31cb84ae05b7d26abde0a6c52a505fdd545ef080
[x262.git] / 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  *          Jason Garrett-Glaser <darkshikari@gmail.com>
9  *
10  * This program is free software; you can redistribute it and/or modify
11  * it under the terms of the GNU General Public License as published by
12  * the Free Software Foundation; either version 2 of the License, or
13  * (at your option) any later version.
14  *
15  * This program is distributed in the hope that it will be useful,
16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
18  * GNU General Public License for more details.
19  *
20  * You should have received a copy of the GNU General Public License
21  * along with this program; if not, write to the Free Software
22  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02111, USA.
23  *****************************************************************************/
24
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     if( h->i_thread_frames > 1 )
431         h->param.nalu_process = NULL;
432
433     if( h->param.b_interlaced )
434     {
435         if( h->param.analyse.i_me_method >= X264_ME_ESA )
436         {
437             x264_log( h, X264_LOG_WARNING, "interlace + me=esa is not implemented\n" );
438             h->param.analyse.i_me_method = X264_ME_UMH;
439         }
440         if( h->param.analyse.i_weighted_pred > 0 )
441         {
442             x264_log( h, X264_LOG_WARNING, "interlace + weightp is not implemented\n" );
443             h->param.analyse.i_weighted_pred = X264_WEIGHTP_NONE;
444         }
445     }
446
447     /* Detect default ffmpeg settings and terminate with an error. */
448     {
449         int score = 0;
450         score += h->param.analyse.i_me_range == 0;
451         score += h->param.rc.i_qp_step == 3;
452         score += h->param.i_keyint_max == 12;
453         score += h->param.rc.i_qp_min == 2;
454         score += h->param.rc.i_qp_max == 31;
455         score += h->param.rc.f_qcompress == 0.5;
456         score += fabs(h->param.rc.f_ip_factor - 1.25) < 0.01;
457         score += fabs(h->param.rc.f_pb_factor - 1.25) < 0.01;
458         score += h->param.analyse.inter == 0 && h->param.analyse.i_subpel_refine == 8;
459         if( score >= 5 )
460         {
461             x264_log( h, X264_LOG_ERROR, "broken ffmpeg default settings detected\n" );
462             x264_log( h, X264_LOG_ERROR, "use an encoding preset (vpre)\n" );
463             return -1;
464         }
465     }
466
467     if( h->param.rc.i_rc_method < 0 || h->param.rc.i_rc_method > 2 )
468     {
469         x264_log( h, X264_LOG_ERROR, "no ratecontrol method specified\n" );
470         return -1;
471     }
472     h->param.rc.f_rf_constant = x264_clip3f( h->param.rc.f_rf_constant, 0, 51 );
473     h->param.rc.i_qp_constant = x264_clip3( h->param.rc.i_qp_constant, 0, 51 );
474     if( h->param.rc.i_rc_method == X264_RC_CRF )
475     {
476         h->param.rc.i_qp_constant = h->param.rc.f_rf_constant;
477         h->param.rc.i_bitrate = 0;
478     }
479     if( (h->param.rc.i_rc_method == X264_RC_CQP || h->param.rc.i_rc_method == X264_RC_CRF)
480         && h->param.rc.i_qp_constant == 0 )
481     {
482         h->mb.b_lossless = 1;
483         h->param.i_cqm_preset = X264_CQM_FLAT;
484         h->param.psz_cqm_file = NULL;
485         h->param.rc.i_rc_method = X264_RC_CQP;
486         h->param.rc.f_ip_factor = 1;
487         h->param.rc.f_pb_factor = 1;
488         h->param.analyse.b_psnr = 0;
489         h->param.analyse.b_ssim = 0;
490         h->param.analyse.i_chroma_qp_offset = 0;
491         h->param.analyse.i_trellis = 0;
492         h->param.analyse.b_fast_pskip = 0;
493         h->param.analyse.i_noise_reduction = 0;
494         h->param.analyse.b_psy = 0;
495         h->param.i_bframe = 0;
496         /* 8x8dct is not useful at all in CAVLC lossless */
497         if( !h->param.b_cabac )
498             h->param.analyse.b_transform_8x8 = 0;
499     }
500     if( h->param.rc.i_rc_method == X264_RC_CQP )
501     {
502         float qp_p = h->param.rc.i_qp_constant;
503         float qp_i = qp_p - 6*log2f( h->param.rc.f_ip_factor );
504         float qp_b = qp_p + 6*log2f( h->param.rc.f_pb_factor );
505         h->param.rc.i_qp_min = x264_clip3( (int)(X264_MIN3( qp_p, qp_i, qp_b )), 0, 51 );
506         h->param.rc.i_qp_max = x264_clip3( (int)(X264_MAX3( qp_p, qp_i, qp_b ) + .999), 0, 51 );
507         h->param.rc.i_aq_mode = 0;
508         h->param.rc.b_mb_tree = 0;
509     }
510     h->param.rc.i_qp_max = x264_clip3( h->param.rc.i_qp_max, 0, 51 );
511     h->param.rc.i_qp_min = x264_clip3( h->param.rc.i_qp_min, 0, h->param.rc.i_qp_max );
512     if( h->param.rc.i_vbv_buffer_size )
513     {
514         if( h->param.rc.i_rc_method == X264_RC_CQP )
515         {
516             x264_log( h, X264_LOG_WARNING, "VBV is incompatible with constant QP, ignored.\n" );
517             h->param.rc.i_vbv_max_bitrate = 0;
518             h->param.rc.i_vbv_buffer_size = 0;
519         }
520         else if( h->param.rc.i_vbv_max_bitrate == 0 )
521         {
522             if( h->param.rc.i_rc_method == X264_RC_ABR )
523             {
524                 x264_log( h, X264_LOG_WARNING, "VBV maxrate unspecified, assuming CBR\n" );
525                 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate;
526             }
527             else
528             {
529                 x264_log( h, X264_LOG_WARNING, "VBV bufsize set but maxrate unspecified, ignored\n" );
530                 h->param.rc.i_vbv_buffer_size = 0;
531             }
532         }
533         else if( h->param.rc.i_vbv_max_bitrate < h->param.rc.i_bitrate &&
534                  h->param.rc.i_rc_method == X264_RC_ABR )
535         {
536             x264_log( h, X264_LOG_WARNING, "max bitrate less than average bitrate, assuming CBR\n" );
537             h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate;
538         }
539     }
540     else if( h->param.rc.i_vbv_max_bitrate )
541     {
542         x264_log( h, X264_LOG_WARNING, "VBV maxrate specified, but no bufsize, ignored\n" );
543         h->param.rc.i_vbv_max_bitrate = 0;
544     }
545
546     if( h->param.b_interlaced && h->param.i_slice_max_size )
547     {
548         x264_log( h, X264_LOG_WARNING, "interlaced + slice-max-size is not implemented\n" );
549         h->param.i_slice_max_size = 0;
550     }
551     if( h->param.b_interlaced && h->param.i_slice_max_mbs )
552     {
553         x264_log( h, X264_LOG_WARNING, "interlaced + slice-max-mbs is not implemented\n" );
554         h->param.i_slice_max_mbs = 0;
555     }
556     int max_slices = (h->param.i_height+((16<<h->param.b_interlaced)-1))/(16<<h->param.b_interlaced);
557     if( h->param.b_sliced_threads )
558         h->param.i_slice_count = x264_clip3( h->param.i_threads, 0, max_slices );
559     else
560     {
561         h->param.i_slice_count = x264_clip3( h->param.i_slice_count, 0, max_slices );
562         h->param.i_slice_max_size = X264_MAX( h->param.i_slice_max_size, 0 );
563         h->param.i_slice_max_mbs = X264_MAX( h->param.i_slice_max_mbs, 0 );
564         if( h->param.i_slice_max_mbs || h->param.i_slice_max_size )
565             h->param.i_slice_count = 0;
566     }
567
568     h->param.i_frame_reference = x264_clip3( h->param.i_frame_reference, 1, 16 );
569     h->param.i_dpb_size = x264_clip3( h->param.i_dpb_size, 1, 16 );
570     h->param.i_keyint_max = x264_clip3( h->param.i_keyint_max, 1, X264_KEYINT_MAX_INFINITE );
571     if( h->param.i_scenecut_threshold < 0 )
572         h->param.i_scenecut_threshold = 0;
573     if( !h->param.analyse.i_subpel_refine && h->param.analyse.i_direct_mv_pred > X264_DIRECT_PRED_SPATIAL )
574     {
575         x264_log( h, X264_LOG_WARNING, "subme=0 + direct=temporal is not supported\n" );
576         h->param.analyse.i_direct_mv_pred = X264_DIRECT_PRED_SPATIAL;
577     }
578     h->param.i_bframe = x264_clip3( h->param.i_bframe, 0, X264_MIN( X264_BFRAME_MAX, h->param.i_keyint_max-1 ) );
579     h->param.i_open_gop = x264_clip3( h->param.i_open_gop, X264_OPEN_GOP_NONE, X264_OPEN_GOP_BLURAY );
580     if( h->param.i_keyint_max == 1 )
581         h->param.b_intra_refresh = 0;
582     h->param.i_bframe_bias = x264_clip3( h->param.i_bframe_bias, -90, 100 );
583     if( h->param.i_bframe <= 1 )
584         h->param.i_bframe_pyramid = X264_B_PYRAMID_NONE;
585     h->param.i_bframe_pyramid = x264_clip3( h->param.i_bframe_pyramid, X264_B_PYRAMID_NONE, X264_B_PYRAMID_NORMAL );
586     if( !h->param.i_bframe )
587     {
588         h->param.i_bframe_adaptive = X264_B_ADAPT_NONE;
589         h->param.analyse.i_direct_mv_pred = 0;
590         h->param.analyse.b_weighted_bipred = 0;
591         h->param.i_open_gop = X264_OPEN_GOP_NONE;
592     }
593     if( h->param.b_intra_refresh && h->param.i_bframe_pyramid == X264_B_PYRAMID_NORMAL )
594     {
595         x264_log( h, X264_LOG_WARNING, "b-pyramid normal + intra-refresh is not supported\n" );
596         h->param.i_bframe_pyramid = X264_B_PYRAMID_STRICT;
597     }
598     if( h->param.b_intra_refresh && (h->param.i_frame_reference > 1 || h->param.i_dpb_size > 1) )
599     {
600         x264_log( h, X264_LOG_WARNING, "ref > 1 + intra-refresh is not supported\n" );
601         h->param.i_frame_reference = 1;
602         h->param.i_dpb_size = 1;
603     }
604     if( h->param.b_intra_refresh && h->param.i_open_gop )
605     {
606         x264_log( h, X264_LOG_WARNING, "intra-refresh is not compatible with open-gop\n" );
607         h->param.i_open_gop = X264_OPEN_GOP_NONE;
608     }
609     if( h->param.i_keyint_min == X264_KEYINT_MIN_AUTO )
610         h->param.i_keyint_min = h->param.i_keyint_max / 10;
611     h->param.i_keyint_min = x264_clip3( h->param.i_keyint_min, 1, h->param.i_keyint_max/2+1 );
612     h->param.rc.i_lookahead = x264_clip3( h->param.rc.i_lookahead, 0, X264_LOOKAHEAD_MAX );
613     {
614         int maxrate = X264_MAX( h->param.rc.i_vbv_max_bitrate, h->param.rc.i_bitrate );
615         float bufsize = maxrate ? (float)h->param.rc.i_vbv_buffer_size / maxrate : 0;
616         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;
617         h->param.rc.i_lookahead = X264_MIN( h->param.rc.i_lookahead, X264_MAX( h->param.i_keyint_max, bufsize*fps ) );
618     }
619
620     if( !h->param.i_timebase_num || !h->param.i_timebase_den )
621     {
622         h->param.i_timebase_num = h->param.i_fps_den;
623         h->param.i_timebase_den = h->param.i_fps_num;
624     }
625
626     h->param.rc.f_qcompress = x264_clip3f( h->param.rc.f_qcompress, 0.0, 1.0 );
627     if( h->param.i_keyint_max == 1 || h->param.rc.f_qcompress == 1 )
628         h->param.rc.b_mb_tree = 0;
629     if( (!h->param.b_intra_refresh && h->param.i_keyint_max != X264_KEYINT_MAX_INFINITE) &&
630         !h->param.rc.i_lookahead && h->param.rc.b_mb_tree )
631     {
632         x264_log( h, X264_LOG_WARNING, "lookaheadless mb-tree requires intra refresh or infinite keyint\n" );
633         h->param.rc.b_mb_tree = 0;
634     }
635     if( h->param.rc.b_stat_read )
636         h->param.rc.i_lookahead = 0;
637 #if HAVE_PTHREAD
638     if( h->param.i_sync_lookahead < 0 )
639         h->param.i_sync_lookahead = h->param.i_bframe + 1;
640     h->param.i_sync_lookahead = X264_MIN( h->param.i_sync_lookahead, X264_LOOKAHEAD_MAX );
641     if( h->param.rc.b_stat_read || h->i_thread_frames == 1 )
642         h->param.i_sync_lookahead = 0;
643 #else
644     h->param.i_sync_lookahead = 0;
645 #endif
646
647     h->param.i_deblocking_filter_alphac0 = x264_clip3( h->param.i_deblocking_filter_alphac0, -6, 6 );
648     h->param.i_deblocking_filter_beta    = x264_clip3( h->param.i_deblocking_filter_beta, -6, 6 );
649     h->param.analyse.i_luma_deadzone[0] = x264_clip3( h->param.analyse.i_luma_deadzone[0], 0, 32 );
650     h->param.analyse.i_luma_deadzone[1] = x264_clip3( h->param.analyse.i_luma_deadzone[1], 0, 32 );
651
652     h->param.i_cabac_init_idc = x264_clip3( h->param.i_cabac_init_idc, 0, 2 );
653
654     if( h->param.i_cqm_preset < X264_CQM_FLAT || h->param.i_cqm_preset > X264_CQM_CUSTOM )
655         h->param.i_cqm_preset = X264_CQM_FLAT;
656
657     if( h->param.analyse.i_me_method < X264_ME_DIA ||
658         h->param.analyse.i_me_method > X264_ME_TESA )
659         h->param.analyse.i_me_method = X264_ME_HEX;
660     if( h->param.analyse.i_me_range < 4 )
661         h->param.analyse.i_me_range = 4;
662     if( h->param.analyse.i_me_range > 16 && h->param.analyse.i_me_method <= X264_ME_HEX )
663         h->param.analyse.i_me_range = 16;
664     if( h->param.analyse.i_me_method == X264_ME_TESA &&
665         (h->mb.b_lossless || h->param.analyse.i_subpel_refine <= 1) )
666         h->param.analyse.i_me_method = X264_ME_ESA;
667     h->param.analyse.i_subpel_refine = x264_clip3( h->param.analyse.i_subpel_refine, 0, 10 );
668     h->param.analyse.b_mixed_references = h->param.analyse.b_mixed_references && h->param.i_frame_reference > 1;
669     h->param.analyse.inter &= X264_ANALYSE_PSUB16x16|X264_ANALYSE_PSUB8x8|X264_ANALYSE_BSUB16x16|
670                               X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
671     h->param.analyse.intra &= X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
672     if( !(h->param.analyse.inter & X264_ANALYSE_PSUB16x16) )
673         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
674     if( !h->param.analyse.b_transform_8x8 )
675     {
676         h->param.analyse.inter &= ~X264_ANALYSE_I8x8;
677         h->param.analyse.intra &= ~X264_ANALYSE_I8x8;
678     }
679     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -12, 12);
680     if( !h->param.b_cabac )
681         h->param.analyse.i_trellis = 0;
682     h->param.analyse.i_trellis = x264_clip3( h->param.analyse.i_trellis, 0, 2 );
683     if( !h->param.analyse.b_psy )
684     {
685         h->param.analyse.f_psy_rd = 0;
686         h->param.analyse.f_psy_trellis = 0;
687     }
688     if( !h->param.analyse.i_trellis )
689         h->param.analyse.f_psy_trellis = 0;
690     h->param.analyse.f_psy_rd = x264_clip3f( h->param.analyse.f_psy_rd, 0, 10 );
691     h->param.analyse.f_psy_trellis = x264_clip3f( h->param.analyse.f_psy_trellis, 0, 10 );
692     if( h->param.analyse.i_subpel_refine < 6 )
693         h->param.analyse.f_psy_rd = 0;
694     h->mb.i_psy_rd = FIX8( h->param.analyse.f_psy_rd );
695     /* Psy RDO increases overall quantizers to improve the quality of luma--this indirectly hurts chroma quality */
696     /* so we lower the chroma QP offset to compensate */
697     /* This can be triggered repeatedly on multiple calls to parameter_validate, but since encoding
698      * uses the pps chroma qp offset not the param chroma qp offset, this is not a problem. */
699     if( h->mb.i_psy_rd )
700         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_rd < 0.25 ? 1 : 2;
701     h->mb.i_psy_trellis = FIX8( h->param.analyse.f_psy_trellis / 4 );
702     /* Psy trellis has a similar effect. */
703     if( h->mb.i_psy_trellis )
704         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_trellis < 0.25 ? 1 : 2;
705     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -12, 12);
706     h->param.rc.i_aq_mode = x264_clip3( h->param.rc.i_aq_mode, 0, 2 );
707     h->param.rc.f_aq_strength = x264_clip3f( h->param.rc.f_aq_strength, 0, 3 );
708     if( h->param.rc.f_aq_strength == 0 )
709         h->param.rc.i_aq_mode = 0;
710     /* MB-tree requires AQ to be on, even if the strength is zero. */
711     if( !h->param.rc.i_aq_mode && h->param.rc.b_mb_tree )
712     {
713         h->param.rc.i_aq_mode = 1;
714         h->param.rc.f_aq_strength = 0;
715     }
716     h->param.analyse.i_noise_reduction = x264_clip3( h->param.analyse.i_noise_reduction, 0, 1<<16 );
717     if( h->param.analyse.i_subpel_refine == 10 && (h->param.analyse.i_trellis != 2 || !h->param.rc.i_aq_mode) )
718         h->param.analyse.i_subpel_refine = 9;
719
720     {
721         const x264_level_t *l = x264_levels;
722         if( h->param.i_level_idc < 0 )
723         {
724             int maxrate_bak = h->param.rc.i_vbv_max_bitrate;
725             if( h->param.rc.i_rc_method == X264_RC_ABR && h->param.rc.i_vbv_buffer_size <= 0 )
726                 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate * 2;
727             h->sps = h->sps_array;
728             x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
729             do h->param.i_level_idc = l->level_idc;
730                 while( l[1].level_idc && x264_validate_levels( h, 0 ) && l++ );
731             h->param.rc.i_vbv_max_bitrate = maxrate_bak;
732         }
733         else
734         {
735             while( l->level_idc && l->level_idc != h->param.i_level_idc )
736                 l++;
737             if( l->level_idc == 0 )
738             {
739                 x264_log( h, X264_LOG_ERROR, "invalid level_idc: %d\n", h->param.i_level_idc );
740                 return -1;
741             }
742         }
743         if( h->param.analyse.i_mv_range <= 0 )
744             h->param.analyse.i_mv_range = l->mv_range >> h->param.b_interlaced;
745         else
746             h->param.analyse.i_mv_range = x264_clip3(h->param.analyse.i_mv_range, 32, 512 >> h->param.b_interlaced);
747     }
748
749     h->param.analyse.i_weighted_pred = x264_clip3( h->param.analyse.i_weighted_pred, 0, X264_WEIGHTP_SMART );
750     if( !h->param.analyse.i_weighted_pred && h->param.rc.b_mb_tree && h->param.analyse.b_psy && !h->param.b_interlaced )
751         h->param.analyse.i_weighted_pred = X264_WEIGHTP_FAKE;
752
753     if( h->i_thread_frames > 1 )
754     {
755         int r = h->param.analyse.i_mv_range_thread;
756         int r2;
757         if( r <= 0 )
758         {
759             // half of the available space is reserved and divided evenly among the threads,
760             // the rest is allocated to whichever thread is far enough ahead to use it.
761             // reserving more space increases quality for some videos, but costs more time
762             // in thread synchronization.
763             int max_range = (h->param.i_height + X264_THREAD_HEIGHT) / h->i_thread_frames - X264_THREAD_HEIGHT;
764             r = max_range / 2;
765         }
766         r = X264_MAX( r, h->param.analyse.i_me_range );
767         r = X264_MIN( r, h->param.analyse.i_mv_range );
768         // round up to use the whole mb row
769         r2 = (r & ~15) + ((-X264_THREAD_HEIGHT) & 15);
770         if( r2 < r )
771             r2 += 16;
772         x264_log( h, X264_LOG_DEBUG, "using mv_range_thread = %d\n", r2 );
773         h->param.analyse.i_mv_range_thread = r2;
774     }
775
776     if( h->param.rc.f_qblur < 0 )
777         h->param.rc.f_qblur = 0;
778     if( h->param.rc.f_complexity_blur < 0 )
779         h->param.rc.f_complexity_blur = 0;
780
781     h->param.i_sps_id &= 31;
782
783     if( h->param.i_log_level < X264_LOG_INFO )
784     {
785         h->param.analyse.b_psnr = 0;
786         h->param.analyse.b_ssim = 0;
787     }
788
789     if( h->param.b_interlaced )
790         h->param.b_pic_struct = 1;
791
792     if( h->param.i_nal_hrd && !h->param.rc.i_vbv_buffer_size )
793     {
794         x264_log( h, X264_LOG_WARNING, "NAL HRD parameters require VBV parameters\n" );
795         h->param.i_nal_hrd = X264_NAL_HRD_NONE;
796     }
797
798     if( h->param.i_nal_hrd == X264_NAL_HRD_CBR &&
799        (h->param.rc.i_bitrate != h->param.rc.i_vbv_max_bitrate || !h->param.rc.i_vbv_max_bitrate) )
800     {
801         x264_log( h, X264_LOG_WARNING, "CBR HRD requires constant bitrate\n" );
802         h->param.i_nal_hrd = X264_NAL_HRD_VBR;
803     }
804
805     /* ensure the booleans are 0 or 1 so they can be used in math */
806 #define BOOLIFY(x) h->param.x = !!h->param.x
807     BOOLIFY( b_cabac );
808     BOOLIFY( b_constrained_intra );
809     BOOLIFY( b_deblocking_filter );
810     BOOLIFY( b_deterministic );
811     BOOLIFY( b_sliced_threads );
812     BOOLIFY( b_interlaced );
813     BOOLIFY( b_intra_refresh );
814     BOOLIFY( b_visualize );
815     BOOLIFY( b_aud );
816     BOOLIFY( b_repeat_headers );
817     BOOLIFY( b_annexb );
818     BOOLIFY( b_vfr_input );
819     BOOLIFY( b_pic_struct );
820     BOOLIFY( b_fake_interlaced );
821     BOOLIFY( analyse.b_transform_8x8 );
822     BOOLIFY( analyse.b_weighted_bipred );
823     BOOLIFY( analyse.b_chroma_me );
824     BOOLIFY( analyse.b_mixed_references );
825     BOOLIFY( analyse.b_fast_pskip );
826     BOOLIFY( analyse.b_dct_decimate );
827     BOOLIFY( analyse.b_psy );
828     BOOLIFY( analyse.b_psnr );
829     BOOLIFY( analyse.b_ssim );
830     BOOLIFY( rc.b_stat_write );
831     BOOLIFY( rc.b_stat_read );
832     BOOLIFY( rc.b_mb_tree );
833 #undef BOOLIFY
834
835     return 0;
836 }
837
838 static void mbcmp_init( x264_t *h )
839 {
840     int satd = !h->mb.b_lossless && h->param.analyse.i_subpel_refine > 1;
841     memcpy( h->pixf.mbcmp, satd ? h->pixf.satd : h->pixf.sad_aligned, sizeof(h->pixf.mbcmp) );
842     memcpy( h->pixf.mbcmp_unaligned, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.mbcmp_unaligned) );
843     h->pixf.intra_mbcmp_x3_16x16 = satd ? h->pixf.intra_satd_x3_16x16 : h->pixf.intra_sad_x3_16x16;
844     h->pixf.intra_mbcmp_x3_8x8c = satd ? h->pixf.intra_satd_x3_8x8c : h->pixf.intra_sad_x3_8x8c;
845     h->pixf.intra_mbcmp_x3_8x8 = satd ? h->pixf.intra_sa8d_x3_8x8 : h->pixf.intra_sad_x3_8x8;
846     h->pixf.intra_mbcmp_x3_4x4 = satd ? h->pixf.intra_satd_x3_4x4 : h->pixf.intra_sad_x3_4x4;
847     satd &= h->param.analyse.i_me_method == X264_ME_TESA;
848     memcpy( h->pixf.fpelcmp, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.fpelcmp) );
849     memcpy( h->pixf.fpelcmp_x3, satd ? h->pixf.satd_x3 : h->pixf.sad_x3, sizeof(h->pixf.fpelcmp_x3) );
850     memcpy( h->pixf.fpelcmp_x4, satd ? h->pixf.satd_x4 : h->pixf.sad_x4, sizeof(h->pixf.fpelcmp_x4) );
851 }
852
853 static void x264_set_aspect_ratio( x264_t *h, x264_param_t *param, int initial )
854 {
855     /* VUI */
856     if( param->vui.i_sar_width > 0 && param->vui.i_sar_height > 0 )
857     {
858         uint32_t i_w = param->vui.i_sar_width;
859         uint32_t i_h = param->vui.i_sar_height;
860         uint32_t old_w = h->param.vui.i_sar_width;
861         uint32_t old_h = h->param.vui.i_sar_height;
862
863         x264_reduce_fraction( &i_w, &i_h );
864
865         while( i_w > 65535 || i_h > 65535 )
866         {
867             i_w /= 2;
868             i_h /= 2;
869         }
870
871         x264_reduce_fraction( &i_w, &i_h );
872
873         if( i_w != old_w || i_h != old_h || initial )
874         {
875             h->param.vui.i_sar_width = 0;
876             h->param.vui.i_sar_height = 0;
877             if( i_w == 0 || i_h == 0 )
878                 x264_log( h, X264_LOG_WARNING, "cannot create valid sample aspect ratio\n" );
879             else
880             {
881                 x264_log( h, initial?X264_LOG_INFO:X264_LOG_DEBUG, "using SAR=%d/%d\n", i_w, i_h );
882                 h->param.vui.i_sar_width = i_w;
883                 h->param.vui.i_sar_height = i_h;
884             }
885         }
886     }
887 }
888
889 /****************************************************************************
890  * x264_encoder_open:
891  ****************************************************************************/
892 x264_t *x264_encoder_open( x264_param_t *param )
893 {
894     x264_t *h;
895     char buf[1000], *p;
896     int qp, i_slicetype_length;
897
898     CHECKED_MALLOCZERO( h, sizeof(x264_t) );
899
900     /* Create a copy of param */
901     memcpy( &h->param, param, sizeof(x264_param_t) );
902
903     if( param->param_free )
904         param->param_free( param );
905
906     if( x264_validate_parameters( h ) < 0 )
907         goto fail;
908
909     if( h->param.psz_cqm_file )
910         if( x264_cqm_parse_file( h, h->param.psz_cqm_file ) < 0 )
911             goto fail;
912
913     if( h->param.rc.psz_stat_out )
914         h->param.rc.psz_stat_out = strdup( h->param.rc.psz_stat_out );
915     if( h->param.rc.psz_stat_in )
916         h->param.rc.psz_stat_in = strdup( h->param.rc.psz_stat_in );
917
918     x264_set_aspect_ratio( h, &h->param, 1 );
919
920     x264_reduce_fraction( &h->param.i_fps_num, &h->param.i_fps_den );
921     x264_reduce_fraction( &h->param.i_timebase_num, &h->param.i_timebase_den );
922
923     /* Init x264_t */
924     h->i_frame = -1;
925     h->i_frame_num = 0;
926     h->i_idr_pic_id = 0;
927     uint64_t new_timebase_den = h->param.i_timebase_den;
928     if( h->param.b_dts_compress )
929     {
930         /* h->i_dts_compress_multiplier == h->frames.i_bframe_delay + 1 */
931         h->i_dts_compress_multiplier = h->param.i_bframe ? (h->param.i_bframe_pyramid ? 3 : 2) : 1;
932         if( h->i_dts_compress_multiplier != 1 )
933         {
934             new_timebase_den = h->param.i_timebase_den * h->i_dts_compress_multiplier;
935             x264_log( h, X264_LOG_DEBUG, "DTS compresion changed timebase: %u/%u -> %u/%"PRIu64"\n",
936                       h->param.i_timebase_num, h->param.i_timebase_den,
937                       h->param.i_timebase_num, new_timebase_den );
938         }
939     }
940     else
941         h->i_dts_compress_multiplier = 1;
942
943     if( new_timebase_den * 2 > UINT32_MAX )
944     {
945         x264_log( h, X264_LOG_ERROR, "Effective timebase denominator %"PRIu64" exceeds H.264 maximum\n", new_timebase_den );
946         goto fail;
947     }
948     h->param.i_timebase_den = new_timebase_den;
949
950     h->sps = &h->sps_array[0];
951     x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
952
953     h->pps = &h->pps_array[0];
954     x264_pps_init( h->pps, h->param.i_sps_id, &h->param, h->sps );
955
956     x264_validate_levels( h, 1 );
957
958     h->chroma_qp_table = i_chroma_qp_table + 12 + h->pps->i_chroma_qp_index_offset;
959
960     if( x264_cqm_init( h ) < 0 )
961         goto fail;
962
963     h->mb.i_mb_width = h->sps->i_mb_width;
964     h->mb.i_mb_height = h->sps->i_mb_height;
965     h->mb.i_mb_count = h->mb.i_mb_width * h->mb.i_mb_height;
966
967     /* Init frames. */
968     if( h->param.i_bframe_adaptive == X264_B_ADAPT_TRELLIS && !h->param.rc.b_stat_read )
969         h->frames.i_delay = X264_MAX(h->param.i_bframe,3)*4;
970     else
971         h->frames.i_delay = h->param.i_bframe;
972     if( h->param.rc.b_mb_tree || h->param.rc.i_vbv_buffer_size )
973         h->frames.i_delay = X264_MAX( h->frames.i_delay, h->param.rc.i_lookahead );
974     i_slicetype_length = h->frames.i_delay;
975     h->frames.i_delay += h->i_thread_frames - 1;
976     h->frames.i_delay += h->param.i_sync_lookahead;
977     h->frames.i_delay += h->param.b_vfr_input && (h->param.rc.i_rc_method == X264_RC_ABR || h->param.rc.b_stat_write
978                                                  || h->param.rc.i_vbv_buffer_size);
979     h->frames.i_bframe_delay = h->param.i_bframe ? (h->param.i_bframe_pyramid ? 2 : 1) : 0;
980
981     h->frames.i_max_ref0 = h->param.i_frame_reference;
982     h->frames.i_max_ref1 = X264_MIN( h->sps->vui.i_num_reorder_frames, h->param.i_frame_reference );
983     h->frames.i_max_dpb  = h->sps->vui.i_max_dec_frame_buffering;
984     h->frames.b_have_lowres = !h->param.rc.b_stat_read
985         && ( h->param.rc.i_rc_method == X264_RC_ABR
986           || h->param.rc.i_rc_method == X264_RC_CRF
987           || h->param.i_bframe_adaptive
988           || h->param.i_scenecut_threshold
989           || h->param.rc.b_mb_tree
990           || h->param.analyse.i_weighted_pred == X264_WEIGHTP_SMART );
991     h->frames.b_have_lowres |= h->param.rc.b_stat_read && h->param.rc.i_vbv_buffer_size > 0;
992     h->frames.b_have_sub8x8_esa = !!(h->param.analyse.inter & X264_ANALYSE_PSUB8x8);
993
994     h->frames.i_last_idr =
995     h->frames.i_last_keyframe = - h->param.i_keyint_max;
996     h->frames.i_input    = 0;
997     h->frames.i_largest_pts = h->frames.i_second_largest_pts = -1;
998     h->frames.i_poc_last_open_gop = -1;
999
1000     CHECKED_MALLOCZERO( h->frames.unused[0], (h->frames.i_delay + 3) * sizeof(x264_frame_t *) );
1001     /* Allocate room for max refs plus a few extra just in case. */
1002     CHECKED_MALLOCZERO( h->frames.unused[1], (h->i_thread_frames + 20) * sizeof(x264_frame_t *) );
1003     CHECKED_MALLOCZERO( h->frames.current, (h->param.i_sync_lookahead + h->param.i_bframe
1004                         + h->i_thread_frames + 3) * sizeof(x264_frame_t *) );
1005     if( h->param.analyse.i_weighted_pred > 0 )
1006         CHECKED_MALLOCZERO( h->frames.blank_unused, h->i_thread_frames * 4 * sizeof(x264_frame_t *) );
1007     h->i_ref0 = 0;
1008     h->i_ref1 = 0;
1009     h->i_cpb_delay = h->i_coded_fields = h->i_disp_fields = h->i_prev_duration = 0;
1010     h->i_disp_fields_last_frame = -1;
1011     x264_rdo_init();
1012
1013     /* init CPU functions */
1014     x264_predict_16x16_init( h->param.cpu, h->predict_16x16 );
1015     x264_predict_8x8c_init( h->param.cpu, h->predict_8x8c );
1016     x264_predict_8x8_init( h->param.cpu, h->predict_8x8, &h->predict_8x8_filter );
1017     x264_predict_4x4_init( h->param.cpu, h->predict_4x4 );
1018     if( !h->param.b_cabac )
1019         x264_init_vlc_tables();
1020     x264_pixel_init( h->param.cpu, &h->pixf );
1021     x264_dct_init( h->param.cpu, &h->dctf );
1022     x264_zigzag_init( h->param.cpu, &h->zigzagf, h->param.b_interlaced );
1023     x264_mc_init( h->param.cpu, &h->mc );
1024     x264_quant_init( h, h->param.cpu, &h->quantf );
1025     x264_deblock_init( h->param.cpu, &h->loopf );
1026     x264_bitstream_init( h->param.cpu, &h->bsf );
1027     x264_dct_init_weights();
1028
1029     mbcmp_init( h );
1030
1031     p = buf + sprintf( buf, "using cpu capabilities:" );
1032     for( int i = 0; x264_cpu_names[i].flags; i++ )
1033     {
1034         if( !strcmp(x264_cpu_names[i].name, "SSE2")
1035             && h->param.cpu & (X264_CPU_SSE2_IS_FAST|X264_CPU_SSE2_IS_SLOW) )
1036             continue;
1037         if( !strcmp(x264_cpu_names[i].name, "SSE3")
1038             && (h->param.cpu & X264_CPU_SSSE3 || !(h->param.cpu & X264_CPU_CACHELINE_64)) )
1039             continue;
1040         if( !strcmp(x264_cpu_names[i].name, "SSE4.1")
1041             && (h->param.cpu & X264_CPU_SSE42) )
1042             continue;
1043         if( (h->param.cpu & x264_cpu_names[i].flags) == x264_cpu_names[i].flags
1044             && (!i || x264_cpu_names[i].flags != x264_cpu_names[i-1].flags) )
1045             p += sprintf( p, " %s", x264_cpu_names[i].name );
1046     }
1047     if( !h->param.cpu )
1048         p += sprintf( p, " none!" );
1049     x264_log( h, X264_LOG_INFO, "%s\n", buf );
1050
1051     for( qp = h->param.rc.i_qp_min; qp <= h->param.rc.i_qp_max; qp++ )
1052         if( x264_analyse_init_costs( h, qp ) )
1053             goto fail;
1054     if( x264_analyse_init_costs( h, X264_LOOKAHEAD_QP ) )
1055         goto fail;
1056
1057     /* Checks for known miscompilation issues. */
1058     if( h->cost_mv[1][2013] != 24 )
1059     {
1060         x264_log( h, X264_LOG_ERROR, "MV cost test failed: x264 has been miscompiled!\n" );
1061         goto fail;
1062     }
1063
1064     /* Must be volatile or else GCC will optimize it out. */
1065     volatile int temp = 392;
1066     if( x264_clz( temp ) != 23 )
1067     {
1068         x264_log( h, X264_LOG_ERROR, "CLZ test failed: x264 has been miscompiled!\n" );
1069 #if ARCH_X86 || ARCH_X86_64
1070         x264_log( h, X264_LOG_ERROR, "Are you attempting to run an SSE4a-targeted build on a CPU that\n" );
1071         x264_log( h, X264_LOG_ERROR, "doesn't support it?\n" );
1072 #endif
1073         goto fail;
1074     }
1075
1076     h->out.i_nal = 0;
1077     h->out.i_bitstream = X264_MAX( 1000000, h->param.i_width * h->param.i_height * 4
1078         * ( h->param.rc.i_rc_method == X264_RC_ABR ? pow( 0.95, h->param.rc.i_qp_min )
1079           : pow( 0.95, h->param.rc.i_qp_constant ) * X264_MAX( 1, h->param.rc.f_ip_factor )));
1080
1081     CHECKED_MALLOC( h->nal_buffer, h->out.i_bitstream * 3/2 + 4 );
1082     h->nal_buffer_size = h->out.i_bitstream * 3/2 + 4;
1083
1084     if( h->param.i_threads > 1 &&
1085         x264_threadpool_init( &h->threadpool, h->param.i_threads, (void*)x264_encoder_thread_init, h ) )
1086         goto fail;
1087
1088     h->thread[0] = h;
1089     for( int i = 1; i < h->param.i_threads + !!h->param.i_sync_lookahead; i++ )
1090         CHECKED_MALLOC( h->thread[i], sizeof(x264_t) );
1091
1092     if( x264_lookahead_init( h, i_slicetype_length ) )
1093         goto fail;
1094
1095     for( int i = 0; i < h->param.i_threads; i++ )
1096     {
1097         int init_nal_count = h->param.i_slice_count + 3;
1098         int allocate_threadlocal_data = !h->param.b_sliced_threads || !i;
1099         if( i > 0 )
1100             *h->thread[i] = *h;
1101
1102         if( allocate_threadlocal_data )
1103         {
1104             h->thread[i]->fdec = x264_frame_pop_unused( h, 1 );
1105             if( !h->thread[i]->fdec )
1106                 goto fail;
1107         }
1108         else
1109             h->thread[i]->fdec = h->thread[0]->fdec;
1110
1111         CHECKED_MALLOC( h->thread[i]->out.p_bitstream, h->out.i_bitstream );
1112         /* Start each thread with room for init_nal_count NAL units; it'll realloc later if needed. */
1113         CHECKED_MALLOC( h->thread[i]->out.nal, init_nal_count*sizeof(x264_nal_t) );
1114         h->thread[i]->out.i_nals_allocated = init_nal_count;
1115
1116         if( allocate_threadlocal_data && x264_macroblock_cache_allocate( h->thread[i] ) < 0 )
1117             goto fail;
1118     }
1119
1120     for( int i = 0; i < h->param.i_threads; i++ )
1121         if( x264_macroblock_thread_allocate( h->thread[i], 0 ) < 0 )
1122             goto fail;
1123
1124     if( x264_ratecontrol_new( h ) < 0 )
1125         goto fail;
1126
1127     if( h->param.i_nal_hrd )
1128     {
1129         x264_log( h, X264_LOG_DEBUG, "HRD bitrate: %i bits/sec\n", h->sps->vui.hrd.i_bit_rate_unscaled );
1130         x264_log( h, X264_LOG_DEBUG, "CPB size: %i bits\n", h->sps->vui.hrd.i_cpb_size_unscaled );
1131     }
1132
1133     if( h->param.psz_dump_yuv )
1134     {
1135         /* create or truncate the reconstructed video file */
1136         FILE *f = fopen( h->param.psz_dump_yuv, "w" );
1137         if( !f )
1138         {
1139             x264_log( h, X264_LOG_ERROR, "dump_yuv: can't write to %s\n", h->param.psz_dump_yuv );
1140             goto fail;
1141         }
1142         else if( !x264_is_regular_file( f ) )
1143         {
1144             x264_log( h, X264_LOG_ERROR, "dump_yuv: incompatible with non-regular file %s\n", h->param.psz_dump_yuv );
1145             goto fail;
1146         }
1147         fclose( f );
1148     }
1149
1150     x264_log( h, X264_LOG_INFO, "profile %s, level %d.%d\n",
1151         h->sps->i_profile_idc == PROFILE_BASELINE ? "Baseline" :
1152         h->sps->i_profile_idc == PROFILE_MAIN ? "Main" :
1153         h->sps->i_profile_idc == PROFILE_HIGH ? "High" :
1154         "High 4:4:4 Predictive", h->sps->i_level_idc/10, h->sps->i_level_idc%10 );
1155
1156     return h;
1157 fail:
1158     x264_free( h );
1159     return NULL;
1160 }
1161
1162 /****************************************************************************
1163  * x264_encoder_reconfig:
1164  ****************************************************************************/
1165 int x264_encoder_reconfig( x264_t *h, x264_param_t *param )
1166 {
1167     int rc_reconfig = 0;
1168     h = h->thread[h->thread[0]->i_thread_phase];
1169     x264_set_aspect_ratio( h, param, 0 );
1170 #define COPY(var) h->param.var = param->var
1171     COPY( i_frame_reference ); // but never uses more refs than initially specified
1172     COPY( i_bframe_bias );
1173     if( h->param.i_scenecut_threshold )
1174         COPY( i_scenecut_threshold ); // can't turn it on or off, only vary the threshold
1175     COPY( b_deblocking_filter );
1176     COPY( i_deblocking_filter_alphac0 );
1177     COPY( i_deblocking_filter_beta );
1178     COPY( analyse.inter );
1179     COPY( analyse.intra );
1180     COPY( analyse.i_direct_mv_pred );
1181     /* Scratch buffer prevents me_range from being increased for esa/tesa */
1182     if( h->param.analyse.i_me_method < X264_ME_ESA || param->analyse.i_me_range < h->param.analyse.i_me_range )
1183         COPY( analyse.i_me_range );
1184     COPY( analyse.i_noise_reduction );
1185     /* We can't switch out of subme=0 during encoding. */
1186     if( h->param.analyse.i_subpel_refine )
1187         COPY( analyse.i_subpel_refine );
1188     COPY( analyse.i_trellis );
1189     COPY( analyse.b_chroma_me );
1190     COPY( analyse.b_dct_decimate );
1191     COPY( analyse.b_fast_pskip );
1192     COPY( analyse.b_mixed_references );
1193     COPY( analyse.f_psy_rd );
1194     COPY( analyse.f_psy_trellis );
1195     // can only twiddle these if they were enabled to begin with:
1196     if( h->param.analyse.i_me_method >= X264_ME_ESA || param->analyse.i_me_method < X264_ME_ESA )
1197         COPY( analyse.i_me_method );
1198     if( h->param.analyse.i_me_method >= X264_ME_ESA && !h->frames.b_have_sub8x8_esa )
1199         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
1200     if( h->pps->b_transform_8x8_mode )
1201         COPY( analyse.b_transform_8x8 );
1202     if( h->frames.i_max_ref1 > 1 )
1203         COPY( i_bframe_pyramid );
1204     COPY( i_slice_max_size );
1205     COPY( i_slice_max_mbs );
1206     COPY( i_slice_count );
1207
1208     /* VBV can't be turned on if it wasn't on to begin with */
1209     if( h->param.rc.i_vbv_max_bitrate > 0 && h->param.rc.i_vbv_buffer_size > 0 &&
1210           param->rc.i_vbv_max_bitrate > 0 &&   param->rc.i_vbv_buffer_size > 0 )
1211     {
1212         COPY( rc.i_vbv_max_bitrate );
1213         COPY( rc.i_vbv_buffer_size );
1214         COPY( rc.i_bitrate );
1215         rc_reconfig = 1;
1216     }
1217     if( h->param.rc.f_rf_constant != param->rc.f_rf_constant )
1218     {
1219         COPY( rc.f_rf_constant );
1220         rc_reconfig = 1;
1221     }
1222     if( h->param.rc.f_rf_constant_max != param->rc.f_rf_constant_max )
1223     {
1224         COPY( rc.f_rf_constant_max );
1225         rc_reconfig = 1;
1226     }
1227
1228 #undef COPY
1229
1230     mbcmp_init( h );
1231
1232     int ret = x264_validate_parameters( h );
1233
1234     /* Supported reconfiguration options (1-pass only):
1235      * vbv-maxrate
1236      * vbv-bufsize
1237      * crf
1238      * bitrate (CBR only) */
1239     if( !ret && rc_reconfig )
1240         x264_ratecontrol_init_reconfigurable( h, 0 );
1241
1242     return ret;
1243 }
1244
1245 /****************************************************************************
1246  * x264_encoder_parameters:
1247  ****************************************************************************/
1248 void x264_encoder_parameters( x264_t *h, x264_param_t *param )
1249 {
1250     memcpy( param, &h->thread[h->i_thread_phase]->param, sizeof(x264_param_t) );
1251 }
1252
1253 /* internal usage */
1254 static void x264_nal_start( x264_t *h, int i_type, int i_ref_idc )
1255 {
1256     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1257
1258     nal->i_ref_idc        = i_ref_idc;
1259     nal->i_type           = i_type;
1260     nal->b_long_startcode = 1;
1261
1262     nal->i_payload= 0;
1263     nal->p_payload= &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8];
1264 }
1265
1266 /* if number of allocated nals is not enough, re-allocate a larger one. */
1267 static int x264_nal_check_buffer( x264_t *h )
1268 {
1269     if( h->out.i_nal >= h->out.i_nals_allocated )
1270     {
1271         x264_nal_t *new_out = x264_malloc( sizeof(x264_nal_t) * (h->out.i_nals_allocated*2) );
1272         if( !new_out )
1273             return -1;
1274         memcpy( new_out, h->out.nal, sizeof(x264_nal_t) * (h->out.i_nals_allocated) );
1275         x264_free( h->out.nal );
1276         h->out.nal = new_out;
1277         h->out.i_nals_allocated *= 2;
1278     }
1279     return 0;
1280 }
1281
1282 static int x264_nal_end( x264_t *h )
1283 {
1284     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1285     nal->i_payload = &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8] - nal->p_payload;
1286     if( h->param.nalu_process )
1287         h->param.nalu_process( h, nal );
1288     h->out.i_nal++;
1289
1290     return x264_nal_check_buffer( h );
1291 }
1292
1293 static int x264_encoder_encapsulate_nals( x264_t *h, int start )
1294 {
1295     int nal_size = 0, previous_nal_size = 0;
1296
1297     if( h->param.nalu_process )
1298     {
1299         for( int i = start; i < h->out.i_nal; i++ )
1300             nal_size += h->out.nal[i].i_payload;
1301         return nal_size;
1302     }
1303
1304     for( int i = 0; i < start; i++ )
1305         previous_nal_size += h->out.nal[i].i_payload;
1306
1307     for( int i = start; i < h->out.i_nal; i++ )
1308         nal_size += h->out.nal[i].i_payload;
1309
1310     /* Worst-case NAL unit escaping: reallocate the buffer if it's too small. */
1311     if( h->nal_buffer_size < nal_size * 3/2 + h->out.i_nal * 4 )
1312     {
1313         uint8_t *buf = x264_malloc( nal_size * 2 + h->out.i_nal * 4 );
1314         if( !buf )
1315             return -1;
1316         if( previous_nal_size )
1317             memcpy( buf, h->nal_buffer, previous_nal_size );
1318         x264_free( h->nal_buffer );
1319         h->nal_buffer = buf;
1320     }
1321
1322     uint8_t *nal_buffer = h->nal_buffer + previous_nal_size;
1323
1324     for( int i = start; i < h->out.i_nal; i++ )
1325     {
1326         h->out.nal[i].b_long_startcode = !i || h->out.nal[i].i_type == NAL_SPS || h->out.nal[i].i_type == NAL_PPS;
1327         x264_nal_encode( h, nal_buffer, &h->out.nal[i] );
1328         nal_buffer += h->out.nal[i].i_payload;
1329     }
1330
1331     x264_emms();
1332
1333     return nal_buffer - (h->nal_buffer + previous_nal_size);
1334 }
1335
1336 /****************************************************************************
1337  * x264_encoder_headers:
1338  ****************************************************************************/
1339 int x264_encoder_headers( x264_t *h, x264_nal_t **pp_nal, int *pi_nal )
1340 {
1341     int frame_size = 0;
1342     /* init bitstream context */
1343     h->out.i_nal = 0;
1344     bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
1345
1346     /* Write SEI, SPS and PPS. */
1347
1348     /* generate sequence parameters */
1349     x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
1350     x264_sps_write( &h->out.bs, h->sps );
1351     if( x264_nal_end( h ) )
1352         return -1;
1353
1354     /* generate picture parameters */
1355     x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
1356     x264_pps_write( &h->out.bs, h->pps );
1357     if( x264_nal_end( h ) )
1358         return -1;
1359
1360     /* identify ourselves */
1361     x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
1362     if( x264_sei_version_write( h, &h->out.bs ) )
1363         return -1;
1364     if( x264_nal_end( h ) )
1365         return -1;
1366
1367     frame_size = x264_encoder_encapsulate_nals( h, 0 );
1368
1369     /* now set output*/
1370     *pi_nal = h->out.i_nal;
1371     *pp_nal = &h->out.nal[0];
1372     h->out.i_nal = 0;
1373
1374     return frame_size;
1375 }
1376
1377 /* Check to see whether we have chosen a reference list ordering different
1378  * from the standard's default. */
1379 static inline void x264_reference_check_reorder( x264_t *h )
1380 {
1381     for( int i = 0; i < h->i_ref0 - 1; i++ )
1382         /* P and B-frames use different default orders. */
1383         if( h->sh.i_type == SLICE_TYPE_P ? h->fref0[i]->i_frame_num < h->fref0[i+1]->i_frame_num
1384                                          : h->fref0[i]->i_poc < h->fref0[i+1]->i_poc )
1385         {
1386             h->b_ref_reorder[0] = 1;
1387             break;
1388         }
1389 }
1390
1391 /* return -1 on failure, else return the index of the new reference frame */
1392 int x264_weighted_reference_duplicate( x264_t *h, int i_ref, const x264_weight_t *w )
1393 {
1394     int i = h->i_ref0;
1395     int j = 1;
1396     x264_frame_t *newframe;
1397     if( i <= 1 ) /* empty list, definitely can't duplicate frame */
1398         return -1;
1399
1400     newframe = x264_frame_pop_blank_unused( h );
1401
1402     //FIXME: probably don't need to copy everything
1403     *newframe = *h->fref0[i_ref];
1404     newframe->i_reference_count = 1;
1405     newframe->orig = h->fref0[i_ref];
1406     newframe->b_duplicate = 1;
1407     memcpy( h->fenc->weight[j], w, sizeof(h->fenc->weight[i]) );
1408
1409     /* shift the frames to make space for the dupe. */
1410     h->b_ref_reorder[0] = 1;
1411     if( h->i_ref0 < 16 )
1412         ++h->i_ref0;
1413     h->fref0[15] = NULL;
1414     x264_frame_unshift( &h->fref0[j], newframe );
1415
1416     return j;
1417 }
1418
1419 static void x264_weighted_pred_init( x264_t *h )
1420 {
1421     /* for now no analysis and set all weights to nothing */
1422     for( int i_ref = 0; i_ref < h->i_ref0; i_ref++ )
1423         h->fenc->weighted[i_ref] = h->fref0[i_ref]->filtered[0];
1424
1425     // FIXME: This only supports weighting of one reference frame
1426     // and duplicates of that frame.
1427     h->fenc->i_lines_weighted = 0;
1428
1429     for( int i_ref = 0; i_ref < (h->i_ref0 << h->sh.b_mbaff); i_ref++ )
1430         for( int i = 0; i < 3; i++ )
1431             h->sh.weight[i_ref][i].weightfn = NULL;
1432
1433
1434     if( h->sh.i_type != SLICE_TYPE_P || h->param.analyse.i_weighted_pred <= 0 )
1435         return;
1436
1437     int i_padv = PADV << h->param.b_interlaced;
1438     int denom = -1;
1439     int weightluma = 0;
1440     int buffer_next = 0;
1441     //FIXME: when chroma support is added, move this into loop
1442     h->sh.weight[0][1].weightfn = h->sh.weight[0][2].weightfn = NULL;
1443     h->sh.weight[0][1].i_denom = h->sh.weight[0][2].i_denom = 0;
1444     for( int j = 0; j < h->i_ref0; j++ )
1445     {
1446         if( h->fenc->weight[j][0].weightfn )
1447         {
1448             h->sh.weight[j][0] = h->fenc->weight[j][0];
1449             // if weight is useless, don't write it to stream
1450             if( h->sh.weight[j][0].i_scale == 1<<h->sh.weight[j][0].i_denom && h->sh.weight[j][0].i_offset == 0 )
1451                 h->sh.weight[j][0].weightfn = NULL;
1452             else
1453             {
1454                 if( !weightluma )
1455                 {
1456                     weightluma = 1;
1457                     h->sh.weight[0][0].i_denom = denom = h->sh.weight[j][0].i_denom;
1458                     assert( x264_clip3( denom, 0, 7 ) == denom );
1459                 }
1460                 assert( h->sh.weight[j][0].i_denom == denom );
1461                 assert( x264_clip3( h->sh.weight[j][0].i_scale, 0, 127 ) == h->sh.weight[j][0].i_scale );
1462                 assert( x264_clip3( h->sh.weight[j][0].i_offset, -128, 127 ) == h->sh.weight[j][0].i_offset );
1463                 h->fenc->weighted[j] = h->mb.p_weight_buf[buffer_next++] +
1464                     h->fenc->i_stride[0] * i_padv + PADH;
1465             }
1466         }
1467
1468         //scale full resolution frame
1469         if( h->sh.weight[j][0].weightfn && h->param.i_threads == 1 )
1470         {
1471             pixel *src = h->fref0[j]->filtered[0] - h->fref0[j]->i_stride[0]*i_padv - PADH;
1472             pixel *dst = h->fenc->weighted[j] - h->fenc->i_stride[0]*i_padv - PADH;
1473             int stride = h->fenc->i_stride[0];
1474             int width = h->fenc->i_width[0] + PADH*2;
1475             int height = h->fenc->i_lines[0] + i_padv*2;
1476             x264_weight_scale_plane( h, dst, stride, src, stride, width, height, &h->sh.weight[j][0] );
1477             h->fenc->i_lines_weighted = height;
1478         }
1479     }
1480     if( !weightluma )
1481         h->sh.weight[0][0].i_denom = 0;
1482 }
1483
1484 static inline void x264_reference_build_list( x264_t *h, int i_poc )
1485 {
1486     int b_ok;
1487
1488     /* build ref list 0/1 */
1489     h->mb.pic.i_fref[0] = h->i_ref0 = 0;
1490     h->mb.pic.i_fref[1] = h->i_ref1 = 0;
1491     if( h->sh.i_type == SLICE_TYPE_I )
1492         return;
1493
1494     for( int i = 0; h->frames.reference[i]; i++ )
1495     {
1496         if( h->frames.reference[i]->b_corrupt )
1497             continue;
1498         if( h->frames.reference[i]->i_poc < i_poc )
1499             h->fref0[h->i_ref0++] = h->frames.reference[i];
1500         else if( h->frames.reference[i]->i_poc > i_poc )
1501             h->fref1[h->i_ref1++] = h->frames.reference[i];
1502     }
1503
1504     /* Order ref0 from higher to lower poc */
1505     do
1506     {
1507         b_ok = 1;
1508         for( int i = 0; i < h->i_ref0 - 1; i++ )
1509         {
1510             if( h->fref0[i]->i_poc < h->fref0[i+1]->i_poc )
1511             {
1512                 XCHG( x264_frame_t*, h->fref0[i], h->fref0[i+1] );
1513                 b_ok = 0;
1514                 break;
1515             }
1516         }
1517     } while( !b_ok );
1518
1519     if( h->sh.i_mmco_remove_from_end )
1520         for( int i = h->i_ref0-1; i >= h->i_ref0 - h->sh.i_mmco_remove_from_end; i-- )
1521         {
1522             int diff = h->i_frame_num - h->fref0[i]->i_frame_num;
1523             h->sh.mmco[h->sh.i_mmco_command_count].i_poc = h->fref0[i]->i_poc;
1524             h->sh.mmco[h->sh.i_mmco_command_count++].i_difference_of_pic_nums = diff;
1525         }
1526
1527     /* Order ref1 from lower to higher poc (bubble sort) for B-frame */
1528     do
1529     {
1530         b_ok = 1;
1531         for( int i = 0; i < h->i_ref1 - 1; i++ )
1532         {
1533             if( h->fref1[i]->i_poc > h->fref1[i+1]->i_poc )
1534             {
1535                 XCHG( x264_frame_t*, h->fref1[i], h->fref1[i+1] );
1536                 b_ok = 0;
1537                 break;
1538             }
1539         }
1540     } while( !b_ok );
1541
1542     x264_reference_check_reorder( h );
1543
1544     h->i_ref1 = X264_MIN( h->i_ref1, h->frames.i_max_ref1 );
1545     h->i_ref0 = X264_MIN( h->i_ref0, h->frames.i_max_ref0 );
1546     h->i_ref0 = X264_MIN( h->i_ref0, h->param.i_frame_reference ); // if reconfig() has lowered the limit
1547
1548     /* add duplicates */
1549     if( h->fenc->i_type == X264_TYPE_P )
1550     {
1551         int idx = -1;
1552         if( h->param.analyse.i_weighted_pred == X264_WEIGHTP_SMART )
1553         {
1554             x264_weight_t w[3];
1555             w[1].weightfn = w[2].weightfn = NULL;
1556             if( h->param.rc.b_stat_read )
1557                 x264_ratecontrol_set_weights( h, h->fenc );
1558
1559             if( !h->fenc->weight[0][0].weightfn )
1560             {
1561                 h->fenc->weight[0][0].i_denom = 0;
1562                 SET_WEIGHT( w[0], 1, 1, 0, -1 );
1563                 idx = x264_weighted_reference_duplicate( h, 0, w );
1564             }
1565             else
1566             {
1567                 if( h->fenc->weight[0][0].i_scale == 1<<h->fenc->weight[0][0].i_denom )
1568                 {
1569                     SET_WEIGHT( h->fenc->weight[0][0], 1, 1, 0, h->fenc->weight[0][0].i_offset );
1570                 }
1571                 x264_weighted_reference_duplicate( h, 0, weight_none );
1572                 if( h->fenc->weight[0][0].i_offset > -128 )
1573                 {
1574                     w[0] = h->fenc->weight[0][0];
1575                     w[0].i_offset--;
1576                     h->mc.weight_cache( h, &w[0] );
1577                     idx = x264_weighted_reference_duplicate( h, 0, w );
1578                 }
1579             }
1580         }
1581         else if( h->param.analyse.i_weighted_pred == X264_WEIGHTP_BLIND )
1582         {
1583             //weighted offset=-1
1584             x264_weight_t w[3];
1585             SET_WEIGHT( w[0], 1, 1, 0, -1 );
1586             h->fenc->weight[0][0].i_denom = 0;
1587             w[1].weightfn = w[2].weightfn = NULL;
1588             idx = x264_weighted_reference_duplicate( h, 0, w );
1589         }
1590         h->mb.ref_blind_dupe = idx;
1591     }
1592
1593     assert( h->i_ref0 + h->i_ref1 <= 16 );
1594     h->mb.pic.i_fref[0] = h->i_ref0;
1595     h->mb.pic.i_fref[1] = h->i_ref1;
1596 }
1597
1598 static void x264_fdec_filter_row( x264_t *h, int mb_y, int b_inloop )
1599 {
1600     /* mb_y is the mb to be encoded next, not the mb to be filtered here */
1601     int b_hpel = h->fdec->b_kept_as_ref;
1602     int b_deblock = h->sh.i_disable_deblocking_filter_idc != 1;
1603     int b_end = mb_y == h->i_threadslice_end;
1604     int b_measure_quality = 1;
1605     int min_y = mb_y - (1 << h->sh.b_mbaff);
1606     int b_start = min_y == h->i_threadslice_start;
1607     int max_y = b_end ? h->i_threadslice_end : mb_y;
1608     b_deblock &= b_hpel || h->param.psz_dump_yuv;
1609     if( h->param.b_sliced_threads && b_start && min_y && !b_inloop )
1610     {
1611         b_deblock = 0;         /* We already deblocked on the inloop pass. */
1612         b_measure_quality = 0; /* We already measured quality on the inloop pass. */
1613     }
1614     if( mb_y & h->sh.b_mbaff )
1615         return;
1616     if( min_y < h->i_threadslice_start )
1617         return;
1618
1619     if( !b_end && b_inloop )
1620         for( int j = 0; j <= h->sh.b_mbaff; j++ )
1621             for( int i = 0; i < 3; i++ )
1622             {
1623                 memcpy( h->intra_border_backup[j][i],
1624                         h->fdec->plane[i] + ((mb_y*16 >> !!i) + j - 1 - h->sh.b_mbaff) * h->fdec->i_stride[i],
1625                         (h->mb.i_mb_width*16 >> !!i) * sizeof(pixel) );
1626             }
1627
1628     if( b_deblock )
1629         for( int y = min_y; y < max_y; y += (1 << h->sh.b_mbaff) )
1630             x264_frame_deblock_row( h, y );
1631
1632     if( b_hpel )
1633     {
1634         int end = mb_y == h->mb.i_mb_height;
1635         x264_frame_expand_border( h, h->fdec, min_y, end );
1636         if( h->param.analyse.i_subpel_refine )
1637         {
1638             x264_frame_filter( h, h->fdec, min_y, end );
1639             x264_frame_expand_border_filtered( h, h->fdec, min_y, end );
1640         }
1641     }
1642
1643     if( h->i_thread_frames > 1 && h->fdec->b_kept_as_ref )
1644         x264_frame_cond_broadcast( h->fdec, mb_y*16 + (b_end ? 10000 : -(X264_THREAD_HEIGHT << h->sh.b_mbaff)) );
1645
1646     min_y = min_y*16 - 8 * !b_start;
1647     max_y = b_end ? X264_MIN( h->i_threadslice_end*16 , h->param.i_height ) : mb_y*16 - 8;
1648
1649     if( b_measure_quality )
1650     {
1651         if( h->param.analyse.b_psnr )
1652             for( int i = 0; i < 3; i++ )
1653                 h->stat.frame.i_ssd[i] +=
1654                     x264_pixel_ssd_wxh( &h->pixf,
1655                         h->fdec->plane[i] + (min_y>>!!i) * h->fdec->i_stride[i], h->fdec->i_stride[i],
1656                         h->fenc->plane[i] + (min_y>>!!i) * h->fenc->i_stride[i], h->fenc->i_stride[i],
1657                         h->param.i_width >> !!i, (max_y-min_y) >> !!i );
1658
1659         if( h->param.analyse.b_ssim )
1660         {
1661             x264_emms();
1662             /* offset by 2 pixels to avoid alignment of ssim blocks with dct blocks,
1663              * and overlap by 4 */
1664             min_y += b_start ? 2 : -6;
1665             h->stat.frame.f_ssim +=
1666                 x264_pixel_ssim_wxh( &h->pixf,
1667                     h->fdec->plane[0] + 2+min_y*h->fdec->i_stride[0], h->fdec->i_stride[0],
1668                     h->fenc->plane[0] + 2+min_y*h->fenc->i_stride[0], h->fenc->i_stride[0],
1669                     h->param.i_width-2, max_y-min_y, h->scratch_buffer );
1670         }
1671     }
1672 }
1673
1674 static inline int x264_reference_update( x264_t *h )
1675 {
1676     if( !h->fdec->b_kept_as_ref )
1677     {
1678         if( h->i_thread_frames > 1 )
1679         {
1680             x264_frame_push_unused( h, h->fdec );
1681             h->fdec = x264_frame_pop_unused( h, 1 );
1682             if( !h->fdec )
1683                 return -1;
1684         }
1685         return 0;
1686     }
1687
1688     /* apply mmco from previous frame. */
1689     for( int i = 0; i < h->sh.i_mmco_command_count; i++ )
1690         for( int j = 0; h->frames.reference[j]; j++ )
1691             if( h->frames.reference[j]->i_poc == h->sh.mmco[i].i_poc )
1692                 x264_frame_push_unused( h, x264_frame_shift( &h->frames.reference[j] ) );
1693
1694     /* move frame in the buffer */
1695     x264_frame_push( h->frames.reference, h->fdec );
1696     if( h->frames.reference[h->sps->i_num_ref_frames] )
1697         x264_frame_push_unused( h, x264_frame_shift( h->frames.reference ) );
1698     h->fdec = x264_frame_pop_unused( h, 1 );
1699     if( !h->fdec )
1700         return -1;
1701     return 0;
1702 }
1703
1704 static inline void x264_reference_reset( x264_t *h )
1705 {
1706     while( h->frames.reference[0] )
1707         x264_frame_push_unused( h, x264_frame_pop( h->frames.reference ) );
1708     h->fdec->i_poc =
1709     h->fenc->i_poc = 0;
1710 }
1711
1712 static inline void x264_reference_hierarchy_reset( x264_t *h )
1713 {
1714     int ref;
1715     int b_hasdelayframe = 0;
1716
1717     /* look for delay frames -- chain must only contain frames that are disposable */
1718     for( int i = 0; h->frames.current[i] && IS_DISPOSABLE( h->frames.current[i]->i_type ); i++ )
1719         b_hasdelayframe |= h->frames.current[i]->i_coded
1720                         != h->frames.current[i]->i_frame + h->sps->vui.i_num_reorder_frames;
1721
1722     /* This function must handle b-pyramid and clear frames for open-gop */
1723     if( h->param.i_bframe_pyramid != X264_B_PYRAMID_STRICT && !b_hasdelayframe && h->frames.i_poc_last_open_gop == -1 )
1724         return;
1725
1726     /* Remove last BREF. There will never be old BREFs in the
1727      * dpb during a BREF decode when pyramid == STRICT */
1728     for( ref = 0; h->frames.reference[ref]; ref++ )
1729     {
1730         if( ( h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT
1731             && h->frames.reference[ref]->i_type == X264_TYPE_BREF )
1732             || ( h->frames.reference[ref]->i_poc < h->frames.i_poc_last_open_gop
1733             && h->sh.i_type != SLICE_TYPE_B ) )
1734         {
1735             int diff = h->i_frame_num - h->frames.reference[ref]->i_frame_num;
1736             h->sh.mmco[h->sh.i_mmco_command_count].i_difference_of_pic_nums = diff;
1737             h->sh.mmco[h->sh.i_mmco_command_count++].i_poc = h->frames.reference[ref]->i_poc;
1738             x264_frame_push_unused( h, x264_frame_shift( &h->frames.reference[ref] ) );
1739             h->b_ref_reorder[0] = 1;
1740             ref--;
1741         }
1742     }
1743
1744     /* Prepare room in the dpb for the delayed display time of the later b-frame's */
1745     if( h->param.i_bframe_pyramid )
1746         h->sh.i_mmco_remove_from_end = X264_MAX( ref + 2 - h->frames.i_max_dpb, 0 );
1747 }
1748
1749 static inline void x264_slice_init( x264_t *h, int i_nal_type, int i_global_qp )
1750 {
1751     /* ------------------------ Create slice header  ----------------------- */
1752     if( i_nal_type == NAL_SLICE_IDR )
1753     {
1754         x264_slice_header_init( h, &h->sh, h->sps, h->pps, h->i_idr_pic_id, h->i_frame_num, i_global_qp );
1755
1756         /* increment id */
1757         h->i_idr_pic_id = ( h->i_idr_pic_id + 1 ) % 65536;
1758     }
1759     else
1760     {
1761         x264_slice_header_init( h, &h->sh, h->sps, h->pps, -1, h->i_frame_num, i_global_qp );
1762
1763         h->sh.i_num_ref_idx_l0_active = h->i_ref0 <= 0 ? 1 : h->i_ref0;
1764         h->sh.i_num_ref_idx_l1_active = h->i_ref1 <= 0 ? 1 : h->i_ref1;
1765         if( h->sh.i_num_ref_idx_l0_active != h->pps->i_num_ref_idx_l0_default_active ||
1766             (h->sh.i_type == SLICE_TYPE_B && h->sh.i_num_ref_idx_l1_active != h->pps->i_num_ref_idx_l1_default_active) )
1767         {
1768             h->sh.b_num_ref_idx_override = 1;
1769         }
1770     }
1771
1772     h->fdec->i_frame_num = h->sh.i_frame_num;
1773
1774     if( h->sps->i_poc_type == 0 )
1775     {
1776         h->sh.i_poc = h->fdec->i_poc;
1777         if( h->param.b_interlaced )
1778         {
1779             h->sh.i_delta_poc_bottom = h->param.b_tff ? 1 : -1;
1780             if( h->sh.i_delta_poc_bottom == -1 )
1781                 h->sh.i_poc = h->fdec->i_poc + 1;
1782         }
1783         else
1784             h->sh.i_delta_poc_bottom = 0;
1785     }
1786     else if( h->sps->i_poc_type == 1 )
1787     {
1788         /* FIXME TODO FIXME */
1789     }
1790     else
1791     {
1792         /* Nothing to do ? */
1793     }
1794
1795     x264_macroblock_slice_init( h );
1796 }
1797
1798 static int x264_slice_write( x264_t *h )
1799 {
1800     int i_skip;
1801     int mb_xy, i_mb_x, i_mb_y;
1802     int i_skip_bak = 0; /* Shut up GCC. */
1803     bs_t bs_bak;
1804     x264_cabac_t cabac_bak;
1805     uint8_t cabac_prevbyte_bak = 0; /* Shut up GCC. */
1806     /* Assume no more than 3 bytes of NALU escaping.
1807      * NALUs other than the first use a 3-byte startcode. */
1808     int overhead_guess = (NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal)) + 3;
1809     int slice_max_size = h->param.i_slice_max_size > 0 ? (h->param.i_slice_max_size-overhead_guess)*8 : INT_MAX;
1810     int starting_bits = bs_pos(&h->out.bs);
1811     int b_deblock = h->sh.i_disable_deblocking_filter_idc != 1;
1812     int b_hpel = h->fdec->b_kept_as_ref;
1813     b_deblock &= b_hpel || h->param.psz_dump_yuv;
1814     bs_realign( &h->out.bs );
1815
1816     /* Slice */
1817     x264_nal_start( h, h->i_nal_type, h->i_nal_ref_idc );
1818     h->out.nal[h->out.i_nal].i_first_mb = h->sh.i_first_mb;
1819
1820     /* Slice header */
1821     x264_macroblock_thread_init( h );
1822
1823     /* If this isn't the first slice in the threadslice, set the slice QP
1824      * equal to the last QP in the previous slice for more accurate
1825      * CABAC initialization. */
1826     if( h->sh.i_first_mb != h->i_threadslice_start * h->mb.i_mb_width )
1827     {
1828         h->sh.i_qp = h->mb.i_last_qp;
1829         h->sh.i_qp_delta = h->sh.i_qp - h->pps->i_pic_init_qp;
1830     }
1831
1832     x264_slice_header_write( &h->out.bs, &h->sh, h->i_nal_ref_idc );
1833     if( h->param.b_cabac )
1834     {
1835         /* alignment needed */
1836         bs_align_1( &h->out.bs );
1837
1838         /* init cabac */
1839         x264_cabac_context_init( &h->cabac, h->sh.i_type, h->sh.i_qp, h->sh.i_cabac_init_idc );
1840         x264_cabac_encode_init ( &h->cabac, h->out.bs.p, h->out.bs.p_end );
1841     }
1842     h->mb.i_last_qp = h->sh.i_qp;
1843     h->mb.i_last_dqp = 0;
1844
1845     i_mb_y = h->sh.i_first_mb / h->mb.i_mb_width;
1846     i_mb_x = h->sh.i_first_mb % h->mb.i_mb_width;
1847     i_skip = 0;
1848
1849     while( (mb_xy = i_mb_x + i_mb_y * h->mb.i_mb_width) <= h->sh.i_last_mb )
1850     {
1851         int mb_spos = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
1852         if( h->param.i_slice_max_size > 0 )
1853         {
1854             /* We don't need the contexts because flushing the CABAC encoder has no context
1855              * dependency and macroblocks are only re-encoded in the case where a slice is
1856              * ended (and thus the content of all contexts are thrown away). */
1857             if( h->param.b_cabac )
1858             {
1859                 memcpy( &cabac_bak, &h->cabac, offsetof(x264_cabac_t, f8_bits_encoded) );
1860                 /* x264's CABAC writer modifies the previous byte during carry, so it has to be
1861                  * backed up. */
1862                 cabac_prevbyte_bak = h->cabac.p[-1];
1863             }
1864             else
1865             {
1866                 bs_bak = h->out.bs;
1867                 i_skip_bak = i_skip;
1868             }
1869         }
1870
1871         if( i_mb_x == 0 && !h->mb.b_reencode_mb )
1872             x264_fdec_filter_row( h, i_mb_y, 1 );
1873
1874         /* load cache */
1875         x264_macroblock_cache_load( h, i_mb_x, i_mb_y );
1876
1877         x264_macroblock_analyse( h );
1878
1879         /* encode this macroblock -> be careful it can change the mb type to P_SKIP if needed */
1880         x264_macroblock_encode( h );
1881
1882         if( x264_bitstream_check_buffer( h ) )
1883             return -1;
1884
1885         if( h->param.b_cabac )
1886         {
1887             if( mb_xy > h->sh.i_first_mb && !(h->sh.b_mbaff && (i_mb_y&1)) )
1888                 x264_cabac_encode_terminal( &h->cabac );
1889
1890             if( IS_SKIP( h->mb.i_type ) )
1891                 x264_cabac_mb_skip( h, 1 );
1892             else
1893             {
1894                 if( h->sh.i_type != SLICE_TYPE_I )
1895                     x264_cabac_mb_skip( h, 0 );
1896                 x264_macroblock_write_cabac( h, &h->cabac );
1897             }
1898         }
1899         else
1900         {
1901             if( IS_SKIP( h->mb.i_type ) )
1902                 i_skip++;
1903             else
1904             {
1905                 if( h->sh.i_type != SLICE_TYPE_I )
1906                 {
1907                     bs_write_ue( &h->out.bs, i_skip );  /* skip run */
1908                     i_skip = 0;
1909                 }
1910                 x264_macroblock_write_cavlc( h );
1911             }
1912         }
1913
1914         int total_bits = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
1915         int mb_size = total_bits - mb_spos;
1916
1917         /* We'll just re-encode this last macroblock if we go over the max slice size. */
1918         if( total_bits - starting_bits > slice_max_size && !h->mb.b_reencode_mb )
1919         {
1920             if( mb_xy != h->sh.i_first_mb )
1921             {
1922                 if( h->param.b_cabac )
1923                 {
1924                     memcpy( &h->cabac, &cabac_bak, offsetof(x264_cabac_t, f8_bits_encoded) );
1925                     h->cabac.p[-1] = cabac_prevbyte_bak;
1926                 }
1927                 else
1928                 {
1929                     h->out.bs = bs_bak;
1930                     i_skip = i_skip_bak;
1931                 }
1932                 h->mb.b_reencode_mb = 1;
1933                 h->sh.i_last_mb = mb_xy-1;
1934                 break;
1935             }
1936             else
1937             {
1938                 h->sh.i_last_mb = mb_xy;
1939                 h->mb.b_reencode_mb = 0;
1940             }
1941         }
1942         else
1943             h->mb.b_reencode_mb = 0;
1944
1945 #if HAVE_VISUALIZE
1946         if( h->param.b_visualize )
1947             x264_visualize_mb( h );
1948 #endif
1949
1950         /* save cache */
1951         x264_macroblock_cache_save( h );
1952
1953         /* accumulate mb stats */
1954         h->stat.frame.i_mb_count[h->mb.i_type]++;
1955
1956         int b_intra = IS_INTRA( h->mb.i_type );
1957         if( h->param.i_log_level >= X264_LOG_INFO || h->param.rc.b_stat_write )
1958         {
1959             if( !b_intra && !IS_SKIP( h->mb.i_type ) && !IS_DIRECT( h->mb.i_type ) )
1960             {
1961                 if( h->mb.i_partition != D_8x8 )
1962                         h->stat.frame.i_mb_partition[h->mb.i_partition] += 4;
1963                     else
1964                         for( int i = 0; i < 4; i++ )
1965                             h->stat.frame.i_mb_partition[h->mb.i_sub_partition[i]] ++;
1966                 if( h->param.i_frame_reference > 1 )
1967                     for( int i_list = 0; i_list <= (h->sh.i_type == SLICE_TYPE_B); i_list++ )
1968                         for( int i = 0; i < 4; i++ )
1969                         {
1970                             int i_ref = h->mb.cache.ref[i_list][ x264_scan8[4*i] ];
1971                             if( i_ref >= 0 )
1972                                 h->stat.frame.i_mb_count_ref[i_list][i_ref] ++;
1973                         }
1974             }
1975         }
1976
1977         if( h->param.i_log_level >= X264_LOG_INFO )
1978         {
1979             if( h->mb.i_cbp_luma | h->mb.i_cbp_chroma )
1980             {
1981                 int cbpsum = (h->mb.i_cbp_luma&1) + ((h->mb.i_cbp_luma>>1)&1)
1982                            + ((h->mb.i_cbp_luma>>2)&1) + (h->mb.i_cbp_luma>>3);
1983                 h->stat.frame.i_mb_cbp[!b_intra + 0] += cbpsum;
1984                 h->stat.frame.i_mb_cbp[!b_intra + 2] += !!h->mb.i_cbp_chroma;
1985                 h->stat.frame.i_mb_cbp[!b_intra + 4] += h->mb.i_cbp_chroma >> 1;
1986             }
1987             if( h->mb.i_cbp_luma && !b_intra )
1988             {
1989                 h->stat.frame.i_mb_count_8x8dct[0] ++;
1990                 h->stat.frame.i_mb_count_8x8dct[1] += h->mb.b_transform_8x8;
1991             }
1992             if( b_intra && h->mb.i_type != I_PCM )
1993             {
1994                 if( h->mb.i_type == I_16x16 )
1995                     h->stat.frame.i_mb_pred_mode[0][h->mb.i_intra16x16_pred_mode]++;
1996                 else if( h->mb.i_type == I_8x8 )
1997                     for( int i = 0; i < 16; i += 4 )
1998                         h->stat.frame.i_mb_pred_mode[1][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
1999                 else //if( h->mb.i_type == I_4x4 )
2000                     for( int i = 0; i < 16; i++ )
2001                         h->stat.frame.i_mb_pred_mode[2][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
2002                 h->stat.frame.i_mb_pred_mode[3][x264_mb_pred_mode8x8c_fix[h->mb.i_chroma_pred_mode]]++;
2003             }
2004         }
2005
2006         /* calculate deblock strength values (actual deblocking is done per-row along with hpel) */
2007         if( b_deblock )
2008         {
2009             int mvy_limit = 4 >> h->sh.b_mbaff;
2010             uint8_t (*bs)[4][4] = h->deblock_strength[h->mb.i_mb_y&h->sh.b_mbaff][h->mb.i_mb_x];
2011             x264_macroblock_cache_load_deblock( h );
2012             if( IS_INTRA( h->mb.type[h->mb.i_mb_xy] ) )
2013                 memset( bs, 3, 2*4*4*sizeof(uint8_t) );
2014             else
2015                 h->loopf.deblock_strength( h->mb.cache.non_zero_count, h->mb.cache.ref, h->mb.cache.mv,
2016                                            bs, mvy_limit, h->sh.i_type == SLICE_TYPE_B );
2017         }
2018
2019         x264_ratecontrol_mb( h, mb_size );
2020
2021         if( h->sh.b_mbaff )
2022         {
2023             i_mb_x += i_mb_y & 1;
2024             i_mb_y ^= i_mb_x < h->mb.i_mb_width;
2025         }
2026         else
2027             i_mb_x++;
2028         if( i_mb_x == h->mb.i_mb_width )
2029         {
2030             i_mb_y++;
2031             i_mb_x = 0;
2032         }
2033     }
2034     h->out.nal[h->out.i_nal].i_last_mb = h->sh.i_last_mb;
2035
2036     if( h->param.b_cabac )
2037     {
2038         x264_cabac_encode_flush( h, &h->cabac );
2039         h->out.bs.p = h->cabac.p;
2040     }
2041     else
2042     {
2043         if( i_skip > 0 )
2044             bs_write_ue( &h->out.bs, i_skip );  /* last skip run */
2045         /* rbsp_slice_trailing_bits */
2046         bs_rbsp_trailing( &h->out.bs );
2047         bs_flush( &h->out.bs );
2048     }
2049     if( x264_nal_end( h ) )
2050         return -1;
2051
2052     if( h->sh.i_last_mb == (h->i_threadslice_end * h->mb.i_mb_width - 1) )
2053     {
2054         h->stat.frame.i_misc_bits = bs_pos( &h->out.bs )
2055                                   + (h->out.i_nal*NALU_OVERHEAD * 8)
2056                                   - h->stat.frame.i_tex_bits
2057                                   - h->stat.frame.i_mv_bits;
2058         x264_fdec_filter_row( h, h->i_threadslice_end, 1 );
2059     }
2060
2061     return 0;
2062 }
2063
2064 static void x264_thread_sync_context( x264_t *dst, x264_t *src )
2065 {
2066     if( dst == src )
2067         return;
2068
2069     // reference counting
2070     for( x264_frame_t **f = src->frames.reference; *f; f++ )
2071         (*f)->i_reference_count++;
2072     for( x264_frame_t **f = dst->frames.reference; *f; f++ )
2073         x264_frame_push_unused( src, *f );
2074     src->fdec->i_reference_count++;
2075     x264_frame_push_unused( src, dst->fdec );
2076
2077     // copy everything except the per-thread pointers and the constants.
2078     memcpy( &dst->i_frame, &src->i_frame, offsetof(x264_t, mb.type) - offsetof(x264_t, i_frame) );
2079     dst->param = src->param;
2080     dst->stat = src->stat;
2081 }
2082
2083 static void x264_thread_sync_stat( x264_t *dst, x264_t *src )
2084 {
2085     if( dst == src )
2086         return;
2087     memcpy( &dst->stat.i_frame_count, &src->stat.i_frame_count, sizeof(dst->stat) - sizeof(dst->stat.frame) );
2088 }
2089
2090 static void *x264_slices_write( x264_t *h )
2091 {
2092     int i_slice_num = 0;
2093     int last_thread_mb = h->sh.i_last_mb;
2094
2095 #if HAVE_VISUALIZE
2096     if( h->param.b_visualize )
2097         if( x264_visualize_init( h ) )
2098             return (void *)-1;
2099 #endif
2100
2101     /* init stats */
2102     memset( &h->stat.frame, 0, sizeof(h->stat.frame) );
2103     h->mb.b_reencode_mb = 0;
2104     while( h->sh.i_first_mb <= last_thread_mb )
2105     {
2106         h->sh.i_last_mb = last_thread_mb;
2107         if( h->param.i_slice_max_mbs )
2108             h->sh.i_last_mb = h->sh.i_first_mb + h->param.i_slice_max_mbs - 1;
2109         else if( h->param.i_slice_count && !h->param.b_sliced_threads )
2110         {
2111             int height = h->mb.i_mb_height >> h->param.b_interlaced;
2112             int width = h->mb.i_mb_width << h->param.b_interlaced;
2113             i_slice_num++;
2114             h->sh.i_last_mb = (height * i_slice_num + h->param.i_slice_count/2) / h->param.i_slice_count * width - 1;
2115         }
2116         h->sh.i_last_mb = X264_MIN( h->sh.i_last_mb, last_thread_mb );
2117         if( x264_stack_align( x264_slice_write, h ) )
2118             return (void *)-1;
2119         h->sh.i_first_mb = h->sh.i_last_mb + 1;
2120     }
2121
2122 #if HAVE_VISUALIZE
2123     if( h->param.b_visualize )
2124     {
2125         x264_visualize_show( h );
2126         x264_visualize_close( h );
2127     }
2128 #endif
2129
2130     return (void *)0;
2131 }
2132
2133 static int x264_threaded_slices_write( x264_t *h )
2134 {
2135     /* set first/last mb and sync contexts */
2136     for( int i = 0; i < h->param.i_threads; i++ )
2137     {
2138         x264_t *t = h->thread[i];
2139         if( i )
2140         {
2141             t->param = h->param;
2142             memcpy( &t->i_frame, &h->i_frame, offsetof(x264_t, rc) - offsetof(x264_t, i_frame) );
2143         }
2144         int height = h->mb.i_mb_height >> h->param.b_interlaced;
2145         t->i_threadslice_start = ((height *  i    + h->param.i_slice_count/2) / h->param.i_threads) << h->param.b_interlaced;
2146         t->i_threadslice_end   = ((height * (i+1) + h->param.i_slice_count/2) / h->param.i_threads) << h->param.b_interlaced;
2147         t->sh.i_first_mb = t->i_threadslice_start * h->mb.i_mb_width;
2148         t->sh.i_last_mb  =   t->i_threadslice_end * h->mb.i_mb_width - 1;
2149     }
2150
2151     x264_stack_align( x264_analyse_weight_frame, h, h->mb.i_mb_height*16 + 16 );
2152
2153     x264_threads_distribute_ratecontrol( h );
2154
2155     /* dispatch */
2156     for( int i = 0; i < h->param.i_threads; i++ )
2157     {
2158         x264_threadpool_run( h->threadpool, (void*)x264_slices_write, h->thread[i] );
2159         h->thread[i]->b_thread_active = 1;
2160     }
2161     for( int i = 0; i < h->param.i_threads; i++ )
2162     {
2163         h->thread[i]->b_thread_active = 0;
2164         if( (intptr_t)x264_threadpool_wait( h->threadpool, h->thread[i] ) )
2165             return -1;
2166     }
2167
2168     /* Go back and fix up the hpel on the borders between slices. */
2169     for( int i = 1; i < h->param.i_threads; i++ )
2170     {
2171         x264_fdec_filter_row( h->thread[i], h->thread[i]->i_threadslice_start + 1, 0 );
2172         if( h->sh.b_mbaff )
2173             x264_fdec_filter_row( h->thread[i], h->thread[i]->i_threadslice_start + 2, 0 );
2174     }
2175
2176     x264_threads_merge_ratecontrol( h );
2177
2178     for( int i = 1; i < h->param.i_threads; i++ )
2179     {
2180         x264_t *t = h->thread[i];
2181         for( int j = 0; j < t->out.i_nal; j++ )
2182         {
2183             h->out.nal[h->out.i_nal] = t->out.nal[j];
2184             h->out.i_nal++;
2185             x264_nal_check_buffer( h );
2186         }
2187         /* All entries in stat.frame are ints except for ssd/ssim. */
2188         for( int j = 0; j < (offsetof(x264_t,stat.frame.i_ssd) - offsetof(x264_t,stat.frame.i_mv_bits)) / sizeof(int); j++ )
2189             ((int*)&h->stat.frame)[j] += ((int*)&t->stat.frame)[j];
2190         for( int j = 0; j < 3; j++ )
2191             h->stat.frame.i_ssd[j] += t->stat.frame.i_ssd[j];
2192         h->stat.frame.f_ssim += t->stat.frame.f_ssim;
2193     }
2194
2195     return 0;
2196 }
2197
2198 void x264_encoder_intra_refresh( x264_t *h )
2199 {
2200     h = h->thread[h->i_thread_phase];
2201     h->b_queued_intra_refresh = 1;
2202 }
2203
2204 int x264_encoder_invalidate_reference( x264_t *h, int64_t pts )
2205 {
2206     if( h->param.i_bframe )
2207     {
2208         x264_log( h, X264_LOG_ERROR, "x264_encoder_invalidate_reference is not supported with B-frames enabled\n" );
2209         return -1;
2210     }
2211     if( h->param.b_intra_refresh )
2212     {
2213         x264_log( h, X264_LOG_ERROR, "x264_encoder_invalidate_reference is not supported with intra refresh enabled\n" );
2214         return -1;
2215     }
2216     h = h->thread[h->i_thread_phase];
2217     h->i_reference_invalidate_pts = pts;
2218     return 0;
2219 }
2220
2221 /****************************************************************************
2222  * x264_encoder_encode:
2223  *  XXX: i_poc   : is the poc of the current given picture
2224  *       i_frame : is the number of the frame being coded
2225  *  ex:  type frame poc
2226  *       I      0   2*0
2227  *       P      1   2*3
2228  *       B      2   2*1
2229  *       B      3   2*2
2230  *       P      4   2*6
2231  *       B      5   2*4
2232  *       B      6   2*5
2233  ****************************************************************************/
2234 int     x264_encoder_encode( x264_t *h,
2235                              x264_nal_t **pp_nal, int *pi_nal,
2236                              x264_picture_t *pic_in,
2237                              x264_picture_t *pic_out )
2238 {
2239     x264_t *thread_current, *thread_prev, *thread_oldest;
2240     int i_nal_type, i_nal_ref_idc, i_global_qp;
2241     int overhead = NALU_OVERHEAD;
2242
2243     if( h->i_thread_frames > 1 )
2244     {
2245         thread_prev    = h->thread[ h->i_thread_phase ];
2246         h->i_thread_phase = (h->i_thread_phase + 1) % h->i_thread_frames;
2247         thread_current = h->thread[ h->i_thread_phase ];
2248         thread_oldest  = h->thread[ (h->i_thread_phase + 1) % h->i_thread_frames ];
2249         x264_thread_sync_context( thread_current, thread_prev );
2250         x264_thread_sync_ratecontrol( thread_current, thread_prev, thread_oldest );
2251         h = thread_current;
2252     }
2253     else
2254     {
2255         thread_current =
2256         thread_oldest  = h;
2257     }
2258 #if HAVE_MMX
2259     if( h->i_thread_frames == 1 && h->param.cpu&X264_CPU_SSE_MISALIGN )
2260         x264_cpu_mask_misalign_sse();
2261 #endif
2262
2263     // ok to call this before encoding any frames, since the initial values of fdec have b_kept_as_ref=0
2264     if( x264_reference_update( h ) )
2265         return -1;
2266     h->fdec->i_lines_completed = -1;
2267
2268     /* no data out */
2269     *pi_nal = 0;
2270     *pp_nal = NULL;
2271
2272     /* ------------------- Setup new frame from picture -------------------- */
2273     if( pic_in != NULL )
2274     {
2275         /* 1: Copy the picture to a frame and move it to a buffer */
2276         x264_frame_t *fenc = x264_frame_pop_unused( h, 0 );
2277         if( !fenc )
2278             return -1;
2279
2280         if( x264_frame_copy_picture( h, fenc, pic_in ) < 0 )
2281             return -1;
2282
2283         if( h->param.i_width != 16 * h->mb.i_mb_width ||
2284             h->param.i_height != 16 * h->mb.i_mb_height )
2285             x264_frame_expand_border_mod16( h, fenc );
2286
2287         fenc->i_frame = h->frames.i_input++;
2288
2289         if( h->frames.i_bframe_delay && fenc->i_frame == h->frames.i_bframe_delay )
2290             h->frames.i_bframe_delay_time = fenc->i_pts;
2291
2292         if( h->param.b_vfr_input && fenc->i_pts <= h->frames.i_largest_pts )
2293             x264_log( h, X264_LOG_WARNING, "non-strictly-monotonic PTS\n" );
2294
2295         h->frames.i_second_largest_pts = h->frames.i_largest_pts;
2296         h->frames.i_largest_pts = fenc->i_pts;
2297
2298         if( (fenc->i_pic_struct < PIC_STRUCT_AUTO) || (fenc->i_pic_struct > PIC_STRUCT_TRIPLE) )
2299             fenc->i_pic_struct = PIC_STRUCT_AUTO;
2300
2301         if( fenc->i_pic_struct == PIC_STRUCT_AUTO )
2302         {
2303             int b_interlaced = fenc->param ? fenc->param->b_interlaced : h->param.b_interlaced;
2304             if( b_interlaced )
2305             {
2306                 int b_tff = fenc->param ? fenc->param->b_tff : h->param.b_tff;
2307                 fenc->i_pic_struct = b_tff ? PIC_STRUCT_TOP_BOTTOM : PIC_STRUCT_BOTTOM_TOP;
2308             }
2309             else
2310                 fenc->i_pic_struct = PIC_STRUCT_PROGRESSIVE;
2311         }
2312
2313         if( h->param.rc.b_mb_tree && h->param.rc.b_stat_read )
2314         {
2315             if( x264_macroblock_tree_read( h, fenc, pic_in->prop.quant_offsets ) )
2316                 return -1;
2317         }
2318         else
2319             x264_adaptive_quant_frame( h, fenc, pic_in->prop.quant_offsets );
2320
2321         if( pic_in->prop.quant_offsets_free )
2322             pic_in->prop.quant_offsets_free( pic_in->prop.quant_offsets );
2323
2324         if( h->frames.b_have_lowres )
2325             x264_frame_init_lowres( h, fenc );
2326
2327         /* 2: Place the frame into the queue for its slice type decision */
2328         x264_lookahead_put_frame( h, fenc );
2329
2330         if( h->frames.i_input <= h->frames.i_delay + 1 - h->i_thread_frames )
2331         {
2332             /* Nothing yet to encode, waiting for filling of buffers */
2333             pic_out->i_type = X264_TYPE_AUTO;
2334             return 0;
2335         }
2336     }
2337     else
2338     {
2339         /* signal kills for lookahead thread */
2340         x264_pthread_mutex_lock( &h->lookahead->ifbuf.mutex );
2341         h->lookahead->b_exit_thread = 1;
2342         x264_pthread_cond_broadcast( &h->lookahead->ifbuf.cv_fill );
2343         x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
2344     }
2345
2346     h->i_frame++;
2347     /* 3: The picture is analyzed in the lookahead */
2348     if( !h->frames.current[0] )
2349         x264_lookahead_get_frames( h );
2350
2351     if( !h->frames.current[0] && x264_lookahead_is_empty( h ) )
2352         return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
2353
2354     /* ------------------- Get frame to be encoded ------------------------- */
2355     /* 4: get picture to encode */
2356     h->fenc = x264_frame_shift( h->frames.current );
2357     if( h->i_frame == h->i_thread_frames - 1 )
2358         h->i_reordered_pts_delay = h->fenc->i_reordered_pts;
2359     if( h->fenc->param )
2360     {
2361         x264_encoder_reconfig( h, h->fenc->param );
2362         if( h->fenc->param->param_free )
2363             h->fenc->param->param_free( h->fenc->param );
2364     }
2365
2366     if( h->i_reference_invalidate_pts )
2367     {
2368         if( h->i_reference_invalidate_pts >= h->i_last_idr_pts )
2369             for( int i = 0; h->frames.reference[i]; i++ )
2370                 if( h->i_reference_invalidate_pts <= h->frames.reference[i]->i_pts )
2371                     h->frames.reference[i]->b_corrupt = 1;
2372         h->i_reference_invalidate_pts = 0;
2373     }
2374
2375     if( !IS_X264_TYPE_I( h->fenc->i_type ) )
2376     {
2377         int valid_refs_left = 0;
2378         for( int i = 0; h->frames.reference[i]; i++ )
2379             if( !h->frames.reference[i]->b_corrupt )
2380                 valid_refs_left++;
2381         /* No valid reference frames left: force an IDR. */
2382         if( !valid_refs_left )
2383         {
2384             h->fenc->b_keyframe = 1;
2385             h->fenc->i_type = X264_TYPE_IDR;
2386         }
2387     }
2388
2389     if( h->fenc->b_keyframe )
2390     {
2391         h->frames.i_last_keyframe = h->fenc->i_frame;
2392         if( h->fenc->i_type == X264_TYPE_IDR )
2393         {
2394             h->i_frame_num = 0;
2395             h->frames.i_last_idr = h->fenc->i_frame;
2396         }
2397     }
2398     h->sh.i_mmco_command_count =
2399     h->sh.i_mmco_remove_from_end = 0;
2400     h->b_ref_reorder[0] =
2401     h->b_ref_reorder[1] = 0;
2402     h->fdec->i_poc =
2403     h->fenc->i_poc = 2 * ( h->fenc->i_frame - X264_MAX( h->frames.i_last_idr, 0 ) );
2404
2405     /* ------------------- Setup frame context ----------------------------- */
2406     /* 5: Init data dependent of frame type */
2407     if( h->fenc->i_type == X264_TYPE_IDR )
2408     {
2409         /* reset ref pictures */
2410         i_nal_type    = NAL_SLICE_IDR;
2411         i_nal_ref_idc = NAL_PRIORITY_HIGHEST;
2412         h->sh.i_type = SLICE_TYPE_I;
2413         x264_reference_reset( h );
2414         h->frames.i_poc_last_open_gop = -1;
2415     }
2416     else if( h->fenc->i_type == X264_TYPE_I )
2417     {
2418         i_nal_type    = NAL_SLICE;
2419         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
2420         h->sh.i_type = SLICE_TYPE_I;
2421         x264_reference_hierarchy_reset( h );
2422         if( h->param.i_open_gop )
2423             h->frames.i_poc_last_open_gop = h->fenc->b_keyframe ? h->fenc->i_poc : -1;
2424     }
2425     else if( h->fenc->i_type == X264_TYPE_P )
2426     {
2427         i_nal_type    = NAL_SLICE;
2428         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
2429         h->sh.i_type = SLICE_TYPE_P;
2430         x264_reference_hierarchy_reset( h );
2431         h->frames.i_poc_last_open_gop = -1;
2432     }
2433     else if( h->fenc->i_type == X264_TYPE_BREF )
2434     {
2435         i_nal_type    = NAL_SLICE;
2436         i_nal_ref_idc = h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT ? NAL_PRIORITY_LOW : NAL_PRIORITY_HIGH;
2437         h->sh.i_type = SLICE_TYPE_B;
2438         x264_reference_hierarchy_reset( h );
2439     }
2440     else    /* B frame */
2441     {
2442         i_nal_type    = NAL_SLICE;
2443         i_nal_ref_idc = NAL_PRIORITY_DISPOSABLE;
2444         h->sh.i_type = SLICE_TYPE_B;
2445     }
2446
2447     h->fdec->i_type = h->fenc->i_type;
2448     h->fdec->i_frame = h->fenc->i_frame;
2449     h->fenc->b_kept_as_ref =
2450     h->fdec->b_kept_as_ref = i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE && h->param.i_keyint_max > 1;
2451
2452     h->fdec->i_pts = h->fenc->i_pts *= h->i_dts_compress_multiplier;
2453     if( h->frames.i_bframe_delay )
2454     {
2455         int64_t *prev_reordered_pts = thread_current->frames.i_prev_reordered_pts;
2456         if( h->i_frame <= h->frames.i_bframe_delay )
2457         {
2458             if( h->i_dts_compress_multiplier == 1 )
2459                 h->fdec->i_dts = h->fenc->i_reordered_pts - h->frames.i_bframe_delay_time;
2460             else
2461             {
2462                 /* DTS compression */
2463                 if( h->i_frame == 1 )
2464                     thread_current->frames.i_init_delta = h->fenc->i_reordered_pts * h->i_dts_compress_multiplier;
2465                 h->fdec->i_dts = h->i_frame * thread_current->frames.i_init_delta / h->i_dts_compress_multiplier;
2466             }
2467         }
2468         else
2469             h->fdec->i_dts = prev_reordered_pts[ (h->i_frame - h->frames.i_bframe_delay) % h->frames.i_bframe_delay ];
2470         prev_reordered_pts[ h->i_frame % h->frames.i_bframe_delay ] = h->fenc->i_reordered_pts * h->i_dts_compress_multiplier;
2471     }
2472     else
2473         h->fdec->i_dts = h->fenc->i_reordered_pts;
2474     if( h->fenc->i_type == X264_TYPE_IDR )
2475         h->i_last_idr_pts = h->fdec->i_pts;
2476
2477     /* ------------------- Init                ----------------------------- */
2478     /* build ref list 0/1 */
2479     x264_reference_build_list( h, h->fdec->i_poc );
2480
2481     /* ---------------------- Write the bitstream -------------------------- */
2482     /* Init bitstream context */
2483     if( h->param.b_sliced_threads )
2484     {
2485         for( int i = 0; i < h->param.i_threads; i++ )
2486         {
2487             bs_init( &h->thread[i]->out.bs, h->thread[i]->out.p_bitstream, h->thread[i]->out.i_bitstream );
2488             h->thread[i]->out.i_nal = 0;
2489         }
2490     }
2491     else
2492     {
2493         bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
2494         h->out.i_nal = 0;
2495     }
2496
2497     if( h->param.b_aud )
2498     {
2499         int pic_type;
2500
2501         if( h->sh.i_type == SLICE_TYPE_I )
2502             pic_type = 0;
2503         else if( h->sh.i_type == SLICE_TYPE_P )
2504             pic_type = 1;
2505         else if( h->sh.i_type == SLICE_TYPE_B )
2506             pic_type = 2;
2507         else
2508             pic_type = 7;
2509
2510         x264_nal_start( h, NAL_AUD, NAL_PRIORITY_DISPOSABLE );
2511         bs_write( &h->out.bs, 3, pic_type );
2512         bs_rbsp_trailing( &h->out.bs );
2513         if( x264_nal_end( h ) )
2514             return -1;
2515         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2516     }
2517
2518     h->i_nal_type = i_nal_type;
2519     h->i_nal_ref_idc = i_nal_ref_idc;
2520
2521     if( h->param.b_intra_refresh )
2522     {
2523         if( IS_X264_TYPE_I( h->fenc->i_type ) )
2524         {
2525             h->fdec->i_frames_since_pir = 0;
2526             h->b_queued_intra_refresh = 0;
2527             /* PIR is currently only supported with ref == 1, so any intra frame effectively refreshes
2528              * the whole frame and counts as an intra refresh. */
2529             h->fdec->f_pir_position = h->mb.i_mb_width;
2530         }
2531         else if( h->fenc->i_type == X264_TYPE_P )
2532         {
2533             int pocdiff = (h->fdec->i_poc - h->fref0[0]->i_poc)/2;
2534             float increment = X264_MAX( ((float)h->mb.i_mb_width-1) / h->param.i_keyint_max, 1 );
2535             h->fdec->f_pir_position = h->fref0[0]->f_pir_position;
2536             h->fdec->i_frames_since_pir = h->fref0[0]->i_frames_since_pir + pocdiff;
2537             if( h->fdec->i_frames_since_pir >= h->param.i_keyint_max ||
2538                 (h->b_queued_intra_refresh && h->fdec->f_pir_position + 0.5 >= h->mb.i_mb_width) )
2539             {
2540                 h->fdec->f_pir_position = 0;
2541                 h->fdec->i_frames_since_pir = 0;
2542                 h->b_queued_intra_refresh = 0;
2543                 h->fenc->b_keyframe = 1;
2544             }
2545             h->fdec->i_pir_start_col = h->fdec->f_pir_position+0.5;
2546             h->fdec->f_pir_position += increment * pocdiff;
2547             h->fdec->i_pir_end_col = h->fdec->f_pir_position+0.5;
2548         }
2549     }
2550
2551     if( h->fenc->b_keyframe )
2552     {
2553         /* Write SPS and PPS */
2554         if( h->param.b_repeat_headers )
2555         {
2556             /* generate sequence parameters */
2557             x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
2558             x264_sps_write( &h->out.bs, h->sps );
2559             if( x264_nal_end( h ) )
2560                 return -1;
2561             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
2562
2563             /* generate picture parameters */
2564             x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
2565             x264_pps_write( &h->out.bs, h->pps );
2566             if( x264_nal_end( h ) )
2567                 return -1;
2568             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
2569         }
2570
2571         /* buffering period sei is written in x264_encoder_frame_end */
2572
2573         if( h->param.b_repeat_headers && h->fenc->i_frame == 0 )
2574         {
2575             /* identify ourself */
2576             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2577             if( x264_sei_version_write( h, &h->out.bs ) )
2578                 return -1;
2579             if( x264_nal_end( h ) )
2580                 return -1;
2581             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2582         }
2583
2584         if( h->fenc->i_type != X264_TYPE_IDR )
2585         {
2586             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;
2587             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2588             x264_sei_recovery_point_write( h, &h->out.bs, time_to_recovery );
2589             x264_nal_end( h );
2590             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2591         }
2592     }
2593
2594     /* generate sei pic timing */
2595     if( h->sps->vui.b_pic_struct_present || h->sps->vui.b_nal_hrd_parameters_present )
2596     {
2597         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2598         x264_sei_pic_timing_write( h, &h->out.bs );
2599         if( x264_nal_end( h ) )
2600             return -1;
2601         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2602     }
2603
2604     /* Init the rate control */
2605     /* FIXME: Include slice header bit cost. */
2606     x264_ratecontrol_start( h, h->fenc->i_qpplus1, overhead*8 );
2607     i_global_qp = x264_ratecontrol_qp( h );
2608
2609     pic_out->i_qpplus1 =
2610     h->fdec->i_qpplus1 = i_global_qp + 1;
2611
2612     if( h->param.rc.b_stat_read && h->sh.i_type != SLICE_TYPE_I )
2613     {
2614         x264_reference_build_list_optimal( h );
2615         x264_reference_check_reorder( h );
2616     }
2617
2618     if( h->i_ref0 )
2619         h->fdec->i_poc_l0ref0 = h->fref0[0]->i_poc;
2620
2621     if( h->sh.i_type == SLICE_TYPE_B )
2622         x264_macroblock_bipred_init( h );
2623
2624     /*------------------------- Weights -------------------------------------*/
2625     x264_weighted_pred_init( h );
2626
2627     /* ------------------------ Create slice header  ----------------------- */
2628     x264_slice_init( h, i_nal_type, i_global_qp );
2629
2630     if( i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE )
2631         h->i_frame_num++;
2632
2633     /* Write frame */
2634     h->i_threadslice_start = 0;
2635     h->i_threadslice_end = h->mb.i_mb_height;
2636     if( h->i_thread_frames > 1 )
2637     {
2638         x264_threadpool_run( h->threadpool, (void*)x264_slices_write, h );
2639         h->b_thread_active = 1;
2640     }
2641     else if( h->param.b_sliced_threads )
2642     {
2643         if( x264_threaded_slices_write( h ) )
2644             return -1;
2645     }
2646     else
2647         if( (intptr_t)x264_slices_write( h ) )
2648             return -1;
2649
2650     return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
2651 }
2652
2653 static int x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
2654                                    x264_nal_t **pp_nal, int *pi_nal,
2655                                    x264_picture_t *pic_out )
2656 {
2657     char psz_message[80];
2658
2659     if( h->b_thread_active )
2660     {
2661         h->b_thread_active = 0;
2662         if( (intptr_t)x264_threadpool_wait( h->threadpool, h ) )
2663             return -1;
2664     }
2665     if( !h->out.i_nal )
2666     {
2667         pic_out->i_type = X264_TYPE_AUTO;
2668         return 0;
2669     }
2670
2671     x264_emms();
2672     /* generate sei buffering period and insert it into place */
2673     if( h->fenc->b_keyframe && h->sps->vui.b_nal_hrd_parameters_present )
2674     {
2675         x264_hrd_fullness( h );
2676         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2677         x264_sei_buffering_period_write( h, &h->out.bs );
2678         if( x264_nal_end( h ) )
2679            return -1;
2680         /* buffering period sei must follow AUD, SPS and PPS and precede all other SEIs */
2681         int idx = 0;
2682         while( h->out.nal[idx].i_type == NAL_AUD ||
2683                h->out.nal[idx].i_type == NAL_SPS ||
2684                h->out.nal[idx].i_type == NAL_PPS )
2685             idx++;
2686         x264_nal_t nal_tmp = h->out.nal[h->out.i_nal-1];
2687         memmove( &h->out.nal[idx+1], &h->out.nal[idx], (h->out.i_nal-idx-1)*sizeof(x264_nal_t) );
2688         h->out.nal[idx] = nal_tmp;
2689     }
2690
2691     int frame_size = x264_encoder_encapsulate_nals( h, 0 );
2692
2693     /* Set output picture properties */
2694     pic_out->i_type = h->fenc->i_type;
2695
2696     pic_out->b_keyframe = h->fenc->b_keyframe;
2697
2698     pic_out->i_pts = h->fdec->i_pts;
2699     pic_out->i_dts = h->fdec->i_dts;
2700
2701     if( pic_out->i_pts < pic_out->i_dts )
2702         x264_log( h, X264_LOG_WARNING, "invalid DTS: PTS is less than DTS\n" );
2703
2704     pic_out->img.i_plane = h->fdec->i_plane;
2705     for( int i = 0; i < 3; i++ )
2706     {
2707         pic_out->img.i_stride[i] = h->fdec->i_stride[i];
2708         pic_out->img.plane[i] = h->fdec->plane[i];
2709     }
2710
2711     x264_frame_push_unused( thread_current, h->fenc );
2712
2713     /* ---------------------- Update encoder state ------------------------- */
2714
2715     /* update rc */
2716     int filler = 0;
2717     if( x264_ratecontrol_end( h, frame_size * 8, &filler ) < 0 )
2718         return -1;
2719
2720     pic_out->hrd_timing = h->fenc->hrd_timing;
2721
2722     while( filler > 0 )
2723     {
2724         int f, overhead;
2725         overhead = (FILLER_OVERHEAD - h->param.b_annexb);
2726         if( h->param.i_slice_max_size && filler > h->param.i_slice_max_size )
2727         {
2728             int next_size = filler - h->param.i_slice_max_size;
2729             int overflow = X264_MAX( overhead - next_size, 0 );
2730             f = h->param.i_slice_max_size - overhead - overflow;
2731         }
2732         else
2733             f = X264_MAX( 0, filler - overhead );
2734
2735         x264_nal_start( h, NAL_FILLER, NAL_PRIORITY_DISPOSABLE );
2736         x264_filler_write( h, &h->out.bs, f );
2737         if( x264_nal_end( h ) )
2738             return -1;
2739         int total_size = x264_encoder_encapsulate_nals( h, h->out.i_nal-1 );
2740         frame_size += total_size;
2741         filler -= total_size;
2742     }
2743
2744     /* End bitstream, set output  */
2745     *pi_nal = h->out.i_nal;
2746     *pp_nal = h->out.nal;
2747
2748     h->out.i_nal = 0;
2749
2750     x264_noise_reduction_update( h );
2751
2752     /* ---------------------- Compute/Print statistics --------------------- */
2753     x264_thread_sync_stat( h, h->thread[0] );
2754
2755     /* Slice stat */
2756     h->stat.i_frame_count[h->sh.i_type]++;
2757     h->stat.i_frame_size[h->sh.i_type] += frame_size;
2758     h->stat.f_frame_qp[h->sh.i_type] += h->fdec->f_qp_avg_aq;
2759
2760     for( int i = 0; i < X264_MBTYPE_MAX; i++ )
2761         h->stat.i_mb_count[h->sh.i_type][i] += h->stat.frame.i_mb_count[i];
2762     for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
2763         h->stat.i_mb_partition[h->sh.i_type][i] += h->stat.frame.i_mb_partition[i];
2764     for( int i = 0; i < 2; i++ )
2765         h->stat.i_mb_count_8x8dct[i] += h->stat.frame.i_mb_count_8x8dct[i];
2766     for( int i = 0; i < 6; i++ )
2767         h->stat.i_mb_cbp[i] += h->stat.frame.i_mb_cbp[i];
2768     for( int i = 0; i < 4; i++ )
2769         for( int j = 0; j < 13; j++ )
2770             h->stat.i_mb_pred_mode[i][j] += h->stat.frame.i_mb_pred_mode[i][j];
2771     if( h->sh.i_type != SLICE_TYPE_I )
2772         for( int i_list = 0; i_list < 2; i_list++ )
2773             for( int i = 0; i < 32; i++ )
2774                 h->stat.i_mb_count_ref[h->sh.i_type][i_list][i] += h->stat.frame.i_mb_count_ref[i_list][i];
2775     if( h->sh.i_type == SLICE_TYPE_P )
2776     {
2777         h->stat.i_consecutive_bframes[h->fdec->i_frame - h->fref0[0]->i_frame - 1]++;
2778         if( h->param.analyse.i_weighted_pred == X264_WEIGHTP_SMART )
2779             for( int i = 0; i < 3; i++ )
2780                 for( int j = 0; j < h->i_ref0; j++ )
2781                     if( h->sh.weight[0][i].i_denom != 0 )
2782                     {
2783                         h->stat.i_wpred[i]++;
2784                         break;
2785                     }
2786     }
2787     if( h->sh.i_type == SLICE_TYPE_B )
2788     {
2789         h->stat.i_direct_frames[ h->sh.b_direct_spatial_mv_pred ] ++;
2790         if( h->mb.b_direct_auto_write )
2791         {
2792             //FIXME somewhat arbitrary time constants
2793             if( h->stat.i_direct_score[0] + h->stat.i_direct_score[1] > h->mb.i_mb_count )
2794                 for( int i = 0; i < 2; i++ )
2795                     h->stat.i_direct_score[i] = h->stat.i_direct_score[i] * 9/10;
2796             for( int i = 0; i < 2; i++ )
2797                 h->stat.i_direct_score[i] += h->stat.frame.i_direct_score[i];
2798         }
2799     }
2800
2801     psz_message[0] = '\0';
2802     if( h->param.analyse.b_psnr )
2803     {
2804         int64_t ssd[3] = {
2805             h->stat.frame.i_ssd[0],
2806             h->stat.frame.i_ssd[1],
2807             h->stat.frame.i_ssd[2],
2808         };
2809
2810         h->stat.i_ssd_global[h->sh.i_type] += ssd[0] + ssd[1] + ssd[2];
2811         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 );
2812         h->stat.f_psnr_mean_y[h->sh.i_type] += x264_psnr( ssd[0], h->param.i_width * h->param.i_height );
2813         h->stat.f_psnr_mean_u[h->sh.i_type] += x264_psnr( ssd[1], h->param.i_width * h->param.i_height / 4 );
2814         h->stat.f_psnr_mean_v[h->sh.i_type] += x264_psnr( ssd[2], h->param.i_width * h->param.i_height / 4 );
2815
2816         snprintf( psz_message, 80, " PSNR Y:%5.2f U:%5.2f V:%5.2f",
2817                   x264_psnr( ssd[0], h->param.i_width * h->param.i_height ),
2818                   x264_psnr( ssd[1], h->param.i_width * h->param.i_height / 4),
2819                   x264_psnr( ssd[2], h->param.i_width * h->param.i_height / 4) );
2820     }
2821
2822     if( h->param.analyse.b_ssim )
2823     {
2824         double ssim_y = h->stat.frame.f_ssim
2825                       / (((h->param.i_width-6)>>2) * ((h->param.i_height-6)>>2));
2826         h->stat.f_ssim_mean_y[h->sh.i_type] += ssim_y;
2827         snprintf( psz_message + strlen(psz_message), 80 - strlen(psz_message),
2828                   " SSIM Y:%.5f", ssim_y );
2829     }
2830     psz_message[79] = '\0';
2831
2832     x264_log( h, X264_LOG_DEBUG,
2833                   "frame=%4d QP=%.2f NAL=%d Slice:%c Poc:%-3d I:%-4d P:%-4d SKIP:%-4d size=%d bytes%s\n",
2834               h->i_frame,
2835               h->fdec->f_qp_avg_aq,
2836               h->i_nal_ref_idc,
2837               h->sh.i_type == SLICE_TYPE_I ? 'I' : (h->sh.i_type == SLICE_TYPE_P ? 'P' : 'B' ),
2838               h->fdec->i_poc,
2839               h->stat.frame.i_mb_count_i,
2840               h->stat.frame.i_mb_count_p,
2841               h->stat.frame.i_mb_count_skip,
2842               frame_size,
2843               psz_message );
2844
2845     // keep stats all in one place
2846     x264_thread_sync_stat( h->thread[0], h );
2847     // for the use of the next frame
2848     x264_thread_sync_stat( thread_current, h );
2849
2850 #ifdef DEBUG_MB_TYPE
2851 {
2852     static const char mb_chars[] = { 'i', 'i', 'I', 'C', 'P', '8', 'S',
2853         'D', '<', 'X', 'B', 'X', '>', 'B', 'B', 'B', 'B', '8', 'S' };
2854     for( int mb_xy = 0; mb_xy < h->mb.i_mb_width * h->mb.i_mb_height; mb_xy++ )
2855     {
2856         if( h->mb.type[mb_xy] < X264_MBTYPE_MAX && h->mb.type[mb_xy] >= 0 )
2857             fprintf( stderr, "%c ", mb_chars[ h->mb.type[mb_xy] ] );
2858         else
2859             fprintf( stderr, "? " );
2860
2861         if( (mb_xy+1) % h->mb.i_mb_width == 0 )
2862             fprintf( stderr, "\n" );
2863     }
2864 }
2865 #endif
2866
2867     /* Remove duplicates, must be done near the end as breaks h->fref0 array
2868      * by freeing some of its pointers. */
2869      for( int i = 0; i < h->i_ref0; i++ )
2870          if( h->fref0[i] && h->fref0[i]->b_duplicate )
2871          {
2872              x264_frame_push_blank_unused( h, h->fref0[i] );
2873              h->fref0[i] = 0;
2874          }
2875
2876     if( h->param.psz_dump_yuv )
2877         x264_frame_dump( h );
2878
2879     return frame_size;
2880 }
2881
2882 static void x264_print_intra( int64_t *i_mb_count, double i_count, int b_print_pcm, char *intra )
2883 {
2884     intra += sprintf( intra, "I16..4%s: %4.1f%% %4.1f%% %4.1f%%",
2885         b_print_pcm ? "..PCM" : "",
2886         i_mb_count[I_16x16]/ i_count,
2887         i_mb_count[I_8x8]  / i_count,
2888         i_mb_count[I_4x4]  / i_count );
2889     if( b_print_pcm )
2890         sprintf( intra, " %4.1f%%", i_mb_count[I_PCM]  / i_count );
2891 }
2892
2893 /****************************************************************************
2894  * x264_encoder_close:
2895  ****************************************************************************/
2896 void    x264_encoder_close  ( x264_t *h )
2897 {
2898     int64_t i_yuv_size = 3 * h->param.i_width * h->param.i_height / 2;
2899     int64_t i_mb_count_size[2][7] = {{0}};
2900     char buf[200];
2901     int b_print_pcm = h->stat.i_mb_count[SLICE_TYPE_I][I_PCM]
2902                    || h->stat.i_mb_count[SLICE_TYPE_P][I_PCM]
2903                    || h->stat.i_mb_count[SLICE_TYPE_B][I_PCM];
2904
2905     x264_lookahead_delete( h );
2906
2907     if( h->param.i_threads > 1 )
2908         x264_threadpool_delete( h->threadpool );
2909     if( h->i_thread_frames > 1 )
2910     {
2911         for( int i = 0; i < h->i_thread_frames; i++ )
2912             if( h->thread[i]->b_thread_active )
2913             {
2914                 assert( h->thread[i]->fenc->i_reference_count == 1 );
2915                 x264_frame_delete( h->thread[i]->fenc );
2916             }
2917
2918         x264_t *thread_prev = h->thread[h->i_thread_phase];
2919         x264_thread_sync_ratecontrol( h, thread_prev, h );
2920         x264_thread_sync_ratecontrol( thread_prev, thread_prev, h );
2921         h->i_frame = thread_prev->i_frame + 1 - h->i_thread_frames;
2922     }
2923     h->i_frame++;
2924
2925     /* Slices used and PSNR */
2926     for( int i = 0; i < 5; i++ )
2927     {
2928         static const uint8_t slice_order[] = { SLICE_TYPE_I, SLICE_TYPE_SI, SLICE_TYPE_P, SLICE_TYPE_SP, SLICE_TYPE_B };
2929         static const char * const slice_name[] = { "P", "B", "I", "SP", "SI" };
2930         int i_slice = slice_order[i];
2931
2932         if( h->stat.i_frame_count[i_slice] > 0 )
2933         {
2934             const int i_count = h->stat.i_frame_count[i_slice];
2935             if( h->param.analyse.b_psnr )
2936             {
2937                 x264_log( h, X264_LOG_INFO,
2938                           "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",
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                           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,
2944                           h->stat.f_psnr_average[i_slice] / i_count,
2945                           x264_psnr( h->stat.i_ssd_global[i_slice], i_count * i_yuv_size ) );
2946             }
2947             else
2948             {
2949                 x264_log( h, X264_LOG_INFO,
2950                           "frame %s:%-5d Avg QP:%5.2f  size:%6.0f\n",
2951                           slice_name[i_slice],
2952                           i_count,
2953                           h->stat.f_frame_qp[i_slice] / i_count,
2954                           (double)h->stat.i_frame_size[i_slice] / i_count );
2955             }
2956         }
2957     }
2958     if( h->param.i_bframe && h->stat.i_frame_count[SLICE_TYPE_P] )
2959     {
2960         char *p = buf;
2961         int den = 0;
2962         // weight by number of frames (including the P-frame) that are in a sequence of N B-frames
2963         for( int i = 0; i <= h->param.i_bframe; i++ )
2964             den += (i+1) * h->stat.i_consecutive_bframes[i];
2965         for( int i = 0; i <= h->param.i_bframe; i++ )
2966             p += sprintf( p, " %4.1f%%", 100. * (i+1) * h->stat.i_consecutive_bframes[i] / den );
2967         x264_log( h, X264_LOG_INFO, "consecutive B-frames:%s\n", buf );
2968     }
2969
2970     for( int i_type = 0; i_type < 2; i_type++ )
2971         for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
2972         {
2973             if( i == D_DIRECT_8x8 ) continue; /* direct is counted as its own type */
2974             i_mb_count_size[i_type][x264_mb_partition_pixel_table[i]] += h->stat.i_mb_partition[i_type][i];
2975         }
2976
2977     /* MB types used */
2978     if( h->stat.i_frame_count[SLICE_TYPE_I] > 0 )
2979     {
2980         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_I];
2981         double i_count = h->stat.i_frame_count[SLICE_TYPE_I] * h->mb.i_mb_count / 100.0;
2982         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
2983         x264_log( h, X264_LOG_INFO, "mb I  %s\n", buf );
2984     }
2985     if( h->stat.i_frame_count[SLICE_TYPE_P] > 0 )
2986     {
2987         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_P];
2988         double i_count = h->stat.i_frame_count[SLICE_TYPE_P] * h->mb.i_mb_count / 100.0;
2989         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_P];
2990         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
2991         x264_log( h, X264_LOG_INFO,
2992                   "mb P  %s  P16..4: %4.1f%% %4.1f%% %4.1f%% %4.1f%% %4.1f%%    skip:%4.1f%%\n",
2993                   buf,
2994                   i_mb_size[PIXEL_16x16] / (i_count*4),
2995                   (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
2996                   i_mb_size[PIXEL_8x8] / (i_count*4),
2997                   (i_mb_size[PIXEL_8x4] + i_mb_size[PIXEL_4x8]) / (i_count*4),
2998                   i_mb_size[PIXEL_4x4] / (i_count*4),
2999                   i_mb_count[P_SKIP] / i_count );
3000     }
3001     if( h->stat.i_frame_count[SLICE_TYPE_B] > 0 )
3002     {
3003         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_B];
3004         double i_count = h->stat.i_frame_count[SLICE_TYPE_B] * h->mb.i_mb_count / 100.0;
3005         double i_mb_list_count;
3006         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_B];
3007         int64_t list_count[3] = {0}; /* 0 == L0, 1 == L1, 2 == BI */
3008         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
3009         for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
3010             for( int j = 0; j < 2; j++ )
3011             {
3012                 int l0 = x264_mb_type_list_table[i][0][j];
3013                 int l1 = x264_mb_type_list_table[i][1][j];
3014                 if( l0 || l1 )
3015                     list_count[l1+l0*l1] += h->stat.i_mb_count[SLICE_TYPE_B][i] * 2;
3016             }
3017         list_count[0] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L0_8x8];
3018         list_count[1] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L1_8x8];
3019         list_count[2] += h->stat.i_mb_partition[SLICE_TYPE_B][D_BI_8x8];
3020         i_mb_count[B_DIRECT] += (h->stat.i_mb_partition[SLICE_TYPE_B][D_DIRECT_8x8]+2)/4;
3021         i_mb_list_count = (list_count[0] + list_count[1] + list_count[2]) / 100.0;
3022         x264_log( h, X264_LOG_INFO,
3023                   "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",
3024                   buf,
3025                   i_mb_size[PIXEL_16x16] / (i_count*4),
3026                   (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
3027                   i_mb_size[PIXEL_8x8] / (i_count*4),
3028                   i_mb_count[B_DIRECT] / i_count,
3029                   i_mb_count[B_SKIP]   / i_count,
3030                   list_count[0] / i_mb_list_count,
3031                   list_count[1] / i_mb_list_count,
3032                   list_count[2] / i_mb_list_count );
3033     }
3034
3035     x264_ratecontrol_summary( h );
3036
3037     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 )
3038     {
3039 #define SUM3(p) (p[SLICE_TYPE_I] + p[SLICE_TYPE_P] + p[SLICE_TYPE_B])
3040 #define SUM3b(p,o) (p[SLICE_TYPE_I][o] + p[SLICE_TYPE_P][o] + p[SLICE_TYPE_B][o])
3041         int64_t i_i8x8 = SUM3b( h->stat.i_mb_count, I_8x8 );
3042         int64_t i_intra = i_i8x8 + SUM3b( h->stat.i_mb_count, I_4x4 )
3043                                  + SUM3b( h->stat.i_mb_count, I_16x16 );
3044         int64_t i_all_intra = i_intra + SUM3b( h->stat.i_mb_count, I_PCM);
3045         const int i_count = h->stat.i_frame_count[SLICE_TYPE_I] +
3046                             h->stat.i_frame_count[SLICE_TYPE_P] +
3047                             h->stat.i_frame_count[SLICE_TYPE_B];
3048         int64_t i_mb_count = i_count * h->mb.i_mb_count;
3049         float fps = (float) h->param.i_fps_num / h->param.i_fps_den;
3050         float f_bitrate;
3051         /* duration algorithm fails with one frame */
3052         if( !h->param.b_vfr_input || i_count == 1 )
3053             f_bitrate = fps * SUM3(h->stat.i_frame_size) / i_count / 125;
3054         else
3055         {
3056             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;
3057             f_bitrate = SUM3(h->stat.i_frame_size) / duration / 125;
3058         }
3059
3060         if( h->pps->b_transform_8x8_mode )
3061         {
3062             buf[0] = 0;
3063             if( h->stat.i_mb_count_8x8dct[0] )
3064                 sprintf( buf, " inter:%.1f%%", 100. * h->stat.i_mb_count_8x8dct[1] / h->stat.i_mb_count_8x8dct[0] );
3065             x264_log( h, X264_LOG_INFO, "8x8 transform intra:%.1f%%%s\n", 100. * i_i8x8 / i_intra, buf );
3066         }
3067
3068         if( (h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO ||
3069             (h->stat.i_direct_frames[0] && h->stat.i_direct_frames[1]))
3070             && h->stat.i_frame_count[SLICE_TYPE_B] )
3071         {
3072             x264_log( h, X264_LOG_INFO, "direct mvs  spatial:%.1f%% temporal:%.1f%%\n",
3073                       h->stat.i_direct_frames[1] * 100. / h->stat.i_frame_count[SLICE_TYPE_B],
3074                       h->stat.i_direct_frames[0] * 100. / h->stat.i_frame_count[SLICE_TYPE_B] );
3075         }
3076
3077         buf[0] = 0;
3078         if( i_mb_count != i_all_intra )
3079             sprintf( buf, " inter: %.1f%% %.1f%% %.1f%%",
3080                      h->stat.i_mb_cbp[1] * 100.0 / ((i_mb_count - i_all_intra)*4),
3081                      h->stat.i_mb_cbp[3] * 100.0 / ((i_mb_count - i_all_intra)  ),
3082                      h->stat.i_mb_cbp[5] * 100.0 / ((i_mb_count - i_all_intra)) );
3083         x264_log( h, X264_LOG_INFO, "coded y,uvDC,uvAC intra: %.1f%% %.1f%% %.1f%%%s\n",
3084                   h->stat.i_mb_cbp[0] * 100.0 / (i_all_intra*4),
3085                   h->stat.i_mb_cbp[2] * 100.0 / (i_all_intra  ),
3086                   h->stat.i_mb_cbp[4] * 100.0 / (i_all_intra  ), buf );
3087
3088         int64_t fixed_pred_modes[4][9] = {{0}};
3089         int64_t sum_pred_modes[4] = {0};
3090         for( int i = 0; i <= I_PRED_16x16_DC_128; i++ )
3091         {
3092             fixed_pred_modes[0][x264_mb_pred_mode16x16_fix[i]] += h->stat.i_mb_pred_mode[0][i];
3093             sum_pred_modes[0] += h->stat.i_mb_pred_mode[0][i];
3094         }
3095         if( sum_pred_modes[0] )
3096             x264_log( h, X264_LOG_INFO, "i16 v,h,dc,p: %2.0f%% %2.0f%% %2.0f%% %2.0f%%\n",
3097                       fixed_pred_modes[0][0] * 100.0 / sum_pred_modes[0],
3098                       fixed_pred_modes[0][1] * 100.0 / sum_pred_modes[0],
3099                       fixed_pred_modes[0][2] * 100.0 / sum_pred_modes[0],
3100                       fixed_pred_modes[0][3] * 100.0 / sum_pred_modes[0] );
3101         for( int i = 1; i <= 2; i++ )
3102         {
3103             for( int j = 0; j <= I_PRED_8x8_DC_128; j++ )
3104             {
3105                 fixed_pred_modes[i][x264_mb_pred_mode4x4_fix(j)] += h->stat.i_mb_pred_mode[i][j];
3106                 sum_pred_modes[i] += h->stat.i_mb_pred_mode[i][j];
3107             }
3108             if( sum_pred_modes[i] )
3109                 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,
3110                           fixed_pred_modes[i][0] * 100.0 / sum_pred_modes[i],
3111                           fixed_pred_modes[i][1] * 100.0 / sum_pred_modes[i],
3112                           fixed_pred_modes[i][2] * 100.0 / sum_pred_modes[i],
3113                           fixed_pred_modes[i][3] * 100.0 / sum_pred_modes[i],
3114                           fixed_pred_modes[i][4] * 100.0 / sum_pred_modes[i],
3115                           fixed_pred_modes[i][5] * 100.0 / sum_pred_modes[i],
3116                           fixed_pred_modes[i][6] * 100.0 / sum_pred_modes[i],
3117                           fixed_pred_modes[i][7] * 100.0 / sum_pred_modes[i],
3118                           fixed_pred_modes[i][8] * 100.0 / sum_pred_modes[i] );
3119         }
3120         for( int i = 0; i <= I_PRED_CHROMA_DC_128; i++ )
3121         {
3122             fixed_pred_modes[3][x264_mb_pred_mode8x8c_fix[i]] += h->stat.i_mb_pred_mode[3][i];
3123             sum_pred_modes[3] += h->stat.i_mb_pred_mode[3][i];
3124         }
3125         if( sum_pred_modes[3] )
3126             x264_log( h, X264_LOG_INFO, "i8c dc,h,v,p: %2.0f%% %2.0f%% %2.0f%% %2.0f%%\n",
3127                       fixed_pred_modes[3][0] * 100.0 / sum_pred_modes[3],
3128                       fixed_pred_modes[3][1] * 100.0 / sum_pred_modes[3],
3129                       fixed_pred_modes[3][2] * 100.0 / sum_pred_modes[3],
3130                       fixed_pred_modes[3][3] * 100.0 / sum_pred_modes[3] );
3131
3132         if( h->param.analyse.i_weighted_pred == X264_WEIGHTP_SMART && h->stat.i_frame_count[SLICE_TYPE_P] > 0 )
3133             x264_log( h, X264_LOG_INFO, "Weighted P-Frames: Y:%.1f%%\n",
3134                       h->stat.i_wpred[0] * 100.0 / h->stat.i_frame_count[SLICE_TYPE_P] );
3135
3136         for( int i_list = 0; i_list < 2; i_list++ )
3137             for( int i_slice = 0; i_slice < 2; i_slice++ )
3138             {
3139                 char *p = buf;
3140                 int64_t i_den = 0;
3141                 int i_max = 0;
3142                 for( int i = 0; i < 32; i++ )
3143                     if( h->stat.i_mb_count_ref[i_slice][i_list][i] )
3144                     {
3145                         i_den += h->stat.i_mb_count_ref[i_slice][i_list][i];
3146                         i_max = i;
3147                     }
3148                 if( i_max == 0 )
3149                     continue;
3150                 for( int i = 0; i <= i_max; i++ )
3151                     p += sprintf( p, " %4.1f%%", 100. * h->stat.i_mb_count_ref[i_slice][i_list][i] / i_den );
3152                 x264_log( h, X264_LOG_INFO, "ref %c L%d:%s\n", "PB"[i_slice], i_list, buf );
3153             }
3154
3155         if( h->param.analyse.b_ssim )
3156         {
3157             float ssim = SUM3( h->stat.f_ssim_mean_y ) / i_count;
3158             x264_log( h, X264_LOG_INFO, "SSIM Mean Y:%.7f (%6.3fdb)\n", ssim, x264_ssim( ssim ) );
3159         }
3160         if( h->param.analyse.b_psnr )
3161         {
3162             x264_log( h, X264_LOG_INFO,
3163                       "PSNR Mean Y:%6.3f U:%6.3f V:%6.3f Avg:%6.3f Global:%6.3f kb/s:%.2f\n",
3164                       SUM3( h->stat.f_psnr_mean_y ) / i_count,
3165                       SUM3( h->stat.f_psnr_mean_u ) / i_count,
3166                       SUM3( h->stat.f_psnr_mean_v ) / i_count,
3167                       SUM3( h->stat.f_psnr_average ) / i_count,
3168                       x264_psnr( SUM3( h->stat.i_ssd_global ), i_count * i_yuv_size ),
3169                       f_bitrate );
3170         }
3171         else
3172             x264_log( h, X264_LOG_INFO, "kb/s:%.2f\n", f_bitrate );
3173     }
3174
3175     /* rc */
3176     x264_ratecontrol_delete( h );
3177
3178     /* param */
3179     if( h->param.rc.psz_stat_out )
3180         free( h->param.rc.psz_stat_out );
3181     if( h->param.rc.psz_stat_in )
3182         free( h->param.rc.psz_stat_in );
3183
3184     x264_cqm_delete( h );
3185     x264_free( h->nal_buffer );
3186     x264_analyse_free_costs( h );
3187
3188     if( h->i_thread_frames > 1)
3189         h = h->thread[h->i_thread_phase];
3190
3191     /* frames */
3192     x264_frame_delete_list( h->frames.unused[0] );
3193     x264_frame_delete_list( h->frames.unused[1] );
3194     x264_frame_delete_list( h->frames.current );
3195     x264_frame_delete_list( h->frames.blank_unused );
3196
3197     h = h->thread[0];
3198
3199     for( int i = h->param.i_threads - 1; i >= 0; i-- )
3200     {
3201         x264_frame_t **frame;
3202
3203         if( !h->param.b_sliced_threads || i == 0 )
3204         {
3205             for( frame = h->thread[i]->frames.reference; *frame; frame++ )
3206             {
3207                 assert( (*frame)->i_reference_count > 0 );
3208                 (*frame)->i_reference_count--;
3209                 if( (*frame)->i_reference_count == 0 )
3210                     x264_frame_delete( *frame );
3211             }
3212             frame = &h->thread[i]->fdec;
3213             assert( (*frame)->i_reference_count > 0 );
3214             (*frame)->i_reference_count--;
3215             if( (*frame)->i_reference_count == 0 )
3216                 x264_frame_delete( *frame );
3217             x264_macroblock_cache_free( h->thread[i] );
3218         }
3219         x264_macroblock_thread_free( h->thread[i], 0 );
3220         x264_free( h->thread[i]->out.p_bitstream );
3221         x264_free( h->thread[i]->out.nal);
3222         x264_free( h->thread[i] );
3223     }
3224 }
3225
3226 /****************************************************************************
3227  * x264_encoder_delayed_frames:
3228  ****************************************************************************/
3229 int x264_encoder_delayed_frames( x264_t *h )
3230 {
3231     int delayed_frames = 0;
3232     if( h->i_thread_frames > 1 )
3233     {
3234         for( int i = 0; i < h->i_thread_frames; i++ )
3235             delayed_frames += h->thread[i]->b_thread_active;
3236         h = h->thread[h->i_thread_phase];
3237     }
3238     for( int i = 0; h->frames.current[i]; i++ )
3239         delayed_frames++;
3240     x264_pthread_mutex_lock( &h->lookahead->ofbuf.mutex );
3241     x264_pthread_mutex_lock( &h->lookahead->ifbuf.mutex );
3242     x264_pthread_mutex_lock( &h->lookahead->next.mutex );
3243     delayed_frames += h->lookahead->ifbuf.i_size + h->lookahead->next.i_size + h->lookahead->ofbuf.i_size;
3244     x264_pthread_mutex_unlock( &h->lookahead->next.mutex );
3245     x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
3246     x264_pthread_mutex_unlock( &h->lookahead->ofbuf.mutex );
3247     return delayed_frames;
3248 }