Remove --visualize option.
[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 //#define DEBUG_MB_TYPE
37
38 #define bs_write_ue bs_write_ue_big
39
40 static int x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
41                                    x264_nal_t **pp_nal, int *pi_nal,
42                                    x264_picture_t *pic_out );
43
44 /****************************************************************************
45  *
46  ******************************* x264 libs **********************************
47  *
48  ****************************************************************************/
49 static double x264_psnr( double sqe, double size )
50 {
51     double mse = sqe / (PIXEL_MAX*PIXEL_MAX * size);
52     if( mse <= 0.0000000001 ) /* Max 100dB */
53         return 100;
54
55     return -10.0 * log10( mse );
56 }
57
58 static double x264_ssim( double ssim )
59 {
60     double inv_ssim = 1 - ssim;
61     if( inv_ssim <= 0.0000000001 ) /* Max 100dB */
62         return 100;
63
64     return -10.0 * log10( inv_ssim );
65 }
66
67 static int x264_threadpool_wait_all( x264_t *h )
68 {
69     for( int i = 0; i < h->param.i_threads; i++ )
70         if( h->thread[i]->b_thread_active )
71         {
72             h->thread[i]->b_thread_active = 0;
73             if( (intptr_t)x264_threadpool_wait( h->threadpool, h->thread[i] ) < 0 )
74                 return -1;
75         }
76     return 0;
77 }
78
79 static void x264_frame_dump( x264_t *h )
80 {
81     FILE *f = x264_fopen( h->param.psz_dump_yuv, "r+b" );
82     if( !f )
83         return;
84
85     /* Wait for the threads to finish deblocking */
86     if( h->param.b_sliced_threads )
87         x264_threadpool_wait_all( h );
88
89     /* Write the frame in display order */
90     int frame_size = FRAME_SIZE( h->param.i_height * h->param.i_width * sizeof(pixel) );
91     fseek( f, (uint64_t)h->fdec->i_frame * frame_size, SEEK_SET );
92     for( int p = 0; p < (CHROMA444 ? 3 : 1); p++ )
93         for( int y = 0; y < h->param.i_height; y++ )
94             fwrite( &h->fdec->plane[p][y*h->fdec->i_stride[p]], sizeof(pixel), h->param.i_width, f );
95     if( !CHROMA444 )
96     {
97         int cw = h->param.i_width>>1;
98         int ch = h->param.i_height>>CHROMA_V_SHIFT;
99         pixel *planeu = x264_malloc( (cw*ch*2+32)*sizeof(pixel) );
100         pixel *planev = planeu + cw*ch + 16;
101         h->mc.plane_copy_deinterleave( planeu, cw, planev, cw, h->fdec->plane[1], h->fdec->i_stride[1], cw, ch );
102         fwrite( planeu, 1, cw*ch*sizeof(pixel), f );
103         fwrite( planev, 1, cw*ch*sizeof(pixel), f );
104         x264_free( planeu );
105     }
106     fclose( f );
107 }
108
109 /* Fill "default" values */
110 static void x264_slice_header_init( x264_t *h, x264_slice_header_t *sh,
111                                     x264_sps_t *sps, x264_pps_t *pps,
112                                     int i_idr_pic_id, int i_frame, int i_qp )
113 {
114     x264_param_t *param = &h->param;
115
116     /* First we fill all fields */
117     sh->sps = sps;
118     sh->pps = pps;
119
120     sh->i_first_mb  = 0;
121     sh->i_last_mb   = h->mb.i_mb_count - 1;
122     sh->i_pps_id    = pps->i_id;
123
124     sh->i_frame_num = i_frame;
125
126     sh->b_mbaff = PARAM_INTERLACED;
127     sh->b_field_pic = 0;    /* no field support for now */
128     sh->b_bottom_field = 0; /* not yet used */
129
130     sh->i_idr_pic_id = i_idr_pic_id;
131
132     /* poc stuff, fixed later */
133     sh->i_poc = 0;
134     sh->i_delta_poc_bottom = 0;
135     sh->i_delta_poc[0] = 0;
136     sh->i_delta_poc[1] = 0;
137
138     sh->i_redundant_pic_cnt = 0;
139
140     h->mb.b_direct_auto_write = h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO
141                                 && h->param.i_bframe
142                                 && ( h->param.rc.b_stat_write || !h->param.rc.b_stat_read );
143
144     if( !h->mb.b_direct_auto_read && sh->i_type == SLICE_TYPE_B )
145     {
146         if( h->fref[1][0]->i_poc_l0ref0 == h->fref[0][0]->i_poc )
147         {
148             if( h->mb.b_direct_auto_write )
149                 sh->b_direct_spatial_mv_pred = ( h->stat.i_direct_score[1] > h->stat.i_direct_score[0] );
150             else
151                 sh->b_direct_spatial_mv_pred = ( param->analyse.i_direct_mv_pred == X264_DIRECT_PRED_SPATIAL );
152         }
153         else
154         {
155             h->mb.b_direct_auto_write = 0;
156             sh->b_direct_spatial_mv_pred = 1;
157         }
158     }
159     /* else b_direct_spatial_mv_pred was read from the 2pass statsfile */
160
161     sh->b_num_ref_idx_override = 0;
162     sh->i_num_ref_idx_l0_active = 1;
163     sh->i_num_ref_idx_l1_active = 1;
164
165     sh->b_ref_pic_list_reordering[0] = h->b_ref_reorder[0];
166     sh->b_ref_pic_list_reordering[1] = h->b_ref_reorder[1];
167
168     /* If the ref list isn't in the default order, construct reordering header */
169     for( int list = 0; list < 2; list++ )
170     {
171         if( sh->b_ref_pic_list_reordering[list] )
172         {
173             int pred_frame_num = i_frame;
174             for( int i = 0; i < h->i_ref[list]; i++ )
175             {
176                 int diff = h->fref[list][i]->i_frame_num - pred_frame_num;
177                 sh->ref_pic_list_order[list][i].idc = ( diff > 0 );
178                 sh->ref_pic_list_order[list][i].arg = (abs(diff) - 1) & ((1 << sps->i_log2_max_frame_num) - 1);
179                 pred_frame_num = h->fref[list][i]->i_frame_num;
180             }
181         }
182     }
183
184     sh->i_cabac_init_idc = param->i_cabac_init_idc;
185
186     sh->i_qp = SPEC_QP(i_qp);
187     sh->i_qp_delta = sh->i_qp - pps->i_pic_init_qp;
188     sh->b_sp_for_swidth = 0;
189     sh->i_qs_delta = 0;
190
191     int deblock_thresh = i_qp + 2 * X264_MIN(param->i_deblocking_filter_alphac0, param->i_deblocking_filter_beta);
192     /* If effective qp <= 15, deblocking would have no effect anyway */
193     if( param->b_deblocking_filter && (h->mb.b_variable_qp || 15 < deblock_thresh ) )
194         sh->i_disable_deblocking_filter_idc = param->b_sliced_threads ? 2 : 0;
195     else
196         sh->i_disable_deblocking_filter_idc = 1;
197     sh->i_alpha_c0_offset = param->i_deblocking_filter_alphac0 << 1;
198     sh->i_beta_offset = param->i_deblocking_filter_beta << 1;
199 }
200
201 static void x264_slice_header_write( bs_t *s, x264_slice_header_t *sh, int i_nal_ref_idc )
202 {
203     if( sh->b_mbaff )
204     {
205         int first_x = sh->i_first_mb % sh->sps->i_mb_width;
206         int first_y = sh->i_first_mb / sh->sps->i_mb_width;
207         assert( (first_y&1) == 0 );
208         bs_write_ue( s, (2*first_x + sh->sps->i_mb_width*(first_y&~1) + (first_y&1)) >> 1 );
209     }
210     else
211         bs_write_ue( s, sh->i_first_mb );
212
213     bs_write_ue( s, sh->i_type + 5 );   /* same type things */
214     bs_write_ue( s, sh->i_pps_id );
215     bs_write( s, sh->sps->i_log2_max_frame_num, sh->i_frame_num & ((1<<sh->sps->i_log2_max_frame_num)-1) );
216
217     if( !sh->sps->b_frame_mbs_only )
218     {
219         bs_write1( s, sh->b_field_pic );
220         if( sh->b_field_pic )
221             bs_write1( s, sh->b_bottom_field );
222     }
223
224     if( sh->i_idr_pic_id >= 0 ) /* NAL IDR */
225         bs_write_ue( s, sh->i_idr_pic_id );
226
227     if( sh->sps->i_poc_type == 0 )
228     {
229         bs_write( s, sh->sps->i_log2_max_poc_lsb, sh->i_poc & ((1<<sh->sps->i_log2_max_poc_lsb)-1) );
230         if( sh->pps->b_pic_order && !sh->b_field_pic )
231             bs_write_se( s, sh->i_delta_poc_bottom );
232     }
233
234     if( sh->pps->b_redundant_pic_cnt )
235         bs_write_ue( s, sh->i_redundant_pic_cnt );
236
237     if( sh->i_type == SLICE_TYPE_B )
238         bs_write1( s, sh->b_direct_spatial_mv_pred );
239
240     if( sh->i_type == SLICE_TYPE_P || sh->i_type == SLICE_TYPE_B )
241     {
242         bs_write1( s, sh->b_num_ref_idx_override );
243         if( sh->b_num_ref_idx_override )
244         {
245             bs_write_ue( s, sh->i_num_ref_idx_l0_active - 1 );
246             if( sh->i_type == SLICE_TYPE_B )
247                 bs_write_ue( s, sh->i_num_ref_idx_l1_active - 1 );
248         }
249     }
250
251     /* ref pic list reordering */
252     if( sh->i_type != SLICE_TYPE_I )
253     {
254         bs_write1( s, sh->b_ref_pic_list_reordering[0] );
255         if( sh->b_ref_pic_list_reordering[0] )
256         {
257             for( int i = 0; i < sh->i_num_ref_idx_l0_active; i++ )
258             {
259                 bs_write_ue( s, sh->ref_pic_list_order[0][i].idc );
260                 bs_write_ue( s, sh->ref_pic_list_order[0][i].arg );
261             }
262             bs_write_ue( s, 3 );
263         }
264     }
265     if( sh->i_type == SLICE_TYPE_B )
266     {
267         bs_write1( s, sh->b_ref_pic_list_reordering[1] );
268         if( sh->b_ref_pic_list_reordering[1] )
269         {
270             for( int i = 0; i < sh->i_num_ref_idx_l1_active; i++ )
271             {
272                 bs_write_ue( s, sh->ref_pic_list_order[1][i].idc );
273                 bs_write_ue( s, sh->ref_pic_list_order[1][i].arg );
274             }
275             bs_write_ue( s, 3 );
276         }
277     }
278
279     sh->b_weighted_pred = 0;
280     if( sh->pps->b_weighted_pred && sh->i_type == SLICE_TYPE_P )
281     {
282         sh->b_weighted_pred = sh->weight[0][0].weightfn || sh->weight[0][1].weightfn || sh->weight[0][2].weightfn;
283         /* pred_weight_table() */
284         bs_write_ue( s, sh->weight[0][0].i_denom );
285         bs_write_ue( s, sh->weight[0][1].i_denom );
286         for( int i = 0; i < sh->i_num_ref_idx_l0_active; i++ )
287         {
288             int luma_weight_l0_flag = !!sh->weight[i][0].weightfn;
289             int chroma_weight_l0_flag = !!sh->weight[i][1].weightfn || !!sh->weight[i][2].weightfn;
290             bs_write1( s, luma_weight_l0_flag );
291             if( luma_weight_l0_flag )
292             {
293                 bs_write_se( s, sh->weight[i][0].i_scale );
294                 bs_write_se( s, sh->weight[i][0].i_offset );
295             }
296             bs_write1( s, chroma_weight_l0_flag );
297             if( chroma_weight_l0_flag )
298             {
299                 for( int j = 1; j < 3; j++ )
300                 {
301                     bs_write_se( s, sh->weight[i][j].i_scale );
302                     bs_write_se( s, sh->weight[i][j].i_offset );
303                 }
304             }
305         }
306     }
307     else if( sh->pps->b_weighted_bipred == 1 && sh->i_type == SLICE_TYPE_B )
308     {
309       /* TODO */
310     }
311
312     if( i_nal_ref_idc != 0 )
313     {
314         if( sh->i_idr_pic_id >= 0 )
315         {
316             bs_write1( s, 0 );  /* no output of prior pics flag */
317             bs_write1( s, 0 );  /* long term reference flag */
318         }
319         else
320         {
321             bs_write1( s, sh->i_mmco_command_count > 0 ); /* adaptive_ref_pic_marking_mode_flag */
322             if( sh->i_mmco_command_count > 0 )
323             {
324                 for( int i = 0; i < sh->i_mmco_command_count; i++ )
325                 {
326                     bs_write_ue( s, 1 ); /* mark short term ref as unused */
327                     bs_write_ue( s, sh->mmco[i].i_difference_of_pic_nums - 1 );
328                 }
329                 bs_write_ue( s, 0 ); /* end command list */
330             }
331         }
332     }
333
334     if( sh->pps->b_cabac && sh->i_type != SLICE_TYPE_I )
335         bs_write_ue( s, sh->i_cabac_init_idc );
336
337     bs_write_se( s, sh->i_qp_delta );      /* slice qp delta */
338
339     if( sh->pps->b_deblocking_filter_control )
340     {
341         bs_write_ue( s, sh->i_disable_deblocking_filter_idc );
342         if( sh->i_disable_deblocking_filter_idc != 1 )
343         {
344             bs_write_se( s, sh->i_alpha_c0_offset >> 1 );
345             bs_write_se( s, sh->i_beta_offset >> 1 );
346         }
347     }
348 }
349
350 /* If we are within a reasonable distance of the end of the memory allocated for the bitstream, */
351 /* reallocate, adding an arbitrary amount of space. */
352 static int x264_bitstream_check_buffer_internal( x264_t *h, int size, int b_cabac, int i_nal )
353 {
354     if( (b_cabac && (h->cabac.p_end - h->cabac.p < size)) ||
355         (h->out.bs.p_end - h->out.bs.p < size) )
356     {
357         int buf_size = h->out.i_bitstream + size;
358         uint8_t *buf = x264_malloc( buf_size );
359         if( !buf )
360             return -1;
361         int aligned_size = h->out.i_bitstream & ~15;
362         h->mc.memcpy_aligned( buf, h->out.p_bitstream, aligned_size );
363         memcpy( buf + aligned_size, h->out.p_bitstream + aligned_size, h->out.i_bitstream - aligned_size );
364
365         intptr_t delta = buf - h->out.p_bitstream;
366
367         h->out.bs.p_start += delta;
368         h->out.bs.p += delta;
369         h->out.bs.p_end = buf + buf_size;
370
371         h->cabac.p_start += delta;
372         h->cabac.p += delta;
373         h->cabac.p_end = buf + buf_size;
374
375         for( int i = 0; i <= i_nal; i++ )
376             h->out.nal[i].p_payload += delta;
377
378         x264_free( h->out.p_bitstream );
379         h->out.p_bitstream = buf;
380         h->out.i_bitstream = buf_size;
381     }
382     return 0;
383 }
384
385 static int x264_bitstream_check_buffer( x264_t *h )
386 {
387     int max_row_size = (2500 << SLICE_MBAFF) * h->mb.i_mb_width;
388     return x264_bitstream_check_buffer_internal( h, max_row_size, h->param.b_cabac, h->out.i_nal );
389 }
390
391 static int x264_bitstream_check_buffer_filler( x264_t *h, int filler )
392 {
393     filler += 32; // add padding for safety
394     return x264_bitstream_check_buffer_internal( h, filler, 0, -1 );
395 }
396
397 #if HAVE_THREAD
398 static void x264_encoder_thread_init( x264_t *h )
399 {
400     if( h->param.i_sync_lookahead )
401         x264_lower_thread_priority( 10 );
402 }
403 #endif
404
405 /****************************************************************************
406  *
407  ****************************************************************************
408  ****************************** External API*********************************
409  ****************************************************************************
410  *
411  ****************************************************************************/
412
413 static int x264_validate_parameters( x264_t *h, int b_open )
414 {
415 #if HAVE_MMX
416     if( b_open )
417     {
418         int cpuflags = x264_cpu_detect();
419         int fail = 0;
420 #ifdef __SSE__
421         if( !(cpuflags & X264_CPU_SSE) )
422         {
423             x264_log( h, X264_LOG_ERROR, "your cpu does not support SSE1, but x264 was compiled with asm\n");
424             fail = 1;
425         }
426 #else
427         if( !(cpuflags & X264_CPU_MMX2) )
428         {
429             x264_log( h, X264_LOG_ERROR, "your cpu does not support MMXEXT, but x264 was compiled with asm\n");
430             fail = 1;
431         }
432 #endif
433         if( !fail && !(cpuflags & X264_CPU_CMOV) )
434         {
435             x264_log( h, X264_LOG_ERROR, "your cpu does not support CMOV, but x264 was compiled with asm\n");
436             fail = 1;
437         }
438         if( fail )
439         {
440             x264_log( h, X264_LOG_ERROR, "to run x264, recompile without asm (configure --disable-asm)\n");
441             return -1;
442         }
443     }
444 #endif
445
446 #if HAVE_INTERLACED
447     h->param.b_interlaced = !!PARAM_INTERLACED;
448 #else
449     if( h->param.b_interlaced )
450     {
451         x264_log( h, X264_LOG_ERROR, "not compiled with interlaced support\n" );
452         return -1;
453     }
454 #endif
455
456     if( h->param.i_width <= 0 || h->param.i_height <= 0 )
457     {
458         x264_log( h, X264_LOG_ERROR, "invalid width x height (%dx%d)\n",
459                   h->param.i_width, h->param.i_height );
460         return -1;
461     }
462
463     int i_csp = h->param.i_csp & X264_CSP_MASK;
464 #if X264_CHROMA_FORMAT
465     if( CHROMA_FORMAT != CHROMA_420 && i_csp >= X264_CSP_I420 && i_csp <= X264_CSP_NV12 )
466     {
467         x264_log( h, X264_LOG_ERROR, "not compiled with 4:2:0 support\n" );
468         return -1;
469     }
470     else if( CHROMA_FORMAT != CHROMA_422 && i_csp >= X264_CSP_I422 && i_csp <= X264_CSP_NV16 )
471     {
472         x264_log( h, X264_LOG_ERROR, "not compiled with 4:2:2 support\n" );
473         return -1;
474     }
475     else if( CHROMA_FORMAT != CHROMA_444 && i_csp >= X264_CSP_I444 && i_csp <= X264_CSP_RGB )
476     {
477         x264_log( h, X264_LOG_ERROR, "not compiled with 4:4:4 support\n" );
478         return -1;
479     }
480 #endif
481     if( i_csp <= X264_CSP_NONE || i_csp >= X264_CSP_MAX )
482     {
483         x264_log( h, X264_LOG_ERROR, "invalid CSP (only I420/YV12/NV12/I422/YV16/NV16/I444/YV24/BGR/BGRA/RGB supported)\n" );
484         return -1;
485     }
486
487     if( i_csp < X264_CSP_I444 && h->param.i_width % 2 )
488     {
489         x264_log( h, X264_LOG_ERROR, "width not divisible by 2 (%dx%d)\n",
490                   h->param.i_width, h->param.i_height );
491         return -1;
492     }
493
494     if( i_csp < X264_CSP_I422 && PARAM_INTERLACED && h->param.i_height % 4 )
495     {
496         x264_log( h, X264_LOG_ERROR, "height not divisible by 4 (%dx%d)\n",
497                   h->param.i_width, h->param.i_height );
498         return -1;
499     }
500
501     if( (i_csp < X264_CSP_I422 || PARAM_INTERLACED) && h->param.i_height % 2 )
502     {
503         x264_log( h, X264_LOG_ERROR, "height not divisible by 2 (%dx%d)\n",
504                   h->param.i_width, h->param.i_height );
505         return -1;
506     }
507
508     if( (h->param.crop_rect.i_left + h->param.crop_rect.i_right ) >= h->param.i_width ||
509         (h->param.crop_rect.i_top  + h->param.crop_rect.i_bottom) >= h->param.i_height )
510     {
511         x264_log( h, X264_LOG_ERROR, "invalid crop-rect %u,%u,%u,%u\n", h->param.crop_rect.i_left,
512                   h->param.crop_rect.i_top, h->param.crop_rect.i_right,  h->param.crop_rect.i_bottom );
513         return -1;
514     }
515
516     if( h->param.vui.i_sar_width <= 0 || h->param.vui.i_sar_height <= 0 )
517     {
518         h->param.vui.i_sar_width = 0;
519         h->param.vui.i_sar_height = 0;
520     }
521
522     if( h->param.i_threads == X264_THREADS_AUTO )
523         h->param.i_threads = x264_cpu_num_processors() * (h->param.b_sliced_threads?2:3)/2;
524     int max_sliced_threads = X264_MAX( 1, (h->param.i_height+15)/16 / 4 );
525     if( h->param.i_threads > 1 )
526     {
527 #if !HAVE_THREAD
528         x264_log( h, X264_LOG_WARNING, "not compiled with thread support!\n");
529         h->param.i_threads = 1;
530 #endif
531         /* Avoid absurdly small thread slices as they can reduce performance
532          * and VBV compliance.  Capped at an arbitrary 4 rows per thread. */
533         if( h->param.b_sliced_threads )
534             h->param.i_threads = X264_MIN( h->param.i_threads, max_sliced_threads );
535     }
536     h->param.i_threads = x264_clip3( h->param.i_threads, 1, X264_THREAD_MAX );
537     if( h->param.i_threads == 1 )
538     {
539         h->param.b_sliced_threads = 0;
540         h->param.i_lookahead_threads = 1;
541     }
542     h->i_thread_frames = h->param.b_sliced_threads ? 1 : h->param.i_threads;
543     if( h->i_thread_frames > 1 )
544         h->param.nalu_process = NULL;
545
546     if( h->param.b_opencl )
547     {
548 #if !HAVE_OPENCL
549         x264_log( h, X264_LOG_WARNING, "OpenCL: not compiled with OpenCL support, disabling\n" );
550         h->param.b_opencl = 0;
551 #elif BIT_DEPTH > 8
552         x264_log( h, X264_LOG_WARNING, "OpenCL lookahead does not support high bit depth, disabling opencl\n" );
553         h->param.b_opencl = 0;
554 #else
555         if( h->param.i_width < 32 || h->param.i_height < 32 )
556         {
557             x264_log( h, X264_LOG_WARNING, "OpenCL: frame size is too small, disabling opencl\n" );
558             h->param.b_opencl = 0;
559         }
560 #endif
561         if( h->param.opencl_device_id && h->param.i_opencl_device )
562         {
563             x264_log( h, X264_LOG_WARNING, "OpenCL: device id and device skip count configured; dropping skip\n" );
564             h->param.i_opencl_device = 0;
565         }
566     }
567
568     h->param.i_keyint_max = x264_clip3( h->param.i_keyint_max, 1, X264_KEYINT_MAX_INFINITE );
569     if( h->param.i_keyint_max == 1 )
570     {
571         h->param.b_intra_refresh = 0;
572         h->param.analyse.i_weighted_pred = 0;
573         h->param.i_frame_reference = 1;
574         h->param.i_dpb_size = 1;
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
609     if( PARAM_INTERLACED )
610         h->param.b_pic_struct = 1;
611
612     if( h->param.b_avcintra_compat )
613     {
614         if( BIT_DEPTH != 10 )
615         {
616             x264_log( h, X264_LOG_ERROR, "%2d-bit AVC-Intra is not widely compatible\n", BIT_DEPTH );
617             x264_log( h, X264_LOG_ERROR, "10-bit x264 is required to encode AVC-Intra\n" );
618             return -1;
619         }
620
621         /* [50/100][res][fps] */
622         static const struct
623         {
624             uint16_t fps_num;
625             uint16_t fps_den;
626             uint8_t interlaced;
627             uint16_t frame_size;
628             const uint8_t *cqm_4ic;
629             const uint8_t *cqm_8iy;
630         } avcintra_lut[2][2][7] =
631         {
632             {{{ 60000, 1001, 0,  912, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy },
633               {    50,    1, 0, 1100, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy },
634               { 30000, 1001, 0,  912, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy },
635               {    25,    1, 0, 1100, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy },
636               { 24000, 1001, 0,  912, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy }},
637              {{ 30000, 1001, 1, 1820, x264_cqm_avci50_4ic, x264_cqm_avci50_1080i_8iy },
638               {    25,    1, 1, 2196, x264_cqm_avci50_4ic, x264_cqm_avci50_1080i_8iy },
639               { 60000, 1001, 0, 1820, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy },
640               { 30000, 1001, 0, 1820, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy },
641               {    50,    1, 0, 2196, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy },
642               {    25,    1, 0, 2196, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy },
643               { 24000, 1001, 0, 1820, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy }}},
644             {{{ 60000, 1001, 0, 1848, x264_cqm_avci100_720p_4ic, x264_cqm_avci100_720p_8iy },
645               {    50,    1, 0, 2224, x264_cqm_avci100_720p_4ic, x264_cqm_avci100_720p_8iy },
646               { 30000, 1001, 0, 1848, x264_cqm_avci100_720p_4ic, x264_cqm_avci100_720p_8iy },
647               {    25,    1, 0, 2224, x264_cqm_avci100_720p_4ic, x264_cqm_avci100_720p_8iy },
648               { 24000, 1001, 0, 1848, x264_cqm_avci100_720p_4ic, x264_cqm_avci100_720p_8iy }},
649              {{ 30000, 1001, 1, 3692, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080i_8iy },
650               {    25,    1, 1, 4444, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080i_8iy },
651               { 60000, 1001, 0, 3692, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080p_8iy },
652               { 30000, 1001, 0, 3692, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080p_8iy },
653               {    50,    1, 0, 4444, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080p_8iy },
654               {    25,    1, 0, 4444, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080p_8iy },
655               { 24000, 1001, 0, 3692, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080p_8iy }}}
656         };
657
658         int res = -1;
659         int type = i_csp == X264_CSP_I422;
660         if( type )
661         {
662             if(      h->param.i_width == 1920 && h->param.i_height == 1080 ) res =  1;
663             else if( h->param.i_width == 1280 && h->param.i_height ==  720 ) res =  0;
664         }
665         else if( i_csp == X264_CSP_I420 )
666         {
667             if(      h->param.i_width == 1440 && h->param.i_height == 1080 ) res =  1;
668             else if( h->param.i_width ==  960 && h->param.i_height ==  720 ) res =  0;
669         }
670         else
671         {
672             x264_log( h, X264_LOG_ERROR, "Invalid colorspace for AVC-Intra\n" );
673             return -1;
674         }
675
676         if( res < 0 )
677         {
678             x264_log( h, X264_LOG_ERROR, "Resolution %dx%d invalid for AVC-Intra %s\n",
679                       h->param.i_width, h->param.i_height, type ? "100" : "50" );
680             return -1;
681         }
682
683         if( h->param.nalu_process )
684         {
685             x264_log( h, X264_LOG_ERROR, "nalu_process is not supported in AVC-Intra mode\n" );
686             return -1;
687         }
688
689         if( !h->param.b_repeat_headers )
690         {
691             x264_log( h, X264_LOG_ERROR, "Separate headers not supported in AVC-Intra mode\n" );
692             return -1;
693         }
694
695         int i;
696         uint32_t fps_num = h->param.i_fps_num, fps_den = h->param.i_fps_den;
697         x264_reduce_fraction( &fps_num, &fps_den );
698         for( i = 0; i < 7; i++ )
699         {
700             if( avcintra_lut[type][res][i].fps_num == fps_num &&
701                 avcintra_lut[type][res][i].fps_den == fps_den &&
702                 avcintra_lut[type][res][i].interlaced == PARAM_INTERLACED )
703             {
704                 break;
705             }
706         }
707         if( i == 7 )
708         {
709             x264_log( h, X264_LOG_ERROR, "FPS %d/%d%c not compatible with AVC-Intra\n",
710                       h->param.i_fps_num, h->param.i_fps_den, PARAM_INTERLACED ? 'i' : 'p' );
711             return -1;
712         }
713
714         h->param.i_keyint_max = 1;
715         h->param.b_intra_refresh = 0;
716         h->param.analyse.i_weighted_pred = 0;
717         h->param.i_frame_reference = 1;
718         h->param.i_dpb_size = 1;
719
720         h->param.b_bluray_compat = 0;
721         h->param.b_vfr_input = 0;
722         h->param.b_aud = 1;
723         h->param.vui.i_chroma_loc = 0;
724         h->param.i_nal_hrd = X264_NAL_HRD_NONE;
725         h->param.b_deblocking_filter = 0;
726         h->param.b_stitchable = 1;
727         h->param.b_pic_struct = 0;
728         h->param.analyse.b_transform_8x8 = 1;
729         h->param.analyse.intra = X264_ANALYSE_I8x8;
730         h->param.analyse.i_chroma_qp_offset = res && type ? 3 : 4;
731         h->param.b_cabac = !type;
732         h->param.rc.i_vbv_buffer_size = avcintra_lut[type][res][i].frame_size;
733         h->param.rc.i_vbv_max_bitrate =
734         h->param.rc.i_bitrate = h->param.rc.i_vbv_buffer_size * fps_num / fps_den;
735         h->param.rc.i_rc_method = X264_RC_ABR;
736         h->param.rc.f_vbv_buffer_init = 1.0;
737         h->param.rc.b_filler = 1;
738         h->param.i_cqm_preset = X264_CQM_CUSTOM;
739         memcpy( h->param.cqm_4iy, x264_cqm_jvt4i, sizeof(h->param.cqm_4iy) );
740         memcpy( h->param.cqm_4ic, avcintra_lut[type][res][i].cqm_4ic, sizeof(h->param.cqm_4ic) );
741         memcpy( h->param.cqm_8iy, avcintra_lut[type][res][i].cqm_8iy, sizeof(h->param.cqm_8iy) );
742
743         /* Need exactly 10 slices of equal MB count... why?  $deity knows... */
744         h->param.i_slice_max_mbs = ((h->param.i_width + 15) / 16) * ((h->param.i_height + 15) / 16) / 10;
745         h->param.i_slice_max_size = 0;
746         /* The slice structure only allows a maximum of 2 threads for 1080i/p
747          * and 1 or 5 threads for 720p */
748         if( h->param.b_sliced_threads )
749         {
750             if( res )
751                 h->param.i_threads = X264_MIN( 2, h->param.i_threads );
752             else
753             {
754                 h->param.i_threads = X264_MIN( 5, h->param.i_threads );
755                 if( h->param.i_threads < 5 )
756                     h->param.i_threads = 1;
757             }
758         }
759
760         if( type )
761             h->param.vui.i_sar_width = h->param.vui.i_sar_height = 1;
762         else
763         {
764             h->param.vui.i_sar_width  = 4;
765             h->param.vui.i_sar_height = 3;
766         }
767
768         /* Avid cannot handle negative QPs */
769         h->param.rc.i_qp_min = X264_MAX( h->param.rc.i_qp_min, QP_BD_OFFSET );
770     }
771
772     h->param.rc.f_rf_constant = x264_clip3f( h->param.rc.f_rf_constant, -QP_BD_OFFSET, 51 );
773     h->param.rc.f_rf_constant_max = x264_clip3f( h->param.rc.f_rf_constant_max, -QP_BD_OFFSET, 51 );
774     h->param.rc.i_qp_constant = x264_clip3( h->param.rc.i_qp_constant, 0, QP_MAX );
775     h->param.analyse.i_subpel_refine = x264_clip3( h->param.analyse.i_subpel_refine, 0, 11 );
776     h->param.rc.f_ip_factor = X264_MAX( h->param.rc.f_ip_factor, 0.01f );
777     h->param.rc.f_pb_factor = X264_MAX( h->param.rc.f_pb_factor, 0.01f );
778     if( h->param.rc.i_rc_method == X264_RC_CRF )
779     {
780         h->param.rc.i_qp_constant = h->param.rc.f_rf_constant + QP_BD_OFFSET;
781         h->param.rc.i_bitrate = 0;
782     }
783     if( b_open && (h->param.rc.i_rc_method == X264_RC_CQP || h->param.rc.i_rc_method == X264_RC_CRF)
784         && h->param.rc.i_qp_constant == 0 )
785     {
786         h->mb.b_lossless = 1;
787         h->param.i_cqm_preset = X264_CQM_FLAT;
788         h->param.psz_cqm_file = NULL;
789         h->param.rc.i_rc_method = X264_RC_CQP;
790         h->param.rc.f_ip_factor = 1;
791         h->param.rc.f_pb_factor = 1;
792         h->param.analyse.b_psnr = 0;
793         h->param.analyse.b_ssim = 0;
794         h->param.analyse.i_chroma_qp_offset = 0;
795         h->param.analyse.i_trellis = 0;
796         h->param.analyse.b_fast_pskip = 0;
797         h->param.analyse.i_noise_reduction = 0;
798         h->param.analyse.b_psy = 0;
799         h->param.i_bframe = 0;
800         /* 8x8dct is not useful without RD in CAVLC lossless */
801         if( !h->param.b_cabac && h->param.analyse.i_subpel_refine < 6 )
802             h->param.analyse.b_transform_8x8 = 0;
803     }
804     if( h->param.rc.i_rc_method == X264_RC_CQP )
805     {
806         float qp_p = h->param.rc.i_qp_constant;
807         float qp_i = qp_p - 6*log2f( h->param.rc.f_ip_factor );
808         float qp_b = qp_p + 6*log2f( h->param.rc.f_pb_factor );
809         h->param.rc.i_qp_min = x264_clip3( (int)(X264_MIN3( qp_p, qp_i, qp_b )), 0, QP_MAX );
810         h->param.rc.i_qp_max = x264_clip3( (int)(X264_MAX3( qp_p, qp_i, qp_b ) + .999), 0, QP_MAX );
811         h->param.rc.i_aq_mode = 0;
812         h->param.rc.b_mb_tree = 0;
813         h->param.rc.i_bitrate = 0;
814     }
815     h->param.rc.i_qp_max = x264_clip3( h->param.rc.i_qp_max, 0, QP_MAX );
816     h->param.rc.i_qp_min = x264_clip3( h->param.rc.i_qp_min, 0, h->param.rc.i_qp_max );
817     h->param.rc.i_qp_step = x264_clip3( h->param.rc.i_qp_step, 2, QP_MAX );
818     h->param.rc.i_bitrate = x264_clip3( h->param.rc.i_bitrate, 0, 2000000 );
819     if( h->param.rc.i_rc_method == X264_RC_ABR && !h->param.rc.i_bitrate )
820     {
821         x264_log( h, X264_LOG_ERROR, "bitrate not specified\n" );
822         return -1;
823     }
824     h->param.rc.i_vbv_buffer_size = x264_clip3( h->param.rc.i_vbv_buffer_size, 0, 2000000 );
825     h->param.rc.i_vbv_max_bitrate = x264_clip3( h->param.rc.i_vbv_max_bitrate, 0, 2000000 );
826     h->param.rc.f_vbv_buffer_init = x264_clip3f( h->param.rc.f_vbv_buffer_init, 0, 2000000 );
827     if( h->param.rc.i_vbv_buffer_size )
828     {
829         if( h->param.rc.i_rc_method == X264_RC_CQP )
830         {
831             x264_log( h, X264_LOG_WARNING, "VBV is incompatible with constant QP, ignored.\n" );
832             h->param.rc.i_vbv_max_bitrate = 0;
833             h->param.rc.i_vbv_buffer_size = 0;
834         }
835         else if( h->param.rc.i_vbv_max_bitrate == 0 )
836         {
837             if( h->param.rc.i_rc_method == X264_RC_ABR )
838             {
839                 x264_log( h, X264_LOG_WARNING, "VBV maxrate unspecified, assuming CBR\n" );
840                 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate;
841             }
842             else
843             {
844                 x264_log( h, X264_LOG_WARNING, "VBV bufsize set but maxrate unspecified, ignored\n" );
845                 h->param.rc.i_vbv_buffer_size = 0;
846             }
847         }
848         else if( h->param.rc.i_vbv_max_bitrate < h->param.rc.i_bitrate &&
849                  h->param.rc.i_rc_method == X264_RC_ABR )
850         {
851             x264_log( h, X264_LOG_WARNING, "max bitrate less than average bitrate, assuming CBR\n" );
852             h->param.rc.i_bitrate = h->param.rc.i_vbv_max_bitrate;
853         }
854     }
855     else if( h->param.rc.i_vbv_max_bitrate )
856     {
857         x264_log( h, X264_LOG_WARNING, "VBV maxrate specified, but no bufsize, ignored\n" );
858         h->param.rc.i_vbv_max_bitrate = 0;
859     }
860
861     h->param.i_slice_max_size = X264_MAX( h->param.i_slice_max_size, 0 );
862     h->param.i_slice_max_mbs = X264_MAX( h->param.i_slice_max_mbs, 0 );
863     h->param.i_slice_min_mbs = X264_MAX( h->param.i_slice_min_mbs, 0 );
864     if( h->param.i_slice_max_mbs )
865         h->param.i_slice_min_mbs = X264_MIN( h->param.i_slice_min_mbs, h->param.i_slice_max_mbs/2 );
866     else if( !h->param.i_slice_max_size )
867         h->param.i_slice_min_mbs = 0;
868     if( PARAM_INTERLACED && h->param.i_slice_min_mbs )
869     {
870         x264_log( h, X264_LOG_WARNING, "interlace + slice-min-mbs is not implemented\n" );
871         h->param.i_slice_min_mbs = 0;
872     }
873     int mb_width = (h->param.i_width+15)/16;
874     if( h->param.i_slice_min_mbs > mb_width )
875     {
876         x264_log( h, X264_LOG_WARNING, "slice-min-mbs > row mb size (%d) not implemented\n", mb_width );
877         h->param.i_slice_min_mbs = mb_width;
878     }
879
880     int max_slices = (h->param.i_height+((16<<PARAM_INTERLACED)-1))/(16<<PARAM_INTERLACED);
881     if( h->param.b_sliced_threads )
882         h->param.i_slice_count = x264_clip3( h->param.i_threads, 0, max_slices );
883     else
884     {
885         h->param.i_slice_count = x264_clip3( h->param.i_slice_count, 0, max_slices );
886         if( h->param.i_slice_max_mbs || h->param.i_slice_max_size )
887             h->param.i_slice_count = 0;
888     }
889     if( h->param.i_slice_count_max > 0 )
890         h->param.i_slice_count_max = X264_MAX( h->param.i_slice_count, h->param.i_slice_count_max );
891
892     if( h->param.b_bluray_compat )
893     {
894         h->param.i_bframe_pyramid = X264_MIN( X264_B_PYRAMID_STRICT, h->param.i_bframe_pyramid );
895         h->param.i_bframe = X264_MIN( h->param.i_bframe, 3 );
896         h->param.b_aud = 1;
897         h->param.i_nal_hrd = X264_MAX( h->param.i_nal_hrd, X264_NAL_HRD_VBR );
898         h->param.i_slice_max_size = 0;
899         h->param.i_slice_max_mbs = 0;
900         h->param.b_intra_refresh = 0;
901         h->param.i_frame_reference = X264_MIN( h->param.i_frame_reference, 6 );
902         h->param.i_dpb_size = X264_MIN( h->param.i_dpb_size, 6 );
903         /* Don't use I-frames, because Blu-ray treats them the same as IDR. */
904         h->param.i_keyint_min = 1;
905         /* Due to the proliferation of broken players that don't handle dupes properly. */
906         h->param.analyse.i_weighted_pred = X264_MIN( h->param.analyse.i_weighted_pred, X264_WEIGHTP_SIMPLE );
907         if( h->param.b_fake_interlaced )
908             h->param.b_pic_struct = 1;
909     }
910
911     h->param.i_frame_reference = x264_clip3( h->param.i_frame_reference, 1, X264_REF_MAX );
912     h->param.i_dpb_size = x264_clip3( h->param.i_dpb_size, 1, X264_REF_MAX );
913     if( h->param.i_scenecut_threshold < 0 )
914         h->param.i_scenecut_threshold = 0;
915     h->param.analyse.i_direct_mv_pred = x264_clip3( h->param.analyse.i_direct_mv_pred, X264_DIRECT_PRED_NONE, X264_DIRECT_PRED_AUTO );
916     if( !h->param.analyse.i_subpel_refine && h->param.analyse.i_direct_mv_pred > X264_DIRECT_PRED_SPATIAL )
917     {
918         x264_log( h, X264_LOG_WARNING, "subme=0 + direct=temporal is not supported\n" );
919         h->param.analyse.i_direct_mv_pred = X264_DIRECT_PRED_SPATIAL;
920     }
921     h->param.i_bframe = x264_clip3( h->param.i_bframe, 0, X264_MIN( X264_BFRAME_MAX, h->param.i_keyint_max-1 ) );
922     h->param.i_bframe_bias = x264_clip3( h->param.i_bframe_bias, -90, 100 );
923     if( h->param.i_bframe <= 1 )
924         h->param.i_bframe_pyramid = X264_B_PYRAMID_NONE;
925     h->param.i_bframe_pyramid = x264_clip3( h->param.i_bframe_pyramid, X264_B_PYRAMID_NONE, X264_B_PYRAMID_NORMAL );
926     h->param.i_bframe_adaptive = x264_clip3( h->param.i_bframe_adaptive, X264_B_ADAPT_NONE, X264_B_ADAPT_TRELLIS );
927     if( !h->param.i_bframe )
928     {
929         h->param.i_bframe_adaptive = X264_B_ADAPT_NONE;
930         h->param.analyse.i_direct_mv_pred = 0;
931         h->param.analyse.b_weighted_bipred = 0;
932         h->param.b_open_gop = 0;
933     }
934     if( h->param.b_intra_refresh && h->param.i_bframe_pyramid == X264_B_PYRAMID_NORMAL )
935     {
936         x264_log( h, X264_LOG_WARNING, "b-pyramid normal + intra-refresh is not supported\n" );
937         h->param.i_bframe_pyramid = X264_B_PYRAMID_STRICT;
938     }
939     if( h->param.b_intra_refresh && (h->param.i_frame_reference > 1 || h->param.i_dpb_size > 1) )
940     {
941         x264_log( h, X264_LOG_WARNING, "ref > 1 + intra-refresh is not supported\n" );
942         h->param.i_frame_reference = 1;
943         h->param.i_dpb_size = 1;
944     }
945     if( h->param.b_intra_refresh && h->param.b_open_gop )
946     {
947         x264_log( h, X264_LOG_WARNING, "intra-refresh is not compatible with open-gop\n" );
948         h->param.b_open_gop = 0;
949     }
950     if( !h->param.i_fps_num || !h->param.i_fps_den )
951     {
952         h->param.i_fps_num = 25;
953         h->param.i_fps_den = 1;
954     }
955     float fps = (float) h->param.i_fps_num / h->param.i_fps_den;
956     if( h->param.i_keyint_min == X264_KEYINT_MIN_AUTO )
957         h->param.i_keyint_min = X264_MIN( h->param.i_keyint_max / 10, fps );
958     h->param.i_keyint_min = x264_clip3( h->param.i_keyint_min, 1, h->param.i_keyint_max/2+1 );
959     h->param.rc.i_lookahead = x264_clip3( h->param.rc.i_lookahead, 0, X264_LOOKAHEAD_MAX );
960     {
961         int maxrate = X264_MAX( h->param.rc.i_vbv_max_bitrate, h->param.rc.i_bitrate );
962         float bufsize = maxrate ? (float)h->param.rc.i_vbv_buffer_size / maxrate : 0;
963         h->param.rc.i_lookahead = X264_MIN( h->param.rc.i_lookahead, X264_MAX( h->param.i_keyint_max, bufsize*fps ) );
964     }
965
966     if( !h->param.i_timebase_num || !h->param.i_timebase_den || !(h->param.b_vfr_input || h->param.b_pulldown) )
967     {
968         h->param.i_timebase_num = h->param.i_fps_den;
969         h->param.i_timebase_den = h->param.i_fps_num;
970     }
971
972     h->param.rc.f_qcompress = x264_clip3f( h->param.rc.f_qcompress, 0.0, 1.0 );
973     if( h->param.i_keyint_max == 1 || h->param.rc.f_qcompress == 1 )
974         h->param.rc.b_mb_tree = 0;
975     if( (!h->param.b_intra_refresh && h->param.i_keyint_max != X264_KEYINT_MAX_INFINITE) &&
976         !h->param.rc.i_lookahead && h->param.rc.b_mb_tree )
977     {
978         x264_log( h, X264_LOG_WARNING, "lookaheadless mb-tree requires intra refresh or infinite keyint\n" );
979         h->param.rc.b_mb_tree = 0;
980     }
981     if( b_open && h->param.rc.b_stat_read )
982         h->param.rc.i_lookahead = 0;
983 #if HAVE_THREAD
984     if( h->param.i_sync_lookahead < 0 )
985         h->param.i_sync_lookahead = h->param.i_bframe + 1;
986     h->param.i_sync_lookahead = X264_MIN( h->param.i_sync_lookahead, X264_LOOKAHEAD_MAX );
987     if( h->param.rc.b_stat_read || h->i_thread_frames == 1 )
988         h->param.i_sync_lookahead = 0;
989 #else
990     h->param.i_sync_lookahead = 0;
991 #endif
992
993     h->param.i_deblocking_filter_alphac0 = x264_clip3( h->param.i_deblocking_filter_alphac0, -6, 6 );
994     h->param.i_deblocking_filter_beta    = x264_clip3( h->param.i_deblocking_filter_beta, -6, 6 );
995     h->param.analyse.i_luma_deadzone[0] = x264_clip3( h->param.analyse.i_luma_deadzone[0], 0, 32 );
996     h->param.analyse.i_luma_deadzone[1] = x264_clip3( h->param.analyse.i_luma_deadzone[1], 0, 32 );
997
998     h->param.i_cabac_init_idc = x264_clip3( h->param.i_cabac_init_idc, 0, 2 );
999
1000     if( h->param.i_cqm_preset < X264_CQM_FLAT || h->param.i_cqm_preset > X264_CQM_CUSTOM )
1001         h->param.i_cqm_preset = X264_CQM_FLAT;
1002
1003     if( h->param.analyse.i_me_method < X264_ME_DIA ||
1004         h->param.analyse.i_me_method > X264_ME_TESA )
1005         h->param.analyse.i_me_method = X264_ME_HEX;
1006     h->param.analyse.i_me_range = x264_clip3( h->param.analyse.i_me_range, 4, 1024 );
1007     if( h->param.analyse.i_me_range > 16 && h->param.analyse.i_me_method <= X264_ME_HEX )
1008         h->param.analyse.i_me_range = 16;
1009     if( h->param.analyse.i_me_method == X264_ME_TESA &&
1010         (h->mb.b_lossless || h->param.analyse.i_subpel_refine <= 1) )
1011         h->param.analyse.i_me_method = X264_ME_ESA;
1012     h->param.analyse.b_mixed_references = h->param.analyse.b_mixed_references && h->param.i_frame_reference > 1;
1013     h->param.analyse.inter &= X264_ANALYSE_PSUB16x16|X264_ANALYSE_PSUB8x8|X264_ANALYSE_BSUB16x16|
1014                               X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
1015     h->param.analyse.intra &= X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
1016     if( !(h->param.analyse.inter & X264_ANALYSE_PSUB16x16) )
1017         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
1018     if( !h->param.analyse.b_transform_8x8 )
1019     {
1020         h->param.analyse.inter &= ~X264_ANALYSE_I8x8;
1021         h->param.analyse.intra &= ~X264_ANALYSE_I8x8;
1022     }
1023     h->param.analyse.i_trellis = x264_clip3( h->param.analyse.i_trellis, 0, 2 );
1024     h->param.rc.i_aq_mode = x264_clip3( h->param.rc.i_aq_mode, 0, 2 );
1025     h->param.rc.f_aq_strength = x264_clip3f( h->param.rc.f_aq_strength, 0, 3 );
1026     if( h->param.rc.f_aq_strength == 0 )
1027         h->param.rc.i_aq_mode = 0;
1028
1029     if( h->param.i_log_level < X264_LOG_INFO )
1030     {
1031         h->param.analyse.b_psnr = 0;
1032         h->param.analyse.b_ssim = 0;
1033     }
1034     /* Warn users trying to measure PSNR/SSIM with psy opts on. */
1035     if( b_open && (h->param.analyse.b_psnr || h->param.analyse.b_ssim) )
1036     {
1037         char *s = NULL;
1038
1039         if( h->param.analyse.b_psy )
1040         {
1041             s = h->param.analyse.b_psnr ? "psnr" : "ssim";
1042             x264_log( h, X264_LOG_WARNING, "--%s used with psy on: results will be invalid!\n", s );
1043         }
1044         else if( !h->param.rc.i_aq_mode && h->param.analyse.b_ssim )
1045         {
1046             x264_log( h, X264_LOG_WARNING, "--ssim used with AQ off: results will be invalid!\n" );
1047             s = "ssim";
1048         }
1049         else if(  h->param.rc.i_aq_mode && h->param.analyse.b_psnr )
1050         {
1051             x264_log( h, X264_LOG_WARNING, "--psnr used with AQ on: results will be invalid!\n" );
1052             s = "psnr";
1053         }
1054         if( s )
1055             x264_log( h, X264_LOG_WARNING, "--tune %s should be used if attempting to benchmark %s!\n", s, s );
1056     }
1057
1058     if( !h->param.analyse.b_psy )
1059     {
1060         h->param.analyse.f_psy_rd = 0;
1061         h->param.analyse.f_psy_trellis = 0;
1062     }
1063     h->param.analyse.f_psy_rd = x264_clip3f( h->param.analyse.f_psy_rd, 0, 10 );
1064     h->param.analyse.f_psy_trellis = x264_clip3f( h->param.analyse.f_psy_trellis, 0, 10 );
1065     h->mb.i_psy_rd = h->param.analyse.i_subpel_refine >= 6 ? FIX8( h->param.analyse.f_psy_rd ) : 0;
1066     h->mb.i_psy_trellis = h->param.analyse.i_trellis ? FIX8( h->param.analyse.f_psy_trellis / 4 ) : 0;
1067     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -32, 32);
1068     /* In 4:4:4 mode, chroma gets twice as much resolution, so we can halve its quality. */
1069     if( b_open && i_csp >= X264_CSP_I444 && i_csp < X264_CSP_BGR && h->param.analyse.b_psy )
1070         h->param.analyse.i_chroma_qp_offset += 6;
1071     /* Psy RDO increases overall quantizers to improve the quality of luma--this indirectly hurts chroma quality */
1072     /* so we lower the chroma QP offset to compensate */
1073     if( b_open && h->mb.i_psy_rd && !h->param.b_avcintra_compat )
1074         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_rd < 0.25 ? 1 : 2;
1075     /* Psy trellis has a similar effect. */
1076     if( b_open && h->mb.i_psy_trellis && !h->param.b_avcintra_compat )
1077         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_trellis < 0.25 ? 1 : 2;
1078     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -12, 12);
1079     /* MB-tree requires AQ to be on, even if the strength is zero. */
1080     if( !h->param.rc.i_aq_mode && h->param.rc.b_mb_tree )
1081     {
1082         h->param.rc.i_aq_mode = 1;
1083         h->param.rc.f_aq_strength = 0;
1084     }
1085     h->param.analyse.i_noise_reduction = x264_clip3( h->param.analyse.i_noise_reduction, 0, 1<<16 );
1086     if( h->param.analyse.i_subpel_refine >= 10 && (h->param.analyse.i_trellis != 2 || !h->param.rc.i_aq_mode) )
1087         h->param.analyse.i_subpel_refine = 9;
1088
1089     {
1090         const x264_level_t *l = x264_levels;
1091         if( h->param.i_level_idc < 0 )
1092         {
1093             int maxrate_bak = h->param.rc.i_vbv_max_bitrate;
1094             if( h->param.rc.i_rc_method == X264_RC_ABR && h->param.rc.i_vbv_buffer_size <= 0 )
1095                 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate * 2;
1096             x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
1097             do h->param.i_level_idc = l->level_idc;
1098                 while( l[1].level_idc && x264_validate_levels( h, 0 ) && l++ );
1099             h->param.rc.i_vbv_max_bitrate = maxrate_bak;
1100         }
1101         else
1102         {
1103             while( l->level_idc && l->level_idc != h->param.i_level_idc )
1104                 l++;
1105             if( l->level_idc == 0 )
1106             {
1107                 x264_log( h, X264_LOG_ERROR, "invalid level_idc: %d\n", h->param.i_level_idc );
1108                 return -1;
1109             }
1110         }
1111         if( h->param.analyse.i_mv_range <= 0 )
1112             h->param.analyse.i_mv_range = l->mv_range >> PARAM_INTERLACED;
1113         else
1114             h->param.analyse.i_mv_range = x264_clip3(h->param.analyse.i_mv_range, 32, 512 >> PARAM_INTERLACED);
1115     }
1116
1117     h->param.analyse.i_weighted_pred = x264_clip3( h->param.analyse.i_weighted_pred, X264_WEIGHTP_NONE, X264_WEIGHTP_SMART );
1118
1119     if( h->param.i_lookahead_threads == X264_THREADS_AUTO )
1120     {
1121         if( h->param.b_sliced_threads )
1122             h->param.i_lookahead_threads = h->param.i_threads;
1123         else
1124         {
1125             /* If we're using much slower lookahead settings than encoding settings, it helps a lot to use
1126              * more lookahead threads.  This typically happens in the first pass of a two-pass encode, so
1127              * try to guess at this sort of case.
1128              *
1129              * Tuned by a little bit of real encoding with the various presets. */
1130             int badapt = h->param.i_bframe_adaptive == X264_B_ADAPT_TRELLIS;
1131             int subme = X264_MIN( h->param.analyse.i_subpel_refine / 3, 3 ) + (h->param.analyse.i_subpel_refine > 1);
1132             int bframes = X264_MIN( (h->param.i_bframe - 1) / 3, 3 );
1133
1134             /* [b-adapt 0/1 vs 2][quantized subme][quantized bframes] */
1135             static const uint8_t lookahead_thread_div[2][5][4] =
1136             {{{6,6,6,6}, {3,3,3,3}, {4,4,4,4}, {6,6,6,6}, {12,12,12,12}},
1137              {{3,2,1,1}, {2,1,1,1}, {4,3,2,1}, {6,4,3,2}, {12, 9, 6, 4}}};
1138
1139             h->param.i_lookahead_threads = h->param.i_threads / lookahead_thread_div[badapt][subme][bframes];
1140             /* Since too many lookahead threads significantly degrades lookahead accuracy, limit auto
1141              * lookahead threads to about 8 macroblock rows high each at worst.  This number is chosen
1142              * pretty much arbitrarily. */
1143             h->param.i_lookahead_threads = X264_MIN( h->param.i_lookahead_threads, h->param.i_height / 128 );
1144         }
1145     }
1146     h->param.i_lookahead_threads = x264_clip3( h->param.i_lookahead_threads, 1, X264_MIN( max_sliced_threads, X264_LOOKAHEAD_THREAD_MAX ) );
1147
1148     if( PARAM_INTERLACED )
1149     {
1150         if( h->param.analyse.i_me_method >= X264_ME_ESA )
1151         {
1152             x264_log( h, X264_LOG_WARNING, "interlace + me=esa is not implemented\n" );
1153             h->param.analyse.i_me_method = X264_ME_UMH;
1154         }
1155         if( h->param.analyse.i_weighted_pred > 0 )
1156         {
1157             x264_log( h, X264_LOG_WARNING, "interlace + weightp is not implemented\n" );
1158             h->param.analyse.i_weighted_pred = X264_WEIGHTP_NONE;
1159         }
1160     }
1161
1162     if( !h->param.analyse.i_weighted_pred && h->param.rc.b_mb_tree && h->param.analyse.b_psy )
1163         h->param.analyse.i_weighted_pred = X264_WEIGHTP_FAKE;
1164
1165     if( h->i_thread_frames > 1 )
1166     {
1167         int r = h->param.analyse.i_mv_range_thread;
1168         int r2;
1169         if( r <= 0 )
1170         {
1171             // half of the available space is reserved and divided evenly among the threads,
1172             // the rest is allocated to whichever thread is far enough ahead to use it.
1173             // reserving more space increases quality for some videos, but costs more time
1174             // in thread synchronization.
1175             int max_range = (h->param.i_height + X264_THREAD_HEIGHT) / h->i_thread_frames - X264_THREAD_HEIGHT;
1176             r = max_range / 2;
1177         }
1178         r = X264_MAX( r, h->param.analyse.i_me_range );
1179         r = X264_MIN( r, h->param.analyse.i_mv_range );
1180         // round up to use the whole mb row
1181         r2 = (r & ~15) + ((-X264_THREAD_HEIGHT) & 15);
1182         if( r2 < r )
1183             r2 += 16;
1184         x264_log( h, X264_LOG_DEBUG, "using mv_range_thread = %d\n", r2 );
1185         h->param.analyse.i_mv_range_thread = r2;
1186     }
1187
1188     if( h->param.rc.f_rate_tolerance < 0 )
1189         h->param.rc.f_rate_tolerance = 0;
1190     if( h->param.rc.f_qblur < 0 )
1191         h->param.rc.f_qblur = 0;
1192     if( h->param.rc.f_complexity_blur < 0 )
1193         h->param.rc.f_complexity_blur = 0;
1194
1195     h->param.i_sps_id &= 31;
1196
1197     h->param.i_nal_hrd = x264_clip3( h->param.i_nal_hrd, X264_NAL_HRD_NONE, X264_NAL_HRD_CBR );
1198
1199     if( h->param.i_nal_hrd && !h->param.rc.i_vbv_buffer_size )
1200     {
1201         x264_log( h, X264_LOG_WARNING, "NAL HRD parameters require VBV parameters\n" );
1202         h->param.i_nal_hrd = X264_NAL_HRD_NONE;
1203     }
1204
1205     if( h->param.i_nal_hrd == X264_NAL_HRD_CBR &&
1206        (h->param.rc.i_bitrate != h->param.rc.i_vbv_max_bitrate || !h->param.rc.i_vbv_max_bitrate) )
1207     {
1208         x264_log( h, X264_LOG_WARNING, "CBR HRD requires constant bitrate\n" );
1209         h->param.i_nal_hrd = X264_NAL_HRD_VBR;
1210     }
1211
1212     if( h->param.i_nal_hrd == X264_NAL_HRD_CBR )
1213         h->param.rc.b_filler = 1;
1214
1215     /* ensure the booleans are 0 or 1 so they can be used in math */
1216 #define BOOLIFY(x) h->param.x = !!h->param.x
1217     BOOLIFY( b_cabac );
1218     BOOLIFY( b_constrained_intra );
1219     BOOLIFY( b_deblocking_filter );
1220     BOOLIFY( b_deterministic );
1221     BOOLIFY( b_sliced_threads );
1222     BOOLIFY( b_interlaced );
1223     BOOLIFY( b_intra_refresh );
1224     BOOLIFY( b_aud );
1225     BOOLIFY( b_repeat_headers );
1226     BOOLIFY( b_annexb );
1227     BOOLIFY( b_vfr_input );
1228     BOOLIFY( b_pulldown );
1229     BOOLIFY( b_tff );
1230     BOOLIFY( b_pic_struct );
1231     BOOLIFY( b_fake_interlaced );
1232     BOOLIFY( b_open_gop );
1233     BOOLIFY( b_bluray_compat );
1234     BOOLIFY( b_stitchable );
1235     BOOLIFY( b_full_recon );
1236     BOOLIFY( b_opencl );
1237     BOOLIFY( b_avcintra_compat );
1238     BOOLIFY( analyse.b_transform_8x8 );
1239     BOOLIFY( analyse.b_weighted_bipred );
1240     BOOLIFY( analyse.b_chroma_me );
1241     BOOLIFY( analyse.b_mixed_references );
1242     BOOLIFY( analyse.b_fast_pskip );
1243     BOOLIFY( analyse.b_dct_decimate );
1244     BOOLIFY( analyse.b_psy );
1245     BOOLIFY( analyse.b_psnr );
1246     BOOLIFY( analyse.b_ssim );
1247     BOOLIFY( rc.b_stat_write );
1248     BOOLIFY( rc.b_stat_read );
1249     BOOLIFY( rc.b_mb_tree );
1250     BOOLIFY( rc.b_filler );
1251 #undef BOOLIFY
1252
1253     return 0;
1254 }
1255
1256 static void mbcmp_init( x264_t *h )
1257 {
1258     int satd = !h->mb.b_lossless && h->param.analyse.i_subpel_refine > 1;
1259     memcpy( h->pixf.mbcmp, satd ? h->pixf.satd : h->pixf.sad_aligned, sizeof(h->pixf.mbcmp) );
1260     memcpy( h->pixf.mbcmp_unaligned, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.mbcmp_unaligned) );
1261     h->pixf.intra_mbcmp_x3_16x16 = satd ? h->pixf.intra_satd_x3_16x16 : h->pixf.intra_sad_x3_16x16;
1262     h->pixf.intra_mbcmp_x3_8x16c = satd ? h->pixf.intra_satd_x3_8x16c : h->pixf.intra_sad_x3_8x16c;
1263     h->pixf.intra_mbcmp_x3_8x8c  = satd ? h->pixf.intra_satd_x3_8x8c  : h->pixf.intra_sad_x3_8x8c;
1264     h->pixf.intra_mbcmp_x3_8x8 = satd ? h->pixf.intra_sa8d_x3_8x8 : h->pixf.intra_sad_x3_8x8;
1265     h->pixf.intra_mbcmp_x3_4x4 = satd ? h->pixf.intra_satd_x3_4x4 : h->pixf.intra_sad_x3_4x4;
1266     h->pixf.intra_mbcmp_x9_4x4 = h->param.b_cpu_independent || h->mb.b_lossless ? NULL
1267                                : satd ? h->pixf.intra_satd_x9_4x4 : h->pixf.intra_sad_x9_4x4;
1268     h->pixf.intra_mbcmp_x9_8x8 = h->param.b_cpu_independent || h->mb.b_lossless ? NULL
1269                                : satd ? h->pixf.intra_sa8d_x9_8x8 : h->pixf.intra_sad_x9_8x8;
1270     satd &= h->param.analyse.i_me_method == X264_ME_TESA;
1271     memcpy( h->pixf.fpelcmp, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.fpelcmp) );
1272     memcpy( h->pixf.fpelcmp_x3, satd ? h->pixf.satd_x3 : h->pixf.sad_x3, sizeof(h->pixf.fpelcmp_x3) );
1273     memcpy( h->pixf.fpelcmp_x4, satd ? h->pixf.satd_x4 : h->pixf.sad_x4, sizeof(h->pixf.fpelcmp_x4) );
1274 }
1275
1276 static void chroma_dsp_init( x264_t *h )
1277 {
1278     memcpy( h->luma2chroma_pixel, x264_luma2chroma_pixel[CHROMA_FORMAT], sizeof(h->luma2chroma_pixel) );
1279
1280     switch( CHROMA_FORMAT )
1281     {
1282         case CHROMA_420:
1283             memcpy( h->predict_chroma, h->predict_8x8c, sizeof(h->predict_chroma) );
1284             h->mc.prefetch_fenc = h->mc.prefetch_fenc_420;
1285             h->loopf.deblock_chroma[0] = h->loopf.deblock_h_chroma_420;
1286             h->loopf.deblock_chroma_intra[0] = h->loopf.deblock_h_chroma_420_intra;
1287             h->loopf.deblock_chroma_mbaff = h->loopf.deblock_chroma_420_mbaff;
1288             h->loopf.deblock_chroma_intra_mbaff = h->loopf.deblock_chroma_420_intra_mbaff;
1289             h->pixf.intra_mbcmp_x3_chroma = h->pixf.intra_mbcmp_x3_8x8c;
1290             h->quantf.coeff_last[DCT_CHROMA_DC] = h->quantf.coeff_last4;
1291             h->quantf.coeff_level_run[DCT_CHROMA_DC] = h->quantf.coeff_level_run4;
1292             break;
1293         case CHROMA_422:
1294             memcpy( h->predict_chroma, h->predict_8x16c, sizeof(h->predict_chroma) );
1295             h->mc.prefetch_fenc = h->mc.prefetch_fenc_422;
1296             h->loopf.deblock_chroma[0] = h->loopf.deblock_h_chroma_422;
1297             h->loopf.deblock_chroma_intra[0] = h->loopf.deblock_h_chroma_422_intra;
1298             h->loopf.deblock_chroma_mbaff = h->loopf.deblock_chroma_422_mbaff;
1299             h->loopf.deblock_chroma_intra_mbaff = h->loopf.deblock_chroma_422_intra_mbaff;
1300             h->pixf.intra_mbcmp_x3_chroma = h->pixf.intra_mbcmp_x3_8x16c;
1301             h->quantf.coeff_last[DCT_CHROMA_DC] = h->quantf.coeff_last8;
1302             h->quantf.coeff_level_run[DCT_CHROMA_DC] = h->quantf.coeff_level_run8;
1303             break;
1304         case CHROMA_444:
1305             h->mc.prefetch_fenc = h->mc.prefetch_fenc_422; /* FIXME: doesn't cover V plane */
1306             h->loopf.deblock_chroma_mbaff = h->loopf.deblock_luma_mbaff;
1307             h->loopf.deblock_chroma_intra_mbaff = h->loopf.deblock_luma_intra_mbaff;
1308             break;
1309     }
1310 }
1311
1312 static void x264_set_aspect_ratio( x264_t *h, x264_param_t *param, int initial )
1313 {
1314     /* VUI */
1315     if( param->vui.i_sar_width > 0 && param->vui.i_sar_height > 0 )
1316     {
1317         uint32_t i_w = param->vui.i_sar_width;
1318         uint32_t i_h = param->vui.i_sar_height;
1319         uint32_t old_w = h->param.vui.i_sar_width;
1320         uint32_t old_h = h->param.vui.i_sar_height;
1321
1322         x264_reduce_fraction( &i_w, &i_h );
1323
1324         while( i_w > 65535 || i_h > 65535 )
1325         {
1326             i_w /= 2;
1327             i_h /= 2;
1328         }
1329
1330         x264_reduce_fraction( &i_w, &i_h );
1331
1332         if( i_w != old_w || i_h != old_h || initial )
1333         {
1334             h->param.vui.i_sar_width = 0;
1335             h->param.vui.i_sar_height = 0;
1336             if( i_w == 0 || i_h == 0 )
1337                 x264_log( h, X264_LOG_WARNING, "cannot create valid sample aspect ratio\n" );
1338             else
1339             {
1340                 x264_log( h, initial?X264_LOG_INFO:X264_LOG_DEBUG, "using SAR=%d/%d\n", i_w, i_h );
1341                 h->param.vui.i_sar_width = i_w;
1342                 h->param.vui.i_sar_height = i_h;
1343             }
1344         }
1345     }
1346 }
1347
1348 /****************************************************************************
1349  * x264_encoder_open:
1350  ****************************************************************************/
1351 x264_t *x264_encoder_open( x264_param_t *param )
1352 {
1353     x264_t *h;
1354     char buf[1000], *p;
1355     int qp, i_slicetype_length;
1356
1357     CHECKED_MALLOCZERO( h, sizeof(x264_t) );
1358
1359     /* Create a copy of param */
1360     memcpy( &h->param, param, sizeof(x264_param_t) );
1361
1362     if( param->param_free )
1363         param->param_free( param );
1364
1365     if( x264_threading_init() )
1366     {
1367         x264_log( h, X264_LOG_ERROR, "unable to initialize threading\n" );
1368         goto fail;
1369     }
1370
1371     if( x264_validate_parameters( h, 1 ) < 0 )
1372         goto fail;
1373
1374     if( h->param.psz_cqm_file )
1375         if( x264_cqm_parse_file( h, h->param.psz_cqm_file ) < 0 )
1376             goto fail;
1377
1378     if( h->param.rc.psz_stat_out )
1379         h->param.rc.psz_stat_out = strdup( h->param.rc.psz_stat_out );
1380     if( h->param.rc.psz_stat_in )
1381         h->param.rc.psz_stat_in = strdup( h->param.rc.psz_stat_in );
1382
1383     x264_reduce_fraction( &h->param.i_fps_num, &h->param.i_fps_den );
1384     x264_reduce_fraction( &h->param.i_timebase_num, &h->param.i_timebase_den );
1385
1386     /* Init x264_t */
1387     h->i_frame = -1;
1388     h->i_frame_num = 0;
1389     h->i_idr_pic_id = 0;
1390
1391     if( (uint64_t)h->param.i_timebase_den * 2 > UINT32_MAX )
1392     {
1393         x264_log( h, X264_LOG_ERROR, "Effective timebase denominator %u exceeds H.264 maximum\n", h->param.i_timebase_den );
1394         goto fail;
1395     }
1396
1397     x264_set_aspect_ratio( h, &h->param, 1 );
1398
1399     x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
1400     x264_pps_init( h->pps, h->param.i_sps_id, &h->param, h->sps );
1401
1402     x264_validate_levels( h, 1 );
1403
1404     h->chroma_qp_table = i_chroma_qp_table + 12 + h->pps->i_chroma_qp_index_offset;
1405
1406     if( x264_cqm_init( h ) < 0 )
1407         goto fail;
1408
1409     h->mb.i_mb_width = h->sps->i_mb_width;
1410     h->mb.i_mb_height = h->sps->i_mb_height;
1411     h->mb.i_mb_count = h->mb.i_mb_width * h->mb.i_mb_height;
1412
1413     h->mb.chroma_h_shift = CHROMA_FORMAT == CHROMA_420 || CHROMA_FORMAT == CHROMA_422;
1414     h->mb.chroma_v_shift = CHROMA_FORMAT == CHROMA_420;
1415
1416     /* Adaptive MBAFF and subme 0 are not supported as we require halving motion
1417      * vectors during prediction, resulting in hpel mvs.
1418      * The chosen solution is to make MBAFF non-adaptive in this case. */
1419     h->mb.b_adaptive_mbaff = PARAM_INTERLACED && h->param.analyse.i_subpel_refine;
1420
1421     /* Init frames. */
1422     if( h->param.i_bframe_adaptive == X264_B_ADAPT_TRELLIS && !h->param.rc.b_stat_read )
1423         h->frames.i_delay = X264_MAX(h->param.i_bframe,3)*4;
1424     else
1425         h->frames.i_delay = h->param.i_bframe;
1426     if( h->param.rc.b_mb_tree || h->param.rc.i_vbv_buffer_size )
1427         h->frames.i_delay = X264_MAX( h->frames.i_delay, h->param.rc.i_lookahead );
1428     i_slicetype_length = h->frames.i_delay;
1429     h->frames.i_delay += h->i_thread_frames - 1;
1430     h->frames.i_delay += h->param.i_sync_lookahead;
1431     h->frames.i_delay += h->param.b_vfr_input;
1432     h->frames.i_bframe_delay = h->param.i_bframe ? (h->param.i_bframe_pyramid ? 2 : 1) : 0;
1433
1434     h->frames.i_max_ref0 = h->param.i_frame_reference;
1435     h->frames.i_max_ref1 = X264_MIN( h->sps->vui.i_num_reorder_frames, h->param.i_frame_reference );
1436     h->frames.i_max_dpb  = h->sps->vui.i_max_dec_frame_buffering;
1437     h->frames.b_have_lowres = !h->param.rc.b_stat_read
1438         && ( h->param.rc.i_rc_method == X264_RC_ABR
1439           || h->param.rc.i_rc_method == X264_RC_CRF
1440           || h->param.i_bframe_adaptive
1441           || h->param.i_scenecut_threshold
1442           || h->param.rc.b_mb_tree
1443           || h->param.analyse.i_weighted_pred );
1444     h->frames.b_have_lowres |= h->param.rc.b_stat_read && h->param.rc.i_vbv_buffer_size > 0;
1445     h->frames.b_have_sub8x8_esa = !!(h->param.analyse.inter & X264_ANALYSE_PSUB8x8);
1446
1447     h->frames.i_last_idr =
1448     h->frames.i_last_keyframe = - h->param.i_keyint_max;
1449     h->frames.i_input    = 0;
1450     h->frames.i_largest_pts = h->frames.i_second_largest_pts = -1;
1451     h->frames.i_poc_last_open_gop = -1;
1452
1453     CHECKED_MALLOCZERO( h->frames.unused[0], (h->frames.i_delay + 3) * sizeof(x264_frame_t *) );
1454     /* Allocate room for max refs plus a few extra just in case. */
1455     CHECKED_MALLOCZERO( h->frames.unused[1], (h->i_thread_frames + X264_REF_MAX + 4) * sizeof(x264_frame_t *) );
1456     CHECKED_MALLOCZERO( h->frames.current, (h->param.i_sync_lookahead + h->param.i_bframe
1457                         + h->i_thread_frames + 3) * sizeof(x264_frame_t *) );
1458     if( h->param.analyse.i_weighted_pred > 0 )
1459         CHECKED_MALLOCZERO( h->frames.blank_unused, h->i_thread_frames * 4 * sizeof(x264_frame_t *) );
1460     h->i_ref[0] = h->i_ref[1] = 0;
1461     h->i_cpb_delay = h->i_coded_fields = h->i_disp_fields = 0;
1462     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);
1463     h->i_disp_fields_last_frame = -1;
1464     x264_rdo_init();
1465
1466     /* init CPU functions */
1467     x264_predict_16x16_init( h->param.cpu, h->predict_16x16 );
1468     x264_predict_8x8c_init( h->param.cpu, h->predict_8x8c );
1469     x264_predict_8x16c_init( h->param.cpu, h->predict_8x16c );
1470     x264_predict_8x8_init( h->param.cpu, h->predict_8x8, &h->predict_8x8_filter );
1471     x264_predict_4x4_init( h->param.cpu, h->predict_4x4 );
1472     x264_pixel_init( h->param.cpu, &h->pixf );
1473     x264_dct_init( h->param.cpu, &h->dctf );
1474     x264_zigzag_init( h->param.cpu, &h->zigzagf_progressive, &h->zigzagf_interlaced );
1475     memcpy( &h->zigzagf, PARAM_INTERLACED ? &h->zigzagf_interlaced : &h->zigzagf_progressive, sizeof(h->zigzagf) );
1476     x264_mc_init( h->param.cpu, &h->mc, h->param.b_cpu_independent );
1477     x264_quant_init( h, h->param.cpu, &h->quantf );
1478     x264_deblock_init( h->param.cpu, &h->loopf, PARAM_INTERLACED );
1479     x264_bitstream_init( h->param.cpu, &h->bsf );
1480     if( h->param.b_cabac )
1481         x264_cabac_init( h );
1482     else
1483         x264_stack_align( x264_cavlc_init, h );
1484
1485     mbcmp_init( h );
1486     chroma_dsp_init( h );
1487
1488     p = buf + sprintf( buf, "using cpu capabilities:" );
1489     for( int i = 0; x264_cpu_names[i].flags; i++ )
1490     {
1491         if( !strcmp(x264_cpu_names[i].name, "SSE")
1492             && h->param.cpu & (X264_CPU_SSE2) )
1493             continue;
1494         if( !strcmp(x264_cpu_names[i].name, "SSE2")
1495             && h->param.cpu & (X264_CPU_SSE2_IS_FAST|X264_CPU_SSE2_IS_SLOW) )
1496             continue;
1497         if( !strcmp(x264_cpu_names[i].name, "SSE3")
1498             && (h->param.cpu & X264_CPU_SSSE3 || !(h->param.cpu & X264_CPU_CACHELINE_64)) )
1499             continue;
1500         if( !strcmp(x264_cpu_names[i].name, "SSE4.1")
1501             && (h->param.cpu & X264_CPU_SSE42) )
1502             continue;
1503         if( !strcmp(x264_cpu_names[i].name, "BMI1")
1504             && (h->param.cpu & X264_CPU_BMI2) )
1505             continue;
1506         if( (h->param.cpu & x264_cpu_names[i].flags) == x264_cpu_names[i].flags
1507             && (!i || x264_cpu_names[i].flags != x264_cpu_names[i-1].flags) )
1508             p += sprintf( p, " %s", x264_cpu_names[i].name );
1509     }
1510     if( !h->param.cpu )
1511         p += sprintf( p, " none!" );
1512     x264_log( h, X264_LOG_INFO, "%s\n", buf );
1513
1514     float *logs = x264_analyse_prepare_costs( h );
1515     if( !logs )
1516         goto fail;
1517     for( qp = X264_MIN( h->param.rc.i_qp_min, QP_MAX_SPEC ); qp <= h->param.rc.i_qp_max; qp++ )
1518         if( x264_analyse_init_costs( h, logs, qp ) )
1519             goto fail;
1520     if( x264_analyse_init_costs( h, logs, X264_LOOKAHEAD_QP ) )
1521         goto fail;
1522     x264_free( logs );
1523
1524     static const uint16_t cost_mv_correct[7] = { 24, 47, 95, 189, 379, 757, 1515 };
1525     /* Checks for known miscompilation issues. */
1526     if( h->cost_mv[X264_LOOKAHEAD_QP][2013] != cost_mv_correct[BIT_DEPTH-8] )
1527     {
1528         x264_log( h, X264_LOG_ERROR, "MV cost test failed: x264 has been miscompiled!\n" );
1529         goto fail;
1530     }
1531
1532     /* Must be volatile or else GCC will optimize it out. */
1533     volatile int temp = 392;
1534     if( x264_clz( temp ) != 23 )
1535     {
1536         x264_log( h, X264_LOG_ERROR, "CLZ test failed: x264 has been miscompiled!\n" );
1537 #if ARCH_X86 || ARCH_X86_64
1538         x264_log( h, X264_LOG_ERROR, "Are you attempting to run an SSE4a/LZCNT-targeted build on a CPU that\n" );
1539         x264_log( h, X264_LOG_ERROR, "doesn't support it?\n" );
1540 #endif
1541         goto fail;
1542     }
1543
1544     h->out.i_nal = 0;
1545     h->out.i_bitstream = X264_MAX( 1000000, h->param.i_width * h->param.i_height * 4
1546         * ( h->param.rc.i_rc_method == X264_RC_ABR ? pow( 0.95, h->param.rc.i_qp_min )
1547           : pow( 0.95, h->param.rc.i_qp_constant ) * X264_MAX( 1, h->param.rc.f_ip_factor )));
1548
1549     h->nal_buffer_size = h->out.i_bitstream * 3/2 + 4 + 64; /* +4 for startcode, +64 for nal_escape assembly padding */
1550     CHECKED_MALLOC( h->nal_buffer, h->nal_buffer_size );
1551
1552     CHECKED_MALLOC( h->reconfig_h, sizeof(x264_t) );
1553
1554     if( h->param.i_threads > 1 &&
1555         x264_threadpool_init( &h->threadpool, h->param.i_threads, (void*)x264_encoder_thread_init, h ) )
1556         goto fail;
1557     if( h->param.i_lookahead_threads > 1 &&
1558         x264_threadpool_init( &h->lookaheadpool, h->param.i_lookahead_threads, NULL, NULL ) )
1559         goto fail;
1560
1561 #if HAVE_OPENCL
1562     if( h->param.b_opencl )
1563     {
1564         h->opencl.ocl = x264_opencl_load_library();
1565         if( !h->opencl.ocl )
1566         {
1567             x264_log( h, X264_LOG_WARNING, "failed to load OpenCL\n" );
1568             h->param.b_opencl = 0;
1569         }
1570     }
1571 #endif
1572
1573     h->thread[0] = h;
1574     for( int i = 1; i < h->param.i_threads + !!h->param.i_sync_lookahead; i++ )
1575         CHECKED_MALLOC( h->thread[i], sizeof(x264_t) );
1576     if( h->param.i_lookahead_threads > 1 )
1577         for( int i = 0; i < h->param.i_lookahead_threads; i++ )
1578         {
1579             CHECKED_MALLOC( h->lookahead_thread[i], sizeof(x264_t) );
1580             *h->lookahead_thread[i] = *h;
1581         }
1582     *h->reconfig_h = *h;
1583
1584     for( int i = 0; i < h->param.i_threads; i++ )
1585     {
1586         int init_nal_count = h->param.i_slice_count + 3;
1587         int allocate_threadlocal_data = !h->param.b_sliced_threads || !i;
1588         if( i > 0 )
1589             *h->thread[i] = *h;
1590
1591         if( x264_pthread_mutex_init( &h->thread[i]->mutex, NULL ) )
1592             goto fail;
1593         if( x264_pthread_cond_init( &h->thread[i]->cv, NULL ) )
1594             goto fail;
1595
1596         if( allocate_threadlocal_data )
1597         {
1598             h->thread[i]->fdec = x264_frame_pop_unused( h, 1 );
1599             if( !h->thread[i]->fdec )
1600                 goto fail;
1601         }
1602         else
1603             h->thread[i]->fdec = h->thread[0]->fdec;
1604
1605         CHECKED_MALLOC( h->thread[i]->out.p_bitstream, h->out.i_bitstream );
1606         /* Start each thread with room for init_nal_count NAL units; it'll realloc later if needed. */
1607         CHECKED_MALLOC( h->thread[i]->out.nal, init_nal_count*sizeof(x264_nal_t) );
1608         h->thread[i]->out.i_nals_allocated = init_nal_count;
1609
1610         if( allocate_threadlocal_data && x264_macroblock_cache_allocate( h->thread[i] ) < 0 )
1611             goto fail;
1612     }
1613
1614 #if HAVE_OPENCL
1615     if( h->param.b_opencl && x264_opencl_lookahead_init( h ) < 0 )
1616         h->param.b_opencl = 0;
1617 #endif
1618
1619     if( x264_lookahead_init( h, i_slicetype_length ) )
1620         goto fail;
1621
1622     for( int i = 0; i < h->param.i_threads; i++ )
1623         if( x264_macroblock_thread_allocate( h->thread[i], 0 ) < 0 )
1624             goto fail;
1625
1626     if( x264_ratecontrol_new( h ) < 0 )
1627         goto fail;
1628
1629     if( h->param.i_nal_hrd )
1630     {
1631         x264_log( h, X264_LOG_DEBUG, "HRD bitrate: %i bits/sec\n", h->sps->vui.hrd.i_bit_rate_unscaled );
1632         x264_log( h, X264_LOG_DEBUG, "CPB size: %i bits\n", h->sps->vui.hrd.i_cpb_size_unscaled );
1633     }
1634
1635     if( h->param.psz_dump_yuv )
1636     {
1637         /* create or truncate the reconstructed video file */
1638         FILE *f = x264_fopen( h->param.psz_dump_yuv, "w" );
1639         if( !f )
1640         {
1641             x264_log( h, X264_LOG_ERROR, "dump_yuv: can't write to %s\n", h->param.psz_dump_yuv );
1642             goto fail;
1643         }
1644         else if( !x264_is_regular_file( f ) )
1645         {
1646             x264_log( h, X264_LOG_ERROR, "dump_yuv: incompatible with non-regular file %s\n", h->param.psz_dump_yuv );
1647             goto fail;
1648         }
1649         fclose( f );
1650     }
1651
1652     const char *profile = h->sps->i_profile_idc == PROFILE_BASELINE ? "Constrained Baseline" :
1653                           h->sps->i_profile_idc == PROFILE_MAIN ? "Main" :
1654                           h->sps->i_profile_idc == PROFILE_HIGH ? "High" :
1655                           h->sps->i_profile_idc == PROFILE_HIGH10 ? (h->sps->b_constraint_set3 == 1 ? "High 10 Intra" : "High 10") :
1656                           h->sps->i_profile_idc == PROFILE_HIGH422 ? (h->sps->b_constraint_set3 == 1 ? "High 4:2:2 Intra" : "High 4:2:2") :
1657                           h->sps->b_constraint_set3 == 1 ? "High 4:4:4 Intra" : "High 4:4:4 Predictive";
1658     char level[4];
1659     snprintf( level, sizeof(level), "%d.%d", h->sps->i_level_idc/10, h->sps->i_level_idc%10 );
1660     if( h->sps->i_level_idc == 9 || ( h->sps->i_level_idc == 11 && h->sps->b_constraint_set3 &&
1661         (h->sps->i_profile_idc == PROFILE_BASELINE || h->sps->i_profile_idc == PROFILE_MAIN) ) )
1662         strcpy( level, "1b" );
1663
1664     if( h->sps->i_profile_idc < PROFILE_HIGH10 )
1665     {
1666         x264_log( h, X264_LOG_INFO, "profile %s, level %s\n",
1667             profile, level );
1668     }
1669     else
1670     {
1671         static const char * const subsampling[4] = { "4:0:0", "4:2:0", "4:2:2", "4:4:4" };
1672         x264_log( h, X264_LOG_INFO, "profile %s, level %s, %s %d-bit\n",
1673             profile, level, subsampling[CHROMA_FORMAT], BIT_DEPTH );
1674     }
1675
1676     return h;
1677 fail:
1678     x264_free( h );
1679     return NULL;
1680 }
1681
1682 /****************************************************************************/
1683 static int x264_encoder_try_reconfig( x264_t *h, x264_param_t *param, int *rc_reconfig )
1684 {
1685     *rc_reconfig = 0;
1686     x264_set_aspect_ratio( h, param, 0 );
1687 #define COPY(var) h->param.var = param->var
1688     COPY( i_frame_reference ); // but never uses more refs than initially specified
1689     COPY( i_bframe_bias );
1690     if( h->param.i_scenecut_threshold )
1691         COPY( i_scenecut_threshold ); // can't turn it on or off, only vary the threshold
1692     COPY( b_deblocking_filter );
1693     COPY( i_deblocking_filter_alphac0 );
1694     COPY( i_deblocking_filter_beta );
1695     COPY( i_frame_packing );
1696     COPY( analyse.inter );
1697     COPY( analyse.intra );
1698     COPY( analyse.i_direct_mv_pred );
1699     /* Scratch buffer prevents me_range from being increased for esa/tesa */
1700     if( h->param.analyse.i_me_method < X264_ME_ESA || param->analyse.i_me_range < h->param.analyse.i_me_range )
1701         COPY( analyse.i_me_range );
1702     COPY( analyse.i_noise_reduction );
1703     /* We can't switch out of subme=0 during encoding. */
1704     if( h->param.analyse.i_subpel_refine )
1705         COPY( analyse.i_subpel_refine );
1706     COPY( analyse.i_trellis );
1707     COPY( analyse.b_chroma_me );
1708     COPY( analyse.b_dct_decimate );
1709     COPY( analyse.b_fast_pskip );
1710     COPY( analyse.b_mixed_references );
1711     COPY( analyse.f_psy_rd );
1712     COPY( analyse.f_psy_trellis );
1713     COPY( crop_rect );
1714     // can only twiddle these if they were enabled to begin with:
1715     if( h->param.analyse.i_me_method >= X264_ME_ESA || param->analyse.i_me_method < X264_ME_ESA )
1716         COPY( analyse.i_me_method );
1717     if( h->param.analyse.i_me_method >= X264_ME_ESA && !h->frames.b_have_sub8x8_esa )
1718         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
1719     if( h->pps->b_transform_8x8_mode )
1720         COPY( analyse.b_transform_8x8 );
1721     if( h->frames.i_max_ref1 > 1 )
1722         COPY( i_bframe_pyramid );
1723     COPY( i_slice_max_size );
1724     COPY( i_slice_max_mbs );
1725     COPY( i_slice_min_mbs );
1726     COPY( i_slice_count );
1727     COPY( i_slice_count_max );
1728     COPY( b_tff );
1729
1730     /* VBV can't be turned on if it wasn't on to begin with */
1731     if( h->param.rc.i_vbv_max_bitrate > 0 && h->param.rc.i_vbv_buffer_size > 0 &&
1732           param->rc.i_vbv_max_bitrate > 0 &&   param->rc.i_vbv_buffer_size > 0 )
1733     {
1734         *rc_reconfig |= h->param.rc.i_vbv_max_bitrate != param->rc.i_vbv_max_bitrate;
1735         *rc_reconfig |= h->param.rc.i_vbv_buffer_size != param->rc.i_vbv_buffer_size;
1736         *rc_reconfig |= h->param.rc.i_bitrate != param->rc.i_bitrate;
1737         COPY( rc.i_vbv_max_bitrate );
1738         COPY( rc.i_vbv_buffer_size );
1739         COPY( rc.i_bitrate );
1740     }
1741     *rc_reconfig |= h->param.rc.f_rf_constant != param->rc.f_rf_constant;
1742     *rc_reconfig |= h->param.rc.f_rf_constant_max != param->rc.f_rf_constant_max;
1743     COPY( rc.f_rf_constant );
1744     COPY( rc.f_rf_constant_max );
1745 #undef COPY
1746
1747     return x264_validate_parameters( h, 0 );
1748 }
1749
1750 int x264_encoder_reconfig_apply( x264_t *h, x264_param_t *param )
1751 {
1752     int rc_reconfig;
1753     int ret = x264_encoder_try_reconfig( h, param, &rc_reconfig );
1754
1755     mbcmp_init( h );
1756     if( !ret )
1757         x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
1758
1759     /* Supported reconfiguration options (1-pass only):
1760      * vbv-maxrate
1761      * vbv-bufsize
1762      * crf
1763      * bitrate (CBR only) */
1764     if( !ret && rc_reconfig )
1765         x264_ratecontrol_init_reconfigurable( h, 0 );
1766
1767     return ret;
1768 }
1769
1770 /****************************************************************************
1771  * x264_encoder_reconfig:
1772  ****************************************************************************/
1773 int x264_encoder_reconfig( x264_t *h, x264_param_t *param )
1774 {
1775     h = h->thread[h->thread[0]->i_thread_phase];
1776     x264_param_t param_save = h->reconfig_h->param;
1777     h->reconfig_h->param = h->param;
1778
1779     int rc_reconfig;
1780     int ret = x264_encoder_try_reconfig( h->reconfig_h, param, &rc_reconfig );
1781     if( !ret )
1782         h->reconfig = 1;
1783     else
1784         h->reconfig_h->param = param_save;
1785
1786     return ret;
1787 }
1788
1789 /****************************************************************************
1790  * x264_encoder_parameters:
1791  ****************************************************************************/
1792 void x264_encoder_parameters( x264_t *h, x264_param_t *param )
1793 {
1794     memcpy( param, &h->thread[h->i_thread_phase]->param, sizeof(x264_param_t) );
1795 }
1796
1797 /* internal usage */
1798 static void x264_nal_start( x264_t *h, int i_type, int i_ref_idc )
1799 {
1800     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1801
1802     nal->i_ref_idc        = i_ref_idc;
1803     nal->i_type           = i_type;
1804     nal->b_long_startcode = 1;
1805
1806     nal->i_payload= 0;
1807     nal->p_payload= &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8];
1808     nal->i_padding= 0;
1809 }
1810
1811 /* if number of allocated nals is not enough, re-allocate a larger one. */
1812 static int x264_nal_check_buffer( x264_t *h )
1813 {
1814     if( h->out.i_nal >= h->out.i_nals_allocated )
1815     {
1816         x264_nal_t *new_out = x264_malloc( sizeof(x264_nal_t) * (h->out.i_nals_allocated*2) );
1817         if( !new_out )
1818             return -1;
1819         memcpy( new_out, h->out.nal, sizeof(x264_nal_t) * (h->out.i_nals_allocated) );
1820         x264_free( h->out.nal );
1821         h->out.nal = new_out;
1822         h->out.i_nals_allocated *= 2;
1823     }
1824     return 0;
1825 }
1826
1827 static int x264_nal_end( x264_t *h )
1828 {
1829     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1830     uint8_t *end = &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8];
1831     nal->i_payload = end - nal->p_payload;
1832     /* Assembly implementation of nal_escape reads past the end of the input.
1833      * While undefined padding wouldn't actually affect the output, it makes valgrind unhappy. */
1834     memset( end, 0xff, 64 );
1835     if( h->param.nalu_process )
1836         h->param.nalu_process( h, nal, h->fenc->opaque );
1837     h->out.i_nal++;
1838
1839     return x264_nal_check_buffer( h );
1840 }
1841
1842 static int x264_check_encapsulated_buffer( x264_t *h, x264_t *h0, int start,
1843                                            int previous_nal_size, int necessary_size )
1844 {
1845     if( h0->nal_buffer_size < necessary_size )
1846     {
1847         necessary_size *= 2;
1848         uint8_t *buf = x264_malloc( necessary_size );
1849         if( !buf )
1850             return -1;
1851         if( previous_nal_size )
1852             memcpy( buf, h0->nal_buffer, previous_nal_size );
1853
1854         intptr_t delta = buf - h0->nal_buffer;
1855         for( int i = 0; i < start; i++ )
1856             h->out.nal[i].p_payload += delta;
1857
1858         x264_free( h0->nal_buffer );
1859         h0->nal_buffer = buf;
1860         h0->nal_buffer_size = necessary_size;
1861     }
1862
1863     return 0;
1864 }
1865
1866 static int x264_encoder_encapsulate_nals( x264_t *h, int start )
1867 {
1868     x264_t *h0 = h->thread[0];
1869     int nal_size = 0, previous_nal_size = 0;
1870
1871     if( h->param.nalu_process )
1872     {
1873         for( int i = start; i < h->out.i_nal; i++ )
1874             nal_size += h->out.nal[i].i_payload;
1875         return nal_size;
1876     }
1877
1878     for( int i = 0; i < start; i++ )
1879         previous_nal_size += h->out.nal[i].i_payload;
1880
1881     for( int i = start; i < h->out.i_nal; i++ )
1882         nal_size += h->out.nal[i].i_payload;
1883
1884     /* Worst-case NAL unit escaping: reallocate the buffer if it's too small. */
1885     int necessary_size = previous_nal_size + nal_size * 3/2 + h->out.i_nal * 4 + 4 + 64;
1886     for( int i = start; i < h->out.i_nal; i++ )
1887         necessary_size += h->out.nal[i].i_padding;
1888     if( x264_check_encapsulated_buffer( h, h0, start, previous_nal_size, necessary_size ) )
1889         return -1;
1890
1891     uint8_t *nal_buffer = h0->nal_buffer + previous_nal_size;
1892
1893     for( int i = start; i < h->out.i_nal; i++ )
1894     {
1895         int old_payload_len = h->out.nal[i].i_payload;
1896         h->out.nal[i].b_long_startcode = !i || h->out.nal[i].i_type == NAL_SPS || h->out.nal[i].i_type == NAL_PPS ||
1897                                          h->param.b_avcintra_compat;
1898         x264_nal_encode( h, nal_buffer, &h->out.nal[i] );
1899         nal_buffer += h->out.nal[i].i_payload;
1900         if( h->param.b_avcintra_compat )
1901         {
1902             h->out.nal[i].i_padding -= h->out.nal[i].i_payload - (old_payload_len + NALU_OVERHEAD);
1903             if( h->out.nal[i].i_padding > 0 )
1904             {
1905                 memset( nal_buffer, 0, h->out.nal[i].i_padding );
1906                 nal_buffer += h->out.nal[i].i_padding;
1907                 h->out.nal[i].i_payload += h->out.nal[i].i_padding;
1908             }
1909             h->out.nal[i].i_padding = X264_MAX( h->out.nal[i].i_padding, 0 );
1910         }
1911     }
1912
1913     x264_emms();
1914
1915     return nal_buffer - (h0->nal_buffer + previous_nal_size);
1916 }
1917
1918 /****************************************************************************
1919  * x264_encoder_headers:
1920  ****************************************************************************/
1921 int x264_encoder_headers( x264_t *h, x264_nal_t **pp_nal, int *pi_nal )
1922 {
1923     int frame_size = 0;
1924     /* init bitstream context */
1925     h->out.i_nal = 0;
1926     bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
1927
1928     /* Write SEI, SPS and PPS. */
1929
1930     /* generate sequence parameters */
1931     x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
1932     x264_sps_write( &h->out.bs, h->sps );
1933     if( x264_nal_end( h ) )
1934         return -1;
1935
1936     /* generate picture parameters */
1937     x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
1938     x264_pps_write( &h->out.bs, h->sps, h->pps );
1939     if( x264_nal_end( h ) )
1940         return -1;
1941
1942     /* identify ourselves */
1943     x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
1944     if( x264_sei_version_write( h, &h->out.bs ) )
1945         return -1;
1946     if( x264_nal_end( h ) )
1947         return -1;
1948
1949     frame_size = x264_encoder_encapsulate_nals( h, 0 );
1950     if( frame_size < 0 )
1951         return -1;
1952
1953     /* now set output*/
1954     *pi_nal = h->out.i_nal;
1955     *pp_nal = &h->out.nal[0];
1956     h->out.i_nal = 0;
1957
1958     return frame_size;
1959 }
1960
1961 /* Check to see whether we have chosen a reference list ordering different
1962  * from the standard's default. */
1963 static inline void x264_reference_check_reorder( x264_t *h )
1964 {
1965     /* The reorder check doesn't check for missing frames, so just
1966      * force a reorder if one of the reference list is corrupt. */
1967     for( int i = 0; h->frames.reference[i]; i++ )
1968         if( h->frames.reference[i]->b_corrupt )
1969         {
1970             h->b_ref_reorder[0] = 1;
1971             return;
1972         }
1973     for( int list = 0; list <= (h->sh.i_type == SLICE_TYPE_B); list++ )
1974         for( int i = 0; i < h->i_ref[list] - 1; i++ )
1975         {
1976             int framenum_diff = h->fref[list][i+1]->i_frame_num - h->fref[list][i]->i_frame_num;
1977             int poc_diff = h->fref[list][i+1]->i_poc - h->fref[list][i]->i_poc;
1978             /* P and B-frames use different default orders. */
1979             if( h->sh.i_type == SLICE_TYPE_P ? framenum_diff > 0 : list == 1 ? poc_diff < 0 : poc_diff > 0 )
1980             {
1981                 h->b_ref_reorder[list] = 1;
1982                 return;
1983             }
1984         }
1985 }
1986
1987 /* return -1 on failure, else return the index of the new reference frame */
1988 int x264_weighted_reference_duplicate( x264_t *h, int i_ref, const x264_weight_t *w )
1989 {
1990     int i = h->i_ref[0];
1991     int j = 1;
1992     x264_frame_t *newframe;
1993     if( i <= 1 ) /* empty list, definitely can't duplicate frame */
1994         return -1;
1995
1996     //Duplication is only used in X264_WEIGHTP_SMART
1997     if( h->param.analyse.i_weighted_pred != X264_WEIGHTP_SMART )
1998         return -1;
1999
2000     /* Duplication is a hack to compensate for crappy rounding in motion compensation.
2001      * With high bit depth, it's not worth doing, so turn it off except in the case of
2002      * unweighted dupes. */
2003     if( BIT_DEPTH > 8 && w != x264_weight_none )
2004         return -1;
2005
2006     newframe = x264_frame_pop_blank_unused( h );
2007     if( !newframe )
2008         return -1;
2009
2010     //FIXME: probably don't need to copy everything
2011     *newframe = *h->fref[0][i_ref];
2012     newframe->i_reference_count = 1;
2013     newframe->orig = h->fref[0][i_ref];
2014     newframe->b_duplicate = 1;
2015     memcpy( h->fenc->weight[j], w, sizeof(h->fenc->weight[i]) );
2016
2017     /* shift the frames to make space for the dupe. */
2018     h->b_ref_reorder[0] = 1;
2019     if( h->i_ref[0] < X264_REF_MAX )
2020         ++h->i_ref[0];
2021     h->fref[0][X264_REF_MAX-1] = NULL;
2022     x264_frame_unshift( &h->fref[0][j], newframe );
2023
2024     return j;
2025 }
2026
2027 static void x264_weighted_pred_init( x264_t *h )
2028 {
2029     /* for now no analysis and set all weights to nothing */
2030     for( int i_ref = 0; i_ref < h->i_ref[0]; i_ref++ )
2031         h->fenc->weighted[i_ref] = h->fref[0][i_ref]->filtered[0][0];
2032
2033     // FIXME: This only supports weighting of one reference frame
2034     // and duplicates of that frame.
2035     h->fenc->i_lines_weighted = 0;
2036
2037     for( int i_ref = 0; i_ref < (h->i_ref[0] << SLICE_MBAFF); i_ref++ )
2038         for( int i = 0; i < 3; i++ )
2039             h->sh.weight[i_ref][i].weightfn = NULL;
2040
2041
2042     if( h->sh.i_type != SLICE_TYPE_P || h->param.analyse.i_weighted_pred <= 0 )
2043         return;
2044
2045     int i_padv = PADV << PARAM_INTERLACED;
2046     int denom = -1;
2047     int weightplane[2] = { 0, 0 };
2048     int buffer_next = 0;
2049     for( int i = 0; i < 3; i++ )
2050     {
2051         for( int j = 0; j < h->i_ref[0]; j++ )
2052         {
2053             if( h->fenc->weight[j][i].weightfn )
2054             {
2055                 h->sh.weight[j][i] = h->fenc->weight[j][i];
2056                 // if weight is useless, don't write it to stream
2057                 if( h->sh.weight[j][i].i_scale == 1<<h->sh.weight[j][i].i_denom && h->sh.weight[j][i].i_offset == 0 )
2058                     h->sh.weight[j][i].weightfn = NULL;
2059                 else
2060                 {
2061                     if( !weightplane[!!i] )
2062                     {
2063                         weightplane[!!i] = 1;
2064                         h->sh.weight[0][!!i].i_denom = denom = h->sh.weight[j][i].i_denom;
2065                         assert( x264_clip3( denom, 0, 7 ) == denom );
2066                     }
2067
2068                     assert( h->sh.weight[j][i].i_denom == denom );
2069                     if( !i )
2070                     {
2071                         h->fenc->weighted[j] = h->mb.p_weight_buf[buffer_next++] + h->fenc->i_stride[0] * i_padv + PADH;
2072                         //scale full resolution frame
2073                         if( h->param.i_threads == 1 )
2074                         {
2075                             pixel *src = h->fref[0][j]->filtered[0][0] - h->fref[0][j]->i_stride[0]*i_padv - PADH;
2076                             pixel *dst = h->fenc->weighted[j] - h->fenc->i_stride[0]*i_padv - PADH;
2077                             int stride = h->fenc->i_stride[0];
2078                             int width = h->fenc->i_width[0] + PADH*2;
2079                             int height = h->fenc->i_lines[0] + i_padv*2;
2080                             x264_weight_scale_plane( h, dst, stride, src, stride, width, height, &h->sh.weight[j][0] );
2081                             h->fenc->i_lines_weighted = height;
2082                         }
2083                     }
2084                 }
2085             }
2086         }
2087     }
2088
2089     if( weightplane[1] )
2090         for( int i = 0; i < h->i_ref[0]; i++ )
2091         {
2092             if( h->sh.weight[i][1].weightfn && !h->sh.weight[i][2].weightfn )
2093             {
2094                 h->sh.weight[i][2].i_scale = 1 << h->sh.weight[0][1].i_denom;
2095                 h->sh.weight[i][2].i_offset = 0;
2096             }
2097             else if( h->sh.weight[i][2].weightfn && !h->sh.weight[i][1].weightfn )
2098             {
2099                 h->sh.weight[i][1].i_scale = 1 << h->sh.weight[0][1].i_denom;
2100                 h->sh.weight[i][1].i_offset = 0;
2101             }
2102         }
2103
2104     if( !weightplane[0] )
2105         h->sh.weight[0][0].i_denom = 0;
2106     if( !weightplane[1] )
2107         h->sh.weight[0][1].i_denom = 0;
2108     h->sh.weight[0][2].i_denom = h->sh.weight[0][1].i_denom;
2109 }
2110
2111 static inline int x264_reference_distance( x264_t *h, x264_frame_t *frame )
2112 {
2113     if( h->param.i_frame_packing == 5 )
2114         return abs((h->fenc->i_frame&~1) - (frame->i_frame&~1)) +
2115                   ((h->fenc->i_frame&1) != (frame->i_frame&1));
2116     else
2117         return abs(h->fenc->i_frame - frame->i_frame);
2118 }
2119
2120 static inline void x264_reference_build_list( x264_t *h, int i_poc )
2121 {
2122     int b_ok;
2123
2124     /* build ref list 0/1 */
2125     h->mb.pic.i_fref[0] = h->i_ref[0] = 0;
2126     h->mb.pic.i_fref[1] = h->i_ref[1] = 0;
2127     if( h->sh.i_type == SLICE_TYPE_I )
2128         return;
2129
2130     for( int i = 0; h->frames.reference[i]; i++ )
2131     {
2132         if( h->frames.reference[i]->b_corrupt )
2133             continue;
2134         if( h->frames.reference[i]->i_poc < i_poc )
2135             h->fref[0][h->i_ref[0]++] = h->frames.reference[i];
2136         else if( h->frames.reference[i]->i_poc > i_poc )
2137             h->fref[1][h->i_ref[1]++] = h->frames.reference[i];
2138     }
2139
2140     /* Order reference lists by distance from the current frame. */
2141     for( int list = 0; list < 2; list++ )
2142     {
2143         h->fref_nearest[list] = h->fref[list][0];
2144         do
2145         {
2146             b_ok = 1;
2147             for( int i = 0; i < h->i_ref[list] - 1; i++ )
2148             {
2149                 if( list ? h->fref[list][i+1]->i_poc < h->fref_nearest[list]->i_poc
2150                          : h->fref[list][i+1]->i_poc > h->fref_nearest[list]->i_poc )
2151                     h->fref_nearest[list] = h->fref[list][i+1];
2152                 if( x264_reference_distance( h, h->fref[list][i] ) > x264_reference_distance( h, h->fref[list][i+1] ) )
2153                 {
2154                     XCHG( x264_frame_t*, h->fref[list][i], h->fref[list][i+1] );
2155                     b_ok = 0;
2156                     break;
2157                 }
2158             }
2159         } while( !b_ok );
2160     }
2161
2162     if( h->sh.i_mmco_remove_from_end )
2163         for( int i = h->i_ref[0]-1; i >= h->i_ref[0] - h->sh.i_mmco_remove_from_end; i-- )
2164         {
2165             int diff = h->i_frame_num - h->fref[0][i]->i_frame_num;
2166             h->sh.mmco[h->sh.i_mmco_command_count].i_poc = h->fref[0][i]->i_poc;
2167             h->sh.mmco[h->sh.i_mmco_command_count++].i_difference_of_pic_nums = diff;
2168         }
2169
2170     x264_reference_check_reorder( h );
2171
2172     h->i_ref[1] = X264_MIN( h->i_ref[1], h->frames.i_max_ref1 );
2173     h->i_ref[0] = X264_MIN( h->i_ref[0], h->frames.i_max_ref0 );
2174     h->i_ref[0] = X264_MIN( h->i_ref[0], h->param.i_frame_reference ); // if reconfig() has lowered the limit
2175
2176     /* For Blu-ray compliance, don't reference frames outside of the minigop. */
2177     if( IS_X264_TYPE_B( h->fenc->i_type ) && h->param.b_bluray_compat )
2178         h->i_ref[0] = X264_MIN( h->i_ref[0], IS_X264_TYPE_B( h->fref[0][0]->i_type ) + 1 );
2179
2180     /* add duplicates */
2181     if( h->fenc->i_type == X264_TYPE_P )
2182     {
2183         int idx = -1;
2184         if( h->param.analyse.i_weighted_pred >= X264_WEIGHTP_SIMPLE )
2185         {
2186             x264_weight_t w[3];
2187             w[1].weightfn = w[2].weightfn = NULL;
2188             if( h->param.rc.b_stat_read )
2189                 x264_ratecontrol_set_weights( h, h->fenc );
2190
2191             if( !h->fenc->weight[0][0].weightfn )
2192             {
2193                 h->fenc->weight[0][0].i_denom = 0;
2194                 SET_WEIGHT( w[0], 1, 1, 0, -1 );
2195                 idx = x264_weighted_reference_duplicate( h, 0, w );
2196             }
2197             else
2198             {
2199                 if( h->fenc->weight[0][0].i_scale == 1<<h->fenc->weight[0][0].i_denom )
2200                 {
2201                     SET_WEIGHT( h->fenc->weight[0][0], 1, 1, 0, h->fenc->weight[0][0].i_offset );
2202                 }
2203                 x264_weighted_reference_duplicate( h, 0, x264_weight_none );
2204                 if( h->fenc->weight[0][0].i_offset > -128 )
2205                 {
2206                     w[0] = h->fenc->weight[0][0];
2207                     w[0].i_offset--;
2208                     h->mc.weight_cache( h, &w[0] );
2209                     idx = x264_weighted_reference_duplicate( h, 0, w );
2210                 }
2211             }
2212         }
2213         h->mb.ref_blind_dupe = idx;
2214     }
2215
2216     assert( h->i_ref[0] + h->i_ref[1] <= X264_REF_MAX );
2217     h->mb.pic.i_fref[0] = h->i_ref[0];
2218     h->mb.pic.i_fref[1] = h->i_ref[1];
2219 }
2220
2221 static void x264_fdec_filter_row( x264_t *h, int mb_y, int pass )
2222 {
2223     /* mb_y is the mb to be encoded next, not the mb to be filtered here */
2224     int b_hpel = h->fdec->b_kept_as_ref;
2225     int b_deblock = h->sh.i_disable_deblocking_filter_idc != 1;
2226     int b_end = mb_y == h->i_threadslice_end;
2227     int b_measure_quality = 1;
2228     int min_y = mb_y - (1 << SLICE_MBAFF);
2229     int b_start = min_y == h->i_threadslice_start;
2230     /* Even in interlaced mode, deblocking never modifies more than 4 pixels
2231      * above each MB, as bS=4 doesn't happen for the top of interlaced mbpairs. */
2232     int minpix_y = min_y*16 - 4 * !b_start;
2233     int maxpix_y = mb_y*16 - 4 * !b_end;
2234     b_deblock &= b_hpel || h->param.b_full_recon || h->param.psz_dump_yuv;
2235     if( h->param.b_sliced_threads )
2236     {
2237         switch( pass )
2238         {
2239             /* During encode: only do deblock if asked for */
2240             default:
2241             case 0:
2242                 b_deblock &= h->param.b_full_recon;
2243                 b_hpel = 0;
2244                 break;
2245             /* During post-encode pass: do deblock if not done yet, do hpel for all
2246              * rows except those between slices. */
2247             case 1:
2248                 b_deblock &= !h->param.b_full_recon;
2249                 b_hpel &= !(b_start && min_y > 0);
2250                 b_measure_quality = 0;
2251                 break;
2252             /* Final pass: do the rows between slices in sequence. */
2253             case 2:
2254                 b_deblock = 0;
2255                 b_measure_quality = 0;
2256                 break;
2257         }
2258     }
2259     if( mb_y & SLICE_MBAFF )
2260         return;
2261     if( min_y < h->i_threadslice_start )
2262         return;
2263
2264     if( b_deblock )
2265         for( int y = min_y; y < mb_y; y += (1 << SLICE_MBAFF) )
2266             x264_frame_deblock_row( h, y );
2267
2268     /* FIXME: Prediction requires different borders for interlaced/progressive mc,
2269      * but the actual image data is equivalent. For now, maintain this
2270      * consistency by copying deblocked pixels between planes. */
2271     if( PARAM_INTERLACED && (!h->param.b_sliced_threads || pass == 1) )
2272         for( int p = 0; p < h->fdec->i_plane; p++ )
2273             for( int i = minpix_y>>(CHROMA_V_SHIFT && p); i < maxpix_y>>(CHROMA_V_SHIFT && p); i++ )
2274                 memcpy( h->fdec->plane_fld[p] + i*h->fdec->i_stride[p],
2275                         h->fdec->plane[p]     + i*h->fdec->i_stride[p],
2276                         h->mb.i_mb_width*16*sizeof(pixel) );
2277
2278     if( h->fdec->b_kept_as_ref && (!h->param.b_sliced_threads || pass == 1) )
2279         x264_frame_expand_border( h, h->fdec, min_y );
2280     if( b_hpel )
2281     {
2282         int end = mb_y == h->mb.i_mb_height;
2283         /* Can't do hpel until the previous slice is done encoding. */
2284         if( h->param.analyse.i_subpel_refine )
2285         {
2286             x264_frame_filter( h, h->fdec, min_y, end );
2287             x264_frame_expand_border_filtered( h, h->fdec, min_y, end );
2288         }
2289     }
2290
2291     if( SLICE_MBAFF && pass == 0 )
2292         for( int i = 0; i < 3; i++ )
2293         {
2294             XCHG( pixel *, h->intra_border_backup[0][i], h->intra_border_backup[3][i] );
2295             XCHG( pixel *, h->intra_border_backup[1][i], h->intra_border_backup[4][i] );
2296         }
2297
2298     if( h->i_thread_frames > 1 && h->fdec->b_kept_as_ref )
2299         x264_frame_cond_broadcast( h->fdec, mb_y*16 + (b_end ? 10000 : -(X264_THREAD_HEIGHT << SLICE_MBAFF)) );
2300
2301     if( b_measure_quality )
2302     {
2303         maxpix_y = X264_MIN( maxpix_y, h->param.i_height );
2304         if( h->param.analyse.b_psnr )
2305         {
2306             for( int p = 0; p < (CHROMA444 ? 3 : 1); p++ )
2307                 h->stat.frame.i_ssd[p] += x264_pixel_ssd_wxh( &h->pixf,
2308                     h->fdec->plane[p] + minpix_y * h->fdec->i_stride[p], h->fdec->i_stride[p],
2309                     h->fenc->plane[p] + minpix_y * h->fenc->i_stride[p], h->fenc->i_stride[p],
2310                     h->param.i_width, maxpix_y-minpix_y );
2311             if( !CHROMA444 )
2312             {
2313                 uint64_t ssd_u, ssd_v;
2314                 int v_shift = CHROMA_V_SHIFT;
2315                 x264_pixel_ssd_nv12( &h->pixf,
2316                     h->fdec->plane[1] + (minpix_y>>v_shift) * h->fdec->i_stride[1], h->fdec->i_stride[1],
2317                     h->fenc->plane[1] + (minpix_y>>v_shift) * h->fenc->i_stride[1], h->fenc->i_stride[1],
2318                     h->param.i_width>>1, (maxpix_y-minpix_y)>>v_shift, &ssd_u, &ssd_v );
2319                 h->stat.frame.i_ssd[1] += ssd_u;
2320                 h->stat.frame.i_ssd[2] += ssd_v;
2321             }
2322         }
2323
2324         if( h->param.analyse.b_ssim )
2325         {
2326             int ssim_cnt;
2327             x264_emms();
2328             /* offset by 2 pixels to avoid alignment of ssim blocks with dct blocks,
2329              * and overlap by 4 */
2330             minpix_y += b_start ? 2 : -6;
2331             h->stat.frame.f_ssim +=
2332                 x264_pixel_ssim_wxh( &h->pixf,
2333                     h->fdec->plane[0] + 2+minpix_y*h->fdec->i_stride[0], h->fdec->i_stride[0],
2334                     h->fenc->plane[0] + 2+minpix_y*h->fenc->i_stride[0], h->fenc->i_stride[0],
2335                     h->param.i_width-2, maxpix_y-minpix_y, h->scratch_buffer, &ssim_cnt );
2336             h->stat.frame.i_ssim_cnt += ssim_cnt;
2337         }
2338     }
2339 }
2340
2341 static inline int x264_reference_update( x264_t *h )
2342 {
2343     if( !h->fdec->b_kept_as_ref )
2344     {
2345         if( h->i_thread_frames > 1 )
2346         {
2347             x264_frame_push_unused( h, h->fdec );
2348             h->fdec = x264_frame_pop_unused( h, 1 );
2349             if( !h->fdec )
2350                 return -1;
2351         }
2352         return 0;
2353     }
2354
2355     /* apply mmco from previous frame. */
2356     for( int i = 0; i < h->sh.i_mmco_command_count; i++ )
2357         for( int j = 0; h->frames.reference[j]; j++ )
2358             if( h->frames.reference[j]->i_poc == h->sh.mmco[i].i_poc )
2359                 x264_frame_push_unused( h, x264_frame_shift( &h->frames.reference[j] ) );
2360
2361     /* move frame in the buffer */
2362     x264_frame_push( h->frames.reference, h->fdec );
2363     if( h->frames.reference[h->sps->i_num_ref_frames] )
2364         x264_frame_push_unused( h, x264_frame_shift( h->frames.reference ) );
2365     h->fdec = x264_frame_pop_unused( h, 1 );
2366     if( !h->fdec )
2367         return -1;
2368     return 0;
2369 }
2370
2371 static inline void x264_reference_reset( x264_t *h )
2372 {
2373     while( h->frames.reference[0] )
2374         x264_frame_push_unused( h, x264_frame_pop( h->frames.reference ) );
2375     h->fdec->i_poc =
2376     h->fenc->i_poc = 0;
2377 }
2378
2379 static inline void x264_reference_hierarchy_reset( x264_t *h )
2380 {
2381     int ref;
2382     int b_hasdelayframe = 0;
2383
2384     /* look for delay frames -- chain must only contain frames that are disposable */
2385     for( int i = 0; h->frames.current[i] && IS_DISPOSABLE( h->frames.current[i]->i_type ); i++ )
2386         b_hasdelayframe |= h->frames.current[i]->i_coded
2387                         != h->frames.current[i]->i_frame + h->sps->vui.i_num_reorder_frames;
2388
2389     /* This function must handle b-pyramid and clear frames for open-gop */
2390     if( h->param.i_bframe_pyramid != X264_B_PYRAMID_STRICT && !b_hasdelayframe && h->frames.i_poc_last_open_gop == -1 )
2391         return;
2392
2393     /* Remove last BREF. There will never be old BREFs in the
2394      * dpb during a BREF decode when pyramid == STRICT */
2395     for( ref = 0; h->frames.reference[ref]; ref++ )
2396     {
2397         if( ( h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT
2398             && h->frames.reference[ref]->i_type == X264_TYPE_BREF )
2399             || ( h->frames.reference[ref]->i_poc < h->frames.i_poc_last_open_gop
2400             && h->sh.i_type != SLICE_TYPE_B ) )
2401         {
2402             int diff = h->i_frame_num - h->frames.reference[ref]->i_frame_num;
2403             h->sh.mmco[h->sh.i_mmco_command_count].i_difference_of_pic_nums = diff;
2404             h->sh.mmco[h->sh.i_mmco_command_count++].i_poc = h->frames.reference[ref]->i_poc;
2405             x264_frame_push_unused( h, x264_frame_shift( &h->frames.reference[ref] ) );
2406             h->b_ref_reorder[0] = 1;
2407             ref--;
2408         }
2409     }
2410
2411     /* Prepare room in the dpb for the delayed display time of the later b-frame's */
2412     if( h->param.i_bframe_pyramid )
2413         h->sh.i_mmco_remove_from_end = X264_MAX( ref + 2 - h->frames.i_max_dpb, 0 );
2414 }
2415
2416 static inline void x264_slice_init( x264_t *h, int i_nal_type, int i_global_qp )
2417 {
2418     /* ------------------------ Create slice header  ----------------------- */
2419     if( i_nal_type == NAL_SLICE_IDR )
2420     {
2421         x264_slice_header_init( h, &h->sh, h->sps, h->pps, h->i_idr_pic_id, h->i_frame_num, i_global_qp );
2422
2423         /* alternate id */
2424         h->i_idr_pic_id ^= 1;
2425     }
2426     else
2427     {
2428         x264_slice_header_init( h, &h->sh, h->sps, h->pps, -1, h->i_frame_num, i_global_qp );
2429
2430         h->sh.i_num_ref_idx_l0_active = h->i_ref[0] <= 0 ? 1 : h->i_ref[0];
2431         h->sh.i_num_ref_idx_l1_active = h->i_ref[1] <= 0 ? 1 : h->i_ref[1];
2432         if( h->sh.i_num_ref_idx_l0_active != h->pps->i_num_ref_idx_l0_default_active ||
2433             (h->sh.i_type == SLICE_TYPE_B && h->sh.i_num_ref_idx_l1_active != h->pps->i_num_ref_idx_l1_default_active) )
2434         {
2435             h->sh.b_num_ref_idx_override = 1;
2436         }
2437     }
2438
2439     if( h->fenc->i_type == X264_TYPE_BREF && h->param.b_bluray_compat && h->sh.i_mmco_command_count )
2440     {
2441         h->b_sh_backup = 1;
2442         h->sh_backup = h->sh;
2443     }
2444
2445     h->fdec->i_frame_num = h->sh.i_frame_num;
2446
2447     if( h->sps->i_poc_type == 0 )
2448     {
2449         h->sh.i_poc = h->fdec->i_poc;
2450         if( PARAM_INTERLACED )
2451         {
2452             h->sh.i_delta_poc_bottom = h->param.b_tff ? 1 : -1;
2453             h->sh.i_poc += h->sh.i_delta_poc_bottom == -1;
2454         }
2455         else
2456             h->sh.i_delta_poc_bottom = 0;
2457         h->fdec->i_delta_poc[0] = h->sh.i_delta_poc_bottom == -1;
2458         h->fdec->i_delta_poc[1] = h->sh.i_delta_poc_bottom ==  1;
2459     }
2460     else
2461     {
2462         /* Nothing to do ? */
2463     }
2464
2465     x264_macroblock_slice_init( h );
2466 }
2467
2468 typedef struct
2469 {
2470     int skip;
2471     uint8_t cabac_prevbyte;
2472     bs_t bs;
2473     x264_cabac_t cabac;
2474     x264_frame_stat_t stat;
2475     int last_qp;
2476     int last_dqp;
2477     int field_decoding_flag;
2478 } x264_bs_bak_t;
2479
2480 static ALWAYS_INLINE void x264_bitstream_backup( x264_t *h, x264_bs_bak_t *bak, int i_skip, int full )
2481 {
2482     if( full )
2483     {
2484         bak->stat = h->stat.frame;
2485         bak->last_qp = h->mb.i_last_qp;
2486         bak->last_dqp = h->mb.i_last_dqp;
2487         bak->field_decoding_flag = h->mb.field_decoding_flag;
2488     }
2489     else
2490     {
2491         bak->stat.i_mv_bits = h->stat.frame.i_mv_bits;
2492         bak->stat.i_tex_bits = h->stat.frame.i_tex_bits;
2493     }
2494     /* In the per-MB backup, we don't need the contexts because flushing the CABAC
2495      * encoder has no context dependency and in this case, a slice is ended (and
2496      * thus the content of all contexts are thrown away). */
2497     if( h->param.b_cabac )
2498     {
2499         if( full )
2500             memcpy( &bak->cabac, &h->cabac, sizeof(x264_cabac_t) );
2501         else
2502             memcpy( &bak->cabac, &h->cabac, offsetof(x264_cabac_t, f8_bits_encoded) );
2503         /* x264's CABAC writer modifies the previous byte during carry, so it has to be
2504          * backed up. */
2505         bak->cabac_prevbyte = h->cabac.p[-1];
2506     }
2507     else
2508     {
2509         bak->bs = h->out.bs;
2510         bak->skip = i_skip;
2511     }
2512 }
2513
2514 static ALWAYS_INLINE void x264_bitstream_restore( x264_t *h, x264_bs_bak_t *bak, int *skip, int full )
2515 {
2516     if( full )
2517     {
2518         h->stat.frame = bak->stat;
2519         h->mb.i_last_qp = bak->last_qp;
2520         h->mb.i_last_dqp = bak->last_dqp;
2521         h->mb.field_decoding_flag = bak->field_decoding_flag;
2522     }
2523     else
2524     {
2525         h->stat.frame.i_mv_bits = bak->stat.i_mv_bits;
2526         h->stat.frame.i_tex_bits = bak->stat.i_tex_bits;
2527     }
2528     if( h->param.b_cabac )
2529     {
2530         if( full )
2531             memcpy( &h->cabac, &bak->cabac, sizeof(x264_cabac_t) );
2532         else
2533             memcpy( &h->cabac, &bak->cabac, offsetof(x264_cabac_t, f8_bits_encoded) );
2534         h->cabac.p[-1] = bak->cabac_prevbyte;
2535     }
2536     else
2537     {
2538         h->out.bs = bak->bs;
2539         *skip = bak->skip;
2540     }
2541 }
2542
2543 static int x264_slice_write( x264_t *h )
2544 {
2545     int i_skip;
2546     int mb_xy, i_mb_x, i_mb_y;
2547     /* NALUs other than the first use a 3-byte startcode.
2548      * Add one extra byte for the rbsp, and one more for the final CABAC putbyte.
2549      * Then add an extra 5 bytes just in case, to account for random NAL escapes and
2550      * other inaccuracies. */
2551     int overhead_guess = (NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal)) + 1 + h->param.b_cabac + 5;
2552     int slice_max_size = h->param.i_slice_max_size > 0 ? (h->param.i_slice_max_size-overhead_guess)*8 : 0;
2553     int back_up_bitstream = slice_max_size || (!h->param.b_cabac && h->sps->i_profile_idc < PROFILE_HIGH);
2554     int starting_bits = bs_pos(&h->out.bs);
2555     int b_deblock = h->sh.i_disable_deblocking_filter_idc != 1;
2556     int b_hpel = h->fdec->b_kept_as_ref;
2557     int orig_last_mb = h->sh.i_last_mb;
2558     int thread_last_mb = h->i_threadslice_end * h->mb.i_mb_width - 1;
2559     uint8_t *last_emu_check;
2560 #define BS_BAK_SLICE_MAX_SIZE 0
2561 #define BS_BAK_SLICE_MIN_MBS  1
2562 #define BS_BAK_ROW_VBV        2
2563     x264_bs_bak_t bs_bak[3];
2564     b_deblock &= b_hpel || h->param.b_full_recon || h->param.psz_dump_yuv;
2565     bs_realign( &h->out.bs );
2566
2567     /* Slice */
2568     x264_nal_start( h, h->i_nal_type, h->i_nal_ref_idc );
2569     h->out.nal[h->out.i_nal].i_first_mb = h->sh.i_first_mb;
2570
2571     /* Slice header */
2572     x264_macroblock_thread_init( h );
2573
2574     /* Set the QP equal to the first QP in the slice for more accurate CABAC initialization. */
2575     h->mb.i_mb_xy = h->sh.i_first_mb;
2576     h->sh.i_qp = x264_ratecontrol_mb_qp( h );
2577     h->sh.i_qp = SPEC_QP( h->sh.i_qp );
2578     h->sh.i_qp_delta = h->sh.i_qp - h->pps->i_pic_init_qp;
2579
2580     x264_slice_header_write( &h->out.bs, &h->sh, h->i_nal_ref_idc );
2581     if( h->param.b_cabac )
2582     {
2583         /* alignment needed */
2584         bs_align_1( &h->out.bs );
2585
2586         /* init cabac */
2587         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 );
2588         x264_cabac_encode_init ( &h->cabac, h->out.bs.p, h->out.bs.p_end );
2589         last_emu_check = h->cabac.p;
2590     }
2591     else
2592         last_emu_check = h->out.bs.p;
2593     h->mb.i_last_qp = h->sh.i_qp;
2594     h->mb.i_last_dqp = 0;
2595     h->mb.field_decoding_flag = 0;
2596
2597     i_mb_y = h->sh.i_first_mb / h->mb.i_mb_width;
2598     i_mb_x = h->sh.i_first_mb % h->mb.i_mb_width;
2599     i_skip = 0;
2600
2601     while( 1 )
2602     {
2603         mb_xy = i_mb_x + i_mb_y * h->mb.i_mb_width;
2604         int mb_spos = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
2605
2606         if( i_mb_x == 0 )
2607         {
2608             if( x264_bitstream_check_buffer( h ) )
2609                 return -1;
2610             if( !(i_mb_y & SLICE_MBAFF) && h->param.rc.i_vbv_buffer_size )
2611                 x264_bitstream_backup( h, &bs_bak[BS_BAK_ROW_VBV], i_skip, 1 );
2612             if( !h->mb.b_reencode_mb )
2613                 x264_fdec_filter_row( h, i_mb_y, 0 );
2614         }
2615
2616         if( !(i_mb_y & SLICE_MBAFF) && back_up_bitstream )
2617         {
2618             x264_bitstream_backup( h, &bs_bak[BS_BAK_SLICE_MAX_SIZE], i_skip, 0 );
2619             if( slice_max_size && (thread_last_mb+1-mb_xy) == h->param.i_slice_min_mbs )
2620                 x264_bitstream_backup( h, &bs_bak[BS_BAK_SLICE_MIN_MBS], i_skip, 0 );
2621         }
2622
2623         if( PARAM_INTERLACED )
2624         {
2625             if( h->mb.b_adaptive_mbaff )
2626             {
2627                 if( !(i_mb_y&1) )
2628                 {
2629                     /* FIXME: VSAD is fast but fairly poor at choosing the best interlace type. */
2630                     h->mb.b_interlaced = x264_field_vsad( h, i_mb_x, i_mb_y );
2631                     memcpy( &h->zigzagf, MB_INTERLACED ? &h->zigzagf_interlaced : &h->zigzagf_progressive, sizeof(h->zigzagf) );
2632                     if( !MB_INTERLACED && (i_mb_y+2) == h->mb.i_mb_height )
2633                         x264_expand_border_mbpair( h, i_mb_x, i_mb_y );
2634                 }
2635             }
2636             h->mb.field[mb_xy] = MB_INTERLACED;
2637         }
2638
2639         /* load cache */
2640         if( SLICE_MBAFF )
2641             x264_macroblock_cache_load_interlaced( h, i_mb_x, i_mb_y );
2642         else
2643             x264_macroblock_cache_load_progressive( h, i_mb_x, i_mb_y );
2644
2645         x264_macroblock_analyse( h );
2646
2647         /* encode this macroblock -> be careful it can change the mb type to P_SKIP if needed */
2648 reencode:
2649         x264_macroblock_encode( h );
2650
2651         if( h->param.b_cabac )
2652         {
2653             if( mb_xy > h->sh.i_first_mb && !(SLICE_MBAFF && (i_mb_y&1)) )
2654                 x264_cabac_encode_terminal( &h->cabac );
2655
2656             if( IS_SKIP( h->mb.i_type ) )
2657                 x264_cabac_mb_skip( h, 1 );
2658             else
2659             {
2660                 if( h->sh.i_type != SLICE_TYPE_I )
2661                     x264_cabac_mb_skip( h, 0 );
2662                 x264_macroblock_write_cabac( h, &h->cabac );
2663             }
2664         }
2665         else
2666         {
2667             if( IS_SKIP( h->mb.i_type ) )
2668                 i_skip++;
2669             else
2670             {
2671                 if( h->sh.i_type != SLICE_TYPE_I )
2672                 {
2673                     bs_write_ue( &h->out.bs, i_skip );  /* skip run */
2674                     i_skip = 0;
2675                 }
2676                 x264_macroblock_write_cavlc( h );
2677                 /* If there was a CAVLC level code overflow, try again at a higher QP. */
2678                 if( h->mb.b_overflow )
2679                 {
2680                     h->mb.i_chroma_qp = h->chroma_qp_table[++h->mb.i_qp];
2681                     h->mb.i_skip_intra = 0;
2682                     h->mb.b_skip_mc = 0;
2683                     h->mb.b_overflow = 0;
2684                     x264_bitstream_restore( h, &bs_bak[BS_BAK_SLICE_MAX_SIZE], &i_skip, 0 );
2685                     goto reencode;
2686                 }
2687             }
2688         }
2689
2690         int total_bits = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
2691         int mb_size = total_bits - mb_spos;
2692
2693         if( slice_max_size && (!SLICE_MBAFF || (i_mb_y&1)) )
2694         {
2695             /* Count the skip run, just in case. */
2696             if( !h->param.b_cabac )
2697                 total_bits += bs_size_ue_big( i_skip );
2698             /* Check for escape bytes. */
2699             uint8_t *end = h->param.b_cabac ? h->cabac.p : h->out.bs.p;
2700             for( ; last_emu_check < end - 2; last_emu_check++ )
2701                 if( last_emu_check[0] == 0 && last_emu_check[1] == 0 && last_emu_check[2] <= 3 )
2702                 {
2703                     slice_max_size -= 8;
2704                     last_emu_check++;
2705                 }
2706             /* We'll just re-encode this last macroblock if we go over the max slice size. */
2707             if( total_bits - starting_bits > slice_max_size && !h->mb.b_reencode_mb )
2708             {
2709                 if( !x264_frame_new_slice( h, h->fdec ) )
2710                 {
2711                     /* Handle the most obnoxious slice-min-mbs edge case: we need to end the slice
2712                      * because it's gone over the maximum size, but doing so would violate slice-min-mbs.
2713                      * If possible, roll back to the last checkpoint and try again.
2714                      * We could try raising QP, but that would break in the case where a slice spans multiple
2715                      * rows, which the re-encoding infrastructure can't currently handle. */
2716                     if( mb_xy <= thread_last_mb && (thread_last_mb+1-mb_xy) < h->param.i_slice_min_mbs )
2717                     {
2718                         if( thread_last_mb-h->param.i_slice_min_mbs < h->sh.i_first_mb+h->param.i_slice_min_mbs )
2719                         {
2720                             x264_log( h, X264_LOG_WARNING, "slice-max-size violated (frame %d, cause: slice-min-mbs)\n", h->i_frame );
2721                             slice_max_size = 0;
2722                             goto cont;
2723                         }
2724                         x264_bitstream_restore( h, &bs_bak[BS_BAK_SLICE_MIN_MBS], &i_skip, 0 );
2725                         h->mb.b_reencode_mb = 1;
2726                         h->sh.i_last_mb = thread_last_mb-h->param.i_slice_min_mbs;
2727                         break;
2728                     }
2729                     if( mb_xy-SLICE_MBAFF*h->mb.i_mb_stride != h->sh.i_first_mb )
2730                     {
2731                         x264_bitstream_restore( h, &bs_bak[BS_BAK_SLICE_MAX_SIZE], &i_skip, 0 );
2732                         h->mb.b_reencode_mb = 1;
2733                         if( SLICE_MBAFF )
2734                         {
2735                             // set to bottom of previous mbpair
2736                             if( i_mb_x )
2737                                 h->sh.i_last_mb = mb_xy-1+h->mb.i_mb_stride*(!(i_mb_y&1));
2738                             else
2739                                 h->sh.i_last_mb = (i_mb_y-2+!(i_mb_y&1))*h->mb.i_mb_stride + h->mb.i_mb_width - 1;
2740                         }
2741                         else
2742                             h->sh.i_last_mb = mb_xy-1;
2743                         break;
2744                     }
2745                     else
2746                         h->sh.i_last_mb = mb_xy;
2747                 }
2748                 else
2749                     slice_max_size = 0;
2750             }
2751         }
2752 cont:
2753         h->mb.b_reencode_mb = 0;
2754
2755         /* save cache */
2756         x264_macroblock_cache_save( h );
2757
2758         if( x264_ratecontrol_mb( h, mb_size ) < 0 )
2759         {
2760             x264_bitstream_restore( h, &bs_bak[BS_BAK_ROW_VBV], &i_skip, 1 );
2761             h->mb.b_reencode_mb = 1;
2762             i_mb_x = 0;
2763             i_mb_y = i_mb_y - SLICE_MBAFF;
2764             h->mb.i_mb_prev_xy = i_mb_y * h->mb.i_mb_stride - 1;
2765             h->sh.i_last_mb = orig_last_mb;
2766             continue;
2767         }
2768
2769         /* accumulate mb stats */
2770         h->stat.frame.i_mb_count[h->mb.i_type]++;
2771
2772         int b_intra = IS_INTRA( h->mb.i_type );
2773         int b_skip = IS_SKIP( h->mb.i_type );
2774         if( h->param.i_log_level >= X264_LOG_INFO || h->param.rc.b_stat_write )
2775         {
2776             if( !b_intra && !b_skip && !IS_DIRECT( h->mb.i_type ) )
2777             {
2778                 if( h->mb.i_partition != D_8x8 )
2779                         h->stat.frame.i_mb_partition[h->mb.i_partition] += 4;
2780                     else
2781                         for( int i = 0; i < 4; i++ )
2782                             h->stat.frame.i_mb_partition[h->mb.i_sub_partition[i]] ++;
2783                 if( h->param.i_frame_reference > 1 )
2784                     for( int i_list = 0; i_list <= (h->sh.i_type == SLICE_TYPE_B); i_list++ )
2785                         for( int i = 0; i < 4; i++ )
2786                         {
2787                             int i_ref = h->mb.cache.ref[i_list][ x264_scan8[4*i] ];
2788                             if( i_ref >= 0 )
2789                                 h->stat.frame.i_mb_count_ref[i_list][i_ref] ++;
2790                         }
2791             }
2792         }
2793
2794         if( h->param.i_log_level >= X264_LOG_INFO )
2795         {
2796             if( h->mb.i_cbp_luma | h->mb.i_cbp_chroma )
2797             {
2798                 if( CHROMA444 )
2799                 {
2800                     for( int i = 0; i < 4; i++ )
2801                         if( h->mb.i_cbp_luma & (1 << i) )
2802                             for( int p = 0; p < 3; p++ )
2803                             {
2804                                 int s8 = i*4+p*16;
2805                                 int nnz8x8 = M16( &h->mb.cache.non_zero_count[x264_scan8[s8]+0] )
2806                                            | M16( &h->mb.cache.non_zero_count[x264_scan8[s8]+8] );
2807                                 h->stat.frame.i_mb_cbp[!b_intra + p*2] += !!nnz8x8;
2808                             }
2809                 }
2810                 else
2811                 {
2812                     int cbpsum = (h->mb.i_cbp_luma&1) + ((h->mb.i_cbp_luma>>1)&1)
2813                                + ((h->mb.i_cbp_luma>>2)&1) + (h->mb.i_cbp_luma>>3);
2814                     h->stat.frame.i_mb_cbp[!b_intra + 0] += cbpsum;
2815                     h->stat.frame.i_mb_cbp[!b_intra + 2] += !!h->mb.i_cbp_chroma;
2816                     h->stat.frame.i_mb_cbp[!b_intra + 4] += h->mb.i_cbp_chroma >> 1;
2817                 }
2818             }
2819             if( h->mb.i_cbp_luma && !b_intra )
2820             {
2821                 h->stat.frame.i_mb_count_8x8dct[0] ++;
2822                 h->stat.frame.i_mb_count_8x8dct[1] += h->mb.b_transform_8x8;
2823             }
2824             if( b_intra && h->mb.i_type != I_PCM )
2825             {
2826                 if( h->mb.i_type == I_16x16 )
2827                     h->stat.frame.i_mb_pred_mode[0][h->mb.i_intra16x16_pred_mode]++;
2828                 else if( h->mb.i_type == I_8x8 )
2829                     for( int i = 0; i < 16; i += 4 )
2830                         h->stat.frame.i_mb_pred_mode[1][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
2831                 else //if( h->mb.i_type == I_4x4 )
2832                     for( int i = 0; i < 16; i++ )
2833                         h->stat.frame.i_mb_pred_mode[2][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
2834                 h->stat.frame.i_mb_pred_mode[3][x264_mb_chroma_pred_mode_fix[h->mb.i_chroma_pred_mode]]++;
2835             }
2836             h->stat.frame.i_mb_field[b_intra?0:b_skip?2:1] += MB_INTERLACED;
2837         }
2838
2839         /* calculate deblock strength values (actual deblocking is done per-row along with hpel) */
2840         if( b_deblock )
2841             x264_macroblock_deblock_strength( h );
2842
2843         if( mb_xy == h->sh.i_last_mb )
2844             break;
2845
2846         if( SLICE_MBAFF )
2847         {
2848             i_mb_x += i_mb_y & 1;
2849             i_mb_y ^= i_mb_x < h->mb.i_mb_width;
2850         }
2851         else
2852             i_mb_x++;
2853         if( i_mb_x == h->mb.i_mb_width )
2854         {
2855             i_mb_y++;
2856             i_mb_x = 0;
2857         }
2858     }
2859     if( h->sh.i_last_mb < h->sh.i_first_mb )
2860         return 0;
2861
2862     h->out.nal[h->out.i_nal].i_last_mb = h->sh.i_last_mb;
2863
2864     if( h->param.b_cabac )
2865     {
2866         x264_cabac_encode_flush( h, &h->cabac );
2867         h->out.bs.p = h->cabac.p;
2868     }
2869     else
2870     {
2871         if( i_skip > 0 )
2872             bs_write_ue( &h->out.bs, i_skip );  /* last skip run */
2873         /* rbsp_slice_trailing_bits */
2874         bs_rbsp_trailing( &h->out.bs );
2875         bs_flush( &h->out.bs );
2876     }
2877     if( x264_nal_end( h ) )
2878         return -1;
2879
2880     if( h->sh.i_last_mb == (h->i_threadslice_end * h->mb.i_mb_width - 1) )
2881     {
2882         h->stat.frame.i_misc_bits = bs_pos( &h->out.bs )
2883                                   + (h->out.i_nal*NALU_OVERHEAD * 8)
2884                                   - h->stat.frame.i_tex_bits
2885                                   - h->stat.frame.i_mv_bits;
2886         x264_fdec_filter_row( h, h->i_threadslice_end, 0 );
2887
2888         if( h->param.b_sliced_threads )
2889         {
2890             /* Tell the main thread we're done. */
2891             x264_threadslice_cond_broadcast( h, 1 );
2892             /* Do hpel now */
2893             for( int mb_y = h->i_threadslice_start; mb_y <= h->i_threadslice_end; mb_y++ )
2894                 x264_fdec_filter_row( h, mb_y, 1 );
2895             x264_threadslice_cond_broadcast( h, 2 );
2896             /* Do the first row of hpel, now that the previous slice is done */
2897             if( h->i_thread_idx > 0 )
2898             {
2899                 x264_threadslice_cond_wait( h->thread[h->i_thread_idx-1], 2 );
2900                 x264_fdec_filter_row( h, h->i_threadslice_start + (1 << SLICE_MBAFF), 2 );
2901             }
2902         }
2903
2904         /* Free mb info after the last thread's done using it */
2905         if( h->fdec->mb_info_free && (!h->param.b_sliced_threads || h->i_thread_idx == (h->param.i_threads-1)) )
2906         {
2907             h->fdec->mb_info_free( h->fdec->mb_info );
2908             h->fdec->mb_info = NULL;
2909             h->fdec->mb_info_free = NULL;
2910         }
2911     }
2912
2913     return 0;
2914 }
2915
2916 static void x264_thread_sync_context( x264_t *dst, x264_t *src )
2917 {
2918     if( dst == src )
2919         return;
2920
2921     // reference counting
2922     for( x264_frame_t **f = src->frames.reference; *f; f++ )
2923         (*f)->i_reference_count++;
2924     for( x264_frame_t **f = dst->frames.reference; *f; f++ )
2925         x264_frame_push_unused( src, *f );
2926     src->fdec->i_reference_count++;
2927     x264_frame_push_unused( src, dst->fdec );
2928
2929     // copy everything except the per-thread pointers and the constants.
2930     memcpy( &dst->i_frame, &src->i_frame, offsetof(x264_t, mb.base) - offsetof(x264_t, i_frame) );
2931     dst->param = src->param;
2932     dst->stat = src->stat;
2933     dst->pixf = src->pixf;
2934     dst->reconfig = src->reconfig;
2935 }
2936
2937 static void x264_thread_sync_stat( x264_t *dst, x264_t *src )
2938 {
2939     if( dst == src )
2940         return;
2941     memcpy( &dst->stat.i_frame_count, &src->stat.i_frame_count, sizeof(dst->stat) - sizeof(dst->stat.frame) );
2942 }
2943
2944 static void *x264_slices_write( x264_t *h )
2945 {
2946     int i_slice_num = 0;
2947     int last_thread_mb = h->sh.i_last_mb;
2948
2949     /* init stats */
2950     memset( &h->stat.frame, 0, sizeof(h->stat.frame) );
2951     h->mb.b_reencode_mb = 0;
2952     while( h->sh.i_first_mb + SLICE_MBAFF*h->mb.i_mb_stride <= last_thread_mb )
2953     {
2954         h->sh.i_last_mb = last_thread_mb;
2955         if( !i_slice_num || !x264_frame_new_slice( h, h->fdec ) )
2956         {
2957             if( h->param.i_slice_max_mbs )
2958             {
2959                 if( SLICE_MBAFF )
2960                 {
2961                     // convert first to mbaff form, add slice-max-mbs, then convert back to normal form
2962                     int last_mbaff = 2*(h->sh.i_first_mb % h->mb.i_mb_width)
2963                         + h->mb.i_mb_width*(h->sh.i_first_mb / h->mb.i_mb_width)
2964                         + h->param.i_slice_max_mbs - 1;
2965                     int last_x = (last_mbaff % (2*h->mb.i_mb_width))/2;
2966                     int last_y = (last_mbaff / (2*h->mb.i_mb_width))*2 + 1;
2967                     h->sh.i_last_mb = last_x + h->mb.i_mb_stride*last_y;
2968                 }
2969                 else
2970                 {
2971                     h->sh.i_last_mb = h->sh.i_first_mb + h->param.i_slice_max_mbs - 1;
2972                     if( h->sh.i_last_mb < last_thread_mb && last_thread_mb - h->sh.i_last_mb < h->param.i_slice_min_mbs )
2973                         h->sh.i_last_mb = last_thread_mb - h->param.i_slice_min_mbs;
2974                 }
2975                 i_slice_num++;
2976             }
2977             else if( h->param.i_slice_count && !h->param.b_sliced_threads )
2978             {
2979                 int height = h->mb.i_mb_height >> PARAM_INTERLACED;
2980                 int width = h->mb.i_mb_width << PARAM_INTERLACED;
2981                 i_slice_num++;
2982                 h->sh.i_last_mb = (height * i_slice_num + h->param.i_slice_count/2) / h->param.i_slice_count * width - 1;
2983             }
2984         }
2985         h->sh.i_last_mb = X264_MIN( h->sh.i_last_mb, last_thread_mb );
2986         if( x264_stack_align( x264_slice_write, h ) )
2987             goto fail;
2988         h->sh.i_first_mb = h->sh.i_last_mb + 1;
2989         // if i_first_mb is not the last mb in a row then go to the next mb in MBAFF order
2990         if( SLICE_MBAFF && h->sh.i_first_mb % h->mb.i_mb_width )
2991             h->sh.i_first_mb -= h->mb.i_mb_stride;
2992     }
2993
2994     return (void *)0;
2995
2996 fail:
2997     /* Tell other threads we're done, so they wouldn't wait for it */
2998     if( h->param.b_sliced_threads )
2999         x264_threadslice_cond_broadcast( h, 2 );
3000     return (void *)-1;
3001 }
3002
3003 static int x264_threaded_slices_write( x264_t *h )
3004 {
3005     /* set first/last mb and sync contexts */
3006     for( int i = 0; i < h->param.i_threads; i++ )
3007     {
3008         x264_t *t = h->thread[i];
3009         if( i )
3010         {
3011             t->param = h->param;
3012             memcpy( &t->i_frame, &h->i_frame, offsetof(x264_t, rc) - offsetof(x264_t, i_frame) );
3013         }
3014         int height = h->mb.i_mb_height >> PARAM_INTERLACED;
3015         t->i_threadslice_start = ((height *  i    + h->param.i_slice_count/2) / h->param.i_threads) << PARAM_INTERLACED;
3016         t->i_threadslice_end   = ((height * (i+1) + h->param.i_slice_count/2) / h->param.i_threads) << PARAM_INTERLACED;
3017         t->sh.i_first_mb = t->i_threadslice_start * h->mb.i_mb_width;
3018         t->sh.i_last_mb  =   t->i_threadslice_end * h->mb.i_mb_width - 1;
3019     }
3020
3021     x264_stack_align( x264_analyse_weight_frame, h, h->mb.i_mb_height*16 + 16 );
3022
3023     x264_threads_distribute_ratecontrol( h );
3024
3025     /* setup */
3026     for( int i = 0; i < h->param.i_threads; i++ )
3027     {
3028         h->thread[i]->i_thread_idx = i;
3029         h->thread[i]->b_thread_active = 1;
3030         x264_threadslice_cond_broadcast( h->thread[i], 0 );
3031     }
3032     /* dispatch */
3033     for( int i = 0; i < h->param.i_threads; i++ )
3034         x264_threadpool_run( h->threadpool, (void*)x264_slices_write, h->thread[i] );
3035     /* wait */
3036     for( int i = 0; i < h->param.i_threads; i++ )
3037         x264_threadslice_cond_wait( h->thread[i], 1 );
3038
3039     x264_threads_merge_ratecontrol( h );
3040
3041     for( int i = 1; i < h->param.i_threads; i++ )
3042     {
3043         x264_t *t = h->thread[i];
3044         for( int j = 0; j < t->out.i_nal; j++ )
3045         {
3046             h->out.nal[h->out.i_nal] = t->out.nal[j];
3047             h->out.i_nal++;
3048             x264_nal_check_buffer( h );
3049         }
3050         /* All entries in stat.frame are ints except for ssd/ssim. */
3051         for( int j = 0; j < (offsetof(x264_t,stat.frame.i_ssd) - offsetof(x264_t,stat.frame.i_mv_bits)) / sizeof(int); j++ )
3052             ((int*)&h->stat.frame)[j] += ((int*)&t->stat.frame)[j];
3053         for( int j = 0; j < 3; j++ )
3054             h->stat.frame.i_ssd[j] += t->stat.frame.i_ssd[j];
3055         h->stat.frame.f_ssim += t->stat.frame.f_ssim;
3056         h->stat.frame.i_ssim_cnt += t->stat.frame.i_ssim_cnt;
3057     }
3058
3059     return 0;
3060 }
3061
3062 void x264_encoder_intra_refresh( x264_t *h )
3063 {
3064     h = h->thread[h->i_thread_phase];
3065     h->b_queued_intra_refresh = 1;
3066 }
3067
3068 int x264_encoder_invalidate_reference( x264_t *h, int64_t pts )
3069 {
3070     if( h->param.i_bframe )
3071     {
3072         x264_log( h, X264_LOG_ERROR, "x264_encoder_invalidate_reference is not supported with B-frames enabled\n" );
3073         return -1;
3074     }
3075     if( h->param.b_intra_refresh )
3076     {
3077         x264_log( h, X264_LOG_ERROR, "x264_encoder_invalidate_reference is not supported with intra refresh enabled\n" );
3078         return -1;
3079     }
3080     h = h->thread[h->i_thread_phase];
3081     if( pts >= h->i_last_idr_pts )
3082     {
3083         for( int i = 0; h->frames.reference[i]; i++ )
3084             if( pts <= h->frames.reference[i]->i_pts )
3085                 h->frames.reference[i]->b_corrupt = 1;
3086         if( pts <= h->fdec->i_pts )
3087             h->fdec->b_corrupt = 1;
3088     }
3089     return 0;
3090 }
3091
3092 /****************************************************************************
3093  * x264_encoder_encode:
3094  *  XXX: i_poc   : is the poc of the current given picture
3095  *       i_frame : is the number of the frame being coded
3096  *  ex:  type frame poc
3097  *       I      0   2*0
3098  *       P      1   2*3
3099  *       B      2   2*1
3100  *       B      3   2*2
3101  *       P      4   2*6
3102  *       B      5   2*4
3103  *       B      6   2*5
3104  ****************************************************************************/
3105 int     x264_encoder_encode( x264_t *h,
3106                              x264_nal_t **pp_nal, int *pi_nal,
3107                              x264_picture_t *pic_in,
3108                              x264_picture_t *pic_out )
3109 {
3110     x264_t *thread_current, *thread_prev, *thread_oldest;
3111     int i_nal_type, i_nal_ref_idc, i_global_qp;
3112     int overhead = NALU_OVERHEAD;
3113
3114 #if HAVE_OPENCL
3115     if( h->opencl.b_fatal_error )
3116         return -1;
3117 #endif
3118
3119     if( h->i_thread_frames > 1 )
3120     {
3121         thread_prev    = h->thread[ h->i_thread_phase ];
3122         h->i_thread_phase = (h->i_thread_phase + 1) % h->i_thread_frames;
3123         thread_current = h->thread[ h->i_thread_phase ];
3124         thread_oldest  = h->thread[ (h->i_thread_phase + 1) % h->i_thread_frames ];
3125         x264_thread_sync_context( thread_current, thread_prev );
3126         x264_thread_sync_ratecontrol( thread_current, thread_prev, thread_oldest );
3127         h = thread_current;
3128     }
3129     else
3130     {
3131         thread_current =
3132         thread_oldest  = h;
3133     }
3134     h->i_cpb_delay_pir_offset = h->i_cpb_delay_pir_offset_next;
3135
3136     /* no data out */
3137     *pi_nal = 0;
3138     *pp_nal = NULL;
3139
3140     /* ------------------- Setup new frame from picture -------------------- */
3141     if( pic_in != NULL )
3142     {
3143         /* 1: Copy the picture to a frame and move it to a buffer */
3144         x264_frame_t *fenc = x264_frame_pop_unused( h, 0 );
3145         if( !fenc )
3146             return -1;
3147
3148         if( x264_frame_copy_picture( h, fenc, pic_in ) < 0 )
3149             return -1;
3150
3151         if( h->param.i_width != 16 * h->mb.i_mb_width ||
3152             h->param.i_height != 16 * h->mb.i_mb_height )
3153             x264_frame_expand_border_mod16( h, fenc );
3154
3155         fenc->i_frame = h->frames.i_input++;
3156
3157         if( fenc->i_frame == 0 )
3158             h->frames.i_first_pts = fenc->i_pts;
3159         if( h->frames.i_bframe_delay && fenc->i_frame == h->frames.i_bframe_delay )
3160             h->frames.i_bframe_delay_time = fenc->i_pts - h->frames.i_first_pts;
3161
3162         if( h->param.b_vfr_input && fenc->i_pts <= h->frames.i_largest_pts )
3163             x264_log( h, X264_LOG_WARNING, "non-strictly-monotonic PTS\n" );
3164
3165         h->frames.i_second_largest_pts = h->frames.i_largest_pts;
3166         h->frames.i_largest_pts = fenc->i_pts;
3167
3168         if( (fenc->i_pic_struct < PIC_STRUCT_AUTO) || (fenc->i_pic_struct > PIC_STRUCT_TRIPLE) )
3169             fenc->i_pic_struct = PIC_STRUCT_AUTO;
3170
3171         if( fenc->i_pic_struct == PIC_STRUCT_AUTO )
3172         {
3173 #if HAVE_INTERLACED
3174             int b_interlaced = fenc->param ? fenc->param->b_interlaced : h->param.b_interlaced;
3175 #else
3176             int b_interlaced = 0;
3177 #endif
3178             if( b_interlaced )
3179             {
3180                 int b_tff = fenc->param ? fenc->param->b_tff : h->param.b_tff;
3181                 fenc->i_pic_struct = b_tff ? PIC_STRUCT_TOP_BOTTOM : PIC_STRUCT_BOTTOM_TOP;
3182             }
3183             else
3184                 fenc->i_pic_struct = PIC_STRUCT_PROGRESSIVE;
3185         }
3186
3187         if( h->param.rc.b_mb_tree && h->param.rc.b_stat_read )
3188         {
3189             if( x264_macroblock_tree_read( h, fenc, pic_in->prop.quant_offsets ) )
3190                 return -1;
3191         }
3192         else
3193             x264_stack_align( x264_adaptive_quant_frame, h, fenc, pic_in->prop.quant_offsets );
3194
3195         if( pic_in->prop.quant_offsets_free )
3196             pic_in->prop.quant_offsets_free( pic_in->prop.quant_offsets );
3197
3198         if( h->frames.b_have_lowres )
3199             x264_frame_init_lowres( h, fenc );
3200
3201         /* 2: Place the frame into the queue for its slice type decision */
3202         x264_lookahead_put_frame( h, fenc );
3203
3204         if( h->frames.i_input <= h->frames.i_delay + 1 - h->i_thread_frames )
3205         {
3206             /* Nothing yet to encode, waiting for filling of buffers */
3207             pic_out->i_type = X264_TYPE_AUTO;
3208             return 0;
3209         }
3210     }
3211     else
3212     {
3213         /* signal kills for lookahead thread */
3214         x264_pthread_mutex_lock( &h->lookahead->ifbuf.mutex );
3215         h->lookahead->b_exit_thread = 1;
3216         x264_pthread_cond_broadcast( &h->lookahead->ifbuf.cv_fill );
3217         x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
3218     }
3219
3220     h->i_frame++;
3221     /* 3: The picture is analyzed in the lookahead */
3222     if( !h->frames.current[0] )
3223         x264_lookahead_get_frames( h );
3224
3225     if( !h->frames.current[0] && x264_lookahead_is_empty( h ) )
3226         return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
3227
3228     /* ------------------- Get frame to be encoded ------------------------- */
3229     /* 4: get picture to encode */
3230     h->fenc = x264_frame_shift( h->frames.current );
3231
3232     /* If applicable, wait for previous frame reconstruction to finish */
3233     if( h->param.b_sliced_threads )
3234         if( x264_threadpool_wait_all( h ) < 0 )
3235             return -1;
3236
3237     if( h->i_frame == h->i_thread_frames - 1 )
3238         h->i_reordered_pts_delay = h->fenc->i_reordered_pts;
3239     if( h->reconfig )
3240     {
3241         x264_encoder_reconfig_apply( h, &h->reconfig_h->param );
3242         h->reconfig = 0;
3243     }
3244     if( h->fenc->param )
3245     {
3246         x264_encoder_reconfig_apply( h, h->fenc->param );
3247         if( h->fenc->param->param_free )
3248         {
3249             h->fenc->param->param_free( h->fenc->param );
3250             h->fenc->param = NULL;
3251         }
3252     }
3253
3254     // ok to call this before encoding any frames, since the initial values of fdec have b_kept_as_ref=0
3255     if( x264_reference_update( h ) )
3256         return -1;
3257     h->fdec->i_lines_completed = -1;
3258
3259     if( !IS_X264_TYPE_I( h->fenc->i_type ) )
3260     {
3261         int valid_refs_left = 0;
3262         for( int i = 0; h->frames.reference[i]; i++ )
3263             if( !h->frames.reference[i]->b_corrupt )
3264                 valid_refs_left++;
3265         /* No valid reference frames left: force an IDR. */
3266         if( !valid_refs_left )
3267         {
3268             h->fenc->b_keyframe = 1;
3269             h->fenc->i_type = X264_TYPE_IDR;
3270         }
3271     }
3272
3273     if( h->fenc->b_keyframe )
3274     {
3275         h->frames.i_last_keyframe = h->fenc->i_frame;
3276         if( h->fenc->i_type == X264_TYPE_IDR )
3277         {
3278             h->i_frame_num = 0;
3279             h->frames.i_last_idr = h->fenc->i_frame;
3280         }
3281     }
3282     h->sh.i_mmco_command_count =
3283     h->sh.i_mmco_remove_from_end = 0;
3284     h->b_ref_reorder[0] =
3285     h->b_ref_reorder[1] = 0;
3286     h->fdec->i_poc =
3287     h->fenc->i_poc = 2 * ( h->fenc->i_frame - X264_MAX( h->frames.i_last_idr, 0 ) );
3288
3289     /* ------------------- Setup frame context ----------------------------- */
3290     /* 5: Init data dependent of frame type */
3291     if( h->fenc->i_type == X264_TYPE_IDR )
3292     {
3293         /* reset ref pictures */
3294         i_nal_type    = NAL_SLICE_IDR;
3295         i_nal_ref_idc = NAL_PRIORITY_HIGHEST;
3296         h->sh.i_type = SLICE_TYPE_I;
3297         x264_reference_reset( h );
3298         h->frames.i_poc_last_open_gop = -1;
3299     }
3300     else if( h->fenc->i_type == X264_TYPE_I )
3301     {
3302         i_nal_type    = NAL_SLICE;
3303         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
3304         h->sh.i_type = SLICE_TYPE_I;
3305         x264_reference_hierarchy_reset( h );
3306         if( h->param.b_open_gop )
3307             h->frames.i_poc_last_open_gop = h->fenc->b_keyframe ? h->fenc->i_poc : -1;
3308     }
3309     else if( h->fenc->i_type == X264_TYPE_P )
3310     {
3311         i_nal_type    = NAL_SLICE;
3312         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
3313         h->sh.i_type = SLICE_TYPE_P;
3314         x264_reference_hierarchy_reset( h );
3315         h->frames.i_poc_last_open_gop = -1;
3316     }
3317     else if( h->fenc->i_type == X264_TYPE_BREF )
3318     {
3319         i_nal_type    = NAL_SLICE;
3320         i_nal_ref_idc = h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT ? NAL_PRIORITY_LOW : NAL_PRIORITY_HIGH;
3321         h->sh.i_type = SLICE_TYPE_B;
3322         x264_reference_hierarchy_reset( h );
3323     }
3324     else    /* B frame */
3325     {
3326         i_nal_type    = NAL_SLICE;
3327         i_nal_ref_idc = NAL_PRIORITY_DISPOSABLE;
3328         h->sh.i_type = SLICE_TYPE_B;
3329     }
3330
3331     h->fdec->i_type = h->fenc->i_type;
3332     h->fdec->i_frame = h->fenc->i_frame;
3333     h->fenc->b_kept_as_ref =
3334     h->fdec->b_kept_as_ref = i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE && h->param.i_keyint_max > 1;
3335
3336     h->fdec->mb_info = h->fenc->mb_info;
3337     h->fdec->mb_info_free = h->fenc->mb_info_free;
3338     h->fenc->mb_info = NULL;
3339     h->fenc->mb_info_free = NULL;
3340
3341     h->fdec->i_pts = h->fenc->i_pts;
3342     if( h->frames.i_bframe_delay )
3343     {
3344         int64_t *prev_reordered_pts = thread_current->frames.i_prev_reordered_pts;
3345         h->fdec->i_dts = h->i_frame > h->frames.i_bframe_delay
3346                        ? prev_reordered_pts[ (h->i_frame - h->frames.i_bframe_delay) % h->frames.i_bframe_delay ]
3347                        : h->fenc->i_reordered_pts - h->frames.i_bframe_delay_time;
3348         prev_reordered_pts[ h->i_frame % h->frames.i_bframe_delay ] = h->fenc->i_reordered_pts;
3349     }
3350     else
3351         h->fdec->i_dts = h->fenc->i_reordered_pts;
3352     if( h->fenc->i_type == X264_TYPE_IDR )
3353         h->i_last_idr_pts = h->fdec->i_pts;
3354
3355     /* ------------------- Init                ----------------------------- */
3356     /* build ref list 0/1 */
3357     x264_reference_build_list( h, h->fdec->i_poc );
3358
3359     /* ---------------------- Write the bitstream -------------------------- */
3360     /* Init bitstream context */
3361     if( h->param.b_sliced_threads )
3362     {
3363         for( int i = 0; i < h->param.i_threads; i++ )
3364         {
3365             bs_init( &h->thread[i]->out.bs, h->thread[i]->out.p_bitstream, h->thread[i]->out.i_bitstream );
3366             h->thread[i]->out.i_nal = 0;
3367         }
3368     }
3369     else
3370     {
3371         bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
3372         h->out.i_nal = 0;
3373     }
3374
3375     if( h->param.b_aud )
3376     {
3377         int pic_type;
3378
3379         if( h->sh.i_type == SLICE_TYPE_I )
3380             pic_type = 0;
3381         else if( h->sh.i_type == SLICE_TYPE_P )
3382             pic_type = 1;
3383         else if( h->sh.i_type == SLICE_TYPE_B )
3384             pic_type = 2;
3385         else
3386             pic_type = 7;
3387
3388         x264_nal_start( h, NAL_AUD, NAL_PRIORITY_DISPOSABLE );
3389         bs_write( &h->out.bs, 3, pic_type );
3390         bs_rbsp_trailing( &h->out.bs );
3391         if( x264_nal_end( h ) )
3392             return -1;
3393         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
3394     }
3395
3396     h->i_nal_type = i_nal_type;
3397     h->i_nal_ref_idc = i_nal_ref_idc;
3398
3399     if( h->param.b_intra_refresh )
3400     {
3401         if( IS_X264_TYPE_I( h->fenc->i_type ) )
3402         {
3403             h->fdec->i_frames_since_pir = 0;
3404             h->b_queued_intra_refresh = 0;
3405             /* PIR is currently only supported with ref == 1, so any intra frame effectively refreshes
3406              * the whole frame and counts as an intra refresh. */
3407             h->fdec->f_pir_position = h->mb.i_mb_width;
3408         }
3409         else if( h->fenc->i_type == X264_TYPE_P )
3410         {
3411             int pocdiff = (h->fdec->i_poc - h->fref[0][0]->i_poc)/2;
3412             float increment = X264_MAX( ((float)h->mb.i_mb_width-1) / h->param.i_keyint_max, 1 );
3413             h->fdec->f_pir_position = h->fref[0][0]->f_pir_position;
3414             h->fdec->i_frames_since_pir = h->fref[0][0]->i_frames_since_pir + pocdiff;
3415             if( h->fdec->i_frames_since_pir >= h->param.i_keyint_max ||
3416                 (h->b_queued_intra_refresh && h->fdec->f_pir_position + 0.5 >= h->mb.i_mb_width) )
3417             {
3418                 h->fdec->f_pir_position = 0;
3419                 h->fdec->i_frames_since_pir = 0;
3420                 h->b_queued_intra_refresh = 0;
3421                 h->fenc->b_keyframe = 1;
3422             }
3423             h->fdec->i_pir_start_col = h->fdec->f_pir_position+0.5;
3424             h->fdec->f_pir_position += increment * pocdiff;
3425             h->fdec->i_pir_end_col = h->fdec->f_pir_position+0.5;
3426             /* If our intra refresh has reached the right side of the frame, we're done. */
3427             if( h->fdec->i_pir_end_col >= h->mb.i_mb_width - 1 )
3428             {
3429                 h->fdec->f_pir_position = h->mb.i_mb_width;
3430                 h->fdec->i_pir_end_col = h->mb.i_mb_width - 1;
3431             }
3432         }
3433     }
3434
3435     if( h->fenc->b_keyframe )
3436     {
3437         /* Write SPS and PPS */
3438         if( h->param.b_repeat_headers )
3439         {
3440             /* generate sequence parameters */
3441             x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
3442             x264_sps_write( &h->out.bs, h->sps );
3443             if( x264_nal_end( h ) )
3444                 return -1;
3445             /* Pad AUD/SPS to 256 bytes like Panasonic */
3446             if( h->param.b_avcintra_compat )
3447                 h->out.nal[h->out.i_nal-1].i_padding = 256 - bs_pos( &h->out.bs ) / 8 - 2*NALU_OVERHEAD;
3448             overhead += h->out.nal[h->out.i_nal-1].i_payload + h->out.nal[h->out.i_nal-1].i_padding + NALU_OVERHEAD;
3449
3450             /* generate picture parameters */
3451             x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
3452             x264_pps_write( &h->out.bs, h->sps, h->pps );
3453             if( x264_nal_end( h ) )
3454                 return -1;
3455             if( h->param.b_avcintra_compat )
3456                 h->out.nal[h->out.i_nal-1].i_padding = 256 - h->out.nal[h->out.i_nal-1].i_payload - NALU_OVERHEAD;
3457             overhead += h->out.nal[h->out.i_nal-1].i_payload + h->out.nal[h->out.i_nal-1].i_padding + NALU_OVERHEAD;
3458         }
3459
3460         /* when frame threading is used, buffering period sei is written in x264_encoder_frame_end */
3461         if( h->i_thread_frames == 1 && h->sps->vui.b_nal_hrd_parameters_present )
3462         {
3463             x264_hrd_fullness( h );
3464             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3465             x264_sei_buffering_period_write( h, &h->out.bs );
3466             if( x264_nal_end( h ) )
3467                return -1;
3468             overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3469         }
3470     }
3471
3472     /* write extra sei */
3473     for( int i = 0; i < h->fenc->extra_sei.num_payloads; i++ )
3474     {
3475         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3476         x264_sei_write( &h->out.bs, h->fenc->extra_sei.payloads[i].payload, h->fenc->extra_sei.payloads[i].payload_size,
3477                         h->fenc->extra_sei.payloads[i].payload_type );
3478         if( x264_nal_end( h ) )
3479             return -1;
3480         overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3481         if( h->fenc->extra_sei.sei_free )
3482         {
3483             h->fenc->extra_sei.sei_free( h->fenc->extra_sei.payloads[i].payload );
3484             h->fenc->extra_sei.payloads[i].payload = NULL;
3485         }
3486     }
3487
3488     if( h->fenc->extra_sei.sei_free )
3489     {
3490         h->fenc->extra_sei.sei_free( h->fenc->extra_sei.payloads );
3491         h->fenc->extra_sei.payloads = NULL;
3492         h->fenc->extra_sei.sei_free = NULL;
3493     }
3494
3495     if( h->fenc->b_keyframe )
3496     {
3497         /* Avid's decoder strictly wants two SEIs for AVC-Intra so we can't insert the x264 SEI */
3498         if( h->param.b_repeat_headers && h->fenc->i_frame == 0 && !h->param.b_avcintra_compat )
3499         {
3500             /* identify ourself */
3501             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3502             if( x264_sei_version_write( h, &h->out.bs ) )
3503                 return -1;
3504             if( x264_nal_end( h ) )
3505                 return -1;
3506             overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3507         }
3508
3509         if( h->fenc->i_type != X264_TYPE_IDR )
3510         {
3511             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;
3512             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3513             x264_sei_recovery_point_write( h, &h->out.bs, time_to_recovery );
3514             if( x264_nal_end( h ) )
3515                 return -1;
3516             overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3517         }
3518
3519         if( h->param.i_frame_packing >= 0 )
3520         {
3521             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3522             x264_sei_frame_packing_write( h, &h->out.bs );
3523             if( x264_nal_end( h ) )
3524                 return -1;
3525             overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3526         }
3527     }
3528
3529     /* generate sei pic timing */
3530     if( h->sps->vui.b_pic_struct_present || h->sps->vui.b_nal_hrd_parameters_present )
3531     {
3532         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3533         x264_sei_pic_timing_write( h, &h->out.bs );
3534         if( x264_nal_end( h ) )
3535             return -1;
3536         overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3537     }
3538
3539     /* As required by Blu-ray. */
3540     if( !IS_X264_TYPE_B( h->fenc->i_type ) && h->b_sh_backup )
3541     {
3542         h->b_sh_backup = 0;
3543         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3544         x264_sei_dec_ref_pic_marking_write( h, &h->out.bs );
3545         if( x264_nal_end( h ) )
3546             return -1;
3547         overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3548     }