     if( h->i_thread_frames > 1 )
4371     {
4372         for( int i = 0; i < h->i_thread_frames; i++ )
4373             delayed_frames += h->thread[i]->b_thread_active;
4374         h = h->thread[h->i_thread_phase];
4375     }
4376     for( int i = 0; h->frames.current[i]; i++ )
4377         delayed_frames++;
4378     x264_pthread_mutex_lock( &h->lookahead->ofbuf.mutex );
4379     x264_pthread_mutex_lock( &h->lookahead->ifbuf.mutex );
4380     x264_pthread_mutex_lock( &h->lookahead->next.mutex );
4381     delayed_frames += h->lookahead->ifbuf.i_size + h->lookahead->next.i_size + h->lookahead->ofbuf.i_size;
4382     x264_pthread_mutex_unlock( &h->lookahead->next.mutex );
4383     x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
4384     x264_pthread_mutex_unlock( &h->lookahead->ofbuf.mutex );
4385     return delayed_frames;
4386 }
4387
4388 int x264_encoder_maximum_delayed_frames( x264_t *h )
4389 {
4390     return h->frames.i_delay;
4391 }