Make x264_encoder_reconfig more threadsafe
[x262.git] / encoder / encoder.c
1 /*****************************************************************************
2  * encoder.c: top-level encoder functions
3  *****************************************************************************
4  * Copyright (C) 2003-2013 x264 project
5  *
6  * Authors: Laurent Aimar <fenrir@via.ecp.fr>
7  *          Loren Merritt <lorenm@u.washington.edu>
8  *          Jason Garrett-Glaser <darkshikari@gmail.com>
9  *
10  * This program is free software; you can redistribute it and/or modify
11  * it under the terms of the GNU General Public License as published by
12  * the Free Software Foundation; either version 2 of the License, or
13  * (at your option) any later version.
14  *
15  * This program is distributed in the hope that it will be useful,
16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
18  * GNU General Public License for more details.
19  *
20  * You should have received a copy of the GNU General Public License
21  * along with this program; if not, write to the Free Software
22  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02111, USA.
23  *
24  * This program is also available under a commercial proprietary license.
25  * For more information, contact us at licensing@x264.com.
26  *****************************************************************************/
27
28 #include "common/common.h"
29
30 #include "set.h"
31 #include "analyse.h"
32 #include "ratecontrol.h"
33 #include "macroblock.h"
34 #include "me.h"
35
36 #if HAVE_VISUALIZE
37 #include "common/visualize.h"
38 #endif
39
40 //#define DEBUG_MB_TYPE
41
42 #define bs_write_ue bs_write_ue_big
43
44 static int x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
45                                    x264_nal_t **pp_nal, int *pi_nal,
46                                    x264_picture_t *pic_out );
47
48 /****************************************************************************
49  *
50  ******************************* x264 libs **********************************
51  *
52  ****************************************************************************/
53 static double x264_psnr( double sqe, double size )
54 {
55     double mse = sqe / (PIXEL_MAX*PIXEL_MAX * size);
56     if( mse <= 0.0000000001 ) /* Max 100dB */
57         return 100;
58
59     return -10.0 * log10( mse );
60 }
61
62 static double x264_ssim( double ssim )
63 {
64     double inv_ssim = 1 - ssim;
65     if( inv_ssim <= 0.0000000001 ) /* Max 100dB */
66         return 100;
67
68     return -10.0 * log10( inv_ssim );
69 }
70
71 static int x264_threadpool_wait_all( x264_t *h )
72 {
73     for( int i = 0; i < h->param.i_threads; i++ )
74         if( h->thread[i]->b_thread_active )
75         {
76             h->thread[i]->b_thread_active = 0;
77             if( (intptr_t)x264_threadpool_wait( h->threadpool, h->thread[i] ) < 0 )
78                 return -1;
79         }
80     return 0;
81 }
82
83 static void x264_frame_dump( x264_t *h )
84 {
85     FILE *f = x264_fopen( h->param.psz_dump_yuv, "r+b" );
86     if( !f )
87         return;
88
89     /* Wait for the threads to finish deblocking */
90     if( h->param.b_sliced_threads )
91         x264_threadpool_wait_all( h );
92
93     /* Write the frame in display order */
94     int frame_size = FRAME_SIZE( h->param.i_height * h->param.i_width * sizeof(pixel) );
95     fseek( f, (uint64_t)h->fdec->i_frame * frame_size, SEEK_SET );
96     for( int p = 0; p < (CHROMA444 ? 3 : 1); p++ )
97         for( int y = 0; y < h->param.i_height; y++ )
98             fwrite( &h->fdec->plane[p][y*h->fdec->i_stride[p]], sizeof(pixel), h->param.i_width, f );
99     if( !CHROMA444 )
100     {
101         int cw = h->param.i_width>>1;
102         int ch = h->param.i_height>>CHROMA_V_SHIFT;
103         pixel *planeu = x264_malloc( (cw*ch*2+32)*sizeof(pixel) );
104         pixel *planev = planeu + cw*ch + 16;
105         h->mc.plane_copy_deinterleave( planeu, cw, planev, cw, h->fdec->plane[1], h->fdec->i_stride[1], cw, ch );
106         fwrite( planeu, 1, cw*ch*sizeof(pixel), f );
107         fwrite( planev, 1, cw*ch*sizeof(pixel), f );
108         x264_free( planeu );
109     }
110     fclose( f );
111 }
112
113 /* Fill "default" values */
114 static void x264_slice_header_init( x264_t *h, x264_slice_header_t *sh,
115                                     x264_sps_t *sps, x264_pps_t *pps,
116                                     int i_idr_pic_id, int i_frame, int i_qp )
117 {
118     x264_param_t *param = &h->param;
119
120     /* First we fill all fields */
121     sh->sps = sps;
122     sh->pps = pps;
123
124     sh->i_first_mb  = 0;
125     sh->i_last_mb   = h->mb.i_mb_count - 1;
126     sh->i_pps_id    = pps->i_id;
127
128     sh->i_frame_num = i_frame;
129
130     sh->b_mbaff = PARAM_INTERLACED;
131     sh->b_field_pic = 0;    /* no field support for now */
132     sh->b_bottom_field = 0; /* not yet used */
133
134     sh->i_idr_pic_id = i_idr_pic_id;
135
136     /* poc stuff, fixed later */
137     sh->i_poc = 0;
138     sh->i_delta_poc_bottom = 0;
139     sh->i_delta_poc[0] = 0;
140     sh->i_delta_poc[1] = 0;
141
142     sh->i_redundant_pic_cnt = 0;
143
144     h->mb.b_direct_auto_write = h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO
145                                 && h->param.i_bframe
146                                 && ( h->param.rc.b_stat_write || !h->param.rc.b_stat_read );
147
148     if( !h->mb.b_direct_auto_read && sh->i_type == SLICE_TYPE_B )
149     {
150         if( h->fref[1][0]->i_poc_l0ref0 == h->fref[0][0]->i_poc )
151         {
152             if( h->mb.b_direct_auto_write )
153                 sh->b_direct_spatial_mv_pred = ( h->stat.i_direct_score[1] > h->stat.i_direct_score[0] );
154             else
155                 sh->b_direct_spatial_mv_pred = ( param->analyse.i_direct_mv_pred == X264_DIRECT_PRED_SPATIAL );
156         }
157         else
158         {
159             h->mb.b_direct_auto_write = 0;
160             sh->b_direct_spatial_mv_pred = 1;
161         }
162     }
163     /* else b_direct_spatial_mv_pred was read from the 2pass statsfile */
164
165     sh->b_num_ref_idx_override = 0;
166     sh->i_num_ref_idx_l0_active = 1;
167     sh->i_num_ref_idx_l1_active = 1;
168
169     sh->b_ref_pic_list_reordering[0] = h->b_ref_reorder[0];
170     sh->b_ref_pic_list_reordering[1] = h->b_ref_reorder[1];
171
172     /* If the ref list isn't in the default order, construct reordering header */
173     for( int list = 0; list < 2; list++ )
174     {
175         if( sh->b_ref_pic_list_reordering[list] )
176         {
177             int pred_frame_num = i_frame;
178             for( int i = 0; i < h->i_ref[list]; i++ )
179             {
180                 int diff = h->fref[list][i]->i_frame_num - pred_frame_num;
181                 sh->ref_pic_list_order[list][i].idc = ( diff > 0 );
182                 sh->ref_pic_list_order[list][i].arg = (abs(diff) - 1) & ((1 << sps->i_log2_max_frame_num) - 1);
183                 pred_frame_num = h->fref[list][i]->i_frame_num;
184             }
185         }
186     }
187
188     sh->i_cabac_init_idc = param->i_cabac_init_idc;
189
190     sh->i_qp = SPEC_QP(i_qp);
191     sh->i_qp_delta = sh->i_qp - pps->i_pic_init_qp;
192     sh->b_sp_for_swidth = 0;
193     sh->i_qs_delta = 0;
194
195     int deblock_thresh = i_qp + 2 * X264_MIN(param->i_deblocking_filter_alphac0, param->i_deblocking_filter_beta);
196     /* If effective qp <= 15, deblocking would have no effect anyway */
197     if( param->b_deblocking_filter && (h->mb.b_variable_qp || 15 < deblock_thresh ) )
198         sh->i_disable_deblocking_filter_idc = param->b_sliced_threads ? 2 : 0;
199     else
200         sh->i_disable_deblocking_filter_idc = 1;
201     sh->i_alpha_c0_offset = param->i_deblocking_filter_alphac0 << 1;
202     sh->i_beta_offset = param->i_deblocking_filter_beta << 1;
203 }
204
205 static void x264_slice_header_write( bs_t *s, x264_slice_header_t *sh, int i_nal_ref_idc )
206 {
207     if( sh->b_mbaff )
208     {
209         int first_x = sh->i_first_mb % sh->sps->i_mb_width;
210         int first_y = sh->i_first_mb / sh->sps->i_mb_width;
211         assert( (first_y&1) == 0 );
212         bs_write_ue( s, (2*first_x + sh->sps->i_mb_width*(first_y&~1) + (first_y&1)) >> 1 );
213     }
214     else
215         bs_write_ue( s, sh->i_first_mb );
216
217     bs_write_ue( s, sh->i_type + 5 );   /* same type things */
218     bs_write_ue( s, sh->i_pps_id );
219     bs_write( s, sh->sps->i_log2_max_frame_num, sh->i_frame_num & ((1<<sh->sps->i_log2_max_frame_num)-1) );
220
221     if( !sh->sps->b_frame_mbs_only )
222     {
223         bs_write1( s, sh->b_field_pic );
224         if( sh->b_field_pic )
225             bs_write1( s, sh->b_bottom_field );
226     }
227
228     if( sh->i_idr_pic_id >= 0 ) /* NAL IDR */
229         bs_write_ue( s, sh->i_idr_pic_id );
230
231     if( sh->sps->i_poc_type == 0 )
232     {
233         bs_write( s, sh->sps->i_log2_max_poc_lsb, sh->i_poc & ((1<<sh->sps->i_log2_max_poc_lsb)-1) );
234         if( sh->pps->b_pic_order && !sh->b_field_pic )
235             bs_write_se( s, sh->i_delta_poc_bottom );
236     }
237
238     if( sh->pps->b_redundant_pic_cnt )
239         bs_write_ue( s, sh->i_redundant_pic_cnt );
240
241     if( sh->i_type == SLICE_TYPE_B )
242         bs_write1( s, sh->b_direct_spatial_mv_pred );
243
244     if( sh->i_type == SLICE_TYPE_P || sh->i_type == SLICE_TYPE_B )
245     {
246         bs_write1( s, sh->b_num_ref_idx_override );
247         if( sh->b_num_ref_idx_override )
248         {
249             bs_write_ue( s, sh->i_num_ref_idx_l0_active - 1 );
250             if( sh->i_type == SLICE_TYPE_B )
251                 bs_write_ue( s, sh->i_num_ref_idx_l1_active - 1 );
252         }
253     }
254
255     /* ref pic list reordering */
256     if( sh->i_type != SLICE_TYPE_I )
257     {
258         bs_write1( s, sh->b_ref_pic_list_reordering[0] );
259         if( sh->b_ref_pic_list_reordering[0] )
260         {
261             for( int i = 0; i < sh->i_num_ref_idx_l0_active; i++ )
262             {
263                 bs_write_ue( s, sh->ref_pic_list_order[0][i].idc );
264                 bs_write_ue( s, sh->ref_pic_list_order[0][i].arg );
265             }
266             bs_write_ue( s, 3 );
267         }
268     }
269     if( sh->i_type == SLICE_TYPE_B )
270     {
271         bs_write1( s, sh->b_ref_pic_list_reordering[1] );
272         if( sh->b_ref_pic_list_reordering[1] )
273         {
274             for( int i = 0; i < sh->i_num_ref_idx_l1_active; i++ )
275             {
276                 bs_write_ue( s, sh->ref_pic_list_order[1][i].idc );
277                 bs_write_ue( s, sh->ref_pic_list_order[1][i].arg );
278             }
279             bs_write_ue( s, 3 );
280         }
281     }
282
283     sh->b_weighted_pred = 0;
284     if( sh->pps->b_weighted_pred && sh->i_type == SLICE_TYPE_P )
285     {
286         sh->b_weighted_pred = sh->weight[0][0].weightfn || sh->weight[0][1].weightfn || sh->weight[0][2].weightfn;
287         /* pred_weight_table() */
288         bs_write_ue( s, sh->weight[0][0].i_denom );
289         bs_write_ue( s, sh->weight[0][1].i_denom );
290         for( int i = 0; i < sh->i_num_ref_idx_l0_active; i++ )
291         {
292             int luma_weight_l0_flag = !!sh->weight[i][0].weightfn;
293             int chroma_weight_l0_flag = !!sh->weight[i][1].weightfn || !!sh->weight[i][2].weightfn;
294             bs_write1( s, luma_weight_l0_flag );
295             if( luma_weight_l0_flag )
296             {
297                 bs_write_se( s, sh->weight[i][0].i_scale );
298                 bs_write_se( s, sh->weight[i][0].i_offset );
299             }
300             bs_write1( s, chroma_weight_l0_flag );
301             if( chroma_weight_l0_flag )
302             {
303                 for( int j = 1; j < 3; j++ )
304                 {
305                     bs_write_se( s, sh->weight[i][j].i_scale );
306                     bs_write_se( s, sh->weight[i][j].i_offset );
307                 }
308             }
309         }
310     }
311     else if( sh->pps->b_weighted_bipred == 1 && sh->i_type == SLICE_TYPE_B )
312     {
313       /* TODO */
314     }
315
316     if( i_nal_ref_idc != 0 )
317     {
318         if( sh->i_idr_pic_id >= 0 )
319         {
320             bs_write1( s, 0 );  /* no output of prior pics flag */
321             bs_write1( s, 0 );  /* long term reference flag */
322         }
323         else
324         {
325             bs_write1( s, sh->i_mmco_command_count > 0 ); /* adaptive_ref_pic_marking_mode_flag */
326             if( sh->i_mmco_command_count > 0 )
327             {
328                 for( int i = 0; i < sh->i_mmco_command_count; i++ )
329                 {
330                     bs_write_ue( s, 1 ); /* mark short term ref as unused */
331                     bs_write_ue( s, sh->mmco[i].i_difference_of_pic_nums - 1 );
332                 }
333                 bs_write_ue( s, 0 ); /* end command list */
334             }
335         }
336     }
337
338     if( sh->pps->b_cabac && sh->i_type != SLICE_TYPE_I )
339         bs_write_ue( s, sh->i_cabac_init_idc );
340
341     bs_write_se( s, sh->i_qp_delta );      /* slice qp delta */
342
343     if( sh->pps->b_deblocking_filter_control )
344     {
345         bs_write_ue( s, sh->i_disable_deblocking_filter_idc );
346         if( sh->i_disable_deblocking_filter_idc != 1 )
347         {
348             bs_write_se( s, sh->i_alpha_c0_offset >> 1 );
349             bs_write_se( s, sh->i_beta_offset >> 1 );
350         }
351     }
352 }
353
354 /* If we are within a reasonable distance of the end of the memory allocated for the bitstream, */
355 /* reallocate, adding an arbitrary amount of space. */
356 static int x264_bitstream_check_buffer_internal( x264_t *h, int size, int b_cabac, int i_nal )
357 {
358     if( (b_cabac && (h->cabac.p_end - h->cabac.p < size)) ||
359         (h->out.bs.p_end - h->out.bs.p < size) )
360     {
361         int buf_size = h->out.i_bitstream + size;
362         uint8_t *buf = x264_malloc( buf_size );
363         if( !buf )
364             return -1;
365         int aligned_size = h->out.i_bitstream & ~15;
366         h->mc.memcpy_aligned( buf, h->out.p_bitstream, aligned_size );
367         memcpy( buf + aligned_size, h->out.p_bitstream + aligned_size, h->out.i_bitstream - aligned_size );
368
369         intptr_t delta = buf - h->out.p_bitstream;
370
371         h->out.bs.p_start += delta;
372         h->out.bs.p += delta;
373         h->out.bs.p_end = buf + buf_size;
374
375         h->cabac.p_start += delta;
376         h->cabac.p += delta;
377         h->cabac.p_end = buf + buf_size;
378
379         for( int i = 0; i <= i_nal; i++ )
380             h->out.nal[i].p_payload += delta;
381
382         x264_free( h->out.p_bitstream );
383         h->out.p_bitstream = buf;
384         h->out.i_bitstream = buf_size;
385     }
386     return 0;
387 }
388
389 static int x264_bitstream_check_buffer( x264_t *h )
390 {
391     int max_row_size = (2500 << SLICE_MBAFF) * h->mb.i_mb_width;
392     return x264_bitstream_check_buffer_internal( h, max_row_size, h->param.b_cabac, h->out.i_nal );
393 }
394
395 static int x264_bitstream_check_buffer_filler( x264_t *h, int filler )
396 {
397     filler += 32; // add padding for safety
398     return x264_bitstream_check_buffer_internal( h, filler, 0, -1 );
399 }
400
401 #if HAVE_THREAD
402 static void x264_encoder_thread_init( x264_t *h )
403 {
404     if( h->param.i_sync_lookahead )
405         x264_lower_thread_priority( 10 );
406 }
407 #endif
408
409 /****************************************************************************
410  *
411  ****************************************************************************
412  ****************************** External API*********************************
413  ****************************************************************************
414  *
415  ****************************************************************************/
416
417 static int x264_validate_parameters( x264_t *h, int b_open )
418 {
419 #if HAVE_MMX
420     if( b_open )
421     {
422         int cpuflags = x264_cpu_detect();
423         int fail = 0;
424 #ifdef __SSE__
425         if( !(cpuflags & X264_CPU_SSE) )
426         {
427             x264_log( h, X264_LOG_ERROR, "your cpu does not support SSE1, but x264 was compiled with asm\n");
428             fail = 1;
429         }
430 #else
431         if( !(cpuflags & X264_CPU_MMX2) )
432         {
433             x264_log( h, X264_LOG_ERROR, "your cpu does not support MMXEXT, but x264 was compiled with asm\n");
434             fail = 1;
435         }
436 #endif
437         if( !fail && !(cpuflags & X264_CPU_CMOV) )
438         {
439             x264_log( h, X264_LOG_ERROR, "your cpu does not support CMOV, but x264 was compiled with asm\n");
440             fail = 1;
441         }
442         if( fail )
443         {
444             x264_log( h, X264_LOG_ERROR, "to run x264, recompile without asm (configure --disable-asm)\n");
445             return -1;
446         }
447     }
448 #endif
449
450 #if HAVE_INTERLACED
451     h->param.b_interlaced = !!PARAM_INTERLACED;
452 #else
453     if( h->param.b_interlaced )
454     {
455         x264_log( h, X264_LOG_ERROR, "not compiled with interlaced support\n" );
456         return -1;
457     }
458 #endif
459
460     if( h->param.i_width <= 0 || h->param.i_height <= 0 )
461     {
462         x264_log( h, X264_LOG_ERROR, "invalid width x height (%dx%d)\n",
463                   h->param.i_width, h->param.i_height );
464         return -1;
465     }
466
467     int i_csp = h->param.i_csp & X264_CSP_MASK;
468 #if X264_CHROMA_FORMAT
469     if( CHROMA_FORMAT != CHROMA_420 && i_csp >= X264_CSP_I420 && i_csp <= X264_CSP_NV12 )
470     {
471         x264_log( h, X264_LOG_ERROR, "not compiled with 4:2:0 support\n" );
472         return -1;
473     }
474     else if( CHROMA_FORMAT != CHROMA_422 && i_csp >= X264_CSP_I422 && i_csp <= X264_CSP_NV16 )
475     {
476         x264_log( h, X264_LOG_ERROR, "not compiled with 4:2:2 support\n" );
477         return -1;
478     }
479     else if( CHROMA_FORMAT != CHROMA_444 && i_csp >= X264_CSP_I444 && i_csp <= X264_CSP_RGB )
480     {
481         x264_log( h, X264_LOG_ERROR, "not compiled with 4:4:4 support\n" );
482         return -1;
483     }
484 #endif
485     if( i_csp <= X264_CSP_NONE || i_csp >= X264_CSP_MAX )
486     {
487         x264_log( h, X264_LOG_ERROR, "invalid CSP (only I420/YV12/NV12/I422/YV16/NV16/I444/YV24/BGR/BGRA/RGB supported)\n" );
488         return -1;
489     }
490
491     if( i_csp < X264_CSP_I444 && h->param.i_width % 2 )
492     {
493         x264_log( h, X264_LOG_ERROR, "width not divisible by 2 (%dx%d)\n",
494                   h->param.i_width, h->param.i_height );
495         return -1;
496     }
497
498     if( i_csp < X264_CSP_I422 && PARAM_INTERLACED && h->param.i_height % 4 )
499     {
500         x264_log( h, X264_LOG_ERROR, "height not divisible by 4 (%dx%d)\n",
501                   h->param.i_width, h->param.i_height );
502         return -1;
503     }
504
505     if( (i_csp < X264_CSP_I422 || PARAM_INTERLACED) && h->param.i_height % 2 )
506     {
507         x264_log( h, X264_LOG_ERROR, "height not divisible by 2 (%dx%d)\n",
508                   h->param.i_width, h->param.i_height );
509         return -1;
510     }
511
512     if( (h->param.crop_rect.i_left + h->param.crop_rect.i_right ) >= h->param.i_width ||
513         (h->param.crop_rect.i_top  + h->param.crop_rect.i_bottom) >= h->param.i_height )
514     {
515         x264_log( h, X264_LOG_ERROR, "invalid crop-rect %u,%u,%u,%u\n", h->param.crop_rect.i_left,
516                   h->param.crop_rect.i_top, h->param.crop_rect.i_right,  h->param.crop_rect.i_bottom );
517         return -1;
518     }
519
520     if( h->param.vui.i_sar_width <= 0 || h->param.vui.i_sar_height <= 0 )
521     {
522         h->param.vui.i_sar_width = 0;
523         h->param.vui.i_sar_height = 0;
524     }
525
526     if( h->param.i_threads == X264_THREADS_AUTO )
527         h->param.i_threads = x264_cpu_num_processors() * (h->param.b_sliced_threads?2:3)/2;
528     int max_sliced_threads = X264_MAX( 1, (h->param.i_height+15)/16 / 4 );
529     if( h->param.i_threads > 1 )
530     {
531 #if !HAVE_THREAD
532         x264_log( h, X264_LOG_WARNING, "not compiled with thread support!\n");
533         h->param.i_threads = 1;
534 #endif
535         /* Avoid absurdly small thread slices as they can reduce performance
536          * and VBV compliance.  Capped at an arbitrary 4 rows per thread. */
537         if( h->param.b_sliced_threads )
538             h->param.i_threads = X264_MIN( h->param.i_threads, max_sliced_threads );
539     }
540     h->param.i_threads = x264_clip3( h->param.i_threads, 1, X264_THREAD_MAX );
541     if( h->param.i_threads == 1 )
542     {
543         h->param.b_sliced_threads = 0;
544         h->param.i_lookahead_threads = 1;
545     }
546     h->i_thread_frames = h->param.b_sliced_threads ? 1 : h->param.i_threads;
547     if( h->i_thread_frames > 1 )
548         h->param.nalu_process = NULL;
549
550     if( h->param.b_opencl )
551     {
552 #if !HAVE_OPENCL
553         x264_log( h, X264_LOG_WARNING, "OpenCL: not compiled with OpenCL support, disabling\n" );
554         h->param.b_opencl = 0;
555 #elif BIT_DEPTH > 8
556         x264_log( h, X264_LOG_WARNING, "OpenCL lookahead does not support high bit depth, disabling opencl\n" );
557         h->param.b_opencl = 0;
558 #else
559         if( h->param.i_width < 32 || h->param.i_height < 32 )
560         {
561             x264_log( h, X264_LOG_WARNING, "OpenCL: frame size is too small, disabling opencl\n" );
562             h->param.b_opencl = 0;
563         }
564 #endif
565         if( h->param.opencl_device_id && h->param.i_opencl_device )
566         {
567             x264_log( h, X264_LOG_WARNING, "OpenCL: device id and device skip count configured; dropping skip\n" );
568             h->param.i_opencl_device = 0;
569         }
570     }
571
572     h->param.i_keyint_max = x264_clip3( h->param.i_keyint_max, 1, X264_KEYINT_MAX_INFINITE );
573     if( h->param.i_keyint_max == 1 )
574     {
575         h->param.b_intra_refresh = 0;
576         h->param.analyse.i_weighted_pred = 0;
577         h->param.i_frame_reference = 1;
578         h->param.i_dpb_size = 1;
579     }
580
581     h->param.i_frame_packing = x264_clip3( h->param.i_frame_packing, -1, 5 );
582
583     /* Detect default ffmpeg settings and terminate with an error. */
584     if( b_open )
585     {
586         int score = 0;
587         score += h->param.analyse.i_me_range == 0;
588         score += h->param.rc.i_qp_step == 3;
589         score += h->param.i_keyint_max == 12;
590         score += h->param.rc.i_qp_min == 2;
591         score += h->param.rc.i_qp_max == 31;
592         score += h->param.rc.f_qcompress == 0.5;
593         score += fabs(h->param.rc.f_ip_factor - 1.25) < 0.01;
594         score += fabs(h->param.rc.f_pb_factor - 1.25) < 0.01;
595         score += h->param.analyse.inter == 0 && h->param.analyse.i_subpel_refine == 8;
596         if( score >= 5 )
597         {
598             x264_log( h, X264_LOG_ERROR, "broken ffmpeg default settings detected\n" );
599             x264_log( h, X264_LOG_ERROR, "use an encoding preset (e.g. -vpre medium)\n" );
600             x264_log( h, X264_LOG_ERROR, "preset usage: -vpre <speed> -vpre <profile>\n" );
601             x264_log( h, X264_LOG_ERROR, "speed presets are listed in x264 --help\n" );
602             x264_log( h, X264_LOG_ERROR, "profile is optional; x264 defaults to high\n" );
603             return -1;
604         }
605     }
606
607     if( h->param.rc.i_rc_method < 0 || h->param.rc.i_rc_method > 2 )
608     {
609         x264_log( h, X264_LOG_ERROR, "no ratecontrol method specified\n" );
610         return -1;
611     }
612
613     if( PARAM_INTERLACED )
614         h->param.b_pic_struct = 1;
615
616     if( h->param.b_avcintra_compat )
617     {
618         if( BIT_DEPTH != 10 )
619         {
620             x264_log( h, X264_LOG_ERROR, "%2d-bit AVC-Intra is not widely compatible\n", BIT_DEPTH );
621             x264_log( h, X264_LOG_ERROR, "10-bit x264 is required to encode AVC-Intra\n" );
622             return -1;
623         }
624
625         /* [50/100][res][fps] */
626         static const struct
627         {
628             uint16_t fps_num;
629             uint16_t fps_den;
630             uint8_t interlaced;
631             uint16_t frame_size;
632             const uint8_t *cqm_4ic;
633             const uint8_t *cqm_8iy;
634         } avcintra_lut[2][2][5] =
635         {
636             {{{ 60000, 1001, 0,  912, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy },
637               {    50,    1, 0, 1100, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy },
638               { 30000, 1001, 0,  912, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy },
639               {    25,    1, 0, 1100, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy },
640               { 24000, 1001, 0,  912, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy }},
641              {{ 30000, 1001, 1, 1820, x264_cqm_avci50_4ic, x264_cqm_avci50_1080i_8iy },
642               {    25,    1, 1, 2196, x264_cqm_avci50_4ic, x264_cqm_avci50_1080i_8iy },
643               { 30000, 1001, 0, 1820, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy },
644               {    25,    1, 0, 2196, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy },
645               { 24000, 1001, 0, 1820, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy }}},
646             {{{ 60000, 1001, 0, 1848, x264_cqm_avci100_720p_4ic, x264_cqm_avci100_720p_8iy },
647               {    50,    1, 0, 2224, x264_cqm_avci100_720p_4ic, x264_cqm_avci100_720p_8iy },
648               { 30000, 1001, 0, 1848, x264_cqm_avci100_720p_4ic, x264_cqm_avci100_720p_8iy },
649               {    25,    1, 0, 2224, x264_cqm_avci100_720p_4ic, x264_cqm_avci100_720p_8iy },
650               { 24000, 1001, 0, 1848, x264_cqm_avci100_720p_4ic, x264_cqm_avci100_720p_8iy }},
651              {{ 30000, 1001, 1, 3692, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080i_8iy },
652               {    25,    1, 1, 4444, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080i_8iy },
653               { 30000, 1001, 0, 3692, 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 < 5; 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 == 5 )
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.i_cqm_preset = X264_CQM_CUSTOM;
738         memcpy( h->param.cqm_4iy, x264_cqm_jvt4i, sizeof(h->param.cqm_4iy) );
739         memcpy( h->param.cqm_4ic, avcintra_lut[type][res][i].cqm_4ic, sizeof(h->param.cqm_4ic) );
740         memcpy( h->param.cqm_8iy, avcintra_lut[type][res][i].cqm_8iy, sizeof(h->param.cqm_8iy) );
741
742         /* Need exactly 10 slices of equal MB count... why?  $deity knows... */
743         h->param.i_slice_max_mbs = ((h->param.i_width + 15) / 16) * ((h->param.i_height + 15) / 16) / 10;
744         h->param.i_slice_max_size = 0;
745         /* The slice structure only allows a maximum of 2 threads for 1080i/p
746          * and 1 or 5 threads for 720p */
747         if( h->param.b_sliced_threads )
748         {
749             if( res )
750                 h->param.i_threads = X264_MIN( 2, h->param.i_threads );
751             else
752             {
753                 h->param.i_threads = X264_MIN( 5, h->param.i_threads );
754                 if( h->param.i_threads < 5 )
755                     h->param.i_threads = 1;
756             }
757         }
758
759         if( type )
760             h->param.vui.i_sar_width = h->param.vui.i_sar_height = 1;
761         else
762         {
763             h->param.vui.i_sar_width  = 4;
764             h->param.vui.i_sar_height = 3;
765         }
766
767         /* Avid cannot handle negative QPs */
768         h->param.rc.i_qp_min = X264_MAX( h->param.rc.i_qp_min, QP_BD_OFFSET );
769     }
770
771     h->param.rc.f_rf_constant = x264_clip3f( h->param.rc.f_rf_constant, -QP_BD_OFFSET, 51 );
772     h->param.rc.f_rf_constant_max = x264_clip3f( h->param.rc.f_rf_constant_max, -QP_BD_OFFSET, 51 );
773     h->param.rc.i_qp_constant = x264_clip3( h->param.rc.i_qp_constant, 0, QP_MAX );
774     h->param.analyse.i_subpel_refine = x264_clip3( h->param.analyse.i_subpel_refine, 0, 11 );
775     h->param.rc.f_ip_factor = X264_MAX( h->param.rc.f_ip_factor, 0.01f );
776     h->param.rc.f_pb_factor = X264_MAX( h->param.rc.f_pb_factor, 0.01f );
777     if( h->param.rc.i_rc_method == X264_RC_CRF )
778     {
779         h->param.rc.i_qp_constant = h->param.rc.f_rf_constant + QP_BD_OFFSET;
780         h->param.rc.i_bitrate = 0;
781     }
782     if( b_open && (h->param.rc.i_rc_method == X264_RC_CQP || h->param.rc.i_rc_method == X264_RC_CRF)
783         && h->param.rc.i_qp_constant == 0 )
784     {
785         h->mb.b_lossless = 1;
786         h->param.i_cqm_preset = X264_CQM_FLAT;
787         h->param.psz_cqm_file = NULL;
788         h->param.rc.i_rc_method = X264_RC_CQP;
789         h->param.rc.f_ip_factor = 1;
790         h->param.rc.f_pb_factor = 1;
791         h->param.analyse.b_psnr = 0;
792         h->param.analyse.b_ssim = 0;
793         h->param.analyse.i_chroma_qp_offset = 0;
794         h->param.analyse.i_trellis = 0;
795         h->param.analyse.b_fast_pskip = 0;
796         h->param.analyse.i_noise_reduction = 0;
797         h->param.analyse.b_psy = 0;
798         h->param.i_bframe = 0;
799         /* 8x8dct is not useful without RD in CAVLC lossless */
800         if( !h->param.b_cabac && h->param.analyse.i_subpel_refine < 6 )
801             h->param.analyse.b_transform_8x8 = 0;
802     }
803     if( h->param.rc.i_rc_method == X264_RC_CQP )
804     {
805         float qp_p = h->param.rc.i_qp_constant;
806         float qp_i = qp_p - 6*log2f( h->param.rc.f_ip_factor );
807         float qp_b = qp_p + 6*log2f( h->param.rc.f_pb_factor );
808         h->param.rc.i_qp_min = x264_clip3( (int)(X264_MIN3( qp_p, qp_i, qp_b )), 0, QP_MAX );
809         h->param.rc.i_qp_max = x264_clip3( (int)(X264_MAX3( qp_p, qp_i, qp_b ) + .999), 0, QP_MAX );
810         h->param.rc.i_aq_mode = 0;
811         h->param.rc.b_mb_tree = 0;
812         h->param.rc.i_bitrate = 0;
813     }
814     h->param.rc.i_qp_max = x264_clip3( h->param.rc.i_qp_max, 0, QP_MAX );
815     h->param.rc.i_qp_min = x264_clip3( h->param.rc.i_qp_min, 0, h->param.rc.i_qp_max );
816     h->param.rc.i_qp_step = x264_clip3( h->param.rc.i_qp_step, 2, QP_MAX );
817     h->param.rc.i_bitrate = x264_clip3( h->param.rc.i_bitrate, 0, 2000000 );
818     if( h->param.rc.i_rc_method == X264_RC_ABR && !h->param.rc.i_bitrate )
819     {
820         x264_log( h, X264_LOG_ERROR, "bitrate not specified\n" );
821         return -1;
822     }
823     h->param.rc.i_vbv_buffer_size = x264_clip3( h->param.rc.i_vbv_buffer_size, 0, 2000000 );
824     h->param.rc.i_vbv_max_bitrate = x264_clip3( h->param.rc.i_vbv_max_bitrate, 0, 2000000 );
825     h->param.rc.f_vbv_buffer_init = x264_clip3f( h->param.rc.f_vbv_buffer_init, 0, 2000000 );
826     if( h->param.rc.i_vbv_buffer_size )
827     {
828         if( h->param.rc.i_rc_method == X264_RC_CQP )
829         {
830             x264_log( h, X264_LOG_WARNING, "VBV is incompatible with constant QP, ignored.\n" );
831             h->param.rc.i_vbv_max_bitrate = 0;
832             h->param.rc.i_vbv_buffer_size = 0;
833         }
834         else if( h->param.rc.i_vbv_max_bitrate == 0 )
835         {
836             if( h->param.rc.i_rc_method == X264_RC_ABR )
837             {
838                 x264_log( h, X264_LOG_WARNING, "VBV maxrate unspecified, assuming CBR\n" );
839                 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate;
840             }
841             else
842             {
843                 x264_log( h, X264_LOG_WARNING, "VBV bufsize set but maxrate unspecified, ignored\n" );
844                 h->param.rc.i_vbv_buffer_size = 0;
845             }
846         }
847         else if( h->param.rc.i_vbv_max_bitrate < h->param.rc.i_bitrate &&
848                  h->param.rc.i_rc_method == X264_RC_ABR )
849         {
850             x264_log( h, X264_LOG_WARNING, "max bitrate less than average bitrate, assuming CBR\n" );
851             h->param.rc.i_bitrate = h->param.rc.i_vbv_max_bitrate;
852         }
853     }
854     else if( h->param.rc.i_vbv_max_bitrate )
855     {
856         x264_log( h, X264_LOG_WARNING, "VBV maxrate specified, but no bufsize, ignored\n" );
857         h->param.rc.i_vbv_max_bitrate = 0;
858     }
859
860     h->param.i_slice_max_size = X264_MAX( h->param.i_slice_max_size, 0 );
861     h->param.i_slice_max_mbs = X264_MAX( h->param.i_slice_max_mbs, 0 );
862     h->param.i_slice_min_mbs = X264_MAX( h->param.i_slice_min_mbs, 0 );
863     if( h->param.i_slice_max_mbs )
864         h->param.i_slice_min_mbs = X264_MIN( h->param.i_slice_min_mbs, h->param.i_slice_max_mbs/2 );
865     else if( !h->param.i_slice_max_size )
866         h->param.i_slice_min_mbs = 0;
867     if( PARAM_INTERLACED && h->param.i_slice_min_mbs )
868     {
869         x264_log( h, X264_LOG_WARNING, "interlace + slice-min-mbs is not implemented\n" );
870         h->param.i_slice_min_mbs = 0;
871     }
872     int mb_width = (h->param.i_width+15)/16;
873     if( h->param.i_slice_min_mbs > mb_width )
874     {
875         x264_log( h, X264_LOG_WARNING, "slice-min-mbs > row mb size (%d) not implemented\n", mb_width );
876         h->param.i_slice_min_mbs = mb_width;
877     }
878
879     int max_slices = (h->param.i_height+((16<<PARAM_INTERLACED)-1))/(16<<PARAM_INTERLACED);
880     if( h->param.b_sliced_threads )
881         h->param.i_slice_count = x264_clip3( h->param.i_threads, 0, max_slices );
882     else
883     {
884         h->param.i_slice_count = x264_clip3( h->param.i_slice_count, 0, max_slices );
885         if( h->param.i_slice_max_mbs || h->param.i_slice_max_size )
886             h->param.i_slice_count = 0;
887     }
888     if( h->param.i_slice_count_max > 0 )
889         h->param.i_slice_count_max = X264_MAX( h->param.i_slice_count, h->param.i_slice_count_max );
890
891     if( h->param.b_bluray_compat )
892     {
893         h->param.i_bframe_pyramid = X264_MIN( X264_B_PYRAMID_STRICT, h->param.i_bframe_pyramid );
894         h->param.i_bframe = X264_MIN( h->param.i_bframe, 3 );
895         h->param.b_aud = 1;
896         h->param.i_nal_hrd = X264_MAX( h->param.i_nal_hrd, X264_NAL_HRD_VBR );
897         h->param.i_slice_max_size = 0;
898         h->param.i_slice_max_mbs = 0;
899         h->param.b_intra_refresh = 0;
900         h->param.i_frame_reference = X264_MIN( h->param.i_frame_reference, 6 );
901         h->param.i_dpb_size = X264_MIN( h->param.i_dpb_size, 6 );
902         /* Don't use I-frames, because Blu-ray treats them the same as IDR. */
903         h->param.i_keyint_min = 1;
904         /* Due to the proliferation of broken players that don't handle dupes properly. */
905         h->param.analyse.i_weighted_pred = X264_MIN( h->param.analyse.i_weighted_pred, X264_WEIGHTP_SIMPLE );
906         if( h->param.b_fake_interlaced )
907             h->param.b_pic_struct = 1;
908     }
909
910     h->param.i_frame_reference = x264_clip3( h->param.i_frame_reference, 1, X264_REF_MAX );
911     h->param.i_dpb_size = x264_clip3( h->param.i_dpb_size, 1, X264_REF_MAX );
912     if( h->param.i_scenecut_threshold < 0 )
913         h->param.i_scenecut_threshold = 0;
914     h->param.analyse.i_direct_mv_pred = x264_clip3( h->param.analyse.i_direct_mv_pred, X264_DIRECT_PRED_NONE, X264_DIRECT_PRED_AUTO );
915     if( !h->param.analyse.i_subpel_refine && h->param.analyse.i_direct_mv_pred > X264_DIRECT_PRED_SPATIAL )
916     {
917         x264_log( h, X264_LOG_WARNING, "subme=0 + direct=temporal is not supported\n" );
918         h->param.analyse.i_direct_mv_pred = X264_DIRECT_PRED_SPATIAL;
919     }
920     h->param.i_bframe = x264_clip3( h->param.i_bframe, 0, X264_MIN( X264_BFRAME_MAX, h->param.i_keyint_max-1 ) );
921     h->param.i_bframe_bias = x264_clip3( h->param.i_bframe_bias, -90, 100 );
922     if( h->param.i_bframe <= 1 )
923         h->param.i_bframe_pyramid = X264_B_PYRAMID_NONE;
924     h->param.i_bframe_pyramid = x264_clip3( h->param.i_bframe_pyramid, X264_B_PYRAMID_NONE, X264_B_PYRAMID_NORMAL );
925     h->param.i_bframe_adaptive = x264_clip3( h->param.i_bframe_adaptive, X264_B_ADAPT_NONE, X264_B_ADAPT_TRELLIS );
926     if( !h->param.i_bframe )
927     {
928         h->param.i_bframe_adaptive = X264_B_ADAPT_NONE;
929         h->param.analyse.i_direct_mv_pred = 0;
930         h->param.analyse.b_weighted_bipred = 0;
931         h->param.b_open_gop = 0;
932     }
933     if( h->param.b_intra_refresh && h->param.i_bframe_pyramid == X264_B_PYRAMID_NORMAL )
934     {
935         x264_log( h, X264_LOG_WARNING, "b-pyramid normal + intra-refresh is not supported\n" );
936         h->param.i_bframe_pyramid = X264_B_PYRAMID_STRICT;
937     }
938     if( h->param.b_intra_refresh && (h->param.i_frame_reference > 1 || h->param.i_dpb_size > 1) )
939     {
940         x264_log( h, X264_LOG_WARNING, "ref > 1 + intra-refresh is not supported\n" );
941         h->param.i_frame_reference = 1;
942         h->param.i_dpb_size = 1;
943     }
944     if( h->param.b_intra_refresh && h->param.b_open_gop )
945     {
946         x264_log( h, X264_LOG_WARNING, "intra-refresh is not compatible with open-gop\n" );
947         h->param.b_open_gop = 0;
948     }
949     if( !h->param.i_fps_num || !h->param.i_fps_den )
950     {
951         h->param.i_fps_num = 25;
952         h->param.i_fps_den = 1;
953     }
954     float fps = (float) h->param.i_fps_num / h->param.i_fps_den;
955     if( h->param.i_keyint_min == X264_KEYINT_MIN_AUTO )
956         h->param.i_keyint_min = X264_MIN( h->param.i_keyint_max / 10, fps );
957     h->param.i_keyint_min = x264_clip3( h->param.i_keyint_min, 1, h->param.i_keyint_max/2+1 );
958     h->param.rc.i_lookahead = x264_clip3( h->param.rc.i_lookahead, 0, X264_LOOKAHEAD_MAX );
959     {
960         int maxrate = X264_MAX( h->param.rc.i_vbv_max_bitrate, h->param.rc.i_bitrate );
961         float bufsize = maxrate ? (float)h->param.rc.i_vbv_buffer_size / maxrate : 0;
962         h->param.rc.i_lookahead = X264_MIN( h->param.rc.i_lookahead, X264_MAX( h->param.i_keyint_max, bufsize*fps ) );
963     }
964
965     if( !h->param.i_timebase_num || !h->param.i_timebase_den || !(h->param.b_vfr_input || h->param.b_pulldown) )
966     {
967         h->param.i_timebase_num = h->param.i_fps_den;
968         h->param.i_timebase_den = h->param.i_fps_num;
969     }
970
971     h->param.rc.f_qcompress = x264_clip3f( h->param.rc.f_qcompress, 0.0, 1.0 );
972     if( h->param.i_keyint_max == 1 || h->param.rc.f_qcompress == 1 )
973         h->param.rc.b_mb_tree = 0;
974     if( (!h->param.b_intra_refresh && h->param.i_keyint_max != X264_KEYINT_MAX_INFINITE) &&
975         !h->param.rc.i_lookahead && h->param.rc.b_mb_tree )
976     {
977         x264_log( h, X264_LOG_WARNING, "lookaheadless mb-tree requires intra refresh or infinite keyint\n" );
978         h->param.rc.b_mb_tree = 0;
979     }
980     if( b_open && h->param.rc.b_stat_read )
981         h->param.rc.i_lookahead = 0;
982 #if HAVE_THREAD
983     if( h->param.i_sync_lookahead < 0 )
984         h->param.i_sync_lookahead = h->param.i_bframe + 1;
985     h->param.i_sync_lookahead = X264_MIN( h->param.i_sync_lookahead, X264_LOOKAHEAD_MAX );
986     if( h->param.rc.b_stat_read || h->i_thread_frames == 1 )
987         h->param.i_sync_lookahead = 0;
988 #else
989     h->param.i_sync_lookahead = 0;
990 #endif
991
992     h->param.i_deblocking_filter_alphac0 = x264_clip3( h->param.i_deblocking_filter_alphac0, -6, 6 );
993     h->param.i_deblocking_filter_beta    = x264_clip3( h->param.i_deblocking_filter_beta, -6, 6 );
994     h->param.analyse.i_luma_deadzone[0] = x264_clip3( h->param.analyse.i_luma_deadzone[0], 0, 32 );
995     h->param.analyse.i_luma_deadzone[1] = x264_clip3( h->param.analyse.i_luma_deadzone[1], 0, 32 );
996
997     h->param.i_cabac_init_idc = x264_clip3( h->param.i_cabac_init_idc, 0, 2 );
998
999     if( h->param.i_cqm_preset < X264_CQM_FLAT || h->param.i_cqm_preset > X264_CQM_CUSTOM )
1000         h->param.i_cqm_preset = X264_CQM_FLAT;
1001
1002     if( h->param.analyse.i_me_method < X264_ME_DIA ||
1003         h->param.analyse.i_me_method > X264_ME_TESA )
1004         h->param.analyse.i_me_method = X264_ME_HEX;
1005     h->param.analyse.i_me_range = x264_clip3( h->param.analyse.i_me_range, 4, 1024 );
1006     if( h->param.analyse.i_me_range > 16 && h->param.analyse.i_me_method <= X264_ME_HEX )
1007         h->param.analyse.i_me_range = 16;
1008     if( h->param.analyse.i_me_method == X264_ME_TESA &&
1009         (h->mb.b_lossless || h->param.analyse.i_subpel_refine <= 1) )
1010         h->param.analyse.i_me_method = X264_ME_ESA;
1011     h->param.analyse.b_mixed_references = h->param.analyse.b_mixed_references && h->param.i_frame_reference > 1;
1012     h->param.analyse.inter &= X264_ANALYSE_PSUB16x16|X264_ANALYSE_PSUB8x8|X264_ANALYSE_BSUB16x16|
1013                               X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
1014     h->param.analyse.intra &= X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
1015     if( !(h->param.analyse.inter & X264_ANALYSE_PSUB16x16) )
1016         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
1017     if( !h->param.analyse.b_transform_8x8 )
1018     {
1019         h->param.analyse.inter &= ~X264_ANALYSE_I8x8;
1020         h->param.analyse.intra &= ~X264_ANALYSE_I8x8;
1021     }
1022     h->param.analyse.i_trellis = x264_clip3( h->param.analyse.i_trellis, 0, 2 );
1023     h->param.rc.i_aq_mode = x264_clip3( h->param.rc.i_aq_mode, 0, 2 );
1024     h->param.rc.f_aq_strength = x264_clip3f( h->param.rc.f_aq_strength, 0, 3 );
1025     if( h->param.rc.f_aq_strength == 0 )
1026         h->param.rc.i_aq_mode = 0;
1027
1028     if( h->param.i_log_level < X264_LOG_INFO )
1029     {
1030         h->param.analyse.b_psnr = 0;
1031         h->param.analyse.b_ssim = 0;
1032     }
1033     /* Warn users trying to measure PSNR/SSIM with psy opts on. */
1034     if( b_open && (h->param.analyse.b_psnr || h->param.analyse.b_ssim) )
1035     {
1036         char *s = NULL;
1037
1038         if( h->param.analyse.b_psy )
1039         {
1040             s = h->param.analyse.b_psnr ? "psnr" : "ssim";
1041             x264_log( h, X264_LOG_WARNING, "--%s used with psy on: results will be invalid!\n", s );
1042         }
1043         else if( !h->param.rc.i_aq_mode && h->param.analyse.b_ssim )
1044         {
1045             x264_log( h, X264_LOG_WARNING, "--ssim used with AQ off: results will be invalid!\n" );
1046             s = "ssim";
1047         }
1048         else if(  h->param.rc.i_aq_mode && h->param.analyse.b_psnr )
1049         {
1050             x264_log( h, X264_LOG_WARNING, "--psnr used with AQ on: results will be invalid!\n" );
1051             s = "psnr";
1052         }
1053         if( s )
1054             x264_log( h, X264_LOG_WARNING, "--tune %s should be used if attempting to benchmark %s!\n", s, s );
1055     }
1056
1057     if( !h->param.analyse.b_psy )
1058     {
1059         h->param.analyse.f_psy_rd = 0;
1060         h->param.analyse.f_psy_trellis = 0;
1061     }
1062     h->param.analyse.f_psy_rd = x264_clip3f( h->param.analyse.f_psy_rd, 0, 10 );
1063     h->param.analyse.f_psy_trellis = x264_clip3f( h->param.analyse.f_psy_trellis, 0, 10 );
1064     h->mb.i_psy_rd = h->param.analyse.i_subpel_refine >= 6 ? FIX8( h->param.analyse.f_psy_rd ) : 0;
1065     h->mb.i_psy_trellis = h->param.analyse.i_trellis ? FIX8( h->param.analyse.f_psy_trellis / 4 ) : 0;
1066     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -32, 32);
1067     /* In 4:4:4 mode, chroma gets twice as much resolution, so we can halve its quality. */
1068     if( b_open && i_csp >= X264_CSP_I444 && i_csp < X264_CSP_BGR && h->param.analyse.b_psy )
1069         h->param.analyse.i_chroma_qp_offset += 6;
1070     /* Psy RDO increases overall quantizers to improve the quality of luma--this indirectly hurts chroma quality */
1071     /* so we lower the chroma QP offset to compensate */
1072     if( b_open && h->mb.i_psy_rd && !h->param.b_avcintra_compat )
1073         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_rd < 0.25 ? 1 : 2;
1074     /* Psy trellis has a similar effect. */
1075     if( b_open && h->mb.i_psy_trellis && !h->param.b_avcintra_compat )
1076         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_trellis < 0.25 ? 1 : 2;
1077     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -12, 12);
1078     /* MB-tree requires AQ to be on, even if the strength is zero. */
1079     if( !h->param.rc.i_aq_mode && h->param.rc.b_mb_tree )
1080     {
1081         h->param.rc.i_aq_mode = 1;
1082         h->param.rc.f_aq_strength = 0;
1083     }
1084     h->param.analyse.i_noise_reduction = x264_clip3( h->param.analyse.i_noise_reduction, 0, 1<<16 );
1085     if( h->param.analyse.i_subpel_refine >= 10 && (h->param.analyse.i_trellis != 2 || !h->param.rc.i_aq_mode) )
1086         h->param.analyse.i_subpel_refine = 9;
1087
1088     {
1089         const x264_level_t *l = x264_levels;
1090         if( h->param.i_level_idc < 0 )
1091         {
1092             int maxrate_bak = h->param.rc.i_vbv_max_bitrate;
1093             if( h->param.rc.i_rc_method == X264_RC_ABR && h->param.rc.i_vbv_buffer_size <= 0 )
1094                 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate * 2;
1095             x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
1096             do h->param.i_level_idc = l->level_idc;
1097                 while( l[1].level_idc && x264_validate_levels( h, 0 ) && l++ );
1098             h->param.rc.i_vbv_max_bitrate = maxrate_bak;
1099         }
1100         else
1101         {
1102             while( l->level_idc && l->level_idc != h->param.i_level_idc )
1103                 l++;
1104             if( l->level_idc == 0 )
1105             {
1106                 x264_log( h, X264_LOG_ERROR, "invalid level_idc: %d\n", h->param.i_level_idc );
1107                 return -1;
1108             }
1109         }
1110         if( h->param.analyse.i_mv_range <= 0 )
1111             h->param.analyse.i_mv_range = l->mv_range >> PARAM_INTERLACED;
1112         else
1113             h->param.analyse.i_mv_range = x264_clip3(h->param.analyse.i_mv_range, 32, 512 >> PARAM_INTERLACED);
1114     }
1115
1116     h->param.analyse.i_weighted_pred = x264_clip3( h->param.analyse.i_weighted_pred, X264_WEIGHTP_NONE, X264_WEIGHTP_SMART );
1117
1118     if( h->param.i_lookahead_threads == X264_THREADS_AUTO )
1119     {
1120         if( h->param.b_sliced_threads )
1121             h->param.i_lookahead_threads = h->param.i_threads;
1122         else
1123         {
1124             /* If we're using much slower lookahead settings than encoding settings, it helps a lot to use
1125              * more lookahead threads.  This typically happens in the first pass of a two-pass encode, so
1126              * try to guess at this sort of case.
1127              *
1128              * Tuned by a little bit of real encoding with the various presets. */
1129             int badapt = h->param.i_bframe_adaptive == X264_B_ADAPT_TRELLIS;
1130             int subme = X264_MIN( h->param.analyse.i_subpel_refine / 3, 3 ) + (h->param.analyse.i_subpel_refine > 1);
1131             int bframes = X264_MIN( (h->param.i_bframe - 1) / 3, 3 );
1132
1133             /* [b-adapt 0/1 vs 2][quantized subme][quantized bframes] */
1134             static const uint8_t lookahead_thread_div[2][5][4] =
1135             {{{6,6,6,6}, {3,3,3,3}, {4,4,4,4}, {6,6,6,6}, {12,12,12,12}},
1136              {{3,2,1,1}, {2,1,1,1}, {4,3,2,1}, {6,4,3,2}, {12, 9, 6, 4}}};
1137
1138             h->param.i_lookahead_threads = h->param.i_threads / lookahead_thread_div[badapt][subme][bframes];
1139             /* Since too many lookahead threads significantly degrades lookahead accuracy, limit auto
1140              * lookahead threads to about 8 macroblock rows high each at worst.  This number is chosen
1141              * pretty much arbitrarily. */
1142             h->param.i_lookahead_threads = X264_MIN( h->param.i_lookahead_threads, h->param.i_height / 128 );
1143         }
1144     }
1145     h->param.i_lookahead_threads = x264_clip3( h->param.i_lookahead_threads, 1, X264_MIN( max_sliced_threads, X264_LOOKAHEAD_THREAD_MAX ) );
1146
1147     if( PARAM_INTERLACED )
1148     {
1149         if( h->param.analyse.i_me_method >= X264_ME_ESA )
1150         {
1151             x264_log( h, X264_LOG_WARNING, "interlace + me=esa is not implemented\n" );
1152             h->param.analyse.i_me_method = X264_ME_UMH;
1153         }
1154         if( h->param.analyse.i_weighted_pred > 0 )
1155         {
1156             x264_log( h, X264_LOG_WARNING, "interlace + weightp is not implemented\n" );
1157             h->param.analyse.i_weighted_pred = X264_WEIGHTP_NONE;
1158         }
1159     }
1160
1161     if( !h->param.analyse.i_weighted_pred && h->param.rc.b_mb_tree && h->param.analyse.b_psy )
1162         h->param.analyse.i_weighted_pred = X264_WEIGHTP_FAKE;
1163
1164     if( h->i_thread_frames > 1 )
1165     {
1166         int r = h->param.analyse.i_mv_range_thread;
1167         int r2;
1168         if( r <= 0 )
1169         {
1170             // half of the available space is reserved and divided evenly among the threads,
1171             // the rest is allocated to whichever thread is far enough ahead to use it.
1172             // reserving more space increases quality for some videos, but costs more time
1173             // in thread synchronization.
1174             int max_range = (h->param.i_height + X264_THREAD_HEIGHT) / h->i_thread_frames - X264_THREAD_HEIGHT;
1175             r = max_range / 2;
1176         }
1177         r = X264_MAX( r, h->param.analyse.i_me_range );
1178         r = X264_MIN( r, h->param.analyse.i_mv_range );
1179         // round up to use the whole mb row
1180         r2 = (r & ~15) + ((-X264_THREAD_HEIGHT) & 15);
1181         if( r2 < r )
1182             r2 += 16;
1183         x264_log( h, X264_LOG_DEBUG, "using mv_range_thread = %d\n", r2 );
1184         h->param.analyse.i_mv_range_thread = r2;
1185     }
1186
1187     if( h->param.rc.f_rate_tolerance < 0 )
1188         h->param.rc.f_rate_tolerance = 0;
1189     if( h->param.rc.f_qblur < 0 )
1190         h->param.rc.f_qblur = 0;
1191     if( h->param.rc.f_complexity_blur < 0 )
1192         h->param.rc.f_complexity_blur = 0;
1193
1194     h->param.i_sps_id &= 31;
1195
1196     h->param.i_nal_hrd = x264_clip3( h->param.i_nal_hrd, X264_NAL_HRD_NONE, X264_NAL_HRD_CBR );
1197
1198     if( h->param.i_nal_hrd && !h->param.rc.i_vbv_buffer_size )
1199     {
1200         x264_log( h, X264_LOG_WARNING, "NAL HRD parameters require VBV parameters\n" );
1201         h->param.i_nal_hrd = X264_NAL_HRD_NONE;
1202     }
1203
1204     if( h->param.i_nal_hrd == X264_NAL_HRD_CBR &&
1205        (h->param.rc.i_bitrate != h->param.rc.i_vbv_max_bitrate || !h->param.rc.i_vbv_max_bitrate) )
1206     {
1207         x264_log( h, X264_LOG_WARNING, "CBR HRD requires constant bitrate\n" );
1208         h->param.i_nal_hrd = X264_NAL_HRD_VBR;
1209     }
1210
1211     /* ensure the booleans are 0 or 1 so they can be used in math */
1212 #define BOOLIFY(x) h->param.x = !!h->param.x
1213     BOOLIFY( b_cabac );
1214     BOOLIFY( b_constrained_intra );
1215     BOOLIFY( b_deblocking_filter );
1216     BOOLIFY( b_deterministic );
1217     BOOLIFY( b_sliced_threads );
1218     BOOLIFY( b_interlaced );
1219     BOOLIFY( b_intra_refresh );
1220     BOOLIFY( b_visualize );
1221     BOOLIFY( b_aud );
1222     BOOLIFY( b_repeat_headers );
1223     BOOLIFY( b_annexb );
1224     BOOLIFY( b_vfr_input );
1225     BOOLIFY( b_pulldown );
1226     BOOLIFY( b_tff );
1227     BOOLIFY( b_pic_struct );
1228     BOOLIFY( b_fake_interlaced );
1229     BOOLIFY( b_open_gop );
1230     BOOLIFY( b_bluray_compat );
1231     BOOLIFY( b_stitchable );
1232     BOOLIFY( b_full_recon );
1233     BOOLIFY( b_opencl );
1234     BOOLIFY( b_avcintra_compat );
1235     BOOLIFY( analyse.b_transform_8x8 );
1236     BOOLIFY( analyse.b_weighted_bipred );
1237     BOOLIFY( analyse.b_chroma_me );
1238     BOOLIFY( analyse.b_mixed_references );
1239     BOOLIFY( analyse.b_fast_pskip );
1240     BOOLIFY( analyse.b_dct_decimate );
1241     BOOLIFY( analyse.b_psy );
1242     BOOLIFY( analyse.b_psnr );
1243     BOOLIFY( analyse.b_ssim );
1244     BOOLIFY( rc.b_stat_write );
1245     BOOLIFY( rc.b_stat_read );
1246     BOOLIFY( rc.b_mb_tree );
1247 #undef BOOLIFY
1248
1249     return 0;
1250 }
1251
1252 static void mbcmp_init( x264_t *h )
1253 {
1254     int satd = !h->mb.b_lossless && h->param.analyse.i_subpel_refine > 1;
1255     memcpy( h->pixf.mbcmp, satd ? h->pixf.satd : h->pixf.sad_aligned, sizeof(h->pixf.mbcmp) );
1256     memcpy( h->pixf.mbcmp_unaligned, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.mbcmp_unaligned) );
1257     h->pixf.intra_mbcmp_x3_16x16 = satd ? h->pixf.intra_satd_x3_16x16 : h->pixf.intra_sad_x3_16x16;
1258     h->pixf.intra_mbcmp_x3_8x16c = satd ? h->pixf.intra_satd_x3_8x16c : h->pixf.intra_sad_x3_8x16c;
1259     h->pixf.intra_mbcmp_x3_8x8c  = satd ? h->pixf.intra_satd_x3_8x8c  : h->pixf.intra_sad_x3_8x8c;
1260     h->pixf.intra_mbcmp_x3_8x8 = satd ? h->pixf.intra_sa8d_x3_8x8 : h->pixf.intra_sad_x3_8x8;
1261     h->pixf.intra_mbcmp_x3_4x4 = satd ? h->pixf.intra_satd_x3_4x4 : h->pixf.intra_sad_x3_4x4;
1262     h->pixf.intra_mbcmp_x9_4x4 = h->param.b_cpu_independent || h->mb.b_lossless ? NULL
1263                                : satd ? h->pixf.intra_satd_x9_4x4 : h->pixf.intra_sad_x9_4x4;
1264     h->pixf.intra_mbcmp_x9_8x8 = h->param.b_cpu_independent || h->mb.b_lossless ? NULL
1265                                : satd ? h->pixf.intra_sa8d_x9_8x8 : h->pixf.intra_sad_x9_8x8;
1266     satd &= h->param.analyse.i_me_method == X264_ME_TESA;
1267     memcpy( h->pixf.fpelcmp, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.fpelcmp) );
1268     memcpy( h->pixf.fpelcmp_x3, satd ? h->pixf.satd_x3 : h->pixf.sad_x3, sizeof(h->pixf.fpelcmp_x3) );
1269     memcpy( h->pixf.fpelcmp_x4, satd ? h->pixf.satd_x4 : h->pixf.sad_x4, sizeof(h->pixf.fpelcmp_x4) );
1270 }
1271
1272 static void chroma_dsp_init( x264_t *h )
1273 {
1274     memcpy( h->luma2chroma_pixel, x264_luma2chroma_pixel[CHROMA_FORMAT], sizeof(h->luma2chroma_pixel) );
1275
1276     switch( CHROMA_FORMAT )
1277     {
1278         case CHROMA_420:
1279             memcpy( h->predict_chroma, h->predict_8x8c, sizeof(h->predict_chroma) );
1280             h->mc.prefetch_fenc = h->mc.prefetch_fenc_420;
1281             h->loopf.deblock_chroma[0] = h->loopf.deblock_h_chroma_420;
1282             h->loopf.deblock_chroma_intra[0] = h->loopf.deblock_h_chroma_420_intra;
1283             h->loopf.deblock_chroma_mbaff = h->loopf.deblock_chroma_420_mbaff;
1284             h->loopf.deblock_chroma_intra_mbaff = h->loopf.deblock_chroma_420_intra_mbaff;
1285             h->pixf.intra_mbcmp_x3_chroma = h->pixf.intra_mbcmp_x3_8x8c;
1286             h->quantf.coeff_last[DCT_CHROMA_DC] = h->quantf.coeff_last4;
1287             h->quantf.coeff_level_run[DCT_CHROMA_DC] = h->quantf.coeff_level_run4;
1288             break;
1289         case CHROMA_422:
1290             memcpy( h->predict_chroma, h->predict_8x16c, sizeof(h->predict_chroma) );
1291             h->mc.prefetch_fenc = h->mc.prefetch_fenc_422;
1292             h->loopf.deblock_chroma[0] = h->loopf.deblock_h_chroma_422;
1293             h->loopf.deblock_chroma_intra[0] = h->loopf.deblock_h_chroma_422_intra;
1294             h->loopf.deblock_chroma_mbaff = h->loopf.deblock_chroma_422_mbaff;
1295             h->loopf.deblock_chroma_intra_mbaff = h->loopf.deblock_chroma_422_intra_mbaff;
1296             h->pixf.intra_mbcmp_x3_chroma = h->pixf.intra_mbcmp_x3_8x16c;
1297             h->quantf.coeff_last[DCT_CHROMA_DC] = h->quantf.coeff_last8;
1298             h->quantf.coeff_level_run[DCT_CHROMA_DC] = h->quantf.coeff_level_run8;
1299             break;
1300         case CHROMA_444:
1301             h->mc.prefetch_fenc = h->mc.prefetch_fenc_422; /* FIXME: doesn't cover V plane */
1302             h->loopf.deblock_chroma_mbaff = h->loopf.deblock_luma_mbaff;
1303             h->loopf.deblock_chroma_intra_mbaff = h->loopf.deblock_luma_intra_mbaff;
1304             break;
1305     }
1306 }
1307
1308 static void x264_set_aspect_ratio( x264_t *h, x264_param_t *param, int initial )
1309 {
1310     /* VUI */
1311     if( param->vui.i_sar_width > 0 && param->vui.i_sar_height > 0 )
1312     {
1313         uint32_t i_w = param->vui.i_sar_width;
1314         uint32_t i_h = param->vui.i_sar_height;
1315         uint32_t old_w = h->param.vui.i_sar_width;
1316         uint32_t old_h = h->param.vui.i_sar_height;
1317
1318         x264_reduce_fraction( &i_w, &i_h );
1319
1320         while( i_w > 65535 || i_h > 65535 )
1321         {
1322             i_w /= 2;
1323             i_h /= 2;
1324         }
1325
1326         x264_reduce_fraction( &i_w, &i_h );
1327
1328         if( i_w != old_w || i_h != old_h || initial )
1329         {
1330             h->param.vui.i_sar_width = 0;
1331             h->param.vui.i_sar_height = 0;
1332             if( i_w == 0 || i_h == 0 )
1333                 x264_log( h, X264_LOG_WARNING, "cannot create valid sample aspect ratio\n" );
1334             else
1335             {
1336                 x264_log( h, initial?X264_LOG_INFO:X264_LOG_DEBUG, "using SAR=%d/%d\n", i_w, i_h );
1337                 h->param.vui.i_sar_width = i_w;
1338                 h->param.vui.i_sar_height = i_h;
1339             }
1340         }
1341     }
1342 }
1343
1344 /****************************************************************************
1345  * x264_encoder_open:
1346  ****************************************************************************/
1347 x264_t *x264_encoder_open( x264_param_t *param )
1348 {
1349     x264_t *h;
1350     char buf[1000], *p;
1351     int qp, i_slicetype_length;
1352
1353     CHECKED_MALLOCZERO( h, sizeof(x264_t) );
1354
1355     /* Create a copy of param */
1356     memcpy( &h->param, param, sizeof(x264_param_t) );
1357
1358     if( param->param_free )
1359         param->param_free( param );
1360
1361     if( x264_threading_init() )
1362     {
1363         x264_log( h, X264_LOG_ERROR, "unable to initialize threading\n" );
1364         goto fail;
1365     }
1366
1367     if( x264_validate_parameters( h, 1 ) < 0 )
1368         goto fail;
1369
1370     if( h->param.psz_cqm_file )
1371         if( x264_cqm_parse_file( h, h->param.psz_cqm_file ) < 0 )
1372             goto fail;
1373
1374     if( h->param.rc.psz_stat_out )
1375         h->param.rc.psz_stat_out = strdup( h->param.rc.psz_stat_out );
1376     if( h->param.rc.psz_stat_in )
1377         h->param.rc.psz_stat_in = strdup( h->param.rc.psz_stat_in );
1378
1379     x264_reduce_fraction( &h->param.i_fps_num, &h->param.i_fps_den );
1380     x264_reduce_fraction( &h->param.i_timebase_num, &h->param.i_timebase_den );
1381
1382     /* Init x264_t */
1383     h->i_frame = -1;
1384     h->i_frame_num = 0;
1385     h->i_idr_pic_id = 0;
1386
1387     if( (uint64_t)h->param.i_timebase_den * 2 > UINT32_MAX )
1388     {
1389         x264_log( h, X264_LOG_ERROR, "Effective timebase denominator %u exceeds H.264 maximum\n", h->param.i_timebase_den );
1390         goto fail;
1391     }
1392
1393     x264_set_aspect_ratio( h, &h->param, 1 );
1394
1395     x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
1396     x264_pps_init( h->pps, h->param.i_sps_id, &h->param, h->sps );
1397
1398     x264_validate_levels( h, 1 );
1399
1400     h->chroma_qp_table = i_chroma_qp_table + 12 + h->pps->i_chroma_qp_index_offset;
1401
1402     if( x264_cqm_init( h ) < 0 )
1403         goto fail;
1404
1405     h->mb.i_mb_width = h->sps->i_mb_width;
1406     h->mb.i_mb_height = h->sps->i_mb_height;
1407     h->mb.i_mb_count = h->mb.i_mb_width * h->mb.i_mb_height;
1408
1409     h->mb.chroma_h_shift = CHROMA_FORMAT == CHROMA_420 || CHROMA_FORMAT == CHROMA_422;
1410     h->mb.chroma_v_shift = CHROMA_FORMAT == CHROMA_420;
1411
1412     /* Adaptive MBAFF and subme 0 are not supported as we require halving motion
1413      * vectors during prediction, resulting in hpel mvs.
1414      * The chosen solution is to make MBAFF non-adaptive in this case. */
1415     h->mb.b_adaptive_mbaff = PARAM_INTERLACED && h->param.analyse.i_subpel_refine;
1416
1417     /* Init frames. */
1418     if( h->param.i_bframe_adaptive == X264_B_ADAPT_TRELLIS && !h->param.rc.b_stat_read )
1419         h->frames.i_delay = X264_MAX(h->param.i_bframe,3)*4;
1420     else
1421         h->frames.i_delay = h->param.i_bframe;
1422     if( h->param.rc.b_mb_tree || h->param.rc.i_vbv_buffer_size )
1423         h->frames.i_delay = X264_MAX( h->frames.i_delay, h->param.rc.i_lookahead );
1424     i_slicetype_length = h->frames.i_delay;
1425     h->frames.i_delay += h->i_thread_frames - 1;
1426     h->frames.i_delay += h->param.i_sync_lookahead;
1427     h->frames.i_delay += h->param.b_vfr_input;
1428     h->frames.i_bframe_delay = h->param.i_bframe ? (h->param.i_bframe_pyramid ? 2 : 1) : 0;
1429
1430     h->frames.i_max_ref0 = h->param.i_frame_reference;
1431     h->frames.i_max_ref1 = X264_MIN( h->sps->vui.i_num_reorder_frames, h->param.i_frame_reference );
1432     h->frames.i_max_dpb  = h->sps->vui.i_max_dec_frame_buffering;
1433     h->frames.b_have_lowres = !h->param.rc.b_stat_read
1434         && ( h->param.rc.i_rc_method == X264_RC_ABR
1435           || h->param.rc.i_rc_method == X264_RC_CRF
1436           || h->param.i_bframe_adaptive
1437           || h->param.i_scenecut_threshold
1438           || h->param.rc.b_mb_tree
1439           || h->param.analyse.i_weighted_pred );
1440     h->frames.b_have_lowres |= h->param.rc.b_stat_read && h->param.rc.i_vbv_buffer_size > 0;
1441     h->frames.b_have_sub8x8_esa = !!(h->param.analyse.inter & X264_ANALYSE_PSUB8x8);
1442
1443     h->frames.i_last_idr =
1444     h->frames.i_last_keyframe = - h->param.i_keyint_max;
1445     h->frames.i_input    = 0;
1446     h->frames.i_largest_pts = h->frames.i_second_largest_pts = -1;
1447     h->frames.i_poc_last_open_gop = -1;
1448
1449     CHECKED_MALLOCZERO( h->frames.unused[0], (h->frames.i_delay + 3) * sizeof(x264_frame_t *) );
1450     /* Allocate room for max refs plus a few extra just in case. */
1451     CHECKED_MALLOCZERO( h->frames.unused[1], (h->i_thread_frames + X264_REF_MAX + 4) * sizeof(x264_frame_t *) );
1452     CHECKED_MALLOCZERO( h->frames.current, (h->param.i_sync_lookahead + h->param.i_bframe
1453                         + h->i_thread_frames + 3) * sizeof(x264_frame_t *) );
1454     if( h->param.analyse.i_weighted_pred > 0 )
1455         CHECKED_MALLOCZERO( h->frames.blank_unused, h->i_thread_frames * 4 * sizeof(x264_frame_t *) );
1456     h->i_ref[0] = h->i_ref[1] = 0;
1457     h->i_cpb_delay = h->i_coded_fields = h->i_disp_fields = 0;
1458     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);
1459     h->i_disp_fields_last_frame = -1;
1460     x264_rdo_init();
1461
1462     /* init CPU functions */
1463     x264_predict_16x16_init( h->param.cpu, h->predict_16x16 );
1464     x264_predict_8x8c_init( h->param.cpu, h->predict_8x8c );
1465     x264_predict_8x16c_init( h->param.cpu, h->predict_8x16c );
1466     x264_predict_8x8_init( h->param.cpu, h->predict_8x8, &h->predict_8x8_filter );
1467     x264_predict_4x4_init( h->param.cpu, h->predict_4x4 );
1468     x264_pixel_init( h->param.cpu, &h->pixf );
1469     x264_dct_init( h->param.cpu, &h->dctf );
1470     x264_zigzag_init( h->param.cpu, &h->zigzagf_progressive, &h->zigzagf_interlaced );
1471     memcpy( &h->zigzagf, PARAM_INTERLACED ? &h->zigzagf_interlaced : &h->zigzagf_progressive, sizeof(h->zigzagf) );
1472     x264_mc_init( h->param.cpu, &h->mc, h->param.b_cpu_independent );
1473     x264_quant_init( h, h->param.cpu, &h->quantf );
1474     x264_deblock_init( h->param.cpu, &h->loopf, PARAM_INTERLACED );
1475     x264_bitstream_init( h->param.cpu, &h->bsf );
1476     if( h->param.b_cabac )
1477         x264_cabac_init( h );
1478     else
1479         x264_stack_align( x264_cavlc_init, h );
1480
1481     mbcmp_init( h );
1482     chroma_dsp_init( h );
1483
1484     p = buf + sprintf( buf, "using cpu capabilities:" );
1485     for( int i = 0; x264_cpu_names[i].flags; i++ )
1486     {
1487         if( !strcmp(x264_cpu_names[i].name, "SSE")
1488             && h->param.cpu & (X264_CPU_SSE2) )
1489             continue;
1490         if( !strcmp(x264_cpu_names[i].name, "SSE2")
1491             && h->param.cpu & (X264_CPU_SSE2_IS_FAST|X264_CPU_SSE2_IS_SLOW) )
1492             continue;
1493         if( !strcmp(x264_cpu_names[i].name, "SSE3")
1494             && (h->param.cpu & X264_CPU_SSSE3 || !(h->param.cpu & X264_CPU_CACHELINE_64)) )
1495             continue;
1496         if( !strcmp(x264_cpu_names[i].name, "SSE4.1")
1497             && (h->param.cpu & X264_CPU_SSE42) )
1498             continue;
1499         if( !strcmp(x264_cpu_names[i].name, "BMI1")
1500             && (h->param.cpu & X264_CPU_BMI2) )
1501             continue;
1502         if( (h->param.cpu & x264_cpu_names[i].flags) == x264_cpu_names[i].flags
1503             && (!i || x264_cpu_names[i].flags != x264_cpu_names[i-1].flags) )
1504             p += sprintf( p, " %s", x264_cpu_names[i].name );
1505     }
1506     if( !h->param.cpu )
1507         p += sprintf( p, " none!" );
1508     x264_log( h, X264_LOG_INFO, "%s\n", buf );
1509
1510     float *logs = x264_analyse_prepare_costs( h );
1511     if( !logs )
1512         goto fail;
1513     for( qp = X264_MIN( h->param.rc.i_qp_min, QP_MAX_SPEC ); qp <= h->param.rc.i_qp_max; qp++ )
1514         if( x264_analyse_init_costs( h, logs, qp ) )
1515             goto fail;
1516     if( x264_analyse_init_costs( h, logs, X264_LOOKAHEAD_QP ) )
1517         goto fail;
1518     x264_free( logs );
1519
1520     static const uint16_t cost_mv_correct[7] = { 24, 47, 95, 189, 379, 757, 1515 };
1521     /* Checks for known miscompilation issues. */
1522     if( h->cost_mv[X264_LOOKAHEAD_QP][2013] != cost_mv_correct[BIT_DEPTH-8] )
1523     {
1524         x264_log( h, X264_LOG_ERROR, "MV cost test failed: x264 has been miscompiled!\n" );
1525         goto fail;
1526     }
1527
1528     /* Must be volatile or else GCC will optimize it out. */
1529     volatile int temp = 392;
1530     if( x264_clz( temp ) != 23 )
1531     {
1532         x264_log( h, X264_LOG_ERROR, "CLZ test failed: x264 has been miscompiled!\n" );
1533 #if ARCH_X86 || ARCH_X86_64
1534         x264_log( h, X264_LOG_ERROR, "Are you attempting to run an SSE4a/LZCNT-targeted build on a CPU that\n" );
1535         x264_log( h, X264_LOG_ERROR, "doesn't support it?\n" );
1536 #endif
1537         goto fail;
1538     }
1539
1540     h->out.i_nal = 0;
1541     h->out.i_bitstream = X264_MAX( 1000000, h->param.i_width * h->param.i_height * 4
1542         * ( h->param.rc.i_rc_method == X264_RC_ABR ? pow( 0.95, h->param.rc.i_qp_min )
1543           : pow( 0.95, h->param.rc.i_qp_constant ) * X264_MAX( 1, h->param.rc.f_ip_factor )));
1544
1545     h->nal_buffer_size = h->out.i_bitstream * 3/2 + 4 + 64; /* +4 for startcode, +64 for nal_escape assembly padding */
1546     CHECKED_MALLOC( h->nal_buffer, h->nal_buffer_size );
1547
1548     CHECKED_MALLOC( h->reconfig_h, sizeof(x264_t) );
1549
1550     if( h->param.i_threads > 1 &&
1551         x264_threadpool_init( &h->threadpool, h->param.i_threads, (void*)x264_encoder_thread_init, h ) )
1552         goto fail;
1553     if( h->param.i_lookahead_threads > 1 &&
1554         x264_threadpool_init( &h->lookaheadpool, h->param.i_lookahead_threads, NULL, NULL ) )
1555         goto fail;
1556
1557 #if HAVE_OPENCL
1558     if( h->param.b_opencl )
1559     {
1560         h->opencl.ocl = x264_opencl_load_library();
1561         if( !h->opencl.ocl )
1562         {
1563             x264_log( h, X264_LOG_WARNING, "failed to load OpenCL\n" );
1564             h->param.b_opencl = 0;
1565         }
1566     }
1567 #endif
1568
1569     h->thread[0] = h;
1570     for( int i = 1; i < h->param.i_threads + !!h->param.i_sync_lookahead; i++ )
1571         CHECKED_MALLOC( h->thread[i], sizeof(x264_t) );
1572     if( h->param.i_lookahead_threads > 1 )
1573         for( int i = 0; i < h->param.i_lookahead_threads; i++ )
1574         {
1575             CHECKED_MALLOC( h->lookahead_thread[i], sizeof(x264_t) );
1576             *h->lookahead_thread[i] = *h;
1577         }
1578     *h->reconfig_h = *h;
1579
1580     for( int i = 0; i < h->param.i_threads; i++ )
1581     {
1582         int init_nal_count = h->param.i_slice_count + 3;
1583         int allocate_threadlocal_data = !h->param.b_sliced_threads || !i;
1584         if( i > 0 )
1585             *h->thread[i] = *h;
1586
1587         if( x264_pthread_mutex_init( &h->thread[i]->mutex, NULL ) )
1588             goto fail;
1589         if( x264_pthread_cond_init( &h->thread[i]->cv, NULL ) )
1590             goto fail;
1591
1592         if( allocate_threadlocal_data )
1593         {
1594             h->thread[i]->fdec = x264_frame_pop_unused( h, 1 );
1595             if( !h->thread[i]->fdec )
1596                 goto fail;
1597         }
1598         else
1599             h->thread[i]->fdec = h->thread[0]->fdec;
1600
1601         CHECKED_MALLOC( h->thread[i]->out.p_bitstream, h->out.i_bitstream );
1602         /* Start each thread with room for init_nal_count NAL units; it'll realloc later if needed. */
1603         CHECKED_MALLOC( h->thread[i]->out.nal, init_nal_count*sizeof(x264_nal_t) );
1604         h->thread[i]->out.i_nals_allocated = init_nal_count;
1605
1606         if( allocate_threadlocal_data && x264_macroblock_cache_allocate( h->thread[i] ) < 0 )
1607             goto fail;
1608     }
1609
1610 #if HAVE_OPENCL
1611     if( h->param.b_opencl && x264_opencl_lookahead_init( h ) < 0 )
1612         h->param.b_opencl = 0;
1613 #endif
1614
1615     if( x264_lookahead_init( h, i_slicetype_length ) )
1616         goto fail;
1617
1618     for( int i = 0; i < h->param.i_threads; i++ )
1619         if( x264_macroblock_thread_allocate( h->thread[i], 0 ) < 0 )
1620             goto fail;
1621
1622     if( x264_ratecontrol_new( h ) < 0 )
1623         goto fail;
1624
1625     if( h->param.i_nal_hrd )
1626     {
1627         x264_log( h, X264_LOG_DEBUG, "HRD bitrate: %i bits/sec\n", h->sps->vui.hrd.i_bit_rate_unscaled );
1628         x264_log( h, X264_LOG_DEBUG, "CPB size: %i bits\n", h->sps->vui.hrd.i_cpb_size_unscaled );
1629     }
1630
1631     if( h->param.psz_dump_yuv )
1632     {
1633         /* create or truncate the reconstructed video file */
1634         FILE *f = x264_fopen( h->param.psz_dump_yuv, "w" );
1635         if( !f )
1636         {
1637             x264_log( h, X264_LOG_ERROR, "dump_yuv: can't write to %s\n", h->param.psz_dump_yuv );
1638             goto fail;
1639         }
1640         else if( !x264_is_regular_file( f ) )
1641         {
1642             x264_log( h, X264_LOG_ERROR, "dump_yuv: incompatible with non-regular file %s\n", h->param.psz_dump_yuv );
1643             goto fail;
1644         }
1645         fclose( f );
1646     }
1647
1648     const char *profile = h->sps->i_profile_idc == PROFILE_BASELINE ? "Constrained Baseline" :
1649                           h->sps->i_profile_idc == PROFILE_MAIN ? "Main" :
1650                           h->sps->i_profile_idc == PROFILE_HIGH ? "High" :
1651                           h->sps->i_profile_idc == PROFILE_HIGH10 ? (h->sps->b_constraint_set3 == 1 ? "High 10 Intra" : "High 10") :
1652                           h->sps->i_profile_idc == PROFILE_HIGH422 ? (h->sps->b_constraint_set3 == 1 ? "High 4:2:2 Intra" : "High 4:2:2") :
1653                           h->sps->b_constraint_set3 == 1 ? "High 4:4:4 Intra" : "High 4:4:4 Predictive";
1654     char level[4];
1655     snprintf( level, sizeof(level), "%d.%d", h->sps->i_level_idc/10, h->sps->i_level_idc%10 );
1656     if( h->sps->i_level_idc == 9 || ( h->sps->i_level_idc == 11 && h->sps->b_constraint_set3 &&
1657         (h->sps->i_profile_idc == PROFILE_BASELINE || h->sps->i_profile_idc == PROFILE_MAIN) ) )
1658         strcpy( level, "1b" );
1659
1660     if( h->sps->i_profile_idc < PROFILE_HIGH10 )
1661     {
1662         x264_log( h, X264_LOG_INFO, "profile %s, level %s\n",
1663             profile, level );
1664     }
1665     else
1666     {
1667         static const char * const subsampling[4] = { "4:0:0", "4:2:0", "4:2:2", "4:4:4" };
1668         x264_log( h, X264_LOG_INFO, "profile %s, level %s, %s %d-bit\n",
1669             profile, level, subsampling[CHROMA_FORMAT], BIT_DEPTH );
1670     }
1671
1672     return h;
1673 fail:
1674     x264_free( h );
1675     return NULL;
1676 }
1677
1678 /****************************************************************************/
1679 static int x264_encoder_try_reconfig( x264_t *h, x264_param_t *param, int *rc_reconfig )
1680 {
1681     *rc_reconfig = 0;
1682     x264_set_aspect_ratio( h, param, 0 );
1683 #define COPY(var) h->param.var = param->var
1684     COPY( i_frame_reference ); // but never uses more refs than initially specified
1685     COPY( i_bframe_bias );
1686     if( h->param.i_scenecut_threshold )
1687         COPY( i_scenecut_threshold ); // can't turn it on or off, only vary the threshold
1688     COPY( b_deblocking_filter );
1689     COPY( i_deblocking_filter_alphac0 );
1690     COPY( i_deblocking_filter_beta );
1691     COPY( i_frame_packing );
1692     COPY( analyse.inter );
1693     COPY( analyse.intra );
1694     COPY( analyse.i_direct_mv_pred );
1695     /* Scratch buffer prevents me_range from being increased for esa/tesa */
1696     if( h->param.analyse.i_me_method < X264_ME_ESA || param->analyse.i_me_range < h->param.analyse.i_me_range )
1697         COPY( analyse.i_me_range );
1698     COPY( analyse.i_noise_reduction );
1699     /* We can't switch out of subme=0 during encoding. */
1700     if( h->param.analyse.i_subpel_refine )
1701         COPY( analyse.i_subpel_refine );
1702     COPY( analyse.i_trellis );
1703     COPY( analyse.b_chroma_me );
1704     COPY( analyse.b_dct_decimate );
1705     COPY( analyse.b_fast_pskip );
1706     COPY( analyse.b_mixed_references );
1707     COPY( analyse.f_psy_rd );
1708     COPY( analyse.f_psy_trellis );
1709     COPY( crop_rect );
1710     // can only twiddle these if they were enabled to begin with:
1711     if( h->param.analyse.i_me_method >= X264_ME_ESA || param->analyse.i_me_method < X264_ME_ESA )
1712         COPY( analyse.i_me_method );
1713     if( h->param.analyse.i_me_method >= X264_ME_ESA && !h->frames.b_have_sub8x8_esa )
1714         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
1715     if( h->pps->b_transform_8x8_mode )
1716         COPY( analyse.b_transform_8x8 );
1717     if( h->frames.i_max_ref1 > 1 )
1718         COPY( i_bframe_pyramid );
1719     COPY( i_slice_max_size );
1720     COPY( i_slice_max_mbs );
1721     COPY( i_slice_min_mbs );
1722     COPY( i_slice_count );
1723     COPY( i_slice_count_max );
1724     COPY( b_tff );
1725
1726     /* VBV can't be turned on if it wasn't on to begin with */
1727     if( h->param.rc.i_vbv_max_bitrate > 0 && h->param.rc.i_vbv_buffer_size > 0 &&
1728           param->rc.i_vbv_max_bitrate > 0 &&   param->rc.i_vbv_buffer_size > 0 )
1729     {
1730         *rc_reconfig |= h->param.rc.i_vbv_max_bitrate != param->rc.i_vbv_max_bitrate;
1731         *rc_reconfig |= h->param.rc.i_vbv_buffer_size != param->rc.i_vbv_buffer_size;
1732         *rc_reconfig |= h->param.rc.i_bitrate != param->rc.i_bitrate;
1733         COPY( rc.i_vbv_max_bitrate );
1734         COPY( rc.i_vbv_buffer_size );
1735         COPY( rc.i_bitrate );
1736     }
1737     *rc_reconfig |= h->param.rc.f_rf_constant != param->rc.f_rf_constant;
1738     *rc_reconfig |= h->param.rc.f_rf_constant_max != param->rc.f_rf_constant_max;
1739     COPY( rc.f_rf_constant );
1740     COPY( rc.f_rf_constant_max );
1741 #undef COPY
1742
1743     return x264_validate_parameters( h, 0 );
1744 }
1745
1746 int x264_encoder_reconfig_apply( x264_t *h, x264_param_t *param )
1747 {
1748     int rc_reconfig;
1749     int ret = x264_encoder_try_reconfig( h, param, &rc_reconfig );
1750
1751     mbcmp_init( h );
1752     if( !ret )
1753         x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
1754
1755     /* Supported reconfiguration options (1-pass only):
1756      * vbv-maxrate
1757      * vbv-bufsize
1758      * crf
1759      * bitrate (CBR only) */
1760     if( !ret && rc_reconfig )
1761         x264_ratecontrol_init_reconfigurable( h, 0 );
1762
1763     return ret;
1764 }
1765
1766 /****************************************************************************
1767  * x264_encoder_reconfig:
1768  ****************************************************************************/
1769 int x264_encoder_reconfig( x264_t *h, x264_param_t *param )
1770 {
1771     h = h->thread[h->thread[0]->i_thread_phase];
1772     x264_param_t param_save = h->reconfig_h->param;
1773     h->reconfig_h->param = h->param;
1774
1775     int rc_reconfig;
1776     int ret = x264_encoder_try_reconfig( h->reconfig_h, param, &rc_reconfig );
1777     if( !ret )
1778         h->reconfig = 1;
1779     else
1780         h->reconfig_h->param = param_save;
1781
1782     return ret;
1783 }
1784
1785 /****************************************************************************
1786  * x264_encoder_parameters:
1787  ****************************************************************************/
1788 void x264_encoder_parameters( x264_t *h, x264_param_t *param )
1789 {
1790     memcpy( param, &h->thread[h->i_thread_phase]->param, sizeof(x264_param_t) );
1791 }
1792
1793 /* internal usage */
1794 static void x264_nal_start( x264_t *h, int i_type, int i_ref_idc )
1795 {
1796     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1797
1798     nal->i_ref_idc        = i_ref_idc;
1799     nal->i_type           = i_type;
1800     nal->b_long_startcode = 1;
1801
1802     nal->i_payload= 0;
1803     nal->p_payload= &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8];
1804     nal->i_padding= 0;
1805 }
1806
1807 /* if number of allocated nals is not enough, re-allocate a larger one. */
1808 static int x264_nal_check_buffer( x264_t *h )
1809 {
1810     if( h->out.i_nal >= h->out.i_nals_allocated )
1811     {
1812         x264_nal_t *new_out = x264_malloc( sizeof(x264_nal_t) * (h->out.i_nals_allocated*2) );
1813         if( !new_out )
1814             return -1;
1815         memcpy( new_out, h->out.nal, sizeof(x264_nal_t) * (h->out.i_nals_allocated) );
1816         x264_free( h->out.nal );
1817         h->out.nal = new_out;
1818         h->out.i_nals_allocated *= 2;
1819     }
1820     return 0;
1821 }
1822
1823 static int x264_nal_end( x264_t *h )
1824 {
1825     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1826     uint8_t *end = &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8];
1827     nal->i_payload = end - nal->p_payload;
1828     /* Assembly implementation of nal_escape reads past the end of the input.
1829      * While undefined padding wouldn't actually affect the output, it makes valgrind unhappy. */
1830     memset( end, 0xff, 64 );
1831     if( h->param.nalu_process )
1832         h->param.nalu_process( h, nal, h->fenc->opaque );
1833     h->out.i_nal++;
1834
1835     return x264_nal_check_buffer( h );
1836 }
1837
1838 static int x264_check_encapsulated_buffer( x264_t *h, x264_t *h0, int start,
1839                                            int previous_nal_size, int necessary_size )
1840 {
1841     if( h0->nal_buffer_size < necessary_size )
1842     {
1843         necessary_size *= 2;
1844         uint8_t *buf = x264_malloc( necessary_size );
1845         if( !buf )
1846             return -1;
1847         if( previous_nal_size )
1848             memcpy( buf, h0->nal_buffer, previous_nal_size );
1849
1850         intptr_t delta = buf - h0->nal_buffer;
1851         for( int i = 0; i < start; i++ )
1852             h->out.nal[i].p_payload += delta;
1853
1854         x264_free( h0->nal_buffer );
1855         h0->nal_buffer = buf;
1856         h0->nal_buffer_size = necessary_size;
1857     }
1858
1859     return 0;
1860 }
1861
1862 static int x264_encoder_encapsulate_nals( x264_t *h, int start )
1863 {
1864     x264_t *h0 = h->thread[0];
1865     int nal_size = 0, previous_nal_size = 0;
1866
1867     if( h->param.nalu_process )
1868     {
1869         for( int i = start; i < h->out.i_nal; i++ )
1870             nal_size += h->out.nal[i].i_payload;
1871         return nal_size;
1872     }
1873
1874     for( int i = 0; i < start; i++ )
1875         previous_nal_size += h->out.nal[i].i_payload;
1876
1877     for( int i = start; i < h->out.i_nal; i++ )
1878         nal_size += h->out.nal[i].i_payload;
1879
1880     /* Worst-case NAL unit escaping: reallocate the buffer if it's too small. */
1881     int necessary_size = previous_nal_size + nal_size * 3/2 + h->out.i_nal * 4 + 4 + 64;
1882     for( int i = start; i < h->out.i_nal; i++ )
1883         necessary_size += h->out.nal[i].i_padding;
1884     if( x264_check_encapsulated_buffer( h, h0, start, previous_nal_size, necessary_size ) )
1885         return -1;
1886
1887     uint8_t *nal_buffer = h0->nal_buffer + previous_nal_size;
1888
1889     for( int i = start; i < h->out.i_nal; i++ )
1890     {
1891         int old_payload_len = h->out.nal[i].i_payload;
1892         h->out.nal[i].b_long_startcode = !i || h->out.nal[i].i_type == NAL_SPS || h->out.nal[i].i_type == NAL_PPS ||
1893                                          h->param.b_avcintra_compat;
1894         x264_nal_encode( h, nal_buffer, &h->out.nal[i] );
1895         nal_buffer += h->out.nal[i].i_payload;
1896         if( h->param.b_avcintra_compat )
1897         {
1898             h->out.nal[i].i_padding -= h->out.nal[i].i_payload - (old_payload_len + NALU_OVERHEAD);
1899             if( h->out.nal[i].i_padding > 0 )
1900             {
1901                 memset( nal_buffer, 0, h->out.nal[i].i_padding );
1902                 nal_buffer += h->out.nal[i].i_padding;
1903                 h->out.nal[i].i_payload += h->out.nal[i].i_padding;
1904             }
1905             h->out.nal[i].i_padding = X264_MAX( h->out.nal[i].i_padding, 0 );
1906         }
1907     }
1908
1909     x264_emms();
1910
1911     return nal_buffer - (h0->nal_buffer + previous_nal_size);
1912 }
1913
1914 /****************************************************************************
1915  * x264_encoder_headers:
1916  ****************************************************************************/
1917 int x264_encoder_headers( x264_t *h, x264_nal_t **pp_nal, int *pi_nal )
1918 {
1919     int frame_size = 0;
1920     /* init bitstream context */
1921     h->out.i_nal = 0;
1922     bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
1923
1924     /* Write SEI, SPS and PPS. */
1925
1926     /* generate sequence parameters */
1927     x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
1928     x264_sps_write( &h->out.bs, h->sps );
1929     if( x264_nal_end( h ) )
1930         return -1;
1931
1932     /* generate picture parameters */
1933     x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
1934     x264_pps_write( &h->out.bs, h->sps, h->pps );
1935     if( x264_nal_end( h ) )
1936         return -1;
1937
1938     /* identify ourselves */
1939     x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
1940     if( x264_sei_version_write( h, &h->out.bs ) )
1941         return -1;
1942     if( x264_nal_end( h ) )
1943         return -1;
1944
1945     frame_size = x264_encoder_encapsulate_nals( h, 0 );
1946     if( frame_size < 0 )
1947         return -1;
1948
1949     /* now set output*/
1950     *pi_nal = h->out.i_nal;
1951     *pp_nal = &h->out.nal[0];
1952     h->out.i_nal = 0;
1953
1954     return frame_size;
1955 }
1956
1957 /* Check to see whether we have chosen a reference list ordering different
1958  * from the standard's default. */
1959 static inline void x264_reference_check_reorder( x264_t *h )
1960 {
1961     /* The reorder check doesn't check for missing frames, so just
1962      * force a reorder if one of the reference list is corrupt. */
1963     for( int i = 0; h->frames.reference[i]; i++ )
1964         if( h->frames.reference[i]->b_corrupt )
1965         {
1966             h->b_ref_reorder[0] = 1;
1967             return;
1968         }
1969     for( int list = 0; list <= (h->sh.i_type == SLICE_TYPE_B); list++ )
1970         for( int i = 0; i < h->i_ref[list] - 1; i++ )
1971         {
1972             int framenum_diff = h->fref[list][i+1]->i_frame_num - h->fref[list][i]->i_frame_num;
1973             int poc_diff = h->fref[list][i+1]->i_poc - h->fref[list][i]->i_poc;
1974             /* P and B-frames use different default orders. */
1975             if( h->sh.i_type == SLICE_TYPE_P ? framenum_diff > 0 : list == 1 ? poc_diff < 0 : poc_diff > 0 )
1976             {
1977                 h->b_ref_reorder[list] = 1;
1978                 return;
1979             }
1980         }
1981 }
1982
1983 /* return -1 on failure, else return the index of the new reference frame */
1984 int x264_weighted_reference_duplicate( x264_t *h, int i_ref, const x264_weight_t *w )
1985 {
1986     int i = h->i_ref[0];
1987     int j = 1;
1988     x264_frame_t *newframe;
1989     if( i <= 1 ) /* empty list, definitely can't duplicate frame */
1990         return -1;
1991
1992     //Duplication is only used in X264_WEIGHTP_SMART
1993     if( h->param.analyse.i_weighted_pred != X264_WEIGHTP_SMART )
1994         return -1;
1995
1996     /* Duplication is a hack to compensate for crappy rounding in motion compensation.
1997      * With high bit depth, it's not worth doing, so turn it off except in the case of
1998      * unweighted dupes. */
1999     if( BIT_DEPTH > 8 && w != x264_weight_none )
2000         return -1;
2001
2002     newframe = x264_frame_pop_blank_unused( h );
2003     if( !newframe )
2004         return -1;
2005
2006     //FIXME: probably don't need to copy everything
2007     *newframe = *h->fref[0][i_ref];
2008     newframe->i_reference_count = 1;
2009     newframe->orig = h->fref[0][i_ref];
2010     newframe->b_duplicate = 1;
2011     memcpy( h->fenc->weight[j], w, sizeof(h->fenc->weight[i]) );
2012
2013     /* shift the frames to make space for the dupe. */
2014     h->b_ref_reorder[0] = 1;
2015     if( h->i_ref[0] < X264_REF_MAX )
2016         ++h->i_ref[0];
2017     h->fref[0][X264_REF_MAX-1] = NULL;
2018     x264_frame_unshift( &h->fref[0][j], newframe );
2019
2020     return j;
2021 }
2022
2023 static void x264_weighted_pred_init( x264_t *h )
2024 {
2025     /* for now no analysis and set all weights to nothing */
2026     for( int i_ref = 0; i_ref < h->i_ref[0]; i_ref++ )
2027         h->fenc->weighted[i_ref] = h->fref[0][i_ref]->filtered[0][0];
2028
2029     // FIXME: This only supports weighting of one reference frame
2030     // and duplicates of that frame.
2031     h->fenc->i_lines_weighted = 0;
2032
2033     for( int i_ref = 0; i_ref < (h->i_ref[0] << SLICE_MBAFF); i_ref++ )
2034         for( int i = 0; i < 3; i++ )
2035             h->sh.weight[i_ref][i].weightfn = NULL;
2036
2037
2038     if( h->sh.i_type != SLICE_TYPE_P || h->param.analyse.i_weighted_pred <= 0 )
2039         return;
2040
2041     int i_padv = PADV << PARAM_INTERLACED;
2042     int denom = -1;
2043     int weightplane[2] = { 0, 0 };
2044     int buffer_next = 0;
2045     for( int i = 0; i < 3; i++ )
2046     {
2047         for( int j = 0; j < h->i_ref[0]; j++ )
2048         {
2049             if( h->fenc->weight[j][i].weightfn )
2050             {
2051                 h->sh.weight[j][i] = h->fenc->weight[j][i];
2052                 // if weight is useless, don't write it to stream
2053                 if( h->sh.weight[j][i].i_scale == 1<<h->sh.weight[j][i].i_denom && h->sh.weight[j][i].i_offset == 0 )
2054                     h->sh.weight[j][i].weightfn = NULL;
2055                 else
2056                 {
2057                     if( !weightplane[!!i] )
2058                     {
2059                         weightplane[!!i] = 1;
2060                         h->sh.weight[0][!!i].i_denom = denom = h->sh.weight[j][i].i_denom;
2061                         assert( x264_clip3( denom, 0, 7 ) == denom );
2062                     }
2063
2064                     assert( h->sh.weight[j][i].i_denom == denom );
2065                     if( !i )
2066                     {
2067                         h->fenc->weighted[j] = h->mb.p_weight_buf[buffer_next++] + h->fenc->i_stride[0] * i_padv + PADH;
2068                         //scale full resolution frame
2069                         if( h->param.i_threads == 1 )
2070                         {
2071                             pixel *src = h->fref[0][j]->filtered[0][0] - h->fref[0][j]->i_stride[0]*i_padv - PADH;
2072                             pixel *dst = h->fenc->weighted[j] - h->fenc->i_stride[0]*i_padv - PADH;
2073                             int stride = h->fenc->i_stride[0];
2074                             int width = h->fenc->i_width[0] + PADH*2;
2075                             int height = h->fenc->i_lines[0] + i_padv*2;
2076                             x264_weight_scale_plane( h, dst, stride, src, stride, width, height, &h->sh.weight[j][0] );
2077                             h->fenc->i_lines_weighted = height;
2078                         }
2079                     }
2080                 }
2081             }
2082         }
2083     }
2084
2085     if( weightplane[1] )
2086         for( int i = 0; i < h->i_ref[0]; i++ )
2087         {
2088             if( h->sh.weight[i][1].weightfn && !h->sh.weight[i][2].weightfn )
2089             {
2090                 h->sh.weight[i][2].i_scale = 1 << h->sh.weight[0][1].i_denom;
2091                 h->sh.weight[i][2].i_offset = 0;
2092             }
2093             else if( h->sh.weight[i][2].weightfn && !h->sh.weight[i][1].weightfn )
2094             {
2095                 h->sh.weight[i][1].i_scale = 1 << h->sh.weight[0][1].i_denom;
2096                 h->sh.weight[i][1].i_offset = 0;
2097             }
2098         }
2099
2100     if( !weightplane[0] )
2101         h->sh.weight[0][0].i_denom = 0;
2102     if( !weightplane[1] )
2103         h->sh.weight[0][1].i_denom = 0;
2104     h->sh.weight[0][2].i_denom = h->sh.weight[0][1].i_denom;
2105 }
2106
2107 static inline int x264_reference_distance( x264_t *h, x264_frame_t *frame )
2108 {
2109     if( h->param.i_frame_packing == 5 )
2110         return abs((h->fenc->i_frame&~1) - (frame->i_frame&~1)) +
2111                   ((h->fenc->i_frame&1) != (frame->i_frame&1));
2112     else
2113         return abs(h->fenc->i_frame - frame->i_frame);
2114 }
2115
2116 static inline void x264_reference_build_list( x264_t *h, int i_poc )
2117 {
2118     int b_ok;
2119
2120     /* build ref list 0/1 */
2121     h->mb.pic.i_fref[0] = h->i_ref[0] = 0;
2122     h->mb.pic.i_fref[1] = h->i_ref[1] = 0;
2123     if( h->sh.i_type == SLICE_TYPE_I )
2124         return;
2125
2126     for( int i = 0; h->frames.reference[i]; i++ )
2127     {
2128         if( h->frames.reference[i]->b_corrupt )
2129             continue;
2130         if( h->frames.reference[i]->i_poc < i_poc )
2131             h->fref[0][h->i_ref[0]++] = h->frames.reference[i];
2132         else if( h->frames.reference[i]->i_poc > i_poc )
2133             h->fref[1][h->i_ref[1]++] = h->frames.reference[i];
2134     }
2135
2136     /* Order reference lists by distance from the current frame. */
2137     for( int list = 0; list < 2; list++ )
2138     {
2139         h->fref_nearest[list] = h->fref[list][0];
2140         do
2141         {
2142             b_ok = 1;
2143             for( int i = 0; i < h->i_ref[list] - 1; i++ )
2144             {
2145                 if( list ? h->fref[list][i+1]->i_poc < h->fref_nearest[list]->i_poc
2146                          : h->fref[list][i+1]->i_poc > h->fref_nearest[list]->i_poc )
2147                     h->fref_nearest[list] = h->fref[list][i+1];
2148                 if( x264_reference_distance( h, h->fref[list][i] ) > x264_reference_distance( h, h->fref[list][i+1] ) )
2149                 {
2150                     XCHG( x264_frame_t*, h->fref[list][i], h->fref[list][i+1] );
2151                     b_ok = 0;
2152                     break;
2153                 }
2154             }
2155         } while( !b_ok );
2156     }
2157
2158     if( h->sh.i_mmco_remove_from_end )
2159         for( int i = h->i_ref[0]-1; i >= h->i_ref[0] - h->sh.i_mmco_remove_from_end; i-- )
2160         {
2161             int diff = h->i_frame_num - h->fref[0][i]->i_frame_num;
2162             h->sh.mmco[h->sh.i_mmco_command_count].i_poc = h->fref[0][i]->i_poc;
2163             h->sh.mmco[h->sh.i_mmco_command_count++].i_difference_of_pic_nums = diff;
2164         }
2165
2166     x264_reference_check_reorder( h );
2167
2168     h->i_ref[1] = X264_MIN( h->i_ref[1], h->frames.i_max_ref1 );
2169     h->i_ref[0] = X264_MIN( h->i_ref[0], h->frames.i_max_ref0 );
2170     h->i_ref[0] = X264_MIN( h->i_ref[0], h->param.i_frame_reference ); // if reconfig() has lowered the limit
2171
2172     /* For Blu-ray compliance, don't reference frames outside of the minigop. */
2173     if( IS_X264_TYPE_B( h->fenc->i_type ) && h->param.b_bluray_compat )
2174         h->i_ref[0] = X264_MIN( h->i_ref[0], IS_X264_TYPE_B( h->fref[0][0]->i_type ) + 1 );
2175
2176     /* add duplicates */
2177     if( h->fenc->i_type == X264_TYPE_P )
2178     {
2179         int idx = -1;
2180         if( h->param.analyse.i_weighted_pred >= X264_WEIGHTP_SIMPLE )
2181         {
2182             x264_weight_t w[3];
2183             w[1].weightfn = w[2].weightfn = NULL;
2184             if( h->param.rc.b_stat_read )
2185                 x264_ratecontrol_set_weights( h, h->fenc );
2186
2187             if( !h->fenc->weight[0][0].weightfn )
2188             {
2189                 h->fenc->weight[0][0].i_denom = 0;
2190                 SET_WEIGHT( w[0], 1, 1, 0, -1 );
2191                 idx = x264_weighted_reference_duplicate( h, 0, w );
2192             }
2193             else
2194             {
2195                 if( h->fenc->weight[0][0].i_scale == 1<<h->fenc->weight[0][0].i_denom )
2196                 {
2197                     SET_WEIGHT( h->fenc->weight[0][0], 1, 1, 0, h->fenc->weight[0][0].i_offset );
2198                 }
2199                 x264_weighted_reference_duplicate( h, 0, x264_weight_none );
2200                 if( h->fenc->weight[0][0].i_offset > -128 )
2201                 {
2202                     w[0] = h->fenc->weight[0][0];
2203                     w[0].i_offset--;
2204                     h->mc.weight_cache( h, &w[0] );
2205                     idx = x264_weighted_reference_duplicate( h, 0, w );
2206                 }
2207             }
2208         }
2209         h->mb.ref_blind_dupe = idx;
2210     }
2211
2212     assert( h->i_ref[0] + h->i_ref[1] <= X264_REF_MAX );
2213     h->mb.pic.i_fref[0] = h->i_ref[0];
2214     h->mb.pic.i_fref[1] = h->i_ref[1];
2215 }
2216
2217 static void x264_fdec_filter_row( x264_t *h, int mb_y, int pass )
2218 {
2219     /* mb_y is the mb to be encoded next, not the mb to be filtered here */
2220     int b_hpel = h->fdec->b_kept_as_ref;
2221     int b_deblock = h->sh.i_disable_deblocking_filter_idc != 1;
2222     int b_end = mb_y == h->i_threadslice_end;
2223     int b_measure_quality = 1;
2224     int min_y = mb_y - (1 << SLICE_MBAFF);
2225     int b_start = min_y == h->i_threadslice_start;
2226     /* Even in interlaced mode, deblocking never modifies more than 4 pixels
2227      * above each MB, as bS=4 doesn't happen for the top of interlaced mbpairs. */
2228     int minpix_y = min_y*16 - 4 * !b_start;
2229     int maxpix_y = mb_y*16 - 4 * !b_end;
2230     b_deblock &= b_hpel || h->param.b_full_recon || h->param.psz_dump_yuv;
2231     if( h->param.b_sliced_threads )
2232     {
2233         switch( pass )
2234         {
2235             /* During encode: only do deblock if asked for */
2236             default:
2237             case 0:
2238                 b_deblock &= h->param.b_full_recon;
2239                 b_hpel = 0;
2240                 break;
2241             /* During post-encode pass: do deblock if not done yet, do hpel for all
2242              * rows except those between slices. */
2243             case 1:
2244                 b_deblock &= !h->param.b_full_recon;
2245                 b_hpel &= !(b_start && min_y > 0);
2246                 b_measure_quality = 0;
2247                 break;
2248             /* Final pass: do the rows between slices in sequence. */
2249             case 2:
2250                 b_deblock = 0;
2251                 b_measure_quality = 0;
2252                 break;
2253         }
2254     }
2255     if( mb_y & SLICE_MBAFF )
2256         return;
2257     if( min_y < h->i_threadslice_start )
2258         return;
2259
2260     if( b_deblock )
2261         for( int y = min_y; y < mb_y; y += (1 << SLICE_MBAFF) )
2262             x264_frame_deblock_row( h, y );
2263
2264     /* FIXME: Prediction requires different borders for interlaced/progressive mc,
2265      * but the actual image data is equivalent. For now, maintain this
2266      * consistency by copying deblocked pixels between planes. */
2267     if( PARAM_INTERLACED && (!h->param.b_sliced_threads || pass == 1) )
2268         for( int p = 0; p < h->fdec->i_plane; p++ )
2269             for( int i = minpix_y>>(CHROMA_V_SHIFT && p); i < maxpix_y>>(CHROMA_V_SHIFT && p); i++ )
2270                 memcpy( h->fdec->plane_fld[p] + i*h->fdec->i_stride[p],
2271                         h->fdec->plane[p]     + i*h->fdec->i_stride[p],
2272                         h->mb.i_mb_width*16*sizeof(pixel) );
2273
2274     if( h->fdec->b_kept_as_ref && (!h->param.b_sliced_threads || pass == 1) )
2275         x264_frame_expand_border( h, h->fdec, min_y );
2276     if( b_hpel )
2277     {
2278         int end = mb_y == h->mb.i_mb_height;
2279         /* Can't do hpel until the previous slice is done encoding. */
2280         if( h->param.analyse.i_subpel_refine )
2281         {
2282             x264_frame_filter( h, h->fdec, min_y, end );
2283             x264_frame_expand_border_filtered( h, h->fdec, min_y, end );
2284         }
2285     }
2286
2287     if( SLICE_MBAFF && pass == 0 )
2288         for( int i = 0; i < 3; i++ )
2289         {
2290             XCHG( pixel *, h->intra_border_backup[0][i], h->intra_border_backup[3][i] );
2291             XCHG( pixel *, h->intra_border_backup[1][i], h->intra_border_backup[4][i] );
2292         }
2293
2294     if( h->i_thread_frames > 1 && h->fdec->b_kept_as_ref )
2295         x264_frame_cond_broadcast( h->fdec, mb_y*16 + (b_end ? 10000 : -(X264_THREAD_HEIGHT << SLICE_MBAFF)) );
2296
2297     if( b_measure_quality )
2298     {
2299         maxpix_y = X264_MIN( maxpix_y, h->param.i_height );
2300         if( h->param.analyse.b_psnr )
2301         {
2302             for( int p = 0; p < (CHROMA444 ? 3 : 1); p++ )
2303                 h->stat.frame.i_ssd[p] += x264_pixel_ssd_wxh( &h->pixf,
2304                     h->fdec->plane[p] + minpix_y * h->fdec->i_stride[p], h->fdec->i_stride[p],
2305                     h->fenc->plane[p] + minpix_y * h->fenc->i_stride[p], h->fenc->i_stride[p],
2306                     h->param.i_width, maxpix_y-minpix_y );
2307             if( !CHROMA444 )
2308             {
2309                 uint64_t ssd_u, ssd_v;
2310                 int v_shift = CHROMA_V_SHIFT;
2311                 x264_pixel_ssd_nv12( &h->pixf,
2312                     h->fdec->plane[1] + (minpix_y>>v_shift) * h->fdec->i_stride[1], h->fdec->i_stride[1],
2313                     h->fenc->plane[1] + (minpix_y>>v_shift) * h->fenc->i_stride[1], h->fenc->i_stride[1],
2314                     h->param.i_width>>1, (maxpix_y-minpix_y)>>v_shift, &ssd_u, &ssd_v );
2315                 h->stat.frame.i_ssd[1] += ssd_u;
2316                 h->stat.frame.i_ssd[2] += ssd_v;
2317             }
2318         }
2319
2320         if( h->param.analyse.b_ssim )
2321         {
2322             int ssim_cnt;
2323             x264_emms();
2324             /* offset by 2 pixels to avoid alignment of ssim blocks with dct blocks,
2325              * and overlap by 4 */
2326             minpix_y += b_start ? 2 : -6;
2327             h->stat.frame.f_ssim +=
2328                 x264_pixel_ssim_wxh( &h->pixf,
2329                     h->fdec->plane[0] + 2+minpix_y*h->fdec->i_stride[0], h->fdec->i_stride[0],
2330                     h->fenc->plane[0] + 2+minpix_y*h->fenc->i_stride[0], h->fenc->i_stride[0],
2331                     h->param.i_width-2, maxpix_y-minpix_y, h->scratch_buffer, &ssim_cnt );
2332             h->stat.frame.i_ssim_cnt += ssim_cnt;
2333         }
2334     }
2335 }
2336
2337 static inline int x264_reference_update( x264_t *h )
2338 {
2339     if( !h->fdec->b_kept_as_ref )
2340     {
2341         if( h->i_thread_frames > 1 )
2342         {
2343             x264_frame_push_unused( h, h->fdec );
2344             h->fdec = x264_frame_pop_unused( h, 1 );
2345             if( !h->fdec )
2346                 return -1;
2347         }
2348         return 0;
2349     }
2350
2351     /* apply mmco from previous frame. */
2352     for( int i = 0; i < h->sh.i_mmco_command_count; i++ )
2353         for( int j = 0; h->frames.reference[j]; j++ )
2354             if( h->frames.reference[j]->i_poc == h->sh.mmco[i].i_poc )
2355                 x264_frame_push_unused( h, x264_frame_shift( &h->frames.reference[j] ) );
2356
2357     /* move frame in the buffer */
2358     x264_frame_push( h->frames.reference, h->fdec );
2359     if( h->frames.reference[h->sps->i_num_ref_frames] )
2360         x264_frame_push_unused( h, x264_frame_shift( h->frames.reference ) );
2361     h->fdec = x264_frame_pop_unused( h, 1 );
2362     if( !h->fdec )
2363         return -1;
2364     return 0;
2365 }
2366
2367 static inline void x264_reference_reset( x264_t *h )
2368 {
2369     while( h->frames.reference[0] )
2370         x264_frame_push_unused( h, x264_frame_pop( h->frames.reference ) );
2371     h->fdec->i_poc =
2372     h->fenc->i_poc = 0;
2373 }
2374
2375 static inline void x264_reference_hierarchy_reset( x264_t *h )
2376 {
2377     int ref;
2378     int b_hasdelayframe = 0;
2379
2380     /* look for delay frames -- chain must only contain frames that are disposable */
2381     for( int i = 0; h->frames.current[i] && IS_DISPOSABLE( h->frames.current[i]->i_type ); i++ )
2382         b_hasdelayframe |= h->frames.current[i]->i_coded
2383                         != h->frames.current[i]->i_frame + h->sps->vui.i_num_reorder_frames;
2384
2385     /* This function must handle b-pyramid and clear frames for open-gop */
2386     if( h->param.i_bframe_pyramid != X264_B_PYRAMID_STRICT && !b_hasdelayframe && h->frames.i_poc_last_open_gop == -1 )
2387         return;
2388
2389     /* Remove last BREF. There will never be old BREFs in the
2390      * dpb during a BREF decode when pyramid == STRICT */
2391     for( ref = 0; h->frames.reference[ref]; ref++ )
2392     {
2393         if( ( h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT
2394             && h->frames.reference[ref]->i_type == X264_TYPE_BREF )
2395             || ( h->frames.reference[ref]->i_poc < h->frames.i_poc_last_open_gop
2396             && h->sh.i_type != SLICE_TYPE_B ) )
2397         {
2398             int diff = h->i_frame_num - h->frames.reference[ref]->i_frame_num;
2399             h->sh.mmco[h->sh.i_mmco_command_count].i_difference_of_pic_nums = diff;
2400             h->sh.mmco[h->sh.i_mmco_command_count++].i_poc = h->frames.reference[ref]->i_poc;
2401             x264_frame_push_unused( h, x264_frame_shift( &h->frames.reference[ref] ) );
2402             h->b_ref_reorder[0] = 1;
2403             ref--;
2404         }
2405     }
2406
2407     /* Prepare room in the dpb for the delayed display time of the later b-frame's */
2408     if( h->param.i_bframe_pyramid )
2409         h->sh.i_mmco_remove_from_end = X264_MAX( ref + 2 - h->frames.i_max_dpb, 0 );
2410 }
2411
2412 static inline void x264_slice_init( x264_t *h, int i_nal_type, int i_global_qp )
2413 {
2414     /* ------------------------ Create slice header  ----------------------- */
2415     if( i_nal_type == NAL_SLICE_IDR )
2416     {
2417         x264_slice_header_init( h, &h->sh, h->sps, h->pps, h->i_idr_pic_id, h->i_frame_num, i_global_qp );
2418
2419         /* alternate id */
2420         h->i_idr_pic_id ^= 1;
2421     }
2422     else
2423     {
2424         x264_slice_header_init( h, &h->sh, h->sps, h->pps, -1, h->i_frame_num, i_global_qp );
2425
2426         h->sh.i_num_ref_idx_l0_active = h->i_ref[0] <= 0 ? 1 : h->i_ref[0];
2427         h->sh.i_num_ref_idx_l1_active = h->i_ref[1] <= 0 ? 1 : h->i_ref[1];
2428         if( h->sh.i_num_ref_idx_l0_active != h->pps->i_num_ref_idx_l0_default_active ||
2429             (h->sh.i_type == SLICE_TYPE_B && h->sh.i_num_ref_idx_l1_active != h->pps->i_num_ref_idx_l1_default_active) )
2430         {
2431             h->sh.b_num_ref_idx_override = 1;
2432         }
2433     }
2434
2435     if( h->fenc->i_type == X264_TYPE_BREF && h->param.b_bluray_compat && h->sh.i_mmco_command_count )
2436     {
2437         h->b_sh_backup = 1;
2438         h->sh_backup = h->sh;
2439     }
2440
2441     h->fdec->i_frame_num = h->sh.i_frame_num;
2442
2443     if( h->sps->i_poc_type == 0 )
2444     {
2445         h->sh.i_poc = h->fdec->i_poc;
2446         if( PARAM_INTERLACED )
2447         {
2448             h->sh.i_delta_poc_bottom = h->param.b_tff ? 1 : -1;
2449             h->sh.i_poc += h->sh.i_delta_poc_bottom == -1;
2450         }
2451         else
2452             h->sh.i_delta_poc_bottom = 0;
2453         h->fdec->i_delta_poc[0] = h->sh.i_delta_poc_bottom == -1;
2454         h->fdec->i_delta_poc[1] = h->sh.i_delta_poc_bottom ==  1;
2455     }
2456     else
2457     {
2458         /* Nothing to do ? */
2459     }
2460
2461     x264_macroblock_slice_init( h );
2462 }
2463
2464 typedef struct
2465 {
2466     int skip;
2467     uint8_t cabac_prevbyte;
2468     bs_t bs;
2469     x264_cabac_t cabac;
2470     x264_frame_stat_t stat;
2471     int last_qp;
2472     int last_dqp;
2473     int field_decoding_flag;
2474 } x264_bs_bak_t;
2475
2476 static ALWAYS_INLINE void x264_bitstream_backup( x264_t *h, x264_bs_bak_t *bak, int i_skip, int full )
2477 {
2478     if( full )
2479     {
2480         bak->stat = h->stat.frame;
2481         bak->last_qp = h->mb.i_last_qp;
2482         bak->last_dqp = h->mb.i_last_dqp;
2483         bak->field_decoding_flag = h->mb.field_decoding_flag;
2484     }
2485     else
2486     {
2487         bak->stat.i_mv_bits = h->stat.frame.i_mv_bits;
2488         bak->stat.i_tex_bits = h->stat.frame.i_tex_bits;
2489     }
2490     /* In the per-MB backup, we don't need the contexts because flushing the CABAC
2491      * encoder has no context dependency and in this case, a slice is ended (and
2492      * thus the content of all contexts are thrown away). */
2493     if( h->param.b_cabac )
2494     {
2495         if( full )
2496             memcpy( &bak->cabac, &h->cabac, sizeof(x264_cabac_t) );
2497         else
2498             memcpy( &bak->cabac, &h->cabac, offsetof(x264_cabac_t, f8_bits_encoded) );
2499         /* x264's CABAC writer modifies the previous byte during carry, so it has to be
2500          * backed up. */
2501         bak->cabac_prevbyte = h->cabac.p[-1];
2502     }
2503     else
2504     {
2505         bak->bs = h->out.bs;
2506         bak->skip = i_skip;
2507     }
2508 }
2509
2510 static ALWAYS_INLINE void x264_bitstream_restore( x264_t *h, x264_bs_bak_t *bak, int *skip, int full )
2511 {
2512     if( full )
2513     {
2514         h->stat.frame = bak->stat;
2515         h->mb.i_last_qp = bak->last_qp;
2516         h->mb.i_last_dqp = bak->last_dqp;
2517         h->mb.field_decoding_flag = bak->field_decoding_flag;
2518     }
2519     else
2520     {
2521         h->stat.frame.i_mv_bits = bak->stat.i_mv_bits;
2522         h->stat.frame.i_tex_bits = bak->stat.i_tex_bits;
2523     }
2524     if( h->param.b_cabac )
2525     {
2526         if( full )
2527             memcpy( &h->cabac, &bak->cabac, sizeof(x264_cabac_t) );
2528         else
2529             memcpy( &h->cabac, &bak->cabac, offsetof(x264_cabac_t, f8_bits_encoded) );
2530         h->cabac.p[-1] = bak->cabac_prevbyte;
2531     }
2532     else
2533     {
2534         h->out.bs = bak->bs;
2535         *skip = bak->skip;
2536     }
2537 }
2538
2539 static int x264_slice_write( x264_t *h )
2540 {
2541     int i_skip;
2542     int mb_xy, i_mb_x, i_mb_y;
2543     /* NALUs other than the first use a 3-byte startcode.
2544      * Add one extra byte for the rbsp, and one more for the final CABAC putbyte.
2545      * Then add an extra 5 bytes just in case, to account for random NAL escapes and
2546      * other inaccuracies. */
2547     int overhead_guess = (NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal)) + 1 + h->param.b_cabac + 5;
2548     int slice_max_size = h->param.i_slice_max_size > 0 ? (h->param.i_slice_max_size-overhead_guess)*8 : 0;
2549     int back_up_bitstream = slice_max_size || (!h->param.b_cabac && h->sps->i_profile_idc < PROFILE_HIGH);
2550     int starting_bits = bs_pos(&h->out.bs);
2551     int b_deblock = h->sh.i_disable_deblocking_filter_idc != 1;
2552     int b_hpel = h->fdec->b_kept_as_ref;
2553     int orig_last_mb = h->sh.i_last_mb;
2554     int thread_last_mb = h->i_threadslice_end * h->mb.i_mb_width - 1;
2555     uint8_t *last_emu_check;
2556 #define BS_BAK_SLICE_MAX_SIZE 0
2557 #define BS_BAK_SLICE_MIN_MBS  1
2558 #define BS_BAK_ROW_VBV        2
2559     x264_bs_bak_t bs_bak[3];
2560     b_deblock &= b_hpel || h->param.b_full_recon || h->param.psz_dump_yuv;
2561     bs_realign( &h->out.bs );
2562
2563     /* Slice */
2564     x264_nal_start( h, h->i_nal_type, h->i_nal_ref_idc );
2565     h->out.nal[h->out.i_nal].i_first_mb = h->sh.i_first_mb;
2566
2567     /* Slice header */
2568     x264_macroblock_thread_init( h );
2569
2570     /* Set the QP equal to the first QP in the slice for more accurate CABAC initialization. */
2571     h->mb.i_mb_xy = h->sh.i_first_mb;
2572     h->sh.i_qp = x264_ratecontrol_mb_qp( h );
2573     h->sh.i_qp = SPEC_QP( h->sh.i_qp );
2574     h->sh.i_qp_delta = h->sh.i_qp - h->pps->i_pic_init_qp;
2575
2576     x264_slice_header_write( &h->out.bs, &h->sh, h->i_nal_ref_idc );
2577     if( h->param.b_cabac )
2578     {
2579         /* alignment needed */
2580         bs_align_1( &h->out.bs );
2581
2582         /* init cabac */
2583         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 );
2584         x264_cabac_encode_init ( &h->cabac, h->out.bs.p, h->out.bs.p_end );
2585         last_emu_check = h->cabac.p;
2586     }
2587     else
2588         last_emu_check = h->out.bs.p;
2589     h->mb.i_last_qp = h->sh.i_qp;
2590     h->mb.i_last_dqp = 0;
2591     h->mb.field_decoding_flag = 0;
2592
2593     i_mb_y = h->sh.i_first_mb / h->mb.i_mb_width;
2594     i_mb_x = h->sh.i_first_mb % h->mb.i_mb_width;
2595     i_skip = 0;
2596
2597     while( 1 )
2598     {
2599         mb_xy = i_mb_x + i_mb_y * h->mb.i_mb_width;
2600         int mb_spos = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
2601
2602         if( i_mb_x == 0 )
2603         {
2604             if( x264_bitstream_check_buffer( h ) )
2605                 return -1;
2606             if( !(i_mb_y & SLICE_MBAFF) && h->param.rc.i_vbv_buffer_size )
2607                 x264_bitstream_backup( h, &bs_bak[BS_BAK_ROW_VBV], i_skip, 1 );
2608             if( !h->mb.b_reencode_mb )
2609                 x264_fdec_filter_row( h, i_mb_y, 0 );
2610         }
2611
2612         if( !(i_mb_y & SLICE_MBAFF) && back_up_bitstream )
2613         {
2614             x264_bitstream_backup( h, &bs_bak[BS_BAK_SLICE_MAX_SIZE], i_skip, 0 );
2615             if( slice_max_size && (thread_last_mb+1-mb_xy) == h->param.i_slice_min_mbs )
2616                 x264_bitstream_backup( h, &bs_bak[BS_BAK_SLICE_MIN_MBS], i_skip, 0 );
2617         }
2618
2619         if( PARAM_INTERLACED )
2620         {
2621             if( h->mb.b_adaptive_mbaff )
2622             {
2623                 if( !(i_mb_y&1) )
2624                 {
2625                     /* FIXME: VSAD is fast but fairly poor at choosing the best interlace type. */
2626                     h->mb.b_interlaced = x264_field_vsad( h, i_mb_x, i_mb_y );
2627                     memcpy( &h->zigzagf, MB_INTERLACED ? &h->zigzagf_interlaced : &h->zigzagf_progressive, sizeof(h->zigzagf) );
2628                     if( !MB_INTERLACED && (i_mb_y+2) == h->mb.i_mb_height )
2629                         x264_expand_border_mbpair( h, i_mb_x, i_mb_y );
2630                 }
2631             }
2632             h->mb.field[mb_xy] = MB_INTERLACED;
2633         }
2634
2635         /* load cache */
2636         if( SLICE_MBAFF )
2637             x264_macroblock_cache_load_interlaced( h, i_mb_x, i_mb_y );
2638         else
2639             x264_macroblock_cache_load_progressive( h, i_mb_x, i_mb_y );
2640
2641         x264_macroblock_analyse( h );
2642
2643         /* encode this macroblock -> be careful it can change the mb type to P_SKIP if needed */
2644 reencode:
2645         x264_macroblock_encode( h );
2646
2647         if( h->param.b_cabac )
2648         {
2649             if( mb_xy > h->sh.i_first_mb && !(SLICE_MBAFF && (i_mb_y&1)) )
2650                 x264_cabac_encode_terminal( &h->cabac );
2651
2652             if( IS_SKIP( h->mb.i_type ) )
2653                 x264_cabac_mb_skip( h, 1 );
2654             else
2655             {
2656                 if( h->sh.i_type != SLICE_TYPE_I )
2657                     x264_cabac_mb_skip( h, 0 );
2658                 x264_macroblock_write_cabac( h, &h->cabac );
2659             }
2660         }
2661         else
2662         {
2663             if( IS_SKIP( h->mb.i_type ) )
2664                 i_skip++;
2665             else
2666             {
2667                 if( h->sh.i_type != SLICE_TYPE_I )
2668                 {
2669                     bs_write_ue( &h->out.bs, i_skip );  /* skip run */
2670                     i_skip = 0;
2671                 }
2672                 x264_macroblock_write_cavlc( h );
2673                 /* If there was a CAVLC level code overflow, try again at a higher QP. */
2674                 if( h->mb.b_overflow )
2675                 {
2676                     h->mb.i_chroma_qp = h->chroma_qp_table[++h->mb.i_qp];
2677                     h->mb.i_skip_intra = 0;
2678                     h->mb.b_skip_mc = 0;
2679                     h->mb.b_overflow = 0;
2680                     x264_bitstream_restore( h, &bs_bak[BS_BAK_SLICE_MAX_SIZE], &i_skip, 0 );
2681                     goto reencode;
2682                 }
2683             }
2684         }
2685
2686         int total_bits = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
2687         int mb_size = total_bits - mb_spos;
2688
2689         if( slice_max_size && (!SLICE_MBAFF || (i_mb_y&1)) )
2690         {
2691             /* Count the skip run, just in case. */
2692             if( !h->param.b_cabac )
2693                 total_bits += bs_size_ue_big( i_skip );
2694             /* Check for escape bytes. */
2695             uint8_t *end = h->param.b_cabac ? h->cabac.p : h->out.bs.p;
2696             for( ; last_emu_check < end - 2; last_emu_check++ )
2697                 if( last_emu_check[0] == 0 && last_emu_check[1] == 0 && last_emu_check[2] <= 3 )
2698                 {
2699                     slice_max_size -= 8;
2700                     last_emu_check++;
2701                 }
2702             /* We'll just re-encode this last macroblock if we go over the max slice size. */
2703             if( total_bits - starting_bits > slice_max_size && !h->mb.b_reencode_mb )
2704             {
2705                 if( !x264_frame_new_slice( h, h->fdec ) )
2706                 {
2707                     /* Handle the most obnoxious slice-min-mbs edge case: we need to end the slice
2708                      * because it's gone over the maximum size, but doing so would violate slice-min-mbs.
2709                      * If possible, roll back to the last checkpoint and try again.
2710                      * We could try raising QP, but that would break in the case where a slice spans multiple
2711                      * rows, which the re-encoding infrastructure can't currently handle. */
2712                     if( mb_xy <= thread_last_mb && (thread_last_mb+1-mb_xy) < h->param.i_slice_min_mbs )
2713                     {
2714                         if( thread_last_mb-h->param.i_slice_min_mbs < h->sh.i_first_mb+h->param.i_slice_min_mbs )
2715                         {
2716                             x264_log( h, X264_LOG_WARNING, "slice-max-size violated (frame %d, cause: slice-min-mbs)\n", h->i_frame );
2717                             slice_max_size = 0;
2718                             goto cont;
2719                         }
2720                         x264_bitstream_restore( h, &bs_bak[BS_BAK_SLICE_MIN_MBS], &i_skip, 0 );
2721                         h->mb.b_reencode_mb = 1;
2722                         h->sh.i_last_mb = thread_last_mb-h->param.i_slice_min_mbs;
2723                         break;
2724                     }
2725                     if( mb_xy-SLICE_MBAFF*h->mb.i_mb_stride != h->sh.i_first_mb )
2726                     {
2727                         x264_bitstream_restore( h, &bs_bak[BS_BAK_SLICE_MAX_SIZE], &i_skip, 0 );
2728                         h->mb.b_reencode_mb = 1;
2729                         if( SLICE_MBAFF )
2730                         {
2731                             // set to bottom of previous mbpair
2732                             if( i_mb_x )
2733                                 h->sh.i_last_mb = mb_xy-1+h->mb.i_mb_stride*(!(i_mb_y&1));
2734                             else
2735                                 h->sh.i_last_mb = (i_mb_y-2+!(i_mb_y&1))*h->mb.i_mb_stride + h->mb.i_mb_width - 1;
2736                         }
2737                         else
2738                             h->sh.i_last_mb = mb_xy-1;
2739                         break;
2740                     }
2741                     else
2742                         h->sh.i_last_mb = mb_xy;
2743                 }
2744                 else
2745                     slice_max_size = 0;
2746             }
2747         }
2748 cont:
2749         h->mb.b_reencode_mb = 0;
2750
2751 #if HAVE_VISUALIZE
2752         if( h->param.b_visualize )
2753             x264_visualize_mb( h );
2754 #endif
2755
2756         /* save cache */
2757         x264_macroblock_cache_save( h );
2758
2759         if( x264_ratecontrol_mb( h, mb_size ) < 0 )
2760         {
2761             x264_bitstream_restore( h, &bs_bak[BS_BAK_ROW_VBV], &i_skip, 1 );
2762             h->mb.b_reencode_mb = 1;
2763             i_mb_x = 0;
2764             i_mb_y = i_mb_y - SLICE_MBAFF;
2765             h->mb.i_mb_prev_xy = i_mb_y * h->mb.i_mb_stride - 1;
2766             h->sh.i_last_mb = orig_last_mb;
2767             continue;
2768         }
2769
2770         /* accumulate mb stats */
2771         h->stat.frame.i_mb_count[h->mb.i_type]++;
2772
2773         int b_intra = IS_INTRA( h->mb.i_type );
2774         int b_skip = IS_SKIP( h->mb.i_type );
2775         if( h->param.i_log_level >= X264_LOG_INFO || h->param.rc.b_stat_write )
2776         {
2777             if( !b_intra && !b_skip && !IS_DIRECT( h->mb.i_type ) )
2778             {
2779                 if( h->mb.i_partition != D_8x8 )
2780                         h->stat.frame.i_mb_partition[h->mb.i_partition] += 4;
2781                     else
2782                         for( int i = 0; i < 4; i++ )
2783                             h->stat.frame.i_mb_partition[h->mb.i_sub_partition[i]] ++;
2784                 if( h->param.i_frame_reference > 1 )
2785                     for( int i_list = 0; i_list <= (h->sh.i_type == SLICE_TYPE_B); i_list++ )
2786                         for( int i = 0; i < 4; i++ )
2787                         {
2788                             int i_ref = h->mb.cache.ref[i_list][ x264_scan8[4*i] ];
2789                             if( i_ref >= 0 )
2790                                 h->stat.frame.i_mb_count_ref[i_list][i_ref] ++;
2791                         }
2792             }
2793         }
2794
2795         if( h->param.i_log_level >= X264_LOG_INFO )
2796         {
2797             if( h->mb.i_cbp_luma | h->mb.i_cbp_chroma )
2798             {
2799                 if( CHROMA444 )
2800                 {
2801                     for( int i = 0; i < 4; i++ )
2802                         if( h->mb.i_cbp_luma & (1 << i) )
2803                             for( int p = 0; p < 3; p++ )
2804                             {
2805                                 int s8 = i*4+p*16;
2806                                 int nnz8x8 = M16( &h->mb.cache.non_zero_count[x264_scan8[s8]+0] )
2807                                            | M16( &h->mb.cache.non_zero_count[x264_scan8[s8]+8] );
2808                                 h->stat.frame.i_mb_cbp[!b_intra + p*2] += !!nnz8x8;
2809                             }
2810                 }
2811                 else
2812                 {
2813                     int cbpsum = (h->mb.i_cbp_luma&1) + ((h->mb.i_cbp_luma>>1)&1)
2814                                + ((h->mb.i_cbp_luma>>2)&1) + (h->mb.i_cbp_luma>>3);
2815                     h->stat.frame.i_mb_cbp[!b_intra + 0] += cbpsum;
2816                     h->stat.frame.i_mb_cbp[!b_intra + 2] += !!h->mb.i_cbp_chroma;
2817                     h->stat.frame.i_mb_cbp[!b_intra + 4] += h->mb.i_cbp_chroma >> 1;
2818                 }
2819             }
2820             if( h->mb.i_cbp_luma && !b_intra )
2821             {
2822                 h->stat.frame.i_mb_count_8x8dct[0] ++;
2823                 h->stat.frame.i_mb_count_8x8dct[1] += h->mb.b_transform_8x8;
2824             }
2825             if( b_intra && h->mb.i_type != I_PCM )
2826             {
2827                 if( h->mb.i_type == I_16x16 )
2828                     h->stat.frame.i_mb_pred_mode[0][h->mb.i_intra16x16_pred_mode]++;
2829                 else if( h->mb.i_type == I_8x8 )
2830                     for( int i = 0; i < 16; i += 4 )
2831                         h->stat.frame.i_mb_pred_mode[1][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
2832                 else //if( h->mb.i_type == I_4x4 )
2833                     for( int i = 0; i < 16; i++ )
2834                         h->stat.frame.i_mb_pred_mode[2][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
2835                 h->stat.frame.i_mb_pred_mode[3][x264_mb_chroma_pred_mode_fix[h->mb.i_chroma_pred_mode]]++;
2836             }
2837             h->stat.frame.i_mb_field[b_intra?0:b_skip?2:1] += MB_INTERLACED;
2838         }
2839
2840         /* calculate deblock strength values (actual deblocking is done per-row along with hpel) */
2841         if( b_deblock )
2842             x264_macroblock_deblock_strength( h );
2843
2844         if( mb_xy == h->sh.i_last_mb )
2845             break;
2846
2847         if( SLICE_MBAFF )
2848         {
2849             i_mb_x += i_mb_y & 1;
2850             i_mb_y ^= i_mb_x < h->mb.i_mb_width;
2851         }
2852         else
2853             i_mb_x++;
2854         if( i_mb_x == h->mb.i_mb_width )
2855         {
2856             i_mb_y++;
2857             i_mb_x = 0;
2858         }
2859     }
2860     if( h->sh.i_last_mb < h->sh.i_first_mb )
2861         return 0;
2862
2863     h->out.nal[h->out.i_nal].i_last_mb = h->sh.i_last_mb;
2864
2865     if( h->param.b_cabac )
2866     {
2867         x264_cabac_encode_flush( h, &h->cabac );
2868         h->out.bs.p = h->cabac.p;
2869     }
2870     else
2871     {
2872         if( i_skip > 0 )
2873             bs_write_ue( &h->out.bs, i_skip );  /* last skip run */
2874         /* rbsp_slice_trailing_bits */
2875         bs_rbsp_trailing( &h->out.bs );
2876         bs_flush( &h->out.bs );
2877     }
2878     if( x264_nal_end( h ) )
2879         return -1;
2880
2881     if( h->sh.i_last_mb == (h->i_threadslice_end * h->mb.i_mb_width - 1) )
2882     {
2883         h->stat.frame.i_misc_bits = bs_pos( &h->out.bs )
2884                                   + (h->out.i_nal*NALU_OVERHEAD * 8)
2885                                   - h->stat.frame.i_tex_bits
2886                                   - h->stat.frame.i_mv_bits;
2887         x264_fdec_filter_row( h, h->i_threadslice_end, 0 );
2888
2889         if( h->param.b_sliced_threads )
2890         {
2891             /* Tell the main thread we're done. */
2892             x264_threadslice_cond_broadcast( h, 1 );
2893             /* Do hpel now */
2894             for( int mb_y = h->i_threadslice_start; mb_y <= h->i_threadslice_end; mb_y++ )
2895                 x264_fdec_filter_row( h, mb_y, 1 );
2896             x264_threadslice_cond_broadcast( h, 2 );
2897             /* Do the first row of hpel, now that the previous slice is done */
2898             if( h->i_thread_idx > 0 )
2899             {
2900                 x264_threadslice_cond_wait( h->thread[h->i_thread_idx-1], 2 );
2901                 x264_fdec_filter_row( h, h->i_threadslice_start + (1 << SLICE_MBAFF), 2 );
2902             }
2903         }
2904
2905         /* Free mb info after the last thread's done using it */
2906         if( h->fdec->mb_info_free && (!h->param.b_sliced_threads || h->i_thread_idx == (h->param.i_threads-1)) )
2907         {
2908             h->fdec->mb_info_free( h->fdec->mb_info );
2909             h->fdec->mb_info = NULL;
2910             h->fdec->mb_info_free = NULL;
2911         }
2912     }
2913
2914     return 0;
2915 }
2916
2917 static void x264_thread_sync_context( x264_t *dst, x264_t *src )
2918 {
2919     if( dst == src )
2920         return;
2921
2922     // reference counting
2923     for( x264_frame_t **f = src->frames.reference; *f; f++ )
2924         (*f)->i_reference_count++;
2925     for( x264_frame_t **f = dst->frames.reference; *f; f++ )
2926         x264_frame_push_unused( src, *f );
2927     src->fdec->i_reference_count++;
2928     x264_frame_push_unused( src, dst->fdec );
2929
2930     // copy everything except the per-thread pointers and the constants.
2931     memcpy( &dst->i_frame, &src->i_frame, offsetof(x264_t, mb.base) - offsetof(x264_t, i_frame) );
2932     dst->param = src->param;
2933     dst->stat = src->stat;
2934     dst->pixf = src->pixf;
2935     dst->reconfig = src->reconfig;
2936 }
2937
2938 static void x264_thread_sync_stat( x264_t *dst, x264_t *src )
2939 {
2940     if( dst == src )
2941         return;
2942     memcpy( &dst->stat.i_frame_count, &src->stat.i_frame_count, sizeof(dst->stat) - sizeof(dst->stat.frame) );
2943 }
2944
2945 static void *x264_slices_write( x264_t *h )
2946 {
2947     int i_slice_num = 0;
2948     int last_thread_mb = h->sh.i_last_mb;
2949
2950 #if HAVE_VISUALIZE
2951     if( h->param.b_visualize )
2952         if( x264_visualize_init( h ) )
2953             goto fail;
2954 #endif
2955
2956     /* init stats */
2957     memset( &h->stat.frame, 0, sizeof(h->stat.frame) );
2958     h->mb.b_reencode_mb = 0;
2959     while( h->sh.i_first_mb + SLICE_MBAFF*h->mb.i_mb_stride <= last_thread_mb )
2960     {
2961         h->sh.i_last_mb = last_thread_mb;
2962         if( !i_slice_num || !x264_frame_new_slice( h, h->fdec ) )
2963         {
2964             if( h->param.i_slice_max_mbs )
2965             {
2966                 if( SLICE_MBAFF )
2967                 {
2968                     // convert first to mbaff form, add slice-max-mbs, then convert back to normal form
2969                     int last_mbaff = 2*(h->sh.i_first_mb % h->mb.i_mb_width)
2970                         + h->mb.i_mb_width*(h->sh.i_first_mb / h->mb.i_mb_width)
2971                         + h->param.i_slice_max_mbs - 1;
2972                     int last_x = (last_mbaff % (2*h->mb.i_mb_width))/2;
2973                     int last_y = (last_mbaff / (2*h->mb.i_mb_width))*2 + 1;
2974                     h->sh.i_last_mb = last_x + h->mb.i_mb_stride*last_y;
2975                 }
2976                 else
2977                 {
2978                     h->sh.i_last_mb = h->sh.i_first_mb + h->param.i_slice_max_mbs - 1;
2979                     if( h->sh.i_last_mb < last_thread_mb && last_thread_mb - h->sh.i_last_mb < h->param.i_slice_min_mbs )
2980                         h->sh.i_last_mb = last_thread_mb - h->param.i_slice_min_mbs;
2981                 }
2982                 i_slice_num++;
2983             }
2984             else if( h->param.i_slice_count && !h->param.b_sliced_threads )
2985             {
2986                 int height = h->mb.i_mb_height >> PARAM_INTERLACED;
2987                 int width = h->mb.i_mb_width << PARAM_INTERLACED;
2988                 i_slice_num++;
2989                 h->sh.i_last_mb = (height * i_slice_num + h->param.i_slice_count/2) / h->param.i_slice_count * width - 1;
2990             }
2991         }
2992         h->sh.i_last_mb = X264_MIN( h->sh.i_last_mb, last_thread_mb );
2993         if( x264_stack_align( x264_slice_write, h ) )
2994             goto fail;
2995         h->sh.i_first_mb = h->sh.i_last_mb + 1;
2996         // if i_first_mb is not the last mb in a row then go to the next mb in MBAFF order
2997         if( SLICE_MBAFF && h->sh.i_first_mb % h->mb.i_mb_width )
2998             h->sh.i_first_mb -= h->mb.i_mb_stride;
2999     }
3000
3001 #if HAVE_VISUALIZE
3002     if( h->param.b_visualize )
3003     {
3004         x264_visualize_show( h );
3005         x264_visualize_close( h );
3006     }
3007 #endif
3008
3009     return (void *)0;
3010
3011 fail:
3012     /* Tell other threads we're done, so they wouldn't wait for it */
3013     if( h->param.b_sliced_threads )
3014         x264_threadslice_cond_broadcast( h, 2 );
3015     return (void *)-1;
3016 }
3017
3018 static int x264_threaded_slices_write( x264_t *h )
3019 {
3020     /* set first/last mb and sync contexts */
3021     for( int i = 0; i < h->param.i_threads; i++ )
3022     {
3023         x264_t *t = h->thread[i];
3024         if( i )
3025         {
3026             t->param = h->param;
3027             memcpy( &t->i_frame, &h->i_frame, offsetof(x264_t, rc) - offsetof(x264_t, i_frame) );
3028         }
3029         int height = h->mb.i_mb_height >> PARAM_INTERLACED;
3030         t->i_threadslice_start = ((height *  i    + h->param.i_slice_count/2) / h->param.i_threads) << PARAM_INTERLACED;
3031         t->i_threadslice_end   = ((height * (i+1) + h->param.i_slice_count/2) / h->param.i_threads) << PARAM_INTERLACED;
3032         t->sh.i_first_mb = t->i_threadslice_start * h->mb.i_mb_width;
3033         t->sh.i_last_mb  =   t->i_threadslice_end * h->mb.i_mb_width - 1;
3034     }
3035
3036     x264_stack_align( x264_analyse_weight_frame, h, h->mb.i_mb_height*16 + 16 );
3037
3038     x264_threads_distribute_ratecontrol( h );
3039
3040     /* setup */
3041     for( int i = 0; i < h->param.i_threads; i++ )
3042     {
3043         h->thread[i]->i_thread_idx = i;
3044         h->thread[i]->b_thread_active = 1;
3045         x264_threadslice_cond_broadcast( h->thread[i], 0 );
3046     }
3047     /* dispatch */
3048     for( int i = 0; i < h->param.i_threads; i++ )
3049         x264_threadpool_run( h->threadpool, (void*)x264_slices_write, h->thread[i] );
3050     /* wait */
3051     for( int i = 0; i < h->param.i_threads; i++ )
3052         x264_threadslice_cond_wait( h->thread[i], 1 );
3053
3054     x264_threads_merge_ratecontrol( h );
3055
3056     for( int i = 1; i < h->param.i_threads; i++ )
3057     {
3058         x264_t *t = h->thread[i];
3059         for( int j = 0; j < t->out.i_nal; j++ )
3060         {
3061             h->out.nal[h->out.i_nal] = t->out.nal[j];
3062             h->out.i_nal++;
3063             x264_nal_check_buffer( h );
3064         }
3065         /* All entries in stat.frame are ints except for ssd/ssim. */
3066         for( int j = 0; j < (offsetof(x264_t,stat.frame.i_ssd) - offsetof(x264_t,stat.frame.i_mv_bits)) / sizeof(int); j++ )
3067             ((int*)&h->stat.frame)[j] += ((int*)&t->stat.frame)[j];
3068         for( int j = 0; j < 3; j++ )
3069             h->stat.frame.i_ssd[j] += t->stat.frame.i_ssd[j];
3070         h->stat.frame.f_ssim += t->stat.frame.f_ssim;
3071         h->stat.frame.i_ssim_cnt += t->stat.frame.i_ssim_cnt;
3072     }
3073
3074     return 0;
3075 }
3076
3077 void x264_encoder_intra_refresh( x264_t *h )
3078 {
3079     h = h->thread[h->i_thread_phase];
3080     h->b_queued_intra_refresh = 1;
3081 }
3082
3083 int x264_encoder_invalidate_reference( x264_t *h, int64_t pts )
3084 {
3085     if( h->param.i_bframe )
3086     {
3087         x264_log( h, X264_LOG_ERROR, "x264_encoder_invalidate_reference is not supported with B-frames enabled\n" );
3088         return -1;
3089     }
3090     if( h->param.b_intra_refresh )
3091     {
3092         x264_log( h, X264_LOG_ERROR, "x264_encoder_invalidate_reference is not supported with intra refresh enabled\n" );
3093         return -1;
3094     }
3095     h = h->thread[h->i_thread_phase];
3096     if( pts >= h->i_last_idr_pts )
3097     {
3098         for( int i = 0; h->frames.reference[i]; i++ )
3099             if( pts <= h->frames.reference[i]->i_pts )
3100                 h->frames.reference[i]->b_corrupt = 1;
3101         if( pts <= h->fdec->i_pts )
3102             h->fdec->b_corrupt = 1;
3103     }
3104     return 0;
3105 }
3106
3107 /****************************************************************************
3108  * x264_encoder_encode:
3109  *  XXX: i_poc   : is the poc of the current given picture
3110  *       i_frame : is the number of the frame being coded
3111  *  ex:  type frame poc
3112  *       I      0   2*0
3113  *       P      1   2*3
3114  *       B      2   2*1
3115  *       B      3   2*2
3116  *       P      4   2*6
3117  *       B      5   2*4
3118  *       B      6   2*5
3119  ****************************************************************************/
3120 int     x264_encoder_encode( x264_t *h,
3121                              x264_nal_t **pp_nal, int *pi_nal,
3122                              x264_picture_t *pic_in,
3123                              x264_picture_t *pic_out )
3124 {
3125     x264_t *thread_current, *thread_prev, *thread_oldest;
3126     int i_nal_type, i_nal_ref_idc, i_global_qp;
3127     int overhead = NALU_OVERHEAD;
3128
3129 #if HAVE_OPENCL
3130     if( h->opencl.b_fatal_error )
3131         return -1;
3132 #endif
3133
3134     if( h->i_thread_frames > 1 )
3135     {
3136         thread_prev    = h->thread[ h->i_thread_phase ];
3137         h->i_thread_phase = (h->i_thread_phase + 1) % h->i_thread_frames;
3138         thread_current = h->thread[ h->i_thread_phase ];
3139         thread_oldest  = h->thread[ (h->i_thread_phase + 1) % h->i_thread_frames ];
3140         x264_thread_sync_context( thread_current, thread_prev );
3141         x264_thread_sync_ratecontrol( thread_current, thread_prev, thread_oldest );
3142         h = thread_current;
3143     }
3144     else
3145     {
3146         thread_current =
3147         thread_oldest  = h;
3148     }
3149     h->i_cpb_delay_pir_offset = h->i_cpb_delay_pir_offset_next;
3150
3151     /* no data out */
3152     *pi_nal = 0;
3153     *pp_nal = NULL;
3154
3155     /* ------------------- Setup new frame from picture -------------------- */
3156     if( pic_in != NULL )
3157     {
3158         /* 1: Copy the picture to a frame and move it to a buffer */
3159         x264_frame_t *fenc = x264_frame_pop_unused( h, 0 );
3160         if( !fenc )
3161             return -1;
3162
3163         if( x264_frame_copy_picture( h, fenc, pic_in ) < 0 )
3164             return -1;
3165
3166         if( h->param.i_width != 16 * h->mb.i_mb_width ||
3167             h->param.i_height != 16 * h->mb.i_mb_height )
3168             x264_frame_expand_border_mod16( h, fenc );
3169
3170         fenc->i_frame = h->frames.i_input++;
3171
3172         if( fenc->i_frame == 0 )
3173             h->frames.i_first_pts = fenc->i_pts;
3174         if( h->frames.i_bframe_delay && fenc->i_frame == h->frames.i_bframe_delay )
3175             h->frames.i_bframe_delay_time = fenc->i_pts - h->frames.i_first_pts;
3176
3177         if( h->param.b_vfr_input && fenc->i_pts <= h->frames.i_largest_pts )
3178             x264_log( h, X264_LOG_WARNING, "non-strictly-monotonic PTS\n" );
3179
3180         h->frames.i_second_largest_pts = h->frames.i_largest_pts;
3181         h->frames.i_largest_pts = fenc->i_pts;
3182
3183         if( (fenc->i_pic_struct < PIC_STRUCT_AUTO) || (fenc->i_pic_struct > PIC_STRUCT_TRIPLE) )
3184             fenc->i_pic_struct = PIC_STRUCT_AUTO;
3185
3186         if( fenc->i_pic_struct == PIC_STRUCT_AUTO )
3187         {
3188 #if HAVE_INTERLACED
3189             int b_interlaced = fenc->param ? fenc->param->b_interlaced : h->param.b_interlaced;
3190 #else
3191             int b_interlaced = 0;
3192 #endif
3193             if( b_interlaced )
3194             {
3195                 int b_tff = fenc->param ? fenc->param->b_tff : h->param.b_tff;
3196                 fenc->i_pic_struct = b_tff ? PIC_STRUCT_TOP_BOTTOM : PIC_STRUCT_BOTTOM_TOP;
3197             }
3198             else
3199                 fenc->i_pic_struct = PIC_STRUCT_PROGRESSIVE;
3200         }
3201
3202         if( h->param.rc.b_mb_tree && h->param.rc.b_stat_read )
3203         {
3204             if( x264_macroblock_tree_read( h, fenc, pic_in->prop.quant_offsets ) )
3205                 return -1;
3206         }
3207         else
3208             x264_stack_align( x264_adaptive_quant_frame, h, fenc, pic_in->prop.quant_offsets );
3209
3210         if( pic_in->prop.quant_offsets_free )
3211             pic_in->prop.quant_offsets_free( pic_in->prop.quant_offsets );
3212
3213         if( h->frames.b_have_lowres )
3214             x264_frame_init_lowres( h, fenc );
3215
3216         /* 2: Place the frame into the queue for its slice type decision */
3217         x264_lookahead_put_frame( h, fenc );
3218
3219         if( h->frames.i_input <= h->frames.i_delay + 1 - h->i_thread_frames )
3220         {
3221             /* Nothing yet to encode, waiting for filling of buffers */
3222             pic_out->i_type = X264_TYPE_AUTO;
3223             return 0;
3224         }
3225     }
3226     else
3227     {
3228         /* signal kills for lookahead thread */
3229         x264_pthread_mutex_lock( &h->lookahead->ifbuf.mutex );
3230         h->lookahead->b_exit_thread = 1;
3231         x264_pthread_cond_broadcast( &h->lookahead->ifbuf.cv_fill );
3232         x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
3233     }
3234
3235     h->i_frame++;
3236     /* 3: The picture is analyzed in the lookahead */
3237     if( !h->frames.current[0] )
3238         x264_lookahead_get_frames( h );
3239
3240     if( !h->frames.current[0] && x264_lookahead_is_empty( h ) )
3241         return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
3242
3243     /* ------------------- Get frame to be encoded ------------------------- */
3244     /* 4: get picture to encode */
3245     h->fenc = x264_frame_shift( h->frames.current );
3246
3247     /* If applicable, wait for previous frame reconstruction to finish */
3248     if( h->param.b_sliced_threads )
3249         if( x264_threadpool_wait_all( h ) < 0 )
3250             return -1;
3251
3252     if( h->i_frame == h->i_thread_frames - 1 )
3253         h->i_reordered_pts_delay = h->fenc->i_reordered_pts;
3254     if( h->reconfig )
3255     {
3256         x264_encoder_reconfig_apply( h, &h->reconfig_h->param );
3257         h->reconfig = 0;
3258     }
3259     if( h->fenc->param )
3260     {
3261         x264_encoder_reconfig_apply( h, h->fenc->param );
3262         if( h->fenc->param->param_free )
3263         {
3264             h->fenc->param->param_free( h->fenc->param );
3265             h->fenc->param = NULL;
3266         }
3267     }
3268
3269     // ok to call this before encoding any frames, since the initial values of fdec have b_kept_as_ref=0
3270     if( x264_reference_update( h ) )
3271         return -1;
3272     h->fdec->i_lines_completed = -1;
3273
3274     if( !IS_X264_TYPE_I( h->fenc->i_type ) )
3275     {
3276         int valid_refs_left = 0;
3277         for( int i = 0; h->frames.reference[i]; i++ )
3278             if( !h->frames.reference[i]->b_corrupt )
3279                 valid_refs_left++;
3280         /* No valid reference frames left: force an IDR. */
3281         if( !valid_refs_left )
3282         {
3283             h->fenc->b_keyframe = 1;
3284             h->fenc->i_type = X264_TYPE_IDR;
3285         }
3286     }
3287
3288     if( h->fenc->b_keyframe )
3289     {
3290         h->frames.i_last_keyframe = h->fenc->i_frame;
3291         if( h->fenc->i_type == X264_TYPE_IDR )
3292         {
3293             h->i_frame_num = 0;
3294             h->frames.i_last_idr = h->fenc->i_frame;
3295         }
3296     }
3297     h->sh.i_mmco_command_count =
3298     h->sh.i_mmco_remove_from_end = 0;
3299     h->b_ref_reorder[0] =
3300     h->b_ref_reorder[1] = 0;
3301     h->fdec->i_poc =
3302     h->fenc->i_poc = 2 * ( h->fenc->i_frame - X264_MAX( h->frames.i_last_idr, 0 ) );
3303
3304     /* ------------------- Setup frame context ----------------------------- */
3305     /* 5: Init data dependent of frame type */
3306     if( h->fenc->i_type == X264_TYPE_IDR )
3307     {
3308         /* reset ref pictures */
3309         i_nal_type    = NAL_SLICE_IDR;
3310         i_nal_ref_idc = NAL_PRIORITY_HIGHEST;
3311         h->sh.i_type = SLICE_TYPE_I;
3312         x264_reference_reset( h );
3313         h->frames.i_poc_last_open_gop = -1;
3314     }
3315     else if( h->fenc->i_type == X264_TYPE_I )
3316     {
3317         i_nal_type    = NAL_SLICE;
3318         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
3319         h->sh.i_type = SLICE_TYPE_I;
3320         x264_reference_hierarchy_reset( h );
3321         if( h->param.b_open_gop )
3322             h->frames.i_poc_last_open_gop = h->fenc->b_keyframe ? h->fenc->i_poc : -1;
3323     }
3324     else if( h->fenc->i_type == X264_TYPE_P )
3325     {
3326         i_nal_type    = NAL_SLICE;
3327         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
3328         h->sh.i_type = SLICE_TYPE_P;
3329         x264_reference_hierarchy_reset( h );
3330         h->frames.i_poc_last_open_gop = -1;
3331     }
3332     else if( h->fenc->i_type == X264_TYPE_BREF )
3333     {
3334         i_nal_type    = NAL_SLICE;
3335         i_nal_ref_idc = h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT ? NAL_PRIORITY_LOW : NAL_PRIORITY_HIGH;
3336         h->sh.i_type = SLICE_TYPE_B;
3337         x264_reference_hierarchy_reset( h );
3338     }
3339     else    /* B frame */
3340     {
3341         i_nal_type    = NAL_SLICE;
3342         i_nal_ref_idc = NAL_PRIORITY_DISPOSABLE;
3343         h->sh.i_type = SLICE_TYPE_B;
3344     }
3345
3346     h->fdec->i_type = h->fenc->i_type;
3347     h->fdec->i_frame = h->fenc->i_frame;
3348     h->fenc->b_kept_as_ref =
3349     h->fdec->b_kept_as_ref = i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE && h->param.i_keyint_max > 1;
3350
3351     h->fdec->mb_info = h->fenc->mb_info;
3352     h->fdec->mb_info_free = h->fenc->mb_info_free;
3353     h->fenc->mb_info = NULL;
3354     h->fenc->mb_info_free = NULL;
3355
3356     h->fdec->i_pts = h->fenc->i_pts;
3357     if( h->frames.i_bframe_delay )
3358     {
3359         int64_t *prev_reordered_pts = thread_current->frames.i_prev_reordered_pts;
3360         h->fdec->i_dts = h->i_frame > h->frames.i_bframe_delay
3361                        ? prev_reordered_pts[ (h->i_frame - h->frames.i_bframe_delay) % h->frames.i_bframe_delay ]
3362                        : h->fenc->i_reordered_pts - h->frames.i_bframe_delay_time;
3363         prev_reordered_pts[ h->i_frame % h->frames.i_bframe_delay ] = h->fenc->i_reordered_pts;
3364     }
3365     else
3366         h->fdec->i_dts = h->fenc->i_reordered_pts;
3367     if( h->fenc->i_type == X264_TYPE_IDR )
3368         h->i_last_idr_pts = h->fdec->i_pts;
3369
3370     /* ------------------- Init                ----------------------------- */
3371     /* build ref list 0/1 */
3372     x264_reference_build_list( h, h->fdec->i_poc );
3373
3374     /* ---------------------- Write the bitstream -------------------------- */
3375     /* Init bitstream context */
3376     if( h->param.b_sliced_threads )
3377     {
3378         for( int i = 0; i < h->param.i_threads; i++ )
3379         {
3380             bs_init( &h->thread[i]->out.bs, h->thread[i]->out.p_bitstream, h->thread[i]->out.i_bitstream );
3381             h->thread[i]->out.i_nal = 0;
3382         }
3383     }
3384     else
3385     {
3386         bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
3387         h->out.i_nal = 0;
3388     }
3389
3390     if( h->param.b_aud )
3391     {
3392         int pic_type;
3393
3394         if( h->sh.i_type == SLICE_TYPE_I )
3395             pic_type = 0;
3396         else if( h->sh.i_type == SLICE_TYPE_P )
3397             pic_type = 1;
3398         else if( h->sh.i_type == SLICE_TYPE_B )
3399             pic_type = 2;
3400         else
3401             pic_type = 7;
3402
3403         x264_nal_start( h, NAL_AUD, NAL_PRIORITY_DISPOSABLE );
3404         bs_write( &h->out.bs, 3, pic_type );
3405         bs_rbsp_trailing( &h->out.bs );
3406         if( x264_nal_end( h ) )
3407             return -1;
3408         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
3409     }
3410
3411     h->i_nal_type = i_nal_type;
3412     h->i_nal_ref_idc = i_nal_ref_idc;
3413
3414     if( h->param.b_intra_refresh )
3415     {
3416         if( IS_X264_TYPE_I( h->fenc->i_type ) )
3417         {
3418             h->fdec->i_frames_since_pir = 0;
3419             h->b_queued_intra_refresh = 0;
3420             /* PIR is currently only supported with ref == 1, so any intra frame effectively refreshes
3421              * the whole frame and counts as an intra refresh. */
3422             h->fdec->f_pir_position = h->mb.i_mb_width;
3423         }
3424         else if( h->fenc->i_type == X264_TYPE_P )
3425         {
3426             int pocdiff = (h->fdec->i_poc - h->fref[0][0]->i_poc)/2;
3427             float increment = X264_MAX( ((float)h->mb.i_mb_width-1) / h->param.i_keyint_max, 1 );
3428             h->fdec->f_pir_position = h->fref[0][0]->f_pir_position;
3429             h->fdec->i_frames_since_pir = h->fref[0][0]->i_frames_since_pir + pocdiff;
3430             if( h->fdec->i_frames_since_pir >= h->param.i_keyint_max ||
3431                 (h->b_queued_intra_refresh && h->fdec->f_pir_position + 0.5 >= h->mb.i_mb_width) )
3432             {
3433                 h->fdec->f_pir_position = 0;
3434                 h->fdec->i_frames_since_pir = 0;
3435                 h->b_queued_intra_refresh = 0;
3436                 h->fenc->b_keyframe = 1;
3437             }
3438             h->fdec->i_pir_start_col = h->fdec->f_pir_position+0.5;
3439             h->fdec->f_pir_position += increment * pocdiff;
3440             h->fdec->i_pir_end_col = h->fdec->f_pir_position+0.5;
3441             /* If our intra refresh has reached the right side of the frame, we're done. */
3442             if( h->fdec->i_pir_end_col >= h->mb.i_mb_width - 1 )
3443             {
3444                 h->fdec->f_pir_position = h->mb.i_mb_width;
3445                 h->fdec->i_pir_end_col = h->mb.i_mb_width - 1;
3446             }
3447         }
3448     }
3449
3450     if( h->fenc->b_keyframe )
3451     {
3452         /* Write SPS and PPS */
3453         if( h->param.b_repeat_headers )
3454         {
3455             /* generate sequence parameters */
3456             x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
3457             x264_sps_write( &h->out.bs, h->sps );
3458             if( x264_nal_end( h ) )
3459                 return -1;
3460             /* Pad AUD/SPS to 256 bytes like Panasonic */
3461             if( h->param.b_avcintra_compat )
3462                 h->out.nal[h->out.i_nal-1].i_padding = 256 - bs_pos( &h->out.bs ) / 8 - 2*NALU_OVERHEAD;
3463             overhead += h->out.nal[h->out.i_nal-1].i_payload + h->out.nal[h->out.i_nal-1].i_padding + NALU_OVERHEAD;
3464
3465             /* generate picture parameters */
3466             x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
3467             x264_pps_write( &h->out.bs, h->sps, h->pps );
3468             if( x264_nal_end( h ) )
3469                 return -1;
3470             if( h->param.b_avcintra_compat )
3471                 h->out.nal[h->out.i_nal-1].i_padding = 256 - h->out.nal[h->out.i_nal-1].i_payload - NALU_OVERHEAD;
3472             overhead += h->out.nal[h->out.i_nal-1].i_payload + h->out.nal[h->out.i_nal-1].i_padding + NALU_OVERHEAD;
3473         }
3474
3475         /* when frame threading is used, buffering period sei is written in x264_encoder_frame_end */
3476         if( h->i_thread_frames == 1 && h->sps->vui.b_nal_hrd_parameters_present )
3477         {
3478             x264_hrd_fullness( h );
3479             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3480             x264_sei_buffering_period_write( h, &h->out.bs );
3481             if( x264_nal_end( h ) )
3482                return -1;
3483             overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3484         }
3485     }
3486
3487     /* write extra sei */
3488     for( int i = 0; i < h->fenc->extra_sei.num_payloads; i++ )
3489     {
3490         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3491         x264_sei_write( &h->out.bs, h->fenc->extra_sei.payloads[i].payload, h->fenc->extra_sei.payloads[i].payload_size,
3492                         h->fenc->extra_sei.payloads[i].payload_type );
3493         if( x264_nal_end( h ) )
3494             return -1;
3495         overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3496         if( h->fenc->extra_sei.sei_free )
3497         {
3498             h->fenc->extra_sei.sei_free( h->fenc->extra_sei.payloads[i].payload );
3499             h->fenc->extra_sei.payloads[i].payload = NULL;
3500         }
3501     }
3502
3503     if( h->fenc->extra_sei.sei_free )
3504     {
3505         h->fenc->extra_sei.sei_free( h->fenc->extra_sei.payloads );
3506         h->fenc->extra_sei.payloads = NULL;
3507         h->fenc->extra_sei.sei_free = NULL;
3508     }
3509
3510     if( h->fenc->b_keyframe )
3511     {
3512         /* Avid's decoder strictly wants two SEIs for AVC-Intra so we can't insert the x264 SEI */
3513         if( h->param.b_repeat_headers && h->fenc->i_frame == 0 && !h->param.b_avcintra_compat )
3514         {
3515             /* identify ourself */
3516             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3517             if( x264_sei_version_write( h, &h->out.bs ) )
3518                 return -1;
3519             if( x264_nal_end( h ) )
3520                 return -1;
3521             overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3522         }
3523
3524         if( h->fenc->i_type != X264_TYPE_IDR )
3525         {
3526             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;
3527             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3528             x264_sei_recovery_point_write( h, &h->out.bs, time_to_recovery );
3529             if( x264_nal_end( h ) )
3530                 return -1;
3531             overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3532         }
3533
3534         if( h->param.i_frame_packing >= 0 )
3535         {
3536             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3537             x264_sei_frame_packing_write( h, &h->out.bs );
3538             if( x264_nal_end( h ) )
3539                 return -1;
3540             overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3541         }
3542     }
3543
3544     /* generate sei pic timing */
3545     if( h->sps->vui.b_pic_struct_present || h->sps->vui.b_nal_hrd_parameters_present )
3546     {
3547         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3548         x264_sei_pic_timing_write( h, &h->out.bs );
3549         if( x264_nal_end( h ) )
3550             return -1;
3551         overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3552     }
3553
3554     /* As required by Blu-ray. */
3555     if( !IS_X264_TYPE_B( h->fenc->i_type ) && h->b_sh_backup )
3556     {
3557         h->b_sh_backup = 0;
3558         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3559         x264_sei_dec_ref_pic_marking_write( h, &h->out.bs );
3560         if( x264_nal_end( h ) )
3561             return -1;
3562         overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3563     }
3564
3565     if( h->fenc->b_keyframe && h->param.b_intra_refresh )
3566         h->i_cpb_delay_pir_offset_next = h->fenc->i_cpb_delay;
3567
3568     /* Filler space: 10 or 18 SEIs' worth of space, depending on resolution */
3569     if( h->param.b_avcintra_compat )
3570     {
3571         /* Write an empty filler NAL to mimic the AUD in the P2 format*/
3572         if( h->param.b_avcintra_compat )
3573         {
3574             x264_nal_start( h, NAL_FILLER, NAL_PRIORITY_DISPOSABLE );
3575             x264_filler_write( h, &h->out.bs, 0 );
3576             if( x264_nal_end( h ) )
3577                 return -1;
3578             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
3579         }
3580
3581         /* All lengths are magic lengths that decoders expect to see */
3582         /* "UMID" SEI */
3583         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3584         if( x264_sei_avcintra_write( h, &h->out.bs, 497, "UMID" ) < 0 )
3585             return -1;
3586         if( x264_nal_end( h ) )
3587             return -1;
3588         overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3589
3590
3591         int unpadded_len;
3592         int total_len;
3593         if( h->param.i_height == 1080 )
3594         {
3595             unpadded_len = 5780;
3596             total_len = 17*512;
3597         }
3598         else
3599         {
3600             unpadded_len = 2900;
3601             total_len = 9*512;
3602         }
3603         /* "VANC" SEI */
3604         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3605         if( x264_sei_avcintra_write( h, &h->out.bs, unpadded_len, "VANC" ) < 0 )
3606             return -1;
3607         if( x264_nal_end( h ) )
3608             return -1;
3609
3610         h->out.nal[h->out.i_nal-1].i_padding = total_len - h->out.nal[h->out.i_nal-1].i_payload - SEI_OVERHEAD;
3611         overhead += h->out.nal[h->out.i_nal-1].i_payload + h->out.nal[h->out.i_nal-1].i_padding + SEI_OVERHEAD;
3612     }
3613
3614     /* Init the rate control */
3615     /* FIXME: Include slice header bit cost. */
3616     x264_ratecontrol_start( h, h->fenc->i_qpplus1, overhead*8 );
3617     i_global_qp = x264_ratecontrol_qp( h );
3618
3619     pic_out->i_qpplus1 =
3620     h->fdec->i_qpplus1 = i_global_qp + 1;
3621
3622     if( h->param.rc.b_stat_read && h->sh.i_type != SLICE_TYPE_I )
3623     {
3624         x264_reference_build_list_optimal( h );
3625         x264_reference_check_reorder( h );
3626     }
3627
3628     if( h->i_ref[0] )
3629         h->fdec->i_poc_l0ref0 = h->fref[0][0]->i_poc;
3630
3631     /* ------------------------ Create slice header  ----------------------- */
3632     x264_slice_init( h, i_nal_type, i_global_qp );
3633
3634     /*------------------------- Weights -------------------------------------*/
3635     if( h->sh.i_type == SLICE_TYPE_B )
3636         x264_macroblock_bipred_init( h );
3637
3638     x264_weighted_pred_init( h );
3639
3640     if( i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE )
3641         h->i_frame_num++;
3642
3643     /* Write frame */
3644     h->i_threadslice_start = 0;
3645     h->i_threadslice_end = h->mb.i_mb_height;
3646     if( h->i_thread_frames > 1 )
3647     {
3648         x264_threadpool_run( h->threadpool, (void*)x264_slices_write, h );
3649         h->b_thread_active = 1;
3650     }
3651     else if( h->param.b_sliced_threads )
3652     {
3653         if( x264_threaded_slices_write( h ) )
3654             return -1;
3655     }
3656     else
3657         if( (intptr_t)x264_slices_write( h ) )
3658             return -1;
3659
3660     return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
3661 }
3662
3663 static int x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
3664                                    x264_nal_t **pp_nal, int *pi_nal,
3665                                    x264_picture_t *pic_out )
3666 {
3667     char psz_message[80];
3668
3669     if( !h->param.b_sliced_threads && h->b_thread_active )
3670     {
3671         h->b_thread_active = 0;
3672         if( (intptr_t)x264_threadpool_wait( h->threadpool, h ) )
3673             return -1;
3674     }
3675     if( !h->out.i_nal )
3676     {
3677         pic_out->i_type = X264_TYPE_AUTO;
3678         return 0;
3679     }
3680
3681     x264_emms();
3682
3683     /* generate buffering period sei and insert it into place */
3684     if( h->i_thread_frames > 1 && h->fenc->b_keyframe && h->sps->vui.b_nal_hrd_parameters_present )
3685     {
3686         x264_hrd_fullness( h );
3687         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3688         x264_sei_buffering_period_write( h, &h->out.bs );
3689         if( x264_nal_end( h ) )
3690            return -1;
3691         /* buffering period sei must follow AUD, SPS and PPS and precede all other SEIs */
3692         int idx = 0;
3693         while( h->out.nal[idx].i_type == NAL_AUD ||
3694                h->out.nal[idx].i_type == NAL_SPS ||
3695                h->out.nal[idx].i_type == NAL_PPS )
3696             idx++;
3697         x264_nal_t nal_tmp = h->out.nal[h->out.i_nal-1];
3698         memmove( &h->out.nal[idx+1], &h->out.nal[idx], (h->out.i_nal-idx-1)*sizeof(x264_nal_t) );
3699         h->out.nal[idx] = nal_tmp;
3700     }
3701
3702     int frame_size = x264_encoder_encapsulate_nals( h, 0 );
3703     if( frame_size < 0 )
3704         return -1;
3705
3706     /* Set output picture properties */
3707     pic_out->i_type = h->fenc->i_type;
3708
3709     pic_out->b_keyframe = h->fenc->b_keyframe;
3710     pic_out->i_pic_struct = h->fenc->i_pic_struct;
3711
3712     pic_out->i_pts = h->fdec->i_pts;
3713     pic_out->i_dts = h->fdec->i_dts;
3714
3715     if( pic_out->i_pts < pic_out->i_dts )
3716         x264_log( h, X264_LOG_WARNING, "invalid DTS: PTS is less than DTS\n" );
3717
3718     pic_out->opaque = h->fenc->opaque;
3719
3720     pic_out->img.i_csp = h->fdec->i_csp;
3721 #if HIGH_BIT_DEPTH
3722     pic_out->img.i_csp |= X264_CSP_HIGH_DEPTH;
3723 #endif
3724     pic_out->img.i_plane = h->fdec->i_plane;
3725     for( int i = 0; i < pic_out->img.i_plane; i++ )
3726     {
3727         pic_out->img.i_stride[i] = h->fdec->i_stride[i] * sizeof(pixel);
3728         pic_out->img.plane[i] = (uint8_t*)h->fdec->plane[i];
3729     }
3730
3731     x264_frame_push_unused( thread_current, h->fenc );
3732
3733     /* ---------------------- Update encoder state ------------------------- */
3734
3735     /* update rc */
3736     int filler = 0;
3737     if( x264_ratecontrol_end( h, frame_size * 8, &filler ) < 0 )
3738         return -1;
3739
3740     pic_out->hrd_timing = h->fenc->hrd_timing;
3741     pic_out->prop.f_crf_avg = h->fdec->f_crf_avg;
3742
3743     /* Filler in AVC-Intra mode is written as zero bytes to the last slice
3744      * We don't know the size of the last slice until encapsulation so we add filler to the encapsulated NAL */
3745     if( h->param.b_avcintra_compat )
3746     {
3747         x264_t *h0 = h->thread[0];
3748         int ret = x264_check_encapsulated_buffer( h, h0, h->out.i_nal, frame_size, frame_size + filler );
3749         if( ret < 0 )
3750             return -1;
3751         memset( h->out.nal[0].p_payload + frame_size, 0, filler );
3752         h->out.nal[h->out.i_nal-1].i_payload += filler;
3753         h->out.nal[h->out.i_nal-1].i_padding = filler;
3754         frame_size += filler;
3755     }
3756     else
3757     {
3758         while( filler > 0 )
3759         {
3760             int f, overhead;
3761             overhead = (FILLER_OVERHEAD - h->param.b_annexb);
3762             if( h->param.i_slice_max_size && filler > h->param.i_slice_max_size )
3763             {
3764                 int next_size = filler - h->param.i_slice_max_size;
3765                 int overflow = X264_MAX( overhead - next_size, 0 );
3766                 f = h->param.i_slice_max_size - overhead - overflow;
3767             }
3768             else
3769                 f = X264_MAX( 0, filler - overhead );
3770
3771             if( x264_bitstream_check_buffer_filler( h, f ) )
3772                 return -1;
3773             x264_nal_start( h, NAL_FILLER, NAL_PRIORITY_DISPOSABLE );
3774             x264_filler_write( h, &h->out.bs, f );
3775             if( x264_nal_end( h ) )
3776                 return -1;
3777             int total_size = x264_encoder_encapsulate_nals( h, h->out.i_nal-1 );
3778             if( total_size < 0 )
3779                 return -1;
3780             frame_size += total_size;
3781             filler -= total_size;
3782         }
3783     }
3784
3785     /* End bitstream, set output  */
3786     *pi_nal = h->out.i_nal;
3787     *pp_nal = h->out.nal;
3788
3789     h->out.i_nal = 0;
3790
3791     x264_noise_reduction_update( h );
3792
3793     /* ---------------------- Compute/Print statistics --------------------- */
3794     x264_thread_sync_stat( h, h->thread[0] );
3795
3796     /* Slice stat */
3797     h->stat.i_frame_count[h->sh.i_type]++;
3798     h->stat.i_frame_size[h->sh.i_type] += frame_size;
3799     h->stat.f_frame_qp[h->sh.i_type] += h->fdec->f_qp_avg_aq;
3800
3801     for( int i = 0; i < X264_MBTYPE_MAX; i++ )
3802         h->stat.i_mb_count[h->sh.i_type][i] += h->stat.frame.i_mb_count[i];
3803     for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
3804         h->stat.i_mb_partition[h->sh.i_type][i] += h->stat.frame.i_mb_partition[i];
3805     for( int i = 0; i < 2; i++ )
3806         h->stat.i_mb_count_8x8dct[i] += h->stat.frame.i_mb_count_8x8dct[i];
3807     for( int i = 0; i < 6; i++ )
3808         h->stat.i_mb_cbp[i] += h->stat.frame.i_mb_cbp[i];
3809     for( int i = 0; i < 4; i++ )
3810         for( int j = 0; j < 13; j++ )
3811             h->stat.i_mb_pred_mode[i][j] += h->stat.frame.i_mb_pred_mode[i][j];
3812     if( h->sh.i_type != SLICE_TYPE_I )
3813         for( int i_list = 0; i_list < 2; i_list++ )
3814             for( int i = 0; i < X264_REF_MAX*2; i++ )
3815                 h->stat.i_mb_count_ref[h->sh.i_type][i_list][i] += h->stat.frame.i_mb_count_ref[i_list][i];
3816     for( int i = 0; i < 3; i++ )
3817         h->stat.i_mb_field[i] += h->stat.frame.i_mb_field[i];
3818     if( h->sh.i_type == SLICE_TYPE_P && h->param.analyse.i_weighted_pred >= X264_WEIGHTP_SIMPLE )
3819     {
3820         h->stat.i_wpred[0] += !!h->sh.weight[0][0].weightfn;
3821         h->stat.i_wpred[1] += !!h->sh.weight[0][1].weightfn || !!h->sh.weight[0][2].weightfn;
3822     }
3823     if( h->sh.i_type == SLICE_TYPE_B )
3824     {
3825         h->stat.i_direct_frames[ h->sh.b_direct_spatial_mv_pred ] ++;
3826         if( h->mb.b_direct_auto_write )
3827         {
3828             //FIXME somewhat arbitrary time constants
3829             if( h->stat.i_direct_score[0] + h->stat.i_direct_score[1] > h->mb.i_mb_count )
3830                 for( int i = 0; i < 2; i++ )
3831                     h->stat.i_direct_score[i] = h->stat.i_direct_score[i] * 9/10;
3832             for( int i = 0; i < 2; i++ )
3833                 h->stat.i_direct_score[i] += h->stat.frame.i_direct_score[i];
3834         }
3835     }
3836     else
3837         h->stat.i_consecutive_bframes[h->fenc->i_bframes]++;
3838
3839     psz_message[0] = '\0';
3840     double dur = h->fenc->f_duration;
3841     h->stat.f_frame_duration[h->sh.i_type] += dur;
3842     if( h->param.analyse.b_psnr )
3843     {
3844         int64_t ssd[3] =
3845         {
3846             h->stat.frame.i_ssd[0],
3847             h->stat.frame.i_ssd[1],
3848             h->stat.frame.i_ssd[2],
3849         };
3850         int luma_size = h->param.i_width * h->param.i_height;
3851         int chroma_size = CHROMA_SIZE( luma_size );
3852         pic_out->prop.f_psnr[0] = x264_psnr( ssd[0], luma_size );
3853         pic_out->prop.f_psnr[1] = x264_psnr( ssd[1], chroma_size );
3854         pic_out->prop.f_psnr[2] = x264_psnr( ssd[2], chroma_size );
3855         pic_out->prop.f_psnr_avg = x264_psnr( ssd[0] + ssd[1] + ssd[2], luma_size + chroma_size*2 );
3856
3857         h->stat.f_ssd_global[h->sh.i_type]   += dur * (ssd[0] + ssd[1] + ssd[2]);
3858         h->stat.f_psnr_average[h->sh.i_type] += dur * pic_out->prop.f_psnr_avg;
3859         h->stat.f_psnr_mean_y[h->sh.i_type]  += dur * pic_out->prop.f_psnr[0];
3860         h->stat.f_psnr_mean_u[h->sh.i_type]  += dur * pic_out->prop.f_psnr[1];
3861         h->stat.f_psnr_mean_v[h->sh.i_type]  += dur * pic_out->prop.f_psnr[2];
3862
3863         snprintf( psz_message, 80, " PSNR Y:%5.2f U:%5.2f V:%5.2f", pic_out->prop.f_psnr[0],
3864                                                                     pic_out->prop.f_psnr[1],
3865                                                                     pic_out->prop.f_psnr[2] );
3866     }
3867
3868     if( h->param.analyse.b_ssim )
3869     {
3870         pic_out->prop.f_ssim = h->stat.frame.f_ssim / h->stat.frame.i_ssim_cnt;
3871         h->stat.f_ssim_mean_y[h->sh.i_type] += pic_out->prop.f_ssim * dur;
3872         snprintf( psz_message + strlen(psz_message), 80 - strlen(psz_message),
3873                   " SSIM Y:%.5f", pic_out->prop.f_ssim );
3874     }
3875     psz_message[79] = '\0';
3876
3877     x264_log( h, X264_LOG_DEBUG,
3878                   "frame=%4d QP=%.2f NAL=%d Slice:%c Poc:%-3d I:%-4d P:%-4d SKIP:%-4d size=%d bytes%s\n",
3879               h->i_frame,
3880               h->fdec->f_qp_avg_aq,
3881               h->i_nal_ref_idc,
3882               h->sh.i_type == SLICE_TYPE_I ? 'I' : (h->sh.i_type == SLICE_TYPE_P ? 'P' : 'B' ),
3883               h->fdec->i_poc,
3884               h->stat.frame.i_mb_count_i,
3885               h->stat.frame.i_mb_count_p,
3886               h->stat.frame.i_mb_count_skip,
3887               frame_size,
3888               psz_message );
3889
3890     // keep stats all in one place
3891     x264_thread_sync_stat( h->thread[0], h );
3892     // for the use of the next frame
3893     x264_thread_sync_stat( thread_current, h );
3894
3895 #ifdef DEBUG_MB_TYPE
3896 {
3897     static const char mb_chars[] = { 'i', 'i', 'I', 'C', 'P', '8', 'S',
3898         'D', '<', 'X', 'B', 'X', '>', 'B', 'B', 'B', 'B', '8', 'S' };
3899     for( int mb_xy = 0; mb_xy < h->mb.i_mb_width * h->mb.i_mb_height; mb_xy++ )
3900     {
3901         if( h->mb.type[mb_xy] < X264_MBTYPE_MAX && h->mb.type[mb_xy] >= 0 )
3902             fprintf( stderr, "%c ", mb_chars[ h->mb.type[mb_xy] ] );
3903         else
3904             fprintf( stderr, "? " );
3905
3906         if( (mb_xy+1) % h->mb.i_mb_width == 0 )
3907             fprintf( stderr, "\n" );
3908     }
3909 }
3910 #endif
3911
3912     /* Remove duplicates, must be done near the end as breaks h->fref0 array
3913      * by freeing some of its pointers. */
3914     for( int i = 0; i < h->i_ref[0]; i++ )
3915         if( h->fref[0][i] && h->fref[0][i]->b_duplicate )
3916         {
3917             x264_frame_push_blank_unused( h, h->fref[0][i] );
3918             h->fref[0][i] = 0;
3919         }
3920
3921     if( h->param.psz_dump_yuv )
3922         x264_frame_dump( h );
3923     x264_emms();
3924
3925     return frame_size;
3926 }
3927
3928 static void x264_print_intra( int64_t *i_mb_count, double i_count, int b_print_pcm, char *intra )
3929 {
3930     intra += sprintf( intra, "I16..4%s: %4.1f%% %4.1f%% %4.1f%%",
3931         b_print_pcm ? "..PCM" : "",
3932         i_mb_count[I_16x16]/ i_count,
3933         i_mb_count[I_8x8]  / i_count,
3934         i_mb_count[I_4x4]  / i_count );
3935     if( b_print_pcm )
3936         sprintf( intra, " %4.1f%%", i_mb_count[I_PCM]  / i_count );
3937 }
3938
3939 /****************************************************************************
3940  * x264_encoder_close:
3941  ****************************************************************************/
3942 void    x264_encoder_close  ( x264_t *h )
3943 {
3944     int64_t i_yuv_size = FRAME_SIZE( h->param.i_width * h->param.i_height );
3945     int64_t i_mb_count_size[2][7] = {{0}};
3946     char buf[200];
3947     int b_print_pcm = h->stat.i_mb_count[SLICE_TYPE_I][I_PCM]
3948                    || h->stat.i_mb_count[SLICE_TYPE_P][I_PCM]
3949                    || h->stat.i_mb_count[SLICE_TYPE_B][I_PCM];
3950
3951     x264_lookahead_delete( h );
3952
3953 #if HAVE_OPENCL
3954     x264_opencl_lookahead_delete( h );
3955     x264_opencl_function_t *ocl = h->opencl.ocl;
3956 #endif
3957
3958     if( h->param.b_sliced_threads )
3959         x264_threadpool_wait_all( h );
3960     if( h->param.i_threads > 1 )
3961         x264_threadpool_delete( h->threadpool );
3962     if( h->param.i_lookahead_threads > 1 )
3963         x264_threadpool_delete( h->lookaheadpool );
3964     if( h->i_thread_frames > 1 )
3965     {
3966         for( int i = 0; i < h->i_thread_frames; i++ )
3967             if( h->thread[i]->b_thread_active )
3968             {
3969                 assert( h->thread[i]->fenc->i_reference_count == 1 );
3970                 x264_frame_delete( h->thread[i]->fenc );
3971             }
3972
3973         x264_t *thread_prev = h->thread[h->i_thread_phase];
3974         x264_thread_sync_ratecontrol( h, thread_prev, h );
3975         x264_thread_sync_ratecontrol( thread_prev, thread_prev, h );
3976         h->i_frame = thread_prev->i_frame + 1 - h->i_thread_frames;
3977     }
3978     h->i_frame++;
3979
3980     /* Slices used and PSNR */
3981     for( int i = 0; i < 3; i++ )
3982     {
3983         static const uint8_t slice_order[] = { SLICE_TYPE_I, SLICE_TYPE_P, SLICE_TYPE_B };
3984         int i_slice = slice_order[i];
3985
3986         if( h->stat.i_frame_count[i_slice] > 0 )
3987         {
3988             int i_count = h->stat.i_frame_count[i_slice];
3989             double dur =  h->stat.f_frame_duration[i_slice];
3990             if( h->param.analyse.b_psnr )
3991             {
3992                 x264_log( h, X264_LOG_INFO,
3993                           "frame %c:%-5d Avg QP:%5.2f  size:%6.0f  PSNR Mean Y:%5.2f U:%5.2f V:%5.2f Avg:%5.2f Global:%5.2f\n",
3994                           slice_type_to_char[i_slice],
3995                           i_count,
3996                           h->stat.f_frame_qp[i_slice] / i_count,
3997                           (double)h->stat.i_frame_size[i_slice] / i_count,
3998                           h->stat.f_psnr_mean_y[i_slice] / dur, h->stat.f_psnr_mean_u[i_slice] / dur, h->stat.f_psnr_mean_v[i_slice] / dur,
3999                           h->stat.f_psnr_average[i_slice] / dur,
4000                           x264_psnr( h->stat.f_ssd_global[i_slice], dur * i_yuv_size ) );
4001             }
4002             else
4003             {
4004                 x264_log( h, X264_LOG_INFO,
4005                           "frame %c:%-5d Avg QP:%5.2f  size:%6.0f\n",
4006                           slice_type_to_char[i_slice],
4007                           i_count,
4008                           h->stat.f_frame_qp[i_slice] / i_count,
4009                           (double)h->stat.i_frame_size[i_slice] / i_count );
4010             }
4011         }
4012     }
4013     if( h->param.i_bframe && h->stat.i_frame_count[SLICE_TYPE_B] )
4014     {
4015         char *p = buf;
4016         int den = 0;
4017         // weight by number of frames (including the I/P-frames) that are in a sequence of N B-frames
4018         for( int i = 0; i <= h->param.i_bframe; i++ )
4019             den += (i+1) * h->stat.i_consecutive_bframes[i];
4020         for( int i = 0; i <= h->param.i_bframe; i++ )
4021             p += sprintf( p, " %4.1f%%", 100. * (i+1) * h->stat.i_consecutive_bframes[i] / den );
4022         x264_log( h, X264_LOG_INFO, "consecutive B-frames:%s\n", buf );
4023     }
4024
4025     for( int i_type = 0; i_type < 2; i_type++ )
4026         for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
4027         {
4028             if( i == D_DIRECT_8x8 ) continue; /* direct is counted as its own type */
4029             i_mb_count_size[i_type][x264_mb_partition_pixel_table[i]] += h->stat.i_mb_partition[i_type][i];
4030         }
4031
4032     /* MB types used */
4033     if( h->stat.i_frame_count[SLICE_TYPE_I] > 0 )
4034     {
4035         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_I];
4036         double i_count = h->stat.i_frame_count[SLICE_TYPE_I] * h->mb.i_mb_count / 100.0;
4037         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
4038         x264_log( h, X264_LOG_INFO, "mb I  %s\n", buf );
4039     }
4040     if( h->stat.i_frame_count[SLICE_TYPE_P] > 0 )
4041     {
4042         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_P];
4043         double i_count = h->stat.i_frame_count[SLICE_TYPE_P] * h->mb.i_mb_count / 100.0;
4044         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_P];
4045         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
4046         x264_log( h, X264_LOG_INFO,
4047                   "mb P  %s  P16..4: %4.1f%% %4.1f%% %4.1f%% %4.1f%% %4.1f%%    skip:%4.1f%%\n",
4048                   buf,
4049                   i_mb_size[PIXEL_16x16] / (i_count*4),
4050                   (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
4051                   i_mb_size[PIXEL_8x8] / (i_count*4),
4052                   (i_mb_size[PIXEL_8x4] + i_mb_size[PIXEL_4x8]) / (i_count*4),
4053                   i_mb_size[PIXEL_4x4] / (i_count*4),
4054                   i_mb_count[P_SKIP] / i_count );
4055     }
4056     if( h->stat.i_frame_count[SLICE_TYPE_B] > 0 )
4057     {
4058         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_B];
4059         double i_count = h->stat.i_frame_count[SLICE_TYPE_B] * h->mb.i_mb_count / 100.0;
4060         double i_mb_list_count;
4061         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_B];
4062         int64_t list_count[3] = {0}; /* 0 == L0, 1 == L1, 2 == BI */
4063         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
4064         for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
4065             for( int j = 0; j < 2; j++ )
4066             {
4067                 int l0 = x264_mb_type_list_table[i][0][j];
4068                 int l1 = x264_mb_type_list_table[i][1][j];
4069                 if( l0 || l1 )
4070                     list_count[l1+l0*l1] += h->stat.i_mb_count[SLICE_TYPE_B][i] * 2;
4071             }
4072         list_count[0] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L0_8x8];
4073         list_count[1] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L1_8x8];
4074         list_count[2] += h->stat.i_mb_partition[SLICE_TYPE_B][D_BI_8x8];
4075         i_mb_count[B_DIRECT] += (h->stat.i_mb_partition[SLICE_TYPE_B][D_DIRECT_8x8]+2)/4;
4076         i_mb_list_count = (list_count[0] + list_count[1] + list_count[2]) / 100.0;
4077         sprintf( buf + strlen(buf), "  B16..8: %4.1f%% %4.1f%% %4.1f%%  direct:%4.1f%%  skip:%4.1f%%",
4078                  i_mb_size[PIXEL_16x16] / (i_count*4),
4079                  (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
4080                  i_mb_size[PIXEL_8x8] / (i_count*4),
4081                  i_mb_count[B_DIRECT] / i_count,
4082                  i_mb_count[B_SKIP]   / i_count );
4083         if( i_mb_list_count != 0 )
4084             sprintf( buf + strlen(buf), "  L0:%4.1f%% L1:%4.1f%% BI:%4.1f%%",
4085                      list_count[0] / i_mb_list_count,
4086                      list_count[1] / i_mb_list_count,
4087                      list_count[2] / i_mb_list_count );
4088         x264_log( h, X264_LOG_INFO, "mb B  %s\n", buf );
4089     }
4090
4091     x264_ratecontrol_summary( h );
4092
4093     if( h->stat.i_frame_count[SLICE_TYPE_I] + h->stat.i_frame_count[SLICE_TYPE_P] + h->stat.i_frame_count[SLICE_TYPE_B] > 0 )
4094     {
4095 #define SUM3(p) (p[SLICE_TYPE_I] + p[SLICE_TYPE_P] + p[SLICE_TYPE_B])
4096 #define SUM3b(p,o) (p[SLICE_TYPE_I][o] + p[SLICE_TYPE_P][o] + p[SLICE_TYPE_B][o])
4097         int64_t i_i8x8 = SUM3b( h->stat.i_mb_count, I_8x8 );
4098         int64_t i_intra = i_i8x8 + SUM3b( h->stat.i_mb_count, I_4x4 )
4099                                  + SUM3b( h->stat.i_mb_count, I_16x16 );
4100         int64_t i_all_intra = i_intra + SUM3b( h->stat.i_mb_count, I_PCM);
4101         int64_t i_skip = SUM3b( h->stat.i_mb_count, P_SKIP )
4102                        + SUM3b( h->stat.i_mb_count, B_SKIP );
4103         const int i_count = h->stat.i_frame_count[SLICE_TYPE_I] +
4104                             h->stat.i_frame_count[SLICE_TYPE_P] +
4105                             h->stat.i_frame_count[SLICE_TYPE_B];
4106         int64_t i_mb_count = (int64_t)i_count * h->mb.i_mb_count;
4107         int64_t i_inter = i_mb_count - i_skip - i_intra;
4108         const double duration = h->stat.f_frame_duration[SLICE_TYPE_I] +
4109                                 h->stat.f_frame_duration[SLICE_TYPE_P] +
4110                                 h->stat.f_frame_duration[SLICE_TYPE_B];
4111         float f_bitrate = SUM3(h->stat.i_frame_size) / duration / 125;
4112
4113         if( PARAM_INTERLACED )
4114         {
4115             char *fieldstats = buf;
4116             fieldstats[0] = 0;
4117             if( i_inter )
4118                 fieldstats += sprintf( fieldstats, " inter:%.1f%%", h->stat.i_mb_field[1] * 100.0 / i_inter );
4119             if( i_skip )
4120                 fieldstats += sprintf( fieldstats, " skip:%.1f%%", h->stat.i_mb_field[2] * 100.0 / i_skip );
4121             x264_log( h, X264_LOG_INFO, "field mbs: intra: %.1f%%%s\n",
4122                       h->stat.i_mb_field[0] * 100.0 / i_intra, buf );
4123         }
4124
4125         if( h->pps->b_transform_8x8_mode )
4126         {
4127             buf[0] = 0;
4128             if( h->stat.i_mb_count_8x8dct[0] )
4129                 sprintf( buf, " inter:%.1f%%", 100. * h->stat.i_mb_count_8x8dct[1] / h->stat.i_mb_count_8x8dct[0] );
4130             x264_log( h, X264_LOG_INFO, "8x8 transform intra:%.1f%%%s\n", 100. * i_i8x8 / i_intra, buf );
4131         }
4132
4133         if( (h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO ||
4134             (h->stat.i_direct_frames[0] && h->stat.i_direct_frames[1]))
4135             && h->stat.i_frame_count[SLICE_TYPE_B] )
4136         {
4137             x264_log( h, X264_LOG_INFO, "direct mvs  spatial:%.1f%% temporal:%.1f%%\n",
4138                       h->stat.i_direct_frames[1] * 100. / h->stat.i_frame_count[SLICE_TYPE_B],
4139                       h->stat.i_direct_frames[0] * 100. / h->stat.i_frame_count[SLICE_TYPE_B] );
4140         }
4141
4142         buf[0] = 0;
4143         int csize = CHROMA444 ? 4 : 1;
4144         if( i_mb_count != i_all_intra )
4145             sprintf( buf, " inter: %.1f%% %.1f%% %.1f%%",
4146                      h->stat.i_mb_cbp[1] * 100.0 / ((i_mb_count - i_all_intra)*4),
4147                      h->stat.i_mb_cbp[3] * 100.0 / ((i_mb_count - i_all_intra)*csize),
4148