Eliminate the possibility of CAVLC level code overflow
[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 = h->param.b_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         assert( sh->i_first_mb % (2*sh->sps->i_mb_width) == 0 );
185         bs_write_ue( s, sh->i_first_mb >> 1 );
186     }
187     else
188         bs_write_ue( s, sh->i_first_mb );
189
190     bs_write_ue( s, sh->i_type + 5 );   /* same type things */
191     bs_write_ue( s, sh->i_pps_id );
192     bs_write( s, sh->sps->i_log2_max_frame_num, sh->i_frame_num & ((1<<sh->sps->i_log2_max_frame_num)-1) );
193
194     if( !sh->sps->b_frame_mbs_only )
195     {
196         bs_write1( s, sh->b_field_pic );
197         if( sh->b_field_pic )
198             bs_write1( s, sh->b_bottom_field );
199     }
200
201     if( sh->i_idr_pic_id >= 0 ) /* NAL IDR */
202         bs_write_ue( s, sh->i_idr_pic_id );
203
204     if( sh->sps->i_poc_type == 0 )
205     {
206         bs_write( s, sh->sps->i_log2_max_poc_lsb, sh->i_poc & ((1<<sh->sps->i_log2_max_poc_lsb)-1) );
207         if( sh->pps->b_pic_order && !sh->b_field_pic )
208             bs_write_se( s, sh->i_delta_poc_bottom );
209     }
210     else if( sh->sps->i_poc_type == 1 && !sh->sps->b_delta_pic_order_always_zero )
211     {
212         bs_write_se( s, sh->i_delta_poc[0] );
213         if( sh->pps->b_pic_order && !sh->b_field_pic )
214             bs_write_se( s, sh->i_delta_poc[1] );
215     }
216
217     if( sh->pps->b_redundant_pic_cnt )
218         bs_write_ue( s, sh->i_redundant_pic_cnt );
219
220     if( sh->i_type == SLICE_TYPE_B )
221         bs_write1( s, sh->b_direct_spatial_mv_pred );
222
223     if( sh->i_type == SLICE_TYPE_P || sh->i_type == SLICE_TYPE_B )
224     {
225         bs_write1( s, sh->b_num_ref_idx_override );
226         if( sh->b_num_ref_idx_override )
227         {
228             bs_write_ue( s, sh->i_num_ref_idx_l0_active - 1 );
229             if( sh->i_type == SLICE_TYPE_B )
230                 bs_write_ue( s, sh->i_num_ref_idx_l1_active - 1 );
231         }
232     }
233
234     /* ref pic list reordering */
235     if( sh->i_type != SLICE_TYPE_I )
236     {
237         bs_write1( s, sh->b_ref_pic_list_reordering[0] );
238         if( sh->b_ref_pic_list_reordering[0] )
239         {
240             for( int i = 0; i < sh->i_num_ref_idx_l0_active; i++ )
241             {
242                 bs_write_ue( s, sh->ref_pic_list_order[0][i].idc );
243                 bs_write_ue( s, sh->ref_pic_list_order[0][i].arg );
244             }
245             bs_write_ue( s, 3 );
246         }
247     }
248     if( sh->i_type == SLICE_TYPE_B )
249     {
250         bs_write1( s, sh->b_ref_pic_list_reordering[1] );
251         if( sh->b_ref_pic_list_reordering[1] )
252         {
253             for( int i = 0; i < sh->i_num_ref_idx_l1_active; i++ )
254             {
255                 bs_write_ue( s, sh->ref_pic_list_order[1][i].idc );
256                 bs_write_ue( s, sh->ref_pic_list_order[1][i].arg );
257             }
258             bs_write_ue( s, 3 );
259         }
260     }
261
262     if( sh->pps->b_weighted_pred && sh->i_type == SLICE_TYPE_P )
263     {
264         /* pred_weight_table() */
265         bs_write_ue( s, sh->weight[0][0].i_denom );
266         bs_write_ue( s, sh->weight[0][1].i_denom );
267         for( int i = 0; i < sh->i_num_ref_idx_l0_active; i++ )
268         {
269             int luma_weight_l0_flag = !!sh->weight[i][0].weightfn;
270             int chroma_weight_l0_flag = !!sh->weight[i][1].weightfn || !!sh->weight[i][2].weightfn;
271             bs_write1( s, luma_weight_l0_flag );
272             if( luma_weight_l0_flag )
273             {
274                 bs_write_se( s, sh->weight[i][0].i_scale );
275                 bs_write_se( s, sh->weight[i][0].i_offset );
276             }
277             bs_write1( s, chroma_weight_l0_flag );
278             if( chroma_weight_l0_flag )
279             {
280                 for( int j = 1; j < 3; j++ )
281                 {
282                     bs_write_se( s, sh->weight[i][j].i_scale );
283                     bs_write_se( s, sh->weight[i][j].i_offset );
284                 }
285             }
286         }
287     }
288     else if( sh->pps->b_weighted_bipred == 1 && sh->i_type == SLICE_TYPE_B )
289     {
290       /* TODO */
291     }
292
293     if( i_nal_ref_idc != 0 )
294     {
295         if( sh->i_idr_pic_id >= 0 )
296         {
297             bs_write1( s, 0 );  /* no output of prior pics flag */
298             bs_write1( s, 0 );  /* long term reference flag */
299         }
300         else
301         {
302             bs_write1( s, sh->i_mmco_command_count > 0 ); /* adaptive_ref_pic_marking_mode_flag */
303             if( sh->i_mmco_command_count > 0 )
304             {
305                 for( int i = 0; i < sh->i_mmco_command_count; i++ )
306                 {
307                     bs_write_ue( s, 1 ); /* mark short term ref as unused */
308                     bs_write_ue( s, sh->mmco[i].i_difference_of_pic_nums - 1 );
309                 }
310                 bs_write_ue( s, 0 ); /* end command list */
311             }
312         }
313     }
314
315     if( sh->pps->b_cabac && sh->i_type != SLICE_TYPE_I )
316         bs_write_ue( s, sh->i_cabac_init_idc );
317
318     bs_write_se( s, sh->i_qp_delta );      /* slice qp delta */
319
320     if( sh->pps->b_deblocking_filter_control )
321     {
322         bs_write_ue( s, sh->i_disable_deblocking_filter_idc );
323         if( sh->i_disable_deblocking_filter_idc != 1 )
324         {
325             bs_write_se( s, sh->i_alpha_c0_offset >> 1 );
326             bs_write_se( s, sh->i_beta_offset >> 1 );
327         }
328     }
329 }
330
331 /* If we are within a reasonable distance of the end of the memory allocated for the bitstream, */
332 /* reallocate, adding an arbitrary amount of space (100 kilobytes). */
333 static int x264_bitstream_check_buffer( x264_t *h )
334 {
335     uint8_t *bs_bak = h->out.p_bitstream;
336     if( (h->param.b_cabac && (h->cabac.p_end - h->cabac.p < 2500)) ||
337         (h->out.bs.p_end - h->out.bs.p < 2500) )
338     {
339         h->out.i_bitstream += 100000;
340         CHECKED_MALLOC( h->out.p_bitstream, h->out.i_bitstream );
341         h->mc.memcpy_aligned( h->out.p_bitstream, bs_bak, (h->out.i_bitstream - 100000) & ~15 );
342         intptr_t delta = h->out.p_bitstream - bs_bak;
343
344         h->out.bs.p_start += delta;
345         h->out.bs.p += delta;
346         h->out.bs.p_end = h->out.p_bitstream + h->out.i_bitstream;
347
348         h->cabac.p_start += delta;
349         h->cabac.p += delta;
350         h->cabac.p_end = h->out.p_bitstream + h->out.i_bitstream;
351
352         for( int i = 0; i <= h->out.i_nal; i++ )
353             h->out.nal[i].p_payload += delta;
354         x264_free( bs_bak );
355     }
356     return 0;
357 fail:
358     x264_free( bs_bak );
359     return -1;
360 }
361
362 #if HAVE_THREAD
363 static void x264_encoder_thread_init( x264_t *h )
364 {
365     if( h->param.i_sync_lookahead )
366         x264_lower_thread_priority( 10 );
367
368 #if HAVE_MMX
369     /* Misalign mask has to be set separately for each thread. */
370     if( h->param.cpu&X264_CPU_SSE_MISALIGN )
371         x264_cpu_mask_misalign_sse();
372 #endif
373 }
374 #endif
375
376 /****************************************************************************
377  *
378  ****************************************************************************
379  ****************************** External API*********************************
380  ****************************************************************************
381  *
382  ****************************************************************************/
383
384 static int x264_validate_parameters( x264_t *h )
385 {
386 #if HAVE_MMX
387 #ifdef __SSE__
388     if( !(x264_cpu_detect() & X264_CPU_SSE) )
389     {
390         x264_log( h, X264_LOG_ERROR, "your cpu does not support SSE1, but x264 was compiled with asm support\n");
391 #else
392     if( !(x264_cpu_detect() & X264_CPU_MMXEXT) )
393     {
394         x264_log( h, X264_LOG_ERROR, "your cpu does not support MMXEXT, but x264 was compiled with asm support\n");
395 #endif
396         x264_log( h, X264_LOG_ERROR, "to run x264, recompile without asm support (configure --disable-asm)\n");
397         return -1;
398     }
399 #endif
400     if( h->param.i_width <= 0 || h->param.i_height <= 0 )
401     {
402         x264_log( h, X264_LOG_ERROR, "invalid width x height (%dx%d)\n",
403                   h->param.i_width, h->param.i_height );
404         return -1;
405     }
406
407     if( h->param.i_width % 2 || h->param.i_height % 2 )
408     {
409         x264_log( h, X264_LOG_ERROR, "width or height not divisible by 2 (%dx%d)\n",
410                   h->param.i_width, h->param.i_height );
411         return -1;
412     }
413     int i_csp = h->param.i_csp & X264_CSP_MASK;
414     if( i_csp <= X264_CSP_NONE || i_csp >= X264_CSP_MAX )
415     {
416         x264_log( h, X264_LOG_ERROR, "invalid CSP (only I420/YV12/NV12 supported)\n" );
417         return -1;
418     }
419
420     if( (h->param.crop_rect.i_left + h->param.crop_rect.i_right ) >= h->param.i_width ||
421         (h->param.crop_rect.i_top  + h->param.crop_rect.i_bottom) >= h->param.i_height )
422     {
423         x264_log( h, X264_LOG_ERROR, "invalid crop-rect %u,%u,%u,%u\n", h->param.crop_rect.i_left,
424                   h->param.crop_rect.i_top, h->param.crop_rect.i_right,  h->param.crop_rect.i_bottom );
425         return -1;
426     }
427
428     if( h->param.i_threads == X264_THREADS_AUTO )
429         h->param.i_threads = x264_cpu_num_processors() * (h->param.b_sliced_threads?2:3)/2;
430     h->param.i_threads = x264_clip3( h->param.i_threads, 1, X264_THREAD_MAX );
431     if( h->param.i_threads > 1 )
432     {
433 #if !HAVE_THREAD
434         x264_log( h, X264_LOG_WARNING, "not compiled with thread support!\n");
435         h->param.i_threads = 1;
436 #endif
437         /* Avoid absurdly small thread slices as they can reduce performance
438          * and VBV compliance.  Capped at an arbitrary 4 rows per thread. */
439         if( h->param.b_sliced_threads )
440         {
441             int max_threads = (h->param.i_height+15)/16 / 4;
442             h->param.i_threads = X264_MIN( h->param.i_threads, max_threads );
443         }
444     }
445     else
446         h->param.b_sliced_threads = 0;
447     h->i_thread_frames = h->param.b_sliced_threads ? 1 : h->param.i_threads;
448     if( h->i_thread_frames > 1 )
449         h->param.nalu_process = NULL;
450
451     h->param.i_keyint_max = x264_clip3( h->param.i_keyint_max, 1, X264_KEYINT_MAX_INFINITE );
452     if( h->param.i_keyint_max == 1 )
453     {
454         h->param.b_intra_refresh = 0;
455         h->param.analyse.i_weighted_pred = 0;
456     }
457
458     if( h->param.b_interlaced )
459     {
460         if( h->param.analyse.i_me_method >= X264_ME_ESA )
461         {
462             x264_log( h, X264_LOG_WARNING, "interlace + me=esa is not implemented\n" );
463             h->param.analyse.i_me_method = X264_ME_UMH;
464         }
465         if( h->param.analyse.i_weighted_pred > 0 )
466         {
467             x264_log( h, X264_LOG_WARNING, "interlace + weightp is not implemented\n" );
468             h->param.analyse.i_weighted_pred = X264_WEIGHTP_NONE;
469         }
470     }
471
472     h->param.i_frame_packing = x264_clip3( h->param.i_frame_packing, -1, 5 );
473
474     /* Detect default ffmpeg settings and terminate with an error. */
475     {
476         int score = 0;
477         score += h->param.analyse.i_me_range == 0;
478         score += h->param.rc.i_qp_step == 3;
479         score += h->param.i_keyint_max == 12;
480         score += h->param.rc.i_qp_min == 2;
481         score += h->param.rc.i_qp_max == 31;
482         score += h->param.rc.f_qcompress == 0.5;
483         score += fabs(h->param.rc.f_ip_factor - 1.25) < 0.01;
484         score += fabs(h->param.rc.f_pb_factor - 1.25) < 0.01;
485         score += h->param.analyse.inter == 0 && h->param.analyse.i_subpel_refine == 8;
486         if( score >= 5 )
487         {
488             x264_log( h, X264_LOG_ERROR, "broken ffmpeg default settings detected\n" );
489             x264_log( h, X264_LOG_ERROR, "use an encoding preset (e.g. -vpre medium)\n" );
490             x264_log( h, X264_LOG_ERROR, "preset usage: -vpre <speed> -vpre <profile>\n" );
491             x264_log( h, X264_LOG_ERROR, "speed presets are listed in x264 --help\n" );
492             x264_log( h, X264_LOG_ERROR, "profile is optional; x264 defaults to high\n" );
493             return -1;
494         }
495     }
496
497     if( h->param.rc.i_rc_method < 0 || h->param.rc.i_rc_method > 2 )
498     {
499         x264_log( h, X264_LOG_ERROR, "no ratecontrol method specified\n" );
500         return -1;
501     }
502     h->param.rc.f_rf_constant = x264_clip3f( h->param.rc.f_rf_constant, -QP_BD_OFFSET, 51 );
503     h->param.rc.i_qp_constant = x264_clip3( h->param.rc.i_qp_constant, 0, QP_MAX );
504     if( h->param.rc.i_rc_method == X264_RC_CRF )
505     {
506         h->param.rc.i_qp_constant = h->param.rc.f_rf_constant + QP_BD_OFFSET;
507         h->param.rc.i_bitrate = 0;
508     }
509     if( (h->param.rc.i_rc_method == X264_RC_CQP || h->param.rc.i_rc_method == X264_RC_CRF)
510         && h->param.rc.i_qp_constant == 0 )
511     {
512         h->mb.b_lossless = 1;
513         h->param.i_cqm_preset = X264_CQM_FLAT;
514         h->param.psz_cqm_file = NULL;
515         h->param.rc.i_rc_method = X264_RC_CQP;
516         h->param.rc.f_ip_factor = 1;
517         h->param.rc.f_pb_factor = 1;
518         h->param.analyse.b_psnr = 0;
519         h->param.analyse.b_ssim = 0;
520         h->param.analyse.i_chroma_qp_offset = 0;
521         h->param.analyse.i_trellis = 0;
522         h->param.analyse.b_fast_pskip = 0;
523         h->param.analyse.i_noise_reduction = 0;
524         h->param.analyse.b_psy = 0;
525         h->param.i_bframe = 0;
526         /* 8x8dct is not useful without RD in CAVLC lossless */
527         if( !h->param.b_cabac && h->param.analyse.i_subpel_refine < 6 )
528             h->param.analyse.b_transform_8x8 = 0;
529     }
530     if( h->param.rc.i_rc_method == X264_RC_CQP )
531     {
532         float qp_p = h->param.rc.i_qp_constant;
533         float qp_i = qp_p - 6*log2f( h->param.rc.f_ip_factor );
534         float qp_b = qp_p + 6*log2f( h->param.rc.f_pb_factor );
535         h->param.rc.i_qp_min = x264_clip3( (int)(X264_MIN3( qp_p, qp_i, qp_b )), 0, QP_MAX );
536         h->param.rc.i_qp_max = x264_clip3( (int)(X264_MAX3( qp_p, qp_i, qp_b ) + .999), 0, QP_MAX );
537         h->param.rc.i_aq_mode = 0;
538         h->param.rc.b_mb_tree = 0;
539     }
540     h->param.rc.i_qp_max = x264_clip3( h->param.rc.i_qp_max, 0, QP_MAX );
541     h->param.rc.i_qp_min = x264_clip3( h->param.rc.i_qp_min, 0, h->param.rc.i_qp_max );
542     if( h->param.rc.i_vbv_buffer_size )
543     {
544         if( h->param.rc.i_rc_method == X264_RC_CQP )
545         {
546             x264_log( h, X264_LOG_WARNING, "VBV is incompatible with constant QP, ignored.\n" );
547             h->param.rc.i_vbv_max_bitrate = 0;
548             h->param.rc.i_vbv_buffer_size = 0;
549         }
550         else if( h->param.rc.i_vbv_max_bitrate == 0 )
551         {
552             if( h->param.rc.i_rc_method == X264_RC_ABR )
553             {
554                 x264_log( h, X264_LOG_WARNING, "VBV maxrate unspecified, assuming CBR\n" );
555                 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate;
556             }
557             else
558             {
559                 x264_log( h, X264_LOG_WARNING, "VBV bufsize set but maxrate unspecified, ignored\n" );
560                 h->param.rc.i_vbv_buffer_size = 0;
561             }
562         }
563         else if( h->param.rc.i_vbv_max_bitrate < h->param.rc.i_bitrate &&
564                  h->param.rc.i_rc_method == X264_RC_ABR )
565         {
566             x264_log( h, X264_LOG_WARNING, "max bitrate less than average bitrate, assuming CBR\n" );
567             h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate;
568         }
569     }
570     else if( h->param.rc.i_vbv_max_bitrate )
571     {
572         x264_log( h, X264_LOG_WARNING, "VBV maxrate specified, but no bufsize, ignored\n" );
573         h->param.rc.i_vbv_max_bitrate = 0;
574     }
575
576     if( h->param.b_interlaced && h->param.i_slice_max_size )
577     {
578         x264_log( h, X264_LOG_WARNING, "interlaced + slice-max-size is not implemented\n" );
579         h->param.i_slice_max_size = 0;
580     }
581     if( h->param.b_interlaced && h->param.i_slice_max_mbs )
582     {
583         x264_log( h, X264_LOG_WARNING, "interlaced + slice-max-mbs is not implemented\n" );
584         h->param.i_slice_max_mbs = 0;
585     }
586     int max_slices = (h->param.i_height+((16<<h->param.b_interlaced)-1))/(16<<h->param.b_interlaced);
587     if( h->param.b_sliced_threads )
588         h->param.i_slice_count = x264_clip3( h->param.i_threads, 0, max_slices );
589     else
590     {
591         h->param.i_slice_count = x264_clip3( h->param.i_slice_count, 0, max_slices );
592         h->param.i_slice_max_size = X264_MAX( h->param.i_slice_max_size, 0 );
593         h->param.i_slice_max_mbs = X264_MAX( h->param.i_slice_max_mbs, 0 );
594         if( h->param.i_slice_max_mbs || h->param.i_slice_max_size )
595             h->param.i_slice_count = 0;
596     }
597
598     h->param.i_frame_reference = x264_clip3( h->param.i_frame_reference, 1, X264_REF_MAX );
599     h->param.i_dpb_size = x264_clip3( h->param.i_dpb_size, 1, X264_REF_MAX );
600     if( h->param.i_scenecut_threshold < 0 )
601         h->param.i_scenecut_threshold = 0;
602     if( !h->param.analyse.i_subpel_refine && h->param.analyse.i_direct_mv_pred > X264_DIRECT_PRED_SPATIAL )
603     {
604         x264_log( h, X264_LOG_WARNING, "subme=0 + direct=temporal is not supported\n" );
605         h->param.analyse.i_direct_mv_pred = X264_DIRECT_PRED_SPATIAL;
606     }
607     h->param.i_bframe = x264_clip3( h->param.i_bframe, 0, X264_MIN( X264_BFRAME_MAX, h->param.i_keyint_max-1 ) );
608     h->param.i_open_gop = x264_clip3( h->param.i_open_gop, X264_OPEN_GOP_NONE, X264_OPEN_GOP_BLURAY );
609     h->param.i_bframe_bias = x264_clip3( h->param.i_bframe_bias, -90, 100 );
610     if( h->param.i_bframe <= 1 )
611         h->param.i_bframe_pyramid = X264_B_PYRAMID_NONE;
612     h->param.i_bframe_pyramid = x264_clip3( h->param.i_bframe_pyramid, X264_B_PYRAMID_NONE, X264_B_PYRAMID_NORMAL );
613     if( !h->param.i_bframe )
614     {
615         h->param.i_bframe_adaptive = X264_B_ADAPT_NONE;
616         h->param.analyse.i_direct_mv_pred = 0;
617         h->param.analyse.b_weighted_bipred = 0;
618         h->param.i_open_gop = X264_OPEN_GOP_NONE;
619     }
620     if( h->param.b_intra_refresh && h->param.i_bframe_pyramid == X264_B_PYRAMID_NORMAL )
621     {
622         x264_log( h, X264_LOG_WARNING, "b-pyramid normal + intra-refresh is not supported\n" );
623         h->param.i_bframe_pyramid = X264_B_PYRAMID_STRICT;
624     }
625     if( h->param.b_intra_refresh && (h->param.i_frame_reference > 1 || h->param.i_dpb_size > 1) )
626     {
627         x264_log( h, X264_LOG_WARNING, "ref > 1 + intra-refresh is not supported\n" );
628         h->param.i_frame_reference = 1;
629         h->param.i_dpb_size = 1;
630     }
631     if( h->param.b_intra_refresh && h->param.i_open_gop )
632     {
633         x264_log( h, X264_LOG_WARNING, "intra-refresh is not compatible with open-gop\n" );
634         h->param.i_open_gop = X264_OPEN_GOP_NONE;
635     }
636     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;
637     if( h->param.i_keyint_min == X264_KEYINT_MIN_AUTO )
638         h->param.i_keyint_min = X264_MIN( h->param.i_keyint_max / 10, fps );
639     h->param.i_keyint_min = x264_clip3( h->param.i_keyint_min, 1, h->param.i_keyint_max/2+1 );
640     h->param.rc.i_lookahead = x264_clip3( h->param.rc.i_lookahead, 0, X264_LOOKAHEAD_MAX );
641     {
642         int maxrate = X264_MAX( h->param.rc.i_vbv_max_bitrate, h->param.rc.i_bitrate );
643         float bufsize = maxrate ? (float)h->param.rc.i_vbv_buffer_size / maxrate : 0;
644         h->param.rc.i_lookahead = X264_MIN( h->param.rc.i_lookahead, X264_MAX( h->param.i_keyint_max, bufsize*fps ) );
645     }
646
647     if( !h->param.i_timebase_num || !h->param.i_timebase_den || !(h->param.b_vfr_input || h->param.b_pulldown) )
648     {
649         h->param.i_timebase_num = h->param.i_fps_den;
650         h->param.i_timebase_den = h->param.i_fps_num;
651     }
652
653     h->param.rc.f_qcompress = x264_clip3f( h->param.rc.f_qcompress, 0.0, 1.0 );
654     if( h->param.i_keyint_max == 1 || h->param.rc.f_qcompress == 1 )
655         h->param.rc.b_mb_tree = 0;
656     if( (!h->param.b_intra_refresh && h->param.i_keyint_max != X264_KEYINT_MAX_INFINITE) &&
657         !h->param.rc.i_lookahead && h->param.rc.b_mb_tree )
658     {
659         x264_log( h, X264_LOG_WARNING, "lookaheadless mb-tree requires intra refresh or infinite keyint\n" );
660         h->param.rc.b_mb_tree = 0;
661     }
662     if( h->param.rc.b_stat_read )
663         h->param.rc.i_lookahead = 0;
664 #if HAVE_THREAD
665     if( h->param.i_sync_lookahead < 0 )
666         h->param.i_sync_lookahead = h->param.i_bframe + 1;
667     h->param.i_sync_lookahead = X264_MIN( h->param.i_sync_lookahead, X264_LOOKAHEAD_MAX );
668     if( h->param.rc.b_stat_read || h->i_thread_frames == 1 )
669         h->param.i_sync_lookahead = 0;
670 #else
671     h->param.i_sync_lookahead = 0;
672 #endif
673
674     h->param.i_deblocking_filter_alphac0 = x264_clip3( h->param.i_deblocking_filter_alphac0, -6, 6 );
675     h->param.i_deblocking_filter_beta    = x264_clip3( h->param.i_deblocking_filter_beta, -6, 6 );
676     h->param.analyse.i_luma_deadzone[0] = x264_clip3( h->param.analyse.i_luma_deadzone[0], 0, 32 );
677     h->param.analyse.i_luma_deadzone[1] = x264_clip3( h->param.analyse.i_luma_deadzone[1], 0, 32 );
678
679     h->param.i_cabac_init_idc = x264_clip3( h->param.i_cabac_init_idc, 0, 2 );
680
681     if( h->param.i_cqm_preset < X264_CQM_FLAT || h->param.i_cqm_preset > X264_CQM_CUSTOM )
682         h->param.i_cqm_preset = X264_CQM_FLAT;
683
684     if( h->param.analyse.i_me_method < X264_ME_DIA ||
685         h->param.analyse.i_me_method > X264_ME_TESA )
686         h->param.analyse.i_me_method = X264_ME_HEX;
687     if( h->param.analyse.i_me_range < 4 )
688         h->param.analyse.i_me_range = 4;
689     if( h->param.analyse.i_me_range > 16 && h->param.analyse.i_me_method <= X264_ME_HEX )
690         h->param.analyse.i_me_range = 16;
691     if( h->param.analyse.i_me_method == X264_ME_TESA &&
692         (h->mb.b_lossless || h->param.analyse.i_subpel_refine <= 1) )
693         h->param.analyse.i_me_method = X264_ME_ESA;
694     h->param.analyse.i_subpel_refine = x264_clip3( h->param.analyse.i_subpel_refine, 0, 10 );
695     h->param.analyse.b_mixed_references = h->param.analyse.b_mixed_references && h->param.i_frame_reference > 1;
696     h->param.analyse.inter &= X264_ANALYSE_PSUB16x16|X264_ANALYSE_PSUB8x8|X264_ANALYSE_BSUB16x16|
697                               X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
698     h->param.analyse.intra &= X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
699     if( !(h->param.analyse.inter & X264_ANALYSE_PSUB16x16) )
700         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
701     if( !h->param.analyse.b_transform_8x8 )
702     {
703         h->param.analyse.inter &= ~X264_ANALYSE_I8x8;
704         h->param.analyse.intra &= ~X264_ANALYSE_I8x8;
705     }
706     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -12, 12);
707     h->param.analyse.i_trellis = x264_clip3( h->param.analyse.i_trellis, 0, 2 );
708     if( !h->param.analyse.b_psy )
709     {
710         h->param.analyse.f_psy_rd = 0;
711         h->param.analyse.f_psy_trellis = 0;
712     }
713     if( !h->param.analyse.i_trellis )
714         h->param.analyse.f_psy_trellis = 0;
715     h->param.analyse.f_psy_rd = x264_clip3f( h->param.analyse.f_psy_rd, 0, 10 );
716     h->param.analyse.f_psy_trellis = x264_clip3f( h->param.analyse.f_psy_trellis, 0, 10 );
717     if( h->param.analyse.i_subpel_refine < 6 )
718         h->param.analyse.f_psy_rd = 0;
719     h->mb.i_psy_rd = FIX8( h->param.analyse.f_psy_rd );
720     /* Psy RDO increases overall quantizers to improve the quality of luma--this indirectly hurts chroma quality */
721     /* so we lower the chroma QP offset to compensate */
722     /* This can be triggered repeatedly on multiple calls to parameter_validate, but since encoding
723      * uses the pps chroma qp offset not the param chroma qp offset, this is not a problem. */
724     if( h->mb.i_psy_rd )
725         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_rd < 0.25 ? 1 : 2;
726     h->mb.i_psy_trellis = FIX8( h->param.analyse.f_psy_trellis / 4 );
727     /* Psy trellis has a similar effect. */
728     if( h->mb.i_psy_trellis )
729         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_trellis < 0.25 ? 1 : 2;
730     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -12, 12);
731     h->param.rc.i_aq_mode = x264_clip3( h->param.rc.i_aq_mode, 0, 2 );
732     h->param.rc.f_aq_strength = x264_clip3f( h->param.rc.f_aq_strength, 0, 3 );
733     if( h->param.rc.f_aq_strength == 0 )
734         h->param.rc.i_aq_mode = 0;
735     /* MB-tree requires AQ to be on, even if the strength is zero. */
736     if( !h->param.rc.i_aq_mode && h->param.rc.b_mb_tree )
737     {
738         h->param.rc.i_aq_mode = 1;
739         h->param.rc.f_aq_strength = 0;
740     }
741     h->param.analyse.i_noise_reduction = x264_clip3( h->param.analyse.i_noise_reduction, 0, 1<<16 );
742     if( h->param.analyse.i_subpel_refine == 10 && (h->param.analyse.i_trellis != 2 || !h->param.rc.i_aq_mode) )
743         h->param.analyse.i_subpel_refine = 9;
744
745     {
746         const x264_level_t *l = x264_levels;
747         if( h->param.i_level_idc < 0 )
748         {
749             int maxrate_bak = h->param.rc.i_vbv_max_bitrate;
750             if( h->param.rc.i_rc_method == X264_RC_ABR && h->param.rc.i_vbv_buffer_size <= 0 )
751                 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate * 2;
752             h->sps = h->sps_array;
753             x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
754             do h->param.i_level_idc = l->level_idc;
755                 while( l[1].level_idc && x264_validate_levels( h, 0 ) && l++ );
756             h->param.rc.i_vbv_max_bitrate = maxrate_bak;
757         }
758         else
759         {
760             while( l->level_idc && l->level_idc != h->param.i_level_idc )
761                 l++;
762             if( l->level_idc == 0 )
763             {
764                 x264_log( h, X264_LOG_ERROR, "invalid level_idc: %d\n", h->param.i_level_idc );
765                 return -1;
766             }
767         }
768         if( h->param.analyse.i_mv_range <= 0 )
769             h->param.analyse.i_mv_range = l->mv_range >> h->param.b_interlaced;
770         else
771             h->param.analyse.i_mv_range = x264_clip3(h->param.analyse.i_mv_range, 32, 512 >> h->param.b_interlaced);
772     }
773
774     h->param.analyse.i_weighted_pred = x264_clip3( h->param.analyse.i_weighted_pred, X264_WEIGHTP_NONE, X264_WEIGHTP_SMART );
775     if( !h->param.analyse.i_weighted_pred && h->param.rc.b_mb_tree && h->param.analyse.b_psy )
776         h->param.analyse.i_weighted_pred = X264_WEIGHTP_FAKE;
777
778     if( h->i_thread_frames > 1 )
779     {
780         int r = h->param.analyse.i_mv_range_thread;
781         int r2;
782         if( r <= 0 )
783         {
784             // half of the available space is reserved and divided evenly among the threads,
785             // the rest is allocated to whichever thread is far enough ahead to use it.
786             // reserving more space increases quality for some videos, but costs more time
787             // in thread synchronization.
788             int max_range = (h->param.i_height + X264_THREAD_HEIGHT) / h->i_thread_frames - X264_THREAD_HEIGHT;
789             r = max_range / 2;
790         }
791         r = X264_MAX( r, h->param.analyse.i_me_range );
792         r = X264_MIN( r, h->param.analyse.i_mv_range );
793         // round up to use the whole mb row
794         r2 = (r & ~15) + ((-X264_THREAD_HEIGHT) & 15);
795         if( r2 < r )
796             r2 += 16;
797         x264_log( h, X264_LOG_DEBUG, "using mv_range_thread = %d\n", r2 );
798         h->param.analyse.i_mv_range_thread = r2;
799     }
800
801     if( h->param.rc.f_qblur < 0 )
802         h->param.rc.f_qblur = 0;
803     if( h->param.rc.f_complexity_blur < 0 )
804         h->param.rc.f_complexity_blur = 0;
805
806     h->param.i_sps_id &= 31;
807
808     if( h->param.i_log_level < X264_LOG_INFO )
809     {
810         h->param.analyse.b_psnr = 0;
811         h->param.analyse.b_ssim = 0;
812     }
813
814     if( h->param.b_interlaced )
815         h->param.b_pic_struct = 1;
816
817     if( h->param.i_nal_hrd && !h->param.rc.i_vbv_buffer_size )
818     {
819         x264_log( h, X264_LOG_WARNING, "NAL HRD parameters require VBV parameters\n" );
820         h->param.i_nal_hrd = X264_NAL_HRD_NONE;
821     }
822
823     if( h->param.i_nal_hrd == X264_NAL_HRD_CBR &&
824        (h->param.rc.i_bitrate != h->param.rc.i_vbv_max_bitrate || !h->param.rc.i_vbv_max_bitrate) )
825     {
826         x264_log( h, X264_LOG_WARNING, "CBR HRD requires constant bitrate\n" );
827         h->param.i_nal_hrd = X264_NAL_HRD_VBR;
828     }
829
830     /* ensure the booleans are 0 or 1 so they can be used in math */
831 #define BOOLIFY(x) h->param.x = !!h->param.x
832     BOOLIFY( b_cabac );
833     BOOLIFY( b_constrained_intra );
834     BOOLIFY( b_deblocking_filter );
835     BOOLIFY( b_deterministic );
836     BOOLIFY( b_sliced_threads );
837     BOOLIFY( b_interlaced );
838     BOOLIFY( b_intra_refresh );
839     BOOLIFY( b_visualize );
840     BOOLIFY( b_aud );
841     BOOLIFY( b_repeat_headers );
842     BOOLIFY( b_annexb );
843     BOOLIFY( b_vfr_input );
844     BOOLIFY( b_pic_struct );
845     BOOLIFY( b_fake_interlaced );
846     BOOLIFY( analyse.b_transform_8x8 );
847     BOOLIFY( analyse.b_weighted_bipred );
848     BOOLIFY( analyse.b_chroma_me );
849     BOOLIFY( analyse.b_mixed_references );
850     BOOLIFY( analyse.b_fast_pskip );
851     BOOLIFY( analyse.b_dct_decimate );
852     BOOLIFY( analyse.b_psy );
853     BOOLIFY( analyse.b_psnr );
854     BOOLIFY( analyse.b_ssim );
855     BOOLIFY( rc.b_stat_write );
856     BOOLIFY( rc.b_stat_read );
857     BOOLIFY( rc.b_mb_tree );
858 #undef BOOLIFY
859
860     return 0;
861 }
862
863 static void mbcmp_init( x264_t *h )
864 {
865     int satd = !h->mb.b_lossless && h->param.analyse.i_subpel_refine > 1;
866     memcpy( h->pixf.mbcmp, satd ? h->pixf.satd : h->pixf.sad_aligned, sizeof(h->pixf.mbcmp) );
867     memcpy( h->pixf.mbcmp_unaligned, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.mbcmp_unaligned) );
868     h->pixf.intra_mbcmp_x3_16x16 = satd ? h->pixf.intra_satd_x3_16x16 : h->pixf.intra_sad_x3_16x16;
869     h->pixf.intra_mbcmp_x3_8x8c = satd ? h->pixf.intra_satd_x3_8x8c : h->pixf.intra_sad_x3_8x8c;
870     h->pixf.intra_mbcmp_x3_8x8 = satd ? h->pixf.intra_sa8d_x3_8x8 : h->pixf.intra_sad_x3_8x8;
871     h->pixf.intra_mbcmp_x3_4x4 = satd ? h->pixf.intra_satd_x3_4x4 : h->pixf.intra_sad_x3_4x4;
872     satd &= h->param.analyse.i_me_method == X264_ME_TESA;
873     memcpy( h->pixf.fpelcmp, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.fpelcmp) );
874     memcpy( h->pixf.fpelcmp_x3, satd ? h->pixf.satd_x3 : h->pixf.sad_x3, sizeof(h->pixf.fpelcmp_x3) );
875     memcpy( h->pixf.fpelcmp_x4, satd ? h->pixf.satd_x4 : h->pixf.sad_x4, sizeof(h->pixf.fpelcmp_x4) );
876 }
877
878 static void x264_set_aspect_ratio( x264_t *h, x264_param_t *param, int initial )
879 {
880     /* VUI */
881     if( param->vui.i_sar_width > 0 && param->vui.i_sar_height > 0 )
882     {
883         uint32_t i_w = param->vui.i_sar_width;
884         uint32_t i_h = param->vui.i_sar_height;
885         uint32_t old_w = h->param.vui.i_sar_width;
886         uint32_t old_h = h->param.vui.i_sar_height;
887
888         x264_reduce_fraction( &i_w, &i_h );
889
890         while( i_w > 65535 || i_h > 65535 )
891         {
892             i_w /= 2;
893             i_h /= 2;
894         }
895
896         x264_reduce_fraction( &i_w, &i_h );
897
898         if( i_w != old_w || i_h != old_h || initial )
899         {
900             h->param.vui.i_sar_width = 0;
901             h->param.vui.i_sar_height = 0;
902             if( i_w == 0 || i_h == 0 )
903                 x264_log( h, X264_LOG_WARNING, "cannot create valid sample aspect ratio\n" );
904             else
905             {
906                 x264_log( h, initial?X264_LOG_INFO:X264_LOG_DEBUG, "using SAR=%d/%d\n", i_w, i_h );
907                 h->param.vui.i_sar_width = i_w;
908                 h->param.vui.i_sar_height = i_h;
909             }
910             x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
911         }
912     }
913 }
914
915 /****************************************************************************
916  * x264_encoder_open:
917  ****************************************************************************/
918 x264_t *x264_encoder_open( x264_param_t *param )
919 {
920     x264_t *h;
921     char buf[1000], *p;
922     int qp, i_slicetype_length;
923
924     CHECKED_MALLOCZERO( h, sizeof(x264_t) );
925
926     /* Create a copy of param */
927     memcpy( &h->param, param, sizeof(x264_param_t) );
928
929     if( param->param_free )
930         param->param_free( param );
931
932     if( x264_threading_init() )
933     {
934         x264_log( h, X264_LOG_ERROR, "unable to initialize threading\n" );
935         goto fail;
936     }
937
938     if( x264_validate_parameters( h ) < 0 )
939         goto fail;
940
941     if( h->param.psz_cqm_file )
942         if( x264_cqm_parse_file( h, h->param.psz_cqm_file ) < 0 )
943             goto fail;
944
945     if( h->param.rc.psz_stat_out )
946         h->param.rc.psz_stat_out = strdup( h->param.rc.psz_stat_out );
947     if( h->param.rc.psz_stat_in )
948         h->param.rc.psz_stat_in = strdup( h->param.rc.psz_stat_in );
949
950     x264_reduce_fraction( &h->param.i_fps_num, &h->param.i_fps_den );
951     x264_reduce_fraction( &h->param.i_timebase_num, &h->param.i_timebase_den );
952
953     /* Init x264_t */
954     h->i_frame = -1;
955     h->i_frame_num = 0;
956     h->i_idr_pic_id = 0;
957
958     if( (uint64_t)h->param.i_timebase_den * 2 > UINT32_MAX )
959     {
960         x264_log( h, X264_LOG_ERROR, "Effective timebase denominator %u exceeds H.264 maximum\n", h->param.i_timebase_den );
961         goto fail;
962     }
963
964     h->sps = &h->sps_array[0];
965     x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
966
967     h->pps = &h->pps_array[0];
968     x264_pps_init( h->pps, h->param.i_sps_id, &h->param, h->sps );
969
970     x264_set_aspect_ratio( h, &h->param, 1 );
971
972     x264_validate_levels( h, 1 );
973
974     h->chroma_qp_table = i_chroma_qp_table + 12 + h->pps->i_chroma_qp_index_offset;
975
976     if( x264_cqm_init( h ) < 0 )
977         goto fail;
978
979     h->mb.i_mb_width = h->sps->i_mb_width;
980     h->mb.i_mb_height = h->sps->i_mb_height;
981     h->mb.i_mb_count = h->mb.i_mb_width * h->mb.i_mb_height;
982
983     /* Init frames. */
984     if( h->param.i_bframe_adaptive == X264_B_ADAPT_TRELLIS && !h->param.rc.b_stat_read )
985         h->frames.i_delay = X264_MAX(h->param.i_bframe,3)*4;
986     else
987         h->frames.i_delay = h->param.i_bframe;
988     if( h->param.rc.b_mb_tree || h->param.rc.i_vbv_buffer_size )
989         h->frames.i_delay = X264_MAX( h->frames.i_delay, h->param.rc.i_lookahead );
990     i_slicetype_length = h->frames.i_delay;
991     h->frames.i_delay += h->i_thread_frames - 1;
992     h->frames.i_delay += h->param.i_sync_lookahead;
993     h->frames.i_delay += h->param.b_vfr_input;
994     h->frames.i_bframe_delay = h->param.i_bframe ? (h->param.i_bframe_pyramid ? 2 : 1) : 0;
995
996     h->frames.i_max_ref0 = h->param.i_frame_reference;
997     h->frames.i_max_ref1 = X264_MIN( h->sps->vui.i_num_reorder_frames, h->param.i_frame_reference );
998     h->frames.i_max_dpb  = h->sps->vui.i_max_dec_frame_buffering;
999     h->frames.b_have_lowres = !h->param.rc.b_stat_read
1000         && ( h->param.rc.i_rc_method == X264_RC_ABR
1001           || h->param.rc.i_rc_method == X264_RC_CRF
1002           || h->param.i_bframe_adaptive
1003           || h->param.i_scenecut_threshold
1004           || h->param.rc.b_mb_tree
1005           || h->param.analyse.i_weighted_pred );
1006     h->frames.b_have_lowres |= h->param.rc.b_stat_read && h->param.rc.i_vbv_buffer_size > 0;
1007     h->frames.b_have_sub8x8_esa = !!(h->param.analyse.inter & X264_ANALYSE_PSUB8x8);
1008
1009     h->frames.i_last_idr =
1010     h->frames.i_last_keyframe = - h->param.i_keyint_max;
1011     h->frames.i_input    = 0;
1012     h->frames.i_largest_pts = h->frames.i_second_largest_pts = -1;
1013     h->frames.i_poc_last_open_gop = -1;
1014
1015     CHECKED_MALLOCZERO( h->frames.unused[0], (h->frames.i_delay + 3) * sizeof(x264_frame_t *) );
1016     /* Allocate room for max refs plus a few extra just in case. */
1017     CHECKED_MALLOCZERO( h->frames.unused[1], (h->i_thread_frames + X264_REF_MAX + 4) * sizeof(x264_frame_t *) );
1018     CHECKED_MALLOCZERO( h->frames.current, (h->param.i_sync_lookahead + h->param.i_bframe
1019                         + h->i_thread_frames + 3) * sizeof(x264_frame_t *) );
1020     if( h->param.analyse.i_weighted_pred > 0 )
1021         CHECKED_MALLOCZERO( h->frames.blank_unused, h->i_thread_frames * 4 * sizeof(x264_frame_t *) );
1022     h->i_ref[0] = h->i_ref[1] = 0;
1023     h->i_cpb_delay = h->i_coded_fields = h->i_disp_fields = 0;
1024     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);
1025     h->i_disp_fields_last_frame = -1;
1026     x264_rdo_init();
1027
1028     /* init CPU functions */
1029     x264_predict_16x16_init( h->param.cpu, h->predict_16x16 );
1030     x264_predict_8x8c_init( h->param.cpu, h->predict_8x8c );
1031     x264_predict_8x8_init( h->param.cpu, h->predict_8x8, &h->predict_8x8_filter );
1032     x264_predict_4x4_init( h->param.cpu, h->predict_4x4 );
1033     if( !h->param.b_cabac )
1034         x264_init_vlc_tables();
1035     x264_pixel_init( h->param.cpu, &h->pixf );
1036     x264_dct_init( h->param.cpu, &h->dctf );
1037     x264_zigzag_init( h->param.cpu, &h->zigzagf, h->param.b_interlaced );
1038     x264_mc_init( h->param.cpu, &h->mc );
1039     x264_quant_init( h, h->param.cpu, &h->quantf );
1040     x264_deblock_init( h->param.cpu, &h->loopf );
1041     x264_bitstream_init( h->param.cpu, &h->bsf );
1042     x264_dct_init_weights();
1043
1044     mbcmp_init( h );
1045
1046     p = buf + sprintf( buf, "using cpu capabilities:" );
1047     for( int i = 0; x264_cpu_names[i].flags; i++ )
1048     {
1049         if( !strcmp(x264_cpu_names[i].name, "SSE2")
1050             && h->param.cpu & (X264_CPU_SSE2_IS_FAST|X264_CPU_SSE2_IS_SLOW) )
1051             continue;
1052         if( !strcmp(x264_cpu_names[i].name, "SSE3")
1053             && (h->param.cpu & X264_CPU_SSSE3 || !(h->param.cpu & X264_CPU_CACHELINE_64)) )
1054             continue;
1055         if( !strcmp(x264_cpu_names[i].name, "SSE4.1")
1056             && (h->param.cpu & X264_CPU_SSE42) )
1057             continue;
1058         if( (h->param.cpu & x264_cpu_names[i].flags) == x264_cpu_names[i].flags
1059             && (!i || x264_cpu_names[i].flags != x264_cpu_names[i-1].flags) )
1060             p += sprintf( p, " %s", x264_cpu_names[i].name );
1061     }
1062     if( !h->param.cpu )
1063         p += sprintf( p, " none!" );
1064     x264_log( h, X264_LOG_INFO, "%s\n", buf );
1065
1066     for( qp = X264_MIN( h->param.rc.i_qp_min, QP_MAX_SPEC ); qp <= h->param.rc.i_qp_max; qp++ )
1067         if( x264_analyse_init_costs( h, qp ) )
1068             goto fail;
1069     if( x264_analyse_init_costs( h, X264_LOOKAHEAD_QP ) )
1070         goto fail;
1071
1072     static const uint16_t cost_mv_correct[7] = { 24, 47, 95, 189, 379, 757, 1515 };
1073     /* Checks for known miscompilation issues. */
1074     if( h->cost_mv[X264_LOOKAHEAD_QP][2013] != cost_mv_correct[BIT_DEPTH-8] )
1075     {
1076         x264_log( h, X264_LOG_ERROR, "MV cost test failed: x264 has been miscompiled!\n" );
1077         goto fail;
1078     }
1079
1080     /* Must be volatile or else GCC will optimize it out. */
1081     volatile int temp = 392;
1082     if( x264_clz( temp ) != 23 )
1083     {
1084         x264_log( h, X264_LOG_ERROR, "CLZ test failed: x264 has been miscompiled!\n" );
1085 #if ARCH_X86 || ARCH_X86_64
1086         x264_log( h, X264_LOG_ERROR, "Are you attempting to run an SSE4a-targeted build on a CPU that\n" );
1087         x264_log( h, X264_LOG_ERROR, "doesn't support it?\n" );
1088 #endif
1089         goto fail;
1090     }
1091
1092     h->out.i_nal = 0;
1093     h->out.i_bitstream = X264_MAX( 1000000, h->param.i_width * h->param.i_height * 4
1094         * ( h->param.rc.i_rc_method == X264_RC_ABR ? pow( 0.95, h->param.rc.i_qp_min )
1095           : pow( 0.95, h->param.rc.i_qp_constant ) * X264_MAX( 1, h->param.rc.f_ip_factor )));
1096
1097     CHECKED_MALLOC( h->nal_buffer, h->out.i_bitstream * 3/2 + 4 );
1098     h->nal_buffer_size = h->out.i_bitstream * 3/2 + 4;
1099
1100     if( h->param.i_threads > 1 &&
1101         x264_threadpool_init( &h->threadpool, h->param.i_threads, (void*)x264_encoder_thread_init, h ) )
1102         goto fail;
1103
1104     h->thread[0] = h;
1105     for( int i = 1; i < h->param.i_threads + !!h->param.i_sync_lookahead; i++ )
1106         CHECKED_MALLOC( h->thread[i], sizeof(x264_t) );
1107
1108     for( int i = 0; i < h->param.i_threads; i++ )
1109     {
1110         int init_nal_count = h->param.i_slice_count + 3;
1111         int allocate_threadlocal_data = !h->param.b_sliced_threads || !i;
1112         if( i > 0 )
1113             *h->thread[i] = *h;
1114
1115         if( allocate_threadlocal_data )
1116         {
1117             h->thread[i]->fdec = x264_frame_pop_unused( h, 1 );
1118             if( !h->thread[i]->fdec )
1119                 goto fail;
1120         }
1121         else
1122             h->thread[i]->fdec = h->thread[0]->fdec;
1123
1124         h->thread[i]->sps = &h->thread[i]->sps_array[0];
1125         h->thread[i]->pps = &h->thread[i]->pps_array[0];
1126
1127         CHECKED_MALLOC( h->thread[i]->out.p_bitstream, h->out.i_bitstream );
1128         /* Start each thread with room for init_nal_count NAL units; it'll realloc later if needed. */
1129         CHECKED_MALLOC( h->thread[i]->out.nal, init_nal_count*sizeof(x264_nal_t) );
1130         h->thread[i]->out.i_nals_allocated = init_nal_count;
1131
1132         if( allocate_threadlocal_data && x264_macroblock_cache_allocate( h->thread[i] ) < 0 )
1133             goto fail;
1134     }
1135
1136     if( x264_lookahead_init( h, i_slicetype_length ) )
1137         goto fail;
1138
1139     for( int i = 0; i < h->param.i_threads; i++ )
1140         if( x264_macroblock_thread_allocate( h->thread[i], 0 ) < 0 )
1141             goto fail;
1142
1143     if( x264_ratecontrol_new( h ) < 0 )
1144         goto fail;
1145
1146     if( h->param.i_nal_hrd )
1147     {
1148         x264_log( h, X264_LOG_DEBUG, "HRD bitrate: %i bits/sec\n", h->sps->vui.hrd.i_bit_rate_unscaled );
1149         x264_log( h, X264_LOG_DEBUG, "CPB size: %i bits\n", h->sps->vui.hrd.i_cpb_size_unscaled );
1150     }
1151
1152     if( h->param.psz_dump_yuv )
1153     {
1154         /* create or truncate the reconstructed video file */
1155         FILE *f = fopen( h->param.psz_dump_yuv, "w" );
1156         if( !f )
1157         {
1158             x264_log( h, X264_LOG_ERROR, "dump_yuv: can't write to %s\n", h->param.psz_dump_yuv );
1159             goto fail;
1160         }
1161         else if( !x264_is_regular_file( f ) )
1162         {
1163             x264_log( h, X264_LOG_ERROR, "dump_yuv: incompatible with non-regular file %s\n", h->param.psz_dump_yuv );
1164             goto fail;
1165         }
1166         fclose( f );
1167     }
1168
1169     const char *profile = h->sps->i_profile_idc == PROFILE_BASELINE ? "Constrained Baseline" :
1170                           h->sps->i_profile_idc == PROFILE_MAIN ? "Main" :
1171                           h->sps->i_profile_idc == PROFILE_HIGH ? "High" :
1172                           h->sps->i_profile_idc == PROFILE_HIGH10 ? (h->sps->b_constraint_set3 == 1 ? "High 10 Intra" : "High 10") :
1173                           "High 4:4:4 Predictive";
1174     char level[4];
1175     snprintf( level, sizeof(level), "%d.%d", h->sps->i_level_idc/10, h->sps->i_level_idc%10 );
1176     if( h->sps->i_level_idc == 9 || ( h->sps->i_level_idc == 11 && h->sps->b_constraint_set3 ) )
1177         strcpy( level, "1b" );
1178
1179     if( h->sps->i_profile_idc < PROFILE_HIGH10 )
1180     {
1181         x264_log( h, X264_LOG_INFO, "profile %s, level %s\n",
1182             profile, level );
1183     }
1184     else
1185     {
1186         x264_log( h, X264_LOG_INFO, "profile %s, level %s, bit depth %d\n",
1187             profile, level, BIT_DEPTH );
1188     }
1189
1190     return h;
1191 fail:
1192     x264_free( h );
1193     return NULL;
1194 }
1195
1196 /****************************************************************************
1197  * x264_encoder_reconfig:
1198  ****************************************************************************/
1199 int x264_encoder_reconfig( x264_t *h, x264_param_t *param )
1200 {
1201     int rc_reconfig = 0;
1202     h = h->thread[h->thread[0]->i_thread_phase];
1203     x264_set_aspect_ratio( h, param, 0 );
1204 #define COPY(var) h->param.var = param->var
1205     COPY( i_frame_reference ); // but never uses more refs than initially specified
1206     COPY( i_bframe_bias );
1207     if( h->param.i_scenecut_threshold )
1208         COPY( i_scenecut_threshold ); // can't turn it on or off, only vary the threshold
1209     COPY( b_deblocking_filter );
1210     COPY( i_deblocking_filter_alphac0 );
1211     COPY( i_deblocking_filter_beta );
1212     COPY( i_frame_packing );
1213     COPY( analyse.inter );
1214     COPY( analyse.intra );
1215     COPY( analyse.i_direct_mv_pred );
1216     /* Scratch buffer prevents me_range from being increased for esa/tesa */
1217     if( h->param.analyse.i_me_method < X264_ME_ESA || param->analyse.i_me_range < h->param.analyse.i_me_range )
1218         COPY( analyse.i_me_range );
1219     COPY( analyse.i_noise_reduction );
1220     /* We can't switch out of subme=0 during encoding. */
1221     if( h->param.analyse.i_subpel_refine )
1222         COPY( analyse.i_subpel_refine );
1223     COPY( analyse.i_trellis );
1224     COPY( analyse.b_chroma_me );
1225     COPY( analyse.b_dct_decimate );
1226     COPY( analyse.b_fast_pskip );
1227     COPY( analyse.b_mixed_references );
1228     COPY( analyse.f_psy_rd );
1229     COPY( analyse.f_psy_trellis );
1230     COPY( crop_rect );
1231     // can only twiddle these if they were enabled to begin with:
1232     if( h->param.analyse.i_me_method >= X264_ME_ESA || param->analyse.i_me_method < X264_ME_ESA )
1233         COPY( analyse.i_me_method );
1234     if( h->param.analyse.i_me_method >= X264_ME_ESA && !h->frames.b_have_sub8x8_esa )
1235         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
1236     if( h->pps->b_transform_8x8_mode )
1237         COPY( analyse.b_transform_8x8 );
1238     if( h->frames.i_max_ref1 > 1 )
1239         COPY( i_bframe_pyramid );
1240     COPY( i_slice_max_size );
1241     COPY( i_slice_max_mbs );
1242     COPY( i_slice_count );
1243     COPY( b_tff );
1244
1245     /* VBV can't be turned on if it wasn't on to begin with */
1246     if( h->param.rc.i_vbv_max_bitrate > 0 && h->param.rc.i_vbv_buffer_size > 0 &&
1247           param->rc.i_vbv_max_bitrate > 0 &&   param->rc.i_vbv_buffer_size > 0 )
1248     {
1249         COPY( rc.i_vbv_max_bitrate );
1250         COPY( rc.i_vbv_buffer_size );
1251         COPY( rc.i_bitrate );
1252         rc_reconfig = 1;
1253     }
1254     if( h->param.rc.f_rf_constant != param->rc.f_rf_constant )
1255     {
1256         COPY( rc.f_rf_constant );
1257         rc_reconfig = 1;
1258     }
1259     if( h->param.rc.f_rf_constant_max != param->rc.f_rf_constant_max )
1260     {
1261         COPY( rc.f_rf_constant_max );
1262         rc_reconfig = 1;
1263     }
1264
1265 #undef COPY
1266
1267     mbcmp_init( h );
1268
1269     int ret = x264_validate_parameters( h );
1270
1271     /* Supported reconfiguration options (1-pass only):
1272      * vbv-maxrate
1273      * vbv-bufsize
1274      * crf
1275      * bitrate (CBR only) */
1276     if( !ret && rc_reconfig )
1277         x264_ratecontrol_init_reconfigurable( h, 0 );
1278
1279     return ret;
1280 }
1281
1282 /****************************************************************************
1283  * x264_encoder_parameters:
1284  ****************************************************************************/
1285 void x264_encoder_parameters( x264_t *h, x264_param_t *param )
1286 {
1287     memcpy( param, &h->thread[h->i_thread_phase]->param, sizeof(x264_param_t) );
1288 }
1289
1290 /* internal usage */
1291 static void x264_nal_start( x264_t *h, int i_type, int i_ref_idc )
1292 {
1293     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1294
1295     nal->i_ref_idc        = i_ref_idc;
1296     nal->i_type           = i_type;
1297     nal->b_long_startcode = 1;
1298
1299     nal->i_payload= 0;
1300     nal->p_payload= &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8];
1301 }
1302
1303 /* if number of allocated nals is not enough, re-allocate a larger one. */
1304 static int x264_nal_check_buffer( x264_t *h )
1305 {
1306     if( h->out.i_nal >= h->out.i_nals_allocated )
1307     {
1308         x264_nal_t *new_out = x264_malloc( sizeof(x264_nal_t) * (h->out.i_nals_allocated*2) );
1309         if( !new_out )
1310             return -1;
1311         memcpy( new_out, h->out.nal, sizeof(x264_nal_t) * (h->out.i_nals_allocated) );
1312         x264_free( h->out.nal );
1313         h->out.nal = new_out;
1314         h->out.i_nals_allocated *= 2;
1315     }
1316     return 0;
1317 }
1318
1319 static int x264_nal_end( x264_t *h )
1320 {
1321     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1322     nal->i_payload = &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8] - nal->p_payload;
1323     if( h->param.nalu_process )
1324         h->param.nalu_process( h, nal );
1325     h->out.i_nal++;
1326
1327     return x264_nal_check_buffer( h );
1328 }
1329
1330 static int x264_encoder_encapsulate_nals( x264_t *h, int start )
1331 {
1332     int nal_size = 0, previous_nal_size = 0;
1333
1334     if( h->param.nalu_process )
1335     {
1336         for( int i = start; i < h->out.i_nal; i++ )
1337             nal_size += h->out.nal[i].i_payload;
1338         return nal_size;
1339     }
1340
1341     for( int i = 0; i < start; i++ )
1342         previous_nal_size += h->out.nal[i].i_payload;
1343
1344     for( int i = start; i < h->out.i_nal; i++ )
1345         nal_size += h->out.nal[i].i_payload;
1346
1347     /* Worst-case NAL unit escaping: reallocate the buffer if it's too small. */
1348     if( h->nal_buffer_size < nal_size * 3/2 + h->out.i_nal * 4 )
1349     {
1350         uint8_t *buf = x264_malloc( nal_size * 2 + h->out.i_nal * 4 );
1351         if( !buf )
1352             return -1;
1353         if( previous_nal_size )
1354             memcpy( buf, h->nal_buffer, previous_nal_size );
1355         x264_free( h->nal_buffer );
1356         h->nal_buffer = buf;
1357     }
1358
1359     uint8_t *nal_buffer = h->nal_buffer + previous_nal_size;
1360
1361     for( int i = start; i < h->out.i_nal; i++ )
1362     {
1363         h->out.nal[i].b_long_startcode = !i || h->out.nal[i].i_type == NAL_SPS || h->out.nal[i].i_type == NAL_PPS;
1364         x264_nal_encode( h, nal_buffer, &h->out.nal[i] );
1365         nal_buffer += h->out.nal[i].i_payload;
1366     }
1367
1368     x264_emms();
1369
1370     return nal_buffer - (h->nal_buffer + previous_nal_size);
1371 }
1372
1373 /****************************************************************************
1374  * x264_encoder_headers:
1375  ****************************************************************************/
1376 int x264_encoder_headers( x264_t *h, x264_nal_t **pp_nal, int *pi_nal )
1377 {
1378     int frame_size = 0;
1379     /* init bitstream context */
1380     h->out.i_nal = 0;
1381     bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
1382
1383     /* Write SEI, SPS and PPS. */
1384
1385     /* generate sequence parameters */
1386     x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
1387     x264_sps_write( &h->out.bs, h->sps );
1388     if( x264_nal_end( h ) )
1389         return -1;
1390
1391     /* generate picture parameters */
1392     x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
1393     x264_pps_write( &h->out.bs, h->pps );
1394     if( x264_nal_end( h ) )
1395         return -1;
1396
1397     /* identify ourselves */
1398     x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
1399     if( x264_sei_version_write( h, &h->out.bs ) )
1400         return -1;
1401     if( x264_nal_end( h ) )
1402         return -1;
1403
1404     frame_size = x264_encoder_encapsulate_nals( h, 0 );
1405
1406     /* now set output*/
1407     *pi_nal = h->out.i_nal;
1408     *pp_nal = &h->out.nal[0];
1409     h->out.i_nal = 0;
1410
1411     return frame_size;
1412 }
1413
1414 /* Check to see whether we have chosen a reference list ordering different
1415  * from the standard's default. */
1416 static inline void x264_reference_check_reorder( x264_t *h )
1417 {
1418     /* The reorder check doesn't check for missing frames, so just
1419      * force a reorder if one of the reference list is corrupt. */
1420     for( int i = 0; h->frames.reference[i]; i++ )
1421         if( h->frames.reference[i]->b_corrupt )
1422         {
1423             h->b_ref_reorder[0] = 1;
1424             return;
1425         }
1426     for( int list = 0; list <= (h->sh.i_type == SLICE_TYPE_B); list++ )
1427         for( int i = 0; i < h->i_ref[list] - 1; i++ )
1428         {
1429             int framenum_diff = h->fref[list][i+1]->i_frame_num - h->fref[list][i]->i_frame_num;
1430             int poc_diff = h->fref[list][i+1]->i_poc - h->fref[list][i]->i_poc;
1431             /* P and B-frames use different default orders. */
1432             if( h->sh.i_type == SLICE_TYPE_P ? framenum_diff > 0 : list == 1 ? poc_diff < 0 : poc_diff > 0 )
1433             {
1434                 h->b_ref_reorder[list] = 1;
1435                 return;
1436             }
1437         }
1438 }
1439
1440 /* return -1 on failure, else return the index of the new reference frame */
1441 int x264_weighted_reference_duplicate( x264_t *h, int i_ref, const x264_weight_t *w )
1442 {
1443     int i = h->i_ref[0];
1444     int j = 1;
1445     x264_frame_t *newframe;
1446     if( i <= 1 ) /* empty list, definitely can't duplicate frame */
1447         return -1;
1448
1449     //Duplication is only used in X264_WEIGHTP_SMART
1450     if( h->param.analyse.i_weighted_pred != X264_WEIGHTP_SMART )
1451         return -1;
1452
1453     /* Duplication is a hack to compensate for crappy rounding in motion compensation.
1454      * With high bit depth, it's not worth doing, so turn it off except in the case of
1455      * unweighted dupes. */
1456     if( BIT_DEPTH > 8 && w != weight_none )
1457         return -1;
1458
1459     newframe = x264_frame_pop_blank_unused( h );
1460     if( !newframe )
1461         return -1;
1462
1463     //FIXME: probably don't need to copy everything
1464     *newframe = *h->fref[0][i_ref];
1465     newframe->i_reference_count = 1;
1466     newframe->orig = h->fref[0][i_ref];
1467     newframe->b_duplicate = 1;
1468     memcpy( h->fenc->weight[j], w, sizeof(h->fenc->weight[i]) );
1469
1470     /* shift the frames to make space for the dupe. */
1471     h->b_ref_reorder[0] = 1;
1472     if( h->i_ref[0] < X264_REF_MAX )
1473         ++h->i_ref[0];
1474     h->fref[0][X264_REF_MAX-1] = NULL;
1475     x264_frame_unshift( &h->fref[0][j], newframe );
1476
1477     return j;
1478 }
1479
1480 static void x264_weighted_pred_init( x264_t *h )
1481 {
1482     /* for now no analysis and set all weights to nothing */
1483     for( int i_ref = 0; i_ref < h->i_ref[0]; i_ref++ )
1484         h->fenc->weighted[i_ref] = h->fref[0][i_ref]->filtered[0];
1485
1486     // FIXME: This only supports weighting of one reference frame
1487     // and duplicates of that frame.
1488     h->fenc->i_lines_weighted = 0;
1489
1490     for( int i_ref = 0; i_ref < (h->i_ref[0] << h->sh.b_mbaff); i_ref++ )
1491         for( int i = 0; i < 3; i++ )
1492             h->sh.weight[i_ref][i].weightfn = NULL;
1493
1494
1495     if( h->sh.i_type != SLICE_TYPE_P || h->param.analyse.i_weighted_pred <= 0 )
1496         return;
1497
1498     int i_padv = PADV << h->param.b_interlaced;
1499     int denom = -1;
1500     int weightplane[2] = { 0, 0 };
1501     int buffer_next = 0;
1502     for( int i = 0; i < 3; i++ )
1503     {
1504         for( int j = 0; j < h->i_ref[0]; j++ )
1505         {
1506             if( h->fenc->weight[j][i].weightfn )
1507             {
1508                 h->sh.weight[j][i] = h->fenc->weight[j][i];
1509                 // if weight is useless, don't write it to stream
1510                 if( h->sh.weight[j][i].i_scale == 1<<h->sh.weight[j][i].i_denom && h->sh.weight[j][i].i_offset == 0 )
1511                     h->sh.weight[j][i].weightfn = NULL;
1512                 else
1513                 {
1514                     if( !weightplane[!!i] )
1515                     {
1516                         weightplane[!!i] = 1;
1517                         h->sh.weight[0][!!i].i_denom = denom = h->sh.weight[j][i].i_denom;
1518                         assert( x264_clip3( denom, 0, 7 ) == denom );
1519                     }
1520
1521                     assert( h->sh.weight[j][i].i_denom == denom );
1522                     if( !i )
1523                     {
1524                         h->fenc->weighted[j] = h->mb.p_weight_buf[buffer_next++] + h->fenc->i_stride[0] * i_padv + PADH;
1525                         //scale full resolution frame
1526                         if( h->param.i_threads == 1 )
1527                         {
1528                             pixel *src = h->fref[0][j]->filtered[0] - h->fref[0][j]->i_stride[0]*i_padv - PADH;
1529                             pixel *dst = h->fenc->weighted[j] - h->fenc->i_stride[0]*i_padv - PADH;
1530                             int stride = h->fenc->i_stride[0];
1531                             int width = h->fenc->i_width[0] + PADH*2;
1532                             int height = h->fenc->i_lines[0] + i_padv*2;
1533                             x264_weight_scale_plane( h, dst, stride, src, stride, width, height, &h->sh.weight[j][0] );
1534                             h->fenc->i_lines_weighted = height;
1535                         }
1536                     }
1537                 }
1538             }
1539         }
1540     }
1541
1542     if( weightplane[1] )
1543         for( int i = 0; i < h->i_ref[0]; i++ )
1544         {
1545             if( h->sh.weight[i][1].weightfn && !h->sh.weight[i][2].weightfn )
1546             {
1547                 h->sh.weight[i][2].i_scale = 1 << h->sh.weight[0][1].i_denom;
1548                 h->sh.weight[i][2].i_offset = 0;
1549             }
1550             else if( h->sh.weight[i][2].weightfn && !h->sh.weight[i][1].weightfn )
1551             {
1552                 h->sh.weight[i][1].i_scale = 1 << h->sh.weight[0][1].i_denom;
1553                 h->sh.weight[i][1].i_offset = 0;
1554             }
1555         }
1556
1557     if( !weightplane[0] )
1558         h->sh.weight[0][0].i_denom = 0;
1559     if( !weightplane[1] )
1560         h->sh.weight[0][1].i_denom = 0;
1561     h->sh.weight[0][2].i_denom = h->sh.weight[0][1].i_denom;
1562 }
1563
1564 static inline int x264_reference_distance( x264_t *h, x264_frame_t *frame )
1565 {
1566     if( h->param.i_frame_packing == 5 )
1567         return abs((h->fenc->i_frame&~1) - (frame->i_frame&~1)) +
1568                   ((h->fenc->i_frame&1) != (frame->i_frame&1));
1569     else
1570         return abs(h->fenc->i_frame - frame->i_frame);
1571 }
1572
1573 static inline void x264_reference_build_list( x264_t *h, int i_poc )
1574 {
1575     int b_ok;
1576
1577     /* build ref list 0/1 */
1578     h->mb.pic.i_fref[0] = h->i_ref[0] = 0;
1579     h->mb.pic.i_fref[1] = h->i_ref[1] = 0;
1580     if( h->sh.i_type == SLICE_TYPE_I )
1581         return;
1582
1583     for( int i = 0; h->frames.reference[i]; i++ )
1584     {
1585         if( h->frames.reference[i]->b_corrupt )
1586             continue;
1587         if( h->frames.reference[i]->i_poc < i_poc )
1588             h->fref[0][h->i_ref[0]++] = h->frames.reference[i];
1589         else if( h->frames.reference[i]->i_poc > i_poc )
1590             h->fref[1][h->i_ref[1]++] = h->frames.reference[i];
1591     }
1592
1593     /* Order reference lists by distance from the current frame. */
1594     for( int list = 0; list < 2; list++ )
1595     {
1596         h->fref_nearest[list] = h->fref[list][0];
1597         do
1598         {
1599             b_ok = 1;
1600             for( int i = 0; i < h->i_ref[list] - 1; i++ )
1601             {
1602                 if( list ? h->fref[list][i+1]->i_poc < h->fref_nearest[list]->i_poc
1603                          : h->fref[list][i+1]->i_poc > h->fref_nearest[list]->i_poc )
1604                     h->fref_nearest[list] = h->fref[list][i+1];
1605                 if( x264_reference_distance( h, h->fref[list][i] ) > x264_reference_distance( h, h->fref[list][i+1] ) )
1606                 {
1607                     XCHG( x264_frame_t*, h->fref[list][i], h->fref[list][i+1] );
1608                     b_ok = 0;
1609                     break;
1610                 }
1611             }
1612         } while( !b_ok );
1613     }
1614
1615     if( h->sh.i_mmco_remove_from_end )
1616         for( int i = h->i_ref[0]-1; i >= h->i_ref[0] - h->sh.i_mmco_remove_from_end; i-- )
1617         {
1618             int diff = h->i_frame_num - h->fref[0][i]->i_frame_num;
1619             h->sh.mmco[h->sh.i_mmco_command_count].i_poc = h->fref[0][i]->i_poc;
1620             h->sh.mmco[h->sh.i_mmco_command_count++].i_difference_of_pic_nums = diff;
1621         }
1622
1623     x264_reference_check_reorder( h );
1624
1625     h->i_ref[1] = X264_MIN( h->i_ref[1], h->frames.i_max_ref1 );
1626     h->i_ref[0] = X264_MIN( h->i_ref[0], h->frames.i_max_ref0 );
1627     h->i_ref[0] = X264_MIN( h->i_ref[0], h->param.i_frame_reference ); // if reconfig() has lowered the limit
1628
1629     /* add duplicates */
1630     if( h->fenc->i_type == X264_TYPE_P )
1631     {
1632         int idx = -1;
1633         if( h->param.analyse.i_weighted_pred >= X264_WEIGHTP_SIMPLE )
1634         {
1635             x264_weight_t w[3];
1636             w[1].weightfn = w[2].weightfn = NULL;
1637             if( h->param.rc.b_stat_read )
1638                 x264_ratecontrol_set_weights( h, h->fenc );
1639
1640             if( !h->fenc->weight[0][0].weightfn )
1641             {
1642                 h->fenc->weight[0][0].i_denom = 0;
1643                 SET_WEIGHT( w[0], 1, 1, 0, -1 );
1644                 idx = x264_weighted_reference_duplicate( h, 0, w );
1645             }
1646             else
1647             {
1648                 if( h->fenc->weight[0][0].i_scale == 1<<h->fenc->weight[0][0].i_denom )
1649                 {
1650                     SET_WEIGHT( h->fenc->weight[0][0], 1, 1, 0, h->fenc->weight[0][0].i_offset );
1651                 }
1652                 x264_weighted_reference_duplicate( h, 0, weight_none );
1653                 if( h->fenc->weight[0][0].i_offset > -128 )
1654                 {
1655                     w[0] = h->fenc->weight[0][0];
1656                     w[0].i_offset--;
1657                     h->mc.weight_cache( h, &w[0] );
1658                     idx = x264_weighted_reference_duplicate( h, 0, w );
1659                 }
1660             }
1661         }
1662         h->mb.ref_blind_dupe = idx;
1663     }
1664
1665     assert( h->i_ref[0] + h->i_ref[1] <= X264_REF_MAX );
1666     h->mb.pic.i_fref[0] = h->i_ref[0];
1667     h->mb.pic.i_fref[1] = h->i_ref[1];
1668 }
1669
1670 static void x264_fdec_filter_row( x264_t *h, int mb_y, int b_inloop )
1671 {
1672     /* mb_y is the mb to be encoded next, not the mb to be filtered here */
1673     int b_hpel = h->fdec->b_kept_as_ref;
1674     int b_deblock = h->sh.i_disable_deblocking_filter_idc != 1;
1675     int b_end = mb_y == h->i_threadslice_end;
1676     int b_measure_quality = 1;
1677     int min_y = mb_y - (1 << h->sh.b_mbaff);
1678     int b_start = min_y == h->i_threadslice_start;
1679     int max_y = b_end ? h->i_threadslice_end : mb_y;
1680     b_deblock &= b_hpel || h->param.psz_dump_yuv;
1681     if( h->param.b_sliced_threads && b_start && min_y && !b_inloop )
1682     {
1683         b_deblock = 0;         /* We already deblocked on the inloop pass. */
1684         b_measure_quality = 0; /* We already measured quality on the inloop pass. */
1685     }
1686     if( mb_y & h->sh.b_mbaff )
1687         return;
1688     if( min_y < h->i_threadslice_start )
1689         return;
1690
1691     if( b_deblock )
1692         for( int y = min_y; y < max_y; y += (1 << h->sh.b_mbaff) )
1693             x264_frame_deblock_row( h, y );
1694
1695     if( b_hpel )
1696     {
1697         int end = mb_y == h->mb.i_mb_height;
1698         x264_frame_expand_border( h, h->fdec, min_y, end );
1699         if( h->param.analyse.i_subpel_refine )
1700         {
1701             x264_frame_filter( h, h->fdec, min_y, end );
1702             x264_frame_expand_border_filtered( h, h->fdec, min_y, end );
1703         }
1704     }
1705
1706     if( h->i_thread_frames > 1 && h->fdec->b_kept_as_ref )
1707         x264_frame_cond_broadcast( h->fdec, mb_y*16 + (b_end ? 10000 : -(X264_THREAD_HEIGHT << h->sh.b_mbaff)) );
1708
1709     min_y = min_y*16 - 8 * !b_start;
1710     max_y = b_end ? X264_MIN( h->i_threadslice_end*16 , h->param.i_height ) : mb_y*16 - 8;
1711
1712     if( b_measure_quality )
1713     {
1714         if( h->param.analyse.b_psnr )
1715         {
1716             uint64_t ssd_y = x264_pixel_ssd_wxh( &h->pixf,
1717                 h->fdec->plane[0] + min_y * h->fdec->i_stride[0], h->fdec->i_stride[0],
1718                 h->fenc->plane[0] + min_y * h->fenc->i_stride[0], h->fenc->i_stride[0],
1719                 h->param.i_width, max_y-min_y );
1720             uint64_t ssd_u, ssd_v;
1721             x264_pixel_ssd_nv12( &h->pixf,
1722                 h->fdec->plane[1] + (min_y>>1) * h->fdec->i_stride[1], h->fdec->i_stride[1],
1723                 h->fenc->plane[1] + (min_y>>1) * h->fenc->i_stride[1], h->fenc->i_stride[1],
1724                 h->param.i_width>>1, (max_y-min_y)>>1, &ssd_u, &ssd_v );
1725             h->stat.frame.i_ssd[0] += ssd_y;
1726             h->stat.frame.i_ssd[1] += ssd_u;
1727             h->stat.frame.i_ssd[2] += ssd_v;
1728         }
1729
1730         if( h->param.analyse.b_ssim )
1731         {
1732             x264_emms();
1733             /* offset by 2 pixels to avoid alignment of ssim blocks with dct blocks,
1734              * and overlap by 4 */
1735             min_y += b_start ? 2 : -6;
1736             h->stat.frame.f_ssim +=
1737                 x264_pixel_ssim_wxh( &h->pixf,
1738                     h->fdec->plane[0] + 2+min_y*h->fdec->i_stride[0], h->fdec->i_stride[0],
1739                     h->fenc->plane[0] + 2+min_y*h->fenc->i_stride[0], h->fenc->i_stride[0],
1740                     h->param.i_width-2, max_y-min_y, h->scratch_buffer );
1741         }
1742     }
1743 }
1744
1745 static inline int x264_reference_update( x264_t *h )
1746 {
1747     if( !h->fdec->b_kept_as_ref )
1748     {
1749         if( h->i_thread_frames > 1 )
1750         {
1751             x264_frame_push_unused( h, h->fdec );
1752             h->fdec = x264_frame_pop_unused( h, 1 );
1753             if( !h->fdec )
1754                 return -1;
1755         }
1756         return 0;
1757     }
1758
1759     /* apply mmco from previous frame. */
1760     for( int i = 0; i < h->sh.i_mmco_command_count; i++ )
1761         for( int j = 0; h->frames.reference[j]; j++ )
1762             if( h->frames.reference[j]->i_poc == h->sh.mmco[i].i_poc )
1763                 x264_frame_push_unused( h, x264_frame_shift( &h->frames.reference[j] ) );
1764
1765     /* move frame in the buffer */
1766     x264_frame_push( h->frames.reference, h->fdec );
1767     if( h->frames.reference[h->sps->i_num_ref_frames] )
1768         x264_frame_push_unused( h, x264_frame_shift( h->frames.reference ) );
1769     h->fdec = x264_frame_pop_unused( h, 1 );
1770     if( !h->fdec )
1771         return -1;
1772     return 0;
1773 }
1774
1775 static inline void x264_reference_reset( x264_t *h )
1776 {
1777     while( h->frames.reference[0] )
1778         x264_frame_push_unused( h, x264_frame_pop( h->frames.reference ) );
1779     h->fdec->i_poc =
1780     h->fenc->i_poc = 0;
1781 }
1782
1783 static inline void x264_reference_hierarchy_reset( x264_t *h )
1784 {
1785     int ref;
1786     int b_hasdelayframe = 0;
1787
1788     /* look for delay frames -- chain must only contain frames that are disposable */
1789     for( int i = 0; h->frames.current[i] && IS_DISPOSABLE( h->frames.current[i]->i_type ); i++ )
1790         b_hasdelayframe |= h->frames.current[i]->i_coded
1791                         != h->frames.current[i]->i_frame + h->sps->vui.i_num_reorder_frames;
1792
1793     /* This function must handle b-pyramid and clear frames for open-gop */
1794     if( h->param.i_bframe_pyramid != X264_B_PYRAMID_STRICT && !b_hasdelayframe && h->frames.i_poc_last_open_gop == -1 )
1795         return;
1796
1797     /* Remove last BREF. There will never be old BREFs in the
1798      * dpb during a BREF decode when pyramid == STRICT */
1799     for( ref = 0; h->frames.reference[ref]; ref++ )
1800     {
1801         if( ( h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT
1802             && h->frames.reference[ref]->i_type == X264_TYPE_BREF )
1803             || ( h->frames.reference[ref]->i_poc < h->frames.i_poc_last_open_gop
1804             && h->sh.i_type != SLICE_TYPE_B ) )
1805         {
1806             int diff = h->i_frame_num - h->frames.reference[ref]->i_frame_num;
1807             h->sh.mmco[h->sh.i_mmco_command_count].i_difference_of_pic_nums = diff;
1808             h->sh.mmco[h->sh.i_mmco_command_count++].i_poc = h->frames.reference[ref]->i_poc;
1809             x264_frame_push_unused( h, x264_frame_shift( &h->frames.reference[ref] ) );
1810             h->b_ref_reorder[0] = 1;
1811             ref--;
1812         }
1813     }
1814
1815     /* Prepare room in the dpb for the delayed display time of the later b-frame's */
1816     if( h->param.i_bframe_pyramid )
1817         h->sh.i_mmco_remove_from_end = X264_MAX( ref + 2 - h->frames.i_max_dpb, 0 );
1818 }
1819
1820 static inline void x264_slice_init( x264_t *h, int i_nal_type, int i_global_qp )
1821 {
1822     /* ------------------------ Create slice header  ----------------------- */
1823     if( i_nal_type == NAL_SLICE_IDR )
1824     {
1825         x264_slice_header_init( h, &h->sh, h->sps, h->pps, h->i_idr_pic_id, h->i_frame_num, i_global_qp );
1826
1827         /* alternate id */
1828         h->i_idr_pic_id ^= 1;
1829     }
1830     else
1831     {
1832         x264_slice_header_init( h, &h->sh, h->sps, h->pps, -1, h->i_frame_num, i_global_qp );
1833
1834         h->sh.i_num_ref_idx_l0_active = h->i_ref[0] <= 0 ? 1 : h->i_ref[0];
1835         h->sh.i_num_ref_idx_l1_active = h->i_ref[1] <= 0 ? 1 : h->i_ref[1];
1836         if( h->sh.i_num_ref_idx_l0_active != h->pps->i_num_ref_idx_l0_default_active ||
1837             (h->sh.i_type == SLICE_TYPE_B && h->sh.i_num_ref_idx_l1_active != h->pps->i_num_ref_idx_l1_default_active) )
1838         {
1839             h->sh.b_num_ref_idx_override = 1;
1840         }
1841     }
1842
1843     h->fdec->i_frame_num = h->sh.i_frame_num;
1844
1845     if( h->sps->i_poc_type == 0 )
1846     {
1847         h->sh.i_poc = h->fdec->i_poc;
1848         if( h->param.b_interlaced )
1849         {
1850             h->sh.i_delta_poc_bottom = h->param.b_tff ? 1 : -1;
1851             h->sh.i_poc += h->sh.i_delta_poc_bottom == -1;
1852         }
1853         else
1854             h->sh.i_delta_poc_bottom = 0;
1855         h->fdec->i_delta_poc[0] = h->sh.i_delta_poc_bottom == -1;
1856         h->fdec->i_delta_poc[1] = h->sh.i_delta_poc_bottom ==  1;
1857     }
1858     else if( h->sps->i_poc_type == 1 )
1859     {
1860         /* FIXME TODO FIXME */
1861     }
1862     else
1863     {
1864         /* Nothing to do ? */
1865     }
1866
1867     x264_macroblock_slice_init( h );
1868 }
1869
1870 static int x264_slice_write( x264_t *h )
1871 {
1872     int i_skip;
1873     int mb_xy, i_mb_x, i_mb_y;
1874     int i_skip_bak = 0; /* Shut up GCC. */
1875     bs_t UNINIT(bs_bak);
1876     x264_cabac_t cabac_bak;
1877     uint8_t cabac_prevbyte_bak = 0; /* Shut up GCC. */
1878     int mv_bits_bak = 0;
1879     int tex_bits_bak = 0;
1880     /* NALUs other than the first use a 3-byte startcode.
1881      * Add one extra byte for the rbsp, and one more for the final CABAC putbyte.
1882      * Then add an extra 5 bytes just in case, to account for random NAL escapes and
1883      * other inaccuracies. */
1884     int overhead_guess = (NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal)) + 1 + h->param.b_cabac + 5;
1885     int slice_max_size = h->param.i_slice_max_size > 0 ? (h->param.i_slice_max_size-overhead_guess)*8 : 0;
1886     int back_up_bitstream = slice_max_size || (!h->param.b_cabac && h->sps->i_profile_idc < PROFILE_HIGH);
1887     int starting_bits = bs_pos(&h->out.bs);
1888     int b_deblock = h->sh.i_disable_deblocking_filter_idc != 1;
1889     int b_hpel = h->fdec->b_kept_as_ref;
1890     uint8_t *last_emu_check;
1891     b_deblock &= b_hpel || h->param.psz_dump_yuv;
1892     bs_realign( &h->out.bs );
1893
1894     /* Slice */
1895     x264_nal_start( h, h->i_nal_type, h->i_nal_ref_idc );
1896     h->out.nal[h->out.i_nal].i_first_mb = h->sh.i_first_mb;
1897
1898     /* Slice header */
1899     x264_macroblock_thread_init( h );
1900
1901     /* If this isn't the first slice in the threadslice, set the slice QP
1902      * equal to the last QP in the previous slice for more accurate
1903      * CABAC initialization. */
1904     if( h->sh.i_first_mb != h->i_threadslice_start * h->mb.i_mb_width )
1905     {
1906         h->sh.i_qp = h->mb.i_last_qp;
1907         h->sh.i_qp_delta = h->sh.i_qp - h->pps->i_pic_init_qp;
1908     }
1909
1910     x264_slice_header_write( &h->out.bs, &h->sh, h->i_nal_ref_idc );
1911     if( h->param.b_cabac )
1912     {
1913         /* alignment needed */
1914         bs_align_1( &h->out.bs );
1915
1916         /* init cabac */
1917         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 );
1918         x264_cabac_encode_init ( &h->cabac, h->out.bs.p, h->out.bs.p_end );
1919         last_emu_check = h->cabac.p;
1920     }
1921     else
1922         last_emu_check = h->out.bs.p;
1923     h->mb.i_last_qp = h->sh.i_qp;
1924     h->mb.i_last_dqp = 0;
1925
1926     i_mb_y = h->sh.i_first_mb / h->mb.i_mb_width;
1927     i_mb_x = h->sh.i_first_mb % h->mb.i_mb_width;
1928     i_skip = 0;
1929
1930     while( (mb_xy = i_mb_x + i_mb_y * h->mb.i_mb_width) <= h->sh.i_last_mb )
1931     {
1932         int mb_spos = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
1933
1934         if( x264_bitstream_check_buffer( h ) )
1935             return -1;
1936
1937         if( back_up_bitstream )
1938         {
1939             mv_bits_bak = h->stat.frame.i_mv_bits;
1940             tex_bits_bak = h->stat.frame.i_tex_bits;
1941             /* We don't need the contexts because flushing the CABAC encoder has no context
1942              * dependency and macroblocks are only re-encoded in the case where a slice is
1943              * ended (and thus the content of all contexts are thrown away). */
1944             if( h->param.b_cabac )
1945             {
1946                 memcpy( &cabac_bak, &h->cabac, offsetof(x264_cabac_t, f8_bits_encoded) );
1947                 /* x264's CABAC writer modifies the previous byte during carry, so it has to be
1948                  * backed up. */
1949                 cabac_prevbyte_bak = h->cabac.p[-1];
1950             }
1951             else
1952             {
1953                 bs_bak = h->out.bs;
1954                 i_skip_bak = i_skip;
1955             }
1956         }
1957
1958         if( i_mb_x == 0 && !h->mb.b_reencode_mb )
1959             x264_fdec_filter_row( h, i_mb_y, 1 );
1960
1961         /* load cache */
1962         x264_macroblock_cache_load( h, i_mb_x, i_mb_y );
1963
1964         x264_macroblock_analyse( h );
1965
1966         /* encode this macroblock -> be careful it can change the mb type to P_SKIP if needed */
1967 reencode:
1968         x264_macroblock_encode( h );
1969
1970         if( h->param.b_cabac )
1971         {
1972             if( mb_xy > h->sh.i_first_mb && !(h->sh.b_mbaff && (i_mb_y&1)) )
1973                 x264_cabac_encode_terminal( &h->cabac );
1974
1975             if( IS_SKIP( h->mb.i_type ) )
1976                 x264_cabac_mb_skip( h, 1 );
1977             else
1978             {
1979                 if( h->sh.i_type != SLICE_TYPE_I )
1980                     x264_cabac_mb_skip( h, 0 );
1981                 x264_macroblock_write_cabac( h, &h->cabac );
1982             }
1983         }
1984         else
1985         {
1986             if( IS_SKIP( h->mb.i_type ) )
1987                 i_skip++;
1988             else
1989             {
1990                 if( h->sh.i_type != SLICE_TYPE_I )
1991                 {
1992                     bs_write_ue( &h->out.bs, i_skip );  /* skip run */
1993                     i_skip = 0;
1994                 }
1995                 x264_macroblock_write_cavlc( h );
1996                 /* If there was a CAVLC level code overflow, try again at a higher QP. */
1997                 if( h->mb.b_overflow )
1998                 {
1999                     h->mb.i_chroma_qp = h->chroma_qp_table[++h->mb.i_qp];
2000                     h->mb.i_skip_intra = 0;
2001                     h->mb.b_skip_mc = 0;
2002                     h->mb.b_overflow = 0;
2003                     h->out.bs = bs_bak;
2004                     i_skip = i_skip_bak;
2005                     h->stat.frame.i_mv_bits = mv_bits_bak;
2006                     h->stat.frame.i_tex_bits = tex_bits_bak;
2007                     goto reencode;
2008                 }
2009             }
2010         }
2011
2012         int total_bits = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
2013         int mb_size = total_bits - mb_spos;
2014
2015         if( slice_max_size )
2016         {
2017             /* Count the skip run, just in case. */
2018             if( !h->param.b_cabac )
2019                 total_bits += bs_size_ue_big( i_skip );
2020             /* Check for escape bytes. */
2021             uint8_t *end = h->param.b_cabac ? h->cabac.p : h->out.bs.p;
2022             for( ; last_emu_check < end - 2; last_emu_check++ )
2023                 if( last_emu_check[0] == 0 && last_emu_check[1] == 0 && last_emu_check[2] <= 3 )
2024                 {
2025                     slice_max_size -= 8;
2026                     last_emu_check++;
2027                 }
2028             /* We'll just re-encode this last macroblock if we go over the max slice size. */
2029             if( total_bits - starting_bits > slice_max_size && !h->mb.b_reencode_mb )
2030             {
2031                 if( mb_xy != h->sh.i_first_mb )
2032                 {
2033                     h->stat.frame.i_mv_bits = mv_bits_bak;
2034                     h->stat.frame.i_tex_bits = tex_bits_bak;
2035                     if( h->param.b_cabac )
2036                     {
2037                         memcpy( &h->cabac, &cabac_bak, offsetof(x264_cabac_t, f8_bits_encoded) );
2038                         h->cabac.p[-1] = cabac_prevbyte_bak;
2039                     }
2040                     else
2041                     {
2042                         h->out.bs = bs_bak;
2043                         i_skip = i_skip_bak;
2044                     }
2045                     h->mb.b_reencode_mb = 1;
2046                     h->sh.i_last_mb = mb_xy-1;
2047                     break;
2048                 }
2049                 else
2050                 {
2051                     h->sh.i_last_mb = mb_xy;
2052                     h->mb.b_reencode_mb = 0;
2053                 }
2054             }
2055             else
2056                 h->mb.b_reencode_mb = 0;
2057         }
2058
2059 #if HAVE_VISUALIZE
2060         if( h->param.b_visualize )
2061             x264_visualize_mb( h );
2062 #endif
2063
2064         /* save cache */
2065         x264_macroblock_cache_save( h );
2066
2067         /* accumulate mb stats */
2068         h->stat.frame.i_mb_count[h->mb.i_type]++;
2069
2070         int b_intra = IS_INTRA( h->mb.i_type );
2071         if( h->param.i_log_level >= X264_LOG_INFO || h->param.rc.b_stat_write )
2072         {
2073             if( !b_intra && !IS_SKIP( h->mb.i_type ) && !IS_DIRECT( h->mb.i_type ) )
2074             {
2075                 if( h->mb.i_partition != D_8x8 )
2076                         h->stat.frame.i_mb_partition[h->mb.i_partition] += 4;
2077                     else
2078                         for( int i = 0; i < 4; i++ )
2079                             h->stat.frame.i_mb_partition[h->mb.i_sub_partition[i]] ++;
2080                 if( h->param.i_frame_reference > 1 )
2081                     for( int i_list = 0; i_list <= (h->sh.i_type == SLICE_TYPE_B); i_list++ )
2082                         for( int i = 0; i < 4; i++ )
2083                         {
2084                             int i_ref = h->mb.cache.ref[i_list][ x264_scan8[4*i] ];
2085                             if( i_ref >= 0 )
2086                                 h->stat.frame.i_mb_count_ref[i_list][i_ref] ++;
2087                         }
2088             }
2089         }
2090
2091         if( h->param.i_log_level >= X264_LOG_INFO )
2092         {
2093             if( h->mb.i_cbp_luma | h->mb.i_cbp_chroma )
2094             {
2095                 int cbpsum = (h->mb.i_cbp_luma&1) + ((h->mb.i_cbp_luma>>1)&1)
2096                            + ((h->mb.i_cbp_luma>>2)&1) + (h->mb.i_cbp_luma>>3);
2097                 h->stat.frame.i_mb_cbp[!b_intra + 0] += cbpsum;
2098                 h->stat.frame.i_mb_cbp[!b_intra + 2] += !!h->mb.i_cbp_chroma;
2099                 h->stat.frame.i_mb_cbp[!b_intra + 4] += h->mb.i_cbp_chroma >> 1;
2100             }
2101             if( h->mb.i_cbp_luma && !b_intra )
2102             {
2103                 h->stat.frame.i_mb_count_8x8dct[0] ++;
2104                 h->stat.frame.i_mb_count_8x8dct[1] += h->mb.b_transform_8x8;
2105             }
2106             if( b_intra && h->mb.i_type != I_PCM )
2107             {
2108                 if( h->mb.i_type == I_16x16 )
2109                     h->stat.frame.i_mb_pred_mode[0][h->mb.i_intra16x16_pred_mode]++;
2110                 else if( h->mb.i_type == I_8x8 )
2111                     for( int i = 0; i < 16; i += 4 )
2112                         h->stat.frame.i_mb_pred_mode[1][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
2113                 else //if( h->mb.i_type == I_4x4 )
2114                     for( int i = 0; i < 16; i++ )
2115                         h->stat.frame.i_mb_pred_mode[2][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
2116                 h->stat.frame.i_mb_pred_mode[3][x264_mb_pred_mode8x8c_fix[h->mb.i_chroma_pred_mode]]++;
2117             }
2118         }
2119
2120         /* calculate deblock strength values (actual deblocking is done per-row along with hpel) */
2121         if( b_deblock )
2122         {
2123             int mvy_limit = 4 >> h->sh.b_mbaff;
2124             uint8_t (*bs)[4][4] = h->deblock_strength[h->mb.i_mb_y&1][h->mb.i_mb_x];
2125             x264_macroblock_cache_load_deblock( h );
2126             if( IS_INTRA( h->mb.type[h->mb.i_mb_xy] ) )
2127                 memset( bs, 3, 2*4*4*sizeof(uint8_t) );
2128             else
2129                 h->loopf.deblock_strength( h->mb.cache.non_zero_count, h->mb.cache.ref, h->mb.cache.mv,
2130                                            bs, mvy_limit, h->sh.i_type == SLICE_TYPE_B );
2131         }
2132
2133         x264_ratecontrol_mb( h, mb_size );
2134
2135         if( h->sh.b_mbaff )
2136         {
2137             i_mb_x += i_mb_y & 1;
2138             i_mb_y ^= i_mb_x < h->mb.i_mb_width;
2139         }
2140         else
2141             i_mb_x++;
2142         if( i_mb_x == h->mb.i_mb_width )
2143         {
2144             i_mb_y++;
2145             i_mb_x = 0;
2146         }
2147     }
2148     h->out.nal[h->out.i_nal].i_last_mb = h->sh.i_last_mb;
2149
2150     if( h->param.b_cabac )
2151     {
2152         x264_cabac_encode_flush( h, &h->cabac );
2153         h->out.bs.p = h->cabac.p;
2154     }
2155     else
2156     {
2157         if( i_skip > 0 )
2158             bs_write_ue( &h->out.bs, i_skip );  /* last skip run */
2159         /* rbsp_slice_trailing_bits */
2160         bs_rbsp_trailing( &h->out.bs );
2161         bs_flush( &h->out.bs );
2162     }
2163     if( x264_nal_end( h ) )
2164         return -1;
2165
2166     if( h->sh.i_last_mb == (h->i_threadslice_end * h->mb.i_mb_width - 1) )
2167     {
2168         h->stat.frame.i_misc_bits = bs_pos( &h->out.bs )
2169                                   + (h->out.i_nal*NALU_OVERHEAD * 8)
2170                                   - h->stat.frame.i_tex_bits
2171                                   - h->stat.frame.i_mv_bits;
2172         x264_fdec_filter_row( h, h->i_threadslice_end, 1 );
2173     }
2174
2175     return 0;
2176 }
2177
2178 static void x264_thread_sync_context( x264_t *dst, x264_t *src )
2179 {
2180     if( dst == src )
2181         return;
2182
2183     // reference counting
2184     for( x264_frame_t **f = src->frames.reference; *f; f++ )
2185         (*f)->i_reference_count++;
2186     for( x264_frame_t **f = dst->frames.reference; *f; f++ )
2187         x264_frame_push_unused( src, *f );
2188     src->fdec->i_reference_count++;
2189     x264_frame_push_unused( src, dst->fdec );
2190
2191     // copy everything except the per-thread pointers and the constants.
2192     memcpy( &dst->i_frame, &src->i_frame, offsetof(x264_t, mb.type) - offsetof(x264_t, i_frame) );
2193     dst->param = src->param;
2194     dst->stat = src->stat;
2195 }
2196
2197 static void x264_thread_sync_stat( x264_t *dst, x264_t *src )
2198 {
2199     if( dst == src )
2200         return;
2201     memcpy( &dst->stat.i_frame_count, &src->stat.i_frame_count, sizeof(dst->stat) - sizeof(dst->stat.frame) );
2202 }
2203
2204 static void *x264_slices_write( x264_t *h )
2205 {
2206     int i_slice_num = 0;
2207     int last_thread_mb = h->sh.i_last_mb;
2208
2209 #if HAVE_VISUALIZE
2210     if( h->param.b_visualize )
2211         if( x264_visualize_init( h ) )
2212             return (void *)-1;
2213 #endif
2214
2215     /* init stats */
2216     memset( &h->stat.frame, 0, sizeof(h->stat.frame) );
2217     h->mb.b_reencode_mb = 0;
2218     while( h->sh.i_first_mb <= last_thread_mb )
2219     {
2220         h->sh.i_last_mb = last_thread_mb;
2221         if( h->param.i_slice_max_mbs )
2222             h->sh.i_last_mb = h->sh.i_first_mb + h->param.i_slice_max_mbs - 1;
2223         else if( h->param.i_slice_count && !h->param.b_sliced_threads )
2224         {
2225             int height = h->mb.i_mb_height >> h->param.b_interlaced;
2226             int width = h->mb.i_mb_width << h->param.b_interlaced;
2227             i_slice_num++;
2228             h->sh.i_last_mb = (height * i_slice_num + h->param.i_slice_count/2) / h->param.i_slice_count * width - 1;
2229         }
2230         h->sh.i_last_mb = X264_MIN( h->sh.i_last_mb, last_thread_mb );
2231         if( x264_stack_align( x264_slice_write, h ) )
2232             return (void *)-1;
2233         h->sh.i_first_mb = h->sh.i_last_mb + 1;
2234     }
2235
2236 #if HAVE_VISUALIZE
2237     if( h->param.b_visualize )
2238     {
2239         x264_visualize_show( h );
2240         x264_visualize_close( h );
2241     }
2242 #endif
2243
2244     return (void *)0;
2245 }
2246
2247 static int x264_threaded_slices_write( x264_t *h )
2248 {
2249     /* set first/last mb and sync contexts */
2250     for( int i = 0; i < h->param.i_threads; i++ )
2251     {
2252         x264_t *t = h->thread[i];
2253         if( i )
2254         {
2255             t->param = h->param;
2256             memcpy( &t->i_frame, &h->i_frame, offsetof(x264_t, rc) - offsetof(x264_t, i_frame) );
2257         }
2258         int height = h->mb.i_mb_height >> h->param.b_interlaced;
2259         t->i_threadslice_start = ((height *  i    + h->param.i_slice_count/2) / h->param.i_threads) << h->param.b_interlaced;
2260         t->i_threadslice_end   = ((height * (i+1) + h->param.i_slice_count/2) / h->param.i_threads) << h->param.b_interlaced;
2261         t->sh.i_first_mb = t->i_threadslice_start * h->mb.i_mb_width;
2262         t->sh.i_last_mb  =   t->i_threadslice_end * h->mb.i_mb_width - 1;
2263     }
2264
2265     x264_stack_align( x264_analyse_weight_frame, h, h->mb.i_mb_height*16 + 16 );
2266
2267     x264_threads_distribute_ratecontrol( h );
2268
2269     /* dispatch */
2270     for( int i = 0; i < h->param.i_threads; i++ )
2271     {
2272         x264_threadpool_run( h->threadpool, (void*)x264_slices_write, h->thread[i] );
2273         h->thread[i]->b_thread_active = 1;
2274     }
2275     for( int i = 0; i < h->param.i_threads; i++ )
2276     {
2277         h->thread[i]->b_thread_active = 0;
2278         if( (intptr_t)x264_threadpool_wait( h->threadpool, h->thread[i] ) )
2279             return -1;
2280     }
2281
2282     /* Go back and fix up the hpel on the borders between slices. */
2283     for( int i = 1; i < h->param.i_threads; i++ )
2284     {
2285         x264_fdec_filter_row( h->thread[i], h->thread[i]->i_threadslice_start + 1, 0 );
2286         if( h->sh.b_mbaff )
2287             x264_fdec_filter_row( h->thread[i], h->thread[i]->i_threadslice_start + 2, 0 );
2288     }
2289
2290     x264_threads_merge_ratecontrol( h );
2291
2292     for( int i = 1; i < h->param.i_threads; i++ )
2293     {
2294         x264_t *t = h->thread[i];
2295         for( int j = 0; j < t->out.i_nal; j++ )
2296         {
2297             h->out.nal[h->out.i_nal] = t->out.nal[j];
2298             h->out.i_nal++;
2299             x264_nal_check_buffer( h );
2300         }
2301         /* All entries in stat.frame are ints except for ssd/ssim. */
2302         for( int j = 0; j < (offsetof(x264_t,stat.frame.i_ssd) - offsetof(x264_t,stat.frame.i_mv_bits)) / sizeof(int); j++ )
2303             ((int*)&h->stat.frame)[j] += ((int*)&t->stat.frame)[j];
2304         for( int j = 0; j < 3; j++ )
2305             h->stat.frame.i_ssd[j] += t->stat.frame.i_ssd[j];
2306         h->stat.frame.f_ssim += t->stat.frame.f_ssim;
2307     }
2308
2309     return 0;
2310 }
2311
2312 void x264_encoder_intra_refresh( x264_t *h )
2313 {
2314     h = h->thread[h->i_thread_phase];
2315     h->b_queued_intra_refresh = 1;
2316 }
2317
2318 int x264_encoder_invalidate_reference( x264_t *h, int64_t pts )
2319 {
2320     if( h->param.i_bframe )
2321     {
2322         x264_log( h, X264_LOG_ERROR, "x264_encoder_invalidate_reference is not supported with B-frames enabled\n" );
2323         return -1;
2324     }
2325     if( h->param.b_intra_refresh )
2326     {
2327         x264_log( h, X264_LOG_ERROR, "x264_encoder_invalidate_reference is not supported with intra refresh enabled\n" );
2328         return -1;
2329     }
2330     h = h->thread[h->i_thread_phase];
2331     if( pts >= h->i_last_idr_pts )
2332     {
2333         for( int i = 0; h->frames.reference[i]; i++ )
2334             if( pts <= h->frames.reference[i]->i_pts )
2335                 h->frames.reference[i]->b_corrupt = 1;
2336         if( pts <= h->fdec->i_pts )
2337             h->fdec->b_corrupt = 1;
2338     }
2339     return 0;
2340 }
2341
2342 /****************************************************************************
2343  * x264_encoder_encode:
2344  *  XXX: i_poc   : is the poc of the current given picture
2345  *       i_frame : is the number of the frame being coded
2346  *  ex:  type frame poc
2347  *       I      0   2*0
2348  *       P      1   2*3
2349  *       B      2   2*1
2350  *       B      3   2*2
2351  *       P      4   2*6
2352  *       B      5   2*4
2353  *       B      6   2*5
2354  ****************************************************************************/
2355 int     x264_encoder_encode( x264_t *h,
2356                              x264_nal_t **pp_nal, int *pi_nal,
2357                              x264_picture_t *pic_in,
2358                              x264_picture_t *pic_out )
2359 {
2360     x264_t *thread_current, *thread_prev, *thread_oldest;
2361     int i_nal_type, i_nal_ref_idc, i_global_qp;
2362     int overhead = NALU_OVERHEAD;
2363
2364     if( h->i_thread_frames > 1 )
2365     {
2366         thread_prev    = h->thread[ h->i_thread_phase ];
2367         h->i_thread_phase = (h->i_thread_phase + 1) % h->i_thread_frames;
2368         thread_current = h->thread[ h->i_thread_phase ];
2369         thread_oldest  = h->thread[ (h->i_thread_phase + 1) % h->i_thread_frames ];
2370         x264_thread_sync_context( thread_current, thread_prev );
2371         x264_thread_sync_ratecontrol( thread_current, thread_prev, thread_oldest );
2372         h = thread_current;
2373     }
2374     else
2375     {
2376         thread_current =
2377         thread_oldest  = h;
2378     }
2379 #if HAVE_MMX
2380     if( h->param.cpu&X264_CPU_SSE_MISALIGN )
2381         x264_cpu_mask_misalign_sse();
2382 #endif
2383
2384     // ok to call this before encoding any frames, since the initial values of fdec have b_kept_as_ref=0
2385     if( x264_reference_update( h ) )
2386         return -1;
2387     h->fdec->i_lines_completed = -1;
2388
2389     /* no data out */
2390     *pi_nal = 0;
2391     *pp_nal = NULL;
2392
2393     /* ------------------- Setup new frame from picture -------------------- */
2394     if( pic_in != NULL )
2395     {
2396         /* 1: Copy the picture to a frame and move it to a buffer */
2397         x264_frame_t *fenc = x264_frame_pop_unused( h, 0 );
2398         if( !fenc )
2399             return -1;
2400
2401         if( x264_frame_copy_picture( h, fenc, pic_in ) < 0 )
2402             return -1;
2403
2404         if( h->param.i_width != 16 * h->mb.i_mb_width ||
2405             h->param.i_height != 16 * h->mb.i_mb_height )
2406             x264_frame_expand_border_mod16( h, fenc );
2407
2408         fenc->i_frame = h->frames.i_input++;
2409
2410         if( fenc->i_frame == 0 )
2411             h->frames.i_first_pts = fenc->i_pts;
2412         if( h->frames.i_bframe_delay && fenc->i_frame == h->frames.i_bframe_delay )
2413             h->frames.i_bframe_delay_time = fenc->i_pts - h->frames.i_first_pts;
2414
2415         if( h->param.b_vfr_input && fenc->i_pts <= h->frames.i_largest_pts )
2416             x264_log( h, X264_LOG_WARNING, "non-strictly-monotonic PTS\n" );
2417
2418         h->frames.i_second_largest_pts = h->frames.i_largest_pts;
2419         h->frames.i_largest_pts = fenc->i_pts;
2420
2421         if( (fenc->i_pic_struct < PIC_STRUCT_AUTO) || (fenc->i_pic_struct > PIC_STRUCT_TRIPLE) )
2422             fenc->i_pic_struct = PIC_STRUCT_AUTO;
2423
2424         if( fenc->i_pic_struct == PIC_STRUCT_AUTO )
2425         {
2426             int b_interlaced = fenc->param ? fenc->param->b_interlaced : h->param.b_interlaced;
2427             if( b_interlaced )
2428             {
2429                 int b_tff = fenc->param ? fenc->param->b_tff : h->param.b_tff;
2430                 fenc->i_pic_struct = b_tff ? PIC_STRUCT_TOP_BOTTOM : PIC_STRUCT_BOTTOM_TOP;
2431             }
2432             else
2433                 fenc->i_pic_struct = PIC_STRUCT_PROGRESSIVE;
2434         }
2435
2436         if( h->param.rc.b_mb_tree && h->param.rc.b_stat_read )
2437         {
2438             if( x264_macroblock_tree_read( h, fenc, pic_in->prop.quant_offsets ) )
2439                 return -1;
2440         }
2441         else
2442             x264_stack_align( x264_adaptive_quant_frame, h, fenc, pic_in->prop.quant_offsets );
2443
2444         if( pic_in->prop.quant_offsets_free )
2445             pic_in->prop.quant_offsets_free( pic_in->prop.quant_offsets );
2446
2447         if( h->frames.b_have_lowres )
2448             x264_frame_init_lowres( h, fenc );
2449
2450         /* 2: Place the frame into the queue for its slice type decision */
2451         x264_lookahead_put_frame( h, fenc );
2452
2453         if( h->frames.i_input <= h->frames.i_delay + 1 - h->i_thread_frames )
2454         {
2455             /* Nothing yet to encode, waiting for filling of buffers */
2456             pic_out->i_type = X264_TYPE_AUTO;
2457             return 0;
2458         }
2459     }
2460     else
2461     {
2462         /* signal kills for lookahead thread */
2463         x264_pthread_mutex_lock( &h->lookahead->ifbuf.mutex );
2464         h->lookahead->b_exit_thread = 1;
2465         x264_pthread_cond_broadcast( &h->lookahead->ifbuf.cv_fill );
2466         x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
2467     }
2468
2469     h->i_frame++;
2470     /* 3: The picture is analyzed in the lookahead */
2471     if( !h->frames.current[0] )
2472         x264_lookahead_get_frames( h );
2473
2474     if( !h->frames.current[0] && x264_lookahead_is_empty( h ) )
2475         return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
2476
2477     /* ------------------- Get frame to be encoded ------------------------- */
2478     /* 4: get picture to encode */
2479     h->fenc = x264_frame_shift( h->frames.current );
2480     if( h->i_frame == h->i_thread_frames - 1 )
2481         h->i_reordered_pts_delay = h->fenc->i_reordered_pts;
2482     if( h->fenc->param )
2483     {
2484         x264_encoder_reconfig( h, h->fenc->param );
2485         if( h->fenc->param->param_free )
2486             h->fenc->param->param_free( h->fenc->param );
2487     }
2488
2489     if( !IS_X264_TYPE_I( h->fenc->i_type ) )
2490     {
2491         int valid_refs_left = 0;
2492         for( int i = 0; h->frames.reference[i]; i++ )
2493             if( !h->frames.reference[i]->b_corrupt )
2494                 valid_refs_left++;
2495         /* No valid reference frames left: force an IDR. */
2496         if( !valid_refs_left )
2497         {
2498             h->fenc->b_keyframe = 1;
2499             h->fenc->i_type = X264_TYPE_IDR;
2500         }
2501     }
2502
2503     if( h->fenc->b_keyframe )
2504     {
2505         h->frames.i_last_keyframe = h->fenc->i_frame;
2506         if( h->fenc->i_type == X264_TYPE_IDR )
2507         {
2508             h->i_frame_num = 0;
2509             h->frames.i_last_idr = h->fenc->i_frame;
2510         }
2511     }
2512     h->sh.i_mmco_command_count =
2513     h->sh.i_mmco_remove_from_end = 0;
2514     h->b_ref_reorder[0] =
2515     h->b_ref_reorder[1] = 0;
2516     h->fdec->i_poc =
2517     h->fenc->i_poc = 2 * ( h->fenc->i_frame - X264_MAX( h->frames.i_last_idr, 0 ) );
2518
2519     /* ------------------- Setup frame context ----------------------------- */
2520     /* 5: Init data dependent of frame type */
2521     if( h->fenc->i_type == X264_TYPE_IDR )
2522     {
2523         /* reset ref pictures */
2524         i_nal_type    = NAL_SLICE_IDR;
2525         i_nal_ref_idc = NAL_PRIORITY_HIGHEST;
2526         h->sh.i_type = SLICE_TYPE_I;
2527         x264_reference_reset( h );
2528         h->frames.i_poc_last_open_gop = -1;
2529     }
2530     else if( h->fenc->i_type == X264_TYPE_I )
2531     {
2532         i_nal_type    = NAL_SLICE;
2533         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
2534         h->sh.i_type = SLICE_TYPE_I;
2535         x264_reference_hierarchy_reset( h );
2536         if( h->param.i_open_gop )
2537             h->frames.i_poc_last_open_gop = h->fenc->b_keyframe ? h->fenc->i_poc : -1;
2538     }
2539     else if( h->fenc->i_type == X264_TYPE_P )
2540     {
2541         i_nal_type    = NAL_SLICE;
2542         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
2543         h->sh.i_type = SLICE_TYPE_P;
2544         x264_reference_hierarchy_reset( h );
2545         h->frames.i_poc_last_open_gop = -1;
2546     }
2547     else if( h->fenc->i_type == X264_TYPE_BREF )
2548     {
2549         i_nal_type    = NAL_SLICE;
2550         i_nal_ref_idc = h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT ? NAL_PRIORITY_LOW : NAL_PRIORITY_HIGH;
2551         h->sh.i_type = SLICE_TYPE_B;
2552         x264_reference_hierarchy_reset( h );
2553     }
2554     else    /* B frame */
2555     {
2556         i_nal_type    = NAL_SLICE;
2557         i_nal_ref_idc = NAL_PRIORITY_DISPOSABLE;
2558         h->sh.i_type = SLICE_TYPE_B;
2559     }
2560
2561     h->fdec->i_type = h->fenc->i_type;
2562     h->fdec->i_frame = h->fenc->i_frame;
2563     h->fenc->b_kept_as_ref =
2564     h->fdec->b_kept_as_ref = i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE && h->param.i_keyint_max > 1;
2565
2566     h->fdec->i_pts = h->fenc->i_pts;
2567     if( h->frames.i_bframe_delay )
2568     {
2569         int64_t *prev_reordered_pts = thread_current->frames.i_prev_reordered_pts;
2570         h->fdec->i_dts = h->i_frame > h->frames.i_bframe_delay
2571                        ? prev_reordered_pts[ (h->i_frame - h->frames.i_bframe_delay) % h->frames.i_bframe_delay ]
2572                        : h->fenc->i_reordered_pts - h->frames.i_bframe_delay_time;
2573         prev_reordered_pts[ h->i_frame % h->frames.i_bframe_delay ] = h->fenc->i_reordered_pts;
2574     }
2575     else
2576         h->fdec->i_dts = h->fenc->i_reordered_pts;
2577     if( h->fenc->i_type == X264_TYPE_IDR )
2578         h->i_last_idr_pts = h->fdec->i_pts;
2579
2580     /* ------------------- Init                ----------------------------- */
2581     /* build ref list 0/1 */
2582     x264_reference_build_list( h, h->fdec->i_poc );
2583
2584     /* ---------------------- Write the bitstream -------------------------- */
2585     /* Init bitstream context */
2586     if( h->param.b_sliced_threads )
2587     {
2588         for( int i = 0; i < h->param.i_threads; i++ )
2589         {
2590             bs_init( &h->thread[i]->out.bs, h->thread[i]->out.p_bitstream, h->thread[i]->out.i_bitstream );
2591             h->thread[i]->out.i_nal = 0;
2592         }
2593     }
2594     else
2595     {
2596         bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
2597         h->out.i_nal = 0;
2598     }
2599
2600     if( h->param.b_aud )
2601     {
2602         int pic_type;
2603
2604         if( h->sh.i_type == SLICE_TYPE_I )
2605             pic_type = 0;
2606         else if( h->sh.i_type == SLICE_TYPE_P )
2607             pic_type = 1;
2608         else if( h->sh.i_type == SLICE_TYPE_B )
2609             pic_type = 2;
2610         else
2611             pic_type = 7;
2612
2613         x264_nal_start( h, NAL_AUD, NAL_PRIORITY_DISPOSABLE );
2614         bs_write( &h->out.bs, 3, pic_type );
2615         bs_rbsp_trailing( &h->out.bs );
2616         if( x264_nal_end( h ) )
2617             return -1;
2618         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2619     }
2620
2621     h->i_nal_type = i_nal_type;
2622     h->i_nal_ref_idc = i_nal_ref_idc;
2623
2624     if( h->param.b_intra_refresh )
2625     {
2626         if( IS_X264_TYPE_I( h->fenc->i_type ) )
2627         {
2628             h->fdec->i_frames_since_pir = 0;
2629             h->b_queued_intra_refresh = 0;
2630             /* PIR is currently only supported with ref == 1, so any intra frame effectively refreshes
2631              * the whole frame and counts as an intra refresh. */
2632             h->fdec->f_pir_position = h->mb.i_mb_width;
2633         }
2634         else if( h->fenc->i_type == X264_TYPE_P )
2635         {
2636             int pocdiff = (h->fdec->i_poc - h->fref[0][0]->i_poc)/2;
2637             float increment = X264_MAX( ((float)h->mb.i_mb_width-1) / h->param.i_keyint_max, 1 );
2638             h->fdec->f_pir_position = h->fref[0][0]->f_pir_position;
2639             h->fdec->i_frames_since_pir = h->fref[0][0]->i_frames_since_pir + pocdiff;
2640             if( h->fdec->i_frames_since_pir >= h->param.i_keyint_max ||
2641                 (h->b_queued_intra_refresh && h->fdec->f_pir_position + 0.5 >= h->mb.i_mb_width) )
2642             {
2643                 h->fdec->f_pir_position = 0;
2644                 h->fdec->i_frames_since_pir = 0;
2645                 h->b_queued_intra_refresh = 0;
2646                 h->fenc->b_keyframe = 1;
2647             }
2648             h->fdec->i_pir_start_col = h->fdec->f_pir_position+0.5;
2649             h->fdec->f_pir_position += increment * pocdiff;
2650             h->fdec->i_pir_end_col = h->fdec->f_pir_position+0.5;
2651             /* If our intra refresh has reached the right side of the frame, we're done. */
2652             if( h->fdec->i_pir_end_col >= h->mb.i_mb_width - 1 )
2653                 h->fdec->f_pir_position = h->mb.i_mb_width;
2654         }
2655     }
2656
2657     if( h->fenc->b_keyframe )
2658     {
2659         /* Write SPS and PPS */
2660         if( h->param.b_repeat_headers )
2661         {
2662             /* generate sequence parameters */
2663             x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
2664             x264_sps_write( &h->out.bs, h->sps );
2665             if( x264_nal_end( h ) )
2666                 return -1;
2667             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
2668
2669             /* generate picture parameters */
2670             x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
2671             x264_pps_write( &h->out.bs, h->pps );
2672             if( x264_nal_end( h ) )
2673                 return -1;
2674             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
2675         }
2676
2677         /* buffering period sei is written in x264_encoder_frame_end */
2678     }
2679
2680     /* write extra sei */
2681     for( int i = 0; i < h->fenc->extra_sei.num_payloads; i++ )
2682     {
2683         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2684         x264_sei_write( &h->out.bs, h->fenc->extra_sei.payloads[i].payload, h->fenc->extra_sei.payloads[i].payload_size,
2685                         h->fenc->extra_sei.payloads[i].payload_type );
2686         if( x264_nal_end( h ) )
2687             return -1;
2688         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2689         if( h->fenc->extra_sei.sei_free && h->fenc->extra_sei.payloads[i].payload )
2690             h->fenc->extra_sei.sei_free( h->fenc->extra_sei.payloads[i].payload );
2691     }
2692
2693     if( h->fenc->extra_sei.sei_free && h->fenc->extra_sei.payloads )
2694         h->fenc->extra_sei.sei_free( h->fenc->extra_sei.payloads );
2695
2696     if( h->fenc->b_keyframe )
2697     {
2698         if( h->param.b_repeat_headers && h->fenc->i_frame == 0 )
2699         {
2700             /* identify ourself */
2701             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2702             if( x264_sei_version_write( h, &h->out.bs ) )
2703                 return -1;
2704             if( x264_nal_end( h ) )
2705                 return -1;
2706             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2707         }
2708
2709         if( h->fenc->i_type != X264_TYPE_IDR )
2710         {
2711             int time_to_recovery = h->param.i_open_gop ? 0 : X264_MIN( h->mb.i_mb_width - 1, h->param.i_keyint_max ) + h->param.i_bframe - 1;
2712             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2713             x264_sei_recovery_point_write( h, &h->out.bs, time_to_recovery );
2714             if( x264_nal_end( h ) )
2715                 return -1;
2716             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2717         }
2718
2719         if ( h->param.i_frame_packing >= 0 )
2720         {
2721             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2722             x264_sei_frame_packing_write( h, &h->out.bs );
2723             if( x264_nal_end( h ) )
2724                 return -1;
2725             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2726         }
2727     }
2728
2729     /* generate sei pic timing */
2730     if( h->sps->vui.b_pic_struct_present || h->sps->vui.b_nal_hrd_parameters_present )
2731     {
2732         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2733         x264_sei_pic_timing_write( h, &h->out.bs );
2734         if( x264_nal_end( h ) )
2735             return -1;
2736         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2737     }
2738
2739     if( h->fenc->b_keyframe && h->param.b_intra_refresh )
2740         h->i_cpb_delay_pir_offset = h->fenc->i_cpb_delay;
2741
2742     /* Init the rate control */
2743     /* FIXME: Include slice header bit cost. */
2744     x264_ratecontrol_start( h, h->fenc->i_qpplus1, overhead*8 );
2745     i_global_qp = x264_ratecontrol_qp( h );
2746
2747     pic_out->i_qpplus1 =
2748     h->fdec->i_qpplus1 = i_global_qp + 1;
2749
2750     if( h->param.rc.b_stat_read && h->sh.i_type != SLICE_TYPE_I )
2751     {
2752         x264_reference_build_list_optimal( h );
2753         x264_reference_check_reorder( h );
2754     }
2755
2756     if( h->i_ref[0] )
2757         h->fdec->i_poc_l0ref0 = h->fref[0][0]->i_poc;
2758
2759     /* ------------------------ Create slice header  ----------------------- */
2760     x264_slice_init( h, i_nal_type, i_global_qp );
2761
2762     /*------------------------- Weights -------------------------------------*/
2763     if( h->sh.i_type == SLICE_TYPE_B )
2764         x264_macroblock_bipred_init( h );
2765
2766     x264_weighted_pred_init( h );
2767
2768     if( i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE )
2769         h->i_frame_num++;
2770
2771     /* Write frame */
2772     h->i_threadslice_start = 0;
2773     h->i_threadslice_end = h->mb.i_mb_height;
2774     if( h->i_thread_frames > 1 )
2775     {
2776         x264_threadpool_run( h->threadpool, (void*)x264_slices_write, h );
2777         h->b_thread_active = 1;
2778     }
2779     else if( h->param.b_sliced_threads )
2780     {
2781         if( x264_threaded_slices_write( h ) )
2782             return -1;
2783     }
2784     else
2785         if( (intptr_t)x264_slices_write( h ) )
2786             return -1;
2787
2788     return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
2789 }
2790
2791 static int x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
2792                                    x264_nal_t **pp_nal, int *pi_nal,
2793                                    x264_picture_t *pic_out )
2794 {
2795     char psz_message[80];
2796
2797     if( h->b_thread_active )
2798     {
2799         h->b_thread_active = 0;
2800         if( (intptr_t)x264_threadpool_wait( h->threadpool, h ) )
2801             return -1;
2802     }
2803     if( !h->out.i_nal )
2804     {
2805         pic_out->i_type = X264_TYPE_AUTO;
2806         return 0;
2807     }
2808
2809     x264_emms();
2810     /* generate sei buffering period and insert it into place */
2811     if( h->fenc->b_keyframe && h->sps->vui.b_nal_hrd_parameters_present )
2812     {
2813         x264_hrd_fullness( h );
2814         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2815         x264_sei_buffering_period_write( h, &h->out.bs );
2816         if( x264_nal_end( h ) )
2817            return -1;
2818         /* buffering period sei must follow AUD, SPS and PPS and precede all other SEIs */
2819         int idx = 0;
2820         while( h->out.nal[idx].i_type == NAL_AUD ||
2821                h->out.nal[idx].i_type == NAL_SPS ||
2822                h->out.nal[idx].i_type == NAL_PPS )
2823             idx++;
2824         x264_nal_t nal_tmp = h->out.nal[h->out.i_nal-1];
2825         memmove( &h->out.nal[idx+1], &h->out.nal[idx], (h->out.i_nal-idx-1)*sizeof(x264_nal_t) );
2826         h->out.nal[idx] = nal_tmp;
2827     }
2828
2829     int frame_size = x264_encoder_encapsulate_nals( h, 0 );
2830
2831     /* Set output picture properties */
2832     pic_out->i_type = h->fenc->i_type;
2833
2834     pic_out->b_keyframe = h->fenc->b_keyframe;
2835     pic_out->i_pic_struct = h->fenc->i_pic_struct;
2836
2837     pic_out->i_pts = h->fdec->i_pts;
2838     pic_out->i_dts = h->fdec->i_dts;
2839
2840     if( pic_out->i_pts < pic_out->i_dts )
2841         x264_log( h, X264_LOG_WARNING, "invalid DTS: PTS is less than DTS\n" );
2842
2843     pic_out->img.i_csp = X264_CSP_NV12;
2844 #if HIGH_BIT_DEPTH
2845     pic_out->img.i_csp |= X264_CSP_HIGH_DEPTH;
2846 #endif
2847     pic_out->img.i_plane = h->fdec->i_plane;
2848     for( int i = 0; i < 2; i++ )
2849     {
2850         pic_out->img.i_stride[i] = h->fdec->i_stride[i] * sizeof(pixel);
2851         pic_out->img.plane[i] = (uint8_t*)h->fdec->plane[i];
2852     }
2853
2854     x264_frame_push_unused( thread_current, h->fenc );
2855
2856     /* ---------------------- Update encoder state ------------------------- */
2857
2858     /* update rc */
2859     int filler = 0;
2860     if( x264_ratecontrol_end( h, frame_size * 8, &filler ) < 0 )
2861         return -1;
2862
2863     pic_out->hrd_timing = h->fenc->hrd_timing;
2864
2865     while( filler > 0 )
2866     {
2867         int f, overhead;
2868         overhead = (FILLER_OVERHEAD - h->param.b_annexb);
2869         if( h->param.i_slice_max_size && filler > h->param.i_slice_max_size )
2870         {
2871             int next_size = filler - h->param.i_slice_max_size;
2872             int overflow = X264_MAX( overhead - next_size, 0 );
2873             f = h->param.i_slice_max_size - overhead - overflow;
2874         }
2875         else
2876             f = X264_MAX( 0, filler - overhead );
2877
2878         x264_nal_start( h, NAL_FILLER, NAL_PRIORITY_DISPOSABLE );
2879         x264_filler_write( h, &h->out.bs, f );
2880         if( x264_nal_end( h ) )
2881             return -1;
2882         int total_size = x264_encoder_encapsulate_nals( h, h->out.i_nal-1 );
2883         frame_size += total_size;
2884         filler -= total_size;
2885     }
2886
2887     /* End bitstream, set output  */
2888     *pi_nal = h->out.i_nal;
2889     *pp_nal = h->out.nal;
2890
2891     h->out.i_nal = 0;
2892
2893     x264_noise_reduction_update( h );
2894
2895     /* ---------------------- Compute/Print statistics --------------------- */
2896     x264_thread_sync_stat( h, h->thread[0] );
2897
2898     /* Slice stat */
2899     h->stat.i_frame_count[h->sh.i_type]++;
2900     h->stat.i_frame_size[h->sh.i_type] += frame_size;
2901     h->stat.f_frame_qp[h->sh.i_type] += h->fdec->f_qp_avg_aq;
2902
2903     for( int i = 0; i < X264_MBTYPE_MAX; i++ )
2904         h->stat.i_mb_count[h->sh.i_type][i] += h->stat.frame.i_mb_count[i];
2905     for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
2906         h->stat.i_mb_partition[h->sh.i_type][i] += h->stat.frame.i_mb_partition[i];
2907     for( int i = 0; i < 2; i++ )
2908         h->stat.i_mb_count_8x8dct[i] += h->stat.frame.i_mb_count_8x8dct[i];
2909     for( int i = 0; i < 6; i++ )
2910         h->stat.i_mb_cbp[i] += h->stat.frame.i_mb_cbp[i];
2911     for( int i = 0; i < 4; i++ )
2912         for( int j = 0; j < 13; j++ )
2913             h->stat.i_mb_pred_mode[i][j] += h->stat.frame.i_mb_pred_mode[i][j];
2914     if( h->sh.i_type != SLICE_TYPE_I )
2915         for( int i_list = 0; i_list < 2; i_list++ )
2916             for( int i = 0; i < X264_REF_MAX*2; i++ )
2917                 h->stat.i_mb_count_ref[h->sh.i_type][i_list][i] += h->stat.frame.i_mb_count_ref[i_list][i];
2918     if( h->sh.i_type == SLICE_TYPE_P && h->param.analyse.i_weighted_pred >= X264_WEIGHTP_SIMPLE )
2919     {
2920         h->stat.i_wpred[0] += !!h->sh.weight[0][0].weightfn;
2921         h->stat.i_wpred[1] += !!h->sh.weight[0][1].weightfn || !!h->sh.weight[0][2].weightfn;
2922     }
2923     if( h->sh.i_type == SLICE_TYPE_B )
2924     {
2925         h->stat.i_direct_frames[ h->sh.b_direct_spatial_mv_pred ] ++;
2926         if( h->mb.b_direct_auto_write )
2927         {
2928             //FIXME somewhat arbitrary time constants
2929             if( h->stat.i_direct_score[0] + h->stat.i_direct_score[1] > h->mb.i_mb_count )
2930                 for( int i = 0; i < 2; i++ )
2931                     h->stat.i_direct_score[i] = h->stat.i_direct_score[i] * 9/10;
2932             for( int i = 0; i < 2; i++ )
2933                 h->stat.i_direct_score[i] += h->stat.frame.i_direct_score[i];
2934         }
2935     }
2936     else
2937         h->stat.i_consecutive_bframes[h->fenc->i_bframes]++;
2938
2939     psz_message[0] = '\0';
2940     double dur = h->fenc->f_duration;
2941     h->stat.f_frame_duration[h->sh.i_type] += dur;
2942     if( h->param.analyse.b_psnr )
2943     {
2944         int64_t ssd[3] =
2945         {
2946             h->stat.frame.i_ssd[0],
2947             h->stat.frame.i_ssd[1],
2948             h->stat.frame.i_ssd[2],
2949         };
2950
2951         h->stat.f_ssd_global[h->sh.i_type]   += dur * (ssd[0] + ssd[1] + ssd[2]);
2952         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 );
2953         h->stat.f_psnr_mean_y[h->sh.i_type]  += dur * x264_psnr( ssd[0], h->param.i_width * h->param.i_height );
2954         h->stat.f_psnr_mean_u[h->sh.i_type]  += dur * x264_psnr( ssd[1], h->param.i_width * h->param.i_height / 4 );
2955         h->stat.f_psnr_mean_v[h->sh.i_type]  += dur * x264_psnr( ssd[2], h->param.i_width * h->param.i_height / 4 );
2956
2957         snprintf( psz_message, 80, " PSNR Y:%5.2f U:%5.2f V:%5.2f",
2958                   x264_psnr( ssd[0], h->param.i_width * h->param.i_height ),
2959                   x264_psnr( ssd[1], h->param.i_width * h->param.i_height / 4),
2960                   x264_psnr( ssd[2], h->param.i_width * h->param.i_height / 4) );
2961     }
2962
2963     if( h->param.analyse.b_ssim )
2964     {
2965         double ssim_y = h->stat.frame.f_ssim
2966                       / (((h->param.i_width-6)>>2) * ((h->param.i_height-6)>>2));
2967         h->stat.f_ssim_mean_y[h->sh.i_type] += ssim_y * dur;
2968         snprintf( psz_message + strlen(psz_message), 80 - strlen(psz_message),
2969                   " SSIM Y:%.5f", ssim_y );
2970     }
2971     psz_message[79] = '\0';
2972
2973     x264_log( h, X264_LOG_DEBUG,
2974                   "frame=%4d QP=%.2f NAL=%d Slice:%c Poc:%-3d I:%-4d P:%-4d SKIP:%-4d size=%d bytes%s\n",
2975               h->i_frame,
2976               h->fdec->f_qp_avg_aq,
2977               h->i_nal_ref_idc,
2978               h->sh.i_type == SLICE_TYPE_I ? 'I' : (h->sh.i_type == SLICE_TYPE_P ? 'P' : 'B' ),
2979               h->fdec->i_poc,
2980               h->stat.frame.i_mb_count_i,
2981               h->stat.frame.i_mb_count_p,
2982               h->stat.frame.i_mb_count_skip,
2983               frame_size,
2984               psz_message );
2985
2986     // keep stats all in one place
2987     x264_thread_sync_stat( h->thread[0], h );
2988     // for the use of the next frame
2989     x264_thread_sync_stat( thread_current, h );
2990
2991 #ifdef DEBUG_MB_TYPE
2992 {
2993     static const char mb_chars[] = { 'i', 'i', 'I', 'C', 'P', '8', 'S',
2994         'D', '<', 'X', 'B', 'X', '>', 'B', 'B', 'B', 'B', '8', 'S' };
2995     for( int mb_xy = 0; mb_xy < h->mb.i_mb_width * h->mb.i_mb_height; mb_xy++ )
2996     {
2997         if( h->mb.type[mb_xy] < X264_MBTYPE_MAX && h->mb.type[mb_xy] >= 0 )
2998             fprintf( stderr, "%c ", mb_chars[ h->mb.type[mb_xy] ] );
2999         else
3000             fprintf( stderr, "? " );
3001
3002         if( (mb_xy+1) % h->mb.i_mb_width == 0 )
3003             fprintf( stderr, "\n" );
3004     }
3005 }
3006 #endif
3007
3008     /* Remove duplicates, must be done near the end as breaks h->fref0 array
3009      * by freeing some of its pointers. */
3010     for( int i = 0; i < h->i_ref[0]; i++ )
3011         if( h->fref[0][i] && h->fref[0][i]->b_duplicate )
3012         {
3013             x264_frame_push_blank_unused( h, h->fref[0][i] );
3014             h->fref[0][i] = 0;
3015         }
3016
3017     if( h->param.psz_dump_yuv )
3018         x264_frame_dump( h );
3019     x264_emms();
3020
3021     return frame_size;
3022 }
3023
3024 static void x264_print_intra( int64_t *i_mb_count, double i_count, int b_print_pcm, char *intra )
3025 {
3026     intra += sprintf( intra, "I16..4%s: %4.1f%% %4.1f%% %4.1f%%",
3027         b_print_pcm ? "..PCM" : "",
3028         i_mb_count[I_16x16]/ i_count,
3029         i_mb_count[I_8x8]  / i_count,
3030         i_mb_count[I_4x4]  / i_count );
3031     if( b_print_pcm )
3032         sprintf( intra, " %4.1f%%", i_mb_count[I_PCM]  / i_count );
3033 }
3034
3035 /****************************************************************************
3036  * x264_encoder_close:
3037  ****************************************************************************/
3038 void    x264_encoder_close  ( x264_t *h )
3039 {
3040     int64_t i_yuv_size = 3 * h->param.i_width * h->param.i_height / 2;
3041     int64_t i_mb_count_size[2][7] = {{0}};
3042     char buf[200];
3043     int b_print_pcm = h->stat.i_mb_count[SLICE_TYPE_I][I_PCM]
3044                    || h->stat.i_mb_count[SLICE_TYPE_P][I_PCM]
3045                    || h->stat.i_mb_count[SLICE_TYPE_B][I_PCM];
3046
3047     x264_lookahead_delete( h );
3048
3049     if( h->param.i_threads > 1 )
3050         x264_threadpool_delete( h->threadpool );
3051     if( h->i_thread_frames > 1 )
3052     {
3053         for( int i = 0; i < h->i_thread_frames; i++ )
3054             if( h->thread[i]->b_thread_active )
3055             {
3056                 assert( h->thread[i]->fenc->i_reference_count == 1 );
3057                 x264_frame_delete( h->thread[i]->fenc );
3058             }
3059
3060         x264_t *thread_prev = h->thread[h->i_thread_phase];
3061         x264_thread_sync_ratecontrol( h, thread_prev, h );
3062         x264_thread_sync_ratecontrol( thread_prev, thread_prev, h );
3063         h->i_frame = thread_prev->i_frame + 1 - h->i_thread_frames;
3064     }
3065     h->i_frame++;
3066
3067     /* Slices used and PSNR */
3068     for( int i = 0; i < 3; i++ )
3069     {
3070         static const uint8_t slice_order[] = { SLICE_TYPE_I, SLICE_TYPE_P, SLICE_TYPE_B };
3071         int i_slice = slice_order[i];
3072
3073         if( h->stat.i_frame_count[i_slice] > 0 )
3074         {
3075             int i_count = h->stat.i_frame_count[i_slice];
3076             double dur =  h->stat.f_frame_duration[i_slice];
3077             if( h->param.analyse.b_psnr )
3078             {
3079                 x264_log( h, X264_LOG_INFO,
3080                           "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",
3081                           slice_type_to_char[i_slice],
3082                           i_count,
3083                           h->stat.f_frame_qp[i_slice] / i_count,
3084                           (double)h->stat.i_frame_size[i_slice] / i_count,
3085                           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,
3086                           h->stat.f_psnr_average[i_slice] / dur,
3087                           x264_psnr( h->stat.f_ssd_global[i_slice], dur * i_yuv_size ) );
3088             }
3089             else
3090             {
3091                 x264_log( h, X264_LOG_INFO,
3092                           "frame %c:%-5d Avg QP:%5.2f  size:%6.0f\n",
3093                           slice_type_to_char[i_slice],
3094                           i_count,
3095                           h->stat.f_frame_qp[i_slice] / i_count,
3096                           (double)h->stat.i_frame_size[i_slice] / i_count );
3097             }
3098         }
3099     }
3100     if( h->param.i_bframe && h->stat.i_frame_count[SLICE_TYPE_B] )
3101     {
3102         char *p = buf;
3103         int den = 0;
3104         // weight by number of frames (including the I/P-frames) that are in a sequence of N B-frames
3105         for( int i = 0; i <= h->param.i_bframe; i++ )
3106             den += (i+1) * h->stat.i_consecutive_bframes[i];
3107         for( int i = 0; i <= h->param.i_bframe; i++ )
3108             p += sprintf( p, " %4.1f%%", 100. * (i+1) * h->stat.i_consecutive_bframes[i] / den );
3109         x264_log( h, X264_LOG_INFO, "consecutive B-frames:%s\n", buf );
3110     }
3111
3112     for( int i_type = 0; i_type < 2; i_type++ )
3113         for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
3114         {
3115             if( i == D_DIRECT_8x8 ) continue; /* direct is counted as its own type */
3116             i_mb_count_size[i_type][x264_mb_partition_pixel_table[i]] += h->stat.i_mb_partition[i_type][i];
3117         }
3118
3119     /* MB types used */
3120     if( h->stat.i_frame_count[SLICE_TYPE_I] > 0 )
3121     {
3122         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_I];
3123         double i_count = h->stat.i_frame_count[SLICE_TYPE_I] * h->mb.i_mb_count / 100.0;
3124         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
3125         x264_log( h, X264_LOG_INFO, "mb I  %s\n", buf );
3126     }
3127     if( h->stat.i_frame_count[SLICE_TYPE_P] > 0 )
3128     {
3129         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_P];
3130         double i_count = h->stat.i_frame_count[SLICE_TYPE_P] * h->mb.i_mb_count / 100.0;
3131         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_P];
3132         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
3133         x264_log( h, X264_LOG_INFO,
3134                   "mb P  %s  P16..4: %4.1f%% %4.1f%% %4.1f%% %4.1f%% %4.1f%%    skip:%4.1f%%\n",
3135                   buf,
3136                   i_mb_size[PIXEL_16x16] / (i_count*4),
3137                   (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
3138                   i_mb_size[PIXEL_8x8] / (i_count*4),
3139                   (i_mb_size[PIXEL_8x4] + i_mb_size[PIXEL_4x8]) / (i_count*4),
3140                   i_mb_size[PIXEL_4x4] / (i_count*4),
3141                   i_mb_count[P_SKIP] / i_count );
3142     }
3143     if( h->stat.i_frame_count[SLICE_TYPE_B] > 0 )
3144     {
3145         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_B];
3146         double i_count = h->stat.i_frame_count[SLICE_TYPE_B] * h->mb.i_mb_count / 100.0;
3147         double i_mb_list_count;
3148         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_B];
3149         int64_t list_count[3] = {0}; /* 0 == L0, 1 == L1, 2 == BI */
3150         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
3151         for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
3152             for( int j = 0; j < 2; j++ )
3153             {
3154                 int l0 = x264_mb_type_list_table[i][0][j];
3155                 int l1 = x264_mb_type_list_table[i][1][j];
3156                 if( l0 || l1 )
3157                     list_count[l1+l0*l1] += h->stat.i_mb_count[SLICE_TYPE_B][i] * 2;
3158             }
3159         list_count[0] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L0_8x8];
3160         list_count[1] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L1_8x8];
3161         list_count[2] += h->stat.i_mb_partition[SLICE_TYPE_B][D_BI_8x8];
3162         i_mb_count[B_DIRECT] += (h->stat.i_mb_partition[SLICE_TYPE_B][D_DIRECT_8x8]+2)/4;
3163         i_mb_list_count = (list_count[0] + list_count[1] + list_count[2]) / 100.0;
3164         x264_log( h, X264_LOG_INFO,
3165                   "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",
3166                   buf,
3167                   i_mb_size[PIXEL_16x16] / (i_count*4),
3168                   (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
3169                   i_mb_size[PIXEL_8x8] / (i_count*4),
3170                   i_mb_count[B_DIRECT] / i_count,
3171                   i_mb_count[B_SKIP]   / i_count,
3172                   list_count[0] / i_mb_list_count,
3173                   list_count[1] / i_mb_list_count,
3174                   list_count[2] / i_mb_list_count );
3175     }
3176
3177     x264_ratecontrol_summary( h );
3178
3179     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 )
3180     {
3181 #define SUM3(p) (p[SLICE_TYPE_I] + p[SLICE_TYPE_P] + p[SLICE_TYPE_B])
3182 #define SUM3b(p,o) (p[SLICE_TYPE_I][o] + p[SLICE_TYPE_P][o] + p[SLICE_TYPE_B][o])
3183         int64_t i_i8x8 = SUM3b( h->stat.i_mb_count, I_8x8 );
3184         int64_t i_intra = i_i8x8 + SUM3b( h->stat.i_mb_count, I_4x4 )
3185                                  + SUM3b( h->stat.i_mb_count, I_16x16 );
3186         int64_t i_all_intra = i_intra + SUM3b( h->stat.i_mb_count, I_PCM);
3187         const int i_count = h->stat.i_frame_count[SLICE_TYPE_I] +
3188                             h->stat.i_frame_count[SLICE_TYPE_P] +
3189                             h->stat.i_frame_count[SLICE_TYPE_B];
3190         const double duration = h->stat.f_frame_duration[SLICE_TYPE_I] +
3191                                 h->stat.f_frame_duration[SLICE_TYPE_P] +
3192                                 h->stat.f_frame_duration[SLICE_TYPE_B];
3193         int64_t i_mb_count = (int64_t)i_count * h->mb.i_mb_count;
3194         float f_bitrate = SUM3(h->stat.i_frame_size) / duration / 125;
3195
3196         if( h->pps->b_transform_8x8_mode )
3197         {
3198             buf[0] = 0;
3199             if( h->stat.i_mb_count_8x8dct[0] )
3200                 sprintf( buf, " inter:%.1f%%", 100. * h->stat.i_mb_count_8x8dct[1] / h->stat.i_mb_count_8x8dct[0] );
3201             x264_log( h, X264_LOG_INFO, "8x8 transform intra:%.1f%%%s\n", 100. * i_i8x8 / i_intra, buf );
3202         }
3203
3204         if( (h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO ||
3205             (h->stat.i_direct_frames[0] && h->stat.i_direct_frames[1]))
3206             && h->stat.i_frame_count[SLICE_TYPE_B] )
3207         {
3208             x264_log( h, X264_LOG_INFO, "direct mvs  spatial:%.1f%% temporal:%.1f%%\n",
3209                       h->stat.i_direct_frames[1] * 100. / h->stat.i_frame_count[SLICE_TYPE_B],
3210                       h->stat.i_direct_frames[0] * 100. / h->stat.i_frame_count[SLICE_TYPE_B] );
3211         }
3212
3213         buf[0] = 0;
3214         if( i_mb_count != i_all_intra )
3215             sprintf( buf, " inter: %.1f%% %.1f%% %.1f%%",
3216                      h->stat.i_mb_cbp[1] * 100.0 / ((i_mb_count - i_all_intra)*4),
3217                      h->stat.i_mb_cbp[3] * 100.0 / ((i_mb_count - i_all_intra)  ),
3218                      h->stat.i_mb_cbp[5] * 100.0 / ((i_mb_count - i_all_intra)) );
3219         x264_log( h, X264_LOG_INFO, "coded y,uvDC,uvAC intra: %.1f%% %.1f%% %.1f%%%s\n",
3220                   h->stat.i_mb_cbp[0] * 100.0 / (i_all_intra*4),
3221                   h->stat.i_mb_cbp[2] * 100.0 / (i_all_intra  ),
3222                   h->stat.i_mb_cbp[4] * 100.0 / (i_all_intra  ), buf );
3223
3224         int64_t fixed_pred_modes[4][9] = {{0}};
3225         int64_t sum_pred_modes[4] = {0};
3226         for( int i = 0; i <= I_PRED_16x16_DC_128; i++ )
3227         {
3228             fixed_pred_modes[0][x264_mb_pred_mode16x16_fix[i]] += h->stat.i_mb_pred_mode[0][i];
3229             sum_pred_modes[0] += h->stat.i_mb_pred_mode[0][i];
3230         }
3231         if( sum_pred_modes[0] )
3232             x264_log( h, X264_LOG_INFO, "i16 v,h,dc,p: %2.0f%% %2.0f%% %2.0f%% %2.0f%%\n",
3233                       fixed_pred_modes[0][0] * 100.0 / sum_pred_modes[0],
3234                       fixed_pred_modes[0][1] * 100.0 / sum_pred_modes[0],
3235                       fixed_pred_modes[0][2] * 100.0 / sum_pred_modes[0],
3236                       fixed_pred_modes[0][3] * 100.0 / sum_pred_modes[0] );
3237         for( int i = 1; i <= 2; i++ )
3238         {
3239             for( int j = 0; j <= I_PRED_8x8_DC_128; j++ )
3240             {
3241                 fixed_pred_modes[i][x264_mb_pred_mode4x4_fix(j)] += h->stat.i_mb_pred_mode[i][j];
3242                 sum_pred_modes[i] += h->stat.i_mb_pred_mode[i][j];
3243             }
3244             if( sum_pred_modes[i] )
3245                 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,
3246                           fixed_pred_modes[i][0] * 100.0 / sum_pred_modes[i],
3247                           fixed_pred_modes[i][1] * 100.0 / sum_pred_modes[i],
3248                           fixed_pred_modes[i][2] * 100.0 / sum_pred_modes[i],
3249                           fixed_pred_modes[i][3] * 100.0 / sum_pred_modes[i],
3250                           fixed_pred_modes[i][4] * 100.0 / sum_pred_modes[i],
3251                           fixed_pred_modes[i][5] * 100.0 / sum_pred_modes[i],
3252                           fixed_pred_modes[i][6] * 100.0 / sum_pred_modes[i],
3253                           fixed_pred_modes[i][7] * 100.0 / sum_pred_modes[i],
3254                           fixed_pred_modes[i][8] * 100.0 / sum_pred_modes[i] );
3255         }
3256         for( int i = 0; i <= I_PRED_CHROMA_DC_128; i++ )
3257         {
3258             fixed_pred_modes[3][x264_mb_pred_mode8x8c_fix[i]] += h->stat.i_mb_pred_mode[3][i];
3259             sum_pred_modes[3] += h->stat.i_mb_pred_mode[3][i];
3260         }
3261         if( sum_pred_modes[3] )
3262             x264_log( h, X264_LOG_INFO, "i8c dc,h,v,p: %2.0f%% %2.0f%% %2.0f%% %2.0f%%\n",
3263                       fixed_pred_modes[3][0] * 100.0 / sum_pred_modes[3],
3264                       fixed_pred_modes[3][1] * 100.0 / sum_pred_modes[3],
3265                       fixed_pred_modes[3][2] * 100.0 / sum_pred_modes[3],
3266                       fixed_pred_modes[3][3] * 100.0 / sum_pred_modes[3] );
3267
3268         if( h->param.analyse.i_weighted_pred >= X264_WEIGHTP_SIMPLE && h->stat.i_frame_count[SLICE_TYPE_P] > 0 )
3269             x264_log( h, X264_LOG_INFO, "Weighted P-Frames: Y:%.1f%% UV:%.1f%%\n",
3270                       h->stat.i_wpred[0] * 100.0 / h->stat.i_frame_count[SLICE_TYPE_P],
3271                       h->stat.i_wpred[1] * 100.0 / h->stat.i_frame_count[SLICE_TYPE_P] );
3272
3273         for( int i_list = 0; i_list < 2; i_list++ )
3274             for( int i_slice = 0; i_slice < 2; i_slice++ )
3275             {
3276                 char *p = buf;
3277                 int64_t i_den = 0;
3278                 int i_max = 0;
3279                 for( int i = 0; i < X264_REF_MAX*2; i++ )
3280                     if( h->stat.i_mb_count_ref[i_slice][i_list][i] )
3281                     {
3282                         i_den += h->stat.i_mb_count_ref[i_slice][i_list][i];
3283                         i_max = i;
3284                     }
3285                 if( i_max == 0 )
3286                     continue;
3287                 for( int i = 0; i <= i_max; i++ )
3288                     p += sprintf( p, " %4.1f%%", 100. * h->stat.i_mb_count_ref[i_slice][i_list][i] / i_den );
3289                 x264_log( h, X264_LOG_INFO, "ref %c L%d:%s\n", "PB"[i_slice], i_list, buf );
3290             }
3291
3292         if( h->param.analyse.b_ssim )
3293         {
3294             float ssim = SUM3( h->stat.f_ssim_mean_y ) / duration;
3295             x264_log( h, X264_LOG_INFO, "SSIM Mean Y:%.7f (%6.3fdb)\n", ssim, x264_ssim( ssim ) );
3296         }
3297         if( h->param.analyse.b_psnr )
3298         {
3299             x264_log( h, X264_LOG_INFO,
3300                       "PSNR Mean Y:%6.3f U:%6.3f V:%6.3f Avg:%6.3f Global:%6.3f kb/s:%.2f\n",
3301                       SUM3( h->stat.f_psnr_mean_y ) / duration,
3302                       SUM3( h->stat.f_psnr_mean_u ) / duration,
3303                       SUM3( h->stat.f_psnr_mean_v ) / duration,
3304                       SUM3( h->stat.f_psnr_average ) / duration,
3305                       x264_psnr( SUM3( h->stat.f_ssd_global ), duration * i_yuv_size ),
3306                       f_bitrate );
3307         }
3308         else
3309             x264_log( h, X264_LOG_INFO, "kb/s:%.2f\n", f_bitrate );
3310     }
3311
3312     /* rc */
3313     x264_ratecontrol_delete( h );
3314
3315     /* param */
3316     if( h->param.rc.psz_stat_out )
3317         free( h->param.rc.psz_stat_out );
3318     if( h->param.rc.psz_stat_in )
3319         free( h->param.rc.psz_stat_in );
3320
3321     x264_cqm_delete( h );
3322     x264_free( h->nal_buffer );
3323     x264_analyse_free_costs( h );
3324
3325     if( h->i_thread_frames > 1)
3326         h = h->thread[h->i_thread_phase];
3327
3328     /* frames */
3329     x264_frame_delete_list( h->frames.unused[0] );
3330     x264_frame_delete_list( h->frames.unused[1] );
3331     x264_frame_delete_list( h->frames.current );
3332     x264_frame_delete_list( h->frames.blank_unused );
3333
3334     h = h->thread[0];
3335
3336     for( int i = 0; i < h->i_thread_frames; i++ )
3337         if( h->thread[i]->b_thread_active )
3338             for( int j = 0; j < h->thread[i]->i_ref[0]; j++ )
3339                 if( h->thread[i]->fref[0][j] && h->thread[i]->fref[0][j]->b_duplicate )
3340                     x264_frame_delete( h->thread[i]->fref[0][j] );
3341
3342     for( int i = h->param.i_threads - 1; i >= 0; i-- )
3343     {
3344         x264_frame_t **frame;
3345
3346         if( !h->param.b_sliced_threads || i == 0 )
3347         {
3348             for( frame = h->thread[i]->frames.reference; *frame; frame++ )
3349             {
3350                 assert( (*frame)->i_reference_count > 0 );
3351                 (*frame)->i_reference_count--;
3352                 if( (*frame)->i_reference_count == 0 )
3353                     x264_frame_delete( *frame );
3354             }
3355             frame = &h->thread[i]->fdec;
3356             if( *frame )
3357             {
3358                 assert( (*frame)->i_reference_count > 0 );
3359                 (*frame)->i_reference_count--;
3360                 if( (*frame)->i_reference_count == 0 )
3361                     x264_frame_delete( *frame );
3362             }
3363             x264_macroblock_cache_free( h->thread[i] );
3364         }
3365         x264_macroblock_thread_free( h->thread[i], 0 );
3366         x264_free( h->thread[i]->out.p_bitstream );
3367         x264_free( h->thread[i]->out.nal);
3368         x264_free( h->thread[i] );
3369     }
3370 }
3371
3372 int x264_encoder_delayed_frames( x264_t *h )
3373 {
3374     int delayed_frames = 0;
3375     if( h->i_thread_frames > 1 )
3376     {
3377         for( int i = 0; i < h->i_thread_frames; i++ )
3378             delayed_frames += h->thread[i]->b_thread_active;
3379         h = h->thread[h->i_thread_phase];
3380     }
3381     for( int i = 0; h->frames.current[i]; i++ )
3382         delayed_frames++;
3383     x264_pthread_mutex_lock( &h->lookahead->ofbuf.mutex );
3384     x264_pthread_mutex_lock( &h->lookahead->ifbuf.mutex );
3385     x264_pthread_mutex_lock( &h->lookahead->next.mutex );
3386     delayed_frames += h->lookahead->ifbuf.i_size + h->lookahead->next.i_size + h->lookahead->ofbuf.i_size;
3387     x264_pthread_mutex_unlock( &h->lookahead->next.mutex );
3388     x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
3389     x264_pthread_mutex_unlock( &h->lookahead->ofbuf.mutex );
3390     return delayed_frames;
3391 }
3392
3393 int x264_encoder_maximum_delayed_frames( x264_t *h )
3394 {
3395     return h->frames.i_delay;
3396 }