2ff45153957d52ac40c39ef0d4c1de143cf7a632
[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][7] =
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               { 60000, 1001, 0, 1820, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy },
644               { 30000, 1001, 0, 1820, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy },
645               {    50,    1, 0, 2196, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy },
646               {    25,    1, 0, 2196, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy },
647               { 24000, 1001, 0, 1820, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy }}},
648             {{{ 60000, 1001, 0, 1848, x264_cqm_avci100_720p_4ic, x264_cqm_avci100_720p_8iy },
649               {    50,    1, 0, 2224, x264_cqm_avci100_720p_4ic, x264_cqm_avci100_720p_8iy },
650               { 30000, 1001, 0, 1848, x264_cqm_avci100_720p_4ic, x264_cqm_avci100_720p_8iy },
651               {    25,    1, 0, 2224, x264_cqm_avci100_720p_4ic, x264_cqm_avci100_720p_8iy },
652               { 24000, 1001, 0, 1848, x264_cqm_avci100_720p_4ic, x264_cqm_avci100_720p_8iy }},
653              {{ 30000, 1001, 1, 3692, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080i_8iy },
654               {    25,    1, 1, 4444, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080i_8iy },
655               { 60000, 1001, 0, 3692, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080p_8iy },
656               { 30000, 1001, 0, 3692, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080p_8iy },
657               {    50,    1, 0, 4444, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080p_8iy },
658               {    25,    1, 0, 4444, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080p_8iy },
659               { 24000, 1001, 0, 3692, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080p_8iy }}}
660         };
661
662         int res = -1;
663         int type = i_csp == X264_CSP_I422;
664         if( type )
665         {
666             if(      h->param.i_width == 1920 && h->param.i_height == 1080 ) res =  1;
667             else if( h->param.i_width == 1280 && h->param.i_height ==  720 ) res =  0;
668         }
669         else if( i_csp == X264_CSP_I420 )
670         {
671             if(      h->param.i_width == 1440 && h->param.i_height == 1080 ) res =  1;
672             else if( h->param.i_width ==  960 && h->param.i_height ==  720 ) res =  0;
673         }
674         else
675         {
676             x264_log( h, X264_LOG_ERROR, "Invalid colorspace for AVC-Intra\n" );
677             return -1;
678         }
679
680         if( res < 0 )
681         {
682             x264_log( h, X264_LOG_ERROR, "Resolution %dx%d invalid for AVC-Intra %s\n",
683                       h->param.i_width, h->param.i_height, type ? "100" : "50" );
684             return -1;
685         }
686
687         if( h->param.nalu_process )
688         {
689             x264_log( h, X264_LOG_ERROR, "nalu_process is not supported in AVC-Intra mode\n" );
690             return -1;
691         }
692
693         if( !h->param.b_repeat_headers )
694         {
695             x264_log( h, X264_LOG_ERROR, "Separate headers not supported in AVC-Intra mode\n" );
696             return -1;
697         }
698
699         int i;
700         uint32_t fps_num = h->param.i_fps_num, fps_den = h->param.i_fps_den;
701         x264_reduce_fraction( &fps_num, &fps_den );
702         for( i = 0; i < 7; i++ )
703         {
704             if( avcintra_lut[type][res][i].fps_num == fps_num &&
705                 avcintra_lut[type][res][i].fps_den == fps_den &&
706                 avcintra_lut[type][res][i].interlaced == PARAM_INTERLACED )
707             {
708                 break;
709             }
710         }
711         if( i == 7 )
712         {
713             x264_log( h, X264_LOG_ERROR, "FPS %d/%d%c not compatible with AVC-Intra\n",
714                       h->param.i_fps_num, h->param.i_fps_den, PARAM_INTERLACED ? 'i' : 'p' );
715             return -1;
716         }
717
718         h->param.i_keyint_max = 1;
719         h->param.b_intra_refresh = 0;
720         h->param.analyse.i_weighted_pred = 0;
721         h->param.i_frame_reference = 1;
722         h->param.i_dpb_size = 1;
723
724         h->param.b_bluray_compat = 0;
725         h->param.b_vfr_input = 0;
726         h->param.b_aud = 1;
727         h->param.vui.i_chroma_loc = 0;
728         h->param.i_nal_hrd = X264_NAL_HRD_NONE;
729         h->param.b_deblocking_filter = 0;
730         h->param.b_stitchable = 1;
731         h->param.b_pic_struct = 0;
732         h->param.analyse.b_transform_8x8 = 1;
733         h->param.analyse.intra = X264_ANALYSE_I8x8;
734         h->param.analyse.i_chroma_qp_offset = res && type ? 3 : 4;
735         h->param.b_cabac = !type;
736         h->param.rc.i_vbv_buffer_size = avcintra_lut[type][res][i].frame_size;
737         h->param.rc.i_vbv_max_bitrate =
738         h->param.rc.i_bitrate = h->param.rc.i_vbv_buffer_size * fps_num / fps_den;
739         h->param.rc.i_rc_method = X264_RC_ABR;
740         h->param.rc.f_vbv_buffer_init = 1.0;
741         h->param.rc.b_filler = 1;
742         h->param.i_cqm_preset = X264_CQM_CUSTOM;
743         memcpy( h->param.cqm_4iy, x264_cqm_jvt4i, sizeof(h->param.cqm_4iy) );
744         memcpy( h->param.cqm_4ic, avcintra_lut[type][res][i].cqm_4ic, sizeof(h->param.cqm_4ic) );
745         memcpy( h->param.cqm_8iy, avcintra_lut[type][res][i].cqm_8iy, sizeof(h->param.cqm_8iy) );
746
747         /* Need exactly 10 slices of equal MB count... why?  $deity knows... */
748         h->param.i_slice_max_mbs = ((h->param.i_width + 15) / 16) * ((h->param.i_height + 15) / 16) / 10;
749         h->param.i_slice_max_size = 0;
750         /* The slice structure only allows a maximum of 2 threads for 1080i/p
751          * and 1 or 5 threads for 720p */
752         if( h->param.b_sliced_threads )
753         {
754             if( res )
755                 h->param.i_threads = X264_MIN( 2, h->param.i_threads );
756             else
757             {
758                 h->param.i_threads = X264_MIN( 5, h->param.i_threads );
759                 if( h->param.i_threads < 5 )
760                     h->param.i_threads = 1;
761             }
762         }
763
764         if( type )
765             h->param.vui.i_sar_width = h->param.vui.i_sar_height = 1;
766         else
767         {
768             h->param.vui.i_sar_width  = 4;
769             h->param.vui.i_sar_height = 3;
770         }
771
772         /* Avid cannot handle negative QPs */
773         h->param.rc.i_qp_min = X264_MAX( h->param.rc.i_qp_min, QP_BD_OFFSET );
774     }
775
776     h->param.rc.f_rf_constant = x264_clip3f( h->param.rc.f_rf_constant, -QP_BD_OFFSET, 51 );
777     h->param.rc.f_rf_constant_max = x264_clip3f( h->param.rc.f_rf_constant_max, -QP_BD_OFFSET, 51 );
778     h->param.rc.i_qp_constant = x264_clip3( h->param.rc.i_qp_constant, 0, QP_MAX );
779     h->param.analyse.i_subpel_refine = x264_clip3( h->param.analyse.i_subpel_refine, 0, 11 );
780     h->param.rc.f_ip_factor = X264_MAX( h->param.rc.f_ip_factor, 0.01f );
781     h->param.rc.f_pb_factor = X264_MAX( h->param.rc.f_pb_factor, 0.01f );
782     if( h->param.rc.i_rc_method == X264_RC_CRF )
783     {
784         h->param.rc.i_qp_constant = h->param.rc.f_rf_constant + QP_BD_OFFSET;
785         h->param.rc.i_bitrate = 0;
786     }
787     if( b_open && (h->param.rc.i_rc_method == X264_RC_CQP || h->param.rc.i_rc_method == X264_RC_CRF)
788         && h->param.rc.i_qp_constant == 0 )
789     {
790         h->mb.b_lossless = 1;
791         h->param.i_cqm_preset = X264_CQM_FLAT;
792         h->param.psz_cqm_file = NULL;
793         h->param.rc.i_rc_method = X264_RC_CQP;
794         h->param.rc.f_ip_factor = 1;
795         h->param.rc.f_pb_factor = 1;
796         h->param.analyse.b_psnr = 0;
797         h->param.analyse.b_ssim = 0;
798         h->param.analyse.i_chroma_qp_offset = 0;
799         h->param.analyse.i_trellis = 0;
800         h->param.analyse.b_fast_pskip = 0;
801         h->param.analyse.i_noise_reduction = 0;
802         h->param.analyse.b_psy = 0;
803         h->param.i_bframe = 0;
804         /* 8x8dct is not useful without RD in CAVLC lossless */
805         if( !h->param.b_cabac && h->param.analyse.i_subpel_refine < 6 )
806             h->param.analyse.b_transform_8x8 = 0;
807     }
808     if( h->param.rc.i_rc_method == X264_RC_CQP )
809     {
810         float qp_p = h->param.rc.i_qp_constant;
811         float qp_i = qp_p - 6*log2f( h->param.rc.f_ip_factor );
812         float qp_b = qp_p + 6*log2f( h->param.rc.f_pb_factor );
813         h->param.rc.i_qp_min = x264_clip3( (int)(X264_MIN3( qp_p, qp_i, qp_b )), 0, QP_MAX );
814         h->param.rc.i_qp_max = x264_clip3( (int)(X264_MAX3( qp_p, qp_i, qp_b ) + .999), 0, QP_MAX );
815         h->param.rc.i_aq_mode = 0;
816         h->param.rc.b_mb_tree = 0;
817         h->param.rc.i_bitrate = 0;
818     }
819     h->param.rc.i_qp_max = x264_clip3( h->param.rc.i_qp_max, 0, QP_MAX );
820     h->param.rc.i_qp_min = x264_clip3( h->param.rc.i_qp_min, 0, h->param.rc.i_qp_max );
821     h->param.rc.i_qp_step = x264_clip3( h->param.rc.i_qp_step, 2, QP_MAX );
822     h->param.rc.i_bitrate = x264_clip3( h->param.rc.i_bitrate, 0, 2000000 );
823     if( h->param.rc.i_rc_method == X264_RC_ABR && !h->param.rc.i_bitrate )
824     {
825         x264_log( h, X264_LOG_ERROR, "bitrate not specified\n" );
826         return -1;
827     }
828     h->param.rc.i_vbv_buffer_size = x264_clip3( h->param.rc.i_vbv_buffer_size, 0, 2000000 );
829     h->param.rc.i_vbv_max_bitrate = x264_clip3( h->param.rc.i_vbv_max_bitrate, 0, 2000000 );
830     h->param.rc.f_vbv_buffer_init = x264_clip3f( h->param.rc.f_vbv_buffer_init, 0, 2000000 );
831     if( h->param.rc.i_vbv_buffer_size )
832     {
833         if( h->param.rc.i_rc_method == X264_RC_CQP )
834         {
835             x264_log( h, X264_LOG_WARNING, "VBV is incompatible with constant QP, ignored.\n" );
836             h->param.rc.i_vbv_max_bitrate = 0;
837             h->param.rc.i_vbv_buffer_size = 0;
838         }
839         else if( h->param.rc.i_vbv_max_bitrate == 0 )
840         {
841             if( h->param.rc.i_rc_method == X264_RC_ABR )
842             {
843                 x264_log( h, X264_LOG_WARNING, "VBV maxrate unspecified, assuming CBR\n" );
844                 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate;
845             }
846             else
847             {
848                 x264_log( h, X264_LOG_WARNING, "VBV bufsize set but maxrate unspecified, ignored\n" );
849                 h->param.rc.i_vbv_buffer_size = 0;
850             }
851         }
852         else if( h->param.rc.i_vbv_max_bitrate < h->param.rc.i_bitrate &&
853                  h->param.rc.i_rc_method == X264_RC_ABR )
854         {
855             x264_log( h, X264_LOG_WARNING, "max bitrate less than average bitrate, assuming CBR\n" );
856             h->param.rc.i_bitrate = h->param.rc.i_vbv_max_bitrate;
857         }
858     }
859     else if( h->param.rc.i_vbv_max_bitrate )
860     {
861         x264_log( h, X264_LOG_WARNING, "VBV maxrate specified, but no bufsize, ignored\n" );
862         h->param.rc.i_vbv_max_bitrate = 0;
863     }
864
865     h->param.i_slice_max_size = X264_MAX( h->param.i_slice_max_size, 0 );
866     h->param.i_slice_max_mbs = X264_MAX( h->param.i_slice_max_mbs, 0 );
867     h->param.i_slice_min_mbs = X264_MAX( h->param.i_slice_min_mbs, 0 );
868     if( h->param.i_slice_max_mbs )
869         h->param.i_slice_min_mbs = X264_MIN( h->param.i_slice_min_mbs, h->param.i_slice_max_mbs/2 );
870     else if( !h->param.i_slice_max_size )
871         h->param.i_slice_min_mbs = 0;
872     if( PARAM_INTERLACED && h->param.i_slice_min_mbs )
873     {
874         x264_log( h, X264_LOG_WARNING, "interlace + slice-min-mbs is not implemented\n" );
875         h->param.i_slice_min_mbs = 0;
876     }
877     int mb_width = (h->param.i_width+15)/16;
878     if( h->param.i_slice_min_mbs > mb_width )
879     {
880         x264_log( h, X264_LOG_WARNING, "slice-min-mbs > row mb size (%d) not implemented\n", mb_width );
881         h->param.i_slice_min_mbs = mb_width;
882     }
883
884     int max_slices = (h->param.i_height+((16<<PARAM_INTERLACED)-1))/(16<<PARAM_INTERLACED);
885     if( h->param.b_sliced_threads )
886         h->param.i_slice_count = x264_clip3( h->param.i_threads, 0, max_slices );
887     else
888     {
889         h->param.i_slice_count = x264_clip3( h->param.i_slice_count, 0, max_slices );
890         if( h->param.i_slice_max_mbs || h->param.i_slice_max_size )
891             h->param.i_slice_count = 0;
892     }
893     if( h->param.i_slice_count_max > 0 )
894         h->param.i_slice_count_max = X264_MAX( h->param.i_slice_count, h->param.i_slice_count_max );
895
896     if( h->param.b_bluray_compat )
897     {
898         h->param.i_bframe_pyramid = X264_MIN( X264_B_PYRAMID_STRICT, h->param.i_bframe_pyramid );
899         h->param.i_bframe = X264_MIN( h->param.i_bframe, 3 );
900         h->param.b_aud = 1;
901         h->param.i_nal_hrd = X264_MAX( h->param.i_nal_hrd, X264_NAL_HRD_VBR );
902         h->param.i_slice_max_size = 0;
903         h->param.i_slice_max_mbs = 0;
904         h->param.b_intra_refresh = 0;
905         h->param.i_frame_reference = X264_MIN( h->param.i_frame_reference, 6 );
906         h->param.i_dpb_size = X264_MIN( h->param.i_dpb_size, 6 );
907         /* Don't use I-frames, because Blu-ray treats them the same as IDR. */
908         h->param.i_keyint_min = 1;
909         /* Due to the proliferation of broken players that don't handle dupes properly. */
910         h->param.analyse.i_weighted_pred = X264_MIN( h->param.analyse.i_weighted_pred, X264_WEIGHTP_SIMPLE );
911         if( h->param.b_fake_interlaced )
912             h->param.b_pic_struct = 1;
913     }
914
915     h->param.i_frame_reference = x264_clip3( h->param.i_frame_reference, 1, X264_REF_MAX );
916     h->param.i_dpb_size = x264_clip3( h->param.i_dpb_size, 1, X264_REF_MAX );
917     if( h->param.i_scenecut_threshold < 0 )
918         h->param.i_scenecut_threshold = 0;
919     h->param.analyse.i_direct_mv_pred = x264_clip3( h->param.analyse.i_direct_mv_pred, X264_DIRECT_PRED_NONE, X264_DIRECT_PRED_AUTO );
920     if( !h->param.analyse.i_subpel_refine && h->param.analyse.i_direct_mv_pred > X264_DIRECT_PRED_SPATIAL )
921     {
922         x264_log( h, X264_LOG_WARNING, "subme=0 + direct=temporal is not supported\n" );
923         h->param.analyse.i_direct_mv_pred = X264_DIRECT_PRED_SPATIAL;
924     }
925     h->param.i_bframe = x264_clip3( h->param.i_bframe, 0, X264_MIN( X264_BFRAME_MAX, h->param.i_keyint_max-1 ) );
926     h->param.i_bframe_bias = x264_clip3( h->param.i_bframe_bias, -90, 100 );
927     if( h->param.i_bframe <= 1 )
928         h->param.i_bframe_pyramid = X264_B_PYRAMID_NONE;
929     h->param.i_bframe_pyramid = x264_clip3( h->param.i_bframe_pyramid, X264_B_PYRAMID_NONE, X264_B_PYRAMID_NORMAL );
930     h->param.i_bframe_adaptive = x264_clip3( h->param.i_bframe_adaptive, X264_B_ADAPT_NONE, X264_B_ADAPT_TRELLIS );
931     if( !h->param.i_bframe )
932     {
933         h->param.i_bframe_adaptive = X264_B_ADAPT_NONE;
934         h->param.analyse.i_direct_mv_pred = 0;
935         h->param.analyse.b_weighted_bipred = 0;
936         h->param.b_open_gop = 0;
937     }
938     if( h->param.b_intra_refresh && h->param.i_bframe_pyramid == X264_B_PYRAMID_NORMAL )
939     {
940         x264_log( h, X264_LOG_WARNING, "b-pyramid normal + intra-refresh is not supported\n" );
941         h->param.i_bframe_pyramid = X264_B_PYRAMID_STRICT;
942     }
943     if( h->param.b_intra_refresh && (h->param.i_frame_reference > 1 || h->param.i_dpb_size > 1) )
944     {
945         x264_log( h, X264_LOG_WARNING, "ref > 1 + intra-refresh is not supported\n" );
946         h->param.i_frame_reference = 1;
947         h->param.i_dpb_size = 1;
948     }
949     if( h->param.b_intra_refresh && h->param.b_open_gop )
950     {
951         x264_log( h, X264_LOG_WARNING, "intra-refresh is not compatible with open-gop\n" );
952         h->param.b_open_gop = 0;
953     }
954     if( !h->param.i_fps_num || !h->param.i_fps_den )
955     {
956         h->param.i_fps_num = 25;
957         h->param.i_fps_den = 1;
958     }
959     float fps = (float) h->param.i_fps_num / h->param.i_fps_den;
960     if( h->param.i_keyint_min == X264_KEYINT_MIN_AUTO )
961         h->param.i_keyint_min = X264_MIN( h->param.i_keyint_max / 10, fps );
962     h->param.i_keyint_min = x264_clip3( h->param.i_keyint_min, 1, h->param.i_keyint_max/2+1 );
963     h->param.rc.i_lookahead = x264_clip3( h->param.rc.i_lookahead, 0, X264_LOOKAHEAD_MAX );
964     {
965         int maxrate = X264_MAX( h->param.rc.i_vbv_max_bitrate, h->param.rc.i_bitrate );
966         float bufsize = maxrate ? (float)h->param.rc.i_vbv_buffer_size / maxrate : 0;
967         h->param.rc.i_lookahead = X264_MIN( h->param.rc.i_lookahead, X264_MAX( h->param.i_keyint_max, bufsize*fps ) );
968     }
969
970     if( !h->param.i_timebase_num || !h->param.i_timebase_den || !(h->param.b_vfr_input || h->param.b_pulldown) )
971     {
972         h->param.i_timebase_num = h->param.i_fps_den;
973         h->param.i_timebase_den = h->param.i_fps_num;
974     }
975
976     h->param.rc.f_qcompress = x264_clip3f( h->param.rc.f_qcompress, 0.0, 1.0 );
977     if( h->param.i_keyint_max == 1 || h->param.rc.f_qcompress == 1 )
978         h->param.rc.b_mb_tree = 0;
979     if( (!h->param.b_intra_refresh && h->param.i_keyint_max != X264_KEYINT_MAX_INFINITE) &&
980         !h->param.rc.i_lookahead && h->param.rc.b_mb_tree )
981     {
982         x264_log( h, X264_LOG_WARNING, "lookaheadless mb-tree requires intra refresh or infinite keyint\n" );
983         h->param.rc.b_mb_tree = 0;
984     }
985     if( b_open && h->param.rc.b_stat_read )
986         h->param.rc.i_lookahead = 0;
987 #if HAVE_THREAD
988     if( h->param.i_sync_lookahead < 0 )
989         h->param.i_sync_lookahead = h->param.i_bframe + 1;
990     h->param.i_sync_lookahead = X264_MIN( h->param.i_sync_lookahead, X264_LOOKAHEAD_MAX );
991     if( h->param.rc.b_stat_read || h->i_thread_frames == 1 )
992         h->param.i_sync_lookahead = 0;
993 #else
994     h->param.i_sync_lookahead = 0;
995 #endif
996
997     h->param.i_deblocking_filter_alphac0 = x264_clip3( h->param.i_deblocking_filter_alphac0, -6, 6 );
998     h->param.i_deblocking_filter_beta    = x264_clip3( h->param.i_deblocking_filter_beta, -6, 6 );
999     h->param.analyse.i_luma_deadzone[0] = x264_clip3( h->param.analyse.i_luma_deadzone[0], 0, 32 );
1000     h->param.analyse.i_luma_deadzone[1] = x264_clip3( h->param.analyse.i_luma_deadzone[1], 0, 32 );
1001
1002     h->param.i_cabac_init_idc = x264_clip3( h->param.i_cabac_init_idc, 0, 2 );
1003
1004     if( h->param.i_cqm_preset < X264_CQM_FLAT || h->param.i_cqm_preset > X264_CQM_CUSTOM )
1005         h->param.i_cqm_preset = X264_CQM_FLAT;
1006
1007     if( h->param.analyse.i_me_method < X264_ME_DIA ||
1008         h->param.analyse.i_me_method > X264_ME_TESA )
1009         h->param.analyse.i_me_method = X264_ME_HEX;
1010     h->param.analyse.i_me_range = x264_clip3( h->param.analyse.i_me_range, 4, 1024 );
1011     if( h->param.analyse.i_me_range > 16 && h->param.analyse.i_me_method <= X264_ME_HEX )
1012         h->param.analyse.i_me_range = 16;
1013     if( h->param.analyse.i_me_method == X264_ME_TESA &&
1014         (h->mb.b_lossless || h->param.analyse.i_subpel_refine <= 1) )
1015         h->param.analyse.i_me_method = X264_ME_ESA;
1016     h->param.analyse.b_mixed_references = h->param.analyse.b_mixed_references && h->param.i_frame_reference > 1;
1017     h->param.analyse.inter &= X264_ANALYSE_PSUB16x16|X264_ANALYSE_PSUB8x8|X264_ANALYSE_BSUB16x16|
1018                               X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
1019     h->param.analyse.intra &= X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
1020     if( !(h->param.analyse.inter & X264_ANALYSE_PSUB16x16) )
1021         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
1022     if( !h->param.analyse.b_transform_8x8 )
1023     {
1024         h->param.analyse.inter &= ~X264_ANALYSE_I8x8;
1025         h->param.analyse.intra &= ~X264_ANALYSE_I8x8;
1026     }
1027     h->param.analyse.i_trellis = x264_clip3( h->param.analyse.i_trellis, 0, 2 );
1028     h->param.rc.i_aq_mode = x264_clip3( h->param.rc.i_aq_mode, 0, 2 );
1029     h->param.rc.f_aq_strength = x264_clip3f( h->param.rc.f_aq_strength, 0, 3 );
1030     if( h->param.rc.f_aq_strength == 0 )
1031         h->param.rc.i_aq_mode = 0;
1032
1033     if( h->param.i_log_level < X264_LOG_INFO )
1034     {
1035         h->param.analyse.b_psnr = 0;
1036         h->param.analyse.b_ssim = 0;
1037     }
1038     /* Warn users trying to measure PSNR/SSIM with psy opts on. */
1039     if( b_open && (h->param.analyse.b_psnr || h->param.analyse.b_ssim) )
1040     {
1041         char *s = NULL;
1042
1043         if( h->param.analyse.b_psy )
1044         {
1045             s = h->param.analyse.b_psnr ? "psnr" : "ssim";
1046             x264_log( h, X264_LOG_WARNING, "--%s used with psy on: results will be invalid!\n", s );
1047         }
1048         else if( !h->param.rc.i_aq_mode && h->param.analyse.b_ssim )
1049         {
1050             x264_log( h, X264_LOG_WARNING, "--ssim used with AQ off: results will be invalid!\n" );
1051             s = "ssim";
1052         }
1053         else if(  h->param.rc.i_aq_mode && h->param.analyse.b_psnr )
1054         {
1055             x264_log( h, X264_LOG_WARNING, "--psnr used with AQ on: results will be invalid!\n" );
1056             s = "psnr";
1057         }
1058         if( s )
1059             x264_log( h, X264_LOG_WARNING, "--tune %s should be used if attempting to benchmark %s!\n", s, s );
1060     }
1061
1062     if( !h->param.analyse.b_psy )
1063     {
1064         h->param.analyse.f_psy_rd = 0;
1065         h->param.analyse.f_psy_trellis = 0;
1066     }
1067     h->param.analyse.f_psy_rd = x264_clip3f( h->param.analyse.f_psy_rd, 0, 10 );
1068     h->param.analyse.f_psy_trellis = x264_clip3f( h->param.analyse.f_psy_trellis, 0, 10 );
1069     h->mb.i_psy_rd = h->param.analyse.i_subpel_refine >= 6 ? FIX8( h->param.analyse.f_psy_rd ) : 0;
1070     h->mb.i_psy_trellis = h->param.analyse.i_trellis ? FIX8( h->param.analyse.f_psy_trellis / 4 ) : 0;
1071     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -32, 32);
1072     /* In 4:4:4 mode, chroma gets twice as much resolution, so we can halve its quality. */
1073     if( b_open && i_csp >= X264_CSP_I444 && i_csp < X264_CSP_BGR && h->param.analyse.b_psy )
1074         h->param.analyse.i_chroma_qp_offset += 6;
1075     /* Psy RDO increases overall quantizers to improve the quality of luma--this indirectly hurts chroma quality */
1076     /* so we lower the chroma QP offset to compensate */
1077     if( b_open && h->mb.i_psy_rd && !h->param.b_avcintra_compat )
1078         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_rd < 0.25 ? 1 : 2;
1079     /* Psy trellis has a similar effect. */
1080     if( b_open && h->mb.i_psy_trellis && !h->param.b_avcintra_compat )
1081         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_trellis < 0.25 ? 1 : 2;
1082     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -12, 12);
1083     /* MB-tree requires AQ to be on, even if the strength is zero. */
1084     if( !h->param.rc.i_aq_mode && h->param.rc.b_mb_tree )
1085     {
1086         h->param.rc.i_aq_mode = 1;
1087         h->param.rc.f_aq_strength = 0;
1088     }
1089     h->param.analyse.i_noise_reduction = x264_clip3( h->param.analyse.i_noise_reduction, 0, 1<<16 );
1090     if( h->param.analyse.i_subpel_refine >= 10 && (h->param.analyse.i_trellis != 2 || !h->param.rc.i_aq_mode) )
1091         h->param.analyse.i_subpel_refine = 9;
1092
1093     {
1094         const x264_level_t *l = x264_levels;
1095         if( h->param.i_level_idc < 0 )
1096         {
1097             int maxrate_bak = h->param.rc.i_vbv_max_bitrate;
1098             if( h->param.rc.i_rc_method == X264_RC_ABR && h->param.rc.i_vbv_buffer_size <= 0 )
1099                 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate * 2;
1100             x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
1101             do h->param.i_level_idc = l->level_idc;
1102                 while( l[1].level_idc && x264_validate_levels( h, 0 ) && l++ );
1103             h->param.rc.i_vbv_max_bitrate = maxrate_bak;
1104         }
1105         else
1106         {
1107             while( l->level_idc && l->level_idc != h->param.i_level_idc )
1108                 l++;
1109             if( l->level_idc == 0 )
1110             {
1111                 x264_log( h, X264_LOG_ERROR, "invalid level_idc: %d\n", h->param.i_level_idc );
1112                 return -1;
1113             }
1114         }
1115         if( h->param.analyse.i_mv_range <= 0 )
1116             h->param.analyse.i_mv_range = l->mv_range >> PARAM_INTERLACED;
1117         else
1118             h->param.analyse.i_mv_range = x264_clip3(h->param.analyse.i_mv_range, 32, 512 >> PARAM_INTERLACED);
1119     }
1120
1121     h->param.analyse.i_weighted_pred = x264_clip3( h->param.analyse.i_weighted_pred, X264_WEIGHTP_NONE, X264_WEIGHTP_SMART );
1122
1123     if( h->param.i_lookahead_threads == X264_THREADS_AUTO )
1124     {
1125         if( h->param.b_sliced_threads )
1126             h->param.i_lookahead_threads = h->param.i_threads;
1127         else
1128         {
1129             /* If we're using much slower lookahead settings than encoding settings, it helps a lot to use
1130              * more lookahead threads.  This typically happens in the first pass of a two-pass encode, so
1131              * try to guess at this sort of case.
1132              *
1133              * Tuned by a little bit of real encoding with the various presets. */
1134             int badapt = h->param.i_bframe_adaptive == X264_B_ADAPT_TRELLIS;
1135             int subme = X264_MIN( h->param.analyse.i_subpel_refine / 3, 3 ) + (h->param.analyse.i_subpel_refine > 1);
1136             int bframes = X264_MIN( (h->param.i_bframe - 1) / 3, 3 );
1137
1138             /* [b-adapt 0/1 vs 2][quantized subme][quantized bframes] */
1139             static const uint8_t lookahead_thread_div[2][5][4] =
1140             {{{6,6,6,6}, {3,3,3,3}, {4,4,4,4}, {6,6,6,6}, {12,12,12,12}},
1141              {{3,2,1,1}, {2,1,1,1}, {4,3,2,1}, {6,4,3,2}, {12, 9, 6, 4}}};
1142
1143             h->param.i_lookahead_threads = h->param.i_threads / lookahead_thread_div[badapt][subme][bframes];
1144             /* Since too many lookahead threads significantly degrades lookahead accuracy, limit auto
1145              * lookahead threads to about 8 macroblock rows high each at worst.  This number is chosen
1146              * pretty much arbitrarily. */
1147             h->param.i_lookahead_threads = X264_MIN( h->param.i_lookahead_threads, h->param.i_height / 128 );
1148         }
1149     }
1150     h->param.i_lookahead_threads = x264_clip3( h->param.i_lookahead_threads, 1, X264_MIN( max_sliced_threads, X264_LOOKAHEAD_THREAD_MAX ) );
1151
1152     if( PARAM_INTERLACED )
1153     {
1154         if( h->param.analyse.i_me_method >= X264_ME_ESA )
1155         {
1156             x264_log( h, X264_LOG_WARNING, "interlace + me=esa is not implemented\n" );
1157             h->param.analyse.i_me_method = X264_ME_UMH;
1158         }
1159         if( h->param.analyse.i_weighted_pred > 0 )
1160         {
1161             x264_log( h, X264_LOG_WARNING, "interlace + weightp is not implemented\n" );
1162             h->param.analyse.i_weighted_pred = X264_WEIGHTP_NONE;
1163         }
1164     }
1165
1166     if( !h->param.analyse.i_weighted_pred && h->param.rc.b_mb_tree && h->param.analyse.b_psy )
1167         h->param.analyse.i_weighted_pred = X264_WEIGHTP_FAKE;
1168
1169     if( h->i_thread_frames > 1 )
1170     {
1171         int r = h->param.analyse.i_mv_range_thread;
1172         int r2;
1173         if( r <= 0 )
1174         {
1175             // half of the available space is reserved and divided evenly among the threads,
1176             // the rest is allocated to whichever thread is far enough ahead to use it.
1177             // reserving more space increases quality for some videos, but costs more time
1178             // in thread synchronization.
1179             int max_range = (h->param.i_height + X264_THREAD_HEIGHT) / h->i_thread_frames - X264_THREAD_HEIGHT;
1180             r = max_range / 2;
1181         }
1182         r = X264_MAX( r, h->param.analyse.i_me_range );
1183         r = X264_MIN( r, h->param.analyse.i_mv_range );
1184         // round up to use the whole mb row
1185         r2 = (r & ~15) + ((-X264_THREAD_HEIGHT) & 15);
1186         if( r2 < r )
1187             r2 += 16;
1188         x264_log( h, X264_LOG_DEBUG, "using mv_range_thread = %d\n", r2 );
1189         h->param.analyse.i_mv_range_thread = r2;
1190     }
1191
1192     if( h->param.rc.f_rate_tolerance < 0 )
1193         h->param.rc.f_rate_tolerance = 0;
1194     if( h->param.rc.f_qblur < 0 )
1195         h->param.rc.f_qblur = 0;
1196     if( h->param.rc.f_complexity_blur < 0 )
1197         h->param.rc.f_complexity_blur = 0;
1198
1199     h->param.i_sps_id &= 31;
1200
1201     h->param.i_nal_hrd = x264_clip3( h->param.i_nal_hrd, X264_NAL_HRD_NONE, X264_NAL_HRD_CBR );
1202
1203     if( h->param.i_nal_hrd && !h->param.rc.i_vbv_buffer_size )
1204     {
1205         x264_log( h, X264_LOG_WARNING, "NAL HRD parameters require VBV parameters\n" );
1206         h->param.i_nal_hrd = X264_NAL_HRD_NONE;
1207     }
1208
1209     if( h->param.i_nal_hrd == X264_NAL_HRD_CBR &&
1210        (h->param.rc.i_bitrate != h->param.rc.i_vbv_max_bitrate || !h->param.rc.i_vbv_max_bitrate) )
1211     {
1212         x264_log( h, X264_LOG_WARNING, "CBR HRD requires constant bitrate\n" );
1213         h->param.i_nal_hrd = X264_NAL_HRD_VBR;
1214     }
1215
1216     if( h->param.i_nal_hrd == X264_NAL_HRD_CBR )
1217         h->param.rc.b_filler = 1;
1218
1219     /* ensure the booleans are 0 or 1 so they can be used in math */
1220 #define BOOLIFY(x) h->param.x = !!h->param.x
1221     BOOLIFY( b_cabac );
1222     BOOLIFY( b_constrained_intra );
1223     BOOLIFY( b_deblocking_filter );
1224     BOOLIFY( b_deterministic );
1225     BOOLIFY( b_sliced_threads );
1226     BOOLIFY( b_interlaced );
1227     BOOLIFY( b_intra_refresh );
1228     BOOLIFY( b_visualize );
1229     BOOLIFY( b_aud );
1230     BOOLIFY( b_repeat_headers );
1231     BOOLIFY( b_annexb );
1232     BOOLIFY( b_vfr_input );
1233     BOOLIFY( b_pulldown );
1234     BOOLIFY( b_tff );
1235     BOOLIFY( b_pic_struct );
1236     BOOLIFY( b_fake_interlaced );
1237     BOOLIFY( b_open_gop );
1238     BOOLIFY( b_bluray_compat );
1239     BOOLIFY( b_stitchable );
1240     BOOLIFY( b_full_recon );
1241     BOOLIFY( b_opencl );
1242     BOOLIFY( b_avcintra_compat );
1243     BOOLIFY( analyse.b_transform_8x8 );
1244     BOOLIFY( analyse.b_weighted_bipred );
1245     BOOLIFY( analyse.b_chroma_me );
1246     BOOLIFY( analyse.b_mixed_references );
1247     BOOLIFY( analyse.b_fast_pskip );
1248     BOOLIFY( analyse.b_dct_decimate );
1249     BOOLIFY( analyse.b_psy );
1250     BOOLIFY( analyse.b_psnr );
1251     BOOLIFY( analyse.b_ssim );
1252     BOOLIFY( rc.b_stat_write );
1253     BOOLIFY( rc.b_stat_read );
1254     BOOLIFY( rc.b_mb_tree );
1255     BOOLIFY( rc.b_filler );
1256 #undef BOOLIFY
1257
1258     return 0;
1259 }
1260
1261 static void mbcmp_init( x264_t *h )
1262 {
1263     int satd = !h->mb.b_lossless && h->param.analyse.i_subpel_refine > 1;
1264     memcpy( h->pixf.mbcmp, satd ? h->pixf.satd : h->pixf.sad_aligned, sizeof(h->pixf.mbcmp) );
1265     memcpy( h->pixf.mbcmp_unaligned, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.mbcmp_unaligned) );
1266     h->pixf.intra_mbcmp_x3_16x16 = satd ? h->pixf.intra_satd_x3_16x16 : h->pixf.intra_sad_x3_16x16;
1267     h->pixf.intra_mbcmp_x3_8x16c = satd ? h->pixf.intra_satd_x3_8x16c : h->pixf.intra_sad_x3_8x16c;
1268     h->pixf.intra_mbcmp_x3_8x8c  = satd ? h->pixf.intra_satd_x3_8x8c  : h->pixf.intra_sad_x3_8x8c;
1269     h->pixf.intra_mbcmp_x3_8x8 = satd ? h->pixf.intra_sa8d_x3_8x8 : h->pixf.intra_sad_x3_8x8;
1270     h->pixf.intra_mbcmp_x3_4x4 = satd ? h->pixf.intra_satd_x3_4x4 : h->pixf.intra_sad_x3_4x4;
1271     h->pixf.intra_mbcmp_x9_4x4 = h->param.b_cpu_independent || h->mb.b_lossless ? NULL
1272                                : satd ? h->pixf.intra_satd_x9_4x4 : h->pixf.intra_sad_x9_4x4;
1273     h->pixf.intra_mbcmp_x9_8x8 = h->param.b_cpu_independent || h->mb.b_lossless ? NULL
1274                                : satd ? h->pixf.intra_sa8d_x9_8x8 : h->pixf.intra_sad_x9_8x8;
1275     satd &= h->param.analyse.i_me_method == X264_ME_TESA;
1276     memcpy( h->pixf.fpelcmp, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.fpelcmp) );
1277     memcpy( h->pixf.fpelcmp_x3, satd ? h->pixf.satd_x3 : h->pixf.sad_x3, sizeof(h->pixf.fpelcmp_x3) );
1278     memcpy( h->pixf.fpelcmp_x4, satd ? h->pixf.satd_x4 : h->pixf.sad_x4, sizeof(h->pixf.fpelcmp_x4) );
1279 }
1280
1281 static void chroma_dsp_init( x264_t *h )
1282 {
1283     memcpy( h->luma2chroma_pixel, x264_luma2chroma_pixel[CHROMA_FORMAT], sizeof(h->luma2chroma_pixel) );
1284
1285     switch( CHROMA_FORMAT )
1286     {
1287         case CHROMA_420:
1288             memcpy( h->predict_chroma, h->predict_8x8c, sizeof(h->predict_chroma) );
1289             h->mc.prefetch_fenc = h->mc.prefetch_fenc_420;
1290             h->loopf.deblock_chroma[0] = h->loopf.deblock_h_chroma_420;
1291             h->loopf.deblock_chroma_intra[0] = h->loopf.deblock_h_chroma_420_intra;
1292             h->loopf.deblock_chroma_mbaff = h->loopf.deblock_chroma_420_mbaff;
1293             h->loopf.deblock_chroma_intra_mbaff = h->loopf.deblock_chroma_420_intra_mbaff;
1294             h->pixf.intra_mbcmp_x3_chroma = h->pixf.intra_mbcmp_x3_8x8c;
1295             h->quantf.coeff_last[DCT_CHROMA_DC] = h->quantf.coeff_last4;
1296             h->quantf.coeff_level_run[DCT_CHROMA_DC] = h->quantf.coeff_level_run4;
1297             break;
1298         case CHROMA_422:
1299             memcpy( h->predict_chroma, h->predict_8x16c, sizeof(h->predict_chroma) );
1300             h->mc.prefetch_fenc = h->mc.prefetch_fenc_422;
1301             h->loopf.deblock_chroma[0] = h->loopf.deblock_h_chroma_422;
1302             h->loopf.deblock_chroma_intra[0] = h->loopf.deblock_h_chroma_422_intra;
1303             h->loopf.deblock_chroma_mbaff = h->loopf.deblock_chroma_422_mbaff;
1304             h->loopf.deblock_chroma_intra_mbaff = h->loopf.deblock_chroma_422_intra_mbaff;
1305             h->pixf.intra_mbcmp_x3_chroma = h->pixf.intra_mbcmp_x3_8x16c;
1306             h->quantf.coeff_last[DCT_CHROMA_DC] = h->quantf.coeff_last8;
1307             h->quantf.coeff_level_run[DCT_CHROMA_DC] = h->quantf.coeff_level_run8;
1308             break;
1309         case CHROMA_444:
1310             h->mc.prefetch_fenc = h->mc.prefetch_fenc_422; /* FIXME: doesn't cover V plane */
1311             h->loopf.deblock_chroma_mbaff = h->loopf.deblock_luma_mbaff;
1312             h->loopf.deblock_chroma_intra_mbaff = h->loopf.deblock_luma_intra_mbaff;
1313             break;
1314     }
1315 }
1316
1317 static void x264_set_aspect_ratio( x264_t *h, x264_param_t *param, int initial )
1318 {
1319     /* VUI */
1320     if( param->vui.i_sar_width > 0 && param->vui.i_sar_height > 0 )
1321     {
1322         uint32_t i_w = param->vui.i_sar_width;
1323         uint32_t i_h = param->vui.i_sar_height;
1324         uint32_t old_w = h->param.vui.i_sar_width;
1325         uint32_t old_h = h->param.vui.i_sar_height;
1326
1327         x264_reduce_fraction( &i_w, &i_h );
1328
1329         while( i_w > 65535 || i_h > 65535 )
1330         {
1331             i_w /= 2;
1332             i_h /= 2;
1333         }
1334
1335         x264_reduce_fraction( &i_w, &i_h );
1336
1337         if( i_w != old_w || i_h != old_h || initial )
1338         {
1339             h->param.vui.i_sar_width = 0;
1340             h->param.vui.i_sar_height = 0;
1341             if( i_w == 0 || i_h == 0 )
1342                 x264_log( h, X264_LOG_WARNING, "cannot create valid sample aspect ratio\n" );
1343             else
1344             {
1345                 x264_log( h, initial?X264_LOG_INFO:X264_LOG_DEBUG, "using SAR=%d/%d\n", i_w, i_h );
1346                 h->param.vui.i_sar_width = i_w;
1347                 h->param.vui.i_sar_height = i_h;
1348             }
1349         }
1350     }
1351 }
1352
1353 /****************************************************************************
1354  * x264_encoder_open:
1355  ****************************************************************************/
1356 x264_t *x264_encoder_open( x264_param_t *param )
1357 {
1358     x264_t *h;
1359     char buf[1000], *p;
1360     int qp, i_slicetype_length;
1361
1362     CHECKED_MALLOCZERO( h, sizeof(x264_t) );
1363
1364     /* Create a copy of param */
1365     memcpy( &h->param, param, sizeof(x264_param_t) );
1366
1367     if( param->param_free )
1368         param->param_free( param );
1369
1370     if( x264_threading_init() )
1371     {
1372         x264_log( h, X264_LOG_ERROR, "unable to initialize threading\n" );
1373         goto fail;
1374     }
1375
1376     if( x264_validate_parameters( h, 1 ) < 0 )
1377         goto fail;
1378
1379     if( h->param.psz_cqm_file )
1380         if( x264_cqm_parse_file( h, h->param.psz_cqm_file ) < 0 )
1381             goto fail;
1382
1383     if( h->param.rc.psz_stat_out )
1384         h->param.rc.psz_stat_out = strdup( h->param.rc.psz_stat_out );
1385     if( h->param.rc.psz_stat_in )
1386         h->param.rc.psz_stat_in = strdup( h->param.rc.psz_stat_in );
1387
1388     x264_reduce_fraction( &h->param.i_fps_num, &h->param.i_fps_den );
1389     x264_reduce_fraction( &h->param.i_timebase_num, &h->param.i_timebase_den );
1390
1391     /* Init x264_t */
1392     h->i_frame = -1;
1393     h->i_frame_num = 0;
1394     h->i_idr_pic_id = 0;
1395
1396     if( (uint64_t)h->param.i_timebase_den * 2 > UINT32_MAX )
1397     {
1398         x264_log( h, X264_LOG_ERROR, "Effective timebase denominator %u exceeds H.264 maximum\n", h->param.i_timebase_den );
1399         goto fail;
1400     }
1401
1402     x264_set_aspect_ratio( h, &h->param, 1 );
1403
1404     x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
1405     x264_pps_init( h->pps, h->param.i_sps_id, &h->param, h->sps );
1406
1407     x264_validate_levels( h, 1 );
1408
1409     h->chroma_qp_table = i_chroma_qp_table + 12 + h->pps->i_chroma_qp_index_offset;
1410
1411     if( x264_cqm_init( h ) < 0 )
1412         goto fail;
1413
1414     h->mb.i_mb_width = h->sps->i_mb_width;
1415     h->mb.i_mb_height = h->sps->i_mb_height;
1416     h->mb.i_mb_count = h->mb.i_mb_width * h->mb.i_mb_height;
1417
1418     h->mb.chroma_h_shift = CHROMA_FORMAT == CHROMA_420 || CHROMA_FORMAT == CHROMA_422;
1419     h->mb.chroma_v_shift = CHROMA_FORMAT == CHROMA_420;
1420
1421     /* Adaptive MBAFF and subme 0 are not supported as we require halving motion
1422      * vectors during prediction, resulting in hpel mvs.
1423      * The chosen solution is to make MBAFF non-adaptive in this case. */
1424     h->mb.b_adaptive_mbaff = PARAM_INTERLACED && h->param.analyse.i_subpel_refine;
1425
1426     /* Init frames. */
1427     if( h->param.i_bframe_adaptive == X264_B_ADAPT_TRELLIS && !h->param.rc.b_stat_read )
1428         h->frames.i_delay = X264_MAX(h->param.i_bframe,3)*4;
1429     else
1430         h->frames.i_delay = h->param.i_bframe;
1431     if( h->param.rc.b_mb_tree || h->param.rc.i_vbv_buffer_size )
1432         h->frames.i_delay = X264_MAX( h->frames.i_delay, h->param.rc.i_lookahead );
1433     i_slicetype_length = h->frames.i_delay;
1434     h->frames.i_delay += h->i_thread_frames - 1;
1435     h->frames.i_delay += h->param.i_sync_lookahead;
1436     h->frames.i_delay += h->param.b_vfr_input;
1437     h->frames.i_bframe_delay = h->param.i_bframe ? (h->param.i_bframe_pyramid ? 2 : 1) : 0;
1438
1439     h->frames.i_max_ref0 = h->param.i_frame_reference;
1440     h->frames.i_max_ref1 = X264_MIN( h->sps->vui.i_num_reorder_frames, h->param.i_frame_reference );
1441     h->frames.i_max_dpb  = h->sps->vui.i_max_dec_frame_buffering;
1442     h->frames.b_have_lowres = !h->param.rc.b_stat_read
1443         && ( h->param.rc.i_rc_method == X264_RC_ABR
1444           || h->param.rc.i_rc_method == X264_RC_CRF
1445           || h->param.i_bframe_adaptive
1446           || h->param.i_scenecut_threshold
1447           || h->param.rc.b_mb_tree
1448           || h->param.analyse.i_weighted_pred );
1449     h->frames.b_have_lowres |= h->param.rc.b_stat_read && h->param.rc.i_vbv_buffer_size > 0;
1450     h->frames.b_have_sub8x8_esa = !!(h->param.analyse.inter & X264_ANALYSE_PSUB8x8);
1451
1452     h->frames.i_last_idr =
1453     h->frames.i_last_keyframe = - h->param.i_keyint_max;
1454     h->frames.i_input    = 0;
1455     h->frames.i_largest_pts = h->frames.i_second_largest_pts = -1;
1456     h->frames.i_poc_last_open_gop = -1;
1457
1458     CHECKED_MALLOCZERO( h->frames.unused[0], (h->frames.i_delay + 3) * sizeof(x264_frame_t *) );
1459     /* Allocate room for max refs plus a few extra just in case. */
1460     CHECKED_MALLOCZERO( h->frames.unused[1], (h->i_thread_frames + X264_REF_MAX + 4) * sizeof(x264_frame_t *) );
1461     CHECKED_MALLOCZERO( h->frames.current, (h->param.i_sync_lookahead + h->param.i_bframe
1462                         + h->i_thread_frames + 3) * sizeof(x264_frame_t *) );
1463     if( h->param.analyse.i_weighted_pred > 0 )
1464         CHECKED_MALLOCZERO( h->frames.blank_unused, h->i_thread_frames * 4 * sizeof(x264_frame_t *) );
1465     h->i_ref[0] = h->i_ref[1] = 0;
1466     h->i_cpb_delay = h->i_coded_fields = h->i_disp_fields = 0;
1467     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);
1468     h->i_disp_fields_last_frame = -1;
1469     x264_rdo_init();
1470
1471     /* init CPU functions */
1472     x264_predict_16x16_init( h->param.cpu, h->predict_16x16 );
1473     x264_predict_8x8c_init( h->param.cpu, h->predict_8x8c );
1474     x264_predict_8x16c_init( h->param.cpu, h->predict_8x16c );
1475     x264_predict_8x8_init( h->param.cpu, h->predict_8x8, &h->predict_8x8_filter );
1476     x264_predict_4x4_init( h->param.cpu, h->predict_4x4 );
1477     x264_pixel_init( h->param.cpu, &h->pixf );
1478     x264_dct_init( h->param.cpu, &h->dctf );
1479     x264_zigzag_init( h->param.cpu, &h->zigzagf_progressive, &h->zigzagf_interlaced );
1480     memcpy( &h->zigzagf, PARAM_INTERLACED ? &h->zigzagf_interlaced : &h->zigzagf_progressive, sizeof(h->zigzagf) );
1481     x264_mc_init( h->param.cpu, &h->mc, h->param.b_cpu_independent );
1482     x264_quant_init( h, h->param.cpu, &h->quantf );
1483     x264_deblock_init( h->param.cpu, &h->loopf, PARAM_INTERLACED );
1484     x264_bitstream_init( h->param.cpu, &h->bsf );
1485     if( h->param.b_cabac )
1486         x264_cabac_init( h );
1487     else
1488         x264_stack_align( x264_cavlc_init, h );
1489
1490     mbcmp_init( h );
1491     chroma_dsp_init( h );
1492
1493     p = buf + sprintf( buf, "using cpu capabilities:" );
1494     for( int i = 0; x264_cpu_names[i].flags; i++ )
1495     {
1496         if( !strcmp(x264_cpu_names[i].name, "SSE")
1497             && h->param.cpu & (X264_CPU_SSE2) )
1498             continue;
1499         if( !strcmp(x264_cpu_names[i].name, "SSE2")
1500             && h->param.cpu & (X264_CPU_SSE2_IS_FAST|X264_CPU_SSE2_IS_SLOW) )
1501             continue;
1502         if( !strcmp(x264_cpu_names[i].name, "SSE3")
1503             && (h->param.cpu & X264_CPU_SSSE3 || !(h->param.cpu & X264_CPU_CACHELINE_64)) )
1504             continue;
1505         if( !strcmp(x264_cpu_names[i].name, "SSE4.1")
1506             && (h->param.cpu & X264_CPU_SSE42) )
1507             continue;
1508         if( !strcmp(x264_cpu_names[i].name, "BMI1")
1509             && (h->param.cpu & X264_CPU_BMI2) )
1510             continue;
1511         if( (h->param.cpu & x264_cpu_names[i].flags) == x264_cpu_names[i].flags
1512             && (!i || x264_cpu_names[i].flags != x264_cpu_names[i-1].flags) )
1513             p += sprintf( p, " %s", x264_cpu_names[i].name );
1514     }
1515     if( !h->param.cpu )
1516         p += sprintf( p, " none!" );
1517     x264_log( h, X264_LOG_INFO, "%s\n", buf );
1518
1519     float *logs = x264_analyse_prepare_costs( h );
1520     if( !logs )
1521         goto fail;
1522     for( qp = X264_MIN( h->param.rc.i_qp_min, QP_MAX_SPEC ); qp <= h->param.rc.i_qp_max; qp++ )
1523         if( x264_analyse_init_costs( h, logs, qp ) )
1524             goto fail;
1525     if( x264_analyse_init_costs( h, logs, X264_LOOKAHEAD_QP ) )
1526         goto fail;
1527     x264_free( logs );
1528
1529     static const uint16_t cost_mv_correct[7] = { 24, 47, 95, 189, 379, 757, 1515 };
1530     /* Checks for known miscompilation issues. */
1531     if( h->cost_mv[X264_LOOKAHEAD_QP][2013] != cost_mv_correct[BIT_DEPTH-8] )
1532     {
1533         x264_log( h, X264_LOG_ERROR, "MV cost test failed: x264 has been miscompiled!\n" );
1534         goto fail;
1535     }
1536
1537     /* Must be volatile or else GCC will optimize it out. */
1538     volatile int temp = 392;
1539     if( x264_clz( temp ) != 23 )
1540     {
1541         x264_log( h, X264_LOG_ERROR, "CLZ test failed: x264 has been miscompiled!\n" );
1542 #if ARCH_X86 || ARCH_X86_64
1543         x264_log( h, X264_LOG_ERROR, "Are you attempting to run an SSE4a/LZCNT-targeted build on a CPU that\n" );
1544         x264_log( h, X264_LOG_ERROR, "doesn't support it?\n" );
1545 #endif
1546         goto fail;
1547     }
1548
1549     h->out.i_nal = 0;
1550     h->out.i_bitstream = X264_MAX( 1000000, h->param.i_width * h->param.i_height * 4
1551         * ( h->param.rc.i_rc_method == X264_RC_ABR ? pow( 0.95, h->param.rc.i_qp_min )
1552           : pow( 0.95, h->param.rc.i_qp_constant ) * X264_MAX( 1, h->param.rc.f_ip_factor )));
1553
1554     h->nal_buffer_size = h->out.i_bitstream * 3/2 + 4 + 64; /* +4 for startcode, +64 for nal_escape assembly padding */
1555     CHECKED_MALLOC( h->nal_buffer, h->nal_buffer_size );
1556
1557     CHECKED_MALLOC( h->reconfig_h, sizeof(x264_t) );
1558
1559     if( h->param.i_threads > 1 &&
1560         x264_threadpool_init( &h->threadpool, h->param.i_threads, (void*)x264_encoder_thread_init, h ) )
1561         goto fail;
1562     if( h->param.i_lookahead_threads > 1 &&
1563         x264_threadpool_init( &h->lookaheadpool, h->param.i_lookahead_threads, NULL, NULL ) )
1564         goto fail;
1565
1566 #if HAVE_OPENCL
1567     if( h->param.b_opencl )
1568     {
1569         h->opencl.ocl = x264_opencl_load_library();
1570         if( !h->opencl.ocl )
1571         {
1572             x264_log( h, X264_LOG_WARNING, "failed to load OpenCL\n" );
1573             h->param.b_opencl = 0;
1574         }
1575     }
1576 #endif
1577
1578     h->thread[0] = h;
1579     for( int i = 1; i < h->param.i_threads + !!h->param.i_sync_lookahead; i++ )
1580         CHECKED_MALLOC( h->thread[i], sizeof(x264_t) );
1581     if( h->param.i_lookahead_threads > 1 )
1582         for( int i = 0; i < h->param.i_lookahead_threads; i++ )
1583         {
1584             CHECKED_MALLOC( h->lookahead_thread[i], sizeof(x264_t) );
1585             *h->lookahead_thread[i] = *h;
1586         }
1587     *h->reconfig_h = *h;
1588
1589     for( int i = 0; i < h->param.i_threads; i++ )
1590     {
1591         int init_nal_count = h->param.i_slice_count + 3;
1592         int allocate_threadlocal_data = !h->param.b_sliced_threads || !i;
1593         if( i > 0 )
1594             *h->thread[i] = *h;
1595
1596         if( x264_pthread_mutex_init( &h->thread[i]->mutex, NULL ) )
1597             goto fail;
1598         if( x264_pthread_cond_init( &h->thread[i]->cv, NULL ) )
1599             goto fail;
1600
1601         if( allocate_threadlocal_data )
1602         {
1603             h->thread[i]->fdec = x264_frame_pop_unused( h, 1 );
1604             if( !h->thread[i]->fdec )
1605                 goto fail;
1606         }
1607         else
1608             h->thread[i]->fdec = h->thread[0]->fdec;
1609
1610         CHECKED_MALLOC( h->thread[i]->out.p_bitstream, h->out.i_bitstream );
1611         /* Start each thread with room for init_nal_count NAL units; it'll realloc later if needed. */
1612         CHECKED_MALLOC( h->thread[i]->out.nal, init_nal_count*sizeof(x264_nal_t) );
1613         h->thread[i]->out.i_nals_allocated = init_nal_count;
1614
1615         if( allocate_threadlocal_data && x264_macroblock_cache_allocate( h->thread[i] ) < 0 )
1616             goto fail;
1617     }
1618
1619 #if HAVE_OPENCL
1620     if( h->param.b_opencl && x264_opencl_lookahead_init( h ) < 0 )
1621         h->param.b_opencl = 0;
1622 #endif
1623
1624     if( x264_lookahead_init( h, i_slicetype_length ) )
1625         goto fail;
1626
1627     for( int i = 0; i < h->param.i_threads; i++ )
1628         if( x264_macroblock_thread_allocate( h->thread[i], 0 ) < 0 )
1629             goto fail;
1630
1631     if( x264_ratecontrol_new( h ) < 0 )
1632         goto fail;
1633
1634     if( h->param.i_nal_hrd )
1635     {
1636         x264_log( h, X264_LOG_DEBUG, "HRD bitrate: %i bits/sec\n", h->sps->vui.hrd.i_bit_rate_unscaled );
1637         x264_log( h, X264_LOG_DEBUG, "CPB size: %i bits\n", h->sps->vui.hrd.i_cpb_size_unscaled );
1638     }
1639
1640     if( h->param.psz_dump_yuv )
1641     {
1642         /* create or truncate the reconstructed video file */
1643         FILE *f = x264_fopen( h->param.psz_dump_yuv, "w" );
1644         if( !f )
1645         {
1646             x264_log( h, X264_LOG_ERROR, "dump_yuv: can't write to %s\n", h->param.psz_dump_yuv );
1647             goto fail;
1648         }
1649         else if( !x264_is_regular_file( f ) )
1650         {
1651             x264_log( h, X264_LOG_ERROR, "dump_yuv: incompatible with non-regular file %s\n", h->param.psz_dump_yuv );
1652             goto fail;
1653         }
1654         fclose( f );
1655     }
1656
1657     const char *profile = h->sps->i_profile_idc == PROFILE_BASELINE ? "Constrained Baseline" :
1658                           h->sps->i_profile_idc == PROFILE_MAIN ? "Main" :
1659                           h->sps->i_profile_idc == PROFILE_HIGH ? "High" :
1660                           h->sps->i_profile_idc == PROFILE_HIGH10 ? (h->sps->b_constraint_set3 == 1 ? "High 10 Intra" : "High 10") :
1661                           h->sps->i_profile_idc == PROFILE_HIGH422 ? (h->sps->b_constraint_set3 == 1 ? "High 4:2:2 Intra" : "High 4:2:2") :
1662                           h->sps->b_constraint_set3 == 1 ? "High 4:4:4 Intra" : "High 4:4:4 Predictive";
1663     char level[4];
1664     snprintf( level, sizeof(level), "%d.%d", h->sps->i_level_idc/10, h->sps->i_level_idc%10 );
1665     if( h->sps->i_level_idc == 9 || ( h->sps->i_level_idc == 11 && h->sps->b_constraint_set3 &&
1666         (h->sps->i_profile_idc == PROFILE_BASELINE || h->sps->i_profile_idc == PROFILE_MAIN) ) )
1667         strcpy( level, "1b" );
1668
1669     if( h->sps->i_profile_idc < PROFILE_HIGH10 )
1670     {
1671         x264_log( h, X264_LOG_INFO, "profile %s, level %s\n",
1672             profile, level );
1673     }
1674     else
1675     {
1676         static const char * const subsampling[4] = { "4:0:0", "4:2:0", "4:2:2", "4:4:4" };
1677         x264_log( h, X264_LOG_INFO, "profile %s, level %s, %s %d-bit\n",
1678             profile, level, subsampling[CHROMA_FORMAT], BIT_DEPTH );
1679     }
1680
1681     return h;
1682 fail:
1683     x264_free( h );
1684     return NULL;
1685 }
1686
1687 /****************************************************************************/
1688 static int x264_encoder_try_reconfig( x264_t *h, x264_param_t *param, int *rc_reconfig )
1689 {
1690     *rc_reconfig = 0;
1691     x264_set_aspect_ratio( h, param, 0 );
1692 #define COPY(var) h->param.var = param->var
1693     COPY( i_frame_reference ); // but never uses more refs than initially specified
1694     COPY( i_bframe_bias );
1695     if( h->param.i_scenecut_threshold )
1696         COPY( i_scenecut_threshold ); // can't turn it on or off, only vary the threshold
1697     COPY( b_deblocking_filter );
1698     COPY( i_deblocking_filter_alphac0 );
1699     COPY( i_deblocking_filter_beta );
1700     COPY( i_frame_packing );
1701     COPY( analyse.inter );
1702     COPY( analyse.intra );
1703     COPY( analyse.i_direct_mv_pred );
1704     /* Scratch buffer prevents me_range from being increased for esa/tesa */
1705     if( h->param.analyse.i_me_method < X264_ME_ESA || param->analyse.i_me_range < h->param.analyse.i_me_range )
1706         COPY( analyse.i_me_range );
1707     COPY( analyse.i_noise_reduction );
1708     /* We can't switch out of subme=0 during encoding. */
1709     if( h->param.analyse.i_subpel_refine )
1710         COPY( analyse.i_subpel_refine );
1711     COPY( analyse.i_trellis );
1712     COPY( analyse.b_chroma_me );
1713     COPY( analyse.b_dct_decimate );
1714     COPY( analyse.b_fast_pskip );
1715     COPY( analyse.b_mixed_references );
1716     COPY( analyse.f_psy_rd );
1717     COPY( analyse.f_psy_trellis );
1718     COPY( crop_rect );
1719     // can only twiddle these if they were enabled to begin with:
1720     if( h->param.analyse.i_me_method >= X264_ME_ESA || param->analyse.i_me_method < X264_ME_ESA )
1721         COPY( analyse.i_me_method );
1722     if( h->param.analyse.i_me_method >= X264_ME_ESA && !h->frames.b_have_sub8x8_esa )
1723         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
1724     if( h->pps->b_transform_8x8_mode )
1725         COPY( analyse.b_transform_8x8 );
1726     if( h->frames.i_max_ref1 > 1 )
1727         COPY( i_bframe_pyramid );
1728     COPY( i_slice_max_size );
1729     COPY( i_slice_max_mbs );
1730     COPY( i_slice_min_mbs );
1731     COPY( i_slice_count );
1732     COPY( i_slice_count_max );
1733     COPY( b_tff );
1734
1735     /* VBV can't be turned on if it wasn't on to begin with */
1736     if( h->param.rc.i_vbv_max_bitrate > 0 && h->param.rc.i_vbv_buffer_size > 0 &&
1737           param->rc.i_vbv_max_bitrate > 0 &&   param->rc.i_vbv_buffer_size > 0 )
1738     {
1739         *rc_reconfig |= h->param.rc.i_vbv_max_bitrate != param->rc.i_vbv_max_bitrate;
1740         *rc_reconfig |= h->param.rc.i_vbv_buffer_size != param->rc.i_vbv_buffer_size;
1741         *rc_reconfig |= h->param.rc.i_bitrate != param->rc.i_bitrate;
1742         COPY( rc.i_vbv_max_bitrate );
1743         COPY( rc.i_vbv_buffer_size );
1744         COPY( rc.i_bitrate );
1745     }
1746     *rc_reconfig |= h->param.rc.f_rf_constant != param->rc.f_rf_constant;
1747     *rc_reconfig |= h->param.rc.f_rf_constant_max != param->rc.f_rf_constant_max;
1748     COPY( rc.f_rf_constant );
1749     COPY( rc.f_rf_constant_max );
1750 #undef COPY
1751
1752     return x264_validate_parameters( h, 0 );
1753 }
1754
1755 int x264_encoder_reconfig_apply( x264_t *h, x264_param_t *param )
1756 {
1757     int rc_reconfig;
1758     int ret = x264_encoder_try_reconfig( h, param, &rc_reconfig );
1759
1760     mbcmp_init( h );
1761     if( !ret )
1762         x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
1763
1764     /* Supported reconfiguration options (1-pass only):
1765      * vbv-maxrate
1766      * vbv-bufsize
1767      * crf
1768      * bitrate (CBR only) */
1769     if( !ret && rc_reconfig )
1770         x264_ratecontrol_init_reconfigurable( h, 0 );
1771
1772     return ret;
1773 }
1774
1775 /****************************************************************************
1776  * x264_encoder_reconfig:
1777  ****************************************************************************/
1778 int x264_encoder_reconfig( x264_t *h, x264_param_t *param )
1779 {
1780     h = h->thread[h->thread[0]->i_thread_phase];
1781     x264_param_t param_save = h->reconfig_h->param;
1782     h->reconfig_h->param = h->param;
1783
1784     int rc_reconfig;
1785     int ret = x264_encoder_try_reconfig( h->reconfig_h, param, &rc_reconfig );
1786     if( !ret )
1787         h->reconfig = 1;
1788     else
1789         h->reconfig_h->param = param_save;
1790
1791     return ret;
1792 }
1793
1794 /****************************************************************************
1795  * x264_encoder_parameters:
1796  ****************************************************************************/
1797 void x264_encoder_parameters( x264_t *h, x264_param_t *param )
1798 {
1799     memcpy( param, &h->thread[h->i_thread_phase]->param, sizeof(x264_param_t) );
1800 }
1801
1802 /* internal usage */
1803 static void x264_nal_start( x264_t *h, int i_type, int i_ref_idc )
1804 {
1805     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1806
1807     nal->i_ref_idc        = i_ref_idc;
1808     nal->i_type           = i_type;
1809     nal->b_long_startcode = 1;
1810
1811     nal->i_payload= 0;
1812     nal->p_payload= &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8];
1813     nal->i_padding= 0;
1814 }
1815
1816 /* if number of allocated nals is not enough, re-allocate a larger one. */
1817 static int x264_nal_check_buffer( x264_t *h )
1818 {
1819     if( h->out.i_nal >= h->out.i_nals_allocated )
1820     {
1821         x264_nal_t *new_out = x264_malloc( sizeof(x264_nal_t) * (h->out.i_nals_allocated*2) );
1822         if( !new_out )
1823             return -1;
1824         memcpy( new_out, h->out.nal, sizeof(x264_nal_t) * (h->out.i_nals_allocated) );
1825         x264_free( h->out.nal );
1826         h->out.nal = new_out;
1827         h->out.i_nals_allocated *= 2;
1828     }
1829     return 0;
1830 }
1831
1832 static int x264_nal_end( x264_t *h )
1833 {
1834     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1835     uint8_t *end = &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8];
1836     nal->i_payload = end - nal->p_payload;
1837     /* Assembly implementation of nal_escape reads past the end of the input.
1838      * While undefined padding wouldn't actually affect the output, it makes valgrind unhappy. */
1839     memset( end, 0xff, 64 );
1840     if( h->param.nalu_process )
1841         h->param.nalu_process( h, nal, h->fenc->opaque );
1842     h->out.i_nal++;
1843
1844     return x264_nal_check_buffer( h );
1845 }
1846
1847 static int x264_check_encapsulated_buffer( x264_t *h, x264_t *h0, int start,
1848                                            int previous_nal_size, int necessary_size )
1849 {
1850     if( h0->nal_buffer_size < necessary_size )
1851     {
1852         necessary_size *= 2;
1853         uint8_t *buf = x264_malloc( necessary_size );
1854         if( !buf )
1855             return -1;
1856         if( previous_nal_size )
1857             memcpy( buf, h0->nal_buffer, previous_nal_size );
1858
1859         intptr_t delta = buf - h0->nal_buffer;
1860         for( int i = 0; i < start; i++ )
1861             h->out.nal[i].p_payload += delta;
1862
1863         x264_free( h0->nal_buffer );
1864         h0->nal_buffer = buf;
1865         h0->nal_buffer_size = necessary_size;
1866     }
1867
1868     return 0;
1869 }
1870
1871 static int x264_encoder_encapsulate_nals( x264_t *h, int start )
1872 {
1873     x264_t *h0 = h->thread[0];
1874     int nal_size = 0, previous_nal_size = 0;
1875
1876     if( h->param.nalu_process )
1877     {
1878         for( int i = start; i < h->out.i_nal; i++ )
1879             nal_size += h->out.nal[i].i_payload;
1880         return nal_size;
1881     }
1882
1883     for( int i = 0; i < start; i++ )
1884         previous_nal_size += h->out.nal[i].i_payload;
1885
1886     for( int i = start; i < h->out.i_nal; i++ )
1887         nal_size += h->out.nal[i].i_payload;
1888
1889     /* Worst-case NAL unit escaping: reallocate the buffer if it's too small. */
1890     int necessary_size = previous_nal_size + nal_size * 3/2 + h->out.i_nal * 4 + 4 + 64;
1891     for( int i = start; i < h->out.i_nal; i++ )
1892         necessary_size += h->out.nal[i].i_padding;
1893     if( x264_check_encapsulated_buffer( h, h0, start, previous_nal_size, necessary_size ) )
1894         return -1;
1895
1896     uint8_t *nal_buffer = h0->nal_buffer + previous_nal_size;
1897
1898     for( int i = start; i < h->out.i_nal; i++ )
1899     {
1900         int old_payload_len = h->out.nal[i].i_payload;
1901         h->out.nal[i].b_long_startcode = !i || h->out.nal[i].i_type == NAL_SPS || h->out.nal[i].i_type == NAL_PPS ||
1902                                          h->param.b_avcintra_compat;
1903         x264_nal_encode( h, nal_buffer, &h->out.nal[i] );
1904         nal_buffer += h->out.nal[i].i_payload;
1905         if( h->param.b_avcintra_compat )
1906         {
1907             h->out.nal[i].i_padding -= h->out.nal[i].i_payload - (old_payload_len + NALU_OVERHEAD);
1908             if( h->out.nal[i].i_padding > 0 )
1909             {
1910                 memset( nal_buffer, 0, h->out.nal[i].i_padding );
1911                 nal_buffer += h->out.nal[i].i_padding;
1912                 h->out.nal[i].i_payload += h->out.nal[i].i_padding;
1913             }
1914             h->out.nal[i].i_padding = X264_MAX( h->out.nal[i].i_padding, 0 );
1915         }
1916     }
1917
1918     x264_emms();
1919
1920     return nal_buffer - (h0->nal_buffer + previous_nal_size);
1921 }
1922
1923 /****************************************************************************
1924  * x264_encoder_headers:
1925  ****************************************************************************/
1926 int x264_encoder_headers( x264_t *h, x264_nal_t **pp_nal, int *pi_nal )
1927 {
1928     int frame_size = 0;
1929     /* init bitstream context */
1930     h->out.i_nal = 0;
1931     bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
1932
1933     /* Write SEI, SPS and PPS. */
1934
1935     /* generate sequence parameters */
1936     x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
1937     x264_sps_write( &h->out.bs, h->sps );
1938     if( x264_nal_end( h ) )
1939         return -1;
1940
1941     /* generate picture parameters */
1942     x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
1943     x264_pps_write( &h->out.bs, h->sps, h->pps );
1944     if( x264_nal_end( h ) )
1945         return -1;
1946
1947     /* identify ourselves */
1948     x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
1949     if( x264_sei_version_write( h, &h->out.bs ) )
1950         return -1;
1951     if( x264_nal_end( h ) )
1952         return -1;
1953
1954     frame_size = x264_encoder_encapsulate_nals( h, 0 );
1955     if( frame_size < 0 )
1956         return -1;
1957
1958     /* now set output*/
1959     *pi_nal = h->out.i_nal;
1960     *pp_nal = &h->out.nal[0];
1961     h->out.i_nal = 0;
1962
1963     return frame_size;
1964 }
1965
1966 /* Check to see whether we have chosen a reference list ordering different
1967  * from the standard's default. */
1968 static inline void x264_reference_check_reorder( x264_t *h )
1969 {
1970     /* The reorder check doesn't check for missing frames, so just
1971      * force a reorder if one of the reference list is corrupt. */
1972     for( int i = 0; h->frames.reference[i]; i++ )
1973         if( h->frames.reference[i]->b_corrupt )
1974         {
1975             h->b_ref_reorder[0] = 1;
1976             return;
1977         }
1978     for( int list = 0; list <= (h->sh.i_type == SLICE_TYPE_B); list++ )
1979         for( int i = 0; i < h->i_ref[list] - 1; i++ )
1980         {
1981             int framenum_diff = h->fref[list][i+1]->i_frame_num - h->fref[list][i]->i_frame_num;
1982             int poc_diff = h->fref[list][i+1]->i_poc - h->fref[list][i]->i_poc;
1983             /* P and B-frames use different default orders. */
1984             if( h->sh.i_type == SLICE_TYPE_P ? framenum_diff > 0 : list == 1 ? poc_diff < 0 : poc_diff > 0 )
1985             {
1986                 h->b_ref_reorder[list] = 1;
1987                 return;
1988             }
1989         }
1990 }
1991
1992 /* return -1 on failure, else return the index of the new reference frame */
1993 int x264_weighted_reference_duplicate( x264_t *h, int i_ref, const x264_weight_t *w )
1994 {
1995     int i = h->i_ref[0];
1996     int j = 1;
1997     x264_frame_t *newframe;
1998     if( i <= 1 ) /* empty list, definitely can't duplicate frame */
1999         return -1;
2000
2001     //Duplication is only used in X264_WEIGHTP_SMART
2002     if( h->param.analyse.i_weighted_pred != X264_WEIGHTP_SMART )
2003         return -1;
2004
2005     /* Duplication is a hack to compensate for crappy rounding in motion compensation.
2006      * With high bit depth, it's not worth doing, so turn it off except in the case of
2007      * unweighted dupes. */
2008     if( BIT_DEPTH > 8 && w != x264_weight_none )
2009         return -1;
2010
2011     newframe = x264_frame_pop_blank_unused( h );
2012     if( !newframe )
2013         return -1;
2014
2015     //FIXME: probably don't need to copy everything
2016     *newframe = *h->fref[0][i_ref];
2017     newframe->i_reference_count = 1;
2018     newframe->orig = h->fref[0][i_ref];
2019     newframe->b_duplicate = 1;
2020     memcpy( h->fenc->weight[j], w, sizeof(h->fenc->weight[i]) );
2021
2022     /* shift the frames to make space for the dupe. */
2023     h->b_ref_reorder[0] = 1;
2024     if( h->i_ref[0] < X264_REF_MAX )
2025         ++h->i_ref[0];
2026     h->fref[0][X264_REF_MAX-1] = NULL;
2027     x264_frame_unshift( &h->fref[0][j], newframe );
2028
2029     return j;
2030 }
2031
2032 static void x264_weighted_pred_init( x264_t *h )
2033 {
2034     /* for now no analysis and set all weights to nothing */
2035     for( int i_ref = 0; i_ref < h->i_ref[0]; i_ref++ )
2036         h->fenc->weighted[i_ref] = h->fref[0][i_ref]->filtered[0][0];
2037
2038     // FIXME: This only supports weighting of one reference frame
2039     // and duplicates of that frame.
2040     h->fenc->i_lines_weighted = 0;
2041
2042     for( int i_ref = 0; i_ref < (h->i_ref[0] << SLICE_MBAFF); i_ref++ )
2043         for( int i = 0; i < 3; i++ )
2044             h->sh.weight[i_ref][i].weightfn = NULL;
2045
2046
2047     if( h->sh.i_type != SLICE_TYPE_P || h->param.analyse.i_weighted_pred <= 0 )
2048         return;
2049
2050     int i_padv = PADV << PARAM_INTERLACED;
2051     int denom = -1;
2052     int weightplane[2] = { 0, 0 };
2053     int buffer_next = 0;
2054     for( int i = 0; i < 3; i++ )
2055     {
2056         for( int j = 0; j < h->i_ref[0]; j++ )
2057         {
2058             if( h->fenc->weight[j][i].weightfn )
2059             {
2060                 h->sh.weight[j][i] = h->fenc->weight[j][i];
2061                 // if weight is useless, don't write it to stream
2062                 if( h->sh.weight[j][i].i_scale == 1<<h->sh.weight[j][i].i_denom && h->sh.weight[j][i].i_offset == 0 )
2063                     h->sh.weight[j][i].weightfn = NULL;
2064                 else
2065                 {
2066                     if( !weightplane[!!i] )
2067                     {
2068                         weightplane[!!i] = 1;
2069                         h->sh.weight[0][!!i].i_denom = denom = h->sh.weight[j][i].i_denom;
2070                         assert( x264_clip3( denom, 0, 7 ) == denom );
2071                     }
2072
2073                     assert( h->sh.weight[j][i].i_denom == denom );
2074                     if( !i )
2075                     {
2076                         h->fenc->weighted[j] = h->mb.p_weight_buf[buffer_next++] + h->fenc->i_stride[0] * i_padv + PADH;
2077                         //scale full resolution frame
2078                         if( h->param.i_threads == 1 )
2079                         {
2080                             pixel *src = h->fref[0][j]->filtered[0][0] - h->fref[0][j]->i_stride[0]*i_padv - PADH;
2081                             pixel *dst = h->fenc->weighted[j] - h->fenc->i_stride[0]*i_padv - PADH;
2082                             int stride = h->fenc->i_stride[0];
2083                             int width = h->fenc->i_width[0] + PADH*2;
2084                             int height = h->fenc->i_lines[0] + i_padv*2;
2085                             x264_weight_scale_plane( h, dst, stride, src, stride, width, height, &h->sh.weight[j][0] );
2086                             h->fenc->i_lines_weighted = height;
2087                         }
2088                     }
2089                 }
2090             }
2091         }
2092     }
2093
2094     if( weightplane[1] )
2095         for( int i = 0; i < h->i_ref[0]; i++ )
2096         {
2097             if( h->sh.weight[i][1].weightfn && !h->sh.weight[i][2].weightfn )
2098             {
2099                 h->sh.weight[i][2].i_scale = 1 << h->sh.weight[0][1].i_denom;
2100                 h->sh.weight[i][2].i_offset = 0;
2101             }
2102             else if( h->sh.weight[i][2].weightfn && !h->sh.weight[i][1].weightfn )
2103             {
2104                 h->sh.weight[i][1].i_scale = 1 << h->sh.weight[0][1].i_denom;
2105                 h->sh.weight[i][1].i_offset = 0;
2106             }
2107         }
2108
2109     if( !weightplane[0] )
2110         h->sh.weight[0][0].i_denom = 0;
2111     if( !weightplane[1] )
2112         h->sh.weight[0][1].i_denom = 0;
2113     h->sh.weight[0][2].i_denom = h->sh.weight[0][1].i_denom;
2114 }
2115
2116 static inline int x264_reference_distance( x264_t *h, x264_frame_t *frame )
2117 {
2118     if( h->param.i_frame_packing == 5 )
2119         return abs((h->fenc->i_frame&~1) - (frame->i_frame&~1)) +
2120                   ((h->fenc->i_frame&1) != (frame->i_frame&1));
2121     else
2122         return abs(h->fenc->i_frame - frame->i_frame);
2123 }
2124
2125 static inline void x264_reference_build_list( x264_t *h, int i_poc )
2126 {
2127     int b_ok;
2128
2129     /* build ref list 0/1 */
2130     h->mb.pic.i_fref[0] = h->i_ref[0] = 0;
2131     h->mb.pic.i_fref[1] = h->i_ref[1] = 0;
2132     if( h->sh.i_type == SLICE_TYPE_I )
2133         return;
2134
2135     for( int i = 0; h->frames.reference[i]; i++ )
2136     {
2137         if( h->frames.reference[i]->b_corrupt )
2138             continue;
2139         if( h->frames.reference[i]->i_poc < i_poc )
2140             h->fref[0][h->i_ref[0]++] = h->frames.reference[i];
2141         else if( h->frames.reference[i]->i_poc > i_poc )
2142             h->fref[1][h->i_ref[1]++] = h->frames.reference[i];
2143     }
2144
2145     /* Order reference lists by distance from the current frame. */
2146     for( int list = 0; list < 2; list++ )
2147     {
2148         h->fref_nearest[list] = h->fref[list][0];
2149         do
2150         {
2151             b_ok = 1;
2152             for( int i = 0; i < h->i_ref[list] - 1; i++ )
2153             {
2154                 if( list ? h->fref[list][i+1]->i_poc < h->fref_nearest[list]->i_poc
2155                          : h->fref[list][i+1]->i_poc > h->fref_nearest[list]->i_poc )
2156                     h->fref_nearest[list] = h->fref[list][i+1];
2157                 if( x264_reference_distance( h, h->fref[list][i] ) > x264_reference_distance( h, h->fref[list][i+1] ) )
2158                 {
2159                     XCHG( x264_frame_t*, h->fref[list][i], h->fref[list][i+1] );
2160                     b_ok = 0;
2161                     break;
2162                 }
2163             }
2164         } while( !b_ok );
2165     }
2166
2167     if( h->sh.i_mmco_remove_from_end )
2168         for( int i = h->i_ref[0]-1; i >= h->i_ref[0] - h->sh.i_mmco_remove_from_end; i-- )
2169         {
2170             int diff = h->i_frame_num - h->fref[0][i]->i_frame_num;
2171             h->sh.mmco[h->sh.i_mmco_command_count].i_poc = h->fref[0][i]->i_poc;
2172             h->sh.mmco[h->sh.i_mmco_command_count++].i_difference_of_pic_nums = diff;
2173         }
2174
2175     x264_reference_check_reorder( h );
2176
2177     h->i_ref[1] = X264_MIN( h->i_ref[1], h->frames.i_max_ref1 );
2178     h->i_ref[0] = X264_MIN( h->i_ref[0], h->frames.i_max_ref0 );
2179     h->i_ref[0] = X264_MIN( h->i_ref[0], h->param.i_frame_reference ); // if reconfig() has lowered the limit
2180
2181     /* For Blu-ray compliance, don't reference frames outside of the minigop. */
2182     if( IS_X264_TYPE_B( h->fenc->i_type ) && h->param.b_bluray_compat )
2183         h->i_ref[0] = X264_MIN( h->i_ref[0], IS_X264_TYPE_B( h->fref[0][0]->i_type ) + 1 );
2184
2185     /* add duplicates */
2186     if( h->fenc->i_type == X264_TYPE_P )
2187     {
2188         int idx = -1;
2189         if( h->param.analyse.i_weighted_pred >= X264_WEIGHTP_SIMPLE )
2190         {
2191             x264_weight_t w[3];
2192             w[1].weightfn = w[2].weightfn = NULL;
2193             if( h->param.rc.b_stat_read )
2194                 x264_ratecontrol_set_weights( h, h->fenc );
2195
2196             if( !h->fenc->weight[0][0].weightfn )
2197             {
2198                 h->fenc->weight[0][0].i_denom = 0;
2199                 SET_WEIGHT( w[0], 1, 1, 0, -1 );
2200                 idx = x264_weighted_reference_duplicate( h, 0, w );
2201             }
2202             else
2203             {
2204                 if( h->fenc->weight[0][0].i_scale == 1<<h->fenc->weight[0][0].i_denom )
2205                 {
2206                     SET_WEIGHT( h->fenc->weight[0][0], 1, 1, 0, h->fenc->weight[0][0].i_offset );
2207                 }
2208                 x264_weighted_reference_duplicate( h, 0, x264_weight_none );
2209                 if( h->fenc->weight[0][0].i_offset > -128 )
2210                 {
2211                     w[0] = h->fenc->weight[0][0];
2212                     w[0].i_offset--;
2213                     h->mc.weight_cache( h, &w[0] );
2214                     idx = x264_weighted_reference_duplicate( h, 0, w );
2215                 }
2216             }
2217         }
2218         h->mb.ref_blind_dupe = idx;
2219     }
2220
2221     assert( h->i_ref[0] + h->i_ref[1] <= X264_REF_MAX );
2222     h->mb.pic.i_fref[0] = h->i_ref[0];
2223     h->mb.pic.i_fref[1] = h->i_ref[1];
2224 }
2225
2226 static void x264_fdec_filter_row( x264_t *h, int mb_y, int pass )
2227 {
2228     /* mb_y is the mb to be encoded next, not the mb to be filtered here */
2229     int b_hpel = h->fdec->b_kept_as_ref;
2230     int b_deblock = h->sh.i_disable_deblocking_filter_idc != 1;
2231     int b_end = mb_y == h->i_threadslice_end;
2232     int b_measure_quality = 1;
2233     int min_y = mb_y - (1 << SLICE_MBAFF);
2234     int b_start = min_y == h->i_threadslice_start;
2235     /* Even in interlaced mode, deblocking never modifies more than 4 pixels
2236      * above each MB, as bS=4 doesn't happen for the top of interlaced mbpairs. */
2237     int minpix_y = min_y*16 - 4 * !b_start;
2238     int maxpix_y = mb_y*16 - 4 * !b_end;
2239     b_deblock &= b_hpel || h->param.b_full_recon || h->param.psz_dump_yuv;
2240     if( h->param.b_sliced_threads )
2241     {
2242         switch( pass )
2243         {
2244             /* During encode: only do deblock if asked for */
2245             default:
2246             case 0:
2247                 b_deblock &= h->param.b_full_recon;
2248                 b_hpel = 0;
2249                 break;
2250             /* During post-encode pass: do deblock if not done yet, do hpel for all
2251              * rows except those between slices. */
2252             case 1:
2253                 b_deblock &= !h->param.b_full_recon;
2254                 b_hpel &= !(b_start && min_y > 0);
2255                 b_measure_quality = 0;
2256                 break;
2257             /* Final pass: do the rows between slices in sequence. */
2258             case 2:
2259                 b_deblock = 0;
2260                 b_measure_quality = 0;
2261                 break;
2262         }
2263     }
2264     if( mb_y & SLICE_MBAFF )
2265         return;
2266     if( min_y < h->i_threadslice_start )
2267         return;
2268
2269     if( b_deblock )
2270         for( int y = min_y; y < mb_y; y += (1 << SLICE_MBAFF) )
2271             x264_frame_deblock_row( h, y );
2272
2273     /* FIXME: Prediction requires different borders for interlaced/progressive mc,
2274      * but the actual image data is equivalent. For now, maintain this
2275      * consistency by copying deblocked pixels between planes. */
2276     if( PARAM_INTERLACED && (!h->param.b_sliced_threads || pass == 1) )
2277         for( int p = 0; p < h->fdec->i_plane; p++ )
2278             for( int i = minpix_y>>(CHROMA_V_SHIFT && p); i < maxpix_y>>(CHROMA_V_SHIFT && p); i++ )
2279                 memcpy( h->fdec->plane_fld[p] + i*h->fdec->i_stride[p],
2280                         h->fdec->plane[p]     + i*h->fdec->i_stride[p],
2281                         h->mb.i_mb_width*16*sizeof(pixel) );
2282
2283     if( h->fdec->b_kept_as_ref && (!h->param.b_sliced_threads || pass == 1) )
2284         x264_frame_expand_border( h, h->fdec, min_y );
2285     if( b_hpel )
2286     {
2287         int end = mb_y == h->mb.i_mb_height;
2288         /* Can't do hpel until the previous slice is done encoding. */
2289         if( h->param.analyse.i_subpel_refine )
2290         {
2291             x264_frame_filter( h, h->fdec, min_y, end );
2292             x264_frame_expand_border_filtered( h, h->fdec, min_y, end );
2293         }
2294     }
2295
2296     if( SLICE_MBAFF && pass == 0 )
2297         for( int i = 0; i < 3; i++ )
2298         {
2299             XCHG( pixel *, h->intra_border_backup[0][i], h->intra_border_backup[3][i] );
2300             XCHG( pixel *, h->intra_border_backup[1][i], h->intra_border_backup[4][i] );
2301         }
2302
2303     if( h->i_thread_frames > 1 && h->fdec->b_kept_as_ref )
2304         x264_frame_cond_broadcast( h->fdec, mb_y*16 + (b_end ? 10000 : -(X264_THREAD_HEIGHT << SLICE_MBAFF)) );
2305
2306     if( b_measure_quality )
2307     {
2308         maxpix_y = X264_MIN( maxpix_y, h->param.i_height );
2309         if( h->param.analyse.b_psnr )
2310         {
2311             for( int p = 0; p < (CHROMA444 ? 3 : 1); p++ )
2312                 h->stat.frame.i_ssd[p] += x264_pixel_ssd_wxh( &h->pixf,
2313                     h->fdec->plane[p] + minpix_y * h->fdec->i_stride[p], h->fdec->i_stride[p],
2314                     h->fenc->plane[p] + minpix_y * h->fenc->i_stride[p], h->fenc->i_stride[p],
2315                     h->param.i_width, maxpix_y-minpix_y );
2316             if( !CHROMA444 )
2317             {
2318                 uint64_t ssd_u, ssd_v;
2319                 int v_shift = CHROMA_V_SHIFT;
2320                 x264_pixel_ssd_nv12( &h->pixf,
2321                     h->fdec->plane[1] + (minpix_y>>v_shift) * h->fdec->i_stride[1], h->fdec->i_stride[1],
2322                     h->fenc->plane[1] + (minpix_y>>v_shift) * h->fenc->i_stride[1], h->fenc->i_stride[1],
2323                     h->param.i_width>>1, (maxpix_y-minpix_y)>>v_shift, &ssd_u, &ssd_v );
2324                 h->stat.frame.i_ssd[1] += ssd_u;
2325                 h->stat.frame.i_ssd[2] += ssd_v;
2326             }
2327         }
2328
2329         if( h->param.analyse.b_ssim )
2330         {
2331             int ssim_cnt;
2332             x264_emms();
2333             /* offset by 2 pixels to avoid alignment of ssim blocks with dct blocks,
2334              * and overlap by 4 */
2335             minpix_y += b_start ? 2 : -6;
2336             h->stat.frame.f_ssim +=
2337                 x264_pixel_ssim_wxh( &h->pixf,
2338                     h->fdec->plane[0] + 2+minpix_y*h->fdec->i_stride[0], h->fdec->i_stride[0],
2339                     h->fenc->plane[0] + 2+minpix_y*h->fenc->i_stride[0], h->fenc->i_stride[0],
2340                     h->param.i_width-2, maxpix_y-minpix_y, h->scratch_buffer, &ssim_cnt );
2341             h->stat.frame.i_ssim_cnt += ssim_cnt;
2342         }
2343     }
2344 }
2345
2346 static inline int x264_reference_update( x264_t *h )
2347 {
2348     if( !h->fdec->b_kept_as_ref )
2349     {
2350         if( h->i_thread_frames > 1 )
2351         {
2352             x264_frame_push_unused( h, h->fdec );
2353             h->fdec = x264_frame_pop_unused( h, 1 );
2354             if( !h->fdec )
2355                 return -1;
2356         }
2357         return 0;
2358     }
2359
2360     /* apply mmco from previous frame. */
2361     for( int i = 0; i < h->sh.i_mmco_command_count; i++ )
2362         for( int j = 0; h->frames.reference[j]; j++ )
2363             if( h->frames.reference[j]->i_poc == h->sh.mmco[i].i_poc )
2364                 x264_frame_push_unused( h, x264_frame_shift( &h->frames.reference[j] ) );
2365
2366     /* move frame in the buffer */
2367     x264_frame_push( h->frames.reference, h->fdec );
2368     if( h->frames.reference[h->sps->i_num_ref_frames] )
2369         x264_frame_push_unused( h, x264_frame_shift( h->frames.reference ) );
2370     h->fdec = x264_frame_pop_unused( h, 1 );
2371     if( !h->fdec )
2372         return -1;
2373     return 0;
2374 }
2375
2376 static inline void x264_reference_reset( x264_t *h )
2377 {
2378     while( h->frames.reference[0] )
2379         x264_frame_push_unused( h, x264_frame_pop( h->frames.reference ) );
2380     h->fdec->i_poc =
2381     h->fenc->i_poc = 0;
2382 }
2383
2384 static inline void x264_reference_hierarchy_reset( x264_t *h )
2385 {
2386     int ref;
2387     int b_hasdelayframe = 0;
2388
2389     /* look for delay frames -- chain must only contain frames that are disposable */
2390     for( int i = 0; h->frames.current[i] && IS_DISPOSABLE( h->frames.current[i]->i_type ); i++ )
2391         b_hasdelayframe |= h->frames.current[i]->i_coded
2392                         != h->frames.current[i]->i_frame + h->sps->vui.i_num_reorder_frames;
2393
2394     /* This function must handle b-pyramid and clear frames for open-gop */
2395     if( h->param.i_bframe_pyramid != X264_B_PYRAMID_STRICT && !b_hasdelayframe && h->frames.i_poc_last_open_gop == -1 )
2396         return;
2397
2398     /* Remove last BREF. There will never be old BREFs in the
2399      * dpb during a BREF decode when pyramid == STRICT */
2400     for( ref = 0; h->frames.reference[ref]; ref++ )
2401     {
2402         if( ( h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT
2403             && h->frames.reference[ref]->i_type == X264_TYPE_BREF )
2404             || ( h->frames.reference[ref]->i_poc < h->frames.i_poc_last_open_gop
2405             && h->sh.i_type != SLICE_TYPE_B ) )
2406         {
2407             int diff = h->i_frame_num - h->frames.reference[ref]->i_frame_num;
2408             h->sh.mmco[h->sh.i_mmco_command_count].i_difference_of_pic_nums = diff;
2409             h->sh.mmco[h->sh.i_mmco_command_count++].i_poc = h->frames.reference[ref]->i_poc;
2410             x264_frame_push_unused( h, x264_frame_shift( &h->frames.reference[ref] ) );
2411             h->b_ref_reorder[0] = 1;
2412             ref--;
2413         }
2414     }
2415
2416     /* Prepare room in the dpb for the delayed display time of the later b-frame's */
2417     if( h->param.i_bframe_pyramid )
2418         h->sh.i_mmco_remove_from_end = X264_MAX( ref + 2 - h->frames.i_max_dpb, 0 );
2419 }
2420
2421 static inline void x264_slice_init( x264_t *h, int i_nal_type, int i_global_qp )
2422 {
2423     /* ------------------------ Create slice header  ----------------------- */
2424     if( i_nal_type == NAL_SLICE_IDR )
2425     {
2426         x264_slice_header_init( h, &h->sh, h->sps, h->pps, h->i_idr_pic_id, h->i_frame_num, i_global_qp );
2427
2428         /* alternate id */
2429         h->i_idr_pic_id ^= 1;
2430     }
2431     else
2432     {
2433         x264_slice_header_init( h, &h->sh, h->sps, h->pps, -1, h->i_frame_num, i_global_qp );
2434
2435         h->sh.i_num_ref_idx_l0_active = h->i_ref[0] <= 0 ? 1 : h->i_ref[0];
2436         h->sh.i_num_ref_idx_l1_active = h->i_ref[1] <= 0 ? 1 : h->i_ref[1];
2437         if( h->sh.i_num_ref_idx_l0_active != h->pps->i_num_ref_idx_l0_default_active ||
2438             (h->sh.i_type == SLICE_TYPE_B && h->sh.i_num_ref_idx_l1_active != h->pps->i_num_ref_idx_l1_default_active) )
2439         {
2440             h->sh.b_num_ref_idx_override = 1;
2441         }
2442     }
2443
2444     if( h->fenc->i_type == X264_TYPE_BREF && h->param.b_bluray_compat && h->sh.i_mmco_command_count )
2445     {
2446         h->b_sh_backup = 1;
2447         h->sh_backup = h->sh;
2448     }
2449
2450     h->fdec->i_frame_num = h->sh.i_frame_num;
2451
2452     if( h->sps->i_poc_type == 0 )
2453     {
2454         h->sh.i_poc = h->fdec->i_poc;
2455         if( PARAM_INTERLACED )
2456         {
2457             h->sh.i_delta_poc_bottom = h->param.b_tff ? 1 : -1;
2458             h->sh.i_poc += h->sh.i_delta_poc_bottom == -1;
2459         }
2460         else
2461             h->sh.i_delta_poc_bottom = 0;
2462         h->fdec->i_delta_poc[0] = h->sh.i_delta_poc_bottom == -1;
2463         h->fdec->i_delta_poc[1] = h->sh.i_delta_poc_bottom ==  1;
2464     }
2465     else
2466     {
2467         /* Nothing to do ? */
2468     }
2469
2470     x264_macroblock_slice_init( h );
2471 }
2472
2473 typedef struct
2474 {
2475     int skip;
2476     uint8_t cabac_prevbyte;
2477     bs_t bs;
2478     x264_cabac_t cabac;
2479     x264_frame_stat_t stat;
2480     int last_qp;
2481     int last_dqp;
2482     int field_decoding_flag;
2483 } x264_bs_bak_t;
2484
2485 static ALWAYS_INLINE void x264_bitstream_backup( x264_t *h, x264_bs_bak_t *bak, int i_skip, int full )
2486 {
2487     if( full )
2488     {
2489         bak->stat = h->stat.frame;
2490         bak->last_qp = h->mb.i_last_qp;
2491         bak->last_dqp = h->mb.i_last_dqp;
2492         bak->field_decoding_flag = h->mb.field_decoding_flag;
2493     }
2494     else
2495     {
2496         bak->stat.i_mv_bits = h->stat.frame.i_mv_bits;
2497         bak->stat.i_tex_bits = h->stat.frame.i_tex_bits;
2498     }
2499     /* In the per-MB backup, we don't need the contexts because flushing the CABAC
2500      * encoder has no context dependency and in this case, a slice is ended (and
2501      * thus the content of all contexts are thrown away). */
2502     if( h->param.b_cabac )
2503     {
2504         if( full )
2505             memcpy( &bak->cabac, &h->cabac, sizeof(x264_cabac_t) );
2506         else
2507             memcpy( &bak->cabac, &h->cabac, offsetof(x264_cabac_t, f8_bits_encoded) );
2508         /* x264's CABAC writer modifies the previous byte during carry, so it has to be
2509          * backed up. */
2510         bak->cabac_prevbyte = h->cabac.p[-1];
2511     }
2512     else
2513     {
2514         bak->bs = h->out.bs;
2515         bak->skip = i_skip;
2516     }
2517 }
2518
2519 static ALWAYS_INLINE void x264_bitstream_restore( x264_t *h, x264_bs_bak_t *bak, int *skip, int full )
2520 {
2521     if( full )
2522     {
2523         h->stat.frame = bak->stat;
2524         h->mb.i_last_qp = bak->last_qp;
2525         h->mb.i_last_dqp = bak->last_dqp;
2526         h->mb.field_decoding_flag = bak->field_decoding_flag;
2527     }
2528     else
2529     {
2530         h->stat.frame.i_mv_bits = bak->stat.i_mv_bits;
2531         h->stat.frame.i_tex_bits = bak->stat.i_tex_bits;
2532     }
2533     if( h->param.b_cabac )
2534     {
2535         if( full )
2536             memcpy( &h->cabac, &bak->cabac, sizeof(x264_cabac_t) );
2537         else
2538             memcpy( &h->cabac, &bak->cabac, offsetof(x264_cabac_t, f8_bits_encoded) );
2539         h->cabac.p[-1] = bak->cabac_prevbyte;
2540     }
2541     else
2542     {
2543         h->out.bs = bak->bs;
2544         *skip = bak->skip;
2545     }
2546 }
2547
2548 static int x264_slice_write( x264_t *h )
2549 {
2550     int i_skip;
2551     int mb_xy, i_mb_x, i_mb_y;
2552     /* NALUs other than the first use a 3-byte startcode.
2553      * Add one extra byte for the rbsp, and one more for the final CABAC putbyte.
2554      * Then add an extra 5 bytes just in case, to account for random NAL escapes and
2555      * other inaccuracies. */
2556     int overhead_guess = (NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal)) + 1 + h->param.b_cabac + 5;
2557     int slice_max_size = h->param.i_slice_max_size > 0 ? (h->param.i_slice_max_size-overhead_guess)*8 : 0;
2558     int back_up_bitstream = slice_max_size || (!h->param.b_cabac && h->sps->i_profile_idc < PROFILE_HIGH);
2559     int starting_bits = bs_pos(&h->out.bs);
2560     int b_deblock = h->sh.i_disable_deblocking_filter_idc != 1;
2561     int b_hpel = h->fdec->b_kept_as_ref;
2562     int orig_last_mb = h->sh.i_last_mb;
2563     int thread_last_mb = h->i_threadslice_end * h->mb.i_mb_width - 1;
2564     uint8_t *last_emu_check;
2565 #define BS_BAK_SLICE_MAX_SIZE 0
2566 #define BS_BAK_SLICE_MIN_MBS  1
2567 #define BS_BAK_ROW_VBV        2
2568     x264_bs_bak_t bs_bak[3];
2569     b_deblock &= b_hpel || h->param.b_full_recon || h->param.psz_dump_yuv;
2570     bs_realign( &h->out.bs );
2571
2572     /* Slice */
2573     x264_nal_start( h, h->i_nal_type, h->i_nal_ref_idc );
2574     h->out.nal[h->out.i_nal].i_first_mb = h->sh.i_first_mb;
2575
2576     /* Slice header */
2577     x264_macroblock_thread_init( h );
2578
2579     /* Set the QP equal to the first QP in the slice for more accurate CABAC initialization. */
2580     h->mb.i_mb_xy = h->sh.i_first_mb;
2581     h->sh.i_qp = x264_ratecontrol_mb_qp( h );
2582     h->sh.i_qp = SPEC_QP( h->sh.i_qp );
2583     h->sh.i_qp_delta = h->sh.i_qp - h->pps->i_pic_init_qp;
2584
2585     x264_slice_header_write( &h->out.bs, &h->sh, h->i_nal_ref_idc );
2586     if( h->param.b_cabac )
2587     {
2588         /* alignment needed */
2589         bs_align_1( &h->out.bs );
2590
2591         /* init cabac */
2592         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 );
2593         x264_cabac_encode_init ( &h->cabac, h->out.bs.p, h->out.bs.p_end );
2594         last_emu_check = h->cabac.p;
2595     }
2596     else
2597         last_emu_check = h->out.bs.p;
2598     h->mb.i_last_qp = h->sh.i_qp;
2599     h->mb.i_last_dqp = 0;
2600     h->mb.field_decoding_flag = 0;
2601
2602     i_mb_y = h->sh.i_first_mb / h->mb.i_mb_width;
2603     i_mb_x = h->sh.i_first_mb % h->mb.i_mb_width;
2604     i_skip = 0;
2605
2606     while( 1 )
2607     {
2608         mb_xy = i_mb_x + i_mb_y * h->mb.i_mb_width;
2609         int mb_spos = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
2610
2611         if( i_mb_x == 0 )
2612         {
2613             if( x264_bitstream_check_buffer( h ) )
2614                 return -1;
2615             if( !(i_mb_y & SLICE_MBAFF) && h->param.rc.i_vbv_buffer_size )
2616                 x264_bitstream_backup( h, &bs_bak[BS_BAK_ROW_VBV], i_skip, 1 );
2617             if( !h->mb.b_reencode_mb )
2618                 x264_fdec_filter_row( h, i_mb_y, 0 );
2619         }
2620
2621         if( !(i_mb_y & SLICE_MBAFF) && back_up_bitstream )
2622         {
2623             x264_bitstream_backup( h, &bs_bak[BS_BAK_SLICE_MAX_SIZE], i_skip, 0 );
2624             if( slice_max_size && (thread_last_mb+1-mb_xy) == h->param.i_slice_min_mbs )
2625                 x264_bitstream_backup( h, &bs_bak[BS_BAK_SLICE_MIN_MBS], i_skip, 0 );
2626         }
2627
2628         if( PARAM_INTERLACED )
2629         {
2630             if( h->mb.b_adaptive_mbaff )
2631             {
2632                 if( !(i_mb_y&1) )
2633                 {
2634                     /* FIXME: VSAD is fast but fairly poor at choosing the best interlace type. */
2635                     h->mb.b_interlaced = x264_field_vsad( h, i_mb_x, i_mb_y );
2636                     memcpy( &h->zigzagf, MB_INTERLACED ? &h->zigzagf_interlaced : &h->zigzagf_progressive, sizeof(h->zigzagf) );
2637                     if( !MB_INTERLACED && (i_mb_y+2) == h->mb.i_mb_height )
2638                         x264_expand_border_mbpair( h, i_mb_x, i_mb_y );
2639                 }
2640             }
2641             h->mb.field[mb_xy] = MB_INTERLACED;
2642         }
2643
2644         /* load cache */
2645         if( SLICE_MBAFF )
2646             x264_macroblock_cache_load_interlaced( h, i_mb_x, i_mb_y );
2647         else
2648             x264_macroblock_cache_load_progressive( h, i_mb_x, i_mb_y );
2649
2650         x264_macroblock_analyse( h );
2651
2652         /* encode this macroblock -> be careful it can change the mb type to P_SKIP if needed */
2653 reencode:
2654         x264_macroblock_encode( h );
2655
2656         if( h->param.b_cabac )
2657         {
2658             if( mb_xy > h->sh.i_first_mb && !(SLICE_MBAFF && (i_mb_y&1)) )
2659                 x264_cabac_encode_terminal( &h->cabac );
2660
2661             if( IS_SKIP( h->mb.i_type ) )
2662                 x264_cabac_mb_skip( h, 1 );
2663             else
2664             {
2665                 if( h->sh.i_type != SLICE_TYPE_I )
2666                     x264_cabac_mb_skip( h, 0 );
2667                 x264_macroblock_write_cabac( h, &h->cabac );
2668             }
2669         }
2670         else
2671         {
2672             if( IS_SKIP( h->mb.i_type ) )
2673                 i_skip++;
2674             else
2675             {
2676                 if( h->sh.i_type != SLICE_TYPE_I )
2677                 {
2678                     bs_write_ue( &h->out.bs, i_skip );  /* skip run */
2679                     i_skip = 0;
2680                 }
2681                 x264_macroblock_write_cavlc( h );
2682                 /* If there was a CAVLC level code overflow, try again at a higher QP. */
2683                 if( h->mb.b_overflow )
2684                 {
2685                     h->mb.i_chroma_qp = h->chroma_qp_table[++h->mb.i_qp];
2686                     h->mb.i_skip_intra = 0;
2687                     h->mb.b_skip_mc = 0;
2688                     h->mb.b_overflow = 0;
2689                     x264_bitstream_restore( h, &bs_bak[BS_BAK_SLICE_MAX_SIZE], &i_skip, 0 );
2690                     goto reencode;
2691                 }
2692             }
2693         }
2694
2695         int total_bits = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
2696         int mb_size = total_bits - mb_spos;
2697
2698         if( slice_max_size && (!SLICE_MBAFF || (i_mb_y&1)) )
2699         {
2700             /* Count the skip run, just in case. */
2701             if( !h->param.b_cabac )
2702                 total_bits += bs_size_ue_big( i_skip );
2703             /* Check for escape bytes. */
2704             uint8_t *end = h->param.b_cabac ? h->cabac.p : h->out.bs.p;
2705             for( ; last_emu_check < end - 2; last_emu_check++ )
2706                 if( last_emu_check[0] == 0 && last_emu_check[1] == 0 && last_emu_check[2] <= 3 )
2707                 {
2708                     slice_max_size -= 8;
2709                     last_emu_check++;
2710                 }
2711             /* We'll just re-encode this last macroblock if we go over the max slice size. */
2712             if( total_bits - starting_bits > slice_max_size && !h->mb.b_reencode_mb )
2713             {
2714                 if( !x264_frame_new_slice( h, h->fdec ) )
2715                 {
2716                     /* Handle the most obnoxious slice-min-mbs edge case: we need to end the slice
2717                      * because it's gone over the maximum size, but doing so would violate slice-min-mbs.
2718                      * If possible, roll back to the last checkpoint and try again.
2719                      * We could try raising QP, but that would break in the case where a slice spans multiple
2720                      * rows, which the re-encoding infrastructure can't currently handle. */
2721                     if( mb_xy <= thread_last_mb && (thread_last_mb+1-mb_xy) < h->param.i_slice_min_mbs )
2722                     {
2723                         if( thread_last_mb-h->param.i_slice_min_mbs < h->sh.i_first_mb+h->param.i_slice_min_mbs )
2724                         {
2725                             x264_log( h, X264_LOG_WARNING, "slice-max-size violated (frame %d, cause: slice-min-mbs)\n", h->i_frame );
2726                             slice_max_size = 0;
2727                             goto cont;
2728                         }
2729                         x264_bitstream_restore( h, &bs_bak[BS_BAK_SLICE_MIN_MBS], &i_skip, 0 );
2730                         h->mb.b_reencode_mb = 1;
2731                         h->sh.i_last_mb = thread_last_mb-h->param.i_slice_min_mbs;
2732                         break;
2733                     }
2734                     if( mb_xy-SLICE_MBAFF*h->mb.i_mb_stride != h->sh.i_first_mb )
2735                     {
2736                         x264_bitstream_restore( h, &bs_bak[BS_BAK_SLICE_MAX_SIZE], &i_skip, 0 );
2737                         h->mb.b_reencode_mb = 1;
2738                         if( SLICE_MBAFF )
2739                         {
2740                             // set to bottom of previous mbpair
2741                             if( i_mb_x )
2742                                 h->sh.i_last_mb = mb_xy-1+h->mb.i_mb_stride*(!(i_mb_y&1));
2743                             else
2744                                 h->sh.i_last_mb = (i_mb_y-2+!(i_mb_y&1))*h->mb.i_mb_stride + h->mb.i_mb_width - 1;
2745                         }
2746                         else
2747                             h->sh.i_last_mb = mb_xy-1;
2748                         break;
2749                     }
2750                     else
2751                         h->sh.i_last_mb = mb_xy;
2752                 }
2753                 else
2754                     slice_max_size = 0;
2755             }
2756         }
2757 cont:
2758         h->mb.b_reencode_mb = 0;
2759
2760 #if HAVE_VISUALIZE
2761         if( h->param.b_visualize )
2762             x264_visualize_mb( h );
2763 #endif
2764
2765         /* save cache */
2766         x264_macroblock_cache_save( h );
2767
2768         if( x264_ratecontrol_mb( h, mb_size ) < 0 )
2769         {
2770             x264_bitstream_restore( h, &bs_bak[BS_BAK_ROW_VBV], &i_skip, 1 );
2771             h->mb.b_reencode_mb = 1;
2772             i_mb_x = 0;
2773             i_mb_y = i_mb_y - SLICE_MBAFF;
2774             h->mb.i_mb_prev_xy = i_mb_y * h->mb.i_mb_stride - 1;
2775             h->sh.i_last_mb = orig_last_mb;
2776             continue;
2777         }
2778
2779         /* accumulate mb stats */
2780         h->stat.frame.i_mb_count[h->mb.i_type]++;
2781
2782         int b_intra = IS_INTRA( h->mb.i_type );
2783         int b_skip = IS_SKIP( h->mb.i_type );
2784         if( h->param.i_log_level >= X264_LOG_INFO || h->param.rc.b_stat_write )
2785         {
2786             if( !b_intra && !b_skip && !IS_DIRECT( h->mb.i_type ) )
2787             {
2788                 if( h->mb.i_partition != D_8x8 )
2789                         h->stat.frame.i_mb_partition[h->mb.i_partition] += 4;
2790                     else
2791                         for( int i = 0; i < 4; i++ )
2792                             h->stat.frame.i_mb_partition[h->mb.i_sub_partition[i]] ++;
2793                 if( h->param.i_frame_reference > 1 )
2794                     for( int i_list = 0; i_list <= (h->sh.i_type == SLICE_TYPE_B); i_list++ )
2795                         for( int i = 0; i < 4; i++ )
2796                         {
2797                             int i_ref = h->mb.cache.ref[i_list][ x264_scan8[4*i] ];
2798                             if( i_ref >= 0 )
2799                                 h->stat.frame.i_mb_count_ref[i_list][i_ref] ++;
2800                         }
2801             }
2802         }
2803
2804         if( h->param.i_log_level >= X264_LOG_INFO )
2805         {
2806             if( h->mb.i_cbp_luma | h->mb.i_cbp_chroma )
2807             {
2808                 if( CHROMA444 )
2809                 {
2810                     for( int i = 0; i < 4; i++ )
2811                         if( h->mb.i_cbp_luma & (1 << i) )
2812                             for( int p = 0; p < 3; p++ )
2813                             {
2814                                 int s8 = i*4+p*16;
2815                                 int nnz8x8 = M16( &h->mb.cache.non_zero_count[x264_scan8[s8]+0] )
2816                                            | M16( &h->mb.cache.non_zero_count[x264_scan8[s8]+8] );
2817                                 h->stat.frame.i_mb_cbp[!b_intra + p*2] += !!nnz8x8;
2818                             }
2819                 }
2820                 else
2821                 {
2822                     int cbpsum = (h->mb.i_cbp_luma&1) + ((h->mb.i_cbp_luma>>1)&1)
2823                                + ((h->mb.i_cbp_luma>>2)&1) + (h->mb.i_cbp_luma>>3);
2824                     h->stat.frame.i_mb_cbp[!b_intra + 0] += cbpsum;
2825                     h->stat.frame.i_mb_cbp[!b_intra + 2] += !!h->mb.i_cbp_chroma;
2826                     h->stat.frame.i_mb_cbp[!b_intra + 4] += h->mb.i_cbp_chroma >> 1;
2827                 }
2828             }
2829             if( h->mb.i_cbp_luma && !b_intra )
2830             {
2831                 h->stat.frame.i_mb_count_8x8dct[0] ++;
2832                 h->stat.frame.i_mb_count_8x8dct[1] += h->mb.b_transform_8x8;
2833             }
2834             if( b_intra && h->mb.i_type != I_PCM )
2835             {
2836                 if( h->mb.i_type == I_16x16 )
2837                     h->stat.frame.i_mb_pred_mode[0][h->mb.i_intra16x16_pred_mode]++;
2838                 else if( h->mb.i_type == I_8x8 )
2839                     for( int i = 0; i < 16; i += 4 )
2840                         h->stat.frame.i_mb_pred_mode[1][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
2841                 else //if( h->mb.i_type == I_4x4 )
2842                     for( int i = 0; i < 16; i++ )
2843                         h->stat.frame.i_mb_pred_mode[2][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
2844                 h->stat.frame.i_mb_pred_mode[3][x264_mb_chroma_pred_mode_fix[h->mb.i_chroma_pred_mode]]++;
2845             }
2846             h->stat.frame.i_mb_field[b_intra?0:b_skip?2:1] += MB_INTERLACED;
2847         }
2848
2849         /* calculate deblock strength values (actual deblocking is done per-row along with hpel) */
2850         if( b_deblock )
2851             x264_macroblock_deblock_strength( h );
2852
2853         if( mb_xy == h->sh.i_last_mb )
2854             break;
2855
2856         if( SLICE_MBAFF )
2857         {
2858             i_mb_x += i_mb_y & 1;
2859             i_mb_y ^= i_mb_x < h->mb.i_mb_width;
2860         }
2861         else
2862             i_mb_x++;
2863         if( i_mb_x == h->mb.i_mb_width )
2864         {
2865             i_mb_y++;
2866             i_mb_x = 0;
2867         }
2868     }
2869     if( h->sh.i_last_mb < h->sh.i_first_mb )
2870         return 0;
2871
2872     h->out.nal[h->out.i_nal].i_last_mb = h->sh.i_last_mb;
2873
2874     if( h->param.b_cabac )
2875     {
2876         x264_cabac_encode_flush( h, &h->cabac );
2877         h->out.bs.p = h->cabac.p;
2878     }
2879     else
2880     {
2881         if( i_skip > 0 )
2882             bs_write_ue( &h->out.bs, i_skip );  /* last skip run */
2883         /* rbsp_slice_trailing_bits */
2884         bs_rbsp_trailing( &h->out.bs );
2885         bs_flush( &h->out.bs );
2886     }
2887     if( x264_nal_end( h ) )
2888         return -1;
2889
2890     if( h->sh.i_last_mb == (h->i_threadslice_end * h->mb.i_mb_width - 1) )
2891     {
2892         h->stat.frame.i_misc_bits = bs_pos( &h->out.bs )
2893                                   + (h->out.i_nal*NALU_OVERHEAD * 8)
2894                                   - h->stat.frame.i_tex_bits
2895                                   - h->stat.frame.i_mv_bits;
2896         x264_fdec_filter_row( h, h->i_threadslice_end, 0 );
2897
2898         if( h->param.b_sliced_threads )
2899         {
2900             /* Tell the main thread we're done. */
2901             x264_threadslice_cond_broadcast( h, 1 );
2902             /* Do hpel now */
2903             for( int mb_y = h->i_threadslice_start; mb_y <= h->i_threadslice_end; mb_y++ )
2904                 x264_fdec_filter_row( h, mb_y, 1 );
2905             x264_threadslice_cond_broadcast( h, 2 );
2906             /* Do the first row of hpel, now that the previous slice is done */
2907             if( h->i_thread_idx > 0 )
2908             {
2909                 x264_threadslice_cond_wait( h->thread[h->i_thread_idx-1], 2 );
2910                 x264_fdec_filter_row( h, h->i_threadslice_start + (1 << SLICE_MBAFF), 2 );
2911             }
2912         }
2913
2914         /* Free mb info after the last thread's done using it */
2915         if( h->fdec->mb_info_free && (!h->param.b_sliced_threads || h->i_thread_idx == (h->param.i_threads-1)) )
2916         {
2917             h->fdec->mb_info_free( h->fdec->mb_info );
2918             h->fdec->mb_info = NULL;
2919             h->fdec->mb_info_free = NULL;
2920         }
2921     }
2922
2923     return 0;
2924 }
2925
2926 static void x264_thread_sync_context( x264_t *dst, x264_t *src )
2927 {
2928     if( dst == src )
2929         return;
2930
2931     // reference counting
2932     for( x264_frame_t **f = src->frames.reference; *f; f++ )
2933         (*f)->i_reference_count++;
2934     for( x264_frame_t **f = dst->frames.reference; *f; f++ )
2935         x264_frame_push_unused( src, *f );
2936     src->fdec->i_reference_count++;
2937     x264_frame_push_unused( src, dst->fdec );
2938
2939     // copy everything except the per-thread pointers and the constants.
2940     memcpy( &dst->i_frame, &src->i_frame, offsetof(x264_t, mb.base) - offsetof(x264_t, i_frame) );
2941     dst->param = src->param;
2942     dst->stat = src->stat;
2943     dst->pixf = src->pixf;
2944     dst->reconfig = src->reconfig;
2945 }
2946
2947 static void x264_thread_sync_stat( x264_t *dst, x264_t *src )
2948 {
2949     if( dst == src )
2950         return;
2951     memcpy( &dst->stat.i_frame_count, &src->stat.i_frame_count, sizeof(dst->stat) - sizeof(dst->stat.frame) );
2952 }
2953
2954 static void *x264_slices_write( x264_t *h )
2955 {
2956     int i_slice_num = 0;
2957     int last_thread_mb = h->sh.i_last_mb;
2958
2959 #if HAVE_VISUALIZE
2960     if( h->param.b_visualize )
2961         if( x264_visualize_init( h ) )
2962             goto fail;
2963 #endif
2964
2965     /* init stats */
2966     memset( &h->stat.frame, 0, sizeof(h->stat.frame) );
2967     h->mb.b_reencode_mb = 0;
2968     while( h->sh.i_first_mb + SLICE_MBAFF*h->mb.i_mb_stride <= last_thread_mb )
2969     {
2970         h->sh.i_last_mb = last_thread_mb;
2971         if( !i_slice_num || !x264_frame_new_slice( h, h->fdec ) )
2972         {
2973             if( h->param.i_slice_max_mbs )
2974             {
2975                 if( SLICE_MBAFF )
2976                 {
2977                     // convert first to mbaff form, add slice-max-mbs, then convert back to normal form
2978                     int last_mbaff = 2*(h->sh.i_first_mb % h->mb.i_mb_width)
2979                         + h->mb.i_mb_width*(h->sh.i_first_mb / h->mb.i_mb_width)
2980                         + h->param.i_slice_max_mbs - 1;
2981                     int last_x = (last_mbaff % (2*h->mb.i_mb_width))/2;
2982                     int last_y = (last_mbaff / (2*h->mb.i_mb_width))*2 + 1;
2983                     h->sh.i_last_mb = last_x + h->mb.i_mb_stride*last_y;
2984                 }
2985                 else
2986                 {
2987                     h->sh.i_last_mb = h->sh.i_first_mb + h->param.i_slice_max_mbs - 1;
2988                     if( h->sh.i_last_mb < last_thread_mb && last_thread_mb - h->sh.i_last_mb < h->param.i_slice_min_mbs )
2989                         h->sh.i_last_mb = last_thread_mb - h->param.i_slice_min_mbs;
2990                 }
2991                 i_slice_num++;
2992             }
2993             else if( h->param.i_slice_count && !h->param.b_sliced_threads )
2994             {
2995                 int height = h->mb.i_mb_height >> PARAM_INTERLACED;
2996                 int width = h->mb.i_mb_width << PARAM_INTERLACED;
2997                 i_slice_num++;
2998                 h->sh.i_last_mb = (height * i_slice_num + h->param.i_slice_count/2) / h->param.i_slice_count * width - 1;
2999             }
3000         }
3001         h->sh.i_last_mb = X264_MIN( h->sh.i_last_mb, last_thread_mb );
3002         if( x264_stack_align( x264_slice_write, h ) )
3003             goto fail;
3004         h->sh.i_first_mb = h->sh.i_last_mb + 1;
3005         // if i_first_mb is not the last mb in a row then go to the next mb in MBAFF order
3006         if( SLICE_MBAFF && h->sh.i_first_mb % h->mb.i_mb_width )
3007             h->sh.i_first_mb -= h->mb.i_mb_stride;
3008     }
3009
3010 #if HAVE_VISUALIZE
3011     if( h->param.b_visualize )
3012     {
3013         x264_visualize_show( h );
3014         x264_visualize_close( h );
3015     }
3016 #endif
3017
3018     return (void *)0;
3019
3020 fail:
3021     /* Tell other threads we're done, so they wouldn't wait for it */
3022     if( h->param.b_sliced_threads )
3023         x264_threadslice_cond_broadcast( h, 2 );
3024     return (void *)-1;
3025 }
3026
3027 static int x264_threaded_slices_write( x264_t *h )
3028 {
3029     /* set first/last mb and sync contexts */
3030     for( int i = 0; i < h->param.i_threads; i++ )
3031     {
3032         x264_t *t = h->thread[i];
3033         if( i )
3034         {
3035             t->param = h->param;
3036             memcpy( &t->i_frame, &h->i_frame, offsetof(x264_t, rc) - offsetof(x264_t, i_frame) );
3037         }
3038         int height = h->mb.i_mb_height >> PARAM_INTERLACED;
3039         t->i_threadslice_start = ((height *  i    + h->param.i_slice_count/2) / h->param.i_threads) << PARAM_INTERLACED;
3040         t->i_threadslice_end   = ((height * (i+1) + h->param.i_slice_count/2) / h->param.i_threads) << PARAM_INTERLACED;
3041         t->sh.i_first_mb = t->i_threadslice_start * h->mb.i_mb_width;
3042         t->sh.i_last_mb  =   t->i_threadslice_end * h->mb.i_mb_width - 1;
3043     }
3044
3045     x264_stack_align( x264_analyse_weight_frame, h, h->mb.i_mb_height*16 + 16 );
3046
3047     x264_threads_distribute_ratecontrol( h );
3048
3049     /* setup */
3050     for( int i = 0; i < h->param.i_threads; i++ )
3051     {
3052         h->thread[i]->i_thread_idx = i;
3053         h->thread[i]->b_thread_active = 1;
3054         x264_threadslice_cond_broadcast( h->thread[i], 0 );
3055     }
3056     /* dispatch */
3057     for( int i = 0; i < h->param.i_threads; i++ )
3058         x264_threadpool_run( h->threadpool, (void*)x264_slices_write, h->thread[i] );
3059     /* wait */
3060     for( int i = 0; i < h->param.i_threads; i++ )
3061         x264_threadslice_cond_wait( h->thread[i], 1 );
3062
3063     x264_threads_merge_ratecontrol( h );
3064
3065     for( int i = 1; i < h->param.i_threads; i++ )
3066     {
3067         x264_t *t = h->thread[i];
3068         for( int j = 0; j < t->out.i_nal; j++ )
3069         {
3070             h->out.nal[h->out.i_nal] = t->out.nal[j];
3071             h->out.i_nal++;
3072             x264_nal_check_buffer( h );
3073         }
3074         /* All entries in stat.frame are ints except for ssd/ssim. */
3075         for( int j = 0; j < (offsetof(x264_t,stat.frame.i_ssd) - offsetof(x264_t,stat.frame.i_mv_bits)) / sizeof(int); j++ )
3076             ((int*)&h->stat.frame)[j] += ((int*)&t->stat.frame)[j];
3077         for( int j = 0; j < 3; j++ )
3078             h->stat.frame.i_ssd[j] += t->stat.frame.i_ssd[j];
3079         h->stat.frame.f_ssim += t->stat.frame.f_ssim;
3080         h->stat.frame.i_ssim_cnt += t->stat.frame.i_ssim_cnt;
3081     }
3082
3083     return 0;
3084 }
3085
3086 void x264_encoder_intra_refresh( x264_t *h )
3087 {
3088     h = h->thread[h->i_thread_phase];
3089     h->b_queued_intra_refresh = 1;
3090 }
3091
3092 int x264_encoder_invalidate_reference( x264_t *h, int64_t pts )
3093 {
3094     if( h->param.i_bframe )
3095     {
3096         x264_log( h, X264_LOG_ERROR, "x264_encoder_invalidate_reference is not supported with B-frames enabled\n" );
3097         return -1;
3098     }
3099     if( h->param.b_intra_refresh )
3100     {
3101         x264_log( h, X264_LOG_ERROR, "x264_encoder_invalidate_reference is not supported with intra refresh enabled\n" );
3102         return -1;
3103     }
3104     h = h->thread[h->i_thread_phase];
3105     if( pts >= h->i_last_idr_pts )
3106     {
3107         for( int i = 0; h->frames.reference[i]; i++ )
3108             if( pts <= h->frames.reference[i]->i_pts )
3109                 h->frames.reference[i]->b_corrupt = 1;
3110         if( pts <= h->fdec->i_pts )
3111             h->fdec->b_corrupt = 1;
3112     }
3113     return 0;
3114 }
3115
3116 /****************************************************************************
3117  * x264_encoder_encode:
3118  *  XXX: i_poc   : is the poc of the current given picture
3119  *       i_frame : is the number of the frame being coded
3120  *  ex:  type frame poc
3121  *       I      0   2*0
3122  *       P      1   2*3
3123  *       B      2   2*1
3124  *       B      3   2*2
3125  *       P      4   2*6
3126  *       B      5   2*4
3127  *       B      6   2*5
3128  ****************************************************************************/
3129 int     x264_encoder_encode( x264_t *h,
3130                              x264_nal_t **pp_nal, int *pi_nal,
3131                              x264_picture_t *pic_in,
3132                              x264_picture_t *pic_out )
3133 {
3134     x264_t *thread_current, *thread_prev, *thread_oldest;
3135     int i_nal_type, i_nal_ref_idc, i_global_qp;
3136     int overhead = NALU_OVERHEAD;
3137
3138 #if HAVE_OPENCL
3139     if( h->opencl.b_fatal_error )
3140         return -1;
3141 #endif
3142
3143     if( h->i_thread_frames > 1 )
3144     {
3145         thread_prev    = h->thread[ h->i_thread_phase ];
3146         h->i_thread_phase = (h->i_thread_phase + 1) % h->i_thread_frames;
3147         thread_current = h->thread[ h->i_thread_phase ];
3148         thread_oldest  = h->thread[ (h->i_thread_phase + 1) % h->i_thread_frames ];
3149         x264_thread_sync_context( thread_current, thread_prev );
3150         x264_thread_sync_ratecontrol( thread_current, thread_prev, thread_oldest );
3151         h = thread_current;
3152     }
3153     else
3154     {
3155         thread_current =
3156         thread_oldest  = h;
3157     }
3158     h->i_cpb_delay_pir_offset = h->i_cpb_delay_pir_offset_next;
3159
3160     /* no data out */
3161     *pi_nal = 0;
3162     *pp_nal = NULL;
3163
3164     /* ------------------- Setup new frame from picture -------------------- */
3165     if( pic_in != NULL )
3166     {
3167         /* 1: Copy the picture to a frame and move it to a buffer */
3168         x264_frame_t *fenc = x264_frame_pop_unused( h, 0 );
3169         if( !fenc )
3170             return -1;
3171
3172         if( x264_frame_copy_picture( h, fenc, pic_in ) < 0 )
3173             return -1;
3174
3175         if( h->param.i_width != 16 * h->mb.i_mb_width ||
3176             h->param.i_height != 16 * h->mb.i_mb_height )
3177             x264_frame_expand_border_mod16( h, fenc );
3178
3179         fenc->i_frame = h->frames.i_input++;
3180
3181         if( fenc->i_frame == 0 )
3182             h->frames.i_first_pts = fenc->i_pts;
3183         if( h->frames.i_bframe_delay && fenc->i_frame == h->frames.i_bframe_delay )
3184             h->frames.i_bframe_delay_time = fenc->i_pts - h->frames.i_first_pts;
3185
3186         if( h->param.b_vfr_input && fenc->i_pts <= h->frames.i_largest_pts )
3187             x264_log( h, X264_LOG_WARNING, "non-strictly-monotonic PTS\n" );
3188
3189         h->frames.i_second_largest_pts = h->frames.i_largest_pts;
3190         h->frames.i_largest_pts = fenc->i_pts;
3191
3192         if( (fenc->i_pic_struct < PIC_STRUCT_AUTO) || (fenc->i_pic_struct > PIC_STRUCT_TRIPLE) )
3193             fenc->i_pic_struct = PIC_STRUCT_AUTO;
3194
3195         if( fenc->i_pic_struct == PIC_STRUCT_AUTO )
3196         {
3197 #if HAVE_INTERLACED
3198             int b_interlaced = fenc->param ? fenc->param->b_interlaced : h->param.b_interlaced;
3199 #else
3200             int b_interlaced = 0;
3201 #endif
3202             if( b_interlaced )
3203             {
3204                 int b_tff = fenc->param ? fenc->param->b_tff : h->param.b_tff;
3205                 fenc->i_pic_struct = b_tff ? PIC_STRUCT_TOP_BOTTOM : PIC_STRUCT_BOTTOM_TOP;
3206             }
3207             else
3208                 fenc->i_pic_struct = PIC_STRUCT_PROGRESSIVE;
3209         }
3210
3211         if( h->param.rc.b_mb_tree && h->param.rc.b_stat_read )
3212         {
3213             if( x264_macroblock_tree_read( h, fenc, pic_in->prop.quant_offsets ) )
3214                 return -1;
3215         }
3216         else
3217             x264_stack_align( x264_adaptive_quant_frame, h, fenc, pic_in->prop.quant_offsets );
3218
3219         if( pic_in->prop.quant_offsets_free )
3220             pic_in->prop.quant_offsets_free( pic_in->prop.quant_offsets );
3221
3222         if( h->frames.b_have_lowres )
3223             x264_frame_init_lowres( h, fenc );
3224
3225         /* 2: Place the frame into the queue for its slice type decision */
3226         x264_lookahead_put_frame( h, fenc );
3227
3228         if( h->frames.i_input <= h->frames.i_delay + 1 - h->i_thread_frames )
3229         {
3230             /* Nothing yet to encode, waiting for filling of buffers */
3231             pic_out->i_type = X264_TYPE_AUTO;
3232             return 0;
3233         }
3234     }
3235     else
3236     {
3237         /* signal kills for lookahead thread */
3238         x264_pthread_mutex_lock( &h->lookahead->ifbuf.mutex );
3239         h->lookahead->b_exit_thread = 1;
3240         x264_pthread_cond_broadcast( &h->lookahead->ifbuf.cv_fill );
3241         x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
3242     }
3243
3244     h->i_frame++;
3245     /* 3: The picture is analyzed in the lookahead */
3246     if( !h->frames.current[0] )
3247         x264_lookahead_get_frames( h );
3248
3249     if( !h->frames.current[0] && x264_lookahead_is_empty( h ) )
3250         return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
3251
3252     /* ------------------- Get frame to be encoded ------------------------- */
3253     /* 4: get picture to encode */
3254     h->fenc = x264_frame_shift( h->frames.current );
3255
3256     /* If applicable, wait for previous frame reconstruction to finish */
3257     if( h->param.b_sliced_threads )
3258         if( x264_threadpool_wait_all( h ) < 0 )
3259             return -1;
3260
3261     if( h->i_frame == h->i_thread_frames - 1 )
3262         h->i_reordered_pts_delay = h->fenc->i_reordered_pts;
3263     if( h->reconfig )
3264     {
3265         x264_encoder_reconfig_apply( h, &h->reconfig_h->param );
3266         h->reconfig = 0;
3267     }
3268     if( h->fenc->param )
3269     {
3270         x264_encoder_reconfig_apply( h, h->fenc->param );
3271         if( h->fenc->param->param_free )
3272         {
3273             h->fenc->param->param_free( h->fenc->param );
3274             h->fenc->param = NULL;
3275         }
3276     }
3277
3278     // ok to call this before encoding any frames, since the initial values of fdec have b_kept_as_ref=0
3279     if( x264_reference_update( h ) )
3280         return -1;
3281     h->fdec->i_lines_completed = -1;
3282
3283     if( !IS_X264_TYPE_I( h->fenc->i_type ) )
3284     {
3285         int valid_refs_left = 0;
3286         for( int i = 0; h->frames.reference[i]; i++ )
3287             if( !h->frames.reference[i]->b_corrupt )
3288                 valid_refs_left++;
3289         /* No valid reference frames left: force an IDR. */
3290         if( !valid_refs_left )
3291         {
3292             h->fenc->b_keyframe = 1;
3293             h->fenc->i_type = X264_TYPE_IDR;
3294         }
3295     }
3296
3297     if( h->fenc->b_keyframe )
3298     {
3299         h->frames.i_last_keyframe = h->fenc->i_frame;
3300         if( h->fenc->i_type == X264_TYPE_IDR )
3301         {
3302             h->i_frame_num = 0;
3303             h->frames.i_last_idr = h->fenc->i_frame;
3304         }
3305     }
3306     h->sh.i_mmco_command_count =
3307     h->sh.i_mmco_remove_from_end = 0;
3308     h->b_ref_reorder[0] =
3309     h->b_ref_reorder[1] = 0;
3310     h->fdec->i_poc =
3311     h->fenc->i_poc = 2 * ( h->fenc->i_frame - X264_MAX( h->frames.i_last_idr, 0 ) );
3312
3313     /* ------------------- Setup frame context ----------------------------- */
3314     /* 5: Init data dependent of frame type */
3315     if( h->fenc->i_type == X264_TYPE_IDR )
3316     {
3317         /* reset ref pictures */
3318         i_nal_type    = NAL_SLICE_IDR;
3319         i_nal_ref_idc = NAL_PRIORITY_HIGHEST;
3320         h->sh.i_type = SLICE_TYPE_I;
3321         x264_reference_reset( h );
3322         h->frames.i_poc_last_open_gop = -1;
3323     }
3324     else if( h->fenc->i_type == X264_TYPE_I )
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_I;
3329         x264_reference_hierarchy_reset( h );
3330         if( h->param.b_open_gop )
3331             h->frames.i_poc_last_open_gop = h->fenc->b_keyframe ? h->fenc->i_poc : -1;
3332     }
3333     else if( h->fenc->i_type == X264_TYPE_P )
3334     {
3335         i_nal_type    = NAL_SLICE;
3336         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
3337         h->sh.i_type = SLICE_TYPE_P;
3338         x264_reference_hierarchy_reset( h );
3339         h->frames.i_poc_last_open_gop = -1;
3340     }
3341     else if( h->fenc->i_type == X264_TYPE_BREF )
3342     {
3343         i_nal_type    = NAL_SLICE;
3344         i_nal_ref_idc = h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT ? NAL_PRIORITY_LOW : NAL_PRIORITY_HIGH;
3345         h->sh.i_type = SLICE_TYPE_B;
3346         x264_reference_hierarchy_reset( h );
3347     }
3348     else    /* B frame */
3349     {
3350         i_nal_type    = NAL_SLICE;
3351         i_nal_ref_idc = NAL_PRIORITY_DISPOSABLE;
3352         h->sh.i_type = SLICE_TYPE_B;
3353     }
3354
3355     h->fdec->i_type = h->fenc->i_type;
3356     h->fdec->i_frame = h->fenc->i_frame;
3357     h->fenc->b_kept_as_ref =
3358     h->fdec->b_kept_as_ref = i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE && h->param.i_keyint_max > 1;
3359
3360     h->fdec->mb_info = h->fenc->mb_info;
3361     h->fdec->mb_info_free = h->fenc->mb_info_free;
3362     h->fenc->mb_info = NULL;
3363     h->fenc->mb_info_free = NULL;
3364
3365     h->fdec->i_pts = h->fenc->i_pts;
3366     if( h->frames.i_bframe_delay )
3367     {
3368         int64_t *prev_reordered_pts = thread_current->frames.i_prev_reordered_pts;
3369         h->fdec->i_dts = h->i_frame > h->frames.i_bframe_delay
3370                        ? prev_reordered_pts[ (h->i_frame - h->frames.i_bframe_delay) % h->frames.i_bframe_delay ]
3371                        : h->fenc->i_reordered_pts - h->frames.i_bframe_delay_time;
3372         prev_reordered_pts[ h->i_frame % h->frames.i_bframe_delay ] = h->fenc->i_reordered_pts;
3373     }
3374     else
3375         h->fdec->i_dts = h->fenc->i_reordered_pts;
3376     if( h->fenc->i_type == X264_TYPE_IDR )
3377         h->i_last_idr_pts = h->fdec->i_pts;
3378
3379     /* ------------------- Init                ----------------------------- */
3380     /* build ref list 0/1 */
3381     x264_reference_build_list( h, h->fdec->i_poc );
3382
3383     /* ---------------------- Write the bitstream -------------------------- */
3384     /* Init bitstream context */
3385     if( h->param.b_sliced_threads )
3386     {
3387         for( int i = 0; i < h->param.i_threads; i++ )
3388         {
3389             bs_init( &h->thread[i]->out.bs, h->thread[i]->out.p_bitstream, h->thread[i]->out.i_bitstream );
3390             h->thread[i]->out.i_nal = 0;
3391         }
3392     }
3393     else
3394     {
3395         bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
3396         h->out.i_nal = 0;
3397     }
3398
3399     if( h->param.b_aud )
3400     {
3401         int pic_type;
3402
3403         if( h->sh.i_type == SLICE_TYPE_I )
3404             pic_type = 0;
3405         else if( h->sh.i_type == SLICE_TYPE_P )
3406             pic_type = 1;
3407         else if( h->sh.i_type == SLICE_TYPE_B )
3408             pic_type = 2;
3409         else
3410             pic_type = 7;
3411
3412         x264_nal_start( h, NAL_AUD, NAL_PRIORITY_DISPOSABLE );
3413         bs_write( &h->out.bs, 3, pic_type );
3414         bs_rbsp_trailing( &h->out.bs );
3415         if( x264_nal_end( h ) )
3416             return -1;
3417         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
3418     }
3419
3420     h->i_nal_type = i_nal_type;
3421     h->i_nal_ref_idc = i_nal_ref_idc;
3422
3423     if( h->param.b_intra_refresh )
3424     {
3425         if( IS_X264_TYPE_I( h->fenc->i_type ) )
3426         {
3427             h->fdec->i_frames_since_pir = 0;
3428             h->b_queued_intra_refresh = 0;
3429             /* PIR is currently only supported with ref == 1, so any intra frame effectively refreshes
3430              * the whole frame and counts as an intra refresh. */
3431             h->fdec->f_pir_position = h->mb.i_mb_width;
3432         }
3433         else if( h->fenc->i_type == X264_TYPE_P )
3434         {
3435             int pocdiff = (h->fdec->i_poc - h->fref[0][0]->i_poc)/2;
3436             float increment = X264_MAX( ((float)h->mb.i_mb_width-1) / h->param.i_keyint_max, 1 );
3437             h->fdec->f_pir_position = h->fref[0][0]->f_pir_position;
3438             h->fdec->i_frames_since_pir = h->fref[0][0]->i_frames_since_pir + pocdiff;
3439             if( h->fdec->i_frames_since_pir >= h->param.i_keyint_max ||
3440                 (h->b_queued_intra_refresh && h->fdec->f_pir_position + 0.5 >= h->mb.i_mb_width) )
3441             {
3442                 h->fdec->f_pir_position = 0;
3443                 h->fdec->i_frames_since_pir = 0;
3444                 h->b_queued_intra_refresh = 0;
3445                 h->fenc->b_keyframe = 1;
3446             }
3447             h->fdec->i_pir_start_col = h->fdec->f_pir_position+0.5;
3448             h->fdec->f_pir_position += increment * pocdiff;
3449             h->fdec->i_pir_end_col = h->fdec->f_pir_position+0.5;
3450             /* If our intra refresh has reached the right side of the frame, we're done. */
3451             if( h->fdec->i_pir_end_col >= h->mb.i_mb_width - 1 )
3452             {
3453                 h->fdec->f_pir_position = h->mb.i_mb_width;
3454                 h->fdec->i_pir_end_col = h->mb.i_mb_width - 1;
3455             }
3456         }
3457     }
3458
3459     if( h->fenc->b_keyframe )
3460     {
3461         /* Write SPS and PPS */
3462         if( h->param.b_repeat_headers )
3463         {
3464             /* generate sequence parameters */
3465             x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
3466             x264_sps_write( &h->out.bs, h->sps );
3467             if( x264_nal_end( h ) )
3468                 return -1;
3469             /* Pad AUD/SPS to 256 bytes like Panasonic */
3470             if( h->param.b_avcintra_compat )
3471                 h->out.nal[h->out.i_nal-1].i_padding = 256 - bs_pos( &h->out.bs ) / 8 - 2*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             /* generate picture parameters */
3475             x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
3476             x264_pps_write( &h->out.bs, h->sps, h->pps );
3477             if( x264_nal_end( h ) )
3478                 return -1;
3479             if( h->param.b_avcintra_compat )
3480                 h->out.nal[h->out.i_nal-1].i_padding = 256 - h->out.nal[h->out.i_nal-1].i_payload - NALU_OVERHEAD;
3481             overhead += h->out.nal[h->out.i_nal-1].i_payload + h->out.nal[h->out.i_nal-1].i_padding + NALU_OVERHEAD;
3482         }
3483
3484         /* when frame threading is used, buffering period sei is written in x264_encoder_frame_end */
3485         if( h->i_thread_frames == 1 && h->sps->vui.b_nal_hrd_parameters_present )
3486         {
3487             x264_hrd_fullness( h );
3488             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3489             x264_sei_buffering_period_write( h, &h->out.bs );
3490             if( x264_nal_end( h ) )
3491                return -1;
3492             overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3493         }
3494     }
3495
3496     /* write extra sei */
3497     for( int i = 0; i < h->fenc->extra_sei.num_payloads; i++ )
3498     {
3499         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3500         x264_sei_write( &h->out.bs, h->fenc->extra_sei.payloads[i].payload, h->fenc->extra_sei.payloads[i].payload_size,
3501                         h->fenc->extra_sei.payloads[i].payload_type );
3502         if( x264_nal_end( h ) )
3503             return -1;
3504         overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3505         if( h->fenc->extra_sei.sei_free )
3506         {
3507             h->fenc->extra_sei.sei_free( h->fenc->extra_sei.payloads[i].payload );
3508             h->fenc->extra_sei.payloads[i].payload = NULL;
3509         }
3510     }
3511
3512     if( h->fenc->extra_sei.sei_free )
3513     {
3514         h->fenc->extra_sei.sei_free( h->fenc->extra_sei.payloads );
3515         h->fenc->extra_sei.payloads = NULL;
3516         h->fenc->extra_sei.sei_free = NULL;
3517     }
3518
3519     if( h->fenc->b_keyframe )
3520     {
3521         /* Avid's decoder strictly wants two SEIs for AVC-Intra so we can't insert the x264 SEI */
3522         if( h->param.b_repeat_headers && h->fenc->i_frame == 0 && !h->param.b_avcintra_compat )
3523         {
3524             /* identify ourself */
3525             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3526             if( x264_sei_version_write( h, &h->out.bs ) )
3527                 return -1;
3528             if( x264_nal_end( h ) )
3529                 return -1;
3530             overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3531         }
3532
3533         if( h->fenc->i_type != X264_TYPE_IDR )
3534         {
3535             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;
3536             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3537             x264_sei_recovery_point_write( h, &h->out.bs, time_to_recovery );
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         if( h->param.i_frame_packing >= 0 )
3544         {
3545             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3546             x264_sei_frame_packing_write( h, &h->out.bs );
3547             if( x264_nal_end( h ) )
3548                 return -1;
3549             overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3550         }
3551     }
3552
3553     /* generate sei pic timing */
3554     if( h->sps->vui.b_pic_struct_present || h->sps->vui.b_nal_hrd_parameters_present )
3555     {
3556         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3557         x264_sei_pic_timing_write( h, &h->out.bs );
3558         if( x264_nal_end( h ) )
3559             return -1;
3560         overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3561     }
3562
3563     /* As required by Blu-ray. */
3564     if( !IS_X264_TYPE_B( h->fenc->i_type ) && h->b_sh_backup )
3565     {
3566         h->b_sh_backup = 0;
3567         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3568         x264_sei_dec_ref_pic_marking_write( h, &h->out.bs );
3569         if( x264_nal_end( h ) )
3570             return -1;
3571         overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3572     }
3573
3574     if( h->fenc->b_keyframe && h->param.b_intra_refresh )
3575         h->i_cpb_delay_pir_offset_next = h->fenc->i_cpb_delay;
3576
3577     /* Filler space: 10 or 18 SEIs' worth of space, depending on resolution */
3578     if( h->param.b_avcintra_compat )
3579     {
3580         /* Write an empty filler NAL to mimic the AUD in the P2 format*/
3581         x264_nal_start( h, NAL_FILLER, NAL_PRIORITY_DISPOSABLE );
3582         x264_filler_write( h, &h->out.bs, 0 );
3583         if( x264_nal_end( h ) )
3584             return -1;
3585         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
3586
3587         /* All lengths are magic lengths that decoders expect to see */
3588         /* "UMID" SEI */
3589         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3590         if( x264_sei_avcintra_umid_write( h, &h->out.bs ) < 0 )
3591             return -1;
3592         if( x264_nal_end( h ) )
3593             return -1;
3594         overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3595
3596         int unpadded_len;
3597         int total_len;
3598         if( h->param.i_height == 1080 )
3599         {
3600             unpadded_len = 5780;
3601             total_len = 17*512;
3602         }
3603         else
3604         {
3605             unpadded_len = 2900;
3606             total_len = 9*512;
3607         }
3608         /* "VANC" SEI */
3609         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3610         if( x264_sei_avcintra_vanc_write( h, &h->out.bs, unpadded_len ) < 0 )
3611             return -1;
3612         if( x264_nal_end( h ) )
3613             return -1;
3614
3615         h->out.nal[h->out.i_nal-1].i_padding = total_len - h->out.nal[h->out.i_nal-1].i_payload - SEI_OVERHEAD;
3616         overhead += h->out.nal[h->out.i_nal-1].i_payload + h->out.nal[h->out.i_nal-1].i_padding + SEI_OVERHEAD;
3617     }
3618
3619     /* Init the rate control */
3620     /* FIXME: Include slice header bit cost. */
3621     x264_ratecontrol_start( h, h->fenc->i_qpplus1, overhead*8 );
3622     i_global_qp = x264_ratecontrol_qp( h );
3623
3624     pic_out->i_qpplus1 =
3625     h->fdec->i_qpplus1 = i_global_qp + 1;
3626
3627     if( h->param.rc.b_stat_read && h->sh.i_type != SLICE_TYPE_I )
3628     {
3629         x264_reference_build_list_optimal( h );
3630         x264_reference_check_reorder( h );
3631     }
3632
3633     if( h->i_ref[0] )
3634         h->fdec->i_poc_l0ref0 = h->fref[0][0]->i_poc;
3635
3636     /* ------------------------ Create slice header  ----------------------- */
3637     x264_slice_init( h, i_nal_type, i_global_qp );
3638
3639     /*------------------------- Weights -------------------------------------*/
3640     if( h->sh.i_type == SLICE_TYPE_B )
3641         x264_macroblock_bipred_init( h );
3642
3643     x264_weighted_pred_init( h );
3644
3645     if( i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE )
3646         h->i_frame_num++;
3647
3648     /* Write frame */
3649     h->i_threadslice_start = 0;
3650     h->i_threadslice_end = h->mb.i_mb_height;
3651     if( h->i_thread_frames > 1 )
3652     {
3653         x264_threadpool_run( h->threadpool, (void*)x264_slices_write, h );
3654         h->b_thread_active = 1;
3655     }
3656     else if( h->param.b_sliced_threads )
3657     {
3658         if( x264_threaded_slices_write( h ) )
3659             return -1;
3660     }
3661     else
3662         if( (intptr_t)x264_slices_write( h ) )
3663             return -1;
3664
3665     return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
3666 }
3667
3668 static int x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
3669                                    x264_nal_t **pp_nal, int *pi_nal,
3670                                    x264_picture_t *pic_out )
3671 {
3672     char psz_message[80];
3673
3674     if( !h->param.b_sliced_threads && h->b_thread_active )
3675     {
3676         h->b_thread_active = 0;
3677         if( (intptr_t)x264_threadpool_wait( h->threadpool, h ) )
3678             return -1;
3679     }
3680     if( !h->out.i_nal )
3681     {
3682         pic_out->i_type = X264_TYPE_AUTO;
3683         return 0;
3684     }
3685
3686     x264_emms();
3687
3688     /* generate buffering period sei and insert it into place */
3689     if( h->i_thread_frames > 1 && h->fenc->b_keyframe && h->sps->vui.b_nal_hrd_parameters_present )
3690     {
3691         x264_hrd_fullness( h );
3692         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3693         x264_sei_buffering_period_write( h, &h->out.bs );
3694         if( x264_nal_end( h ) )
3695            return -1;
3696         /* buffering period sei must follow AUD, SPS and PPS and precede all other SEIs */
3697         int idx = 0;
3698         while( h->out.nal[idx].i_type == NAL_AUD ||
3699                h->out.nal[idx].i_type == NAL_SPS ||
3700                h->out.nal[idx].i_type == NAL_PPS )
3701             idx++;
3702         x264_nal_t nal_tmp = h->out.nal[h->out.i_nal-1];
3703         memmove( &h->out.nal[idx+1], &h->out.nal[idx], (h->out.i_nal-idx-1)*sizeof(x264_nal_t) );
3704         h->out.nal[idx] = nal_tmp;
3705     }
3706
3707     int frame_size = x264_encoder_encapsulate_nals( h, 0 );
3708     if( frame_size < 0 )
3709         return -1;
3710
3711     /* Set output picture properties */
3712     pic_out->i_type = h->fenc->i_type;
3713
3714     pic_out->b_keyframe = h->fenc->b_keyframe;
3715     pic_out->i_pic_struct = h->fenc->i_pic_struct;
3716
3717     pic_out->i_pts = h->fdec->i_pts;
3718     pic_out->i_dts = h->fdec->i_dts;
3719
3720     if( pic_out->i_pts < pic_out->i_dts )
3721         x264_log( h, X264_LOG_WARNING, "invalid DTS: PTS is less than DTS\n" );
3722
3723     pic_out->opaque = h->fenc->opaque;
3724
3725     pic_out->img.i_csp = h->fdec->i_csp;
3726 #if HIGH_BIT_DEPTH
3727     pic_out->img.i_csp |= X264_CSP_HIGH_DEPTH;
3728 #endif
3729     pic_out->img.i_plane = h->fdec->i_plane;
3730     for( int i = 0; i < pic_out->img.i_plane; i++ )
3731     {
3732         pic_out->img.i_stride[i] = h->fdec->i_stride[i] * sizeof(pixel);
3733         pic_out->img.plane[i] = (uint8_t*)h->fdec->plane[i];
3734     }
3735
3736     x264_frame_push_unused( thread_current, h->fenc );
3737
3738     /* ---------------------- Update encoder state ------------------------- */
3739
3740     /* update rc */
3741     int filler = 0;
3742     if( x264_ratecontrol_end( h, frame_size * 8, &filler ) < 0 )
3743         return -1;
3744
3745     pic_out->hrd_timing = h->fenc->hrd_timing;
3746     pic_out->prop.f_crf_avg = h->fdec->f_crf_avg;
3747
3748     /* Filler in AVC-Intra mode is written as zero bytes to the last slice
3749      * We don't know the size of the last slice until encapsulation so we add filler to the encapsulated NAL */
3750     if( h->param.b_avcintra_compat )
3751     {
3752         x264_t *h0 = h->thread[0];
3753         int ret = x264_check_encapsulated_buffer( h, h0, h->out.i_nal, frame_size, frame_size + filler );
3754         if( ret < 0 )
3755             return -1;
3756         memset( h->out.nal[0].p_payload + frame_size, 0, filler );
3757         h->out.nal[h->out.i_nal-1].i_payload += filler;
3758         h->out.nal[h->out.i_nal-1].i_padding = filler;
3759         frame_size += filler;
3760     }
3761     else
3762     {
3763         while( filler > 0 )
3764         {
3765             int f, overhead;
3766             overhead = (FILLER_OVERHEAD - h->param.b_annexb);
3767             if( h->param.i_slice_max_size && filler > h->param.i_slice_max_size )
3768             {
3769                 int next_size = filler - h->param.i_slice_max_size;
3770                 int overflow = X264_MAX( overhead - next_size, 0 );
3771                 f = h->param.i_slice_max_size - overhead - overflow;
3772             }
3773             else
3774                 f = X264_MAX( 0, filler - overhead );
3775
3776             if( x264_bitstream_check_buffer_filler( h, f ) )
3777                 return -1;
3778             x264_nal_start( h, NAL_FILLER, NAL_PRIORITY_DISPOSABLE );
3779             x264_filler_write( h, &h->out.bs, f );
3780             if( x264_nal_end( h ) )
3781                 return -1;
3782             int total_size = x264_encoder_encapsulate_nals( h, h->out.i_nal-1 );
3783             if( total_size < 0 )
3784                 return -1;
3785             frame_size += total_size;
3786             filler -= total_size;
3787         }
3788     }
3789
3790     /* End bitstream, set output  */
3791     *pi_nal = h->out.i_nal;
3792     *pp_nal = h->out.nal;
3793
3794     h->out.i_nal = 0;
3795
3796     x264_noise_reduction_update( h );
3797
3798     /* ---------------------- Compute/Print statistics --------------------- */
3799     x264_thread_sync_stat( h, h->thread[0] );
3800
3801     /* Slice stat */
3802     h->stat.i_frame_count[h->sh.i_type]++;
3803     h->stat.i_frame_size[h->sh.i_type] += frame_size;
3804     h->stat.f_frame_qp[h->sh.i_type] += h->fdec->f_qp_avg_aq;
3805
3806     for( int i = 0; i < X264_MBTYPE_MAX; i++ )
3807         h->stat.i_mb_count[h->sh.i_type][i] += h->stat.frame.i_mb_count[i];
3808     for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
3809         h->stat.i_mb_partition[h->sh.i_type][i] += h->stat.frame.i_mb_partition[i];
3810     for( int i = 0; i < 2; i++ )
3811         h->stat.i_mb_count_8x8dct[i] += h->stat.frame.i_mb_count_8x8dct[i];
3812     for( int i = 0; i < 6; i++ )
3813         h->stat.i_mb_cbp[i] += h->stat.frame.i_mb_cbp[i];
3814     for( int i = 0; i < 4; i++ )
3815         for( int j = 0; j < 13; j++ )
3816             h->stat.i_mb_pred_mode[i][j] += h->stat.frame.i_mb_pred_mode[i][j];
3817     if( h->sh.i_type != SLICE_TYPE_I )
3818         for( int i_list = 0; i_list < 2; i_list++ )
3819             for( int i = 0; i < X264_REF_MAX*2; i++ )
3820                 h->stat.i_mb_count_ref[h->sh.i_type][i_list][i] += h->stat.frame.i_mb_count_ref[i_list][i];
3821     for( int i = 0; i < 3; i++ )
3822         h->stat.i_mb_field[i] += h->stat.frame.i_mb_field[i];
3823     if( h->sh.i_type == SLICE_TYPE_P && h->param.analyse.i_weighted_pred >= X264_WEIGHTP_SIMPLE )
3824     {
3825         h->stat.i_wpred[0] += !!h->sh.weight[0][0].weightfn;
3826         h->stat.i_wpred[1] += !!h->sh.weight[0][1].weightfn || !!h->sh.weight[0][2].weightfn;
3827     }
3828     if( h->sh.i_type == SLICE_TYPE_B )
3829     {
3830         h->stat.i_direct_frames[ h->sh.b_direct_spatial_mv_pred ] ++;
3831         if( h->mb.b_direct_auto_write )
3832         {
3833             //FIXME somewhat arbitrary time constants
3834             if( h->stat.i_direct_score[0] + h->stat.i_direct_score[1] > h->mb.i_mb_count )
3835                 for( int i = 0; i < 2; i++ )
3836                     h->stat.i_direct_score[i] = h->stat.i_direct_score[i] * 9/10;
3837             for( int i = 0; i < 2; i++ )
3838                 h->stat.i_direct_score[i] += h->stat.frame.i_direct_score[i];
3839         }
3840     }
3841     else
3842         h->stat.i_consecutive_bframes[h->fenc->i_bframes]++;
3843
3844     psz_message[0] = '\0';
3845     double dur = h->fenc->f_duration;
3846     h->stat.f_frame_duration[h->sh.i_type] += dur;
3847     if( h->param.analyse.b_psnr )
3848     {
3849         int64_t ssd[3] =
3850         {
3851             h->stat.frame.i_ssd[0],
3852             h->stat.frame.i_ssd[1],
3853             h->stat.frame.i_ssd[2],
3854         };
3855         int luma_size = h->param.i_width * h->param.i_height;
3856         int chroma_size = CHROMA_SIZE( luma_size );
3857         pic_out->prop.f_psnr[0] = x264_psnr( ssd[0], luma_size );
3858         pic_out->prop.f_psnr[1] = x264_psnr( ssd[1], chroma_size );
3859         pic_out->prop.f_psnr[2] = x264_psnr( ssd[2], chroma_size );
3860         pic_out->prop.f_psnr_avg = x264_psnr( ssd[0] + ssd[1] + ssd[2], luma_size + chroma_size*2 );
3861
3862         h->stat.f_ssd_global[h->sh.i_type]   += dur * (ssd[0] + ssd[1] + ssd[2]);
3863         h->stat.f_psnr_average[h->sh.i_type] += dur * pic_out->prop.f_psnr_avg;
3864         h->stat.f_psnr_mean_y[h->sh.i_type]  += dur * pic_out->prop.f_psnr[0];
3865         h->stat.f_psnr_mean_u[h->sh.i_type]  += dur * pic_out->prop.f_psnr[1];
3866         h->stat.f_psnr_mean_v[h->sh.i_type]  += dur * pic_out->prop.f_psnr[2];
3867
3868         snprintf( psz_message, 80, " PSNR Y:%5.2f U:%5.2f V:%5.2f", pic_out->prop.f_psnr[0],
3869                                                                     pic_out->prop.f_psnr[1],
3870                                                                     pic_out->prop.f_psnr[2] );
3871     }
3872
3873     if( h->param.analyse.b_ssim )
3874     {
3875         pic_out->prop.f_ssim = h->stat.frame.f_ssim / h->stat.frame.i_ssim_cnt;
3876         h->stat.f_ssim_mean_y[h->sh.i_type] += pic_out->prop.f_ssim * dur;
3877         snprintf( psz_message + strlen(psz_message), 80 - strlen(psz_message),
3878                   " SSIM Y:%.5f", pic_out->prop.f_ssim );
3879     }
3880     psz_message[79] = '\0';
3881
3882     x264_log( h, X264_LOG_DEBUG,
3883                   "frame=%4d QP=%.2f NAL=%d Slice:%c Poc:%-3d I:%-4d P:%-4d SKIP:%-4d size=%d bytes%s\n",
3884               h->i_frame,
3885               h->fdec->f_qp_avg_aq,
3886               h->i_nal_ref_idc,
3887               h->sh.i_type == SLICE_TYPE_I ? 'I' : (h->sh.i_type == SLICE_TYPE_P ? 'P' : 'B' ),
3888               h->fdec->i_poc,
3889               h->stat.frame.i_mb_count_i,
3890               h->stat.frame.i_mb_count_p,
3891               h->stat.frame.i_mb_count_skip,
3892               frame_size,
3893               psz_message );
3894
3895     // keep stats all in one place
3896     x264_thread_sync_stat( h->thread[0], h );
3897     // for the use of the next frame
3898     x264_thread_sync_stat( thread_current, h );
3899
3900 #ifdef DEBUG_MB_TYPE
3901 {
3902     static const char mb_chars[] = { 'i', 'i', 'I', 'C', 'P', '8', 'S',
3903         'D', '<', 'X', 'B', 'X', '>', 'B', 'B', 'B', 'B', '8', 'S' };
3904     for( int mb_xy = 0; mb_xy < h->mb.i_mb_width * h->mb.i_mb_height; mb_xy++ )
3905     {
3906         if( h->mb.type[mb_xy] < X264_MBTYPE_MAX && h->mb.type[mb_xy] >= 0 )
3907             fprintf( stderr, "%c ", mb_chars[ h->mb.type[mb_xy] ] );
3908         else
3909             fprintf( stderr, "? " );
3910
3911         if( (mb_xy+1) % h->mb.i_mb_width == 0 )
3912             fprintf( stderr, "\n" );
3913     }
3914 }
3915 #endif
3916
3917     /* Remove duplicates, must be done near the end as breaks h->fref0 array
3918      * by freeing some of its pointers. */
3919     for( int i = 0; i < h->i_ref[0]; i++ )
3920         if( h->fref[0][i] && h->fref[0][i]->b_duplicate )
3921         {
3922             x264_frame_push_blank_unused( h, h->fref[0][i] );
3923             h->fref[0][i] = 0;
3924         }
3925
3926     if( h->param.psz_dump_yuv )
3927         x264_frame_dump( h );
3928     x264_emms();
3929
3930     return frame_size;
3931 }
3932
3933 static void x264_print_intra( int64_t *i_mb_count, double i_count, int b_print_pcm, char *intra )
3934 {
3935     intra += sprintf( intra, "I16..4%s: %4.1f%% %4.1f%% %4.1f%%",
3936         b_print_pcm ? "..PCM" : "",
3937         i_mb_count[I_16x16]/ i_count,
3938         i_mb_count[I_8x8]  / i_count,
3939         i_mb_count[I_4x4]  / i_count );
3940     if( b_print_pcm )
3941         sprintf( intra, " %4.1f%%", i_mb_count[I_PCM]  / i_count );
3942 }
3943
3944 /****************************************************************************
3945  * x264_encoder_close:
3946  ****************************************************************************/
3947 void    x264_encoder_close  ( x264_t *h )
3948 {
3949     int64_t i_yuv_size = FRAME_SIZE( h->param.i_width * h->param.i_height );
3950     int64_t i_mb_count_size[2][7] = {{0}};
3951     char buf[200];
3952     int b_print_pcm = h->stat.i_mb_count[SLICE_TYPE_I][I_PCM]
3953                    || h->stat.i_mb_count[SLICE_TYPE_P][I_PCM]
3954                    || h->stat.i_mb_count[SLICE_TYPE_B][I_PCM];
3955
3956     x264_lookahead_delete( h );
3957
3958 #if HAVE_OPENCL
3959     x264_opencl_lookahead_delete( h );
3960     x264_opencl_function_t *ocl = h->opencl.ocl;
3961 #endif
3962
3963     if( h->param.b_sliced_threads )
3964         x264_threadpool_wait_all( h );
3965     if( h->param.i_threads > 1 )
3966         x264_threadpool_delete( h->threadpool );
3967     if( h->param.i_lookahead_threads > 1 )
3968         x264_threadpool_delete( h->lookaheadpool );
3969     if( h->i_thread_frames > 1 )
3970     {
3971         for( int i = 0; i < h->i_thread_frames; i++ )
3972             if( h->thread[i]->b_thread_active )
3973             {
3974                 assert( h->thread[i]->fenc->i_reference_count == 1 );
3975                 x264_frame_delete( h->thread[i]->fenc );
3976             }
3977
3978         x264_t *thread_prev = h->thread[h->i_thread_phase];
3979         x264_thread_sync_ratecontrol( h, thread_prev, h );
3980         x264_thread_sync_ratecontrol( thread_prev, thread_prev, h );
3981         h->i_frame = thread_prev->i_frame + 1 - h->i_thread_frames;
3982     }
3983     h->i_frame++;
3984
3985     /* Slices used and PSNR */
3986     for( int i = 0; i < 3; i++ )
3987     {
3988         static const uint8_t slice_order[] = { SLICE_TYPE_I, SLICE_TYPE_P, SLICE_TYPE_B };
3989         int i_slice = slice_order[i];
3990
3991         if( h->stat.i_frame_count[i_slice] > 0 )
3992         {
3993             int i_count = h->stat.i_frame_count[i_slice];
3994             double dur =  h->stat.f_frame_duration[i_slice];
3995             if( h->param.analyse.b_psnr )
3996             {
3997                 x264_log( h, X264_LOG_INFO,
3998                           "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",
3999                           slice_type_to_char[i_slice],
4000                           i_count,
4001                           h->stat.f_frame_qp[i_slice] / i_count,
4002                           (double)h->stat.i_frame_size[i_slice] / i_count,
4003                           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,
4004                           h->stat.f_psnr_average[i_slice] / dur,
4005                           x264_psnr( h->stat.f_ssd_global[i_slice], dur * i_yuv_size ) );
4006             }
4007             else
4008             {
4009                 x264_log( h, X264_LOG_INFO,
4010                           "frame %c:%-5d Avg QP:%5.2f  size:%6.0f\n",
4011                           slice_type_to_char[i_slice],
4012                           i_count,
4013                           h->stat.f_frame_qp[i_slice] / i_count,
4014                           (double)h->stat.i_frame_size[i_slice] / i_count );
4015             }
4016         }
4017     }
4018     if( h->param.i_bframe && h->stat.i_frame_count[SLICE_TYPE_B] )
4019     {
4020         char *p = buf;
4021         int den = 0;
4022         // weight by number of frames (including the I/P-frames) that are in a sequence of N B-frames
4023         for( int i = 0; i <= h->param.i_bframe; i++ )
4024             den += (i+1) * h->stat.i_consecutive_bframes[i];
4025         for( int i = 0; i <= h->param.i_bframe; i++ )
4026             p += sprintf( p, " %4.1f%%", 100. * (i+1) * h->stat.i_consecutive_bframes[i] / den );
4027         x264_log( h, X264_LOG_INFO, "consecutive B-frames:%s\n", buf );
4028     }
4029
4030     for( int i_type = 0; i_type < 2; i_type++ )
4031         for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
4032         {
4033             if( i == D_DIRECT_8x8 ) continue; /* direct is counted as its own type */
4034             i_mb_count_size[i_type][x264_mb_partition_pixel_table[i]] += h->stat.i_mb_partition[i_type][i];
4035         }
4036
4037     /* MB types used */
4038     if( h->stat.i_frame_count[SLICE_TYPE_I] > 0 )
4039     {
4040         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_I];
4041         double i_count = h->stat.i_frame_count[SLICE_TYPE_I] * h->mb.i_mb_count / 100.0;
4042         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
4043         x264_log( h, X264_LOG_INFO, "mb I  %s\n", buf );
4044     }
4045     if( h->stat.i_frame_count[SLICE_TYPE_P] > 0 )
4046     {
4047         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_P];
4048         double i_count = h->stat.i_frame_count[SLICE_TYPE_P] * h->mb.i_mb_count / 100.0;
4049         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_P];
4050         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
4051         x264_log( h, X264_LOG_INFO,
4052                   "mb P  %s  P16..4: %4.1f%% %4.1f%% %4.1f%% %4.1f%% %4.1f%%    skip:%4.1f%%\n",
4053                   buf,
4054                   i_mb_size[PIXEL_16x16] / (i_count*4),
4055                   (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
4056                   i_mb_size[PIXEL_8x8] / (i_count*4),
4057                   (i_mb_size[PIXEL_8x4] + i_mb_size[PIXEL_4x8]) / (i_count*4),
4058                   i_mb_size[PIXEL_4x4] / (i_count*4),
4059                   i_mb_count[P_SKIP] / i_count );
4060     }
4061     if( h->stat.i_frame_count[SLICE_TYPE_B] > 0 )
4062     {
4063         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_B];
4064         double i_count = h->stat.i_frame_count[SLICE_TYPE_B] * h->mb.i_mb_count / 100.0;
4065         double i_mb_list_count;
4066         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_B];
4067         int64_t list_count[3] = {0}; /* 0 == L0, 1 == L1, 2 == BI */
4068         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
4069         for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
4070             for( int j = 0; j < 2; j++ )
4071             {
4072                 int l0 = x264_mb_type_list_table[i][0][j];
4073                 int l1 = x264_mb_type_list_table[i][1][j];
4074                 if( l0 || l1 )
4075                     list_count[l1+l0*l1] += h->stat.i_mb_count[SLICE_TYPE_B][i] * 2;
4076             }
4077         list_count[0] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L0_8x8];
4078         list_count[1] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L1_8x8];
4079         list_count[2] += h->stat.i_mb_partition[SLICE_TYPE_B][D_BI_8x8];
4080         i_mb_count[B_DIRECT] += (h->stat.i_mb_partition[SLICE_TYPE_B][D_DIRECT_8x8]+2)/4;
4081         i_mb_list_count = (list_count[0] + list_count[1] + list_count[2]) / 100.0;
4082         sprintf( buf + strlen(buf), "  B16..8: %4.1f%% %4.1f%% %4.1f%%  direct:%4.1f%%  skip:%4.1f%%",
4083                  i_mb_size[PIXEL_16x16] / (i_count*4),
4084                  (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
4085                  i_mb_size[PIXEL_8x8] / (i_count*4),
4086                  i_mb_count[B_DIRECT] / i_count,
4087                  i_mb_count[B_SKIP]   / i_count );
4088         if( i_mb_list_count != 0 )
4089             sprintf( buf + strlen(buf), "  L0:%4.1f%% L1:%4.1f%% BI:%4.1f%%",
4090                      list_count[0] / i_mb_list_count,
4091                      list_count[1] / i_mb_list_count,
4092                      list_count[2] / i_mb_list_count );
4093         x264_log( h, X264_LOG_INFO, "mb B  %s\n", buf );
4094     }
4095
4096     x264_ratecontrol_summary( h );
4097
4098     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 )
4099     {
4100 #define SUM3(p) (p[SLICE_TYPE_I] + p[SLICE_TYPE_P] + p[SLICE_TYPE_B])
4101 #define SUM3b(p,o) (p[SLICE_TYPE_I][o] + p[SLICE_TYPE_P][o] + p[SLICE_TYPE_B][o])
4102         int64_t i_i8x8 = SUM3b( h->stat.i_mb_count, I_8x8 );
4103         int64_t i_intra = i_i8x8 + SUM3b( h->stat.i_mb_count, I_4x4 )
4104                                  + SUM3b( h->stat.i_mb_count, I_16x16 );
4105         int64_t i_all_intra = i_intra + SUM3b( h->stat.i_mb_count, I_PCM);
4106         int64_t i_skip = SUM3b( h->stat.i_mb_count, P_SKIP )
4107                        + SUM3b( h->stat.i_mb_count, B_SKIP );
4108         const int i_count = h->stat.i_frame_count[SLICE_TYPE_I] +
4109                             h->stat.i_frame_count[SLICE_TYPE_P] +
4110                             h->stat.i_frame_count[SLICE_TYPE_B];
4111         int64_t i_mb_count = (int64_t)i_count * h->mb.i_mb_count;
4112         int64_t i_inter = i_mb_count - i_skip - i_intra;
4113         const double duration = h->stat.f_frame_duration[SLICE_TYPE_I] +
4114                                 h->stat.f_frame_duration[SLICE_TYPE_P] +
4115                                 h->stat.f_frame_duration[SLICE_TYPE_B];
4116         float f_bitrate = SUM3(h->stat.i_frame_size) / duration / 125;
4117
4118         if( PARAM_INTERLACED )
4119         {
4120             char *fieldstats = buf;
4121             fieldstats[0] = 0;
4122             if( i_inter )
4123                 fieldstats += sprintf( fieldstats, " inter:%.1f%%", h->stat.i_mb_field[1] * 100.0 / i_inter );
4124             if( i_skip )
4125                 fieldstats += sprintf( fieldstats, " skip:%.1f%%", h->stat.i_mb_field[2] * 100.0 / i_skip );
4126             x264_log( h, X264_LOG_INFO, "field mbs: intra: %.1f%%%s\n",
4127                       h->stat.i_mb_field[0] * 100.0 / i_intra, buf );
4128         }
4129
4130         if( h->pps->b_transform_8x8_mode )
4131         {
4132             buf[0] = 0;
4133             if( h->stat.i_mb_count_8x8dct[0] )
4134                 sprintf( buf, " inter:%.1f%%", 100. * h->stat.i_mb_count_8x8dct[1] / h->stat.i_mb_count_8x8dct[0] );
4135             x264_log( h, X264_LOG_INFO, "8x8 transform intra:%.1f%%%s\n", 100. * i_i8x8 / i_intra, buf );
4136         }
4137
4138         if( (h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO ||
4139             (h->stat.i_direct_frames[0] && h->stat.i_direct_frames[1]))
4140             && h->stat.i_frame_count[SLICE_TYPE_B] )
4141         {
4142             x264_log( h, X264_LOG_INFO, "direct mvs  spatial:%.1f%% temporal:%.1f%%\n",
4143                       h->stat.i_direct_frames[1] * 100. / h->stat.i_frame_count[SLICE_TYPE_B],
4144                       h->stat.i_direct_frames[0] * 100. / h->stat.i_frame_count[SLICE_TYPE_B] );
4145         }
4146
4147         buf[0] = 0;
4148         int csize = CHROMA444 ? 4 : 1;
4149         if( i_mb_count != i_all_intra )
4150             sprintf( buf, " inter: %.1f%% %.1f%% %.1f%%",
4151                      h->stat.i_mb_cbp[1] * 100.0 / ((i_mb_count - i_all_intra)*4),
4152                      h->stat.i_mb_cbp[3] * 100.0 / ((i_mb_count - i_all_intra)*csize),
4153                      h->stat.i_mb_cbp[5] * 100.0 / ((i_mb_count - i_all_intra)*csize) );
4154         x264_log( h, X264_LOG_INFO, "coded y,%s,%s intra: %.1f%% %.1f%% %.1f%%%s\n",
4155                   CHROMA444?"u":"uvDC", CHROMA444?"v":"uvAC",
4156                   h->stat.i_mb_cbp[0] * 100.0 / (i_all_intra*4),
4157                   h->stat.i_mb_cbp[2] * 100.0 / (i_all_intra*csize),
4158                   h->stat.i_mb_cbp[4] * 100.0 / (i_all_intra*csize), buf );
4159
4160         int64_t fixed_pred_modes[4][9] = {{0}};
4161         int64_t sum_pred_modes[4] = {0};
4162         for( int i = 0; i <= I_PRED_16x16_DC_128; i++ )
4163         {
4164             fixed_pred_modes[0][x264_mb_pred_mode16x16_fix[i]] += h->stat.i_mb_pred_mode[0][i];
4165 &nb