Fix two bugs in slice-min-mbs and slices-max
[x262.git] / encoder / encoder.c
1 /*****************************************************************************
2  * encoder.c: top-level encoder functions
3  *****************************************************************************
4  * Copyright (C) 2003-2013 x264 project
5  *
6  * Authors: Laurent Aimar <fenrir@via.ecp.fr>
7  *          Loren Merritt <lorenm@u.washington.edu>
8  *          Jason Garrett-Glaser <darkshikari@gmail.com>
9  *
10  * This program is free software; you can redistribute it and/or modify
11  * it under the terms of the GNU General Public License as published by
12  * the Free Software Foundation; either version 2 of the License, or
13  * (at your option) any later version.
14  *
15  * This program is distributed in the hope that it will be useful,
16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
18  * GNU General Public License for more details.
19  *
20  * You should have received a copy of the GNU General Public License
21  * along with this program; if not, write to the Free Software
22  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02111, USA.
23  *
24  * This program is also available under a commercial proprietary license.
25  * For more information, contact us at licensing@x264.com.
26  *****************************************************************************/
27
28 #include "common/common.h"
29
30 #include "set.h"
31 #include "analyse.h"
32 #include "ratecontrol.h"
33 #include "macroblock.h"
34 #include "me.h"
35
36 #if HAVE_VISUALIZE
37 #include "common/visualize.h"
38 #endif
39
40 #if HAVE_OPENCL
41 #include "common/opencl.h"
42 #endif
43
44 //#define DEBUG_MB_TYPE
45
46 #define bs_write_ue bs_write_ue_big
47
48 static int x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
49                                    x264_nal_t **pp_nal, int *pi_nal,
50                                    x264_picture_t *pic_out );
51
52 /****************************************************************************
53  *
54  ******************************* x264 libs **********************************
55  *
56  ****************************************************************************/
57 static double x264_psnr( double sqe, double size )
58 {
59     double mse = sqe / (PIXEL_MAX*PIXEL_MAX * size);
60     if( mse <= 0.0000000001 ) /* Max 100dB */
61         return 100;
62
63     return -10.0 * log10( mse );
64 }
65
66 static double x264_ssim( double ssim )
67 {
68     double inv_ssim = 1 - ssim;
69     if( inv_ssim <= 0.0000000001 ) /* Max 100dB */
70         return 100;
71
72     return -10.0 * log10( inv_ssim );
73 }
74
75 static int x264_threadpool_wait_all( x264_t *h )
76 {
77     for( int i = 0; i < h->param.i_threads; i++ )
78         if( h->thread[i]->b_thread_active )
79         {
80             h->thread[i]->b_thread_active = 0;
81             if( (intptr_t)x264_threadpool_wait( h->threadpool, h->thread[i] ) < 0 )
82                 return -1;
83         }
84     return 0;
85 }
86
87 static void x264_frame_dump( x264_t *h )
88 {
89     FILE *f = fopen( h->param.psz_dump_yuv, "r+b" );
90     if( !f )
91         return;
92
93     /* Wait for the threads to finish deblocking */
94     if( h->param.b_sliced_threads )
95         x264_threadpool_wait_all( h );
96
97     /* Write the frame in display order */
98     int frame_size = FRAME_SIZE( h->param.i_height * h->param.i_width * sizeof(pixel) );
99     fseek( f, (uint64_t)h->fdec->i_frame * frame_size, SEEK_SET );
100     for( int p = 0; p < (CHROMA444 ? 3 : 1); p++ )
101         for( int y = 0; y < h->param.i_height; y++ )
102             fwrite( &h->fdec->plane[p][y*h->fdec->i_stride[p]], sizeof(pixel), h->param.i_width, f );
103     if( !CHROMA444 )
104     {
105         int cw = h->param.i_width>>1;
106         int ch = h->param.i_height>>CHROMA_V_SHIFT;
107         pixel *planeu = x264_malloc( (cw*ch*2+32)*sizeof(pixel) );
108         pixel *planev = planeu + cw*ch + 16;
109         h->mc.plane_copy_deinterleave( planeu, cw, planev, cw, h->fdec->plane[1], h->fdec->i_stride[1], cw, ch );
110         fwrite( planeu, 1, cw*ch*sizeof(pixel), f );
111         fwrite( planev, 1, cw*ch*sizeof(pixel), f );
112         x264_free( planeu );
113     }
114     fclose( f );
115 }
116
117 /* Fill "default" values */
118 static void x264_slice_header_init( x264_t *h, x264_slice_header_t *sh,
119                                     x264_sps_t *sps, x264_pps_t *pps,
120                                     int i_idr_pic_id, int i_frame, int i_qp )
121 {
122     x264_param_t *param = &h->param;
123
124     /* First we fill all fields */
125     sh->sps = sps;
126     sh->pps = pps;
127
128     sh->i_first_mb  = 0;
129     sh->i_last_mb   = h->mb.i_mb_count - 1;
130     sh->i_pps_id    = pps->i_id;
131
132     sh->i_frame_num = i_frame;
133
134     sh->b_mbaff = PARAM_INTERLACED;
135     sh->b_field_pic = 0;    /* no field support for now */
136     sh->b_bottom_field = 0; /* not yet used */
137
138     sh->i_idr_pic_id = i_idr_pic_id;
139
140     /* poc stuff, fixed later */
141     sh->i_poc = 0;
142     sh->i_delta_poc_bottom = 0;
143     sh->i_delta_poc[0] = 0;
144     sh->i_delta_poc[1] = 0;
145
146     sh->i_redundant_pic_cnt = 0;
147
148     h->mb.b_direct_auto_write = h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO
149                                 && h->param.i_bframe
150                                 && ( h->param.rc.b_stat_write || !h->param.rc.b_stat_read );
151
152     if( !h->mb.b_direct_auto_read && sh->i_type == SLICE_TYPE_B )
153     {
154         if( h->fref[1][0]->i_poc_l0ref0 == h->fref[0][0]->i_poc )
155         {
156             if( h->mb.b_direct_auto_write )
157                 sh->b_direct_spatial_mv_pred = ( h->stat.i_direct_score[1] > h->stat.i_direct_score[0] );
158             else
159                 sh->b_direct_spatial_mv_pred = ( param->analyse.i_direct_mv_pred == X264_DIRECT_PRED_SPATIAL );
160         }
161         else
162         {
163             h->mb.b_direct_auto_write = 0;
164             sh->b_direct_spatial_mv_pred = 1;
165         }
166     }
167     /* else b_direct_spatial_mv_pred was read from the 2pass statsfile */
168
169     sh->b_num_ref_idx_override = 0;
170     sh->i_num_ref_idx_l0_active = 1;
171     sh->i_num_ref_idx_l1_active = 1;
172
173     sh->b_ref_pic_list_reordering[0] = h->b_ref_reorder[0];
174     sh->b_ref_pic_list_reordering[1] = h->b_ref_reorder[1];
175
176     /* If the ref list isn't in the default order, construct reordering header */
177     for( int list = 0; list < 2; list++ )
178     {
179         if( sh->b_ref_pic_list_reordering[list] )
180         {
181             int pred_frame_num = i_frame;
182             for( int i = 0; i < h->i_ref[list]; i++ )
183             {
184                 int diff = h->fref[list][i]->i_frame_num - pred_frame_num;
185                 sh->ref_pic_list_order[list][i].idc = ( diff > 0 );
186                 sh->ref_pic_list_order[list][i].arg = (abs(diff) - 1) & ((1 << sps->i_log2_max_frame_num) - 1);
187                 pred_frame_num = h->fref[list][i]->i_frame_num;
188             }
189         }
190     }
191
192     sh->i_cabac_init_idc = param->i_cabac_init_idc;
193
194     sh->i_qp = SPEC_QP(i_qp);
195     sh->i_qp_delta = sh->i_qp - pps->i_pic_init_qp;
196     sh->b_sp_for_swidth = 0;
197     sh->i_qs_delta = 0;
198
199     int deblock_thresh = i_qp + 2 * X264_MIN(param->i_deblocking_filter_alphac0, param->i_deblocking_filter_beta);
200     /* If effective qp <= 15, deblocking would have no effect anyway */
201     if( param->b_deblocking_filter && (h->mb.b_variable_qp || 15 < deblock_thresh ) )
202         sh->i_disable_deblocking_filter_idc = param->b_sliced_threads ? 2 : 0;
203     else
204         sh->i_disable_deblocking_filter_idc = 1;
205     sh->i_alpha_c0_offset = param->i_deblocking_filter_alphac0 << 1;
206     sh->i_beta_offset = param->i_deblocking_filter_beta << 1;
207 }
208
209 static void x264_slice_header_write( bs_t *s, x264_slice_header_t *sh, int i_nal_ref_idc )
210 {
211     if( sh->b_mbaff )
212     {
213         int first_x = sh->i_first_mb % sh->sps->i_mb_width;
214         int first_y = sh->i_first_mb / sh->sps->i_mb_width;
215         assert( (first_y&1) == 0 );
216         bs_write_ue( s, (2*first_x + sh->sps->i_mb_width*(first_y&~1) + (first_y&1)) >> 1 );
217     }
218     else
219         bs_write_ue( s, sh->i_first_mb );
220
221     bs_write_ue( s, sh->i_type + 5 );   /* same type things */
222     bs_write_ue( s, sh->i_pps_id );
223     bs_write( s, sh->sps->i_log2_max_frame_num, sh->i_frame_num & ((1<<sh->sps->i_log2_max_frame_num)-1) );
224
225     if( !sh->sps->b_frame_mbs_only )
226     {
227         bs_write1( s, sh->b_field_pic );
228         if( sh->b_field_pic )
229             bs_write1( s, sh->b_bottom_field );
230     }
231
232     if( sh->i_idr_pic_id >= 0 ) /* NAL IDR */
233         bs_write_ue( s, sh->i_idr_pic_id );
234
235     if( sh->sps->i_poc_type == 0 )
236     {
237         bs_write( s, sh->sps->i_log2_max_poc_lsb, sh->i_poc & ((1<<sh->sps->i_log2_max_poc_lsb)-1) );
238         if( sh->pps->b_pic_order && !sh->b_field_pic )
239             bs_write_se( s, sh->i_delta_poc_bottom );
240     }
241
242     if( sh->pps->b_redundant_pic_cnt )
243         bs_write_ue( s, sh->i_redundant_pic_cnt );
244
245     if( sh->i_type == SLICE_TYPE_B )
246         bs_write1( s, sh->b_direct_spatial_mv_pred );
247
248     if( sh->i_type == SLICE_TYPE_P || sh->i_type == SLICE_TYPE_B )
249     {
250         bs_write1( s, sh->b_num_ref_idx_override );
251         if( sh->b_num_ref_idx_override )
252         {
253             bs_write_ue( s, sh->i_num_ref_idx_l0_active - 1 );
254             if( sh->i_type == SLICE_TYPE_B )
255                 bs_write_ue( s, sh->i_num_ref_idx_l1_active - 1 );
256         }
257     }
258
259     /* ref pic list reordering */
260     if( sh->i_type != SLICE_TYPE_I )
261     {
262         bs_write1( s, sh->b_ref_pic_list_reordering[0] );
263         if( sh->b_ref_pic_list_reordering[0] )
264         {
265             for( int i = 0; i < sh->i_num_ref_idx_l0_active; i++ )
266             {
267                 bs_write_ue( s, sh->ref_pic_list_order[0][i].idc );
268                 bs_write_ue( s, sh->ref_pic_list_order[0][i].arg );
269             }
270             bs_write_ue( s, 3 );
271         }
272     }
273     if( sh->i_type == SLICE_TYPE_B )
274     {
275         bs_write1( s, sh->b_ref_pic_list_reordering[1] );
276         if( sh->b_ref_pic_list_reordering[1] )
277         {
278             for( int i = 0; i < sh->i_num_ref_idx_l1_active; i++ )
279             {
280                 bs_write_ue( s, sh->ref_pic_list_order[1][i].idc );
281                 bs_write_ue( s, sh->ref_pic_list_order[1][i].arg );
282             }
283             bs_write_ue( s, 3 );
284         }
285     }
286
287     sh->b_weighted_pred = 0;
288     if( sh->pps->b_weighted_pred && sh->i_type == SLICE_TYPE_P )
289     {
290         sh->b_weighted_pred = sh->weight[0][0].weightfn || sh->weight[0][1].weightfn || sh->weight[0][2].weightfn;
291         /* pred_weight_table() */
292         bs_write_ue( s, sh->weight[0][0].i_denom );
293         bs_write_ue( s, sh->weight[0][1].i_denom );
294         for( int i = 0; i < sh->i_num_ref_idx_l0_active; i++ )
295         {
296             int luma_weight_l0_flag = !!sh->weight[i][0].weightfn;
297             int chroma_weight_l0_flag = !!sh->weight[i][1].weightfn || !!sh->weight[i][2].weightfn;
298             bs_write1( s, luma_weight_l0_flag );
299             if( luma_weight_l0_flag )
300             {
301                 bs_write_se( s, sh->weight[i][0].i_scale );
302                 bs_write_se( s, sh->weight[i][0].i_offset );
303             }
304             bs_write1( s, chroma_weight_l0_flag );
305             if( chroma_weight_l0_flag )
306             {
307                 for( int j = 1; j < 3; j++ )
308                 {
309                     bs_write_se( s, sh->weight[i][j].i_scale );
310                     bs_write_se( s, sh->weight[i][j].i_offset );
311                 }
312             }
313         }
314     }
315     else if( sh->pps->b_weighted_bipred == 1 && sh->i_type == SLICE_TYPE_B )
316     {
317       /* TODO */
318     }
319
320     if( i_nal_ref_idc != 0 )
321     {
322         if( sh->i_idr_pic_id >= 0 )
323         {
324             bs_write1( s, 0 );  /* no output of prior pics flag */
325             bs_write1( s, 0 );  /* long term reference flag */
326         }
327         else
328         {
329             bs_write1( s, sh->i_mmco_command_count > 0 ); /* adaptive_ref_pic_marking_mode_flag */
330             if( sh->i_mmco_command_count > 0 )
331             {
332                 for( int i = 0; i < sh->i_mmco_command_count; i++ )
333                 {
334                     bs_write_ue( s, 1 ); /* mark short term ref as unused */
335                     bs_write_ue( s, sh->mmco[i].i_difference_of_pic_nums - 1 );
336                 }
337                 bs_write_ue( s, 0 ); /* end command list */
338             }
339         }
340     }
341
342     if( sh->pps->b_cabac && sh->i_type != SLICE_TYPE_I )
343         bs_write_ue( s, sh->i_cabac_init_idc );
344
345     bs_write_se( s, sh->i_qp_delta );      /* slice qp delta */
346
347     if( sh->pps->b_deblocking_filter_control )
348     {
349         bs_write_ue( s, sh->i_disable_deblocking_filter_idc );
350         if( sh->i_disable_deblocking_filter_idc != 1 )
351         {
352             bs_write_se( s, sh->i_alpha_c0_offset >> 1 );
353             bs_write_se( s, sh->i_beta_offset >> 1 );
354         }
355     }
356 }
357
358 /* If we are within a reasonable distance of the end of the memory allocated for the bitstream, */
359 /* reallocate, adding an arbitrary amount of space. */
360 static int x264_bitstream_check_buffer( x264_t *h )
361 {
362     uint8_t *bs_bak = h->out.p_bitstream;
363     int max_row_size = (2500 << SLICE_MBAFF) * h->mb.i_mb_width;
364     if( (h->param.b_cabac && (h->cabac.p_end - h->cabac.p < max_row_size)) ||
365         (h->out.bs.p_end - h->out.bs.p < max_row_size) )
366     {
367         h->out.i_bitstream += max_row_size;
368         CHECKED_MALLOC( h->out.p_bitstream, h->out.i_bitstream );
369         h->mc.memcpy_aligned( h->out.p_bitstream, bs_bak, (h->out.i_bitstream - max_row_size) & ~15 );
370         intptr_t delta = h->out.p_bitstream - bs_bak;
371
372         h->out.bs.p_start += delta;
373         h->out.bs.p += delta;
374         h->out.bs.p_end = h->out.p_bitstream + h->out.i_bitstream;
375
376         h->cabac.p_start += delta;
377         h->cabac.p += delta;
378         h->cabac.p_end = h->out.p_bitstream + h->out.i_bitstream;
379
380         for( int i = 0; i <= h->out.i_nal; i++ )
381             h->out.nal[i].p_payload += delta;
382         x264_free( bs_bak );
383     }
384     return 0;
385 fail:
386     x264_free( bs_bak );
387     return -1;
388 }
389
390 #if HAVE_THREAD
391 static void x264_encoder_thread_init( x264_t *h )
392 {
393     if( h->param.i_sync_lookahead )
394         x264_lower_thread_priority( 10 );
395
396 #if HAVE_MMX
397     /* Misalign mask has to be set separately for each thread. */
398     if( h->param.cpu&X264_CPU_SSE_MISALIGN )
399         x264_cpu_mask_misalign_sse();
400 #endif
401 }
402
403 static void x264_lookahead_thread_init( x264_t *h )
404 {
405 #if HAVE_MMX
406     /* Misalign mask has to be set separately for each thread. */
407     if( h->param.cpu&X264_CPU_SSE_MISALIGN )
408         x264_cpu_mask_misalign_sse();
409 #endif
410 }
411 #endif
412
413 /****************************************************************************
414  *
415  ****************************************************************************
416  ****************************** External API*********************************
417  ****************************************************************************
418  *
419  ****************************************************************************/
420
421 static int x264_validate_parameters( x264_t *h, int b_open )
422 {
423 #if HAVE_MMX
424     if( b_open )
425     {
426         int cpuflags = x264_cpu_detect();
427         int fail = 0;
428 #ifdef __SSE__
429         if( !(cpuflags & X264_CPU_SSE) )
430         {
431             x264_log( h, X264_LOG_ERROR, "your cpu does not support SSE1, but x264 was compiled with asm\n");
432             fail = 1;
433         }
434 #else
435         if( !(cpuflags & X264_CPU_MMX2) )
436         {
437             x264_log( h, X264_LOG_ERROR, "your cpu does not support MMXEXT, but x264 was compiled with asm\n");
438             fail = 1;
439         }
440 #endif
441         if( !fail && !(cpuflags & X264_CPU_CMOV) )
442         {
443             x264_log( h, X264_LOG_ERROR, "your cpu does not support CMOV, but x264 was compiled with asm\n");
444             fail = 1;
445         }
446         if( fail )
447         {
448             x264_log( h, X264_LOG_ERROR, "to run x264, recompile without asm (configure --disable-asm)\n");
449             return -1;
450         }
451     }
452 #endif
453
454 #if HAVE_INTERLACED
455     h->param.b_interlaced = !!PARAM_INTERLACED;
456 #else
457     if( h->param.b_interlaced )
458     {
459         x264_log( h, X264_LOG_ERROR, "not compiled with interlaced support\n" );
460         return -1;
461     }
462 #endif
463
464     if( h->param.i_width <= 0 || h->param.i_height <= 0 )
465     {
466         x264_log( h, X264_LOG_ERROR, "invalid width x height (%dx%d)\n",
467                   h->param.i_width, h->param.i_height );
468         return -1;
469     }
470
471     int i_csp = h->param.i_csp & X264_CSP_MASK;
472 #if X264_CHROMA_FORMAT
473     if( CHROMA_FORMAT != CHROMA_420 && i_csp >= X264_CSP_I420 && i_csp <= X264_CSP_NV12 )
474     {
475         x264_log( h, X264_LOG_ERROR, "not compiled with 4:2:0 support\n" );
476         return -1;
477     }
478     else if( CHROMA_FORMAT != CHROMA_422 && i_csp >= X264_CSP_I422 && i_csp <= X264_CSP_NV16 )
479     {
480         x264_log( h, X264_LOG_ERROR, "not compiled with 4:2:2 support\n" );
481         return -1;
482     }
483     else if( CHROMA_FORMAT != CHROMA_444 && i_csp >= X264_CSP_I444 && i_csp <= X264_CSP_RGB )
484     {
485         x264_log( h, X264_LOG_ERROR, "not compiled with 4:4:4 support\n" );
486         return -1;
487     }
488 #endif
489     if( i_csp <= X264_CSP_NONE || i_csp >= X264_CSP_MAX )
490     {
491         x264_log( h, X264_LOG_ERROR, "invalid CSP (only I420/YV12/NV12/I422/YV16/NV16/I444/YV24/BGR/BGRA/RGB supported)\n" );
492         return -1;
493     }
494
495     if( i_csp < X264_CSP_I444 && h->param.i_width % 2 )
496     {
497         x264_log( h, X264_LOG_ERROR, "width not divisible by 2 (%dx%d)\n",
498                   h->param.i_width, h->param.i_height );
499         return -1;
500     }
501
502     if( i_csp < X264_CSP_I422 && PARAM_INTERLACED && h->param.i_height % 4 )
503     {
504         x264_log( h, X264_LOG_ERROR, "height not divisible by 4 (%dx%d)\n",
505                   h->param.i_width, h->param.i_height );
506         return -1;
507     }
508
509     if( (i_csp < X264_CSP_I422 || PARAM_INTERLACED) && h->param.i_height % 2 )
510     {
511         x264_log( h, X264_LOG_ERROR, "height not divisible by 2 (%dx%d)\n",
512                   h->param.i_width, h->param.i_height );
513         return -1;
514     }
515
516     if( (h->param.crop_rect.i_left + h->param.crop_rect.i_right ) >= h->param.i_width ||
517         (h->param.crop_rect.i_top  + h->param.crop_rect.i_bottom) >= h->param.i_height )
518     {
519         x264_log( h, X264_LOG_ERROR, "invalid crop-rect %u,%u,%u,%u\n", h->param.crop_rect.i_left,
520                   h->param.crop_rect.i_top, h->param.crop_rect.i_right,  h->param.crop_rect.i_bottom );
521         return -1;
522     }
523
524     if( h->param.i_threads == X264_THREADS_AUTO )
525         h->param.i_threads = x264_cpu_num_processors() * (h->param.b_sliced_threads?2:3)/2;
526     int max_sliced_threads = X264_MAX( 1, (h->param.i_height+15)/16 / 4 );
527     if( h->param.i_threads > 1 )
528     {
529 #if !HAVE_THREAD
530         x264_log( h, X264_LOG_WARNING, "not compiled with thread support!\n");
531         h->param.i_threads = 1;
532 #endif
533         /* Avoid absurdly small thread slices as they can reduce performance
534          * and VBV compliance.  Capped at an arbitrary 4 rows per thread. */
535         if( h->param.b_sliced_threads )
536             h->param.i_threads = X264_MIN( h->param.i_threads, max_sliced_threads );
537     }
538     h->param.i_threads = x264_clip3( h->param.i_threads, 1, X264_THREAD_MAX );
539     if( h->param.i_threads == 1 )
540     {
541         h->param.b_sliced_threads = 0;
542         h->param.i_lookahead_threads = 1;
543     }
544     h->i_thread_frames = h->param.b_sliced_threads ? 1 : h->param.i_threads;
545     if( h->i_thread_frames > 1 )
546         h->param.nalu_process = NULL;
547
548     if( h->param.b_opencl )
549     {
550 #if !HAVE_OPENCL
551         x264_log( h, X264_LOG_WARNING, "OpenCL: not compiled with OpenCL support, disabling\n" );
552         h->param.b_opencl = 0;
553 #elif BIT_DEPTH > 8
554         x264_log( h, X264_LOG_WARNING, "OpenCL lookahead does not support high bit depth, disabling opencl\n" );
555         h->param.b_opencl = 0;
556 #else
557         if( h->param.i_width < 32 || h->param.i_height < 32 )
558         {
559             x264_log( h, X264_LOG_WARNING, "OpenCL: frame size is too small, disabling opencl\n" );
560             h->param.b_opencl = 0;
561         }
562 #endif
563         if( h->param.opencl_device_id && h->param.i_opencl_device )
564         {
565             x264_log( h, X264_LOG_WARNING, "OpenCL: device id and device skip count configured; dropping skip\n" );
566             h->param.i_opencl_device = 0;
567         }
568     }
569
570     h->param.i_keyint_max = x264_clip3( h->param.i_keyint_max, 1, X264_KEYINT_MAX_INFINITE );
571     if( h->param.i_keyint_max == 1 )
572     {
573         h->param.b_intra_refresh = 0;
574         h->param.analyse.i_weighted_pred = 0;
575     }
576
577     h->param.i_frame_packing = x264_clip3( h->param.i_frame_packing, -1, 5 );
578
579     /* Detect default ffmpeg settings and terminate with an error. */
580     if( b_open )
581     {
582         int score = 0;
583         score += h->param.analyse.i_me_range == 0;
584         score += h->param.rc.i_qp_step == 3;
585         score += h->param.i_keyint_max == 12;
586         score += h->param.rc.i_qp_min == 2;
587         score += h->param.rc.i_qp_max == 31;
588         score += h->param.rc.f_qcompress == 0.5;
589         score += fabs(h->param.rc.f_ip_factor - 1.25) < 0.01;
590         score += fabs(h->param.rc.f_pb_factor - 1.25) < 0.01;
591         score += h->param.analyse.inter == 0 && h->param.analyse.i_subpel_refine == 8;
592         if( score >= 5 )
593         {
594             x264_log( h, X264_LOG_ERROR, "broken ffmpeg default settings detected\n" );
595             x264_log( h, X264_LOG_ERROR, "use an encoding preset (e.g. -vpre medium)\n" );
596             x264_log( h, X264_LOG_ERROR, "preset usage: -vpre <speed> -vpre <profile>\n" );
597             x264_log( h, X264_LOG_ERROR, "speed presets are listed in x264 --help\n" );
598             x264_log( h, X264_LOG_ERROR, "profile is optional; x264 defaults to high\n" );
599             return -1;
600         }
601     }
602
603     if( h->param.rc.i_rc_method < 0 || h->param.rc.i_rc_method > 2 )
604     {
605         x264_log( h, X264_LOG_ERROR, "no ratecontrol method specified\n" );
606         return -1;
607     }
608     h->param.rc.f_rf_constant = x264_clip3f( h->param.rc.f_rf_constant, -QP_BD_OFFSET, 51 );
609     h->param.rc.f_rf_constant_max = x264_clip3f( h->param.rc.f_rf_constant_max, -QP_BD_OFFSET, 51 );
610     h->param.rc.i_qp_constant = x264_clip3( h->param.rc.i_qp_constant, 0, QP_MAX );
611     h->param.analyse.i_subpel_refine = x264_clip3( h->param.analyse.i_subpel_refine, 0, 11 );
612     h->param.rc.f_ip_factor = X264_MAX( h->param.rc.f_ip_factor, 0.01f );
613     h->param.rc.f_pb_factor = X264_MAX( h->param.rc.f_pb_factor, 0.01f );
614     if( h->param.rc.i_rc_method == X264_RC_CRF )
615     {
616         h->param.rc.i_qp_constant = h->param.rc.f_rf_constant + QP_BD_OFFSET;
617         h->param.rc.i_bitrate = 0;
618     }
619     if( b_open && (h->param.rc.i_rc_method == X264_RC_CQP || h->param.rc.i_rc_method == X264_RC_CRF)
620         && h->param.rc.i_qp_constant == 0 )
621     {
622         h->mb.b_lossless = 1;
623         h->param.i_cqm_preset = X264_CQM_FLAT;
624         h->param.psz_cqm_file = NULL;
625         h->param.rc.i_rc_method = X264_RC_CQP;
626         h->param.rc.f_ip_factor = 1;
627         h->param.rc.f_pb_factor = 1;
628         h->param.analyse.b_psnr = 0;
629         h->param.analyse.b_ssim = 0;
630         h->param.analyse.i_chroma_qp_offset = 0;
631         h->param.analyse.i_trellis = 0;
632         h->param.analyse.b_fast_pskip = 0;
633         h->param.analyse.i_noise_reduction = 0;
634         h->param.analyse.b_psy = 0;
635         h->param.i_bframe = 0;
636         /* 8x8dct is not useful without RD in CAVLC lossless */
637         if( !h->param.b_cabac && h->param.analyse.i_subpel_refine < 6 )
638             h->param.analyse.b_transform_8x8 = 0;
639     }
640     if( h->param.rc.i_rc_method == X264_RC_CQP )
641     {
642         float qp_p = h->param.rc.i_qp_constant;
643         float qp_i = qp_p - 6*log2f( h->param.rc.f_ip_factor );
644         float qp_b = qp_p + 6*log2f( h->param.rc.f_pb_factor );
645         h->param.rc.i_qp_min = x264_clip3( (int)(X264_MIN3( qp_p, qp_i, qp_b )), 0, QP_MAX );
646         h->param.rc.i_qp_max = x264_clip3( (int)(X264_MAX3( qp_p, qp_i, qp_b ) + .999), 0, QP_MAX );
647         h->param.rc.i_aq_mode = 0;
648         h->param.rc.b_mb_tree = 0;
649         h->param.rc.i_bitrate = 0;
650     }
651     h->param.rc.i_qp_max = x264_clip3( h->param.rc.i_qp_max, 0, QP_MAX );
652     h->param.rc.i_qp_min = x264_clip3( h->param.rc.i_qp_min, 0, h->param.rc.i_qp_max );
653     h->param.rc.i_qp_step = x264_clip3( h->param.rc.i_qp_step, 2, QP_MAX );
654     h->param.rc.i_bitrate = x264_clip3( h->param.rc.i_bitrate, 0, 2000000 );
655     if( h->param.rc.i_rc_method == X264_RC_ABR && !h->param.rc.i_bitrate )
656     {
657         x264_log( h, X264_LOG_ERROR, "bitrate not specified\n" );
658         return -1;
659     }
660     h->param.rc.i_vbv_buffer_size = x264_clip3( h->param.rc.i_vbv_buffer_size, 0, 2000000 );
661     h->param.rc.i_vbv_max_bitrate = x264_clip3( h->param.rc.i_vbv_max_bitrate, 0, 2000000 );
662     h->param.rc.f_vbv_buffer_init = x264_clip3f( h->param.rc.f_vbv_buffer_init, 0, 2000000 );
663     if( h->param.rc.i_vbv_buffer_size )
664     {
665         if( h->param.rc.i_rc_method == X264_RC_CQP )
666         {
667             x264_log( h, X264_LOG_WARNING, "VBV is incompatible with constant QP, ignored.\n" );
668             h->param.rc.i_vbv_max_bitrate = 0;
669             h->param.rc.i_vbv_buffer_size = 0;
670         }
671         else if( h->param.rc.i_vbv_max_bitrate == 0 )
672         {
673             if( h->param.rc.i_rc_method == X264_RC_ABR )
674             {
675                 x264_log( h, X264_LOG_WARNING, "VBV maxrate unspecified, assuming CBR\n" );
676                 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate;
677             }
678             else
679             {
680                 x264_log( h, X264_LOG_WARNING, "VBV bufsize set but maxrate unspecified, ignored\n" );
681                 h->param.rc.i_vbv_buffer_size = 0;
682             }
683         }
684         else if( h->param.rc.i_vbv_max_bitrate < h->param.rc.i_bitrate &&
685                  h->param.rc.i_rc_method == X264_RC_ABR )
686         {
687             x264_log( h, X264_LOG_WARNING, "max bitrate less than average bitrate, assuming CBR\n" );
688             h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate;
689         }
690     }
691     else if( h->param.rc.i_vbv_max_bitrate )
692     {
693         x264_log( h, X264_LOG_WARNING, "VBV maxrate specified, but no bufsize, ignored\n" );
694         h->param.rc.i_vbv_max_bitrate = 0;
695     }
696
697     h->param.i_slice_max_size = X264_MAX( h->param.i_slice_max_size, 0 );
698     h->param.i_slice_max_mbs = X264_MAX( h->param.i_slice_max_mbs, 0 );
699     h->param.i_slice_min_mbs = X264_MAX( h->param.i_slice_min_mbs, 0 );
700     if( h->param.i_slice_max_mbs )
701         h->param.i_slice_min_mbs = X264_MIN( h->param.i_slice_min_mbs, h->param.i_slice_max_mbs/2 );
702     else if( !h->param.i_slice_max_size )
703         h->param.i_slice_min_mbs = 0;
704     if( PARAM_INTERLACED && h->param.i_slice_min_mbs )
705     {
706         x264_log( h, X264_LOG_WARNING, "interlace + slice-min-mbs is not implemented\n" );
707         h->param.i_slice_min_mbs = 0;
708     }
709     int mb_width = (h->param.i_width+15)/16;
710     if( h->param.i_slice_min_mbs > mb_width )
711     {
712         x264_log( h, X264_LOG_WARNING, "slice-min-mbs > row mb size (%d) not implemented\n", mb_width );
713         h->param.i_slice_min_mbs = mb_width;
714     }
715
716     int max_slices = (h->param.i_height+((16<<PARAM_INTERLACED)-1))/(16<<PARAM_INTERLACED);
717     if( h->param.b_sliced_threads )
718         h->param.i_slice_count = x264_clip3( h->param.i_threads, 0, max_slices );
719     else
720     {
721         h->param.i_slice_count = x264_clip3( h->param.i_slice_count, 0, max_slices );
722         if( h->param.i_slice_max_mbs || h->param.i_slice_max_size )
723             h->param.i_slice_count = 0;
724     }
725     if( h->param.i_slice_count_max > 0 )
726         h->param.i_slice_count_max = X264_MAX( h->param.i_slice_count, h->param.i_slice_count_max );
727
728     if( h->param.b_bluray_compat )
729     {
730         h->param.i_bframe_pyramid = X264_MIN( X264_B_PYRAMID_STRICT, h->param.i_bframe_pyramid );
731         h->param.i_bframe = X264_MIN( h->param.i_bframe, 3 );
732         h->param.b_aud = 1;
733         h->param.i_nal_hrd = X264_MAX( h->param.i_nal_hrd, X264_NAL_HRD_VBR );
734         h->param.i_slice_max_size = 0;
735         h->param.i_slice_max_mbs = 0;
736         h->param.b_intra_refresh = 0;
737         h->param.i_frame_reference = X264_MIN( h->param.i_frame_reference, 6 );
738         h->param.i_dpb_size = X264_MIN( h->param.i_dpb_size, 6 );
739         /* Don't use I-frames, because Blu-ray treats them the same as IDR. */
740         h->param.i_keyint_min = 1;
741         /* Due to the proliferation of broken players that don't handle dupes properly. */
742         h->param.analyse.i_weighted_pred = X264_MIN( h->param.analyse.i_weighted_pred, X264_WEIGHTP_SIMPLE );
743         if( h->param.b_fake_interlaced )
744             h->param.b_pic_struct = 1;
745     }
746
747     h->param.i_frame_reference = x264_clip3( h->param.i_frame_reference, 1, X264_REF_MAX );
748     h->param.i_dpb_size = x264_clip3( h->param.i_dpb_size, 1, X264_REF_MAX );
749     if( h->param.i_scenecut_threshold < 0 )
750         h->param.i_scenecut_threshold = 0;
751     h->param.analyse.i_direct_mv_pred = x264_clip3( h->param.analyse.i_direct_mv_pred, X264_DIRECT_PRED_NONE, X264_DIRECT_PRED_AUTO );
752     if( !h->param.analyse.i_subpel_refine && h->param.analyse.i_direct_mv_pred > X264_DIRECT_PRED_SPATIAL )
753     {
754         x264_log( h, X264_LOG_WARNING, "subme=0 + direct=temporal is not supported\n" );
755         h->param.analyse.i_direct_mv_pred = X264_DIRECT_PRED_SPATIAL;
756     }
757     h->param.i_bframe = x264_clip3( h->param.i_bframe, 0, X264_MIN( X264_BFRAME_MAX, h->param.i_keyint_max-1 ) );
758     h->param.i_bframe_bias = x264_clip3( h->param.i_bframe_bias, -90, 100 );
759     if( h->param.i_bframe <= 1 )
760         h->param.i_bframe_pyramid = X264_B_PYRAMID_NONE;
761     h->param.i_bframe_pyramid = x264_clip3( h->param.i_bframe_pyramid, X264_B_PYRAMID_NONE, X264_B_PYRAMID_NORMAL );
762     h->param.i_bframe_adaptive = x264_clip3( h->param.i_bframe_adaptive, X264_B_ADAPT_NONE, X264_B_ADAPT_TRELLIS );
763     if( !h->param.i_bframe )
764     {
765         h->param.i_bframe_adaptive = X264_B_ADAPT_NONE;
766         h->param.analyse.i_direct_mv_pred = 0;
767         h->param.analyse.b_weighted_bipred = 0;
768         h->param.b_open_gop = 0;
769     }
770     if( h->param.b_intra_refresh && h->param.i_bframe_pyramid == X264_B_PYRAMID_NORMAL )
771     {
772         x264_log( h, X264_LOG_WARNING, "b-pyramid normal + intra-refresh is not supported\n" );
773         h->param.i_bframe_pyramid = X264_B_PYRAMID_STRICT;
774     }
775     if( h->param.b_intra_refresh && (h->param.i_frame_reference > 1 || h->param.i_dpb_size > 1) )
776     {
777         x264_log( h, X264_LOG_WARNING, "ref > 1 + intra-refresh is not supported\n" );
778         h->param.i_frame_reference = 1;
779         h->param.i_dpb_size = 1;
780     }
781     if( h->param.b_intra_refresh && h->param.b_open_gop )
782     {
783         x264_log( h, X264_LOG_WARNING, "intra-refresh is not compatible with open-gop\n" );
784         h->param.b_open_gop = 0;
785     }
786     if( !h->param.i_fps_num || !h->param.i_fps_den )
787     {
788         h->param.i_fps_num = 25;
789         h->param.i_fps_den = 1;
790     }
791     float fps = (float) h->param.i_fps_num / h->param.i_fps_den;
792     if( h->param.i_keyint_min == X264_KEYINT_MIN_AUTO )
793         h->param.i_keyint_min = X264_MIN( h->param.i_keyint_max / 10, fps );
794     h->param.i_keyint_min = x264_clip3( h->param.i_keyint_min, 1, h->param.i_keyint_max/2+1 );
795     h->param.rc.i_lookahead = x264_clip3( h->param.rc.i_lookahead, 0, X264_LOOKAHEAD_MAX );
796     {
797         int maxrate = X264_MAX( h->param.rc.i_vbv_max_bitrate, h->param.rc.i_bitrate );
798         float bufsize = maxrate ? (float)h->param.rc.i_vbv_buffer_size / maxrate : 0;
799         h->param.rc.i_lookahead = X264_MIN( h->param.rc.i_lookahead, X264_MAX( h->param.i_keyint_max, bufsize*fps ) );
800     }
801
802     if( !h->param.i_timebase_num || !h->param.i_timebase_den || !(h->param.b_vfr_input || h->param.b_pulldown) )
803     {
804         h->param.i_timebase_num = h->param.i_fps_den;
805         h->param.i_timebase_den = h->param.i_fps_num;
806     }
807
808     h->param.rc.f_qcompress = x264_clip3f( h->param.rc.f_qcompress, 0.0, 1.0 );
809     if( h->param.i_keyint_max == 1 || h->param.rc.f_qcompress == 1 )
810         h->param.rc.b_mb_tree = 0;
811     if( (!h->param.b_intra_refresh && h->param.i_keyint_max != X264_KEYINT_MAX_INFINITE) &&
812         !h->param.rc.i_lookahead && h->param.rc.b_mb_tree )
813     {
814         x264_log( h, X264_LOG_WARNING, "lookaheadless mb-tree requires intra refresh or infinite keyint\n" );
815         h->param.rc.b_mb_tree = 0;
816     }
817     if( b_open && h->param.rc.b_stat_read )
818         h->param.rc.i_lookahead = 0;
819 #if HAVE_THREAD
820     if( h->param.i_sync_lookahead < 0 )
821         h->param.i_sync_lookahead = h->param.i_bframe + 1;
822     h->param.i_sync_lookahead = X264_MIN( h->param.i_sync_lookahead, X264_LOOKAHEAD_MAX );
823     if( h->param.rc.b_stat_read || h->i_thread_frames == 1 )
824         h->param.i_sync_lookahead = 0;
825 #else
826     h->param.i_sync_lookahead = 0;
827 #endif
828
829     h->param.i_deblocking_filter_alphac0 = x264_clip3( h->param.i_deblocking_filter_alphac0, -6, 6 );
830     h->param.i_deblocking_filter_beta    = x264_clip3( h->param.i_deblocking_filter_beta, -6, 6 );
831     h->param.analyse.i_luma_deadzone[0] = x264_clip3( h->param.analyse.i_luma_deadzone[0], 0, 32 );
832     h->param.analyse.i_luma_deadzone[1] = x264_clip3( h->param.analyse.i_luma_deadzone[1], 0, 32 );
833
834     h->param.i_cabac_init_idc = x264_clip3( h->param.i_cabac_init_idc, 0, 2 );
835
836     if( h->param.i_cqm_preset < X264_CQM_FLAT || h->param.i_cqm_preset > X264_CQM_CUSTOM )
837         h->param.i_cqm_preset = X264_CQM_FLAT;
838
839     if( h->param.analyse.i_me_method < X264_ME_DIA ||
840         h->param.analyse.i_me_method > X264_ME_TESA )
841         h->param.analyse.i_me_method = X264_ME_HEX;
842     h->param.analyse.i_me_range = x264_clip3( h->param.analyse.i_me_range, 4, 1024 );
843     if( h->param.analyse.i_me_range > 16 && h->param.analyse.i_me_method <= X264_ME_HEX )
844         h->param.analyse.i_me_range = 16;
845     if( h->param.analyse.i_me_method == X264_ME_TESA &&
846         (h->mb.b_lossless || h->param.analyse.i_subpel_refine <= 1) )
847         h->param.analyse.i_me_method = X264_ME_ESA;
848     h->param.analyse.b_mixed_references = h->param.analyse.b_mixed_references && h->param.i_frame_reference > 1;
849     h->param.analyse.inter &= X264_ANALYSE_PSUB16x16|X264_ANALYSE_PSUB8x8|X264_ANALYSE_BSUB16x16|
850                               X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
851     h->param.analyse.intra &= X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
852     if( !(h->param.analyse.inter & X264_ANALYSE_PSUB16x16) )
853         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
854     if( !h->param.analyse.b_transform_8x8 )
855     {
856         h->param.analyse.inter &= ~X264_ANALYSE_I8x8;
857         h->param.analyse.intra &= ~X264_ANALYSE_I8x8;
858     }
859     h->param.analyse.i_trellis = x264_clip3( h->param.analyse.i_trellis, 0, 2 );
860     h->param.rc.i_aq_mode = x264_clip3( h->param.rc.i_aq_mode, 0, 2 );
861     h->param.rc.f_aq_strength = x264_clip3f( h->param.rc.f_aq_strength, 0, 3 );
862     if( h->param.rc.f_aq_strength == 0 )
863         h->param.rc.i_aq_mode = 0;
864
865     if( h->param.i_log_level < X264_LOG_INFO )
866     {
867         h->param.analyse.b_psnr = 0;
868         h->param.analyse.b_ssim = 0;
869     }
870     /* Warn users trying to measure PSNR/SSIM with psy opts on. */
871     if( b_open && (h->param.analyse.b_psnr || h->param.analyse.b_ssim) )
872     {
873         char *s = NULL;
874
875         if( h->param.analyse.b_psy )
876         {
877             s = h->param.analyse.b_psnr ? "psnr" : "ssim";
878             x264_log( h, X264_LOG_WARNING, "--%s used with psy on: results will be invalid!\n", s );
879         }
880         else if( !h->param.rc.i_aq_mode && h->param.analyse.b_ssim )
881         {
882             x264_log( h, X264_LOG_WARNING, "--ssim used with AQ off: results will be invalid!\n" );
883             s = "ssim";
884         }
885         else if(  h->param.rc.i_aq_mode && h->param.analyse.b_psnr )
886         {
887             x264_log( h, X264_LOG_WARNING, "--psnr used with AQ on: results will be invalid!\n" );
888             s = "psnr";
889         }
890         if( s )
891             x264_log( h, X264_LOG_WARNING, "--tune %s should be used if attempting to benchmark %s!\n", s, s );
892     }
893
894     if( !h->param.analyse.b_psy )
895     {
896         h->param.analyse.f_psy_rd = 0;
897         h->param.analyse.f_psy_trellis = 0;
898     }
899     h->param.analyse.f_psy_rd = x264_clip3f( h->param.analyse.f_psy_rd, 0, 10 );
900     h->param.analyse.f_psy_trellis = x264_clip3f( h->param.analyse.f_psy_trellis, 0, 10 );
901     h->mb.i_psy_rd = h->param.analyse.i_subpel_refine >= 6 ? FIX8( h->param.analyse.f_psy_rd ) : 0;
902     h->mb.i_psy_trellis = h->param.analyse.i_trellis ? FIX8( h->param.analyse.f_psy_trellis / 4 ) : 0;
903     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -32, 32);
904     /* In 4:4:4 mode, chroma gets twice as much resolution, so we can halve its quality. */
905     if( b_open && i_csp >= X264_CSP_I444 && i_csp < X264_CSP_BGR && h->param.analyse.b_psy )
906         h->param.analyse.i_chroma_qp_offset += 6;
907     /* Psy RDO increases overall quantizers to improve the quality of luma--this indirectly hurts chroma quality */
908     /* so we lower the chroma QP offset to compensate */
909     if( b_open && h->mb.i_psy_rd )
910         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_rd < 0.25 ? 1 : 2;
911     /* Psy trellis has a similar effect. */
912     if( b_open && h->mb.i_psy_trellis )
913         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_trellis < 0.25 ? 1 : 2;
914     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -12, 12);
915     /* MB-tree requires AQ to be on, even if the strength is zero. */
916     if( !h->param.rc.i_aq_mode && h->param.rc.b_mb_tree )
917     {
918         h->param.rc.i_aq_mode = 1;
919         h->param.rc.f_aq_strength = 0;
920     }
921     h->param.analyse.i_noise_reduction = x264_clip3( h->param.analyse.i_noise_reduction, 0, 1<<16 );
922     if( h->param.analyse.i_subpel_refine >= 10 && (h->param.analyse.i_trellis != 2 || !h->param.rc.i_aq_mode) )
923         h->param.analyse.i_subpel_refine = 9;
924
925     {
926         const x264_level_t *l = x264_levels;
927         if( h->param.i_level_idc < 0 )
928         {
929             int maxrate_bak = h->param.rc.i_vbv_max_bitrate;
930             if( h->param.rc.i_rc_method == X264_RC_ABR && h->param.rc.i_vbv_buffer_size <= 0 )
931                 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate * 2;
932             x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
933             do h->param.i_level_idc = l->level_idc;
934                 while( l[1].level_idc && x264_validate_levels( h, 0 ) && l++ );
935             h->param.rc.i_vbv_max_bitrate = maxrate_bak;
936         }
937         else
938         {
939             while( l->level_idc && l->level_idc != h->param.i_level_idc )
940                 l++;
941             if( l->level_idc == 0 )
942             {
943                 x264_log( h, X264_LOG_ERROR, "invalid level_idc: %d\n", h->param.i_level_idc );
944                 return -1;
945             }
946         }
947         if( h->param.analyse.i_mv_range <= 0 )
948             h->param.analyse.i_mv_range = l->mv_range >> PARAM_INTERLACED;
949         else
950             h->param.analyse.i_mv_range = x264_clip3(h->param.analyse.i_mv_range, 32, 512 >> PARAM_INTERLACED);
951     }
952
953     h->param.analyse.i_weighted_pred = x264_clip3( h->param.analyse.i_weighted_pred, X264_WEIGHTP_NONE, X264_WEIGHTP_SMART );
954
955     if( h->param.i_lookahead_threads == X264_THREADS_AUTO )
956     {
957         if( h->param.b_sliced_threads )
958             h->param.i_lookahead_threads = h->param.i_threads;
959         else
960         {
961             /* If we're using much slower lookahead settings than encoding settings, it helps a lot to use
962              * more lookahead threads.  This typically happens in the first pass of a two-pass encode, so
963              * try to guess at this sort of case.
964              *
965              * Tuned by a little bit of real encoding with the various presets. */
966             int badapt = h->param.i_bframe_adaptive == X264_B_ADAPT_TRELLIS;
967             int subme = X264_MIN( h->param.analyse.i_subpel_refine / 3, 3 ) + (h->param.analyse.i_subpel_refine > 1);
968             int bframes = X264_MIN( (h->param.i_bframe - 1) / 3, 3 );
969
970             /* [b-adapt 0/1 vs 2][quantized subme][quantized bframes] */
971             static const uint8_t lookahead_thread_div[2][5][4] =
972             {{{6,6,6,6}, {3,3,3,3}, {4,4,4,4}, {6,6,6,6}, {12,12,12,12}},
973              {{3,2,1,1}, {2,1,1,1}, {4,3,2,1}, {6,4,3,2}, {12, 9, 6, 4}}};
974
975             h->param.i_lookahead_threads = h->param.i_threads / lookahead_thread_div[badapt][subme][bframes];
976             /* Since too many lookahead threads significantly degrades lookahead accuracy, limit auto
977              * lookahead threads to about 8 macroblock rows high each at worst.  This number is chosen
978              * pretty much arbitrarily. */
979             h->param.i_lookahead_threads = X264_MIN( h->param.i_lookahead_threads, h->param.i_height / 128 );
980         }
981     }
982     h->param.i_lookahead_threads = x264_clip3( h->param.i_lookahead_threads, 1, X264_MIN( max_sliced_threads, X264_LOOKAHEAD_THREAD_MAX ) );
983
984     if( PARAM_INTERLACED )
985     {
986         if( h->param.analyse.i_me_method >= X264_ME_ESA )
987         {
988             x264_log( h, X264_LOG_WARNING, "interlace + me=esa is not implemented\n" );
989             h->param.analyse.i_me_method = X264_ME_UMH;
990         }
991         if( h->param.analyse.i_weighted_pred > 0 )
992         {
993             x264_log( h, X264_LOG_WARNING, "interlace + weightp is not implemented\n" );
994             h->param.analyse.i_weighted_pred = X264_WEIGHTP_NONE;
995         }
996     }
997
998     if( !h->param.analyse.i_weighted_pred && h->param.rc.b_mb_tree && h->param.analyse.b_psy )
999         h->param.analyse.i_weighted_pred = X264_WEIGHTP_FAKE;
1000
1001     if( h->i_thread_frames > 1 )
1002     {
1003         int r = h->param.analyse.i_mv_range_thread;
1004         int r2;
1005         if( r <= 0 )
1006         {
1007             // half of the available space is reserved and divided evenly among the threads,
1008             // the rest is allocated to whichever thread is far enough ahead to use it.
1009             // reserving more space increases quality for some videos, but costs more time
1010             // in thread synchronization.
1011             int max_range = (h->param.i_height + X264_THREAD_HEIGHT) / h->i_thread_frames - X264_THREAD_HEIGHT;
1012             r = max_range / 2;
1013         }
1014         r = X264_MAX( r, h->param.analyse.i_me_range );
1015         r = X264_MIN( r, h->param.analyse.i_mv_range );
1016         // round up to use the whole mb row
1017         r2 = (r & ~15) + ((-X264_THREAD_HEIGHT) & 15);
1018         if( r2 < r )
1019             r2 += 16;
1020         x264_log( h, X264_LOG_DEBUG, "using mv_range_thread = %d\n", r2 );
1021         h->param.analyse.i_mv_range_thread = r2;
1022     }
1023
1024     if( h->param.rc.f_rate_tolerance < 0 )
1025         h->param.rc.f_rate_tolerance = 0;
1026     if( h->param.rc.f_qblur < 0 )
1027         h->param.rc.f_qblur = 0;
1028     if( h->param.rc.f_complexity_blur < 0 )
1029         h->param.rc.f_complexity_blur = 0;
1030
1031     h->param.i_sps_id &= 31;
1032
1033     if( PARAM_INTERLACED )
1034         h->param.b_pic_struct = 1;
1035
1036     h->param.i_nal_hrd = x264_clip3( h->param.i_nal_hrd, X264_NAL_HRD_NONE, X264_NAL_HRD_CBR );
1037
1038     if( h->param.i_nal_hrd && !h->param.rc.i_vbv_buffer_size )
1039     {
1040         x264_log( h, X264_LOG_WARNING, "NAL HRD parameters require VBV parameters\n" );
1041         h->param.i_nal_hrd = X264_NAL_HRD_NONE;
1042     }
1043
1044     if( h->param.i_nal_hrd == X264_NAL_HRD_CBR &&
1045        (h->param.rc.i_bitrate != h->param.rc.i_vbv_max_bitrate || !h->param.rc.i_vbv_max_bitrate) )
1046     {
1047         x264_log( h, X264_LOG_WARNING, "CBR HRD requires constant bitrate\n" );
1048         h->param.i_nal_hrd = X264_NAL_HRD_VBR;
1049     }
1050
1051     /* ensure the booleans are 0 or 1 so they can be used in math */
1052 #define BOOLIFY(x) h->param.x = !!h->param.x
1053     BOOLIFY( b_cabac );
1054     BOOLIFY( b_constrained_intra );
1055     BOOLIFY( b_deblocking_filter );
1056     BOOLIFY( b_deterministic );
1057     BOOLIFY( b_sliced_threads );
1058     BOOLIFY( b_interlaced );
1059     BOOLIFY( b_intra_refresh );
1060     BOOLIFY( b_visualize );
1061     BOOLIFY( b_aud );
1062     BOOLIFY( b_repeat_headers );
1063     BOOLIFY( b_annexb );
1064     BOOLIFY( b_vfr_input );
1065     BOOLIFY( b_pulldown );
1066     BOOLIFY( b_tff );
1067     BOOLIFY( b_pic_struct );
1068     BOOLIFY( b_fake_interlaced );
1069     BOOLIFY( b_open_gop );
1070     BOOLIFY( b_bluray_compat );
1071     BOOLIFY( b_full_recon );
1072     BOOLIFY( b_opencl );
1073     BOOLIFY( analyse.b_transform_8x8 );
1074     BOOLIFY( analyse.b_weighted_bipred );
1075     BOOLIFY( analyse.b_chroma_me );
1076     BOOLIFY( analyse.b_mixed_references );
1077     BOOLIFY( analyse.b_fast_pskip );
1078     BOOLIFY( analyse.b_dct_decimate );
1079     BOOLIFY( analyse.b_psy );
1080     BOOLIFY( analyse.b_psnr );
1081     BOOLIFY( analyse.b_ssim );
1082     BOOLIFY( rc.b_stat_write );
1083     BOOLIFY( rc.b_stat_read );
1084     BOOLIFY( rc.b_mb_tree );
1085 #undef BOOLIFY
1086
1087     return 0;
1088 }
1089
1090 static void mbcmp_init( x264_t *h )
1091 {
1092     int satd = !h->mb.b_lossless && h->param.analyse.i_subpel_refine > 1;
1093     memcpy( h->pixf.mbcmp, satd ? h->pixf.satd : h->pixf.sad_aligned, sizeof(h->pixf.mbcmp) );
1094     memcpy( h->pixf.mbcmp_unaligned, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.mbcmp_unaligned) );
1095     h->pixf.intra_mbcmp_x3_16x16 = satd ? h->pixf.intra_satd_x3_16x16 : h->pixf.intra_sad_x3_16x16;
1096     h->pixf.intra_mbcmp_x3_8x16c = satd ? h->pixf.intra_satd_x3_8x16c : h->pixf.intra_sad_x3_8x16c;
1097     h->pixf.intra_mbcmp_x3_8x8c  = satd ? h->pixf.intra_satd_x3_8x8c  : h->pixf.intra_sad_x3_8x8c;
1098     h->pixf.intra_mbcmp_x3_8x8 = satd ? h->pixf.intra_sa8d_x3_8x8 : h->pixf.intra_sad_x3_8x8;
1099     h->pixf.intra_mbcmp_x3_4x4 = satd ? h->pixf.intra_satd_x3_4x4 : h->pixf.intra_sad_x3_4x4;
1100     h->pixf.intra_mbcmp_x9_4x4 = h->param.b_cpu_independent || h->mb.b_lossless ? NULL
1101                                : satd ? h->pixf.intra_satd_x9_4x4 : h->pixf.intra_sad_x9_4x4;
1102     h->pixf.intra_mbcmp_x9_8x8 = h->param.b_cpu_independent || h->mb.b_lossless ? NULL
1103                                : satd ? h->pixf.intra_sa8d_x9_8x8 : h->pixf.intra_sad_x9_8x8;
1104     satd &= h->param.analyse.i_me_method == X264_ME_TESA;
1105     memcpy( h->pixf.fpelcmp, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.fpelcmp) );
1106     memcpy( h->pixf.fpelcmp_x3, satd ? h->pixf.satd_x3 : h->pixf.sad_x3, sizeof(h->pixf.fpelcmp_x3) );
1107     memcpy( h->pixf.fpelcmp_x4, satd ? h->pixf.satd_x4 : h->pixf.sad_x4, sizeof(h->pixf.fpelcmp_x4) );
1108 }
1109
1110 static void chroma_dsp_init( x264_t *h )
1111 {
1112     memcpy( h->luma2chroma_pixel, x264_luma2chroma_pixel[CHROMA_FORMAT], sizeof(h->luma2chroma_pixel) );
1113
1114     switch( CHROMA_FORMAT )
1115     {
1116         case CHROMA_420:
1117             memcpy( h->predict_chroma, h->predict_8x8c, sizeof(h->predict_chroma) );
1118             h->mc.prefetch_fenc = h->mc.prefetch_fenc_420;
1119             h->loopf.deblock_chroma[0] = h->loopf.deblock_h_chroma_420;
1120             h->loopf.deblock_chroma_intra[0] = h->loopf.deblock_h_chroma_420_intra;
1121             h->loopf.deblock_chroma_mbaff = h->loopf.deblock_chroma_420_mbaff;
1122             h->loopf.deblock_chroma_intra_mbaff = h->loopf.deblock_chroma_420_intra_mbaff;
1123             h->pixf.intra_mbcmp_x3_chroma = h->pixf.intra_mbcmp_x3_8x8c;
1124             h->quantf.coeff_last[DCT_CHROMA_DC] = h->quantf.coeff_last4;
1125             h->quantf.coeff_level_run[DCT_CHROMA_DC] = h->quantf.coeff_level_run4;
1126             break;
1127         case CHROMA_422:
1128             memcpy( h->predict_chroma, h->predict_8x16c, sizeof(h->predict_chroma) );
1129             h->mc.prefetch_fenc = h->mc.prefetch_fenc_422;
1130             h->loopf.deblock_chroma[0] = h->loopf.deblock_h_chroma_422;
1131             h->loopf.deblock_chroma_intra[0] = h->loopf.deblock_h_chroma_422_intra;
1132             h->loopf.deblock_chroma_mbaff = h->loopf.deblock_chroma_422_mbaff;
1133             h->loopf.deblock_chroma_intra_mbaff = h->loopf.deblock_chroma_422_intra_mbaff;
1134             h->pixf.intra_mbcmp_x3_chroma = h->pixf.intra_mbcmp_x3_8x16c;
1135             h->quantf.coeff_last[DCT_CHROMA_DC] = h->quantf.coeff_last8;
1136             h->quantf.coeff_level_run[DCT_CHROMA_DC] = h->quantf.coeff_level_run8;
1137             break;
1138         case CHROMA_444:
1139             h->mc.prefetch_fenc = h->mc.prefetch_fenc_422; /* FIXME: doesn't cover V plane */
1140             h->loopf.deblock_chroma_mbaff = h->loopf.deblock_luma_mbaff;
1141             h->loopf.deblock_chroma_intra_mbaff = h->loopf.deblock_luma_intra_mbaff;
1142             break;
1143     }
1144 }
1145
1146 static void x264_set_aspect_ratio( x264_t *h, x264_param_t *param, int initial )
1147 {
1148     /* VUI */
1149     if( param->vui.i_sar_width > 0 && param->vui.i_sar_height > 0 )
1150     {
1151         uint32_t i_w = param->vui.i_sar_width;
1152         uint32_t i_h = param->vui.i_sar_height;
1153         uint32_t old_w = h->param.vui.i_sar_width;
1154         uint32_t old_h = h->param.vui.i_sar_height;
1155
1156         x264_reduce_fraction( &i_w, &i_h );
1157
1158         while( i_w > 65535 || i_h > 65535 )
1159         {
1160             i_w /= 2;
1161             i_h /= 2;
1162         }
1163
1164         x264_reduce_fraction( &i_w, &i_h );
1165
1166         if( i_w != old_w || i_h != old_h || initial )
1167         {
1168             h->param.vui.i_sar_width = 0;
1169             h->param.vui.i_sar_height = 0;
1170             if( i_w == 0 || i_h == 0 )
1171                 x264_log( h, X264_LOG_WARNING, "cannot create valid sample aspect ratio\n" );
1172             else
1173             {
1174                 x264_log( h, initial?X264_LOG_INFO:X264_LOG_DEBUG, "using SAR=%d/%d\n", i_w, i_h );
1175                 h->param.vui.i_sar_width = i_w;
1176                 h->param.vui.i_sar_height = i_h;
1177             }
1178             x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
1179         }
1180     }
1181 }
1182
1183 /****************************************************************************
1184  * x264_encoder_open:
1185  ****************************************************************************/
1186 x264_t *x264_encoder_open( x264_param_t *param )
1187 {
1188     x264_t *h;
1189     char buf[1000], *p;
1190     int qp, i_slicetype_length;
1191
1192     CHECKED_MALLOCZERO( h, sizeof(x264_t) );
1193
1194     /* Create a copy of param */
1195     memcpy( &h->param, param, sizeof(x264_param_t) );
1196
1197     if( param->param_free )
1198         param->param_free( param );
1199
1200     if( x264_threading_init() )
1201     {
1202         x264_log( h, X264_LOG_ERROR, "unable to initialize threading\n" );
1203         goto fail;
1204     }
1205
1206     if( x264_validate_parameters( h, 1 ) < 0 )
1207         goto fail;
1208
1209     if( h->param.psz_cqm_file )
1210         if( x264_cqm_parse_file( h, h->param.psz_cqm_file ) < 0 )
1211             goto fail;
1212
1213     if( h->param.rc.psz_stat_out )
1214         h->param.rc.psz_stat_out = strdup( h->param.rc.psz_stat_out );
1215     if( h->param.rc.psz_stat_in )
1216         h->param.rc.psz_stat_in = strdup( h->param.rc.psz_stat_in );
1217
1218     x264_reduce_fraction( &h->param.i_fps_num, &h->param.i_fps_den );
1219     x264_reduce_fraction( &h->param.i_timebase_num, &h->param.i_timebase_den );
1220
1221     /* Init x264_t */
1222     h->i_frame = -1;
1223     h->i_frame_num = 0;
1224     h->i_idr_pic_id = 0;
1225
1226     if( (uint64_t)h->param.i_timebase_den * 2 > UINT32_MAX )
1227     {
1228         x264_log( h, X264_LOG_ERROR, "Effective timebase denominator %u exceeds H.264 maximum\n", h->param.i_timebase_den );
1229         goto fail;
1230     }
1231
1232     x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
1233     x264_pps_init( h->pps, h->param.i_sps_id, &h->param, h->sps );
1234
1235     x264_set_aspect_ratio( h, &h->param, 1 );
1236
1237     x264_validate_levels( h, 1 );
1238
1239     h->chroma_qp_table = i_chroma_qp_table + 12 + h->pps->i_chroma_qp_index_offset;
1240
1241     if( x264_cqm_init( h ) < 0 )
1242         goto fail;
1243
1244     h->mb.i_mb_width = h->sps->i_mb_width;
1245     h->mb.i_mb_height = h->sps->i_mb_height;
1246     h->mb.i_mb_count = h->mb.i_mb_width * h->mb.i_mb_height;
1247
1248     h->mb.chroma_h_shift = CHROMA_FORMAT == CHROMA_420 || CHROMA_FORMAT == CHROMA_422;
1249     h->mb.chroma_v_shift = CHROMA_FORMAT == CHROMA_420;
1250
1251     /* Adaptive MBAFF and subme 0 are not supported as we require halving motion
1252      * vectors during prediction, resulting in hpel mvs.
1253      * The chosen solution is to make MBAFF non-adaptive in this case. */
1254     h->mb.b_adaptive_mbaff = PARAM_INTERLACED && h->param.analyse.i_subpel_refine;
1255
1256     /* Init frames. */
1257     if( h->param.i_bframe_adaptive == X264_B_ADAPT_TRELLIS && !h->param.rc.b_stat_read )
1258         h->frames.i_delay = X264_MAX(h->param.i_bframe,3)*4;
1259     else
1260         h->frames.i_delay = h->param.i_bframe;
1261     if( h->param.rc.b_mb_tree || h->param.rc.i_vbv_buffer_size )
1262         h->frames.i_delay = X264_MAX( h->frames.i_delay, h->param.rc.i_lookahead );
1263     i_slicetype_length = h->frames.i_delay;
1264     h->frames.i_delay += h->i_thread_frames - 1;
1265     h->frames.i_delay += h->param.i_sync_lookahead;
1266     h->frames.i_delay += h->param.b_vfr_input;
1267     h->frames.i_bframe_delay = h->param.i_bframe ? (h->param.i_bframe_pyramid ? 2 : 1) : 0;
1268
1269     h->frames.i_max_ref0 = h->param.i_frame_reference;
1270     h->frames.i_max_ref1 = X264_MIN( h->sps->vui.i_num_reorder_frames, h->param.i_frame_reference );
1271     h->frames.i_max_dpb  = h->sps->vui.i_max_dec_frame_buffering;
1272     h->frames.b_have_lowres = !h->param.rc.b_stat_read
1273         && ( h->param.rc.i_rc_method == X264_RC_ABR
1274           || h->param.rc.i_rc_method == X264_RC_CRF
1275           || h->param.i_bframe_adaptive
1276           || h->param.i_scenecut_threshold
1277           || h->param.rc.b_mb_tree
1278           || h->param.analyse.i_weighted_pred );
1279     h->frames.b_have_lowres |= h->param.rc.b_stat_read && h->param.rc.i_vbv_buffer_size > 0;
1280     h->frames.b_have_sub8x8_esa = !!(h->param.analyse.inter & X264_ANALYSE_PSUB8x8);
1281
1282     h->frames.i_last_idr =
1283     h->frames.i_last_keyframe = - h->param.i_keyint_max;
1284     h->frames.i_input    = 0;
1285     h->frames.i_largest_pts = h->frames.i_second_largest_pts = -1;
1286     h->frames.i_poc_last_open_gop = -1;
1287
1288     CHECKED_MALLOCZERO( h->frames.unused[0], (h->frames.i_delay + 3) * sizeof(x264_frame_t *) );
1289     /* Allocate room for max refs plus a few extra just in case. */
1290     CHECKED_MALLOCZERO( h->frames.unused[1], (h->i_thread_frames + X264_REF_MAX + 4) * sizeof(x264_frame_t *) );
1291     CHECKED_MALLOCZERO( h->frames.current, (h->param.i_sync_lookahead + h->param.i_bframe
1292                         + h->i_thread_frames + 3) * sizeof(x264_frame_t *) );
1293     if( h->param.analyse.i_weighted_pred > 0 )
1294         CHECKED_MALLOCZERO( h->frames.blank_unused, h->i_thread_frames * 4 * sizeof(x264_frame_t *) );
1295     h->i_ref[0] = h->i_ref[1] = 0;
1296     h->i_cpb_delay = h->i_coded_fields = h->i_disp_fields = 0;
1297     h->i_prev_duration = ((uint64_t)h->param.i_fps_den * h->sps->vui.i_time_scale) / ((uint64_t)h->param.i_fps_num * h->sps->vui.i_num_units_in_tick);
1298     h->i_disp_fields_last_frame = -1;
1299     x264_rdo_init();
1300
1301     /* init CPU functions */
1302     x264_predict_16x16_init( h->param.cpu, h->predict_16x16 );
1303     x264_predict_8x8c_init( h->param.cpu, h->predict_8x8c );
1304     x264_predict_8x16c_init( h->param.cpu, h->predict_8x16c );
1305     x264_predict_8x8_init( h->param.cpu, h->predict_8x8, &h->predict_8x8_filter );
1306     x264_predict_4x4_init( h->param.cpu, h->predict_4x4 );
1307     x264_pixel_init( h->param.cpu, &h->pixf );
1308     x264_dct_init( h->param.cpu, &h->dctf );
1309     x264_zigzag_init( h->param.cpu, &h->zigzagf_progressive, &h->zigzagf_interlaced );
1310     memcpy( &h->zigzagf, PARAM_INTERLACED ? &h->zigzagf_interlaced : &h->zigzagf_progressive, sizeof(h->zigzagf) );
1311     x264_mc_init( h->param.cpu, &h->mc, h->param.b_cpu_independent );
1312     x264_quant_init( h, h->param.cpu, &h->quantf );
1313     x264_deblock_init( h->param.cpu, &h->loopf, PARAM_INTERLACED );
1314     x264_bitstream_init( h->param.cpu, &h->bsf );
1315     if( h->param.b_cabac )
1316         x264_cabac_init( h );
1317     else
1318         x264_stack_align( x264_cavlc_init, h );
1319
1320     mbcmp_init( h );
1321     chroma_dsp_init( h );
1322
1323     p = buf + sprintf( buf, "using cpu capabilities:" );
1324     for( int i = 0; x264_cpu_names[i].flags; i++ )
1325     {
1326         if( !strcmp(x264_cpu_names[i].name, "SSE")
1327             && h->param.cpu & (X264_CPU_SSE2) )
1328             continue;
1329         if( !strcmp(x264_cpu_names[i].name, "SSE2")
1330             && h->param.cpu & (X264_CPU_SSE2_IS_FAST|X264_CPU_SSE2_IS_SLOW) )
1331             continue;
1332         if( !strcmp(x264_cpu_names[i].name, "SSE3")
1333             && (h->param.cpu & X264_CPU_SSSE3 || !(h->param.cpu & X264_CPU_CACHELINE_64)) )
1334             continue;
1335         if( !strcmp(x264_cpu_names[i].name, "SSE4.1")
1336             && (h->param.cpu & X264_CPU_SSE42) )
1337             continue;
1338         if( !strcmp(x264_cpu_names[i].name, "BMI1")
1339             && (h->param.cpu & X264_CPU_BMI2) )
1340             continue;
1341         if( (h->param.cpu & x264_cpu_names[i].flags) == x264_cpu_names[i].flags
1342             && (!i || x264_cpu_names[i].flags != x264_cpu_names[i-1].flags) )
1343             p += sprintf( p, " %s", x264_cpu_names[i].name );
1344     }
1345     if( !h->param.cpu )
1346         p += sprintf( p, " none!" );
1347     x264_log( h, X264_LOG_INFO, "%s\n", buf );
1348
1349     float *logs = x264_analyse_prepare_costs( h );
1350     if( !logs )
1351         goto fail;
1352     for( qp = X264_MIN( h->param.rc.i_qp_min, QP_MAX_SPEC ); qp <= h->param.rc.i_qp_max; qp++ )
1353         if( x264_analyse_init_costs( h, logs, qp ) )
1354             goto fail;
1355     if( x264_analyse_init_costs( h, logs, X264_LOOKAHEAD_QP ) )
1356         goto fail;
1357     x264_free( logs );
1358
1359     static const uint16_t cost_mv_correct[7] = { 24, 47, 95, 189, 379, 757, 1515 };
1360     /* Checks for known miscompilation issues. */
1361     if( h->cost_mv[X264_LOOKAHEAD_QP][2013] != cost_mv_correct[BIT_DEPTH-8] )
1362     {
1363         x264_log( h, X264_LOG_ERROR, "MV cost test failed: x264 has been miscompiled!\n" );
1364         goto fail;
1365     }
1366
1367     /* Must be volatile or else GCC will optimize it out. */
1368     volatile int temp = 392;
1369     if( x264_clz( temp ) != 23 )
1370     {
1371         x264_log( h, X264_LOG_ERROR, "CLZ test failed: x264 has been miscompiled!\n" );
1372 #if ARCH_X86 || ARCH_X86_64
1373         x264_log( h, X264_LOG_ERROR, "Are you attempting to run an SSE4a/LZCNT-targeted build on a CPU that\n" );
1374         x264_log( h, X264_LOG_ERROR, "doesn't support it?\n" );
1375 #endif
1376         goto fail;
1377     }
1378
1379     h->out.i_nal = 0;
1380     h->out.i_bitstream = X264_MAX( 1000000, h->param.i_width * h->param.i_height * 4
1381         * ( h->param.rc.i_rc_method == X264_RC_ABR ? pow( 0.95, h->param.rc.i_qp_min )
1382           : pow( 0.95, h->param.rc.i_qp_constant ) * X264_MAX( 1, h->param.rc.f_ip_factor )));
1383
1384     h->nal_buffer_size = h->out.i_bitstream * 3/2 + 4 + 64; /* +4 for startcode, +64 for nal_escape assembly padding */
1385     CHECKED_MALLOC( h->nal_buffer, h->nal_buffer_size );
1386
1387     if( h->param.i_threads > 1 &&
1388         x264_threadpool_init( &h->threadpool, h->param.i_threads, (void*)x264_encoder_thread_init, h ) )
1389         goto fail;
1390     if( h->param.i_lookahead_threads > 1 &&
1391         x264_threadpool_init( &h->lookaheadpool, h->param.i_lookahead_threads, (void*)x264_lookahead_thread_init, h ) )
1392         goto fail;
1393
1394     h->thread[0] = h;
1395     for( int i = 1; i < h->param.i_threads + !!h->param.i_sync_lookahead; i++ )
1396         CHECKED_MALLOC( h->thread[i], sizeof(x264_t) );
1397     if( h->param.i_lookahead_threads > 1 )
1398         for( int i = 0; i < h->param.i_lookahead_threads; i++ )
1399         {
1400             CHECKED_MALLOC( h->lookahead_thread[i], sizeof(x264_t) );
1401             *h->lookahead_thread[i] = *h;
1402         }
1403
1404     for( int i = 0; i < h->param.i_threads; i++ )
1405     {
1406         int init_nal_count = h->param.i_slice_count + 3;
1407         int allocate_threadlocal_data = !h->param.b_sliced_threads || !i;
1408         if( i > 0 )
1409             *h->thread[i] = *h;
1410
1411         if( x264_pthread_mutex_init( &h->thread[i]->mutex, NULL ) )
1412             goto fail;
1413         if( x264_pthread_cond_init( &h->thread[i]->cv, NULL ) )
1414             goto fail;
1415
1416         if( allocate_threadlocal_data )
1417         {
1418             h->thread[i]->fdec = x264_frame_pop_unused( h, 1 );
1419             if( !h->thread[i]->fdec )
1420                 goto fail;
1421         }
1422         else
1423             h->thread[i]->fdec = h->thread[0]->fdec;
1424
1425         CHECKED_MALLOC( h->thread[i]->out.p_bitstream, h->out.i_bitstream );
1426         /* Start each thread with room for init_nal_count NAL units; it'll realloc later if needed. */
1427         CHECKED_MALLOC( h->thread[i]->out.nal, init_nal_count*sizeof(x264_nal_t) );
1428         h->thread[i]->out.i_nals_allocated = init_nal_count;
1429
1430         if( allocate_threadlocal_data && x264_macroblock_cache_allocate( h->thread[i] ) < 0 )
1431             goto fail;
1432     }
1433
1434 #if HAVE_OPENCL
1435     if( h->param.b_opencl && x264_opencl_init( h ) < 0 )
1436         h->param.b_opencl = 0;
1437 #endif
1438
1439     if( x264_lookahead_init( h, i_slicetype_length ) )
1440         goto fail;
1441
1442     for( int i = 0; i < h->param.i_threads; i++ )
1443         if( x264_macroblock_thread_allocate( h->thread[i], 0 ) < 0 )
1444             goto fail;
1445
1446     if( x264_ratecontrol_new( h ) < 0 )
1447         goto fail;
1448
1449     if( h->param.i_nal_hrd )
1450     {
1451         x264_log( h, X264_LOG_DEBUG, "HRD bitrate: %i bits/sec\n", h->sps->vui.hrd.i_bit_rate_unscaled );
1452         x264_log( h, X264_LOG_DEBUG, "CPB size: %i bits\n", h->sps->vui.hrd.i_cpb_size_unscaled );
1453     }
1454
1455     if( h->param.psz_dump_yuv )
1456     {
1457         /* create or truncate the reconstructed video file */
1458         FILE *f = fopen( h->param.psz_dump_yuv, "w" );
1459         if( !f )
1460         {
1461             x264_log( h, X264_LOG_ERROR, "dump_yuv: can't write to %s\n", h->param.psz_dump_yuv );
1462             goto fail;
1463         }
1464         else if( !x264_is_regular_file( f ) )
1465         {
1466             x264_log( h, X264_LOG_ERROR, "dump_yuv: incompatible with non-regular file %s\n", h->param.psz_dump_yuv );
1467             goto fail;
1468         }
1469         fclose( f );
1470     }
1471
1472     const char *profile = h->sps->i_profile_idc == PROFILE_BASELINE ? "Constrained Baseline" :
1473                           h->sps->i_profile_idc == PROFILE_MAIN ? "Main" :
1474                           h->sps->i_profile_idc == PROFILE_HIGH ? "High" :
1475                           h->sps->i_profile_idc == PROFILE_HIGH10 ? (h->sps->b_constraint_set3 == 1 ? "High 10 Intra" : "High 10") :
1476                           h->sps->i_profile_idc == PROFILE_HIGH422 ? (h->sps->b_constraint_set3 == 1 ? "High 4:2:2 Intra" : "High 4:2:2") :
1477                           h->sps->b_constraint_set3 == 1 ? "High 4:4:4 Intra" : "High 4:4:4 Predictive";
1478     char level[4];
1479     snprintf( level, sizeof(level), "%d.%d", h->sps->i_level_idc/10, h->sps->i_level_idc%10 );
1480     if( h->sps->i_level_idc == 9 || ( h->sps->i_level_idc == 11 && h->sps->b_constraint_set3 &&
1481         (h->sps->i_profile_idc == PROFILE_BASELINE || h->sps->i_profile_idc == PROFILE_MAIN) ) )
1482         strcpy( level, "1b" );
1483
1484     if( h->sps->i_profile_idc < PROFILE_HIGH10 )
1485     {
1486         x264_log( h, X264_LOG_INFO, "profile %s, level %s\n",
1487             profile, level );
1488     }
1489     else
1490     {
1491         static const char * const subsampling[4] = { "4:0:0", "4:2:0", "4:2:2", "4:4:4" };
1492         x264_log( h, X264_LOG_INFO, "profile %s, level %s, %s %d-bit\n",
1493             profile, level, subsampling[CHROMA_FORMAT], BIT_DEPTH );
1494     }
1495
1496     return h;
1497 fail:
1498     x264_free( h );
1499     return NULL;
1500 }
1501
1502 /****************************************************************************
1503  * x264_encoder_reconfig:
1504  ****************************************************************************/
1505 int x264_encoder_reconfig( x264_t *h, x264_param_t *param )
1506 {
1507     /* If the previous frame isn't done encoding, reconfiguring is probably dangerous. */
1508     if( h->param.b_sliced_threads )
1509         if( x264_threadpool_wait_all( h ) < 0 )
1510             return -1;
1511
1512     int rc_reconfig = 0;
1513     h = h->thread[h->thread[0]->i_thread_phase];
1514     x264_set_aspect_ratio( h, param, 0 );
1515 #define COPY(var) h->param.var = param->var
1516     COPY( i_frame_reference ); // but never uses more refs than initially specified
1517     COPY( i_bframe_bias );
1518     if( h->param.i_scenecut_threshold )
1519         COPY( i_scenecut_threshold ); // can't turn it on or off, only vary the threshold
1520     COPY( b_deblocking_filter );
1521     COPY( i_deblocking_filter_alphac0 );
1522     COPY( i_deblocking_filter_beta );
1523     COPY( i_frame_packing );
1524     COPY( analyse.inter );
1525     COPY( analyse.intra );
1526     COPY( analyse.i_direct_mv_pred );
1527     /* Scratch buffer prevents me_range from being increased for esa/tesa */
1528     if( h->param.analyse.i_me_method < X264_ME_ESA || param->analyse.i_me_range < h->param.analyse.i_me_range )
1529         COPY( analyse.i_me_range );
1530     COPY( analyse.i_noise_reduction );
1531     /* We can't switch out of subme=0 during encoding. */
1532     if( h->param.analyse.i_subpel_refine )
1533         COPY( analyse.i_subpel_refine );
1534     COPY( analyse.i_trellis );
1535     COPY( analyse.b_chroma_me );
1536     COPY( analyse.b_dct_decimate );
1537     COPY( analyse.b_fast_pskip );
1538     COPY( analyse.b_mixed_references );
1539     COPY( analyse.f_psy_rd );
1540     COPY( analyse.f_psy_trellis );
1541     COPY( crop_rect );
1542     // can only twiddle these if they were enabled to begin with:
1543     if( h->param.analyse.i_me_method >= X264_ME_ESA || param->analyse.i_me_method < X264_ME_ESA )
1544         COPY( analyse.i_me_method );
1545     if( h->param.analyse.i_me_method >= X264_ME_ESA && !h->frames.b_have_sub8x8_esa )
1546         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
1547     if( h->pps->b_transform_8x8_mode )
1548         COPY( analyse.b_transform_8x8 );
1549     if( h->frames.i_max_ref1 > 1 )
1550         COPY( i_bframe_pyramid );
1551     COPY( i_slice_max_size );
1552     COPY( i_slice_max_mbs );
1553     COPY( i_slice_min_mbs );
1554     COPY( i_slice_count );
1555     COPY( i_slice_count_max );
1556     COPY( b_tff );
1557
1558     /* VBV can't be turned on if it wasn't on to begin with */
1559     if( h->param.rc.i_vbv_max_bitrate > 0 && h->param.rc.i_vbv_buffer_size > 0 &&
1560           param->rc.i_vbv_max_bitrate > 0 &&   param->rc.i_vbv_buffer_size > 0 )
1561     {
1562         rc_reconfig |= h->param.rc.i_vbv_max_bitrate != param->rc.i_vbv_max_bitrate;
1563         rc_reconfig |= h->param.rc.i_vbv_buffer_size != param->rc.i_vbv_buffer_size;
1564         rc_reconfig |= h->param.rc.i_bitrate != param->rc.i_bitrate;
1565         COPY( rc.i_vbv_max_bitrate );
1566         COPY( rc.i_vbv_buffer_size );
1567         COPY( rc.i_bitrate );
1568     }
1569     rc_reconfig |= h->param.rc.f_rf_constant != param->rc.f_rf_constant;
1570     rc_reconfig |= h->param.rc.f_rf_constant_max != param->rc.f_rf_constant_max;
1571     COPY( rc.f_rf_constant );
1572     COPY( rc.f_rf_constant_max );
1573 #undef COPY
1574
1575     mbcmp_init( h );
1576
1577     int ret = x264_validate_parameters( h, 0 );
1578
1579     /* Supported reconfiguration options (1-pass only):
1580      * vbv-maxrate
1581      * vbv-bufsize
1582      * crf
1583      * bitrate (CBR only) */
1584     if( !ret && rc_reconfig )
1585         x264_ratecontrol_init_reconfigurable( h, 0 );
1586
1587     return ret;
1588 }
1589
1590 /****************************************************************************
1591  * x264_encoder_parameters:
1592  ****************************************************************************/
1593 void x264_encoder_parameters( x264_t *h, x264_param_t *param )
1594 {
1595     memcpy( param, &h->thread[h->i_thread_phase]->param, sizeof(x264_param_t) );
1596 }
1597
1598 /* internal usage */
1599 static void x264_nal_start( x264_t *h, int i_type, int i_ref_idc )
1600 {
1601     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1602
1603     nal->i_ref_idc        = i_ref_idc;
1604     nal->i_type           = i_type;
1605     nal->b_long_startcode = 1;
1606
1607     nal->i_payload= 0;
1608     nal->p_payload= &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8];
1609 }
1610
1611 /* if number of allocated nals is not enough, re-allocate a larger one. */
1612 static int x264_nal_check_buffer( x264_t *h )
1613 {
1614     if( h->out.i_nal >= h->out.i_nals_allocated )
1615     {
1616         x264_nal_t *new_out = x264_malloc( sizeof(x264_nal_t) * (h->out.i_nals_allocated*2) );
1617         if( !new_out )
1618             return -1;
1619         memcpy( new_out, h->out.nal, sizeof(x264_nal_t) * (h->out.i_nals_allocated) );
1620         x264_free( h->out.nal );
1621         h->out.nal = new_out;
1622         h->out.i_nals_allocated *= 2;
1623     }
1624     return 0;
1625 }
1626
1627 static int x264_nal_end( x264_t *h )
1628 {
1629     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1630     uint8_t *end = &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8];
1631     nal->i_payload = end - nal->p_payload;
1632     /* Assembly implementation of nal_escape reads past the end of the input.
1633      * While undefined padding wouldn't actually affect the output, it makes valgrind unhappy. */
1634     memset( end, 0xff, 64 );
1635     if( h->param.nalu_process )
1636         h->param.nalu_process( h, nal, h->fenc->opaque );
1637     h->out.i_nal++;
1638
1639     return x264_nal_check_buffer( h );
1640 }
1641
1642 static int x264_encoder_encapsulate_nals( x264_t *h, int start )
1643 {
1644     int nal_size = 0, previous_nal_size = 0;
1645
1646     if( h->param.nalu_process )
1647     {
1648         for( int i = start; i < h->out.i_nal; i++ )
1649             nal_size += h->out.nal[i].i_payload;
1650         return nal_size;
1651     }
1652
1653     for( int i = 0; i < start; i++ )
1654         previous_nal_size += h->out.nal[i].i_payload;
1655
1656     for( int i = start; i < h->out.i_nal; i++ )
1657         nal_size += h->out.nal[i].i_payload;
1658
1659     /* Worst-case NAL unit escaping: reallocate the buffer if it's too small. */
1660     int necessary_size = nal_size * 3/2 + h->out.i_nal * 4 + 4 + 64;
1661     if( h->nal_buffer_size < necessary_size )
1662     {
1663         h->nal_buffer_size = necessary_size * 2;
1664         uint8_t *buf = x264_malloc( h->nal_buffer_size );
1665         if( !buf )
1666             return -1;
1667         if( previous_nal_size )
1668             memcpy( buf, h->nal_buffer, previous_nal_size );
1669         x264_free( h->nal_buffer );
1670         h->nal_buffer = buf;
1671     }
1672
1673     uint8_t *nal_buffer = h->nal_buffer + previous_nal_size;
1674
1675     for( int i = start; i < h->out.i_nal; i++ )
1676     {
1677         h->out.nal[i].b_long_startcode = !i || h->out.nal[i].i_type == NAL_SPS || h->out.nal[i].i_type == NAL_PPS;
1678         x264_nal_encode( h, nal_buffer, &h->out.nal[i] );
1679         nal_buffer += h->out.nal[i].i_payload;
1680     }
1681
1682     x264_emms();
1683
1684     return nal_buffer - (h->nal_buffer + previous_nal_size);
1685 }
1686
1687 /****************************************************************************
1688  * x264_encoder_headers:
1689  ****************************************************************************/
1690 int x264_encoder_headers( x264_t *h, x264_nal_t **pp_nal, int *pi_nal )
1691 {
1692     int frame_size = 0;
1693     /* init bitstream context */
1694     h->out.i_nal = 0;
1695     bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
1696
1697     /* Write SEI, SPS and PPS. */
1698
1699     /* generate sequence parameters */
1700     x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
1701     x264_sps_write( &h->out.bs, h->sps );
1702     if( x264_nal_end( h ) )
1703         return -1;
1704
1705     /* generate picture parameters */
1706     x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
1707     x264_pps_write( &h->out.bs, h->sps, h->pps );
1708     if( x264_nal_end( h ) )
1709         return -1;
1710
1711     /* identify ourselves */
1712     x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
1713     if( x264_sei_version_write( h, &h->out.bs ) )
1714         return -1;
1715     if( x264_nal_end( h ) )
1716         return -1;
1717
1718     frame_size = x264_encoder_encapsulate_nals( h, 0 );
1719     if( frame_size < 0 )
1720         return -1;
1721
1722     /* now set output*/
1723     *pi_nal = h->out.i_nal;
1724     *pp_nal = &h->out.nal[0];
1725     h->out.i_nal = 0;
1726
1727     return frame_size;
1728 }
1729
1730 /* Check to see whether we have chosen a reference list ordering different
1731  * from the standard's default. */
1732 static inline void x264_reference_check_reorder( x264_t *h )
1733 {
1734     /* The reorder check doesn't check for missing frames, so just
1735      * force a reorder if one of the reference list is corrupt. */
1736     for( int i = 0; h->frames.reference[i]; i++ )
1737         if( h->frames.reference[i]->b_corrupt )
1738         {
1739             h->b_ref_reorder[0] = 1;
1740             return;
1741         }
1742     for( int list = 0; list <= (h->sh.i_type == SLICE_TYPE_B); list++ )
1743         for( int i = 0; i < h->i_ref[list] - 1; i++ )
1744         {
1745             int framenum_diff = h->fref[list][i+1]->i_frame_num - h->fref[list][i]->i_frame_num;
1746             int poc_diff = h->fref[list][i+1]->i_poc - h->fref[list][i]->i_poc;
1747             /* P and B-frames use different default orders. */
1748             if( h->sh.i_type == SLICE_TYPE_P ? framenum_diff > 0 : list == 1 ? poc_diff < 0 : poc_diff > 0 )
1749             {
1750                 h->b_ref_reorder[list] = 1;
1751                 return;
1752             }
1753         }
1754 }
1755
1756 /* return -1 on failure, else return the index of the new reference frame */
1757 int x264_weighted_reference_duplicate( x264_t *h, int i_ref, const x264_weight_t *w )
1758 {
1759     int i = h->i_ref[0];
1760     int j = 1;
1761     x264_frame_t *newframe;
1762     if( i <= 1 ) /* empty list, definitely can't duplicate frame */
1763         return -1;
1764
1765     //Duplication is only used in X264_WEIGHTP_SMART
1766     if( h->param.analyse.i_weighted_pred != X264_WEIGHTP_SMART )
1767         return -1;
1768
1769     /* Duplication is a hack to compensate for crappy rounding in motion compensation.
1770      * With high bit depth, it's not worth doing, so turn it off except in the case of
1771      * unweighted dupes. */
1772     if( BIT_DEPTH > 8 && w != x264_weight_none )
1773         return -1;
1774
1775     newframe = x264_frame_pop_blank_unused( h );
1776     if( !newframe )
1777         return -1;
1778
1779     //FIXME: probably don't need to copy everything
1780     *newframe = *h->fref[0][i_ref];
1781     newframe->i_reference_count = 1;
1782     newframe->orig = h->fref[0][i_ref];
1783     newframe->b_duplicate = 1;
1784     memcpy( h->fenc->weight[j], w, sizeof(h->fenc->weight[i]) );
1785
1786     /* shift the frames to make space for the dupe. */
1787     h->b_ref_reorder[0] = 1;
1788     if( h->i_ref[0] < X264_REF_MAX )
1789         ++h->i_ref[0];
1790     h->fref[0][X264_REF_MAX-1] = NULL;
1791     x264_frame_unshift( &h->fref[0][j], newframe );
1792
1793     return j;
1794 }
1795
1796 static void x264_weighted_pred_init( x264_t *h )
1797 {
1798     /* for now no analysis and set all weights to nothing */
1799     for( int i_ref = 0; i_ref < h->i_ref[0]; i_ref++ )
1800         h->fenc->weighted[i_ref] = h->fref[0][i_ref]->filtered[0][0];
1801
1802     // FIXME: This only supports weighting of one reference frame
1803     // and duplicates of that frame.
1804     h->fenc->i_lines_weighted = 0;
1805
1806     for( int i_ref = 0; i_ref < (h->i_ref[0] << SLICE_MBAFF); i_ref++ )
1807         for( int i = 0; i < 3; i++ )
1808             h->sh.weight[i_ref][i].weightfn = NULL;
1809
1810
1811     if( h->sh.i_type != SLICE_TYPE_P || h->param.analyse.i_weighted_pred <= 0 )
1812         return;
1813
1814     int i_padv = PADV << PARAM_INTERLACED;
1815     int denom = -1;
1816     int weightplane[2] = { 0, 0 };
1817     int buffer_next = 0;
1818     for( int i = 0; i < 3; i++ )
1819     {
1820         for( int j = 0; j < h->i_ref[0]; j++ )
1821         {
1822             if( h->fenc->weight[j][i].weightfn )
1823             {
1824                 h->sh.weight[j][i] = h->fenc->weight[j][i];
1825                 // if weight is useless, don't write it to stream
1826                 if( h->sh.weight[j][i].i_scale == 1<<h->sh.weight[j][i].i_denom && h->sh.weight[j][i].i_offset == 0 )
1827                     h->sh.weight[j][i].weightfn = NULL;
1828                 else
1829                 {
1830                     if( !weightplane[!!i] )
1831                     {
1832                         weightplane[!!i] = 1;
1833                         h->sh.weight[0][!!i].i_denom = denom = h->sh.weight[j][i].i_denom;
1834                         assert( x264_clip3( denom, 0, 7 ) == denom );
1835                     }
1836
1837                     assert( h->sh.weight[j][i].i_denom == denom );
1838                     if( !i )
1839                     {
1840                         h->fenc->weighted[j] = h->mb.p_weight_buf[buffer_next++] + h->fenc->i_stride[0] * i_padv + PADH;
1841                         //scale full resolution frame
1842                         if( h->param.i_threads == 1 )
1843                         {
1844                             pixel *src = h->fref[0][j]->filtered[0][0] - h->fref[0][j]->i_stride[0]*i_padv - PADH;
1845                             pixel *dst = h->fenc->weighted[j] - h->fenc->i_stride[0]*i_padv - PADH;
1846                             int stride = h->fenc->i_stride[0];
1847                             int width = h->fenc->i_width[0] + PADH*2;
1848                             int height = h->fenc->i_lines[0] + i_padv*2;
1849                             x264_weight_scale_plane( h, dst, stride, src, stride, width, height, &h->sh.weight[j][0] );
1850                             h->fenc->i_lines_weighted = height;
1851                         }
1852                     }
1853                 }
1854             }
1855         }
1856     }
1857
1858     if( weightplane[1] )
1859         for( int i = 0; i < h->i_ref[0]; i++ )
1860         {
1861             if( h->sh.weight[i][1].weightfn && !h->sh.weight[i][2].weightfn )
1862             {
1863                 h->sh.weight[i][2].i_scale = 1 << h->sh.weight[0][1].i_denom;
1864                 h->sh.weight[i][2].i_offset = 0;
1865             }
1866             else if( h->sh.weight[i][2].weightfn && !h->sh.weight[i][1].weightfn )
1867             {
1868                 h->sh.weight[i][1].i_scale = 1 << h->sh.weight[0][1].i_denom;
1869                 h->sh.weight[i][1].i_offset = 0;
1870             }
1871         }
1872
1873     if( !weightplane[0] )
1874         h->sh.weight[0][0].i_denom = 0;
1875     if( !weightplane[1] )
1876         h->sh.weight[0][1].i_denom = 0;
1877     h->sh.weight[0][2].i_denom = h->sh.weight[0][1].i_denom;
1878 }
1879
1880 static inline int x264_reference_distance( x264_t *h, x264_frame_t *frame )
1881 {
1882     if( h->param.i_frame_packing == 5 )
1883         return abs((h->fenc->i_frame&~1) - (frame->i_frame&~1)) +
1884                   ((h->fenc->i_frame&1) != (frame->i_frame&1));
1885     else
1886         return abs(h->fenc->i_frame - frame->i_frame);
1887 }
1888
1889 static inline void x264_reference_build_list( x264_t *h, int i_poc )
1890 {
1891     int b_ok;
1892
1893     /* build ref list 0/1 */
1894     h->mb.pic.i_fref[0] = h->i_ref[0] = 0;
1895     h->mb.pic.i_fref[1] = h->i_ref[1] = 0;
1896     if( h->sh.i_type == SLICE_TYPE_I )
1897         return;
1898
1899     for( int i = 0; h->frames.reference[i]; i++ )
1900     {
1901         if( h->frames.reference[i]->b_corrupt )
1902             continue;
1903         if( h->frames.reference[i]->i_poc < i_poc )
1904             h->fref[0][h->i_ref[0]++] = h->frames.reference[i];
1905         else if( h->frames.reference[i]->i_poc > i_poc )
1906             h->fref[1][h->i_ref[1]++] = h->frames.reference[i];
1907     }
1908
1909     /* Order reference lists by distance from the current frame. */
1910     for( int list = 0; list < 2; list++ )
1911     {
1912         h->fref_nearest[list] = h->fref[list][0];
1913         do
1914         {
1915             b_ok = 1;
1916             for( int i = 0; i < h->i_ref[list] - 1; i++ )
1917             {
1918                 if( list ? h->fref[list][i+1]->i_poc < h->fref_nearest[list]->i_poc
1919                          : h->fref[list][i+1]->i_poc > h->fref_nearest[list]->i_poc )
1920                     h->fref_nearest[list] = h->fref[list][i+1];
1921                 if( x264_reference_distance( h, h->fref[list][i] ) > x264_reference_distance( h, h->fref[list][i+1] ) )
1922                 {
1923                     XCHG( x264_frame_t*, h->fref[list][i], h->fref[list][i+1] );
1924                     b_ok = 0;
1925                     break;
1926                 }
1927             }
1928         } while( !b_ok );
1929     }
1930
1931     if( h->sh.i_mmco_remove_from_end )
1932         for( int i = h->i_ref[0]-1; i >= h->i_ref[0] - h->sh.i_mmco_remove_from_end; i-- )
1933         {
1934             int diff = h->i_frame_num - h->fref[0][i]->i_frame_num;
1935             h->sh.mmco[h->sh.i_mmco_command_count].i_poc = h->fref[0][i]->i_poc;
1936             h->sh.mmco[h->sh.i_mmco_command_count++].i_difference_of_pic_nums = diff;
1937         }
1938
1939     x264_reference_check_reorder( h );
1940
1941     h->i_ref[1] = X264_MIN( h->i_ref[1], h->frames.i_max_ref1 );
1942     h->i_ref[0] = X264_MIN( h->i_ref[0], h->frames.i_max_ref0 );
1943     h->i_ref[0] = X264_MIN( h->i_ref[0], h->param.i_frame_reference ); // if reconfig() has lowered the limit
1944
1945     /* For Blu-ray compliance, don't reference frames outside of the minigop. */
1946     if( IS_X264_TYPE_B( h->fenc->i_type ) && h->param.b_bluray_compat )
1947         h->i_ref[0] = X264_MIN( h->i_ref[0], IS_X264_TYPE_B( h->fref[0][0]->i_type ) + 1 );
1948
1949     /* add duplicates */
1950     if( h->fenc->i_type == X264_TYPE_P )
1951     {
1952         int idx = -1;
1953         if( h->param.analyse.i_weighted_pred >= X264_WEIGHTP_SIMPLE )
1954         {
1955             x264_weight_t w[3];
1956             w[1].weightfn = w[2].weightfn = NULL;
1957             if( h->param.rc.b_stat_read )
1958                 x264_ratecontrol_set_weights( h, h->fenc );
1959
1960             if( !h->fenc->weight[0][0].weightfn )
1961             {
1962                 h->fenc->weight[0][0].i_denom = 0;
1963                 SET_WEIGHT( w[0], 1, 1, 0, -1 );
1964                 idx = x264_weighted_reference_duplicate( h, 0, w );
1965             }
1966             else
1967             {
1968                 if( h->fenc->weight[0][0].i_scale == 1<<h->fenc->weight[0][0].i_denom )
1969                 {
1970                     SET_WEIGHT( h->fenc->weight[0][0], 1, 1, 0, h->fenc->weight[0][0].i_offset );
1971                 }
1972                 x264_weighted_reference_duplicate( h, 0, x264_weight_none );
1973                 if( h->fenc->weight[0][0].i_offset > -128 )
1974                 {
1975                     w[0] = h->fenc->weight[0][0];
1976                     w[0].i_offset--;
1977                     h->mc.weight_cache( h, &w[0] );
1978                     idx = x264_weighted_reference_duplicate( h, 0, w );
1979                 }
1980             }
1981         }
1982         h->mb.ref_blind_dupe = idx;
1983     }
1984
1985     assert( h->i_ref[0] + h->i_ref[1] <= X264_REF_MAX );
1986     h->mb.pic.i_fref[0] = h->i_ref[0];
1987     h->mb.pic.i_fref[1] = h->i_ref[1];
1988 }
1989
1990 static void x264_fdec_filter_row( x264_t *h, int mb_y, int pass )
1991 {
1992     /* mb_y is the mb to be encoded next, not the mb to be filtered here */
1993     int b_hpel = h->fdec->b_kept_as_ref;
1994     int b_deblock = h->sh.i_disable_deblocking_filter_idc != 1;
1995     int b_end = mb_y == h->i_threadslice_end;
1996     int b_measure_quality = 1;
1997     int min_y = mb_y - (1 << SLICE_MBAFF);
1998     int b_start = min_y == h->i_threadslice_start;
1999     /* Even in interlaced mode, deblocking never modifies more than 4 pixels
2000      * above each MB, as bS=4 doesn't happen for the top of interlaced mbpairs. */
2001     int minpix_y = min_y*16 - 4 * !b_start;
2002     int maxpix_y = mb_y*16 - 4 * !b_end;
2003     b_deblock &= b_hpel || h->param.b_full_recon || h->param.psz_dump_yuv;
2004     if( h->param.b_sliced_threads )
2005     {
2006         switch( pass )
2007         {
2008             /* During encode: only do deblock if asked for */
2009             default:
2010             case 0:
2011                 b_deblock &= h->param.b_full_recon;
2012                 b_hpel = 0;
2013                 break;
2014             /* During post-encode pass: do deblock if not done yet, do hpel for all
2015              * rows except those between slices. */
2016             case 1:
2017                 b_deblock &= !h->param.b_full_recon;
2018                 b_hpel &= !(b_start && min_y > 0);
2019                 b_measure_quality = 0;
2020                 break;
2021             /* Final pass: do the rows between slices in sequence. */
2022             case 2:
2023                 b_deblock = 0;
2024                 b_measure_quality = 0;
2025                 break;
2026         }
2027     }
2028     if( mb_y & SLICE_MBAFF )
2029         return;
2030     if( min_y < h->i_threadslice_start )
2031         return;
2032
2033     if( b_deblock )
2034         for( int y = min_y; y < mb_y; y += (1 << SLICE_MBAFF) )
2035             x264_frame_deblock_row( h, y );
2036
2037     /* FIXME: Prediction requires different borders for interlaced/progressive mc,
2038      * but the actual image data is equivalent. For now, maintain this
2039      * consistency by copying deblocked pixels between planes. */
2040     if( PARAM_INTERLACED && (!h->param.b_sliced_threads || pass == 1) )
2041         for( int p = 0; p < h->fdec->i_plane; p++ )
2042             for( int i = minpix_y>>(CHROMA_V_SHIFT && p); i < maxpix_y>>(CHROMA_V_SHIFT && p); i++ )
2043                 memcpy( h->fdec->plane_fld[p] + i*h->fdec->i_stride[p],
2044                         h->fdec->plane[p]     + i*h->fdec->i_stride[p],
2045                         h->mb.i_mb_width*16*sizeof(pixel) );
2046
2047     if( h->fdec->b_kept_as_ref && (!h->param.b_sliced_threads || pass == 1) )
2048         x264_frame_expand_border( h, h->fdec, min_y );
2049     if( b_hpel )
2050     {
2051         int end = mb_y == h->mb.i_mb_height;
2052         /* Can't do hpel until the previous slice is done encoding. */
2053         if( h->param.analyse.i_subpel_refine )
2054         {
2055             x264_frame_filter( h, h->fdec, min_y, end );
2056             x264_frame_expand_border_filtered( h, h->fdec, min_y, end );
2057         }
2058     }
2059
2060     if( SLICE_MBAFF && pass == 0 )
2061         for( int i = 0; i < 3; i++ )
2062         {
2063             XCHG( pixel *, h->intra_border_backup[0][i], h->intra_border_backup[3][i] );
2064             XCHG( pixel *, h->intra_border_backup[1][i], h->intra_border_backup[4][i] );
2065         }
2066
2067     if( h->i_thread_frames > 1 && h->fdec->b_kept_as_ref )
2068         x264_frame_cond_broadcast( h->fdec, mb_y*16 + (b_end ? 10000 : -(X264_THREAD_HEIGHT << SLICE_MBAFF)) );
2069
2070     if( b_measure_quality )
2071     {
2072         maxpix_y = X264_MIN( maxpix_y, h->param.i_height );
2073         if( h->param.analyse.b_psnr )
2074         {
2075             for( int p = 0; p < (CHROMA444 ? 3 : 1); p++ )
2076                 h->stat.frame.i_ssd[p] += x264_pixel_ssd_wxh( &h->pixf,
2077                     h->fdec->plane[p] + minpix_y * h->fdec->i_stride[p], h->fdec->i_stride[p],
2078                     h->fenc->plane[p] + minpix_y * h->fenc->i_stride[p], h->fenc->i_stride[p],
2079                     h->param.i_width, maxpix_y-minpix_y );
2080             if( !CHROMA444 )
2081             {
2082                 uint64_t ssd_u, ssd_v;
2083                 int v_shift = CHROMA_V_SHIFT;
2084                 x264_pixel_ssd_nv12( &h->pixf,
2085                     h->fdec->plane[1] + (minpix_y>>v_shift) * h->fdec->i_stride[1], h->fdec->i_stride[1],
2086                     h->fenc->plane[1] + (minpix_y>>v_shift) * h->fenc->i_stride[1], h->fenc->i_stride[1],
2087                     h->param.i_width>>1, (maxpix_y-minpix_y)>>v_shift, &ssd_u, &ssd_v );
2088                 h->stat.frame.i_ssd[1] += ssd_u;
2089                 h->stat.frame.i_ssd[2] += ssd_v;
2090             }
2091         }
2092
2093         if( h->param.analyse.b_ssim )
2094         {
2095             int ssim_cnt;
2096             x264_emms();
2097             /* offset by 2 pixels to avoid alignment of ssim blocks with dct blocks,
2098              * and overlap by 4 */
2099             minpix_y += b_start ? 2 : -6;
2100             h->stat.frame.f_ssim +=
2101                 x264_pixel_ssim_wxh( &h->pixf,
2102                     h->fdec->plane[0] + 2+minpix_y*h->fdec->i_stride[0], h->fdec->i_stride[0],
2103                     h->fenc->plane[0] + 2+minpix_y*h->fenc->i_stride[0], h->fenc->i_stride[0],
2104                     h->param.i_width-2, maxpix_y-minpix_y, h->scratch_buffer, &ssim_cnt );
2105             h->stat.frame.i_ssim_cnt += ssim_cnt;
2106         }
2107     }
2108 }
2109
2110 static inline int x264_reference_update( x264_t *h )
2111 {
2112     if( !h->fdec->b_kept_as_ref )
2113     {
2114         if( h->i_thread_frames > 1 )
2115         {
2116             x264_frame_push_unused( h, h->fdec );
2117             h->fdec = x264_frame_pop_unused( h, 1 );
2118             if( !h->fdec )
2119                 return -1;
2120         }
2121         return 0;
2122     }
2123
2124     /* apply mmco from previous frame. */
2125     for( int i = 0; i < h->sh.i_mmco_command_count; i++ )
2126         for( int j = 0; h->frames.reference[j]; j++ )
2127             if( h->frames.reference[j]->i_poc == h->sh.mmco[i].i_poc )
2128                 x264_frame_push_unused( h, x264_frame_shift( &h->frames.reference[j] ) );
2129
2130     /* move frame in the buffer */
2131     x264_frame_push( h->frames.reference, h->fdec );
2132     if( h->frames.reference[h->sps->i_num_ref_frames] )
2133         x264_frame_push_unused( h, x264_frame_shift( h->frames.reference ) );
2134     h->fdec = x264_frame_pop_unused( h, 1 );
2135     if( !h->fdec )
2136         return -1;
2137     return 0;
2138 }
2139
2140 static inline void x264_reference_reset( x264_t *h )
2141 {
2142     while( h->frames.reference[0] )
2143         x264_frame_push_unused( h, x264_frame_pop( h->frames.reference ) );
2144     h->fdec->i_poc =
2145     h->fenc->i_poc = 0;
2146 }
2147
2148 static inline void x264_reference_hierarchy_reset( x264_t *h )
2149 {
2150     int ref;
2151     int b_hasdelayframe = 0;
2152
2153     /* look for delay frames -- chain must only contain frames that are disposable */
2154     for( int i = 0; h->frames.current[i] && IS_DISPOSABLE( h->frames.current[i]->i_type ); i++ )
2155         b_hasdelayframe |= h->frames.current[i]->i_coded
2156                         != h->frames.current[i]->i_frame + h->sps->vui.i_num_reorder_frames;
2157
2158     /* This function must handle b-pyramid and clear frames for open-gop */
2159     if( h->param.i_bframe_pyramid != X264_B_PYRAMID_STRICT && !b_hasdelayframe && h->frames.i_poc_last_open_gop == -1 )
2160         return;
2161
2162     /* Remove last BREF. There will never be old BREFs in the
2163      * dpb during a BREF decode when pyramid == STRICT */
2164     for( ref = 0; h->frames.reference[ref]; ref++ )
2165     {
2166         if( ( h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT
2167             && h->frames.reference[ref]->i_type == X264_TYPE_BREF )
2168             || ( h->frames.reference[ref]->i_poc < h->frames.i_poc_last_open_gop
2169             && h->sh.i_type != SLICE_TYPE_B ) )
2170         {
2171             int diff = h->i_frame_num - h->frames.reference[ref]->i_frame_num;
2172             h->sh.mmco[h->sh.i_mmco_command_count].i_difference_of_pic_nums = diff;
2173             h->sh.mmco[h->sh.i_mmco_command_count++].i_poc = h->frames.reference[ref]->i_poc;
2174             x264_frame_push_unused( h, x264_frame_shift( &h->frames.reference[ref] ) );
2175             h->b_ref_reorder[0] = 1;
2176             ref--;
2177         }
2178     }
2179
2180     /* Prepare room in the dpb for the delayed display time of the later b-frame's */
2181     if( h->param.i_bframe_pyramid )
2182         h->sh.i_mmco_remove_from_end = X264_MAX( ref + 2 - h->frames.i_max_dpb, 0 );
2183 }
2184
2185 static inline void x264_slice_init( x264_t *h, int i_nal_type, int i_global_qp )
2186 {
2187     /* ------------------------ Create slice header  ----------------------- */
2188     if( i_nal_type == NAL_SLICE_IDR )
2189     {
2190         x264_slice_header_init( h, &h->sh, h->sps, h->pps, h->i_idr_pic_id, h->i_frame_num, i_global_qp );
2191
2192         /* alternate id */
2193         h->i_idr_pic_id ^= 1;
2194     }
2195     else
2196     {
2197         x264_slice_header_init( h, &h->sh, h->sps, h->pps, -1, h->i_frame_num, i_global_qp );
2198
2199         h->sh.i_num_ref_idx_l0_active = h->i_ref[0] <= 0 ? 1 : h->i_ref[0];
2200         h->sh.i_num_ref_idx_l1_active = h->i_ref[1] <= 0 ? 1 : h->i_ref[1];
2201         if( h->sh.i_num_ref_idx_l0_active != h->pps->i_num_ref_idx_l0_default_active ||
2202             (h->sh.i_type == SLICE_TYPE_B && h->sh.i_num_ref_idx_l1_active != h->pps->i_num_ref_idx_l1_default_active) )
2203         {
2204             h->sh.b_num_ref_idx_override = 1;
2205         }
2206     }
2207
2208     if( h->fenc->i_type == X264_TYPE_BREF && h->param.b_bluray_compat && h->sh.i_mmco_command_count )
2209     {
2210         h->b_sh_backup = 1;
2211         h->sh_backup = h->sh;
2212     }
2213
2214     h->fdec->i_frame_num = h->sh.i_frame_num;
2215
2216     if( h->sps->i_poc_type == 0 )
2217     {
2218         h->sh.i_poc = h->fdec->i_poc;
2219         if( PARAM_INTERLACED )
2220         {
2221             h->sh.i_delta_poc_bottom = h->param.b_tff ? 1 : -1;
2222             h->sh.i_poc += h->sh.i_delta_poc_bottom == -1;
2223         }
2224         else
2225             h->sh.i_delta_poc_bottom = 0;
2226         h->fdec->i_delta_poc[0] = h->sh.i_delta_poc_bottom == -1;
2227         h->fdec->i_delta_poc[1] = h->sh.i_delta_poc_bottom ==  1;
2228     }
2229     else
2230     {
2231         /* Nothing to do ? */
2232     }
2233
2234     x264_macroblock_slice_init( h );
2235 }
2236
2237 typedef struct
2238 {
2239     int skip;
2240     uint8_t cabac_prevbyte;
2241     bs_t bs;
2242     x264_cabac_t cabac;
2243     x264_frame_stat_t stat;
2244     int last_qp;
2245     int last_dqp;
2246     int field_decoding_flag;
2247 } x264_bs_bak_t;
2248
2249 static ALWAYS_INLINE void x264_bitstream_backup( x264_t *h, x264_bs_bak_t *bak, int i_skip, int full )
2250 {
2251     if( full )
2252     {
2253         bak->stat = h->stat.frame;
2254         bak->last_qp = h->mb.i_last_qp;
2255         bak->last_dqp = h->mb.i_last_dqp;
2256         bak->field_decoding_flag = h->mb.field_decoding_flag;
2257     }
2258     else
2259     {
2260         bak->stat.i_mv_bits = h->stat.frame.i_mv_bits;
2261         bak->stat.i_tex_bits = h->stat.frame.i_tex_bits;
2262     }
2263     /* In the per-MB backup, we don't need the contexts because flushing the CABAC
2264      * encoder has no context dependency and in this case, a slice is ended (and
2265      * thus the content of all contexts are thrown away). */
2266     if( h->param.b_cabac )
2267     {
2268         if( full )
2269             memcpy( &bak->cabac, &h->cabac, sizeof(x264_cabac_t) );
2270         else
2271             memcpy( &bak->cabac, &h->cabac, offsetof(x264_cabac_t, f8_bits_encoded) );
2272         /* x264's CABAC writer modifies the previous byte during carry, so it has to be
2273          * backed up. */
2274         bak->cabac_prevbyte = h->cabac.p[-1];
2275     }
2276     else
2277     {
2278         bak->bs = h->out.bs;
2279         bak->skip = i_skip;
2280     }
2281 }
2282
2283 static ALWAYS_INLINE void x264_bitstream_restore( x264_t *h, x264_bs_bak_t *bak, int *skip, int full )
2284 {
2285     if( full )
2286     {
2287         h->stat.frame = bak->stat;
2288         h->mb.i_last_qp = bak->last_qp;
2289         h->mb.i_last_dqp = bak->last_dqp;
2290         h->mb.field_decoding_flag = bak->field_decoding_flag;
2291     }
2292     else
2293     {
2294         h->stat.frame.i_mv_bits = bak->stat.i_mv_bits;
2295         h->stat.frame.i_tex_bits = bak->stat.i_tex_bits;
2296     }
2297     if( h->param.b_cabac )
2298     {
2299         if( full )
2300             memcpy( &h->cabac, &bak->cabac, sizeof(x264_cabac_t) );
2301         else
2302             memcpy( &h->cabac, &bak->cabac, offsetof(x264_cabac_t, f8_bits_encoded) );
2303         h->cabac.p[-1] = bak->cabac_prevbyte;
2304     }
2305     else
2306     {
2307         h->out.bs = bak->bs;
2308         *skip = bak->skip;
2309     }
2310 }
2311
2312 static int x264_slice_write( x264_t *h )
2313 {
2314     int i_skip;
2315     int mb_xy, i_mb_x, i_mb_y;
2316     /* NALUs other than the first use a 3-byte startcode.
2317      * Add one extra byte for the rbsp, and one more for the final CABAC putbyte.
2318      * Then add an extra 5 bytes just in case, to account for random NAL escapes and
2319      * other inaccuracies. */
2320     int overhead_guess = (NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal)) + 1 + h->param.b_cabac + 5;
2321     int slice_max_size = h->param.i_slice_max_size > 0 ? (h->param.i_slice_max_size-overhead_guess)*8 : 0;
2322     int back_up_bitstream = slice_max_size || (!h->param.b_cabac && h->sps->i_profile_idc < PROFILE_HIGH);
2323     int starting_bits = bs_pos(&h->out.bs);
2324     int b_deblock = h->sh.i_disable_deblocking_filter_idc != 1;
2325     int b_hpel = h->fdec->b_kept_as_ref;
2326     int orig_last_mb = h->sh.i_last_mb;
2327     int thread_last_mb = h->i_threadslice_end * h->mb.i_mb_width - 1;
2328     uint8_t *last_emu_check;
2329 #define BS_BAK_SLICE_MAX_SIZE 0
2330 #define BS_BAK_SLICE_MIN_MBS  1
2331 #define BS_BAK_ROW_VBV        2
2332     x264_bs_bak_t bs_bak[3];
2333     b_deblock &= b_hpel || h->param.b_full_recon || h->param.psz_dump_yuv;
2334     bs_realign( &h->out.bs );
2335
2336     /* Slice */
2337     x264_nal_start( h, h->i_nal_type, h->i_nal_ref_idc );
2338     h->out.nal[h->out.i_nal].i_first_mb = h->sh.i_first_mb;
2339
2340     /* Slice header */
2341     x264_macroblock_thread_init( h );
2342
2343     /* Set the QP equal to the first QP in the slice for more accurate CABAC initialization. */
2344     h->mb.i_mb_xy = h->sh.i_first_mb;
2345     h->sh.i_qp = x264_ratecontrol_mb_qp( h );
2346     h->sh.i_qp = SPEC_QP( h->sh.i_qp );
2347     h->sh.i_qp_delta = h->sh.i_qp - h->pps->i_pic_init_qp;
2348
2349     x264_slice_header_write( &h->out.bs, &h->sh, h->i_nal_ref_idc );
2350     if( h->param.b_cabac )
2351     {
2352         /* alignment needed */
2353         bs_align_1( &h->out.bs );
2354
2355         /* init cabac */
2356         x264_cabac_context_init( h, &h->cabac, h->sh.i_type, x264_clip3( h->sh.i_qp-QP_BD_OFFSET, 0, 51 ), h->sh.i_cabac_init_idc );
2357         x264_cabac_encode_init ( &h->cabac, h->out.bs.p, h->out.bs.p_end );
2358         last_emu_check = h->cabac.p;
2359     }
2360     else
2361         last_emu_check = h->out.bs.p;
2362     h->mb.i_last_qp = h->sh.i_qp;
2363     h->mb.i_last_dqp = 0;
2364     h->mb.field_decoding_flag = 0;
2365
2366     i_mb_y = h->sh.i_first_mb / h->mb.i_mb_width;
2367     i_mb_x = h->sh.i_first_mb % h->mb.i_mb_width;
2368     i_skip = 0;
2369
2370     while( 1 )
2371     {
2372         mb_xy = i_mb_x + i_mb_y * h->mb.i_mb_width;
2373         int mb_spos = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
2374
2375         if( i_mb_x == 0 )
2376         {
2377             if( x264_bitstream_check_buffer( h ) )
2378                 return -1;
2379             if( !(i_mb_y & SLICE_MBAFF) && h->param.rc.i_vbv_buffer_size )
2380                 x264_bitstream_backup( h, &bs_bak[BS_BAK_ROW_VBV], i_skip, 1 );
2381             if( !h->mb.b_reencode_mb )
2382                 x264_fdec_filter_row( h, i_mb_y, 0 );
2383         }
2384
2385         if( !(i_mb_y & SLICE_MBAFF) && back_up_bitstream )
2386         {
2387             x264_bitstream_backup( h, &bs_bak[BS_BAK_SLICE_MAX_SIZE], i_skip, 0 );
2388             if( slice_max_size && (thread_last_mb+1-mb_xy) == h->param.i_slice_min_mbs )
2389                 x264_bitstream_backup( h, &bs_bak[BS_BAK_SLICE_MIN_MBS], i_skip, 0 );
2390         }
2391
2392         if( PARAM_INTERLACED )
2393         {
2394             if( h->mb.b_adaptive_mbaff )
2395             {
2396                 if( !(i_mb_y&1) )
2397                 {
2398                     /* FIXME: VSAD is fast but fairly poor at choosing the best interlace type. */
2399                     h->mb.b_interlaced = x264_field_vsad( h, i_mb_x, i_mb_y );
2400                     memcpy( &h->zigzagf, MB_INTERLACED ? &h->zigzagf_interlaced : &h->zigzagf_progressive, sizeof(h->zigzagf) );
2401                     if( !MB_INTERLACED && (i_mb_y+2) == h->mb.i_mb_height )
2402                         x264_expand_border_mbpair( h, i_mb_x, i_mb_y );
2403                 }
2404             }
2405             h->mb.field[mb_xy] = MB_INTERLACED;
2406         }
2407
2408         /* load cache */
2409         if( SLICE_MBAFF )
2410             x264_macroblock_cache_load_interlaced( h, i_mb_x, i_mb_y );
2411         else
2412             x264_macroblock_cache_load_progressive( h, i_mb_x, i_mb_y );
2413
2414         x264_macroblock_analyse( h );
2415
2416         /* encode this macroblock -> be careful it can change the mb type to P_SKIP if needed */
2417 reencode:
2418         x264_macroblock_encode( h );
2419
2420         if( h->param.b_cabac )
2421         {
2422             if( mb_xy > h->sh.i_first_mb && !(SLICE_MBAFF && (i_mb_y&1)) )
2423                 x264_cabac_encode_terminal( &h->cabac );
2424
2425             if( IS_SKIP( h->mb.i_type ) )
2426                 x264_cabac_mb_skip( h, 1 );
2427             else
2428             {
2429                 if( h->sh.i_type != SLICE_TYPE_I )
2430                     x264_cabac_mb_skip( h, 0 );
2431                 x264_macroblock_write_cabac( h, &h->cabac );
2432             }
2433         }
2434         else
2435         {
2436             if( IS_SKIP( h->mb.i_type ) )
2437                 i_skip++;
2438             else
2439             {
2440                 if( h->sh.i_type != SLICE_TYPE_I )
2441                 {
2442                     bs_write_ue( &h->out.bs, i_skip );  /* skip run */
2443                     i_skip = 0;
2444                 }
2445                 x264_macroblock_write_cavlc( h );
2446                 /* If there was a CAVLC level code overflow, try again at a higher QP. */
2447                 if( h->mb.b_overflow )
2448                 {
2449                     h->mb.i_chroma_qp = h->chroma_qp_table[++h->mb.i_qp];
2450                     h->mb.i_skip_intra = 0;
2451                     h->mb.b_skip_mc = 0;
2452                     h->mb.b_overflow = 0;
2453                     x264_bitstream_restore( h, &bs_bak[BS_BAK_SLICE_MAX_SIZE], &i_skip, 0 );
2454                     goto reencode;
2455                 }
2456             }
2457         }
2458
2459         int total_bits = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
2460         int mb_size = total_bits - mb_spos;
2461
2462         if( slice_max_size && (!SLICE_MBAFF || (i_mb_y&1)) )
2463         {
2464             /* Count the skip run, just in case. */
2465             if( !h->param.b_cabac )
2466                 total_bits += bs_size_ue_big( i_skip );
2467             /* Check for escape bytes. */
2468             uint8_t *end = h->param.b_cabac ? h->cabac.p : h->out.bs.p;
2469             for( ; last_emu_check < end - 2; last_emu_check++ )
2470                 if( last_emu_check[0] == 0 && last_emu_check[1] == 0 && last_emu_check[2] <= 3 )
2471                 {
2472                     slice_max_size -= 8;
2473                     last_emu_check++;
2474                 }
2475             /* We'll just re-encode this last macroblock if we go over the max slice size. */
2476             if( total_bits - starting_bits > slice_max_size && !h->mb.b_reencode_mb )
2477             {
2478                 if( !x264_frame_new_slice( h, h->fdec ) )
2479                 {
2480                     /* Handle the most obnoxious slice-min-mbs edge case: we need to end the slice
2481                      * because it's gone over the maximum size, but doing so would violate slice-min-mbs.
2482                      * If possible, roll back to the last checkpoint and try again.
2483                      * We could try raising QP, but that would break in the case where a slice spans multiple
2484                      * rows, which the re-encoding infrastructure can't currently handle. */
2485                     if( mb_xy <= thread_last_mb && (thread_last_mb+1-mb_xy) < h->param.i_slice_min_mbs )
2486                     {
2487                         if( thread_last_mb-h->param.i_slice_min_mbs < h->sh.i_first_mb+h->param.i_slice_min_mbs )
2488                         {
2489                             x264_log( h, X264_LOG_WARNING, "slice-max-size violated (frame %d, cause: slice-min-mbs)\n", h->i_frame );
2490                             slice_max_size = 0;
2491                             goto cont;
2492                         }
2493                         x264_bitstream_restore( h, &bs_bak[BS_BAK_SLICE_MIN_MBS], &i_skip, 0 );
2494                         h->mb.b_reencode_mb = 1;
2495                         h->sh.i_last_mb = thread_last_mb-h->param.i_slice_min_mbs;
2496                         break;
2497                     }
2498                     if( mb_xy-SLICE_MBAFF*h->mb.i_mb_stride != h->sh.i_first_mb )
2499                     {
2500                         x264_bitstream_restore( h, &bs_bak[BS_BAK_SLICE_MAX_SIZE], &i_skip, 0 );
2501                         h->mb.b_reencode_mb = 1;
2502                         if( SLICE_MBAFF )
2503                         {
2504                             // set to bottom of previous mbpair
2505                             if( i_mb_x )
2506                                 h->sh.i_last_mb = mb_xy-1+h->mb.i_mb_stride*(!(i_mb_y&1));
2507                             else
2508                                 h->sh.i_last_mb = (i_mb_y-2+!(i_mb_y&1))*h->mb.i_mb_stride + h->mb.i_mb_width - 1;
2509                         }
2510                         else
2511                             h->sh.i_last_mb = mb_xy-1;
2512                         break;
2513                     }
2514                     else
2515                         h->sh.i_last_mb = mb_xy;
2516                 }
2517                 else
2518                     slice_max_size = 0;
2519             }
2520         }
2521 cont:
2522         h->mb.b_reencode_mb = 0;
2523
2524 #if HAVE_VISUALIZE
2525         if( h->param.b_visualize )
2526             x264_visualize_mb( h );
2527 #endif
2528
2529         /* save cache */
2530         x264_macroblock_cache_save( h );
2531
2532         if( x264_ratecontrol_mb( h, mb_size ) < 0 )
2533         {
2534             x264_bitstream_restore( h, &bs_bak[BS_BAK_ROW_VBV], &i_skip, 1 );
2535             h->mb.b_reencode_mb = 1;
2536             i_mb_x = 0;
2537             i_mb_y = i_mb_y - SLICE_MBAFF;
2538             h->mb.i_mb_prev_xy = i_mb_y * h->mb.i_mb_stride - 1;
2539             h->sh.i_last_mb = orig_last_mb;
2540             continue;
2541         }
2542
2543         /* accumulate mb stats */
2544         h->stat.frame.i_mb_count[h->mb.i_type]++;
2545
2546         int b_intra = IS_INTRA( h->mb.i_type );
2547         int b_skip = IS_SKIP( h->mb.i_type );
2548         if( h->param.i_log_level >= X264_LOG_INFO || h->param.rc.b_stat_write )
2549         {
2550             if( !b_intra && !b_skip && !IS_DIRECT( h->mb.i_type ) )
2551             {
2552                 if( h->mb.i_partition != D_8x8 )
2553                         h->stat.frame.i_mb_partition[h->mb.i_partition] += 4;
2554                     else
2555                         for( int i = 0; i < 4; i++ )
2556                             h->stat.frame.i_mb_partition[h->mb.i_sub_partition[i]] ++;
2557                 if( h->param.i_frame_reference > 1 )
2558                     for( int i_list = 0; i_list <= (h->sh.i_type == SLICE_TYPE_B); i_list++ )
2559                         for( int i = 0; i < 4; i++ )
2560                         {
2561                             int i_ref = h->mb.cache.ref[i_list][ x264_scan8[4*i] ];
2562                             if( i_ref >= 0 )
2563                                 h->stat.frame.i_mb_count_ref[i_list][i_ref] ++;
2564                         }
2565             }
2566         }
2567
2568         if( h->param.i_log_level >= X264_LOG_INFO )
2569         {
2570             if( h->mb.i_cbp_luma | h->mb.i_cbp_chroma )
2571             {
2572                 if( CHROMA444 )
2573                 {
2574                     for( int i = 0; i < 4; i++ )
2575                         if( h->mb.i_cbp_luma & (1 << i) )
2576                             for( int p = 0; p < 3; p++ )
2577                             {
2578                                 int s8 = i*4+p*16;
2579                                 int nnz8x8 = M16( &h->mb.cache.non_zero_count[x264_scan8[s8]+0] )
2580                                            | M16( &h->mb.cache.non_zero_count[x264_scan8[s8]+8] );
2581                                 h->stat.frame.i_mb_cbp[!b_intra + p*2] += !!nnz8x8;
2582                             }
2583                 }
2584                 else
2585                 {
2586                     int cbpsum = (h->mb.i_cbp_luma&1) + ((h->mb.i_cbp_luma>>1)&1)
2587                                + ((h->mb.i_cbp_luma>>2)&1) + (h->mb.i_cbp_luma>>3);
2588                     h->stat.frame.i_mb_cbp[!b_intra + 0] += cbpsum;
2589                     h->stat.frame.i_mb_cbp[!b_intra + 2] += !!h->mb.i_cbp_chroma;
2590                     h->stat.frame.i_mb_cbp[!b_intra + 4] += h->mb.i_cbp_chroma >> 1;
2591                 }
2592             }
2593             if( h->mb.i_cbp_luma && !b_intra )
2594             {
2595                 h->stat.frame.i_mb_count_8x8dct[0] ++;
2596                 h->stat.frame.i_mb_count_8x8dct[1] += h->mb.b_transform_8x8;
2597             }
2598             if( b_intra && h->mb.i_type != I_PCM )
2599             {
2600                 if( h->mb.i_type == I_16x16 )
2601                     h->stat.frame.i_mb_pred_mode[0][h->mb.i_intra16x16_pred_mode]++;
2602                 else if( h->mb.i_type == I_8x8 )
2603                     for( int i = 0; i < 16; i += 4 )
2604                         h->stat.frame.i_mb_pred_mode[1][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
2605                 else //if( h->mb.i_type == I_4x4 )
2606                     for( int i = 0; i < 16; i++ )
2607                         h->stat.frame.i_mb_pred_mode[2][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
2608                 h->stat.frame.i_mb_pred_mode[3][x264_mb_chroma_pred_mode_fix[h->mb.i_chroma_pred_mode]]++;
2609             }
2610             h->stat.frame.i_mb_field[b_intra?0:b_skip?2:1] += MB_INTERLACED;
2611         }
2612
2613         /* calculate deblock strength values (actual deblocking is done per-row along with hpel) */
2614         if( b_deblock )
2615             x264_macroblock_deblock_strength( h );
2616
2617         if( mb_xy == h->sh.i_last_mb )
2618             break;
2619
2620         if( SLICE_MBAFF )
2621         {
2622             i_mb_x += i_mb_y & 1;
2623             i_mb_y ^= i_mb_x < h->mb.i_mb_width;
2624         }
2625         else
2626             i_mb_x++;
2627         if( i_mb_x == h->mb.i_mb_width )
2628         {
2629             i_mb_y++;
2630             i_mb_x = 0;
2631         }
2632     }
2633     if( h->sh.i_last_mb < h->sh.i_first_mb )
2634         return 0;
2635
2636     h->out.nal[h->out.i_nal].i_last_mb = h->sh.i_last_mb;
2637
2638     if( h->param.b_cabac )
2639     {
2640         x264_cabac_encode_flush( h, &h->cabac );
2641         h->out.bs.p = h->cabac.p;
2642     }
2643     else
2644     {
2645         if( i_skip > 0 )
2646             bs_write_ue( &h->out.bs, i_skip );  /* last skip run */
2647         /* rbsp_slice_trailing_bits */
2648         bs_rbsp_trailing( &h->out.bs );
2649         bs_flush( &h->out.bs );
2650     }
2651     if( x264_nal_end( h ) )
2652         return -1;
2653
2654     if( h->sh.i_last_mb == (h->i_threadslice_end * h->mb.i_mb_width - 1) )
2655     {
2656         h->stat.frame.i_misc_bits = bs_pos( &h->out.bs )
2657                                   + (h->out.i_nal*NALU_OVERHEAD * 8)
2658                                   - h->stat.frame.i_tex_bits
2659                                   - h->stat.frame.i_mv_bits;
2660         x264_fdec_filter_row( h, h->i_threadslice_end, 0 );
2661
2662         if( h->param.b_sliced_threads )
2663         {
2664             /* Tell the main thread we're done. */
2665             x264_threadslice_cond_broadcast( h, 1 );
2666             /* Do hpel now */
2667             for( int mb_y = h->i_threadslice_start; mb_y <= h->i_threadslice_end; mb_y++ )
2668                 x264_fdec_filter_row( h, mb_y, 1 );
2669             x264_threadslice_cond_broadcast( h, 2 );
2670             /* Do the first row of hpel, now that the previous slice is done */
2671             if( h->i_thread_idx > 0 )
2672             {
2673                 x264_threadslice_cond_wait( h->thread[h->i_thread_idx-1], 2 );
2674                 x264_fdec_filter_row( h, h->i_threadslice_start + (1 << SLICE_MBAFF), 2 );
2675             }
2676         }
2677
2678         /* Free mb info after the last thread's done using it */
2679         if( h->fdec->mb_info_free && (!h->param.b_sliced_threads || h->i_thread_idx == (h->param.i_threads-1)) )
2680         {
2681             h->fdec->mb_info_free( h->fdec->mb_info );
2682             h->fdec->mb_info = NULL;
2683             h->fdec->mb_info_free = NULL;
2684         }
2685     }
2686
2687     return 0;
2688 }
2689
2690 static void x264_thread_sync_context( x264_t *dst, x264_t *src )
2691 {
2692     if( dst == src )
2693         return;
2694
2695     // reference counting
2696     for( x264_frame_t **f = src->frames.reference; *f; f++ )
2697         (*f)->i_reference_count++;
2698     for( x264_frame_t **f = dst->frames.reference; *f; f++ )
2699         x264_frame_push_unused( src, *f );
2700     src->fdec->i_reference_count++;
2701     x264_frame_push_unused( src, dst->fdec );
2702
2703     // copy everything except the per-thread pointers and the constants.
2704     memcpy( &dst->i_frame, &src->i_frame, offsetof(x264_t, mb.type) - offsetof(x264_t, i_frame) );
2705     dst->param = src->param;
2706     dst->stat = src->stat;
2707     dst->pixf = src->pixf;
2708 }
2709
2710 static void x264_thread_sync_stat( x264_t *dst, x264_t *src )
2711 {
2712     if( dst == src )
2713         return;
2714     memcpy( &dst->stat.i_frame_count, &src->stat.i_frame_count, sizeof(dst->stat) - sizeof(dst->stat.frame) );
2715 }
2716
2717 static void *x264_slices_write( x264_t *h )
2718 {
2719     int i_slice_num = 0;
2720     int last_thread_mb = h->sh.i_last_mb;
2721
2722 #if HAVE_VISUALIZE
2723     if( h->param.b_visualize )
2724         if( x264_visualize_init( h ) )
2725             goto fail;
2726 #endif
2727
2728     /* init stats */
2729     memset( &h->stat.frame, 0, sizeof(h->stat.frame) );
2730     h->mb.b_reencode_mb = 0;
2731     while( h->sh.i_first_mb + SLICE_MBAFF*h->mb.i_mb_stride <= last_thread_mb )
2732     {
2733         h->sh.i_last_mb = last_thread_mb;
2734         if( !i_slice_num || !x264_frame_new_slice( h, h->fdec ) )
2735         {
2736             if( h->param.i_slice_max_mbs )
2737             {
2738                 if( SLICE_MBAFF )
2739                 {
2740                     // convert first to mbaff form, add slice-max-mbs, then convert back to normal form
2741                     int last_mbaff = 2*(h->sh.i_first_mb % h->mb.i_mb_width)
2742                         + h->mb.i_mb_width*(h->sh.i_first_mb / h->mb.i_mb_width)
2743                         + h->param.i_slice_max_mbs - 1;
2744                     int last_x = (last_mbaff % (2*h->mb.i_mb_width))/2;
2745                     int last_y = (last_mbaff / (2*h->mb.i_mb_width))*2 + 1;
2746                     h->sh.i_last_mb = last_x + h->mb.i_mb_stride*last_y;
2747                 }
2748                 else
2749                 {
2750                     h->sh.i_last_mb = h->sh.i_first_mb + h->param.i_slice_max_mbs - 1;
2751                     if( h->sh.i_last_mb < last_thread_mb && last_thread_mb - h->sh.i_last_mb < h->param.i_slice_min_mbs )
2752                         h->sh.i_last_mb = last_thread_mb - h->param.i_slice_min_mbs;
2753                 }
2754                 i_slice_num++;
2755             }
2756             else if( h->param.i_slice_count && !h->param.b_sliced_threads )
2757             {
2758                 int height = h->mb.i_mb_height >> PARAM_INTERLACED;
2759                 int width = h->mb.i_mb_width << PARAM_INTERLACED;
2760                 i_slice_num++;
2761                 h->sh.i_last_mb = (height * i_slice_num + h->param.i_slice_count/2) / h->param.i_slice_count * width - 1;
2762             }
2763         }
2764         h->sh.i_last_mb = X264_MIN( h->sh.i_last_mb, last_thread_mb );
2765         if( x264_stack_align( x264_slice_write, h ) )
2766             goto fail;
2767         h->sh.i_first_mb = h->sh.i_last_mb + 1;
2768         // if i_first_mb is not the last mb in a row then go to the next mb in MBAFF order
2769         if( SLICE_MBAFF && h->sh.i_first_mb % h->mb.i_mb_width )
2770             h->sh.i_first_mb -= h->mb.i_mb_stride;
2771     }
2772
2773 #if HAVE_VISUALIZE
2774     if( h->param.b_visualize )
2775     {
2776         x264_visualize_show( h );
2777         x264_visualize_close( h );
2778     }
2779 #endif
2780
2781     return (void *)0;
2782
2783 fail:
2784     /* Tell other threads we're done, so they wouldn't wait for it */
2785     if( h->param.b_sliced_threads )
2786         x264_threadslice_cond_broadcast( h, 2 );
2787     return (void *)-1;
2788 }
2789
2790 static int x264_threaded_slices_write( x264_t *h )
2791 {
2792     /* set first/last mb and sync contexts */
2793     for( int i = 0; i < h->param.i_threads; i++ )
2794     {
2795         x264_t *t = h->thread[i];
2796         if( i )
2797         {
2798             t->param = h->param;
2799             memcpy( &t->i_frame, &h->i_frame, offsetof(x264_t, rc) - offsetof(x264_t, i_frame) );
2800         }
2801         int height = h->mb.i_mb_height >> PARAM_INTERLACED;
2802         t->i_threadslice_start = ((height *  i    + h->param.i_slice_count/2) / h->param.i_threads) << PARAM_INTERLACED;
2803         t->i_threadslice_end   = ((height * (i+1) + h->param.i_slice_count/2) / h->param.i_threads) << PARAM_INTERLACED;
2804         t->sh.i_first_mb = t->i_threadslice_start * h->mb.i_mb_width;
2805         t->sh.i_last_mb  =   t->i_threadslice_end * h->mb.i_mb_width - 1;
2806     }
2807
2808     x264_stack_align( x264_analyse_weight_frame, h, h->mb.i_mb_height*16 + 16 );
2809
2810     x264_threads_distribute_ratecontrol( h );
2811
2812     /* setup */
2813     for( int i = 0; i < h->param.i_threads; i++ )
2814     {
2815         h->thread[i]->i_thread_idx = i;
2816         h->thread[i]->b_thread_active = 1;
2817         x264_threadslice_cond_broadcast( h->thread[i], 0 );
2818     }
2819     /* dispatch */
2820     for( int i = 0; i < h->param.i_threads; i++ )
2821         x264_threadpool_run( h->threadpool, (void*)x264_slices_write, h->thread[i] );
2822     /* wait */
2823     for( int i = 0; i < h->param.i_threads; i++ )
2824         x264_threadslice_cond_wait( h->thread[i], 1 );
2825
2826     x264_threads_merge_ratecontrol( h );
2827
2828     for( int i = 1; i < h->param.i_threads; i++ )
2829     {
2830         x264_t *t = h->thread[i];
2831         for( int j = 0; j < t->out.i_nal; j++ )
2832         {
2833             h->out.nal[h->out.i_nal] = t->out.nal[j];
2834             h->out.i_nal++;
2835             x264_nal_check_buffer( h );
2836         }
2837         /* All entries in stat.frame are ints except for ssd/ssim. */
2838         for( int j = 0; j < (offsetof(x264_t,stat.frame.i_ssd) - offsetof(x264_t,stat.frame.i_mv_bits)) / sizeof(int); j++ )
2839             ((int*)&h->stat.frame)[j] += ((int*)&t->stat.frame)[j];
2840         for( int j = 0; j < 3; j++ )
2841             h->stat.frame.i_ssd[j] += t->stat.frame.i_ssd[j];
2842         h->stat.frame.f_ssim += t->stat.frame.f_ssim;
2843         h->stat.frame.i_ssim_cnt += t->stat.frame.i_ssim_cnt;
2844     }
2845
2846     return 0;
2847 }
2848
2849 void x264_encoder_intra_refresh( x264_t *h )
2850 {
2851     h = h->thread[h->i_thread_phase];
2852     h->b_queued_intra_refresh = 1;
2853 }
2854
2855 int x264_encoder_invalidate_reference( x264_t *h, int64_t pts )
2856 {
2857     if( h->param.i_bframe )
2858     {
2859         x264_log( h, X264_LOG_ERROR, "x264_encoder_invalidate_reference is not supported with B-frames enabled\n" );
2860         return -1;
2861     }
2862     if( h->param.b_intra_refresh )
2863     {
2864         x264_log( h, X264_LOG_ERROR, "x264_encoder_invalidate_reference is not supported with intra refresh enabled\n" );
2865         return -1;
2866     }
2867     h = h->thread[h->i_thread_phase];
2868     if( pts >= h->i_last_idr_pts )
2869     {
2870         for( int i = 0; h->frames.reference[i]; i++ )
2871             if( pts <= h->frames.reference[i]->i_pts )
2872                 h->frames.reference[i]->b_corrupt = 1;
2873         if( pts <= h->fdec->i_pts )
2874             h->fdec->b_corrupt = 1;
2875     }
2876     return 0;
2877 }
2878
2879 /****************************************************************************
2880  * x264_encoder_encode:
2881  *  XXX: i_poc   : is the poc of the current given picture
2882  *       i_frame : is the number of the frame being coded
2883  *  ex:  type frame poc
2884  *       I      0   2*0
2885  *       P      1   2*3
2886  *       B      2   2*1
2887  *       B      3   2*2
2888  *       P      4   2*6
2889  *       B      5   2*4
2890  *       B      6   2*5
2891  ****************************************************************************/
2892 int     x264_encoder_encode( x264_t *h,
2893                              x264_nal_t **pp_nal, int *pi_nal,
2894                              x264_picture_t *pic_in,
2895                              x264_picture_t *pic_out )
2896 {
2897     x264_t *thread_current, *thread_prev, *thread_oldest;
2898     int i_nal_type, i_nal_ref_idc, i_global_qp;
2899     int overhead = NALU_OVERHEAD;
2900
2901 #if HAVE_OPENCL
2902     if( h->opencl.b_fatal_error )
2903         return -1;
2904 #endif
2905
2906     if( h->i_thread_frames > 1 )
2907     {
2908         thread_prev    = h->thread[ h->i_thread_phase ];
2909         h->i_thread_phase = (h->i_thread_phase + 1) % h->i_thread_frames;
2910         thread_current = h->thread[ h->i_thread_phase ];
2911         thread_oldest  = h->thread[ (h->i_thread_phase + 1) % h->i_thread_frames ];
2912         x264_thread_sync_context( thread_current, thread_prev );
2913         x264_thread_sync_ratecontrol( thread_current, thread_prev, thread_oldest );
2914         h = thread_current;
2915     }
2916     else
2917     {
2918         thread_current =
2919         thread_oldest  = h;
2920     }
2921 #if HAVE_MMX
2922     if( h->param.cpu&X264_CPU_SSE_MISALIGN )
2923         x264_cpu_mask_misalign_sse();
2924 #endif
2925     h->i_cpb_delay_pir_offset = h->i_cpb_delay_pir_offset_next;
2926
2927     /* no data out */
2928     *pi_nal = 0;
2929     *pp_nal = NULL;
2930
2931     /* ------------------- Setup new frame from picture -------------------- */
2932     if( pic_in != NULL )
2933     {
2934         /* 1: Copy the picture to a frame and move it to a buffer */
2935         x264_frame_t *fenc = x264_frame_pop_unused( h, 0 );
2936         if( !fenc )
2937             return -1;
2938
2939         if( x264_frame_copy_picture( h, fenc, pic_in ) < 0 )
2940             return -1;
2941
2942         if( h->param.i_width != 16 * h->mb.i_mb_width ||
2943             h->param.i_height != 16 * h->mb.i_mb_height )
2944             x264_frame_expand_border_mod16( h, fenc );
2945
2946         fenc->i_frame = h->frames.i_input++;
2947
2948         if( fenc->i_frame == 0 )
2949             h->frames.i_first_pts = fenc->i_pts;
2950         if( h->frames.i_bframe_delay && fenc->i_frame == h->frames.i_bframe_delay )
2951             h->frames.i_bframe_delay_time = fenc->i_pts - h->frames.i_first_pts;
2952
2953         if( h->param.b_vfr_input && fenc->i_pts <= h->frames.i_largest_pts )
2954             x264_log( h, X264_LOG_WARNING, "non-strictly-monotonic PTS\n" );
2955
2956         h->frames.i_second_largest_pts = h->frames.i_largest_pts;
2957         h->frames.i_largest_pts = fenc->i_pts;
2958
2959         if( (fenc->i_pic_struct < PIC_STRUCT_AUTO) || (fenc->i_pic_struct > PIC_STRUCT_TRIPLE) )
2960             fenc->i_pic_struct = PIC_STRUCT_AUTO;
2961
2962         if( fenc->i_pic_struct == PIC_STRUCT_AUTO )
2963         {
2964 #if HAVE_INTERLACED
2965             int b_interlaced = fenc->param ? fenc->param->b_interlaced : h->param.b_interlaced;
2966 #else
2967             int b_interlaced = 0;
2968 #endif
2969             if( b_interlaced )
2970             {
2971                 int b_tff = fenc->param ? fenc->param->b_tff : h->param.b_tff;
2972                 fenc->i_pic_struct = b_tff ? PIC_STRUCT_TOP_BOTTOM : PIC_STRUCT_BOTTOM_TOP;
2973             }
2974             else
2975                 fenc->i_pic_struct = PIC_STRUCT_PROGRESSIVE;
2976         }
2977
2978         if( h->param.rc.b_mb_tree && h->param.rc.b_stat_read )
2979         {
2980             if( x264_macroblock_tree_read( h, fenc, pic_in->prop.quant_offsets ) )
2981                 return -1;
2982         }
2983         else
2984             x264_stack_align( x264_adaptive_quant_frame, h, fenc, pic_in->prop.quant_offsets );
2985
2986         if( pic_in->prop.quant_offsets_free )
2987             pic_in->prop.quant_offsets_free( pic_in->prop.quant_offsets );
2988
2989         if( h->frames.b_have_lowres )
2990             x264_frame_init_lowres( h, fenc );
2991
2992         /* 2: Place the frame into the queue for its slice type decision */
2993         x264_lookahead_put_frame( h, fenc );
2994
2995         if( h->frames.i_input <= h->frames.i_delay + 1 - h->i_thread_frames )
2996         {
2997             /* Nothing yet to encode, waiting for filling of buffers */
2998             pic_out->i_type = X264_TYPE_AUTO;
2999             return 0;
3000         }
3001     }
3002     else
3003     {
3004         /* signal kills for lookahead thread */
3005         x264_pthread_mutex_lock( &h->lookahead->ifbuf.mutex );
3006         h->lookahead->b_exit_thread = 1;
3007         x264_pthread_cond_broadcast( &h->lookahead->ifbuf.cv_fill );
3008         x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
3009     }
3010
3011     h->i_frame++;
3012     /* 3: The picture is analyzed in the lookahead */
3013     if( !h->frames.current[0] )
3014         x264_lookahead_get_frames( h );
3015
3016     if( !h->frames.current[0] && x264_lookahead_is_empty( h ) )
3017         return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
3018
3019     /* ------------------- Get frame to be encoded ------------------------- */
3020     /* 4: get picture to encode */
3021     h->fenc = x264_frame_shift( h->frames.current );
3022
3023     /* If applicable, wait for previous frame reconstruction to finish */
3024     if( h->param.b_sliced_threads )
3025         if( x264_threadpool_wait_all( h ) < 0 )
3026             return -1;
3027
3028     if( h->i_frame == h->i_thread_frames - 1 )
3029         h->i_reordered_pts_delay = h->fenc->i_reordered_pts;
3030     if( h->fenc->param )
3031     {
3032         x264_encoder_reconfig( h, h->fenc->param );
3033         if( h->fenc->param->param_free )
3034         {
3035             h->fenc->param->param_free( h->fenc->param );
3036             h->fenc->param = NULL;
3037         }
3038     }
3039
3040     // ok to call this before encoding any frames, since the initial values of fdec have b_kept_as_ref=0
3041     if( x264_reference_update( h ) )
3042         return -1;
3043     h->fdec->i_lines_completed = -1;
3044
3045     if( !IS_X264_TYPE_I( h->fenc->i_type ) )
3046     {
3047         int valid_refs_left = 0;
3048         for( int i = 0; h->frames.reference[i]; i++ )
3049             if( !h->frames.reference[i]->b_corrupt )
3050                 valid_refs_left++;
3051         /* No valid reference frames left: force an IDR. */
3052         if( !valid_refs_left )
3053         {
3054             h->fenc->b_keyframe = 1;
3055             h->fenc->i_type = X264_TYPE_IDR;
3056         }
3057     }
3058
3059     if( h->fenc->b_keyframe )
3060     {
3061         h->frames.i_last_keyframe = h->fenc->i_frame;
3062         if( h->fenc->i_type == X264_TYPE_IDR )
3063         {
3064             h->i_frame_num = 0;
3065             h->frames.i_last_idr = h->fenc->i_frame;
3066         }
3067     }
3068     h->sh.i_mmco_command_count =
3069     h->sh.i_mmco_remove_from_end = 0;
3070     h->b_ref_reorder[0] =
3071     h->b_ref_reorder[1] = 0;
3072     h->fdec->i_poc =
3073     h->fenc->i_poc = 2 * ( h->fenc->i_frame - X264_MAX( h->frames.i_last_idr, 0 ) );
3074
3075     /* ------------------- Setup frame context ----------------------------- */
3076     /* 5: Init data dependent of frame type */
3077     if( h->fenc->i_type == X264_TYPE_IDR )
3078     {
3079         /* reset ref pictures */
3080         i_nal_type    = NAL_SLICE_IDR;
3081         i_nal_ref_idc = NAL_PRIORITY_HIGHEST;
3082         h->sh.i_type = SLICE_TYPE_I;
3083         x264_reference_reset( h );
3084         h->frames.i_poc_last_open_gop = -1;
3085     }
3086     else if( h->fenc->i_type == X264_TYPE_I )
3087     {
3088         i_nal_type    = NAL_SLICE;
3089         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
3090         h->sh.i_type = SLICE_TYPE_I;
3091         x264_reference_hierarchy_reset( h );
3092         if( h->param.b_open_gop )
3093             h->frames.i_poc_last_open_gop = h->fenc->b_keyframe ? h->fenc->i_poc : -1;
3094     }
3095     else if( h->fenc->i_type == X264_TYPE_P )
3096     {
3097         i_nal_type    = NAL_SLICE;
3098         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
3099         h->sh.i_type = SLICE_TYPE_P;
3100         x264_reference_hierarchy_reset( h );
3101         h->frames.i_poc_last_open_gop = -1;
3102     }
3103     else if( h->fenc->i_type == X264_TYPE_BREF )
3104     {
3105         i_nal_type    = NAL_SLICE;
3106         i_nal_ref_idc = h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT ? NAL_PRIORITY_LOW : NAL_PRIORITY_HIGH;
3107         h->sh.i_type = SLICE_TYPE_B;
3108         x264_reference_hierarchy_reset( h );
3109     }
3110     else    /* B frame */
3111     {
3112         i_nal_type    = NAL_SLICE;
3113         i_nal_ref_idc = NAL_PRIORITY_DISPOSABLE;
3114         h->sh.i_type = SLICE_TYPE_B;
3115     }
3116
3117     h->fdec->i_type = h->fenc->i_type;
3118     h->fdec->i_frame = h->fenc->i_frame;
3119     h->fenc->b_kept_as_ref =
3120     h->fdec->b_kept_as_ref = i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE && h->param.i_keyint_max > 1;
3121
3122     h->fdec->mb_info = h->fenc->mb_info;
3123     h->fdec->mb_info_free = h->fenc->mb_info_free;
3124     h->fenc->mb_info = NULL;
3125     h->fenc->mb_info_free = NULL;
3126
3127     h->fdec->i_pts = h->fenc->i_pts;
3128     if( h->frames.i_bframe_delay )
3129     {
3130         int64_t *prev_reordered_pts = thread_current->frames.i_prev_reordered_pts;
3131         h->fdec->i_dts = h->i_frame > h->frames.i_bframe_delay
3132                        ? prev_reordered_pts[ (h->i_frame - h->frames.i_bframe_delay) % h->frames.i_bframe_delay ]
3133                        : h->fenc->i_reordered_pts - h->frames.i_bframe_delay_time;
3134         prev_reordered_pts[ h->i_frame % h->frames.i_bframe_delay ] = h->fenc->i_reordered_pts;
3135     }
3136     else
3137         h->fdec->i_dts = h->fenc->i_reordered_pts;
3138     if( h->fenc->i_type == X264_TYPE_IDR )
3139         h->i_last_idr_pts = h->fdec->i_pts;
3140
3141     /* ------------------- Init                ----------------------------- */
3142     /* build ref list 0/1 */
3143     x264_reference_build_list( h, h->fdec->i_poc );
3144
3145     /* ---------------------- Write the bitstream -------------------------- */
3146     /* Init bitstream context */
3147     if( h->param.b_sliced_threads )
3148     {
3149         for( int i = 0; i < h->param.i_threads; i++ )
3150         {
3151             bs_init( &h->thread[i]->out.bs, h->thread[i]->out.p_bitstream, h->thread[i]->out.i_bitstream );
3152             h->thread[i]->out.i_nal = 0;
3153         }
3154     }
3155     else
3156     {
3157         bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
3158         h->out.i_nal = 0;
3159     }
3160
3161     if( h->param.b_aud )
3162     {
3163         int pic_type;
3164
3165         if( h->sh.i_type == SLICE_TYPE_I )
3166             pic_type = 0;
3167         else if( h->sh.i_type == SLICE_TYPE_P )
3168             pic_type = 1;
3169         else if( h->sh.i_type == SLICE_TYPE_B )
3170             pic_type = 2;
3171         else
3172             pic_type = 7;
3173
3174         x264_nal_start( h, NAL_AUD, NAL_PRIORITY_DISPOSABLE );
3175         bs_write( &h->out.bs, 3, pic_type );
3176         bs_rbsp_trailing( &h->out.bs );
3177         if( x264_nal_end( h ) )
3178             return -1;
3179         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
3180     }
3181
3182     h->i_nal_type = i_nal_type;
3183     h->i_nal_ref_idc = i_nal_ref_idc;
3184
3185     if( h->param.b_intra_refresh )
3186     {
3187         if( IS_X264_TYPE_I( h->fenc->i_type ) )
3188         {
3189             h->fdec->i_frames_since_pir = 0;
3190             h->b_queued_intra_refresh = 0;
3191             /* PIR is currently only supported with ref == 1, so any intra frame effectively refreshes
3192              * the whole frame and counts as an intra refresh. */
3193             h->fdec->f_pir_position = h->mb.i_mb_width;
3194         }
3195         else if( h->fenc->i_type == X264_TYPE_P )
3196         {
3197             int pocdiff = (h->fdec->i_poc - h->fref[0][0]->i_poc)/2;
3198             float increment = X264_MAX( ((float)h->mb.i_mb_width-1) / h->param.i_keyint_max, 1 );
3199             h->fdec->f_pir_position = h->fref[0][0]->f_pir_position;
3200             h->fdec->i_frames_since_pir = h->fref[0][0]->i_frames_since_pir + pocdiff;
3201             if( h->fdec->i_frames_since_pir >= h->param.i_keyint_max ||
3202                 (h->b_queued_intra_refresh && h->fdec->f_pir_position + 0.5 >= h->mb.i_mb_width) )
3203             {
3204                 h->fdec->f_pir_position = 0;
3205                 h->fdec->i_frames_since_pir = 0;
3206                 h->b_queued_intra_refresh = 0;
3207                 h->fenc->b_keyframe = 1;
3208             }
3209             h->fdec->i_pir_start_col = h->fdec->f_pir_position+0.5;
3210             h->fdec->f_pir_position += increment * pocdiff;
3211             h->fdec->i_pir_end_col = h->fdec->f_pir_position+0.5;
3212             /* If our intra refresh has reached the right side of the frame, we're done. */
3213             if( h->fdec->i_pir_end_col >= h->mb.i_mb_width - 1 )
3214             {
3215                 h->fdec->f_pir_position = h->mb.i_mb_width;
3216                 h->fdec->i_pir_end_col = h->mb.i_mb_width - 1;
3217             }
3218         }
3219     }
3220
3221     if( h->fenc->b_keyframe )
3222     {
3223         /* Write SPS and PPS */
3224         if( h->param.b_repeat_headers )
3225         {
3226             /* generate sequence parameters */
3227             x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
3228             x264_sps_write( &h->out.bs, h->sps );
3229             if( x264_nal_end( h ) )
3230                 return -1;
3231             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
3232
3233             /* generate picture parameters */
3234             x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
3235             x264_pps_write( &h->out.bs, h->sps, h->pps );
3236             if( x264_nal_end( h ) )
3237                 return -1;
3238             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
3239         }
3240
3241         /* when frame threading is used, buffering period sei is written in x264_encoder_frame_end */
3242         if( h->i_thread_frames == 1 && h->sps->vui.b_nal_hrd_parameters_present )
3243         {
3244             x264_hrd_fullness( h );
3245             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3246             x264_sei_buffering_period_write( h, &h->out.bs );
3247             if( x264_nal_end( h ) )
3248                return -1;
3249             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
3250         }
3251     }
3252
3253     /* write extra sei */
3254     for( int i = 0; i < h->fenc->extra_sei.num_payloads; i++ )
3255     {
3256         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3257         x264_sei_write( &h->out.bs, h->fenc->extra_sei.payloads[i].payload, h->fenc->extra_sei.payloads[i].payload_size,
3258                         h->fenc->extra_sei.payloads[i].payload_type );
3259         if( x264_nal_end( h ) )
3260             return -1;
3261         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
3262         if( h->fenc->extra_sei.sei_free )
3263         {
3264             h->fenc->extra_sei.sei_free( h->fenc->extra_sei.payloads[i].payload );
3265             h->fenc->extra_sei.payloads[i].payload = NULL;
3266         }
3267     }
3268
3269     if( h->fenc->extra_sei.sei_free )
3270     {
3271         h->fenc->extra_sei.sei_free( h->fenc->extra_sei.payloads );
3272         h->fenc->extra_sei.payloads = NULL;
3273         h->fenc->extra_sei.sei_free = NULL;
3274     }
3275
3276     if( h->fenc->b_keyframe )
3277     {
3278         if( h->param.b_repeat_headers && h->fenc->i_frame == 0 )
3279         {
3280             /* identify ourself */
3281             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3282             if( x264_sei_version_write( h, &h->out.bs ) )
3283                 return -1;
3284             if( x264_nal_end( h ) )
3285                 return -1;
3286             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
3287         }
3288
3289         if( h->fenc->i_type != X264_TYPE_IDR )
3290         {
3291             int time_to_recovery = h->param.b_open_gop ? 0 : X264_MIN( h->mb.i_mb_width - 1, h->param.i_keyint_max ) + h->param.i_bframe - 1;
3292             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3293             x264_sei_recovery_point_write( h, &h->out.bs, time_to_recovery );
3294             if( x264_nal_end( h ) )
3295                 return -1;
3296             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
3297         }
3298
3299         if ( h->param.i_frame_packing >= 0 )
3300         {
3301             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3302             x264_sei_frame_packing_write( h, &h->out.bs );
3303             if( x264_nal_end( h ) )
3304                 return -1;
3305             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
3306         }
3307     }
3308
3309     /* generate sei pic timing */
3310     if( h->sps->vui.b_pic_struct_present || h->sps->vui.b_nal_hrd_parameters_present )
3311     {
3312         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3313         x264_sei_pic_timing_write( h, &h->out.bs );
3314         if( x264_nal_end( h ) )
3315             return -1;
3316         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
3317     }
3318
3319     /* As required by Blu-ray. */
3320     if( !IS_X264_TYPE_B( h->fenc->i_type ) && h->b_sh_backup )
3321     {
3322         h->b_sh_backup = 0;
3323         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3324         x264_sei_dec_ref_pic_marking_write( h, &h->out.bs );
3325         if( x264_nal_end( h ) )
3326             return -1;
3327         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
3328     }
3329
3330     if( h->fenc->b_keyframe && h->param.b_intra_refresh )
3331         h->i_cpb_delay_pir_offset_next = h->fenc->i_cpb_delay;
3332
3333     /* Init the rate control */
3334     /* FIXME: Include slice header bit cost. */
3335     x264_ratecontrol_start( h, h->fenc->i_qpplus1, overhead*8 );
3336     i_global_qp = x264_ratecontrol_qp( h );
3337
3338     pic_out->i_qpplus1 =
3339     h->fdec->i_qpplus1 = i_global_qp + 1;
3340
3341     if( h->param.rc.b_stat_read && h->sh.i_type != SLICE_TYPE_I )
3342     {
3343         x264_reference_build_list_optimal( h );
3344         x264_reference_check_reorder( h );
3345     }
3346
3347     if( h->i_ref[0] )
3348         h->fdec->i_poc_l0ref0 = h->fref[0][0]->i_poc;
3349
3350     /* ------------------------ Create slice header  ----------------------- */
3351     x264_slice_init( h, i_nal_type, i_global_qp );
3352
3353     /*------------------------- Weights -------------------------------------*/
3354     if( h->sh.i_type == SLICE_TYPE_B )
3355         x264_macroblock_bipred_init( h );
3356
3357     x264_weighted_pred_init( h );
3358
3359     if( i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE )
3360         h->i_frame_num++;
3361
3362     /* Write frame */
3363     h->i_threadslice_start = 0;
3364     h->i_threadslice_end = h->mb.i_mb_height;
3365     if( h->i_thread_frames > 1 )
3366     {
3367         x264_threadpool_run( h->threadpool, (void*)x264_slices_write, h );
3368         h->b_thread_active = 1;
3369     }
3370     else if( h->param.b_sliced_threads )
3371     {
3372         if( x264_threaded_slices_write( h ) )
3373             return -1;
3374     }
3375     else
3376         if( (intptr_t)x264_slices_write( h ) )
3377             return -1;
3378
3379     return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
3380 }
3381
3382 static int x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
3383                                    x264_nal_t **pp_nal, int *pi_nal,
3384                                    x264_picture_t *pic_out )
3385 {
3386     char psz_message[80];
3387
3388     if( !h->param.b_sliced_threads && h->b_thread_active )
3389     {
3390         h->b_thread_active = 0;
3391         if( (intptr_t)x264_threadpool_wait( h->threadpool, h ) )
3392             return -1;
3393     }
3394     if( !h->out.i_nal )
3395     {
3396         pic_out->i_type = X264_TYPE_AUTO;
3397         return 0;
3398     }
3399
3400     x264_emms();
3401
3402     /* generate buffering period sei and insert it into place */
3403     if( h->i_thread_frames > 1 && h->fenc->b_keyframe && h->sps->vui.b_nal_hrd_parameters_present )
3404     {
3405         x264_hrd_fullness( h );
3406         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3407         x264_sei_buffering_period_write( h, &h->out.bs );
3408         if( x264_nal_end( h ) )
3409            return -1;
3410         /* buffering period sei must follow AUD, SPS and PPS and precede all other SEIs */
3411         int idx = 0;
3412         while( h->out.nal[idx].i_type == NAL_AUD ||
3413                h->out.nal[idx].i_type == NAL_SPS ||
3414                h->out.nal[idx].i_type == NAL_PPS )
3415             idx++;
3416         x264_nal_t nal_tmp = h->out.nal[h->out.i_nal-1];
3417         memmove( &h->out.nal[idx+1], &h->out.nal[idx], (h->out.i_nal-idx-1)*sizeof(x264_nal_t) );
3418         h->out.nal[idx] = nal_tmp;
3419     }
3420
3421     int frame_size = x264_encoder_encapsulate_nals( h, 0 );
3422     if( frame_size < 0 )
3423         return -1;
3424
3425     /* Set output picture properties */
3426     pic_out->i_type = h->fenc->i_type;
3427
3428     pic_out->b_keyframe = h->fenc->b_keyframe;
3429     pic_out->i_pic_struct = h->fenc->i_pic_struct;
3430
3431     pic_out->i_pts = h->fdec->i_pts;
3432     pic_out->i_dts = h->fdec->i_dts;
3433
3434     if( pic_out->i_pts < pic_out->i_dts )
3435         x264_log( h, X264_LOG_WARNING, "invalid DTS: PTS is less than DTS\n" );
3436
3437     pic_out->opaque = h->fenc->opaque;
3438
3439     pic_out->img.i_csp = h->fdec->i_csp;
3440 #if HIGH_BIT_DEPTH
3441     pic_out->img.i_csp |= X264_CSP_HIGH_DEPTH;
3442 #endif
3443     pic_out->img.i_plane = h->fdec->i_plane;
3444     for( int i = 0; i < pic_out->img.i_plane; i++ )
3445     {
3446         pic_out->img.i_stride[i] = h->fdec->i_stride[i] * sizeof(pixel);
3447         pic_out->img.plane[i] = (uint8_t*)h->fdec->plane[i];
3448     }
3449
3450     x264_frame_push_unused( thread_current, h->fenc );
3451
3452     /* ---------------------- Update encoder state ------------------------- */
3453
3454     /* update rc */
3455     int filler = 0;
3456     if( x264_ratecontrol_end( h, frame_size * 8, &filler ) < 0 )
3457         return -1;
3458
3459     pic_out->hrd_timing = h->fenc->hrd_timing;
3460     pic_out->prop.f_crf_avg = h->fdec->f_crf_avg;
3461
3462     while( filler > 0 )
3463     {
3464         int f, overhead;
3465         overhead = (FILLER_OVERHEAD - h->param.b_annexb);
3466         if( h->param.i_slice_max_size && filler > h->param.i_slice_max_size )
3467         {
3468             int next_size = filler - h->param.i_slice_max_size;
3469             int overflow = X264_MAX( overhead - next_size, 0 );
3470             f = h->param.i_slice_max_size - overhead - overflow;
3471         }
3472         else
3473             f = X264_MAX( 0, filler - overhead );
3474
3475         x264_nal_start( h, NAL_FILLER, NAL_PRIORITY_DISPOSABLE );
3476         x264_filler_write( h, &h->out.bs, f );
3477         if( x264_nal_end( h ) )
3478             return -1;
3479         int total_size = x264_encoder_encapsulate_nals( h, h->out.i_nal-1 );
3480         if( total_size < 0 )
3481             return -1;
3482         frame_size += total_size;
3483         filler -= total_size;
3484     }
3485
3486     /* End bitstream, set output  */
3487     *pi_nal = h->out.i_nal;
3488     *pp_nal = h->out.nal;
3489
3490     h->out.i_nal = 0;
3491
3492     x264_noise_reduction_update( h );
3493
3494     /* ---------------------- Compute/Print statistics --------------------- */
3495     x264_thread_sync_stat( h, h->thread[0] );
3496
3497     /* Slice stat */
3498     h->stat.i_frame_count[h->sh.i_type]++;
3499     h->stat.i_frame_size[h->sh.i_type] += frame_size;
3500     h->stat.f_frame_qp[h->sh.i_type] += h->fdec->f_qp_avg_aq;
3501
3502     for( int i = 0; i < X264_MBTYPE_MAX; i++ )
3503         h->stat.i_mb_count[h->sh.i_type][i] += h->stat.frame.i_mb_count[i];
3504     for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
3505         h->stat.i_mb_partition[h->sh.i_type][i] += h->stat.frame.i_mb_partition[i];
3506     for( int i = 0; i < 2; i++ )
3507         h->stat.i_mb_count_8x8dct[i] += h->stat.frame.i_mb_count_8x8dct[i];
3508     for( int i = 0; i < 6; i++ )
3509         h->stat.i_mb_cbp[i] += h->stat.frame.i_mb_cbp[i];
3510     for( int i = 0; i < 4; i++ )
3511         for( int j = 0; j < 13; j++ )
3512             h->stat.i_mb_pred_mode[i][j] += h->stat.frame.i_mb_pred_mode[i][j];
3513     if( h->sh.i_type != SLICE_TYPE_I )
3514         for( int i_list = 0; i_list < 2; i_list++ )
3515             for( int i = 0; i < X264_REF_MAX*2; i++ )
3516                 h->stat.i_mb_count_ref[h->sh.i_type][i_list][i] += h->stat.frame.i_mb_count_ref[i_list][i];
3517     for( int i = 0; i < 3; i++ )
3518         h->stat.i_mb_field[i] += h->stat.frame.i_mb_field[i];
3519     if( h->sh.i_type == SLICE_TYPE_P && h->param.analyse.i_weighted_pred >= X264_WEIGHTP_SIMPLE )
3520     {
3521         h->stat.i_wpred[0] += !!h->sh.weight[0][0].weightfn;
3522         h->stat.i_wpred[1] += !!h->sh.weight[0][1].weightfn || !!h->sh.weight[0][2].weightfn;
3523     }
3524     if( h->sh.i_type == SLICE_TYPE_B )
3525     {
3526         h->stat.i_direct_frames[ h->sh.b_direct_spatial_mv_pred ] ++;
3527         if( h->mb.b_direct_auto_write )
3528         {
3529             //FIXME somewhat arbitrary time constants
3530             if( h->stat.i_direct_score[0] + h->stat.i_direct_score[1] > h->mb.i_mb_count )
3531                 for( int i = 0; i < 2; i++ )
3532                     h->stat.i_direct_score[i] = h->stat.i_direct_score[i] * 9/10;
3533             for( int i = 0; i < 2; i++ )
3534                 h->stat.i_direct_score[i] += h->stat.frame.i_direct_score[i];
3535         }
3536     }
3537     else
3538         h->stat.i_consecutive_bframes[h->fenc->i_bframes]++;
3539
3540     psz_message[0] = '\0';
3541     double dur = h->fenc->f_duration;
3542     h->stat.f_frame_duration[h->sh.i_type] += dur;
3543     if( h->param.analyse.b_psnr )
3544     {
3545         int64_t ssd[3] =
3546         {
3547             h->stat.frame.i_ssd[0],
3548             h->stat.frame.i_ssd[1],
3549             h->stat.frame.i_ssd[2],
3550         };
3551         int luma_size = h->param.i_width * h->param.i_height;
3552         int chroma_size = CHROMA_SIZE( luma_size );
3553         pic_out->prop.f_psnr[0] = x264_psnr( ssd[0], luma_size );
3554         pic_out->prop.f_psnr[1] = x264_psnr( ssd[1], chroma_size );
3555         pic_out->prop.f_psnr[2] = x264_psnr( ssd[2], chroma_size );
3556         pic_out->prop.f_psnr_avg = x264_psnr( ssd[0] + ssd[1] + ssd[2], luma_size + chroma_size*2 );
3557
3558         h->stat.f_ssd_global[h->sh.i_type]   += dur * (ssd[0] + ssd[1] + ssd[2]);
3559         h->stat.f_psnr_average[h->sh.i_type] += dur * pic_out->prop.f_psnr_avg;
3560         h->stat.f_psnr_mean_y[h->sh.i_type]  += dur * pic_out->prop.f_psnr[0];
3561         h->stat.f_psnr_mean_u[h->sh.i_type]  += dur * pic_out->prop.f_psnr[1];
3562         h->stat.f_psnr_mean_v[h->sh.i_type]  += dur * pic_out->prop.f_psnr[2];
3563
3564         snprintf( psz_message, 80, " PSNR Y:%5.2f U:%5.2f V:%5.2f", pic_out->prop.f_psnr[0],
3565                                                                     pic_out->prop.f_psnr[1],
3566                                                                     pic_out->prop.f_psnr[2] );
3567     }
3568
3569     if( h->param.analyse.b_ssim )
3570     {
3571         pic_out->prop.f_ssim = h->stat.frame.f_ssim / h->stat.frame.i_ssim_cnt;
3572         h->stat.f_ssim_mean_y[h->sh.i_type] += pic_out->prop.f_ssim * dur;
3573         snprintf( psz_message + strlen(psz_message), 80 - strlen(psz_message),
3574                   " SSIM Y:%.5f", pic_out->prop.f_ssim );
3575     }
3576     psz_message[79] = '\0';
3577
3578     x264_log( h, X264_LOG_DEBUG,
3579                   "frame=%4d QP=%.2f NAL=%d Slice:%c Poc:%-3d I:%-4d P:%-4d SKIP:%-4d size=%d bytes%s\n",
3580               h->i_frame,
3581               h->fdec->f_qp_avg_aq,
3582               h->i_nal_ref_idc,
3583               h->sh.i_type == SLICE_TYPE_I ? 'I' : (h->sh.i_type == SLICE_TYPE_P ? 'P' : 'B' ),
3584               h->fdec->i_poc,
3585               h->stat.frame.i_mb_count_i,
3586               h->stat.frame.i_mb_count_p,
3587               h->stat.frame.i_mb_count_skip,
3588               frame_size,
3589               psz_message );
3590
3591     // keep stats all in one place
3592     x264_thread_sync_stat( h->thread[0], h );
3593     // for the use of the next frame
3594     x264_thread_sync_stat( thread_current, h );
3595
3596 #ifdef DEBUG_MB_TYPE
3597 {
3598     static const char mb_chars[] = { 'i', 'i', 'I', 'C', 'P', '8', 'S',
3599         'D', '<', 'X', 'B', 'X', '>', 'B', 'B', 'B', 'B', '8', 'S' };
3600     for( int mb_xy = 0; mb_xy < h->mb.i_mb_width * h->mb.i_mb_height; mb_xy++ )
3601     {
3602         if( h->mb.type[mb_xy] < X264_MBTYPE_MAX && h->mb.type[mb_xy] >= 0 )
3603             fprintf( stderr, "%c ", mb_chars[ h->mb.type[mb_xy] ] );
3604         else
3605             fprintf( stderr, "? " );
3606
3607         if( (mb_xy+1) % h->mb.i_mb_width == 0 )
3608             fprintf( stderr, "\n" );
3609     }
3610 }
3611 #endif
3612
3613     /* Remove duplicates, must be done near the end as breaks h->fref0 array
3614      * by freeing some of its pointers. */
3615     for( int i = 0; i < h->i_ref[0]; i++ )
3616         if( h->fref[0][i] && h->fref[0][i]->b_duplicate )
3617         {
3618             x264_frame_push_blank_unused( h, h->fref[0][i] );
3619             h->fref[0][i] = 0;
3620         }
3621
3622     if( h->param.psz_dump_yuv )
3623         x264_frame_dump( h );
3624     x264_emms();
3625
3626     return frame_size;
3627 }
3628
3629 static void x264_print_intra( int64_t *i_mb_count, double i_count, int b_print_pcm, char *intra )
3630 {
3631     intra += sprintf( intra, "I16..4%s: %4.1f%% %4.1f%% %4.1f%%",
3632         b_print_pcm ? "..PCM" : "",
3633         i_mb_count[I_16x16]/ i_count,
3634         i_mb_count[I_8x8]  / i_count,
3635         i_mb_count[I_4x4]  / i_count );
3636     if( b_print_pcm )
3637         sprintf( intra, " %4.1f%%", i_mb_count[I_PCM]  / i_count );
3638 }
3639
3640 /****************************************************************************
3641  * x264_encoder_close:
3642  ****************************************************************************/
3643 void    x264_encoder_close  ( x264_t *h )
3644 {
3645     int64_t i_yuv_size = FRAME_SIZE( h->param.i_width * h->param.i_height );
3646     int64_t i_mb_count_size[2][7] = {{0}};
3647     char buf[200];
3648     int b_print_pcm = h->stat.i_mb_count[SLICE_TYPE_I][I_PCM]
3649                    || h->stat.i_mb_count[SLICE_TYPE_P][I_PCM]
3650                    || h->stat.i_mb_count[SLICE_TYPE_B][I_PCM];
3651
3652 #if HAVE_OPENCL
3653     x264_opencl_free( h );
3654 #endif
3655
3656     x264_lookahead_delete( h );
3657
3658     if( h->param.b_sliced_threads )
3659         x264_threadpool_wait_all( h );
3660     if( h->param.i_threads > 1 )
3661         x264_threadpool_delete( h->threadpool );
3662     if( h->param.i_lookahead_threads > 1 )
3663         x264_threadpool_delete( h->lookaheadpool );
3664     if( h->i_thread_frames > 1 )
3665     {
3666         for( int i = 0; i < h->i_thread_frames; i++ )
3667             if( h->thread[i]->b_thread_active )
3668             {
3669                 assert( h->thread[i]->fenc->i_reference_count == 1 );
3670                 x264_frame_delete( h->thread[i]->fenc );
3671             }
3672
3673         x264_t *thread_prev = h->thread[h->i_thread_phase];
3674         x264_thread_sync_ratecontrol( h, thread_prev, h );
3675         x264_thread_sync_ratecontrol( thread_prev, thread_prev, h );
3676         h->i_frame = thread_prev->i_frame + 1 - h->i_thread_frames;
3677     }
3678     h->i_frame++;
3679
3680     /* Slices used and PSNR */
3681     for( int i = 0; i < 3; i++ )
3682     {
3683         static const uint8_t slice_order[] = { SLICE_TYPE_I, SLICE_TYPE_P, SLICE_TYPE_B };
3684         int i_slice = slice_order[i];
3685
3686         if( h->stat.i_frame_count[i_slice] > 0 )
3687         {
3688             int i_count = h->stat.i_frame_count[i_slice];
3689             double dur =  h->stat.f_frame_duration[i_slice];
3690             if( h->param.analyse.b_psnr )
3691             {
3692                 x264_log( h, X264_LOG_INFO,
3693                           "frame %c:%-5d Avg QP:%5.2f  size:%6.0f  PSNR Mean Y:%5.2f U:%5.2f V:%5.2f Avg:%5.2f Global:%5.2f\n",
3694                           slice_type_to_char[i_slice],
3695                           i_count,
3696                           h->stat.f_frame_qp[i_slice] / i_count,
3697                           (double)h->stat.i_frame_size[i_slice] / i_count,
3698                           h->stat.f_psnr_mean_y[i_slice] / dur, h->stat.f_psnr_mean_u[i_slice] / dur, h->stat.f_psnr_mean_v[i_slice] / dur,
3699                           h->stat.f_psnr_average[i_slice] / dur,
3700                           x264_psnr( h->stat.f_ssd_global[i_slice], dur * i_yuv_size ) );
3701             }
3702             else
3703             {
3704                 x264_log( h, X264_LOG_INFO,
3705                           "frame %c:%-5d Avg QP:%5.2f  size:%6.0f\n",
3706                           slice_type_to_char[i_slice],
3707                           i_count,
3708                           h->stat.f_frame_qp[i_slice] / i_count,
3709                           (double)h->stat.i_frame_size[i_slice] / i_count );
3710             }
3711         }
3712     }
3713     if( h->param.i_bframe && h->stat.i_frame_count[SLICE_TYPE_B] )
3714     {
3715         char *p = buf;
3716         int den = 0;
3717         // weight by number of frames (including the I/P-frames) that are in a sequence of N B-frames
3718         for( int i = 0; i <= h->param.i_bframe; i++ )
3719             den += (i+1) * h->stat.i_consecutive_bframes[i];
3720         for( int i = 0; i <= h->param.i_bframe; i++ )
3721             p += sprintf( p, " %4.1f%%", 100. * (i+1) * h->stat.i_consecutive_bframes[i] / den );
3722         x264_log( h, X264_LOG_INFO, "consecutive B-frames:%s\n", buf );
3723     }
3724
3725     for( int i_type = 0; i_type < 2; i_type++ )
3726         for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
3727         {
3728             if( i == D_DIRECT_8x8 ) continue; /* direct is counted as its own type */
3729             i_mb_count_size[i_type][x264_mb_partition_pixel_table[i]] += h->stat.i_mb_partition[i_type][i];
3730         }
3731
3732     /* MB types used */
3733     if( h->stat.i_frame_count[SLICE_TYPE_I] > 0 )
3734     {
3735         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_I];
3736         double i_count = h->stat.i_frame_count[SLICE_TYPE_I] * h->mb.i_mb_count / 100.0;
3737         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
3738         x264_log( h, X264_LOG_INFO, "mb I  %s\n", buf );
3739     }
3740     if( h->stat.i_frame_count[SLICE_TYPE_P] > 0 )
3741     {
3742         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_P];
3743         double i_count = h->stat.i_frame_count[SLICE_TYPE_P] * h->mb.i_mb_count / 100.0;
3744         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_P];
3745         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
3746         x264_log( h, X264_LOG_INFO,
3747                   "mb P  %s  P16..4: %4.1f%% %4.1f%% %4.1f%% %4.1f%% %4.1f%%    skip:%4.1f%%\n",
3748                   buf,
3749                   i_mb_size[PIXEL_16x16] / (i_count*4),
3750                   (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
3751                   i_mb_size[PIXEL_8x8] / (i_count*4),
3752                   (i_mb_size[PIXEL_8x4] + i_mb_size[PIXEL_4x8]) / (i_count*4),
3753                   i_mb_size[PIXEL_4x4] / (i_count*4),
3754                   i_mb_count[P_SKIP] / i_count );
3755     }
3756     if( h->stat.i_frame_count[SLICE_TYPE_B] > 0 )
3757     {
3758         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_B];
3759         double i_count = h->stat.i_frame_count[SLICE_TYPE_B] * h->mb.i_mb_count / 100.0;
3760         double i_mb_list_count;
3761         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_B];
3762         int64_t list_count[3] = {0}; /* 0 == L0, 1 == L1, 2 == BI */
3763         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
3764         for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
3765             for( int j = 0; j < 2; j++ )
3766             {
3767                 int l0 = x264_mb_type_list_table[i][0][j];
3768                 int l1 = x264_mb_type_list_table[i][1][j];
3769                 if( l0 || l1 )
3770                     list_count[l1+l0*l1] += h->stat.i_mb_count[SLICE_TYPE_B][i] * 2;
3771             }
3772         list_count[0] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L0_8x8];
3773         list_count[1] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L1_8x8];
3774         list_count[2] += h->stat.i_mb_partition[SLICE_TYPE_B][D_BI_8x8];
3775         i_mb_count[B_DIRECT] += (h->stat.i_mb_partition[SLICE_TYPE_B][D_DIRECT_8x8]+2)/4;
3776         i_mb_list_count = (list_count[0] + list_count[1] + list_count[2]) / 100.0;
3777         sprintf( buf + strlen(buf), "  B16..8: %4.1f%% %4.1f%% %4.1f%%  direct:%4.1f%%  skip:%4.1f%%",
3778                  i_mb_size[PIXEL_16x16] / (i_count*4),
3779                  (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
3780                  i_mb_size[PIXEL_8x8] / (i_count*4),
3781                  i_mb_count[B_DIRECT] / i_count,
3782                  i_mb_count[B_SKIP]   / i_count );
3783         if( i_mb_list_count != 0 )
3784             sprintf( buf + strlen(buf), "  L0:%4.1f%% L1:%4.1f%% BI:%4.1f%%",
3785                      list_count[0] / i_mb_list_count,
3786                      list_count[1] / i_mb_list_count,
3787                      list_count[2] / i_mb_list_count );
3788         x264_log( h, X264_LOG_INFO, "mb B  %s\n", buf );
3789     }
3790
3791     x264_ratecontrol_summary( h );
3792
3793     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 )
3794     {
3795 #define SUM3(p) (p[SLICE_TYPE_I] + p[SLICE_TYPE_P] + p[SLICE_TYPE_B])
3796 #define SUM3b(p,o) (p[SLICE_TYPE_I][o] + p[SLICE_TYPE_P][o] + p[SLICE_TYPE_B][o])
3797         int64_t i_i8x8 = SUM3b( h->stat.i_mb_count, I_8x8 );
3798         int64_t i_intra = i_i8x8 + SUM3b( h->stat.i_mb_count, I_4x4 )
3799                                  + SUM3b( h->stat.i_mb_count, I_16x16 );
3800         int64_t i_all_intra = i_intra + SUM3b( h->stat.i_mb_count, I_PCM);
3801         int64_t i_skip = SUM3b( h->stat.i_mb_count, P_SKIP )
3802                        + SUM3b( h->stat.i_mb_count, B_SKIP );
3803         const int i_count = h->stat.i_frame_count[SLICE_TYPE_I] +
3804                             h->stat.i_frame_count[SLICE_TYPE_P] +
3805                             h->stat.i_frame_count[SLICE_TYPE_B];
3806         int64_t i_mb_count = (int64_t)i_count * h->mb.i_mb_count;
3807         int64_t i_inter = i_mb_count - i_skip - i_intra;
3808         const double duration = h->stat.f_frame_duration[SLICE_TYPE_I] +
3809                                 h->stat.f_frame_duration[SLICE_TYPE_P] +
3810                                 h->stat.f_frame_duration[SLICE_TYPE_B];
3811         float f_bitrate = SUM3(h->stat.i_frame_size) / duration / 125;
3812
3813         if( PARAM_INTERLACED )
3814         {
3815             char *fieldstats = buf;
3816             fieldstats[0] = 0;
3817             if( i_inter )
3818                 fieldstats += sprintf( fieldstats, " inter:%.1f%%", h->stat.i_mb_field[1] * 100.0 / i_inter );
3819             if( i_skip )
3820                 fieldstats += sprintf( fieldstats, " skip:%.1f%%", h->stat.i_mb_field[2] * 100.0 / i_skip );
3821             x264_log( h, X264_LOG_INFO, "field mbs: intra: %.1f%%%s\n",
3822                       h->stat.i_mb_field[0] * 100.0 / i_intra, buf );
3823         }
3824
3825         if( h->pps->b_transform_8x8_mode )
3826         {
3827             buf[0] = 0;
3828             if( h->stat.i_mb_count_8x8dct[0] )
3829                 sprintf( buf, " inter:%.1f%%", 100. * h->stat.i_mb_count_8x8dct[1] / h->stat.i_mb_count_8x8dct[0] );
3830             x264_log( h, X264_LOG_INFO, "8x8 transform intra:%.1f%%%s\n", 100. * i_i8x8 / i_intra, buf );
3831         }
3832
3833         if( (h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO ||
3834             (h->stat.i_direct_frames[0] && h->stat.i_direct_frames[1]))
3835             && h->stat.i_frame_count[SLICE_TYPE_B] )
3836         {
3837             x264_log( h, X264_LOG_INFO, "direct mvs  spatial:%.1f%% temporal:%.1f%%\n",
3838                       h->stat.i_direct_frames[1] * 100. / h->stat.i_frame_count[SLICE_TYPE_B],
3839                       h->stat.i_direct_frames[0] * 100. / h->stat.i_frame_count[SLICE_TYPE_B] );
3840         }
3841
3842         buf[0] = 0;
3843         int csize = CHROMA444 ? 4 : 1;
3844         if( i_mb_count != i_all_intra )
3845             sprintf( buf, " inter: %.1f%% %.1f%% %.1f%%",
3846                      h->stat.i_mb_cbp[1] * 100.0 / ((i_mb_count - i_all_intra)*4),
3847                      h->stat.i_mb_cbp[3] * 100.0 / ((i_mb_count - i_all_intra)*csize),
3848                      h->stat.i_mb_cbp[5] * 100.0 / ((i_mb_count - i_all_intra)*csize) );
3849         x264_log( h, X264_LOG_INFO, "coded y,%s,%s intra: %.1f%% %.1f%% %.1f%%%s\n",
3850                   CHROMA444?"u":"uvDC", CHROMA444?"v":"uvAC",
3851                   h->stat.i_mb_cbp[0] * 100.0 / (i_all_intra*4),
3852                   h->stat.i_mb_cbp[2] * 100.0 / (i_all_intra*csize),
3853                   h->stat.i_mb_cbp[4] * 100.0 / (i_all_intra*csize), buf );
3854
3855         int64_t fixed_pred_modes[4][9] = {{0}};
3856         int64_t sum_pred_modes[4] = {0};
3857         for( int i = 0; i <= I_PRED_16x16_DC_128; i++ )
3858         {
3859             fixed_pred_modes[0][x264_mb_pred_mode16x16_fix[i]] += h->stat.i_mb_pred_mode[0][i];
3860             sum_pred_modes[0] += h->stat.i_mb_pred_mode[0][i];
3861         }
3862         if( sum_pred_modes[0] )
3863             x264_log( h, X264_LOG_INFO, "i16 v,h,dc,p: %2.0f%% %2.0f%% %2.0f%% %2.0f%%\n",
3864                       fixed_pred_modes[0][0] * 100.0 / sum_pred_modes[0],
3865                       fixed_pred_modes[0][1] * 100.0 / sum_pred_modes[0],
3866                       fixed_pred_modes[0][2] * 100.0 / sum_pred_modes[0],
3867                       fixed_pred_modes[0][3] * 100.0 / sum_pred_modes[0] );
3868         for( int i = 1; i <= 2; i++ )
3869         {
3870             for( int j = 0; j <= I_PRED_8x8_DC_128; j++ )
3871             {
3872                 fixed_pred_modes[i][x264_mb_pred_mode4x4_fix(j)] += h->stat.i_mb_pred_mode[i][j];
3873                 sum_pred_modes[i] += h->stat.i_mb_pred_mode[i][j];
3874             }
3875             if( sum_pred_modes[i] )
3876                 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,
3877                           fixed_pred_modes[i][0] * 100.0 / sum_pred_modes[i],
3878                           fixed_pred_modes[i][1] * 100.0 / sum_pred_modes[i],
3879                           fixed_pred_modes[i][2] * 100.0 / sum_pred_modes[i],
3880                           fixed_pred_modes[i][3] * 100.0 / sum_pred_modes[i],
3881                           fixed_pred_modes[i][4] * 100.0 / sum_pred_modes[i],
3882                           fixed_pred_modes[i][5] * 100.0 / sum_pred_modes[i],
3883                           fixed_pred_modes[i][6] * 100.0 / sum_pred_modes[i],
3884                           fixed_pred_modes[i][7] * 100.0 / sum_pred_modes[i],
3885                           fixed_pred_modes[i][8] * 100.0 / sum_pred_modes[i] );
3886         }
3887         for( int i = 0; i <= I_PRED_CHROMA_DC_128; i++ )
3888         {
3889             fixed_pred_modes[3][x264_mb_chroma_pred_mode_fix[i]] += h->stat.i_mb_pred_mode[3][i];
3890             sum_pred_modes[3] += h->stat.i_mb_pred_mode[3][i];
3891         }
3892         if( sum_pred_modes[3] && !CHROMA444 )
3893             x264_log( h, X264_LOG_INFO, "i8c dc,h,v,p: %2.0f%% %2.0f%% %2.0f%% %2.0f%%\n",
3894                       fixed_pred_modes[3][0] * 100.0 / sum_pred_modes[3],
3895                       fixed_pred_modes[3][1] * 100.0 / sum_pred_modes[3],
3896                       fixed_pred_modes[3][2] * 100.0 / sum_pred_modes[3],
3897                       fixed_pred_modes[3][3] * 100.0 / sum_pred_modes[3] );
3898
3899         if( h->param.analyse.i_weighted_pred >= X264_WEIGHTP_SIMPLE && h->stat.i_frame_count[SLICE_TYPE_P] > 0 )
3900             x264_log( h, X264_LOG_INFO, "Weighted P-Frames: Y:%.1f%% UV:%.1f%%\n",
3901                       h->stat.i_wpred[0] * 100.0 / h->stat.i_frame_count[SLICE_TYPE_P],
3902                       h->stat.i_wpred[1] * 100.0 / h->stat.i_frame_count[SLICE_TYPE_P] );
3903
3904         for( int i_list = 0; i_list < 2; i_list++ )
3905             for( int i_slice = 0; i_slice < 2; i_slice++ )
3906             {
3907                 char *p = buf;
3908                 int64_t i_den = 0;
3909                 int i_max = 0;
3910                 for( int i = 0; i < X264_REF_MAX*2; i++ )
3911                     if( h->stat.i_mb_count_ref[i_slice][i_list][i] )
3912                     {
3913                         i_den += h->stat.i_mb_count_ref[i_slice][i_list][i];
3914                         i_max = i;
3915                     }
3916                 if( i_max == 0 )
3917                     continue;
3918                 for( int i = 0; i <= i_max; i++ )
3919                     p += sprintf( p, " %4.1f%%", 100. * h->stat.i_mb_count_ref[i_slice][i_list][i] / i_den );
3920                 x264_log( h, X264_LOG_INFO, "ref %c L%d:%s\n", "PB"[i_slice], i_list, buf );
3921             }
3922
3923         if( h->param.analyse.b_ssim )
3924         {
3925             float ssim = SUM3( h->stat.f_ssim_mean_y ) / duration;
3926             x264_log( h, X264_LOG_INFO, "SSIM Mean Y:%.7f (%6.3fdb)\n", ssim, x264_ssim( ssim ) );
3927         }
3928         if( h->param.analyse.b_psnr )
3929         {
3930             x264_log( h, X264_LOG_INFO,
3931                       "PSNR Mean Y:%6.3f U:%6.3f V:%6.3f Avg:%6.3f Global:%6.3f kb/s:%.2f\n",
3932                       SUM3( h->stat.f_psnr_mean_y ) / duration,
3933                       SUM3( h->stat.f_psnr_mean_u ) / duration,
3934                       SUM3( h->stat.f_psnr_mean_v ) / duration,
3935                       SUM3( h->stat.f_psnr_average ) / duration,
3936                       x264_psnr( SUM3( h->stat.f_ssd_global ), duration * i_yuv_size ),
3937                       f_bitrate );
3938         }
3939         else
3940             x264_log( h, X264_LOG_INFO, "kb/s:%.2f\n", f_bitrate );
3941     }
3942
3943     /* rc */
3944     x264_ratecontrol_delete( h );
3945
3946     /* param */
3947     if( h->param.rc.psz_stat_out )
3948         free( h->param.rc.psz_stat_out );
3949     if( h->param.rc.psz_stat_in )
3950         free( h->param.rc.psz_stat_in );
3951
3952     x264_cqm_delete( h );
3953     x264_free( h->nal_buffer );
3954     x264_analyse_free_costs( h );
3955
3956     if( h->i_thread_frames > 1 )
3957         h = h->thread[h->i_thread_phase];
3958
3959     /* frames */
3960     x264_frame_delete_list( h->frames.unused[0] );
3961     x264_frame_delete_list( h->frames.unused[1] );
3962     x264_frame_delete_list( h->frames.current );
3963     x264_frame_delete_list( h->frames.blank_unused );
3964
3965     h = h->thread[0];
3966
3967     for( int i = 0; i < h->i_thread_frames; i++ )
3968         if( h->thread[i]->b_thread_active )
3969             for( int j = 0; j < h->thread[i]->i_ref[0]; j++ )
3970                 if( h->thread[i]->fref[0][j] && h->thread[i]->fref[0][j]->b_duplicate )
3971                     x264_frame_delete( h->thread[i]->fref[0][j] );
3972
3973     if( h->param.i_lookahead_threads > 1 )
3974         for( int i = 0; i < h->param.i_lookahead_threads; i++ )
3975             x264_free( h->lookahead_thread[i] );
3976
3977     for( int i = h->param.i_threads - 1; i >= 0; i-- )
3978     {
3979         x264_frame_t **frame;
3980
3981         if( !h->param.b_sliced_threads || i == 0 )
3982         {
3983             for( frame = h->thread[i]->frames.reference; *frame; frame++ )
3984             {
3985                 assert( (*frame)->i_reference_count > 0 );
3986                 (*frame)->i_reference_count--;
3987                 if( (*frame)->i_reference_count == 0 )
3988                     x264_frame_delete( *frame );
3989             }
3990             frame = &h->thread[i]->fdec;
3991             if( *frame )
3992             {
3993                 assert( (*frame)->i_reference_count > 0 );
3994                 (*frame)->i_reference_count--;
3995                 if( (*frame)->i_reference_count == 0 )
3996                     x264_frame_delete( *frame );
3997             }
3998             x264_macroblock_cache_free( h->thread[i] );
3999         }
4000         x264_macroblock_thread_free( h->thread[i], 0 );
4001         x264_free( h->thread[i]->out.p_bitstream );
4002         x264_free( h->thread[i]->out.nal );
4003         x264_pthread_mutex_destroy( &h->thread[i]->mutex );
4004         x264_pthread_cond_destroy( &h->thread[i]->cv );
4005         x264_free( h->thread[i] );
4006     }
4007 }
4008
4009 int x264_encoder_delayed_frames( x264_t *h )
4010 {
4011     int delayed_frames = 0;
4012     if( h->i_thread_frames > 1 )
4013     {
4014         for( int i = 0; i < h->i_thread_frames; i++ )
4015             delayed_frames += h->thread[i]->b_thread_active;
4016         h = h->thread[h->i_thread_phase];
4017     }
4018     for( int i = 0; h->frames.current[i]; i++ )
4019         delayed_frames++;
4020     x264_pthread_mutex_lock( &h->lookahead->ofbuf.mutex );
4021     x264_pthread_mutex_lock( &h->lookahead->ifbuf.mutex );
4022     x264_pthread_mutex_lock( &h->lookahead->next.mutex );
4023     delayed_frames += h->lookahead->ifbuf.i_size + h->lookahead->next.i_size + h->lookahead->ofbuf.i_size;
4024     x264_pthread_mutex_unlock( &h->lookahead->next.mutex );
4025     x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
4026     x264_pthread_mutex_unlock( &h->lookahead->ofbuf.mutex );
4027     return delayed_frames;
4028 }
4029
4030 int x264_encoder_maximum_delayed_frames( x264_t *h )
4031 {
4032     return h->frames.i_delay;
4033 }