RGB encoding support
[x262.git] / encoder / encoder.c
1 /*****************************************************************************
2  * encoder.c: top-level encoder functions
3  *****************************************************************************
4  * Copyright (C) 2003-2011 x264 project
5  *
6  * Authors: Laurent Aimar <fenrir@via.ecp.fr>
7  *          Loren Merritt <lorenm@u.washington.edu>
8  *          Jason Garrett-Glaser <darkshikari@gmail.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 #if HAVE_VISUALIZE
37 #include "common/visualize.h"
38 #endif
39
40 //#define DEBUG_MB_TYPE
41
42 #define bs_write_ue bs_write_ue_big
43
44 static int x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
45                                    x264_nal_t **pp_nal, int *pi_nal,
46                                    x264_picture_t *pic_out );
47
48 /****************************************************************************
49  *
50  ******************************* x264 libs **********************************
51  *
52  ****************************************************************************/
53 static double x264_psnr( double sqe, double size )
54 {
55     double mse = sqe / (PIXEL_MAX*PIXEL_MAX * size);
56     if( mse <= 0.0000000001 ) /* Max 100dB */
57         return 100;
58
59     return -10.0 * log10( mse );
60 }
61
62 static double x264_ssim( double ssim )
63 {
64     return -10.0 * log10( 1 - ssim );
65 }
66
67 static void x264_frame_dump( x264_t *h )
68 {
69     FILE *f = fopen( h->param.psz_dump_yuv, "r+b" );
70     if( !f )
71         return;
72
73     /* Write the frame in display order */
74     int frame_size = h->param.i_height * h->param.i_width * (3<<CHROMA444)/2 * sizeof(pixel);
75     fseek( f, (uint64_t)h->fdec->i_frame * frame_size, SEEK_SET );
76     for( int p = 0; p < (CHROMA444 ? 3 : 1); p++ )
77         for( int y = 0; y < h->param.i_height; y++ )
78             fwrite( &h->fdec->plane[p][y*h->fdec->i_stride[p]], sizeof(pixel), h->param.i_width, f );
79     if( !CHROMA444 )
80     {
81         int cw = h->param.i_width>>1;
82         int ch = h->param.i_height>>1;
83         pixel *planeu = x264_malloc( (cw*ch*2+32)*sizeof(pixel) );
84         pixel *planev = planeu + cw*ch + 16;
85         h->mc.plane_copy_deinterleave( planeu, cw, planev, cw, h->fdec->plane[1], h->fdec->i_stride[1], cw, ch );
86         fwrite( planeu, 1, cw*ch*sizeof(pixel), f );
87         fwrite( planev, 1, cw*ch*sizeof(pixel), f );
88         x264_free( planeu );
89     }
90     fclose( f );
91 }
92
93
94 /* Fill "default" values */
95 static void x264_slice_header_init( x264_t *h, x264_slice_header_t *sh,
96                                     x264_sps_t *sps, x264_pps_t *pps,
97                                     int i_idr_pic_id, int i_frame, int i_qp )
98 {
99     x264_param_t *param = &h->param;
100
101     /* First we fill all fields */
102     sh->sps = sps;
103     sh->pps = pps;
104
105     sh->i_first_mb  = 0;
106     sh->i_last_mb   = h->mb.i_mb_count - 1;
107     sh->i_pps_id    = pps->i_id;
108
109     sh->i_frame_num = i_frame;
110
111     sh->b_mbaff = PARAM_INTERLACED;
112     sh->b_field_pic = 0;    /* no field support for now */
113     sh->b_bottom_field = 0; /* not yet used */
114
115     sh->i_idr_pic_id = i_idr_pic_id;
116
117     /* poc stuff, fixed later */
118     sh->i_poc = 0;
119     sh->i_delta_poc_bottom = 0;
120     sh->i_delta_poc[0] = 0;
121     sh->i_delta_poc[1] = 0;
122
123     sh->i_redundant_pic_cnt = 0;
124
125     h->mb.b_direct_auto_write = h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO
126                                 && h->param.i_bframe
127                                 && ( h->param.rc.b_stat_write || !h->param.rc.b_stat_read );
128
129     if( !h->mb.b_direct_auto_read && sh->i_type == SLICE_TYPE_B )
130     {
131         if( h->fref[1][0]->i_poc_l0ref0 == h->fref[0][0]->i_poc )
132         {
133             if( h->mb.b_direct_auto_write )
134                 sh->b_direct_spatial_mv_pred = ( h->stat.i_direct_score[1] > h->stat.i_direct_score[0] );
135             else
136                 sh->b_direct_spatial_mv_pred = ( param->analyse.i_direct_mv_pred == X264_DIRECT_PRED_SPATIAL );
137         }
138         else
139         {
140             h->mb.b_direct_auto_write = 0;
141             sh->b_direct_spatial_mv_pred = 1;
142         }
143     }
144     /* else b_direct_spatial_mv_pred was read from the 2pass statsfile */
145
146     sh->b_num_ref_idx_override = 0;
147     sh->i_num_ref_idx_l0_active = 1;
148     sh->i_num_ref_idx_l1_active = 1;
149
150     sh->b_ref_pic_list_reordering[0] = h->b_ref_reorder[0];
151     sh->b_ref_pic_list_reordering[1] = h->b_ref_reorder[1];
152
153     /* If the ref list isn't in the default order, construct reordering header */
154     for( int list = 0; list < 2; list++ )
155     {
156         if( sh->b_ref_pic_list_reordering[list] )
157         {
158             int pred_frame_num = i_frame;
159             for( int i = 0; i < h->i_ref[list]; i++ )
160             {
161                 int diff = h->fref[list][i]->i_frame_num - pred_frame_num;
162                 sh->ref_pic_list_order[list][i].idc = ( diff > 0 );
163                 sh->ref_pic_list_order[list][i].arg = (abs(diff) - 1) & ((1 << sps->i_log2_max_frame_num) - 1);
164                 pred_frame_num = h->fref[list][i]->i_frame_num;
165             }
166         }
167     }
168
169     sh->i_cabac_init_idc = param->i_cabac_init_idc;
170
171     sh->i_qp = SPEC_QP(i_qp);
172     sh->i_qp_delta = sh->i_qp - pps->i_pic_init_qp;
173     sh->b_sp_for_swidth = 0;
174     sh->i_qs_delta = 0;
175
176     int deblock_thresh = i_qp + 2 * X264_MIN(param->i_deblocking_filter_alphac0, param->i_deblocking_filter_beta);
177     /* If effective qp <= 15, deblocking would have no effect anyway */
178     if( param->b_deblocking_filter && (h->mb.b_variable_qp || 15 < deblock_thresh ) )
179         sh->i_disable_deblocking_filter_idc = param->b_sliced_threads ? 2 : 0;
180     else
181         sh->i_disable_deblocking_filter_idc = 1;
182     sh->i_alpha_c0_offset = param->i_deblocking_filter_alphac0 << 1;
183     sh->i_beta_offset = param->i_deblocking_filter_beta << 1;
184 }
185
186 static void x264_slice_header_write( bs_t *s, x264_slice_header_t *sh, int i_nal_ref_idc )
187 {
188     if( sh->b_mbaff )
189     {
190         int first_x = sh->i_first_mb % sh->sps->i_mb_width;
191         int first_y = sh->i_first_mb / sh->sps->i_mb_width;
192         assert( (first_y&1) == 0 );
193         bs_write_ue( s, (2*first_x + sh->sps->i_mb_width*(first_y&~1) + (first_y&1)) >> 1 );
194     }
195     else
196         bs_write_ue( s, sh->i_first_mb );
197
198     bs_write_ue( s, sh->i_type + 5 );   /* same type things */
199     bs_write_ue( s, sh->i_pps_id );
200     bs_write( s, sh->sps->i_log2_max_frame_num, sh->i_frame_num & ((1<<sh->sps->i_log2_max_frame_num)-1) );
201
202     if( !sh->sps->b_frame_mbs_only )
203     {
204         bs_write1( s, sh->b_field_pic );
205         if( sh->b_field_pic )
206             bs_write1( s, sh->b_bottom_field );
207     }
208
209     if( sh->i_idr_pic_id >= 0 ) /* NAL IDR */
210         bs_write_ue( s, sh->i_idr_pic_id );
211
212     if( sh->sps->i_poc_type == 0 )
213     {
214         bs_write( s, sh->sps->i_log2_max_poc_lsb, sh->i_poc & ((1<<sh->sps->i_log2_max_poc_lsb)-1) );
215         if( sh->pps->b_pic_order && !sh->b_field_pic )
216             bs_write_se( s, sh->i_delta_poc_bottom );
217     }
218
219     if( sh->pps->b_redundant_pic_cnt )
220         bs_write_ue( s, sh->i_redundant_pic_cnt );
221
222     if( sh->i_type == SLICE_TYPE_B )
223         bs_write1( s, sh->b_direct_spatial_mv_pred );
224
225     if( sh->i_type == SLICE_TYPE_P || sh->i_type == SLICE_TYPE_B )
226     {
227         bs_write1( s, sh->b_num_ref_idx_override );
228         if( sh->b_num_ref_idx_override )
229         {
230             bs_write_ue( s, sh->i_num_ref_idx_l0_active - 1 );
231             if( sh->i_type == SLICE_TYPE_B )
232                 bs_write_ue( s, sh->i_num_ref_idx_l1_active - 1 );
233         }
234     }
235
236     /* ref pic list reordering */
237     if( sh->i_type != SLICE_TYPE_I )
238     {
239         bs_write1( s, sh->b_ref_pic_list_reordering[0] );
240         if( sh->b_ref_pic_list_reordering[0] )
241         {
242             for( int i = 0; i < sh->i_num_ref_idx_l0_active; i++ )
243             {
244                 bs_write_ue( s, sh->ref_pic_list_order[0][i].idc );
245                 bs_write_ue( s, sh->ref_pic_list_order[0][i].arg );
246             }
247             bs_write_ue( s, 3 );
248         }
249     }
250     if( sh->i_type == SLICE_TYPE_B )
251     {
252         bs_write1( s, sh->b_ref_pic_list_reordering[1] );
253         if( sh->b_ref_pic_list_reordering[1] )
254         {
255             for( int i = 0; i < sh->i_num_ref_idx_l1_active; i++ )
256             {
257                 bs_write_ue( s, sh->ref_pic_list_order[1][i].idc );
258                 bs_write_ue( s, sh->ref_pic_list_order[1][i].arg );
259             }
260             bs_write_ue( s, 3 );
261         }
262     }
263
264     if( sh->pps->b_weighted_pred && sh->i_type == SLICE_TYPE_P )
265     {
266         /* pred_weight_table() */
267         bs_write_ue( s, sh->weight[0][0].i_denom );
268         bs_write_ue( s, sh->weight[0][1].i_denom );
269         for( int i = 0; i < sh->i_num_ref_idx_l0_active; i++ )
270         {
271             int luma_weight_l0_flag = !!sh->weight[i][0].weightfn;
272             int chroma_weight_l0_flag = !!sh->weight[i][1].weightfn || !!sh->weight[i][2].weightfn;
273             bs_write1( s, luma_weight_l0_flag );
274             if( luma_weight_l0_flag )
275             {
276                 bs_write_se( s, sh->weight[i][0].i_scale );
277                 bs_write_se( s, sh->weight[i][0].i_offset );
278             }
279             bs_write1( s, chroma_weight_l0_flag );
280             if( chroma_weight_l0_flag )
281             {
282                 for( int j = 1; j < 3; j++ )
283                 {
284                     bs_write_se( s, sh->weight[i][j].i_scale );
285                     bs_write_se( s, sh->weight[i][j].i_offset );
286                 }
287             }
288         }
289     }
290     else if( sh->pps->b_weighted_bipred == 1 && sh->i_type == SLICE_TYPE_B )
291     {
292       /* TODO */
293     }
294
295     if( i_nal_ref_idc != 0 )
296     {
297         if( sh->i_idr_pic_id >= 0 )
298         {
299             bs_write1( s, 0 );  /* no output of prior pics flag */
300             bs_write1( s, 0 );  /* long term reference flag */
301         }
302         else
303         {
304             bs_write1( s, sh->i_mmco_command_count > 0 ); /* adaptive_ref_pic_marking_mode_flag */
305             if( sh->i_mmco_command_count > 0 )
306             {
307                 for( int i = 0; i < sh->i_mmco_command_count; i++ )
308                 {
309                     bs_write_ue( s, 1 ); /* mark short term ref as unused */
310                     bs_write_ue( s, sh->mmco[i].i_difference_of_pic_nums - 1 );
311                 }
312                 bs_write_ue( s, 0 ); /* end command list */
313             }
314         }
315     }
316
317     if( sh->pps->b_cabac && sh->i_type != SLICE_TYPE_I )
318         bs_write_ue( s, sh->i_cabac_init_idc );
319
320     bs_write_se( s, sh->i_qp_delta );      /* slice qp delta */
321
322     if( sh->pps->b_deblocking_filter_control )
323     {
324         bs_write_ue( s, sh->i_disable_deblocking_filter_idc );
325         if( sh->i_disable_deblocking_filter_idc != 1 )
326         {
327             bs_write_se( s, sh->i_alpha_c0_offset >> 1 );
328             bs_write_se( s, sh->i_beta_offset >> 1 );
329         }
330     }
331 }
332
333 /* If we are within a reasonable distance of the end of the memory allocated for the bitstream, */
334 /* reallocate, adding an arbitrary amount of space (100 kilobytes). */
335 static int x264_bitstream_check_buffer( x264_t *h )
336 {
337     uint8_t *bs_bak = h->out.p_bitstream;
338     int max_mb_size = 2500 << SLICE_MBAFF;
339     if( (h->param.b_cabac && (h->cabac.p_end - h->cabac.p < max_mb_size)) ||
340         (h->out.bs.p_end - h->out.bs.p < max_mb_size) )
341     {
342         h->out.i_bitstream += 100000;
343         CHECKED_MALLOC( h->out.p_bitstream, h->out.i_bitstream );
344         h->mc.memcpy_aligned( h->out.p_bitstream, bs_bak, (h->out.i_bitstream - 100000) & ~15 );
345         intptr_t delta = h->out.p_bitstream - bs_bak;
346
347         h->out.bs.p_start += delta;
348         h->out.bs.p += delta;
349         h->out.bs.p_end = h->out.p_bitstream + h->out.i_bitstream;
350
351         h->cabac.p_start += delta;
352         h->cabac.p += delta;
353         h->cabac.p_end = h->out.p_bitstream + h->out.i_bitstream;
354
355         for( int i = 0; i <= h->out.i_nal; i++ )
356             h->out.nal[i].p_payload += delta;
357         x264_free( bs_bak );
358     }
359     return 0;
360 fail:
361     x264_free( bs_bak );
362     return -1;
363 }
364
365 #if HAVE_THREAD
366 static void x264_encoder_thread_init( x264_t *h )
367 {
368     if( h->param.i_sync_lookahead )
369         x264_lower_thread_priority( 10 );
370
371 #if HAVE_MMX
372     /* Misalign mask has to be set separately for each thread. */
373     if( h->param.cpu&X264_CPU_SSE_MISALIGN )
374         x264_cpu_mask_misalign_sse();
375 #endif
376 }
377 #endif
378
379 /****************************************************************************
380  *
381  ****************************************************************************
382  ****************************** External API*********************************
383  ****************************************************************************
384  *
385  ****************************************************************************/
386
387 static int x264_validate_parameters( x264_t *h, int b_open )
388 {
389 #if HAVE_MMX
390 #ifdef __SSE__
391     if( b_open && !(x264_cpu_detect() & X264_CPU_SSE) )
392     {
393         x264_log( h, X264_LOG_ERROR, "your cpu does not support SSE1, but x264 was compiled with asm support\n");
394 #else
395     if( b_open && !(x264_cpu_detect() & X264_CPU_MMXEXT) )
396     {
397         x264_log( h, X264_LOG_ERROR, "your cpu does not support MMXEXT, but x264 was compiled with asm support\n");
398 #endif
399         x264_log( h, X264_LOG_ERROR, "to run x264, recompile without asm support (configure --disable-asm)\n");
400         return -1;
401     }
402 #endif
403     if( h->param.i_width <= 0 || h->param.i_height <= 0 )
404     {
405         x264_log( h, X264_LOG_ERROR, "invalid width x height (%dx%d)\n",
406                   h->param.i_width, h->param.i_height );
407         return -1;
408     }
409
410     int i_csp = h->param.i_csp & X264_CSP_MASK;
411     if( i_csp <= X264_CSP_NONE || i_csp >= X264_CSP_MAX )
412     {
413         x264_log( h, X264_LOG_ERROR, "invalid CSP (only I420/YV12/NV12/I444/YV24/BGR/BGRA/RGB supported)\n" );
414         return -1;
415     }
416
417     if( i_csp < X264_CSP_I444 && (h->param.i_width % 2 || h->param.i_height % 2) )
418     {
419         x264_log( h, X264_LOG_ERROR, "width or height not divisible by 2 (%dx%d)\n",
420                   h->param.i_width, h->param.i_height );
421         return -1;
422     }
423
424 #if HAVE_INTERLACED
425     h->param.b_interlaced = !!PARAM_INTERLACED;
426 #else
427     if( h->param.b_interlaced )
428     {
429         x264_log( h, X264_LOG_ERROR, "not compiled with interlaced support\n" );
430         return -1;
431     }
432 #endif
433
434     if( (h->param.crop_rect.i_left + h->param.crop_rect.i_right ) >= h->param.i_width ||
435         (h->param.crop_rect.i_top  + h->param.crop_rect.i_bottom) >= h->param.i_height )
436     {
437         x264_log( h, X264_LOG_ERROR, "invalid crop-rect %u,%u,%u,%u\n", h->param.crop_rect.i_left,
438                   h->param.crop_rect.i_top, h->param.crop_rect.i_right,  h->param.crop_rect.i_bottom );
439         return -1;
440     }
441
442     if( h->param.i_threads == X264_THREADS_AUTO )
443         h->param.i_threads = x264_cpu_num_processors() * (h->param.b_sliced_threads?2:3)/2;
444     h->param.i_threads = x264_clip3( h->param.i_threads, 1, X264_THREAD_MAX );
445     if( h->param.i_threads > 1 )
446     {
447 #if !HAVE_THREAD
448         x264_log( h, X264_LOG_WARNING, "not compiled with thread support!\n");
449         h->param.i_threads = 1;
450 #endif
451         /* Avoid absurdly small thread slices as they can reduce performance
452          * and VBV compliance.  Capped at an arbitrary 4 rows per thread. */
453         if( h->param.b_sliced_threads )
454         {
455             int max_threads = (h->param.i_height+15)/16 / 4;
456             h->param.i_threads = X264_MIN( h->param.i_threads, max_threads );
457         }
458     }
459     else
460         h->param.b_sliced_threads = 0;
461     h->i_thread_frames = h->param.b_sliced_threads ? 1 : h->param.i_threads;
462     if( h->i_thread_frames > 1 )
463         h->param.nalu_process = NULL;
464
465     h->param.i_keyint_max = x264_clip3( h->param.i_keyint_max, 1, X264_KEYINT_MAX_INFINITE );
466     if( h->param.i_keyint_max == 1 )
467     {
468         h->param.b_intra_refresh = 0;
469         h->param.analyse.i_weighted_pred = 0;
470     }
471
472     h->param.i_frame_packing = x264_clip3( h->param.i_frame_packing, -1, 5 );
473
474     /* Detect default ffmpeg settings and terminate with an error. */
475     if( b_open )
476     {
477         int score = 0;
478         score += h->param.analyse.i_me_range == 0;
479         score += h->param.rc.i_qp_step == 3;
480         score += h->param.i_keyint_max == 12;
481         score += h->param.rc.i_qp_min == 2;
482         score += h->param.rc.i_qp_max == 31;
483         score += h->param.rc.f_qcompress == 0.5;
484         score += fabs(h->param.rc.f_ip_factor - 1.25) < 0.01;
485         score += fabs(h->param.rc.f_pb_factor - 1.25) < 0.01;
486         score += h->param.analyse.inter == 0 && h->param.analyse.i_subpel_refine == 8;
487         if( score >= 5 )
488         {
489             x264_log( h, X264_LOG_ERROR, "broken ffmpeg default settings detected\n" );
490             x264_log( h, X264_LOG_ERROR, "use an encoding preset (e.g. -vpre medium)\n" );
491             x264_log( h, X264_LOG_ERROR, "preset usage: -vpre <speed> -vpre <profile>\n" );
492             x264_log( h, X264_LOG_ERROR, "speed presets are listed in x264 --help\n" );
493             x264_log( h, X264_LOG_ERROR, "profile is optional; x264 defaults to high\n" );
494             return -1;
495         }
496     }
497
498     if( h->param.rc.i_rc_method < 0 || h->param.rc.i_rc_method > 2 )
499     {
500         x264_log( h, X264_LOG_ERROR, "no ratecontrol method specified\n" );
501         return -1;
502     }
503     h->param.rc.f_rf_constant = x264_clip3f( h->param.rc.f_rf_constant, -QP_BD_OFFSET, 51 );
504     h->param.rc.f_rf_constant_max = x264_clip3f( h->param.rc.f_rf_constant_max, -QP_BD_OFFSET, 51 );
505     h->param.rc.i_qp_constant = x264_clip3( h->param.rc.i_qp_constant, 0, QP_MAX );
506     h->param.analyse.i_subpel_refine = x264_clip3( h->param.analyse.i_subpel_refine, 0, 10 );
507     h->param.rc.f_ip_factor = X264_MAX( h->param.rc.f_ip_factor, 0.01f );
508     h->param.rc.f_pb_factor = X264_MAX( h->param.rc.f_pb_factor, 0.01f );
509     if( h->param.rc.i_rc_method == X264_RC_CRF )
510     {
511         h->param.rc.i_qp_constant = h->param.rc.f_rf_constant + QP_BD_OFFSET;
512         h->param.rc.i_bitrate = 0;
513     }
514     if( (h->param.rc.i_rc_method == X264_RC_CQP || h->param.rc.i_rc_method == X264_RC_CRF)
515         && h->param.rc.i_qp_constant == 0 )
516     {
517         h->mb.b_lossless = 1;
518         h->param.i_cqm_preset = X264_CQM_FLAT;
519         h->param.psz_cqm_file = NULL;
520         h->param.rc.i_rc_method = X264_RC_CQP;
521         h->param.rc.f_ip_factor = 1;
522         h->param.rc.f_pb_factor = 1;
523         h->param.analyse.b_psnr = 0;
524         h->param.analyse.b_ssim = 0;
525         h->param.analyse.i_chroma_qp_offset = 0;
526         h->param.analyse.i_trellis = 0;
527         h->param.analyse.b_fast_pskip = 0;
528         h->param.analyse.i_noise_reduction = 0;
529         h->param.analyse.b_psy = 0;
530         h->param.i_bframe = 0;
531         /* 8x8dct is not useful without RD in CAVLC lossless */
532         if( !h->param.b_cabac && h->param.analyse.i_subpel_refine < 6 )
533             h->param.analyse.b_transform_8x8 = 0;
534     }
535     if( h->param.rc.i_rc_method == X264_RC_CQP )
536     {
537         float qp_p = h->param.rc.i_qp_constant;
538         float qp_i = qp_p - 6*log2f( h->param.rc.f_ip_factor );
539         float qp_b = qp_p + 6*log2f( h->param.rc.f_pb_factor );
540         h->param.rc.i_qp_min = x264_clip3( (int)(X264_MIN3( qp_p, qp_i, qp_b )), 0, QP_MAX );
541         h->param.rc.i_qp_max = x264_clip3( (int)(X264_MAX3( qp_p, qp_i, qp_b ) + .999), 0, QP_MAX );
542         h->param.rc.i_aq_mode = 0;
543         h->param.rc.b_mb_tree = 0;
544         h->param.rc.i_bitrate = 0;
545     }
546     h->param.rc.i_qp_max = x264_clip3( h->param.rc.i_qp_max, 0, QP_MAX );
547     h->param.rc.i_qp_min = x264_clip3( h->param.rc.i_qp_min, 0, h->param.rc.i_qp_max );
548     h->param.rc.i_qp_step = x264_clip3( h->param.rc.i_qp_step, 0, QP_MAX );
549     h->param.rc.i_bitrate = x264_clip3( h->param.rc.i_bitrate, 0, 2000000 );
550     h->param.rc.i_vbv_buffer_size = x264_clip3( h->param.rc.i_vbv_buffer_size, 0, 2000000 );
551     h->param.rc.i_vbv_max_bitrate = x264_clip3( h->param.rc.i_vbv_max_bitrate, 0, 2000000 );
552     h->param.rc.f_vbv_buffer_init = x264_clip3f( h->param.rc.f_vbv_buffer_init, 0, 2000000 );
553     if( h->param.rc.i_vbv_buffer_size )
554     {
555         if( h->param.rc.i_rc_method == X264_RC_CQP )
556         {
557             x264_log( h, X264_LOG_WARNING, "VBV is incompatible with constant QP, ignored.\n" );
558             h->param.rc.i_vbv_max_bitrate = 0;
559             h->param.rc.i_vbv_buffer_size = 0;
560         }
561         else if( h->param.rc.i_vbv_max_bitrate == 0 )
562         {
563             if( h->param.rc.i_rc_method == X264_RC_ABR )
564             {
565                 x264_log( h, X264_LOG_WARNING, "VBV maxrate unspecified, assuming CBR\n" );
566                 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate;
567             }
568             else
569             {
570                 x264_log( h, X264_LOG_WARNING, "VBV bufsize set but maxrate unspecified, ignored\n" );
571                 h->param.rc.i_vbv_buffer_size = 0;
572             }
573         }
574         else if( h->param.rc.i_vbv_max_bitrate < h->param.rc.i_bitrate &&
575                  h->param.rc.i_rc_method == X264_RC_ABR )
576         {
577             x264_log( h, X264_LOG_WARNING, "max bitrate less than average bitrate, assuming CBR\n" );
578             h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate;
579         }
580     }
581     else if( h->param.rc.i_vbv_max_bitrate )
582     {
583         x264_log( h, X264_LOG_WARNING, "VBV maxrate specified, but no bufsize, ignored\n" );
584         h->param.rc.i_vbv_max_bitrate = 0;
585     }
586
587     h->param.i_slice_max_size = X264_MAX( h->param.i_slice_max_size, 0 );
588     h->param.i_slice_max_mbs = X264_MAX( h->param.i_slice_max_mbs, 0 );
589
590     int max_slices = (h->param.i_height+((16<<PARAM_INTERLACED)-1))/(16<<PARAM_INTERLACED);
591     if( h->param.b_sliced_threads )
592         h->param.i_slice_count = x264_clip3( h->param.i_threads, 0, max_slices );
593     else
594     {
595         h->param.i_slice_count = x264_clip3( h->param.i_slice_count, 0, max_slices );
596         if( h->param.i_slice_max_mbs || h->param.i_slice_max_size )
597             h->param.i_slice_count = 0;
598     }
599
600     if( h->param.b_bluray_compat )
601     {
602         h->param.i_bframe_pyramid = X264_MIN( X264_B_PYRAMID_STRICT, h->param.i_bframe_pyramid );
603         h->param.i_bframe = X264_MIN( h->param.i_bframe, 3 );
604         h->param.b_aud = 1;
605         h->param.i_nal_hrd = X264_MAX( h->param.i_nal_hrd, X264_NAL_HRD_VBR );
606         h->param.i_slice_max_size = 0;
607         h->param.i_slice_max_mbs = 0;
608         h->param.b_intra_refresh = 0;
609         h->param.i_frame_reference = X264_MIN( h->param.i_frame_reference, 6 );
610         h->param.i_dpb_size = X264_MIN( h->param.i_dpb_size, 6 );
611         /* Due to the proliferation of broken players that don't handle dupes properly. */
612         h->param.analyse.i_weighted_pred = X264_MIN( h->param.analyse.i_weighted_pred, X264_WEIGHTP_SIMPLE );
613         if( h->param.b_fake_interlaced )
614             h->param.b_pic_struct = 1;
615     }
616
617     h->param.i_frame_reference = x264_clip3( h->param.i_frame_reference, 1, X264_REF_MAX );
618     h->param.i_dpb_size = x264_clip3( h->param.i_dpb_size, 1, X264_REF_MAX );
619     if( h->param.i_scenecut_threshold < 0 )
620         h->param.i_scenecut_threshold = 0;
621     h->param.analyse.i_direct_mv_pred = x264_clip3( h->param.analyse.i_direct_mv_pred, X264_DIRECT_PRED_NONE, X264_DIRECT_PRED_AUTO );
622     if( !h->param.analyse.i_subpel_refine && h->param.analyse.i_direct_mv_pred > X264_DIRECT_PRED_SPATIAL )
623     {
624         x264_log( h, X264_LOG_WARNING, "subme=0 + direct=temporal is not supported\n" );
625         h->param.analyse.i_direct_mv_pred = X264_DIRECT_PRED_SPATIAL;
626     }
627     h->param.i_bframe = x264_clip3( h->param.i_bframe, 0, X264_MIN( X264_BFRAME_MAX, h->param.i_keyint_max-1 ) );
628     h->param.i_bframe_bias = x264_clip3( h->param.i_bframe_bias, -90, 100 );
629     if( h->param.i_bframe <= 1 )
630         h->param.i_bframe_pyramid = X264_B_PYRAMID_NONE;
631     h->param.i_bframe_pyramid = x264_clip3( h->param.i_bframe_pyramid, X264_B_PYRAMID_NONE, X264_B_PYRAMID_NORMAL );
632     h->param.i_bframe_adaptive = x264_clip3( h->param.i_bframe_adaptive, X264_B_ADAPT_NONE, X264_B_ADAPT_TRELLIS );
633     if( !h->param.i_bframe )
634     {
635         h->param.i_bframe_adaptive = X264_B_ADAPT_NONE;
636         h->param.analyse.i_direct_mv_pred = 0;
637         h->param.analyse.b_weighted_bipred = 0;
638         h->param.b_open_gop = 0;
639     }
640     if( h->param.b_intra_refresh && h->param.i_bframe_pyramid == X264_B_PYRAMID_NORMAL )
641     {
642         x264_log( h, X264_LOG_WARNING, "b-pyramid normal + intra-refresh is not supported\n" );
643         h->param.i_bframe_pyramid = X264_B_PYRAMID_STRICT;
644     }
645     if( h->param.b_intra_refresh && (h->param.i_frame_reference > 1 || h->param.i_dpb_size > 1) )
646     {
647         x264_log( h, X264_LOG_WARNING, "ref > 1 + intra-refresh is not supported\n" );
648         h->param.i_frame_reference = 1;
649         h->param.i_dpb_size = 1;
650     }
651     if( h->param.b_intra_refresh && h->param.b_open_gop )
652     {
653         x264_log( h, X264_LOG_WARNING, "intra-refresh is not compatible with open-gop\n" );
654         h->param.b_open_gop = 0;
655     }
656     float fps = h->param.i_fps_num > 0 && h->param.i_fps_den > 0 ? (float) h->param.i_fps_num / h->param.i_fps_den : 25.0;
657     if( h->param.i_keyint_min == X264_KEYINT_MIN_AUTO )
658         h->param.i_keyint_min = X264_MIN( h->param.i_keyint_max / 10, fps );
659     h->param.i_keyint_min = x264_clip3( h->param.i_keyint_min, 1, h->param.i_keyint_max/2+1 );
660     h->param.rc.i_lookahead = x264_clip3( h->param.rc.i_lookahead, 0, X264_LOOKAHEAD_MAX );
661     {
662         int maxrate = X264_MAX( h->param.rc.i_vbv_max_bitrate, h->param.rc.i_bitrate );
663         float bufsize = maxrate ? (float)h->param.rc.i_vbv_buffer_size / maxrate : 0;
664         h->param.rc.i_lookahead = X264_MIN( h->param.rc.i_lookahead, X264_MAX( h->param.i_keyint_max, bufsize*fps ) );
665     }
666
667     if( !h->param.i_timebase_num || !h->param.i_timebase_den || !(h->param.b_vfr_input || h->param.b_pulldown) )
668     {
669         h->param.i_timebase_num = h->param.i_fps_den;
670         h->param.i_timebase_den = h->param.i_fps_num;
671     }
672
673     h->param.rc.f_qcompress = x264_clip3f( h->param.rc.f_qcompress, 0.0, 1.0 );
674     if( h->param.i_keyint_max == 1 || h->param.rc.f_qcompress == 1 )
675         h->param.rc.b_mb_tree = 0;
676     if( (!h->param.b_intra_refresh && h->param.i_keyint_max != X264_KEYINT_MAX_INFINITE) &&
677         !h->param.rc.i_lookahead && h->param.rc.b_mb_tree )
678     {
679         x264_log( h, X264_LOG_WARNING, "lookaheadless mb-tree requires intra refresh or infinite keyint\n" );
680         h->param.rc.b_mb_tree = 0;
681     }
682     if( h->param.rc.b_stat_read )
683         h->param.rc.i_lookahead = 0;
684 #if HAVE_THREAD
685     if( h->param.i_sync_lookahead < 0 )
686         h->param.i_sync_lookahead = h->param.i_bframe + 1;
687     h->param.i_sync_lookahead = X264_MIN( h->param.i_sync_lookahead, X264_LOOKAHEAD_MAX );
688     if( h->param.rc.b_stat_read || h->i_thread_frames == 1 )
689         h->param.i_sync_lookahead = 0;
690 #else
691     h->param.i_sync_lookahead = 0;
692 #endif
693
694     h->param.i_deblocking_filter_alphac0 = x264_clip3( h->param.i_deblocking_filter_alphac0, -6, 6 );
695     h->param.i_deblocking_filter_beta    = x264_clip3( h->param.i_deblocking_filter_beta, -6, 6 );
696     h->param.analyse.i_luma_deadzone[0] = x264_clip3( h->param.analyse.i_luma_deadzone[0], 0, 32 );
697     h->param.analyse.i_luma_deadzone[1] = x264_clip3( h->param.analyse.i_luma_deadzone[1], 0, 32 );
698
699     h->param.i_cabac_init_idc = x264_clip3( h->param.i_cabac_init_idc, 0, 2 );
700
701     if( h->param.i_cqm_preset < X264_CQM_FLAT || h->param.i_cqm_preset > X264_CQM_CUSTOM )
702         h->param.i_cqm_preset = X264_CQM_FLAT;
703
704     if( h->param.analyse.i_me_method < X264_ME_DIA ||
705         h->param.analyse.i_me_method > X264_ME_TESA )
706         h->param.analyse.i_me_method = X264_ME_HEX;
707     h->param.analyse.i_me_range = x264_clip3( h->param.analyse.i_me_range, 4, 1024 );
708     if( h->param.analyse.i_me_range > 16 && h->param.analyse.i_me_method <= X264_ME_HEX )
709         h->param.analyse.i_me_range = 16;
710     if( h->param.analyse.i_me_method == X264_ME_TESA &&
711         (h->mb.b_lossless || h->param.analyse.i_subpel_refine <= 1) )
712         h->param.analyse.i_me_method = X264_ME_ESA;
713     h->param.analyse.b_mixed_references = h->param.analyse.b_mixed_references && h->param.i_frame_reference > 1;
714     h->param.analyse.inter &= X264_ANALYSE_PSUB16x16|X264_ANALYSE_PSUB8x8|X264_ANALYSE_BSUB16x16|
715                               X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
716     h->param.analyse.intra &= X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
717     if( !(h->param.analyse.inter & X264_ANALYSE_PSUB16x16) )
718         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
719     if( !h->param.analyse.b_transform_8x8 )
720     {
721         h->param.analyse.inter &= ~X264_ANALYSE_I8x8;
722         h->param.analyse.intra &= ~X264_ANALYSE_I8x8;
723     }
724     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -12, 12);
725     h->param.analyse.i_trellis = x264_clip3( h->param.analyse.i_trellis, 0, 2 );
726     h->param.rc.i_aq_mode = x264_clip3( h->param.rc.i_aq_mode, 0, 2 );
727     h->param.rc.f_aq_strength = x264_clip3f( h->param.rc.f_aq_strength, 0, 3 );
728     if( h->param.rc.f_aq_strength == 0 )
729         h->param.rc.i_aq_mode = 0;
730
731     if( h->param.i_log_level < X264_LOG_INFO )
732     {
733         h->param.analyse.b_psnr = 0;
734         h->param.analyse.b_ssim = 0;
735     }
736     /* Warn users trying to measure PSNR/SSIM with psy opts on. */
737     if( b_open && (h->param.analyse.b_psnr || h->param.analyse.b_ssim) )
738     {
739         char *s = NULL;
740
741         if( h->param.analyse.b_psy )
742         {
743             s = h->param.analyse.b_psnr ? "psnr" : "ssim";
744             x264_log( h, X264_LOG_WARNING, "--%s used with psy on: results will be invalid!\n", s );
745         }
746         else if( !h->param.rc.i_aq_mode && h->param.analyse.b_ssim )
747         {
748             x264_log( h, X264_LOG_WARNING, "--ssim used with AQ off: results will be invalid!\n" );
749             s = "ssim";
750         }
751         else if(  h->param.rc.i_aq_mode && h->param.analyse.b_psnr )
752         {
753             x264_log( h, X264_LOG_WARNING, "--psnr used with AQ on: results will be invalid!\n" );
754             s = "psnr";
755         }
756         if( s )
757             x264_log( h, X264_LOG_WARNING, "--tune %s should be used if attempting to benchmark %s!\n", s, s );
758     }
759
760     if( !h->param.analyse.b_psy )
761     {
762         h->param.analyse.f_psy_rd = 0;
763         h->param.analyse.f_psy_trellis = 0;
764     }
765     h->param.analyse.f_psy_rd = x264_clip3f( h->param.analyse.f_psy_rd, 0, 10 );
766     h->param.analyse.f_psy_trellis = x264_clip3f( h->param.analyse.f_psy_trellis, 0, 10 );
767     h->mb.i_psy_rd = h->param.analyse.i_subpel_refine >= 6 ? FIX8( h->param.analyse.f_psy_rd ) : 0;
768     h->mb.i_psy_trellis = h->param.analyse.i_trellis ? FIX8( h->param.analyse.f_psy_trellis / 4 ) : 0;
769     /* In 4:4:4 mode, chroma gets twice as much resolution, so we can halve its quality. */
770     if( b_open && i_csp >= X264_CSP_I444 && i_csp < X264_CSP_BGR && h->param.analyse.b_psy )
771         h->param.analyse.i_chroma_qp_offset += 6;
772     /* Psy RDO increases overall quantizers to improve the quality of luma--this indirectly hurts chroma quality */
773     /* so we lower the chroma QP offset to compensate */
774     if( b_open && h->mb.i_psy_rd )
775         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_rd < 0.25 ? 1 : 2;
776     /* Psy trellis has a similar effect. */
777     if( b_open && h->mb.i_psy_trellis )
778         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_trellis < 0.25 ? 1 : 2;
779     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -12, 12);
780     /* MB-tree requires AQ to be on, even if the strength is zero. */
781     if( !h->param.rc.i_aq_mode && h->param.rc.b_mb_tree )
782     {
783         h->param.rc.i_aq_mode = 1;
784         h->param.rc.f_aq_strength = 0;
785     }
786     h->param.analyse.i_noise_reduction = x264_clip3( h->param.analyse.i_noise_reduction, 0, 1<<16 );
787     if( h->param.analyse.i_subpel_refine == 10 && (h->param.analyse.i_trellis != 2 || !h->param.rc.i_aq_mode) )
788         h->param.analyse.i_subpel_refine = 9;
789
790     {
791         const x264_level_t *l = x264_levels;
792         if( h->param.i_level_idc < 0 )
793         {
794             int maxrate_bak = h->param.rc.i_vbv_max_bitrate;
795             if( h->param.rc.i_rc_method == X264_RC_ABR && h->param.rc.i_vbv_buffer_size <= 0 )
796                 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate * 2;
797             x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
798             do h->param.i_level_idc = l->level_idc;
799                 while( l[1].level_idc && x264_validate_levels( h, 0 ) && l++ );
800             h->param.rc.i_vbv_max_bitrate = maxrate_bak;
801         }
802         else
803         {
804             while( l->level_idc && l->level_idc != h->param.i_level_idc )
805                 l++;
806             if( l->level_idc == 0 )
807             {
808                 x264_log( h, X264_LOG_ERROR, "invalid level_idc: %d\n", h->param.i_level_idc );
809                 return -1;
810             }
811         }
812         if( h->param.analyse.i_mv_range <= 0 )
813             h->param.analyse.i_mv_range = l->mv_range >> PARAM_INTERLACED;
814         else
815             h->param.analyse.i_mv_range = x264_clip3(h->param.analyse.i_mv_range, 32, 512 >> PARAM_INTERLACED);
816     }
817
818     h->param.analyse.i_weighted_pred = x264_clip3( h->param.analyse.i_weighted_pred, X264_WEIGHTP_NONE, X264_WEIGHTP_SMART );
819
820     if( PARAM_INTERLACED )
821     {
822         if( h->param.analyse.i_me_method >= X264_ME_ESA )
823         {
824             x264_log( h, X264_LOG_WARNING, "interlace + me=esa is not implemented\n" );
825             h->param.analyse.i_me_method = X264_ME_UMH;
826         }
827         if( h->param.analyse.i_weighted_pred > 0 )
828         {
829             x264_log( h, X264_LOG_WARNING, "interlace + weightp is not implemented\n" );
830             h->param.analyse.i_weighted_pred = X264_WEIGHTP_NONE;
831         }
832     }
833
834     if( !h->param.analyse.i_weighted_pred && h->param.rc.b_mb_tree && h->param.analyse.b_psy )
835         h->param.analyse.i_weighted_pred = X264_WEIGHTP_FAKE;
836
837     if( h->i_thread_frames > 1 )
838     {
839         int r = h->param.analyse.i_mv_range_thread;
840         int r2;
841         if( r <= 0 )
842         {
843             // half of the available space is reserved and divided evenly among the threads,
844             // the rest is allocated to whichever thread is far enough ahead to use it.
845             // reserving more space increases quality for some videos, but costs more time
846             // in thread synchronization.
847             int max_range = (h->param.i_height + X264_THREAD_HEIGHT) / h->i_thread_frames - X264_THREAD_HEIGHT;
848             r = max_range / 2;
849         }
850         r = X264_MAX( r, h->param.analyse.i_me_range );
851         r = X264_MIN( r, h->param.analyse.i_mv_range );
852         // round up to use the whole mb row
853         r2 = (r & ~15) + ((-X264_THREAD_HEIGHT) & 15);
854         if( r2 < r )
855             r2 += 16;
856         x264_log( h, X264_LOG_DEBUG, "using mv_range_thread = %d\n", r2 );
857         h->param.analyse.i_mv_range_thread = r2;
858     }
859
860     if( h->param.rc.f_rate_tolerance < 0 )
861         h->param.rc.f_rate_tolerance = 0;
862     if( h->param.rc.f_qblur < 0 )
863         h->param.rc.f_qblur = 0;
864     if( h->param.rc.f_complexity_blur < 0 )
865         h->param.rc.f_complexity_blur = 0;
866
867     h->param.i_sps_id &= 31;
868
869     if( PARAM_INTERLACED )
870         h->param.b_pic_struct = 1;
871
872     h->param.i_nal_hrd = x264_clip3( h->param.i_nal_hrd, X264_NAL_HRD_NONE, X264_NAL_HRD_CBR );
873
874     if( h->param.i_nal_hrd && !h->param.rc.i_vbv_buffer_size )
875     {
876         x264_log( h, X264_LOG_WARNING, "NAL HRD parameters require VBV parameters\n" );
877         h->param.i_nal_hrd = X264_NAL_HRD_NONE;
878     }
879
880     if( h->param.i_nal_hrd == X264_NAL_HRD_CBR &&
881        (h->param.rc.i_bitrate != h->param.rc.i_vbv_max_bitrate || !h->param.rc.i_vbv_max_bitrate) )
882     {
883         x264_log( h, X264_LOG_WARNING, "CBR HRD requires constant bitrate\n" );
884         h->param.i_nal_hrd = X264_NAL_HRD_VBR;
885     }
886
887     /* ensure the booleans are 0 or 1 so they can be used in math */
888 #define BOOLIFY(x) h->param.x = !!h->param.x
889     BOOLIFY( b_cabac );
890     BOOLIFY( b_constrained_intra );
891     BOOLIFY( b_deblocking_filter );
892     BOOLIFY( b_deterministic );
893     BOOLIFY( b_sliced_threads );
894     BOOLIFY( b_interlaced );
895     BOOLIFY( b_intra_refresh );
896     BOOLIFY( b_visualize );
897     BOOLIFY( b_aud );
898     BOOLIFY( b_repeat_headers );
899     BOOLIFY( b_annexb );
900     BOOLIFY( b_vfr_input );
901     BOOLIFY( b_pulldown );
902     BOOLIFY( b_tff );
903     BOOLIFY( b_pic_struct );
904     BOOLIFY( b_fake_interlaced );
905     BOOLIFY( b_open_gop );
906     BOOLIFY( b_bluray_compat );
907     BOOLIFY( analyse.b_transform_8x8 );
908     BOOLIFY( analyse.b_weighted_bipred );
909     BOOLIFY( analyse.b_chroma_me );
910     BOOLIFY( analyse.b_mixed_references );
911     BOOLIFY( analyse.b_fast_pskip );
912     BOOLIFY( analyse.b_dct_decimate );
913     BOOLIFY( analyse.b_psy );
914     BOOLIFY( analyse.b_psnr );
915     BOOLIFY( analyse.b_ssim );
916     BOOLIFY( rc.b_stat_write );
917     BOOLIFY( rc.b_stat_read );
918     BOOLIFY( rc.b_mb_tree );
919 #undef BOOLIFY
920
921     return 0;
922 }
923
924 static void mbcmp_init( x264_t *h )
925 {
926     int satd = !h->mb.b_lossless && h->param.analyse.i_subpel_refine > 1;
927     memcpy( h->pixf.mbcmp, satd ? h->pixf.satd : h->pixf.sad_aligned, sizeof(h->pixf.mbcmp) );
928     memcpy( h->pixf.mbcmp_unaligned, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.mbcmp_unaligned) );
929     h->pixf.intra_mbcmp_x3_16x16 = satd ? h->pixf.intra_satd_x3_16x16 : h->pixf.intra_sad_x3_16x16;
930     h->pixf.intra_mbcmp_x3_8x8c = satd ? h->pixf.intra_satd_x3_8x8c : h->pixf.intra_sad_x3_8x8c;
931     h->pixf.intra_mbcmp_x3_8x8 = satd ? h->pixf.intra_sa8d_x3_8x8 : h->pixf.intra_sad_x3_8x8;
932     h->pixf.intra_mbcmp_x3_4x4 = satd ? h->pixf.intra_satd_x3_4x4 : h->pixf.intra_sad_x3_4x4;
933     satd &= h->param.analyse.i_me_method == X264_ME_TESA;
934     memcpy( h->pixf.fpelcmp, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.fpelcmp) );
935     memcpy( h->pixf.fpelcmp_x3, satd ? h->pixf.satd_x3 : h->pixf.sad_x3, sizeof(h->pixf.fpelcmp_x3) );
936     memcpy( h->pixf.fpelcmp_x4, satd ? h->pixf.satd_x4 : h->pixf.sad_x4, sizeof(h->pixf.fpelcmp_x4) );
937 }
938
939 static void x264_set_aspect_ratio( x264_t *h, x264_param_t *param, int initial )
940 {
941     /* VUI */
942     if( param->vui.i_sar_width > 0 && param->vui.i_sar_height > 0 )
943     {
944         uint32_t i_w = param->vui.i_sar_width;
945         uint32_t i_h = param->vui.i_sar_height;
946         uint32_t old_w = h->param.vui.i_sar_width;
947         uint32_t old_h = h->param.vui.i_sar_height;
948
949         x264_reduce_fraction( &i_w, &i_h );
950
951         while( i_w > 65535 || i_h > 65535 )
952         {
953             i_w /= 2;
954             i_h /= 2;
955         }
956
957         x264_reduce_fraction( &i_w, &i_h );
958
959         if( i_w != old_w || i_h != old_h || initial )
960         {
961             h->param.vui.i_sar_width = 0;
962             h->param.vui.i_sar_height = 0;
963             if( i_w == 0 || i_h == 0 )
964                 x264_log( h, X264_LOG_WARNING, "cannot create valid sample aspect ratio\n" );
965             else
966             {
967                 x264_log( h, initial?X264_LOG_INFO:X264_LOG_DEBUG, "using SAR=%d/%d\n", i_w, i_h );
968                 h->param.vui.i_sar_width = i_w;
969                 h->param.vui.i_sar_height = i_h;
970             }
971             x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
972         }
973     }
974 }
975
976 /****************************************************************************
977  * x264_encoder_open:
978  ****************************************************************************/
979 x264_t *x264_encoder_open( x264_param_t *param )
980 {
981     x264_t *h;
982     char buf[1000], *p;
983     int qp, i_slicetype_length;
984
985     CHECKED_MALLOCZERO( h, sizeof(x264_t) );
986
987     /* Create a copy of param */
988     memcpy( &h->param, param, sizeof(x264_param_t) );
989
990     if( param->param_free )
991         param->param_free( param );
992
993     if( x264_threading_init() )
994     {
995         x264_log( h, X264_LOG_ERROR, "unable to initialize threading\n" );
996         goto fail;
997     }
998
999     if( x264_validate_parameters( h, 1 ) < 0 )
1000         goto fail;
1001
1002     if( h->param.psz_cqm_file )
1003         if( x264_cqm_parse_file( h, h->param.psz_cqm_file ) < 0 )
1004             goto fail;
1005
1006     if( h->param.rc.psz_stat_out )
1007         h->param.rc.psz_stat_out = strdup( h->param.rc.psz_stat_out );
1008     if( h->param.rc.psz_stat_in )
1009         h->param.rc.psz_stat_in = strdup( h->param.rc.psz_stat_in );
1010
1011     x264_reduce_fraction( &h->param.i_fps_num, &h->param.i_fps_den );
1012     x264_reduce_fraction( &h->param.i_timebase_num, &h->param.i_timebase_den );
1013
1014     /* Init x264_t */
1015     h->i_frame = -1;
1016     h->i_frame_num = 0;
1017     h->i_idr_pic_id = 0;
1018
1019     if( (uint64_t)h->param.i_timebase_den * 2 > UINT32_MAX )
1020     {
1021         x264_log( h, X264_LOG_ERROR, "Effective timebase denominator %u exceeds H.264 maximum\n", h->param.i_timebase_den );
1022         goto fail;
1023     }
1024
1025     x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
1026     x264_pps_init( h->pps, h->param.i_sps_id, &h->param, h->sps );
1027
1028     x264_set_aspect_ratio( h, &h->param, 1 );
1029
1030     x264_validate_levels( h, 1 );
1031
1032     h->chroma_qp_table = i_chroma_qp_table + 12 + h->pps->i_chroma_qp_index_offset;
1033
1034     if( x264_cqm_init( h ) < 0 )
1035         goto fail;
1036
1037     h->mb.i_mb_width = h->sps->i_mb_width;
1038     h->mb.i_mb_height = h->sps->i_mb_height;
1039     h->mb.i_mb_count = h->mb.i_mb_width * h->mb.i_mb_height;
1040     /* Adaptive MBAFF and subme 0 are not supported as we require halving motion
1041      * vectors during prediction, resulting in hpel mvs.
1042      * The chosen solution is to make MBAFF non-adaptive in this case. */
1043     h->mb.b_adaptive_mbaff = PARAM_INTERLACED && h->param.analyse.i_subpel_refine;
1044
1045     /* Init frames. */
1046     if( h->param.i_bframe_adaptive == X264_B_ADAPT_TRELLIS && !h->param.rc.b_stat_read )
1047         h->frames.i_delay = X264_MAX(h->param.i_bframe,3)*4;
1048     else
1049         h->frames.i_delay = h->param.i_bframe;
1050     if( h->param.rc.b_mb_tree || h->param.rc.i_vbv_buffer_size )
1051         h->frames.i_delay = X264_MAX( h->frames.i_delay, h->param.rc.i_lookahead );
1052     i_slicetype_length = h->frames.i_delay;
1053     h->frames.i_delay += h->i_thread_frames - 1;
1054     h->frames.i_delay += h->param.i_sync_lookahead;
1055     h->frames.i_delay += h->param.b_vfr_input;
1056     h->frames.i_bframe_delay = h->param.i_bframe ? (h->param.i_bframe_pyramid ? 2 : 1) : 0;
1057
1058     h->frames.i_max_ref0 = h->param.i_frame_reference;
1059     h->frames.i_max_ref1 = X264_MIN( h->sps->vui.i_num_reorder_frames, h->param.i_frame_reference );
1060     h->frames.i_max_dpb  = h->sps->vui.i_max_dec_frame_buffering;
1061     h->frames.b_have_lowres = !h->param.rc.b_stat_read
1062         && ( h->param.rc.i_rc_method == X264_RC_ABR
1063           || h->param.rc.i_rc_method == X264_RC_CRF
1064           || h->param.i_bframe_adaptive
1065           || h->param.i_scenecut_threshold
1066           || h->param.rc.b_mb_tree
1067           || h->param.analyse.i_weighted_pred );
1068     h->frames.b_have_lowres |= h->param.rc.b_stat_read && h->param.rc.i_vbv_buffer_size > 0;
1069     h->frames.b_have_sub8x8_esa = !!(h->param.analyse.inter & X264_ANALYSE_PSUB8x8);
1070
1071     h->frames.i_last_idr =
1072     h->frames.i_last_keyframe = - h->param.i_keyint_max;
1073     h->frames.i_input    = 0;
1074     h->frames.i_largest_pts = h->frames.i_second_largest_pts = -1;
1075     h->frames.i_poc_last_open_gop = -1;
1076
1077     CHECKED_MALLOCZERO( h->frames.unused[0], (h->frames.i_delay + 3) * sizeof(x264_frame_t *) );
1078     /* Allocate room for max refs plus a few extra just in case. */
1079     CHECKED_MALLOCZERO( h->frames.unused[1], (h->i_thread_frames + X264_REF_MAX + 4) * sizeof(x264_frame_t *) );
1080     CHECKED_MALLOCZERO( h->frames.current, (h->param.i_sync_lookahead + h->param.i_bframe
1081                         + h->i_thread_frames + 3) * sizeof(x264_frame_t *) );
1082     if( h->param.analyse.i_weighted_pred > 0 )
1083         CHECKED_MALLOCZERO( h->frames.blank_unused, h->i_thread_frames * 4 * sizeof(x264_frame_t *) );
1084     h->i_ref[0] = h->i_ref[1] = 0;
1085     h->i_cpb_delay = h->i_coded_fields = h->i_disp_fields = 0;
1086     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);
1087     h->i_disp_fields_last_frame = -1;
1088     x264_rdo_init();
1089
1090     /* init CPU functions */
1091     x264_predict_16x16_init( h->param.cpu, h->predict_16x16 );
1092     x264_predict_8x8c_init( h->param.cpu, h->predict_8x8c );
1093     x264_predict_8x8_init( h->param.cpu, h->predict_8x8, &h->predict_8x8_filter );
1094     x264_predict_4x4_init( h->param.cpu, h->predict_4x4 );
1095     if( h->param.b_cabac )
1096         x264_cabac_init( h );
1097     else
1098         x264_cavlc_init();
1099     x264_pixel_init( h->param.cpu, &h->pixf );
1100     x264_dct_init( h->param.cpu, &h->dctf );
1101     x264_zigzag_init( h->param.cpu, &h->zigzagf_progressive, &h->zigzagf_interlaced );
1102     memcpy( &h->zigzagf, PARAM_INTERLACED ? &h->zigzagf_interlaced : &h->zigzagf_progressive, sizeof(h->zigzagf) );
1103     x264_mc_init( h->param.cpu, &h->mc );
1104     x264_quant_init( h, h->param.cpu, &h->quantf );
1105     x264_deblock_init( h->param.cpu, &h->loopf, PARAM_INTERLACED );
1106     x264_bitstream_init( h->param.cpu, &h->bsf );
1107     x264_dct_init_weights();
1108
1109     mbcmp_init( h );
1110
1111     p = buf + sprintf( buf, "using cpu capabilities:" );
1112     for( int i = 0; x264_cpu_names[i].flags; i++ )
1113     {
1114         if( !strcmp(x264_cpu_names[i].name, "SSE2")
1115             && h->param.cpu & (X264_CPU_SSE2_IS_FAST|X264_CPU_SSE2_IS_SLOW) )
1116             continue;
1117         if( !strcmp(x264_cpu_names[i].name, "SSE3")
1118             && (h->param.cpu & X264_CPU_SSSE3 || !(h->param.cpu & X264_CPU_CACHELINE_64)) )
1119             continue;
1120         if( !strcmp(x264_cpu_names[i].name, "SSE4.1")
1121             && (h->param.cpu & X264_CPU_SSE42) )
1122             continue;
1123         if( (h->param.cpu & x264_cpu_names[i].flags) == x264_cpu_names[i].flags
1124             && (!i || x264_cpu_names[i].flags != x264_cpu_names[i-1].flags) )
1125             p += sprintf( p, " %s", x264_cpu_names[i].name );
1126     }
1127     if( !h->param.cpu )
1128         p += sprintf( p, " none!" );
1129     x264_log( h, X264_LOG_INFO, "%s\n", buf );
1130
1131     float *logs = x264_analyse_prepare_costs( h );
1132     if( !logs )
1133         goto fail;
1134     for( qp = X264_MIN( h->param.rc.i_qp_min, QP_MAX_SPEC ); qp <= h->param.rc.i_qp_max; qp++ )
1135         if( x264_analyse_init_costs( h, logs, qp ) )
1136             goto fail;
1137     if( x264_analyse_init_costs( h, logs, X264_LOOKAHEAD_QP ) )
1138         goto fail;
1139     x264_free( logs );
1140
1141     static const uint16_t cost_mv_correct[7] = { 24, 47, 95, 189, 379, 757, 1515 };
1142     /* Checks for known miscompilation issues. */
1143     if( h->cost_mv[X264_LOOKAHEAD_QP][2013] != cost_mv_correct[BIT_DEPTH-8] )
1144     {
1145         x264_log( h, X264_LOG_ERROR, "MV cost test failed: x264 has been miscompiled!\n" );
1146         goto fail;
1147     }
1148
1149     /* Must be volatile or else GCC will optimize it out. */
1150     volatile int temp = 392;
1151     if( x264_clz( temp ) != 23 )
1152     {
1153         x264_log( h, X264_LOG_ERROR, "CLZ test failed: x264 has been miscompiled!\n" );
1154 #if ARCH_X86 || ARCH_X86_64
1155         x264_log( h, X264_LOG_ERROR, "Are you attempting to run an SSE4a-targeted build on a CPU that\n" );
1156         x264_log( h, X264_LOG_ERROR, "doesn't support it?\n" );
1157 #endif
1158         goto fail;
1159     }
1160
1161     h->out.i_nal = 0;
1162     h->out.i_bitstream = X264_MAX( 1000000, h->param.i_width * h->param.i_height * 4
1163         * ( h->param.rc.i_rc_method == X264_RC_ABR ? pow( 0.95, h->param.rc.i_qp_min )
1164           : pow( 0.95, h->param.rc.i_qp_constant ) * X264_MAX( 1, h->param.rc.f_ip_factor )));
1165
1166     h->nal_buffer_size = h->out.i_bitstream * 3/2 + 4;
1167     CHECKED_MALLOC( h->nal_buffer, h->nal_buffer_size );
1168
1169     if( h->param.i_threads > 1 &&
1170         x264_threadpool_init( &h->threadpool, h->param.i_threads, (void*)x264_encoder_thread_init, h ) )
1171         goto fail;
1172
1173     h->thread[0] = h;
1174     for( int i = 1; i < h->param.i_threads + !!h->param.i_sync_lookahead; i++ )
1175         CHECKED_MALLOC( h->thread[i], sizeof(x264_t) );
1176
1177     for( int i = 0; i < h->param.i_threads; i++ )
1178     {
1179         int init_nal_count = h->param.i_slice_count + 3;
1180         int allocate_threadlocal_data = !h->param.b_sliced_threads || !i;
1181         if( i > 0 )
1182             *h->thread[i] = *h;
1183
1184         if( allocate_threadlocal_data )
1185         {
1186             h->thread[i]->fdec = x264_frame_pop_unused( h, 1 );
1187             if( !h->thread[i]->fdec )
1188                 goto fail;
1189         }
1190         else
1191             h->thread[i]->fdec = h->thread[0]->fdec;
1192
1193         CHECKED_MALLOC( h->thread[i]->out.p_bitstream, h->out.i_bitstream );
1194         /* Start each thread with room for init_nal_count NAL units; it'll realloc later if needed. */
1195         CHECKED_MALLOC( h->thread[i]->out.nal, init_nal_count*sizeof(x264_nal_t) );
1196         h->thread[i]->out.i_nals_allocated = init_nal_count;
1197
1198         if( allocate_threadlocal_data && x264_macroblock_cache_allocate( h->thread[i] ) < 0 )
1199             goto fail;
1200     }
1201
1202     if( x264_lookahead_init( h, i_slicetype_length ) )
1203         goto fail;
1204
1205     for( int i = 0; i < h->param.i_threads; i++ )
1206         if( x264_macroblock_thread_allocate( h->thread[i], 0 ) < 0 )
1207             goto fail;
1208
1209     if( x264_ratecontrol_new( h ) < 0 )
1210         goto fail;
1211
1212     if( h->param.i_nal_hrd )
1213     {
1214         x264_log( h, X264_LOG_DEBUG, "HRD bitrate: %i bits/sec\n", h->sps->vui.hrd.i_bit_rate_unscaled );
1215         x264_log( h, X264_LOG_DEBUG, "CPB size: %i bits\n", h->sps->vui.hrd.i_cpb_size_unscaled );
1216     }
1217
1218     if( h->param.psz_dump_yuv )
1219     {
1220         /* create or truncate the reconstructed video file */
1221         FILE *f = fopen( h->param.psz_dump_yuv, "w" );
1222         if( !f )
1223         {
1224             x264_log( h, X264_LOG_ERROR, "dump_yuv: can't write to %s\n", h->param.psz_dump_yuv );
1225             goto fail;
1226         }
1227         else if( !x264_is_regular_file( f ) )
1228         {
1229             x264_log( h, X264_LOG_ERROR, "dump_yuv: incompatible with non-regular file %s\n", h->param.psz_dump_yuv );
1230             goto fail;
1231         }
1232         fclose( f );
1233     }
1234
1235     const char *profile = h->sps->i_profile_idc == PROFILE_BASELINE ? "Constrained Baseline" :
1236                           h->sps->i_profile_idc == PROFILE_MAIN ? "Main" :
1237                           h->sps->i_profile_idc == PROFILE_HIGH ? "High" :
1238                           h->sps->i_profile_idc == PROFILE_HIGH10 ? (h->sps->b_constraint_set3 == 1 ? "High 10 Intra" : "High 10") :
1239                           h->sps->b_constraint_set3 == 1 ? "High 4:4:4 Intra" : "High 4:4:4 Predictive";
1240     char level[4];
1241     snprintf( level, sizeof(level), "%d.%d", h->sps->i_level_idc/10, h->sps->i_level_idc%10 );
1242     if( h->sps->i_level_idc == 9 || ( h->sps->i_level_idc == 11 && h->sps->b_constraint_set3 &&
1243         (h->sps->i_profile_idc >= PROFILE_BASELINE && h->sps->i_profile_idc <= PROFILE_EXTENDED) ) )
1244         strcpy( level, "1b" );
1245
1246     if( h->sps->i_profile_idc < PROFILE_HIGH10 )
1247     {
1248         x264_log( h, X264_LOG_INFO, "profile %s, level %s\n",
1249             profile, level );
1250     }
1251     else
1252     {
1253         x264_log( h, X264_LOG_INFO, "profile %s, level %s, %s %d-bit\n",
1254             profile, level, CHROMA444 ? "4:4:4" : "4:2:0", BIT_DEPTH );
1255     }
1256
1257     return h;
1258 fail:
1259     x264_free( h );
1260     return NULL;
1261 }
1262
1263 /****************************************************************************
1264  * x264_encoder_reconfig:
1265  ****************************************************************************/
1266 int x264_encoder_reconfig( x264_t *h, x264_param_t *param )
1267 {
1268     int rc_reconfig = 0;
1269     h = h->thread[h->thread[0]->i_thread_phase];
1270     x264_set_aspect_ratio( h, param, 0 );
1271 #define COPY(var) h->param.var = param->var
1272     COPY( i_frame_reference ); // but never uses more refs than initially specified
1273     COPY( i_bframe_bias );
1274     if( h->param.i_scenecut_threshold )
1275         COPY( i_scenecut_threshold ); // can't turn it on or off, only vary the threshold
1276     COPY( b_deblocking_filter );
1277     COPY( i_deblocking_filter_alphac0 );
1278     COPY( i_deblocking_filter_beta );
1279     COPY( i_frame_packing );
1280     COPY( analyse.inter );
1281     COPY( analyse.intra );
1282     COPY( analyse.i_direct_mv_pred );
1283     /* Scratch buffer prevents me_range from being increased for esa/tesa */
1284     if( h->param.analyse.i_me_method < X264_ME_ESA || param->analyse.i_me_range < h->param.analyse.i_me_range )
1285         COPY( analyse.i_me_range );
1286     COPY( analyse.i_noise_reduction );
1287     /* We can't switch out of subme=0 during encoding. */
1288     if( h->param.analyse.i_subpel_refine )
1289         COPY( analyse.i_subpel_refine );
1290     COPY( analyse.i_trellis );
1291     COPY( analyse.b_chroma_me );
1292     COPY( analyse.b_dct_decimate );
1293     COPY( analyse.b_fast_pskip );
1294     COPY( analyse.b_mixed_references );
1295     COPY( analyse.f_psy_rd );
1296     COPY( analyse.f_psy_trellis );
1297     COPY( crop_rect );
1298     // can only twiddle these if they were enabled to begin with:
1299     if( h->param.analyse.i_me_method >= X264_ME_ESA || param->analyse.i_me_method < X264_ME_ESA )
1300         COPY( analyse.i_me_method );
1301     if( h->param.analyse.i_me_method >= X264_ME_ESA && !h->frames.b_have_sub8x8_esa )
1302         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
1303     if( h->pps->b_transform_8x8_mode )
1304         COPY( analyse.b_transform_8x8 );
1305     if( h->frames.i_max_ref1 > 1 )
1306         COPY( i_bframe_pyramid );
1307     COPY( i_slice_max_size );
1308     COPY( i_slice_max_mbs );
1309     COPY( i_slice_count );
1310     COPY( b_tff );
1311
1312     /* VBV can't be turned on if it wasn't on to begin with */
1313     if( h->param.rc.i_vbv_max_bitrate > 0 && h->param.rc.i_vbv_buffer_size > 0 &&
1314           param->rc.i_vbv_max_bitrate > 0 &&   param->rc.i_vbv_buffer_size > 0 )
1315     {
1316         rc_reconfig |= h->param.rc.i_vbv_max_bitrate != param->rc.i_vbv_max_bitrate;
1317         rc_reconfig |= h->param.rc.i_vbv_buffer_size != param->rc.i_vbv_buffer_size;
1318         rc_reconfig |= h->param.rc.i_bitrate != param->rc.i_bitrate;
1319         COPY( rc.i_vbv_max_bitrate );
1320         COPY( rc.i_vbv_buffer_size );
1321         COPY( rc.i_bitrate );
1322     }
1323     rc_reconfig |= h->param.rc.f_rf_constant != param->rc.f_rf_constant;
1324     rc_reconfig |= h->param.rc.f_rf_constant_max != param->rc.f_rf_constant_max;
1325     COPY( rc.f_rf_constant );
1326     COPY( rc.f_rf_constant_max );
1327 #undef COPY
1328
1329     mbcmp_init( h );
1330
1331     int ret = x264_validate_parameters( h, 0 );
1332
1333     /* Supported reconfiguration options (1-pass only):
1334      * vbv-maxrate
1335      * vbv-bufsize
1336      * crf
1337      * bitrate (CBR only) */
1338     if( !ret && rc_reconfig )
1339         x264_ratecontrol_init_reconfigurable( h, 0 );
1340
1341     return ret;
1342 }
1343
1344 /****************************************************************************
1345  * x264_encoder_parameters:
1346  ****************************************************************************/
1347 void x264_encoder_parameters( x264_t *h, x264_param_t *param )
1348 {
1349     memcpy( param, &h->thread[h->i_thread_phase]->param, sizeof(x264_param_t) );
1350 }
1351
1352 /* internal usage */
1353 static void x264_nal_start( x264_t *h, int i_type, int i_ref_idc )
1354 {
1355     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1356
1357     nal->i_ref_idc        = i_ref_idc;
1358     nal->i_type           = i_type;
1359     nal->b_long_startcode = 1;
1360
1361     nal->i_payload= 0;
1362     nal->p_payload= &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8];
1363 }
1364
1365 /* if number of allocated nals is not enough, re-allocate a larger one. */
1366 static int x264_nal_check_buffer( x264_t *h )
1367 {
1368     if( h->out.i_nal >= h->out.i_nals_allocated )
1369     {
1370         x264_nal_t *new_out = x264_malloc( sizeof(x264_nal_t) * (h->out.i_nals_allocated*2) );
1371         if( !new_out )
1372             return -1;
1373         memcpy( new_out, h->out.nal, sizeof(x264_nal_t) * (h->out.i_nals_allocated) );
1374         x264_free( h->out.nal );
1375         h->out.nal = new_out;
1376         h->out.i_nals_allocated *= 2;
1377     }
1378     return 0;
1379 }
1380
1381 static int x264_nal_end( x264_t *h )
1382 {
1383     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1384     nal->i_payload = &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8] - nal->p_payload;
1385     if( h->param.nalu_process )
1386         h->param.nalu_process( h, nal );
1387     h->out.i_nal++;
1388
1389     return x264_nal_check_buffer( h );
1390 }
1391
1392 static int x264_encoder_encapsulate_nals( x264_t *h, int start )
1393 {
1394     int nal_size = 0, previous_nal_size = 0;
1395
1396     if( h->param.nalu_process )
1397     {
1398         for( int i = start; i < h->out.i_nal; i++ )
1399             nal_size += h->out.nal[i].i_payload;
1400         return nal_size;
1401     }
1402
1403     for( int i = 0; i < start; i++ )
1404         previous_nal_size += h->out.nal[i].i_payload;
1405
1406     for( int i = start; i < h->out.i_nal; i++ )
1407         nal_size += h->out.nal[i].i_payload;
1408
1409     /* Worst-case NAL unit escaping: reallocate the buffer if it's too small. */
1410     int necessary_size = nal_size * 3/2 + h->out.i_nal * 4;
1411     if( h->nal_buffer_size < necessary_size )
1412     {
1413         h->nal_buffer_size = necessary_size * 2;
1414         uint8_t *buf = x264_malloc( h->nal_buffer_size );
1415         if( !buf )
1416             return -1;
1417         if( previous_nal_size )
1418             memcpy( buf, h->nal_buffer, previous_nal_size );
1419         x264_free( h->nal_buffer );
1420         h->nal_buffer = buf;
1421     }
1422
1423     uint8_t *nal_buffer = h->nal_buffer + previous_nal_size;
1424
1425     for( int i = start; i < h->out.i_nal; i++ )
1426     {
1427         h->out.nal[i].b_long_startcode = !i || h->out.nal[i].i_type == NAL_SPS || h->out.nal[i].i_type == NAL_PPS;
1428         x264_nal_encode( h, nal_buffer, &h->out.nal[i] );
1429         nal_buffer += h->out.nal[i].i_payload;
1430     }
1431
1432     x264_emms();
1433
1434     return nal_buffer - (h->nal_buffer + previous_nal_size);
1435 }
1436
1437 /****************************************************************************
1438  * x264_encoder_headers:
1439  ****************************************************************************/
1440 int x264_encoder_headers( x264_t *h, x264_nal_t **pp_nal, int *pi_nal )
1441 {
1442     int frame_size = 0;
1443     /* init bitstream context */
1444     h->out.i_nal = 0;
1445     bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
1446
1447     /* Write SEI, SPS and PPS. */
1448
1449     /* generate sequence parameters */
1450     x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
1451     x264_sps_write( &h->out.bs, h->sps );
1452     if( x264_nal_end( h ) )
1453         return -1;
1454
1455     /* generate picture parameters */
1456     x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
1457     x264_pps_write( &h->out.bs, h->sps, h->pps );
1458     if( x264_nal_end( h ) )
1459         return -1;
1460
1461     /* identify ourselves */
1462     x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
1463     if( x264_sei_version_write( h, &h->out.bs ) )
1464         return -1;
1465     if( x264_nal_end( h ) )
1466         return -1;
1467
1468     frame_size = x264_encoder_encapsulate_nals( h, 0 );
1469     if( frame_size < 0 )
1470         return -1;
1471
1472     /* now set output*/
1473     *pi_nal = h->out.i_nal;
1474     *pp_nal = &h->out.nal[0];
1475     h->out.i_nal = 0;
1476
1477     return frame_size;
1478 }
1479
1480 /* Check to see whether we have chosen a reference list ordering different
1481  * from the standard's default. */
1482 static inline void x264_reference_check_reorder( x264_t *h )
1483 {
1484     /* The reorder check doesn't check for missing frames, so just
1485      * force a reorder if one of the reference list is corrupt. */
1486     for( int i = 0; h->frames.reference[i]; i++ )
1487         if( h->frames.reference[i]->b_corrupt )
1488         {
1489             h->b_ref_reorder[0] = 1;
1490             return;
1491         }
1492     for( int list = 0; list <= (h->sh.i_type == SLICE_TYPE_B); list++ )
1493         for( int i = 0; i < h->i_ref[list] - 1; i++ )
1494         {
1495             int framenum_diff = h->fref[list][i+1]->i_frame_num - h->fref[list][i]->i_frame_num;
1496             int poc_diff = h->fref[list][i+1]->i_poc - h->fref[list][i]->i_poc;
1497             /* P and B-frames use different default orders. */
1498             if( h->sh.i_type == SLICE_TYPE_P ? framenum_diff > 0 : list == 1 ? poc_diff < 0 : poc_diff > 0 )
1499             {
1500                 h->b_ref_reorder[list] = 1;
1501                 return;
1502             }
1503         }
1504 }
1505
1506 /* return -1 on failure, else return the index of the new reference frame */
1507 int x264_weighted_reference_duplicate( x264_t *h, int i_ref, const x264_weight_t *w )
1508 {
1509     int i = h->i_ref[0];
1510     int j = 1;
1511     x264_frame_t *newframe;
1512     if( i <= 1 ) /* empty list, definitely can't duplicate frame */
1513         return -1;
1514
1515     //Duplication is only used in X264_WEIGHTP_SMART
1516     if( h->param.analyse.i_weighted_pred != X264_WEIGHTP_SMART )
1517         return -1;
1518
1519     /* Duplication is a hack to compensate for crappy rounding in motion compensation.
1520      * With high bit depth, it's not worth doing, so turn it off except in the case of
1521      * unweighted dupes. */
1522     if( BIT_DEPTH > 8 && w != weight_none )
1523         return -1;
1524
1525     newframe = x264_frame_pop_blank_unused( h );
1526     if( !newframe )
1527         return -1;
1528
1529     //FIXME: probably don't need to copy everything
1530     *newframe = *h->fref[0][i_ref];
1531     newframe->i_reference_count = 1;
1532     newframe->orig = h->fref[0][i_ref];
1533     newframe->b_duplicate = 1;
1534     memcpy( h->fenc->weight[j], w, sizeof(h->fenc->weight[i]) );
1535
1536     /* shift the frames to make space for the dupe. */
1537     h->b_ref_reorder[0] = 1;
1538     if( h->i_ref[0] < X264_REF_MAX )
1539         ++h->i_ref[0];
1540     h->fref[0][X264_REF_MAX-1] = NULL;
1541     x264_frame_unshift( &h->fref[0][j], newframe );
1542
1543     return j;
1544 }
1545
1546 static void x264_weighted_pred_init( x264_t *h )
1547 {
1548     /* for now no analysis and set all weights to nothing */
1549     for( int i_ref = 0; i_ref < h->i_ref[0]; i_ref++ )
1550         h->fenc->weighted[i_ref] = h->fref[0][i_ref]->filtered[0][0];
1551
1552     // FIXME: This only supports weighting of one reference frame
1553     // and duplicates of that frame.
1554     h->fenc->i_lines_weighted = 0;
1555
1556     for( int i_ref = 0; i_ref < (h->i_ref[0] << SLICE_MBAFF); i_ref++ )
1557         for( int i = 0; i < 3; i++ )
1558             h->sh.weight[i_ref][i].weightfn = NULL;
1559
1560
1561     if( h->sh.i_type != SLICE_TYPE_P || h->param.analyse.i_weighted_pred <= 0 )
1562         return;
1563
1564     int i_padv = PADV << PARAM_INTERLACED;
1565     int denom = -1;
1566     int weightplane[2] = { 0, 0 };
1567     int buffer_next = 0;
1568     for( int i = 0; i < 3; i++ )
1569     {
1570         for( int j = 0; j < h->i_ref[0]; j++ )
1571         {
1572             if( h->fenc->weight[j][i].weightfn )
1573             {
1574                 h->sh.weight[j][i] = h->fenc->weight[j][i];
1575                 // if weight is useless, don't write it to stream
1576                 if( h->sh.weight[j][i].i_scale == 1<<h->sh.weight[j][i].i_denom && h->sh.weight[j][i].i_offset == 0 )
1577                     h->sh.weight[j][i].weightfn = NULL;
1578                 else
1579                 {
1580                     if( !weightplane[!!i] )
1581                     {
1582                         weightplane[!!i] = 1;
1583                         h->sh.weight[0][!!i].i_denom = denom = h->sh.weight[j][i].i_denom;
1584                         assert( x264_clip3( denom, 0, 7 ) == denom );
1585                     }
1586
1587                     assert( h->sh.weight[j][i].i_denom == denom );
1588                     if( !i )
1589                     {
1590                         h->fenc->weighted[j] = h->mb.p_weight_buf[buffer_next++] + h->fenc->i_stride[0] * i_padv + PADH;
1591                         //scale full resolution frame
1592                         if( h->param.i_threads == 1 )
1593                         {
1594                             pixel *src = h->fref[0][j]->filtered[0][0] - h->fref[0][j]->i_stride[0]*i_padv - PADH;
1595                             pixel *dst = h->fenc->weighted[j] - h->fenc->i_stride[0]*i_padv - PADH;
1596                             int stride = h->fenc->i_stride[0];
1597                             int width = h->fenc->i_width[0] + PADH*2;
1598                             int height = h->fenc->i_lines[0] + i_padv*2;
1599                             x264_weight_scale_plane( h, dst, stride, src, stride, width, height, &h->sh.weight[j][0] );
1600                             h->fenc->i_lines_weighted = height;
1601                         }
1602                     }
1603                 }
1604             }
1605         }
1606     }
1607
1608     if( weightplane[1] )
1609         for( int i = 0; i < h->i_ref[0]; i++ )
1610         {
1611             if( h->sh.weight[i][1].weightfn && !h->sh.weight[i][2].weightfn )
1612             {
1613                 h->sh.weight[i][2].i_scale = 1 << h->sh.weight[0][1].i_denom;
1614                 h->sh.weight[i][2].i_offset = 0;
1615             }
1616             else if( h->sh.weight[i][2].weightfn && !h->sh.weight[i][1].weightfn )
1617             {
1618                 h->sh.weight[i][1].i_scale = 1 << h->sh.weight[0][1].i_denom;
1619                 h->sh.weight[i][1].i_offset = 0;
1620             }
1621         }
1622
1623     if( !weightplane[0] )
1624         h->sh.weight[0][0].i_denom = 0;
1625     if( !weightplane[1] )
1626         h->sh.weight[0][1].i_denom = 0;
1627     h->sh.weight[0][2].i_denom = h->sh.weight[0][1].i_denom;
1628 }
1629
1630 static inline int x264_reference_distance( x264_t *h, x264_frame_t *frame )
1631 {
1632     if( h->param.i_frame_packing == 5 )
1633         return abs((h->fenc->i_frame&~1) - (frame->i_frame&~1)) +
1634                   ((h->fenc->i_frame&1) != (frame->i_frame&1));
1635     else
1636         return abs(h->fenc->i_frame - frame->i_frame);
1637 }
1638
1639 static inline void x264_reference_build_list( x264_t *h, int i_poc )
1640 {
1641     int b_ok;
1642
1643     /* build ref list 0/1 */
1644     h->mb.pic.i_fref[0] = h->i_ref[0] = 0;
1645     h->mb.pic.i_fref[1] = h->i_ref[1] = 0;
1646     if( h->sh.i_type == SLICE_TYPE_I )
1647         return;
1648
1649     for( int i = 0; h->frames.reference[i]; i++ )
1650     {
1651         if( h->frames.reference[i]->b_corrupt )
1652             continue;
1653         if( h->frames.reference[i]->i_poc < i_poc )
1654             h->fref[0][h->i_ref[0]++] = h->frames.reference[i];
1655         else if( h->frames.reference[i]->i_poc > i_poc )
1656             h->fref[1][h->i_ref[1]++] = h->frames.reference[i];
1657     }
1658
1659     /* Order reference lists by distance from the current frame. */
1660     for( int list = 0; list < 2; list++ )
1661     {
1662         h->fref_nearest[list] = h->fref[list][0];
1663         do
1664         {
1665             b_ok = 1;
1666             for( int i = 0; i < h->i_ref[list] - 1; i++ )
1667             {
1668                 if( list ? h->fref[list][i+1]->i_poc < h->fref_nearest[list]->i_poc
1669                          : h->fref[list][i+1]->i_poc > h->fref_nearest[list]->i_poc )
1670                     h->fref_nearest[list] = h->fref[list][i+1];
1671                 if( x264_reference_distance( h, h->fref[list][i] ) > x264_reference_distance( h, h->fref[list][i+1] ) )
1672                 {
1673                     XCHG( x264_frame_t*, h->fref[list][i], h->fref[list][i+1] );
1674                     b_ok = 0;
1675                     break;
1676                 }
1677             }
1678         } while( !b_ok );
1679     }
1680
1681     if( h->sh.i_mmco_remove_from_end )
1682         for( int i = h->i_ref[0]-1; i >= h->i_ref[0] - h->sh.i_mmco_remove_from_end; i-- )
1683         {
1684             int diff = h->i_frame_num - h->fref[0][i]->i_frame_num;
1685             h->sh.mmco[h->sh.i_mmco_command_count].i_poc = h->fref[0][i]->i_poc;
1686             h->sh.mmco[h->sh.i_mmco_command_count++].i_difference_of_pic_nums = diff;
1687         }
1688
1689     x264_reference_check_reorder( h );
1690
1691     h->i_ref[1] = X264_MIN( h->i_ref[1], h->frames.i_max_ref1 );
1692     h->i_ref[0] = X264_MIN( h->i_ref[0], h->frames.i_max_ref0 );
1693     h->i_ref[0] = X264_MIN( h->i_ref[0], h->param.i_frame_reference ); // if reconfig() has lowered the limit
1694
1695     /* For Blu-ray compliance, don't reference frames outside of the minigop. */
1696     if( IS_X264_TYPE_B( h->fenc->i_type ) && h->param.b_bluray_compat )
1697         h->i_ref[0] = X264_MIN( h->i_ref[0], IS_X264_TYPE_B( h->fref[0][0]->i_type ) + 1 );
1698
1699     /* add duplicates */
1700     if( h->fenc->i_type == X264_TYPE_P )
1701     {
1702         int idx = -1;
1703         if( h->param.analyse.i_weighted_pred >= X264_WEIGHTP_SIMPLE )
1704         {
1705             x264_weight_t w[3];
1706             w[1].weightfn = w[2].weightfn = NULL;
1707             if( h->param.rc.b_stat_read )
1708                 x264_ratecontrol_set_weights( h, h->fenc );
1709
1710             if( !h->fenc->weight[0][0].weightfn )
1711             {
1712                 h->fenc->weight[0][0].i_denom = 0;
1713                 SET_WEIGHT( w[0], 1, 1, 0, -1 );
1714                 idx = x264_weighted_reference_duplicate( h, 0, w );
1715             }
1716             else
1717             {
1718                 if( h->fenc->weight[0][0].i_scale == 1<<h->fenc->weight[0][0].i_denom )
1719                 {
1720                     SET_WEIGHT( h->fenc->weight[0][0], 1, 1, 0, h->fenc->weight[0][0].i_offset );
1721                 }
1722                 x264_weighted_reference_duplicate( h, 0, weight_none );
1723                 if( h->fenc->weight[0][0].i_offset > -128 )
1724                 {
1725                     w[0] = h->fenc->weight[0][0];
1726                     w[0].i_offset--;
1727                     h->mc.weight_cache( h, &w[0] );
1728                     idx = x264_weighted_reference_duplicate( h, 0, w );
1729                 }
1730             }
1731         }
1732         h->mb.ref_blind_dupe = idx;
1733     }
1734
1735     assert( h->i_ref[0] + h->i_ref[1] <= X264_REF_MAX );
1736     h->mb.pic.i_fref[0] = h->i_ref[0];
1737     h->mb.pic.i_fref[1] = h->i_ref[1];
1738 }
1739
1740 static void x264_fdec_filter_row( x264_t *h, int mb_y, int b_inloop )
1741 {
1742     /* mb_y is the mb to be encoded next, not the mb to be filtered here */
1743     int b_hpel = h->fdec->b_kept_as_ref;
1744     int b_deblock = h->sh.i_disable_deblocking_filter_idc != 1;
1745     int b_end = mb_y == h->i_threadslice_end;
1746     int b_measure_quality = 1;
1747     int min_y = mb_y - (1 << SLICE_MBAFF);
1748     int b_start = min_y == h->i_threadslice_start;
1749     /* Even in interlaced mode, deblocking never modifies more than 4 pixels
1750      * above each MB, as bS=4 doesn't happen for the top of interlaced mbpairs. */
1751     int minpix_y = min_y*16 - 4 * !b_start;
1752     int maxpix_y = mb_y*16 - 4 * !b_end;
1753     b_deblock &= b_hpel || h->param.psz_dump_yuv;
1754     if( h->param.b_sliced_threads && b_start && min_y && !b_inloop )
1755     {
1756         b_deblock = 0;         /* We already deblocked on the inloop pass. */
1757         b_measure_quality = 0; /* We already measured quality on the inloop pass. */
1758     }
1759     if( mb_y & SLICE_MBAFF )
1760         return;
1761     if( min_y < h->i_threadslice_start )
1762         return;
1763
1764     if( b_deblock )
1765         for( int y = min_y; y < mb_y; y += (1 << SLICE_MBAFF) )
1766             x264_frame_deblock_row( h, y );
1767
1768     /* FIXME: Prediction requires different borders for interlaced/progressive mc,
1769      * but the actual image data is equivalent. For now, maintain this
1770      * consistency by copying deblocked pixels between planes. */
1771     if( PARAM_INTERLACED )
1772         for( int p = 0; p < h->fdec->i_plane; p++ )
1773             for( int i = minpix_y>>(!CHROMA444 && p); i < maxpix_y>>(!CHROMA444 && p); i++ )
1774                 memcpy( h->fdec->plane_fld[p] + i*h->fdec->i_stride[p],
1775                         h->fdec->plane[p]     + i*h->fdec->i_stride[p],
1776                         h->mb.i_mb_width*16*sizeof(pixel) );
1777
1778     if( b_hpel )
1779     {
1780         int end = mb_y == h->mb.i_mb_height;
1781         x264_frame_expand_border( h, h->fdec, min_y, end );
1782         if( h->param.analyse.i_subpel_refine )
1783         {
1784             x264_frame_filter( h, h->fdec, min_y, end );
1785             x264_frame_expand_border_filtered( h, h->fdec, min_y, end );
1786         }
1787     }
1788
1789     if( SLICE_MBAFF )
1790         for( int i = 0; i < 3; i++ )
1791         {
1792             XCHG( pixel *, h->intra_border_backup[0][i], h->intra_border_backup[3][i] );
1793             XCHG( pixel *, h->intra_border_backup[1][i], h->intra_border_backup[4][i] );
1794         }
1795
1796     if( h->i_thread_frames > 1 && h->fdec->b_kept_as_ref )
1797         x264_frame_cond_broadcast( h->fdec, mb_y*16 + (b_end ? 10000 : -(X264_THREAD_HEIGHT << SLICE_MBAFF)) );
1798
1799     if( b_measure_quality )
1800     {
1801         maxpix_y = X264_MIN( maxpix_y, h->param.i_height );
1802         if( h->param.analyse.b_psnr )
1803         {
1804             for( int p = 0; p < (CHROMA444 ? 3 : 1); p++ )
1805                 h->stat.frame.i_ssd[p] += x264_pixel_ssd_wxh( &h->pixf,
1806                     h->fdec->plane[p] + minpix_y * h->fdec->i_stride[p], h->fdec->i_stride[p],
1807                     h->fenc->plane[p] + minpix_y * h->fenc->i_stride[p], h->fenc->i_stride[p],
1808                     h->param.i_width, maxpix_y-minpix_y );
1809             if( !CHROMA444 )
1810             {
1811                 uint64_t ssd_u, ssd_v;
1812                 x264_pixel_ssd_nv12( &h->pixf,
1813                     h->fdec->plane[1] + (minpix_y>>1) * h->fdec->i_stride[1], h->fdec->i_stride[1],
1814                     h->fenc->plane[1] + (minpix_y>>1) * h->fenc->i_stride[1], h->fenc->i_stride[1],
1815                     h->param.i_width>>1, (maxpix_y-minpix_y)>>1, &ssd_u, &ssd_v );
1816                 h->stat.frame.i_ssd[1] += ssd_u;
1817                 h->stat.frame.i_ssd[2] += ssd_v;
1818             }
1819         }
1820
1821         if( h->param.analyse.b_ssim )
1822         {
1823             int ssim_cnt;
1824             x264_emms();
1825             /* offset by 2 pixels to avoid alignment of ssim blocks with dct blocks,
1826              * and overlap by 4 */
1827             minpix_y += b_start ? 2 : -6;
1828             h->stat.frame.f_ssim +=
1829                 x264_pixel_ssim_wxh( &h->pixf,
1830                     h->fdec->plane[0] + 2+minpix_y*h->fdec->i_stride[0], h->fdec->i_stride[0],
1831                     h->fenc->plane[0] + 2+minpix_y*h->fenc->i_stride[0], h->fenc->i_stride[0],
1832                     h->param.i_width-2, maxpix_y-minpix_y, h->scratch_buffer, &ssim_cnt );
1833             h->stat.frame.i_ssim_cnt += ssim_cnt;
1834         }
1835     }
1836 }
1837
1838 static inline int x264_reference_update( x264_t *h )
1839 {
1840     if( !h->fdec->b_kept_as_ref )
1841     {
1842         if( h->i_thread_frames > 1 )
1843         {
1844             x264_frame_push_unused( h, h->fdec );
1845             h->fdec = x264_frame_pop_unused( h, 1 );
1846             if( !h->fdec )
1847                 return -1;
1848         }
1849         return 0;
1850     }
1851
1852     /* apply mmco from previous frame. */
1853     for( int i = 0; i < h->sh.i_mmco_command_count; i++ )
1854         for( int j = 0; h->frames.reference[j]; j++ )
1855             if( h->frames.reference[j]->i_poc == h->sh.mmco[i].i_poc )
1856                 x264_frame_push_unused( h, x264_frame_shift( &h->frames.reference[j] ) );
1857
1858     /* move frame in the buffer */
1859     x264_frame_push( h->frames.reference, h->fdec );
1860     if( h->frames.reference[h->sps->i_num_ref_frames] )
1861         x264_frame_push_unused( h, x264_frame_shift( h->frames.reference ) );
1862     h->fdec = x264_frame_pop_unused( h, 1 );
1863     if( !h->fdec )
1864         return -1;
1865     return 0;
1866 }
1867
1868 static inline void x264_reference_reset( x264_t *h )
1869 {
1870     while( h->frames.reference[0] )
1871         x264_frame_push_unused( h, x264_frame_pop( h->frames.reference ) );
1872     h->fdec->i_poc =
1873     h->fenc->i_poc = 0;
1874 }
1875
1876 static inline void x264_reference_hierarchy_reset( x264_t *h )
1877 {
1878     int ref;
1879     int b_hasdelayframe = 0;
1880
1881     /* look for delay frames -- chain must only contain frames that are disposable */
1882     for( int i = 0; h->frames.current[i] && IS_DISPOSABLE( h->frames.current[i]->i_type ); i++ )
1883         b_hasdelayframe |= h->frames.current[i]->i_coded
1884                         != h->frames.current[i]->i_frame + h->sps->vui.i_num_reorder_frames;
1885
1886     /* This function must handle b-pyramid and clear frames for open-gop */
1887     if( h->param.i_bframe_pyramid != X264_B_PYRAMID_STRICT && !b_hasdelayframe && h->frames.i_poc_last_open_gop == -1 )
1888         return;
1889
1890     /* Remove last BREF. There will never be old BREFs in the
1891      * dpb during a BREF decode when pyramid == STRICT */
1892     for( ref = 0; h->frames.reference[ref]; ref++ )
1893     {
1894         if( ( h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT
1895             && h->frames.reference[ref]->i_type == X264_TYPE_BREF )
1896             || ( h->frames.reference[ref]->i_poc < h->frames.i_poc_last_open_gop
1897             && h->sh.i_type != SLICE_TYPE_B ) )
1898         {
1899             int diff = h->i_frame_num - h->frames.reference[ref]->i_frame_num;
1900             h->sh.mmco[h->sh.i_mmco_command_count].i_difference_of_pic_nums = diff;
1901             h->sh.mmco[h->sh.i_mmco_command_count++].i_poc = h->frames.reference[ref]->i_poc;
1902             x264_frame_push_unused( h, x264_frame_shift( &h->frames.reference[ref] ) );
1903             h->b_ref_reorder[0] = 1;
1904             ref--;
1905         }
1906     }
1907
1908     /* Prepare room in the dpb for the delayed display time of the later b-frame's */
1909     if( h->param.i_bframe_pyramid )
1910         h->sh.i_mmco_remove_from_end = X264_MAX( ref + 2 - h->frames.i_max_dpb, 0 );
1911 }
1912
1913 static inline void x264_slice_init( x264_t *h, int i_nal_type, int i_global_qp )
1914 {
1915     /* ------------------------ Create slice header  ----------------------- */
1916     if( i_nal_type == NAL_SLICE_IDR )
1917     {
1918         x264_slice_header_init( h, &h->sh, h->sps, h->pps, h->i_idr_pic_id, h->i_frame_num, i_global_qp );
1919
1920         /* alternate id */
1921         h->i_idr_pic_id ^= 1;
1922     }
1923     else
1924     {
1925         x264_slice_header_init( h, &h->sh, h->sps, h->pps, -1, h->i_frame_num, i_global_qp );
1926
1927         h->sh.i_num_ref_idx_l0_active = h->i_ref[0] <= 0 ? 1 : h->i_ref[0];
1928         h->sh.i_num_ref_idx_l1_active = h->i_ref[1] <= 0 ? 1 : h->i_ref[1];
1929         if( h->sh.i_num_ref_idx_l0_active != h->pps->i_num_ref_idx_l0_default_active ||
1930             (h->sh.i_type == SLICE_TYPE_B && h->sh.i_num_ref_idx_l1_active != h->pps->i_num_ref_idx_l1_default_active) )
1931         {
1932             h->sh.b_num_ref_idx_override = 1;
1933         }
1934     }
1935
1936     if( h->fenc->i_type == X264_TYPE_BREF && h->param.b_bluray_compat && h->sh.i_mmco_command_count )
1937     {
1938         h->b_sh_backup = 1;
1939         h->sh_backup = h->sh;
1940     }
1941
1942     h->fdec->i_frame_num = h->sh.i_frame_num;
1943
1944     if( h->sps->i_poc_type == 0 )
1945     {
1946         h->sh.i_poc = h->fdec->i_poc;
1947         if( PARAM_INTERLACED )
1948         {
1949             h->sh.i_delta_poc_bottom = h->param.b_tff ? 1 : -1;
1950             h->sh.i_poc += h->sh.i_delta_poc_bottom == -1;
1951         }
1952         else
1953             h->sh.i_delta_poc_bottom = 0;
1954         h->fdec->i_delta_poc[0] = h->sh.i_delta_poc_bottom == -1;
1955         h->fdec->i_delta_poc[1] = h->sh.i_delta_poc_bottom ==  1;
1956     }
1957     else
1958     {
1959         /* Nothing to do ? */
1960     }
1961
1962     x264_macroblock_slice_init( h );
1963 }
1964
1965 static int x264_slice_write( x264_t *h )
1966 {
1967     int i_skip;
1968     int mb_xy, i_mb_x, i_mb_y;
1969     int i_skip_bak = 0; /* Shut up GCC. */
1970     bs_t UNINIT(bs_bak);
1971     x264_cabac_t cabac_bak;
1972     uint8_t cabac_prevbyte_bak = 0; /* Shut up GCC. */
1973     int mv_bits_bak = 0;
1974     int tex_bits_bak = 0;
1975     /* NALUs other than the first use a 3-byte startcode.
1976      * Add one extra byte for the rbsp, and one more for the final CABAC putbyte.
1977      * Then add an extra 5 bytes just in case, to account for random NAL escapes and
1978      * other inaccuracies. */
1979     int overhead_guess = (NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal)) + 1 + h->param.b_cabac + 5;
1980     int slice_max_size = h->param.i_slice_max_size > 0 ? (h->param.i_slice_max_size-overhead_guess)*8 : 0;
1981     int back_up_bitstream = slice_max_size || (!h->param.b_cabac && h->sps->i_profile_idc < PROFILE_HIGH);
1982     int starting_bits = bs_pos(&h->out.bs);
1983     int b_deblock = h->sh.i_disable_deblocking_filter_idc != 1;
1984     int b_hpel = h->fdec->b_kept_as_ref;
1985     uint8_t *last_emu_check;
1986     b_deblock &= b_hpel || h->param.psz_dump_yuv;
1987     bs_realign( &h->out.bs );
1988
1989     /* Slice */
1990     x264_nal_start( h, h->i_nal_type, h->i_nal_ref_idc );
1991     h->out.nal[h->out.i_nal].i_first_mb = h->sh.i_first_mb;
1992
1993     /* Slice header */
1994     x264_macroblock_thread_init( h );
1995
1996     /* If this isn't the first slice in the threadslice, set the slice QP
1997      * equal to the last QP in the previous slice for more accurate
1998      * CABAC initialization. */
1999     if( h->sh.i_first_mb != h->i_threadslice_start * h->mb.i_mb_width )
2000     {
2001         h->sh.i_qp = h->mb.i_last_qp;
2002         h->sh.i_qp_delta = h->sh.i_qp - h->pps->i_pic_init_qp;
2003     }
2004
2005     x264_slice_header_write( &h->out.bs, &h->sh, h->i_nal_ref_idc );
2006     if( h->param.b_cabac )
2007     {
2008         /* alignment needed */
2009         bs_align_1( &h->out.bs );
2010
2011         /* init cabac */
2012         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 );
2013         x264_cabac_encode_init ( &h->cabac, h->out.bs.p, h->out.bs.p_end );
2014         last_emu_check = h->cabac.p;
2015     }
2016     else
2017         last_emu_check = h->out.bs.p;
2018     h->mb.i_last_qp = h->sh.i_qp;
2019     h->mb.i_last_dqp = 0;
2020     h->mb.field_decoding_flag = 0;
2021
2022     i_mb_y = h->sh.i_first_mb / h->mb.i_mb_width;
2023     i_mb_x = h->sh.i_first_mb % h->mb.i_mb_width;
2024     i_skip = 0;
2025
2026     while( 1 )
2027     {
2028         mb_xy = i_mb_x + i_mb_y * h->mb.i_mb_width;
2029         int mb_spos = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
2030
2031         if( !(i_mb_y & SLICE_MBAFF) )
2032         {
2033             if( x264_bitstream_check_buffer( h ) )
2034                 return -1;
2035
2036             if( back_up_bitstream )
2037             {
2038                 mv_bits_bak = h->stat.frame.i_mv_bits;
2039                 tex_bits_bak = h->stat.frame.i_tex_bits;
2040                 /* We don't need the contexts because flushing the CABAC encoder has no context
2041                  * dependency and macroblocks are only re-encoded in the case where a slice is
2042                  * ended (and thus the content of all contexts are thrown away). */
2043                 if( h->param.b_cabac )
2044                 {
2045                     memcpy( &cabac_bak, &h->cabac, offsetof(x264_cabac_t, f8_bits_encoded) );
2046                     /* x264's CABAC writer modifies the previous byte during carry, so it has to be
2047                      * backed up. */
2048                     cabac_prevbyte_bak = h->cabac.p[-1];
2049                 }
2050                 else
2051                 {
2052                     bs_bak = h->out.bs;
2053                     i_skip_bak = i_skip;
2054                 }
2055             }
2056         }
2057
2058         if( i_mb_x == 0 && !h->mb.b_reencode_mb )
2059             x264_fdec_filter_row( h, i_mb_y, 1 );
2060
2061         if( PARAM_INTERLACED )
2062         {
2063             if( h->mb.b_adaptive_mbaff )
2064             {
2065                 if( !(i_mb_y&1) )
2066                 {
2067                     /* FIXME: VSAD is fast but fairly poor at choosing the best interlace type. */
2068                     h->mb.b_interlaced = x264_field_vsad( h, i_mb_x, i_mb_y );
2069                     memcpy( &h->zigzagf, MB_INTERLACED ? &h->zigzagf_interlaced : &h->zigzagf_progressive, sizeof(h->zigzagf) );
2070                     if( !MB_INTERLACED && (i_mb_y+2) == h->mb.i_mb_height )
2071                         x264_expand_border_mbpair( h, i_mb_x, i_mb_y );
2072                 }
2073             }
2074             h->mb.field[mb_xy] = MB_INTERLACED;
2075         }
2076
2077         /* load cache */
2078         if( SLICE_MBAFF )
2079             x264_macroblock_cache_load_interlaced( h, i_mb_x, i_mb_y );
2080         else
2081             x264_macroblock_cache_load_progressive( h, i_mb_x, i_mb_y );
2082
2083         x264_macroblock_analyse( h );
2084
2085         /* encode this macroblock -> be careful it can change the mb type to P_SKIP if needed */
2086 reencode:
2087         x264_macroblock_encode( h );
2088
2089         if( h->param.b_cabac )
2090         {
2091             if( mb_xy > h->sh.i_first_mb && !(SLICE_MBAFF && (i_mb_y&1)) )
2092                 x264_cabac_encode_terminal( &h->cabac );
2093
2094             if( IS_SKIP( h->mb.i_type ) )
2095                 x264_cabac_mb_skip( h, 1 );
2096             else
2097             {
2098                 if( h->sh.i_type != SLICE_TYPE_I )
2099                     x264_cabac_mb_skip( h, 0 );
2100                 x264_macroblock_write_cabac( h, &h->cabac );
2101             }
2102         }
2103         else
2104         {
2105             if( IS_SKIP( h->mb.i_type ) )
2106                 i_skip++;
2107             else
2108             {
2109                 if( h->sh.i_type != SLICE_TYPE_I )
2110                 {
2111                     bs_write_ue( &h->out.bs, i_skip );  /* skip run */
2112                     i_skip = 0;
2113                 }
2114                 x264_macroblock_write_cavlc( h );
2115                 /* If there was a CAVLC level code overflow, try again at a higher QP. */
2116                 if( h->mb.b_overflow )
2117                 {
2118                     h->mb.i_chroma_qp = h->chroma_qp_table[++h->mb.i_qp];
2119                     h->mb.i_skip_intra = 0;
2120                     h->mb.b_skip_mc = 0;
2121                     h->mb.b_overflow = 0;
2122                     h->out.bs = bs_bak;
2123                     i_skip = i_skip_bak;
2124                     h->stat.frame.i_mv_bits = mv_bits_bak;
2125                     h->stat.frame.i_tex_bits = tex_bits_bak;
2126                     goto reencode;
2127                 }
2128             }
2129         }
2130
2131         int total_bits = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
2132         int mb_size = total_bits - mb_spos;
2133
2134         if( slice_max_size )
2135         {
2136             /* Count the skip run, just in case. */
2137             if( !h->param.b_cabac )
2138                 total_bits += bs_size_ue_big( i_skip );
2139             /* Check for escape bytes. */
2140             uint8_t *end = h->param.b_cabac ? h->cabac.p : h->out.bs.p;
2141             for( ; last_emu_check < end - 2; last_emu_check++ )
2142                 if( last_emu_check[0] == 0 && last_emu_check[1] == 0 && last_emu_check[2] <= 3 )
2143                 {
2144                     slice_max_size -= 8;
2145                     last_emu_check++;
2146                 }
2147             /* We'll just re-encode this last macroblock if we go over the max slice size. */
2148             if( total_bits - starting_bits > slice_max_size && !h->mb.b_reencode_mb )
2149             {
2150                 if( mb_xy != h->sh.i_first_mb )
2151                 {
2152                     h->stat.frame.i_mv_bits = mv_bits_bak;
2153                     h->stat.frame.i_tex_bits = tex_bits_bak;
2154                     if( h->param.b_cabac )
2155                     {
2156                         memcpy( &h->cabac, &cabac_bak, offsetof(x264_cabac_t, f8_bits_encoded) );
2157                         h->cabac.p[-1] = cabac_prevbyte_bak;
2158                     }
2159                     else
2160                     {
2161                         h->out.bs = bs_bak;
2162                         i_skip = i_skip_bak;
2163                     }
2164                     h->mb.b_reencode_mb = 1;
2165                     if( SLICE_MBAFF )
2166                     {
2167                         // set to bottom of previous mbpair
2168                         if( i_mb_x )
2169                             h->sh.i_last_mb = mb_xy-1+h->mb.i_mb_stride*(!(i_mb_y&1));
2170                         else
2171                             h->sh.i_last_mb = (i_mb_y-2+!(i_mb_y&1))*h->mb.i_mb_stride + h->mb.i_mb_width - 1;
2172                     }
2173                     else
2174                         h->sh.i_last_mb = mb_xy-1;
2175                     break;
2176                 }
2177                 else
2178                 {
2179                     h->sh.i_last_mb = mb_xy;
2180                     h->mb.b_reencode_mb = 0;
2181                 }
2182             }
2183             else
2184                 h->mb.b_reencode_mb = 0;
2185         }
2186
2187 #if HAVE_VISUALIZE
2188         if( h->param.b_visualize )
2189             x264_visualize_mb( h );
2190 #endif
2191
2192         /* save cache */
2193         x264_macroblock_cache_save( h );
2194
2195         /* accumulate mb stats */
2196         h->stat.frame.i_mb_count[h->mb.i_type]++;
2197
2198         int b_intra = IS_INTRA( h->mb.i_type );
2199         int b_skip = IS_SKIP( h->mb.i_type );
2200         if( h->param.i_log_level >= X264_LOG_INFO || h->param.rc.b_stat_write )
2201         {
2202             if( !b_intra && !b_skip && !IS_DIRECT( h->mb.i_type ) )
2203             {
2204                 if( h->mb.i_partition != D_8x8 )
2205                         h->stat.frame.i_mb_partition[h->mb.i_partition] += 4;
2206                     else
2207                         for( int i = 0; i < 4; i++ )
2208                             h->stat.frame.i_mb_partition[h->mb.i_sub_partition[i]] ++;
2209                 if( h->param.i_frame_reference > 1 )
2210                     for( int i_list = 0; i_list <= (h->sh.i_type == SLICE_TYPE_B); i_list++ )
2211                         for( int i = 0; i < 4; i++ )
2212                         {
2213                             int i_ref = h->mb.cache.ref[i_list][ x264_scan8[4*i] ];
2214                             if( i_ref >= 0 )
2215                                 h->stat.frame.i_mb_count_ref[i_list][i_ref] ++;
2216                         }
2217             }
2218         }
2219
2220         if( h->param.i_log_level >= X264_LOG_INFO )
2221         {
2222             if( h->mb.i_cbp_luma | h->mb.i_cbp_chroma )
2223             {
2224                 if( CHROMA444 )
2225                 {
2226                     for( int i = 0; i < 4; i++ )
2227                         if( h->mb.i_cbp_luma & (1 << i) )
2228                             for( int p = 0; p < 3; p++ )
2229                             {
2230                                 int s8 = i*4+p*16;
2231                                 int nnz8x8 = M16( &h->mb.cache.non_zero_count[x264_scan8[s8]+0] )
2232                                            | M16( &h->mb.cache.non_zero_count[x264_scan8[s8]+8] );
2233                                 h->stat.frame.i_mb_cbp[!b_intra + p*2] += !!nnz8x8;
2234                             }
2235                 }
2236                 else
2237                 {
2238                     int cbpsum = (h->mb.i_cbp_luma&1) + ((h->mb.i_cbp_luma>>1)&1)
2239                                + ((h->mb.i_cbp_luma>>2)&1) + (h->mb.i_cbp_luma>>3);
2240                     h->stat.frame.i_mb_cbp[!b_intra + 0] += cbpsum;
2241                     h->stat.frame.i_mb_cbp[!b_intra + 2] += !!h->mb.i_cbp_chroma;
2242                     h->stat.frame.i_mb_cbp[!b_intra + 4] += h->mb.i_cbp_chroma >> 1;
2243                 }
2244             }
2245             if( h->mb.i_cbp_luma && !b_intra )
2246             {
2247                 h->stat.frame.i_mb_count_8x8dct[0] ++;
2248                 h->stat.frame.i_mb_count_8x8dct[1] += h->mb.b_transform_8x8;
2249             }
2250             if( b_intra && h->mb.i_type != I_PCM )
2251             {
2252                 if( h->mb.i_type == I_16x16 )
2253                     h->stat.frame.i_mb_pred_mode[0][h->mb.i_intra16x16_pred_mode]++;
2254                 else if( h->mb.i_type == I_8x8 )
2255                     for( int i = 0; i < 16; i += 4 )
2256                         h->stat.frame.i_mb_pred_mode[1][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
2257                 else //if( h->mb.i_type == I_4x4 )
2258                     for( int i = 0; i < 16; i++ )
2259                         h->stat.frame.i_mb_pred_mode[2][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
2260                 h->stat.frame.i_mb_pred_mode[3][x264_mb_pred_mode8x8c_fix[h->mb.i_chroma_pred_mode]]++;
2261             }
2262             h->stat.frame.i_mb_field[b_intra?0:b_skip?2:1] += MB_INTERLACED;
2263         }
2264
2265         /* calculate deblock strength values (actual deblocking is done per-row along with hpel) */
2266         if( b_deblock )
2267             x264_macroblock_deblock_strength( h );
2268
2269         x264_ratecontrol_mb( h, mb_size );
2270
2271         if( mb_xy == h->sh.i_last_mb )
2272             break;
2273
2274         if( SLICE_MBAFF )
2275         {
2276             i_mb_x += i_mb_y & 1;
2277             i_mb_y ^= i_mb_x < h->mb.i_mb_width;
2278         }
2279         else
2280             i_mb_x++;
2281         if( i_mb_x == h->mb.i_mb_width )
2282         {
2283             i_mb_y++;
2284             i_mb_x = 0;
2285         }
2286     }
2287     h->out.nal[h->out.i_nal].i_last_mb = h->sh.i_last_mb;
2288
2289     if( h->param.b_cabac )
2290     {
2291         x264_cabac_encode_flush( h, &h->cabac );
2292         h->out.bs.p = h->cabac.p;
2293     }
2294     else
2295     {
2296         if( i_skip > 0 )
2297             bs_write_ue( &h->out.bs, i_skip );  /* last skip run */
2298         /* rbsp_slice_trailing_bits */
2299         bs_rbsp_trailing( &h->out.bs );
2300         bs_flush( &h->out.bs );
2301     }
2302     if( x264_nal_end( h ) )
2303         return -1;
2304
2305     if( h->sh.i_last_mb == (h->i_threadslice_end * h->mb.i_mb_width - 1) )
2306     {
2307         h->stat.frame.i_misc_bits = bs_pos( &h->out.bs )
2308                                   + (h->out.i_nal*NALU_OVERHEAD * 8)
2309                                   - h->stat.frame.i_tex_bits
2310                                   - h->stat.frame.i_mv_bits;
2311         x264_fdec_filter_row( h, h->i_threadslice_end, 1 );
2312     }
2313
2314     return 0;
2315 }
2316
2317 static void x264_thread_sync_context( x264_t *dst, x264_t *src )
2318 {
2319     if( dst == src )
2320         return;
2321
2322     // reference counting
2323     for( x264_frame_t **f = src->frames.reference; *f; f++ )
2324         (*f)->i_reference_count++;
2325     for( x264_frame_t **f = dst->frames.reference; *f; f++ )
2326         x264_frame_push_unused( src, *f );
2327     src->fdec->i_reference_count++;
2328     x264_frame_push_unused( src, dst->fdec );
2329
2330     // copy everything except the per-thread pointers and the constants.
2331     memcpy( &dst->i_frame, &src->i_frame, offsetof(x264_t, mb.type) - offsetof(x264_t, i_frame) );
2332     dst->param = src->param;
2333     dst->stat = src->stat;
2334     dst->pixf = src->pixf;
2335 }
2336
2337 static void x264_thread_sync_stat( x264_t *dst, x264_t *src )
2338 {
2339     if( dst == src )
2340         return;
2341     memcpy( &dst->stat.i_frame_count, &src->stat.i_frame_count, sizeof(dst->stat) - sizeof(dst->stat.frame) );
2342 }
2343
2344 static void *x264_slices_write( x264_t *h )
2345 {
2346     int i_slice_num = 0;
2347     int last_thread_mb = h->sh.i_last_mb;
2348
2349 #if HAVE_VISUALIZE
2350     if( h->param.b_visualize )
2351         if( x264_visualize_init( h ) )
2352             return (void *)-1;
2353 #endif
2354
2355     /* init stats */
2356     memset( &h->stat.frame, 0, sizeof(h->stat.frame) );
2357     h->mb.b_reencode_mb = 0;
2358     while( h->sh.i_first_mb + SLICE_MBAFF*h->mb.i_mb_stride <= last_thread_mb )
2359     {
2360         h->sh.i_last_mb = last_thread_mb;
2361         if( h->param.i_slice_max_mbs )
2362         {
2363             if( SLICE_MBAFF )
2364             {
2365                 // convert first to mbaff form, add slice-max-mbs, then convert back to normal form
2366                 int last_mbaff = 2*(h->sh.i_first_mb % h->mb.i_mb_width)
2367                     + h->mb.i_mb_width*(h->sh.i_first_mb / h->mb.i_mb_width)
2368                     + h->param.i_slice_max_mbs - 1;
2369                 int last_x = (last_mbaff % (2*h->mb.i_mb_width))/2;
2370                 int last_y = (last_mbaff / (2*h->mb.i_mb_width))*2 + 1;
2371                 h->sh.i_last_mb = last_x + h->mb.i_mb_stride*last_y;
2372             }
2373             else
2374                 h->sh.i_last_mb = h->sh.i_first_mb + h->param.i_slice_max_mbs - 1;
2375         }
2376         else if( h->param.i_slice_count && !h->param.b_sliced_threads )
2377         {
2378             int height = h->mb.i_mb_height >> PARAM_INTERLACED;
2379             int width = h->mb.i_mb_width << PARAM_INTERLACED;
2380             i_slice_num++;
2381             h->sh.i_last_mb = (height * i_slice_num + h->param.i_slice_count/2) / h->param.i_slice_count * width - 1;
2382         }
2383         h->sh.i_last_mb = X264_MIN( h->sh.i_last_mb, last_thread_mb );
2384         if( x264_stack_align( x264_slice_write, h ) )
2385             return (void *)-1;
2386         h->sh.i_first_mb = h->sh.i_last_mb + 1;
2387         // if i_first_mb is not the last mb in a row then go to the next mb in MBAFF order
2388         if( SLICE_MBAFF && h->sh.i_first_mb % h->mb.i_mb_width )
2389             h->sh.i_first_mb -= h->mb.i_mb_stride;
2390     }
2391
2392 #if HAVE_VISUALIZE
2393     if( h->param.b_visualize )
2394     {
2395         x264_visualize_show( h );
2396         x264_visualize_close( h );
2397     }
2398 #endif
2399
2400     return (void *)0;
2401 }
2402
2403 static int x264_threaded_slices_write( x264_t *h )
2404 {
2405     /* set first/last mb and sync contexts */
2406     for( int i = 0; i < h->param.i_threads; i++ )
2407     {
2408         x264_t *t = h->thread[i];
2409         if( i )
2410         {
2411             t->param = h->param;
2412             memcpy( &t->i_frame, &h->i_frame, offsetof(x264_t, rc) - offsetof(x264_t, i_frame) );
2413         }
2414         int height = h->mb.i_mb_height >> PARAM_INTERLACED;
2415         t->i_threadslice_start = ((height *  i    + h->param.i_slice_count/2) / h->param.i_threads) << PARAM_INTERLACED;
2416         t->i_threadslice_end   = ((height * (i+1) + h->param.i_slice_count/2) / h->param.i_threads) << PARAM_INTERLACED;
2417         t->sh.i_first_mb = t->i_threadslice_start * h->mb.i_mb_width;
2418         t->sh.i_last_mb  =   t->i_threadslice_end * h->mb.i_mb_width - 1;
2419     }
2420
2421     x264_stack_align( x264_analyse_weight_frame, h, h->mb.i_mb_height*16 + 16 );
2422
2423     x264_threads_distribute_ratecontrol( h );
2424
2425     /* dispatch */
2426     for( int i = 0; i < h->param.i_threads; i++ )
2427     {
2428         x264_threadpool_run( h->threadpool, (void*)x264_slices_write, h->thread[i] );
2429         h->thread[i]->b_thread_active = 1;
2430     }
2431     for( int i = 0; i < h->param.i_threads; i++ )
2432     {
2433         h->thread[i]->b_thread_active = 0;
2434         if( (intptr_t)x264_threadpool_wait( h->threadpool, h->thread[i] ) )
2435             return -1;
2436     }
2437
2438     /* Go back and fix up the hpel on the borders between slices. */
2439     for( int i = 1; i < h->param.i_threads; i++ )
2440     {
2441         x264_fdec_filter_row( h->thread[i], h->thread[i]->i_threadslice_start + 1, 0 );
2442         if( SLICE_MBAFF )
2443             x264_fdec_filter_row( h->thread[i], h->thread[i]->i_threadslice_start + 2, 0 );
2444     }
2445
2446     x264_threads_merge_ratecontrol( h );
2447
2448     for( int i = 1; i < h->param.i_threads; i++ )
2449     {
2450         x264_t *t = h->thread[i];
2451         for( int j = 0; j < t->out.i_nal; j++ )
2452         {
2453             h->out.nal[h->out.i_nal] = t->out.nal[j];
2454             h->out.i_nal++;
2455             x264_nal_check_buffer( h );
2456         }
2457         /* All entries in stat.frame are ints except for ssd/ssim. */
2458         for( int j = 0; j < (offsetof(x264_t,stat.frame.i_ssd) - offsetof(x264_t,stat.frame.i_mv_bits)) / sizeof(int); j++ )
2459             ((int*)&h->stat.frame)[j] += ((int*)&t->stat.frame)[j];
2460         for( int j = 0; j < 3; j++ )
2461             h->stat.frame.i_ssd[j] += t->stat.frame.i_ssd[j];
2462         h->stat.frame.f_ssim += t->stat.frame.f_ssim;
2463         h->stat.frame.i_ssim_cnt += t->stat.frame.i_ssim_cnt;
2464     }
2465
2466     return 0;
2467 }
2468
2469 void x264_encoder_intra_refresh( x264_t *h )
2470 {
2471     h = h->thread[h->i_thread_phase];
2472     h->b_queued_intra_refresh = 1;
2473 }
2474
2475 int x264_encoder_invalidate_reference( x264_t *h, int64_t pts )
2476 {
2477     if( h->param.i_bframe )
2478     {
2479         x264_log( h, X264_LOG_ERROR, "x264_encoder_invalidate_reference is not supported with B-frames enabled\n" );
2480         return -1;
2481     }
2482     if( h->param.b_intra_refresh )
2483     {
2484         x264_log( h, X264_LOG_ERROR, "x264_encoder_invalidate_reference is not supported with intra refresh enabled\n" );
2485         return -1;
2486     }
2487     h = h->thread[h->i_thread_phase];
2488     if( pts >= h->i_last_idr_pts )
2489     {
2490         for( int i = 0; h->frames.reference[i]; i++ )
2491             if( pts <= h->frames.reference[i]->i_pts )
2492                 h->frames.reference[i]->b_corrupt = 1;
2493         if( pts <= h->fdec->i_pts )
2494             h->fdec->b_corrupt = 1;
2495     }
2496     return 0;
2497 }
2498
2499 /****************************************************************************
2500  * x264_encoder_encode:
2501  *  XXX: i_poc   : is the poc of the current given picture
2502  *       i_frame : is the number of the frame being coded
2503  *  ex:  type frame poc
2504  *       I      0   2*0
2505  *       P      1   2*3
2506  *       B      2   2*1
2507  *       B      3   2*2
2508  *       P      4   2*6
2509  *       B      5   2*4
2510  *       B      6   2*5
2511  ****************************************************************************/
2512 int     x264_encoder_encode( x264_t *h,
2513                              x264_nal_t **pp_nal, int *pi_nal,
2514                              x264_picture_t *pic_in,
2515                              x264_picture_t *pic_out )
2516 {
2517     x264_t *thread_current, *thread_prev, *thread_oldest;
2518     int i_nal_type, i_nal_ref_idc, i_global_qp;
2519     int overhead = NALU_OVERHEAD;
2520
2521     if( h->i_thread_frames > 1 )
2522     {
2523         thread_prev    = h->thread[ h->i_thread_phase ];
2524         h->i_thread_phase = (h->i_thread_phase + 1) % h->i_thread_frames;
2525         thread_current = h->thread[ h->i_thread_phase ];
2526         thread_oldest  = h->thread[ (h->i_thread_phase + 1) % h->i_thread_frames ];
2527         x264_thread_sync_context( thread_current, thread_prev );
2528         x264_thread_sync_ratecontrol( thread_current, thread_prev, thread_oldest );
2529         h = thread_current;
2530     }
2531     else
2532     {
2533         thread_current =
2534         thread_oldest  = h;
2535     }
2536 #if HAVE_MMX
2537     if( h->param.cpu&X264_CPU_SSE_MISALIGN )
2538         x264_cpu_mask_misalign_sse();
2539 #endif
2540
2541     // ok to call this before encoding any frames, since the initial values of fdec have b_kept_as_ref=0
2542     if( x264_reference_update( h ) )
2543         return -1;
2544     h->fdec->i_lines_completed = -1;
2545
2546     /* no data out */
2547     *pi_nal = 0;
2548     *pp_nal = NULL;
2549
2550     /* ------------------- Setup new frame from picture -------------------- */
2551     if( pic_in != NULL )
2552     {
2553         /* 1: Copy the picture to a frame and move it to a buffer */
2554         x264_frame_t *fenc = x264_frame_pop_unused( h, 0 );
2555         if( !fenc )
2556             return -1;
2557
2558         if( x264_frame_copy_picture( h, fenc, pic_in ) < 0 )
2559             return -1;
2560
2561         if( h->param.i_width != 16 * h->mb.i_mb_width ||
2562             h->param.i_height != 16 * h->mb.i_mb_height )
2563             x264_frame_expand_border_mod16( h, fenc );
2564
2565         fenc->i_frame = h->frames.i_input++;
2566
2567         if( fenc->i_frame == 0 )
2568             h->frames.i_first_pts = fenc->i_pts;
2569         if( h->frames.i_bframe_delay && fenc->i_frame == h->frames.i_bframe_delay )
2570             h->frames.i_bframe_delay_time = fenc->i_pts - h->frames.i_first_pts;
2571
2572         if( h->param.b_vfr_input && fenc->i_pts <= h->frames.i_largest_pts )
2573             x264_log( h, X264_LOG_WARNING, "non-strictly-monotonic PTS\n" );
2574
2575         h->frames.i_second_largest_pts = h->frames.i_largest_pts;
2576         h->frames.i_largest_pts = fenc->i_pts;
2577
2578         if( (fenc->i_pic_struct < PIC_STRUCT_AUTO) || (fenc->i_pic_struct > PIC_STRUCT_TRIPLE) )
2579             fenc->i_pic_struct = PIC_STRUCT_AUTO;
2580
2581         if( fenc->i_pic_struct == PIC_STRUCT_AUTO )
2582         {
2583 #if HAVE_INTERLACED
2584             int b_interlaced = fenc->param ? fenc->param->b_interlaced : h->param.b_interlaced;
2585 #else
2586             int b_interlaced = 0;
2587 #endif
2588             if( b_interlaced )
2589             {
2590                 int b_tff = fenc->param ? fenc->param->b_tff : h->param.b_tff;
2591                 fenc->i_pic_struct = b_tff ? PIC_STRUCT_TOP_BOTTOM : PIC_STRUCT_BOTTOM_TOP;
2592             }
2593             else
2594                 fenc->i_pic_struct = PIC_STRUCT_PROGRESSIVE;
2595         }
2596
2597         if( h->param.rc.b_mb_tree && h->param.rc.b_stat_read )
2598         {
2599             if( x264_macroblock_tree_read( h, fenc, pic_in->prop.quant_offsets ) )
2600                 return -1;
2601         }
2602         else
2603             x264_stack_align( x264_adaptive_quant_frame, h, fenc, pic_in->prop.quant_offsets );
2604
2605         if( pic_in->prop.quant_offsets_free )
2606             pic_in->prop.quant_offsets_free( pic_in->prop.quant_offsets );
2607
2608         if( h->frames.b_have_lowres )
2609             x264_frame_init_lowres( h, fenc );
2610
2611         /* 2: Place the frame into the queue for its slice type decision */
2612         x264_lookahead_put_frame( h, fenc );
2613
2614         if( h->frames.i_input <= h->frames.i_delay + 1 - h->i_thread_frames )
2615         {
2616             /* Nothing yet to encode, waiting for filling of buffers */
2617             pic_out->i_type = X264_TYPE_AUTO;
2618             return 0;
2619         }
2620     }
2621     else
2622     {
2623         /* signal kills for lookahead thread */
2624         x264_pthread_mutex_lock( &h->lookahead->ifbuf.mutex );
2625         h->lookahead->b_exit_thread = 1;
2626         x264_pthread_cond_broadcast( &h->lookahead->ifbuf.cv_fill );
2627         x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
2628     }
2629
2630     h->i_frame++;
2631     /* 3: The picture is analyzed in the lookahead */
2632     if( !h->frames.current[0] )
2633         x264_lookahead_get_frames( h );
2634
2635     if( !h->frames.current[0] && x264_lookahead_is_empty( h ) )
2636         return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
2637
2638     /* ------------------- Get frame to be encoded ------------------------- */
2639     /* 4: get picture to encode */
2640     h->fenc = x264_frame_shift( h->frames.current );
2641     if( h->i_frame == h->i_thread_frames - 1 )
2642         h->i_reordered_pts_delay = h->fenc->i_reordered_pts;
2643     if( h->fenc->param )
2644     {
2645         x264_encoder_reconfig( h, h->fenc->param );
2646         if( h->fenc->param->param_free )
2647             h->fenc->param->param_free( h->fenc->param );
2648     }
2649
2650     if( !IS_X264_TYPE_I( h->fenc->i_type ) )
2651     {
2652         int valid_refs_left = 0;
2653         for( int i = 0; h->frames.reference[i]; i++ )
2654             if( !h->frames.reference[i]->b_corrupt )
2655                 valid_refs_left++;
2656         /* No valid reference frames left: force an IDR. */
2657         if( !valid_refs_left )
2658         {
2659             h->fenc->b_keyframe = 1;
2660             h->fenc->i_type = X264_TYPE_IDR;
2661         }
2662     }
2663
2664     if( h->fenc->b_keyframe )
2665     {
2666         h->frames.i_last_keyframe = h->fenc->i_frame;
2667         if( h->fenc->i_type == X264_TYPE_IDR )
2668         {
2669             h->i_frame_num = 0;
2670             h->frames.i_last_idr = h->fenc->i_frame;
2671         }
2672     }
2673     h->sh.i_mmco_command_count =
2674     h->sh.i_mmco_remove_from_end = 0;
2675     h->b_ref_reorder[0] =
2676     h->b_ref_reorder[1] = 0;
2677     h->fdec->i_poc =
2678     h->fenc->i_poc = 2 * ( h->fenc->i_frame - X264_MAX( h->frames.i_last_idr, 0 ) );
2679
2680     /* ------------------- Setup frame context ----------------------------- */
2681     /* 5: Init data dependent of frame type */
2682     if( h->fenc->i_type == X264_TYPE_IDR )
2683     {
2684         /* reset ref pictures */
2685         i_nal_type    = NAL_SLICE_IDR;
2686         i_nal_ref_idc = NAL_PRIORITY_HIGHEST;
2687         h->sh.i_type = SLICE_TYPE_I;
2688         x264_reference_reset( h );
2689         h->frames.i_poc_last_open_gop = -1;
2690     }
2691     else if( h->fenc->i_type == X264_TYPE_I )
2692     {
2693         i_nal_type    = NAL_SLICE;
2694         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
2695         h->sh.i_type = SLICE_TYPE_I;
2696         x264_reference_hierarchy_reset( h );
2697         if( h->param.b_open_gop )
2698             h->frames.i_poc_last_open_gop = h->fenc->b_keyframe ? h->fenc->i_poc : -1;
2699     }
2700     else if( h->fenc->i_type == X264_TYPE_P )
2701     {
2702         i_nal_type    = NAL_SLICE;
2703         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
2704         h->sh.i_type = SLICE_TYPE_P;
2705         x264_reference_hierarchy_reset( h );
2706         h->frames.i_poc_last_open_gop = -1;
2707     }
2708     else if( h->fenc->i_type == X264_TYPE_BREF )
2709     {
2710         i_nal_type    = NAL_SLICE;
2711         i_nal_ref_idc = h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT ? NAL_PRIORITY_LOW : NAL_PRIORITY_HIGH;
2712         h->sh.i_type = SLICE_TYPE_B;
2713         x264_reference_hierarchy_reset( h );
2714     }
2715     else    /* B frame */
2716     {
2717         i_nal_type    = NAL_SLICE;
2718         i_nal_ref_idc = NAL_PRIORITY_DISPOSABLE;
2719         h->sh.i_type = SLICE_TYPE_B;
2720     }
2721
2722     h->fdec->i_type = h->fenc->i_type;
2723     h->fdec->i_frame = h->fenc->i_frame;
2724     h->fenc->b_kept_as_ref =
2725     h->fdec->b_kept_as_ref = i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE && h->param.i_keyint_max > 1;
2726
2727     h->fdec->i_pts = h->fenc->i_pts;
2728     if( h->frames.i_bframe_delay )
2729     {
2730         int64_t *prev_reordered_pts = thread_current->frames.i_prev_reordered_pts;
2731         h->fdec->i_dts = h->i_frame > h->frames.i_bframe_delay
2732                        ? prev_reordered_pts[ (h->i_frame - h->frames.i_bframe_delay) % h->frames.i_bframe_delay ]
2733                        : h->fenc->i_reordered_pts - h->frames.i_bframe_delay_time;
2734         prev_reordered_pts[ h->i_frame % h->frames.i_bframe_delay ] = h->fenc->i_reordered_pts;
2735     }
2736     else
2737         h->fdec->i_dts = h->fenc->i_reordered_pts;
2738     if( h->fenc->i_type == X264_TYPE_IDR )
2739         h->i_last_idr_pts = h->fdec->i_pts;
2740
2741     /* ------------------- Init                ----------------------------- */
2742     /* build ref list 0/1 */
2743     x264_reference_build_list( h, h->fdec->i_poc );
2744
2745     /* ---------------------- Write the bitstream -------------------------- */
2746     /* Init bitstream context */
2747     if( h->param.b_sliced_threads )
2748     {
2749         for( int i = 0; i < h->param.i_threads; i++ )
2750         {
2751             bs_init( &h->thread[i]->out.bs, h->thread[i]->out.p_bitstream, h->thread[i]->out.i_bitstream );
2752             h->thread[i]->out.i_nal = 0;
2753         }
2754     }
2755     else
2756     {
2757         bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
2758         h->out.i_nal = 0;
2759     }
2760
2761     if( h->param.b_aud )
2762     {
2763         int pic_type;
2764
2765         if( h->sh.i_type == SLICE_TYPE_I )
2766             pic_type = 0;
2767         else if( h->sh.i_type == SLICE_TYPE_P )
2768             pic_type = 1;
2769         else if( h->sh.i_type == SLICE_TYPE_B )
2770             pic_type = 2;
2771         else
2772             pic_type = 7;
2773
2774         x264_nal_start( h, NAL_AUD, NAL_PRIORITY_DISPOSABLE );
2775         bs_write( &h->out.bs, 3, pic_type );
2776         bs_rbsp_trailing( &h->out.bs );
2777         if( x264_nal_end( h ) )
2778             return -1;
2779         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2780     }
2781
2782     h->i_nal_type = i_nal_type;
2783     h->i_nal_ref_idc = i_nal_ref_idc;
2784
2785     if( h->param.b_intra_refresh )
2786     {
2787         if( IS_X264_TYPE_I( h->fenc->i_type ) )
2788         {
2789             h->fdec->i_frames_since_pir = 0;
2790             h->b_queued_intra_refresh = 0;
2791             /* PIR is currently only supported with ref == 1, so any intra frame effectively refreshes
2792              * the whole frame and counts as an intra refresh. */
2793             h->fdec->f_pir_position = h->mb.i_mb_width;
2794         }
2795         else if( h->fenc->i_type == X264_TYPE_P )
2796         {
2797             int pocdiff = (h->fdec->i_poc - h->fref[0][0]->i_poc)/2;
2798             float increment = X264_MAX( ((float)h->mb.i_mb_width-1) / h->param.i_keyint_max, 1 );
2799             h->fdec->f_pir_position = h->fref[0][0]->f_pir_position;
2800             h->fdec->i_frames_since_pir = h->fref[0][0]->i_frames_since_pir + pocdiff;
2801             if( h->fdec->i_frames_since_pir >= h->param.i_keyint_max ||
2802                 (h->b_queued_intra_refresh && h->fdec->f_pir_position + 0.5 >= h->mb.i_mb_width) )
2803             {
2804                 h->fdec->f_pir_position = 0;
2805                 h->fdec->i_frames_since_pir = 0;
2806                 h->b_queued_intra_refresh = 0;
2807                 h->fenc->b_keyframe = 1;
2808             }
2809             h->fdec->i_pir_start_col = h->fdec->f_pir_position+0.5;
2810             h->fdec->f_pir_position += increment * pocdiff;
2811             h->fdec->i_pir_end_col = h->fdec->f_pir_position+0.5;
2812             /* If our intra refresh has reached the right side of the frame, we're done. */
2813             if( h->fdec->i_pir_end_col >= h->mb.i_mb_width - 1 )
2814                 h->fdec->f_pir_position = h->mb.i_mb_width;
2815         }
2816     }
2817
2818     if( h->fenc->b_keyframe )
2819     {
2820         /* Write SPS and PPS */
2821         if( h->param.b_repeat_headers )
2822         {
2823             /* generate sequence parameters */
2824             x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
2825             x264_sps_write( &h->out.bs, h->sps );
2826             if( x264_nal_end( h ) )
2827                 return -1;
2828             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
2829
2830             /* generate picture parameters */
2831             x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
2832             x264_pps_write( &h->out.bs, h->sps, h->pps );
2833             if( x264_nal_end( h ) )
2834                 return -1;
2835             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
2836         }
2837
2838         /* buffering period sei is written in x264_encoder_frame_end */
2839     }
2840
2841     /* write extra sei */
2842     for( int i = 0; i < h->fenc->extra_sei.num_payloads; i++ )
2843     {
2844         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2845         x264_sei_write( &h->out.bs, h->fenc->extra_sei.payloads[i].payload, h->fenc->extra_sei.payloads[i].payload_size,
2846                         h->fenc->extra_sei.payloads[i].payload_type );
2847         if( x264_nal_end( h ) )
2848             return -1;
2849         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2850         if( h->fenc->extra_sei.sei_free && h->fenc->extra_sei.payloads[i].payload )
2851             h->fenc->extra_sei.sei_free( h->fenc->extra_sei.payloads[i].payload );
2852     }
2853
2854     if( h->fenc->extra_sei.sei_free && h->fenc->extra_sei.payloads )
2855         h->fenc->extra_sei.sei_free( h->fenc->extra_sei.payloads );
2856
2857     if( h->fenc->b_keyframe )
2858     {
2859         if( h->param.b_repeat_headers && h->fenc->i_frame == 0 )
2860         {
2861             /* identify ourself */
2862             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2863             if( x264_sei_version_write( h, &h->out.bs ) )
2864                 return -1;
2865             if( x264_nal_end( h ) )
2866                 return -1;
2867             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2868         }
2869
2870         if( h->fenc->i_type != X264_TYPE_IDR )
2871         {
2872             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;
2873             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2874             x264_sei_recovery_point_write( h, &h->out.bs, time_to_recovery );
2875             if( x264_nal_end( h ) )
2876                 return -1;
2877             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2878         }
2879
2880         if ( h->param.i_frame_packing >= 0 )
2881         {
2882             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2883             x264_sei_frame_packing_write( h, &h->out.bs );
2884             if( x264_nal_end( h ) )
2885                 return -1;
2886             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2887         }
2888     }
2889
2890     /* generate sei pic timing */
2891     if( h->sps->vui.b_pic_struct_present || h->sps->vui.b_nal_hrd_parameters_present )
2892     {
2893         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2894         x264_sei_pic_timing_write( h, &h->out.bs );
2895         if( x264_nal_end( h ) )
2896             return -1;
2897         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2898     }
2899
2900     /* As required by Blu-ray. */
2901     if( !IS_X264_TYPE_B( h->fenc->i_type ) && h->b_sh_backup )
2902     {
2903         h->b_sh_backup = 0;
2904         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2905         x264_sei_dec_ref_pic_marking_write( h, &h->out.bs );
2906         if( x264_nal_end( h ) )
2907             return -1;
2908         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2909     }
2910
2911     if( h->fenc->b_keyframe && h->param.b_intra_refresh )
2912         h->i_cpb_delay_pir_offset = h->fenc->i_cpb_delay;
2913
2914     /* Init the rate control */
2915     /* FIXME: Include slice header bit cost. */
2916     x264_ratecontrol_start( h, h->fenc->i_qpplus1, overhead*8 );
2917     i_global_qp = x264_ratecontrol_qp( h );
2918
2919     pic_out->i_qpplus1 =
2920     h->fdec->i_qpplus1 = i_global_qp + 1;
2921
2922     if( h->param.rc.b_stat_read && h->sh.i_type != SLICE_TYPE_I )
2923     {
2924         x264_reference_build_list_optimal( h );
2925         x264_reference_check_reorder( h );
2926     }
2927
2928     if( h->i_ref[0] )
2929         h->fdec->i_poc_l0ref0 = h->fref[0][0]->i_poc;
2930
2931     /* ------------------------ Create slice header  ----------------------- */
2932     x264_slice_init( h, i_nal_type, i_global_qp );
2933
2934     /*------------------------- Weights -------------------------------------*/
2935     if( h->sh.i_type == SLICE_TYPE_B )
2936         x264_macroblock_bipred_init( h );
2937
2938     x264_weighted_pred_init( h );
2939
2940     if( i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE )
2941         h->i_frame_num++;
2942
2943     /* Write frame */
2944     h->i_threadslice_start = 0;
2945     h->i_threadslice_end = h->mb.i_mb_height;
2946     if( h->i_thread_frames > 1 )
2947     {
2948         x264_threadpool_run( h->threadpool, (void*)x264_slices_write, h );
2949         h->b_thread_active = 1;
2950     }
2951     else if( h->param.b_sliced_threads )
2952     {
2953         if( x264_threaded_slices_write( h ) )
2954             return -1;
2955     }
2956     else
2957         if( (intptr_t)x264_slices_write( h ) )
2958             return -1;
2959
2960     return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
2961 }
2962
2963 static int x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
2964                                    x264_nal_t **pp_nal, int *pi_nal,
2965                                    x264_picture_t *pic_out )
2966 {
2967     char psz_message[80];
2968
2969     if( h->b_thread_active )
2970     {
2971         h->b_thread_active = 0;
2972         if( (intptr_t)x264_threadpool_wait( h->threadpool, h ) )
2973             return -1;
2974     }
2975     if( !h->out.i_nal )
2976     {
2977         pic_out->i_type = X264_TYPE_AUTO;
2978         return 0;
2979     }
2980
2981     x264_emms();
2982     /* generate sei buffering period and insert it into place */
2983     if( h->fenc->b_keyframe && h->sps->vui.b_nal_hrd_parameters_present )
2984     {
2985         x264_hrd_fullness( h );
2986         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2987         x264_sei_buffering_period_write( h, &h->out.bs );
2988         if( x264_nal_end( h ) )
2989            return -1;
2990         /* buffering period sei must follow AUD, SPS and PPS and precede all other SEIs */
2991         int idx = 0;
2992         while( h->out.nal[idx].i_type == NAL_AUD ||
2993                h->out.nal[idx].i_type == NAL_SPS ||
2994                h->out.nal[idx].i_type == NAL_PPS )
2995             idx++;
2996         x264_nal_t nal_tmp = h->out.nal[h->out.i_nal-1];
2997         memmove( &h->out.nal[idx+1], &h->out.nal[idx], (h->out.i_nal-idx-1)*sizeof(x264_nal_t) );
2998         h->out.nal[idx] = nal_tmp;
2999     }
3000
3001     int frame_size = x264_encoder_encapsulate_nals( h, 0 );
3002     if( frame_size < 0 )
3003         return -1;
3004
3005     /* Set output picture properties */
3006     pic_out->i_type = h->fenc->i_type;
3007
3008     pic_out->b_keyframe = h->fenc->b_keyframe;
3009     pic_out->i_pic_struct = h->fenc->i_pic_struct;
3010
3011     pic_out->i_pts = h->fdec->i_pts;
3012     pic_out->i_dts = h->fdec->i_dts;
3013
3014     if( pic_out->i_pts < pic_out->i_dts )
3015         x264_log( h, X264_LOG_WARNING, "invalid DTS: PTS is less than DTS\n" );
3016
3017     pic_out->img.i_csp = X264_CSP_NV12;
3018 #if HIGH_BIT_DEPTH
3019     pic_out->img.i_csp |= X264_CSP_HIGH_DEPTH;
3020 #endif
3021     pic_out->img.i_plane = h->fdec->i_plane;
3022     for( int i = 0; i < 2; i++ )
3023     {
3024         pic_out->img.i_stride[i] = h->fdec->i_stride[i] * sizeof(pixel);
3025         pic_out->img.plane[i] = (uint8_t*)h->fdec->plane[i];
3026     }
3027
3028     x264_frame_push_unused( thread_current, h->fenc );
3029
3030     /* ---------------------- Update encoder state ------------------------- */
3031
3032     /* update rc */
3033     int filler = 0;
3034     if( x264_ratecontrol_end( h, frame_size * 8, &filler ) < 0 )
3035         return -1;
3036
3037     pic_out->hrd_timing = h->fenc->hrd_timing;
3038
3039     while( filler > 0 )
3040     {
3041         int f, overhead;
3042         overhead = (FILLER_OVERHEAD - h->param.b_annexb);
3043         if( h->param.i_slice_max_size && filler > h->param.i_slice_max_size )
3044         {
3045             int next_size = filler - h->param.i_slice_max_size;
3046             int overflow = X264_MAX( overhead - next_size, 0 );
3047             f = h->param.i_slice_max_size - overhead - overflow;
3048         }
3049         else
3050             f = X264_MAX( 0, filler - overhead );
3051
3052         x264_nal_start( h, NAL_FILLER, NAL_PRIORITY_DISPOSABLE );
3053         x264_filler_write( h, &h->out.bs, f );
3054         if( x264_nal_end( h ) )
3055             return -1;
3056         int total_size = x264_encoder_encapsulate_nals( h, h->out.i_nal-1 );
3057         if( total_size < 0 )
3058             return -1;
3059         frame_size += total_size;
3060         filler -= total_size;
3061     }
3062
3063     /* End bitstream, set output  */
3064     *pi_nal = h->out.i_nal;
3065     *pp_nal = h->out.nal;
3066
3067     h->out.i_nal = 0;
3068
3069     x264_noise_reduction_update( h );
3070
3071     /* ---------------------- Compute/Print statistics --------------------- */
3072     x264_thread_sync_stat( h, h->thread[0] );
3073
3074     /* Slice stat */
3075     h->stat.i_frame_count[h->sh.i_type]++;
3076     h->stat.i_frame_size[h->sh.i_type] += frame_size;
3077     h->stat.f_frame_qp[h->sh.i_type] += h->fdec->f_qp_avg_aq;
3078
3079     for( int i = 0; i < X264_MBTYPE_MAX; i++ )
3080         h->stat.i_mb_count[h->sh.i_type][i] += h->stat.frame.i_mb_count[i];
3081     for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
3082         h->stat.i_mb_partition[h->sh.i_type][i] += h->stat.frame.i_mb_partition[i];
3083     for( int i = 0; i < 2; i++ )
3084         h->stat.i_mb_count_8x8dct[i] += h->stat.frame.i_mb_count_8x8dct[i];
3085     for( int i = 0; i < 6; i++ )
3086         h->stat.i_mb_cbp[i] += h->stat.frame.i_mb_cbp[i];
3087     for( int i = 0; i < 4; i++ )
3088         for( int j = 0; j < 13; j++ )
3089             h->stat.i_mb_pred_mode[i][j] += h->stat.frame.i_mb_pred_mode[i][j];
3090     if( h->sh.i_type != SLICE_TYPE_I )
3091         for( int i_list = 0; i_list < 2; i_list++ )
3092             for( int i = 0; i < X264_REF_MAX*2; i++ )
3093                 h->stat.i_mb_count_ref[h->sh.i_type][i_list][i] += h->stat.frame.i_mb_count_ref[i_list][i];
3094     for( int i = 0; i < 3; i++ )
3095         h->stat.i_mb_field[i] += h->stat.frame.i_mb_field[i];
3096     if( h->sh.i_type == SLICE_TYPE_P && h->param.analyse.i_weighted_pred >= X264_WEIGHTP_SIMPLE )
3097     {
3098         h->stat.i_wpred[0] += !!h->sh.weight[0][0].weightfn;
3099         h->stat.i_wpred[1] += !!h->sh.weight[0][1].weightfn || !!h->sh.weight[0][2].weightfn;
3100     }
3101     if( h->sh.i_type == SLICE_TYPE_B )
3102     {
3103         h->stat.i_direct_frames[ h->sh.b_direct_spatial_mv_pred ] ++;
3104         if( h->mb.b_direct_auto_write )
3105         {
3106             //FIXME somewhat arbitrary time constants
3107             if( h->stat.i_direct_score[0] + h->stat.i_direct_score[1] > h->mb.i_mb_count )
3108                 for( int i = 0; i < 2; i++ )
3109                     h->stat.i_direct_score[i] = h->stat.i_direct_score[i] * 9/10;
3110             for( int i = 0; i < 2; i++ )
3111                 h->stat.i_direct_score[i] += h->stat.frame.i_direct_score[i];
3112         }
3113     }
3114     else
3115         h->stat.i_consecutive_bframes[h->fenc->i_bframes]++;
3116
3117     psz_message[0] = '\0';
3118     double dur = h->fenc->f_duration;
3119     h->stat.f_frame_duration[h->sh.i_type] += dur;
3120     if( h->param.analyse.b_psnr )
3121     {
3122         int64_t ssd[3] =
3123         {
3124             h->stat.frame.i_ssd[0],
3125             h->stat.frame.i_ssd[1],
3126             h->stat.frame.i_ssd[2],
3127         };
3128         int luma_size = h->param.i_width * h->param.i_height;
3129         int chroma_size = h->param.i_width * h->param.i_height >> (!CHROMA444 * 2);
3130         double psnr_y = x264_psnr( ssd[0], luma_size );
3131         double psnr_u = x264_psnr( ssd[1], chroma_size );
3132         double psnr_v = x264_psnr( ssd[2], chroma_size );
3133
3134         h->stat.f_ssd_global[h->sh.i_type]   += dur * (ssd[0] + ssd[1] + ssd[2]);
3135         h->stat.f_psnr_average[h->sh.i_type] += dur * x264_psnr( ssd[0] + ssd[1] + ssd[2], luma_size + chroma_size*2 );
3136         h->stat.f_psnr_mean_y[h->sh.i_type]  += dur * psnr_y;
3137         h->stat.f_psnr_mean_u[h->sh.i_type]  += dur * psnr_u;
3138         h->stat.f_psnr_mean_v[h->sh.i_type]  += dur * psnr_v;
3139
3140         snprintf( psz_message, 80, " PSNR Y:%5.2f U:%5.2f V:%5.2f", psnr_y, psnr_u, psnr_v );
3141     }
3142
3143     if( h->param.analyse.b_ssim )
3144     {
3145         double ssim_y = h->stat.frame.f_ssim
3146                       / h->stat.frame.i_ssim_cnt;
3147         h->stat.f_ssim_mean_y[h->sh.i_type] += ssim_y * dur;
3148         snprintf( psz_message + strlen(psz_message), 80 - strlen(psz_message),
3149                   " SSIM Y:%.5f", ssim_y );
3150     }
3151     psz_message[79] = '\0';
3152
3153     x264_log( h, X264_LOG_DEBUG,
3154                   "frame=%4d QP=%.2f NAL=%d Slice:%c Poc:%-3d I:%-4d P:%-4d SKIP:%-4d size=%d bytes%s\n",
3155               h->i_frame,
3156               h->fdec->f_qp_avg_aq,
3157               h->i_nal_ref_idc,
3158               h->sh.i_type == SLICE_TYPE_I ? 'I' : (h->sh.i_type == SLICE_TYPE_P ? 'P' : 'B' ),
3159               h->fdec->i_poc,
3160               h->stat.frame.i_mb_count_i,
3161               h->stat.frame.i_mb_count_p,
3162               h->stat.frame.i_mb_count_skip,
3163               frame_size,
3164               psz_message );
3165
3166     // keep stats all in one place
3167     x264_thread_sync_stat( h->thread[0], h );
3168     // for the use of the next frame
3169     x264_thread_sync_stat( thread_current, h );
3170
3171 #ifdef DEBUG_MB_TYPE
3172 {
3173     static const char mb_chars[] = { 'i', 'i', 'I', 'C', 'P', '8', 'S',
3174         'D', '<', 'X', 'B', 'X', '>', 'B', 'B', 'B', 'B', '8', 'S' };
3175     for( int mb_xy = 0; mb_xy < h->mb.i_mb_width * h->mb.i_mb_height; mb_xy++ )
3176     {
3177         if( h->mb.type[mb_xy] < X264_MBTYPE_MAX && h->mb.type[mb_xy] >= 0 )
3178             fprintf( stderr, "%c ", mb_chars[ h->mb.type[mb_xy] ] );
3179         else
3180             fprintf( stderr, "? " );
3181
3182         if( (mb_xy+1) % h->mb.i_mb_width == 0 )
3183             fprintf( stderr, "\n" );
3184     }
3185 }
3186 #endif
3187
3188     /* Remove duplicates, must be done near the end as breaks h->fref0 array
3189      * by freeing some of its pointers. */
3190     for( int i = 0; i < h->i_ref[0]; i++ )
3191         if( h->fref[0][i] && h->fref[0][i]->b_duplicate )
3192         {
3193             x264_frame_push_blank_unused( h, h->fref[0][i] );
3194             h->fref[0][i] = 0;
3195         }
3196
3197     if( h->param.psz_dump_yuv )
3198         x264_frame_dump( h );
3199     x264_emms();
3200
3201     return frame_size;
3202 }
3203
3204 static void x264_print_intra( int64_t *i_mb_count, double i_count, int b_print_pcm, char *intra )
3205 {
3206     intra += sprintf( intra, "I16..4%s: %4.1f%% %4.1f%% %4.1f%%",
3207         b_print_pcm ? "..PCM" : "",
3208         i_mb_count[I_16x16]/ i_count,
3209         i_mb_count[I_8x8]  / i_count,
3210         i_mb_count[I_4x4]  / i_count );
3211     if( b_print_pcm )
3212         sprintf( intra, " %4.1f%%", i_mb_count[I_PCM]  / i_count );
3213 }
3214
3215 /****************************************************************************
3216  * x264_encoder_close:
3217  ****************************************************************************/
3218 void    x264_encoder_close  ( x264_t *h )
3219 {
3220     int luma_size = h->param.i_width * h->param.i_height;
3221     int chroma_size = h->param.i_width * h->param.i_height >> (!CHROMA444 * 2);
3222     int64_t i_yuv_size = luma_size + chroma_size * 2;
3223     int64_t i_mb_count_size[2][7] = {{0}};
3224     char buf[200];
3225     int b_print_pcm = h->stat.i_mb_count[SLICE_TYPE_I][I_PCM]
3226                    || h->stat.i_mb_count[SLICE_TYPE_P][I_PCM]
3227                    || h->stat.i_mb_count[SLICE_TYPE_B][I_PCM];
3228
3229     x264_lookahead_delete( h );
3230
3231     if( h->param.i_threads > 1 )
3232         x264_threadpool_delete( h->threadpool );
3233     if( h->i_thread_frames > 1 )
3234     {
3235         for( int i = 0; i < h->i_thread_frames; i++ )
3236             if( h->thread[i]->b_thread_active )
3237             {
3238                 assert( h->thread[i]->fenc->i_reference_count == 1 );
3239                 x264_frame_delete( h->thread[i]->fenc );
3240             }
3241
3242         x264_t *thread_prev = h->thread[h->i_thread_phase];
3243         x264_thread_sync_ratecontrol( h, thread_prev, h );
3244         x264_thread_sync_ratecontrol( thread_prev, thread_prev, h );
3245         h->i_frame = thread_prev->i_frame + 1 - h->i_thread_frames;
3246     }
3247     h->i_frame++;
3248
3249     /* Slices used and PSNR */
3250     for( int i = 0; i < 3; i++ )
3251     {
3252         static const uint8_t slice_order[] = { SLICE_TYPE_I, SLICE_TYPE_P, SLICE_TYPE_B };
3253         int i_slice = slice_order[i];
3254
3255         if( h->stat.i_frame_count[i_slice] > 0 )
3256         {
3257             int i_count = h->stat.i_frame_count[i_slice];
3258             double dur =  h->stat.f_frame_duration[i_slice];
3259             if( h->param.analyse.b_psnr )
3260             {
3261                 x264_log( h, X264_LOG_INFO,
3262                           "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",
3263                           slice_type_to_char[i_slice],
3264                           i_count,
3265                           h->stat.f_frame_qp[i_slice] / i_count,
3266                           (double)h->stat.i_frame_size[i_slice] / i_count,
3267                           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,
3268                           h->stat.f_psnr_average[i_slice] / dur,
3269                           x264_psnr( h->stat.f_ssd_global[i_slice], dur * i_yuv_size ) );
3270             }
3271             else
3272             {
3273                 x264_log( h, X264_LOG_INFO,
3274                           "frame %c:%-5d Avg QP:%5.2f  size:%6.0f\n",
3275                           slice_type_to_char[i_slice],
3276                           i_count,
3277                           h->stat.f_frame_qp[i_slice] / i_count,
3278                           (double)h->stat.i_frame_size[i_slice] / i_count );
3279             }
3280         }
3281     }
3282     if( h->param.i_bframe && h->stat.i_frame_count[SLICE_TYPE_B] )
3283     {
3284         char *p = buf;
3285         int den = 0;
3286         // weight by number of frames (including the I/P-frames) that are in a sequence of N B-frames
3287         for( int i = 0; i <= h->param.i_bframe; i++ )
3288             den += (i+1) * h->stat.i_consecutive_bframes[i];
3289         for( int i = 0; i <= h->param.i_bframe; i++ )
3290             p += sprintf( p, " %4.1f%%", 100. * (i+1) * h->stat.i_consecutive_bframes[i] / den );
3291         x264_log( h, X264_LOG_INFO, "consecutive B-frames:%s\n", buf );
3292     }
3293
3294     for( int i_type = 0; i_type < 2; i_type++ )
3295         for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
3296         {
3297             if( i == D_DIRECT_8x8 ) continue; /* direct is counted as its own type */
3298             i_mb_count_size[i_type][x264_mb_partition_pixel_table[i]] += h->stat.i_mb_partition[i_type][i];
3299         }
3300
3301     /* MB types used */
3302     if( h->stat.i_frame_count[SLICE_TYPE_I] > 0 )
3303     {
3304         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_I];
3305         double i_count = h->stat.i_frame_count[SLICE_TYPE_I] * h->mb.i_mb_count / 100.0;
3306         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
3307         x264_log( h, X264_LOG_INFO, "mb I  %s\n", buf );
3308     }
3309     if( h->stat.i_frame_count[SLICE_TYPE_P] > 0 )
3310     {
3311         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_P];
3312         double i_count = h->stat.i_frame_count[SLICE_TYPE_P] * h->mb.i_mb_count / 100.0;
3313         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_P];
3314         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
3315         x264_log( h, X264_LOG_INFO,
3316                   "mb P  %s  P16..4: %4.1f%% %4.1f%% %4.1f%% %4.1f%% %4.1f%%    skip:%4.1f%%\n",
3317                   buf,
3318                   i_mb_size[PIXEL_16x16] / (i_count*4),
3319                   (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
3320                   i_mb_size[PIXEL_8x8] / (i_count*4),
3321                   (i_mb_size[PIXEL_8x4] + i_mb_size[PIXEL_4x8]) / (i_count*4),
3322                   i_mb_size[PIXEL_4x4] / (i_count*4),
3323                   i_mb_count[P_SKIP] / i_count );
3324     }
3325     if( h->stat.i_frame_count[SLICE_TYPE_B] > 0 )
3326     {
3327         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_B];
3328         double i_count = h->stat.i_frame_count[SLICE_TYPE_B] * h->mb.i_mb_count / 100.0;
3329         double i_mb_list_count;
3330         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_B];
3331         int64_t list_count[3] = {0}; /* 0 == L0, 1 == L1, 2 == BI */
3332         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
3333         for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
3334             for( int j = 0; j < 2; j++ )
3335             {
3336                 int l0 = x264_mb_type_list_table[i][0][j];
3337                 int l1 = x264_mb_type_list_table[i][1][j];
3338                 if( l0 || l1 )
3339                     list_count[l1+l0*l1] += h->stat.i_mb_count[SLICE_TYPE_B][i] * 2;
3340             }
3341         list_count[0] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L0_8x8];
3342         list_count[1] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L1_8x8];
3343         list_count[2] += h->stat.i_mb_partition[SLICE_TYPE_B][D_BI_8x8];
3344         i_mb_count[B_DIRECT] += (h->stat.i_mb_partition[SLICE_TYPE_B][D_DIRECT_8x8]+2)/4;
3345         i_mb_list_count = (list_count[0] + list_count[1] + list_count[2]) / 100.0;
3346         sprintf( buf + strlen(buf), "  B16..8: %4.1f%% %4.1f%% %4.1f%%  direct:%4.1f%%  skip:%4.1f%%",
3347                  i_mb_size[PIXEL_16x16] / (i_count*4),
3348                  (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
3349                  i_mb_size[PIXEL_8x8] / (i_count*4),
3350                  i_mb_count[B_DIRECT] / i_count,
3351                  i_mb_count[B_SKIP]   / i_count );
3352         if( i_mb_list_count != 0 )
3353             sprintf( buf + strlen(buf), "  L0:%4.1f%% L1:%4.1f%% BI:%4.1f%%",
3354                      list_count[0] / i_mb_list_count,
3355                      list_count[1] / i_mb_list_count,
3356                      list_count[2] / i_mb_list_count );
3357         x264_log( h, X264_LOG_INFO, "mb B  %s\n", buf );
3358     }
3359
3360     x264_ratecontrol_summary( h );
3361
3362     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 )
3363     {
3364 #define SUM3(p) (p[SLICE_TYPE_I] + p[SLICE_TYPE_P] + p[SLICE_TYPE_B])
3365 #define SUM3b(p,o) (p[SLICE_TYPE_I][o] + p[SLICE_TYPE_P][o] + p[SLICE_TYPE_B][o])
3366         int64_t i_i8x8 = SUM3b( h->stat.i_mb_count, I_8x8 );
3367         int64_t i_intra = i_i8x8 + SUM3b( h->stat.i_mb_count, I_4x4 )
3368                                  + SUM3b( h->stat.i_mb_count, I_16x16 );
3369         int64_t i_all_intra = i_intra + SUM3b( h->stat.i_mb_count, I_PCM);
3370         int64_t i_skip = SUM3b( h->stat.i_mb_count, P_SKIP )
3371                        + SUM3b( h->stat.i_mb_count, B_SKIP );
3372         const int i_count = h->stat.i_frame_count[SLICE_TYPE_I] +
3373                             h->stat.i_frame_count[SLICE_TYPE_P] +
3374                             h->stat.i_frame_count[SLICE_TYPE_B];
3375         int64_t i_mb_count = (int64_t)i_count * h->mb.i_mb_count;
3376         int64_t i_inter = i_mb_count - i_skip - i_intra;
3377         const double duration = h->stat.f_frame_duration[SLICE_TYPE_I] +
3378                                 h->stat.f_frame_duration[SLICE_TYPE_P] +
3379                                 h->stat.f_frame_duration[SLICE_TYPE_B];
3380         float f_bitrate = SUM3(h->stat.i_frame_size) / duration / 125;
3381
3382         if( PARAM_INTERLACED )
3383         {
3384             char *fieldstats = buf;
3385             fieldstats[0] = 0;
3386             if( i_inter )
3387                 fieldstats += sprintf( fieldstats, " inter:%.1f%%", h->stat.i_mb_field[1] * 100.0 / i_inter );
3388             if( i_skip )
3389                 fieldstats += sprintf( fieldstats, " skip:%.1f%%", h->stat.i_mb_field[2] * 100.0 / i_skip );
3390             x264_log( h, X264_LOG_INFO, "field mbs: intra: %.1f%%%s\n",
3391                       h->stat.i_mb_field[0] * 100.0 / i_intra, buf );
3392         }
3393
3394         if( h->pps->b_transform_8x8_mode )
3395         {
3396             buf[0] = 0;
3397             if( h->stat.i_mb_count_8x8dct[0] )
3398                 sprintf( buf, " inter:%.1f%%", 100. * h->stat.i_mb_count_8x8dct[1] / h->stat.i_mb_count_8x8dct[0] );
3399             x264_log( h, X264_LOG_INFO, "8x8 transform intra:%.1f%%%s\n", 100. * i_i8x8 / i_intra, buf );
3400         }
3401
3402         if( (h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO ||
3403             (h->stat.i_direct_frames[0] && h->stat.i_direct_frames[1]))
3404             && h->stat.i_frame_count[SLICE_TYPE_B] )
3405         {
3406             x264_log( h, X264_LOG_INFO, "direct mvs  spatial:%.1f%% temporal:%.1f%%\n",
3407                       h->stat.i_direct_frames[1] * 100. / h->stat.i_frame_count[SLICE_TYPE_B],
3408                       h->stat.i_direct_frames[0] * 100. / h->stat.i_frame_count[SLICE_TYPE_B] );
3409         }
3410
3411         buf[0] = 0;
3412         int csize = CHROMA444 ? 4 : 1;
3413         if( i_mb_count != i_all_intra )
3414             sprintf( buf, " inter: %.1f%% %.1f%% %.1f%%",
3415                      h->stat.i_mb_cbp[1] * 100.0 / ((i_mb_count - i_all_intra)*4),
3416                      h->stat.i_mb_cbp[3] * 100.0 / ((i_mb_count - i_all_intra)*csize),
3417                      h->stat.i_mb_cbp[5] * 100.0 / ((i_mb_count - i_all_intra)*csize) );
3418         x264_log( h, X264_LOG_INFO, "coded y,%s,%s intra: %.1f%% %.1f%% %.1f%%%s\n",
3419                   CHROMA444?"u":"uvDC", CHROMA444?"v":"uvAC",
3420                   h->stat.i_mb_cbp[0] * 100.0 / (i_all_intra*4),
3421                   h->stat.i_mb_cbp[2] * 100.0 / (i_all_intra*csize),
3422                   h->stat.i_mb_cbp[4] * 100.0 / (i_all_intra*csize), buf );
3423
3424         int64_t fixed_pred_modes[4][9] = {{0}};
3425         int64_t sum_pred_modes[4] = {0};
3426         for( int i = 0; i <= I_PRED_16x16_DC_128; i++ )
3427         {
3428             fixed_pred_modes[0][x264_mb_pred_mode16x16_fix[i]] += h->stat.i_mb_pred_mode[0][i];
3429             sum_pred_modes[0] += h->stat.i_mb_pred_mode[0][i];
3430         }
3431         if( sum_pred_modes[0] )
3432             x264_log( h, X264_LOG_INFO, "i16 v,h,dc,p: %2.0f%% %2.0f%% %2.0f%% %2.0f%%\n",
3433                       fixed_pred_modes[0][0] * 100.0 / sum_pred_modes[0],
3434                       fixed_pred_modes[0][1] * 100.0 / sum_pred_modes[0],
3435                       fixed_pred_modes[0][2] * 100.0 / sum_pred_modes[0],
3436                       fixed_pred_modes[0][3] * 100.0 / sum_pred_modes[0] );
3437         for( int i = 1; i <= 2; i++ )
3438         {
3439             for( int j = 0; j <= I_PRED_8x8_DC_128; j++ )
3440             {
3441                 fixed_pred_modes[i][x264_mb_pred_mode4x4_fix(j)] += h->stat.i_mb_pred_mode[i][j];
3442                 sum_pred_modes[i] += h->stat.i_mb_pred_mode[i][j];
3443             }
3444             if( sum_pred_modes[i] )
3445                 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,
3446                           fixed_pred_modes[i][0] * 100.0 / sum_pred_modes[i],
3447                           fixed_pred_modes[i][1] * 100.0 / sum_pred_modes[i],
3448                           fixed_pred_modes[i][2] * 100.0 / sum_pred_modes[i],
3449                           fixed_pred_modes[i][3] * 100.0 / sum_pred_modes[i],
3450                           fixed_pred_modes[i][4] * 100.0 / sum_pred_modes[i],
3451                           fixed_pred_modes[i][5] * 100.0 / sum_pred_modes[i],
3452                           fixed_pred_modes[i][6] * 100.0 / sum_pred_modes[i],
3453                           fixed_pred_modes[i][7] * 100.0 / sum_pred_modes[i],
3454                           fixed_pred_modes[i][8] * 100.0 / sum_pred_modes[i] );
3455         }
3456         for( int i = 0; i <= I_PRED_CHROMA_DC_128; i++ )
3457         {
3458             fixed_pred_modes[3][x264_mb_pred_mode8x8c_fix[i]] += h->stat.i_mb_pred_mode[3][i];
3459             sum_pred_modes[3] += h->stat.i_mb_pred_mode[3][i];
3460         }
3461         if( sum_pred_modes[3] && !CHROMA444 )
3462             x264_log( h, X264_LOG_INFO, "i8c dc,h,v,p: %2.0f%% %2.0f%% %2.0f%% %2.0f%%\n",
3463                       fixed_pred_modes[3][0] * 100.0 / sum_pred_modes[3],
3464                       fixed_pred_modes[3][1] * 100.0 / sum_pred_modes[3],
3465                       fixed_pred_modes[3][2] * 100.0 / sum_pred_modes[3],
3466                       fixed_pred_modes[3][3] * 100.0 / sum_pred_modes[3] );
3467
3468         if( h->param.analyse.i_weighted_pred >= X264_WEIGHTP_SIMPLE && h->stat.i_frame_count[SLICE_TYPE_P] > 0 )
3469             x264_log( h, X264_LOG_INFO, "Weighted P-Frames: Y:%.1f%% UV:%.1f%%\n",
3470                       h->stat.i_wpred[0] * 100.0 / h->stat.i_frame_count[SLICE_TYPE_P],
3471                       h->stat.i_wpred[1] * 100.0 / h->stat.i_frame_count[SLICE_TYPE_P] );
3472
3473         for( int i_list = 0; i_list < 2; i_list++ )
3474             for( int i_slice = 0; i_slice < 2; i_slice++ )
3475             {
3476                 char *p = buf;
3477                 int64_t i_den = 0;
3478                 int i_max = 0;
3479                 for( int i = 0; i < X264_REF_MAX*2; i++ )
3480                     if( h->stat.i_mb_count_ref[i_slice][i_list][i] )
3481                     {
3482                         i_den += h->stat.i_mb_count_ref[i_slice][i_list][i];
3483                         i_max = i;
3484                     }
3485                 if( i_max == 0 )
3486                     continue;
3487                 for( int i = 0; i <= i_max; i++ )
3488                     p += sprintf( p, " %4.1f%%", 100. * h->stat.i_mb_count_ref[i_slice][i_list][i] / i_den );
3489                 x264_log( h, X264_LOG_INFO, "ref %c L%d:%s\n", "PB"[i_slice], i_list, buf );
3490             }
3491
3492         if( h->param.analyse.b_ssim )
3493         {
3494             float ssim = SUM3( h->stat.f_ssim_mean_y ) / duration;
3495             x264_log( h, X264_LOG_INFO, "SSIM Mean Y:%.7f (%6.3fdb)\n", ssim, x264_ssim( ssim ) );
3496         }
3497         if( h->param.analyse.b_psnr )
3498         {
3499             x264_log( h, X264_LOG_INFO,
3500                       "PSNR Mean Y:%6.3f U:%6.3f V:%6.3f Avg:%6.3f Global:%6.3f kb/s:%.2f\n",
3501                       SUM3( h->stat.f_psnr_mean_y ) / duration,
3502                       SUM3( h->stat.f_psnr_mean_u ) / duration,
3503                       SUM3( h->stat.f_psnr_mean_v ) / duration,
3504                       SUM3( h->stat.f_psnr_average ) / duration,
3505                       x264_psnr( SUM3( h->stat.f_ssd_global ), duration * i_yuv_size ),
3506                       f_bitrate );
3507         }
3508         else
3509             x264_log( h, X264_LOG_INFO, "kb/s:%.2f\n", f_bitrate );
3510     }
3511
3512     /* rc */
3513     x264_ratecontrol_delete( h );
3514
3515     /* param */
3516     if( h->param.rc.psz_stat_out )
3517         free( h->param.rc.psz_stat_out );
3518     if( h->param.rc.psz_stat_in )
3519         free( h->param.rc.psz_stat_in );
3520
3521     x264_cqm_delete( h );
3522     x264_free( h->nal_buffer );
3523     x264_analyse_free_costs( h );
3524
3525     if( h->i_thread_frames > 1)
3526         h = h->thread[h->i_thread_phase];
3527
3528     /* frames */
3529     x264_frame_delete_list( h->frames.unused[0] );
3530     x264_frame_delete_list( h->frames.unused[1] );
3531     x264_frame_delete_list( h->frames.current );
3532     x264_frame_delete_list( h->frames.blank_unused );
3533
3534     h = h->thread[0];
3535
3536     for( int i = 0; i < h->i_thread_frames; i++ )
3537         if( h->thread[i]->b_thread_active )
3538             for( int j = 0; j < h->thread[i]->i_ref[0]; j++ )
3539                 if( h->thread[i]->fref[0][j] && h->thread[i]->fref[0][j]->b_duplicate )
3540                     x264_frame_delete( h->thread[i]->fref[0][j] );
3541
3542     for( int i = h->param.i_threads - 1; i >= 0; i-- )
3543     {
3544         x264_frame_t **frame;
3545
3546         if( !h->param.b_sliced_threads || i == 0 )
3547         {
3548             for( frame = h->thread[i]->frames.reference; *frame; frame++ )
3549             {
3550                 assert( (*frame)->i_reference_count > 0 );
3551                 (*frame)->i_reference_count--;
3552                 if( (*frame)->i_reference_count == 0 )
3553                     x264_frame_delete( *frame );
3554             }
3555             frame = &h->thread[i]->fdec;
3556             if( *frame )
3557             {
3558                 assert( (*frame)->i_reference_count > 0 );
3559                 (*frame)->i_reference_count--;
3560                 if( (*frame)->i_reference_count == 0 )
3561                     x264_frame_delete( *frame );
3562             }
3563             x264_macroblock_cache_free( h->thread[i] );
3564         }
3565         x264_macroblock_thread_free( h->thread[i], 0 );
3566         x264_free( h->thread[i]->out.p_bitstream );
3567         x264_free( h->thread[i]->out.nal);
3568         x264_free( h->thread[i] );
3569     }
3570 }
3571
3572 int x264_encoder_delayed_frames( x264_t *h )
3573 {
3574     int delayed_frames = 0;
3575     if( h->i_thread_frames > 1 )
3576     {
3577         for( int i = 0; i < h->i_thread_frames; i++ )
3578             delayed_frames += h->thread[i]->b_thread_active;
3579         h = h->thread[h->i_thread_phase];
3580     }
3581     for( int i = 0; h->frames.current[i]; i++ )
3582         delayed_frames++;
3583     x264_pthread_mutex_lock( &h->lookahead->ofbuf.mutex );
3584     x264_pthread_mutex_lock( &h->lookahead->ifbuf.mutex );
3585     x264_pthread_mutex_lock( &h->lookahead->next.mutex );
3586     delayed_frames += h->lookahead->ifbuf.i_size + h->lookahead->next.i_size + h->lookahead->ofbuf.i_size;
3587     x264_pthread_mutex_unlock( &h->lookahead->next.mutex );
3588     x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
3589     x264_pthread_mutex_unlock( &h->lookahead->ofbuf.mutex );
3590     return delayed_frames;
3591 }
3592
3593 int x264_encoder_maximum_delayed_frames( x264_t *h )
3594 {
3595     return h->frames.i_delay;
3596 }