MBAFF: Add frame/field MB 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     if( (h->param.b_cabac && (h->cabac.p_end - h->cabac.p < 2500)) ||
339         (h->out.bs.p_end - h->out.bs.p < 2500) )
340     {
341         h->out.i_bitstream += 100000;
342         CHECKED_MALLOC( h->out.p_bitstream, h->out.i_bitstream );
343         h->mc.memcpy_aligned( h->out.p_bitstream, bs_bak, (h->out.i_bitstream - 100000) & ~15 );
344         intptr_t delta = h->out.p_bitstream - bs_bak;
345
346         h->out.bs.p_start += delta;
347         h->out.bs.p += delta;
348         h->out.bs.p_end = h->out.p_bitstream + h->out.i_bitstream;
349
350         h->cabac.p_start += delta;
351         h->cabac.p += delta;
352         h->cabac.p_end = h->out.p_bitstream + h->out.i_bitstream;
353
354         for( int i = 0; i <= h->out.i_nal; i++ )
355             h->out.nal[i].p_payload += delta;
356         x264_free( bs_bak );
357     }
358     return 0;
359 fail:
360     x264_free( bs_bak );
361     return -1;
362 }
363
364 #if HAVE_THREAD
365 static void x264_encoder_thread_init( x264_t *h )
366 {
367     if( h->param.i_sync_lookahead )
368         x264_lower_thread_priority( 10 );
369
370 #if HAVE_MMX
371     /* Misalign mask has to be set separately for each thread. */
372     if( h->param.cpu&X264_CPU_SSE_MISALIGN )
373         x264_cpu_mask_misalign_sse();
374 #endif
375 }
376 #endif
377
378 /****************************************************************************
379  *
380  ****************************************************************************
381  ****************************** External API*********************************
382  ****************************************************************************
383  *
384  ****************************************************************************/
385
386 static int x264_validate_parameters( x264_t *h, int b_open )
387 {
388 #if HAVE_MMX
389 #ifdef __SSE__
390     if( b_open && !(x264_cpu_detect() & X264_CPU_SSE) )
391     {
392         x264_log( h, X264_LOG_ERROR, "your cpu does not support SSE1, but x264 was compiled with asm support\n");
393 #else
394     if( b_open && !(x264_cpu_detect() & X264_CPU_MMXEXT) )
395     {
396         x264_log( h, X264_LOG_ERROR, "your cpu does not support MMXEXT, but x264 was compiled with asm support\n");
397 #endif
398         x264_log( h, X264_LOG_ERROR, "to run x264, recompile without asm support (configure --disable-asm)\n");
399         return -1;
400     }
401 #endif
402     if( h->param.i_width <= 0 || h->param.i_height <= 0 )
403     {
404         x264_log( h, X264_LOG_ERROR, "invalid width x height (%dx%d)\n",
405                   h->param.i_width, h->param.i_height );
406         return -1;
407     }
408
409     if( h->param.i_width % 2 || h->param.i_height % 2 )
410     {
411         x264_log( h, X264_LOG_ERROR, "width or height not divisible by 2 (%dx%d)\n",
412                   h->param.i_width, h->param.i_height );
413         return -1;
414     }
415     int i_csp = h->param.i_csp & X264_CSP_MASK;
416     if( i_csp <= X264_CSP_NONE || i_csp >= X264_CSP_MAX )
417     {
418         x264_log( h, X264_LOG_ERROR, "invalid CSP (only I420/YV12/NV12 supported)\n" );
419         return -1;
420     }
421
422 #if HAVE_INTERLACED
423     h->param.b_interlaced = !!PARAM_INTERLACED;
424 #else
425     if( h->param.b_interlaced )
426     {
427         x264_log( h, X264_LOG_ERROR, "not compiled with interlaced support\n" );
428         return -1;
429     }
430 #endif
431
432     if( (h->param.crop_rect.i_left + h->param.crop_rect.i_right ) >= h->param.i_width ||
433         (h->param.crop_rect.i_top  + h->param.crop_rect.i_bottom) >= h->param.i_height )
434     {
435         x264_log( h, X264_LOG_ERROR, "invalid crop-rect %u,%u,%u,%u\n", h->param.crop_rect.i_left,
436                   h->param.crop_rect.i_top, h->param.crop_rect.i_right,  h->param.crop_rect.i_bottom );
437         return -1;
438     }
439
440     if( h->param.i_threads == X264_THREADS_AUTO )
441         h->param.i_threads = x264_cpu_num_processors() * (h->param.b_sliced_threads?2:3)/2;
442     h->param.i_threads = x264_clip3( h->param.i_threads, 1, X264_THREAD_MAX );
443     if( h->param.i_threads > 1 )
444     {
445 #if !HAVE_THREAD
446         x264_log( h, X264_LOG_WARNING, "not compiled with thread support!\n");
447         h->param.i_threads = 1;
448 #endif
449         /* Avoid absurdly small thread slices as they can reduce performance
450          * and VBV compliance.  Capped at an arbitrary 4 rows per thread. */
451         if( h->param.b_sliced_threads )
452         {
453             int max_threads = (h->param.i_height+15)/16 / 4;
454             h->param.i_threads = X264_MIN( h->param.i_threads, max_threads );
455         }
456     }
457     else
458         h->param.b_sliced_threads = 0;
459     h->i_thread_frames = h->param.b_sliced_threads ? 1 : h->param.i_threads;
460     if( h->i_thread_frames > 1 )
461         h->param.nalu_process = NULL;
462
463     h->param.i_keyint_max = x264_clip3( h->param.i_keyint_max, 1, X264_KEYINT_MAX_INFINITE );
464     if( h->param.i_keyint_max == 1 )
465     {
466         h->param.b_intra_refresh = 0;
467         h->param.analyse.i_weighted_pred = 0;
468     }
469
470     h->param.i_frame_packing = x264_clip3( h->param.i_frame_packing, -1, 5 );
471
472     /* Detect default ffmpeg settings and terminate with an error. */
473     if( b_open )
474     {
475         int score = 0;
476         score += h->param.analyse.i_me_range == 0;
477         score += h->param.rc.i_qp_step == 3;
478         score += h->param.i_keyint_max == 12;
479         score += h->param.rc.i_qp_min == 2;
480         score += h->param.rc.i_qp_max == 31;
481         score += h->param.rc.f_qcompress == 0.5;
482         score += fabs(h->param.rc.f_ip_factor - 1.25) < 0.01;
483         score += fabs(h->param.rc.f_pb_factor - 1.25) < 0.01;
484         score += h->param.analyse.inter == 0 && h->param.analyse.i_subpel_refine == 8;
485         if( score >= 5 )
486         {
487             x264_log( h, X264_LOG_ERROR, "broken ffmpeg default settings detected\n" );
488             x264_log( h, X264_LOG_ERROR, "use an encoding preset (e.g. -vpre medium)\n" );
489             x264_log( h, X264_LOG_ERROR, "preset usage: -vpre <speed> -vpre <profile>\n" );
490             x264_log( h, X264_LOG_ERROR, "speed presets are listed in x264 --help\n" );
491             x264_log( h, X264_LOG_ERROR, "profile is optional; x264 defaults to high\n" );
492             return -1;
493         }
494     }
495
496     if( h->param.rc.i_rc_method < 0 || h->param.rc.i_rc_method > 2 )
497     {
498         x264_log( h, X264_LOG_ERROR, "no ratecontrol method specified\n" );
499         return -1;
500     }
501     h->param.rc.f_rf_constant = x264_clip3f( h->param.rc.f_rf_constant, -QP_BD_OFFSET, 51 );
502     h->param.rc.f_rf_constant_max = x264_clip3f( h->param.rc.f_rf_constant_max, -QP_BD_OFFSET, 51 );
503     h->param.rc.i_qp_constant = x264_clip3( h->param.rc.i_qp_constant, 0, QP_MAX );
504     h->param.analyse.i_subpel_refine = x264_clip3( h->param.analyse.i_subpel_refine, 0, 10 );
505     h->param.rc.f_ip_factor = X264_MAX( h->param.rc.f_ip_factor, 0.01f );
506     h->param.rc.f_pb_factor = X264_MAX( h->param.rc.f_pb_factor, 0.01f );
507     if( h->param.rc.i_rc_method == X264_RC_CRF )
508     {
509         h->param.rc.i_qp_constant = h->param.rc.f_rf_constant + QP_BD_OFFSET;
510         h->param.rc.i_bitrate = 0;
511     }
512     if( (h->param.rc.i_rc_method == X264_RC_CQP || h->param.rc.i_rc_method == X264_RC_CRF)
513         && h->param.rc.i_qp_constant == 0 )
514     {
515         h->mb.b_lossless = 1;
516         h->param.i_cqm_preset = X264_CQM_FLAT;
517         h->param.psz_cqm_file = NULL;
518         h->param.rc.i_rc_method = X264_RC_CQP;
519         h->param.rc.f_ip_factor = 1;
520         h->param.rc.f_pb_factor = 1;
521         h->param.analyse.b_psnr = 0;
522         h->param.analyse.b_ssim = 0;
523         h->param.analyse.i_chroma_qp_offset = 0;
524         h->param.analyse.i_trellis = 0;
525         h->param.analyse.b_fast_pskip = 0;
526         h->param.analyse.i_noise_reduction = 0;
527         h->param.analyse.b_psy = 0;
528         h->param.i_bframe = 0;
529         /* 8x8dct is not useful without RD in CAVLC lossless */
530         if( !h->param.b_cabac && h->param.analyse.i_subpel_refine < 6 )
531             h->param.analyse.b_transform_8x8 = 0;
532     }
533     if( h->param.rc.i_rc_method == X264_RC_CQP )
534     {
535         float qp_p = h->param.rc.i_qp_constant;
536         float qp_i = qp_p - 6*log2f( h->param.rc.f_ip_factor );
537         float qp_b = qp_p + 6*log2f( h->param.rc.f_pb_factor );
538         h->param.rc.i_qp_min = x264_clip3( (int)(X264_MIN3( qp_p, qp_i, qp_b )), 0, QP_MAX );
539         h->param.rc.i_qp_max = x264_clip3( (int)(X264_MAX3( qp_p, qp_i, qp_b ) + .999), 0, QP_MAX );
540         h->param.rc.i_aq_mode = 0;
541         h->param.rc.b_mb_tree = 0;
542         h->param.rc.i_bitrate = 0;
543     }
544     h->param.rc.i_qp_max = x264_clip3( h->param.rc.i_qp_max, 0, QP_MAX );
545     h->param.rc.i_qp_min = x264_clip3( h->param.rc.i_qp_min, 0, h->param.rc.i_qp_max );
546     h->param.rc.i_qp_step = x264_clip3( h->param.rc.i_qp_step, 0, QP_MAX );
547     h->param.rc.i_bitrate = x264_clip3( h->param.rc.i_bitrate, 0, 2000000 );
548     h->param.rc.i_vbv_buffer_size = x264_clip3( h->param.rc.i_vbv_buffer_size, 0, 2000000 );
549     h->param.rc.i_vbv_max_bitrate = x264_clip3( h->param.rc.i_vbv_max_bitrate, 0, 2000000 );
550     h->param.rc.f_vbv_buffer_init = x264_clip3f( h->param.rc.f_vbv_buffer_init, 0, 2000000 );
551     if( h->param.rc.i_vbv_buffer_size )
552     {
553         if( h->param.rc.i_rc_method == X264_RC_CQP )
554         {
555             x264_log( h, X264_LOG_WARNING, "VBV is incompatible with constant QP, ignored.\n" );
556             h->param.rc.i_vbv_max_bitrate = 0;
557             h->param.rc.i_vbv_buffer_size = 0;
558         }
559         else if( h->param.rc.i_vbv_max_bitrate == 0 )
560         {
561             if( h->param.rc.i_rc_method == X264_RC_ABR )
562             {
563                 x264_log( h, X264_LOG_WARNING, "VBV maxrate unspecified, assuming CBR\n" );
564                 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate;
565             }
566             else
567             {
568                 x264_log( h, X264_LOG_WARNING, "VBV bufsize set but maxrate unspecified, ignored\n" );
569                 h->param.rc.i_vbv_buffer_size = 0;
570             }
571         }
572         else if( h->param.rc.i_vbv_max_bitrate < h->param.rc.i_bitrate &&
573                  h->param.rc.i_rc_method == X264_RC_ABR )
574         {
575             x264_log( h, X264_LOG_WARNING, "max bitrate less than average bitrate, assuming CBR\n" );
576             h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate;
577         }
578     }
579     else if( h->param.rc.i_vbv_max_bitrate )
580     {
581         x264_log( h, X264_LOG_WARNING, "VBV maxrate specified, but no bufsize, ignored\n" );
582         h->param.rc.i_vbv_max_bitrate = 0;
583     }
584
585     h->param.i_slice_max_size = X264_MAX( h->param.i_slice_max_size, 0 );
586     h->param.i_slice_max_mbs = X264_MAX( h->param.i_slice_max_mbs, 0 );
587
588     int max_slices = (h->param.i_height+((16<<PARAM_INTERLACED)-1))/(16<<PARAM_INTERLACED);
589     if( h->param.b_sliced_threads )
590         h->param.i_slice_count = x264_clip3( h->param.i_threads, 0, max_slices );
591     else
592     {
593         h->param.i_slice_count = x264_clip3( h->param.i_slice_count, 0, max_slices );
594         if( h->param.i_slice_max_mbs || h->param.i_slice_max_size )
595             h->param.i_slice_count = 0;
596     }
597
598     if( h->param.b_bluray_compat )
599     {
600         h->param.i_bframe_pyramid = X264_MIN( X264_B_PYRAMID_STRICT, h->param.i_bframe_pyramid );
601         h->param.i_bframe = X264_MIN( h->param.i_bframe, 3 );
602         h->param.b_aud = 1;
603         h->param.i_nal_hrd = X264_MAX( h->param.i_nal_hrd, X264_NAL_HRD_VBR );
604         h->param.i_slice_max_size = 0;
605         h->param.i_slice_max_mbs = 0;
606         h->param.b_intra_refresh = 0;
607         h->param.i_frame_reference = X264_MIN( h->param.i_frame_reference, 6 );
608         h->param.i_dpb_size = X264_MIN( h->param.i_dpb_size, 6 );
609         /* Due to the proliferation of broken players that don't handle dupes properly. */
610         h->param.analyse.i_weighted_pred = X264_MIN( h->param.analyse.i_weighted_pred, X264_WEIGHTP_SIMPLE );
611         if( h->param.b_fake_interlaced )
612             h->param.b_pic_struct = 1;
613     }
614
615     h->param.i_frame_reference = x264_clip3( h->param.i_frame_reference, 1, X264_REF_MAX );
616     h->param.i_dpb_size = x264_clip3( h->param.i_dpb_size, 1, X264_REF_MAX );
617     if( h->param.i_scenecut_threshold < 0 )
618         h->param.i_scenecut_threshold = 0;
619     h->param.analyse.i_direct_mv_pred = x264_clip3( h->param.analyse.i_direct_mv_pred, X264_DIRECT_PRED_NONE, X264_DIRECT_PRED_AUTO );
620     if( !h->param.analyse.i_subpel_refine && h->param.analyse.i_direct_mv_pred > X264_DIRECT_PRED_SPATIAL )
621     {
622         x264_log( h, X264_LOG_WARNING, "subme=0 + direct=temporal is not supported\n" );
623         h->param.analyse.i_direct_mv_pred = X264_DIRECT_PRED_SPATIAL;
624     }
625     h->param.i_bframe = x264_clip3( h->param.i_bframe, 0, X264_MIN( X264_BFRAME_MAX, h->param.i_keyint_max-1 ) );
626     h->param.i_bframe_bias = x264_clip3( h->param.i_bframe_bias, -90, 100 );
627     if( h->param.i_bframe <= 1 )
628         h->param.i_bframe_pyramid = X264_B_PYRAMID_NONE;
629     h->param.i_bframe_pyramid = x264_clip3( h->param.i_bframe_pyramid, X264_B_PYRAMID_NONE, X264_B_PYRAMID_NORMAL );
630     h->param.i_bframe_adaptive = x264_clip3( h->param.i_bframe_adaptive, X264_B_ADAPT_NONE, X264_B_ADAPT_TRELLIS );
631     if( !h->param.i_bframe )
632     {
633         h->param.i_bframe_adaptive = X264_B_ADAPT_NONE;
634         h->param.analyse.i_direct_mv_pred = 0;
635         h->param.analyse.b_weighted_bipred = 0;
636         h->param.b_open_gop = 0;
637     }
638     if( h->param.b_intra_refresh && h->param.i_bframe_pyramid == X264_B_PYRAMID_NORMAL )
639     {
640         x264_log( h, X264_LOG_WARNING, "b-pyramid normal + intra-refresh is not supported\n" );
641         h->param.i_bframe_pyramid = X264_B_PYRAMID_STRICT;
642     }
643     if( h->param.b_intra_refresh && (h->param.i_frame_reference > 1 || h->param.i_dpb_size > 1) )
644     {
645         x264_log( h, X264_LOG_WARNING, "ref > 1 + intra-refresh is not supported\n" );
646         h->param.i_frame_reference = 1;
647         h->param.i_dpb_size = 1;
648     }
649     if( h->param.b_intra_refresh && h->param.b_open_gop )
650     {
651         x264_log( h, X264_LOG_WARNING, "intra-refresh is not compatible with open-gop\n" );
652         h->param.b_open_gop = 0;
653     }
654     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;
655     if( h->param.i_keyint_min == X264_KEYINT_MIN_AUTO )
656         h->param.i_keyint_min = X264_MIN( h->param.i_keyint_max / 10, fps );
657     h->param.i_keyint_min = x264_clip3( h->param.i_keyint_min, 1, h->param.i_keyint_max/2+1 );
658     h->param.rc.i_lookahead = x264_clip3( h->param.rc.i_lookahead, 0, X264_LOOKAHEAD_MAX );
659     {
660         int maxrate = X264_MAX( h->param.rc.i_vbv_max_bitrate, h->param.rc.i_bitrate );
661         float bufsize = maxrate ? (float)h->param.rc.i_vbv_buffer_size / maxrate : 0;
662         h->param.rc.i_lookahead = X264_MIN( h->param.rc.i_lookahead, X264_MAX( h->param.i_keyint_max, bufsize*fps ) );
663     }
664
665     if( !h->param.i_timebase_num || !h->param.i_timebase_den || !(h->param.b_vfr_input || h->param.b_pulldown) )
666     {
667         h->param.i_timebase_num = h->param.i_fps_den;
668         h->param.i_timebase_den = h->param.i_fps_num;
669     }
670
671     h->param.rc.f_qcompress = x264_clip3f( h->param.rc.f_qcompress, 0.0, 1.0 );
672     if( h->param.i_keyint_max == 1 || h->param.rc.f_qcompress == 1 )
673         h->param.rc.b_mb_tree = 0;
674     if( (!h->param.b_intra_refresh && h->param.i_keyint_max != X264_KEYINT_MAX_INFINITE) &&
675         !h->param.rc.i_lookahead && h->param.rc.b_mb_tree )
676     {
677         x264_log( h, X264_LOG_WARNING, "lookaheadless mb-tree requires intra refresh or infinite keyint\n" );
678         h->param.rc.b_mb_tree = 0;
679     }
680     if( h->param.rc.b_stat_read )
681         h->param.rc.i_lookahead = 0;
682 #if HAVE_THREAD
683     if( h->param.i_sync_lookahead < 0 )
684         h->param.i_sync_lookahead = h->param.i_bframe + 1;
685     h->param.i_sync_lookahead = X264_MIN( h->param.i_sync_lookahead, X264_LOOKAHEAD_MAX );
686     if( h->param.rc.b_stat_read || h->i_thread_frames == 1 )
687         h->param.i_sync_lookahead = 0;
688 #else
689     h->param.i_sync_lookahead = 0;
690 #endif
691
692     h->param.i_deblocking_filter_alphac0 = x264_clip3( h->param.i_deblocking_filter_alphac0, -6, 6 );
693     h->param.i_deblocking_filter_beta    = x264_clip3( h->param.i_deblocking_filter_beta, -6, 6 );
694     h->param.analyse.i_luma_deadzone[0] = x264_clip3( h->param.analyse.i_luma_deadzone[0], 0, 32 );
695     h->param.analyse.i_luma_deadzone[1] = x264_clip3( h->param.analyse.i_luma_deadzone[1], 0, 32 );
696
697     h->param.i_cabac_init_idc = x264_clip3( h->param.i_cabac_init_idc, 0, 2 );
698
699     if( h->param.i_cqm_preset < X264_CQM_FLAT || h->param.i_cqm_preset > X264_CQM_CUSTOM )
700         h->param.i_cqm_preset = X264_CQM_FLAT;
701
702     if( h->param.analyse.i_me_method < X264_ME_DIA ||
703         h->param.analyse.i_me_method > X264_ME_TESA )
704         h->param.analyse.i_me_method = X264_ME_HEX;
705     h->param.analyse.i_me_range = x264_clip3( h->param.analyse.i_me_range, 4, 1024 );
706     if( h->param.analyse.i_me_range > 16 && h->param.analyse.i_me_method <= X264_ME_HEX )
707         h->param.analyse.i_me_range = 16;
708     if( h->param.analyse.i_me_method == X264_ME_TESA &&
709         (h->mb.b_lossless || h->param.analyse.i_subpel_refine <= 1) )
710         h->param.analyse.i_me_method = X264_ME_ESA;
711     h->param.analyse.b_mixed_references = h->param.analyse.b_mixed_references && h->param.i_frame_reference > 1;
712     h->param.analyse.inter &= X264_ANALYSE_PSUB16x16|X264_ANALYSE_PSUB8x8|X264_ANALYSE_BSUB16x16|
713                               X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
714     h->param.analyse.intra &= X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
715     if( !(h->param.analyse.inter & X264_ANALYSE_PSUB16x16) )
716         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
717     if( !h->param.analyse.b_transform_8x8 )
718     {
719         h->param.analyse.inter &= ~X264_ANALYSE_I8x8;
720         h->param.analyse.intra &= ~X264_ANALYSE_I8x8;
721     }
722     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -12, 12);
723     h->param.analyse.i_trellis = x264_clip3( h->param.analyse.i_trellis, 0, 2 );
724     h->param.rc.i_aq_mode = x264_clip3( h->param.rc.i_aq_mode, 0, 2 );
725     h->param.rc.f_aq_strength = x264_clip3f( h->param.rc.f_aq_strength, 0, 3 );
726     if( h->param.rc.f_aq_strength == 0 )
727         h->param.rc.i_aq_mode = 0;
728
729     if( h->param.i_log_level < X264_LOG_INFO )
730     {
731         h->param.analyse.b_psnr = 0;
732         h->param.analyse.b_ssim = 0;
733     }
734     /* Warn users trying to measure PSNR/SSIM with psy opts on. */
735     if( b_open && (h->param.analyse.b_psnr || h->param.analyse.b_ssim) )
736     {
737         char *s = NULL;
738
739         if( h->param.analyse.b_psy )
740         {
741             s = h->param.analyse.b_psnr ? "psnr" : "ssim";
742             x264_log( h, X264_LOG_WARNING, "--%s used with psy on: results will be invalid!\n", s );
743         }
744         else if( !h->param.rc.i_aq_mode && h->param.analyse.b_ssim )
745         {
746             x264_log( h, X264_LOG_WARNING, "--ssim used with AQ off: results will be invalid!\n" );
747             s = "ssim";
748         }
749         else if(  h->param.rc.i_aq_mode && h->param.analyse.b_psnr )
750         {
751             x264_log( h, X264_LOG_WARNING, "--psnr used with AQ on: results will be invalid!\n" );
752             s = "psnr";
753         }
754         if( s )
755             x264_log( h, X264_LOG_WARNING, "--tune %s should be used if attempting to benchmark %s!\n", s, s );
756     }
757
758     if( !h->param.analyse.b_psy )
759     {
760         h->param.analyse.f_psy_rd = 0;
761         h->param.analyse.f_psy_trellis = 0;
762     }
763     h->param.analyse.f_psy_rd = x264_clip3f( h->param.analyse.f_psy_rd, 0, 10 );
764     h->param.analyse.f_psy_trellis = x264_clip3f( h->param.analyse.f_psy_trellis, 0, 10 );
765     h->mb.i_psy_rd = h->param.analyse.i_subpel_refine >= 6 ? FIX8( h->param.analyse.f_psy_rd ) : 0;
766     h->mb.i_psy_trellis = h->param.analyse.i_trellis ? FIX8( h->param.analyse.f_psy_trellis / 4 ) : 0;
767     /* Psy RDO increases overall quantizers to improve the quality of luma--this indirectly hurts chroma quality */
768     /* so we lower the chroma QP offset to compensate */
769     if( b_open && h->mb.i_psy_rd )
770         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_rd < 0.25 ? 1 : 2;
771     /* Psy trellis has a similar effect. */
772     if( b_open && h->mb.i_psy_trellis )
773         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_trellis < 0.25 ? 1 : 2;
774     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -12, 12);
775     /* MB-tree requires AQ to be on, even if the strength is zero. */
776     if( !h->param.rc.i_aq_mode && h->param.rc.b_mb_tree )
777     {
778         h->param.rc.i_aq_mode = 1;
779         h->param.rc.f_aq_strength = 0;
780     }
781     h->param.analyse.i_noise_reduction = x264_clip3( h->param.analyse.i_noise_reduction, 0, 1<<16 );
782     if( h->param.analyse.i_subpel_refine == 10 && (h->param.analyse.i_trellis != 2 || !h->param.rc.i_aq_mode) )
783         h->param.analyse.i_subpel_refine = 9;
784
785     {
786         const x264_level_t *l = x264_levels;
787         if( h->param.i_level_idc < 0 )
788         {
789             int maxrate_bak = h->param.rc.i_vbv_max_bitrate;
790             if( h->param.rc.i_rc_method == X264_RC_ABR && h->param.rc.i_vbv_buffer_size <= 0 )
791                 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate * 2;
792             h->sps = h->sps_array;
793             x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
794             do h->param.i_level_idc = l->level_idc;
795                 while( l[1].level_idc && x264_validate_levels( h, 0 ) && l++ );
796             h->param.rc.i_vbv_max_bitrate = maxrate_bak;
797         }
798         else
799         {
800             while( l->level_idc && l->level_idc != h->param.i_level_idc )
801                 l++;
802             if( l->level_idc == 0 )
803             {
804                 x264_log( h, X264_LOG_ERROR, "invalid level_idc: %d\n", h->param.i_level_idc );
805                 return -1;
806             }
807         }
808         if( h->param.analyse.i_mv_range <= 0 )
809             h->param.analyse.i_mv_range = l->mv_range >> PARAM_INTERLACED;
810         else
811             h->param.analyse.i_mv_range = x264_clip3(h->param.analyse.i_mv_range, 32, 512 >> PARAM_INTERLACED);
812     }
813
814     h->param.analyse.i_weighted_pred = x264_clip3( h->param.analyse.i_weighted_pred, X264_WEIGHTP_NONE, X264_WEIGHTP_SMART );
815
816     if( PARAM_INTERLACED )
817     {
818         if( h->param.analyse.i_me_method >= X264_ME_ESA )
819         {
820             x264_log( h, X264_LOG_WARNING, "interlace + me=esa is not implemented\n" );
821             h->param.analyse.i_me_method = X264_ME_UMH;
822         }
823         if( h->param.analyse.i_weighted_pred > 0 )
824         {
825             x264_log( h, X264_LOG_WARNING, "interlace + weightp is not implemented\n" );
826             h->param.analyse.i_weighted_pred = X264_WEIGHTP_NONE;
827         }
828     }
829
830     if( !h->param.analyse.i_weighted_pred && h->param.rc.b_mb_tree && h->param.analyse.b_psy )
831         h->param.analyse.i_weighted_pred = X264_WEIGHTP_FAKE;
832
833     if( h->i_thread_frames > 1 )
834     {
835         int r = h->param.analyse.i_mv_range_thread;
836         int r2;
837         if( r <= 0 )
838         {
839             // half of the available space is reserved and divided evenly among the threads,
840             // the rest is allocated to whichever thread is far enough ahead to use it.
841             // reserving more space increases quality for some videos, but costs more time
842             // in thread synchronization.
843             int max_range = (h->param.i_height + X264_THREAD_HEIGHT) / h->i_thread_frames - X264_THREAD_HEIGHT;
844             r = max_range / 2;
845         }
846         r = X264_MAX( r, h->param.analyse.i_me_range );
847         r = X264_MIN( r, h->param.analyse.i_mv_range );
848         // round up to use the whole mb row
849         r2 = (r & ~15) + ((-X264_THREAD_HEIGHT) & 15);
850         if( r2 < r )
851             r2 += 16;
852         x264_log( h, X264_LOG_DEBUG, "using mv_range_thread = %d\n", r2 );
853         h->param.analyse.i_mv_range_thread = r2;
854     }
855
856     if( h->param.rc.f_rate_tolerance < 0 )
857         h->param.rc.f_rate_tolerance = 0;
858     if( h->param.rc.f_qblur < 0 )
859         h->param.rc.f_qblur = 0;
860     if( h->param.rc.f_complexity_blur < 0 )
861         h->param.rc.f_complexity_blur = 0;
862
863     h->param.i_sps_id &= 31;
864
865     if( PARAM_INTERLACED )
866         h->param.b_pic_struct = 1;
867
868     h->param.i_nal_hrd = x264_clip3( h->param.i_nal_hrd, X264_NAL_HRD_NONE, X264_NAL_HRD_CBR );
869
870     if( h->param.i_nal_hrd && !h->param.rc.i_vbv_buffer_size )
871     {
872         x264_log( h, X264_LOG_WARNING, "NAL HRD parameters require VBV parameters\n" );
873         h->param.i_nal_hrd = X264_NAL_HRD_NONE;
874     }
875
876     if( h->param.i_nal_hrd == X264_NAL_HRD_CBR &&
877        (h->param.rc.i_bitrate != h->param.rc.i_vbv_max_bitrate || !h->param.rc.i_vbv_max_bitrate) )
878     {
879         x264_log( h, X264_LOG_WARNING, "CBR HRD requires constant bitrate\n" );
880         h->param.i_nal_hrd = X264_NAL_HRD_VBR;
881     }
882
883     /* ensure the booleans are 0 or 1 so they can be used in math */
884 #define BOOLIFY(x) h->param.x = !!h->param.x
885     BOOLIFY( b_cabac );
886     BOOLIFY( b_constrained_intra );
887     BOOLIFY( b_deblocking_filter );
888     BOOLIFY( b_deterministic );
889     BOOLIFY( b_sliced_threads );
890     BOOLIFY( b_interlaced );
891     BOOLIFY( b_intra_refresh );
892     BOOLIFY( b_visualize );
893     BOOLIFY( b_aud );
894     BOOLIFY( b_repeat_headers );
895     BOOLIFY( b_annexb );
896     BOOLIFY( b_vfr_input );
897     BOOLIFY( b_pulldown );
898     BOOLIFY( b_tff );
899     BOOLIFY( b_pic_struct );
900     BOOLIFY( b_fake_interlaced );
901     BOOLIFY( b_open_gop );
902     BOOLIFY( b_bluray_compat );
903     BOOLIFY( analyse.b_transform_8x8 );
904     BOOLIFY( analyse.b_weighted_bipred );
905     BOOLIFY( analyse.b_chroma_me );
906     BOOLIFY( analyse.b_mixed_references );
907     BOOLIFY( analyse.b_fast_pskip );
908     BOOLIFY( analyse.b_dct_decimate );
909     BOOLIFY( analyse.b_psy );
910     BOOLIFY( analyse.b_psnr );
911     BOOLIFY( analyse.b_ssim );
912     BOOLIFY( rc.b_stat_write );
913     BOOLIFY( rc.b_stat_read );
914     BOOLIFY( rc.b_mb_tree );
915 #undef BOOLIFY
916
917     return 0;
918 }
919
920 static void mbcmp_init( x264_t *h )
921 {
922     int satd = !h->mb.b_lossless && h->param.analyse.i_subpel_refine > 1;
923     memcpy( h->pixf.mbcmp, satd ? h->pixf.satd : h->pixf.sad_aligned, sizeof(h->pixf.mbcmp) );
924     memcpy( h->pixf.mbcmp_unaligned, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.mbcmp_unaligned) );
925     h->pixf.intra_mbcmp_x3_16x16 = satd ? h->pixf.intra_satd_x3_16x16 : h->pixf.intra_sad_x3_16x16;
926     h->pixf.intra_mbcmp_x3_8x8c = satd ? h->pixf.intra_satd_x3_8x8c : h->pixf.intra_sad_x3_8x8c;
927     h->pixf.intra_mbcmp_x3_8x8 = satd ? h->pixf.intra_sa8d_x3_8x8 : h->pixf.intra_sad_x3_8x8;
928     h->pixf.intra_mbcmp_x3_4x4 = satd ? h->pixf.intra_satd_x3_4x4 : h->pixf.intra_sad_x3_4x4;
929     satd &= h->param.analyse.i_me_method == X264_ME_TESA;
930     memcpy( h->pixf.fpelcmp, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.fpelcmp) );
931     memcpy( h->pixf.fpelcmp_x3, satd ? h->pixf.satd_x3 : h->pixf.sad_x3, sizeof(h->pixf.fpelcmp_x3) );
932     memcpy( h->pixf.fpelcmp_x4, satd ? h->pixf.satd_x4 : h->pixf.sad_x4, sizeof(h->pixf.fpelcmp_x4) );
933 }
934
935 static void x264_set_aspect_ratio( x264_t *h, x264_param_t *param, int initial )
936 {
937     /* VUI */
938     if( param->vui.i_sar_width > 0 && param->vui.i_sar_height > 0 )
939     {
940         uint32_t i_w = param->vui.i_sar_width;
941         uint32_t i_h = param->vui.i_sar_height;
942         uint32_t old_w = h->param.vui.i_sar_width;
943         uint32_t old_h = h->param.vui.i_sar_height;
944
945         x264_reduce_fraction( &i_w, &i_h );
946
947         while( i_w > 65535 || i_h > 65535 )
948         {
949             i_w /= 2;
950             i_h /= 2;
951         }
952
953         x264_reduce_fraction( &i_w, &i_h );
954
955         if( i_w != old_w || i_h != old_h || initial )
956         {
957             h->param.vui.i_sar_width = 0;
958             h->param.vui.i_sar_height = 0;
959             if( i_w == 0 || i_h == 0 )
960                 x264_log( h, X264_LOG_WARNING, "cannot create valid sample aspect ratio\n" );
961             else
962             {
963                 x264_log( h, initial?X264_LOG_INFO:X264_LOG_DEBUG, "using SAR=%d/%d\n", i_w, i_h );
964                 h->param.vui.i_sar_width = i_w;
965                 h->param.vui.i_sar_height = i_h;
966             }
967             x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
968         }
969     }
970 }
971
972 /****************************************************************************
973  * x264_encoder_open:
974  ****************************************************************************/
975 x264_t *x264_encoder_open( x264_param_t *param )
976 {
977     x264_t *h;
978     char buf[1000], *p;
979     int qp, i_slicetype_length;
980
981     CHECKED_MALLOCZERO( h, sizeof(x264_t) );
982
983     /* Create a copy of param */
984     memcpy( &h->param, param, sizeof(x264_param_t) );
985
986     if( param->param_free )
987         param->param_free( param );
988
989     if( x264_threading_init() )
990     {
991         x264_log( h, X264_LOG_ERROR, "unable to initialize threading\n" );
992         goto fail;
993     }
994
995     if( x264_validate_parameters( h, 1 ) < 0 )
996         goto fail;
997
998     if( h->param.psz_cqm_file )
999         if( x264_cqm_parse_file( h, h->param.psz_cqm_file ) < 0 )
1000             goto fail;
1001
1002     if( h->param.rc.psz_stat_out )
1003         h->param.rc.psz_stat_out = strdup( h->param.rc.psz_stat_out );
1004     if( h->param.rc.psz_stat_in )
1005         h->param.rc.psz_stat_in = strdup( h->param.rc.psz_stat_in );
1006
1007     x264_reduce_fraction( &h->param.i_fps_num, &h->param.i_fps_den );
1008     x264_reduce_fraction( &h->param.i_timebase_num, &h->param.i_timebase_den );
1009
1010     /* Init x264_t */
1011     h->i_frame = -1;
1012     h->i_frame_num = 0;
1013     h->i_idr_pic_id = 0;
1014
1015     if( (uint64_t)h->param.i_timebase_den * 2 > UINT32_MAX )
1016     {
1017         x264_log( h, X264_LOG_ERROR, "Effective timebase denominator %u exceeds H.264 maximum\n", h->param.i_timebase_den );
1018         goto fail;
1019     }
1020
1021     h->sps = &h->sps_array[0];
1022     x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
1023
1024     h->pps = &h->pps_array[0];
1025     x264_pps_init( h->pps, h->param.i_sps_id, &h->param, h->sps );
1026
1027     x264_set_aspect_ratio( h, &h->param, 1 );
1028
1029     x264_validate_levels( h, 1 );
1030
1031     h->chroma_qp_table = i_chroma_qp_table + 12 + h->pps->i_chroma_qp_index_offset;
1032
1033     if( x264_cqm_init( h ) < 0 )
1034         goto fail;
1035
1036     h->mb.i_mb_width = h->sps->i_mb_width;
1037     h->mb.i_mb_height = h->sps->i_mb_height;
1038     h->mb.i_mb_count = h->mb.i_mb_width * h->mb.i_mb_height;
1039     /* Adaptive MBAFF and subme 0 are not supported as we require halving motion
1040      * vectors during prediction, resulting in hpel mvs.
1041      * The chosen solution is to make MBAFF non-adaptive in this case. */
1042     h->mb.b_adaptive_mbaff = PARAM_INTERLACED && h->param.analyse.i_subpel_refine;
1043
1044     /* Init frames. */
1045     if( h->param.i_bframe_adaptive == X264_B_ADAPT_TRELLIS && !h->param.rc.b_stat_read )
1046         h->frames.i_delay = X264_MAX(h->param.i_bframe,3)*4;
1047     else
1048         h->frames.i_delay = h->param.i_bframe;
1049     if( h->param.rc.b_mb_tree || h->param.rc.i_vbv_buffer_size )
1050         h->frames.i_delay = X264_MAX( h->frames.i_delay, h->param.rc.i_lookahead );
1051     i_slicetype_length = h->frames.i_delay;
1052     h->frames.i_delay += h->i_thread_frames - 1;
1053     h->frames.i_delay += h->param.i_sync_lookahead;
1054     h->frames.i_delay += h->param.b_vfr_input;
1055     h->frames.i_bframe_delay = h->param.i_bframe ? (h->param.i_bframe_pyramid ? 2 : 1) : 0;
1056
1057     h->frames.i_max_ref0 = h->param.i_frame_reference;
1058     h->frames.i_max_ref1 = X264_MIN( h->sps->vui.i_num_reorder_frames, h->param.i_frame_reference );
1059     h->frames.i_max_dpb  = h->sps->vui.i_max_dec_frame_buffering;
1060     h->frames.b_have_lowres = !h->param.rc.b_stat_read
1061         && ( h->param.rc.i_rc_method == X264_RC_ABR
1062           || h->param.rc.i_rc_method == X264_RC_CRF
1063           || h->param.i_bframe_adaptive
1064           || h->param.i_scenecut_threshold
1065           || h->param.rc.b_mb_tree
1066           || h->param.analyse.i_weighted_pred );
1067     h->frames.b_have_lowres |= h->param.rc.b_stat_read && h->param.rc.i_vbv_buffer_size > 0;
1068     h->frames.b_have_sub8x8_esa = !!(h->param.analyse.inter & X264_ANALYSE_PSUB8x8);
1069
1070     h->frames.i_last_idr =
1071     h->frames.i_last_keyframe = - h->param.i_keyint_max;
1072     h->frames.i_input    = 0;
1073     h->frames.i_largest_pts = h->frames.i_second_largest_pts = -1;
1074     h->frames.i_poc_last_open_gop = -1;
1075
1076     CHECKED_MALLOCZERO( h->frames.unused[0], (h->frames.i_delay + 3) * sizeof(x264_frame_t *) );
1077     /* Allocate room for max refs plus a few extra just in case. */
1078     CHECKED_MALLOCZERO( h->frames.unused[1], (h->i_thread_frames + X264_REF_MAX + 4) * sizeof(x264_frame_t *) );
1079     CHECKED_MALLOCZERO( h->frames.current, (h->param.i_sync_lookahead + h->param.i_bframe
1080                         + h->i_thread_frames + 3) * sizeof(x264_frame_t *) );
1081     if( h->param.analyse.i_weighted_pred > 0 )
1082         CHECKED_MALLOCZERO( h->frames.blank_unused, h->i_thread_frames * 4 * sizeof(x264_frame_t *) );
1083     h->i_ref[0] = h->i_ref[1] = 0;
1084     h->i_cpb_delay = h->i_coded_fields = h->i_disp_fields = 0;
1085     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);
1086     h->i_disp_fields_last_frame = -1;
1087     x264_rdo_init();
1088
1089     /* init CPU functions */
1090     x264_predict_16x16_init( h->param.cpu, h->predict_16x16 );
1091     x264_predict_8x8c_init( h->param.cpu, h->predict_8x8c );
1092     x264_predict_8x8_init( h->param.cpu, h->predict_8x8, &h->predict_8x8_filter );
1093     x264_predict_4x4_init( h->param.cpu, h->predict_4x4 );
1094     if( h->param.b_cabac )
1095         x264_cabac_init();
1096     else
1097         x264_cavlc_init();
1098     x264_pixel_init( h->param.cpu, &h->pixf );
1099     x264_dct_init( h->param.cpu, &h->dctf );
1100     x264_zigzag_init( h->param.cpu, &h->zigzagf_progressive, &h->zigzagf_interlaced );
1101     memcpy( &h->zigzagf, PARAM_INTERLACED ? &h->zigzagf_interlaced : &h->zigzagf_progressive, sizeof(h->zigzagf) );
1102     x264_mc_init( h->param.cpu, &h->mc );
1103     x264_quant_init( h, h->param.cpu, &h->quantf );
1104     x264_deblock_init( h->param.cpu, &h->loopf, PARAM_INTERLACED );
1105     x264_bitstream_init( h->param.cpu, &h->bsf );
1106     x264_dct_init_weights();
1107
1108     mbcmp_init( h );
1109
1110     p = buf + sprintf( buf, "using cpu capabilities:" );
1111     for( int i = 0; x264_cpu_names[i].flags; i++ )
1112     {
1113         if( !strcmp(x264_cpu_names[i].name, "SSE2")
1114             && h->param.cpu & (X264_CPU_SSE2_IS_FAST|X264_CPU_SSE2_IS_SLOW) )
1115             continue;
1116         if( !strcmp(x264_cpu_names[i].name, "SSE3")
1117             && (h->param.cpu & X264_CPU_SSSE3 || !(h->param.cpu & X264_CPU_CACHELINE_64)) )
1118             continue;
1119         if( !strcmp(x264_cpu_names[i].name, "SSE4.1")
1120             && (h->param.cpu & X264_CPU_SSE42) )
1121             continue;
1122         if( (h->param.cpu & x264_cpu_names[i].flags) == x264_cpu_names[i].flags
1123             && (!i || x264_cpu_names[i].flags != x264_cpu_names[i-1].flags) )
1124             p += sprintf( p, " %s", x264_cpu_names[i].name );
1125     }
1126     if( !h->param.cpu )
1127         p += sprintf( p, " none!" );
1128     x264_log( h, X264_LOG_INFO, "%s\n", buf );
1129
1130     float *logs = x264_analyse_prepare_costs( h );
1131     if( !logs )
1132         goto fail;
1133     for( qp = X264_MIN( h->param.rc.i_qp_min, QP_MAX_SPEC ); qp <= h->param.rc.i_qp_max; qp++ )
1134         if( x264_analyse_init_costs( h, logs, qp ) )
1135             goto fail;
1136     if( x264_analyse_init_costs( h, logs, X264_LOOKAHEAD_QP ) )
1137         goto fail;
1138     x264_free( logs );
1139
1140     static const uint16_t cost_mv_correct[7] = { 24, 47, 95, 189, 379, 757, 1515 };
1141     /* Checks for known miscompilation issues. */
1142     if( h->cost_mv[X264_LOOKAHEAD_QP][2013] != cost_mv_correct[BIT_DEPTH-8] )
1143     {
1144         x264_log( h, X264_LOG_ERROR, "MV cost test failed: x264 has been miscompiled!\n" );
1145         goto fail;
1146     }
1147
1148     /* Must be volatile or else GCC will optimize it out. */
1149     volatile int temp = 392;
1150     if( x264_clz( temp ) != 23 )
1151     {
1152         x264_log( h, X264_LOG_ERROR, "CLZ test failed: x264 has been miscompiled!\n" );
1153 #if ARCH_X86 || ARCH_X86_64
1154         x264_log( h, X264_LOG_ERROR, "Are you attempting to run an SSE4a-targeted build on a CPU that\n" );
1155         x264_log( h, X264_LOG_ERROR, "doesn't support it?\n" );
1156 #endif
1157         goto fail;
1158     }
1159
1160     h->out.i_nal = 0;
1161     h->out.i_bitstream = X264_MAX( 1000000, h->param.i_width * h->param.i_height * 4
1162         * ( h->param.rc.i_rc_method == X264_RC_ABR ? pow( 0.95, h->param.rc.i_qp_min )
1163           : pow( 0.95, h->param.rc.i_qp_constant ) * X264_MAX( 1, h->param.rc.f_ip_factor )));
1164
1165     h->nal_buffer_size = h->out.i_bitstream * 3/2 + 4;
1166     CHECKED_MALLOC( h->nal_buffer, h->nal_buffer_size );
1167
1168     if( h->param.i_threads > 1 &&
1169         x264_threadpool_init( &h->threadpool, h->param.i_threads, (void*)x264_encoder_thread_init, h ) )
1170         goto fail;
1171
1172     h->thread[0] = h;
1173     for( int i = 1; i < h->param.i_threads + !!h->param.i_sync_lookahead; i++ )
1174         CHECKED_MALLOC( h->thread[i], sizeof(x264_t) );
1175
1176     for( int i = 0; i < h->param.i_threads; i++ )
1177     {
1178         int init_nal_count = h->param.i_slice_count + 3;
1179         int allocate_threadlocal_data = !h->param.b_sliced_threads || !i;
1180         if( i > 0 )
1181             *h->thread[i] = *h;
1182
1183         if( allocate_threadlocal_data )
1184         {
1185             h->thread[i]->fdec = x264_frame_pop_unused( h, 1 );
1186             if( !h->thread[i]->fdec )
1187                 goto fail;
1188         }
1189         else
1190             h->thread[i]->fdec = h->thread[0]->fdec;
1191
1192         h->thread[i]->sps = &h->thread[i]->sps_array[0];
1193         h->thread[i]->pps = &h->thread[i]->pps_array[0];
1194
1195         CHECKED_MALLOC( h->thread[i]->out.p_bitstream, h->out.i_bitstream );
1196         /* Start each thread with room for init_nal_count NAL units; it'll realloc later if needed. */
1197         CHECKED_MALLOC( h->thread[i]->out.nal, init_nal_count*sizeof(x264_nal_t) );
1198         h->thread[i]->out.i_nals_allocated = init_nal_count;
1199
1200         if( allocate_threadlocal_data && x264_macroblock_cache_allocate( h->thread[i] ) < 0 )
1201             goto fail;
1202     }
1203
1204     if( x264_lookahead_init( h, i_slicetype_length ) )
1205         goto fail;
1206
1207     for( int i = 0; i < h->param.i_threads; i++ )
1208         if( x264_macroblock_thread_allocate( h->thread[i], 0 ) < 0 )
1209             goto fail;
1210
1211     if( x264_ratecontrol_new( h ) < 0 )
1212         goto fail;
1213
1214     if( h->param.i_nal_hrd )
1215     {
1216         x264_log( h, X264_LOG_DEBUG, "HRD bitrate: %i bits/sec\n", h->sps->vui.hrd.i_bit_rate_unscaled );
1217         x264_log( h, X264_LOG_DEBUG, "CPB size: %i bits\n", h->sps->vui.hrd.i_cpb_size_unscaled );
1218     }
1219
1220     if( h->param.psz_dump_yuv )
1221     {
1222         /* create or truncate the reconstructed video file */
1223         FILE *f = fopen( h->param.psz_dump_yuv, "w" );
1224         if( !f )
1225         {
1226             x264_log( h, X264_LOG_ERROR, "dump_yuv: can't write to %s\n", h->param.psz_dump_yuv );
1227             goto fail;
1228         }
1229         else if( !x264_is_regular_file( f ) )
1230         {
1231             x264_log( h, X264_LOG_ERROR, "dump_yuv: incompatible with non-regular file %s\n", h->param.psz_dump_yuv );
1232             goto fail;
1233         }
1234         fclose( f );
1235     }
1236
1237     const char *profile = h->sps->i_profile_idc == PROFILE_BASELINE ? "Constrained Baseline" :
1238                           h->sps->i_profile_idc == PROFILE_MAIN ? "Main" :
1239                           h->sps->i_profile_idc == PROFILE_HIGH ? "High" :
1240                           h->sps->i_profile_idc == PROFILE_HIGH10 ? (h->sps->b_constraint_set3 == 1 ? "High 10 Intra" : "High 10") :
1241                           "High 4:4:4 Predictive";
1242     char level[4];
1243     snprintf( level, sizeof(level), "%d.%d", h->sps->i_level_idc/10, h->sps->i_level_idc%10 );
1244     if( h->sps->i_level_idc == 9 || ( h->sps->i_level_idc == 11 && h->sps->b_constraint_set3 ) )
1245         strcpy( level, "1b" );
1246
1247     if( h->sps->i_profile_idc < PROFILE_HIGH10 )
1248     {
1249         x264_log( h, X264_LOG_INFO, "profile %s, level %s\n",
1250             profile, level );
1251     }
1252     else
1253     {
1254         x264_log( h, X264_LOG_INFO, "profile %s, level %s, bit depth %d\n",
1255             profile, level, BIT_DEPTH );
1256     }
1257
1258     return h;
1259 fail:
1260     x264_free( h );
1261     return NULL;
1262 }
1263
1264 /****************************************************************************
1265  * x264_encoder_reconfig:
1266  ****************************************************************************/
1267 int x264_encoder_reconfig( x264_t *h, x264_param_t *param )
1268 {
1269     int rc_reconfig = 0;
1270     h = h->thread[h->thread[0]->i_thread_phase];
1271     x264_set_aspect_ratio( h, param, 0 );
1272 #define COPY(var) h->param.var = param->var
1273     COPY( i_frame_reference ); // but never uses more refs than initially specified
1274     COPY( i_bframe_bias );
1275     if( h->param.i_scenecut_threshold )
1276         COPY( i_scenecut_threshold ); // can't turn it on or off, only vary the threshold
1277     COPY( b_deblocking_filter );
1278     COPY( i_deblocking_filter_alphac0 );
1279     COPY( i_deblocking_filter_beta );
1280     COPY( i_frame_packing );
1281     COPY( analyse.inter );
1282     COPY( analyse.intra );
1283     COPY( analyse.i_direct_mv_pred );
1284     /* Scratch buffer prevents me_range from being increased for esa/tesa */
1285     if( h->param.analyse.i_me_method < X264_ME_ESA || param->analyse.i_me_range < h->param.analyse.i_me_range )
1286         COPY( analyse.i_me_range );
1287     COPY( analyse.i_noise_reduction );
1288     /* We can't switch out of subme=0 during encoding. */
1289     if( h->param.analyse.i_subpel_refine )
1290         COPY( analyse.i_subpel_refine );
1291     COPY( analyse.i_trellis );
1292     COPY( analyse.b_chroma_me );
1293     COPY( analyse.b_dct_decimate );
1294     COPY( analyse.b_fast_pskip );
1295     COPY( analyse.b_mixed_references );
1296     COPY( analyse.f_psy_rd );
1297     COPY( analyse.f_psy_trellis );
1298     COPY( crop_rect );
1299     // can only twiddle these if they were enabled to begin with:
1300     if( h->param.analyse.i_me_method >= X264_ME_ESA || param->analyse.i_me_method < X264_ME_ESA )
1301         COPY( analyse.i_me_method );
1302     if( h->param.analyse.i_me_method >= X264_ME_ESA && !h->frames.b_have_sub8x8_esa )
1303         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
1304     if( h->pps->b_transform_8x8_mode )
1305         COPY( analyse.b_transform_8x8 );
1306     if( h->frames.i_max_ref1 > 1 )
1307         COPY( i_bframe_pyramid );
1308     COPY( i_slice_max_size );
1309     COPY( i_slice_max_mbs );
1310     COPY( i_slice_count );
1311     COPY( b_tff );
1312
1313     /* VBV can't be turned on if it wasn't on to begin with */
1314     if( h->param.rc.i_vbv_max_bitrate > 0 && h->param.rc.i_vbv_buffer_size > 0 &&
1315           param->rc.i_vbv_max_bitrate > 0 &&   param->rc.i_vbv_buffer_size > 0 )
1316     {
1317         rc_reconfig |= h->param.rc.i_vbv_max_bitrate != param->rc.i_vbv_max_bitrate;
1318         rc_reconfig |= h->param.rc.i_vbv_buffer_size != param->rc.i_vbv_buffer_size;
1319         rc_reconfig |= h->param.rc.i_bitrate != param->rc.i_bitrate;
1320         COPY( rc.i_vbv_max_bitrate );
1321         COPY( rc.i_vbv_buffer_size );
1322         COPY( rc.i_bitrate );
1323     }
1324     rc_reconfig |= h->param.rc.f_rf_constant != param->rc.f_rf_constant;
1325     rc_reconfig |= h->param.rc.f_rf_constant_max != param->rc.f_rf_constant_max;
1326     COPY( rc.f_rf_constant );
1327     COPY( rc.f_rf_constant_max );
1328 #undef COPY
1329
1330     mbcmp_init( h );
1331
1332     int ret = x264_validate_parameters( h, 0 );
1333
1334     /* Supported reconfiguration options (1-pass only):
1335      * vbv-maxrate
1336      * vbv-bufsize
1337      * crf
1338      * bitrate (CBR only) */
1339     if( !ret && rc_reconfig )
1340         x264_ratecontrol_init_reconfigurable( h, 0 );
1341
1342     return ret;
1343 }
1344
1345 /****************************************************************************
1346  * x264_encoder_parameters:
1347  ****************************************************************************/
1348 void x264_encoder_parameters( x264_t *h, x264_param_t *param )
1349 {
1350     memcpy( param, &h->thread[h->i_thread_phase]->param, sizeof(x264_param_t) );
1351 }
1352
1353 /* internal usage */
1354 static void x264_nal_start( x264_t *h, int i_type, int i_ref_idc )
1355 {
1356     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1357
1358     nal->i_ref_idc        = i_ref_idc;
1359     nal->i_type           = i_type;
1360     nal->b_long_startcode = 1;
1361
1362     nal->i_payload= 0;
1363     nal->p_payload= &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8];
1364 }
1365
1366 /* if number of allocated nals is not enough, re-allocate a larger one. */
1367 static int x264_nal_check_buffer( x264_t *h )
1368 {
1369     if( h->out.i_nal >= h->out.i_nals_allocated )
1370     {
1371         x264_nal_t *new_out = x264_malloc( sizeof(x264_nal_t) * (h->out.i_nals_allocated*2) );
1372         if( !new_out )
1373             return -1;
1374         memcpy( new_out, h->out.nal, sizeof(x264_nal_t) * (h->out.i_nals_allocated) );
1375         x264_free( h->out.nal );
1376         h->out.nal = new_out;
1377         h->out.i_nals_allocated *= 2;
1378     }
1379     return 0;
1380 }
1381
1382 static int x264_nal_end( x264_t *h )
1383 {
1384     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1385     nal->i_payload = &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8] - nal->p_payload;
1386     if( h->param.nalu_process )
1387         h->param.nalu_process( h, nal );
1388     h->out.i_nal++;
1389
1390     return x264_nal_check_buffer( h );
1391 }
1392
1393 static int x264_encoder_encapsulate_nals( x264_t *h, int start )
1394 {
1395     int nal_size = 0, previous_nal_size = 0;
1396
1397     if( h->param.nalu_process )
1398     {
1399         for( int i = start; i < h->out.i_nal; i++ )
1400             nal_size += h->out.nal[i].i_payload;
1401         return nal_size;
1402     }
1403
1404     for( int i = 0; i < start; i++ )
1405         previous_nal_size += h->out.nal[i].i_payload;
1406
1407     for( int i = start; i < h->out.i_nal; i++ )
1408         nal_size += h->out.nal[i].i_payload;
1409
1410     /* Worst-case NAL unit escaping: reallocate the buffer if it's too small. */
1411     int necessary_size = nal_size * 3/2 + h->out.i_nal * 4;
1412     if( h->nal_buffer_size < necessary_size )
1413     {
1414         h->nal_buffer_size = necessary_size * 2;
1415         uint8_t *buf = x264_malloc( h->nal_buffer_size );
1416         if( !buf )
1417             return -1;
1418         if( previous_nal_size )
1419             memcpy( buf, h->nal_buffer, previous_nal_size );
1420         x264_free( h->nal_buffer );
1421         h->nal_buffer = buf;
1422     }
1423
1424     uint8_t *nal_buffer = h->nal_buffer + previous_nal_size;
1425
1426     for( int i = start; i < h->out.i_nal; i++ )
1427     {
1428         h->out.nal[i].b_long_startcode = !i || h->out.nal[i].i_type == NAL_SPS || h->out.nal[i].i_type == NAL_PPS;
1429         x264_nal_encode( h, nal_buffer, &h->out.nal[i] );
1430         nal_buffer += h->out.nal[i].i_payload;
1431     }
1432
1433     x264_emms();
1434
1435     return nal_buffer - (h->nal_buffer + previous_nal_size);
1436 }
1437
1438 /****************************************************************************
1439  * x264_encoder_headers:
1440  ****************************************************************************/
1441 int x264_encoder_headers( x264_t *h, x264_nal_t **pp_nal, int *pi_nal )
1442 {
1443     int frame_size = 0;
1444     /* init bitstream context */
1445     h->out.i_nal = 0;
1446     bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
1447
1448     /* Write SEI, SPS and PPS. */
1449
1450     /* generate sequence parameters */
1451     x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
1452     x264_sps_write( &h->out.bs, h->sps );
1453     if( x264_nal_end( h ) )
1454         return -1;
1455
1456     /* generate picture parameters */
1457     x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
1458     x264_pps_write( &h->out.bs, h->pps );
1459     if( x264_nal_end( h ) )
1460         return -1;
1461
1462     /* identify ourselves */
1463     x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
1464     if( x264_sei_version_write( h, &h->out.bs ) )
1465         return -1;
1466     if( x264_nal_end( h ) )
1467         return -1;
1468
1469     frame_size = x264_encoder_encapsulate_nals( h, 0 );
1470     if( frame_size < 0 )
1471         return -1;
1472
1473     /* now set output*/
1474     *pi_nal = h->out.i_nal;
1475     *pp_nal = &h->out.nal[0];
1476     h->out.i_nal = 0;
1477
1478     return frame_size;
1479 }
1480
1481 /* Check to see whether we have chosen a reference list ordering different
1482  * from the standard's default. */
1483 static inline void x264_reference_check_reorder( x264_t *h )
1484 {
1485     /* The reorder check doesn't check for missing frames, so just
1486      * force a reorder if one of the reference list is corrupt. */
1487     for( int i = 0; h->frames.reference[i]; i++ )
1488         if( h->frames.reference[i]->b_corrupt )
1489         {
1490             h->b_ref_reorder[0] = 1;
1491             return;
1492         }
1493     for( int list = 0; list <= (h->sh.i_type == SLICE_TYPE_B); list++ )
1494         for( int i = 0; i < h->i_ref[list] - 1; i++ )
1495         {
1496             int framenum_diff = h->fref[list][i+1]->i_frame_num - h->fref[list][i]->i_frame_num;
1497             int poc_diff = h->fref[list][i+1]->i_poc - h->fref[list][i]->i_poc;
1498             /* P and B-frames use different default orders. */
1499             if( h->sh.i_type == SLICE_TYPE_P ? framenum_diff > 0 : list == 1 ? poc_diff < 0 : poc_diff > 0 )
1500             {
1501                 h->b_ref_reorder[list] = 1;
1502                 return;
1503             }
1504         }
1505 }
1506
1507 /* return -1 on failure, else return the index of the new reference frame */
1508 int x264_weighted_reference_duplicate( x264_t *h, int i_ref, const x264_weight_t *w )
1509 {
1510     int i = h->i_ref[0];
1511     int j = 1;
1512     x264_frame_t *newframe;
1513     if( i <= 1 ) /* empty list, definitely can't duplicate frame */
1514         return -1;
1515
1516     //Duplication is only used in X264_WEIGHTP_SMART
1517     if( h->param.analyse.i_weighted_pred != X264_WEIGHTP_SMART )
1518         return -1;
1519
1520     /* Duplication is a hack to compensate for crappy rounding in motion compensation.
1521      * With high bit depth, it's not worth doing, so turn it off except in the case of
1522      * unweighted dupes. */
1523     if( BIT_DEPTH > 8 && w != weight_none )
1524         return -1;
1525
1526     newframe = x264_frame_pop_blank_unused( h );
1527     if( !newframe )
1528         return -1;
1529
1530     //FIXME: probably don't need to copy everything
1531     *newframe = *h->fref[0][i_ref];
1532     newframe->i_reference_count = 1;
1533     newframe->orig = h->fref[0][i_ref];
1534     newframe->b_duplicate = 1;
1535     memcpy( h->fenc->weight[j], w, sizeof(h->fenc->weight[i]) );
1536
1537     /* shift the frames to make space for the dupe. */
1538     h->b_ref_reorder[0] = 1;
1539     if( h->i_ref[0] < X264_REF_MAX )
1540         ++h->i_ref[0];
1541     h->fref[0][X264_REF_MAX-1] = NULL;
1542     x264_frame_unshift( &h->fref[0][j], newframe );
1543
1544     return j;
1545 }
1546
1547 static void x264_weighted_pred_init( x264_t *h )
1548 {
1549     /* for now no analysis and set all weights to nothing */
1550     for( int i_ref = 0; i_ref < h->i_ref[0]; i_ref++ )
1551         h->fenc->weighted[i_ref] = h->fref[0][i_ref]->filtered[0];
1552
1553     // FIXME: This only supports weighting of one reference frame
1554     // and duplicates of that frame.
1555     h->fenc->i_lines_weighted = 0;
1556
1557     for( int i_ref = 0; i_ref < (h->i_ref[0] << SLICE_MBAFF); i_ref++ )
1558         for( int i = 0; i < 3; i++ )
1559             h->sh.weight[i_ref][i].weightfn = NULL;
1560
1561
1562     if( h->sh.i_type != SLICE_TYPE_P || h->param.analyse.i_weighted_pred <= 0 )
1563         return;
1564
1565     int i_padv = PADV << PARAM_INTERLACED;
1566     int denom = -1;
1567     int weightplane[2] = { 0, 0 };
1568     int buffer_next = 0;
1569     for( int i = 0; i < 3; i++ )
1570     {
1571         for( int j = 0; j < h->i_ref[0]; j++ )
1572         {
1573             if( h->fenc->weight[j][i].weightfn )
1574             {
1575                 h->sh.weight[j][i] = h->fenc->weight[j][i];
1576                 // if weight is useless, don't write it to stream
1577                 if( h->sh.weight[j][i].i_scale == 1<<h->sh.weight[j][i].i_denom && h->sh.weight[j][i].i_offset == 0 )
1578                     h->sh.weight[j][i].weightfn = NULL;
1579                 else
1580                 {
1581                     if( !weightplane[!!i] )
1582                     {
1583                         weightplane[!!i] = 1;
1584                         h->sh.weight[0][!!i].i_denom = denom = h->sh.weight[j][i].i_denom;
1585                         assert( x264_clip3( denom, 0, 7 ) == denom );
1586                     }
1587
1588                     assert( h->sh.weight[j][i].i_denom == denom );
1589                     if( !i )
1590                     {
1591                         h->fenc->weighted[j] = h->mb.p_weight_buf[buffer_next++] + h->fenc->i_stride[0] * i_padv + PADH;
1592                         //scale full resolution frame
1593                         if( h->param.i_threads == 1 )
1594                         {
1595                             pixel *src = h->fref[0][j]->filtered[0] - h->fref[0][j]->i_stride[0]*i_padv - PADH;
1596                             pixel *dst = h->fenc->weighted[j] - h->fenc->i_stride[0]*i_padv - PADH;
1597                             int stride = h->fenc->i_stride[0];
1598                             int width = h->fenc->i_width[0] + PADH*2;
1599                             int height = h->fenc->i_lines[0] + i_padv*2;
1600                             x264_weight_scale_plane( h, dst, stride, src, stride, width, height, &h->sh.weight[j][0] );
1601                             h->fenc->i_lines_weighted = height;
1602                         }
1603                     }
1604                 }
1605             }
1606         }
1607     }
1608
1609     if( weightplane[1] )
1610         for( int i = 0; i < h->i_ref[0]; i++ )
1611         {
1612             if( h->sh.weight[i][1].weightfn && !h->sh.weight[i][2].weightfn )
1613             {
1614                 h->sh.weight[i][2].i_scale = 1 << h->sh.weight[0][1].i_denom;
1615                 h->sh.weight[i][2].i_offset = 0;
1616             }
1617             else if( h->sh.weight[i][2].weightfn && !h->sh.weight[i][1].weightfn )
1618             {
1619                 h->sh.weight[i][1].i_scale = 1 << h->sh.weight[0][1].i_denom;
1620                 h->sh.weight[i][1].i_offset = 0;
1621             }
1622         }
1623
1624     if( !weightplane[0] )
1625         h->sh.weight[0][0].i_denom = 0;
1626     if( !weightplane[1] )
1627         h->sh.weight[0][1].i_denom = 0;
1628     h->sh.weight[0][2].i_denom = h->sh.weight[0][1].i_denom;
1629 }
1630
1631 static inline int x264_reference_distance( x264_t *h, x264_frame_t *frame )
1632 {
1633     if( h->param.i_frame_packing == 5 )
1634         return abs((h->fenc->i_frame&~1) - (frame->i_frame&~1)) +
1635                   ((h->fenc->i_frame&1) != (frame->i_frame&1));
1636     else
1637         return abs(h->fenc->i_frame - frame->i_frame);
1638 }
1639
1640 static inline void x264_reference_build_list( x264_t *h, int i_poc )
1641 {
1642     int b_ok;
1643
1644     /* build ref list 0/1 */
1645     h->mb.pic.i_fref[0] = h->i_ref[0] = 0;
1646     h->mb.pic.i_fref[1] = h->i_ref[1] = 0;
1647     if( h->sh.i_type == SLICE_TYPE_I )
1648         return;
1649
1650     for( int i = 0; h->frames.reference[i]; i++ )
1651     {
1652         if( h->frames.reference[i]->b_corrupt )
1653             continue;
1654         if( h->frames.reference[i]->i_poc < i_poc )
1655             h->fref[0][h->i_ref[0]++] = h->frames.reference[i];
1656         else if( h->frames.reference[i]->i_poc > i_poc )
1657             h->fref[1][h->i_ref[1]++] = h->frames.reference[i];
1658     }
1659
1660     /* Order reference lists by distance from the current frame. */
1661     for( int list = 0; list < 2; list++ )
1662     {
1663         h->fref_nearest[list] = h->fref[list][0];
1664         do
1665         {
1666             b_ok = 1;
1667             for( int i = 0; i < h->i_ref[list] - 1; i++ )
1668             {
1669                 if( list ? h->fref[list][i+1]->i_poc < h->fref_nearest[list]->i_poc
1670                          : h->fref[list][i+1]->i_poc > h->fref_nearest[list]->i_poc )
1671                     h->fref_nearest[list] = h->fref[list][i+1];
1672                 if( x264_reference_distance( h, h->fref[list][i] ) > x264_reference_distance( h, h->fref[list][i+1] ) )
1673                 {
1674                     XCHG( x264_frame_t*, h->fref[list][i], h->fref[list][i+1] );
1675                     b_ok = 0;
1676                     break;
1677                 }
1678             }
1679         } while( !b_ok );
1680     }
1681
1682     if( h->sh.i_mmco_remove_from_end )
1683         for( int i = h->i_ref[0]-1; i >= h->i_ref[0] - h->sh.i_mmco_remove_from_end; i-- )
1684         {
1685             int diff = h->i_frame_num - h->fref[0][i]->i_frame_num;
1686             h->sh.mmco[h->sh.i_mmco_command_count].i_poc = h->fref[0][i]->i_poc;
1687             h->sh.mmco[h->sh.i_mmco_command_count++].i_difference_of_pic_nums = diff;
1688         }
1689
1690     x264_reference_check_reorder( h );
1691
1692     h->i_ref[1] = X264_MIN( h->i_ref[1], h->frames.i_max_ref1 );
1693     h->i_ref[0] = X264_MIN( h->i_ref[0], h->frames.i_max_ref0 );
1694     h->i_ref[0] = X264_MIN( h->i_ref[0], h->param.i_frame_reference ); // if reconfig() has lowered the limit
1695
1696     /* For Blu-ray compliance, don't reference frames outside of the minigop. */
1697     if( IS_X264_TYPE_B( h->fenc->i_type ) && h->param.b_bluray_compat )
1698         h->i_ref[0] = X264_MIN( h->i_ref[0], IS_X264_TYPE_B( h->fref[0][0]->i_type ) + 1 );
1699
1700     /* add duplicates */
1701     if( h->fenc->i_type == X264_TYPE_P )
1702     {
1703         int idx = -1;
1704         if( h->param.analyse.i_weighted_pred >= X264_WEIGHTP_SIMPLE )
1705         {
1706             x264_weight_t w[3];
1707             w[1].weightfn = w[2].weightfn = NULL;
1708             if( h->param.rc.b_stat_read )
1709                 x264_ratecontrol_set_weights( h, h->fenc );
1710
1711             if( !h->fenc->weight[0][0].weightfn )
1712             {
1713                 h->fenc->weight[0][0].i_denom = 0;
1714                 SET_WEIGHT( w[0], 1, 1, 0, -1 );
1715                 idx = x264_weighted_reference_duplicate( h, 0, w );
1716             }
1717             else
1718             {
1719                 if( h->fenc->weight[0][0].i_scale == 1<<h->fenc->weight[0][0].i_denom )
1720                 {
1721                     SET_WEIGHT( h->fenc->weight[0][0], 1, 1, 0, h->fenc->weight[0][0].i_offset );
1722                 }
1723                 x264_weighted_reference_duplicate( h, 0, weight_none );
1724                 if( h->fenc->weight[0][0].i_offset > -128 )
1725                 {
1726                     w[0] = h->fenc->weight[0][0];
1727                     w[0].i_offset--;
1728                     h->mc.weight_cache( h, &w[0] );
1729                     idx = x264_weighted_reference_duplicate( h, 0, w );
1730                 }
1731             }
1732         }
1733         h->mb.ref_blind_dupe = idx;
1734     }
1735
1736     assert( h->i_ref[0] + h->i_ref[1] <= X264_REF_MAX );
1737     h->mb.pic.i_fref[0] = h->i_ref[0];
1738     h->mb.pic.i_fref[1] = h->i_ref[1];
1739 }
1740
1741 static void x264_fdec_filter_row( x264_t *h, int mb_y, int b_inloop )
1742 {
1743     /* mb_y is the mb to be encoded next, not the mb to be filtered here */
1744     int b_hpel = h->fdec->b_kept_as_ref;
1745     int b_deblock = h->sh.i_disable_deblocking_filter_idc != 1;
1746     int b_end = mb_y == h->i_threadslice_end;
1747     int b_measure_quality = 1;
1748     int min_y = mb_y - (1 << SLICE_MBAFF);
1749     int b_start = min_y == h->i_threadslice_start;
1750     /* Even in interlaced mode, deblocking never modifies more than 4 pixels
1751      * above each MB, as bS=4 doesn't happen for the top of interlaced mbpairs. */
1752     int minpix_y = min_y*16 - 4 * !b_start;
1753     int maxpix_y = mb_y*16 - 4 * !b_end;
1754     b_deblock &= b_hpel || h->param.psz_dump_yuv;
1755     if( h->param.b_sliced_threads && b_start && min_y && !b_inloop )
1756     {
1757         b_deblock = 0;         /* We already deblocked on the inloop pass. */
1758         b_measure_quality = 0; /* We already measured quality on the inloop pass. */
1759     }
1760     if( mb_y & SLICE_MBAFF )
1761         return;
1762     if( min_y < h->i_threadslice_start )
1763         return;
1764
1765     if( b_deblock )
1766         for( int y = min_y; y < mb_y; y += (1 << SLICE_MBAFF) )
1767             x264_frame_deblock_row( h, y );
1768
1769     /* FIXME: Prediction requires different borders for interlaced/progressive mc,
1770      * but the actual image data is equivalent. For now, maintain this
1771      * consistency by copying deblocked pixels between planes. */
1772     if( PARAM_INTERLACED )
1773         for( int p = 0; p < 2; p++ )
1774             for( int i = minpix_y>>p; i < maxpix_y>>p; i++ )
1775                 memcpy( h->fdec->plane_fld[p] + i*h->fdec->i_stride[p],
1776                         h->fdec->plane[p]     + i*h->fdec->i_stride[p],
1777                         h->mb.i_mb_width*16*sizeof(pixel) );
1778
1779     if( b_hpel )
1780     {
1781         int end = mb_y == h->mb.i_mb_height;
1782         x264_frame_expand_border( h, h->fdec, min_y, end );
1783         if( h->param.analyse.i_subpel_refine )
1784         {
1785             x264_frame_filter( h, h->fdec, min_y, end );
1786             x264_frame_expand_border_filtered( h, h->fdec, min_y, end );
1787         }
1788     }
1789
1790     if( SLICE_MBAFF )
1791         for( int i = 0; i < 2; i++ )
1792         {
1793             XCHG( pixel *, h->intra_border_backup[0][i], h->intra_border_backup[3][i] );
1794             XCHG( pixel *, h->intra_border_backup[1][i], h->intra_border_backup[4][i] );
1795         }
1796
1797     if( h->i_thread_frames > 1 && h->fdec->b_kept_as_ref )
1798         x264_frame_cond_broadcast( h->fdec, mb_y*16 + (b_end ? 10000 : -(X264_THREAD_HEIGHT << SLICE_MBAFF)) );
1799
1800     if( b_measure_quality )
1801     {
1802         maxpix_y = X264_MIN( maxpix_y, h->param.i_height );
1803         if( h->param.analyse.b_psnr )
1804         {
1805             uint64_t ssd_y = x264_pixel_ssd_wxh( &h->pixf,
1806                 h->fdec->plane[0] + minpix_y * h->fdec->i_stride[0], h->fdec->i_stride[0],
1807                 h->fenc->plane[0] + minpix_y * h->fenc->i_stride[0], h->fenc->i_stride[0],
1808                 h->param.i_width, maxpix_y-minpix_y );
1809             uint64_t ssd_u, ssd_v;
1810             x264_pixel_ssd_nv12( &h->pixf,
1811                 h->fdec->plane[1] + (minpix_y>>1) * h->fdec->i_stride[1], h->fdec->i_stride[1],
1812                 h->fenc->plane[1] + (minpix_y>>1) * h->fenc->i_stride[1], h->fenc->i_stride[1],
1813                 h->param.i_width>>1, (maxpix_y-minpix_y)>>1, &ssd_u, &ssd_v );
1814             h->stat.frame.i_ssd[0] += ssd_y;
1815             h->stat.frame.i_ssd[1] += ssd_u;
1816             h->stat.frame.i_ssd[2] += ssd_v;
1817         }
1818
1819         if( h->param.analyse.b_ssim )
1820         {
1821             x264_emms();
1822             /* offset by 2 pixels to avoid alignment of ssim blocks with dct blocks,
1823              * and overlap by 4 */
1824             minpix_y += b_start ? 2 : -6;
1825             h->stat.frame.f_ssim +=
1826                 x264_pixel_ssim_wxh( &h->pixf,
1827                     h->fdec->plane[0] + 2+minpix_y*h->fdec->i_stride[0], h->fdec->i_stride[0],
1828                     h->fenc->plane[0] + 2+minpix_y*h->fenc->i_stride[0], h->fenc->i_stride[0],
1829                     h->param.i_width-2, maxpix_y-minpix_y, h->scratch_buffer );
1830         }
1831     }
1832 }
1833
1834 static inline int x264_reference_update( x264_t *h )
1835 {
1836     if( !h->fdec->b_kept_as_ref )
1837     {
1838         if( h->i_thread_frames > 1 )
1839         {
1840             x264_frame_push_unused( h, h->fdec );
1841             h->fdec = x264_frame_pop_unused( h, 1 );
1842             if( !h->fdec )
1843                 return -1;
1844         }
1845         return 0;
1846     }
1847
1848     /* apply mmco from previous frame. */
1849     for( int i = 0; i < h->sh.i_mmco_command_count; i++ )
1850         for( int j = 0; h->frames.reference[j]; j++ )
1851             if( h->frames.reference[j]->i_poc == h->sh.mmco[i].i_poc )
1852                 x264_frame_push_unused( h, x264_frame_shift( &h->frames.reference[j] ) );
1853
1854     /* move frame in the buffer */
1855     x264_frame_push( h->frames.reference, h->fdec );
1856     if( h->frames.reference[h->sps->i_num_ref_frames] )
1857         x264_frame_push_unused( h, x264_frame_shift( h->frames.reference ) );
1858     h->fdec = x264_frame_pop_unused( h, 1 );
1859     if( !h->fdec )
1860         return -1;
1861     return 0;
1862 }
1863
1864 static inline void x264_reference_reset( x264_t *h )
1865 {
1866     while( h->frames.reference[0] )
1867         x264_frame_push_unused( h, x264_frame_pop( h->frames.reference ) );
1868     h->fdec->i_poc =
1869     h->fenc->i_poc = 0;
1870 }
1871
1872 static inline void x264_reference_hierarchy_reset( x264_t *h )
1873 {
1874     int ref;
1875     int b_hasdelayframe = 0;
1876
1877     /* look for delay frames -- chain must only contain frames that are disposable */
1878     for( int i = 0; h->frames.current[i] && IS_DISPOSABLE( h->frames.current[i]->i_type ); i++ )
1879         b_hasdelayframe |= h->frames.current[i]->i_coded
1880                         != h->frames.current[i]->i_frame + h->sps->vui.i_num_reorder_frames;
1881
1882     /* This function must handle b-pyramid and clear frames for open-gop */
1883     if( h->param.i_bframe_pyramid != X264_B_PYRAMID_STRICT && !b_hasdelayframe && h->frames.i_poc_last_open_gop == -1 )
1884         return;
1885
1886     /* Remove last BREF. There will never be old BREFs in the
1887      * dpb during a BREF decode when pyramid == STRICT */
1888     for( ref = 0; h->frames.reference[ref]; ref++ )
1889     {
1890         if( ( h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT
1891             && h->frames.reference[ref]->i_type == X264_TYPE_BREF )
1892             || ( h->frames.reference[ref]->i_poc < h->frames.i_poc_last_open_gop
1893             && h->sh.i_type != SLICE_TYPE_B ) )
1894         {
1895             int diff = h->i_frame_num - h->frames.reference[ref]->i_frame_num;
1896             h->sh.mmco[h->sh.i_mmco_command_count].i_difference_of_pic_nums = diff;
1897             h->sh.mmco[h->sh.i_mmco_command_count++].i_poc = h->frames.reference[ref]->i_poc;
1898             x264_frame_push_unused( h, x264_frame_shift( &h->frames.reference[ref] ) );
1899             h->b_ref_reorder[0] = 1;
1900             ref--;
1901         }
1902     }
1903
1904     /* Prepare room in the dpb for the delayed display time of the later b-frame's */
1905     if( h->param.i_bframe_pyramid )
1906         h->sh.i_mmco_remove_from_end = X264_MAX( ref + 2 - h->frames.i_max_dpb, 0 );
1907 }
1908
1909 static inline void x264_slice_init( x264_t *h, int i_nal_type, int i_global_qp )
1910 {
1911     /* ------------------------ Create slice header  ----------------------- */
1912     if( i_nal_type == NAL_SLICE_IDR )
1913     {
1914         x264_slice_header_init( h, &h->sh, h->sps, h->pps, h->i_idr_pic_id, h->i_frame_num, i_global_qp );
1915
1916         /* alternate id */
1917         h->i_idr_pic_id ^= 1;
1918     }
1919     else
1920     {
1921         x264_slice_header_init( h, &h->sh, h->sps, h->pps, -1, h->i_frame_num, i_global_qp );
1922
1923         h->sh.i_num_ref_idx_l0_active = h->i_ref[0] <= 0 ? 1 : h->i_ref[0];
1924         h->sh.i_num_ref_idx_l1_active = h->i_ref[1] <= 0 ? 1 : h->i_ref[1];
1925         if( h->sh.i_num_ref_idx_l0_active != h->pps->i_num_ref_idx_l0_default_active ||
1926             (h->sh.i_type == SLICE_TYPE_B && h->sh.i_num_ref_idx_l1_active != h->pps->i_num_ref_idx_l1_default_active) )
1927         {
1928             h->sh.b_num_ref_idx_override = 1;
1929         }
1930     }
1931
1932     if( h->fenc->i_type == X264_TYPE_BREF && h->param.b_bluray_compat && h->sh.i_mmco_command_count )
1933     {
1934         h->b_sh_backup = 1;
1935         h->sh_backup = h->sh;
1936     }
1937
1938     h->fdec->i_frame_num = h->sh.i_frame_num;
1939
1940     if( h->sps->i_poc_type == 0 )
1941     {
1942         h->sh.i_poc = h->fdec->i_poc;
1943         if( PARAM_INTERLACED )
1944         {
1945             h->sh.i_delta_poc_bottom = h->param.b_tff ? 1 : -1;
1946             h->sh.i_poc += h->sh.i_delta_poc_bottom == -1;
1947         }
1948         else
1949             h->sh.i_delta_poc_bottom = 0;
1950         h->fdec->i_delta_poc[0] = h->sh.i_delta_poc_bottom == -1;
1951         h->fdec->i_delta_poc[1] = h->sh.i_delta_poc_bottom ==  1;
1952     }
1953     else if( h->sps->i_poc_type == 1 )
1954     {
1955         /* FIXME TODO FIXME */
1956     }
1957     else
1958     {
1959         /* Nothing to do ? */
1960     }
1961
1962     x264_macroblock_slice_init( h );
1963 }
1964
1965 static int x264_slice_write( x264_t *h )
1966 {
1967     int i_skip;
1968     int mb_xy, i_mb_x, i_mb_y;
1969     int i_skip_bak = 0; /* Shut up GCC. */
1970     bs_t UNINIT(bs_bak);
1971     x264_cabac_t cabac_bak;
1972     uint8_t cabac_prevbyte_bak = 0; /* Shut up GCC. */
1973     int mv_bits_bak = 0;
1974     int tex_bits_bak = 0;
1975     /* NALUs other than the first use a 3-byte startcode.
1976      * Add one extra byte for the rbsp, and one more for the final CABAC putbyte.
1977      * Then add an extra 5 bytes just in case, to account for random NAL escapes and
1978      * other inaccuracies. */
1979     int overhead_guess = (NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal)) + 1 + h->param.b_cabac + 5;
1980     int slice_max_size = h->param.i_slice_max_size > 0 ? (h->param.i_slice_max_size-overhead_guess)*8 : 0;
1981     int back_up_bitstream = slice_max_size || (!h->param.b_cabac && h->sps->i_profile_idc < PROFILE_HIGH);
1982     int starting_bits = bs_pos(&h->out.bs);
1983     int b_deblock = h->sh.i_disable_deblocking_filter_idc != 1;
1984     int b_hpel = h->fdec->b_kept_as_ref;
1985     uint8_t *last_emu_check;
1986     b_deblock &= b_hpel || h->param.psz_dump_yuv;
1987     bs_realign( &h->out.bs );
1988
1989     /* Slice */
1990     x264_nal_start( h, h->i_nal_type, h->i_nal_ref_idc );
1991     h->out.nal[h->out.i_nal].i_first_mb = h->sh.i_first_mb;
1992
1993     /* Slice header */
1994     x264_macroblock_thread_init( h );
1995
1996     /* If this isn't the first slice in the threadslice, set the slice QP
1997      * equal to the last QP in the previous slice for more accurate
1998      * CABAC initialization. */
1999     if( h->sh.i_first_mb != h->i_threadslice_start * h->mb.i_mb_width )
2000     {
2001         h->sh.i_qp = h->mb.i_last_qp;
2002         h->sh.i_qp_delta = h->sh.i_qp - h->pps->i_pic_init_qp;
2003     }
2004
2005     x264_slice_header_write( &h->out.bs, &h->sh, h->i_nal_ref_idc );
2006     if( h->param.b_cabac )
2007     {
2008         /* alignment needed */
2009         bs_align_1( &h->out.bs );
2010
2011         /* init cabac */
2012         x264_cabac_context_init( &h->cabac, h->sh.i_type, x264_clip3( h->sh.i_qp-QP_BD_OFFSET, 0, 51 ), h->sh.i_cabac_init_idc );
2013         x264_cabac_encode_init ( &h->cabac, h->out.bs.p, h->out.bs.p_end );
2014         last_emu_check = h->cabac.p;
2015     }
2016     else
2017         last_emu_check = h->out.bs.p;
2018     h->mb.i_last_qp = h->sh.i_qp;
2019     h->mb.i_last_dqp = 0;
2020     h->mb.field_decoding_flag = 0;
2021
2022     i_mb_y = h->sh.i_first_mb / h->mb.i_mb_width;
2023     i_mb_x = h->sh.i_first_mb % h->mb.i_mb_width;
2024     i_skip = 0;
2025     int mb_size[2];
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( x264_bitstream_check_buffer( h ) )
2033             return -1;
2034
2035         if( back_up_bitstream && (!SLICE_MBAFF || (i_mb_y&1) == 0) )
2036         {
2037             mv_bits_bak = h->stat.frame.i_mv_bits;
2038             tex_bits_bak = h->stat.frame.i_tex_bits;
2039             /* We don't need the contexts because flushing the CABAC encoder has no context
2040              * dependency and macroblocks are only re-encoded in the case where a slice is
2041              * ended (and thus the content of all contexts are thrown away). */
2042             if( h->param.b_cabac )
2043             {
2044                 memcpy( &cabac_bak, &h->cabac, offsetof(x264_cabac_t, f8_bits_encoded) );
2045                 /* x264's CABAC writer modifies the previous byte during carry, so it has to be
2046                  * backed up. */
2047                 cabac_prevbyte_bak = h->cabac.p[-1];
2048             }
2049             else
2050             {
2051                 bs_bak = h->out.bs;
2052                 i_skip_bak = i_skip;
2053             }
2054         }
2055
2056         if( i_mb_x == 0 && !h->mb.b_reencode_mb )
2057             x264_fdec_filter_row( h, i_mb_y, 1 );
2058
2059         if( PARAM_INTERLACED )
2060         {
2061             if( h->mb.b_adaptive_mbaff )
2062             {
2063                 if( !(i_mb_y&1) )
2064                 {
2065                     /* FIXME: VSAD is fast but fairly poor at choosing the best interlace type. */
2066                     int stride = h->fenc->i_stride[0];
2067                     pixel *fenc = h->fenc->plane[0] + 16 * (i_mb_x + i_mb_y * stride);
2068                     h->mb.b_interlaced = x264_field_vsad( h, fenc, stride );
2069                     memcpy( &h->zigzagf, MB_INTERLACED ? &h->zigzagf_interlaced : &h->zigzagf_progressive, sizeof(h->zigzagf) );
2070                 }
2071             }
2072             h->mb.field[mb_xy] = MB_INTERLACED;
2073         }
2074
2075         /* load cache */
2076         if( SLICE_MBAFF )
2077             x264_macroblock_cache_load_interlaced( h, i_mb_x, i_mb_y );
2078         else
2079             x264_macroblock_cache_load_progressive( h, i_mb_x, i_mb_y );
2080
2081         x264_macroblock_analyse( h );
2082
2083         /* encode this macroblock -> be careful it can change the mb type to P_SKIP if needed */
2084 reencode:
2085         x264_macroblock_encode( h );
2086
2087         if( h->param.b_cabac )
2088         {
2089             if( mb_xy > h->sh.i_first_mb && !(SLICE_MBAFF && (i_mb_y&1)) )
2090                 x264_cabac_encode_terminal( &h->cabac );
2091
2092             if( IS_SKIP( h->mb.i_type ) )
2093                 x264_cabac_mb_skip( h, 1 );
2094             else
2095             {
2096                 if( h->sh.i_type != SLICE_TYPE_I )
2097                     x264_cabac_mb_skip( h, 0 );
2098                 x264_macroblock_write_cabac( h, &h->cabac );
2099             }
2100         }
2101         else
2102         {
2103             if( IS_SKIP( h->mb.i_type ) )
2104                 i_skip++;
2105             else
2106             {
2107                 if( h->sh.i_type != SLICE_TYPE_I )
2108                 {
2109                     bs_write_ue( &h->out.bs, i_skip );  /* skip run */
2110                     i_skip = 0;
2111                 }
2112                 x264_macroblock_write_cavlc( h );
2113                 /* If there was a CAVLC level code overflow, try again at a higher QP. */
2114                 if( h->mb.b_overflow )
2115                 {
2116                     h->mb.i_chroma_qp = h->chroma_qp_table[++h->mb.i_qp];
2117                     h->mb.i_skip_intra = 0;
2118                     h->mb.b_skip_mc = 0;
2119                     h->mb.b_overflow = 0;
2120                     h->out.bs = bs_bak;
2121                     i_skip = i_skip_bak;
2122                     h->stat.frame.i_mv_bits = mv_bits_bak;
2123                     h->stat.frame.i_tex_bits = tex_bits_bak;
2124                     goto reencode;
2125                 }
2126             }
2127         }
2128
2129         int total_bits = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
2130         mb_size[i_mb_y&1] = total_bits - mb_spos;
2131
2132         if( slice_max_size )
2133         {
2134             /* Count the skip run, just in case. */
2135             if( !h->param.b_cabac )
2136                 total_bits += bs_size_ue_big( i_skip );
2137             /* Check for escape bytes. */
2138             uint8_t *end = h->param.b_cabac ? h->cabac.p : h->out.bs.p;
2139             for( ; last_emu_check < end - 2; last_emu_check++ )
2140                 if( last_emu_check[0] == 0 && last_emu_check[1] == 0 && last_emu_check[2] <= 3 )
2141                 {
2142                     slice_max_size -= 8;
2143                     last_emu_check++;
2144                 }
2145             /* We'll just re-encode this last macroblock if we go over the max slice size. */
2146             if( total_bits - starting_bits > slice_max_size && !h->mb.b_reencode_mb )
2147             {
2148                 if( mb_xy != h->sh.i_first_mb )
2149                 {
2150                     h->stat.frame.i_mv_bits = mv_bits_bak;
2151                     h->stat.frame.i_tex_bits = tex_bits_bak;
2152                     if( h->param.b_cabac )
2153                     {
2154                         memcpy( &h->cabac, &cabac_bak, offsetof(x264_cabac_t, f8_bits_encoded) );
2155                         h->cabac.p[-1] = cabac_prevbyte_bak;
2156                     }
2157                     else
2158                     {
2159                         h->out.bs = bs_bak;
2160                         i_skip = i_skip_bak;
2161                     }
2162                     h->mb.b_reencode_mb = 1;
2163                     if( SLICE_MBAFF )
2164                     {
2165                         // set to bottom of previous mbpair
2166                         if( i_mb_x )
2167                             h->sh.i_last_mb = mb_xy-1+h->mb.i_mb_stride*(!(i_mb_y&1));
2168                         else
2169                             h->sh.i_last_mb = (i_mb_y-2+!(i_mb_y&1))*h->mb.i_mb_stride + h->mb.i_mb_width - 1;
2170                     }
2171                     else
2172                         h->sh.i_last_mb = mb_xy-1;
2173                     break;
2174                 }
2175                 else
2176                 {
2177                     h->sh.i_last_mb = mb_xy;
2178                     h->mb.b_reencode_mb = 0;
2179                 }
2180             }
2181             else
2182                 h->mb.b_reencode_mb = 0;
2183         }
2184
2185 #if HAVE_VISUALIZE
2186         if( h->param.b_visualize )
2187             x264_visualize_mb( h );
2188 #endif
2189
2190         /* save cache */
2191         x264_macroblock_cache_save( h );
2192
2193         /* accumulate mb stats */
2194         h->stat.frame.i_mb_count[h->mb.i_type]++;
2195
2196         int b_intra = IS_INTRA( h->mb.i_type );
2197         int b_skip = IS_SKIP( h->mb.i_type );
2198         if( h->param.i_log_level >= X264_LOG_INFO || h->param.rc.b_stat_write )
2199         {
2200             if( !b_intra && !b_skip && !IS_DIRECT( h->mb.i_type ) )
2201             {
2202                 if( h->mb.i_partition != D_8x8 )
2203                         h->stat.frame.i_mb_partition[h->mb.i_partition] += 4;
2204                     else
2205                         for( int i = 0; i < 4; i++ )
2206                             h->stat.frame.i_mb_partition[h->mb.i_sub_partition[i]] ++;
2207                 if( h->param.i_frame_reference > 1 )
2208                     for( int i_list = 0; i_list <= (h->sh.i_type == SLICE_TYPE_B); i_list++ )
2209                         for( int i = 0; i < 4; i++ )
2210                         {
2211                             int i_ref = h->mb.cache.ref[i_list][ x264_scan8[4*i] ];
2212                             if( i_ref >= 0 )
2213                                 h->stat.frame.i_mb_count_ref[i_list][i_ref] ++;
2214                         }
2215             }
2216         }
2217
2218         if( h->param.i_log_level >= X264_LOG_INFO )
2219         {
2220             if( h->mb.i_cbp_luma | h->mb.i_cbp_chroma )
2221             {
2222                 int cbpsum = (h->mb.i_cbp_luma&1) + ((h->mb.i_cbp_luma>>1)&1)
2223                            + ((h->mb.i_cbp_luma>>2)&1) + (h->mb.i_cbp_luma>>3);
2224                 h->stat.frame.i_mb_cbp[!b_intra + 0] += cbpsum;
2225                 h->stat.frame.i_mb_cbp[!b_intra + 2] += !!h->mb.i_cbp_chroma;
2226                 h->stat.frame.i_mb_cbp[!b_intra + 4] += h->mb.i_cbp_chroma >> 1;
2227             }
2228             if( h->mb.i_cbp_luma && !b_intra )
2229             {
2230                 h->stat.frame.i_mb_count_8x8dct[0] ++;
2231                 h->stat.frame.i_mb_count_8x8dct[1] += h->mb.b_transform_8x8;
2232             }
2233             if( b_intra && h->mb.i_type != I_PCM )
2234             {
2235                 if( h->mb.i_type == I_16x16 )
2236                     h->stat.frame.i_mb_pred_mode[0][h->mb.i_intra16x16_pred_mode]++;
2237                 else if( h->mb.i_type == I_8x8 )
2238                     for( int i = 0; i < 16; i += 4 )
2239                         h->stat.frame.i_mb_pred_mode[1][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
2240                 else //if( h->mb.i_type == I_4x4 )
2241                     for( int i = 0; i < 16; i++ )
2242                         h->stat.frame.i_mb_pred_mode[2][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
2243                 h->stat.frame.i_mb_pred_mode[3][x264_mb_pred_mode8x8c_fix[h->mb.i_chroma_pred_mode]]++;
2244             }
2245             h->stat.frame.i_mb_field[b_intra?0:b_skip?2:1] += MB_INTERLACED;
2246         }
2247
2248         /* calculate deblock strength values (actual deblocking is done per-row along with hpel) */
2249         if( b_deblock )
2250             x264_macroblock_deblock_strength( h );
2251
2252         if( SLICE_MBAFF )
2253         {
2254             /* update ratecontrol per-mbpair in MBAFF */
2255             if( i_mb_y&1 )
2256                 x264_ratecontrol_mb( h, mb_size[0]+mb_size[1] );
2257         }
2258         else
2259             x264_ratecontrol_mb( h, mb_size[i_mb_y&1] );
2260
2261         if( mb_xy == h->sh.i_last_mb )
2262             break;
2263
2264         if( SLICE_MBAFF )
2265         {
2266             i_mb_x += i_mb_y & 1;
2267             i_mb_y ^= i_mb_x < h->mb.i_mb_width;
2268         }
2269         else
2270             i_mb_x++;
2271         if( i_mb_x == h->mb.i_mb_width )
2272         {
2273             i_mb_y++;
2274             i_mb_x = 0;
2275         }
2276     }
2277     h->out.nal[h->out.i_nal].i_last_mb = h->sh.i_last_mb;
2278
2279     if( h->param.b_cabac )
2280     {
2281         x264_cabac_encode_flush( h, &h->cabac );
2282         h->out.bs.p = h->cabac.p;
2283     }
2284     else
2285     {
2286         if( i_skip > 0 )
2287             bs_write_ue( &h->out.bs, i_skip );  /* last skip run */
2288         /* rbsp_slice_trailing_bits */
2289         bs_rbsp_trailing( &h->out.bs );
2290         bs_flush( &h->out.bs );
2291     }
2292     if( x264_nal_end( h ) )
2293         return -1;
2294
2295     if( h->sh.i_last_mb == (h->i_threadslice_end * h->mb.i_mb_width - 1) )
2296     {
2297         h->stat.frame.i_misc_bits = bs_pos( &h->out.bs )
2298                                   + (h->out.i_nal*NALU_OVERHEAD * 8)
2299                                   - h->stat.frame.i_tex_bits
2300                                   - h->stat.frame.i_mv_bits;
2301         x264_fdec_filter_row( h, h->i_threadslice_end, 1 );
2302     }
2303
2304     return 0;
2305 }
2306
2307 static void x264_thread_sync_context( x264_t *dst, x264_t *src )
2308 {
2309     if( dst == src )
2310         return;
2311
2312     // reference counting
2313     for( x264_frame_t **f = src->frames.reference; *f; f++ )
2314         (*f)->i_reference_count++;
2315     for( x264_frame_t **f = dst->frames.reference; *f; f++ )
2316         x264_frame_push_unused( src, *f );
2317     src->fdec->i_reference_count++;
2318     x264_frame_push_unused( src, dst->fdec );
2319
2320     // copy everything except the per-thread pointers and the constants.
2321     memcpy( &dst->i_frame, &src->i_frame, offsetof(x264_t, mb.type) - offsetof(x264_t, i_frame) );
2322     dst->param = src->param;
2323     dst->stat = src->stat;
2324     dst->pixf = src->pixf;
2325 }
2326
2327 static void x264_thread_sync_stat( x264_t *dst, x264_t *src )
2328 {
2329     if( dst == src )
2330         return;
2331     memcpy( &dst->stat.i_frame_count, &src->stat.i_frame_count, sizeof(dst->stat) - sizeof(dst->stat.frame) );
2332 }
2333
2334 static void *x264_slices_write( x264_t *h )
2335 {
2336     int i_slice_num = 0;
2337     int last_thread_mb = h->sh.i_last_mb;
2338
2339 #if HAVE_VISUALIZE
2340     if( h->param.b_visualize )
2341         if( x264_visualize_init( h ) )
2342             return (void *)-1;
2343 #endif
2344
2345     /* init stats */
2346     memset( &h->stat.frame, 0, sizeof(h->stat.frame) );
2347     h->mb.b_reencode_mb = 0;
2348     while( h->sh.i_first_mb + SLICE_MBAFF*h->mb.i_mb_stride <= last_thread_mb )
2349     {
2350         h->sh.i_last_mb = last_thread_mb;
2351         if( h->param.i_slice_max_mbs )
2352         {
2353             if( SLICE_MBAFF )
2354             {
2355                 // convert first to mbaff form, add slice-max-mbs, then convert back to normal form
2356                 int last_mbaff = 2*(h->sh.i_first_mb % h->mb.i_mb_width)
2357                     + h->mb.i_mb_width*(h->sh.i_first_mb / h->mb.i_mb_width)
2358                     + h->param.i_slice_max_mbs - 1;
2359                 int last_x = (last_mbaff % (2*h->mb.i_mb_width))/2;
2360                 int last_y = (last_mbaff / (2*h->mb.i_mb_width))*2 + 1;
2361                 h->sh.i_last_mb = last_x + h->mb.i_mb_stride*last_y;
2362             }
2363             else
2364                 h->sh.i_last_mb = h->sh.i_first_mb + h->param.i_slice_max_mbs - 1;
2365         }
2366         else if( h->param.i_slice_count && !h->param.b_sliced_threads )
2367         {
2368             int height = h->mb.i_mb_height >> PARAM_INTERLACED;
2369             int width = h->mb.i_mb_width << PARAM_INTERLACED;
2370             i_slice_num++;
2371             h->sh.i_last_mb = (height * i_slice_num + h->param.i_slice_count/2) / h->param.i_slice_count * width - 1;
2372         }
2373         h->sh.i_last_mb = X264_MIN( h->sh.i_last_mb, last_thread_mb );
2374         if( x264_stack_align( x264_slice_write, h ) )
2375             return (void *)-1;
2376         h->sh.i_first_mb = h->sh.i_last_mb + 1;
2377         // if i_first_mb is not the last mb in a row then go to the next mb in MBAFF order
2378         if( SLICE_MBAFF && h->sh.i_first_mb % h->mb.i_mb_width )
2379             h->sh.i_first_mb -= h->mb.i_mb_stride;
2380     }
2381
2382 #if HAVE_VISUALIZE
2383     if( h->param.b_visualize )
2384     {
2385         x264_visualize_show( h );
2386         x264_visualize_close( h );
2387     }
2388 #endif
2389
2390     return (void *)0;
2391 }
2392
2393 static int x264_threaded_slices_write( x264_t *h )
2394 {
2395     /* set first/last mb and sync contexts */
2396     for( int i = 0; i < h->param.i_threads; i++ )
2397     {
2398         x264_t *t = h->thread[i];
2399         if( i )
2400         {
2401             t->param = h->param;
2402             memcpy( &t->i_frame, &h->i_frame, offsetof(x264_t, rc) - offsetof(x264_t, i_frame) );
2403         }
2404         int height = h->mb.i_mb_height >> PARAM_INTERLACED;
2405         t->i_threadslice_start = ((height *  i    + h->param.i_slice_count/2) / h->param.i_threads) << PARAM_INTERLACED;
2406         t->i_threadslice_end   = ((height * (i+1) + h->param.i_slice_count/2) / h->param.i_threads) << PARAM_INTERLACED;
2407         t->sh.i_first_mb = t->i_threadslice_start * h->mb.i_mb_width;
2408         t->sh.i_last_mb  =   t->i_threadslice_end * h->mb.i_mb_width - 1;
2409     }
2410
2411     x264_stack_align( x264_analyse_weight_frame, h, h->mb.i_mb_height*16 + 16 );
2412
2413     x264_threads_distribute_ratecontrol( h );
2414
2415     /* dispatch */
2416     for( int i = 0; i < h->param.i_threads; i++ )
2417     {
2418         x264_threadpool_run( h->threadpool, (void*)x264_slices_write, h->thread[i] );
2419         h->thread[i]->b_thread_active = 1;
2420     }
2421     for( int i = 0; i < h->param.i_threads; i++ )
2422     {
2423         h->thread[i]->b_thread_active = 0;
2424         if( (intptr_t)x264_threadpool_wait( h->threadpool, h->thread[i] ) )
2425             return -1;
2426     }
2427
2428     /* Go back and fix up the hpel on the borders between slices. */
2429     for( int i = 1; i < h->param.i_threads; i++ )
2430     {
2431         x264_fdec_filter_row( h->thread[i], h->thread[i]->i_threadslice_start + 1, 0 );
2432         if( SLICE_MBAFF )
2433             x264_fdec_filter_row( h->thread[i], h->thread[i]->i_threadslice_start + 2, 0 );
2434     }
2435
2436     x264_threads_merge_ratecontrol( h );
2437
2438     for( int i = 1; i < h->param.i_threads; i++ )
2439     {
2440         x264_t *t = h->thread[i];
2441         for( int j = 0; j < t->out.i_nal; j++ )
2442         {
2443             h->out.nal[h->out.i_nal] = t->out.nal[j];
2444             h->out.i_nal++;
2445             x264_nal_check_buffer( h );
2446         }
2447         /* All entries in stat.frame are ints except for ssd/ssim. */
2448         for( int j = 0; j < (offsetof(x264_t,stat.frame.i_ssd) - offsetof(x264_t,stat.frame.i_mv_bits)) / sizeof(int); j++ )
2449             ((int*)&h->stat.frame)[j] += ((int*)&t->stat.frame)[j];
2450         for( int j = 0; j < 3; j++ )
2451             h->stat.frame.i_ssd[j] += t->stat.frame.i_ssd[j];
2452         h->stat.frame.f_ssim += t->stat.frame.f_ssim;
2453     }
2454
2455     return 0;
2456 }
2457
2458 void x264_encoder_intra_refresh( x264_t *h )
2459 {
2460     h = h->thread[h->i_thread_phase];
2461     h->b_queued_intra_refresh = 1;
2462 }
2463
2464 int x264_encoder_invalidate_reference( x264_t *h, int64_t pts )
2465 {
2466     if( h->param.i_bframe )
2467     {
2468         x264_log( h, X264_LOG_ERROR, "x264_encoder_invalidate_reference is not supported with B-frames enabled\n" );
2469         return -1;
2470     }
2471     if( h->param.b_intra_refresh )
2472     {
2473         x264_log( h, X264_LOG_ERROR, "x264_encoder_invalidate_reference is not supported with intra refresh enabled\n" );
2474         return -1;
2475     }
2476     h = h->thread[h->i_thread_phase];
2477     if( pts >= h->i_last_idr_pts )
2478     {
2479         for( int i = 0; h->frames.reference[i]; i++ )
2480             if( pts <= h->frames.reference[i]->i_pts )
2481                 h->frames.reference[i]->b_corrupt = 1;
2482         if( pts <= h->fdec->i_pts )
2483             h->fdec->b_corrupt = 1;
2484     }
2485     return 0;
2486 }
2487
2488 /****************************************************************************
2489  * x264_encoder_encode:
2490  *  XXX: i_poc   : is the poc of the current given picture
2491  *       i_frame : is the number of the frame being coded
2492  *  ex:  type frame poc
2493  *       I      0   2*0
2494  *       P      1   2*3
2495  *       B      2   2*1
2496  *       B      3   2*2
2497  *       P      4   2*6
2498  *       B      5   2*4
2499  *       B      6   2*5
2500  ****************************************************************************/
2501 int     x264_encoder_encode( x264_t *h,
2502                              x264_nal_t **pp_nal, int *pi_nal,
2503                              x264_picture_t *pic_in,
2504                              x264_picture_t *pic_out )
2505 {
2506     x264_t *thread_current, *thread_prev, *thread_oldest;
2507     int i_nal_type, i_nal_ref_idc, i_global_qp;
2508     int overhead = NALU_OVERHEAD;
2509
2510     if( h->i_thread_frames > 1 )
2511     {
2512         thread_prev    = h->thread[ h->i_thread_phase ];
2513         h->i_thread_phase = (h->i_thread_phase + 1) % h->i_thread_frames;
2514         thread_current = h->thread[ h->i_thread_phase ];
2515         thread_oldest  = h->thread[ (h->i_thread_phase + 1) % h->i_thread_frames ];
2516         x264_thread_sync_context( thread_current, thread_prev );
2517         x264_thread_sync_ratecontrol( thread_current, thread_prev, thread_oldest );
2518         h = thread_current;
2519     }
2520     else
2521     {
2522         thread_current =
2523         thread_oldest  = h;
2524     }
2525 #if HAVE_MMX
2526     if( h->param.cpu&X264_CPU_SSE_MISALIGN )
2527         x264_cpu_mask_misalign_sse();
2528 #endif
2529
2530     // ok to call this before encoding any frames, since the initial values of fdec have b_kept_as_ref=0
2531     if( x264_reference_update( h ) )
2532         return -1;
2533     h->fdec->i_lines_completed = -1;
2534
2535     /* no data out */
2536     *pi_nal = 0;
2537     *pp_nal = NULL;
2538
2539     /* ------------------- Setup new frame from picture -------------------- */
2540     if( pic_in != NULL )
2541     {
2542         /* 1: Copy the picture to a frame and move it to a buffer */
2543         x264_frame_t *fenc = x264_frame_pop_unused( h, 0 );
2544         if( !fenc )
2545             return -1;
2546
2547         if( x264_frame_copy_picture( h, fenc, pic_in ) < 0 )
2548             return -1;
2549
2550         if( h->param.i_width != 16 * h->mb.i_mb_width ||
2551             h->param.i_height != 16 * h->mb.i_mb_height )
2552             x264_frame_expand_border_mod16( h, fenc );
2553
2554         fenc->i_frame = h->frames.i_input++;
2555
2556         if( fenc->i_frame == 0 )
2557             h->frames.i_first_pts = fenc->i_pts;
2558         if( h->frames.i_bframe_delay && fenc->i_frame == h->frames.i_bframe_delay )
2559             h->frames.i_bframe_delay_time = fenc->i_pts - h->frames.i_first_pts;
2560
2561         if( h->param.b_vfr_input && fenc->i_pts <= h->frames.i_largest_pts )
2562             x264_log( h, X264_LOG_WARNING, "non-strictly-monotonic PTS\n" );
2563
2564         h->frames.i_second_largest_pts = h->frames.i_largest_pts;
2565         h->frames.i_largest_pts = fenc->i_pts;
2566
2567         if( (fenc->i_pic_struct < PIC_STRUCT_AUTO) || (fenc->i_pic_struct > PIC_STRUCT_TRIPLE) )
2568             fenc->i_pic_struct = PIC_STRUCT_AUTO;
2569
2570         if( fenc->i_pic_struct == PIC_STRUCT_AUTO )
2571         {
2572 #if HAVE_INTERLACED
2573             int b_interlaced = fenc->param ? fenc->param->b_interlaced : h->param.b_interlaced;
2574 #else
2575             int b_interlaced = 0;
2576 #endif
2577             if( b_interlaced )
2578             {
2579                 int b_tff = fenc->param ? fenc->param->b_tff : h->param.b_tff;
2580                 fenc->i_pic_struct = b_tff ? PIC_STRUCT_TOP_BOTTOM : PIC_STRUCT_BOTTOM_TOP;
2581             }
2582             else
2583                 fenc->i_pic_struct = PIC_STRUCT_PROGRESSIVE;
2584         }
2585
2586         if( h->param.rc.b_mb_tree && h->param.rc.b_stat_read )
2587         {
2588             if( x264_macroblock_tree_read( h, fenc, pic_in->prop.quant_offsets ) )
2589                 return -1;
2590         }
2591         else
2592             x264_stack_align( x264_adaptive_quant_frame, h, fenc, pic_in->prop.quant_offsets );
2593
2594         if( pic_in->prop.quant_offsets_free )
2595             pic_in->prop.quant_offsets_free( pic_in->prop.quant_offsets );
2596
2597         if( h->frames.b_have_lowres )
2598             x264_frame_init_lowres( h, fenc );
2599
2600         /* 2: Place the frame into the queue for its slice type decision */
2601         x264_lookahead_put_frame( h, fenc );
2602
2603         if( h->frames.i_input <= h->frames.i_delay + 1 - h->i_thread_frames )
2604         {
2605             /* Nothing yet to encode, waiting for filling of buffers */
2606             pic_out->i_type = X264_TYPE_AUTO;
2607             return 0;
2608         }
2609     }
2610     else
2611     {
2612         /* signal kills for lookahead thread */
2613         x264_pthread_mutex_lock( &h->lookahead->ifbuf.mutex );
2614         h->lookahead->b_exit_thread = 1;
2615         x264_pthread_cond_broadcast( &h->lookahead->ifbuf.cv_fill );
2616         x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
2617     }
2618
2619     h->i_frame++;
2620     /* 3: The picture is analyzed in the lookahead */
2621     if( !h->frames.current[0] )
2622         x264_lookahead_get_frames( h );
2623
2624     if( !h->frames.current[0] && x264_lookahead_is_empty( h ) )
2625         return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
2626
2627     /* ------------------- Get frame to be encoded ------------------------- */
2628     /* 4: get picture to encode */
2629     h->fenc = x264_frame_shift( h->frames.current );
2630     if( h->i_frame == h->i_thread_frames - 1 )
2631         h->i_reordered_pts_delay = h->fenc->i_reordered_pts;
2632     if( h->fenc->param )
2633     {
2634         x264_encoder_reconfig( h, h->fenc->param );
2635         if( h->fenc->param->param_free )
2636             h->fenc->param->param_free( h->fenc->param );
2637     }
2638
2639     if( !IS_X264_TYPE_I( h->fenc->i_type ) )
2640     {
2641         int valid_refs_left = 0;
2642         for( int i = 0; h->frames.reference[i]; i++ )
2643             if( !h->frames.reference[i]->b_corrupt )
2644                 valid_refs_left++;
2645         /* No valid reference frames left: force an IDR. */
2646         if( !valid_refs_left )
2647         {
2648             h->fenc->b_keyframe = 1;
2649             h->fenc->i_type = X264_TYPE_IDR;
2650         }
2651     }
2652
2653     if( h->fenc->b_keyframe )
2654     {
2655         h->frames.i_last_keyframe = h->fenc->i_frame;
2656         if( h->fenc->i_type == X264_TYPE_IDR )
2657         {
2658             h->i_frame_num = 0;
2659             h->frames.i_last_idr = h->fenc->i_frame;
2660         }
2661     }
2662     h->sh.i_mmco_command_count =
2663     h->sh.i_mmco_remove_from_end = 0;
2664     h->b_ref_reorder[0] =
2665     h->b_ref_reorder[1] = 0;
2666     h->fdec->i_poc =
2667     h->fenc->i_poc = 2 * ( h->fenc->i_frame - X264_MAX( h->frames.i_last_idr, 0 ) );
2668
2669     /* ------------------- Setup frame context ----------------------------- */
2670     /* 5: Init data dependent of frame type */
2671     if( h->fenc->i_type == X264_TYPE_IDR )
2672     {
2673         /* reset ref pictures */
2674         i_nal_type    = NAL_SLICE_IDR;
2675         i_nal_ref_idc = NAL_PRIORITY_HIGHEST;
2676         h->sh.i_type = SLICE_TYPE_I;
2677         x264_reference_reset( h );
2678         h->frames.i_poc_last_open_gop = -1;
2679     }
2680     else if( h->fenc->i_type == X264_TYPE_I )
2681     {
2682         i_nal_type    = NAL_SLICE;
2683         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
2684         h->sh.i_type = SLICE_TYPE_I;
2685         x264_reference_hierarchy_reset( h );
2686         if( h->param.b_open_gop )
2687             h->frames.i_poc_last_open_gop = h->fenc->b_keyframe ? h->fenc->i_poc : -1;
2688     }
2689     else if( h->fenc->i_type == X264_TYPE_P )
2690     {
2691         i_nal_type    = NAL_SLICE;
2692         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
2693         h->sh.i_type = SLICE_TYPE_P;
2694         x264_reference_hierarchy_reset( h );
2695         h->frames.i_poc_last_open_gop = -1;
2696     }
2697     else if( h->fenc->i_type == X264_TYPE_BREF )
2698     {
2699         i_nal_type    = NAL_SLICE;
2700         i_nal_ref_idc = h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT ? NAL_PRIORITY_LOW : NAL_PRIORITY_HIGH;
2701         h->sh.i_type = SLICE_TYPE_B;
2702         x264_reference_hierarchy_reset( h );
2703     }
2704     else    /* B frame */
2705     {
2706         i_nal_type    = NAL_SLICE;
2707         i_nal_ref_idc = NAL_PRIORITY_DISPOSABLE;
2708         h->sh.i_type = SLICE_TYPE_B;
2709     }
2710
2711     h->fdec->i_type = h->fenc->i_type;
2712     h->fdec->i_frame = h->fenc->i_frame;
2713     h->fenc->b_kept_as_ref =
2714     h->fdec->b_kept_as_ref = i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE && h->param.i_keyint_max > 1;
2715
2716     h->fdec->i_pts = h->fenc->i_pts;
2717     if( h->frames.i_bframe_delay )
2718     {
2719         int64_t *prev_reordered_pts = thread_current->frames.i_prev_reordered_pts;
2720         h->fdec->i_dts = h->i_frame > h->frames.i_bframe_delay
2721                        ? prev_reordered_pts[ (h->i_frame - h->frames.i_bframe_delay) % h->frames.i_bframe_delay ]
2722                        : h->fenc->i_reordered_pts - h->frames.i_bframe_delay_time;
2723         prev_reordered_pts[ h->i_frame % h->frames.i_bframe_delay ] = h->fenc->i_reordered_pts;
2724     }
2725     else
2726         h->fdec->i_dts = h->fenc->i_reordered_pts;
2727     if( h->fenc->i_type == X264_TYPE_IDR )
2728         h->i_last_idr_pts = h->fdec->i_pts;
2729
2730     /* ------------------- Init                ----------------------------- */
2731     /* build ref list 0/1 */
2732     x264_reference_build_list( h, h->fdec->i_poc );
2733
2734     /* ---------------------- Write the bitstream -------------------------- */
2735     /* Init bitstream context */
2736     if( h->param.b_sliced_threads )
2737     {
2738         for( int i = 0; i < h->param.i_threads; i++ )
2739         {
2740             bs_init( &h->thread[i]->out.bs, h->thread[i]->out.p_bitstream, h->thread[i]->out.i_bitstream );
2741             h->thread[i]->out.i_nal = 0;
2742         }
2743     }
2744     else
2745     {
2746         bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
2747         h->out.i_nal = 0;
2748     }
2749
2750     if( h->param.b_aud )
2751     {
2752         int pic_type;
2753
2754         if( h->sh.i_type == SLICE_TYPE_I )
2755             pic_type = 0;
2756         else if( h->sh.i_type == SLICE_TYPE_P )
2757             pic_type = 1;
2758         else if( h->sh.i_type == SLICE_TYPE_B )
2759             pic_type = 2;
2760         else
2761             pic_type = 7;
2762
2763         x264_nal_start( h, NAL_AUD, NAL_PRIORITY_DISPOSABLE );
2764         bs_write( &h->out.bs, 3, pic_type );
2765         bs_rbsp_trailing( &h->out.bs );
2766         if( x264_nal_end( h ) )
2767             return -1;
2768         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2769     }
2770
2771     h->i_nal_type = i_nal_type;
2772     h->i_nal_ref_idc = i_nal_ref_idc;
2773
2774     if( h->param.b_intra_refresh )
2775     {
2776         if( IS_X264_TYPE_I( h->fenc->i_type ) )
2777         {
2778             h->fdec->i_frames_since_pir = 0;
2779             h->b_queued_intra_refresh = 0;
2780             /* PIR is currently only supported with ref == 1, so any intra frame effectively refreshes
2781              * the whole frame and counts as an intra refresh. */
2782             h->fdec->f_pir_position = h->mb.i_mb_width;
2783         }
2784         else if( h->fenc->i_type == X264_TYPE_P )
2785         {
2786             int pocdiff = (h->fdec->i_poc - h->fref[0][0]->i_poc)/2;
2787             float increment = X264_MAX( ((float)h->mb.i_mb_width-1) / h->param.i_keyint_max, 1 );
2788             h->fdec->f_pir_position = h->fref[0][0]->f_pir_position;
2789             h->fdec->i_frames_since_pir = h->fref[0][0]->i_frames_since_pir + pocdiff;
2790             if( h->fdec->i_frames_since_pir >= h->param.i_keyint_max ||
2791                 (h->b_queued_intra_refresh && h->fdec->f_pir_position + 0.5 >= h->mb.i_mb_width) )
2792             {
2793                 h->fdec->f_pir_position = 0;
2794                 h->fdec->i_frames_since_pir = 0;
2795                 h->b_queued_intra_refresh = 0;
2796                 h->fenc->b_keyframe = 1;
2797             }
2798             h->fdec->i_pir_start_col = h->fdec->f_pir_position+0.5;
2799             h->fdec->f_pir_position += increment * pocdiff;
2800             h->fdec->i_pir_end_col = h->fdec->f_pir_position+0.5;
2801             /* If our intra refresh has reached the right side of the frame, we're done. */
2802             if( h->fdec->i_pir_end_col >= h->mb.i_mb_width - 1 )
2803                 h->fdec->f_pir_position = h->mb.i_mb_width;
2804         }
2805     }
2806
2807     if( h->fenc->b_keyframe )
2808     {
2809         /* Write SPS and PPS */
2810         if( h->param.b_repeat_headers )
2811         {
2812             /* generate sequence parameters */
2813             x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
2814             x264_sps_write( &h->out.bs, h->sps );
2815             if( x264_nal_end( h ) )
2816                 return -1;
2817             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
2818
2819             /* generate picture parameters */
2820             x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
2821             x264_pps_write( &h->out.bs, h->pps );
2822             if( x264_nal_end( h ) )
2823                 return -1;
2824             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
2825         }
2826
2827         /* buffering period sei is written in x264_encoder_frame_end */
2828     }
2829
2830     /* write extra sei */
2831     for( int i = 0; i < h->fenc->extra_sei.num_payloads; i++ )
2832     {
2833         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2834         x264_sei_write( &h->out.bs, h->fenc->extra_sei.payloads[i].payload, h->fenc->extra_sei.payloads[i].payload_size,
2835                         h->fenc->extra_sei.payloads[i].payload_type );
2836         if( x264_nal_end( h ) )
2837             return -1;
2838         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2839         if( h->fenc->extra_sei.sei_free && h->fenc->extra_sei.payloads[i].payload )
2840             h->fenc->extra_sei.sei_free( h->fenc->extra_sei.payloads[i].payload );
2841     }
2842
2843     if( h->fenc->extra_sei.sei_free && h->fenc->extra_sei.payloads )
2844         h->fenc->extra_sei.sei_free( h->fenc->extra_sei.payloads );
2845
2846     if( h->fenc->b_keyframe )
2847     {
2848         if( h->param.b_repeat_headers && h->fenc->i_frame == 0 )
2849         {
2850             /* identify ourself */
2851             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2852             if( x264_sei_version_write( h, &h->out.bs ) )
2853                 return -1;
2854             if( x264_nal_end( h ) )
2855                 return -1;
2856             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2857         }
2858
2859         if( h->fenc->i_type != X264_TYPE_IDR )
2860         {
2861             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;
2862             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2863             x264_sei_recovery_point_write( h, &h->out.bs, time_to_recovery );
2864             if( x264_nal_end( h ) )
2865                 return -1;
2866             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2867         }
2868
2869         if ( h->param.i_frame_packing >= 0 )
2870         {
2871             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2872             x264_sei_frame_packing_write( h, &h->out.bs );
2873             if( x264_nal_end( h ) )
2874                 return -1;
2875             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2876         }
2877     }
2878
2879     /* generate sei pic timing */
2880     if( h->sps->vui.b_pic_struct_present || h->sps->vui.b_nal_hrd_parameters_present )
2881     {
2882         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2883         x264_sei_pic_timing_write( h, &h->out.bs );
2884         if( x264_nal_end( h ) )
2885             return -1;
2886         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2887     }
2888
2889     /* As required by Blu-ray. */
2890     if( !IS_X264_TYPE_B( h->fenc->i_type ) && h->b_sh_backup )
2891     {
2892         h->b_sh_backup = 0;
2893         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2894         x264_sei_dec_ref_pic_marking_write( h, &h->out.bs );
2895         if( x264_nal_end( h ) )
2896             return -1;
2897         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2898     }
2899
2900     if( h->fenc->b_keyframe && h->param.b_intra_refresh )
2901         h->i_cpb_delay_pir_offset = h->fenc->i_cpb_delay;
2902
2903     /* Init the rate control */
2904     /* FIXME: Include slice header bit cost. */
2905     x264_ratecontrol_start( h, h->fenc->i_qpplus1, overhead*8 );
2906     i_global_qp = x264_ratecontrol_qp( h );
2907
2908     pic_out->i_qpplus1 =
2909     h->fdec->i_qpplus1 = i_global_qp + 1;
2910
2911     if( h->param.rc.b_stat_read && h->sh.i_type != SLICE_TYPE_I )
2912     {
2913         x264_reference_build_list_optimal( h );
2914         x264_reference_check_reorder( h );
2915     }
2916
2917     if( h->i_ref[0] )
2918         h->fdec->i_poc_l0ref0 = h->fref[0][0]->i_poc;
2919
2920     /* ------------------------ Create slice header  ----------------------- */
2921     x264_slice_init( h, i_nal_type, i_global_qp );
2922
2923     /*------------------------- Weights -------------------------------------*/
2924     if( h->sh.i_type == SLICE_TYPE_B )
2925         x264_macroblock_bipred_init( h );
2926
2927     x264_weighted_pred_init( h );
2928
2929     if( i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE )
2930         h->i_frame_num++;
2931
2932     /* Write frame */
2933     h->i_threadslice_start = 0;
2934     h->i_threadslice_end = h->mb.i_mb_height;
2935     if( h->i_thread_frames > 1 )
2936     {
2937         x264_threadpool_run( h->threadpool, (void*)x264_slices_write, h );
2938         h->b_thread_active = 1;
2939     }
2940     else if( h->param.b_sliced_threads )
2941     {
2942         if( x264_threaded_slices_write( h ) )
2943             return -1;
2944     }
2945     else
2946         if( (intptr_t)x264_slices_write( h ) )
2947             return -1;
2948
2949     return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
2950 }
2951
2952 static int x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
2953                                    x264_nal_t **pp_nal, int *pi_nal,
2954                                    x264_picture_t *pic_out )
2955 {
2956     char psz_message[80];
2957
2958     if( h->b_thread_active )
2959     {
2960         h->b_thread_active = 0;
2961         if( (intptr_t)x264_threadpool_wait( h->threadpool, h ) )
2962             return -1;
2963     }
2964     if( !h->out.i_nal )
2965     {
2966         pic_out->i_type = X264_TYPE_AUTO;
2967         return 0;
2968     }
2969
2970     x264_emms();
2971     /* generate sei buffering period and insert it into place */
2972     if( h->fenc->b_keyframe && h->sps->vui.b_nal_hrd_parameters_present )
2973     {
2974         x264_hrd_fullness( h );
2975         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2976         x264_sei_buffering_period_write( h, &h->out.bs );
2977         if( x264_nal_end( h ) )
2978            return -1;
2979         /* buffering period sei must follow AUD, SPS and PPS and precede all other SEIs */
2980         int idx = 0;
2981         while( h->out.nal[idx].i_type == NAL_AUD ||
2982                h->out.nal[idx].i_type == NAL_SPS ||
2983                h->out.nal[idx].i_type == NAL_PPS )
2984             idx++;
2985         x264_nal_t nal_tmp = h->out.nal[h->out.i_nal-1];
2986         memmove( &h->out.nal[idx+1], &h->out.nal[idx], (h->out.i_nal-idx-1)*sizeof(x264_nal_t) );
2987         h->out.nal[idx] = nal_tmp;
2988     }
2989
2990     int frame_size = x264_encoder_encapsulate_nals( h, 0 );
2991     if( frame_size < 0 )
2992         return -1;
2993
2994     /* Set output picture properties */
2995     pic_out->i_type = h->fenc->i_type;
2996
2997     pic_out->b_keyframe = h->fenc->b_keyframe;
2998     pic_out->i_pic_struct = h->fenc->i_pic_struct;
2999
3000     pic_out->i_pts = h->fdec->i_pts;
3001     pic_out->i_dts = h->fdec->i_dts;
3002
3003     if( pic_out->i_pts < pic_out->i_dts )
3004         x264_log( h, X264_LOG_WARNING, "invalid DTS: PTS is less than DTS\n" );
3005
3006     pic_out->img.i_csp = X264_CSP_NV12;
3007 #if HIGH_BIT_DEPTH
3008     pic_out->img.i_csp |= X264_CSP_HIGH_DEPTH;
3009 #endif
3010     pic_out->img.i_plane = h->fdec->i_plane;
3011     for( int i = 0; i < 2; i++ )
3012     {
3013         pic_out->img.i_stride[i] = h->fdec->i_stride[i] * sizeof(pixel);
3014         pic_out->img.plane[i] = (uint8_t*)h->fdec->plane[i];
3015     }
3016
3017     x264_frame_push_unused( thread_current, h->fenc );
3018
3019     /* ---------------------- Update encoder state ------------------------- */
3020
3021     /* update rc */
3022     int filler = 0;
3023     if( x264_ratecontrol_end( h, frame_size * 8, &filler ) < 0 )
3024         return -1;
3025
3026     pic_out->hrd_timing = h->fenc->hrd_timing;
3027
3028     while( filler > 0 )
3029     {
3030         int f, overhead;
3031         overhead = (FILLER_OVERHEAD - h->param.b_annexb);
3032         if( h->param.i_slice_max_size && filler > h->param.i_slice_max_size )
3033         {
3034             int next_size = filler - h->param.i_slice_max_size;
3035             int overflow = X264_MAX( overhead - next_size, 0 );
3036             f = h->param.i_slice_max_size - overhead - overflow;
3037         }
3038         else
3039             f = X264_MAX( 0, filler - overhead );
3040
3041         x264_nal_start( h, NAL_FILLER, NAL_PRIORITY_DISPOSABLE );
3042         x264_filler_write( h, &h->out.bs, f );
3043         if( x264_nal_end( h ) )
3044             return -1;
3045         int total_size = x264_encoder_encapsulate_nals( h, h->out.i_nal-1 );
3046         if( total_size < 0 )
3047             return -1;
3048         frame_size += total_size;
3049         filler -= total_size;
3050     }
3051
3052     /* End bitstream, set output  */
3053     *pi_nal = h->out.i_nal;
3054     *pp_nal = h->out.nal;
3055
3056     h->out.i_nal = 0;
3057
3058     x264_noise_reduction_update( h );
3059
3060     /* ---------------------- Compute/Print statistics --------------------- */
3061     x264_thread_sync_stat( h, h->thread[0] );
3062
3063     /* Slice stat */
3064     h->stat.i_frame_count[h->sh.i_type]++;
3065     h->stat.i_frame_size[h->sh.i_type] += frame_size;
3066     h->stat.f_frame_qp[h->sh.i_type] += h->fdec->f_qp_avg_aq;
3067
3068     for( int i = 0; i < X264_MBTYPE_MAX; i++ )
3069         h->stat.i_mb_count[h->sh.i_type][i] += h->stat.frame.i_mb_count[i];
3070     for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
3071         h->stat.i_mb_partition[h->sh.i_type][i] += h->stat.frame.i_mb_partition[i];
3072     for( int i = 0; i < 2; i++ )
3073         h->stat.i_mb_count_8x8dct[i] += h->stat.frame.i_mb_count_8x8dct[i];
3074     for( int i = 0; i < 6; i++ )
3075         h->stat.i_mb_cbp[i] += h->stat.frame.i_mb_cbp[i];
3076     for( int i = 0; i < 4; i++ )
3077         for( int j = 0; j < 13; j++ )
3078             h->stat.i_mb_pred_mode[i][j] += h->stat.frame.i_mb_pred_mode[i][j];
3079     if( h->sh.i_type != SLICE_TYPE_I )
3080         for( int i_list = 0; i_list < 2; i_list++ )
3081             for( int i = 0; i < X264_REF_MAX*2; i++ )
3082                 h->stat.i_mb_count_ref[h->sh.i_type][i_list][i] += h->stat.frame.i_mb_count_ref[i_list][i];
3083     for( int i = 0; i < 3; i++ )
3084         h->stat.i_mb_field[i] += h->stat.frame.i_mb_field[i];
3085     if( h->sh.i_type == SLICE_TYPE_P && h->param.analyse.i_weighted_pred >= X264_WEIGHTP_SIMPLE )
3086     {
3087         h->stat.i_wpred[0] += !!h->sh.weight[0][0].weightfn;
3088         h->stat.i_wpred[1] += !!h->sh.weight[0][1].weightfn || !!h->sh.weight[0][2].weightfn;
3089     }
3090     if( h->sh.i_type == SLICE_TYPE_B )
3091     {
3092         h->stat.i_direct_frames[ h->sh.b_direct_spatial_mv_pred ] ++;
3093         if( h->mb.b_direct_auto_write )
3094         {
3095             //FIXME somewhat arbitrary time constants
3096             if( h->stat.i_direct_score[0] + h->stat.i_direct_score[1] > h->mb.i_mb_count )
3097                 for( int i = 0; i < 2; i++ )
3098                     h->stat.i_direct_score[i] = h->stat.i_direct_score[i] * 9/10;
3099             for( int i = 0; i < 2; i++ )
3100                 h->stat.i_direct_score[i] += h->stat.frame.i_direct_score[i];
3101         }
3102     }
3103     else
3104         h->stat.i_consecutive_bframes[h->fenc->i_bframes]++;
3105
3106     psz_message[0] = '\0';
3107     double dur = h->fenc->f_duration;
3108     h->stat.f_frame_duration[h->sh.i_type] += dur;
3109     if( h->param.analyse.b_psnr )
3110     {
3111         int64_t ssd[3] =
3112         {
3113             h->stat.frame.i_ssd[0],
3114             h->stat.frame.i_ssd[1],
3115             h->stat.frame.i_ssd[2],
3116         };
3117
3118         h->stat.f_ssd_global[h->sh.i_type]   += dur * (ssd[0] + ssd[1] + ssd[2]);
3119         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 );
3120         h->stat.f_psnr_mean_y[h->sh.i_type]  += dur * x264_psnr( ssd[0], h->param.i_width * h->param.i_height );
3121         h->stat.f_psnr_mean_u[h->sh.i_type]  += dur * x264_psnr( ssd[1], h->param.i_width * h->param.i_height / 4 );
3122         h->stat.f_psnr_mean_v[h->sh.i_type]  += dur * x264_psnr( ssd[2], h->param.i_width * h->param.i_height / 4 );
3123
3124         snprintf( psz_message, 80, " PSNR Y:%5.2f U:%5.2f V:%5.2f",
3125                   x264_psnr( ssd[0], h->param.i_width * h->param.i_height ),
3126                   x264_psnr( ssd[1], h->param.i_width * h->param.i_height / 4),
3127                   x264_psnr( ssd[2], h->param.i_width * h->param.i_height / 4) );
3128     }
3129
3130     if( h->param.analyse.b_ssim )
3131     {
3132         double ssim_y = h->stat.frame.f_ssim
3133                       / (((h->param.i_width-6)>>2) * ((h->param.i_height-6)>>2));
3134         h->stat.f_ssim_mean_y[h->sh.i_type] += ssim_y * dur;
3135         snprintf( psz_message + strlen(psz_message), 80 - strlen(psz_message),
3136                   " SSIM Y:%.5f", ssim_y );
3137     }
3138     psz_message[79] = '\0';
3139
3140     x264_log( h, X264_LOG_DEBUG,
3141                   "frame=%4d QP=%.2f NAL=%d Slice:%c Poc:%-3d I:%-4d P:%-4d SKIP:%-4d size=%d bytes%s\n",
3142               h->i_frame,
3143               h->fdec->f_qp_avg_aq,
3144               h->i_nal_ref_idc,
3145               h->sh.i_type == SLICE_TYPE_I ? 'I' : (h->sh.i_type == SLICE_TYPE_P ? 'P' : 'B' ),
3146               h->fdec->i_poc,
3147               h->stat.frame.i_mb_count_i,
3148               h->stat.frame.i_mb_count_p,
3149               h->stat.frame.i_mb_count_skip,
3150               frame_size,
3151               psz_message );
3152
3153     // keep stats all in one place
3154     x264_thread_sync_stat( h->thread[0], h );
3155     // for the use of the next frame
3156     x264_thread_sync_stat( thread_current, h );
3157
3158 #ifdef DEBUG_MB_TYPE
3159 {
3160     static const char mb_chars[] = { 'i', 'i', 'I', 'C', 'P', '8', 'S',
3161         'D', '<', 'X', 'B', 'X', '>', 'B', 'B', 'B', 'B', '8', 'S' };
3162     for( int mb_xy = 0; mb_xy < h->mb.i_mb_width * h->mb.i_mb_height; mb_xy++ )
3163     {
3164         if( h->mb.type[mb_xy] < X264_MBTYPE_MAX && h->mb.type[mb_xy] >= 0 )
3165             fprintf( stderr, "%c ", mb_chars[ h->mb.type[mb_xy] ] );
3166         else
3167             fprintf( stderr, "? " );
3168
3169         if( (mb_xy+1) % h->mb.i_mb_width == 0 )
3170             fprintf( stderr, "\n" );
3171     }
3172 }
3173 #endif
3174
3175     /* Remove duplicates, must be done near the end as breaks h->fref0 array
3176      * by freeing some of its pointers. */
3177     for( int i = 0; i < h->i_ref[0]; i++ )
3178         if( h->fref[0][i] && h->fref[0][i]->b_duplicate )
3179         {
3180             x264_frame_push_blank_unused( h, h->fref[0][i] );
3181             h->fref[0][i] = 0;
3182         }
3183
3184     if( h->param.psz_dump_yuv )
3185         x264_frame_dump( h );
3186     x264_emms();
3187
3188     return frame_size;
3189 }
3190
3191 static void x264_print_intra( int64_t *i_mb_count, double i_count, int b_print_pcm, char *intra )
3192 {
3193     intra += sprintf( intra, "I16..4%s: %4.1f%% %4.1f%% %4.1f%%",
3194         b_print_pcm ? "..PCM" : "",
3195         i_mb_count[I_16x16]/ i_count,
3196         i_mb_count[I_8x8]  / i_count,
3197         i_mb_count[I_4x4]  / i_count );
3198     if( b_print_pcm )
3199         sprintf( intra, " %4.1f%%", i_mb_count[I_PCM]  / i_count );
3200 }
3201
3202 /****************************************************************************
3203  * x264_encoder_close:
3204  ****************************************************************************/
3205 void    x264_encoder_close  ( x264_t *h )
3206 {
3207     int64_t i_yuv_size = 3 * h->param.i_width * h->param.i_height / 2;
3208     int64_t i_mb_count_size[2][7] = {{0}};
3209     char buf[200];
3210     int b_print_pcm = h->stat.i_mb_count[SLICE_TYPE_I][I_PCM]
3211                    || h->stat.i_mb_count[SLICE_TYPE_P][I_PCM]
3212                    || h->stat.i_mb_count[SLICE_TYPE_B][I_PCM];
3213
3214     x264_lookahead_delete( h );
3215
3216     if( h->param.i_threads > 1 )
3217         x264_threadpool_delete( h->threadpool );
3218     if( h->i_thread_frames > 1 )
3219     {
3220         for( int i = 0; i < h->i_thread_frames; i++ )
3221             if( h->thread[i]->b_thread_active )
3222             {
3223                 assert( h->thread[i]->fenc->i_reference_count == 1 );
3224                 x264_frame_delete( h->thread[i]->fenc );
3225             }
3226
3227         x264_t *thread_prev = h->thread[h->i_thread_phase];
3228         x264_thread_sync_ratecontrol( h, thread_prev, h );
3229         x264_thread_sync_ratecontrol( thread_prev, thread_prev, h );
3230         h->i_frame = thread_prev->i_frame + 1 - h->i_thread_frames;
3231     }
3232     h->i_frame++;
3233
3234     /* Slices used and PSNR */
3235     for( int i = 0; i < 3; i++ )
3236     {
3237         static const uint8_t slice_order[] = { SLICE_TYPE_I, SLICE_TYPE_P, SLICE_TYPE_B };
3238         int i_slice = slice_order[i];
3239
3240         if( h->stat.i_frame_count[i_slice] > 0 )
3241         {
3242             int i_count = h->stat.i_frame_count[i_slice];
3243             double dur =  h->stat.f_frame_duration[i_slice];
3244             if( h->param.analyse.b_psnr )
3245             {
3246                 x264_log( h, X264_LOG_INFO,
3247                           "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",
3248                           slice_type_to_char[i_slice],
3249                           i_count,
3250                           h->stat.f_frame_qp[i_slice] / i_count,
3251                           (double)h->stat.i_frame_size[i_slice] / i_count,
3252                           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,
3253                           h->stat.f_psnr_average[i_slice] / dur,
3254                           x264_psnr( h->stat.f_ssd_global[i_slice], dur * i_yuv_size ) );
3255             }
3256             else
3257             {
3258                 x264_log( h, X264_LOG_INFO,
3259                           "frame %c:%-5d Avg QP:%5.2f  size:%6.0f\n",
3260                           slice_type_to_char[i_slice],
3261                           i_count,
3262                           h->stat.f_frame_qp[i_slice] / i_count,
3263                           (double)h->stat.i_frame_size[i_slice] / i_count );
3264             }
3265         }
3266     }
3267     if( h->param.i_bframe && h->stat.i_frame_count[SLICE_TYPE_B] )
3268     {
3269         char *p = buf;
3270         int den = 0;
3271         // weight by number of frames (including the I/P-frames) that are in a sequence of N B-frames
3272         for( int i = 0; i <= h->param.i_bframe; i++ )
3273             den += (i+1) * h->stat.i_consecutive_bframes[i];
3274         for( int i = 0; i <= h->param.i_bframe; i++ )
3275             p += sprintf( p, " %4.1f%%", 100. * (i+1) * h->stat.i_consecutive_bframes[i] / den );
3276         x264_log( h, X264_LOG_INFO, "consecutive B-frames:%s\n", buf );
3277     }
3278
3279     for( int i_type = 0; i_type < 2; i_type++ )
3280         for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
3281         {
3282             if( i == D_DIRECT_8x8 ) continue; /* direct is counted as its own type */
3283             i_mb_count_size[i_type][x264_mb_partition_pixel_table[i]] += h->stat.i_mb_partition[i_type][i];
3284         }
3285
3286     /* MB types used */
3287     if( h->stat.i_frame_count[SLICE_TYPE_I] > 0 )
3288     {
3289         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_I];
3290         double i_count = h->stat.i_frame_count[SLICE_TYPE_I] * h->mb.i_mb_count / 100.0;
3291         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
3292         x264_log( h, X264_LOG_INFO, "mb I  %s\n", buf );
3293     }
3294     if( h->stat.i_frame_count[SLICE_TYPE_P] > 0 )
3295     {
3296         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_P];
3297         double i_count = h->stat.i_frame_count[SLICE_TYPE_P] * h->mb.i_mb_count / 100.0;
3298         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_P];
3299         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
3300         x264_log( h, X264_LOG_INFO,
3301                   "mb P  %s  P16..4: %4.1f%% %4.1f%% %4.1f%% %4.1f%% %4.1f%%    skip:%4.1f%%\n",
3302                   buf,
3303                   i_mb_size[PIXEL_16x16] / (i_count*4),
3304                   (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
3305                   i_mb_size[PIXEL_8x8] / (i_count*4),
3306                   (i_mb_size[PIXEL_8x4] + i_mb_size[PIXEL_4x8]) / (i_count*4),
3307                   i_mb_size[PIXEL_4x4] / (i_count*4),
3308                   i_mb_count[P_SKIP] / i_count );
3309     }
3310     if( h->stat.i_frame_count[SLICE_TYPE_B] > 0 )
3311     {
3312         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_B];
3313         double i_count = h->stat.i_frame_count[SLICE_TYPE_B] * h->mb.i_mb_count / 100.0;
3314         double i_mb_list_count;
3315         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_B];
3316         int64_t list_count[3] = {0}; /* 0 == L0, 1 == L1, 2 == BI */
3317         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
3318         for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
3319             for( int j = 0; j < 2; j++ )
3320             {
3321                 int l0 = x264_mb_type_list_table[i][0][j];
3322                 int l1 = x264_mb_type_list_table[i][1][j];
3323                 if( l0 || l1 )
3324                     list_count[l1+l0*l1] += h->stat.i_mb_count[SLICE_TYPE_B][i] * 2;
3325             }
3326         list_count[0] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L0_8x8];
3327         list_count[1] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L1_8x8];
3328         list_count[2] += h->stat.i_mb_partition[SLICE_TYPE_B][D_BI_8x8];
3329         i_mb_count[B_DIRECT] += (h->stat.i_mb_partition[SLICE_TYPE_B][D_DIRECT_8x8]+2)/4;
3330         i_mb_list_count = (list_count[0] + list_count[1] + list_count[2]) / 100.0;
3331         x264_log( h, X264_LOG_INFO,
3332                   "mb B  %s  B16..8: %4.1f%% %4.1f%% %4.1f%%  direct:%4.1f%%  skip:%4.1f%%  L0:%4.1f%% L1:%4.1f%% BI:%4.1f%%\n",
3333                   buf,
3334                   i_mb_size[PIXEL_16x16] / (i_count*4),
3335                   (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
3336                   i_mb_size[PIXEL_8x8] / (i_count*4),
3337                   i_mb_count[B_DIRECT] / i_count,
3338                   i_mb_count[B_SKIP]   / i_count,
3339                   list_count[0] / i_mb_list_count,
3340                   list_count[1] / i_mb_list_count,
3341                   list_count[2] / i_mb_list_count );
3342     }
3343
3344     x264_ratecontrol_summary( h );
3345
3346     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 )
3347     {
3348 #define SUM3(p) (p[SLICE_TYPE_I] + p[SLICE_TYPE_P] + p[SLICE_TYPE_B])
3349 #define SUM3b(p,o) (p[SLICE_TYPE_I][o] + p[SLICE_TYPE_P][o] + p[SLICE_TYPE_B][o])
3350         int64_t i_i8x8 = SUM3b( h->stat.i_mb_count, I_8x8 );
3351         int64_t i_intra = i_i8x8 + SUM3b( h->stat.i_mb_count, I_4x4 )
3352                                  + SUM3b( h->stat.i_mb_count, I_16x16 );
3353         int64_t i_all_intra = i_intra + SUM3b( h->stat.i_mb_count, I_PCM);
3354         int64_t i_skip = SUM3b( h->stat.i_mb_count, P_SKIP )
3355                        + SUM3b( h->stat.i_mb_count, B_SKIP );
3356         const int i_count = h->stat.i_frame_count[SLICE_TYPE_I] +
3357                             h->stat.i_frame_count[SLICE_TYPE_P] +
3358                             h->stat.i_frame_count[SLICE_TYPE_B];
3359         int64_t i_mb_count = (int64_t)i_count * h->mb.i_mb_count;
3360         int64_t i_inter = i_mb_count - i_skip - i_intra;
3361         const double duration = h->stat.f_frame_duration[SLICE_TYPE_I] +
3362                                 h->stat.f_frame_duration[SLICE_TYPE_P] +
3363                                 h->stat.f_frame_duration[SLICE_TYPE_B];
3364         float f_bitrate = SUM3(h->stat.i_frame_size) / duration / 125;
3365
3366         if( PARAM_INTERLACED )
3367         {
3368             char *fieldstats = buf;
3369             fieldstats[0] = 0;
3370             if( i_inter )
3371                 fieldstats += sprintf( fieldstats, " inter:%.1f%%", h->stat.i_mb_field[1] * 100.0 / i_inter );
3372             if( i_skip )
3373                 fieldstats += sprintf( fieldstats, " skip:%.1f%%", h->stat.i_mb_field[2] * 100.0 / i_skip );
3374             x264_log( h, X264_LOG_INFO, "field mbs: intra: %.1f%%%s\n",
3375                       h->stat.i_mb_field[0] * 100.0 / i_intra, buf );
3376         }
3377
3378         if( h->pps->b_transform_8x8_mode )
3379         {
3380             buf[0] = 0;
3381             if( h->stat.i_mb_count_8x8dct[0] )
3382                 sprintf( buf, " inter:%.1f%%", 100. * h->stat.i_mb_count_8x8dct[1] / h->stat.i_mb_count_8x8dct[0] );
3383             x264_log( h, X264_LOG_INFO, "8x8 transform intra:%.1f%%%s\n", 100. * i_i8x8 / i_intra, buf );
3384         }
3385
3386         if( (h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO ||
3387             (h->stat.i_direct_frames[0] && h->stat.i_direct_frames[1]))
3388             && h->stat.i_frame_count[SLICE_TYPE_B] )
3389         {
3390             x264_log( h, X264_LOG_INFO, "direct mvs  spatial:%.1f%% temporal:%.1f%%\n",
3391                       h->stat.i_direct_frames[1] * 100. / h->stat.i_frame_count[SLICE_TYPE_B],
3392                       h->stat.i_direct_frames[0] * 100. / h->stat.i_frame_count[SLICE_TYPE_B] );
3393         }
3394
3395         buf[0] = 0;
3396         if( i_mb_count != i_all_intra )
3397             sprintf( buf, " inter: %.1f%% %.1f%% %.1f%%",
3398                      h->stat.i_mb_cbp[1] * 100.0 / ((i_mb_count - i_all_intra)*4),
3399                      h->stat.i_mb_cbp[3] * 100.0 / ((i_mb_count - i_all_intra)  ),
3400                      h->stat.i_mb_cbp[5] * 100.0 / ((i_mb_count - i_all_intra)) );
3401         x264_log( h, X264_LOG_INFO, "coded y,uvDC,uvAC intra: %.1f%% %.1f%% %.1f%%%s\n",
3402                   h->stat.i_mb_cbp[0] * 100.0 / (i_all_intra*4),
3403                   h->stat.i_mb_cbp[2] * 100.0 / (i_all_intra  ),
3404                   h->stat.i_mb_cbp[4] * 100.0 / (i_all_intra  ), buf );
3405
3406         int64_t fixed_pred_modes[4][9] = {{0}};
3407         int64_t sum_pred_modes[4] = {0};
3408         for( int i = 0; i <= I_PRED_16x16_DC_128; i++ )
3409         {
3410             fixed_pred_modes[0][x264_mb_pred_mode16x16_fix[i]] += h->stat.i_mb_pred_mode[0][i];
3411             sum_pred_modes[0] += h->stat.i_mb_pred_mode[0][i];
3412         }
3413         if( sum_pred_modes[0] )
3414             x264_log( h, X264_LOG_INFO, "i16 v,h,dc,p: %2.0f%% %2.0f%% %2.0f%% %2.0f%%\n",
3415                       fixed_pred_modes[0][0] * 100.0 / sum_pred_modes[0],
3416                       fixed_pred_modes[0][1] * 100.0 / sum_pred_modes[0],
3417                       fixed_pred_modes[0][2] * 100.0 / sum_pred_modes[0],
3418                       fixed_pred_modes[0][3] * 100.0 / sum_pred_modes[0] );
3419         for( int i = 1; i <= 2; i++ )
3420         {
3421             for( int j = 0; j <= I_PRED_8x8_DC_128; j++ )
3422             {
3423                 fixed_pred_modes[i][x264_mb_pred_mode4x4_fix(j)] += h->stat.i_mb_pred_mode[i][j];
3424                 sum_pred_modes[i] += h->stat.i_mb_pred_mode[i][j];
3425             }
3426             if( sum_pred_modes[i] )
3427                 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,
3428                           fixed_pred_modes[i][0] * 100.0 / sum_pred_modes[i],
3429                           fixed_pred_modes[i][1] * 100.0 / sum_pred_modes[i],
3430                           fixed_pred_modes[i][2] * 100.0 / sum_pred_modes[i],
3431                           fixed_pred_modes[i][3] * 100.0 / sum_pred_modes[i],
3432                           fixed_pred_modes[i][4] * 100.0 / sum_pred_modes[i],
3433                           fixed_pred_modes[i][5] * 100.0 / sum_pred_modes[i],
3434                           fixed_pred_modes[i][6] * 100.0 / sum_pred_modes[i],
3435                           fixed_pred_modes[i][7] * 100.0 / sum_pred_modes[i],
3436                           fixed_pred_modes[i][8] * 100.0 / sum_pred_modes[i] );
3437         }
3438         for( int i = 0; i <= I_PRED_CHROMA_DC_128; i++ )
3439         {
3440             fixed_pred_modes[3][x264_mb_pred_mode8x8c_fix[i]] += h->stat.i_mb_pred_mode[3][i];
3441             sum_pred_modes[3] += h->stat.i_mb_pred_mode[3][i];
3442         }
3443         if( sum_pred_modes[3] )
3444             x264_log( h, X264_LOG_INFO, "i8c dc,h,v,p: %2.0f%% %2.0f%% %2.0f%% %2.0f%%\n",
3445                       fixed_pred_modes[3][0] * 100.0 / sum_pred_modes[3],
3446                       fixed_pred_modes[3][1] * 100.0 / sum_pred_modes[3],
3447                       fixed_pred_modes[3][2] * 100.0 / sum_pred_modes[3],
3448                       fixed_pred_modes[3][3] * 100.0 / sum_pred_modes[3] );
3449
3450         if( h->param.analyse.i_weighted_pred >= X264_WEIGHTP_SIMPLE && h->stat.i_frame_count[SLICE_TYPE_P] > 0 )
3451             x264_log( h, X264_LOG_INFO, "Weighted P-Frames: Y:%.1f%% UV:%.1f%%\n",
3452                       h->stat.i_wpred[0] * 100.0 / h->stat.i_frame_count[SLICE_TYPE_P],
3453                       h->stat.i_wpred[1] * 100.0 / h->stat.i_frame_count[SLICE_TYPE_P] );
3454
3455         for( int i_list = 0; i_list < 2; i_list++ )
3456             for( int i_slice = 0; i_slice < 2; i_slice++ )
3457             {
3458                 char *p = buf;
3459                 int64_t i_den = 0;
3460                 int i_max = 0;
3461                 for( int i = 0; i < X264_REF_MAX*2; i++ )
3462                     if( h->stat.i_mb_count_ref[i_slice][i_list][i] )
3463                     {
3464                         i_den += h->stat.i_mb_count_ref[i_slice][i_list][i];
3465                         i_max = i;
3466                     }
3467                 if( i_max == 0 )
3468                     continue;
3469                 for( int i = 0; i <= i_max; i++ )
3470                     p += sprintf( p, " %4.1f%%", 100. * h->stat.i_mb_count_ref[i_slice][i_list][i] / i_den );
3471                 x264_log( h, X264_LOG_INFO, "ref %c L%d:%s\n", "PB"[i_slice], i_list, buf );
3472             }
3473
3474         if( h->param.analyse.b_ssim )
3475         {
3476             float ssim = SUM3( h->stat.f_ssim_mean_y ) / duration;
3477             x264_log( h, X264_LOG_INFO, "SSIM Mean Y:%.7f (%6.3fdb)\n", ssim, x264_ssim( ssim ) );
3478         }
3479         if( h->param.analyse.b_psnr )
3480         {
3481             x264_log( h, X264_LOG_INFO,
3482                       "PSNR Mean Y:%6.3f U:%6.3f V:%6.3f Avg:%6.3f Global:%6.3f kb/s:%.2f\n",
3483                       SUM3( h->stat.f_psnr_mean_y ) / duration,
3484                       SUM3( h->stat.f_psnr_mean_u ) / duration,
3485                       SUM3( h->stat.f_psnr_mean_v ) / duration,
3486                       SUM3( h->stat.f_psnr_average ) / duration,
3487                       x264_psnr( SUM3( h->stat.f_ssd_global ), duration * i_yuv_size ),
3488                       f_bitrate );
3489         }
3490         else
3491             x264_log( h, X264_LOG_INFO, "kb/s:%.2f\n", f_bitrate );
3492     }
3493
3494     /* rc */
3495     x264_ratecontrol_delete( h );
3496
3497     /* param */
3498     if( h->param.rc.psz_stat_out )
3499         free( h->param.rc.psz_stat_out );
3500     if( h->param.rc.psz_stat_in )
3501         free( h->param.rc.psz_stat_in );
3502
3503     x264_cqm_delete( h );
3504     x264_free( h->nal_buffer );
3505     x264_analyse_free_costs( h );
3506
3507     if( h->i_thread_frames > 1)
3508         h = h->thread[h->i_thread_phase];
3509
3510     /* frames */
3511     x264_frame_delete_list( h->frames.unused[0] );
3512     x264_frame_delete_list( h->frames.unused[1] );
3513     x264_frame_delete_list( h->frames.current );
3514     x264_frame_delete_list( h->frames.blank_unused );
3515
3516     h = h->thread[0];
3517
3518     for( int i = 0; i < h->i_thread_frames; i++ )
3519         if( h->thread[i]->b_thread_active )
3520             for( int j = 0; j < h->thread[i]->i_ref[0]; j++ )
3521                 if( h->thread[i]->fref[0][j] && h->thread[i]->fref[0][j]->b_duplicate )
3522                     x264_frame_delete( h->thread[i]->fref[0][j] );
3523
3524     for( int i = h->param.i_threads - 1; i >= 0; i-- )
3525     {
3526         x264_frame_t **frame;
3527
3528         if( !h->param.b_sliced_threads || i == 0 )
3529         {
3530             for( frame = h->thread[i]->frames.reference; *frame; frame++ )
3531             {
3532                 assert( (*frame)->i_reference_count > 0 );
3533                 (*frame)->i_reference_count--;
3534                 if( (*frame)->i_reference_count == 0 )
3535                     x264_frame_delete( *frame );
3536             }
3537             frame = &h->thread[i]->fdec;
3538             if( *frame )
3539             {
3540                 assert( (*frame)->i_reference_count > 0 );
3541                 (*frame)->i_reference_count--;
3542                 if( (*frame)->i_reference_count == 0 )
3543                     x264_frame_delete( *frame );
3544             }
3545             x264_macroblock_cache_free( h->thread[i] );
3546         }
3547         x264_macroblock_thread_free( h->thread[i], 0 );
3548         x264_free( h->thread[i]->out.p_bitstream );
3549         x264_free( h->thread[i]->out.nal);
3550         x264_free( h->thread[i] );
3551     }