]> git.sesse.net Git - x264/blob - encoder/encoder.c
Interactive encoder control: error resilience
[x264] / encoder / encoder.c
1 /*****************************************************************************
2  * x264: h264 encoder
3  *****************************************************************************
4  * Copyright (C) 2003-2008 x264 project
5  *
6  * Authors: Laurent Aimar <fenrir@via.ecp.fr>
7  *          Loren Merritt <lorenm@u.washington.edu>
8  *          Fiona Glaser <fiona@x264.com>
9  *
10  * This program is free software; you can redistribute it and/or modify
11  * it under the terms of the GNU General Public License as published by
12  * the Free Software Foundation; either version 2 of the License, or
13  * (at your option) any later version.
14  *
15  * This program is distributed in the hope that it will be useful,
16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
18  * GNU General Public License for more details.
19  *
20  * You should have received a copy of the GNU General Public License
21  * along with this program; if not, write to the Free Software
22  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02111, USA.
23  *****************************************************************************/
24
25 #include <math.h>
26
27 #include "common/common.h"
28
29 #include "set.h"
30 #include "analyse.h"
31 #include "ratecontrol.h"
32 #include "macroblock.h"
33 #include "me.h"
34
35 #if HAVE_VISUALIZE
36 #include "common/visualize.h"
37 #endif
38
39 //#define DEBUG_MB_TYPE
40
41 #define bs_write_ue bs_write_ue_big
42
43 static int x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
44                                    x264_nal_t **pp_nal, int *pi_nal,
45                                    x264_picture_t *pic_out );
46
47 /****************************************************************************
48  *
49  ******************************* x264 libs **********************************
50  *
51  ****************************************************************************/
52 static float x264_psnr( int64_t i_sqe, int64_t i_size )
53 {
54     double f_mse = (double)i_sqe / ((double)65025.0 * (double)i_size);
55     if( f_mse <= 0.0000000001 ) /* Max 100dB */
56         return 100;
57
58     return -10.0 * log10( f_mse );
59 }
60
61 static float x264_ssim( float ssim )
62 {
63     return -10.0 * log10( 1 - ssim );
64 }
65
66 static void x264_frame_dump( x264_t *h )
67 {
68     FILE *f = fopen( h->param.psz_dump_yuv, "r+b" );
69     if( !f )
70         return;
71     /* Write the frame in display order */
72     fseek( f, (uint64_t)h->fdec->i_frame * h->param.i_height * h->param.i_width * 3/2, SEEK_SET );
73     for( int i = 0; i < h->fdec->i_plane; i++ )
74         for( int y = 0; y < h->param.i_height >> !!i; y++ )
75             fwrite( &h->fdec->plane[i][y*h->fdec->i_stride[i]], 1, h->param.i_width >> !!i, f );
76     fclose( f );
77 }
78
79
80 /* Fill "default" values */
81 static void x264_slice_header_init( x264_t *h, x264_slice_header_t *sh,
82                                     x264_sps_t *sps, x264_pps_t *pps,
83                                     int i_idr_pic_id, int i_frame, int i_qp )
84 {
85     x264_param_t *param = &h->param;
86
87     /* First we fill all fields */
88     sh->sps = sps;
89     sh->pps = pps;
90
91     sh->i_first_mb  = 0;
92     sh->i_last_mb   = h->mb.i_mb_count - 1;
93     sh->i_pps_id    = pps->i_id;
94
95     sh->i_frame_num = i_frame;
96
97     sh->b_mbaff = h->param.b_interlaced;
98     sh->b_field_pic = 0;    /* no field support for now */
99     sh->b_bottom_field = 0; /* not yet used */
100
101     sh->i_idr_pic_id = i_idr_pic_id;
102
103     /* poc stuff, fixed later */
104     sh->i_poc = 0;
105     sh->i_delta_poc_bottom = 0;
106     sh->i_delta_poc[0] = 0;
107     sh->i_delta_poc[1] = 0;
108
109     sh->i_redundant_pic_cnt = 0;
110
111     h->mb.b_direct_auto_write = h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO
112                                 && h->param.i_bframe
113                                 && ( h->param.rc.b_stat_write || !h->param.rc.b_stat_read );
114
115     if( !h->mb.b_direct_auto_read && sh->i_type == SLICE_TYPE_B )
116     {
117         if( h->fref1[0]->i_poc_l0ref0 == h->fref0[0]->i_poc )
118         {
119             if( h->mb.b_direct_auto_write )
120                 sh->b_direct_spatial_mv_pred = ( h->stat.i_direct_score[1] > h->stat.i_direct_score[0] );
121             else
122                 sh->b_direct_spatial_mv_pred = ( param->analyse.i_direct_mv_pred == X264_DIRECT_PRED_SPATIAL );
123         }
124         else
125         {
126             h->mb.b_direct_auto_write = 0;
127             sh->b_direct_spatial_mv_pred = 1;
128         }
129     }
130     /* else b_direct_spatial_mv_pred was read from the 2pass statsfile */
131
132     sh->b_num_ref_idx_override = 0;
133     sh->i_num_ref_idx_l0_active = 1;
134     sh->i_num_ref_idx_l1_active = 1;
135
136     sh->b_ref_pic_list_reordering_l0 = h->b_ref_reorder[0];
137     sh->b_ref_pic_list_reordering_l1 = h->b_ref_reorder[1];
138
139     /* If the ref list isn't in the default order, construct reordering header */
140     /* List1 reordering isn't needed yet */
141     if( sh->b_ref_pic_list_reordering_l0 )
142     {
143         int pred_frame_num = i_frame;
144         for( int i = 0; i < h->i_ref0; i++ )
145         {
146             int diff = h->fref0[i]->i_frame_num - pred_frame_num;
147             sh->ref_pic_list_order[0][i].idc = ( diff > 0 );
148             sh->ref_pic_list_order[0][i].arg = (abs(diff) - 1) & ((1 << sps->i_log2_max_frame_num) - 1);
149             pred_frame_num = h->fref0[i]->i_frame_num;
150         }
151     }
152
153     sh->i_cabac_init_idc = param->i_cabac_init_idc;
154
155     sh->i_qp = i_qp;
156     sh->i_qp_delta = i_qp - pps->i_pic_init_qp;
157     sh->b_sp_for_swidth = 0;
158     sh->i_qs_delta = 0;
159
160     int deblock_thresh = i_qp + 2 * X264_MIN(param->i_deblocking_filter_alphac0, param->i_deblocking_filter_beta);
161     /* If effective qp <= 15, deblocking would have no effect anyway */
162     if( param->b_deblocking_filter && (h->mb.b_variable_qp || 15 < deblock_thresh ) )
163         sh->i_disable_deblocking_filter_idc = param->b_sliced_threads ? 2 : 0;
164     else
165         sh->i_disable_deblocking_filter_idc = 1;
166     sh->i_alpha_c0_offset = param->i_deblocking_filter_alphac0 << 1;
167     sh->i_beta_offset = param->i_deblocking_filter_beta << 1;
168 }
169
170 static void x264_slice_header_write( bs_t *s, x264_slice_header_t *sh, int i_nal_ref_idc )
171 {
172     if( sh->b_mbaff )
173     {
174         assert( sh->i_first_mb % (2*sh->sps->i_mb_width) == 0 );
175         bs_write_ue( s, sh->i_first_mb >> 1 );
176     }
177     else
178         bs_write_ue( s, sh->i_first_mb );
179
180     bs_write_ue( s, sh->i_type + 5 );   /* same type things */
181     bs_write_ue( s, sh->i_pps_id );
182     bs_write( s, sh->sps->i_log2_max_frame_num, sh->i_frame_num & ((1<<sh->sps->i_log2_max_frame_num)-1) );
183
184     if( !sh->sps->b_frame_mbs_only )
185     {
186         bs_write1( s, sh->b_field_pic );
187         if( sh->b_field_pic )
188             bs_write1( s, sh->b_bottom_field );
189     }
190
191     if( sh->i_idr_pic_id >= 0 ) /* NAL IDR */
192         bs_write_ue( s, sh->i_idr_pic_id );
193
194     if( sh->sps->i_poc_type == 0 )
195     {
196         bs_write( s, sh->sps->i_log2_max_poc_lsb, sh->i_poc & ((1<<sh->sps->i_log2_max_poc_lsb)-1) );
197         if( sh->pps->b_pic_order && !sh->b_field_pic )
198             bs_write_se( s, sh->i_delta_poc_bottom );
199     }
200     else if( sh->sps->i_poc_type == 1 && !sh->sps->b_delta_pic_order_always_zero )
201     {
202         bs_write_se( s, sh->i_delta_poc[0] );
203         if( sh->pps->b_pic_order && !sh->b_field_pic )
204             bs_write_se( s, sh->i_delta_poc[1] );
205     }
206
207     if( sh->pps->b_redundant_pic_cnt )
208         bs_write_ue( s, sh->i_redundant_pic_cnt );
209
210     if( sh->i_type == SLICE_TYPE_B )
211         bs_write1( s, sh->b_direct_spatial_mv_pred );
212
213     if( sh->i_type == SLICE_TYPE_P || sh->i_type == SLICE_TYPE_SP || sh->i_type == SLICE_TYPE_B )
214     {
215         bs_write1( s, sh->b_num_ref_idx_override );
216         if( sh->b_num_ref_idx_override )
217         {
218             bs_write_ue( s, sh->i_num_ref_idx_l0_active - 1 );
219             if( sh->i_type == SLICE_TYPE_B )
220                 bs_write_ue( s, sh->i_num_ref_idx_l1_active - 1 );
221         }
222     }
223
224     /* ref pic list reordering */
225     if( sh->i_type != SLICE_TYPE_I )
226     {
227         bs_write1( s, sh->b_ref_pic_list_reordering_l0 );
228         if( sh->b_ref_pic_list_reordering_l0 )
229         {
230             for( int i = 0; i < sh->i_num_ref_idx_l0_active; i++ )
231             {
232                 bs_write_ue( s, sh->ref_pic_list_order[0][i].idc );
233                 bs_write_ue( s, sh->ref_pic_list_order[0][i].arg );
234             }
235             bs_write_ue( s, 3 );
236         }
237     }
238     if( sh->i_type == SLICE_TYPE_B )
239     {
240         bs_write1( s, sh->b_ref_pic_list_reordering_l1 );
241         if( sh->b_ref_pic_list_reordering_l1 )
242         {
243             for( int i = 0; i < sh->i_num_ref_idx_l1_active; i++ )
244             {
245                 bs_write_ue( s, sh->ref_pic_list_order[1][i].idc );
246                 bs_write_ue( s, sh->ref_pic_list_order[1][i].arg );
247             }
248             bs_write_ue( s, 3 );
249         }
250     }
251
252     if( sh->pps->b_weighted_pred && ( sh->i_type == SLICE_TYPE_P || sh->i_type == SLICE_TYPE_SP ) )
253     {
254         /* pred_weight_table() */
255         bs_write_ue( s, sh->weight[0][0].i_denom );
256         bs_write_ue( s, sh->weight[0][1].i_denom );
257         for( int i = 0; i < sh->i_num_ref_idx_l0_active; i++ )
258         {
259             int luma_weight_l0_flag = !!sh->weight[i][0].weightfn;
260             int chroma_weight_l0_flag = !!sh->weight[i][1].weightfn || !!sh->weight[i][2].weightfn;
261             bs_write1( s, luma_weight_l0_flag );
262             if( luma_weight_l0_flag )
263             {
264                 bs_write_se( s, sh->weight[i][0].i_scale );
265                 bs_write_se( s, sh->weight[i][0].i_offset );
266             }
267             bs_write1( s, chroma_weight_l0_flag );
268             if( chroma_weight_l0_flag )
269             {
270                 for( int j = 1; j < 3; j++ )
271                 {
272                     bs_write_se( s, sh->weight[i][j].i_scale );
273                     bs_write_se( s, sh->weight[i][j].i_offset );
274                 }
275             }
276         }
277     }
278     else if( sh->pps->b_weighted_bipred == 1 && sh->i_type == SLICE_TYPE_B )
279     {
280       /* TODO */
281     }
282
283     if( i_nal_ref_idc != 0 )
284     {
285         if( sh->i_idr_pic_id >= 0 )
286         {
287             bs_write1( s, 0 );  /* no output of prior pics flag */
288             bs_write1( s, 0 );  /* long term reference flag */
289         }
290         else
291         {
292             bs_write1( s, sh->i_mmco_command_count > 0 ); /* adaptive_ref_pic_marking_mode_flag */
293             if( sh->i_mmco_command_count > 0 )
294             {
295                 for( int i = 0; i < sh->i_mmco_command_count; i++ )
296                 {
297                     bs_write_ue( s, 1 ); /* mark short term ref as unused */
298                     bs_write_ue( s, sh->mmco[i].i_difference_of_pic_nums - 1 );
299                 }
300                 bs_write_ue( s, 0 ); /* end command list */
301             }
302         }
303     }
304
305     if( sh->pps->b_cabac && sh->i_type != SLICE_TYPE_I )
306         bs_write_ue( s, sh->i_cabac_init_idc );
307
308     bs_write_se( s, sh->i_qp_delta );      /* slice qp delta */
309
310     if( sh->pps->b_deblocking_filter_control )
311     {
312         bs_write_ue( s, sh->i_disable_deblocking_filter_idc );
313         if( sh->i_disable_deblocking_filter_idc != 1 )
314         {
315             bs_write_se( s, sh->i_alpha_c0_offset >> 1 );
316             bs_write_se( s, sh->i_beta_offset >> 1 );
317         }
318     }
319 }
320
321 /* If we are within a reasonable distance of the end of the memory allocated for the bitstream, */
322 /* reallocate, adding an arbitrary amount of space (100 kilobytes). */
323 static int x264_bitstream_check_buffer( x264_t *h )
324 {
325     uint8_t *bs_bak = h->out.p_bitstream;
326     if( (h->param.b_cabac && (h->cabac.p_end - h->cabac.p < 2500)) ||
327         (h->out.bs.p_end - h->out.bs.p < 2500) )
328     {
329         h->out.i_bitstream += 100000;
330         CHECKED_MALLOC( h->out.p_bitstream, h->out.i_bitstream );
331         h->mc.memcpy_aligned( h->out.p_bitstream, bs_bak, (h->out.i_bitstream - 100000) & ~15 );
332         intptr_t delta = h->out.p_bitstream - bs_bak;
333
334         h->out.bs.p_start += delta;
335         h->out.bs.p += delta;
336         h->out.bs.p_end = h->out.p_bitstream + h->out.i_bitstream;
337
338         h->cabac.p_start += delta;
339         h->cabac.p += delta;
340         h->cabac.p_end = h->out.p_bitstream + h->out.i_bitstream;
341
342         for( int i = 0; i <= h->out.i_nal; i++ )
343             h->out.nal[i].p_payload += delta;
344         x264_free( bs_bak );
345     }
346     return 0;
347 fail:
348     x264_free( bs_bak );
349     return -1;
350 }
351
352 #if HAVE_PTHREAD
353 static void x264_encoder_thread_init( x264_t *h )
354 {
355     if( h->param.i_sync_lookahead )
356         x264_lower_thread_priority( 10 );
357
358 #if HAVE_MMX
359     /* Misalign mask has to be set separately for each thread. */
360     if( h->param.cpu&X264_CPU_SSE_MISALIGN )
361         x264_cpu_mask_misalign_sse();
362 #endif
363 }
364 #endif
365
366 /****************************************************************************
367  *
368  ****************************************************************************
369  ****************************** External API*********************************
370  ****************************************************************************
371  *
372  ****************************************************************************/
373
374 static int x264_validate_parameters( x264_t *h )
375 {
376 #if HAVE_MMX
377 #ifdef __SSE__
378     if( !(x264_cpu_detect() & X264_CPU_SSE) )
379     {
380         x264_log( h, X264_LOG_ERROR, "your cpu does not support SSE1, but x264 was compiled with asm support\n");
381 #else
382     if( !(x264_cpu_detect() & X264_CPU_MMXEXT) )
383     {
384         x264_log( h, X264_LOG_ERROR, "your cpu does not support MMXEXT, but x264 was compiled with asm support\n");
385 #endif
386         x264_log( h, X264_LOG_ERROR, "to run x264, recompile without asm support (configure --disable-asm)\n");
387         return -1;
388     }
389 #endif
390     if( h->param.i_width <= 0 || h->param.i_height <= 0 )
391     {
392         x264_log( h, X264_LOG_ERROR, "invalid width x height (%dx%d)\n",
393                   h->param.i_width, h->param.i_height );
394         return -1;
395     }
396
397     if( h->param.i_width % 2 || h->param.i_height % 2 )
398     {
399         x264_log( h, X264_LOG_ERROR, "width or height not divisible by 2 (%dx%d)\n",
400                   h->param.i_width, h->param.i_height );
401         return -1;
402     }
403     int i_csp = h->param.i_csp & X264_CSP_MASK;
404     if( i_csp != X264_CSP_I420 && i_csp != X264_CSP_YV12 )
405     {
406         x264_log( h, X264_LOG_ERROR, "invalid CSP (only I420/YV12 supported)\n" );
407         return -1;
408     }
409
410     if( h->param.i_threads == X264_THREADS_AUTO )
411         h->param.i_threads = x264_cpu_num_processors() * (h->param.b_sliced_threads?2:3)/2;
412     h->param.i_threads = x264_clip3( h->param.i_threads, 1, X264_THREAD_MAX );
413     if( h->param.i_threads > 1 )
414     {
415 #if !HAVE_PTHREAD
416         x264_log( h, X264_LOG_WARNING, "not compiled with pthread support!\n");
417         h->param.i_threads = 1;
418 #endif
419         /* Avoid absurdly small thread slices as they can reduce performance
420          * and VBV compliance.  Capped at an arbitrary 4 rows per thread. */
421         if( h->param.b_sliced_threads )
422         {
423             int max_threads = (h->param.i_height+15)/16 / 4;
424             h->param.i_threads = X264_MIN( h->param.i_threads, max_threads );
425         }
426     }
427     else
428         h->param.b_sliced_threads = 0;
429     h->i_thread_frames = h->param.b_sliced_threads ? 1 : h->param.i_threads;
430
431     if( h->param.b_interlaced )
432     {
433         if( h->param.analyse.i_me_method >= X264_ME_ESA )
434         {
435             x264_log( h, X264_LOG_WARNING, "interlace + me=esa is not implemented\n" );
436             h->param.analyse.i_me_method = X264_ME_UMH;
437         }
438         if( h->param.analyse.i_weighted_pred > 0 )
439         {
440             x264_log( h, X264_LOG_WARNING, "interlace + weightp is not implemented\n" );
441             h->param.analyse.i_weighted_pred = X264_WEIGHTP_NONE;
442         }
443     }
444
445     /* Detect default ffmpeg settings and terminate with an error. */
446     {
447         int score = 0;
448         score += h->param.analyse.i_me_range == 0;
449         score += h->param.rc.i_qp_step == 3;
450         score += h->param.i_keyint_max == 12;
451         score += h->param.rc.i_qp_min == 2;
452         score += h->param.rc.i_qp_max == 31;
453         score += h->param.rc.f_qcompress == 0.5;
454         score += fabs(h->param.rc.f_ip_factor - 1.25) < 0.01;
455         score += fabs(h->param.rc.f_pb_factor - 1.25) < 0.01;
456         score += h->param.analyse.inter == 0 && h->param.analyse.i_subpel_refine == 8;
457         if( score >= 5 )
458         {
459             x264_log( h, X264_LOG_ERROR, "broken ffmpeg default settings detected\n" );
460             x264_log( h, X264_LOG_ERROR, "use an encoding preset (vpre)\n" );
461             return -1;
462         }
463     }
464
465     if( h->param.rc.i_rc_method < 0 || h->param.rc.i_rc_method > 2 )
466     {
467         x264_log( h, X264_LOG_ERROR, "no ratecontrol method specified\n" );
468         return -1;
469     }
470     h->param.rc.f_rf_constant = x264_clip3f( h->param.rc.f_rf_constant, 0, 51 );
471     h->param.rc.i_qp_constant = x264_clip3( h->param.rc.i_qp_constant, 0, 51 );
472     if( h->param.rc.i_rc_method == X264_RC_CRF )
473     {
474         h->param.rc.i_qp_constant = h->param.rc.f_rf_constant;
475         h->param.rc.i_bitrate = 0;
476     }
477     if( (h->param.rc.i_rc_method == X264_RC_CQP || h->param.rc.i_rc_method == X264_RC_CRF)
478         && h->param.rc.i_qp_constant == 0 )
479     {
480         h->mb.b_lossless = 1;
481         h->param.i_cqm_preset = X264_CQM_FLAT;
482         h->param.psz_cqm_file = NULL;
483         h->param.rc.i_rc_method = X264_RC_CQP;
484         h->param.rc.f_ip_factor = 1;
485         h->param.rc.f_pb_factor = 1;
486         h->param.analyse.b_psnr = 0;
487         h->param.analyse.b_ssim = 0;
488         h->param.analyse.i_chroma_qp_offset = 0;
489         h->param.analyse.i_trellis = 0;
490         h->param.analyse.b_fast_pskip = 0;
491         h->param.analyse.i_noise_reduction = 0;
492         h->param.analyse.b_psy = 0;
493         h->param.i_bframe = 0;
494         /* 8x8dct is not useful at all in CAVLC lossless */
495         if( !h->param.b_cabac )
496             h->param.analyse.b_transform_8x8 = 0;
497     }
498     if( h->param.rc.i_rc_method == X264_RC_CQP )
499     {
500         float qp_p = h->param.rc.i_qp_constant;
501         float qp_i = qp_p - 6*log2f( h->param.rc.f_ip_factor );
502         float qp_b = qp_p + 6*log2f( h->param.rc.f_pb_factor );
503         h->param.rc.i_qp_min = x264_clip3( (int)(X264_MIN3( qp_p, qp_i, qp_b )), 0, 51 );
504         h->param.rc.i_qp_max = x264_clip3( (int)(X264_MAX3( qp_p, qp_i, qp_b ) + .999), 0, 51 );
505         h->param.rc.i_aq_mode = 0;
506         h->param.rc.b_mb_tree = 0;
507     }
508     h->param.rc.i_qp_max = x264_clip3( h->param.rc.i_qp_max, 0, 51 );
509     h->param.rc.i_qp_min = x264_clip3( h->param.rc.i_qp_min, 0, h->param.rc.i_qp_max );
510     if( h->param.rc.i_vbv_buffer_size )
511     {
512         if( h->param.rc.i_rc_method == X264_RC_CQP )
513         {
514             x264_log( h, X264_LOG_WARNING, "VBV is incompatible with constant QP, ignored.\n" );
515             h->param.rc.i_vbv_max_bitrate = 0;
516             h->param.rc.i_vbv_buffer_size = 0;
517         }
518         else if( h->param.rc.i_vbv_max_bitrate == 0 )
519         {
520             if( h->param.rc.i_rc_method == X264_RC_ABR )
521             {
522                 x264_log( h, X264_LOG_WARNING, "VBV maxrate unspecified, assuming CBR\n" );
523                 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate;
524             }
525             else
526             {
527                 x264_log( h, X264_LOG_WARNING, "VBV bufsize set but maxrate unspecified, ignored\n" );
528                 h->param.rc.i_vbv_buffer_size = 0;
529             }
530         }
531         else if( h->param.rc.i_vbv_max_bitrate < h->param.rc.i_bitrate &&
532                  h->param.rc.i_rc_method == X264_RC_ABR )
533         {
534             x264_log( h, X264_LOG_WARNING, "max bitrate less than average bitrate, assuming CBR\n" );
535             h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate;
536         }
537     }
538     else if( h->param.rc.i_vbv_max_bitrate )
539     {
540         x264_log( h, X264_LOG_WARNING, "VBV maxrate specified, but no bufsize, ignored\n" );
541         h->param.rc.i_vbv_max_bitrate = 0;
542     }
543
544     if( h->param.b_interlaced && h->param.i_slice_max_size )
545     {
546         x264_log( h, X264_LOG_WARNING, "interlaced + slice-max-size is not implemented\n" );
547         h->param.i_slice_max_size = 0;
548     }
549     if( h->param.b_interlaced && h->param.i_slice_max_mbs )
550     {
551         x264_log( h, X264_LOG_WARNING, "interlaced + slice-max-mbs is not implemented\n" );
552         h->param.i_slice_max_mbs = 0;
553     }
554     int max_slices = (h->param.i_height+((16<<h->param.b_interlaced)-1))/(16<<h->param.b_interlaced);
555     if( h->param.b_sliced_threads )
556         h->param.i_slice_count = x264_clip3( h->param.i_threads, 0, max_slices );
557     else
558     {
559         h->param.i_slice_count = x264_clip3( h->param.i_slice_count, 0, max_slices );
560         h->param.i_slice_max_size = X264_MAX( h->param.i_slice_max_size, 0 );
561         h->param.i_slice_max_mbs = X264_MAX( h->param.i_slice_max_mbs, 0 );
562         if( h->param.i_slice_max_mbs || h->param.i_slice_max_size )
563             h->param.i_slice_count = 0;
564     }
565
566     h->param.i_frame_reference = x264_clip3( h->param.i_frame_reference, 1, 16 );
567     h->param.i_dpb_size = x264_clip3( h->param.i_dpb_size, 1, 16 );
568     if( h->param.i_keyint_max <= 0 )
569         h->param.i_keyint_max = 1;
570     if( h->param.i_scenecut_threshold < 0 )
571         h->param.i_scenecut_threshold = 0;
572     if( !h->param.analyse.i_subpel_refine && h->param.analyse.i_direct_mv_pred > X264_DIRECT_PRED_SPATIAL )
573     {
574         x264_log( h, X264_LOG_WARNING, "subme=0 + direct=temporal is not supported\n" );
575         h->param.analyse.i_direct_mv_pred = X264_DIRECT_PRED_SPATIAL;
576     }
577     h->param.i_bframe = x264_clip3( h->param.i_bframe, 0, X264_MIN( X264_BFRAME_MAX, h->param.i_keyint_max-1 ) );
578     h->param.i_open_gop = x264_clip3( h->param.i_open_gop, X264_OPEN_GOP_NONE, X264_OPEN_GOP_CODED_ORDER );
579     if( h->param.i_keyint_max == 1 )
580         h->param.b_intra_refresh = 0;
581     h->param.i_bframe_bias = x264_clip3( h->param.i_bframe_bias, -90, 100 );
582     if( h->param.i_bframe <= 1 )
583         h->param.i_bframe_pyramid = X264_B_PYRAMID_NONE;
584     h->param.i_bframe_pyramid = x264_clip3( h->param.i_bframe_pyramid, X264_B_PYRAMID_NONE, X264_B_PYRAMID_NORMAL );
585     if( !h->param.i_bframe )
586     {
587         h->param.i_bframe_adaptive = X264_B_ADAPT_NONE;
588         h->param.analyse.i_direct_mv_pred = 0;
589         h->param.analyse.b_weighted_bipred = 0;
590         h->param.i_open_gop = X264_OPEN_GOP_NONE;
591     }
592     if( h->param.b_intra_refresh && h->param.i_bframe_pyramid == X264_B_PYRAMID_NORMAL )
593     {
594         x264_log( h, X264_LOG_WARNING, "b-pyramid normal + intra-refresh is not supported\n" );
595         h->param.i_bframe_pyramid = X264_B_PYRAMID_STRICT;
596     }
597     if( h->param.b_intra_refresh && (h->param.i_frame_reference > 1 || h->param.i_dpb_size > 1) )
598     {
599         x264_log( h, X264_LOG_WARNING, "ref > 1 + intra-refresh is not supported\n" );
600         h->param.i_frame_reference = 1;
601         h->param.i_dpb_size = 1;
602     }
603     if( h->param.b_intra_refresh && h->param.i_open_gop )
604     {
605         x264_log( h, X264_LOG_WARNING, "intra-refresh is not compatible with open-gop\n" );
606         h->param.i_open_gop = X264_OPEN_GOP_NONE;
607     }
608     if( h->param.i_keyint_min == X264_KEYINT_MIN_AUTO )
609         h->param.i_keyint_min = h->param.i_keyint_max / 10;
610     h->param.i_keyint_min = x264_clip3( h->param.i_keyint_min, 1, h->param.i_keyint_max/2+1 );
611     h->param.rc.i_lookahead = x264_clip3( h->param.rc.i_lookahead, 0, X264_LOOKAHEAD_MAX );
612     {
613         int maxrate = X264_MAX( h->param.rc.i_vbv_max_bitrate, h->param.rc.i_bitrate );
614         float bufsize = maxrate ? (float)h->param.rc.i_vbv_buffer_size / maxrate : 0;
615         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;
616         h->param.rc.i_lookahead = X264_MIN( h->param.rc.i_lookahead, X264_MAX( h->param.i_keyint_max, bufsize*fps ) );
617     }
618
619     if( !h->param.i_timebase_num || !h->param.i_timebase_den )
620     {
621         h->param.i_timebase_num = h->param.i_fps_den;
622         h->param.i_timebase_den = h->param.i_fps_num;
623     }
624
625     h->param.rc.f_qcompress = x264_clip3f( h->param.rc.f_qcompress, 0.0, 1.0 );
626     if( h->param.i_keyint_max == 1 || h->param.rc.f_qcompress == 1 )
627         h->param.rc.b_mb_tree = 0;
628     if( !h->param.rc.i_lookahead && !h->param.b_intra_refresh && h->param.rc.b_mb_tree )
629     {
630         x264_log( h, X264_LOG_WARNING, "lookaheadless mb-tree requires intra refresh\n" );
631         h->param.rc.b_mb_tree = 0;
632     }
633     if( h->param.rc.b_stat_read )
634         h->param.rc.i_lookahead = 0;
635 #if HAVE_PTHREAD
636     if( h->param.i_sync_lookahead < 0 )
637         h->param.i_sync_lookahead = h->param.i_bframe + 1;
638     h->param.i_sync_lookahead = X264_MIN( h->param.i_sync_lookahead, X264_LOOKAHEAD_MAX );
639     if( h->param.rc.b_stat_read || h->i_thread_frames == 1 )
640         h->param.i_sync_lookahead = 0;
641 #else
642     h->param.i_sync_lookahead = 0;
643 #endif
644
645     h->param.i_deblocking_filter_alphac0 = x264_clip3( h->param.i_deblocking_filter_alphac0, -6, 6 );
646     h->param.i_deblocking_filter_beta    = x264_clip3( h->param.i_deblocking_filter_beta, -6, 6 );
647     h->param.analyse.i_luma_deadzone[0] = x264_clip3( h->param.analyse.i_luma_deadzone[0], 0, 32 );
648     h->param.analyse.i_luma_deadzone[1] = x264_clip3( h->param.analyse.i_luma_deadzone[1], 0, 32 );
649
650     h->param.i_cabac_init_idc = x264_clip3( h->param.i_cabac_init_idc, 0, 2 );
651
652     if( h->param.i_cqm_preset < X264_CQM_FLAT || h->param.i_cqm_preset > X264_CQM_CUSTOM )
653         h->param.i_cqm_preset = X264_CQM_FLAT;
654
655     if( h->param.analyse.i_me_method < X264_ME_DIA ||
656         h->param.analyse.i_me_method > X264_ME_TESA )
657         h->param.analyse.i_me_method = X264_ME_HEX;
658     if( h->param.analyse.i_me_range < 4 )
659         h->param.analyse.i_me_range = 4;
660     if( h->param.analyse.i_me_range > 16 && h->param.analyse.i_me_method <= X264_ME_HEX )
661         h->param.analyse.i_me_range = 16;
662     if( h->param.analyse.i_me_method == X264_ME_TESA &&
663         (h->mb.b_lossless || h->param.analyse.i_subpel_refine <= 1) )
664         h->param.analyse.i_me_method = X264_ME_ESA;
665     h->param.analyse.i_subpel_refine = x264_clip3( h->param.analyse.i_subpel_refine, 0, 10 );
666     h->param.analyse.b_mixed_references = h->param.analyse.b_mixed_references && h->param.i_frame_reference > 1;
667     h->param.analyse.inter &= X264_ANALYSE_PSUB16x16|X264_ANALYSE_PSUB8x8|X264_ANALYSE_BSUB16x16|
668                               X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
669     h->param.analyse.intra &= X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
670     if( !(h->param.analyse.inter & X264_ANALYSE_PSUB16x16) )
671         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
672     if( !h->param.analyse.b_transform_8x8 )
673     {
674         h->param.analyse.inter &= ~X264_ANALYSE_I8x8;
675         h->param.analyse.intra &= ~X264_ANALYSE_I8x8;
676     }
677     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -12, 12);
678     if( !h->param.b_cabac )
679         h->param.analyse.i_trellis = 0;
680     h->param.analyse.i_trellis = x264_clip3( h->param.analyse.i_trellis, 0, 2 );
681     if( !h->param.analyse.b_psy )
682     {
683         h->param.analyse.f_psy_rd = 0;
684         h->param.analyse.f_psy_trellis = 0;
685     }
686     if( !h->param.analyse.i_trellis )
687         h->param.analyse.f_psy_trellis = 0;
688     h->param.analyse.f_psy_rd = x264_clip3f( h->param.analyse.f_psy_rd, 0, 10 );
689     h->param.analyse.f_psy_trellis = x264_clip3f( h->param.analyse.f_psy_trellis, 0, 10 );
690     if( h->param.analyse.i_subpel_refine < 6 )
691         h->param.analyse.f_psy_rd = 0;
692     h->mb.i_psy_rd = FIX8( h->param.analyse.f_psy_rd );
693     /* Psy RDO increases overall quantizers to improve the quality of luma--this indirectly hurts chroma quality */
694     /* so we lower the chroma QP offset to compensate */
695     /* This can be triggered repeatedly on multiple calls to parameter_validate, but since encoding
696      * uses the pps chroma qp offset not the param chroma qp offset, this is not a problem. */
697     if( h->mb.i_psy_rd )
698         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_rd < 0.25 ? 1 : 2;
699     h->mb.i_psy_trellis = FIX8( h->param.analyse.f_psy_trellis / 4 );
700     /* Psy trellis has a similar effect. */
701     if( h->mb.i_psy_trellis )
702         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_trellis < 0.25 ? 1 : 2;
703     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -12, 12);
704     h->param.rc.i_aq_mode = x264_clip3( h->param.rc.i_aq_mode, 0, 2 );
705     h->param.rc.f_aq_strength = x264_clip3f( h->param.rc.f_aq_strength, 0, 3 );
706     if( h->param.rc.f_aq_strength == 0 )
707         h->param.rc.i_aq_mode = 0;
708     /* MB-tree requires AQ to be on, even if the strength is zero. */
709     if( !h->param.rc.i_aq_mode && h->param.rc.b_mb_tree )
710     {
711         h->param.rc.i_aq_mode = 1;
712         h->param.rc.f_aq_strength = 0;
713     }
714     h->param.analyse.i_noise_reduction = x264_clip3( h->param.analyse.i_noise_reduction, 0, 1<<16 );
715     if( h->param.analyse.i_subpel_refine == 10 && (h->param.analyse.i_trellis != 2 || !h->param.rc.i_aq_mode) )
716         h->param.analyse.i_subpel_refine = 9;
717
718     {
719         const x264_level_t *l = x264_levels;
720         if( h->param.i_level_idc < 0 )
721         {
722             int maxrate_bak = h->param.rc.i_vbv_max_bitrate;
723             if( h->param.rc.i_rc_method == X264_RC_ABR && h->param.rc.i_vbv_buffer_size <= 0 )
724                 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate * 2;
725             h->sps = h->sps_array;
726             x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
727             do h->param.i_level_idc = l->level_idc;
728                 while( l[1].level_idc && x264_validate_levels( h, 0 ) && l++ );
729             h->param.rc.i_vbv_max_bitrate = maxrate_bak;
730         }
731         else
732         {
733             while( l->level_idc && l->level_idc != h->param.i_level_idc )
734                 l++;
735             if( l->level_idc == 0 )
736             {
737                 x264_log( h, X264_LOG_ERROR, "invalid level_idc: %d\n", h->param.i_level_idc );
738                 return -1;
739             }
740         }
741         if( h->param.analyse.i_mv_range <= 0 )
742             h->param.analyse.i_mv_range = l->mv_range >> h->param.b_interlaced;
743         else
744             h->param.analyse.i_mv_range = x264_clip3(h->param.analyse.i_mv_range, 32, 512 >> h->param.b_interlaced);
745     }
746
747     h->param.analyse.i_weighted_pred = x264_clip3( h->param.analyse.i_weighted_pred, 0, X264_WEIGHTP_SMART );
748     if( !h->param.analyse.i_weighted_pred && h->param.rc.b_mb_tree && h->param.analyse.b_psy && !h->param.b_interlaced )
749         h->param.analyse.i_weighted_pred = X264_WEIGHTP_FAKE;
750
751     if( h->i_thread_frames > 1 )
752     {
753         int r = h->param.analyse.i_mv_range_thread;
754         int r2;
755         if( r <= 0 )
756         {
757             // half of the available space is reserved and divided evenly among the threads,
758             // the rest is allocated to whichever thread is far enough ahead to use it.
759             // reserving more space increases quality for some videos, but costs more time
760             // in thread synchronization.
761             int max_range = (h->param.i_height + X264_THREAD_HEIGHT) / h->i_thread_frames - X264_THREAD_HEIGHT;
762             r = max_range / 2;
763         }
764         r = X264_MAX( r, h->param.analyse.i_me_range );
765         r = X264_MIN( r, h->param.analyse.i_mv_range );
766         // round up to use the whole mb row
767         r2 = (r & ~15) + ((-X264_THREAD_HEIGHT) & 15);
768         if( r2 < r )
769             r2 += 16;
770         x264_log( h, X264_LOG_DEBUG, "using mv_range_thread = %d\n", r2 );
771         h->param.analyse.i_mv_range_thread = r2;
772     }
773
774     if( h->param.rc.f_qblur < 0 )
775         h->param.rc.f_qblur = 0;
776     if( h->param.rc.f_complexity_blur < 0 )
777         h->param.rc.f_complexity_blur = 0;
778
779     h->param.i_sps_id &= 31;
780
781     if( h->param.i_log_level < X264_LOG_INFO )
782     {
783         h->param.analyse.b_psnr = 0;
784         h->param.analyse.b_ssim = 0;
785     }
786
787     if( h->param.b_interlaced )
788         h->param.b_pic_struct = 1;
789
790     if( h->param.i_nal_hrd && !h->param.rc.i_vbv_buffer_size )
791     {
792         x264_log( h, X264_LOG_WARNING, "NAL HRD parameters require VBV parameters\n" );
793         h->param.i_nal_hrd = X264_NAL_HRD_NONE;
794     }
795
796     if( h->param.i_nal_hrd == X264_NAL_HRD_CBR &&
797        (h->param.rc.i_bitrate != h->param.rc.i_vbv_max_bitrate || !h->param.rc.i_vbv_max_bitrate) )
798     {
799         x264_log( h, X264_LOG_WARNING, "CBR HRD requires constant bitrate\n" );
800         h->param.i_nal_hrd = X264_NAL_HRD_VBR;
801     }
802
803     /* ensure the booleans are 0 or 1 so they can be used in math */
804 #define BOOLIFY(x) h->param.x = !!h->param.x
805     BOOLIFY( b_cabac );
806     BOOLIFY( b_constrained_intra );
807     BOOLIFY( b_deblocking_filter );
808     BOOLIFY( b_deterministic );
809     BOOLIFY( b_sliced_threads );
810     BOOLIFY( b_interlaced );
811     BOOLIFY( b_intra_refresh );
812     BOOLIFY( b_visualize );
813     BOOLIFY( b_aud );
814     BOOLIFY( b_repeat_headers );
815     BOOLIFY( b_annexb );
816     BOOLIFY( b_vfr_input );
817     BOOLIFY( b_pic_struct );
818     BOOLIFY( b_fake_interlaced );
819     BOOLIFY( analyse.b_transform_8x8 );
820     BOOLIFY( analyse.b_weighted_bipred );
821     BOOLIFY( analyse.b_chroma_me );
822     BOOLIFY( analyse.b_mixed_references );
823     BOOLIFY( analyse.b_fast_pskip );
824     BOOLIFY( analyse.b_dct_decimate );
825     BOOLIFY( analyse.b_psy );
826     BOOLIFY( analyse.b_psnr );
827     BOOLIFY( analyse.b_ssim );
828     BOOLIFY( rc.b_stat_write );
829     BOOLIFY( rc.b_stat_read );
830     BOOLIFY( rc.b_mb_tree );
831 #undef BOOLIFY
832
833     return 0;
834 }
835
836 static void mbcmp_init( x264_t *h )
837 {
838     int satd = !h->mb.b_lossless && h->param.analyse.i_subpel_refine > 1;
839     memcpy( h->pixf.mbcmp, satd ? h->pixf.satd : h->pixf.sad_aligned, sizeof(h->pixf.mbcmp) );
840     memcpy( h->pixf.mbcmp_unaligned, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.mbcmp_unaligned) );
841     h->pixf.intra_mbcmp_x3_16x16 = satd ? h->pixf.intra_satd_x3_16x16 : h->pixf.intra_sad_x3_16x16;
842     h->pixf.intra_mbcmp_x3_8x8c = satd ? h->pixf.intra_satd_x3_8x8c : h->pixf.intra_sad_x3_8x8c;
843     h->pixf.intra_mbcmp_x3_8x8 = satd ? h->pixf.intra_sa8d_x3_8x8 : h->pixf.intra_sad_x3_8x8;
844     h->pixf.intra_mbcmp_x3_4x4 = satd ? h->pixf.intra_satd_x3_4x4 : h->pixf.intra_sad_x3_4x4;
845     satd &= h->param.analyse.i_me_method == X264_ME_TESA;
846     memcpy( h->pixf.fpelcmp, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.fpelcmp) );
847     memcpy( h->pixf.fpelcmp_x3, satd ? h->pixf.satd_x3 : h->pixf.sad_x3, sizeof(h->pixf.fpelcmp_x3) );
848     memcpy( h->pixf.fpelcmp_x4, satd ? h->pixf.satd_x4 : h->pixf.sad_x4, sizeof(h->pixf.fpelcmp_x4) );
849 }
850
851 static void x264_set_aspect_ratio( x264_t *h, x264_param_t *param, int initial )
852 {
853     /* VUI */
854     if( param->vui.i_sar_width > 0 && param->vui.i_sar_height > 0 )
855     {
856         uint32_t i_w = param->vui.i_sar_width;
857         uint32_t i_h = param->vui.i_sar_height;
858         uint32_t old_w = h->param.vui.i_sar_width;
859         uint32_t old_h = h->param.vui.i_sar_height;
860
861         x264_reduce_fraction( &i_w, &i_h );
862
863         while( i_w > 65535 || i_h > 65535 )
864         {
865             i_w /= 2;
866             i_h /= 2;
867         }
868
869         x264_reduce_fraction( &i_w, &i_h );
870
871         if( i_w != old_w || i_h != old_h || initial )
872         {
873             h->param.vui.i_sar_width = 0;
874             h->param.vui.i_sar_height = 0;
875             if( i_w == 0 || i_h == 0 )
876                 x264_log( h, X264_LOG_WARNING, "cannot create valid sample aspect ratio\n" );
877             else
878             {
879                 x264_log( h, initial?X264_LOG_INFO:X264_LOG_DEBUG, "using SAR=%d/%d\n", i_w, i_h );
880                 h->param.vui.i_sar_width = i_w;
881                 h->param.vui.i_sar_height = i_h;
882             }
883         }
884     }
885 }
886
887 /****************************************************************************
888  * x264_encoder_open:
889  ****************************************************************************/
890 x264_t *x264_encoder_open( x264_param_t *param )
891 {
892     x264_t *h;
893     char buf[1000], *p;
894     int qp, i_slicetype_length;
895
896     CHECKED_MALLOCZERO( h, sizeof(x264_t) );
897
898     /* Create a copy of param */
899     memcpy( &h->param, param, sizeof(x264_param_t) );
900
901     if( param->param_free )
902         param->param_free( param );
903
904     if( x264_validate_parameters( h ) < 0 )
905         goto fail;
906
907     if( h->param.psz_cqm_file )
908         if( x264_cqm_parse_file( h, h->param.psz_cqm_file ) < 0 )
909             goto fail;
910
911     if( h->param.rc.psz_stat_out )
912         h->param.rc.psz_stat_out = strdup( h->param.rc.psz_stat_out );
913     if( h->param.rc.psz_stat_in )
914         h->param.rc.psz_stat_in = strdup( h->param.rc.psz_stat_in );
915
916     x264_set_aspect_ratio( h, &h->param, 1 );
917
918     x264_reduce_fraction( &h->param.i_fps_num, &h->param.i_fps_den );
919     x264_reduce_fraction( &h->param.i_timebase_num, &h->param.i_timebase_den );
920
921     /* Init x264_t */
922     h->i_frame = -1;
923     h->i_frame_num = 0;
924     h->i_idr_pic_id = 0;
925     uint64_t new_timebase_den = h->param.i_timebase_den;
926     if( h->param.b_dts_compress )
927     {
928         /* h->i_dts_compress_multiplier == h->frames.i_bframe_delay + 1 */
929         h->i_dts_compress_multiplier = h->param.i_bframe ? (h->param.i_bframe_pyramid ? 3 : 2) : 1;
930         if( h->i_dts_compress_multiplier != 1 )
931         {
932             new_timebase_den = h->param.i_timebase_den * h->i_dts_compress_multiplier;
933             x264_log( h, X264_LOG_DEBUG, "DTS compresion changed timebase: %u/%u -> %u/%"PRIu64"\n",
934                       h->param.i_timebase_num, h->param.i_timebase_den,
935                       h->param.i_timebase_num, new_timebase_den );
936         }
937     }
938     else
939         h->i_dts_compress_multiplier = 1;
940
941     if( new_timebase_den * 2 > UINT32_MAX )
942     {
943         x264_log( h, X264_LOG_ERROR, "Effective timebase denominator %"PRIu64" exceeds H.264 maximum\n", new_timebase_den );
944         goto fail;
945     }
946     h->param.i_timebase_den = new_timebase_den;
947
948     h->sps = &h->sps_array[0];
949     x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
950
951     h->pps = &h->pps_array[0];
952     x264_pps_init( h->pps, h->param.i_sps_id, &h->param, h->sps );
953
954     x264_validate_levels( h, 1 );
955
956     h->chroma_qp_table = i_chroma_qp_table + 12 + h->pps->i_chroma_qp_index_offset;
957
958     if( x264_cqm_init( h ) < 0 )
959         goto fail;
960
961     h->mb.i_mb_width = h->sps->i_mb_width;
962     h->mb.i_mb_height = h->sps->i_mb_height;
963     h->mb.i_mb_count = h->mb.i_mb_width * h->mb.i_mb_height;
964
965     /* Init frames. */
966     if( h->param.i_bframe_adaptive == X264_B_ADAPT_TRELLIS && !h->param.rc.b_stat_read )
967         h->frames.i_delay = X264_MAX(h->param.i_bframe,3)*4;
968     else
969         h->frames.i_delay = h->param.i_bframe;
970     if( h->param.rc.b_mb_tree || h->param.rc.i_vbv_buffer_size )
971         h->frames.i_delay = X264_MAX( h->frames.i_delay, h->param.rc.i_lookahead );
972     i_slicetype_length = h->frames.i_delay;
973     h->frames.i_delay += h->i_thread_frames - 1;
974     h->frames.i_delay += h->param.i_sync_lookahead;
975     h->frames.i_delay += h->param.b_vfr_input && (h->param.rc.i_rc_method == X264_RC_ABR || h->param.rc.b_stat_write
976                                                  || h->param.rc.i_vbv_buffer_size);
977     h->frames.i_bframe_delay = h->param.i_bframe ? (h->param.i_bframe_pyramid ? 2 : 1) : 0;
978
979     h->frames.i_max_ref0 = h->param.i_frame_reference;
980     h->frames.i_max_ref1 = X264_MIN( h->sps->vui.i_num_reorder_frames, h->param.i_frame_reference );
981     h->frames.i_max_dpb  = h->sps->vui.i_max_dec_frame_buffering;
982     h->frames.b_have_lowres = !h->param.rc.b_stat_read
983         && ( h->param.rc.i_rc_method == X264_RC_ABR
984           || h->param.rc.i_rc_method == X264_RC_CRF
985           || h->param.i_bframe_adaptive
986           || h->param.i_scenecut_threshold
987           || h->param.rc.b_mb_tree
988           || h->param.analyse.i_weighted_pred == X264_WEIGHTP_SMART );
989     h->frames.b_have_lowres |= h->param.rc.b_stat_read && h->param.rc.i_vbv_buffer_size > 0;
990     h->frames.b_have_sub8x8_esa = !!(h->param.analyse.inter & X264_ANALYSE_PSUB8x8);
991
992     h->frames.i_last_idr =
993     h->frames.i_last_keyframe = - h->param.i_keyint_max;
994     h->frames.i_input    = 0;
995     h->frames.i_largest_pts = h->frames.i_second_largest_pts = -1;
996     h->frames.i_poc_last_open_gop = -1;
997
998     CHECKED_MALLOCZERO( h->frames.unused[0], (h->frames.i_delay + 3) * sizeof(x264_frame_t *) );
999     /* Allocate room for max refs plus a few extra just in case. */
1000     CHECKED_MALLOCZERO( h->frames.unused[1], (h->i_thread_frames + 20) * sizeof(x264_frame_t *) );
1001     CHECKED_MALLOCZERO( h->frames.current, (h->param.i_sync_lookahead + h->param.i_bframe
1002                         + h->i_thread_frames + 3) * sizeof(x264_frame_t *) );
1003     if( h->param.analyse.i_weighted_pred > 0 )
1004         CHECKED_MALLOCZERO( h->frames.blank_unused, h->i_thread_frames * 4 * sizeof(x264_frame_t *) );
1005     h->i_ref0 = 0;
1006     h->i_ref1 = 0;
1007     h->i_cpb_delay = h->i_coded_fields = h->i_disp_fields = h->i_prev_duration = 0;
1008     h->i_disp_fields_last_frame = -1;
1009     x264_rdo_init();
1010
1011     /* init CPU functions */
1012     x264_predict_16x16_init( h->param.cpu, h->predict_16x16 );
1013     x264_predict_8x8c_init( h->param.cpu, h->predict_8x8c );
1014     x264_predict_8x8_init( h->param.cpu, h->predict_8x8, &h->predict_8x8_filter );
1015     x264_predict_4x4_init( h->param.cpu, h->predict_4x4 );
1016     if( !h->param.b_cabac )
1017         x264_init_vlc_tables();
1018     x264_pixel_init( h->param.cpu, &h->pixf );
1019     x264_dct_init( h->param.cpu, &h->dctf );
1020     x264_zigzag_init( h->param.cpu, &h->zigzagf, h->param.b_interlaced );
1021     x264_mc_init( h->param.cpu, &h->mc );
1022     x264_quant_init( h, h->param.cpu, &h->quantf );
1023     x264_deblock_init( h->param.cpu, &h->loopf );
1024     x264_bitstream_init( h->param.cpu, &h->bsf );
1025     x264_dct_init_weights();
1026
1027     mbcmp_init( h );
1028
1029     p = buf + sprintf( buf, "using cpu capabilities:" );
1030     for( int i = 0; x264_cpu_names[i].flags; i++ )
1031     {
1032         if( !strcmp(x264_cpu_names[i].name, "SSE2")
1033             && h->param.cpu & (X264_CPU_SSE2_IS_FAST|X264_CPU_SSE2_IS_SLOW) )
1034             continue;
1035         if( !strcmp(x264_cpu_names[i].name, "SSE3")
1036             && (h->param.cpu & X264_CPU_SSSE3 || !(h->param.cpu & X264_CPU_CACHELINE_64)) )
1037             continue;
1038         if( !strcmp(x264_cpu_names[i].name, "SSE4.1")
1039             && (h->param.cpu & X264_CPU_SSE42) )
1040             continue;
1041         if( (h->param.cpu & x264_cpu_names[i].flags) == x264_cpu_names[i].flags
1042             && (!i || x264_cpu_names[i].flags != x264_cpu_names[i-1].flags) )
1043             p += sprintf( p, " %s", x264_cpu_names[i].name );
1044     }
1045     if( !h->param.cpu )
1046         p += sprintf( p, " none!" );
1047     x264_log( h, X264_LOG_INFO, "%s\n", buf );
1048
1049     for( qp = h->param.rc.i_qp_min; qp <= h->param.rc.i_qp_max; qp++ )
1050         if( x264_analyse_init_costs( h, qp ) )
1051             goto fail;
1052     if( x264_analyse_init_costs( h, X264_LOOKAHEAD_QP ) )
1053         goto fail;
1054
1055     /* Checks for known miscompilation issues. */
1056     if( h->cost_mv[1][2013] != 24 )
1057     {
1058         x264_log( h, X264_LOG_ERROR, "MV cost test failed: x264 has been miscompiled!\n" );
1059         goto fail;
1060     }
1061
1062     /* Must be volatile or else GCC will optimize it out. */
1063     volatile int temp = 392;
1064     if( x264_clz( temp ) != 23 )
1065     {
1066         x264_log( h, X264_LOG_ERROR, "CLZ test failed: x264 has been miscompiled!\n" );
1067 #if ARCH_X86 || ARCH_X86_64
1068         x264_log( h, X264_LOG_ERROR, "Are you attempting to run an SSE4a-targeted build on a CPU that\n" );
1069         x264_log( h, X264_LOG_ERROR, "doesn't support it?\n" );
1070 #endif
1071         goto fail;
1072     }
1073
1074     h->out.i_nal = 0;
1075     h->out.i_bitstream = X264_MAX( 1000000, h->param.i_width * h->param.i_height * 4
1076         * ( h->param.rc.i_rc_method == X264_RC_ABR ? pow( 0.95, h->param.rc.i_qp_min )
1077           : pow( 0.95, h->param.rc.i_qp_constant ) * X264_MAX( 1, h->param.rc.f_ip_factor )));
1078
1079     CHECKED_MALLOC( h->nal_buffer, h->out.i_bitstream * 3/2 + 4 );
1080     h->nal_buffer_size = h->out.i_bitstream * 3/2 + 4;
1081
1082     if( h->param.i_threads > 1 &&
1083         x264_threadpool_init( &h->threadpool, h->param.i_threads, (void*)x264_encoder_thread_init, h ) )
1084         goto fail;
1085
1086     h->thread[0] = h;
1087     for( int i = 1; i < h->param.i_threads + !!h->param.i_sync_lookahead; i++ )
1088         CHECKED_MALLOC( h->thread[i], sizeof(x264_t) );
1089
1090     if( x264_lookahead_init( h, i_slicetype_length ) )
1091         goto fail;
1092
1093     for( int i = 0; i < h->param.i_threads; i++ )
1094     {
1095         int init_nal_count = h->param.i_slice_count + 3;
1096         int allocate_threadlocal_data = !h->param.b_sliced_threads || !i;
1097         if( i > 0 )
1098             *h->thread[i] = *h;
1099
1100         if( allocate_threadlocal_data )
1101         {
1102             h->thread[i]->fdec = x264_frame_pop_unused( h, 1 );
1103             if( !h->thread[i]->fdec )
1104                 goto fail;
1105         }
1106         else
1107             h->thread[i]->fdec = h->thread[0]->fdec;
1108
1109         CHECKED_MALLOC( h->thread[i]->out.p_bitstream, h->out.i_bitstream );
1110         /* Start each thread with room for init_nal_count NAL units; it'll realloc later if needed. */
1111         CHECKED_MALLOC( h->thread[i]->out.nal, init_nal_count*sizeof(x264_nal_t) );
1112         h->thread[i]->out.i_nals_allocated = init_nal_count;
1113
1114         if( allocate_threadlocal_data && x264_macroblock_cache_allocate( h->thread[i] ) < 0 )
1115             goto fail;
1116     }
1117
1118     for( int i = 0; i < h->param.i_threads; i++ )
1119         if( x264_macroblock_thread_allocate( h->thread[i], 0 ) < 0 )
1120             goto fail;
1121
1122     if( x264_ratecontrol_new( h ) < 0 )
1123         goto fail;
1124
1125     if( h->param.i_nal_hrd )
1126     {
1127         x264_log( h, X264_LOG_DEBUG, "HRD bitrate: %i bits/sec\n", h->sps->vui.hrd.i_bit_rate_unscaled );
1128         x264_log( h, X264_LOG_DEBUG, "CPB size: %i bits\n", h->sps->vui.hrd.i_cpb_size_unscaled );
1129     }
1130
1131     if( h->param.psz_dump_yuv )
1132     {
1133         /* create or truncate the reconstructed video file */
1134         FILE *f = fopen( h->param.psz_dump_yuv, "w" );
1135         if( !f )
1136         {
1137             x264_log( h, X264_LOG_ERROR, "dump_yuv: can't write to %s\n", h->param.psz_dump_yuv );
1138             goto fail;
1139         }
1140         else if( !x264_is_regular_file( f ) )
1141         {
1142             x264_log( h, X264_LOG_ERROR, "dump_yuv: incompatible with non-regular file %s\n", h->param.psz_dump_yuv );
1143             goto fail;
1144         }
1145         fclose( f );
1146     }
1147
1148     x264_log( h, X264_LOG_INFO, "profile %s, level %d.%d\n",
1149         h->sps->i_profile_idc == PROFILE_BASELINE ? "Baseline" :
1150         h->sps->i_profile_idc == PROFILE_MAIN ? "Main" :
1151         h->sps->i_profile_idc == PROFILE_HIGH ? "High" :
1152         "High 4:4:4 Predictive", h->sps->i_level_idc/10, h->sps->i_level_idc%10 );
1153
1154     return h;
1155 fail:
1156     x264_free( h );
1157     return NULL;
1158 }
1159
1160 /****************************************************************************
1161  * x264_encoder_reconfig:
1162  ****************************************************************************/
1163 int x264_encoder_reconfig( x264_t *h, x264_param_t *param )
1164 {
1165     int rc_reconfig = 0;
1166     h = h->thread[h->thread[0]->i_thread_phase];
1167     x264_set_aspect_ratio( h, param, 0 );
1168 #define COPY(var) h->param.var = param->var
1169     COPY( i_frame_reference ); // but never uses more refs than initially specified
1170     COPY( i_bframe_bias );
1171     if( h->param.i_scenecut_threshold )
1172         COPY( i_scenecut_threshold ); // can't turn it on or off, only vary the threshold
1173     COPY( b_deblocking_filter );
1174     COPY( i_deblocking_filter_alphac0 );
1175     COPY( i_deblocking_filter_beta );
1176     COPY( analyse.inter );
1177     COPY( analyse.intra );
1178     COPY( analyse.i_direct_mv_pred );
1179     /* Scratch buffer prevents me_range from being increased for esa/tesa */
1180     if( h->param.analyse.i_me_method < X264_ME_ESA || param->analyse.i_me_range < h->param.analyse.i_me_range )
1181         COPY( analyse.i_me_range );
1182     COPY( analyse.i_noise_reduction );
1183     /* We can't switch out of subme=0 during encoding. */
1184     if( h->param.analyse.i_subpel_refine )
1185         COPY( analyse.i_subpel_refine );
1186     COPY( analyse.i_trellis );
1187     COPY( analyse.b_chroma_me );
1188     COPY( analyse.b_dct_decimate );
1189     COPY( analyse.b_fast_pskip );
1190     COPY( analyse.b_mixed_references );
1191     COPY( analyse.f_psy_rd );
1192     COPY( analyse.f_psy_trellis );
1193     // can only twiddle these if they were enabled to begin with:
1194     if( h->param.analyse.i_me_method >= X264_ME_ESA || param->analyse.i_me_method < X264_ME_ESA )
1195         COPY( analyse.i_me_method );
1196     if( h->param.analyse.i_me_method >= X264_ME_ESA && !h->frames.b_have_sub8x8_esa )
1197         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
1198     if( h->pps->b_transform_8x8_mode )
1199         COPY( analyse.b_transform_8x8 );
1200     if( h->frames.i_max_ref1 > 1 )
1201         COPY( i_bframe_pyramid );
1202     COPY( i_slice_max_size );
1203     COPY( i_slice_max_mbs );
1204     COPY( i_slice_count );
1205
1206     /* VBV can't be turned on if it wasn't on to begin with */
1207     if( h->param.rc.i_vbv_max_bitrate > 0 && h->param.rc.i_vbv_buffer_size > 0 &&
1208           param->rc.i_vbv_max_bitrate > 0 &&   param->rc.i_vbv_buffer_size > 0 )
1209     {
1210         COPY( rc.i_vbv_max_bitrate );
1211         COPY( rc.i_vbv_buffer_size );
1212         COPY( rc.i_bitrate );
1213         rc_reconfig = 1;
1214     }
1215     if( h->param.rc.f_rf_constant != param->rc.f_rf_constant )
1216     {
1217         COPY( rc.f_rf_constant );
1218         rc_reconfig = 1;
1219     }
1220     if( h->param.rc.f_rf_constant_max != param->rc.f_rf_constant_max )
1221     {
1222         COPY( rc.f_rf_constant_max );
1223         rc_reconfig = 1;
1224     }
1225
1226 #undef COPY
1227
1228     mbcmp_init( h );
1229
1230     int ret = x264_validate_parameters( h );
1231
1232     /* Supported reconfiguration options (1-pass only):
1233      * vbv-maxrate
1234      * vbv-bufsize
1235      * crf
1236      * bitrate (CBR only) */
1237     if( !ret && rc_reconfig )
1238         x264_ratecontrol_init_reconfigurable( h, 0 );
1239
1240     return ret;
1241 }
1242
1243 /****************************************************************************
1244  * x264_encoder_parameters:
1245  ****************************************************************************/
1246 void x264_encoder_parameters( x264_t *h, x264_param_t *param )
1247 {
1248     memcpy( param, &h->thread[h->i_thread_phase]->param, sizeof(x264_param_t) );
1249 }
1250
1251 /* internal usage */
1252 static void x264_nal_start( x264_t *h, int i_type, int i_ref_idc )
1253 {
1254     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1255
1256     nal->i_ref_idc = i_ref_idc;
1257     nal->i_type    = i_type;
1258
1259     nal->i_payload= 0;
1260     nal->p_payload= &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8];
1261 }
1262
1263 /* if number of allocated nals is not enough, re-allocate a larger one. */
1264 static int x264_nal_check_buffer( x264_t *h )
1265 {
1266     if( h->out.i_nal >= h->out.i_nals_allocated )
1267     {
1268         x264_nal_t *new_out = x264_malloc( sizeof(x264_nal_t) * (h->out.i_nals_allocated*2) );
1269         if( !new_out )
1270             return -1;
1271         memcpy( new_out, h->out.nal, sizeof(x264_nal_t) * (h->out.i_nals_allocated) );
1272         x264_free( h->out.nal );
1273         h->out.nal = new_out;
1274         h->out.i_nals_allocated *= 2;
1275     }
1276     return 0;
1277 }
1278
1279 static int x264_nal_end( x264_t *h )
1280 {
1281     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1282     nal->i_payload = &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8] - nal->p_payload;
1283     h->out.i_nal++;
1284
1285     return x264_nal_check_buffer( h );
1286 }
1287
1288 static int x264_encoder_encapsulate_nals( x264_t *h, int start )
1289 {
1290     int nal_size = 0, previous_nal_size = 0;
1291
1292     for( int i = 0; i < start; i++ )
1293         previous_nal_size += h->out.nal[i].i_payload;
1294
1295     for( int i = start; i < h->out.i_nal; i++ )
1296         nal_size += h->out.nal[i].i_payload;
1297
1298     /* Worst-case NAL unit escaping: reallocate the buffer if it's too small. */
1299     if( h->nal_buffer_size < nal_size * 3/2 + h->out.i_nal * 4 )
1300     {
1301         uint8_t *buf = x264_malloc( nal_size * 2 + h->out.i_nal * 4 );
1302         if( !buf )
1303             return -1;
1304         if( previous_nal_size )
1305             memcpy( buf, h->nal_buffer, previous_nal_size );
1306         x264_free( h->nal_buffer );
1307         h->nal_buffer = buf;
1308     }
1309
1310     uint8_t *nal_buffer = h->nal_buffer + previous_nal_size;
1311
1312     for( int i = start; i < h->out.i_nal; i++ )
1313     {
1314         int long_startcode = !i || h->out.nal[i].i_type == NAL_SPS || h->out.nal[i].i_type == NAL_PPS;
1315         int size = x264_nal_encode( h, nal_buffer, &h->out.nal[i], long_startcode );
1316         h->out.nal[i].i_payload = size;
1317         h->out.nal[i].p_payload = nal_buffer;
1318         nal_buffer += size;
1319     }
1320
1321     x264_emms();
1322
1323     return nal_buffer - (h->nal_buffer + previous_nal_size);
1324 }
1325
1326 /****************************************************************************
1327  * x264_encoder_headers:
1328  ****************************************************************************/
1329 int x264_encoder_headers( x264_t *h, x264_nal_t **pp_nal, int *pi_nal )
1330 {
1331     int frame_size = 0;
1332     /* init bitstream context */
1333     h->out.i_nal = 0;
1334     bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
1335
1336     /* Write SEI, SPS and PPS. */
1337
1338     /* generate sequence parameters */
1339     x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
1340     x264_sps_write( &h->out.bs, h->sps );
1341     if( x264_nal_end( h ) )
1342         return -1;
1343
1344     /* generate picture parameters */
1345     x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
1346     x264_pps_write( &h->out.bs, h->pps );
1347     if( x264_nal_end( h ) )
1348         return -1;
1349
1350     /* identify ourselves */
1351     x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
1352     if( x264_sei_version_write( h, &h->out.bs ) )
1353         return -1;
1354     if( x264_nal_end( h ) )
1355         return -1;
1356
1357     frame_size = x264_encoder_encapsulate_nals( h, 0 );
1358
1359     /* now set output*/
1360     *pi_nal = h->out.i_nal;
1361     *pp_nal = &h->out.nal[0];
1362     h->out.i_nal = 0;
1363
1364     return frame_size;
1365 }
1366
1367 /* Check to see whether we have chosen a reference list ordering different
1368  * from the standard's default. */
1369 static inline void x264_reference_check_reorder( x264_t *h )
1370 {
1371     for( int i = 0; i < h->i_ref0 - 1; i++ )
1372         /* P and B-frames use different default orders. */
1373         if( h->sh.i_type == SLICE_TYPE_P ? h->fref0[i]->i_frame_num < h->fref0[i+1]->i_frame_num
1374                                          : h->fref0[i]->i_poc < h->fref0[i+1]->i_poc )
1375         {
1376             h->b_ref_reorder[0] = 1;
1377             break;
1378         }
1379 }
1380
1381 /* return -1 on failure, else return the index of the new reference frame */
1382 int x264_weighted_reference_duplicate( x264_t *h, int i_ref, const x264_weight_t *w )
1383 {
1384     int i = h->i_ref0;
1385     int j = 1;
1386     x264_frame_t *newframe;
1387     if( i <= 1 ) /* empty list, definitely can't duplicate frame */
1388         return -1;
1389
1390     newframe = x264_frame_pop_blank_unused( h );
1391
1392     //FIXME: probably don't need to copy everything
1393     *newframe = *h->fref0[i_ref];
1394     newframe->i_reference_count = 1;
1395     newframe->orig = h->fref0[i_ref];
1396     newframe->b_duplicate = 1;
1397     memcpy( h->fenc->weight[j], w, sizeof(h->fenc->weight[i]) );
1398
1399     /* shift the frames to make space for the dupe. */
1400     h->b_ref_reorder[0] = 1;
1401     if( h->i_ref0 < 16 )
1402         ++h->i_ref0;
1403     h->fref0[15] = NULL;
1404     x264_frame_unshift( &h->fref0[j], newframe );
1405
1406     return j;
1407 }
1408
1409 static void x264_weighted_pred_init( x264_t *h )
1410 {
1411     /* for now no analysis and set all weights to nothing */
1412     for( int i_ref = 0; i_ref < h->i_ref0; i_ref++ )
1413         h->fenc->weighted[i_ref] = h->fref0[i_ref]->filtered[0];
1414
1415     // FIXME: This only supports weighting of one reference frame
1416     // and duplicates of that frame.
1417     h->fenc->i_lines_weighted = 0;
1418
1419     for( int i_ref = 0; i_ref < (h->i_ref0 << h->sh.b_mbaff); i_ref++ )
1420         for( int i = 0; i < 3; i++ )
1421             h->sh.weight[i_ref][i].weightfn = NULL;
1422
1423
1424     if( h->sh.i_type != SLICE_TYPE_P || h->param.analyse.i_weighted_pred <= 0 )
1425         return;
1426
1427     int i_padv = PADV << h->param.b_interlaced;
1428     int denom = -1;
1429     int weightluma = 0;
1430     int buffer_next = 0;
1431     //FIXME: when chroma support is added, move this into loop
1432     h->sh.weight[0][1].weightfn = h->sh.weight[0][2].weightfn = NULL;
1433     h->sh.weight[0][1].i_denom = h->sh.weight[0][2].i_denom = 0;
1434     for( int j = 0; j < h->i_ref0; j++ )
1435     {
1436         if( h->fenc->weight[j][0].weightfn )
1437         {
1438             h->sh.weight[j][0] = h->fenc->weight[j][0];
1439             // if weight is useless, don't write it to stream
1440             if( h->sh.weight[j][0].i_scale == 1<<h->sh.weight[j][0].i_denom && h->sh.weight[j][0].i_offset == 0 )
1441                 h->sh.weight[j][0].weightfn = NULL;
1442             else
1443             {
1444                 if( !weightluma )
1445                 {
1446                     weightluma = 1;
1447                     h->sh.weight[0][0].i_denom = denom = h->sh.weight[j][0].i_denom;
1448                     assert( x264_clip3( denom, 0, 7 ) == denom );
1449                 }
1450                 assert( h->sh.weight[j][0].i_denom == denom );
1451                 assert( x264_clip3( h->sh.weight[j][0].i_scale, 0, 127 ) == h->sh.weight[j][0].i_scale );
1452                 assert( x264_clip3( h->sh.weight[j][0].i_offset, -128, 127 ) == h->sh.weight[j][0].i_offset );
1453                 h->fenc->weighted[j] = h->mb.p_weight_buf[buffer_next++] +
1454                     h->fenc->i_stride[0] * i_padv + PADH;
1455             }
1456         }
1457
1458         //scale full resolution frame
1459         if( h->sh.weight[j][0].weightfn && h->param.i_threads == 1 )
1460         {
1461             pixel *src = h->fref0[j]->filtered[0] - h->fref0[j]->i_stride[0]*i_padv - PADH;
1462             pixel *dst = h->fenc->weighted[j] - h->fenc->i_stride[0]*i_padv - PADH;
1463             int stride = h->fenc->i_stride[0];
1464             int width = h->fenc->i_width[0] + PADH*2;
1465             int height = h->fenc->i_lines[0] + i_padv*2;
1466             x264_weight_scale_plane( h, dst, stride, src, stride, width, height, &h->sh.weight[j][0] );
1467             h->fenc->i_lines_weighted = height;
1468         }
1469     }
1470     if( !weightluma )
1471         h->sh.weight[0][0].i_denom = 0;
1472 }
1473
1474 static inline void x264_reference_build_list( x264_t *h, int i_poc )
1475 {
1476     int b_ok;
1477
1478     /* build ref list 0/1 */
1479     h->mb.pic.i_fref[0] = h->i_ref0 = 0;
1480     h->mb.pic.i_fref[1] = h->i_ref1 = 0;
1481     if( h->sh.i_type == SLICE_TYPE_I )
1482         return;
1483
1484     for( int i = 0; h->frames.reference[i]; i++ )
1485     {
1486         if( h->frames.reference[i]->b_corrupt )
1487             continue;
1488         if( h->frames.reference[i]->i_poc < i_poc )
1489             h->fref0[h->i_ref0++] = h->frames.reference[i];
1490         else if( h->frames.reference[i]->i_poc > i_poc )
1491             h->fref1[h->i_ref1++] = h->frames.reference[i];
1492     }
1493
1494     /* Order ref0 from higher to lower poc */
1495     do
1496     {
1497         b_ok = 1;
1498         for( int i = 0; i < h->i_ref0 - 1; i++ )
1499         {
1500             if( h->fref0[i]->i_poc < h->fref0[i+1]->i_poc )
1501             {
1502                 XCHG( x264_frame_t*, h->fref0[i], h->fref0[i+1] );
1503                 b_ok = 0;
1504                 break;
1505             }
1506         }
1507     } while( !b_ok );
1508
1509     if( h->sh.i_mmco_remove_from_end )
1510         for( int i = h->i_ref0-1; i >= h->i_ref0 - h->sh.i_mmco_remove_from_end; i-- )
1511         {
1512             int diff = h->i_frame_num - h->fref0[i]->i_frame_num;
1513             h->sh.mmco[h->sh.i_mmco_command_count].i_poc = h->fref0[i]->i_poc;
1514             h->sh.mmco[h->sh.i_mmco_command_count++].i_difference_of_pic_nums = diff;
1515         }
1516
1517     /* Order ref1 from lower to higher poc (bubble sort) for B-frame */
1518     do
1519     {
1520         b_ok = 1;
1521         for( int i = 0; i < h->i_ref1 - 1; i++ )
1522         {
1523             if( h->fref1[i]->i_poc > h->fref1[i+1]->i_poc )
1524             {
1525                 XCHG( x264_frame_t*, h->fref1[i], h->fref1[i+1] );
1526                 b_ok = 0;
1527                 break;
1528             }
1529         }
1530     } while( !b_ok );
1531
1532     x264_reference_check_reorder( h );
1533
1534     h->i_ref1 = X264_MIN( h->i_ref1, h->frames.i_max_ref1 );
1535     h->i_ref0 = X264_MIN( h->i_ref0, h->frames.i_max_ref0 );
1536     h->i_ref0 = X264_MIN( h->i_ref0, h->param.i_frame_reference ); // if reconfig() has lowered the limit
1537
1538     /* add duplicates */
1539     if( h->fenc->i_type == X264_TYPE_P )
1540     {
1541         int idx = -1;
1542         if( h->param.analyse.i_weighted_pred == X264_WEIGHTP_SMART )
1543         {
1544             x264_weight_t w[3];
1545             w[1].weightfn = w[2].weightfn = NULL;
1546             if( h->param.rc.b_stat_read )
1547                 x264_ratecontrol_set_weights( h, h->fenc );
1548
1549             if( !h->fenc->weight[0][0].weightfn )
1550             {
1551                 h->fenc->weight[0][0].i_denom = 0;
1552                 SET_WEIGHT( w[0], 1, 1, 0, -1 );
1553                 idx = x264_weighted_reference_duplicate( h, 0, w );
1554             }
1555             else
1556             {
1557                 if( h->fenc->weight[0][0].i_scale == 1<<h->fenc->weight[0][0].i_denom )
1558                 {
1559                     SET_WEIGHT( h->fenc->weight[0][0], 1, 1, 0, h->fenc->weight[0][0].i_offset );
1560                 }
1561                 x264_weighted_reference_duplicate( h, 0, weight_none );
1562                 if( h->fenc->weight[0][0].i_offset > -128 )
1563                 {
1564                     w[0] = h->fenc->weight[0][0];
1565                     w[0].i_offset--;
1566                     h->mc.weight_cache( h, &w[0] );
1567                     idx = x264_weighted_reference_duplicate( h, 0, w );
1568                 }
1569             }
1570         }
1571         else if( h->param.analyse.i_weighted_pred == X264_WEIGHTP_BLIND )
1572         {
1573             //weighted offset=-1
1574             x264_weight_t w[3];
1575             SET_WEIGHT( w[0], 1, 1, 0, -1 );
1576             h->fenc->weight[0][0].i_denom = 0;
1577             w[1].weightfn = w[2].weightfn = NULL;
1578             idx = x264_weighted_reference_duplicate( h, 0, w );
1579         }
1580         h->mb.ref_blind_dupe = idx;
1581     }
1582
1583     assert( h->i_ref0 + h->i_ref1 <= 16 );
1584     h->mb.pic.i_fref[0] = h->i_ref0;
1585     h->mb.pic.i_fref[1] = h->i_ref1;
1586 }
1587
1588 static void x264_fdec_filter_row( x264_t *h, int mb_y, int b_inloop )
1589 {
1590     /* mb_y is the mb to be encoded next, not the mb to be filtered here */
1591     int b_hpel = h->fdec->b_kept_as_ref;
1592     int b_deblock = h->sh.i_disable_deblocking_filter_idc != 1;
1593     int b_end = mb_y == h->i_threadslice_end;
1594     int b_measure_quality = 1;
1595     int min_y = mb_y - (1 << h->sh.b_mbaff);
1596     int b_start = min_y == h->i_threadslice_start;
1597     int max_y = b_end ? h->i_threadslice_end : mb_y;
1598     b_deblock &= b_hpel || h->param.psz_dump_yuv;
1599     if( h->param.b_sliced_threads && b_start && min_y && !b_inloop )
1600     {
1601         b_deblock = 0;         /* We already deblocked on the inloop pass. */
1602         b_measure_quality = 0; /* We already measured quality on the inloop pass. */
1603     }
1604     if( mb_y & h->sh.b_mbaff )
1605         return;
1606     if( min_y < h->i_threadslice_start )
1607         return;
1608
1609     if( !b_end && b_inloop )
1610         for( int j = 0; j <= h->sh.b_mbaff; j++ )
1611             for( int i = 0; i < 3; i++ )
1612             {
1613                 memcpy( h->intra_border_backup[j][i],
1614                         h->fdec->plane[i] + ((mb_y*16 >> !!i) + j - 1 - h->sh.b_mbaff) * h->fdec->i_stride[i],
1615                         (h->mb.i_mb_width*16 >> !!i) * sizeof(pixel) );
1616             }
1617
1618     if( b_deblock )
1619         for( int y = min_y; y < max_y; y += (1 << h->sh.b_mbaff) )
1620             x264_frame_deblock_row( h, y );
1621
1622     if( b_hpel )
1623     {
1624         int end = mb_y == h->mb.i_mb_height;
1625         x264_frame_expand_border( h, h->fdec, min_y, end );
1626         if( h->param.analyse.i_subpel_refine )
1627         {
1628             x264_frame_filter( h, h->fdec, min_y, end );
1629             x264_frame_expand_border_filtered( h, h->fdec, min_y, end );
1630         }
1631     }
1632
1633     if( h->i_thread_frames > 1 && h->fdec->b_kept_as_ref )
1634         x264_frame_cond_broadcast( h->fdec, mb_y*16 + (b_end ? 10000 : -(X264_THREAD_HEIGHT << h->sh.b_mbaff)) );
1635
1636     min_y = min_y*16 - 8 * !b_start;
1637     max_y = b_end ? X264_MIN( h->i_threadslice_end*16 , h->param.i_height ) : mb_y*16 - 8;
1638
1639     if( b_measure_quality )
1640     {
1641         if( h->param.analyse.b_psnr )
1642             for( int i = 0; i < 3; i++ )
1643                 h->stat.frame.i_ssd[i] +=
1644                     x264_pixel_ssd_wxh( &h->pixf,
1645                         h->fdec->plane[i] + (min_y>>!!i) * h->fdec->i_stride[i], h->fdec->i_stride[i],
1646                         h->fenc->plane[i] + (min_y>>!!i) * h->fenc->i_stride[i], h->fenc->i_stride[i],
1647                         h->param.i_width >> !!i, (max_y-min_y) >> !!i );
1648
1649         if( h->param.analyse.b_ssim )
1650         {
1651             x264_emms();
1652             /* offset by 2 pixels to avoid alignment of ssim blocks with dct blocks,
1653              * and overlap by 4 */
1654             min_y += b_start ? 2 : -6;
1655             h->stat.frame.f_ssim +=
1656                 x264_pixel_ssim_wxh( &h->pixf,
1657                     h->fdec->plane[0] + 2+min_y*h->fdec->i_stride[0], h->fdec->i_stride[0],
1658                     h->fenc->plane[0] + 2+min_y*h->fenc->i_stride[0], h->fenc->i_stride[0],
1659                     h->param.i_width-2, max_y-min_y, h->scratch_buffer );
1660         }
1661     }
1662 }
1663
1664 static inline int x264_reference_update( x264_t *h )
1665 {
1666     if( !h->fdec->b_kept_as_ref )
1667     {
1668         if( h->i_thread_frames > 1 )
1669         {
1670             x264_frame_push_unused( h, h->fdec );
1671             h->fdec = x264_frame_pop_unused( h, 1 );
1672             if( !h->fdec )
1673                 return -1;
1674         }
1675         return 0;
1676     }
1677
1678     /* apply mmco from previous frame. */
1679     for( int i = 0; i < h->sh.i_mmco_command_count; i++ )
1680         for( int j = 0; h->frames.reference[j]; j++ )
1681             if( h->frames.reference[j]->i_poc == h->sh.mmco[i].i_poc )
1682                 x264_frame_push_unused( h, x264_frame_shift( &h->frames.reference[j] ) );
1683
1684     /* move frame in the buffer */
1685     x264_frame_push( h->frames.reference, h->fdec );
1686     if( h->frames.reference[h->sps->i_num_ref_frames] )
1687         x264_frame_push_unused( h, x264_frame_shift( h->frames.reference ) );
1688     h->fdec = x264_frame_pop_unused( h, 1 );
1689     if( !h->fdec )
1690         return -1;
1691     return 0;
1692 }
1693
1694 static inline void x264_reference_reset( x264_t *h )
1695 {
1696     while( h->frames.reference[0] )
1697         x264_frame_push_unused( h, x264_frame_pop( h->frames.reference ) );
1698     h->fdec->i_poc =
1699     h->fenc->i_poc = 0;
1700 }
1701
1702 static inline void x264_reference_hierarchy_reset( x264_t *h )
1703 {
1704     int ref;
1705     int b_hasdelayframe = 0;
1706
1707     /* look for delay frames -- chain must only contain frames that are disposable */
1708     for( int i = 0; h->frames.current[i] && IS_DISPOSABLE( h->frames.current[i]->i_type ); i++ )
1709         b_hasdelayframe |= h->frames.current[i]->i_coded
1710                         != h->frames.current[i]->i_frame + h->sps->vui.i_num_reorder_frames;
1711
1712     /* This function must handle b-pyramid and clear frames for open-gop */
1713     if( h->param.i_bframe_pyramid != X264_B_PYRAMID_STRICT && !b_hasdelayframe && h->frames.i_poc_last_open_gop == -1 )
1714         return;
1715
1716     /* Remove last BREF. There will never be old BREFs in the
1717      * dpb during a BREF decode when pyramid == STRICT */
1718     for( ref = 0; h->frames.reference[ref]; ref++ )
1719     {
1720         if( ( h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT
1721             && h->frames.reference[ref]->i_type == X264_TYPE_BREF )
1722             || ( h->frames.reference[ref]->i_poc < h->frames.i_poc_last_open_gop
1723             && h->sh.i_type != SLICE_TYPE_B ) )
1724         {
1725             int diff = h->i_frame_num - h->frames.reference[ref]->i_frame_num;
1726             h->sh.mmco[h->sh.i_mmco_command_count].i_difference_of_pic_nums = diff;
1727             h->sh.mmco[h->sh.i_mmco_command_count++].i_poc = h->frames.reference[ref]->i_poc;
1728             x264_frame_push_unused( h, x264_frame_shift( &h->frames.reference[ref] ) );
1729             h->b_ref_reorder[0] = 1;
1730             ref--;
1731         }
1732     }
1733
1734     /* Prepare room in the dpb for the delayed display time of the later b-frame's */
1735     if( h->param.i_bframe_pyramid )
1736         h->sh.i_mmco_remove_from_end = X264_MAX( ref + 2 - h->frames.i_max_dpb, 0 );
1737 }
1738
1739 static inline void x264_slice_init( x264_t *h, int i_nal_type, int i_global_qp )
1740 {
1741     /* ------------------------ Create slice header  ----------------------- */
1742     if( i_nal_type == NAL_SLICE_IDR )
1743     {
1744         x264_slice_header_init( h, &h->sh, h->sps, h->pps, h->i_idr_pic_id, h->i_frame_num, i_global_qp );
1745
1746         /* increment id */
1747         h->i_idr_pic_id = ( h->i_idr_pic_id + 1 ) % 65536;
1748     }
1749     else
1750     {
1751         x264_slice_header_init( h, &h->sh, h->sps, h->pps, -1, h->i_frame_num, i_global_qp );
1752
1753         h->sh.i_num_ref_idx_l0_active = h->i_ref0 <= 0 ? 1 : h->i_ref0;
1754         h->sh.i_num_ref_idx_l1_active = h->i_ref1 <= 0 ? 1 : h->i_ref1;
1755         if( h->sh.i_num_ref_idx_l0_active != h->pps->i_num_ref_idx_l0_default_active ||
1756             (h->sh.i_type == SLICE_TYPE_B && h->sh.i_num_ref_idx_l1_active != h->pps->i_num_ref_idx_l1_default_active) )
1757         {
1758             h->sh.b_num_ref_idx_override = 1;
1759         }
1760     }
1761
1762     h->fdec->i_frame_num = h->sh.i_frame_num;
1763
1764     if( h->sps->i_poc_type == 0 )
1765     {
1766         h->sh.i_poc = h->fdec->i_poc;
1767         if( h->param.b_interlaced )
1768         {
1769             h->sh.i_delta_poc_bottom = h->param.b_tff ? 1 : -1;
1770             if( h->sh.i_delta_poc_bottom == -1 )
1771                 h->sh.i_poc = h->fdec->i_poc + 1;
1772         }
1773         else
1774             h->sh.i_delta_poc_bottom = 0;
1775     }
1776     else if( h->sps->i_poc_type == 1 )
1777     {
1778         /* FIXME TODO FIXME */
1779     }
1780     else
1781     {
1782         /* Nothing to do ? */
1783     }
1784
1785     x264_macroblock_slice_init( h );
1786 }
1787
1788 static int x264_slice_write( x264_t *h )
1789 {
1790     int i_skip;
1791     int mb_xy, i_mb_x, i_mb_y;
1792     int i_skip_bak = 0; /* Shut up GCC. */
1793     bs_t bs_bak;
1794     x264_cabac_t cabac_bak;
1795     uint8_t cabac_prevbyte_bak = 0; /* Shut up GCC. */
1796     /* Assume no more than 3 bytes of NALU escaping.
1797      * NALUs other than the first use a 3-byte startcode. */
1798     int overhead_guess = (NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal)) + 3;
1799     int slice_max_size = h->param.i_slice_max_size > 0 ? (h->param.i_slice_max_size-overhead_guess)*8 : INT_MAX;
1800     int starting_bits = bs_pos(&h->out.bs);
1801     int b_deblock = h->sh.i_disable_deblocking_filter_idc != 1;
1802     int b_hpel = h->fdec->b_kept_as_ref;
1803     b_deblock &= b_hpel || h->param.psz_dump_yuv;
1804     bs_realign( &h->out.bs );
1805
1806     /* Slice */
1807     x264_nal_start( h, h->i_nal_type, h->i_nal_ref_idc );
1808
1809     /* Slice header */
1810     x264_macroblock_thread_init( h );
1811
1812     /* If this isn't the first slice in the threadslice, set the slice QP
1813      * equal to the last QP in the previous slice for more accurate
1814      * CABAC initialization. */
1815     if( h->sh.i_first_mb != h->i_threadslice_start * h->mb.i_mb_width )
1816     {
1817         h->sh.i_qp = h->mb.i_last_qp;
1818         h->sh.i_qp_delta = h->sh.i_qp - h->pps->i_pic_init_qp;
1819     }
1820
1821     x264_slice_header_write( &h->out.bs, &h->sh, h->i_nal_ref_idc );
1822     if( h->param.b_cabac )
1823     {
1824         /* alignment needed */
1825         bs_align_1( &h->out.bs );
1826
1827         /* init cabac */
1828         x264_cabac_context_init( &h->cabac, h->sh.i_type, h->sh.i_qp, h->sh.i_cabac_init_idc );
1829         x264_cabac_encode_init ( &h->cabac, h->out.bs.p, h->out.bs.p_end );
1830     }
1831     h->mb.i_last_qp = h->sh.i_qp;
1832     h->mb.i_last_dqp = 0;
1833
1834     i_mb_y = h->sh.i_first_mb / h->mb.i_mb_width;
1835     i_mb_x = h->sh.i_first_mb % h->mb.i_mb_width;
1836     i_skip = 0;
1837
1838     while( (mb_xy = i_mb_x + i_mb_y * h->mb.i_mb_width) <= h->sh.i_last_mb )
1839     {
1840         int mb_spos = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
1841         if( h->param.i_slice_max_size > 0 )
1842         {
1843             /* We don't need the contexts because flushing the CABAC encoder has no context
1844              * dependency and macroblocks are only re-encoded in the case where a slice is
1845              * ended (and thus the content of all contexts are thrown away). */
1846             if( h->param.b_cabac )
1847             {
1848                 memcpy( &cabac_bak, &h->cabac, offsetof(x264_cabac_t, f8_bits_encoded) );
1849                 /* x264's CABAC writer modifies the previous byte during carry, so it has to be
1850                  * backed up. */
1851                 cabac_prevbyte_bak = h->cabac.p[-1];
1852             }
1853             else
1854             {
1855                 bs_bak = h->out.bs;
1856                 i_skip_bak = i_skip;
1857             }
1858         }
1859
1860         if( i_mb_x == 0 && !h->mb.b_reencode_mb )
1861             x264_fdec_filter_row( h, i_mb_y, 1 );
1862
1863         /* load cache */
1864         x264_macroblock_cache_load( h, i_mb_x, i_mb_y );
1865
1866         x264_macroblock_analyse( h );
1867
1868         /* encode this macroblock -> be careful it can change the mb type to P_SKIP if needed */
1869         x264_macroblock_encode( h );
1870
1871         if( x264_bitstream_check_buffer( h ) )
1872             return -1;
1873
1874         if( h->param.b_cabac )
1875         {
1876             if( mb_xy > h->sh.i_first_mb && !(h->sh.b_mbaff && (i_mb_y&1)) )
1877                 x264_cabac_encode_terminal( &h->cabac );
1878
1879             if( IS_SKIP( h->mb.i_type ) )
1880                 x264_cabac_mb_skip( h, 1 );
1881             else
1882             {
1883                 if( h->sh.i_type != SLICE_TYPE_I )
1884                     x264_cabac_mb_skip( h, 0 );
1885                 x264_macroblock_write_cabac( h, &h->cabac );
1886             }
1887         }
1888         else
1889         {
1890             if( IS_SKIP( h->mb.i_type ) )
1891                 i_skip++;
1892             else
1893             {
1894                 if( h->sh.i_type != SLICE_TYPE_I )
1895                 {
1896                     bs_write_ue( &h->out.bs, i_skip );  /* skip run */
1897                     i_skip = 0;
1898                 }
1899                 x264_macroblock_write_cavlc( h );
1900             }
1901         }
1902
1903         int total_bits = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
1904         int mb_size = total_bits - mb_spos;
1905
1906         /* We'll just re-encode this last macroblock if we go over the max slice size. */
1907         if( total_bits - starting_bits > slice_max_size && !h->mb.b_reencode_mb )
1908         {
1909             if( mb_xy != h->sh.i_first_mb )
1910             {
1911                 if( h->param.b_cabac )
1912                 {
1913                     memcpy( &h->cabac, &cabac_bak, offsetof(x264_cabac_t, f8_bits_encoded) );
1914                     h->cabac.p[-1] = cabac_prevbyte_bak;
1915                 }
1916                 else
1917                 {
1918                     h->out.bs = bs_bak;
1919                     i_skip = i_skip_bak;
1920                 }
1921                 h->mb.b_reencode_mb = 1;
1922                 h->sh.i_last_mb = mb_xy-1;
1923                 break;
1924             }
1925             else
1926             {
1927                 h->sh.i_last_mb = mb_xy;
1928                 h->mb.b_reencode_mb = 0;
1929             }
1930         }
1931         else
1932             h->mb.b_reencode_mb = 0;
1933
1934 #if HAVE_VISUALIZE
1935         if( h->param.b_visualize )
1936             x264_visualize_mb( h );
1937 #endif
1938
1939         /* save cache */
1940         x264_macroblock_cache_save( h );
1941
1942         /* accumulate mb stats */
1943         h->stat.frame.i_mb_count[h->mb.i_type]++;
1944
1945         int b_intra = IS_INTRA( h->mb.i_type );
1946         if( h->param.i_log_level >= X264_LOG_INFO || h->param.rc.b_stat_write )
1947         {
1948             if( !b_intra && !IS_SKIP( h->mb.i_type ) && !IS_DIRECT( h->mb.i_type ) )
1949             {
1950                 if( h->mb.i_partition != D_8x8 )
1951                         h->stat.frame.i_mb_partition[h->mb.i_partition] += 4;
1952                     else
1953                         for( int i = 0; i < 4; i++ )
1954                             h->stat.frame.i_mb_partition[h->mb.i_sub_partition[i]] ++;
1955                 if( h->param.i_frame_reference > 1 )
1956                     for( int i_list = 0; i_list <= (h->sh.i_type == SLICE_TYPE_B); i_list++ )
1957                         for( int i = 0; i < 4; i++ )
1958                         {
1959                             int i_ref = h->mb.cache.ref[i_list][ x264_scan8[4*i] ];
1960                             if( i_ref >= 0 )
1961                                 h->stat.frame.i_mb_count_ref[i_list][i_ref] ++;
1962                         }
1963             }
1964         }
1965
1966         if( h->param.i_log_level >= X264_LOG_INFO )
1967         {
1968             if( h->mb.i_cbp_luma | h->mb.i_cbp_chroma )
1969             {
1970                 int cbpsum = (h->mb.i_cbp_luma&1) + ((h->mb.i_cbp_luma>>1)&1)
1971                            + ((h->mb.i_cbp_luma>>2)&1) + (h->mb.i_cbp_luma>>3);
1972                 h->stat.frame.i_mb_cbp[!b_intra + 0] += cbpsum;
1973                 h->stat.frame.i_mb_cbp[!b_intra + 2] += !!h->mb.i_cbp_chroma;
1974                 h->stat.frame.i_mb_cbp[!b_intra + 4] += h->mb.i_cbp_chroma >> 1;
1975             }
1976             if( h->mb.i_cbp_luma && !b_intra )
1977             {
1978                 h->stat.frame.i_mb_count_8x8dct[0] ++;
1979                 h->stat.frame.i_mb_count_8x8dct[1] += h->mb.b_transform_8x8;
1980             }
1981             if( b_intra && h->mb.i_type != I_PCM )
1982             {
1983                 if( h->mb.i_type == I_16x16 )
1984                     h->stat.frame.i_mb_pred_mode[0][h->mb.i_intra16x16_pred_mode]++;
1985                 else if( h->mb.i_type == I_8x8 )
1986                     for( int i = 0; i < 16; i += 4 )
1987                         h->stat.frame.i_mb_pred_mode[1][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
1988                 else //if( h->mb.i_type == I_4x4 )
1989                     for( int i = 0; i < 16; i++ )
1990                         h->stat.frame.i_mb_pred_mode[2][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
1991                 h->stat.frame.i_mb_pred_mode[3][x264_mb_pred_mode8x8c_fix[h->mb.i_chroma_pred_mode]]++;
1992             }
1993         }
1994
1995         /* calculate deblock strength values (actual deblocking is done per-row along with hpel) */
1996         if( b_deblock )
1997         {
1998             int mvy_limit = 4 >> h->sh.b_mbaff;
1999             uint8_t (*bs)[4][4] = h->deblock_strength[h->mb.i_mb_y&h->sh.b_mbaff][h->mb.i_mb_x];
2000             x264_macroblock_cache_load_deblock( h );
2001             if( IS_INTRA( h->mb.type[h->mb.i_mb_xy] ) )
2002                 memset( bs, 3, 2*4*4*sizeof(uint8_t) );
2003             else
2004                 h->loopf.deblock_strength( h->mb.cache.non_zero_count, h->mb.cache.ref, h->mb.cache.mv,
2005                                            bs, mvy_limit, h->sh.i_type == SLICE_TYPE_B );
2006         }
2007
2008         x264_ratecontrol_mb( h, mb_size );
2009
2010         if( h->sh.b_mbaff )
2011         {
2012             i_mb_x += i_mb_y & 1;
2013             i_mb_y ^= i_mb_x < h->mb.i_mb_width;
2014         }
2015         else
2016             i_mb_x++;
2017         if( i_mb_x == h->mb.i_mb_width )
2018         {
2019             i_mb_y++;
2020             i_mb_x = 0;
2021         }
2022     }
2023
2024     if( h->param.b_cabac )
2025     {
2026         x264_cabac_encode_flush( h, &h->cabac );
2027         h->out.bs.p = h->cabac.p;
2028     }
2029     else
2030     {
2031         if( i_skip > 0 )
2032             bs_write_ue( &h->out.bs, i_skip );  /* last skip run */
2033         /* rbsp_slice_trailing_bits */
2034         bs_rbsp_trailing( &h->out.bs );
2035         bs_flush( &h->out.bs );
2036     }
2037     if( x264_nal_end( h ) )
2038         return -1;
2039
2040     if( h->sh.i_last_mb == (h->i_threadslice_end * h->mb.i_mb_width - 1) )
2041     {
2042         h->stat.frame.i_misc_bits = bs_pos( &h->out.bs )
2043                                   + (h->out.i_nal*NALU_OVERHEAD * 8)
2044                                   - h->stat.frame.i_tex_bits
2045                                   - h->stat.frame.i_mv_bits;
2046         x264_fdec_filter_row( h, h->i_threadslice_end, 1 );
2047     }
2048
2049     return 0;
2050 }
2051
2052 static void x264_thread_sync_context( x264_t *dst, x264_t *src )
2053 {
2054     if( dst == src )
2055         return;
2056
2057     // reference counting
2058     for( x264_frame_t **f = src->frames.reference; *f; f++ )
2059         (*f)->i_reference_count++;
2060     for( x264_frame_t **f = dst->frames.reference; *f; f++ )
2061         x264_frame_push_unused( src, *f );
2062     src->fdec->i_reference_count++;
2063     x264_frame_push_unused( src, dst->fdec );
2064
2065     // copy everything except the per-thread pointers and the constants.
2066     memcpy( &dst->i_frame, &src->i_frame, offsetof(x264_t, mb.type) - offsetof(x264_t, i_frame) );
2067     dst->param = src->param;
2068     dst->stat = src->stat;
2069 }
2070
2071 static void x264_thread_sync_stat( x264_t *dst, x264_t *src )
2072 {
2073     if( dst == src )
2074         return;
2075     memcpy( &dst->stat.i_frame_count, &src->stat.i_frame_count, sizeof(dst->stat) - sizeof(dst->stat.frame) );
2076 }
2077
2078 static void *x264_slices_write( x264_t *h )
2079 {
2080     int i_slice_num = 0;
2081     int last_thread_mb = h->sh.i_last_mb;
2082
2083 #if HAVE_VISUALIZE
2084     if( h->param.b_visualize )
2085         if( x264_visualize_init( h ) )
2086             return (void *)-1;
2087 #endif
2088
2089     /* init stats */
2090     memset( &h->stat.frame, 0, sizeof(h->stat.frame) );
2091     h->mb.b_reencode_mb = 0;
2092     while( h->sh.i_first_mb <= last_thread_mb )
2093     {
2094         h->sh.i_last_mb = last_thread_mb;
2095         if( h->param.i_slice_max_mbs )
2096             h->sh.i_last_mb = h->sh.i_first_mb + h->param.i_slice_max_mbs - 1;
2097         else if( h->param.i_slice_count && !h->param.b_sliced_threads )
2098         {
2099             int height = h->mb.i_mb_height >> h->param.b_interlaced;
2100             int width = h->mb.i_mb_width << h->param.b_interlaced;
2101             i_slice_num++;
2102             h->sh.i_last_mb = (height * i_slice_num + h->param.i_slice_count/2) / h->param.i_slice_count * width - 1;
2103         }
2104         h->sh.i_last_mb = X264_MIN( h->sh.i_last_mb, last_thread_mb );
2105         if( x264_stack_align( x264_slice_write, h ) )
2106             return (void *)-1;
2107         h->sh.i_first_mb = h->sh.i_last_mb + 1;
2108     }
2109
2110 #if HAVE_VISUALIZE
2111     if( h->param.b_visualize )
2112     {
2113         x264_visualize_show( h );
2114         x264_visualize_close( h );
2115     }
2116 #endif
2117
2118     return (void *)0;
2119 }
2120
2121 static int x264_threaded_slices_write( x264_t *h )
2122 {
2123     /* set first/last mb and sync contexts */
2124     for( int i = 0; i < h->param.i_threads; i++ )
2125     {
2126         x264_t *t = h->thread[i];
2127         if( i )
2128         {
2129             t->param = h->param;
2130             memcpy( &t->i_frame, &h->i_frame, offsetof(x264_t, rc) - offsetof(x264_t, i_frame) );
2131         }
2132         int height = h->mb.i_mb_height >> h->param.b_interlaced;
2133         t->i_threadslice_start = ((height *  i    + h->param.i_slice_count/2) / h->param.i_threads) << h->param.b_interlaced;
2134         t->i_threadslice_end   = ((height * (i+1) + h->param.i_slice_count/2) / h->param.i_threads) << h->param.b_interlaced;
2135         t->sh.i_first_mb = t->i_threadslice_start * h->mb.i_mb_width;
2136         t->sh.i_last_mb  =   t->i_threadslice_end * h->mb.i_mb_width - 1;
2137     }
2138
2139     x264_stack_align( x264_analyse_weight_frame, h, h->mb.i_mb_height*16 + 16 );
2140
2141     x264_threads_distribute_ratecontrol( h );
2142
2143     /* dispatch */
2144     for( int i = 0; i < h->param.i_threads; i++ )
2145     {
2146         x264_threadpool_run( h->threadpool, (void*)x264_slices_write, h->thread[i] );
2147         h->thread[i]->b_thread_active = 1;
2148     }
2149     for( int i = 0; i < h->param.i_threads; i++ )
2150     {
2151         h->thread[i]->b_thread_active = 0;
2152         if( (intptr_t)x264_threadpool_wait( h->threadpool, h->thread[i] ) )
2153             return -1;
2154     }
2155
2156     /* Go back and fix up the hpel on the borders between slices. */
2157     for( int i = 1; i < h->param.i_threads; i++ )
2158     {
2159         x264_fdec_filter_row( h->thread[i], h->thread[i]->i_threadslice_start + 1, 0 );
2160         if( h->sh.b_mbaff )
2161             x264_fdec_filter_row( h->thread[i], h->thread[i]->i_threadslice_start + 2, 0 );
2162     }
2163
2164     x264_threads_merge_ratecontrol( h );
2165
2166     for( int i = 1; i < h->param.i_threads; i++ )
2167     {
2168         x264_t *t = h->thread[i];
2169         for( int j = 0; j < t->out.i_nal; j++ )
2170         {
2171             h->out.nal[h->out.i_nal] = t->out.nal[j];
2172             h->out.i_nal++;
2173             x264_nal_check_buffer( h );
2174         }
2175         /* All entries in stat.frame are ints except for ssd/ssim. */
2176         for( int j = 0; j < (offsetof(x264_t,stat.frame.i_ssd) - offsetof(x264_t,stat.frame.i_mv_bits)) / sizeof(int); j++ )
2177             ((int*)&h->stat.frame)[j] += ((int*)&t->stat.frame)[j];
2178         for( int j = 0; j < 3; j++ )
2179             h->stat.frame.i_ssd[j] += t->stat.frame.i_ssd[j];
2180         h->stat.frame.f_ssim += t->stat.frame.f_ssim;
2181     }
2182
2183     return 0;
2184 }
2185
2186 void x264_encoder_intra_refresh( x264_t *h )
2187 {
2188     h = h->thread[h->i_thread_phase];
2189     h->b_queued_intra_refresh = 1;
2190 }
2191
2192 int x264_encoder_invalidate_reference( x264_t *h, int64_t pts )
2193 {
2194     if( h->param.i_bframe )
2195     {
2196         x264_log( h, X264_LOG_ERROR, "x264_encoder_invalidate_reference is not supported with B-frames enabled\n" );
2197         return -1;
2198     }
2199     if( h->param.b_intra_refresh )
2200     {
2201         x264_log( h, X264_LOG_ERROR, "x264_encoder_invalidate_reference is not supported with intra refresh enabled\n" );
2202         return -1;
2203     }
2204     h = h->thread[h->i_thread_phase];
2205     h->i_reference_invalidate_pts = pts;
2206     return 0;
2207 }
2208
2209 /****************************************************************************
2210  * x264_encoder_encode:
2211  *  XXX: i_poc   : is the poc of the current given picture
2212  *       i_frame : is the number of the frame being coded
2213  *  ex:  type frame poc
2214  *       I      0   2*0
2215  *       P      1   2*3
2216  *       B      2   2*1
2217  *       B      3   2*2
2218  *       P      4   2*6
2219  *       B      5   2*4
2220  *       B      6   2*5
2221  ****************************************************************************/
2222 int     x264_encoder_encode( x264_t *h,
2223                              x264_nal_t **pp_nal, int *pi_nal,
2224                              x264_picture_t *pic_in,
2225                              x264_picture_t *pic_out )
2226 {
2227     x264_t *thread_current, *thread_prev, *thread_oldest;
2228     int i_nal_type, i_nal_ref_idc, i_global_qp;
2229     int overhead = NALU_OVERHEAD;
2230
2231     if( h->i_thread_frames > 1 )
2232     {
2233         thread_prev    = h->thread[ h->i_thread_phase ];
2234         h->i_thread_phase = (h->i_thread_phase + 1) % h->i_thread_frames;
2235         thread_current = h->thread[ h->i_thread_phase ];
2236         thread_oldest  = h->thread[ (h->i_thread_phase + 1) % h->i_thread_frames ];
2237         x264_thread_sync_context( thread_current, thread_prev );
2238         x264_thread_sync_ratecontrol( thread_current, thread_prev, thread_oldest );
2239         h = thread_current;
2240     }
2241     else
2242     {
2243         thread_current =
2244         thread_oldest  = h;
2245     }
2246 #if HAVE_MMX
2247     if( h->i_thread_frames == 1 && h->param.cpu&X264_CPU_SSE_MISALIGN )
2248         x264_cpu_mask_misalign_sse();
2249 #endif
2250
2251     // ok to call this before encoding any frames, since the initial values of fdec have b_kept_as_ref=0
2252     if( x264_reference_update( h ) )
2253         return -1;
2254     h->fdec->i_lines_completed = -1;
2255
2256     /* no data out */
2257     *pi_nal = 0;
2258     *pp_nal = NULL;
2259
2260     /* ------------------- Setup new frame from picture -------------------- */
2261     if( pic_in != NULL )
2262     {
2263         /* 1: Copy the picture to a frame and move it to a buffer */
2264         x264_frame_t *fenc = x264_frame_pop_unused( h, 0 );
2265         if( !fenc )
2266             return -1;
2267
2268         if( x264_frame_copy_picture( h, fenc, pic_in ) < 0 )
2269             return -1;
2270
2271         if( h->param.i_width != 16 * h->mb.i_mb_width ||
2272             h->param.i_height != 16 * h->mb.i_mb_height )
2273             x264_frame_expand_border_mod16( h, fenc );
2274
2275         fenc->i_frame = h->frames.i_input++;
2276
2277         if( h->frames.i_bframe_delay && fenc->i_frame == h->frames.i_bframe_delay )
2278             h->frames.i_bframe_delay_time = fenc->i_pts;
2279
2280         if( h->param.b_vfr_input && fenc->i_pts <= h->frames.i_largest_pts )
2281             x264_log( h, X264_LOG_WARNING, "non-strictly-monotonic PTS\n" );
2282
2283         h->frames.i_second_largest_pts = h->frames.i_largest_pts;
2284         h->frames.i_largest_pts = fenc->i_pts;
2285
2286         if( (fenc->i_pic_struct < PIC_STRUCT_AUTO) || (fenc->i_pic_struct > PIC_STRUCT_TRIPLE) )
2287             fenc->i_pic_struct = PIC_STRUCT_AUTO;
2288
2289         if( fenc->i_pic_struct == PIC_STRUCT_AUTO )
2290         {
2291             int b_interlaced = fenc->param ? fenc->param->b_interlaced : h->param.b_interlaced;
2292             if( b_interlaced )
2293             {
2294                 int b_tff = fenc->param ? fenc->param->b_tff : h->param.b_tff;
2295                 fenc->i_pic_struct = b_tff ? PIC_STRUCT_TOP_BOTTOM : PIC_STRUCT_BOTTOM_TOP;
2296             }
2297             else
2298                 fenc->i_pic_struct = PIC_STRUCT_PROGRESSIVE;
2299         }
2300
2301         if( h->param.rc.b_mb_tree && h->param.rc.b_stat_read )
2302         {
2303             if( x264_macroblock_tree_read( h, fenc, pic_in->prop.quant_offsets ) )
2304                 return -1;
2305         }
2306         else
2307             x264_adaptive_quant_frame( h, fenc, pic_in->prop.quant_offsets );
2308
2309         if( pic_in->prop.quant_offsets_free )
2310             pic_in->prop.quant_offsets_free( pic_in->prop.quant_offsets );
2311
2312         if( h->frames.b_have_lowres )
2313             x264_frame_init_lowres( h, fenc );
2314
2315         /* 2: Place the frame into the queue for its slice type decision */
2316         x264_lookahead_put_frame( h, fenc );
2317
2318         if( h->frames.i_input <= h->frames.i_delay + 1 - h->i_thread_frames )
2319         {
2320             /* Nothing yet to encode, waiting for filling of buffers */
2321             pic_out->i_type = X264_TYPE_AUTO;
2322             return 0;
2323         }
2324     }
2325     else
2326     {
2327         /* signal kills for lookahead thread */
2328         x264_pthread_mutex_lock( &h->lookahead->ifbuf.mutex );
2329         h->lookahead->b_exit_thread = 1;
2330         x264_pthread_cond_broadcast( &h->lookahead->ifbuf.cv_fill );
2331         x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
2332     }
2333
2334     h->i_frame++;
2335     /* 3: The picture is analyzed in the lookahead */
2336     if( !h->frames.current[0] )
2337         x264_lookahead_get_frames( h );
2338
2339     if( !h->frames.current[0] && x264_lookahead_is_empty( h ) )
2340         return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
2341
2342     /* ------------------- Get frame to be encoded ------------------------- */
2343     /* 4: get picture to encode */
2344     h->fenc = x264_frame_shift( h->frames.current );
2345     if( h->i_frame == h->i_thread_frames - 1 )
2346         h->i_reordered_pts_delay = h->fenc->i_reordered_pts;
2347     if( h->fenc->param )
2348     {
2349         x264_encoder_reconfig( h, h->fenc->param );
2350         if( h->fenc->param->param_free )
2351             h->fenc->param->param_free( h->fenc->param );
2352     }
2353
2354     if( h->i_reference_invalidate_pts )
2355     {
2356         if( h->i_reference_invalidate_pts >= h->i_last_idr_pts )
2357             for( int i = 0; h->frames.reference[i]; i++ )
2358                 if( h->i_reference_invalidate_pts <= h->frames.reference[i]->i_pts )
2359                     h->frames.reference[i]->b_corrupt = 1;
2360         h->i_reference_invalidate_pts = 0;
2361     }
2362
2363     if( !IS_X264_TYPE_I( h->fenc->i_type ) )
2364     {
2365         int valid_refs_left = 0;
2366         for( int i = 0; h->frames.reference[i]; i++ )
2367             if( !h->frames.reference[i]->b_corrupt )
2368                 valid_refs_left++;
2369         /* No valid reference frames left: force an IDR. */
2370         if( !valid_refs_left )
2371         {
2372             h->fenc->b_keyframe = 1;
2373             h->fenc->i_type = X264_TYPE_IDR;
2374         }
2375     }
2376
2377     if( h->fenc->b_keyframe )
2378     {
2379         h->frames.i_last_keyframe = h->fenc->i_frame;
2380         if( h->fenc->i_type == X264_TYPE_IDR )
2381         {
2382             h->i_frame_num = 0;
2383             h->frames.i_last_idr = h->fenc->i_frame;
2384         }
2385     }
2386     h->sh.i_mmco_command_count =
2387     h->sh.i_mmco_remove_from_end = 0;
2388     h->b_ref_reorder[0] =
2389     h->b_ref_reorder[1] = 0;
2390     h->fdec->i_poc =
2391     h->fenc->i_poc = 2 * ( h->fenc->i_frame - X264_MAX( h->frames.i_last_idr, 0 ) );
2392
2393     /* ------------------- Setup frame context ----------------------------- */
2394     /* 5: Init data dependent of frame type */
2395     if( h->fenc->i_type == X264_TYPE_IDR )
2396     {
2397         /* reset ref pictures */
2398         i_nal_type    = NAL_SLICE_IDR;
2399         i_nal_ref_idc = NAL_PRIORITY_HIGHEST;
2400         h->sh.i_type = SLICE_TYPE_I;
2401         x264_reference_reset( h );
2402         h->frames.i_poc_last_open_gop = -1;
2403     }
2404     else if( h->fenc->i_type == X264_TYPE_I )
2405     {
2406         i_nal_type    = NAL_SLICE;
2407         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
2408         h->sh.i_type = SLICE_TYPE_I;
2409         x264_reference_hierarchy_reset( h );
2410         if( h->param.i_open_gop )
2411             h->frames.i_poc_last_open_gop = h->fenc->b_keyframe ? h->fenc->i_poc : -1;
2412     }
2413     else if( h->fenc->i_type == X264_TYPE_P )
2414     {
2415         i_nal_type    = NAL_SLICE;
2416         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
2417         h->sh.i_type = SLICE_TYPE_P;
2418         x264_reference_hierarchy_reset( h );
2419         h->frames.i_poc_last_open_gop = -1;
2420     }
2421     else if( h->fenc->i_type == X264_TYPE_BREF )
2422     {
2423         i_nal_type    = NAL_SLICE;
2424         i_nal_ref_idc = h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT ? NAL_PRIORITY_LOW : NAL_PRIORITY_HIGH;
2425         h->sh.i_type = SLICE_TYPE_B;
2426         x264_reference_hierarchy_reset( h );
2427     }
2428     else    /* B frame */
2429     {
2430         i_nal_type    = NAL_SLICE;
2431         i_nal_ref_idc = NAL_PRIORITY_DISPOSABLE;
2432         h->sh.i_type = SLICE_TYPE_B;
2433     }
2434
2435     h->fdec->i_type = h->fenc->i_type;
2436     h->fdec->i_frame = h->fenc->i_frame;
2437     h->fenc->b_kept_as_ref =
2438     h->fdec->b_kept_as_ref = i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE && h->param.i_keyint_max > 1;
2439
2440     h->fdec->i_pts = h->fenc->i_pts *= h->i_dts_compress_multiplier;
2441     if( h->frames.i_bframe_delay )
2442     {
2443         int64_t *prev_reordered_pts = thread_current->frames.i_prev_reordered_pts;
2444         if( h->i_frame <= h->frames.i_bframe_delay )
2445         {
2446             if( h->i_dts_compress_multiplier == 1 )
2447                 h->fdec->i_dts = h->fenc->i_reordered_pts - h->frames.i_bframe_delay_time;
2448             else
2449             {
2450                 /* DTS compression */
2451                 if( h->i_frame == 1 )
2452                     thread_current->frames.i_init_delta = h->fenc->i_reordered_pts * h->i_dts_compress_multiplier;
2453                 h->fdec->i_dts = h->i_frame * thread_current->frames.i_init_delta / h->i_dts_compress_multiplier;
2454             }
2455         }
2456         else
2457             h->fdec->i_dts = prev_reordered_pts[ (h->i_frame - h->frames.i_bframe_delay) % h->frames.i_bframe_delay ];
2458         prev_reordered_pts[ h->i_frame % h->frames.i_bframe_delay ] = h->fenc->i_reordered_pts * h->i_dts_compress_multiplier;
2459     }
2460     else
2461         h->fdec->i_dts = h->fenc->i_reordered_pts;
2462     if( h->fenc->i_type == X264_TYPE_IDR )
2463         h->i_last_idr_pts = h->fdec->i_pts;
2464
2465     /* ------------------- Init                ----------------------------- */
2466     /* build ref list 0/1 */
2467     x264_reference_build_list( h, h->fdec->i_poc );
2468
2469     /* ---------------------- Write the bitstream -------------------------- */
2470     /* Init bitstream context */
2471     if( h->param.b_sliced_threads )
2472     {
2473         for( int i = 0; i < h->param.i_threads; i++ )
2474         {
2475             bs_init( &h->thread[i]->out.bs, h->thread[i]->out.p_bitstream, h->thread[i]->out.i_bitstream );
2476             h->thread[i]->out.i_nal = 0;
2477         }
2478     }
2479     else
2480     {
2481         bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
2482         h->out.i_nal = 0;
2483     }
2484
2485     if( h->param.b_aud )
2486     {
2487         int pic_type;
2488
2489         if( h->sh.i_type == SLICE_TYPE_I )
2490             pic_type = 0;
2491         else if( h->sh.i_type == SLICE_TYPE_P )
2492             pic_type = 1;
2493         else if( h->sh.i_type == SLICE_TYPE_B )
2494             pic_type = 2;
2495         else
2496             pic_type = 7;
2497
2498         x264_nal_start( h, NAL_AUD, NAL_PRIORITY_DISPOSABLE );
2499         bs_write( &h->out.bs, 3, pic_type );
2500         bs_rbsp_trailing( &h->out.bs );
2501         if( x264_nal_end( h ) )
2502             return -1;
2503         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2504     }
2505
2506     h->i_nal_type = i_nal_type;
2507     h->i_nal_ref_idc = i_nal_ref_idc;
2508
2509     if( h->param.b_intra_refresh )
2510     {
2511         if( IS_X264_TYPE_I( h->fenc->i_type ) )
2512         {
2513             h->fdec->i_frames_since_pir = 0;
2514             h->b_queued_intra_refresh = 0;
2515             /* PIR is currently only supported with ref == 1, so any intra frame effectively refreshes
2516              * the whole frame and counts as an intra refresh. */
2517             h->fdec->f_pir_position = h->mb.i_mb_width;
2518         }
2519         else if( h->fenc->i_type == X264_TYPE_P )
2520         {
2521             int pocdiff = (h->fdec->i_poc - h->fref0[0]->i_poc)/2;
2522             float increment = X264_MAX( ((float)h->mb.i_mb_width-1) / h->param.i_keyint_max, 1 );
2523             h->fdec->f_pir_position = h->fref0[0]->f_pir_position;
2524             h->fdec->i_frames_since_pir = h->fref0[0]->i_frames_since_pir + pocdiff;
2525             if( h->fdec->i_frames_since_pir >= h->param.i_keyint_max ||
2526                 (h->b_queued_intra_refresh && h->fdec->f_pir_position + 0.5 >= h->mb.i_mb_width) )
2527             {
2528                 h->fdec->f_pir_position = 0;
2529                 h->fdec->i_frames_since_pir = 0;
2530                 h->b_queued_intra_refresh = 0;
2531                 h->fenc->b_keyframe = 1;
2532             }
2533             h->fdec->i_pir_start_col = h->fdec->f_pir_position+0.5;
2534             h->fdec->f_pir_position += increment * pocdiff;
2535             h->fdec->i_pir_end_col = h->fdec->f_pir_position+0.5;
2536         }
2537     }
2538
2539     if( h->fenc->b_keyframe )
2540     {
2541         /* Write SPS and PPS */
2542         if( h->param.b_repeat_headers )
2543         {
2544             /* generate sequence parameters */
2545             x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
2546             x264_sps_write( &h->out.bs, h->sps );
2547             if( x264_nal_end( h ) )
2548                 return -1;
2549             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
2550
2551             /* generate picture parameters */
2552             x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
2553             x264_pps_write( &h->out.bs, h->pps );
2554             if( x264_nal_end( h ) )
2555                 return -1;
2556             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
2557         }
2558
2559         /* buffering period sei is written in x264_encoder_frame_end */
2560
2561         if( h->param.b_repeat_headers && h->fenc->i_frame == 0 )
2562         {
2563             /* identify ourself */
2564             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2565             if( x264_sei_version_write( h, &h->out.bs ) )
2566                 return -1;
2567             if( x264_nal_end( h ) )
2568                 return -1;
2569             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2570         }
2571
2572         if( h->fenc->i_type != X264_TYPE_IDR )
2573         {
2574             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;
2575             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2576             x264_sei_recovery_point_write( h, &h->out.bs, time_to_recovery );
2577             x264_nal_end( h );
2578             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2579         }
2580     }
2581
2582     /* generate sei pic timing */
2583     if( h->sps->vui.b_pic_struct_present || h->sps->vui.b_nal_hrd_parameters_present )
2584     {
2585         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2586         x264_sei_pic_timing_write( h, &h->out.bs );
2587         if( x264_nal_end( h ) )
2588             return -1;
2589         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2590     }
2591
2592     /* Init the rate control */
2593     /* FIXME: Include slice header bit cost. */
2594     x264_ratecontrol_start( h, h->fenc->i_qpplus1, overhead*8 );
2595     i_global_qp = x264_ratecontrol_qp( h );
2596
2597     pic_out->i_qpplus1 =
2598     h->fdec->i_qpplus1 = i_global_qp + 1;
2599
2600     if( h->param.rc.b_stat_read && h->sh.i_type != SLICE_TYPE_I )
2601     {
2602         x264_reference_build_list_optimal( h );
2603         x264_reference_check_reorder( h );
2604     }
2605
2606     if( h->i_ref0 )
2607         h->fdec->i_poc_l0ref0 = h->fref0[0]->i_poc;
2608
2609     if( h->sh.i_type == SLICE_TYPE_B )
2610         x264_macroblock_bipred_init( h );
2611
2612     /*------------------------- Weights -------------------------------------*/
2613     x264_weighted_pred_init( h );
2614
2615     /* ------------------------ Create slice header  ----------------------- */
2616     x264_slice_init( h, i_nal_type, i_global_qp );
2617
2618     if( i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE )
2619         h->i_frame_num++;
2620
2621     /* Write frame */
2622     h->i_threadslice_start = 0;
2623     h->i_threadslice_end = h->mb.i_mb_height;
2624     if( h->i_thread_frames > 1 )
2625     {
2626         x264_threadpool_run( h->threadpool, (void*)x264_slices_write, h );
2627         h->b_thread_active = 1;
2628     }
2629     else if( h->param.b_sliced_threads )
2630     {
2631         if( x264_threaded_slices_write( h ) )
2632             return -1;
2633     }
2634     else
2635         if( (intptr_t)x264_slices_write( h ) )
2636             return -1;
2637
2638     return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
2639 }
2640
2641 static int x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
2642                                    x264_nal_t **pp_nal, int *pi_nal,
2643                                    x264_picture_t *pic_out )
2644 {
2645     char psz_message[80];
2646
2647     if( h->b_thread_active )
2648     {
2649         h->b_thread_active = 0;
2650         if( (intptr_t)x264_threadpool_wait( h->threadpool, h ) )
2651             return -1;
2652     }
2653     if( !h->out.i_nal )
2654     {
2655         pic_out->i_type = X264_TYPE_AUTO;
2656         return 0;
2657     }
2658
2659     x264_emms();
2660     /* generate sei buffering period and insert it into place */
2661     if( h->fenc->b_keyframe && h->sps->vui.b_nal_hrd_parameters_present )
2662     {
2663         x264_hrd_fullness( h );
2664         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2665         x264_sei_buffering_period_write( h, &h->out.bs );
2666         if( x264_nal_end( h ) )
2667            return -1;
2668         /* buffering period sei must follow AUD, SPS and PPS and precede all other SEIs */
2669         int idx = 0;
2670         while( h->out.nal[idx].i_type == NAL_AUD ||
2671                h->out.nal[idx].i_type == NAL_SPS ||
2672                h->out.nal[idx].i_type == NAL_PPS )
2673             idx++;
2674         x264_nal_t nal_tmp = h->out.nal[h->out.i_nal-1];
2675         memmove( &h->out.nal[idx+1], &h->out.nal[idx], (h->out.i_nal-idx-1)*sizeof(x264_nal_t) );
2676         h->out.nal[idx] = nal_tmp;
2677     }
2678
2679     int frame_size = x264_encoder_encapsulate_nals( h, 0 );
2680
2681     /* Set output picture properties */
2682     pic_out->i_type = h->fenc->i_type;
2683
2684     pic_out->b_keyframe = h->fenc->b_keyframe;
2685
2686     pic_out->i_pts = h->fdec->i_pts;
2687     pic_out->i_dts = h->fdec->i_dts;
2688
2689     if( pic_out->i_pts < pic_out->i_dts )
2690         x264_log( h, X264_LOG_WARNING, "invalid DTS: PTS is less than DTS\n" );
2691
2692     pic_out->img.i_plane = h->fdec->i_plane;
2693     for( int i = 0; i < 3; i++ )
2694     {
2695         pic_out->img.i_stride[i] = h->fdec->i_stride[i];
2696         pic_out->img.plane[i] = h->fdec->plane[i];
2697     }
2698
2699     x264_frame_push_unused( thread_current, h->fenc );
2700
2701     /* ---------------------- Update encoder state ------------------------- */
2702
2703     /* update rc */
2704     int filler = 0;
2705     if( x264_ratecontrol_end( h, frame_size * 8, &filler ) < 0 )
2706         return -1;
2707
2708     pic_out->hrd_timing = h->fenc->hrd_timing;
2709
2710     while( filler > 0 )
2711     {
2712         int f, overhead;
2713         overhead = (FILLER_OVERHEAD - h->param.b_annexb);
2714         if( h->param.i_slice_max_size && filler > h->param.i_slice_max_size )
2715         {
2716             int next_size = filler - h->param.i_slice_max_size;
2717             int overflow = X264_MAX( overhead - next_size, 0 );
2718             f = h->param.i_slice_max_size - overhead - overflow;
2719         }
2720         else
2721             f = X264_MAX( 0, filler - overhead );
2722
2723         x264_nal_start( h, NAL_FILLER, NAL_PRIORITY_DISPOSABLE );
2724         x264_filler_write( h, &h->out.bs, f );
2725         if( x264_nal_end( h ) )
2726             return -1;
2727         int total_size = x264_encoder_encapsulate_nals( h, h->out.i_nal-1 );
2728         frame_size += total_size;
2729         filler -= total_size;
2730     }
2731
2732     /* End bitstream, set output  */
2733     *pi_nal = h->out.i_nal;
2734     *pp_nal = h->out.nal;
2735
2736     h->out.i_nal = 0;
2737
2738     x264_noise_reduction_update( h );
2739
2740     /* ---------------------- Compute/Print statistics --------------------- */
2741     x264_thread_sync_stat( h, h->thread[0] );
2742
2743     /* Slice stat */
2744     h->stat.i_frame_count[h->sh.i_type]++;
2745     h->stat.i_frame_size[h->sh.i_type] += frame_size;
2746     h->stat.f_frame_qp[h->sh.i_type] += h->fdec->f_qp_avg_aq;
2747
2748     for( int i = 0; i < X264_MBTYPE_MAX; i++ )
2749         h->stat.i_mb_count[h->sh.i_type][i] += h->stat.frame.i_mb_count[i];
2750     for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
2751         h->stat.i_mb_partition[h->sh.i_type][i] += h->stat.frame.i_mb_partition[i];
2752     for( int i = 0; i < 2; i++ )
2753         h->stat.i_mb_count_8x8dct[i] += h->stat.frame.i_mb_count_8x8dct[i];
2754     for( int i = 0; i < 6; i++ )
2755         h->stat.i_mb_cbp[i] += h->stat.frame.i_mb_cbp[i];
2756     for( int i = 0; i < 4; i++ )
2757         for( int j = 0; j < 13; j++ )
2758             h->stat.i_mb_pred_mode[i][j] += h->stat.frame.i_mb_pred_mode[i][j];
2759     if( h->sh.i_type != SLICE_TYPE_I )
2760         for( int i_list = 0; i_list < 2; i_list++ )
2761             for( int i = 0; i < 32; i++ )
2762                 h->stat.i_mb_count_ref[h->sh.i_type][i_list][i] += h->stat.frame.i_mb_count_ref[i_list][i];
2763     if( h->sh.i_type == SLICE_TYPE_P )
2764     {
2765         h->stat.i_consecutive_bframes[h->fdec->i_frame - h->fref0[0]->i_frame - 1]++;
2766         if( h->param.analyse.i_weighted_pred == X264_WEIGHTP_SMART )
2767             for( int i = 0; i < 3; i++ )
2768                 for( int j = 0; j < h->i_ref0; j++ )
2769                     if( h->sh.weight[0][i].i_denom != 0 )
2770                     {
2771                         h->stat.i_wpred[i]++;
2772                         break;
2773                     }
2774     }
2775     if( h->sh.i_type == SLICE_TYPE_B )
2776     {
2777         h->stat.i_direct_frames[ h->sh.b_direct_spatial_mv_pred ] ++;
2778         if( h->mb.b_direct_auto_write )
2779         {
2780             //FIXME somewhat arbitrary time constants
2781             if( h->stat.i_direct_score[0] + h->stat.i_direct_score[1] > h->mb.i_mb_count )
2782                 for( int i = 0; i < 2; i++ )
2783                     h->stat.i_direct_score[i] = h->stat.i_direct_score[i] * 9/10;
2784             for( int i = 0; i < 2; i++ )
2785                 h->stat.i_direct_score[i] += h->stat.frame.i_direct_score[i];
2786         }
2787     }
2788
2789     psz_message[0] = '\0';
2790     if( h->param.analyse.b_psnr )
2791     {
2792         int64_t ssd[3] = {
2793             h->stat.frame.i_ssd[0],
2794             h->stat.frame.i_ssd[1],
2795             h->stat.frame.i_ssd[2],
2796         };
2797
2798         h->stat.i_ssd_global[h->sh.i_type] += ssd[0] + ssd[1] + ssd[2];
2799         h->stat.f_psnr_average[h->sh.i_type] += x264_psnr( ssd[0] + ssd[1] + ssd[2], 3 * h->param.i_width * h->param.i_height / 2 );
2800         h->stat.f_psnr_mean_y[h->sh.i_type] += x264_psnr( ssd[0], h->param.i_width * h->param.i_height );
2801         h->stat.f_psnr_mean_u[h->sh.i_type] += x264_psnr( ssd[1], h->param.i_width * h->param.i_height / 4 );
2802         h->stat.f_psnr_mean_v[h->sh.i_type] += x264_psnr( ssd[2], h->param.i_width * h->param.i_height / 4 );
2803
2804         snprintf( psz_message, 80, " PSNR Y:%5.2f U:%5.2f V:%5.2f",
2805                   x264_psnr( ssd[0], h->param.i_width * h->param.i_height ),
2806                   x264_psnr( ssd[1], h->param.i_width * h->param.i_height / 4),
2807                   x264_psnr( ssd[2], h->param.i_width * h->param.i_height / 4) );
2808     }
2809
2810     if( h->param.analyse.b_ssim )
2811     {
2812         double ssim_y = h->stat.frame.f_ssim
2813                       / (((h->param.i_width-6)>>2) * ((h->param.i_height-6)>>2));
2814         h->stat.f_ssim_mean_y[h->sh.i_type] += ssim_y;
2815         snprintf( psz_message + strlen(psz_message), 80 - strlen(psz_message),
2816                   " SSIM Y:%.5f", ssim_y );
2817     }
2818     psz_message[79] = '\0';
2819
2820     x264_log( h, X264_LOG_DEBUG,
2821                   "frame=%4d QP=%.2f NAL=%d Slice:%c Poc:%-3d I:%-4d P:%-4d SKIP:%-4d size=%d bytes%s\n",
2822               h->i_frame,
2823               h->fdec->f_qp_avg_aq,
2824               h->i_nal_ref_idc,
2825               h->sh.i_type == SLICE_TYPE_I ? 'I' : (h->sh.i_type == SLICE_TYPE_P ? 'P' : 'B' ),
2826               h->fdec->i_poc,
2827               h->stat.frame.i_mb_count_i,
2828               h->stat.frame.i_mb_count_p,
2829               h->stat.frame.i_mb_count_skip,
2830               frame_size,
2831               psz_message );
2832
2833     // keep stats all in one place
2834     x264_thread_sync_stat( h->thread[0], h );
2835     // for the use of the next frame
2836     x264_thread_sync_stat( thread_current, h );
2837
2838 #ifdef DEBUG_MB_TYPE
2839 {
2840     static const char mb_chars[] = { 'i', 'i', 'I', 'C', 'P', '8', 'S',
2841         'D', '<', 'X', 'B', 'X', '>', 'B', 'B', 'B', 'B', '8', 'S' };
2842     for( int mb_xy = 0; mb_xy < h->mb.i_mb_width * h->mb.i_mb_height; mb_xy++ )
2843     {
2844         if( h->mb.type[mb_xy] < X264_MBTYPE_MAX && h->mb.type[mb_xy] >= 0 )
2845             fprintf( stderr, "%c ", mb_chars[ h->mb.type[mb_xy] ] );
2846         else
2847             fprintf( stderr, "? " );
2848
2849         if( (mb_xy+1) % h->mb.i_mb_width == 0 )
2850             fprintf( stderr, "\n" );
2851     }
2852 }
2853 #endif
2854
2855     /* Remove duplicates, must be done near the end as breaks h->fref0 array
2856      * by freeing some of its pointers. */
2857      for( int i = 0; i < h->i_ref0; i++ )
2858          if( h->fref0[i] && h->fref0[i]->b_duplicate )
2859          {
2860              x264_frame_push_blank_unused( h, h->fref0[i] );
2861              h->fref0[i] = 0;
2862          }
2863
2864     if( h->param.psz_dump_yuv )
2865         x264_frame_dump( h );
2866
2867     return frame_size;
2868 }
2869
2870 static void x264_print_intra( int64_t *i_mb_count, double i_count, int b_print_pcm, char *intra )
2871 {
2872     intra += sprintf( intra, "I16..4%s: %4.1f%% %4.1f%% %4.1f%%",
2873         b_print_pcm ? "..PCM" : "",
2874         i_mb_count[I_16x16]/ i_count,
2875         i_mb_count[I_8x8]  / i_count,
2876         i_mb_count[I_4x4]  / i_count );
2877     if( b_print_pcm )
2878         sprintf( intra, " %4.1f%%", i_mb_count[I_PCM]  / i_count );
2879 }
2880
2881 /****************************************************************************
2882  * x264_encoder_close:
2883  ****************************************************************************/
2884 void    x264_encoder_close  ( x264_t *h )
2885 {
2886     int64_t i_yuv_size = 3 * h->param.i_width * h->param.i_height / 2;
2887     int64_t i_mb_count_size[2][7] = {{0}};
2888     char buf[200];
2889     int b_print_pcm = h->stat.i_mb_count[SLICE_TYPE_I][I_PCM]
2890                    || h->stat.i_mb_count[SLICE_TYPE_P][I_PCM]
2891                    || h->stat.i_mb_count[SLICE_TYPE_B][I_PCM];
2892
2893     x264_lookahead_delete( h );
2894
2895     if( h->param.i_threads > 1 )
2896         x264_threadpool_delete( h->threadpool );
2897     if( h->i_thread_frames > 1 )
2898     {
2899         for( int i = 0; i < h->i_thread_frames; i++ )
2900             if( h->thread[i]->b_thread_active )
2901             {
2902                 assert( h->thread[i]->fenc->i_reference_count == 1 );
2903                 x264_frame_delete( h->thread[i]->fenc );
2904             }
2905
2906         x264_t *thread_prev = h->thread[h->i_thread_phase];
2907         x264_thread_sync_ratecontrol( h, thread_prev, h );
2908         x264_thread_sync_ratecontrol( thread_prev, thread_prev, h );
2909         h->i_frame = thread_prev->i_frame + 1 - h->i_thread_frames;
2910     }
2911     h->i_frame++;
2912
2913     /* Slices used and PSNR */
2914     for( int i = 0; i < 5; i++ )
2915     {
2916         static const uint8_t slice_order[] = { SLICE_TYPE_I, SLICE_TYPE_SI, SLICE_TYPE_P, SLICE_TYPE_SP, SLICE_TYPE_B };
2917         static const char * const slice_name[] = { "P", "B", "I", "SP", "SI" };
2918         int i_slice = slice_order[i];
2919
2920         if( h->stat.i_frame_count[i_slice] > 0 )
2921         {
2922             const int i_count = h->stat.i_frame_count[i_slice];
2923             if( h->param.analyse.b_psnr )
2924             {
2925                 x264_log( h, X264_LOG_INFO,
2926                           "frame %s:%-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",
2927                           slice_name[i_slice],
2928                           i_count,
2929                           h->stat.f_frame_qp[i_slice] / i_count,
2930                           (double)h->stat.i_frame_size[i_slice] / i_count,
2931                           h->stat.f_psnr_mean_y[i_slice] / i_count, h->stat.f_psnr_mean_u[i_slice] / i_count, h->stat.f_psnr_mean_v[i_slice] / i_count,
2932                           h->stat.f_psnr_average[i_slice] / i_count,
2933                           x264_psnr( h->stat.i_ssd_global[i_slice], i_count * i_yuv_size ) );
2934             }
2935             else
2936             {
2937                 x264_log( h, X264_LOG_INFO,
2938                           "frame %s:%-5d Avg QP:%5.2f  size:%6.0f\n",
2939                           slice_name[i_slice],
2940                           i_count,
2941                           h->stat.f_frame_qp[i_slice] / i_count,
2942                           (double)h->stat.i_frame_size[i_slice] / i_count );
2943             }
2944         }
2945     }
2946     if( h->param.i_bframe && h->stat.i_frame_count[SLICE_TYPE_P] )
2947     {
2948         char *p = buf;
2949         int den = 0;
2950         // weight by number of frames (including the P-frame) that are in a sequence of N B-frames
2951         for( int i = 0; i <= h->param.i_bframe; i++ )
2952             den += (i+1) * h->stat.i_consecutive_bframes[i];
2953         for( int i = 0; i <= h->param.i_bframe; i++ )
2954             p += sprintf( p, " %4.1f%%", 100. * (i+1) * h->stat.i_consecutive_bframes[i] / den );
2955         x264_log( h, X264_LOG_INFO, "consecutive B-frames:%s\n", buf );
2956     }
2957
2958     for( int i_type = 0; i_type < 2; i_type++ )
2959         for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
2960         {
2961             if( i == D_DIRECT_8x8 ) continue; /* direct is counted as its own type */
2962             i_mb_count_size[i_type][x264_mb_partition_pixel_table[i]] += h->stat.i_mb_partition[i_type][i];
2963         }
2964
2965     /* MB types used */
2966     if( h->stat.i_frame_count[SLICE_TYPE_I] > 0 )
2967     {
2968         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_I];
2969         double i_count = h->stat.i_frame_count[SLICE_TYPE_I] * h->mb.i_mb_count / 100.0;
2970         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
2971         x264_log( h, X264_LOG_INFO, "mb I  %s\n", buf );
2972     }
2973     if( h->stat.i_frame_count[SLICE_TYPE_P] > 0 )
2974     {
2975         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_P];
2976         double i_count = h->stat.i_frame_count[SLICE_TYPE_P] * h->mb.i_mb_count / 100.0;
2977         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_P];
2978         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
2979         x264_log( h, X264_LOG_INFO,
2980                   "mb P  %s  P16..4: %4.1f%% %4.1f%% %4.1f%% %4.1f%% %4.1f%%    skip:%4.1f%%\n",
2981                   buf,
2982                   i_mb_size[PIXEL_16x16] / (i_count*4),
2983                   (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
2984                   i_mb_size[PIXEL_8x8] / (i_count*4),
2985                   (i_mb_size[PIXEL_8x4] + i_mb_size[PIXEL_4x8]) / (i_count*4),
2986                   i_mb_size[PIXEL_4x4] / (i_count*4),
2987                   i_mb_count[P_SKIP] / i_count );
2988     }
2989     if( h->stat.i_frame_count[SLICE_TYPE_B] > 0 )
2990     {
2991         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_B];
2992         double i_count = h->stat.i_frame_count[SLICE_TYPE_B] * h->mb.i_mb_count / 100.0;
2993         double i_mb_list_count;
2994         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_B];
2995         int64_t list_count[3] = {0}; /* 0 == L0, 1 == L1, 2 == BI */
2996         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
2997         for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
2998             for( int j = 0; j < 2; j++ )
2999             {
3000                 int l0 = x264_mb_type_list_table[i][0][j];
3001                 int l1 = x264_mb_type_list_table[i][1][j];
3002                 if( l0 || l1 )
3003                     list_count[l1+l0*l1] += h->stat.i_mb_count[SLICE_TYPE_B][i] * 2;
3004             }
3005         list_count[0] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L0_8x8];
3006         list_count[1] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L1_8x8];
3007         list_count[2] += h->stat.i_mb_partition[SLICE_TYPE_B][D_BI_8x8];
3008         i_mb_count[B_DIRECT] += (h->stat.i_mb_partition[SLICE_TYPE_B][D_DIRECT_8x8]+2)/4;
3009         i_mb_list_count = (list_count[0] + list_count[1] + list_count[2]) / 100.0;
3010         x264_log( h, X264_LOG_INFO,
3011                   "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",
3012                   buf,
3013                   i_mb_size[PIXEL_16x16] / (i_count*4),
3014                   (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
3015                   i_mb_size[PIXEL_8x8] / (i_count*4),
3016                   i_mb_count[B_DIRECT] / i_count,
3017                   i_mb_count[B_SKIP]   / i_count,
3018                   list_count[0] / i_mb_list_count,
3019                   list_count[1] / i_mb_list_count,
3020                   list_count[2] / i_mb_list_count );
3021     }
3022
3023     x264_ratecontrol_summary( h );
3024
3025     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 )
3026     {
3027 #define SUM3(p) (p[SLICE_TYPE_I] + p[SLICE_TYPE_P] + p[SLICE_TYPE_B])
3028 #define SUM3b(p,o) (p[SLICE_TYPE_I][o] + p[SLICE_TYPE_P][o] + p[SLICE_TYPE_B][o])
3029         int64_t i_i8x8 = SUM3b( h->stat.i_mb_count, I_8x8 );
3030         int64_t i_intra = i_i8x8 + SUM3b( h->stat.i_mb_count, I_4x4 )
3031                                  + SUM3b( h->stat.i_mb_count, I_16x16 );
3032         int64_t i_all_intra = i_intra + SUM3b( h->stat.i_mb_count, I_PCM);
3033         const int i_count = h->stat.i_frame_count[SLICE_TYPE_I] +
3034                             h->stat.i_frame_count[SLICE_TYPE_P] +
3035                             h->stat.i_frame_count[SLICE_TYPE_B];
3036         int64_t i_mb_count = i_count * h->mb.i_mb_count;
3037         float fps = (float) h->param.i_fps_num / h->param.i_fps_den;
3038         float f_bitrate;
3039         /* duration algorithm fails with one frame */
3040         if( !h->param.b_vfr_input || i_count == 1 )
3041             f_bitrate = fps * SUM3(h->stat.i_frame_size) / i_count / 125;
3042         else
3043         {
3044             float duration = (float)(2 * h->frames.i_largest_pts - h->frames.i_second_largest_pts) * h->param.i_timebase_num / h->param.i_timebase_den;
3045             f_bitrate = SUM3(h->stat.i_frame_size) / duration / 125;
3046         }
3047
3048         if( h->pps->b_transform_8x8_mode )
3049         {
3050             buf[0] = 0;
3051             if( h->stat.i_mb_count_8x8dct[0] )
3052                 sprintf( buf, " inter:%.1f%%", 100. * h->stat.i_mb_count_8x8dct[1] / h->stat.i_mb_count_8x8dct[0] );
3053             x264_log( h, X264_LOG_INFO, "8x8 transform intra:%.1f%%%s\n", 100. * i_i8x8 / i_intra, buf );
3054         }
3055
3056         if( (h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO ||
3057             (h->stat.i_direct_frames[0] && h->stat.i_direct_frames[1]))
3058             && h->stat.i_frame_count[SLICE_TYPE_B] )
3059         {
3060             x264_log( h, X264_LOG_INFO, "direct mvs  spatial:%.1f%% temporal:%.1f%%\n",
3061                       h->stat.i_direct_frames[1] * 100. / h->stat.i_frame_count[SLICE_TYPE_B],
3062                       h->stat.i_direct_frames[0] * 100. / h->stat.i_frame_count[SLICE_TYPE_B] );
3063         }
3064
3065         buf[0] = 0;
3066         if( i_mb_count != i_all_intra )
3067             sprintf( buf, " inter: %.1f%% %.1f%% %.1f%%",
3068                      h->stat.i_mb_cbp[1] * 100.0 / ((i_mb_count - i_all_intra)*4),
3069                      h->stat.i_mb_cbp[3] * 100.0 / ((i_mb_count - i_all_intra)  ),
3070                      h->stat.i_mb_cbp[5] * 100.0 / ((i_mb_count - i_all_intra)) );
3071         x264_log( h, X264_LOG_INFO, "coded y,uvDC,uvAC intra: %.1f%% %.1f%% %.1f%%%s\n",
3072                   h->stat.i_mb_cbp[0] * 100.0 / (i_all_intra*4),
3073                   h->stat.i_mb_cbp[2] * 100.0 / (i_all_intra  ),
3074                   h->stat.i_mb_cbp[4] * 100.0 / (i_all_intra  ), buf );
3075
3076         int64_t fixed_pred_modes[4][9] = {{0}};
3077         int64_t sum_pred_modes[4] = {0};
3078         for( int i = 0; i <= I_PRED_16x16_DC_128; i++ )
3079         {
3080             fixed_pred_modes[0][x264_mb_pred_mode16x16_fix[i]] += h->stat.i_mb_pred_mode[0][i];
3081             sum_pred_modes[0] += h->stat.i_mb_pred_mode[0][i];
3082         }
3083         if( sum_pred_modes[0] )
3084             x264_log( h, X264_LOG_INFO, "i16 v,h,dc,p: %2.0f%% %2.0f%% %2.0f%% %2.0f%%\n",
3085                       fixed_pred_modes[0][0] * 100.0 / sum_pred_modes[0],
3086                       fixed_pred_modes[0][1] * 100.0 / sum_pred_modes[0],
3087                       fixed_pred_modes[0][2] * 100.0 / sum_pred_modes[0],
3088                       fixed_pred_modes[0][3] * 100.0 / sum_pred_modes[0] );
3089         for( int i = 1; i <= 2; i++ )
3090         {
3091             for( int j = 0; j <= I_PRED_8x8_DC_128; j++ )
3092             {
3093                 fixed_pred_modes[i][x264_mb_pred_mode4x4_fix(j)] += h->stat.i_mb_pred_mode[i][j];
3094                 sum_pred_modes[i] += h->stat.i_mb_pred_mode[i][j];
3095             }
3096             if( sum_pred_modes[i] )
3097                 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,
3098                           fixed_pred_modes[i][0] * 100.0 / sum_pred_modes[i],
3099                           fixed_pred_modes[i][1] * 100.0 / sum_pred_modes[i],
3100                           fixed_pred_modes[i][2] * 100.0 / sum_pred_modes[i],
3101                           fixed_pred_modes[i][3] * 100.0 / sum_pred_modes[i],
3102                           fixed_pred_modes[i][4] * 100.0 / sum_pred_modes[i],
3103                           fixed_pred_modes[i][5] * 100.0 / sum_pred_modes[i],
3104                           fixed_pred_modes[i][6] * 100.0 / sum_pred_modes[i],
3105                           fixed_pred_modes[i][7] * 100.0 / sum_pred_modes[i],
3106                           fixed_pred_modes[i][8] * 100.0 / sum_pred_modes[i] );
3107         }
3108         for( int i = 0; i <= I_PRED_CHROMA_DC_128; i++ )
3109         {
3110             fixed_pred_modes[3][x264_mb_pred_mode8x8c_fix[i]] += h->stat.i_mb_pred_mode[3][i];
3111             sum_pred_modes[3] += h->stat.i_mb_pred_mode[3][i];
3112         }
3113         if( sum_pred_modes[3] )
3114             x264_log( h, X264_LOG_INFO, "i8c dc,h,v,p: %2.0f%% %2.0f%% %2.0f%% %2.0f%%\n",
3115                       fixed_pred_modes[3][0] * 100.0 / sum_pred_modes[3],
3116                       fixed_pred_modes[3][1] * 100.0 / sum_pred_modes[3],
3117                       fixed_pred_modes[3][2] * 100.0 / sum_pred_modes[3],
3118                       fixed_pred_modes[3][3] * 100.0 / sum_pred_modes[3] );
3119
3120         if( h->param.analyse.i_weighted_pred == X264_WEIGHTP_SMART && h->stat.i_frame_count[SLICE_TYPE_P] > 0 )
3121             x264_log( h, X264_LOG_INFO, "Weighted P-Frames: Y:%.1f%%\n",
3122                       h->stat.i_wpred[0] * 100.0 / h->stat.i_frame_count[SLICE_TYPE_P] );
3123
3124         for( int i_list = 0; i_list < 2; i_list++ )
3125             for( int i_slice = 0; i_slice < 2; i_slice++ )
3126             {
3127                 char *p = buf;
3128                 int64_t i_den = 0;
3129                 int i_max = 0;
3130                 for( int i = 0; i < 32; i++ )
3131                     if( h->stat.i_mb_count_ref[i_slice][i_list][i] )
3132                     {
3133                         i_den += h->stat.i_mb_count_ref[i_slice][i_list][i];
3134                         i_max = i;
3135                     }
3136                 if( i_max == 0 )
3137                     continue;
3138                 for( int i = 0; i <= i_max; i++ )
3139                     p += sprintf( p, " %4.1f%%", 100. * h->stat.i_mb_count_ref[i_slice][i_list][i] / i_den );
3140                 x264_log( h, X264_LOG_INFO, "ref %c L%d:%s\n", "PB"[i_slice], i_list, buf );
3141             }
3142
3143         if( h->param.analyse.b_ssim )
3144         {
3145             float ssim = SUM3( h->stat.f_ssim_mean_y ) / i_count;
3146             x264_log( h, X264_LOG_INFO, "SSIM Mean Y:%.7f (%6.3fdb)\n", ssim, x264_ssim( ssim ) );
3147         }
3148         if( h->param.analyse.b_psnr )
3149         {
3150             x264_log( h, X264_LOG_INFO,
3151                       "PSNR Mean Y:%6.3f U:%6.3f V:%6.3f Avg:%6.3f Global:%6.3f kb/s:%.2f\n",
3152                       SUM3( h->stat.f_psnr_mean_y ) / i_count,
3153                       SUM3( h->stat.f_psnr_mean_u ) / i_count,
3154                       SUM3( h->stat.f_psnr_mean_v ) / i_count,
3155                       SUM3( h->stat.f_psnr_average ) / i_count,
3156                       x264_psnr( SUM3( h->stat.i_ssd_global ), i_count * i_yuv_size ),
3157                       f_bitrate );
3158         }
3159         else
3160             x264_log( h, X264_LOG_INFO, "kb/s:%.2f\n", f_bitrate );
3161     }
3162
3163     /* rc */
3164     x264_ratecontrol_delete( h );
3165
3166     /* param */
3167     if( h->param.rc.psz_stat_out )
3168         free( h->param.rc.psz_stat_out );
3169     if( h->param.rc.psz_stat_in )
3170         free( h->param.rc.psz_stat_in );
3171
3172     x264_cqm_delete( h );
3173     x264_free( h->nal_buffer );
3174     x264_analyse_free_costs( h );
3175
3176     if( h->i_thread_frames > 1)
3177         h = h->thread[h->i_thread_phase];
3178
3179     /* frames */
3180     x264_frame_delete_list( h->frames.unused[0] );
3181     x264_frame_delete_list( h->frames.unused[1] );
3182     x264_frame_delete_list( h->frames.current );
3183     x264_frame_delete_list( h->frames.blank_unused );
3184
3185     h = h->thread[0];
3186
3187     for( int i = h->param.i_threads - 1; i >= 0; i-- )
3188     {
3189         x264_frame_t **frame;
3190
3191         if( !h->param.b_sliced_threads || i == 0 )
3192         {
3193             for( frame = h->thread[i]->frames.reference; *frame; frame++ )
3194             {
3195                 assert( (*frame)->i_reference_count > 0 );
3196                 (*frame)->i_reference_count--;
3197                 if( (*frame)->i_reference_count == 0 )
3198                     x264_frame_delete( *frame );
3199             }
3200             frame = &h->thread[i]->fdec;
3201             assert( (*frame)->i_reference_count > 0 );
3202             (*frame)->i_reference_count--;
3203             if( (*frame)->i_reference_count == 0 )
3204                 x264_frame_delete( *frame );
3205             x264_macroblock_cache_free( h->thread[i] );
3206         }
3207         x264_macroblock_thread_free( h->thread[i], 0 );
3208         x264_free( h->thread[i]->out.p_bitstream );
3209         x264_free( h->thread[i]->out.nal);
3210         x264_free( h->thread[i] );
3211     }
3212 }
3213
3214 /****************************************************************************
3215  * x264_encoder_delayed_frames:
3216  ****************************************************************************/
3217 int x264_encoder_delayed_frames( x264_t *h )
3218 {
3219     int delayed_frames = 0;
3220     if( h->i_thread_frames > 1 )
3221     {
3222         for( int i = 0; i < h->i_thread_frames; i++ )
3223             delayed_frames += h->thread[i]->b_thread_active;
3224         h = h->thread[h->i_thread_phase];
3225     }
3226     for( int i = 0; h->frames.current[i]; i++ )
3227         delayed_frames++;
3228     x264_pthread_mutex_lock( &h->lookahead->ofbuf.mutex );
3229     x264_pthread_mutex_lock( &h->lookahead->ifbuf.mutex );
3230     x264_pthread_mutex_lock( &h->lookahead->next.mutex );
3231     delayed_frames += h->lookahead->ifbuf.i_size + h->lookahead->next.i_size + h->lookahead->ofbuf.i_size;
3232     x264_pthread_mutex_unlock( &h->lookahead->next.mutex );
3233     x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
3234     x264_pthread_mutex_unlock( &h->lookahead->ofbuf.mutex );
3235     return delayed_frames;
3236 }