Avoid possible NaNs in B-frame output stats
[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     /* Write the frame in display order */
73     fseek( f, (uint64_t)h->fdec->i_frame * h->param.i_height * h->param.i_width * 3/2 * sizeof(pixel), SEEK_SET );
74     for( int y = 0; y < h->param.i_height; y++ )
75         fwrite( &h->fdec->plane[0][y*h->fdec->i_stride[0]], sizeof(pixel), h->param.i_width, f );
76     int cw = h->param.i_width>>1;
77     int ch = h->param.i_height>>1;
78     pixel *planeu = x264_malloc( (cw*ch*2+32)*sizeof(pixel) );
79     pixel *planev = planeu + cw*ch + 16;
80     h->mc.plane_copy_deinterleave( planeu, cw, planev, cw, h->fdec->plane[1], h->fdec->i_stride[1], cw, ch );
81     fwrite( planeu, 1, cw*ch*sizeof(pixel), f );
82     fwrite( planev, 1, cw*ch*sizeof(pixel), f );
83     x264_free( planeu );
84     fclose( f );
85 }
86
87
88 /* Fill "default" values */
89 static void x264_slice_header_init( x264_t *h, x264_slice_header_t *sh,
90                                     x264_sps_t *sps, x264_pps_t *pps,
91                                     int i_idr_pic_id, int i_frame, int i_qp )
92 {
93     x264_param_t *param = &h->param;
94
95     /* First we fill all fields */
96     sh->sps = sps;
97     sh->pps = pps;
98
99     sh->i_first_mb  = 0;
100     sh->i_last_mb   = h->mb.i_mb_count - 1;
101     sh->i_pps_id    = pps->i_id;
102
103     sh->i_frame_num = i_frame;
104
105     sh->b_mbaff = PARAM_INTERLACED;
106     sh->b_field_pic = 0;    /* no field support for now */
107     sh->b_bottom_field = 0; /* not yet used */
108
109     sh->i_idr_pic_id = i_idr_pic_id;
110
111     /* poc stuff, fixed later */
112     sh->i_poc = 0;
113     sh->i_delta_poc_bottom = 0;
114     sh->i_delta_poc[0] = 0;
115     sh->i_delta_poc[1] = 0;
116
117     sh->i_redundant_pic_cnt = 0;
118
119     h->mb.b_direct_auto_write = h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO
120                                 && h->param.i_bframe
121                                 && ( h->param.rc.b_stat_write || !h->param.rc.b_stat_read );
122
123     if( !h->mb.b_direct_auto_read && sh->i_type == SLICE_TYPE_B )
124     {
125         if( h->fref[1][0]->i_poc_l0ref0 == h->fref[0][0]->i_poc )
126         {
127             if( h->mb.b_direct_auto_write )
128                 sh->b_direct_spatial_mv_pred = ( h->stat.i_direct_score[1] > h->stat.i_direct_score[0] );
129             else
130                 sh->b_direct_spatial_mv_pred = ( param->analyse.i_direct_mv_pred == X264_DIRECT_PRED_SPATIAL );
131         }
132         else
133         {
134             h->mb.b_direct_auto_write = 0;
135             sh->b_direct_spatial_mv_pred = 1;
136         }
137     }
138     /* else b_direct_spatial_mv_pred was read from the 2pass statsfile */
139
140     sh->b_num_ref_idx_override = 0;
141     sh->i_num_ref_idx_l0_active = 1;
142     sh->i_num_ref_idx_l1_active = 1;
143
144     sh->b_ref_pic_list_reordering[0] = h->b_ref_reorder[0];
145     sh->b_ref_pic_list_reordering[1] = h->b_ref_reorder[1];
146
147     /* If the ref list isn't in the default order, construct reordering header */
148     for( int list = 0; list < 2; list++ )
149     {
150         if( sh->b_ref_pic_list_reordering[list] )
151         {
152             int pred_frame_num = i_frame;
153             for( int i = 0; i < h->i_ref[list]; i++ )
154             {
155                 int diff = h->fref[list][i]->i_frame_num - pred_frame_num;
156                 sh->ref_pic_list_order[list][i].idc = ( diff > 0 );
157                 sh->ref_pic_list_order[list][i].arg = (abs(diff) - 1) & ((1 << sps->i_log2_max_frame_num) - 1);
158                 pred_frame_num = h->fref[list][i]->i_frame_num;
159             }
160         }
161     }
162
163     sh->i_cabac_init_idc = param->i_cabac_init_idc;
164
165     sh->i_qp = SPEC_QP(i_qp);
166     sh->i_qp_delta = sh->i_qp - pps->i_pic_init_qp;
167     sh->b_sp_for_swidth = 0;
168     sh->i_qs_delta = 0;
169
170     int deblock_thresh = i_qp + 2 * X264_MIN(param->i_deblocking_filter_alphac0, param->i_deblocking_filter_beta);
171     /* If effective qp <= 15, deblocking would have no effect anyway */
172     if( param->b_deblocking_filter && (h->mb.b_variable_qp || 15 < deblock_thresh ) )
173         sh->i_disable_deblocking_filter_idc = param->b_sliced_threads ? 2 : 0;
174     else
175         sh->i_disable_deblocking_filter_idc = 1;
176     sh->i_alpha_c0_offset = param->i_deblocking_filter_alphac0 << 1;
177     sh->i_beta_offset = param->i_deblocking_filter_beta << 1;
178 }
179
180 static void x264_slice_header_write( bs_t *s, x264_slice_header_t *sh, int i_nal_ref_idc )
181 {
182     if( sh->b_mbaff )
183     {
184         int first_x = sh->i_first_mb % sh->sps->i_mb_width;
185         int first_y = sh->i_first_mb / sh->sps->i_mb_width;
186         assert( (first_y&1) == 0 );
187         bs_write_ue( s, (2*first_x + sh->sps->i_mb_width*(first_y&~1) + (first_y&1)) >> 1 );
188     }
189     else
190         bs_write_ue( s, sh->i_first_mb );
191
192     bs_write_ue( s, sh->i_type + 5 );   /* same type things */
193     bs_write_ue( s, sh->i_pps_id );
194     bs_write( s, sh->sps->i_log2_max_frame_num, sh->i_frame_num & ((1<<sh->sps->i_log2_max_frame_num)-1) );
195
196     if( !sh->sps->b_frame_mbs_only )
197     {
198         bs_write1( s, sh->b_field_pic );
199         if( sh->b_field_pic )
200             bs_write1( s, sh->b_bottom_field );
201     }
202
203     if( sh->i_idr_pic_id >= 0 ) /* NAL IDR */
204         bs_write_ue( s, sh->i_idr_pic_id );
205
206     if( sh->sps->i_poc_type == 0 )
207     {
208         bs_write( s, sh->sps->i_log2_max_poc_lsb, sh->i_poc & ((1<<sh->sps->i_log2_max_poc_lsb)-1) );
209         if( sh->pps->b_pic_order && !sh->b_field_pic )
210             bs_write_se( s, sh->i_delta_poc_bottom );
211     }
212     else if( sh->sps->i_poc_type == 1 && !sh->sps->b_delta_pic_order_always_zero )
213     {
214         bs_write_se( s, sh->i_delta_poc[0] );
215         if( sh->pps->b_pic_order && !sh->b_field_pic )
216             bs_write_se( s, sh->i_delta_poc[1] );
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     if( h->param.i_width % 2 || h->param.i_height % 2 )
411     {
412         x264_log( h, X264_LOG_ERROR, "width or height not divisible by 2 (%dx%d)\n",
413                   h->param.i_width, h->param.i_height );
414         return -1;
415     }
416     int i_csp = h->param.i_csp & X264_CSP_MASK;
417     if( i_csp <= X264_CSP_NONE || i_csp >= X264_CSP_MAX )
418     {
419         x264_log( h, X264_LOG_ERROR, "invalid CSP (only I420/YV12/NV12 supported)\n" );
420         return -1;
421     }
422
423 #if HAVE_INTERLACED
424     h->param.b_interlaced = !!PARAM_INTERLACED;
425 #else
426     if( h->param.b_interlaced )
427     {
428         x264_log( h, X264_LOG_ERROR, "not compiled with interlaced support\n" );
429         return -1;
430     }
431 #endif
432
433     if( (h->param.crop_rect.i_left + h->param.crop_rect.i_right ) >= h->param.i_width ||
434         (h->param.crop_rect.i_top  + h->param.crop_rect.i_bottom) >= h->param.i_height )
435     {
436         x264_log( h, X264_LOG_ERROR, "invalid crop-rect %u,%u,%u,%u\n", h->param.crop_rect.i_left,
437                   h->param.crop_rect.i_top, h->param.crop_rect.i_right,  h->param.crop_rect.i_bottom );
438         return -1;
439     }
440
441     if( h->param.i_threads == X264_THREADS_AUTO )
442         h->param.i_threads = x264_cpu_num_processors() * (h->param.b_sliced_threads?2:3)/2;
443     h->param.i_threads = x264_clip3( h->param.i_threads, 1, X264_THREAD_MAX );
444     if( h->param.i_threads > 1 )
445     {
446 #if !HAVE_THREAD
447         x264_log( h, X264_LOG_WARNING, "not compiled with thread support!\n");
448         h->param.i_threads = 1;
449 #endif
450         /* Avoid absurdly small thread slices as they can reduce performance
451          * and VBV compliance.  Capped at an arbitrary 4 rows per thread. */
452         if( h->param.b_sliced_threads )
453         {
454             int max_threads = (h->param.i_height+15)/16 / 4;
455             h->param.i_threads = X264_MIN( h->param.i_threads, max_threads );
456         }
457     }
458     else
459         h->param.b_sliced_threads = 0;
460     h->i_thread_frames = h->param.b_sliced_threads ? 1 : h->param.i_threads;
461     if( h->i_thread_frames > 1 )
462         h->param.nalu_process = NULL;
463
464     h->param.i_keyint_max = x264_clip3( h->param.i_keyint_max, 1, X264_KEYINT_MAX_INFINITE );
465     if( h->param.i_keyint_max == 1 )
466     {
467         h->param.b_intra_refresh = 0;
468         h->param.analyse.i_weighted_pred = 0;
469     }
470
471     h->param.i_frame_packing = x264_clip3( h->param.i_frame_packing, -1, 5 );
472
473     /* Detect default ffmpeg settings and terminate with an error. */
474     if( b_open )
475     {
476         int score = 0;
477         score += h->param.analyse.i_me_range == 0;
478         score += h->param.rc.i_qp_step == 3;
479         score += h->param.i_keyint_max == 12;
480         score += h->param.rc.i_qp_min == 2;
481         score += h->param.rc.i_qp_max == 31;
482         score += h->param.rc.f_qcompress == 0.5;
483         score += fabs(h->param.rc.f_ip_factor - 1.25) < 0.01;
484         score += fabs(h->param.rc.f_pb_factor - 1.25) < 0.01;
485         score += h->param.analyse.inter == 0 && h->param.analyse.i_subpel_refine == 8;
486         if( score >= 5 )
487         {
488             x264_log( h, X264_LOG_ERROR, "broken ffmpeg default settings detected\n" );
489             x264_log( h, X264_LOG_ERROR, "use an encoding preset (e.g. -vpre medium)\n" );
490             x264_log( h, X264_LOG_ERROR, "preset usage: -vpre <speed> -vpre <profile>\n" );
491             x264_log( h, X264_LOG_ERROR, "speed presets are listed in x264 --help\n" );
492             x264_log( h, X264_LOG_ERROR, "profile is optional; x264 defaults to high\n" );
493             return -1;
494         }
495     }
496
497     if( h->param.rc.i_rc_method < 0 || h->param.rc.i_rc_method > 2 )
498     {
499         x264_log( h, X264_LOG_ERROR, "no ratecontrol method specified\n" );
500         return -1;
501     }
502     h->param.rc.f_rf_constant = x264_clip3f( h->param.rc.f_rf_constant, -QP_BD_OFFSET, 51 );
503     h->param.rc.f_rf_constant_max = x264_clip3f( h->param.rc.f_rf_constant_max, -QP_BD_OFFSET, 51 );
504     h->param.rc.i_qp_constant = x264_clip3( h->param.rc.i_qp_constant, 0, QP_MAX );
505     h->param.analyse.i_subpel_refine = x264_clip3( h->param.analyse.i_subpel_refine, 0, 10 );
506     h->param.rc.f_ip_factor = X264_MAX( h->param.rc.f_ip_factor, 0.01f );
507     h->param.rc.f_pb_factor = X264_MAX( h->param.rc.f_pb_factor, 0.01f );
508     if( h->param.rc.i_rc_method == X264_RC_CRF )
509     {
510         h->param.rc.i_qp_constant = h->param.rc.f_rf_constant + QP_BD_OFFSET;
511         h->param.rc.i_bitrate = 0;
512     }
513     if( (h->param.rc.i_rc_method == X264_RC_CQP || h->param.rc.i_rc_method == X264_RC_CRF)
514         && h->param.rc.i_qp_constant == 0 )
515     {
516         h->mb.b_lossless = 1;
517         h->param.i_cqm_preset = X264_CQM_FLAT;
518         h->param.psz_cqm_file = NULL;
519         h->param.rc.i_rc_method = X264_RC_CQP;
520         h->param.rc.f_ip_factor = 1;
521         h->param.rc.f_pb_factor = 1;
522         h->param.analyse.b_psnr = 0;
523         h->param.analyse.b_ssim = 0;
524         h->param.analyse.i_chroma_qp_offset = 0;
525         h->param.analyse.i_trellis = 0;
526         h->param.analyse.b_fast_pskip = 0;
527         h->param.analyse.i_noise_reduction = 0;
528         h->param.analyse.b_psy = 0;
529         h->param.i_bframe = 0;
530         /* 8x8dct is not useful without RD in CAVLC lossless */
531         if( !h->param.b_cabac && h->param.analyse.i_subpel_refine < 6 )
532             h->param.analyse.b_transform_8x8 = 0;
533     }
534     if( h->param.rc.i_rc_method == X264_RC_CQP )
535     {
536         float qp_p = h->param.rc.i_qp_constant;
537         float qp_i = qp_p - 6*log2f( h->param.rc.f_ip_factor );
538         float qp_b = qp_p + 6*log2f( h->param.rc.f_pb_factor );
539         h->param.rc.i_qp_min = x264_clip3( (int)(X264_MIN3( qp_p, qp_i, qp_b )), 0, QP_MAX );
540         h->param.rc.i_qp_max = x264_clip3( (int)(X264_MAX3( qp_p, qp_i, qp_b ) + .999), 0, QP_MAX );
541         h->param.rc.i_aq_mode = 0;
542         h->param.rc.b_mb_tree = 0;
543         h->param.rc.i_bitrate = 0;
544     }
545     h->param.rc.i_qp_max = x264_clip3( h->param.rc.i_qp_max, 0, QP_MAX );
546     h->param.rc.i_qp_min = x264_clip3( h->param.rc.i_qp_min, 0, h->param.rc.i_qp_max );
547     h->param.rc.i_qp_step = x264_clip3( h->param.rc.i_qp_step, 0, QP_MAX );
548     h->param.rc.i_bitrate = x264_clip3( h->param.rc.i_bitrate, 0, 2000000 );
549     h->param.rc.i_vbv_buffer_size = x264_clip3( h->param.rc.i_vbv_buffer_size, 0, 2000000 );
550     h->param.rc.i_vbv_max_bitrate = x264_clip3( h->param.rc.i_vbv_max_bitrate, 0, 2000000 );
551     h->param.rc.f_vbv_buffer_init = x264_clip3f( h->param.rc.f_vbv_buffer_init, 0, 2000000 );
552     if( h->param.rc.i_vbv_buffer_size )
553     {
554         if( h->param.rc.i_rc_method == X264_RC_CQP )
555         {
556             x264_log( h, X264_LOG_WARNING, "VBV is incompatible with constant QP, ignored.\n" );
557             h->param.rc.i_vbv_max_bitrate = 0;
558             h->param.rc.i_vbv_buffer_size = 0;
559         }
560         else if( h->param.rc.i_vbv_max_bitrate == 0 )
561         {
562             if( h->param.rc.i_rc_method == X264_RC_ABR )
563             {
564                 x264_log( h, X264_LOG_WARNING, "VBV maxrate unspecified, assuming CBR\n" );
565                 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate;
566             }
567             else
568             {
569                 x264_log( h, X264_LOG_WARNING, "VBV bufsize set but maxrate unspecified, ignored\n" );
570                 h->param.rc.i_vbv_buffer_size = 0;
571             }
572         }
573         else if( h->param.rc.i_vbv_max_bitrate < h->param.rc.i_bitrate &&
574                  h->param.rc.i_rc_method == X264_RC_ABR )
575         {
576             x264_log( h, X264_LOG_WARNING, "max bitrate less than average bitrate, assuming CBR\n" );
577             h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate;
578         }
579     }
580     else if( h->param.rc.i_vbv_max_bitrate )
581     {
582         x264_log( h, X264_LOG_WARNING, "VBV maxrate specified, but no bufsize, ignored\n" );
583         h->param.rc.i_vbv_max_bitrate = 0;
584     }
585
586     h->param.i_slice_max_size = X264_MAX( h->param.i_slice_max_size, 0 );
587     h->param.i_slice_max_mbs = X264_MAX( h->param.i_slice_max_mbs, 0 );
588
589     int max_slices = (h->param.i_height+((16<<PARAM_INTERLACED)-1))/(16<<PARAM_INTERLACED);
590     if( h->param.b_sliced_threads )
591         h->param.i_slice_count = x264_clip3( h->param.i_threads, 0, max_slices );
592     else
593     {
594         h->param.i_slice_count = x264_clip3( h->param.i_slice_count, 0, max_slices );
595         if( h->param.i_slice_max_mbs || h->param.i_slice_max_size )
596             h->param.i_slice_count = 0;
597     }
598
599     if( h->param.b_bluray_compat )
600     {
601         h->param.i_bframe_pyramid = X264_MIN( X264_B_PYRAMID_STRICT, h->param.i_bframe_pyramid );
602         h->param.i_bframe = X264_MIN( h->param.i_bframe, 3 );
603         h->param.b_aud = 1;
604         h->param.i_nal_hrd = X264_MAX( h->param.i_nal_hrd, X264_NAL_HRD_VBR );
605         h->param.i_slice_max_size = 0;
606         h->param.i_slice_max_mbs = 0;
607         h->param.b_intra_refresh = 0;
608         h->param.i_frame_reference = X264_MIN( h->param.i_frame_reference, 6 );
609         h->param.i_dpb_size = X264_MIN( h->param.i_dpb_size, 6 );
610         /* Due to the proliferation of broken players that don't handle dupes properly. */
611         h->param.analyse.i_weighted_pred = X264_MIN( h->param.analyse.i_weighted_pred, X264_WEIGHTP_SIMPLE );
612         if( h->param.b_fake_interlaced )
613             h->param.b_pic_struct = 1;
614     }
615
616     h->param.i_frame_reference = x264_clip3( h->param.i_frame_reference, 1, X264_REF_MAX );
617     h->param.i_dpb_size = x264_clip3( h->param.i_dpb_size, 1, X264_REF_MAX );
618     if( h->param.i_scenecut_threshold < 0 )
619         h->param.i_scenecut_threshold = 0;
620     h->param.analyse.i_direct_mv_pred = x264_clip3( h->param.analyse.i_direct_mv_pred, X264_DIRECT_PRED_NONE, X264_DIRECT_PRED_AUTO );
621     if( !h->param.analyse.i_subpel_refine && h->param.analyse.i_direct_mv_pred > X264_DIRECT_PRED_SPATIAL )
622     {
623         x264_log( h, X264_LOG_WARNING, "subme=0 + direct=temporal is not supported\n" );
624         h->param.analyse.i_direct_mv_pred = X264_DIRECT_PRED_SPATIAL;
625     }
626     h->param.i_bframe = x264_clip3( h->param.i_bframe, 0, X264_MIN( X264_BFRAME_MAX, h->param.i_keyint_max-1 ) );
627     h->param.i_bframe_bias = x264_clip3( h->param.i_bframe_bias, -90, 100 );
628     if( h->param.i_bframe <= 1 )
629         h->param.i_bframe_pyramid = X264_B_PYRAMID_NONE;
630     h->param.i_bframe_pyramid = x264_clip3( h->param.i_bframe_pyramid, X264_B_PYRAMID_NONE, X264_B_PYRAMID_NORMAL );
631     h->param.i_bframe_adaptive = x264_clip3( h->param.i_bframe_adaptive, X264_B_ADAPT_NONE, X264_B_ADAPT_TRELLIS );
632     if( !h->param.i_bframe )
633     {
634         h->param.i_bframe_adaptive = X264_B_ADAPT_NONE;
635         h->param.analyse.i_direct_mv_pred = 0;
636         h->param.analyse.b_weighted_bipred = 0;
637         h->param.b_open_gop = 0;
638     }
639     if( h->param.b_intra_refresh && h->param.i_bframe_pyramid == X264_B_PYRAMID_NORMAL )
640     {
641         x264_log( h, X264_LOG_WARNING, "b-pyramid normal + intra-refresh is not supported\n" );
642         h->param.i_bframe_pyramid = X264_B_PYRAMID_STRICT;
643     }
644     if( h->param.b_intra_refresh && (h->param.i_frame_reference > 1 || h->param.i_dpb_size > 1) )
645     {
646         x264_log( h, X264_LOG_WARNING, "ref > 1 + intra-refresh is not supported\n" );
647         h->param.i_frame_reference = 1;
648         h->param.i_dpb_size = 1;
649     }
650     if( h->param.b_intra_refresh && h->param.b_open_gop )
651     {
652         x264_log( h, X264_LOG_WARNING, "intra-refresh is not compatible with open-gop\n" );
653         h->param.b_open_gop = 0;
654     }
655     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;
656     if( h->param.i_keyint_min == X264_KEYINT_MIN_AUTO )
657         h->param.i_keyint_min = X264_MIN( h->param.i_keyint_max / 10, fps );
658     h->param.i_keyint_min = x264_clip3( h->param.i_keyint_min, 1, h->param.i_keyint_max/2+1 );
659     h->param.rc.i_lookahead = x264_clip3( h->param.rc.i_lookahead, 0, X264_LOOKAHEAD_MAX );
660     {
661         int maxrate = X264_MAX( h->param.rc.i_vbv_max_bitrate, h->param.rc.i_bitrate );
662         float bufsize = maxrate ? (float)h->param.rc.i_vbv_buffer_size / maxrate : 0;
663         h->param.rc.i_lookahead = X264_MIN( h->param.rc.i_lookahead, X264_MAX( h->param.i_keyint_max, bufsize*fps ) );
664     }
665
666     if( !h->param.i_timebase_num || !h->param.i_timebase_den || !(h->param.b_vfr_input || h->param.b_pulldown) )
667     {
668         h->param.i_timebase_num = h->param.i_fps_den;
669         h->param.i_timebase_den = h->param.i_fps_num;
670     }
671
672     h->param.rc.f_qcompress = x264_clip3f( h->param.rc.f_qcompress, 0.0, 1.0 );
673     if( h->param.i_keyint_max == 1 || h->param.rc.f_qcompress == 1 )
674         h->param.rc.b_mb_tree = 0;
675     if( (!h->param.b_intra_refresh && h->param.i_keyint_max != X264_KEYINT_MAX_INFINITE) &&
676         !h->param.rc.i_lookahead && h->param.rc.b_mb_tree )
677     {
678         x264_log( h, X264_LOG_WARNING, "lookaheadless mb-tree requires intra refresh or infinite keyint\n" );
679         h->param.rc.b_mb_tree = 0;
680     }
681     if( h->param.rc.b_stat_read )
682         h->param.rc.i_lookahead = 0;
683 #if HAVE_THREAD
684     if( h->param.i_sync_lookahead < 0 )
685         h->param.i_sync_lookahead = h->param.i_bframe + 1;
686     h->param.i_sync_lookahead = X264_MIN( h->param.i_sync_lookahead, X264_LOOKAHEAD_MAX );
687     if( h->param.rc.b_stat_read || h->i_thread_frames == 1 )
688         h->param.i_sync_lookahead = 0;
689 #else
690     h->param.i_sync_lookahead = 0;
691 #endif
692
693     h->param.i_deblocking_filter_alphac0 = x264_clip3( h->param.i_deblocking_filter_alphac0, -6, 6 );
694     h->param.i_deblocking_filter_beta    = x264_clip3( h->param.i_deblocking_filter_beta, -6, 6 );
695     h->param.analyse.i_luma_deadzone[0] = x264_clip3( h->param.analyse.i_luma_deadzone[0], 0, 32 );
696     h->param.analyse.i_luma_deadzone[1] = x264_clip3( h->param.analyse.i_luma_deadzone[1], 0, 32 );
697
698     h->param.i_cabac_init_idc = x264_clip3( h->param.i_cabac_init_idc, 0, 2 );
699
700     if( h->param.i_cqm_preset < X264_CQM_FLAT || h->param.i_cqm_preset > X264_CQM_CUSTOM )
701         h->param.i_cqm_preset = X264_CQM_FLAT;
702
703     if( h->param.analyse.i_me_method < X264_ME_DIA ||
704         h->param.analyse.i_me_method > X264_ME_TESA )
705         h->param.analyse.i_me_method = X264_ME_HEX;
706     h->param.analyse.i_me_range = x264_clip3( h->param.analyse.i_me_range, 4, 1024 );
707     if( h->param.analyse.i_me_range > 16 && h->param.analyse.i_me_method <= X264_ME_HEX )
708         h->param.analyse.i_me_range = 16;
709     if( h->param.analyse.i_me_method == X264_ME_TESA &&
710         (h->mb.b_lossless || h->param.analyse.i_subpel_refine <= 1) )
711         h->param.analyse.i_me_method = X264_ME_ESA;
712     h->param.analyse.b_mixed_references = h->param.analyse.b_mixed_references && h->param.i_frame_reference > 1;
713     h->param.analyse.inter &= X264_ANALYSE_PSUB16x16|X264_ANALYSE_PSUB8x8|X264_ANALYSE_BSUB16x16|
714                               X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
715     h->param.analyse.intra &= X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
716     if( !(h->param.analyse.inter & X264_ANALYSE_PSUB16x16) )
717         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
718     if( !h->param.analyse.b_transform_8x8 )
719     {
720         h->param.analyse.inter &= ~X264_ANALYSE_I8x8;
721         h->param.analyse.intra &= ~X264_ANALYSE_I8x8;
722     }
723     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -12, 12);
724     h->param.analyse.i_trellis = x264_clip3( h->param.analyse.i_trellis, 0, 2 );
725     h->param.rc.i_aq_mode = x264_clip3( h->param.rc.i_aq_mode, 0, 2 );
726     h->param.rc.f_aq_strength = x264_clip3f( h->param.rc.f_aq_strength, 0, 3 );
727     if( h->param.rc.f_aq_strength == 0 )
728         h->param.rc.i_aq_mode = 0;
729
730     if( h->param.i_log_level < X264_LOG_INFO )
731     {
732         h->param.analyse.b_psnr = 0;
733         h->param.analyse.b_ssim = 0;
734     }
735     /* Warn users trying to measure PSNR/SSIM with psy opts on. */
736     if( b_open && (h->param.analyse.b_psnr || h->param.analyse.b_ssim) )
737     {
738         char *s = NULL;
739
740         if( h->param.analyse.b_psy )
741         {
742             s = h->param.analyse.b_psnr ? "psnr" : "ssim";
743             x264_log( h, X264_LOG_WARNING, "--%s used with psy on: results will be invalid!\n", s );
744         }
745         else if( !h->param.rc.i_aq_mode && h->param.analyse.b_ssim )
746         {
747             x264_log( h, X264_LOG_WARNING, "--ssim used with AQ off: results will be invalid!\n" );
748             s = "ssim";
749         }
750         else if(  h->param.rc.i_aq_mode && h->param.analyse.b_psnr )
751         {
752             x264_log( h, X264_LOG_WARNING, "--psnr used with AQ on: results will be invalid!\n" );
753             s = "psnr";
754         }
755         if( s )
756             x264_log( h, X264_LOG_WARNING, "--tune %s should be used if attempting to benchmark %s!\n", s, s );
757     }
758
759     if( !h->param.analyse.b_psy )
760     {
761         h->param.analyse.f_psy_rd = 0;
762         h->param.analyse.f_psy_trellis = 0;
763     }
764     h->param.analyse.f_psy_rd = x264_clip3f( h->param.analyse.f_psy_rd, 0, 10 );
765     h->param.analyse.f_psy_trellis = x264_clip3f( h->param.analyse.f_psy_trellis, 0, 10 );
766     h->mb.i_psy_rd = h->param.analyse.i_subpel_refine >= 6 ? FIX8( h->param.analyse.f_psy_rd ) : 0;
767     h->mb.i_psy_trellis = h->param.analyse.i_trellis ? FIX8( h->param.analyse.f_psy_trellis / 4 ) : 0;
768     /* Psy RDO increases overall quantizers to improve the quality of luma--this indirectly hurts chroma quality */
769     /* so we lower the chroma QP offset to compensate */
770     if( b_open && h->mb.i_psy_rd )
771         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_rd < 0.25 ? 1 : 2;
772     /* Psy trellis has a similar effect. */
773     if( b_open && h->mb.i_psy_trellis )
774         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_trellis < 0.25 ? 1 : 2;
775     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -12, 12);
776     /* MB-tree requires AQ to be on, even if the strength is zero. */
777     if( !h->param.rc.i_aq_mode && h->param.rc.b_mb_tree )
778     {
779         h->param.rc.i_aq_mode = 1;
780         h->param.rc.f_aq_strength = 0;
781     }
782     h->param.analyse.i_noise_reduction = x264_clip3( h->param.analyse.i_noise_reduction, 0, 1<<16 );
783     if( h->param.analyse.i_subpel_refine == 10 && (h->param.analyse.i_trellis != 2 || !h->param.rc.i_aq_mode) )
784         h->param.analyse.i_subpel_refine = 9;
785
786     {
787         const x264_level_t *l = x264_levels;
788         if( h->param.i_level_idc < 0 )
789         {
790             int maxrate_bak = h->param.rc.i_vbv_max_bitrate;
791             if( h->param.rc.i_rc_method == X264_RC_ABR && h->param.rc.i_vbv_buffer_size <= 0 )
792                 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate * 2;
793             h->sps = h->sps_array;
794             x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
795             do h->param.i_level_idc = l->level_idc;
796                 while( l[1].level_idc && x264_validate_levels( h, 0 ) && l++ );
797             h->param.rc.i_vbv_max_bitrate = maxrate_bak;
798         }
799         else
800         {
801             while( l->level_idc && l->level_idc != h->param.i_level_idc )
802                 l++;
803             if( l->level_idc == 0 )
804             {
805                 x264_log( h, X264_LOG_ERROR, "invalid level_idc: %d\n", h->param.i_level_idc );
806                 return -1;
807             }
808         }
809         if( h->param.analyse.i_mv_range <= 0 )
810             h->param.analyse.i_mv_range = l->mv_range >> PARAM_INTERLACED;
811         else
812             h->param.analyse.i_mv_range = x264_clip3(h->param.analyse.i_mv_range, 32, 512 >> PARAM_INTERLACED);
813     }
814
815     h->param.analyse.i_weighted_pred = x264_clip3( h->param.analyse.i_weighted_pred, X264_WEIGHTP_NONE, X264_WEIGHTP_SMART );
816
817     if( PARAM_INTERLACED )
818     {
819         if( h->param.analyse.i_me_method >= X264_ME_ESA )
820         {
821             x264_log( h, X264_LOG_WARNING, "interlace + me=esa is not implemented\n" );
822             h->param.analyse.i_me_method = X264_ME_UMH;
823         }
824         if( h->param.analyse.i_weighted_pred > 0 )
825         {
826             x264_log( h, X264_LOG_WARNING, "interlace + weightp is not implemented\n" );
827             h->param.analyse.i_weighted_pred = X264_WEIGHTP_NONE;
828         }
829     }
830
831     if( !h->param.analyse.i_weighted_pred && h->param.rc.b_mb_tree && h->param.analyse.b_psy )
832         h->param.analyse.i_weighted_pred = X264_WEIGHTP_FAKE;
833
834     if( h->i_thread_frames > 1 )
835     {
836         int r = h->param.analyse.i_mv_range_thread;
837         int r2;
838         if( r <= 0 )
839         {
840             // half of the available space is reserved and divided evenly among the threads,
841             // the rest is allocated to whichever thread is far enough ahead to use it.
842             // reserving more space increases quality for some videos, but costs more time
843             // in thread synchronization.
844             int max_range = (h->param.i_height + X264_THREAD_HEIGHT) / h->i_thread_frames - X264_THREAD_HEIGHT;
845             r = max_range / 2;
846         }
847         r = X264_MAX( r, h->param.analyse.i_me_range );
848         r = X264_MIN( r, h->param.analyse.i_mv_range );
849         // round up to use the whole mb row
850         r2 = (r & ~15) + ((-X264_THREAD_HEIGHT) & 15);
851         if( r2 < r )
852             r2 += 16;
853         x264_log( h, X264_LOG_DEBUG, "using mv_range_thread = %d\n", r2 );
854         h->param.analyse.i_mv_range_thread = r2;
855     }
856
857     if( h->param.rc.f_rate_tolerance < 0 )
858         h->param.rc.f_rate_tolerance = 0;
859     if( h->param.rc.f_qblur < 0 )
860         h->param.rc.f_qblur = 0;
861     if( h->param.rc.f_complexity_blur < 0 )
862         h->param.rc.f_complexity_blur = 0;
863
864     h->param.i_sps_id &= 31;
865
866     if( PARAM_INTERLACED )
867         h->param.b_pic_struct = 1;
868
869     h->param.i_nal_hrd = x264_clip3( h->param.i_nal_hrd, X264_NAL_HRD_NONE, X264_NAL_HRD_CBR );
870
871     if( h->param.i_nal_hrd && !h->param.rc.i_vbv_buffer_size )
872     {
873         x264_log( h, X264_LOG_WARNING, "NAL HRD parameters require VBV parameters\n" );
874         h->param.i_nal_hrd = X264_NAL_HRD_NONE;
875     }
876
877     if( h->param.i_nal_hrd == X264_NAL_HRD_CBR &&
878        (h->param.rc.i_bitrate != h->param.rc.i_vbv_max_bitrate || !h->param.rc.i_vbv_max_bitrate) )
879     {
880         x264_log( h, X264_LOG_WARNING, "CBR HRD requires constant bitrate\n" );
881         h->param.i_nal_hrd = X264_NAL_HRD_VBR;
882     }
883
884     /* ensure the booleans are 0 or 1 so they can be used in math */
885 #define BOOLIFY(x) h->param.x = !!h->param.x
886     BOOLIFY( b_cabac );
887     BOOLIFY( b_constrained_intra );
888     BOOLIFY( b_deblocking_filter );
889     BOOLIFY( b_deterministic );
890     BOOLIFY( b_sliced_threads );
891     BOOLIFY( b_interlaced );
892     BOOLIFY( b_intra_refresh );
893     BOOLIFY( b_visualize );
894     BOOLIFY( b_aud );
895     BOOLIFY( b_repeat_headers );
896     BOOLIFY( b_annexb );
897     BOOLIFY( b_vfr_input );
898     BOOLIFY( b_pulldown );
899     BOOLIFY( b_tff );
900     BOOLIFY( b_pic_struct );
901     BOOLIFY( b_fake_interlaced );
902     BOOLIFY( b_open_gop );
903     BOOLIFY( b_bluray_compat );
904     BOOLIFY( analyse.b_transform_8x8 );
905     BOOLIFY( analyse.b_weighted_bipred );
906     BOOLIFY( analyse.b_chroma_me );
907     BOOLIFY( analyse.b_mixed_references );
908     BOOLIFY( analyse.b_fast_pskip );
909     BOOLIFY( analyse.b_dct_decimate );
910     BOOLIFY( analyse.b_psy );
911     BOOLIFY( analyse.b_psnr );
912     BOOLIFY( analyse.b_ssim );
913     BOOLIFY( rc.b_stat_write );
914     BOOLIFY( rc.b_stat_read );
915     BOOLIFY( rc.b_mb_tree );
916 #undef BOOLIFY
917
918     return 0;
919 }
920
921 static void mbcmp_init( x264_t *h )
922 {
923     int satd = !h->mb.b_lossless && h->param.analyse.i_subpel_refine > 1;
924     memcpy( h->pixf.mbcmp, satd ? h->pixf.satd : h->pixf.sad_aligned, sizeof(h->pixf.mbcmp) );
925     memcpy( h->pixf.mbcmp_unaligned, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.mbcmp_unaligned) );
926     h->pixf.intra_mbcmp_x3_16x16 = satd ? h->pixf.intra_satd_x3_16x16 : h->pixf.intra_sad_x3_16x16;
927     h->pixf.intra_mbcmp_x3_8x8c = satd ? h->pixf.intra_satd_x3_8x8c : h->pixf.intra_sad_x3_8x8c;
928     h->pixf.intra_mbcmp_x3_8x8 = satd ? h->pixf.intra_sa8d_x3_8x8 : h->pixf.intra_sad_x3_8x8;
929     h->pixf.intra_mbcmp_x3_4x4 = satd ? h->pixf.intra_satd_x3_4x4 : h->pixf.intra_sad_x3_4x4;
930     satd &= h->param.analyse.i_me_method == X264_ME_TESA;
931     memcpy( h->pixf.fpelcmp, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.fpelcmp) );
932     memcpy( h->pixf.fpelcmp_x3, satd ? h->pixf.satd_x3 : h->pixf.sad_x3, sizeof(h->pixf.fpelcmp_x3) );
933     memcpy( h->pixf.fpelcmp_x4, satd ? h->pixf.satd_x4 : h->pixf.sad_x4, sizeof(h->pixf.fpelcmp_x4) );
934 }
935
936 static void x264_set_aspect_ratio( x264_t *h, x264_param_t *param, int initial )
937 {
938     /* VUI */
939     if( param->vui.i_sar_width > 0 && param->vui.i_sar_height > 0 )
940     {
941         uint32_t i_w = param->vui.i_sar_width;
942         uint32_t i_h = param->vui.i_sar_height;
943         uint32_t old_w = h->param.vui.i_sar_width;
944         uint32_t old_h = h->param.vui.i_sar_height;
945
946         x264_reduce_fraction( &i_w, &i_h );
947
948         while( i_w > 65535 || i_h > 65535 )
949         {
950             i_w /= 2;
951             i_h /= 2;
952         }
953
954         x264_reduce_fraction( &i_w, &i_h );
955
956         if( i_w != old_w || i_h != old_h || initial )
957         {
958             h->param.vui.i_sar_width = 0;
959             h->param.vui.i_sar_height = 0;
960             if( i_w == 0 || i_h == 0 )
961                 x264_log( h, X264_LOG_WARNING, "cannot create valid sample aspect ratio\n" );
962             else
963             {
964                 x264_log( h, initial?X264_LOG_INFO:X264_LOG_DEBUG, "using SAR=%d/%d\n", i_w, i_h );
965                 h->param.vui.i_sar_width = i_w;
966                 h->param.vui.i_sar_height = i_h;
967             }
968             x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
969         }
970     }
971 }
972
973 /****************************************************************************
974  * x264_encoder_open:
975  ****************************************************************************/
976 x264_t *x264_encoder_open( x264_param_t *param )
977 {
978     x264_t *h;
979     char buf[1000], *p;
980     int qp, i_slicetype_length;
981
982     CHECKED_MALLOCZERO( h, sizeof(x264_t) );
983
984     /* Create a copy of param */
985     memcpy( &h->param, param, sizeof(x264_param_t) );
986
987     if( param->param_free )
988         param->param_free( param );
989
990     if( x264_threading_init() )
991     {
992         x264_log( h, X264_LOG_ERROR, "unable to initialize threading\n" );
993         goto fail;
994     }
995
996     if( x264_validate_parameters( h, 1 ) < 0 )
997         goto fail;
998
999     if( h->param.psz_cqm_file )
1000         if( x264_cqm_parse_file( h, h->param.psz_cqm_file ) < 0 )
1001             goto fail;
1002
1003     if( h->param.rc.psz_stat_out )
1004         h->param.rc.psz_stat_out = strdup( h->param.rc.psz_stat_out );
1005     if( h->param.rc.psz_stat_in )
1006         h->param.rc.psz_stat_in = strdup( h->param.rc.psz_stat_in );
1007
1008     x264_reduce_fraction( &h->param.i_fps_num, &h->param.i_fps_den );
1009     x264_reduce_fraction( &h->param.i_timebase_num, &h->param.i_timebase_den );
1010
1011     /* Init x264_t */
1012     h->i_frame = -1;
1013     h->i_frame_num = 0;
1014     h->i_idr_pic_id = 0;
1015
1016     if( (uint64_t)h->param.i_timebase_den * 2 > UINT32_MAX )
1017     {
1018         x264_log( h, X264_LOG_ERROR, "Effective timebase denominator %u exceeds H.264 maximum\n", h->param.i_timebase_den );
1019         goto fail;
1020     }
1021
1022     h->sps = &h->sps_array[0];
1023     x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
1024
1025     h->pps = &h->pps_array[0];
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();
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         h->thread[i]->sps = &h->thread[i]->sps_array[0];
1194         h->thread[i]->pps = &h->thread[i]->pps_array[0];
1195
1196         CHECKED_MALLOC( h->thread[i]->out.p_bitstream, h->out.i_bitstream );
1197         /* Start each thread with room for init_nal_count NAL units; it'll realloc later if needed. */
1198         CHECKED_MALLOC( h->thread[i]->out.nal, init_nal_count*sizeof(x264_nal_t) );
1199         h->thread[i]->out.i_nals_allocated = init_nal_count;
1200
1201         if( allocate_threadlocal_data && x264_macroblock_cache_allocate( h->thread[i] ) < 0 )
1202             goto fail;
1203     }
1204
1205     if( x264_lookahead_init( h, i_slicetype_length ) )
1206         goto fail;
1207
1208     for( int i = 0; i < h->param.i_threads; i++ )
1209         if( x264_macroblock_thread_allocate( h->thread[i], 0 ) < 0 )
1210             goto fail;
1211
1212     if( x264_ratecontrol_new( h ) < 0 )
1213         goto fail;
1214
1215     if( h->param.i_nal_hrd )
1216     {
1217         x264_log( h, X264_LOG_DEBUG, "HRD bitrate: %i bits/sec\n", h->sps->vui.hrd.i_bit_rate_unscaled );
1218         x264_log( h, X264_LOG_DEBUG, "CPB size: %i bits\n", h->sps->vui.hrd.i_cpb_size_unscaled );
1219     }
1220
1221     if( h->param.psz_dump_yuv )
1222     {
1223         /* create or truncate the reconstructed video file */
1224         FILE *f = fopen( h->param.psz_dump_yuv, "w" );
1225         if( !f )
1226         {
1227             x264_log( h, X264_LOG_ERROR, "dump_yuv: can't write to %s\n", h->param.psz_dump_yuv );
1228             goto fail;
1229         }
1230         else if( !x264_is_regular_file( f ) )
1231         {
1232             x264_log( h, X264_LOG_ERROR, "dump_yuv: incompatible with non-regular file %s\n", h->param.psz_dump_yuv );
1233             goto fail;
1234         }
1235         fclose( f );
1236     }
1237
1238     const char *profile = h->sps->i_profile_idc == PROFILE_BASELINE ? "Constrained Baseline" :
1239                           h->sps->i_profile_idc == PROFILE_MAIN ? "Main" :
1240                           h->sps->i_profile_idc == PROFILE_HIGH ? "High" :
1241                           h->sps->i_profile_idc == PROFILE_HIGH10 ? (h->sps->b_constraint_set3 == 1 ? "High 10 Intra" : "High 10") :
1242                           "High 4:4:4 Predictive";
1243     char level[4];
1244     snprintf( level, sizeof(level), "%d.%d", h->sps->i_level_idc/10, h->sps->i_level_idc%10 );
1245     if( h->sps->i_level_idc == 9 || ( h->sps->i_level_idc == 11 && h->sps->b_constraint_set3 ) )
1246         strcpy( level, "1b" );
1247
1248     if( h->sps->i_profile_idc < PROFILE_HIGH10 )
1249     {
1250         x264_log( h, X264_LOG_INFO, "profile %s, level %s\n",
1251             profile, level );
1252     }
1253     else
1254     {
1255         x264_log( h, X264_LOG_INFO, "profile %s, level %s, bit depth %d\n",
1256             profile, level, BIT_DEPTH );
1257     }
1258
1259     return h;
1260 fail:
1261     x264_free( h );
1262     return NULL;
1263 }
1264
1265 /****************************************************************************
1266  * x264_encoder_reconfig:
1267  ****************************************************************************/
1268 int x264_encoder_reconfig( x264_t *h, x264_param_t *param )
1269 {
1270     int rc_reconfig = 0;
1271     h = h->thread[h->thread[0]->i_thread_phase];
1272     x264_set_aspect_ratio( h, param, 0 );
1273 #define COPY(var) h->param.var = param->var
1274     COPY( i_frame_reference ); // but never uses more refs than initially specified
1275     COPY( i_bframe_bias );
1276     if( h->param.i_scenecut_threshold )
1277         COPY( i_scenecut_threshold ); // can't turn it on or off, only vary the threshold
1278     COPY( b_deblocking_filter );
1279     COPY( i_deblocking_filter_alphac0 );
1280     COPY( i_deblocking_filter_beta );
1281     COPY( i_frame_packing );
1282     COPY( analyse.inter );
1283     COPY( analyse.intra );
1284     COPY( analyse.i_direct_mv_pred );
1285     /* Scratch buffer prevents me_range from being increased for esa/tesa */
1286     if( h->param.analyse.i_me_method < X264_ME_ESA || param->analyse.i_me_range < h->param.analyse.i_me_range )
1287         COPY( analyse.i_me_range );
1288     COPY( analyse.i_noise_reduction );
1289     /* We can't switch out of subme=0 during encoding. */
1290     if( h->param.analyse.i_subpel_refine )
1291         COPY( analyse.i_subpel_refine );
1292     COPY( analyse.i_trellis );
1293     COPY( analyse.b_chroma_me );
1294     COPY( analyse.b_dct_decimate );
1295     COPY( analyse.b_fast_pskip );
1296     COPY( analyse.b_mixed_references );
1297     COPY( analyse.f_psy_rd );
1298     COPY( analyse.f_psy_trellis );
1299     COPY( crop_rect );
1300     // can only twiddle these if they were enabled to begin with:
1301     if( h->param.analyse.i_me_method >= X264_ME_ESA || param->analyse.i_me_method < X264_ME_ESA )
1302         COPY( analyse.i_me_method );
1303     if( h->param.analyse.i_me_method >= X264_ME_ESA && !h->frames.b_have_sub8x8_esa )
1304         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
1305     if( h->pps->b_transform_8x8_mode )
1306         COPY( analyse.b_transform_8x8 );
1307     if( h->frames.i_max_ref1 > 1 )
1308         COPY( i_bframe_pyramid );
1309     COPY( i_slice_max_size );
1310     COPY( i_slice_max_mbs );
1311     COPY( i_slice_count );
1312     COPY( b_tff );
1313
1314     /* VBV can't be turned on if it wasn't on to begin with */
1315     if( h->param.rc.i_vbv_max_bitrate > 0 && h->param.rc.i_vbv_buffer_size > 0 &&
1316           param->rc.i_vbv_max_bitrate > 0 &&   param->rc.i_vbv_buffer_size > 0 )
1317     {
1318         rc_reconfig |= h->param.rc.i_vbv_max_bitrate != param->rc.i_vbv_max_bitrate;
1319         rc_reconfig |= h->param.rc.i_vbv_buffer_size != param->rc.i_vbv_buffer_size;
1320         rc_reconfig |= h->param.rc.i_bitrate != param->rc.i_bitrate;
1321         COPY( rc.i_vbv_max_bitrate );
1322         COPY( rc.i_vbv_buffer_size );
1323         COPY( rc.i_bitrate );
1324     }
1325     rc_reconfig |= h->param.rc.f_rf_constant != param->rc.f_rf_constant;
1326     rc_reconfig |= h->param.rc.f_rf_constant_max != param->rc.f_rf_constant_max;
1327     COPY( rc.f_rf_constant );
1328     COPY( rc.f_rf_constant_max );
1329 #undef COPY
1330
1331     mbcmp_init( h );
1332
1333     int ret = x264_validate_parameters( h, 0 );
1334
1335     /* Supported reconfiguration options (1-pass only):
1336      * vbv-maxrate
1337      * vbv-bufsize
1338      * crf
1339      * bitrate (CBR only) */
1340     if( !ret && rc_reconfig )
1341         x264_ratecontrol_init_reconfigurable( h, 0 );
1342
1343     return ret;
1344 }
1345
1346 /****************************************************************************
1347  * x264_encoder_parameters:
1348  ****************************************************************************/
1349 void x264_encoder_parameters( x264_t *h, x264_param_t *param )
1350 {
1351     memcpy( param, &h->thread[h->i_thread_phase]->param, sizeof(x264_param_t) );
1352 }
1353
1354 /* internal usage */
1355 static void x264_nal_start( x264_t *h, int i_type, int i_ref_idc )
1356 {
1357     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1358
1359     nal->i_ref_idc        = i_ref_idc;
1360     nal->i_type           = i_type;
1361     nal->b_long_startcode = 1;
1362
1363     nal->i_payload= 0;
1364     nal->p_payload= &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8];
1365 }
1366
1367 /* if number of allocated nals is not enough, re-allocate a larger one. */
1368 static int x264_nal_check_buffer( x264_t *h )
1369 {
1370     if( h->out.i_nal >= h->out.i_nals_allocated )
1371     {
1372         x264_nal_t *new_out = x264_malloc( sizeof(x264_nal_t) * (h->out.i_nals_allocated*2) );
1373         if( !new_out )
1374             return -1;
1375         memcpy( new_out, h->out.nal, sizeof(x264_nal_t) * (h->out.i_nals_allocated) );
1376         x264_free( h->out.nal );
1377         h->out.nal = new_out;
1378         h->out.i_nals_allocated *= 2;
1379     }
1380     return 0;
1381 }
1382
1383 static int x264_nal_end( x264_t *h )
1384 {
1385     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1386     nal->i_payload = &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8] - nal->p_payload;
1387     if( h->param.nalu_process )
1388         h->param.nalu_process( h, nal );
1389     h->out.i_nal++;
1390
1391     return x264_nal_check_buffer( h );
1392 }
1393
1394 static int x264_encoder_encapsulate_nals( x264_t *h, int start )
1395 {
1396     int nal_size = 0, previous_nal_size = 0;
1397
1398     if( h->param.nalu_process )
1399     {
1400         for( int i = start; i < h->out.i_nal; i++ )
1401             nal_size += h->out.nal[i].i_payload;
1402         return nal_size;
1403     }
1404
1405     for( int i = 0; i < start; i++ )
1406         previous_nal_size += h->out.nal[i].i_payload;
1407
1408     for( int i = start; i < h->out.i_nal; i++ )
1409         nal_size += h->out.nal[i].i_payload;
1410
1411     /* Worst-case NAL unit escaping: reallocate the buffer if it's too small. */
1412     int necessary_size = nal_size * 3/2 + h->out.i_nal * 4;
1413     if( h->nal_buffer_size < necessary_size )
1414     {
1415         h->nal_buffer_size = necessary_size * 2;
1416         uint8_t *buf = x264_malloc( h->nal_buffer_size );
1417         if( !buf )
1418             return -1;
1419         if( previous_nal_size )
1420             memcpy( buf, h->nal_buffer, previous_nal_size );
1421         x264_free( h->nal_buffer );
1422         h->nal_buffer = buf;
1423     }
1424
1425     uint8_t *nal_buffer = h->nal_buffer + previous_nal_size;
1426
1427     for( int i = start; i < h->out.i_nal; i++ )
1428     {
1429         h->out.nal[i].b_long_startcode = !i || h->out.nal[i].i_type == NAL_SPS || h->out.nal[i].i_type == NAL_PPS;
1430         x264_nal_encode( h, nal_buffer, &h->out.nal[i] );
1431         nal_buffer += h->out.nal[i].i_payload;
1432     }
1433
1434     x264_emms();
1435
1436     return nal_buffer - (h->nal_buffer + previous_nal_size);
1437 }
1438
1439 /****************************************************************************
1440  * x264_encoder_headers:
1441  ****************************************************************************/
1442 int x264_encoder_headers( x264_t *h, x264_nal_t **pp_nal, int *pi_nal )
1443 {
1444     int frame_size = 0;
1445     /* init bitstream context */
1446     h->out.i_nal = 0;
1447     bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
1448
1449     /* Write SEI, SPS and PPS. */
1450
1451     /* generate sequence parameters */
1452     x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
1453     x264_sps_write( &h->out.bs, h->sps );
1454     if( x264_nal_end( h ) )
1455         return -1;
1456
1457     /* generate picture parameters */
1458     x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
1459     x264_pps_write( &h->out.bs, h->pps );
1460     if( x264_nal_end( h ) )
1461         return -1;
1462
1463     /* identify ourselves */
1464     x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
1465     if( x264_sei_version_write( h, &h->out.bs ) )
1466         return -1;
1467     if( x264_nal_end( h ) )
1468         return -1;
1469
1470     frame_size = x264_encoder_encapsulate_nals( h, 0 );
1471     if( frame_size < 0 )
1472         return -1;
1473
1474     /* now set output*/
1475     *pi_nal = h->out.i_nal;
1476     *pp_nal = &h->out.nal[0];
1477     h->out.i_nal = 0;
1478
1479     return frame_size;
1480 }
1481
1482 /* Check to see whether we have chosen a reference list ordering different
1483  * from the standard's default. */
1484 static inline void x264_reference_check_reorder( x264_t *h )
1485 {
1486     /* The reorder check doesn't check for missing frames, so just
1487      * force a reorder if one of the reference list is corrupt. */
1488     for( int i = 0; h->frames.reference[i]; i++ )
1489         if( h->frames.reference[i]->b_corrupt )
1490         {
1491             h->b_ref_reorder[0] = 1;
1492             return;
1493         }
1494     for( int list = 0; list <= (h->sh.i_type == SLICE_TYPE_B); list++ )
1495         for( int i = 0; i < h->i_ref[list] - 1; i++ )
1496         {
1497             int framenum_diff = h->fref[list][i+1]->i_frame_num - h->fref[list][i]->i_frame_num;
1498             int poc_diff = h->fref[list][i+1]->i_poc - h->fref[list][i]->i_poc;
1499             /* P and B-frames use different default orders. */
1500             if( h->sh.i_type == SLICE_TYPE_P ? framenum_diff > 0 : list == 1 ? poc_diff < 0 : poc_diff > 0 )
1501             {
1502                 h->b_ref_reorder[list] = 1;
1503                 return;
1504             }
1505         }
1506 }
1507
1508 /* return -1 on failure, else return the index of the new reference frame */
1509 int x264_weighted_reference_duplicate( x264_t *h, int i_ref, const x264_weight_t *w )
1510 {
1511     int i = h->i_ref[0];
1512     int j = 1;
1513     x264_frame_t *newframe;
1514     if( i <= 1 ) /* empty list, definitely can't duplicate frame */
1515         return -1;
1516
1517     //Duplication is only used in X264_WEIGHTP_SMART
1518     if( h->param.analyse.i_weighted_pred != X264_WEIGHTP_SMART )
1519         return -1;
1520
1521     /* Duplication is a hack to compensate for crappy rounding in motion compensation.
1522      * With high bit depth, it's not worth doing, so turn it off except in the case of
1523      * unweighted dupes. */
1524     if( BIT_DEPTH > 8 && w != weight_none )
1525         return -1;
1526
1527     newframe = x264_frame_pop_blank_unused( h );
1528     if( !newframe )
1529         return -1;
1530
1531     //FIXME: probably don't need to copy everything
1532     *newframe = *h->fref[0][i_ref];
1533     newframe->i_reference_count = 1;
1534     newframe->orig = h->fref[0][i_ref];
1535     newframe->b_duplicate = 1;
1536     memcpy( h->fenc->weight[j], w, sizeof(h->fenc->weight[i]) );
1537
1538     /* shift the frames to make space for the dupe. */
1539     h->b_ref_reorder[0] = 1;
1540     if( h->i_ref[0] < X264_REF_MAX )
1541         ++h->i_ref[0];
1542     h->fref[0][X264_REF_MAX-1] = NULL;
1543     x264_frame_unshift( &h->fref[0][j], newframe );
1544
1545     return j;
1546 }
1547
1548 static void x264_weighted_pred_init( x264_t *h )
1549 {
1550     /* for now no analysis and set all weights to nothing */
1551     for( int i_ref = 0; i_ref < h->i_ref[0]; i_ref++ )
1552         h->fenc->weighted[i_ref] = h->fref[0][i_ref]->filtered[0];
1553
1554     // FIXME: This only supports weighting of one reference frame
1555     // and duplicates of that frame.
1556     h->fenc->i_lines_weighted = 0;
1557
1558     for( int i_ref = 0; i_ref < (h->i_ref[0] << SLICE_MBAFF); i_ref++ )
1559         for( int i = 0; i < 3; i++ )
1560             h->sh.weight[i_ref][i].weightfn = NULL;
1561
1562
1563     if( h->sh.i_type != SLICE_TYPE_P || h->param.analyse.i_weighted_pred <= 0 )
1564         return;
1565
1566     int i_padv = PADV << PARAM_INTERLACED;
1567     int denom = -1;
1568     int weightplane[2] = { 0, 0 };
1569     int buffer_next = 0;
1570     for( int i = 0; i < 3; i++ )
1571     {
1572         for( int j = 0; j < h->i_ref[0]; j++ )
1573         {
1574             if( h->fenc->weight[j][i].weightfn )
1575             {
1576                 h->sh.weight[j][i] = h->fenc->weight[j][i];
1577                 // if weight is useless, don't write it to stream
1578                 if( h->sh.weight[j][i].i_scale == 1<<h->sh.weight[j][i].i_denom && h->sh.weight[j][i].i_offset == 0 )
1579                     h->sh.weight[j][i].weightfn = NULL;
1580                 else
1581                 {
1582                     if( !weightplane[!!i] )
1583                     {
1584                         weightplane[!!i] = 1;
1585                         h->sh.weight[0][!!i].i_denom = denom = h->sh.weight[j][i].i_denom;
1586                         assert( x264_clip3( denom, 0, 7 ) == denom );
1587                     }
1588
1589                     assert( h->sh.weight[j][i].i_denom == denom );
1590                     if( !i )
1591                     {
1592                         h->fenc->weighted[j] = h->mb.p_weight_buf[buffer_next++] + h->fenc->i_stride[0] * i_padv + PADH;
1593                         //scale full resolution frame
1594                         if( h->param.i_threads == 1 )
1595                         {
1596                             pixel *src = h->fref[0][j]->filtered[0] - h->fref[0][j]->i_stride[0]*i_padv - PADH;
1597                             pixel *dst = h->fenc->weighted[j] - h->fenc->i_stride[0]*i_padv - PADH;
1598                             int stride = h->fenc->i_stride[0];
1599                             int width = h->fenc->i_width[0] + PADH*2;
1600                             int height = h->fenc->i_lines[0] + i_padv*2;
1601                             x264_weight_scale_plane( h, dst, stride, src, stride, width, height, &h->sh.weight[j][0] );
1602                             h->fenc->i_lines_weighted = height;
1603                         }
1604                     }
1605                 }
1606             }
1607         }
1608     }
1609
1610     if( weightplane[1] )
1611         for( int i = 0; i < h->i_ref[0]; i++ )
1612         {
1613             if( h->sh.weight[i][1].weightfn && !h->sh.weight[i][2].weightfn )
1614             {
1615                 h->sh.weight[i][2].i_scale = 1 << h->sh.weight[0][1].i_denom;
1616                 h->sh.weight[i][2].i_offset = 0;
1617             }
1618             else if( h->sh.weight[i][2].weightfn && !h->sh.weight[i][1].weightfn )
1619             {
1620                 h->sh.weight[i][1].i_scale = 1 << h->sh.weight[0][1].i_denom;
1621                 h->sh.weight[i][1].i_offset = 0;
1622             }
1623         }
1624
1625     if( !weightplane[0] )
1626         h->sh.weight[0][0].i_denom = 0;
1627     if( !weightplane[1] )
1628         h->sh.weight[0][1].i_denom = 0;
1629     h->sh.weight[0][2].i_denom = h->sh.weight[0][1].i_denom;
1630 }
1631
1632 static inline int x264_reference_distance( x264_t *h, x264_frame_t *frame )
1633 {
1634     if( h->param.i_frame_packing == 5 )
1635         return abs((h->fenc->i_frame&~1) - (frame->i_frame&~1)) +
1636                   ((h->fenc->i_frame&1) != (frame->i_frame&1));
1637     else
1638         return abs(h->fenc->i_frame - frame->i_frame);
1639 }
1640
1641 static inline void x264_reference_build_list( x264_t *h, int i_poc )
1642 {
1643     int b_ok;
1644
1645     /* build ref list 0/1 */
1646     h->mb.pic.i_fref[0] = h->i_ref[0] = 0;
1647     h->mb.pic.i_fref[1] = h->i_ref[1] = 0;
1648     if( h->sh.i_type == SLICE_TYPE_I )
1649         return;
1650
1651     for( int i = 0; h->frames.reference[i]; i++ )
1652     {
1653         if( h->frames.reference[i]->b_corrupt )
1654             continue;
1655         if( h->frames.reference[i]->i_poc < i_poc )
1656             h->fref[0][h->i_ref[0]++] = h->frames.reference[i];
1657         else if( h->frames.reference[i]->i_poc > i_poc )
1658             h->fref[1][h->i_ref[1]++] = h->frames.reference[i];
1659     }
1660
1661     /* Order reference lists by distance from the current frame. */
1662     for( int list = 0; list < 2; list++ )
1663     {
1664         h->fref_nearest[list] = h->fref[list][0];
1665         do
1666         {
1667             b_ok = 1;
1668             for( int i = 0; i < h->i_ref[list] - 1; i++ )
1669             {
1670                 if( list ? h->fref[list][i+1]->i_poc < h->fref_nearest[list]->i_poc
1671                          : h->fref[list][i+1]->i_poc > h->fref_nearest[list]->i_poc )
1672                     h->fref_nearest[list] = h->fref[list][i+1];
1673                 if( x264_reference_distance( h, h->fref[list][i] ) > x264_reference_distance( h, h->fref[list][i+1] ) )
1674                 {
1675                     XCHG( x264_frame_t*, h->fref[list][i], h->fref[list][i+1] );
1676                     b_ok = 0;
1677                     break;
1678                 }
1679             }
1680         } while( !b_ok );
1681     }
1682
1683     if( h->sh.i_mmco_remove_from_end )
1684         for( int i = h->i_ref[0]-1; i >= h->i_ref[0] - h->sh.i_mmco_remove_from_end; i-- )
1685         {
1686             int diff = h->i_frame_num - h->fref[0][i]->i_frame_num;
1687             h->sh.mmco[h->sh.i_mmco_command_count].i_poc = h->fref[0][i]->i_poc;
1688             h->sh.mmco[h->sh.i_mmco_command_count++].i_difference_of_pic_nums = diff;
1689         }
1690
1691     x264_reference_check_reorder( h );
1692
1693     h->i_ref[1] = X264_MIN( h->i_ref[1], h->frames.i_max_ref1 );
1694     h->i_ref[0] = X264_MIN( h->i_ref[0], h->frames.i_max_ref0 );
1695     h->i_ref[0] = X264_MIN( h->i_ref[0], h->param.i_frame_reference ); // if reconfig() has lowered the limit
1696
1697     /* For Blu-ray compliance, don't reference frames outside of the minigop. */
1698     if( IS_X264_TYPE_B( h->fenc->i_type ) && h->param.b_bluray_compat )
1699         h->i_ref[0] = X264_MIN( h->i_ref[0], IS_X264_TYPE_B( h->fref[0][0]->i_type ) + 1 );
1700
1701     /* add duplicates */
1702     if( h->fenc->i_type == X264_TYPE_P )
1703     {
1704         int idx = -1;
1705         if( h->param.analyse.i_weighted_pred >= X264_WEIGHTP_SIMPLE )
1706         {
1707             x264_weight_t w[3];
1708             w[1].weightfn = w[2].weightfn = NULL;
1709             if( h->param.rc.b_stat_read )
1710                 x264_ratecontrol_set_weights( h, h->fenc );
1711
1712             if( !h->fenc->weight[0][0].weightfn )
1713             {
1714                 h->fenc->weight[0][0].i_denom = 0;
1715                 SET_WEIGHT( w[0], 1, 1, 0, -1 );
1716                 idx = x264_weighted_reference_duplicate( h, 0, w );
1717             }
1718             else
1719             {
1720                 if( h->fenc->weight[0][0].i_scale == 1<<h->fenc->weight[0][0].i_denom )
1721                 {
1722                     SET_WEIGHT( h->fenc->weight[0][0], 1, 1, 0, h->fenc->weight[0][0].i_offset );
1723                 }
1724                 x264_weighted_reference_duplicate( h, 0, weight_none );
1725                 if( h->fenc->weight[0][0].i_offset > -128 )
1726                 {
1727                     w[0] = h->fenc->weight[0][0];
1728                     w[0].i_offset--;
1729                     h->mc.weight_cache( h, &w[0] );
1730                     idx = x264_weighted_reference_duplicate( h, 0, w );
1731                 }
1732             }
1733         }
1734         h->mb.ref_blind_dupe = idx;
1735     }
1736
1737     assert( h->i_ref[0] + h->i_ref[1] <= X264_REF_MAX );
1738     h->mb.pic.i_fref[0] = h->i_ref[0];
1739     h->mb.pic.i_fref[1] = h->i_ref[1];
1740 }
1741
1742 static void x264_fdec_filter_row( x264_t *h, int mb_y, int b_inloop )
1743 {
1744     /* mb_y is the mb to be encoded next, not the mb to be filtered here */
1745     int b_hpel = h->fdec->b_kept_as_ref;
1746     int b_deblock = h->sh.i_disable_deblocking_filter_idc != 1;
1747     int b_end = mb_y == h->i_threadslice_end;
1748     int b_measure_quality = 1;
1749     int min_y = mb_y - (1 << SLICE_MBAFF);
1750     int b_start = min_y == h->i_threadslice_start;
1751     /* Even in interlaced mode, deblocking never modifies more than 4 pixels
1752      * above each MB, as bS=4 doesn't happen for the top of interlaced mbpairs. */
1753     int minpix_y = min_y*16 - 4 * !b_start;
1754     int maxpix_y = mb_y*16 - 4 * !b_end;
1755     b_deblock &= b_hpel || h->param.psz_dump_yuv;
1756     if( h->param.b_sliced_threads && b_start && min_y && !b_inloop )
1757     {
1758         b_deblock = 0;         /* We already deblocked on the inloop pass. */
1759         b_measure_quality = 0; /* We already measured quality on the inloop pass. */
1760     }
1761     if( mb_y & SLICE_MBAFF )
1762         return;
1763     if( min_y < h->i_threadslice_start )
1764         return;
1765
1766     if( b_deblock )
1767         for( int y = min_y; y < mb_y; y += (1 << SLICE_MBAFF) )
1768             x264_frame_deblock_row( h, y );
1769
1770     /* FIXME: Prediction requires different borders for interlaced/progressive mc,
1771      * but the actual image data is equivalent. For now, maintain this
1772      * consistency by copying deblocked pixels between planes. */
1773     if( PARAM_INTERLACED )
1774         for( int p = 0; p < 2; p++ )
1775             for( int i = minpix_y>>p; i < maxpix_y>>p; i++ )
1776                 memcpy( h->fdec->plane_fld[p] + i*h->fdec->i_stride[p],
1777                         h->fdec->plane[p]     + i*h->fdec->i_stride[p],
1778                         h->mb.i_mb_width*16*sizeof(pixel) );
1779
1780     if( b_hpel )
1781     {
1782         int end = mb_y == h->mb.i_mb_height;
1783         x264_frame_expand_border( h, h->fdec, min_y, end );
1784         if( h->param.analyse.i_subpel_refine )
1785         {
1786             x264_frame_filter( h, h->fdec, min_y, end );
1787             x264_frame_expand_border_filtered( h, h->fdec, min_y, end );
1788         }
1789     }
1790
1791     if( SLICE_MBAFF )
1792         for( int i = 0; i < 2; i++ )
1793         {
1794             XCHG( pixel *, h->intra_border_backup[0][i], h->intra_border_backup[3][i] );
1795             XCHG( pixel *, h->intra_border_backup[1][i], h->intra_border_backup[4][i] );
1796         }
1797
1798     if( h->i_thread_frames > 1 && h->fdec->b_kept_as_ref )
1799         x264_frame_cond_broadcast( h->fdec, mb_y*16 + (b_end ? 10000 : -(X264_THREAD_HEIGHT << SLICE_MBAFF)) );
1800
1801     if( b_measure_quality )
1802     {
1803         maxpix_y = X264_MIN( maxpix_y, h->param.i_height );
1804         if( h->param.analyse.b_psnr )
1805         {
1806             uint64_t ssd_y = x264_pixel_ssd_wxh( &h->pixf,
1807                 h->fdec->plane[0] + minpix_y * h->fdec->i_stride[0], h->fdec->i_stride[0],
1808                 h->fenc->plane[0] + minpix_y * h->fenc->i_stride[0], h->fenc->i_stride[0],
1809                 h->param.i_width, maxpix_y-minpix_y );
1810             uint64_t ssd_u, ssd_v;
1811             x264_pixel_ssd_nv12( &h->pixf,
1812                 h->fdec->plane[1] + (minpix_y>>1) * h->fdec->i_stride[1], h->fdec->i_stride[1],
1813                 h->fenc->plane[1] + (minpix_y>>1) * h->fenc->i_stride[1], h->fenc->i_stride[1],
1814                 h->param.i_width>>1, (maxpix_y-minpix_y)>>1, &ssd_u, &ssd_v );
1815             h->stat.frame.i_ssd[0] += ssd_y;
1816             h->stat.frame.i_ssd[1] += ssd_u;
1817             h->stat.frame.i_ssd[2] += ssd_v;
1818         }
1819
1820         if( h->param.analyse.b_ssim )
1821         {
1822             x264_emms();
1823             /* offset by 2 pixels to avoid alignment of ssim blocks with dct blocks,
1824              * and overlap by 4 */
1825             minpix_y += b_start ? 2 : -6;
1826             h->stat.frame.f_ssim +=
1827                 x264_pixel_ssim_wxh( &h->pixf,
1828                     h->fdec->plane[0] + 2+minpix_y*h->fdec->i_stride[0], h->fdec->i_stride[0],
1829                     h->fenc->plane[0] + 2+minpix_y*h->fenc->i_stride[0], h->fenc->i_stride[0],
1830                     h->param.i_width-2, maxpix_y-minpix_y, h->scratch_buffer );
1831         }
1832     }
1833 }
1834
1835 static inline int x264_reference_update( x264_t *h )
1836 {
1837     if( !h->fdec->b_kept_as_ref )
1838     {
1839         if( h->i_thread_frames > 1 )
1840         {
1841             x264_frame_push_unused( h, h->fdec );
1842             h->fdec = x264_frame_pop_unused( h, 1 );
1843             if( !h->fdec )
1844                 return -1;
1845         }
1846         return 0;
1847     }
1848
1849     /* apply mmco from previous frame. */
1850     for( int i = 0; i < h->sh.i_mmco_command_count; i++ )
1851         for( int j = 0; h->frames.reference[j]; j++ )
1852             if( h->frames.reference[j]->i_poc == h->sh.mmco[i].i_poc )
1853                 x264_frame_push_unused( h, x264_frame_shift( &h->frames.reference[j] ) );
1854
1855     /* move frame in the buffer */
1856     x264_frame_push( h->frames.reference, h->fdec );
1857     if( h->frames.reference[h->sps->i_num_ref_frames] )
1858         x264_frame_push_unused( h, x264_frame_shift( h->frames.reference ) );
1859     h->fdec = x264_frame_pop_unused( h, 1 );
1860     if( !h->fdec )
1861         return -1;
1862     return 0;
1863 }
1864
1865 static inline void x264_reference_reset( x264_t *h )
1866 {
1867     while( h->frames.reference[0] )
1868         x264_frame_push_unused( h, x264_frame_pop( h->frames.reference ) );
1869     h->fdec->i_poc =
1870     h->fenc->i_poc = 0;
1871 }
1872
1873 static inline void x264_reference_hierarchy_reset( x264_t *h )
1874 {
1875     int ref;
1876     int b_hasdelayframe = 0;
1877
1878     /* look for delay frames -- chain must only contain frames that are disposable */
1879     for( int i = 0; h->frames.current[i] && IS_DISPOSABLE( h->frames.current[i]->i_type ); i++ )
1880         b_hasdelayframe |= h->frames.current[i]->i_coded
1881                         != h->frames.current[i]->i_frame + h->sps->vui.i_num_reorder_frames;
1882
1883     /* This function must handle b-pyramid and clear frames for open-gop */
1884     if( h->param.i_bframe_pyramid != X264_B_PYRAMID_STRICT && !b_hasdelayframe && h->frames.i_poc_last_open_gop == -1 )
1885         return;
1886
1887     /* Remove last BREF. There will never be old BREFs in the
1888      * dpb during a BREF decode when pyramid == STRICT */
1889     for( ref = 0; h->frames.reference[ref]; ref++ )
1890     {
1891         if( ( h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT
1892             && h->frames.reference[ref]->i_type == X264_TYPE_BREF )
1893             || ( h->frames.reference[ref]->i_poc < h->frames.i_poc_last_open_gop
1894             && h->sh.i_type != SLICE_TYPE_B ) )
1895         {
1896             int diff = h->i_frame_num - h->frames.reference[ref]->i_frame_num;
1897             h->sh.mmco[h->sh.i_mmco_command_count].i_difference_of_pic_nums = diff;
1898             h->sh.mmco[h->sh.i_mmco_command_count++].i_poc = h->frames.reference[ref]->i_poc;
1899             x264_frame_push_unused( h, x264_frame_shift( &h->frames.reference[ref] ) );
1900             h->b_ref_reorder[0] = 1;
1901             ref--;
1902         }
1903     }
1904
1905     /* Prepare room in the dpb for the delayed display time of the later b-frame's */
1906     if( h->param.i_bframe_pyramid )
1907         h->sh.i_mmco_remove_from_end = X264_MAX( ref + 2 - h->frames.i_max_dpb, 0 );
1908 }
1909
1910 static inline void x264_slice_init( x264_t *h, int i_nal_type, int i_global_qp )
1911 {
1912     /* ------------------------ Create slice header  ----------------------- */
1913     if( i_nal_type == NAL_SLICE_IDR )
1914     {
1915         x264_slice_header_init( h, &h->sh, h->sps, h->pps, h->i_idr_pic_id, h->i_frame_num, i_global_qp );
1916
1917         /* alternate id */
1918         h->i_idr_pic_id ^= 1;
1919     }
1920     else
1921     {
1922         x264_slice_header_init( h, &h->sh, h->sps, h->pps, -1, h->i_frame_num, i_global_qp );
1923
1924         h->sh.i_num_ref_idx_l0_active = h->i_ref[0] <= 0 ? 1 : h->i_ref[0];
1925         h->sh.i_num_ref_idx_l1_active = h->i_ref[1] <= 0 ? 1 : h->i_ref[1];
1926         if( h->sh.i_num_ref_idx_l0_active != h->pps->i_num_ref_idx_l0_default_active ||
1927             (h->sh.i_type == SLICE_TYPE_B && h->sh.i_num_ref_idx_l1_active != h->pps->i_num_ref_idx_l1_default_active) )
1928         {
1929             h->sh.b_num_ref_idx_override = 1;
1930         }
1931     }
1932
1933     if( h->fenc->i_type == X264_TYPE_BREF && h->param.b_bluray_compat && h->sh.i_mmco_command_count )
1934     {
1935         h->b_sh_backup = 1;
1936         h->sh_backup = h->sh;
1937     }
1938
1939     h->fdec->i_frame_num = h->sh.i_frame_num;
1940
1941     if( h->sps->i_poc_type == 0 )
1942     {
1943         h->sh.i_poc = h->fdec->i_poc;
1944         if( PARAM_INTERLACED )
1945         {
1946             h->sh.i_delta_poc_bottom = h->param.b_tff ? 1 : -1;
1947             h->sh.i_poc += h->sh.i_delta_poc_bottom == -1;
1948         }
1949         else
1950             h->sh.i_delta_poc_bottom = 0;
1951         h->fdec->i_delta_poc[0] = h->sh.i_delta_poc_bottom == -1;
1952         h->fdec->i_delta_poc[1] = h->sh.i_delta_poc_bottom ==  1;
1953     }
1954     else if( h->sps->i_poc_type == 1 )
1955     {
1956         /* FIXME TODO FIXME */
1957     }
1958     else
1959     {
1960         /* Nothing to do ? */
1961     }
1962
1963     x264_macroblock_slice_init( h );
1964 }
1965
1966 static int x264_slice_write( x264_t *h )
1967 {
1968     int i_skip;
1969     int mb_xy, i_mb_x, i_mb_y;
1970     int i_skip_bak = 0; /* Shut up GCC. */
1971     bs_t UNINIT(bs_bak);
1972     x264_cabac_t cabac_bak;
1973     uint8_t cabac_prevbyte_bak = 0; /* Shut up GCC. */
1974     int mv_bits_bak = 0;
1975     int tex_bits_bak = 0;
1976     /* NALUs other than the first use a 3-byte startcode.
1977      * Add one extra byte for the rbsp, and one more for the final CABAC putbyte.
1978      * Then add an extra 5 bytes just in case, to account for random NAL escapes and
1979      * other inaccuracies. */
1980     int overhead_guess = (NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal)) + 1 + h->param.b_cabac + 5;
1981     int slice_max_size = h->param.i_slice_max_size > 0 ? (h->param.i_slice_max_size-overhead_guess)*8 : 0;
1982     int back_up_bitstream = slice_max_size || (!h->param.b_cabac && h->sps->i_profile_idc < PROFILE_HIGH);
1983     int starting_bits = bs_pos(&h->out.bs);
1984     int b_deblock = h->sh.i_disable_deblocking_filter_idc != 1;
1985     int b_hpel = h->fdec->b_kept_as_ref;
1986     uint8_t *last_emu_check;
1987     b_deblock &= b_hpel || h->param.psz_dump_yuv;
1988     bs_realign( &h->out.bs );
1989
1990     /* Slice */
1991     x264_nal_start( h, h->i_nal_type, h->i_nal_ref_idc );
1992     h->out.nal[h->out.i_nal].i_first_mb = h->sh.i_first_mb;
1993
1994     /* Slice header */
1995     x264_macroblock_thread_init( h );
1996
1997     /* If this isn't the first slice in the threadslice, set the slice QP
1998      * equal to the last QP in the previous slice for more accurate
1999      * CABAC initialization. */
2000     if( h->sh.i_first_mb != h->i_threadslice_start * h->mb.i_mb_width )
2001     {
2002         h->sh.i_qp = h->mb.i_last_qp;
2003         h->sh.i_qp_delta = h->sh.i_qp - h->pps->i_pic_init_qp;
2004     }
2005
2006     x264_slice_header_write( &h->out.bs, &h->sh, h->i_nal_ref_idc );
2007     if( h->param.b_cabac )
2008     {
2009         /* alignment needed */
2010         bs_align_1( &h->out.bs );
2011
2012         /* init cabac */
2013         x264_cabac_context_init( &h->cabac, h->sh.i_type, x264_clip3( h->sh.i_qp-QP_BD_OFFSET, 0, 51 ), h->sh.i_cabac_init_idc );
2014         x264_cabac_encode_init ( &h->cabac, h->out.bs.p, h->out.bs.p_end );
2015         last_emu_check = h->cabac.p;
2016     }
2017     else
2018         last_emu_check = h->out.bs.p;
2019     h->mb.i_last_qp = h->sh.i_qp;
2020     h->mb.i_last_dqp = 0;
2021     h->mb.field_decoding_flag = 0;
2022
2023     i_mb_y = h->sh.i_first_mb / h->mb.i_mb_width;
2024     i_mb_x = h->sh.i_first_mb % h->mb.i_mb_width;
2025     i_skip = 0;
2026
2027     while( 1 )
2028     {
2029         mb_xy = i_mb_x + i_mb_y * h->mb.i_mb_width;
2030         int mb_spos = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
2031
2032         if( !(i_mb_y & SLICE_MBAFF) )
2033         {
2034             if( x264_bitstream_check_buffer( h ) )
2035                 return -1;
2036
2037             if( back_up_bitstream )
2038             {
2039                 mv_bits_bak = h->stat.frame.i_mv_bits;
2040                 tex_bits_bak = h->stat.frame.i_tex_bits;
2041                 /* We don't need the contexts because flushing the CABAC encoder has no context
2042                  * dependency and macroblocks are only re-encoded in the case where a slice is
2043                  * ended (and thus the content of all contexts are thrown away). */
2044                 if( h->param.b_cabac )
2045                 {
2046                     memcpy( &cabac_bak, &h->cabac, offsetof(x264_cabac_t, f8_bits_encoded) );
2047                     /* x264's CABAC writer modifies the previous byte during carry, so it has to be
2048                      * backed up. */
2049                     cabac_prevbyte_bak = h->cabac.p[-1];
2050                 }
2051                 else
2052                 {
2053                     bs_bak = h->out.bs;
2054                     i_skip_bak = i_skip;
2055                 }
2056             }
2057         }
2058
2059         if( i_mb_x == 0 && !h->mb.b_reencode_mb )
2060             x264_fdec_filter_row( h, i_mb_y, 1 );
2061
2062         if( PARAM_INTERLACED )
2063         {
2064             if( h->mb.b_adaptive_mbaff )
2065             {
2066                 if( !(i_mb_y&1) )
2067                 {
2068                     /* FIXME: VSAD is fast but fairly poor at choosing the best interlace type. */
2069                     h->mb.b_interlaced = x264_field_vsad( h, i_mb_x, i_mb_y );
2070                     memcpy( &h->zigzagf, MB_INTERLACED ? &h->zigzagf_interlaced : &h->zigzagf_progressive, sizeof(h->zigzagf) );
2071                     if( !MB_INTERLACED && (i_mb_y+2) == h->mb.i_mb_height )
2072                         x264_expand_border_mbpair( h, i_mb_x, i_mb_y );
2073                 }
2074             }
2075             h->mb.field[mb_xy] = MB_INTERLACED;
2076         }
2077
2078         /* load cache */
2079         if( SLICE_MBAFF )
2080             x264_macroblock_cache_load_interlaced( h, i_mb_x, i_mb_y );
2081         else
2082             x264_macroblock_cache_load_progressive( h, i_mb_x, i_mb_y );
2083
2084         x264_macroblock_analyse( h );
2085
2086         /* encode this macroblock -> be careful it can change the mb type to P_SKIP if needed */
2087 reencode:
2088         x264_macroblock_encode( h );
2089
2090         if( h->param.b_cabac )
2091         {
2092             if( mb_xy > h->sh.i_first_mb && !(SLICE_MBAFF && (i_mb_y&1)) )
2093                 x264_cabac_encode_terminal( &h->cabac );
2094
2095             if( IS_SKIP( h->mb.i_type ) )
2096                 x264_cabac_mb_skip( h, 1 );
2097             else
2098             {
2099                 if( h->sh.i_type != SLICE_TYPE_I )
2100                     x264_cabac_mb_skip( h, 0 );
2101                 x264_macroblock_write_cabac( h, &h->cabac );
2102             }
2103         }
2104         else
2105         {
2106             if( IS_SKIP( h->mb.i_type ) )
2107                 i_skip++;
2108             else
2109             {
2110                 if( h->sh.i_type != SLICE_TYPE_I )
2111                 {
2112                     bs_write_ue( &h->out.bs, i_skip );  /* skip run */
2113                     i_skip = 0;
2114                 }
2115                 x264_macroblock_write_cavlc( h );
2116                 /* If there was a CAVLC level code overflow, try again at a higher QP. */
2117                 if( h->mb.b_overflow )
2118                 {
2119                     h->mb.i_chroma_qp = h->chroma_qp_table[++h->mb.i_qp];
2120                     h->mb.i_skip_intra = 0;
2121                     h->mb.b_skip_mc = 0;
2122                     h->mb.b_overflow = 0;
2123                     h->out.bs = bs_bak;
2124                     i_skip = i_skip_bak;
2125                     h->stat.frame.i_mv_bits = mv_bits_bak;
2126                     h->stat.frame.i_tex_bits = tex_bits_bak;
2127                     goto reencode;
2128                 }
2129             }
2130         }
2131
2132         int total_bits = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
2133         int mb_size = total_bits - mb_spos;
2134
2135         if( slice_max_size )
2136         {
2137             /* Count the skip run, just in case. */
2138             if( !h->param.b_cabac )
2139                 total_bits += bs_size_ue_big( i_skip );
2140             /* Check for escape bytes. */
2141             uint8_t *end = h->param.b_cabac ? h->cabac.p : h->out.bs.p;
2142             for( ; last_emu_check < end - 2; last_emu_check++ )
2143                 if( last_emu_check[0] == 0 && last_emu_check[1] == 0 && last_emu_check[2] <= 3 )
2144                 {
2145                     slice_max_size -= 8;
2146                     last_emu_check++;
2147                 }
2148             /* We'll just re-encode this last macroblock if we go over the max slice size. */
2149             if( total_bits - starting_bits > slice_max_size && !h->mb.b_reencode_mb )
2150             {
2151                 if( mb_xy != h->sh.i_first_mb )
2152                 {
2153                     h->stat.frame.i_mv_bits = mv_bits_bak;
2154                     h->stat.frame.i_tex_bits = tex_bits_bak;
2155                     if( h->param.b_cabac )
2156                     {
2157                         memcpy( &h->cabac, &cabac_bak, offsetof(x264_cabac_t, f8_bits_encoded) );
2158                         h->cabac.p[-1] = cabac_prevbyte_bak;
2159                     }
2160                     else
2161                     {
2162                         h->out.bs = bs_bak;
2163                         i_skip = i_skip_bak;
2164                     }
2165                     h->mb.b_reencode_mb = 1;
2166                     if( SLICE_MBAFF )
2167                     {
2168                         // set to bottom of previous mbpair
2169                         if( i_mb_x )
2170                             h->sh.i_last_mb = mb_xy-1+h->mb.i_mb_stride*(!(i_mb_y&1));
2171                         else
2172                             h->sh.i_last_mb = (i_mb_y-2+!(i_mb_y&1))*h->mb.i_mb_stride + h->mb.i_mb_width - 1;
2173                     }
2174                     else
2175                         h->sh.i_last_mb = mb_xy-1;
2176                     break;
2177                 }
2178                 else
2179                 {
2180                     h->sh.i_last_mb = mb_xy;
2181                     h->mb.b_reencode_mb = 0;
2182                 }
2183             }
2184             else
2185                 h->mb.b_reencode_mb = 0;
2186         }
2187
2188 #if HAVE_VISUALIZE
2189         if( h->param.b_visualize )
2190             x264_visualize_mb( h );
2191 #endif
2192
2193         /* save cache */
2194         x264_macroblock_cache_save( h );
2195
2196         /* accumulate mb stats */
2197         h->stat.frame.i_mb_count[h->mb.i_type]++;
2198
2199         int b_intra = IS_INTRA( h->mb.i_type );
2200         int b_skip = IS_SKIP( h->mb.i_type );
2201         if( h->param.i_log_level >= X264_LOG_INFO || h->param.rc.b_stat_write )
2202         {
2203             if( !b_intra && !b_skip && !IS_DIRECT( h->mb.i_type ) )
2204             {
2205                 if( h->mb.i_partition != D_8x8 )
2206                         h->stat.frame.i_mb_partition[h->mb.i_partition] += 4;
2207                     else
2208                         for( int i = 0; i < 4; i++ )
2209                             h->stat.frame.i_mb_partition[h->mb.i_sub_partition[i]] ++;
2210                 if( h->param.i_frame_reference > 1 )
2211                     for( int i_list = 0; i_list <= (h->sh.i_type == SLICE_TYPE_B); i_list++ )
2212                         for( int i = 0; i < 4; i++ )
2213                         {
2214                             int i_ref = h->mb.cache.ref[i_list][ x264_scan8[4*i] ];
2215                             if( i_ref >= 0 )
2216                                 h->stat.frame.i_mb_count_ref[i_list][i_ref] ++;
2217                         }
2218             }
2219         }
2220
2221         if( h->param.i_log_level >= X264_LOG_INFO )
2222         {
2223             if( h->mb.i_cbp_luma | h->mb.i_cbp_chroma )
2224             {
2225                 int cbpsum = (h->mb.i_cbp_luma&1) + ((h->mb.i_cbp_luma>>1)&1)
2226                            + ((h->mb.i_cbp_luma>>2)&1) + (h->mb.i_cbp_luma>>3);
2227                 h->stat.frame.i_mb_cbp[!b_intra + 0] += cbpsum;
2228                 h->stat.frame.i_mb_cbp[!b_intra + 2] += !!h->mb.i_cbp_chroma;
2229                 h->stat.frame.i_mb_cbp[!b_intra + 4] += h->mb.i_cbp_chroma >> 1;
2230             }
2231             if( h->mb.i_cbp_luma && !b_intra )
2232             {
2233                 h->stat.frame.i_mb_count_8x8dct[0] ++;
2234                 h->stat.frame.i_mb_count_8x8dct[1] += h->mb.b_transform_8x8;
2235             }
2236             if( b_intra && h->mb.i_type != I_PCM )
2237             {
2238                 if( h->mb.i_type == I_16x16 )
2239                     h->stat.frame.i_mb_pred_mode[0][h->mb.i_intra16x16_pred_mode]++;
2240                 else if( h->mb.i_type == I_8x8 )
2241                     for( int i = 0; i < 16; i += 4 )
2242                         h->stat.frame.i_mb_pred_mode[1][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
2243                 else //if( h->mb.i_type == I_4x4 )
2244                     for( int i = 0; i < 16; i++ )
2245                         h->stat.frame.i_mb_pred_mode[2][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
2246                 h->stat.frame.i_mb_pred_mode[3][x264_mb_pred_mode8x8c_fix[h->mb.i_chroma_pred_mode]]++;
2247             }
2248             h->stat.frame.i_mb_field[b_intra?0:b_skip?2:1] += MB_INTERLACED;
2249         }
2250
2251         /* calculate deblock strength values (actual deblocking is done per-row along with hpel) */
2252         if( b_deblock )
2253             x264_macroblock_deblock_strength( h );
2254
2255         x264_ratecontrol_mb( h, mb_size );
2256
2257         if( mb_xy == h->sh.i_last_mb )
2258             break;
2259
2260         if( SLICE_MBAFF )
2261         {
2262             i_mb_x += i_mb_y & 1;
2263             i_mb_y ^= i_mb_x < h->mb.i_mb_width;
2264         }
2265         else
2266             i_mb_x++;
2267         if( i_mb_x == h->mb.i_mb_width )
2268         {
2269             i_mb_y++;
2270             i_mb_x = 0;
2271         }
2272     }
2273     h->out.nal[h->out.i_nal].i_last_mb = h->sh.i_last_mb;
2274
2275     if( h->param.b_cabac )
2276     {
2277         x264_cabac_encode_flush( h, &h->cabac );
2278         h->out.bs.p = h->cabac.p;
2279     }
2280     else
2281     {
2282         if( i_skip > 0 )
2283             bs_write_ue( &h->out.bs, i_skip );  /* last skip run */
2284         /* rbsp_slice_trailing_bits */
2285         bs_rbsp_trailing( &h->out.bs );
2286         bs_flush( &h->out.bs );
2287     }
2288     if( x264_nal_end( h ) )
2289         return -1;
2290
2291     if( h->sh.i_last_mb == (h->i_threadslice_end * h->mb.i_mb_width - 1) )
2292     {
2293         h->stat.frame.i_misc_bits = bs_pos( &h->out.bs )
2294                                   + (h->out.i_nal*NALU_OVERHEAD * 8)
2295                                   - h->stat.frame.i_tex_bits
2296                                   - h->stat.frame.i_mv_bits;
2297         x264_fdec_filter_row( h, h->i_threadslice_end, 1 );
2298     }
2299
2300     return 0;
2301 }
2302
2303 static void x264_thread_sync_context( x264_t *dst, x264_t *src )
2304 {
2305     if( dst == src )
2306         return;
2307
2308     // reference counting
2309     for( x264_frame_t **f = src->frames.reference; *f; f++ )
2310         (*f)->i_reference_count++;
2311     for( x264_frame_t **f = dst->frames.reference; *f; f++ )
2312         x264_frame_push_unused( src, *f );
2313     src->fdec->i_reference_count++;
2314     x264_frame_push_unused( src, dst->fdec );
2315
2316     // copy everything except the per-thread pointers and the constants.
2317     memcpy( &dst->i_frame, &src->i_frame, offsetof(x264_t, mb.type) - offsetof(x264_t, i_frame) );
2318     dst->param = src->param;
2319     dst->stat = src->stat;
2320     dst->pixf = src->pixf;
2321 }
2322
2323 static void x264_thread_sync_stat( x264_t *dst, x264_t *src )
2324 {
2325     if( dst == src )
2326         return;
2327     memcpy( &dst->stat.i_frame_count, &src->stat.i_frame_count, sizeof(dst->stat) - sizeof(dst->stat.frame) );
2328 }
2329
2330 static void *x264_slices_write( x264_t *h )
2331 {
2332     int i_slice_num = 0;
2333     int last_thread_mb = h->sh.i_last_mb;
2334
2335 #if HAVE_VISUALIZE
2336     if( h->param.b_visualize )
2337         if( x264_visualize_init( h ) )
2338             return (void *)-1;
2339 #endif
2340
2341     /* init stats */
2342     memset( &h->stat.frame, 0, sizeof(h->stat.frame) );
2343     h->mb.b_reencode_mb = 0;
2344     while( h->sh.i_first_mb + SLICE_MBAFF*h->mb.i_mb_stride <= last_thread_mb )
2345     {
2346         h->sh.i_last_mb = last_thread_mb;
2347         if( h->param.i_slice_max_mbs )
2348         {
2349             if( SLICE_MBAFF )
2350             {
2351                 // convert first to mbaff form, add slice-max-mbs, then convert back to normal form
2352                 int last_mbaff = 2*(h->sh.i_first_mb % h->mb.i_mb_width)
2353                     + h->mb.i_mb_width*(h->sh.i_first_mb / h->mb.i_mb_width)
2354                     + h->param.i_slice_max_mbs - 1;
2355                 int last_x = (last_mbaff % (2*h->mb.i_mb_width))/2;
2356                 int last_y = (last_mbaff / (2*h->mb.i_mb_width))*2 + 1;
2357                 h->sh.i_last_mb = last_x + h->mb.i_mb_stride*last_y;
2358             }
2359             else
2360                 h->sh.i_last_mb = h->sh.i_first_mb + h->param.i_slice_max_mbs - 1;
2361         }
2362         else if( h->param.i_slice_count && !h->param.b_sliced_threads )
2363         {
2364             int height = h->mb.i_mb_height >> PARAM_INTERLACED;
2365             int width = h->mb.i_mb_width << PARAM_INTERLACED;
2366             i_slice_num++;
2367             h->sh.i_last_mb = (height * i_slice_num + h->param.i_slice_count/2) / h->param.i_slice_count * width - 1;
2368         }
2369         h->sh.i_last_mb = X264_MIN( h->sh.i_last_mb, last_thread_mb );
2370         if( x264_stack_align( x264_slice_write, h ) )
2371             return (void *)-1;
2372         h->sh.i_first_mb = h->sh.i_last_mb + 1;
2373         // if i_first_mb is not the last mb in a row then go to the next mb in MBAFF order
2374         if( SLICE_MBAFF && h->sh.i_first_mb % h->mb.i_mb_width )
2375             h->sh.i_first_mb -= h->mb.i_mb_stride;
2376     }
2377
2378 #if HAVE_VISUALIZE
2379     if( h->param.b_visualize )
2380     {
2381         x264_visualize_show( h );
2382         x264_visualize_close( h );
2383     }
2384 #endif
2385
2386     return (void *)0;
2387 }
2388
2389 static int x264_threaded_slices_write( x264_t *h )
2390 {
2391     /* set first/last mb and sync contexts */
2392     for( int i = 0; i < h->param.i_threads; i++ )
2393     {
2394         x264_t *t = h->thread[i];
2395         if( i )
2396         {
2397             t->param = h->param;
2398             memcpy( &t->i_frame, &h->i_frame, offsetof(x264_t, rc) - offsetof(x264_t, i_frame) );
2399         }
2400         int height = h->mb.i_mb_height >> PARAM_INTERLACED;
2401         t->i_threadslice_start = ((height *  i    + h->param.i_slice_count/2) / h->param.i_threads) << PARAM_INTERLACED;
2402         t->i_threadslice_end   = ((height * (i+1) + h->param.i_slice_count/2) / h->param.i_threads) << PARAM_INTERLACED;
2403         t->sh.i_first_mb = t->i_threadslice_start * h->mb.i_mb_width;
2404         t->sh.i_last_mb  =   t->i_threadslice_end * h->mb.i_mb_width - 1;
2405     }
2406
2407     x264_stack_align( x264_analyse_weight_frame, h, h->mb.i_mb_height*16 + 16 );
2408
2409     x264_threads_distribute_ratecontrol( h );
2410
2411     /* dispatch */
2412     for( int i = 0; i < h->param.i_threads; i++ )
2413     {
2414         x264_threadpool_run( h->threadpool, (void*)x264_slices_write, h->thread[i] );
2415         h->thread[i]->b_thread_active = 1;
2416     }
2417     for( int i = 0; i < h->param.i_threads; i++ )
2418     {
2419         h->thread[i]->b_thread_active = 0;
2420         if( (intptr_t)x264_threadpool_wait( h->threadpool, h->thread[i] ) )
2421             return -1;
2422     }
2423
2424     /* Go back and fix up the hpel on the borders between slices. */
2425     for( int i = 1; i < h->param.i_threads; i++ )
2426     {
2427         x264_fdec_filter_row( h->thread[i], h->thread[i]->i_threadslice_start + 1, 0 );
2428         if( SLICE_MBAFF )
2429             x264_fdec_filter_row( h->thread[i], h->thread[i]->i_threadslice_start + 2, 0 );
2430     }
2431
2432     x264_threads_merge_ratecontrol( h );
2433
2434     for( int i = 1; i < h->param.i_threads; i++ )
2435     {
2436         x264_t *t = h->thread[i];
2437         for( int j = 0; j < t->out.i_nal; j++ )
2438         {
2439             h->out.nal[h->out.i_nal] = t->out.nal[j];
2440             h->out.i_nal++;
2441             x264_nal_check_buffer( h );
2442         }
2443         /* All entries in stat.frame are ints except for ssd/ssim. */
2444         for( int j = 0; j < (offsetof(x264_t,stat.frame.i_ssd) - offsetof(x264_t,stat.frame.i_mv_bits)) / sizeof(int); j++ )
2445             ((int*)&h->stat.frame)[j] += ((int*)&t->stat.frame)[j];
2446         for( int j = 0; j < 3; j++ )
2447             h->stat.frame.i_ssd[j] += t->stat.frame.i_ssd[j];
2448         h->stat.frame.f_ssim += t->stat.frame.f_ssim;
2449     }
2450
2451     return 0;
2452 }
2453
2454 void x264_encoder_intra_refresh( x264_t *h )
2455 {
2456     h = h->thread[h->i_thread_phase];
2457     h->b_queued_intra_refresh = 1;
2458 }
2459
2460 int x264_encoder_invalidate_reference( x264_t *h, int64_t pts )
2461 {
2462     if( h->param.i_bframe )
2463     {
2464         x264_log( h, X264_LOG_ERROR, "x264_encoder_invalidate_reference is not supported with B-frames enabled\n" );
2465         return -1;
2466     }
2467     if( h->param.b_intra_refresh )
2468     {
2469         x264_log( h, X264_LOG_ERROR, "x264_encoder_invalidate_reference is not supported with intra refresh enabled\n" );
2470         return -1;
2471     }
2472     h = h->thread[h->i_thread_phase];
2473     if( pts >= h->i_last_idr_pts )
2474     {
2475         for( int i = 0; h->frames.reference[i]; i++ )
2476             if( pts <= h->frames.reference[i]->i_pts )
2477                 h->frames.reference[i]->b_corrupt = 1;
2478         if( pts <= h->fdec->i_pts )
2479             h->fdec->b_corrupt = 1;
2480     }
2481     return 0;
2482 }
2483
2484 /****************************************************************************
2485  * x264_encoder_encode:
2486  *  XXX: i_poc   : is the poc of the current given picture
2487  *       i_frame : is the number of the frame being coded
2488  *  ex:  type frame poc
2489  *       I      0   2*0
2490  *       P      1   2*3
2491  *       B      2   2*1
2492  *       B      3   2*2
2493  *       P      4   2*6
2494  *       B      5   2*4
2495  *       B      6   2*5
2496  ****************************************************************************/
2497 int     x264_encoder_encode( x264_t *h,
2498                              x264_nal_t **pp_nal, int *pi_nal,
2499                              x264_picture_t *pic_in,
2500                              x264_picture_t *pic_out )
2501 {
2502     x264_t *thread_current, *thread_prev, *thread_oldest;
2503     int i_nal_type, i_nal_ref_idc, i_global_qp;
2504     int overhead = NALU_OVERHEAD;
2505
2506     if( h->i_thread_frames > 1 )
2507     {
2508         thread_prev    = h->thread[ h->i_thread_phase ];
2509         h->i_thread_phase = (h->i_thread_phase + 1) % h->i_thread_frames;
2510         thread_current = h->thread[ h->i_thread_phase ];
2511         thread_oldest  = h->thread[ (h->i_thread_phase + 1) % h->i_thread_frames ];
2512         x264_thread_sync_context( thread_current, thread_prev );
2513         x264_thread_sync_ratecontrol( thread_current, thread_prev, thread_oldest );
2514         h = thread_current;
2515     }
2516     else
2517     {
2518         thread_current =
2519         thread_oldest  = h;
2520     }
2521 #if HAVE_MMX
2522     if( h->param.cpu&X264_CPU_SSE_MISALIGN )
2523         x264_cpu_mask_misalign_sse();
2524 #endif
2525
2526     // ok to call this before encoding any frames, since the initial values of fdec have b_kept_as_ref=0
2527     if( x264_reference_update( h ) )
2528         return -1;
2529     h->fdec->i_lines_completed = -1;
2530
2531     /* no data out */
2532     *pi_nal = 0;
2533     *pp_nal = NULL;
2534
2535     /* ------------------- Setup new frame from picture -------------------- */
2536     if( pic_in != NULL )
2537     {
2538         /* 1: Copy the picture to a frame and move it to a buffer */
2539         x264_frame_t *fenc = x264_frame_pop_unused( h, 0 );
2540         if( !fenc )
2541             return -1;
2542
2543         if( x264_frame_copy_picture( h, fenc, pic_in ) < 0 )
2544             return -1;
2545
2546         if( h->param.i_width != 16 * h->mb.i_mb_width ||
2547             h->param.i_height != 16 * h->mb.i_mb_height )
2548             x264_frame_expand_border_mod16( h, fenc );
2549
2550         fenc->i_frame = h->frames.i_input++;
2551
2552         if( fenc->i_frame == 0 )
2553             h->frames.i_first_pts = fenc->i_pts;
2554         if( h->frames.i_bframe_delay && fenc->i_frame == h->frames.i_bframe_delay )
2555             h->frames.i_bframe_delay_time = fenc->i_pts - h->frames.i_first_pts;
2556
2557         if( h->param.b_vfr_input && fenc->i_pts <= h->frames.i_largest_pts )
2558             x264_log( h, X264_LOG_WARNING, "non-strictly-monotonic PTS\n" );
2559
2560         h->frames.i_second_largest_pts = h->frames.i_largest_pts;
2561         h->frames.i_largest_pts = fenc->i_pts;
2562
2563         if( (fenc->i_pic_struct < PIC_STRUCT_AUTO) || (fenc->i_pic_struct > PIC_STRUCT_TRIPLE) )
2564             fenc->i_pic_struct = PIC_STRUCT_AUTO;
2565
2566         if( fenc->i_pic_struct == PIC_STRUCT_AUTO )
2567         {
2568 #if HAVE_INTERLACED
2569             int b_interlaced = fenc->param ? fenc->param->b_interlaced : h->param.b_interlaced;
2570 #else
2571             int b_interlaced = 0;
2572 #endif
2573             if( b_interlaced )
2574             {
2575                 int b_tff = fenc->param ? fenc->param->b_tff : h->param.b_tff;
2576                 fenc->i_pic_struct = b_tff ? PIC_STRUCT_TOP_BOTTOM : PIC_STRUCT_BOTTOM_TOP;
2577             }
2578             else
2579                 fenc->i_pic_struct = PIC_STRUCT_PROGRESSIVE;
2580         }
2581
2582         if( h->param.rc.b_mb_tree && h->param.rc.b_stat_read )
2583         {
2584             if( x264_macroblock_tree_read( h, fenc, pic_in->prop.quant_offsets ) )
2585                 return -1;
2586         }
2587         else
2588             x264_stack_align( x264_adaptive_quant_frame, h, fenc, pic_in->prop.quant_offsets );
2589
2590         if( pic_in->prop.quant_offsets_free )
2591             pic_in->prop.quant_offsets_free( pic_in->prop.quant_offsets );
2592
2593         if( h->frames.b_have_lowres )
2594             x264_frame_init_lowres( h, fenc );
2595
2596         /* 2: Place the frame into the queue for its slice type decision */
2597         x264_lookahead_put_frame( h, fenc );
2598
2599         if( h->frames.i_input <= h->frames.i_delay + 1 - h->i_thread_frames )
2600         {
2601             /* Nothing yet to encode, waiting for filling of buffers */
2602             pic_out->i_type = X264_TYPE_AUTO;
2603             return 0;
2604         }
2605     }
2606     else
2607     {
2608         /* signal kills for lookahead thread */
2609         x264_pthread_mutex_lock( &h->lookahead->ifbuf.mutex );
2610         h->lookahead->b_exit_thread = 1;
2611         x264_pthread_cond_broadcast( &h->lookahead->ifbuf.cv_fill );
2612         x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
2613     }
2614
2615     h->i_frame++;
2616     /* 3: The picture is analyzed in the lookahead */
2617     if( !h->frames.current[0] )
2618         x264_lookahead_get_frames( h );
2619
2620     if( !h->frames.current[0] && x264_lookahead_is_empty( h ) )
2621         return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
2622
2623     /* ------------------- Get frame to be encoded ------------------------- */
2624     /* 4: get picture to encode */
2625     h->fenc = x264_frame_shift( h->frames.current );
2626     if( h->i_frame == h->i_thread_frames - 1 )
2627         h->i_reordered_pts_delay = h->fenc->i_reordered_pts;
2628     if( h->fenc->param )
2629     {
2630         x264_encoder_reconfig( h, h->fenc->param );
2631         if( h->fenc->param->param_free )
2632             h->fenc->param->param_free( h->fenc->param );
2633     }
2634
2635     if( !IS_X264_TYPE_I( h->fenc->i_type ) )
2636     {
2637         int valid_refs_left = 0;
2638         for( int i = 0; h->frames.reference[i]; i++ )
2639             if( !h->frames.reference[i]->b_corrupt )
2640                 valid_refs_left++;
2641         /* No valid reference frames left: force an IDR. */
2642         if( !valid_refs_left )
2643         {
2644             h->fenc->b_keyframe = 1;
2645             h->fenc->i_type = X264_TYPE_IDR;
2646         }
2647     }
2648
2649     if( h->fenc->b_keyframe )
2650     {
2651         h->frames.i_last_keyframe = h->fenc->i_frame;
2652         if( h->fenc->i_type == X264_TYPE_IDR )
2653         {
2654             h->i_frame_num = 0;
2655             h->frames.i_last_idr = h->fenc->i_frame;
2656         }
2657     }
2658     h->sh.i_mmco_command_count =
2659     h->sh.i_mmco_remove_from_end = 0;
2660     h->b_ref_reorder[0] =
2661     h->b_ref_reorder[1] = 0;
2662     h->fdec->i_poc =
2663     h->fenc->i_poc = 2 * ( h->fenc->i_frame - X264_MAX( h->frames.i_last_idr, 0 ) );
2664
2665     /* ------------------- Setup frame context ----------------------------- */
2666     /* 5: Init data dependent of frame type */
2667     if( h->fenc->i_type == X264_TYPE_IDR )
2668     {
2669         /* reset ref pictures */
2670         i_nal_type    = NAL_SLICE_IDR;
2671         i_nal_ref_idc = NAL_PRIORITY_HIGHEST;
2672         h->sh.i_type = SLICE_TYPE_I;
2673         x264_reference_reset( h );
2674         h->frames.i_poc_last_open_gop = -1;
2675     }
2676     else if( h->fenc->i_type == X264_TYPE_I )
2677     {
2678         i_nal_type    = NAL_SLICE;
2679         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
2680         h->sh.i_type = SLICE_TYPE_I;
2681         x264_reference_hierarchy_reset( h );
2682         if( h->param.b_open_gop )
2683             h->frames.i_poc_last_open_gop = h->fenc->b_keyframe ? h->fenc->i_poc : -1;
2684     }
2685     else if( h->fenc->i_type == X264_TYPE_P )
2686     {
2687         i_nal_type    = NAL_SLICE;
2688         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
2689         h->sh.i_type = SLICE_TYPE_P;
2690         x264_reference_hierarchy_reset( h );
2691         h->frames.i_poc_last_open_gop = -1;
2692     }
2693     else if( h->fenc->i_type == X264_TYPE_BREF )
2694     {
2695         i_nal_type    = NAL_SLICE;
2696         i_nal_ref_idc = h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT ? NAL_PRIORITY_LOW : NAL_PRIORITY_HIGH;
2697         h->sh.i_type = SLICE_TYPE_B;
2698         x264_reference_hierarchy_reset( h );
2699     }
2700     else    /* B frame */
2701     {
2702         i_nal_type    = NAL_SLICE;
2703         i_nal_ref_idc = NAL_PRIORITY_DISPOSABLE;
2704         h->sh.i_type = SLICE_TYPE_B;
2705     }
2706
2707     h->fdec->i_type = h->fenc->i_type;
2708     h->fdec->i_frame = h->fenc->i_frame;
2709     h->fenc->b_kept_as_ref =
2710     h->fdec->b_kept_as_ref = i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE && h->param.i_keyint_max > 1;
2711
2712     h->fdec->i_pts = h->fenc->i_pts;
2713     if( h->frames.i_bframe_delay )
2714     {
2715         int64_t *prev_reordered_pts = thread_current->frames.i_prev_reordered_pts;
2716         h->fdec->i_dts = h->i_frame > h->frames.i_bframe_delay
2717                        ? prev_reordered_pts[ (h->i_frame - h->frames.i_bframe_delay) % h->frames.i_bframe_delay ]
2718                        : h->fenc->i_reordered_pts - h->frames.i_bframe_delay_time;
2719         prev_reordered_pts[ h->i_frame % h->frames.i_bframe_delay ] = h->fenc->i_reordered_pts;
2720     }
2721     else
2722         h->fdec->i_dts = h->fenc->i_reordered_pts;
2723     if( h->fenc->i_type == X264_TYPE_IDR )
2724         h->i_last_idr_pts = h->fdec->i_pts;
2725
2726     /* ------------------- Init                ----------------------------- */
2727     /* build ref list 0/1 */
2728     x264_reference_build_list( h, h->fdec->i_poc );
2729
2730     /* ---------------------- Write the bitstream -------------------------- */
2731     /* Init bitstream context */
2732     if( h->param.b_sliced_threads )
2733     {
2734         for( int i = 0; i < h->param.i_threads; i++ )
2735         {
2736             bs_init( &h->thread[i]->out.bs, h->thread[i]->out.p_bitstream, h->thread[i]->out.i_bitstream );
2737             h->thread[i]->out.i_nal = 0;
2738         }
2739     }
2740     else
2741     {
2742         bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
2743         h->out.i_nal = 0;
2744     }
2745
2746     if( h->param.b_aud )
2747     {
2748         int pic_type;
2749
2750         if( h->sh.i_type == SLICE_TYPE_I )
2751             pic_type = 0;
2752         else if( h->sh.i_type == SLICE_TYPE_P )
2753             pic_type = 1;
2754         else if( h->sh.i_type == SLICE_TYPE_B )
2755             pic_type = 2;
2756         else
2757             pic_type = 7;
2758
2759         x264_nal_start( h, NAL_AUD, NAL_PRIORITY_DISPOSABLE );
2760         bs_write( &h->out.bs, 3, pic_type );
2761         bs_rbsp_trailing( &h->out.bs );
2762         if( x264_nal_end( h ) )
2763             return -1;
2764         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2765     }
2766
2767     h->i_nal_type = i_nal_type;
2768     h->i_nal_ref_idc = i_nal_ref_idc;
2769
2770     if( h->param.b_intra_refresh )
2771     {
2772         if( IS_X264_TYPE_I( h->fenc->i_type ) )
2773         {
2774             h->fdec->i_frames_since_pir = 0;
2775             h->b_queued_intra_refresh = 0;
2776             /* PIR is currently only supported with ref == 1, so any intra frame effectively refreshes
2777              * the whole frame and counts as an intra refresh. */
2778             h->fdec->f_pir_position = h->mb.i_mb_width;
2779         }
2780         else if( h->fenc->i_type == X264_TYPE_P )
2781         {
2782             int pocdiff = (h->fdec->i_poc - h->fref[0][0]->i_poc)/2;
2783             float increment = X264_MAX( ((float)h->mb.i_mb_width-1) / h->param.i_keyint_max, 1 );
2784             h->fdec->f_pir_position = h->fref[0][0]->f_pir_position;
2785             h->fdec->i_frames_since_pir = h->fref[0][0]->i_frames_since_pir + pocdiff;
2786             if( h->fdec->i_frames_since_pir >= h->param.i_keyint_max ||
2787                 (h->b_queued_intra_refresh && h->fdec->f_pir_position + 0.5 >= h->mb.i_mb_width) )
2788             {
2789                 h->fdec->f_pir_position = 0;
2790                 h->fdec->i_frames_since_pir = 0;
2791                 h->b_queued_intra_refresh = 0;
2792                 h->fenc->b_keyframe = 1;
2793             }
2794             h->fdec->i_pir_start_col = h->fdec->f_pir_position+0.5;
2795             h->fdec->f_pir_position += increment * pocdiff;
2796             h->fdec->i_pir_end_col = h->fdec->f_pir_position+0.5;
2797             /* If our intra refresh has reached the right side of the frame, we're done. */
2798             if( h->fdec->i_pir_end_col >= h->mb.i_mb_width - 1 )
2799                 h->fdec->f_pir_position = h->mb.i_mb_width;
2800         }
2801     }
2802
2803     if( h->fenc->b_keyframe )
2804     {
2805         /* Write SPS and PPS */
2806         if( h->param.b_repeat_headers )
2807         {
2808             /* generate sequence parameters */
2809             x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
2810             x264_sps_write( &h->out.bs, h->sps );
2811             if( x264_nal_end( h ) )
2812                 return -1;
2813             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
2814
2815             /* generate picture parameters */
2816             x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
2817             x264_pps_write( &h->out.bs, h->pps );
2818             if( x264_nal_end( h ) )
2819                 return -1;
2820             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
2821         }
2822
2823         /* buffering period sei is written in x264_encoder_frame_end */
2824     }
2825
2826     /* write extra sei */
2827     for( int i = 0; i < h->fenc->extra_sei.num_payloads; i++ )
2828     {
2829         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2830         x264_sei_write( &h->out.bs, h->fenc->extra_sei.payloads[i].payload, h->fenc->extra_sei.payloads[i].payload_size,
2831                         h->fenc->extra_sei.payloads[i].payload_type );
2832         if( x264_nal_end( h ) )
2833             return -1;
2834         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2835         if( h->fenc->extra_sei.sei_free && h->fenc->extra_sei.payloads[i].payload )
2836             h->fenc->extra_sei.sei_free( h->fenc->extra_sei.payloads[i].payload );
2837     }
2838
2839     if( h->fenc->extra_sei.sei_free && h->fenc->extra_sei.payloads )
2840         h->fenc->extra_sei.sei_free( h->fenc->extra_sei.payloads );
2841
2842     if( h->fenc->b_keyframe )
2843     {
2844         if( h->param.b_repeat_headers && h->fenc->i_frame == 0 )
2845         {
2846             /* identify ourself */
2847             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2848             if( x264_sei_version_write( h, &h->out.bs ) )
2849                 return -1;
2850             if( x264_nal_end( h ) )
2851                 return -1;
2852             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2853         }
2854
2855         if( h->fenc->i_type != X264_TYPE_IDR )
2856         {
2857             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;
2858             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2859             x264_sei_recovery_point_write( h, &h->out.bs, time_to_recovery );
2860             if( x264_nal_end( h ) )
2861                 return -1;
2862             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2863         }
2864
2865         if ( h->param.i_frame_packing >= 0 )
2866         {
2867             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2868             x264_sei_frame_packing_write( h, &h->out.bs );
2869             if( x264_nal_end( h ) )
2870                 return -1;
2871             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2872         }
2873     }
2874
2875     /* generate sei pic timing */
2876     if( h->sps->vui.b_pic_struct_present || h->sps->vui.b_nal_hrd_parameters_present )
2877     {
2878         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2879         x264_sei_pic_timing_write( h, &h->out.bs );
2880         if( x264_nal_end( h ) )
2881             return -1;
2882         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2883     }
2884
2885     /* As required by Blu-ray. */
2886     if( !IS_X264_TYPE_B( h->fenc->i_type ) && h->b_sh_backup )
2887     {
2888         h->b_sh_backup = 0;
2889         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2890         x264_sei_dec_ref_pic_marking_write( h, &h->out.bs );
2891         if( x264_nal_end( h ) )
2892             return -1;
2893         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2894     }
2895
2896     if( h->fenc->b_keyframe && h->param.b_intra_refresh )
2897         h->i_cpb_delay_pir_offset = h->fenc->i_cpb_delay;
2898
2899     /* Init the rate control */
2900     /* FIXME: Include slice header bit cost. */
2901     x264_ratecontrol_start( h, h->fenc->i_qpplus1, overhead*8 );
2902     i_global_qp = x264_ratecontrol_qp( h );
2903
2904     pic_out->i_qpplus1 =
2905     h->fdec->i_qpplus1 = i_global_qp + 1;
2906
2907     if( h->param.rc.b_stat_read && h->sh.i_type != SLICE_TYPE_I )
2908     {
2909         x264_reference_build_list_optimal( h );
2910         x264_reference_check_reorder( h );
2911     }
2912
2913     if( h->i_ref[0] )
2914         h->fdec->i_poc_l0ref0 = h->fref[0][0]->i_poc;
2915
2916     /* ------------------------ Create slice header  ----------------------- */
2917     x264_slice_init( h, i_nal_type, i_global_qp );
2918
2919     /*------------------------- Weights -------------------------------------*/
2920     if( h->sh.i_type == SLICE_TYPE_B )
2921         x264_macroblock_bipred_init( h );
2922
2923     x264_weighted_pred_init( h );
2924
2925     if( i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE )
2926         h->i_frame_num++;
2927
2928     /* Write frame */
2929     h->i_threadslice_start = 0;
2930     h->i_threadslice_end = h->mb.i_mb_height;
2931     if( h->i_thread_frames > 1 )
2932     {
2933         x264_threadpool_run( h->threadpool, (void*)x264_slices_write, h );
2934         h->b_thread_active = 1;
2935     }
2936     else if( h->param.b_sliced_threads )
2937     {
2938         if( x264_threaded_slices_write( h ) )
2939             return -1;
2940     }
2941     else
2942         if( (intptr_t)x264_slices_write( h ) )
2943             return -1;
2944
2945     return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
2946 }
2947
2948 static int x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
2949                                    x264_nal_t **pp_nal, int *pi_nal,
2950                                    x264_picture_t *pic_out )
2951 {
2952     char psz_message[80];
2953
2954     if( h->b_thread_active )
2955     {
2956         h->b_thread_active = 0;
2957         if( (intptr_t)x264_threadpool_wait( h->threadpool, h ) )
2958             return -1;
2959     }
2960     if( !h->out.i_nal )
2961     {
2962         pic_out->i_type = X264_TYPE_AUTO;
2963         return 0;
2964     }
2965
2966     x264_emms();
2967     /* generate sei buffering period and insert it into place */
2968     if( h->fenc->b_keyframe && h->sps->vui.b_nal_hrd_parameters_present )
2969     {
2970         x264_hrd_fullness( h );
2971         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2972         x264_sei_buffering_period_write( h, &h->out.bs );
2973         if( x264_nal_end( h ) )
2974            return -1;
2975         /* buffering period sei must follow AUD, SPS and PPS and precede all other SEIs */
2976         int idx = 0;
2977         while( h->out.nal[idx].i_type == NAL_AUD ||
2978                h->out.nal[idx].i_type == NAL_SPS ||
2979                h->out.nal[idx].i_type == NAL_PPS )
2980             idx++;
2981         x264_nal_t nal_tmp = h->out.nal[h->out.i_nal-1];
2982         memmove( &h->out.nal[idx+1], &h->out.nal[idx], (h->out.i_nal-idx-1)*sizeof(x264_nal_t) );
2983         h->out.nal[idx] = nal_tmp;
2984     }
2985
2986     int frame_size = x264_encoder_encapsulate_nals( h, 0 );
2987     if( frame_size < 0 )
2988         return -1;
2989
2990     /* Set output picture properties */
2991     pic_out->i_type = h->fenc->i_type;
2992
2993     pic_out->b_keyframe = h->fenc->b_keyframe;
2994     pic_out->i_pic_struct = h->fenc->i_pic_struct;
2995
2996     pic_out->i_pts = h->fdec->i_pts;
2997     pic_out->i_dts = h->fdec->i_dts;
2998
2999     if( pic_out->i_pts < pic_out->i_dts )
3000         x264_log( h, X264_LOG_WARNING, "invalid DTS: PTS is less than DTS\n" );
3001
3002     pic_out->img.i_csp = X264_CSP_NV12;
3003 #if HIGH_BIT_DEPTH
3004     pic_out->img.i_csp |= X264_CSP_HIGH_DEPTH;
3005 #endif
3006     pic_out->img.i_plane = h->fdec->i_plane;
3007     for( int i = 0; i < 2; i++ )
3008     {
3009         pic_out->img.i_stride[i] = h->fdec->i_stride[i] * sizeof(pixel);
3010         pic_out->img.plane[i] = (uint8_t*)h->fdec->plane[i];
3011     }
3012
3013     x264_frame_push_unused( thread_current, h->fenc );
3014
3015     /* ---------------------- Update encoder state ------------------------- */
3016
3017     /* update rc */
3018     int filler = 0;
3019     if( x264_ratecontrol_end( h, frame_size * 8, &filler ) < 0 )
3020         return -1;
3021
3022     pic_out->hrd_timing = h->fenc->hrd_timing;
3023
3024     while( filler > 0 )
3025     {
3026         int f, overhead;
3027         overhead = (FILLER_OVERHEAD - h->param.b_annexb);
3028         if( h->param.i_slice_max_size && filler > h->param.i_slice_max_size )
3029         {
3030             int next_size = filler - h->param.i_slice_max_size;
3031             int overflow = X264_MAX( overhead - next_size, 0 );
3032             f = h->param.i_slice_max_size - overhead - overflow;
3033         }
3034         else
3035             f = X264_MAX( 0, filler - overhead );
3036
3037         x264_nal_start( h, NAL_FILLER, NAL_PRIORITY_DISPOSABLE );
3038         x264_filler_write( h, &h->out.bs, f );
3039         if( x264_nal_end( h ) )
3040             return -1;
3041         int total_size = x264_encoder_encapsulate_nals( h, h->out.i_nal-1 );
3042         if( total_size < 0 )
3043             return -1;
3044         frame_size += total_size;
3045         filler -= total_size;
3046     }
3047
3048     /* End bitstream, set output  */
3049     *pi_nal = h->out.i_nal;
3050     *pp_nal = h->out.nal;
3051
3052     h->out.i_nal = 0;
3053
3054     x264_noise_reduction_update( h );
3055
3056     /* ---------------------- Compute/Print statistics --------------------- */
3057     x264_thread_sync_stat( h, h->thread[0] );
3058
3059     /* Slice stat */
3060     h->stat.i_frame_count[h->sh.i_type]++;
3061     h->stat.i_frame_size[h->sh.i_type] += frame_size;
3062     h->stat.f_frame_qp[h->sh.i_type] += h->fdec->f_qp_avg_aq;
3063
3064     for( int i = 0; i < X264_MBTYPE_MAX; i++ )
3065         h->stat.i_mb_count[h->sh.i_type][i] += h->stat.frame.i_mb_count[i];
3066     for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
3067         h->stat.i_mb_partition[h->sh.i_type][i] += h->stat.frame.i_mb_partition[i];
3068     for( int i = 0; i < 2; i++ )
3069         h->stat.i_mb_count_8x8dct[i] += h->stat.frame.i_mb_count_8x8dct[i];
3070     for( int i = 0; i < 6; i++ )
3071         h->stat.i_mb_cbp[i] += h->stat.frame.i_mb_cbp[i];
3072     for( int i = 0; i < 4; i++ )
3073         for( int j = 0; j < 13; j++ )
3074             h->stat.i_mb_pred_mode[i][j] += h->stat.frame.i_mb_pred_mode[i][j];
3075     if( h->sh.i_type != SLICE_TYPE_I )
3076         for( int i_list = 0; i_list < 2; i_list++ )
3077             for( int i = 0; i < X264_REF_MAX*2; i++ )
3078                 h->stat.i_mb_count_ref[h->sh.i_type][i_list][i] += h->stat.frame.i_mb_count_ref[i_list][i];
3079     for( int i = 0; i < 3; i++ )
3080         h->stat.i_mb_field[i] += h->stat.frame.i_mb_field[i];
3081     if( h->sh.i_type == SLICE_TYPE_P && h->param.analyse.i_weighted_pred >= X264_WEIGHTP_SIMPLE )
3082     {
3083         h->stat.i_wpred[0] += !!h->sh.weight[0][0].weightfn;
3084         h->stat.i_wpred[1] += !!h->sh.weight[0][1].weightfn || !!h->sh.weight[0][2].weightfn;
3085     }
3086     if( h->sh.i_type == SLICE_TYPE_B )
3087     {
3088         h->stat.i_direct_frames[ h->sh.b_direct_spatial_mv_pred ] ++;
3089         if( h->mb.b_direct_auto_write )
3090         {
3091             //FIXME somewhat arbitrary time constants
3092             if( h->stat.i_direct_score[0] + h->stat.i_direct_score[1] > h->mb.i_mb_count )
3093                 for( int i = 0; i < 2; i++ )
3094                     h->stat.i_direct_score[i] = h->stat.i_direct_score[i] * 9/10;
3095             for( int i = 0; i < 2; i++ )
3096                 h->stat.i_direct_score[i] += h->stat.frame.i_direct_score[i];
3097         }
3098     }
3099     else
3100         h->stat.i_consecutive_bframes[h->fenc->i_bframes]++;
3101
3102     psz_message[0] = '\0';
3103     double dur = h->fenc->f_duration;
3104     h->stat.f_frame_duration[h->sh.i_type] += dur;
3105     if( h->param.analyse.b_psnr )
3106     {
3107         int64_t ssd[3] =
3108         {
3109             h->stat.frame.i_ssd[0],
3110             h->stat.frame.i_ssd[1],
3111             h->stat.frame.i_ssd[2],
3112         };
3113
3114         h->stat.f_ssd_global[h->sh.i_type]   += dur * (ssd[0] + ssd[1] + ssd[2]);
3115         h->stat.f_psnr_average[h->sh.i_type] += dur * x264_psnr( ssd[0] + ssd[1] + ssd[2], 3 * h->param.i_width * h->param.i_height / 2 );
3116         h->stat.f_psnr_mean_y[h->sh.i_type]  += dur * x264_psnr( ssd[0], h->param.i_width * h->param.i_height );
3117         h->stat.f_psnr_mean_u[h->sh.i_type]  += dur * x264_psnr( ssd[1], h->param.i_width * h->param.i_height / 4 );
3118         h->stat.f_psnr_mean_v[h->sh.i_type]  += dur * x264_psnr( ssd[2], h->param.i_width * h->param.i_height / 4 );
3119
3120         snprintf( psz_message, 80, " PSNR Y:%5.2f U:%5.2f V:%5.2f",
3121                   x264_psnr( ssd[0], h->param.i_width * h->param.i_height ),
3122                   x264_psnr( ssd[1], h->param.i_width * h->param.i_height / 4),
3123                   x264_psnr( ssd[2], h->param.i_width * h->param.i_height / 4) );
3124     }
3125
3126     if( h->param.analyse.b_ssim )
3127     {
3128         double ssim_y = h->stat.frame.f_ssim
3129                       / (((h->param.i_width-6)>>2) * ((h->param.i_height-6)>>2));
3130         h->stat.f_ssim_mean_y[h->sh.i_type] += ssim_y * dur;
3131         snprintf( psz_message + strlen(psz_message), 80 - strlen(psz_message),
3132                   " SSIM Y:%.5f", ssim_y );
3133     }
3134     psz_message[79] = '\0';
3135
3136     x264_log( h, X264_LOG_DEBUG,
3137                   "frame=%4d QP=%.2f NAL=%d Slice:%c Poc:%-3d I:%-4d P:%-4d SKIP:%-4d size=%d bytes%s\n",
3138               h->i_frame,
3139               h->fdec->f_qp_avg_aq,
3140               h->i_nal_ref_idc,
3141               h->sh.i_type == SLICE_TYPE_I ? 'I' : (h->sh.i_type == SLICE_TYPE_P ? 'P' : 'B' ),
3142               h->fdec->i_poc,
3143               h->stat.frame.i_mb_count_i,
3144               h->stat.frame.i_mb_count_p,
3145               h->stat.frame.i_mb_count_skip,
3146               frame_size,
3147               psz_message );
3148
3149     // keep stats all in one place
3150     x264_thread_sync_stat( h->thread[0], h );
3151     // for the use of the next frame
3152     x264_thread_sync_stat( thread_current, h );
3153
3154 #ifdef DEBUG_MB_TYPE
3155 {
3156     static const char mb_chars[] = { 'i', 'i', 'I', 'C', 'P', '8', 'S',
3157         'D', '<', 'X', 'B', 'X', '>', 'B', 'B', 'B', 'B', '8', 'S' };
3158     for( int mb_xy = 0; mb_xy < h->mb.i_mb_width * h->mb.i_mb_height; mb_xy++ )
3159     {
3160         if( h->mb.type[mb_xy] < X264_MBTYPE_MAX && h->mb.type[mb_xy] >= 0 )
3161             fprintf( stderr, "%c ", mb_chars[ h->mb.type[mb_xy] ] );
3162         else
3163             fprintf( stderr, "? " );
3164
3165         if( (mb_xy+1) % h->mb.i_mb_width == 0 )
3166             fprintf( stderr, "\n" );
3167     }
3168 }
3169 #endif
3170
3171     /* Remove duplicates, must be done near the end as breaks h->fref0 array
3172      * by freeing some of its pointers. */
3173     for( int i = 0; i < h->i_ref[0]; i++ )
3174         if( h->fref[0][i] && h->fref[0][i]->b_duplicate )
3175         {
3176             x264_frame_push_blank_unused( h, h->fref[0][i] );
3177             h->fref[0][i] = 0;
3178         }
3179
3180     if( h->param.psz_dump_yuv )
3181         x264_frame_dump( h );
3182     x264_emms();
3183
3184     return frame_size;
3185 }
3186
3187 static void x264_print_intra( int64_t *i_mb_count, double i_count, int b_print_pcm, char *intra )
3188 {
3189     intra += sprintf( intra, "I16..4%s: %4.1f%% %4.1f%% %4.1f%%",
3190         b_print_pcm ? "..PCM" : "",
3191         i_mb_count[I_16x16]/ i_count,
3192         i_mb_count[I_8x8]  / i_count,
3193         i_mb_count[I_4x4]  / i_count );
3194     if( b_print_pcm )
3195         sprintf( intra, " %4.1f%%", i_mb_count[I_PCM]  / i_count );
3196 }
3197
3198 /****************************************************************************
3199  * x264_encoder_close:
3200  ****************************************************************************/
3201 void    x264_encoder_close  ( x264_t *h )
3202 {
3203     int64_t i_yuv_size = 3 * h->param.i_width * h->param.i_height / 2;
3204     int64_t i_mb_count_size[2][7] = {{0}};
3205     char buf[200];
3206     int b_print_pcm = h->stat.i_mb_count[SLICE_TYPE_I][I_PCM]
3207                    || h->stat.i_mb_count[SLICE_TYPE_P][I_PCM]
3208                    || h->stat.i_mb_count[SLICE_TYPE_B][I_PCM];
3209
3210     x264_lookahead_delete( h );
3211
3212     if( h->param.i_threads > 1 )
3213         x264_threadpool_delete( h->threadpool );
3214     if( h->i_thread_frames > 1 )
3215     {
3216         for( int i = 0; i < h->i_thread_frames; i++ )
3217             if( h->thread[i]->b_thread_active )
3218             {
3219                 assert( h->thread[i]->fenc->i_reference_count == 1 );
3220                 x264_frame_delete( h->thread[i]->fenc );
3221             }
3222
3223         x264_t *thread_prev = h->thread[h->i_thread_phase];
3224         x264_thread_sync_ratecontrol( h, thread_prev, h );
3225         x264_thread_sync_ratecontrol( thread_prev, thread_prev, h );
3226         h->i_frame = thread_prev->i_frame + 1 - h->i_thread_frames;
3227     }
3228     h->i_frame++;
3229
3230     /* Slices used and PSNR */
3231     for( int i = 0; i < 3; i++ )
3232     {
3233         static const uint8_t slice_order[] = { SLICE_TYPE_I, SLICE_TYPE_P, SLICE_TYPE_B };
3234         int i_slice = slice_order[i];
3235
3236         if( h->stat.i_frame_count[i_slice] > 0 )
3237         {
3238             int i_count = h->stat.i_frame_count[i_slice];
3239             double dur =  h->stat.f_frame_duration[i_slice];
3240             if( h->param.analyse.b_psnr )
3241             {
3242                 x264_log( h, X264_LOG_INFO,
3243                           "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",
3244                           slice_type_to_char[i_slice],
3245                           i_count,
3246                           h->stat.f_frame_qp[i_slice] / i_count,
3247                           (double)h->stat.i_frame_size[i_slice] / i_count,
3248                           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,
3249                           h->stat.f_psnr_average[i_slice] / dur,
3250                           x264_psnr( h->stat.f_ssd_global[i_slice], dur * i_yuv_size ) );
3251             }
3252             else
3253             {
3254                 x264_log( h, X264_LOG_INFO,
3255                           "frame %c:%-5d Avg QP:%5.2f  size:%6.0f\n",
3256                           slice_type_to_char[i_slice],
3257                           i_count,
3258                           h->stat.f_frame_qp[i_slice] / i_count,
3259                           (double)h->stat.i_frame_size[i_slice] / i_count );
3260             }
3261         }
3262     }
3263     if( h->param.i_bframe && h->stat.i_frame_count[SLICE_TYPE_B] )
3264     {
3265         char *p = buf;
3266         int den = 0;
3267         // weight by number of frames (including the I/P-frames) that are in a sequence of N B-frames
3268         for( int i = 0; i <= h->param.i_bframe; i++ )
3269             den += (i+1) * h->stat.i_consecutive_bframes[i];
3270         for( int i = 0; i <= h->param.i_bframe; i++ )
3271             p += sprintf( p, " %4.1f%%", 100. * (i+1) * h->stat.i_consecutive_bframes[i] / den );
3272         x264_log( h, X264_LOG_INFO, "consecutive B-frames:%s\n", buf );
3273     }
3274
3275     for( int i_type = 0; i_type < 2; i_type++ )
3276         for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
3277         {
3278             if( i == D_DIRECT_8x8 ) continue; /* direct is counted as its own type */
3279             i_mb_count_size[i_type][x264_mb_partition_pixel_table[i]] += h->stat.i_mb_partition[i_type][i];
3280         }
3281
3282     /* MB types used */
3283     if( h->stat.i_frame_count[SLICE_TYPE_I] > 0 )
3284     {
3285         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_I];
3286         double i_count = h->stat.i_frame_count[SLICE_TYPE_I] * h->mb.i_mb_count / 100.0;
3287         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
3288         x264_log( h, X264_LOG_INFO, "mb I  %s\n", buf );
3289     }
3290     if( h->stat.i_frame_count[SLICE_TYPE_P] > 0 )
3291     {
3292         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_P];
3293         double i_count = h->stat.i_frame_count[SLICE_TYPE_P] * h->mb.i_mb_count / 100.0;
3294         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_P];
3295         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
3296         x264_log( h, X264_LOG_INFO,
3297                   "mb P  %s  P16..4: %4.1f%% %4.1f%% %4.1f%% %4.1f%% %4.1f%%    skip:%4.1f%%\n",
3298                   buf,
3299                   i_mb_size[PIXEL_16x16] / (i_count*4),
3300                   (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
3301                   i_mb_size[PIXEL_8x8] / (i_count*4),
3302                   (i_mb_size[PIXEL_8x4] + i_mb_size[PIXEL_4x8]) / (i_count*4),
3303                   i_mb_size[PIXEL_4x4] / (i_count*4),
3304                   i_mb_count[P_SKIP] / i_count );
3305     }
3306     if( h->stat.i_frame_count[SLICE_TYPE_B] > 0 )
3307     {
3308         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_B];
3309         double i_count = h->stat.i_frame_count[SLICE_TYPE_B] * h->mb.i_mb_count / 100.0;
3310         double i_mb_list_count;
3311         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_B];
3312         int64_t list_count[3] = {0}; /* 0 == L0, 1 == L1, 2 == BI */
3313         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
3314         for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
3315             for( int j = 0; j < 2; j++ )
3316             {
3317                 int l0 = x264_mb_type_list_table[i][0][j];
3318                 int l1 = x264_mb_type_list_table[i][1][j];
3319                 if( l0 || l1 )
3320                     list_count[l1+l0*l1] += h->stat.i_mb_count[SLICE_TYPE_B][i] * 2;
3321             }
3322         list_count[0] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L0_8x8];
3323         list_count[1] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L1_8x8];
3324         list_count[2] += h->stat.i_mb_partition[SLICE_TYPE_B][D_BI_8x8];
3325         i_mb_count[B_DIRECT] += (h->stat.i_mb_partition[SLICE_TYPE_B][D_DIRECT_8x8]+2)/4;
3326         i_mb_list_count = (list_count[0] + list_count[1] + list_count[2]) / 100.0;
3327         sprintf( buf + strlen(buf), "  B16..8: %4.1f%% %4.1f%% %4.1f%%  direct:%4.1f%%  skip:%4.1f%%",
3328                  i_mb_size[PIXEL_16x16] / (i_count*4),
3329                  (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
3330                  i_mb_size[PIXEL_8x8] / (i_count*4),
3331                  i_mb_count[B_DIRECT] / i_count,
3332                  i_mb_count[B_SKIP]   / i_count );
3333         if( i_mb_list_count != 0 )
3334             sprintf( buf + strlen(buf), "  L0:%4.1f%% L1:%4.1f%% BI:%4.1f%%",
3335                      list_count[0] / i_mb_list_count,
3336                      list_count[1] / i_mb_list_count,
3337                      list_count[2] / i_mb_list_count );
3338         x264_log( h, X264_LOG_INFO, "mb B  %s\n", buf );
3339     }
3340
3341     x264_ratecontrol_summary( h );
3342
3343     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 )
3344     {
3345 #define SUM3(p) (p[SLICE_TYPE_I] + p[SLICE_TYPE_P] + p[SLICE_TYPE_B])
3346 #define SUM3b(p,o) (p[SLICE_TYPE_I][o] + p[SLICE_TYPE_P][o] + p[SLICE_TYPE_B][o])
3347         int64_t i_i8x8 = SUM3b( h->stat.i_mb_count, I_8x8 );
3348         int64_t i_intra = i_i8x8 + SUM3b( h->stat.i_mb_count, I_4x4 )
3349                                  + SUM3b( h->stat.i_mb_count, I_16x16 );
3350         int64_t i_all_intra = i_intra + SUM3b( h->stat.i_mb_count, I_PCM);
3351         int64_t i_skip = SUM3b( h->stat.i_mb_count, P_SKIP )
3352                        + SUM3b( h->stat.i_mb_count, B_SKIP );
3353         const int i_count = h->stat.i_frame_count[SLICE_TYPE_I] +
3354                             h->stat.i_frame_count[SLICE_TYPE_P] +
3355                             h->stat.i_frame_count[SLICE_TYPE_B];
3356         int64_t i_mb_count = (int64_t)i_count * h->mb.i_mb_count;
3357         int64_t i_inter = i_mb_count - i_skip - i_intra;
3358         const double duration = h->stat.f_frame_duration[SLICE_TYPE_I] +
3359                                 h->stat.f_frame_duration[SLICE_TYPE_P] +
3360                                 h->stat.f_frame_duration[SLICE_TYPE_B];
3361         float f_bitrate = SUM3(h->stat.i_frame_size) / duration / 125;
3362
3363         if( PARAM_INTERLACED )
3364         {
3365             char *fieldstats = buf;
3366             fieldstats[0] = 0;
3367             if( i_inter )
3368                 fieldstats += sprintf( fieldstats, " inter:%.1f%%", h->stat.i_mb_field[1] * 100.0 / i_inter );
3369             if( i_skip )
3370                 fieldstats += sprintf( fieldstats, " skip:%.1f%%", h->stat.i_mb_field[2] * 100.0 / i_skip );
3371             x264_log( h, X264_LOG_INFO, "field mbs: intra: %.1f%%%s\n",
3372                       h->stat.i_mb_field[0] * 100.0 / i_intra, buf );
3373         }
3374
3375         if( h->pps->b_transform_8x8_mode )
3376         {
3377             buf[0] = 0;
3378             if( h->stat.i_mb_count_8x8dct[0] )
3379                 sprintf( buf, " inter:%.1f%%", 100. * h->stat.i_mb_count_8x8dct[1] / h->stat.i_mb_count_8x8dct[0] );
3380             x264_log( h, X264_LOG_INFO, "8x8 transform intra:%.1f%%%s\n", 100. * i_i8x8 / i_intra, buf );
3381         }
3382
3383         if( (h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO ||
3384             (h->stat.i_direct_frames[0] && h->stat.i_direct_frames[1]))
3385             && h->stat.i_frame_count[SLICE_TYPE_B] )
3386         {
3387             x264_log( h, X264_LOG_INFO, "direct mvs  spatial:%.1f%% temporal:%.1f%%\n",
3388                       h->stat.i_direct_frames[1] * 100. / h->stat.i_frame_count[SLICE_TYPE_B],
3389                       h->stat.i_direct_frames[0] * 100. / h->stat.i_frame_count[SLICE_TYPE_B] );
3390         }
3391
3392         buf[0] = 0;
3393         if( i_mb_count != i_all_intra )
3394             sprintf( buf, " inter: %.1f%% %.1f%% %.1f%%",
3395                      h->stat.i_mb_cbp[1] * 100.0 / ((i_mb_count - i_all_intra)*4),
3396                      h->stat.i_mb_cbp[3] * 100.0 / ((i_mb_count - i_all_intra)  ),
3397                      h->stat.i_mb_cbp[5] * 100.0 / ((i_mb_count - i_all_intra)) );
3398         x264_log( h, X264_LOG_INFO, "coded y,uvDC,uvAC intra: %.1f%% %.1f%% %.1f%%%s\n",
3399                   h->stat.i_mb_cbp[0] * 100.0 / (i_all_intra*4),
3400                   h->stat.i_mb_cbp[2] * 100.0 / (i_all_intra  ),
3401                   h->stat.i_mb_cbp[4] * 100.0 / (i_all_intra  ), buf );
3402
3403         int64_t fixed_pred_modes[4][9] = {{0}};
3404         int64_t sum_pred_modes[4] = {0};
3405         for( int i = 0; i <= I_PRED_16x16_DC_128; i++ )
3406         {
3407             fixed_pred_modes[0][x264_mb_pred_mode16x16_fix[i]] += h->stat.i_mb_pred_mode[0][i];
3408             sum_pred_modes[0] += h->stat.i_mb_pred_mode[0][i];
3409         }
3410         if( sum_pred_modes[0] )
3411             x264_log( h, X264_LOG_INFO, "i16 v,h,dc,p: %2.0f%% %2.0f%% %2.0f%% %2.0f%%\n",
3412                       fixed_pred_modes[0][0] * 100.0 / sum_pred_modes[0],
3413                       fixed_pred_modes[0][1] * 100.0 / sum_pred_modes[0],
3414                       fixed_pred_modes[0][2] * 100.0 / sum_pred_modes[0],
3415                       fixed_pred_modes[0][3] * 100.0 / sum_pred_modes[0] );
3416         for( int i = 1; i <= 2; i++ )
3417         {
3418             for( int j = 0; j <= I_PRED_8x8_DC_128; j++ )
3419             {
3420                 fixed_pred_modes[i][x264_mb_pred_mode4x4_fix(j)] += h->stat.i_mb_pred_mode[i][j];
3421                 sum_pred_modes[i] += h->stat.i_mb_pred_mode[i][j];
3422             }
3423             if( sum_pred_modes[i] )
3424                 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,
3425                           fixed_pred_modes[i][0] * 100.0 / sum_pred_modes[i],
3426                           fixed_pred_modes[i][1] * 100.0 / sum_pred_modes[i],
3427                           fixed_pred_modes[i][2] * 100.0 / sum_pred_modes[i],
3428                           fixed_pred_modes[i][3] * 100.0 / sum_pred_modes[i],
3429                           fixed_pred_modes[i][4] * 100.0 / sum_pred_modes[i],
3430                           fixed_pred_modes[i][5] * 100.0 / sum_pred_modes[i],
3431                           fixed_pred_modes[i][6] * 100.0 / sum_pred_modes[i],
3432                           fixed_pred_modes[i][7] * 100.0 / sum_pred_modes[i],
3433                           fixed_pred_modes[i][8] * 100.0 / sum_pred_modes[i] );
3434         }
3435         for( int i = 0; i <= I_PRED_CHROMA_DC_128; i++ )
3436         {
3437             fixed_pred_modes[3][x264_mb_pred_mode8x8c_fix[i]] += h->stat.i_mb_pred_mode[3][i];
3438             sum_pred_modes[3] += h->stat.i_mb_pred_mode[3][i];
3439         }
3440         if( sum_pred_modes[3] )
3441             x264_log( h, X264_LOG_INFO, "i8c dc,h,v,p: %2.0f%% %2.0f%% %2.0f%% %2.0f%%\n",
3442                       fixed_pred_modes[3][0] * 100.0 / sum_pred_modes[3],
3443                       fixed_pred_modes[3][1] * 100.0 / sum_pred_modes[3],
3444                       fixed_pred_modes[3][2] * 100.0 / sum_pred_modes[3],
3445                       fixed_pred_modes[3][3] * 100.0 / sum_pred_modes[3] );
3446
3447         if( h->param.analyse.i_weighted_pred >= X264_WEIGHTP_SIMPLE && h->stat.i_frame_count[SLICE_TYPE_P] > 0 )
3448             x264_log( h, X264_LOG_INFO, "Weighted P-Frames: Y:%.1f%% UV:%.1f%%\n",
3449                       h->stat.i_wpred[0] * 100.0 / h->stat.i_frame_count[SLICE_TYPE_P],
3450                       h->stat.i_wpred[1] * 100.0 / h->stat.i_frame_count[SLICE_TYPE_P] );
3451
3452         for( int i_list = 0; i_list < 2; i_list++ )
3453             for( int i_slice = 0; i_slice < 2; i_slice++ )
3454             {
3455                 char *p = buf;
3456                 int64_t i_den = 0;
3457                 int i_max = 0;
3458                 for( int i = 0; i < X264_REF_MAX*2; i++ )
3459                     if( h->stat.i_mb_count_ref[i_slice][i_list][i] )
3460                     {
3461                         i_den += h->stat.i_mb_count_ref[i_slice][i_list][i];
3462                         i_max = i;
3463                     }
3464                 if( i_max == 0 )
3465                     continue;
3466                 for( int i = 0; i <= i_max; i++ )
3467                     p += sprintf( p, " %4.1f%%", 100. * h->stat.i_mb_count_ref[i_slice][i_list][i] / i_den );
3468                 x264_log( h, X264_LOG_INFO, "ref %c L%d:%s\n", "PB"[i_slice], i_list, buf );
3469             }
3470
3471         if( h->param.analyse.b_ssim )
3472         {
3473             float ssim = SUM3( h->stat.f_ssim_mean_y ) / duration;
3474             x264_log( h, X264_LOG_INFO, "SSIM Mean Y:%.7f (%6.3fdb)\n", ssim, x264_ssim( ssim ) );
3475         }
3476         if( h->param.analyse.b_psnr )
3477         {
3478             x264_log( h, X264_LOG_INFO,
3479                       "PSNR Mean Y:%6.3f U:%6.3f V:%6.3f Avg:%6.3f Global:%6.3f kb/s:%.2f\n",
3480                       SUM3( h->stat.f_psnr_mean_y ) / duration,
3481                       SUM3( h->stat.f_psnr_mean_u ) / duration,
3482                       SUM3( h->stat.f_psnr_mean_v ) / duration,
3483                       SUM3( h->stat.f_psnr_average ) / duration,
3484                       x264_psnr( SUM3( h->stat.f_ssd_global ), duration * i_yuv_size ),
3485                       f_bitrate );
3486         }
3487         else
3488             x264_log( h, X264_LOG_INFO, "kb/s:%.2f\n", f_bitrate );
3489     }
3490
3491     /* rc */
3492     x264_ratecontrol_delete( h );
3493
3494     /* param */
3495     if( h->param.rc.psz_stat_out )
3496         free( h->param.rc.psz_stat_out );
3497     if( h->param.rc.psz_stat_in )
3498         free( h->param.rc.psz_stat_in );
3499
3500     x264_cqm_delete( h );
3501     x264_free( h->nal_buffer );
3502     x264_analyse_free_costs( h );
3503
3504     if( h->i_thread_frames > 1)
3505         h = h->thread[h->i_thread_phase];
3506
3507     /* frames */
3508     x264_frame_delete_list( h->frames.unused[0] );
3509     x264_frame_delete_list( h->frames.unused[1] );
3510     x264_frame_delete_list( h->frames.current );
3511     x264_frame_delete_list( h->frames.blank_unused );
3512
3513     h = h->thread[0];
3514
3515     for( int i = 0; i < h->i_thread_frames; i++ )
3516         if( h->thread[i]->b_thread_active )
3517             for( int j = 0; j < h->thread[i]->i_ref[0]; j++ )
3518                 if( h->thread[i]->fref[0][j] && h->thread[i]->fref[0][j]->b_duplicate )
3519                     x264_frame_delete( h->thread[i]->fref[0][j] );
3520
3521     for( int i = h->param.i_threads - 1; i >= 0; i-- )
3522     {
3523         x264_frame_t **frame;
3524
3525         if( !h->param.b_sliced_threads || i == 0 )
3526         {
3527             for( frame = h->thread[i]->frames.reference; *frame; frame++ )
3528             {
3529                 assert( (*frame)->i_reference_count > 0 );
3530                 (*frame)->i_reference_count--;
3531                 if( (*frame)->i_reference_count == 0 )
3532                     x264_frame_delete( *frame );
3533             }
3534             frame = &h->thread[i]->fdec;
3535             if( *frame )
3536             {
3537                 assert( (*frame)->i_reference_count > 0 );
3538                 (*frame)->i_reference_count--;
3539                 if( (*frame)->i_reference_count == 0 )
3540                     x264_frame_delete( *frame );
3541             }
3542             x264_macroblock_cache_free( h->thread[i] );
3543         }
3544         x264_macroblock_thread_free( h->thread[i], 0 );
3545         x264_free( h->thread[i]->out.p_bitstream );
3546         x264_free( h->thread[i]->out.nal);
3547         x264_free( h->thread[i] );
3548     }
3549 }
3550
3551 int x264_encoder_delayed_frames( x264_t *h )
3552 {
3553     int delayed_frames = 0;
3554     if( h->i_thread_frames > 1 )
3555     {
3556         for( int i = 0; i < h->i_thread_frames; i++ )
3557             delayed_frames += h->thread[i]->b_thread_active;
3558         h = h->thread[h->i_thread_phase];
3559     }
3560     for( int i = 0; h->frames.current[i]; i++ )
3561         delayed_frames++;
3562     x264_pthread_mutex_lock( &h->lookahead->ofbuf.mutex );
3563     x264_pthread_mutex_lock( &h->lookahead->ifbuf.mutex );
3564     x264_pthread_mutex_lock( &h->lookahead->next.mutex );
3565     delayed_frames += h->lookahead->ifbuf.i_size + h->lookahead->next.i_size + h->lookahead->ofbuf.i_size;
3566     x264_pthread_mutex_unlock( &h->lookahead->next.mutex );
3567     x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
3568     x264_pthread_mutex_unlock( &h->lookahead->ofbuf.mutex );
3569     return delayed_frames;
3570 }
3571
3572 int x264_encoder_maximum_delayed_frames( x264_t *h )
3573 {
3574     return h->frames.i_delay;
3575 }