Add --filler option
[x262.git] / encoder / encoder.c
1 /*****************************************************************************
2  * encoder.c: top-level encoder functions
3  *****************************************************************************
4  * Copyright (C) 2003-2013 x264 project
5  *
6  * Authors: Laurent Aimar <fenrir@via.ecp.fr>
7  *          Loren Merritt <lorenm@u.washington.edu>
8  *          Jason Garrett-Glaser <darkshikari@gmail.com>
9  *
10  * This program is free software; you can redistribute it and/or modify
11  * it under the terms of the GNU General Public License as published by
12  * the Free Software Foundation; either version 2 of the License, or
13  * (at your option) any later version.
14  *
15  * This program is distributed in the hope that it will be useful,
16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
18  * GNU General Public License for more details.
19  *
20  * You should have received a copy of the GNU General Public License
21  * along with this program; if not, write to the Free Software
22  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02111, USA.
23  *
24  * This program is also available under a commercial proprietary license.
25  * For more information, contact us at licensing@x264.com.
26  *****************************************************************************/
27
28 #include "common/common.h"
29
30 #include "set.h"
31 #include "analyse.h"
32 #include "ratecontrol.h"
33 #include "macroblock.h"
34 #include "me.h"
35
36 #if HAVE_VISUALIZE
37 #include "common/visualize.h"
38 #endif
39
40 //#define DEBUG_MB_TYPE
41
42 #define bs_write_ue bs_write_ue_big
43
44 static int x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
45                                    x264_nal_t **pp_nal, int *pi_nal,
46                                    x264_picture_t *pic_out );
47
48 /****************************************************************************
49  *
50  ******************************* x264 libs **********************************
51  *
52  ****************************************************************************/
53 static double x264_psnr( double sqe, double size )
54 {
55     double mse = sqe / (PIXEL_MAX*PIXEL_MAX * size);
56     if( mse <= 0.0000000001 ) /* Max 100dB */
57         return 100;
58
59     return -10.0 * log10( mse );
60 }
61
62 static double x264_ssim( double ssim )
63 {
64     double inv_ssim = 1 - ssim;
65     if( inv_ssim <= 0.0000000001 ) /* Max 100dB */
66         return 100;
67
68     return -10.0 * log10( inv_ssim );
69 }
70
71 static int x264_threadpool_wait_all( x264_t *h )
72 {
73     for( int i = 0; i < h->param.i_threads; i++ )
74         if( h->thread[i]->b_thread_active )
75         {
76             h->thread[i]->b_thread_active = 0;
77             if( (intptr_t)x264_threadpool_wait( h->threadpool, h->thread[i] ) < 0 )
78                 return -1;
79         }
80     return 0;
81 }
82
83 static void x264_frame_dump( x264_t *h )
84 {
85     FILE *f = x264_fopen( h->param.psz_dump_yuv, "r+b" );
86     if( !f )
87         return;
88
89     /* Wait for the threads to finish deblocking */
90     if( h->param.b_sliced_threads )
91         x264_threadpool_wait_all( h );
92
93     /* Write the frame in display order */
94     int frame_size = FRAME_SIZE( h->param.i_height * h->param.i_width * sizeof(pixel) );
95     fseek( f, (uint64_t)h->fdec->i_frame * frame_size, SEEK_SET );
96     for( int p = 0; p < (CHROMA444 ? 3 : 1); p++ )
97         for( int y = 0; y < h->param.i_height; y++ )
98             fwrite( &h->fdec->plane[p][y*h->fdec->i_stride[p]], sizeof(pixel), h->param.i_width, f );
99     if( !CHROMA444 )
100     {
101         int cw = h->param.i_width>>1;
102         int ch = h->param.i_height>>CHROMA_V_SHIFT;
103         pixel *planeu = x264_malloc( (cw*ch*2+32)*sizeof(pixel) );
104         pixel *planev = planeu + cw*ch + 16;
105         h->mc.plane_copy_deinterleave( planeu, cw, planev, cw, h->fdec->plane[1], h->fdec->i_stride[1], cw, ch );
106         fwrite( planeu, 1, cw*ch*sizeof(pixel), f );
107         fwrite( planev, 1, cw*ch*sizeof(pixel), f );
108         x264_free( planeu );
109     }
110     fclose( f );
111 }
112
113 /* Fill "default" values */
114 static void x264_slice_header_init( x264_t *h, x264_slice_header_t *sh,
115                                     x264_sps_t *sps, x264_pps_t *pps,
116                                     int i_idr_pic_id, int i_frame, int i_qp )
117 {
118     x264_param_t *param = &h->param;
119
120     /* First we fill all fields */
121     sh->sps = sps;
122     sh->pps = pps;
123
124     sh->i_first_mb  = 0;
125     sh->i_last_mb   = h->mb.i_mb_count - 1;
126     sh->i_pps_id    = pps->i_id;
127
128     sh->i_frame_num = i_frame;
129
130     sh->b_mbaff = PARAM_INTERLACED;
131     sh->b_field_pic = 0;    /* no field support for now */
132     sh->b_bottom_field = 0; /* not yet used */
133
134     sh->i_idr_pic_id = i_idr_pic_id;
135
136     /* poc stuff, fixed later */
137     sh->i_poc = 0;
138     sh->i_delta_poc_bottom = 0;
139     sh->i_delta_poc[0] = 0;
140     sh->i_delta_poc[1] = 0;
141
142     sh->i_redundant_pic_cnt = 0;
143
144     h->mb.b_direct_auto_write = h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO
145                                 && h->param.i_bframe
146                                 && ( h->param.rc.b_stat_write || !h->param.rc.b_stat_read );
147
148     if( !h->mb.b_direct_auto_read && sh->i_type == SLICE_TYPE_B )
149     {
150         if( h->fref[1][0]->i_poc_l0ref0 == h->fref[0][0]->i_poc )
151         {
152             if( h->mb.b_direct_auto_write )
153                 sh->b_direct_spatial_mv_pred = ( h->stat.i_direct_score[1] > h->stat.i_direct_score[0] );
154             else
155                 sh->b_direct_spatial_mv_pred = ( param->analyse.i_direct_mv_pred == X264_DIRECT_PRED_SPATIAL );
156         }
157         else
158         {
159             h->mb.b_direct_auto_write = 0;
160             sh->b_direct_spatial_mv_pred = 1;
161         }
162     }
163     /* else b_direct_spatial_mv_pred was read from the 2pass statsfile */
164
165     sh->b_num_ref_idx_override = 0;
166     sh->i_num_ref_idx_l0_active = 1;
167     sh->i_num_ref_idx_l1_active = 1;
168
169     sh->b_ref_pic_list_reordering[0] = h->b_ref_reorder[0];
170     sh->b_ref_pic_list_reordering[1] = h->b_ref_reorder[1];
171
172     /* If the ref list isn't in the default order, construct reordering header */
173     for( int list = 0; list < 2; list++ )
174     {
175         if( sh->b_ref_pic_list_reordering[list] )
176         {
177             int pred_frame_num = i_frame;
178             for( int i = 0; i < h->i_ref[list]; i++ )
179             {
180                 int diff = h->fref[list][i]->i_frame_num - pred_frame_num;
181                 sh->ref_pic_list_order[list][i].idc = ( diff > 0 );
182                 sh->ref_pic_list_order[list][i].arg = (abs(diff) - 1) & ((1 << sps->i_log2_max_frame_num) - 1);
183                 pred_frame_num = h->fref[list][i]->i_frame_num;
184             }
185         }
186     }
187
188     sh->i_cabac_init_idc = param->i_cabac_init_idc;
189
190     sh->i_qp = SPEC_QP(i_qp);
191     sh->i_qp_delta = sh->i_qp - pps->i_pic_init_qp;
192     sh->b_sp_for_swidth = 0;
193     sh->i_qs_delta = 0;
194
195     int deblock_thresh = i_qp + 2 * X264_MIN(param->i_deblocking_filter_alphac0, param->i_deblocking_filter_beta);
196     /* If effective qp <= 15, deblocking would have no effect anyway */
197     if( param->b_deblocking_filter && (h->mb.b_variable_qp || 15 < deblock_thresh ) )
198         sh->i_disable_deblocking_filter_idc = param->b_sliced_threads ? 2 : 0;
199     else
200         sh->i_disable_deblocking_filter_idc = 1;
201     sh->i_alpha_c0_offset = param->i_deblocking_filter_alphac0 << 1;
202     sh->i_beta_offset = param->i_deblocking_filter_beta << 1;
203 }
204
205 static void x264_slice_header_write( bs_t *s, x264_slice_header_t *sh, int i_nal_ref_idc )
206 {
207     if( sh->b_mbaff )
208     {
209         int first_x = sh->i_first_mb % sh->sps->i_mb_width;
210         int first_y = sh->i_first_mb / sh->sps->i_mb_width;
211         assert( (first_y&1) == 0 );
212         bs_write_ue( s, (2*first_x + sh->sps->i_mb_width*(first_y&~1) + (first_y&1)) >> 1 );
213     }
214     else
215         bs_write_ue( s, sh->i_first_mb );
216
217     bs_write_ue( s, sh->i_type + 5 );   /* same type things */
218     bs_write_ue( s, sh->i_pps_id );
219     bs_write( s, sh->sps->i_log2_max_frame_num, sh->i_frame_num & ((1<<sh->sps->i_log2_max_frame_num)-1) );
220
221     if( !sh->sps->b_frame_mbs_only )
222     {
223         bs_write1( s, sh->b_field_pic );
224         if( sh->b_field_pic )
225             bs_write1( s, sh->b_bottom_field );
226     }
227
228     if( sh->i_idr_pic_id >= 0 ) /* NAL IDR */
229         bs_write_ue( s, sh->i_idr_pic_id );
230
231     if( sh->sps->i_poc_type == 0 )
232     {
233         bs_write( s, sh->sps->i_log2_max_poc_lsb, sh->i_poc & ((1<<sh->sps->i_log2_max_poc_lsb)-1) );
234         if( sh->pps->b_pic_order && !sh->b_field_pic )
235             bs_write_se( s, sh->i_delta_poc_bottom );
236     }
237
238     if( sh->pps->b_redundant_pic_cnt )
239         bs_write_ue( s, sh->i_redundant_pic_cnt );
240
241     if( sh->i_type == SLICE_TYPE_B )
242         bs_write1( s, sh->b_direct_spatial_mv_pred );
243
244     if( sh->i_type == SLICE_TYPE_P || sh->i_type == SLICE_TYPE_B )
245     {
246         bs_write1( s, sh->b_num_ref_idx_override );
247         if( sh->b_num_ref_idx_override )
248         {
249             bs_write_ue( s, sh->i_num_ref_idx_l0_active - 1 );
250             if( sh->i_type == SLICE_TYPE_B )
251                 bs_write_ue( s, sh->i_num_ref_idx_l1_active - 1 );
252         }
253     }
254
255     /* ref pic list reordering */
256     if( sh->i_type != SLICE_TYPE_I )
257     {
258         bs_write1( s, sh->b_ref_pic_list_reordering[0] );
259         if( sh->b_ref_pic_list_reordering[0] )
260         {
261             for( int i = 0; i < sh->i_num_ref_idx_l0_active; i++ )
262             {
263                 bs_write_ue( s, sh->ref_pic_list_order[0][i].idc );
264                 bs_write_ue( s, sh->ref_pic_list_order[0][i].arg );
265             }
266             bs_write_ue( s, 3 );
267         }
268     }
269     if( sh->i_type == SLICE_TYPE_B )
270     {
271         bs_write1( s, sh->b_ref_pic_list_reordering[1] );
272         if( sh->b_ref_pic_list_reordering[1] )
273         {
274             for( int i = 0; i < sh->i_num_ref_idx_l1_active; i++ )
275             {
276                 bs_write_ue( s, sh->ref_pic_list_order[1][i].idc );
277                 bs_write_ue( s, sh->ref_pic_list_order[1][i].arg );
278             }
279             bs_write_ue( s, 3 );
280         }
281     }
282
283     sh->b_weighted_pred = 0;
284     if( sh->pps->b_weighted_pred && sh->i_type == SLICE_TYPE_P )
285     {
286         sh->b_weighted_pred = sh->weight[0][0].weightfn || sh->weight[0][1].weightfn || sh->weight[0][2].weightfn;
287         /* pred_weight_table() */
288         bs_write_ue( s, sh->weight[0][0].i_denom );
289         bs_write_ue( s, sh->weight[0][1].i_denom );
290         for( int i = 0; i < sh->i_num_ref_idx_l0_active; i++ )
291         {
292             int luma_weight_l0_flag = !!sh->weight[i][0].weightfn;
293             int chroma_weight_l0_flag = !!sh->weight[i][1].weightfn || !!sh->weight[i][2].weightfn;
294             bs_write1( s, luma_weight_l0_flag );
295             if( luma_weight_l0_flag )
296             {
297                 bs_write_se( s, sh->weight[i][0].i_scale );
298                 bs_write_se( s, sh->weight[i][0].i_offset );
299             }
300             bs_write1( s, chroma_weight_l0_flag );
301             if( chroma_weight_l0_flag )
302             {
303                 for( int j = 1; j < 3; j++ )
304                 {
305                     bs_write_se( s, sh->weight[i][j].i_scale );
306                     bs_write_se( s, sh->weight[i][j].i_offset );
307                 }
308             }
309         }
310     }
311     else if( sh->pps->b_weighted_bipred == 1 && sh->i_type == SLICE_TYPE_B )
312     {
313       /* TODO */
314     }
315
316     if( i_nal_ref_idc != 0 )
317     {
318         if( sh->i_idr_pic_id >= 0 )
319         {
320             bs_write1( s, 0 );  /* no output of prior pics flag */
321             bs_write1( s, 0 );  /* long term reference flag */
322         }
323         else
324         {
325             bs_write1( s, sh->i_mmco_command_count > 0 ); /* adaptive_ref_pic_marking_mode_flag */
326             if( sh->i_mmco_command_count > 0 )
327             {
328                 for( int i = 0; i < sh->i_mmco_command_count; i++ )
329                 {
330                     bs_write_ue( s, 1 ); /* mark short term ref as unused */
331                     bs_write_ue( s, sh->mmco[i].i_difference_of_pic_nums - 1 );
332                 }
333                 bs_write_ue( s, 0 ); /* end command list */
334             }
335         }
336     }
337
338     if( sh->pps->b_cabac && sh->i_type != SLICE_TYPE_I )
339         bs_write_ue( s, sh->i_cabac_init_idc );
340
341     bs_write_se( s, sh->i_qp_delta );      /* slice qp delta */
342
343     if( sh->pps->b_deblocking_filter_control )
344     {
345         bs_write_ue( s, sh->i_disable_deblocking_filter_idc );
346         if( sh->i_disable_deblocking_filter_idc != 1 )
347         {
348             bs_write_se( s, sh->i_alpha_c0_offset >> 1 );
349             bs_write_se( s, sh->i_beta_offset >> 1 );
350         }
351     }
352 }
353
354 /* If we are within a reasonable distance of the end of the memory allocated for the bitstream, */
355 /* reallocate, adding an arbitrary amount of space. */
356 static int x264_bitstream_check_buffer_internal( x264_t *h, int size, int b_cabac, int i_nal )
357 {
358     if( (b_cabac && (h->cabac.p_end - h->cabac.p < size)) ||
359         (h->out.bs.p_end - h->out.bs.p < size) )
360     {
361         int buf_size = h->out.i_bitstream + size;
362         uint8_t *buf = x264_malloc( buf_size );
363         if( !buf )
364             return -1;
365         int aligned_size = h->out.i_bitstream & ~15;
366         h->mc.memcpy_aligned( buf, h->out.p_bitstream, aligned_size );
367         memcpy( buf + aligned_size, h->out.p_bitstream + aligned_size, h->out.i_bitstream - aligned_size );
368
369         intptr_t delta = buf - h->out.p_bitstream;
370
371         h->out.bs.p_start += delta;
372         h->out.bs.p += delta;
373         h->out.bs.p_end = buf + buf_size;
374
375         h->cabac.p_start += delta;
376         h->cabac.p += delta;
377         h->cabac.p_end = buf + buf_size;
378
379         for( int i = 0; i <= i_nal; i++ )
380             h->out.nal[i].p_payload += delta;
381
382         x264_free( h->out.p_bitstream );
383         h->out.p_bitstream = buf;
384         h->out.i_bitstream = buf_size;
385     }
386     return 0;
387 }
388
389 static int x264_bitstream_check_buffer( x264_t *h )
390 {
391     int max_row_size = (2500 << SLICE_MBAFF) * h->mb.i_mb_width;
392     return x264_bitstream_check_buffer_internal( h, max_row_size, h->param.b_cabac, h->out.i_nal );
393 }
394
395 static int x264_bitstream_check_buffer_filler( x264_t *h, int filler )
396 {
397     filler += 32; // add padding for safety
398     return x264_bitstream_check_buffer_internal( h, filler, 0, -1 );
399 }
400
401 #if HAVE_THREAD
402 static void x264_encoder_thread_init( x264_t *h )
403 {
404     if( h->param.i_sync_lookahead )
405         x264_lower_thread_priority( 10 );
406 }
407 #endif
408
409 /****************************************************************************
410  *
411  ****************************************************************************
412  ****************************** External API*********************************
413  ****************************************************************************
414  *
415  ****************************************************************************/
416
417 static int x264_validate_parameters( x264_t *h, int b_open )
418 {
419 #if HAVE_MMX
420     if( b_open )
421     {
422         int cpuflags = x264_cpu_detect();
423         int fail = 0;
424 #ifdef __SSE__
425         if( !(cpuflags & X264_CPU_SSE) )
426         {
427             x264_log( h, X264_LOG_ERROR, "your cpu does not support SSE1, but x264 was compiled with asm\n");
428             fail = 1;
429         }
430 #else
431         if( !(cpuflags & X264_CPU_MMX2) )
432         {
433             x264_log( h, X264_LOG_ERROR, "your cpu does not support MMXEXT, but x264 was compiled with asm\n");
434             fail = 1;
435         }
436 #endif
437         if( !fail && !(cpuflags & X264_CPU_CMOV) )
438         {
439             x264_log( h, X264_LOG_ERROR, "your cpu does not support CMOV, but x264 was compiled with asm\n");
440             fail = 1;
441         }
442         if( fail )
443         {
444             x264_log( h, X264_LOG_ERROR, "to run x264, recompile without asm (configure --disable-asm)\n");
445             return -1;
446         }
447     }
448 #endif
449
450 #if HAVE_INTERLACED
451     h->param.b_interlaced = !!PARAM_INTERLACED;
452 #else
453     if( h->param.b_interlaced )
454     {
455         x264_log( h, X264_LOG_ERROR, "not compiled with interlaced support\n" );
456         return -1;
457     }
458 #endif
459
460     if( h->param.i_width <= 0 || h->param.i_height <= 0 )
461     {
462         x264_log( h, X264_LOG_ERROR, "invalid width x height (%dx%d)\n",
463                   h->param.i_width, h->param.i_height );
464         return -1;
465     }
466
467     int i_csp = h->param.i_csp & X264_CSP_MASK;
468 #if X264_CHROMA_FORMAT
469     if( CHROMA_FORMAT != CHROMA_420 && i_csp >= X264_CSP_I420 && i_csp <= X264_CSP_NV12 )
470     {
471         x264_log( h, X264_LOG_ERROR, "not compiled with 4:2:0 support\n" );
472         return -1;
473     }
474     else if( CHROMA_FORMAT != CHROMA_422 && i_csp >= X264_CSP_I422 && i_csp <= X264_CSP_NV16 )
475     {
476         x264_log( h, X264_LOG_ERROR, "not compiled with 4:2:2 support\n" );
477         return -1;
478     }
479     else if( CHROMA_FORMAT != CHROMA_444 && i_csp >= X264_CSP_I444 && i_csp <= X264_CSP_RGB )
480     {
481         x264_log( h, X264_LOG_ERROR, "not compiled with 4:4:4 support\n" );
482         return -1;
483     }
484 #endif
485     if( i_csp <= X264_CSP_NONE || i_csp >= X264_CSP_MAX )
486     {
487         x264_log( h, X264_LOG_ERROR, "invalid CSP (only I420/YV12/NV12/I422/YV16/NV16/I444/YV24/BGR/BGRA/RGB supported)\n" );
488         return -1;
489     }
490
491     if( i_csp < X264_CSP_I444 && h->param.i_width % 2 )
492     {
493         x264_log( h, X264_LOG_ERROR, "width not divisible by 2 (%dx%d)\n",
494                   h->param.i_width, h->param.i_height );
495         return -1;
496     }
497
498     if( i_csp < X264_CSP_I422 && PARAM_INTERLACED && h->param.i_height % 4 )
499     {
500         x264_log( h, X264_LOG_ERROR, "height not divisible by 4 (%dx%d)\n",
501                   h->param.i_width, h->param.i_height );
502         return -1;
503     }
504
505     if( (i_csp < X264_CSP_I422 || PARAM_INTERLACED) && h->param.i_height % 2 )
506     {
507         x264_log( h, X264_LOG_ERROR, "height not divisible by 2 (%dx%d)\n",
508                   h->param.i_width, h->param.i_height );
509         return -1;
510     }
511
512     if( (h->param.crop_rect.i_left + h->param.crop_rect.i_right ) >= h->param.i_width ||
513         (h->param.crop_rect.i_top  + h->param.crop_rect.i_bottom) >= h->param.i_height )
514     {
515         x264_log( h, X264_LOG_ERROR, "invalid crop-rect %u,%u,%u,%u\n", h->param.crop_rect.i_left,
516                   h->param.crop_rect.i_top, h->param.crop_rect.i_right,  h->param.crop_rect.i_bottom );
517         return -1;
518     }
519
520     if( h->param.vui.i_sar_width <= 0 || h->param.vui.i_sar_height <= 0 )
521     {
522         h->param.vui.i_sar_width = 0;
523         h->param.vui.i_sar_height = 0;
524     }
525
526     if( h->param.i_threads == X264_THREADS_AUTO )
527         h->param.i_threads = x264_cpu_num_processors() * (h->param.b_sliced_threads?2:3)/2;
528     int max_sliced_threads = X264_MAX( 1, (h->param.i_height+15)/16 / 4 );
529     if( h->param.i_threads > 1 )
530     {
531 #if !HAVE_THREAD
532         x264_log( h, X264_LOG_WARNING, "not compiled with thread support!\n");
533         h->param.i_threads = 1;
534 #endif
535         /* Avoid absurdly small thread slices as they can reduce performance
536          * and VBV compliance.  Capped at an arbitrary 4 rows per thread. */
537         if( h->param.b_sliced_threads )
538             h->param.i_threads = X264_MIN( h->param.i_threads, max_sliced_threads );
539     }
540     h->param.i_threads = x264_clip3( h->param.i_threads, 1, X264_THREAD_MAX );
541     if( h->param.i_threads == 1 )
542     {
543         h->param.b_sliced_threads = 0;
544         h->param.i_lookahead_threads = 1;
545     }
546     h->i_thread_frames = h->param.b_sliced_threads ? 1 : h->param.i_threads;
547     if( h->i_thread_frames > 1 )
548         h->param.nalu_process = NULL;
549
550     if( h->param.b_opencl )
551     {
552 #if !HAVE_OPENCL
553         x264_log( h, X264_LOG_WARNING, "OpenCL: not compiled with OpenCL support, disabling\n" );
554         h->param.b_opencl = 0;
555 #elif BIT_DEPTH > 8
556         x264_log( h, X264_LOG_WARNING, "OpenCL lookahead does not support high bit depth, disabling opencl\n" );
557         h->param.b_opencl = 0;
558 #else
559         if( h->param.i_width < 32 || h->param.i_height < 32 )
560         {
561             x264_log( h, X264_LOG_WARNING, "OpenCL: frame size is too small, disabling opencl\n" );
562             h->param.b_opencl = 0;
563         }
564 #endif
565         if( h->param.opencl_device_id && h->param.i_opencl_device )
566         {
567             x264_log( h, X264_LOG_WARNING, "OpenCL: device id and device skip count configured; dropping skip\n" );
568             h->param.i_opencl_device = 0;
569         }
570     }
571
572     h->param.i_keyint_max = x264_clip3( h->param.i_keyint_max, 1, X264_KEYINT_MAX_INFINITE );
573     if( h->param.i_keyint_max == 1 )
574     {
575         h->param.b_intra_refresh = 0;
576         h->param.analyse.i_weighted_pred = 0;
577         h->param.i_frame_reference = 1;
578         h->param.i_dpb_size = 1;
579     }
580
581     h->param.i_frame_packing = x264_clip3( h->param.i_frame_packing, -1, 5 );
582
583     /* Detect default ffmpeg settings and terminate with an error. */
584     if( b_open )
585     {
586         int score = 0;
587         score += h->param.analyse.i_me_range == 0;
588         score += h->param.rc.i_qp_step == 3;
589         score += h->param.i_keyint_max == 12;
590         score += h->param.rc.i_qp_min == 2;
591         score += h->param.rc.i_qp_max == 31;
592         score += h->param.rc.f_qcompress == 0.5;
593         score += fabs(h->param.rc.f_ip_factor - 1.25) < 0.01;
594         score += fabs(h->param.rc.f_pb_factor - 1.25) < 0.01;
595         score += h->param.analyse.inter == 0 && h->param.analyse.i_subpel_refine == 8;
596         if( score >= 5 )
597         {
598             x264_log( h, X264_LOG_ERROR, "broken ffmpeg default settings detected\n" );
599             x264_log( h, X264_LOG_ERROR, "use an encoding preset (e.g. -vpre medium)\n" );
600             x264_log( h, X264_LOG_ERROR, "preset usage: -vpre <speed> -vpre <profile>\n" );
601             x264_log( h, X264_LOG_ERROR, "speed presets are listed in x264 --help\n" );
602             x264_log( h, X264_LOG_ERROR, "profile is optional; x264 defaults to high\n" );
603             return -1;
604         }
605     }
606
607     if( h->param.rc.i_rc_method < 0 || h->param.rc.i_rc_method > 2 )
608     {
609         x264_log( h, X264_LOG_ERROR, "no ratecontrol method specified\n" );
610         return -1;
611     }
612
613     if( PARAM_INTERLACED )
614         h->param.b_pic_struct = 1;
615
616     if( h->param.b_avcintra_compat )
617     {
618         if( BIT_DEPTH != 10 )
619         {
620             x264_log( h, X264_LOG_ERROR, "%2d-bit AVC-Intra is not widely compatible\n", BIT_DEPTH );
621             x264_log( h, X264_LOG_ERROR, "10-bit x264 is required to encode AVC-Intra\n" );
622             return -1;
623         }
624
625         /* [50/100][res][fps] */
626         static const struct
627         {
628             uint16_t fps_num;
629             uint16_t fps_den;
630             uint8_t interlaced;
631             uint16_t frame_size;
632             const uint8_t *cqm_4ic;
633             const uint8_t *cqm_8iy;
634         } avcintra_lut[2][2][5] =
635         {
636             {{{ 60000, 1001, 0,  912, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy },
637               {    50,    1, 0, 1100, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy },
638               { 30000, 1001, 0,  912, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy },
639               {    25,    1, 0, 1100, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy },
640               { 24000, 1001, 0,  912, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy }},
641              {{ 30000, 1001, 1, 1820, x264_cqm_avci50_4ic, x264_cqm_avci50_1080i_8iy },
642               {    25,    1, 1, 2196, x264_cqm_avci50_4ic, x264_cqm_avci50_1080i_8iy },
643               { 30000, 1001, 0, 1820, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy },
644               {    25,    1, 0, 2196, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy },
645               { 24000, 1001, 0, 1820, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy }}},
646             {{{ 60000, 1001, 0, 1848, x264_cqm_avci100_720p_4ic, x264_cqm_avci100_720p_8iy },
647               {    50,    1, 0, 2224, x264_cqm_avci100_720p_4ic, x264_cqm_avci100_720p_8iy },
648               { 30000, 1001, 0, 1848, x264_cqm_avci100_720p_4ic, x264_cqm_avci100_720p_8iy },
649               {    25,    1, 0, 2224, x264_cqm_avci100_720p_4ic, x264_cqm_avci100_720p_8iy },
650               { 24000, 1001, 0, 1848, x264_cqm_avci100_720p_4ic, x264_cqm_avci100_720p_8iy }},
651              {{ 30000, 1001, 1, 3692, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080i_8iy },
652               {    25,    1, 1, 4444, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080i_8iy },
653               { 30000, 1001, 0, 3692, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080p_8iy },
654               {    25,    1, 0, 4444, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080p_8iy },
655               { 24000, 1001, 0, 3692, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080p_8iy }}}
656         };
657
658         int res = -1;
659         int type = i_csp == X264_CSP_I422;
660         if( type )
661         {
662             if(      h->param.i_width == 1920 && h->param.i_height == 1080 ) res =  1;
663             else if( h->param.i_width == 1280 && h->param.i_height ==  720 ) res =  0;
664         }
665         else if( i_csp == X264_CSP_I420 )
666         {
667             if(      h->param.i_width == 1440 && h->param.i_height == 1080 ) res =  1;
668             else if( h->param.i_width ==  960 && h->param.i_height ==  720 ) res =  0;
669         }
670         else
671         {
672             x264_log( h, X264_LOG_ERROR, "Invalid colorspace for AVC-Intra\n" );
673             return -1;
674         }
675
676         if( res < 0 )
677         {
678             x264_log( h, X264_LOG_ERROR, "Resolution %dx%d invalid for AVC-Intra %s\n",
679                       h->param.i_width, h->param.i_height, type ? "100" : "50" );
680             return -1;
681         }
682
683         if( h->param.nalu_process )
684         {
685             x264_log( h, X264_LOG_ERROR, "nalu_process is not supported in AVC-Intra mode\n" );
686             return -1;
687         }
688
689         if( !h->param.b_repeat_headers )
690         {
691             x264_log( h, X264_LOG_ERROR, "Separate headers not supported in AVC-Intra mode\n" );
692             return -1;
693         }
694
695         int i;
696         uint32_t fps_num = h->param.i_fps_num, fps_den = h->param.i_fps_den;
697         x264_reduce_fraction( &fps_num, &fps_den );
698         for( i = 0; i < 5; i++ )
699         {
700             if( avcintra_lut[type][res][i].fps_num == fps_num &&
701                 avcintra_lut[type][res][i].fps_den == fps_den &&
702                 avcintra_lut[type][res][i].interlaced == PARAM_INTERLACED )
703             {
704                 break;
705             }
706         }
707         if( i == 5 )
708         {
709             x264_log( h, X264_LOG_ERROR, "FPS %d/%d%c not compatible with AVC-Intra\n",
710                       h->param.i_fps_num, h->param.i_fps_den, PARAM_INTERLACED ? 'i' : 'p' );
711             return -1;
712         }
713
714         h->param.i_keyint_max = 1;
715         h->param.b_intra_refresh = 0;
716         h->param.analyse.i_weighted_pred = 0;
717         h->param.i_frame_reference = 1;
718         h->param.i_dpb_size = 1;
719
720         h->param.b_bluray_compat = 0;
721         h->param.b_vfr_input = 0;
722         h->param.b_aud = 1;
723         h->param.vui.i_chroma_loc = 0;
724         h->param.i_nal_hrd = X264_NAL_HRD_NONE;
725         h->param.b_deblocking_filter = 0;
726         h->param.b_stitchable = 1;
727         h->param.b_pic_struct = 0;
728         h->param.analyse.b_transform_8x8 = 1;
729         h->param.analyse.intra = X264_ANALYSE_I8x8;
730         h->param.analyse.i_chroma_qp_offset = res && type ? 3 : 4;
731         h->param.b_cabac = !type;
732         h->param.rc.i_vbv_buffer_size = avcintra_lut[type][res][i].frame_size;
733         h->param.rc.i_vbv_max_bitrate =
734         h->param.rc.i_bitrate = h->param.rc.i_vbv_buffer_size * fps_num / fps_den;
735         h->param.rc.i_rc_method = X264_RC_ABR;
736         h->param.rc.f_vbv_buffer_init = 1.0;
737         h->param.rc.b_filler = 1;
738         h->param.i_cqm_preset = X264_CQM_CUSTOM;
739         memcpy( h->param.cqm_4iy, x264_cqm_jvt4i, sizeof(h->param.cqm_4iy) );
740         memcpy( h->param.cqm_4ic, avcintra_lut[type][res][i].cqm_4ic, sizeof(h->param.cqm_4ic) );
741         memcpy( h->param.cqm_8iy, avcintra_lut[type][res][i].cqm_8iy, sizeof(h->param.cqm_8iy) );
742
743         /* Need exactly 10 slices of equal MB count... why?  $deity knows... */
744         h->param.i_slice_max_mbs = ((h->param.i_width + 15) / 16) * ((h->param.i_height + 15) / 16) / 10;
745         h->param.i_slice_max_size = 0;
746         /* The slice structure only allows a maximum of 2 threads for 1080i/p
747          * and 1 or 5 threads for 720p */
748         if( h->param.b_sliced_threads )
749         {
750             if( res )
751                 h->param.i_threads = X264_MIN( 2, h->param.i_threads );
752             else
753             {
754                 h->param.i_threads = X264_MIN( 5, h->param.i_threads );
755                 if( h->param.i_threads < 5 )
756                     h->param.i_threads = 1;
757             }
758         }
759
760         if( type )
761             h->param.vui.i_sar_width = h->param.vui.i_sar_height = 1;
762         else
763         {
764             h->param.vui.i_sar_width  = 4;
765             h->param.vui.i_sar_height = 3;
766         }
767
768         /* Avid cannot handle negative QPs */
769         h->param.rc.i_qp_min = X264_MAX( h->param.rc.i_qp_min, QP_BD_OFFSET );
770     }
771
772     h->param.rc.f_rf_constant = x264_clip3f( h->param.rc.f_rf_constant, -QP_BD_OFFSET, 51 );
773     h->param.rc.f_rf_constant_max = x264_clip3f( h->param.rc.f_rf_constant_max, -QP_BD_OFFSET, 51 );
774     h->param.rc.i_qp_constant = x264_clip3( h->param.rc.i_qp_constant, 0, QP_MAX );
775     h->param.analyse.i_subpel_refine = x264_clip3( h->param.analyse.i_subpel_refine, 0, 11 );
776     h->param.rc.f_ip_factor = X264_MAX( h->param.rc.f_ip_factor, 0.01f );
777     h->param.rc.f_pb_factor = X264_MAX( h->param.rc.f_pb_factor, 0.01f );
778     if( h->param.rc.i_rc_method == X264_RC_CRF )
779     {
780         h->param.rc.i_qp_constant = h->param.rc.f_rf_constant + QP_BD_OFFSET;
781         h->param.rc.i_bitrate = 0;
782     }
783     if( b_open && (h->param.rc.i_rc_method == X264_RC_CQP || h->param.rc.i_rc_method == X264_RC_CRF)
784         && h->param.rc.i_qp_constant == 0 )
785     {
786         h->mb.b_lossless = 1;
787         h->param.i_cqm_preset = X264_CQM_FLAT;
788         h->param.psz_cqm_file = NULL;
789         h->param.rc.i_rc_method = X264_RC_CQP;
790         h->param.rc.f_ip_factor = 1;
791         h->param.rc.f_pb_factor = 1;
792         h->param.analyse.b_psnr = 0;
793         h->param.analyse.b_ssim = 0;
794         h->param.analyse.i_chroma_qp_offset = 0;
795         h->param.analyse.i_trellis = 0;
796         h->param.analyse.b_fast_pskip = 0;
797         h->param.analyse.i_noise_reduction = 0;
798         h->param.analyse.b_psy = 0;
799         h->param.i_bframe = 0;
800         /* 8x8dct is not useful without RD in CAVLC lossless */
801         if( !h->param.b_cabac && h->param.analyse.i_subpel_refine < 6 )
802             h->param.analyse.b_transform_8x8 = 0;
803     }
804     if( h->param.rc.i_rc_method == X264_RC_CQP )
805     {
806         float qp_p = h->param.rc.i_qp_constant;
807         float qp_i = qp_p - 6*log2f( h->param.rc.f_ip_factor );
808         float qp_b = qp_p + 6*log2f( h->param.rc.f_pb_factor );
809         h->param.rc.i_qp_min = x264_clip3( (int)(X264_MIN3( qp_p, qp_i, qp_b )), 0, QP_MAX );
810         h->param.rc.i_qp_max = x264_clip3( (int)(X264_MAX3( qp_p, qp_i, qp_b ) + .999), 0, QP_MAX );
811         h->param.rc.i_aq_mode = 0;
812         h->param.rc.b_mb_tree = 0;
813         h->param.rc.i_bitrate = 0;
814     }
815     h->param.rc.i_qp_max = x264_clip3( h->param.rc.i_qp_max, 0, QP_MAX );
816     h->param.rc.i_qp_min = x264_clip3( h->param.rc.i_qp_min, 0, h->param.rc.i_qp_max );
817     h->param.rc.i_qp_step = x264_clip3( h->param.rc.i_qp_step, 2, QP_MAX );
818     h->param.rc.i_bitrate = x264_clip3( h->param.rc.i_bitrate, 0, 2000000 );
819     if( h->param.rc.i_rc_method == X264_RC_ABR && !h->param.rc.i_bitrate )
820     {
821         x264_log( h, X264_LOG_ERROR, "bitrate not specified\n" );
822         return -1;
823     }
824     h->param.rc.i_vbv_buffer_size = x264_clip3( h->param.rc.i_vbv_buffer_size, 0, 2000000 );
825     h->param.rc.i_vbv_max_bitrate = x264_clip3( h->param.rc.i_vbv_max_bitrate, 0, 2000000 );
826     h->param.rc.f_vbv_buffer_init = x264_clip3f( h->param.rc.f_vbv_buffer_init, 0, 2000000 );
827     if( h->param.rc.i_vbv_buffer_size )
828     {
829         if( h->param.rc.i_rc_method == X264_RC_CQP )
830         {
831             x264_log( h, X264_LOG_WARNING, "VBV is incompatible with constant QP, ignored.\n" );
832             h->param.rc.i_vbv_max_bitrate = 0;
833             h->param.rc.i_vbv_buffer_size = 0;
834         }
835         else if( h->param.rc.i_vbv_max_bitrate == 0 )
836         {
837             if( h->param.rc.i_rc_method == X264_RC_ABR )
838             {
839                 x264_log( h, X264_LOG_WARNING, "VBV maxrate unspecified, assuming CBR\n" );
840                 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate;
841             }
842             else
843             {
844                 x264_log( h, X264_LOG_WARNING, "VBV bufsize set but maxrate unspecified, ignored\n" );
845                 h->param.rc.i_vbv_buffer_size = 0;
846             }
847         }
848         else if( h->param.rc.i_vbv_max_bitrate < h->param.rc.i_bitrate &&
849                  h->param.rc.i_rc_method == X264_RC_ABR )
850         {
851             x264_log( h, X264_LOG_WARNING, "max bitrate less than average bitrate, assuming CBR\n" );
852             h->param.rc.i_bitrate = h->param.rc.i_vbv_max_bitrate;
853         }
854     }
855     else if( h->param.rc.i_vbv_max_bitrate )
856     {
857         x264_log( h, X264_LOG_WARNING, "VBV maxrate specified, but no bufsize, ignored\n" );
858         h->param.rc.i_vbv_max_bitrate = 0;
859     }
860
861     h->param.i_slice_max_size = X264_MAX( h->param.i_slice_max_size, 0 );
862     h->param.i_slice_max_mbs = X264_MAX( h->param.i_slice_max_mbs, 0 );
863     h->param.i_slice_min_mbs = X264_MAX( h->param.i_slice_min_mbs, 0 );
864     if( h->param.i_slice_max_mbs )
865         h->param.i_slice_min_mbs = X264_MIN( h->param.i_slice_min_mbs, h->param.i_slice_max_mbs/2 );
866     else if( !h->param.i_slice_max_size )
867         h->param.i_slice_min_mbs = 0;
868     if( PARAM_INTERLACED && h->param.i_slice_min_mbs )
869     {
870         x264_log( h, X264_LOG_WARNING, "interlace + slice-min-mbs is not implemented\n" );
871         h->param.i_slice_min_mbs = 0;
872     }
873     int mb_width = (h->param.i_width+15)/16;
874     if( h->param.i_slice_min_mbs > mb_width )
875     {
876         x264_log( h, X264_LOG_WARNING, "slice-min-mbs > row mb size (%d) not implemented\n", mb_width );
877         h->param.i_slice_min_mbs = mb_width;
878     }
879
880     int max_slices = (h->param.i_height+((16<<PARAM_INTERLACED)-1))/(16<<PARAM_INTERLACED);
881     if( h->param.b_sliced_threads )
882         h->param.i_slice_count = x264_clip3( h->param.i_threads, 0, max_slices );
883     else
884     {
885         h->param.i_slice_count = x264_clip3( h->param.i_slice_count, 0, max_slices );
886         if( h->param.i_slice_max_mbs || h->param.i_slice_max_size )
887             h->param.i_slice_count = 0;
888     }
889     if( h->param.i_slice_count_max > 0 )
890         h->param.i_slice_count_max = X264_MAX( h->param.i_slice_count, h->param.i_slice_count_max );
891
892     if( h->param.b_bluray_compat )
893     {
894         h->param.i_bframe_pyramid = X264_MIN( X264_B_PYRAMID_STRICT, h->param.i_bframe_pyramid );
895         h->param.i_bframe = X264_MIN( h->param.i_bframe, 3 );
896         h->param.b_aud = 1;
897         h->param.i_nal_hrd = X264_MAX( h->param.i_nal_hrd, X264_NAL_HRD_VBR );
898         h->param.i_slice_max_size = 0;
899         h->param.i_slice_max_mbs = 0;
900         h->param.b_intra_refresh = 0;
901         h->param.i_frame_reference = X264_MIN( h->param.i_frame_reference, 6 );
902         h->param.i_dpb_size = X264_MIN( h->param.i_dpb_size, 6 );
903         /* Don't use I-frames, because Blu-ray treats them the same as IDR. */
904         h->param.i_keyint_min = 1;
905         /* Due to the proliferation of broken players that don't handle dupes properly. */
906         h->param.analyse.i_weighted_pred = X264_MIN( h->param.analyse.i_weighted_pred, X264_WEIGHTP_SIMPLE );
907         if( h->param.b_fake_interlaced )
908             h->param.b_pic_struct = 1;
909     }
910
911     h->param.i_frame_reference = x264_clip3( h->param.i_frame_reference, 1, X264_REF_MAX );
912     h->param.i_dpb_size = x264_clip3( h->param.i_dpb_size, 1, X264_REF_MAX );
913     if( h->param.i_scenecut_threshold < 0 )
914         h->param.i_scenecut_threshold = 0;
915     h->param.analyse.i_direct_mv_pred = x264_clip3( h->param.analyse.i_direct_mv_pred, X264_DIRECT_PRED_NONE, X264_DIRECT_PRED_AUTO );
916     if( !h->param.analyse.i_subpel_refine && h->param.analyse.i_direct_mv_pred > X264_DIRECT_PRED_SPATIAL )
917     {
918         x264_log( h, X264_LOG_WARNING, "subme=0 + direct=temporal is not supported\n" );
919         h->param.analyse.i_direct_mv_pred = X264_DIRECT_PRED_SPATIAL;
920     }
921     h->param.i_bframe = x264_clip3( h->param.i_bframe, 0, X264_MIN( X264_BFRAME_MAX, h->param.i_keyint_max-1 ) );
922     h->param.i_bframe_bias = x264_clip3( h->param.i_bframe_bias, -90, 100 );
923     if( h->param.i_bframe <= 1 )
924         h->param.i_bframe_pyramid = X264_B_PYRAMID_NONE;
925     h->param.i_bframe_pyramid = x264_clip3( h->param.i_bframe_pyramid, X264_B_PYRAMID_NONE, X264_B_PYRAMID_NORMAL );
926     h->param.i_bframe_adaptive = x264_clip3( h->param.i_bframe_adaptive, X264_B_ADAPT_NONE, X264_B_ADAPT_TRELLIS );
927     if( !h->param.i_bframe )
928     {
929         h->param.i_bframe_adaptive = X264_B_ADAPT_NONE;
930         h->param.analyse.i_direct_mv_pred = 0;
931         h->param.analyse.b_weighted_bipred = 0;
932         h->param.b_open_gop = 0;
933     }
934     if( h->param.b_intra_refresh && h->param.i_bframe_pyramid == X264_B_PYRAMID_NORMAL )
935     {
936         x264_log( h, X264_LOG_WARNING, "b-pyramid normal + intra-refresh is not supported\n" );
937         h->param.i_bframe_pyramid = X264_B_PYRAMID_STRICT;
938     }
939     if( h->param.b_intra_refresh && (h->param.i_frame_reference > 1 || h->param.i_dpb_size > 1) )
940     {
941         x264_log( h, X264_LOG_WARNING, "ref > 1 + intra-refresh is not supported\n" );
942         h->param.i_frame_reference = 1;
943         h->param.i_dpb_size = 1;
944     }
945     if( h->param.b_intra_refresh && h->param.b_open_gop )
946     {
947         x264_log( h, X264_LOG_WARNING, "intra-refresh is not compatible with open-gop\n" );
948         h->param.b_open_gop = 0;
949     }
950     if( !h->param.i_fps_num || !h->param.i_fps_den )
951     {
952         h->param.i_fps_num = 25;
953         h->param.i_fps_den = 1;
954     }
955     float fps = (float) h->param.i_fps_num / h->param.i_fps_den;
956     if( h->param.i_keyint_min == X264_KEYINT_MIN_AUTO )
957         h->param.i_keyint_min = X264_MIN( h->param.i_keyint_max / 10, fps );
958     h->param.i_keyint_min = x264_clip3( h->param.i_keyint_min, 1, h->param.i_keyint_max/2+1 );
959     h->param.rc.i_lookahead = x264_clip3( h->param.rc.i_lookahead, 0, X264_LOOKAHEAD_MAX );
960     {
961         int maxrate = X264_MAX( h->param.rc.i_vbv_max_bitrate, h->param.rc.i_bitrate );
962         float bufsize = maxrate ? (float)h->param.rc.i_vbv_buffer_size / maxrate : 0;
963         h->param.rc.i_lookahead = X264_MIN( h->param.rc.i_lookahead, X264_MAX( h->param.i_keyint_max, bufsize*fps ) );
964     }
965
966     if( !h->param.i_timebase_num || !h->param.i_timebase_den || !(h->param.b_vfr_input || h->param.b_pulldown) )
967     {
968         h->param.i_timebase_num = h->param.i_fps_den;
969         h->param.i_timebase_den = h->param.i_fps_num;
970     }
971
972     h->param.rc.f_qcompress = x264_clip3f( h->param.rc.f_qcompress, 0.0, 1.0 );
973     if( h->param.i_keyint_max == 1 || h->param.rc.f_qcompress == 1 )
974         h->param.rc.b_mb_tree = 0;
975     if( (!h->param.b_intra_refresh && h->param.i_keyint_max != X264_KEYINT_MAX_INFINITE) &&
976         !h->param.rc.i_lookahead && h->param.rc.b_mb_tree )
977     {
978         x264_log( h, X264_LOG_WARNING, "lookaheadless mb-tree requires intra refresh or infinite keyint\n" );
979         h->param.rc.b_mb_tree = 0;
980     }
981     if( b_open && h->param.rc.b_stat_read )
982         h->param.rc.i_lookahead = 0;
983 #if HAVE_THREAD
984     if( h->param.i_sync_lookahead < 0 )
985         h->param.i_sync_lookahead = h->param.i_bframe + 1;
986     h->param.i_sync_lookahead = X264_MIN( h->param.i_sync_lookahead, X264_LOOKAHEAD_MAX );
987     if( h->param.rc.b_stat_read || h->i_thread_frames == 1 )
988         h->param.i_sync_lookahead = 0;
989 #else
990     h->param.i_sync_lookahead = 0;
991 #endif
992
993     h->param.i_deblocking_filter_alphac0 = x264_clip3( h->param.i_deblocking_filter_alphac0, -6, 6 );
994     h->param.i_deblocking_filter_beta    = x264_clip3( h->param.i_deblocking_filter_beta, -6, 6 );
995     h->param.analyse.i_luma_deadzone[0] = x264_clip3( h->param.analyse.i_luma_deadzone[0], 0, 32 );
996     h->param.analyse.i_luma_deadzone[1] = x264_clip3( h->param.analyse.i_luma_deadzone[1], 0, 32 );
997
998     h->param.i_cabac_init_idc = x264_clip3( h->param.i_cabac_init_idc, 0, 2 );
999
1000     if( h->param.i_cqm_preset < X264_CQM_FLAT || h->param.i_cqm_preset > X264_CQM_CUSTOM )
1001         h->param.i_cqm_preset = X264_CQM_FLAT;
1002
1003     if( h->param.analyse.i_me_method < X264_ME_DIA ||
1004         h->param.analyse.i_me_method > X264_ME_TESA )
1005         h->param.analyse.i_me_method = X264_ME_HEX;
1006     h->param.analyse.i_me_range = x264_clip3( h->param.analyse.i_me_range, 4, 1024 );
1007     if( h->param.analyse.i_me_range > 16 && h->param.analyse.i_me_method <= X264_ME_HEX )
1008         h->param.analyse.i_me_range = 16;
1009     if( h->param.analyse.i_me_method == X264_ME_TESA &&
1010         (h->mb.b_lossless || h->param.analyse.i_subpel_refine <= 1) )
1011         h->param.analyse.i_me_method = X264_ME_ESA;
1012     h->param.analyse.b_mixed_references = h->param.analyse.b_mixed_references && h->param.i_frame_reference > 1;
1013     h->param.analyse.inter &= X264_ANALYSE_PSUB16x16|X264_ANALYSE_PSUB8x8|X264_ANALYSE_BSUB16x16|
1014                               X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
1015     h->param.analyse.intra &= X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
1016     if( !(h->param.analyse.inter & X264_ANALYSE_PSUB16x16) )
1017         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
1018     if( !h->param.analyse.b_transform_8x8 )
1019     {
1020         h->param.analyse.inter &= ~X264_ANALYSE_I8x8;
1021         h->param.analyse.intra &= ~X264_ANALYSE_I8x8;
1022     }
1023     h->param.analyse.i_trellis = x264_clip3( h->param.analyse.i_trellis, 0, 2 );
1024     h->param.rc.i_aq_mode = x264_clip3( h->param.rc.i_aq_mode, 0, 2 );
1025     h->param.rc.f_aq_strength = x264_clip3f( h->param.rc.f_aq_strength, 0, 3 );
1026     if( h->param.rc.f_aq_strength == 0 )
1027         h->param.rc.i_aq_mode = 0;
1028
1029     if( h->param.i_log_level < X264_LOG_INFO )
1030     {
1031         h->param.analyse.b_psnr = 0;
1032         h->param.analyse.b_ssim = 0;
1033     }
1034     /* Warn users trying to measure PSNR/SSIM with psy opts on. */
1035     if( b_open && (h->param.analyse.b_psnr || h->param.analyse.b_ssim) )
1036     {
1037         char *s = NULL;
1038
1039         if( h->param.analyse.b_psy )
1040         {
1041             s = h->param.analyse.b_psnr ? "psnr" : "ssim";
1042             x264_log( h, X264_LOG_WARNING, "--%s used with psy on: results will be invalid!\n", s );
1043         }
1044         else if( !h->param.rc.i_aq_mode && h->param.analyse.b_ssim )
1045         {
1046             x264_log( h, X264_LOG_WARNING, "--ssim used with AQ off: results will be invalid!\n" );
1047             s = "ssim";
1048         }
1049         else if(  h->param.rc.i_aq_mode && h->param.analyse.b_psnr )
1050         {
1051             x264_log( h, X264_LOG_WARNING, "--psnr used with AQ on: results will be invalid!\n" );
1052             s = "psnr";
1053         }
1054         if( s )
1055             x264_log( h, X264_LOG_WARNING, "--tune %s should be used if attempting to benchmark %s!\n", s, s );
1056     }
1057
1058     if( !h->param.analyse.b_psy )
1059     {
1060         h->param.analyse.f_psy_rd = 0;
1061         h->param.analyse.f_psy_trellis = 0;
1062     }
1063     h->param.analyse.f_psy_rd = x264_clip3f( h->param.analyse.f_psy_rd, 0, 10 );
1064     h->param.analyse.f_psy_trellis = x264_clip3f( h->param.analyse.f_psy_trellis, 0, 10 );
1065     h->mb.i_psy_rd = h->param.analyse.i_subpel_refine >= 6 ? FIX8( h->param.analyse.f_psy_rd ) : 0;
1066     h->mb.i_psy_trellis = h->param.analyse.i_trellis ? FIX8( h->param.analyse.f_psy_trellis / 4 ) : 0;
1067     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -32, 32);
1068     /* In 4:4:4 mode, chroma gets twice as much resolution, so we can halve its quality. */
1069     if( b_open && i_csp >= X264_CSP_I444 && i_csp < X264_CSP_BGR && h->param.analyse.b_psy )
1070         h->param.analyse.i_chroma_qp_offset += 6;
1071     /* Psy RDO increases overall quantizers to improve the quality of luma--this indirectly hurts chroma quality */
1072     /* so we lower the chroma QP offset to compensate */
1073     if( b_open && h->mb.i_psy_rd && !h->param.b_avcintra_compat )
1074         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_rd < 0.25 ? 1 : 2;
1075     /* Psy trellis has a similar effect. */
1076     if( b_open && h->mb.i_psy_trellis && !h->param.b_avcintra_compat )
1077         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_trellis < 0.25 ? 1 : 2;
1078     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -12, 12);
1079     /* MB-tree requires AQ to be on, even if the strength is zero. */
1080     if( !h->param.rc.i_aq_mode && h->param.rc.b_mb_tree )
1081     {
1082         h->param.rc.i_aq_mode = 1;
1083         h->param.rc.f_aq_strength = 0;
1084     }
1085     h->param.analyse.i_noise_reduction = x264_clip3( h->param.analyse.i_noise_reduction, 0, 1<<16 );
1086     if( h->param.analyse.i_subpel_refine >= 10 && (h->param.analyse.i_trellis != 2 || !h->param.rc.i_aq_mode) )
1087         h->param.analyse.i_subpel_refine = 9;
1088
1089     {
1090         const x264_level_t *l = x264_levels;
1091         if( h->param.i_level_idc < 0 )
1092         {
1093             int maxrate_bak = h->param.rc.i_vbv_max_bitrate;
1094             if( h->param.rc.i_rc_method == X264_RC_ABR && h->param.rc.i_vbv_buffer_size <= 0 )
1095                 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate * 2;
1096             x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
1097             do h->param.i_level_idc = l->level_idc;
1098                 while( l[1].level_idc && x264_validate_levels( h, 0 ) && l++ );
1099             h->param.rc.i_vbv_max_bitrate = maxrate_bak;
1100         }
1101         else
1102         {
1103             while( l->level_idc && l->level_idc != h->param.i_level_idc )
1104                 l++;
1105             if( l->level_idc == 0 )
1106             {
1107                 x264_log( h, X264_LOG_ERROR, "invalid level_idc: %d\n", h->param.i_level_idc );
1108                 return -1;
1109             }
1110         }
1111         if( h->param.analyse.i_mv_range <= 0 )
1112             h->param.analyse.i_mv_range = l->mv_range >> PARAM_INTERLACED;
1113         else
1114             h->param.analyse.i_mv_range = x264_clip3(h->param.analyse.i_mv_range, 32, 512 >> PARAM_INTERLACED);
1115     }
1116
1117     h->param.analyse.i_weighted_pred = x264_clip3( h->param.analyse.i_weighted_pred, X264_WEIGHTP_NONE, X264_WEIGHTP_SMART );
1118
1119     if( h->param.i_lookahead_threads == X264_THREADS_AUTO )
1120     {
1121         if( h->param.b_sliced_threads )
1122             h->param.i_lookahead_threads = h->param.i_threads;
1123         else
1124         {
1125             /* If we're using much slower lookahead settings than encoding settings, it helps a lot to use
1126              * more lookahead threads.  This typically happens in the first pass of a two-pass encode, so
1127              * try to guess at this sort of case.
1128              *
1129              * Tuned by a little bit of real encoding with the various presets. */
1130             int badapt = h->param.i_bframe_adaptive == X264_B_ADAPT_TRELLIS;
1131             int subme = X264_MIN( h->param.analyse.i_subpel_refine / 3, 3 ) + (h->param.analyse.i_subpel_refine > 1);
1132             int bframes = X264_MIN( (h->param.i_bframe - 1) / 3, 3 );
1133
1134             /* [b-adapt 0/1 vs 2][quantized subme][quantized bframes] */
1135             static const uint8_t lookahead_thread_div[2][5][4] =
1136             {{{6,6,6,6}, {3,3,3,3}, {4,4,4,4}, {6,6,6,6}, {12,12,12,12}},
1137              {{3,2,1,1}, {2,1,1,1}, {4,3,2,1}, {6,4,3,2}, {12, 9, 6, 4}}};
1138
1139             h->param.i_lookahead_threads = h->param.i_threads / lookahead_thread_div[badapt][subme][bframes];
1140             /* Since too many lookahead threads significantly degrades lookahead accuracy, limit auto
1141              * lookahead threads to about 8 macroblock rows high each at worst.  This number is chosen
1142              * pretty much arbitrarily. */
1143             h->param.i_lookahead_threads = X264_MIN( h->param.i_lookahead_threads, h->param.i_height / 128 );
1144         }
1145     }
1146     h->param.i_lookahead_threads = x264_clip3( h->param.i_lookahead_threads, 1, X264_MIN( max_sliced_threads, X264_LOOKAHEAD_THREAD_MAX ) );
1147
1148     if( PARAM_INTERLACED )
1149     {
1150         if( h->param.analyse.i_me_method >= X264_ME_ESA )
1151         {
1152             x264_log( h, X264_LOG_WARNING, "interlace + me=esa is not implemented\n" );
1153             h->param.analyse.i_me_method = X264_ME_UMH;
1154         }
1155         if( h->param.analyse.i_weighted_pred > 0 )
1156         {
1157             x264_log( h, X264_LOG_WARNING, "interlace + weightp is not implemented\n" );
1158             h->param.analyse.i_weighted_pred = X264_WEIGHTP_NONE;
1159         }
1160     }
1161
1162     if( !h->param.analyse.i_weighted_pred && h->param.rc.b_mb_tree && h->param.analyse.b_psy )
1163         h->param.analyse.i_weighted_pred = X264_WEIGHTP_FAKE;
1164
1165     if( h->i_thread_frames > 1 )
1166     {
1167         int r = h->param.analyse.i_mv_range_thread;
1168         int r2;
1169         if( r <= 0 )
1170         {
1171             // half of the available space is reserved and divided evenly among the threads,
1172             // the rest is allocated to whichever thread is far enough ahead to use it.
1173             // reserving more space increases quality for some videos, but costs more time
1174             // in thread synchronization.
1175             int max_range = (h->param.i_height + X264_THREAD_HEIGHT) / h->i_thread_frames - X264_THREAD_HEIGHT;
1176             r = max_range / 2;
1177         }
1178         r = X264_MAX( r, h->param.analyse.i_me_range );
1179         r = X264_MIN( r, h->param.analyse.i_mv_range );
1180         // round up to use the whole mb row
1181         r2 = (r & ~15) + ((-X264_THREAD_HEIGHT) & 15);
1182         if( r2 < r )
1183             r2 += 16;
1184         x264_log( h, X264_LOG_DEBUG, "using mv_range_thread = %d\n", r2 );
1185         h->param.analyse.i_mv_range_thread = r2;
1186     }
1187
1188     if( h->param.rc.f_rate_tolerance < 0 )
1189         h->param.rc.f_rate_tolerance = 0;
1190     if( h->param.rc.f_qblur < 0 )
1191         h->param.rc.f_qblur = 0;
1192     if( h->param.rc.f_complexity_blur < 0 )
1193         h->param.rc.f_complexity_blur = 0;
1194
1195     h->param.i_sps_id &= 31;
1196
1197     h->param.i_nal_hrd = x264_clip3( h->param.i_nal_hrd, X264_NAL_HRD_NONE, X264_NAL_HRD_CBR );
1198
1199     if( h->param.i_nal_hrd && !h->param.rc.i_vbv_buffer_size )
1200     {
1201         x264_log( h, X264_LOG_WARNING, "NAL HRD parameters require VBV parameters\n" );
1202         h->param.i_nal_hrd = X264_NAL_HRD_NONE;
1203     }
1204
1205     if( h->param.i_nal_hrd == X264_NAL_HRD_CBR &&
1206        (h->param.rc.i_bitrate != h->param.rc.i_vbv_max_bitrate || !h->param.rc.i_vbv_max_bitrate) )
1207     {
1208         x264_log( h, X264_LOG_WARNING, "CBR HRD requires constant bitrate\n" );
1209         h->param.i_nal_hrd = X264_NAL_HRD_VBR;
1210     }
1211
1212     if( h->param.i_nal_hrd == X264_NAL_HRD_CBR )
1213         h->param.rc.b_filler = 1;
1214
1215     /* ensure the booleans are 0 or 1 so they can be used in math */
1216 #define BOOLIFY(x) h->param.x = !!h->param.x
1217     BOOLIFY( b_cabac );
1218     BOOLIFY( b_constrained_intra );
1219     BOOLIFY( b_deblocking_filter );
1220     BOOLIFY( b_deterministic );
1221     BOOLIFY( b_sliced_threads );
1222     BOOLIFY( b_interlaced );
1223     BOOLIFY( b_intra_refresh );
1224     BOOLIFY( b_visualize );
1225     BOOLIFY( b_aud );
1226     BOOLIFY( b_repeat_headers );
1227     BOOLIFY( b_annexb );
1228     BOOLIFY( b_vfr_input );
1229     BOOLIFY( b_pulldown );
1230     BOOLIFY( b_tff );
1231     BOOLIFY( b_pic_struct );
1232     BOOLIFY( b_fake_interlaced );
1233     BOOLIFY( b_open_gop );
1234     BOOLIFY( b_bluray_compat );
1235     BOOLIFY( b_stitchable );
1236     BOOLIFY( b_full_recon );
1237     BOOLIFY( b_opencl );
1238     BOOLIFY( b_avcintra_compat );
1239     BOOLIFY( analyse.b_transform_8x8 );
1240     BOOLIFY( analyse.b_weighted_bipred );
1241     BOOLIFY( analyse.b_chroma_me );
1242     BOOLIFY( analyse.b_mixed_references );
1243     BOOLIFY( analyse.b_fast_pskip );
1244     BOOLIFY( analyse.b_dct_decimate );
1245     BOOLIFY( analyse.b_psy );
1246     BOOLIFY( analyse.b_psnr );
1247     BOOLIFY( analyse.b_ssim );
1248     BOOLIFY( rc.b_stat_write );
1249     BOOLIFY( rc.b_stat_read );
1250     BOOLIFY( rc.b_mb_tree );
1251     BOOLIFY( rc.b_filler );
1252 #undef BOOLIFY
1253
1254     return 0;
1255 }
1256
1257 static void mbcmp_init( x264_t *h )
1258 {
1259     int satd = !h->mb.b_lossless && h->param.analyse.i_subpel_refine > 1;
1260     memcpy( h->pixf.mbcmp, satd ? h->pixf.satd : h->pixf.sad_aligned, sizeof(h->pixf.mbcmp) );
1261     memcpy( h->pixf.mbcmp_unaligned, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.mbcmp_unaligned) );
1262     h->pixf.intra_mbcmp_x3_16x16 = satd ? h->pixf.intra_satd_x3_16x16 : h->pixf.intra_sad_x3_16x16;
1263     h->pixf.intra_mbcmp_x3_8x16c = satd ? h->pixf.intra_satd_x3_8x16c : h->pixf.intra_sad_x3_8x16c;
1264     h->pixf.intra_mbcmp_x3_8x8c  = satd ? h->pixf.intra_satd_x3_8x8c  : h->pixf.intra_sad_x3_8x8c;
1265     h->pixf.intra_mbcmp_x3_8x8 = satd ? h->pixf.intra_sa8d_x3_8x8 : h->pixf.intra_sad_x3_8x8;
1266     h->pixf.intra_mbcmp_x3_4x4 = satd ? h->pixf.intra_satd_x3_4x4 : h->pixf.intra_sad_x3_4x4;
1267     h->pixf.intra_mbcmp_x9_4x4 = h->param.b_cpu_independent || h->mb.b_lossless ? NULL
1268                                : satd ? h->pixf.intra_satd_x9_4x4 : h->pixf.intra_sad_x9_4x4;
1269     h->pixf.intra_mbcmp_x9_8x8 = h->param.b_cpu_independent || h->mb.b_lossless ? NULL
1270                                : satd ? h->pixf.intra_sa8d_x9_8x8 : h->pixf.intra_sad_x9_8x8;
1271     satd &= h->param.analyse.i_me_method == X264_ME_TESA;
1272     memcpy( h->pixf.fpelcmp, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.fpelcmp) );
1273     memcpy( h->pixf.fpelcmp_x3, satd ? h->pixf.satd_x3 : h->pixf.sad_x3, sizeof(h->pixf.fpelcmp_x3) );
1274     memcpy( h->pixf.fpelcmp_x4, satd ? h->pixf.satd_x4 : h->pixf.sad_x4, sizeof(h->pixf.fpelcmp_x4) );
1275 }
1276
1277 static void chroma_dsp_init( x264_t *h )
1278 {
1279     memcpy( h->luma2chroma_pixel, x264_luma2chroma_pixel[CHROMA_FORMAT], sizeof(h->luma2chroma_pixel) );
1280
1281     switch( CHROMA_FORMAT )
1282     {
1283         case CHROMA_420:
1284             memcpy( h->predict_chroma, h->predict_8x8c, sizeof(h->predict_chroma) );
1285             h->mc.prefetch_fenc = h->mc.prefetch_fenc_420;
1286             h->loopf.deblock_chroma[0] = h->loopf.deblock_h_chroma_420;
1287             h->loopf.deblock_chroma_intra[0] = h->loopf.deblock_h_chroma_420_intra;
1288             h->loopf.deblock_chroma_mbaff = h->loopf.deblock_chroma_420_mbaff;
1289             h->loopf.deblock_chroma_intra_mbaff = h->loopf.deblock_chroma_420_intra_mbaff;
1290             h->pixf.intra_mbcmp_x3_chroma = h->pixf.intra_mbcmp_x3_8x8c;
1291             h->quantf.coeff_last[DCT_CHROMA_DC] = h->quantf.coeff_last4;
1292             h->quantf.coeff_level_run[DCT_CHROMA_DC] = h->quantf.coeff_level_run4;
1293             break;
1294         case CHROMA_422:
1295             memcpy( h->predict_chroma, h->predict_8x16c, sizeof(h->predict_chroma) );
1296             h->mc.prefetch_fenc = h->mc.prefetch_fenc_422;
1297             h->loopf.deblock_chroma[0] = h->loopf.deblock_h_chroma_422;
1298             h->loopf.deblock_chroma_intra[0] = h->loopf.deblock_h_chroma_422_intra;
1299             h->loopf.deblock_chroma_mbaff = h->loopf.deblock_chroma_422_mbaff;
1300             h->loopf.deblock_chroma_intra_mbaff = h->loopf.deblock_chroma_422_intra_mbaff;
1301             h->pixf.intra_mbcmp_x3_chroma = h->pixf.intra_mbcmp_x3_8x16c;
1302             h->quantf.coeff_last[DCT_CHROMA_DC] = h->quantf.coeff_last8;
1303             h->quantf.coeff_level_run[DCT_CHROMA_DC] = h->quantf.coeff_level_run8;
1304             break;
1305         case CHROMA_444:
1306             h->mc.prefetch_fenc = h->mc.prefetch_fenc_422; /* FIXME: doesn't cover V plane */
1307             h->loopf.deblock_chroma_mbaff = h->loopf.deblock_luma_mbaff;
1308             h->loopf.deblock_chroma_intra_mbaff = h->loopf.deblock_luma_intra_mbaff;
1309             break;
1310     }
1311 }
1312
1313 static void x264_set_aspect_ratio( x264_t *h, x264_param_t *param, int initial )
1314 {
1315     /* VUI */
1316     if( param->vui.i_sar_width > 0 && param->vui.i_sar_height > 0 )
1317     {
1318         uint32_t i_w = param->vui.i_sar_width;
1319         uint32_t i_h = param->vui.i_sar_height;
1320         uint32_t old_w = h->param.vui.i_sar_width;
1321         uint32_t old_h = h->param.vui.i_sar_height;
1322
1323         x264_reduce_fraction( &i_w, &i_h );
1324
1325         while( i_w > 65535 || i_h > 65535 )
1326         {
1327             i_w /= 2;
1328             i_h /= 2;
1329         }
1330
1331         x264_reduce_fraction( &i_w, &i_h );
1332
1333         if( i_w != old_w || i_h != old_h || initial )
1334         {
1335             h->param.vui.i_sar_width = 0;
1336             h->param.vui.i_sar_height = 0;
1337             if( i_w == 0 || i_h == 0 )
1338                 x264_log( h, X264_LOG_WARNING, "cannot create valid sample aspect ratio\n" );
1339             else
1340             {
1341                 x264_log( h, initial?X264_LOG_INFO:X264_LOG_DEBUG, "using SAR=%d/%d\n", i_w, i_h );
1342                 h->param.vui.i_sar_width = i_w;
1343                 h->param.vui.i_sar_height = i_h;
1344             }
1345         }
1346     }
1347 }
1348
1349 /****************************************************************************
1350  * x264_encoder_open:
1351  ****************************************************************************/
1352 x264_t *x264_encoder_open( x264_param_t *param )
1353 {
1354     x264_t *h;
1355     char buf[1000], *p;
1356     int qp, i_slicetype_length;
1357
1358     CHECKED_MALLOCZERO( h, sizeof(x264_t) );
1359
1360     /* Create a copy of param */
1361     memcpy( &h->param, param, sizeof(x264_param_t) );
1362
1363     if( param->param_free )
1364         param->param_free( param );
1365
1366     if( x264_threading_init() )
1367     {
1368         x264_log( h, X264_LOG_ERROR, "unable to initialize threading\n" );
1369         goto fail;
1370     }
1371
1372     if( x264_validate_parameters( h, 1 ) < 0 )
1373         goto fail;
1374
1375     if( h->param.psz_cqm_file )
1376         if( x264_cqm_parse_file( h, h->param.psz_cqm_file ) < 0 )
1377             goto fail;
1378
1379     if( h->param.rc.psz_stat_out )
1380         h->param.rc.psz_stat_out = strdup( h->param.rc.psz_stat_out );
1381     if( h->param.rc.psz_stat_in )
1382         h->param.rc.psz_stat_in = strdup( h->param.rc.psz_stat_in );
1383
1384     x264_reduce_fraction( &h->param.i_fps_num, &h->param.i_fps_den );
1385     x264_reduce_fraction( &h->param.i_timebase_num, &h->param.i_timebase_den );
1386
1387     /* Init x264_t */
1388     h->i_frame = -1;
1389     h->i_frame_num = 0;
1390     h->i_idr_pic_id = 0;
1391
1392     if( (uint64_t)h->param.i_timebase_den * 2 > UINT32_MAX )
1393     {
1394         x264_log( h, X264_LOG_ERROR, "Effective timebase denominator %u exceeds H.264 maximum\n", h->param.i_timebase_den );
1395         goto fail;
1396     }
1397
1398     x264_set_aspect_ratio( h, &h->param, 1 );
1399
1400     x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
1401     x264_pps_init( h->pps, h->param.i_sps_id, &h->param, h->sps );
1402
1403     x264_validate_levels( h, 1 );
1404
1405     h->chroma_qp_table = i_chroma_qp_table + 12 + h->pps->i_chroma_qp_index_offset;
1406
1407     if( x264_cqm_init( h ) < 0 )
1408         goto fail;
1409
1410     h->mb.i_mb_width = h->sps->i_mb_width;
1411     h->mb.i_mb_height = h->sps->i_mb_height;
1412     h->mb.i_mb_count = h->mb.i_mb_width * h->mb.i_mb_height;
1413
1414     h->mb.chroma_h_shift = CHROMA_FORMAT == CHROMA_420 || CHROMA_FORMAT == CHROMA_422;
1415     h->mb.chroma_v_shift = CHROMA_FORMAT == CHROMA_420;
1416
1417     /* Adaptive MBAFF and subme 0 are not supported as we require halving motion
1418      * vectors during prediction, resulting in hpel mvs.
1419      * The chosen solution is to make MBAFF non-adaptive in this case. */
1420     h->mb.b_adaptive_mbaff = PARAM_INTERLACED && h->param.analyse.i_subpel_refine;
1421
1422     /* Init frames. */
1423     if( h->param.i_bframe_adaptive == X264_B_ADAPT_TRELLIS && !h->param.rc.b_stat_read )
1424         h->frames.i_delay = X264_MAX(h->param.i_bframe,3)*4;
1425     else
1426         h->frames.i_delay = h->param.i_bframe;
1427     if( h->param.rc.b_mb_tree || h->param.rc.i_vbv_buffer_size )
1428         h->frames.i_delay = X264_MAX( h->frames.i_delay, h->param.rc.i_lookahead );
1429     i_slicetype_length = h->frames.i_delay;
1430     h->frames.i_delay += h->i_thread_frames - 1;
1431     h->frames.i_delay += h->param.i_sync_lookahead;
1432     h->frames.i_delay += h->param.b_vfr_input;
1433     h->frames.i_bframe_delay = h->param.i_bframe ? (h->param.i_bframe_pyramid ? 2 : 1) : 0;
1434
1435     h->frames.i_max_ref0 = h->param.i_frame_reference;
1436     h->frames.i_max_ref1 = X264_MIN( h->sps->vui.i_num_reorder_frames, h->param.i_frame_reference );
1437     h->frames.i_max_dpb  = h->sps->vui.i_max_dec_frame_buffering;
1438     h->frames.b_have_lowres = !h->param.rc.b_stat_read
1439         && ( h->param.rc.i_rc_method == X264_RC_ABR
1440           || h->param.rc.i_rc_method == X264_RC_CRF
1441           || h->param.i_bframe_adaptive
1442           || h->param.i_scenecut_threshold
1443           || h->param.rc.b_mb_tree
1444           || h->param.analyse.i_weighted_pred );
1445     h->frames.b_have_lowres |= h->param.rc.b_stat_read && h->param.rc.i_vbv_buffer_size > 0;
1446     h->frames.b_have_sub8x8_esa = !!(h->param.analyse.inter & X264_ANALYSE_PSUB8x8);
1447
1448     h->frames.i_last_idr =
1449     h->frames.i_last_keyframe = - h->param.i_keyint_max;
1450     h->frames.i_input    = 0;
1451     h->frames.i_largest_pts = h->frames.i_second_largest_pts = -1;
1452     h->frames.i_poc_last_open_gop = -1;
1453
1454     CHECKED_MALLOCZERO( h->frames.unused[0], (h->frames.i_delay + 3) * sizeof(x264_frame_t *) );
1455     /* Allocate room for max refs plus a few extra just in case. */
1456     CHECKED_MALLOCZERO( h->frames.unused[1], (h->i_thread_frames + X264_REF_MAX + 4) * sizeof(x264_frame_t *) );
1457     CHECKED_MALLOCZERO( h->frames.current, (h->param.i_sync_lookahead + h->param.i_bframe
1458                         + h->i_thread_frames + 3) * sizeof(x264_frame_t *) );
1459     if( h->param.analyse.i_weighted_pred > 0 )
1460         CHECKED_MALLOCZERO( h->frames.blank_unused, h->i_thread_frames * 4 * sizeof(x264_frame_t *) );
1461     h->i_ref[0] = h->i_ref[1] = 0;
1462     h->i_cpb_delay = h->i_coded_fields = h->i_disp_fields = 0;
1463     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);
1464     h->i_disp_fields_last_frame = -1;
1465     x264_rdo_init();
1466
1467     /* init CPU functions */
1468     x264_predict_16x16_init( h->param.cpu, h->predict_16x16 );
1469     x264_predict_8x8c_init( h->param.cpu, h->predict_8x8c );
1470     x264_predict_8x16c_init( h->param.cpu, h->predict_8x16c );
1471     x264_predict_8x8_init( h->param.cpu, h->predict_8x8, &h->predict_8x8_filter );
1472     x264_predict_4x4_init( h->param.cpu, h->predict_4x4 );
1473     x264_pixel_init( h->param.cpu, &h->pixf );
1474     x264_dct_init( h->param.cpu, &h->dctf );
1475     x264_zigzag_init( h->param.cpu, &h->zigzagf_progressive, &h->zigzagf_interlaced );
1476     memcpy( &h->zigzagf, PARAM_INTERLACED ? &h->zigzagf_interlaced : &h->zigzagf_progressive, sizeof(h->zigzagf) );
1477     x264_mc_init( h->param.cpu, &h->mc, h->param.b_cpu_independent );
1478     x264_quant_init( h, h->param.cpu, &h->quantf );
1479     x264_deblock_init( h->param.cpu, &h->loopf, PARAM_INTERLACED );
1480     x264_bitstream_init( h->param.cpu, &h->bsf );
1481     if( h->param.b_cabac )
1482         x264_cabac_init( h );
1483     else
1484         x264_stack_align( x264_cavlc_init, h );
1485
1486     mbcmp_init( h );
1487     chroma_dsp_init( h );
1488
1489     p = buf + sprintf( buf, "using cpu capabilities:" );
1490     for( int i = 0; x264_cpu_names[i].flags; i++ )
1491     {
1492         if( !strcmp(x264_cpu_names[i].name, "SSE")
1493             && h->param.cpu & (X264_CPU_SSE2) )
1494             continue;
1495         if( !strcmp(x264_cpu_names[i].name, "SSE2")
1496             && h->param.cpu & (X264_CPU_SSE2_IS_FAST|X264_CPU_SSE2_IS_SLOW) )
1497             continue;
1498         if( !strcmp(x264_cpu_names[i].name, "SSE3")
1499             && (h->param.cpu & X264_CPU_SSSE3 || !(h->param.cpu & X264_CPU_CACHELINE_64)) )
1500             continue;
1501         if( !strcmp(x264_cpu_names[i].name, "SSE4.1")
1502             && (h->param.cpu & X264_CPU_SSE42) )
1503             continue;
1504         if( !strcmp(x264_cpu_names[i].name, "BMI1")
1505             && (h->param.cpu & X264_CPU_BMI2) )
1506             continue;
1507         if( (h->param.cpu & x264_cpu_names[i].flags) == x264_cpu_names[i].flags
1508             && (!i || x264_cpu_names[i].flags != x264_cpu_names[i-1].flags) )
1509             p += sprintf( p, " %s", x264_cpu_names[i].name );
1510     }
1511     if( !h->param.cpu )
1512         p += sprintf( p, " none!" );
1513     x264_log( h, X264_LOG_INFO, "%s\n", buf );
1514
1515     float *logs = x264_analyse_prepare_costs( h );
1516     if( !logs )
1517         goto fail;
1518     for( qp = X264_MIN( h->param.rc.i_qp_min, QP_MAX_SPEC ); qp <= h->param.rc.i_qp_max; qp++ )
1519         if( x264_analyse_init_costs( h, logs, qp ) )
1520             goto fail;
1521     if( x264_analyse_init_costs( h, logs, X264_LOOKAHEAD_QP ) )
1522         goto fail;
1523     x264_free( logs );
1524
1525     static const uint16_t cost_mv_correct[7] = { 24, 47, 95, 189, 379, 757, 1515 };
1526     /* Checks for known miscompilation issues. */
1527     if( h->cost_mv[X264_LOOKAHEAD_QP][2013] != cost_mv_correct[BIT_DEPTH-8] )
1528     {
1529         x264_log( h, X264_LOG_ERROR, "MV cost test failed: x264 has been miscompiled!\n" );
1530         goto fail;
1531     }
1532
1533     /* Must be volatile or else GCC will optimize it out. */
1534     volatile int temp = 392;
1535     if( x264_clz( temp ) != 23 )
1536     {
1537         x264_log( h, X264_LOG_ERROR, "CLZ test failed: x264 has been miscompiled!\n" );
1538 #if ARCH_X86 || ARCH_X86_64
1539         x264_log( h, X264_LOG_ERROR, "Are you attempting to run an SSE4a/LZCNT-targeted build on a CPU that\n" );
1540         x264_log( h, X264_LOG_ERROR, "doesn't support it?\n" );
1541 #endif
1542         goto fail;
1543     }
1544
1545     h->out.i_nal = 0;
1546     h->out.i_bitstream = X264_MAX( 1000000, h->param.i_width * h->param.i_height * 4
1547         * ( h->param.rc.i_rc_method == X264_RC_ABR ? pow( 0.95, h->param.rc.i_qp_min )
1548           : pow( 0.95, h->param.rc.i_qp_constant ) * X264_MAX( 1, h->param.rc.f_ip_factor )));
1549
1550     h->nal_buffer_size = h->out.i_bitstream * 3/2 + 4 + 64; /* +4 for startcode, +64 for nal_escape assembly padding */
1551     CHECKED_MALLOC( h->nal_buffer, h->nal_buffer_size );
1552
1553     CHECKED_MALLOC( h->reconfig_h, sizeof(x264_t) );
1554
1555     if( h->param.i_threads > 1 &&
1556         x264_threadpool_init( &h->threadpool, h->param.i_threads, (void*)x264_encoder_thread_init, h ) )
1557         goto fail;
1558     if( h->param.i_lookahead_threads > 1 &&
1559         x264_threadpool_init( &h->lookaheadpool, h->param.i_lookahead_threads, NULL, NULL ) )
1560         goto fail;
1561
1562 #if HAVE_OPENCL
1563     if( h->param.b_opencl )
1564     {
1565         h->opencl.ocl = x264_opencl_load_library();
1566         if( !h->opencl.ocl )
1567         {
1568             x264_log( h, X264_LOG_WARNING, "failed to load OpenCL\n" );
1569             h->param.b_opencl = 0;
1570         }
1571     }
1572 #endif
1573
1574     h->thread[0] = h;
1575     for( int i = 1; i < h->param.i_threads + !!h->param.i_sync_lookahead; i++ )
1576         CHECKED_MALLOC( h->thread[i], sizeof(x264_t) );
1577     if( h->param.i_lookahead_threads > 1 )
1578         for( int i = 0; i < h->param.i_lookahead_threads; i++ )
1579         {
1580             CHECKED_MALLOC( h->lookahead_thread[i], sizeof(x264_t) );
1581             *h->lookahead_thread[i] = *h;
1582         }
1583     *h->reconfig_h = *h;
1584
1585     for( int i = 0; i < h->param.i_threads; i++ )
1586     {
1587         int init_nal_count = h->param.i_slice_count + 3;
1588         int allocate_threadlocal_data = !h->param.b_sliced_threads || !i;
1589         if( i > 0 )
1590             *h->thread[i] = *h;
1591
1592         if( x264_pthread_mutex_init( &h->thread[i]->mutex, NULL ) )
1593             goto fail;
1594         if( x264_pthread_cond_init( &h->thread[i]->cv, NULL ) )
1595             goto fail;
1596
1597         if( allocate_threadlocal_data )
1598         {
1599             h->thread[i]->fdec = x264_frame_pop_unused( h, 1 );
1600             if( !h->thread[i]->fdec )
1601                 goto fail;
1602         }
1603         else
1604             h->thread[i]->fdec = h->thread[0]->fdec;
1605
1606         CHECKED_MALLOC( h->thread[i]->out.p_bitstream, h->out.i_bitstream );
1607         /* Start each thread with room for init_nal_count NAL units; it'll realloc later if needed. */
1608         CHECKED_MALLOC( h->thread[i]->out.nal, init_nal_count*sizeof(x264_nal_t) );
1609         h->thread[i]->out.i_nals_allocated = init_nal_count;
1610
1611         if( allocate_threadlocal_data && x264_macroblock_cache_allocate( h->thread[i] ) < 0 )
1612             goto fail;
1613     }
1614
1615 #if HAVE_OPENCL
1616     if( h->param.b_opencl && x264_opencl_lookahead_init( h ) < 0 )
1617         h->param.b_opencl = 0;
1618 #endif
1619
1620     if( x264_lookahead_init( h, i_slicetype_length ) )
1621         goto fail;
1622
1623     for( int i = 0; i < h->param.i_threads; i++ )
1624         if( x264_macroblock_thread_allocate( h->thread[i], 0 ) < 0 )
1625             goto fail;
1626
1627     if( x264_ratecontrol_new( h ) < 0 )
1628         goto fail;
1629
1630     if( h->param.i_nal_hrd )
1631     {
1632         x264_log( h, X264_LOG_DEBUG, "HRD bitrate: %i bits/sec\n", h->sps->vui.hrd.i_bit_rate_unscaled );
1633         x264_log( h, X264_LOG_DEBUG, "CPB size: %i bits\n", h->sps->vui.hrd.i_cpb_size_unscaled );
1634     }
1635
1636     if( h->param.psz_dump_yuv )
1637     {
1638         /* create or truncate the reconstructed video file */
1639         FILE *f = x264_fopen( h->param.psz_dump_yuv, "w" );
1640         if( !f )
1641         {
1642             x264_log( h, X264_LOG_ERROR, "dump_yuv: can't write to %s\n", h->param.psz_dump_yuv );
1643             goto fail;
1644         }
1645         else if( !x264_is_regular_file( f ) )
1646         {
1647             x264_log( h, X264_LOG_ERROR, "dump_yuv: incompatible with non-regular file %s\n", h->param.psz_dump_yuv );
1648             goto fail;
1649         }
1650         fclose( f );
1651     }
1652
1653     const char *profile = h->sps->i_profile_idc == PROFILE_BASELINE ? "Constrained Baseline" :
1654                           h->sps->i_profile_idc == PROFILE_MAIN ? "Main" :
1655                           h->sps->i_profile_idc == PROFILE_HIGH ? "High" :
1656                           h->sps->i_profile_idc == PROFILE_HIGH10 ? (h->sps->b_constraint_set3 == 1 ? "High 10 Intra" : "High 10") :
1657                           h->sps->i_profile_idc == PROFILE_HIGH422 ? (h->sps->b_constraint_set3 == 1 ? "High 4:2:2 Intra" : "High 4:2:2") :
1658                           h->sps->b_constraint_set3 == 1 ? "High 4:4:4 Intra" : "High 4:4:4 Predictive";
1659     char level[4];
1660     snprintf( level, sizeof(level), "%d.%d", h->sps->i_level_idc/10, h->sps->i_level_idc%10 );
1661     if( h->sps->i_level_idc == 9 || ( h->sps->i_level_idc == 11 && h->sps->b_constraint_set3 &&
1662         (h->sps->i_profile_idc == PROFILE_BASELINE || h->sps->i_profile_idc == PROFILE_MAIN) ) )
1663         strcpy( level, "1b" );
1664
1665     if( h->sps->i_profile_idc < PROFILE_HIGH10 )
1666     {
1667         x264_log( h, X264_LOG_INFO, "profile %s, level %s\n",
1668             profile, level );
1669     }
1670     else
1671     {
1672         static const char * const subsampling[4] = { "4:0:0", "4:2:0", "4:2:2", "4:4:4" };
1673         x264_log( h, X264_LOG_INFO, "profile %s, level %s, %s %d-bit\n",
1674             profile, level, subsampling[CHROMA_FORMAT], BIT_DEPTH );
1675     }
1676
1677     return h;
1678 fail:
1679     x264_free( h );
1680     return NULL;
1681 }
1682
1683 /****************************************************************************/
1684 static int x264_encoder_try_reconfig( x264_t *h, x264_param_t *param, int *rc_reconfig )
1685 {
1686     *rc_reconfig = 0;
1687     x264_set_aspect_ratio( h, param, 0 );
1688 #define COPY(var) h->param.var = param->var
1689     COPY( i_frame_reference ); // but never uses more refs than initially specified
1690     COPY( i_bframe_bias );
1691     if( h->param.i_scenecut_threshold )
1692         COPY( i_scenecut_threshold ); // can't turn it on or off, only vary the threshold
1693     COPY( b_deblocking_filter );
1694     COPY( i_deblocking_filter_alphac0 );
1695     COPY( i_deblocking_filter_beta );
1696     COPY( i_frame_packing );
1697     COPY( analyse.inter );
1698     COPY( analyse.intra );
1699     COPY( analyse.i_direct_mv_pred );
1700     /* Scratch buffer prevents me_range from being increased for esa/tesa */
1701     if( h->param.analyse.i_me_method < X264_ME_ESA || param->analyse.i_me_range < h->param.analyse.i_me_range )
1702         COPY( analyse.i_me_range );
1703     COPY( analyse.i_noise_reduction );
1704     /* We can't switch out of subme=0 during encoding. */
1705     if( h->param.analyse.i_subpel_refine )
1706         COPY( analyse.i_subpel_refine );
1707     COPY( analyse.i_trellis );
1708     COPY( analyse.b_chroma_me );
1709     COPY( analyse.b_dct_decimate );
1710     COPY( analyse.b_fast_pskip );
1711     COPY( analyse.b_mixed_references );
1712     COPY( analyse.f_psy_rd );
1713     COPY( analyse.f_psy_trellis );
1714     COPY( crop_rect );
1715     // can only twiddle these if they were enabled to begin with:
1716     if( h->param.analyse.i_me_method >= X264_ME_ESA || param->analyse.i_me_method < X264_ME_ESA )
1717         COPY( analyse.i_me_method );
1718     if( h->param.analyse.i_me_method >= X264_ME_ESA && !h->frames.b_have_sub8x8_esa )
1719         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
1720     if( h->pps->b_transform_8x8_mode )
1721         COPY( analyse.b_transform_8x8 );
1722     if( h->frames.i_max_ref1 > 1 )
1723         COPY( i_bframe_pyramid );
1724     COPY( i_slice_max_size );
1725     COPY( i_slice_max_mbs );
1726     COPY( i_slice_min_mbs );
1727     COPY( i_slice_count );
1728     COPY( i_slice_count_max );
1729     COPY( b_tff );
1730
1731     /* VBV can't be turned on if it wasn't on to begin with */
1732     if( h->param.rc.i_vbv_max_bitrate > 0 && h->param.rc.i_vbv_buffer_size > 0 &&
1733           param->rc.i_vbv_max_bitrate > 0 &&   param->rc.i_vbv_buffer_size > 0 )
1734     {
1735         *rc_reconfig |= h->param.rc.i_vbv_max_bitrate != param->rc.i_vbv_max_bitrate;
1736         *rc_reconfig |= h->param.rc.i_vbv_buffer_size != param->rc.i_vbv_buffer_size;
1737         *rc_reconfig |= h->param.rc.i_bitrate != param->rc.i_bitrate;
1738         COPY( rc.i_vbv_max_bitrate );
1739         COPY( rc.i_vbv_buffer_size );
1740         COPY( rc.i_bitrate );
1741     }
1742     *rc_reconfig |= h->param.rc.f_rf_constant != param->rc.f_rf_constant;
1743     *rc_reconfig |= h->param.rc.f_rf_constant_max != param->rc.f_rf_constant_max;
1744     COPY( rc.f_rf_constant );
1745     COPY( rc.f_rf_constant_max );
1746 #undef COPY
1747
1748     return x264_validate_parameters( h, 0 );
1749 }
1750
1751 int x264_encoder_reconfig_apply( x264_t *h, x264_param_t *param )
1752 {
1753     int rc_reconfig;
1754     int ret = x264_encoder_try_reconfig( h, param, &rc_reconfig );
1755
1756     mbcmp_init( h );
1757     if( !ret )
1758         x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
1759
1760     /* Supported reconfiguration options (1-pass only):
1761      * vbv-maxrate
1762      * vbv-bufsize
1763      * crf
1764      * bitrate (CBR only) */
1765     if( !ret && rc_reconfig )
1766         x264_ratecontrol_init_reconfigurable( h, 0 );
1767
1768     return ret;
1769 }
1770
1771 /****************************************************************************
1772  * x264_encoder_reconfig:
1773  ****************************************************************************/
1774 int x264_encoder_reconfig( x264_t *h, x264_param_t *param )
1775 {
1776     h = h->thread[h->thread[0]->i_thread_phase];
1777     x264_param_t param_save = h->reconfig_h->param;
1778     h->reconfig_h->param = h->param;
1779
1780     int rc_reconfig;
1781     int ret = x264_encoder_try_reconfig( h->reconfig_h, param, &rc_reconfig );
1782     if( !ret )
1783         h->reconfig = 1;
1784     else
1785         h->reconfig_h->param = param_save;
1786
1787     return ret;
1788 }
1789
1790 /****************************************************************************
1791  * x264_encoder_parameters:
1792  ****************************************************************************/
1793 void x264_encoder_parameters( x264_t *h, x264_param_t *param )
1794 {
1795     memcpy( param, &h->thread[h->i_thread_phase]->param, sizeof(x264_param_t) );
1796 }
1797
1798 /* internal usage */
1799 static void x264_nal_start( x264_t *h, int i_type, int i_ref_idc )
1800 {
1801     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1802
1803     nal->i_ref_idc        = i_ref_idc;
1804     nal->i_type           = i_type;
1805     nal->b_long_startcode = 1;
1806
1807     nal->i_payload= 0;
1808     nal->p_payload= &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8];
1809     nal->i_padding= 0;
1810 }
1811
1812 /* if number of allocated nals is not enough, re-allocate a larger one. */
1813 static int x264_nal_check_buffer( x264_t *h )
1814 {
1815     if( h->out.i_nal >= h->out.i_nals_allocated )
1816     {
1817         x264_nal_t *new_out = x264_malloc( sizeof(x264_nal_t) * (h->out.i_nals_allocated*2) );
1818         if( !new_out )
1819             return -1;
1820         memcpy( new_out, h->out.nal, sizeof(x264_nal_t) * (h->out.i_nals_allocated) );
1821         x264_free( h->out.nal );
1822         h->out.nal = new_out;
1823         h->out.i_nals_allocated *= 2;
1824     }
1825     return 0;
1826 }
1827
1828 static int x264_nal_end( x264_t *h )
1829 {
1830     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1831     uint8_t *end = &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8];
1832     nal->i_payload = end - nal->p_payload;
1833     /* Assembly implementation of nal_escape reads past the end of the input.
1834      * While undefined padding wouldn't actually affect the output, it makes valgrind unhappy. */
1835     memset( end, 0xff, 64 );
1836     if( h->param.nalu_process )
1837         h->param.nalu_process( h, nal, h->fenc->opaque );
1838     h->out.i_nal++;
1839
1840     return x264_nal_check_buffer( h );
1841 }
1842
1843 static int x264_check_encapsulated_buffer( x264_t *h, x264_t *h0, int start,
1844                                            int previous_nal_size, int necessary_size )
1845 {
1846     if( h0->nal_buffer_size < necessary_size )
1847     {
1848         necessary_size *= 2;
1849         uint8_t *buf = x264_malloc( necessary_size );
1850         if( !buf )
1851             return -1;
1852         if( previous_nal_size )
1853             memcpy( buf, h0->nal_buffer, previous_nal_size );
1854
1855         intptr_t delta = buf - h0->nal_buffer;
1856         for( int i = 0; i < start; i++ )
1857             h->out.nal[i].p_payload += delta;
1858
1859         x264_free( h0->nal_buffer );
1860         h0->nal_buffer = buf;
1861         h0->nal_buffer_size = necessary_size;
1862     }
1863
1864     return 0;
1865 }
1866
1867 static int x264_encoder_encapsulate_nals( x264_t *h, int start )
1868 {
1869     x264_t *h0 = h->thread[0];
1870     int nal_size = 0, previous_nal_size = 0;
1871
1872     if( h->param.nalu_process )
1873     {
1874         for( int i = start; i < h->out.i_nal; i++ )
1875             nal_size += h->out.nal[i].i_payload;
1876         return nal_size;
1877     }
1878
1879     for( int i = 0; i < start; i++ )
1880         previous_nal_size += h->out.nal[i].i_payload;
1881
1882     for( int i = start; i < h->out.i_nal; i++ )
1883         nal_size += h->out.nal[i].i_payload;
1884
1885     /* Worst-case NAL unit escaping: reallocate the buffer if it's too small. */
1886     int necessary_size = previous_nal_size + nal_size * 3/2 + h->out.i_nal * 4 + 4 + 64;
1887     for( int i = start; i < h->out.i_nal; i++ )
1888         necessary_size += h->out.nal[i].i_padding;
1889     if( x264_check_encapsulated_buffer( h, h0, start, previous_nal_size, necessary_size ) )
1890         return -1;
1891
1892     uint8_t *nal_buffer = h0->nal_buffer + previous_nal_size;
1893
1894     for( int i = start; i < h->out.i_nal; i++ )
1895     {
1896         int old_payload_len = h->out.nal[i].i_payload;
1897         h->out.nal[i].b_long_startcode = !i || h->out.nal[i].i_type == NAL_SPS || h->out.nal[i].i_type == NAL_PPS ||
1898                                          h->param.b_avcintra_compat;
1899         x264_nal_encode( h, nal_buffer, &h->out.nal[i] );
1900         nal_buffer += h->out.nal[i].i_payload;
1901         if( h->param.b_avcintra_compat )
1902         {
1903             h->out.nal[i].i_padding -= h->out.nal[i].i_payload - (old_payload_len + NALU_OVERHEAD);
1904             if( h->out.nal[i].i_padding > 0 )
1905             {
1906                 memset( nal_buffer, 0, h->out.nal[i].i_padding );
1907                 nal_buffer += h->out.nal[i].i_padding;
1908                 h->out.nal[i].i_payload += h->out.nal[i].i_padding;
1909             }
1910             h->out.nal[i].i_padding = X264_MAX( h->out.nal[i].i_padding, 0 );
1911         }
1912     }
1913
1914     x264_emms();
1915
1916     return nal_buffer - (h0->nal_buffer + previous_nal_size);
1917 }
1918
1919 /****************************************************************************
1920  * x264_encoder_headers:
1921  ****************************************************************************/
1922 int x264_encoder_headers( x264_t *h, x264_nal_t **pp_nal, int *pi_nal )
1923 {
1924     int frame_size = 0;
1925     /* init bitstream context */
1926     h->out.i_nal = 0;
1927     bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
1928
1929     /* Write SEI, SPS and PPS. */
1930
1931     /* generate sequence parameters */
1932     x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
1933     x264_sps_write( &h->out.bs, h->sps );
1934     if( x264_nal_end( h ) )
1935         return -1;
1936
1937     /* generate picture parameters */
1938     x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
1939     x264_pps_write( &h->out.bs, h->sps, h->pps );
1940     if( x264_nal_end( h ) )
1941         return -1;
1942
1943     /* identify ourselves */
1944     x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
1945     if( x264_sei_version_write( h, &h->out.bs ) )
1946         return -1;
1947     if( x264_nal_end( h ) )
1948         return -1;
1949
1950     frame_size = x264_encoder_encapsulate_nals( h, 0 );
1951     if( frame_size < 0 )
1952         return -1;
1953
1954     /* now set output*/
1955     *pi_nal = h->out.i_nal;
1956     *pp_nal = &h->out.nal[0];
1957     h->out.i_nal = 0;
1958
1959     return frame_size;
1960 }
1961
1962 /* Check to see whether we have chosen a reference list ordering different
1963  * from the standard's default. */
1964 static inline void x264_reference_check_reorder( x264_t *h )
1965 {
1966     /* The reorder check doesn't check for missing frames, so just
1967      * force a reorder if one of the reference list is corrupt. */
1968     for( int i = 0; h->frames.reference[i]; i++ )
1969         if( h->frames.reference[i]->b_corrupt )
1970         {
1971             h->b_ref_reorder[0] = 1;
1972             return;
1973         }
1974     for( int list = 0; list <= (h->sh.i_type == SLICE_TYPE_B); list++ )
1975         for( int i = 0; i < h->i_ref[list] - 1; i++ )
1976         {
1977             int framenum_diff = h->fref[list][i+1]->i_frame_num - h->fref[list][i]->i_frame_num;
1978             int poc_diff = h->fref[list][i+1]->i_poc - h->fref[list][i]->i_poc;
1979             /* P and B-frames use different default orders. */
1980             if( h->sh.i_type == SLICE_TYPE_P ? framenum_diff > 0 : list == 1 ? poc_diff < 0 : poc_diff > 0 )
1981             {
1982                 h->b_ref_reorder[list] = 1;
1983                 return;
1984             }
1985         }
1986 }
1987
1988 /* return -1 on failure, else return the index of the new reference frame */
1989 int x264_weighted_reference_duplicate( x264_t *h, int i_ref, const x264_weight_t *w )
1990 {
1991     int i = h->i_ref[0];
1992     int j = 1;
1993     x264_frame_t *newframe;
1994     if( i <= 1 ) /* empty list, definitely can't duplicate frame */
1995         return -1;
1996
1997     //Duplication is only used in X264_WEIGHTP_SMART
1998     if( h->param.analyse.i_weighted_pred != X264_WEIGHTP_SMART )
1999         return -1;
2000
2001     /* Duplication is a hack to compensate for crappy rounding in motion compensation.
2002      * With high bit depth, it's not worth doing, so turn it off except in the case of
2003      * unweighted dupes. */
2004     if( BIT_DEPTH > 8 && w != x264_weight_none )
2005         return -1;
2006
2007     newframe = x264_frame_pop_blank_unused( h );
2008     if( !newframe )
2009         return -1;
2010
2011     //FIXME: probably don't need to copy everything
2012     *newframe = *h->fref[0][i_ref];
2013     newframe->i_reference_count = 1;
2014     newframe->orig = h->fref[0][i_ref];
2015     newframe->b_duplicate = 1;
2016     memcpy( h->fenc->weight[j], w, sizeof(h->fenc->weight[i]) );
2017
2018     /* shift the frames to make space for the dupe. */
2019     h->b_ref_reorder[0] = 1;
2020     if( h->i_ref[0] < X264_REF_MAX )
2021         ++h->i_ref[0];
2022     h->fref[0][X264_REF_MAX-1] = NULL;
2023     x264_frame_unshift( &h->fref[0][j], newframe );
2024
2025     return j;
2026 }
2027
2028 static void x264_weighted_pred_init( x264_t *h )
2029 {
2030     /* for now no analysis and set all weights to nothing */
2031     for( int i_ref = 0; i_ref < h->i_ref[0]; i_ref++ )
2032         h->fenc->weighted[i_ref] = h->fref[0][i_ref]->filtered[0][0];
2033
2034     // FIXME: This only supports weighting of one reference frame
2035     // and duplicates of that frame.
2036     h->fenc->i_lines_weighted = 0;
2037
2038     for( int i_ref = 0; i_ref < (h->i_ref[0] << SLICE_MBAFF); i_ref++ )
2039         for( int i = 0; i < 3; i++ )
2040             h->sh.weight[i_ref][i].weightfn = NULL;
2041
2042
2043     if( h->sh.i_type != SLICE_TYPE_P || h->param.analyse.i_weighted_pred <= 0 )
2044         return;
2045
2046     int i_padv = PADV << PARAM_INTERLACED;
2047     int denom = -1;
2048     int weightplane[2] = { 0, 0 };
2049     int buffer_next = 0;
2050     for( int i = 0; i < 3; i++ )
2051     {
2052         for( int j = 0; j < h->i_ref[0]; j++ )
2053         {
2054             if( h->fenc->weight[j][i].weightfn )
2055             {
2056                 h->sh.weight[j][i] = h->fenc->weight[j][i];
2057                 // if weight is useless, don't write it to stream
2058                 if( h->sh.weight[j][i].i_scale == 1<<h->sh.weight[j][i].i_denom && h->sh.weight[j][i].i_offset == 0 )
2059                     h->sh.weight[j][i].weightfn = NULL;
2060                 else
2061                 {
2062                     if( !weightplane[!!i] )
2063                     {
2064                         weightplane[!!i] = 1;
2065                         h->sh.weight[0][!!i].i_denom = denom = h->sh.weight[j][i].i_denom;
2066                         assert( x264_clip3( denom, 0, 7 ) == denom );
2067                     }
2068
2069                     assert( h->sh.weight[j][i].i_denom == denom );
2070                     if( !i )
2071                     {
2072                         h->fenc->weighted[j] = h->mb.p_weight_buf[buffer_next++] + h->fenc->i_stride[0] * i_padv + PADH;
2073                         //scale full resolution frame
2074                         if( h->param.i_threads == 1 )
2075                         {
2076                             pixel *src = h->fref[0][j]->filtered[0][0] - h->fref[0][j]->i_stride[0]*i_padv - PADH;
2077                             pixel *dst = h->fenc->weighted[j] - h->fenc->i_stride[0]*i_padv - PADH;
2078                             int stride = h->fenc->i_stride[0];
2079                             int width = h->fenc->i_width[0] + PADH*2;
2080                             int height = h->fenc->i_lines[0] + i_padv*2;
2081                             x264_weight_scale_plane( h, dst, stride, src, stride, width, height, &h->sh.weight[j][0] );
2082                             h->fenc->i_lines_weighted = height;
2083                         }
2084                     }
2085                 }
2086             }
2087         }
2088     }
2089
2090     if( weightplane[1] )
2091         for( int i = 0; i < h->i_ref[0]; i++ )
2092         {
2093             if( h->sh.weight[i][1].weightfn && !h->sh.weight[i][2].weightfn )
2094             {
2095                 h->sh.weight[i][2].i_scale = 1 << h->sh.weight[0][1].i_denom;
2096                 h->sh.weight[i][2].i_offset = 0;
2097             }
2098             else if( h->sh.weight[i][2].weightfn && !h->sh.weight[i][1].weightfn )
2099             {
2100                 h->sh.weight[i][1].i_scale = 1 << h->sh.weight[0][1].i_denom;
2101                 h->sh.weight[i][1].i_offset = 0;
2102             }
2103         }
2104
2105     if( !weightplane[0] )
2106         h->sh.weight[0][0].i_denom = 0;
2107     if( !weightplane[1] )
2108         h->sh.weight[0][1].i_denom = 0;
2109     h->sh.weight[0][2].i_denom = h->sh.weight[0][1].i_denom;
2110 }
2111
2112 static inline int x264_reference_distance( x264_t *h, x264_frame_t *frame )
2113 {
2114     if( h->param.i_frame_packing == 5 )
2115         return abs((h->fenc->i_frame&~1) - (frame->i_frame&~1)) +
2116                   ((h->fenc->i_frame&1) != (frame->i_frame&1));
2117     else
2118         return abs(h->fenc->i_frame - frame->i_frame);
2119 }
2120
2121 static inline void x264_reference_build_list( x264_t *h, int i_poc )
2122 {
2123     int b_ok;
2124
2125     /* build ref list 0/1 */
2126     h->mb.pic.i_fref[0] = h->i_ref[0] = 0;
2127     h->mb.pic.i_fref[1] = h->i_ref[1] = 0;
2128     if( h->sh.i_type == SLICE_TYPE_I )
2129         return;
2130
2131     for( int i = 0; h->frames.reference[i]; i++ )
2132     {
2133         if( h->frames.reference[i]->b_corrupt )
2134             continue;
2135         if( h->frames.reference[i]->i_poc < i_poc )
2136             h->fref[0][h->i_ref[0]++] = h->frames.reference[i];
2137         else if( h->frames.reference[i]->i_poc > i_poc )
2138             h->fref[1][h->i_ref[1]++] = h->frames.reference[i];
2139     }
2140
2141     /* Order reference lists by distance from the current frame. */
2142     for( int list = 0; list < 2; list++ )
2143     {
2144         h->fref_nearest[list] = h->fref[list][0];
2145         do
2146         {
2147             b_ok = 1;
2148             for( int i = 0; i < h->i_ref[list] - 1; i++ )
2149             {
2150                 if( list ? h->fref[list][i+1]->i_poc < h->fref_nearest[list]->i_poc
2151                          : h->fref[list][i+1]->i_poc > h->fref_nearest[list]->i_poc )
2152                     h->fref_nearest[list] = h->fref[list][i+1];
2153                 if( x264_reference_distance( h, h->fref[list][i] ) > x264_reference_distance( h, h->fref[list][i+1] ) )
2154                 {
2155                     XCHG( x264_frame_t*, h->fref[list][i], h->fref[list][i+1] );
2156                     b_ok = 0;
2157                     break;
2158                 }
2159             }
2160         } while( !b_ok );
2161     }
2162
2163     if( h->sh.i_mmco_remove_from_end )
2164         for( int i = h->i_ref[0]-1; i >= h->i_ref[0] - h->sh.i_mmco_remove_from_end; i-- )
2165         {
2166             int diff = h->i_frame_num - h->fref[0][i]->i_frame_num;
2167             h->sh.mmco[h->sh.i_mmco_command_count].i_poc = h->fref[0][i]->i_poc;
2168             h->sh.mmco[h->sh.i_mmco_command_count++].i_difference_of_pic_nums = diff;
2169         }
2170
2171     x264_reference_check_reorder( h );
2172
2173     h->i_ref[1] = X264_MIN( h->i_ref[1], h->frames.i_max_ref1 );
2174     h->i_ref[0] = X264_MIN( h->i_ref[0], h->frames.i_max_ref0 );
2175     h->i_ref[0] = X264_MIN( h->i_ref[0], h->param.i_frame_reference ); // if reconfig() has lowered the limit
2176
2177     /* For Blu-ray compliance, don't reference frames outside of the minigop. */
2178     if( IS_X264_TYPE_B( h->fenc->i_type ) && h->param.b_bluray_compat )
2179         h->i_ref[0] = X264_MIN( h->i_ref[0], IS_X264_TYPE_B( h->fref[0][0]->i_type ) + 1 );
2180
2181     /* add duplicates */
2182     if( h->fenc->i_type == X264_TYPE_P )
2183     {
2184         int idx = -1;
2185         if( h->param.analyse.i_weighted_pred >= X264_WEIGHTP_SIMPLE )
2186         {
2187             x264_weight_t w[3];
2188             w[1].weightfn = w[2].weightfn = NULL;
2189             if( h->param.rc.b_stat_read )
2190                 x264_ratecontrol_set_weights( h, h->fenc );
2191
2192             if( !h->fenc->weight[0][0].weightfn )
2193             {
2194                 h->fenc->weight[0][0].i_denom = 0;
2195                 SET_WEIGHT( w[0], 1, 1, 0, -1 );
2196                 idx = x264_weighted_reference_duplicate( h, 0, w );
2197             }
2198             else
2199             {
2200                 if( h->fenc->weight[0][0].i_scale == 1<<h->fenc->weight[0][0].i_denom )
2201                 {
2202                     SET_WEIGHT( h->fenc->weight[0][0], 1, 1, 0, h->fenc->weight[0][0].i_offset );
2203                 }
2204                 x264_weighted_reference_duplicate( h, 0, x264_weight_none );
2205                 if( h->fenc->weight[0][0].i_offset > -128 )
2206                 {
2207                     w[0] = h->fenc->weight[0][0];
2208                     w[0].i_offset--;
2209                     h->mc.weight_cache( h, &w[0] );
2210                     idx = x264_weighted_reference_duplicate( h, 0, w );
2211                 }
2212             }
2213         }
2214         h->mb.ref_blind_dupe = idx;
2215     }
2216
2217     assert( h->i_ref[0] + h->i_ref[1] <= X264_REF_MAX );
2218     h->mb.pic.i_fref[0] = h->i_ref[0];
2219     h->mb.pic.i_fref[1] = h->i_ref[1];
2220 }
2221
2222 static void x264_fdec_filter_row( x264_t *h, int mb_y, int pass )
2223 {
2224     /* mb_y is the mb to be encoded next, not the mb to be filtered here */
2225     int b_hpel = h->fdec->b_kept_as_ref;
2226     int b_deblock = h->sh.i_disable_deblocking_filter_idc != 1;
2227     int b_end = mb_y == h->i_threadslice_end;
2228     int b_measure_quality = 1;
2229     int min_y = mb_y - (1 << SLICE_MBAFF);
2230     int b_start = min_y == h->i_threadslice_start;
2231     /* Even in interlaced mode, deblocking never modifies more than 4 pixels
2232      * above each MB, as bS=4 doesn't happen for the top of interlaced mbpairs. */
2233     int minpix_y = min_y*16 - 4 * !b_start;
2234     int maxpix_y = mb_y*16 - 4 * !b_end;
2235     b_deblock &= b_hpel || h->param.b_full_recon || h->param.psz_dump_yuv;
2236     if( h->param.b_sliced_threads )
2237     {
2238         switch( pass )
2239         {
2240             /* During encode: only do deblock if asked for */
2241             default:
2242             case 0:
2243                 b_deblock &= h->param.b_full_recon;
2244                 b_hpel = 0;
2245                 break;
2246             /* During post-encode pass: do deblock if not done yet, do hpel for all
2247              * rows except those between slices. */
2248             case 1:
2249                 b_deblock &= !h->param.b_full_recon;
2250                 b_hpel &= !(b_start && min_y > 0);
2251                 b_measure_quality = 0;
2252                 break;
2253             /* Final pass: do the rows between slices in sequence. */
2254             case 2:
2255                 b_deblock = 0;
2256                 b_measure_quality = 0;
2257                 break;
2258         }
2259     }
2260     if( mb_y & SLICE_MBAFF )
2261         return;
2262     if( min_y < h->i_threadslice_start )
2263         return;
2264
2265     if( b_deblock )
2266         for( int y = min_y; y < mb_y; y += (1 << SLICE_MBAFF) )
2267             x264_frame_deblock_row( h, y );
2268
2269     /* FIXME: Prediction requires different borders for interlaced/progressive mc,
2270      * but the actual image data is equivalent. For now, maintain this
2271      * consistency by copying deblocked pixels between planes. */
2272     if( PARAM_INTERLACED && (!h->param.b_sliced_threads || pass == 1) )
2273         for( int p = 0; p < h->fdec->i_plane; p++ )
2274             for( int i = minpix_y>>(CHROMA_V_SHIFT && p); i < maxpix_y>>(CHROMA_V_SHIFT && p); i++ )
2275                 memcpy( h->fdec->plane_fld[p] + i*h->fdec->i_stride[p],
2276                         h->fdec->plane[p]     + i*h->fdec->i_stride[p],
2277                         h->mb.i_mb_width*16*sizeof(pixel) );
2278
2279     if( h->fdec->b_kept_as_ref && (!h->param.b_sliced_threads || pass == 1) )
2280         x264_frame_expand_border( h, h->fdec, min_y );
2281     if( b_hpel )
2282     {
2283         int end = mb_y == h->mb.i_mb_height;
2284         /* Can't do hpel until the previous slice is done encoding. */
2285         if( h->param.analyse.i_subpel_refine )
2286         {
2287             x264_frame_filter( h, h->fdec, min_y, end );
2288             x264_frame_expand_border_filtered( h, h->fdec, min_y, end );
2289         }
2290     }
2291
2292     if( SLICE_MBAFF && pass == 0 )
2293         for( int i = 0; i < 3; i++ )
2294         {
2295             XCHG( pixel *, h->intra_border_backup[0][i], h->intra_border_backup[3][i] );
2296             XCHG( pixel *, h->intra_border_backup[1][i], h->intra_border_backup[4][i] );
2297         }
2298
2299     if( h->i_thread_frames > 1 && h->fdec->b_kept_as_ref )
2300         x264_frame_cond_broadcast( h->fdec, mb_y*16 + (b_end ? 10000 : -(X264_THREAD_HEIGHT << SLICE_MBAFF)) );
2301
2302     if( b_measure_quality )
2303     {
2304         maxpix_y = X264_MIN( maxpix_y, h->param.i_height );
2305         if( h->param.analyse.b_psnr )
2306         {
2307             for( int p = 0; p < (CHROMA444 ? 3 : 1); p++ )
2308                 h->stat.frame.i_ssd[p] += x264_pixel_ssd_wxh( &h->pixf,
2309                     h->fdec->plane[p] + minpix_y * h->fdec->i_stride[p], h->fdec->i_stride[p],
2310                     h->fenc->plane[p] + minpix_y * h->fenc->i_stride[p], h->fenc->i_stride[p],
2311                     h->param.i_width, maxpix_y-minpix_y );
2312             if( !CHROMA444 )
2313             {
2314                 uint64_t ssd_u, ssd_v;
2315                 int v_shift = CHROMA_V_SHIFT;
2316                 x264_pixel_ssd_nv12( &h->pixf,
2317                     h->fdec->plane[1] + (minpix_y>>v_shift) * h->fdec->i_stride[1], h->fdec->i_stride[1],
2318                     h->fenc->plane[1] + (minpix_y>>v_shift) * h->fenc->i_stride[1], h->fenc->i_stride[1],
2319                     h->param.i_width>>1, (maxpix_y-minpix_y)>>v_shift, &ssd_u, &ssd_v );
2320                 h->stat.frame.i_ssd[1] += ssd_u;
2321                 h->stat.frame.i_ssd[2] += ssd_v;
2322             }
2323         }
2324
2325         if( h->param.analyse.b_ssim )
2326         {
2327             int ssim_cnt;
2328             x264_emms();
2329             /* offset by 2 pixels to avoid alignment of ssim blocks with dct blocks,
2330              * and overlap by 4 */
2331             minpix_y += b_start ? 2 : -6;
2332             h->stat.frame.f_ssim +=
2333                 x264_pixel_ssim_wxh( &h->pixf,
2334                     h->fdec->plane[0] + 2+minpix_y*h->fdec->i_stride[0], h->fdec->i_stride[0],
2335                     h->fenc->plane[0] + 2+minpix_y*h->fenc->i_stride[0], h->fenc->i_stride[0],
2336                     h->param.i_width-2, maxpix_y-minpix_y, h->scratch_buffer, &ssim_cnt );
2337             h->stat.frame.i_ssim_cnt += ssim_cnt;
2338         }
2339     }
2340 }
2341
2342 static inline int x264_reference_update( x264_t *h )
2343 {
2344     if( !h->fdec->b_kept_as_ref )
2345     {
2346         if( h->i_thread_frames > 1 )
2347         {
2348             x264_frame_push_unused( h, h->fdec );
2349             h->fdec = x264_frame_pop_unused( h, 1 );
2350             if( !h->fdec )
2351                 return -1;
2352         }
2353         return 0;
2354     }
2355
2356     /* apply mmco from previous frame. */
2357     for( int i = 0; i < h->sh.i_mmco_command_count; i++ )
2358         for( int j = 0; h->frames.reference[j]; j++ )
2359             if( h->frames.reference[j]->i_poc == h->sh.mmco[i].i_poc )
2360                 x264_frame_push_unused( h, x264_frame_shift( &h->frames.reference[j] ) );
2361
2362     /* move frame in the buffer */
2363     x264_frame_push( h->frames.reference, h->fdec );
2364     if( h->frames.reference[h->sps->i_num_ref_frames] )
2365         x264_frame_push_unused( h, x264_frame_shift( h->frames.reference ) );
2366     h->fdec = x264_frame_pop_unused( h, 1 );
2367     if( !h->fdec )
2368         return -1;
2369     return 0;
2370 }
2371
2372 static inline void x264_reference_reset( x264_t *h )
2373 {
2374     while( h->frames.reference[0] )
2375         x264_frame_push_unused( h, x264_frame_pop( h->frames.reference ) );
2376     h->fdec->i_poc =
2377     h->fenc->i_poc = 0;
2378 }
2379
2380 static inline void x264_reference_hierarchy_reset( x264_t *h )
2381 {
2382     int ref;
2383     int b_hasdelayframe = 0;
2384
2385     /* look for delay frames -- chain must only contain frames that are disposable */
2386     for( int i = 0; h->frames.current[i] && IS_DISPOSABLE( h->frames.current[i]->i_type ); i++ )
2387         b_hasdelayframe |= h->frames.current[i]->i_coded
2388                         != h->frames.current[i]->i_frame + h->sps->vui.i_num_reorder_frames;
2389
2390     /* This function must handle b-pyramid and clear frames for open-gop */
2391     if( h->param.i_bframe_pyramid != X264_B_PYRAMID_STRICT && !b_hasdelayframe && h->frames.i_poc_last_open_gop == -1 )
2392         return;
2393
2394     /* Remove last BREF. There will never be old BREFs in the
2395      * dpb during a BREF decode when pyramid == STRICT */
2396     for( ref = 0; h->frames.reference[ref]; ref++ )
2397     {
2398         if( ( h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT
2399             && h->frames.reference[ref]->i_type == X264_TYPE_BREF )
2400             || ( h->frames.reference[ref]->i_poc < h->frames.i_poc_last_open_gop
2401             && h->sh.i_type != SLICE_TYPE_B ) )
2402         {
2403             int diff = h->i_frame_num - h->frames.reference[ref]->i_frame_num;
2404             h->sh.mmco[h->sh.i_mmco_command_count].i_difference_of_pic_nums = diff;
2405             h->sh.mmco[h->sh.i_mmco_command_count++].i_poc = h->frames.reference[ref]->i_poc;
2406             x264_frame_push_unused( h, x264_frame_shift( &h->frames.reference[ref] ) );
2407             h->b_ref_reorder[0] = 1;
2408             ref--;
2409         }
2410     }
2411
2412     /* Prepare room in the dpb for the delayed display time of the later b-frame's */
2413     if( h->param.i_bframe_pyramid )
2414         h->sh.i_mmco_remove_from_end = X264_MAX( ref + 2 - h->frames.i_max_dpb, 0 );
2415 }
2416
2417 static inline void x264_slice_init( x264_t *h, int i_nal_type, int i_global_qp )
2418 {
2419     /* ------------------------ Create slice header  ----------------------- */
2420     if( i_nal_type == NAL_SLICE_IDR )
2421     {
2422         x264_slice_header_init( h, &h->sh, h->sps, h->pps, h->i_idr_pic_id, h->i_frame_num, i_global_qp );
2423
2424         /* alternate id */
2425         h->i_idr_pic_id ^= 1;
2426     }
2427     else
2428     {
2429         x264_slice_header_init( h, &h->sh, h->sps, h->pps, -1, h->i_frame_num, i_global_qp );
2430
2431         h->sh.i_num_ref_idx_l0_active = h->i_ref[0] <= 0 ? 1 : h->i_ref[0];
2432         h->sh.i_num_ref_idx_l1_active = h->i_ref[1] <= 0 ? 1 : h->i_ref[1];
2433         if( h->sh.i_num_ref_idx_l0_active != h->pps->i_num_ref_idx_l0_default_active ||
2434             (h->sh.i_type == SLICE_TYPE_B && h->sh.i_num_ref_idx_l1_active != h->pps->i_num_ref_idx_l1_default_active) )
2435         {
2436             h->sh.b_num_ref_idx_override = 1;
2437         }
2438     }
2439
2440     if( h->fenc->i_type == X264_TYPE_BREF && h->param.b_bluray_compat && h->sh.i_mmco_command_count )
2441     {
2442         h->b_sh_backup = 1;
2443         h->sh_backup = h->sh;
2444     }
2445
2446     h->fdec->i_frame_num = h->sh.i_frame_num;
2447
2448     if( h->sps->i_poc_type == 0 )
2449     {
2450         h->sh.i_poc = h->fdec->i_poc;
2451         if( PARAM_INTERLACED )
2452         {
2453             h->sh.i_delta_poc_bottom = h->param.b_tff ? 1 : -1;
2454             h->sh.i_poc += h->sh.i_delta_poc_bottom == -1;
2455         }
2456         else
2457             h->sh.i_delta_poc_bottom = 0;
2458         h->fdec->i_delta_poc[0] = h->sh.i_delta_poc_bottom == -1;
2459         h->fdec->i_delta_poc[1] = h->sh.i_delta_poc_bottom ==  1;
2460     }
2461     else
2462     {
2463         /* Nothing to do ? */
2464     }
2465
2466     x264_macroblock_slice_init( h );
2467 }
2468
2469 typedef struct
2470 {
2471     int skip;
2472     uint8_t cabac_prevbyte;
2473     bs_t bs;
2474     x264_cabac_t cabac;
2475     x264_frame_stat_t stat;
2476     int last_qp;
2477     int last_dqp;
2478     int field_decoding_flag;
2479 } x264_bs_bak_t;
2480
2481 static ALWAYS_INLINE void x264_bitstream_backup( x264_t *h, x264_bs_bak_t *bak, int i_skip, int full )
2482 {
2483     if( full )
2484     {
2485         bak->stat = h->stat.frame;
2486         bak->last_qp = h->mb.i_last_qp;
2487         bak->last_dqp = h->mb.i_last_dqp;
2488         bak->field_decoding_flag = h->mb.field_decoding_flag;
2489     }
2490     else
2491     {
2492         bak->stat.i_mv_bits = h->stat.frame.i_mv_bits;
2493         bak->stat.i_tex_bits = h->stat.frame.i_tex_bits;
2494     }
2495     /* In the per-MB backup, we don't need the contexts because flushing the CABAC
2496      * encoder has no context dependency and in this case, a slice is ended (and
2497      * thus the content of all contexts are thrown away). */
2498     if( h->param.b_cabac )
2499     {
2500         if( full )
2501             memcpy( &bak->cabac, &h->cabac, sizeof(x264_cabac_t) );
2502         else
2503             memcpy( &bak->cabac, &h->cabac, offsetof(x264_cabac_t, f8_bits_encoded) );
2504         /* x264's CABAC writer modifies the previous byte during carry, so it has to be
2505          * backed up. */
2506         bak->cabac_prevbyte = h->cabac.p[-1];
2507     }
2508     else
2509     {
2510         bak->bs = h->out.bs;
2511         bak->skip = i_skip;
2512     }
2513 }
2514
2515 static ALWAYS_INLINE void x264_bitstream_restore( x264_t *h, x264_bs_bak_t *bak, int *skip, int full )
2516 {
2517     if( full )
2518     {
2519         h->stat.frame = bak->stat;
2520         h->mb.i_last_qp = bak->last_qp;
2521         h->mb.i_last_dqp = bak->last_dqp;
2522         h->mb.field_decoding_flag = bak->field_decoding_flag;
2523     }
2524     else
2525     {
2526         h->stat.frame.i_mv_bits = bak->stat.i_mv_bits;
2527         h->stat.frame.i_tex_bits = bak->stat.i_tex_bits;
2528     }
2529     if( h->param.b_cabac )
2530     {
2531         if( full )
2532             memcpy( &h->cabac, &bak->cabac, sizeof(x264_cabac_t) );
2533         else
2534             memcpy( &h->cabac, &bak->cabac, offsetof(x264_cabac_t, f8_bits_encoded) );
2535         h->cabac.p[-1] = bak->cabac_prevbyte;
2536     }
2537     else
2538     {
2539         h->out.bs = bak->bs;
2540         *skip = bak->skip;
2541     }
2542 }
2543
2544 static int x264_slice_write( x264_t *h )
2545 {
2546     int i_skip;
2547     int mb_xy, i_mb_x, i_mb_y;
2548     /* NALUs other than the first use a 3-byte startcode.
2549      * Add one extra byte for the rbsp, and one more for the final CABAC putbyte.
2550      * Then add an extra 5 bytes just in case, to account for random NAL escapes and
2551      * other inaccuracies. */
2552     int overhead_guess = (NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal)) + 1 + h->param.b_cabac + 5;
2553     int slice_max_size = h->param.i_slice_max_size > 0 ? (h->param.i_slice_max_size-overhead_guess)*8 : 0;
2554     int back_up_bitstream = slice_max_size || (!h->param.b_cabac && h->sps->i_profile_idc < PROFILE_HIGH);
2555     int starting_bits = bs_pos(&h->out.bs);
2556     int b_deblock = h->sh.i_disable_deblocking_filter_idc != 1;
2557     int b_hpel = h->fdec->b_kept_as_ref;
2558     int orig_last_mb = h->sh.i_last_mb;
2559     int thread_last_mb = h->i_threadslice_end * h->mb.i_mb_width - 1;
2560     uint8_t *last_emu_check;
2561 #define BS_BAK_SLICE_MAX_SIZE 0
2562 #define BS_BAK_SLICE_MIN_MBS  1
2563 #define BS_BAK_ROW_VBV        2
2564     x264_bs_bak_t bs_bak[3];
2565     b_deblock &= b_hpel || h->param.b_full_recon || h->param.psz_dump_yuv;
2566     bs_realign( &h->out.bs );
2567
2568     /* Slice */
2569     x264_nal_start( h, h->i_nal_type, h->i_nal_ref_idc );
2570     h->out.nal[h->out.i_nal].i_first_mb = h->sh.i_first_mb;
2571
2572     /* Slice header */
2573     x264_macroblock_thread_init( h );
2574
2575     /* Set the QP equal to the first QP in the slice for more accurate CABAC initialization. */
2576     h->mb.i_mb_xy = h->sh.i_first_mb;
2577     h->sh.i_qp = x264_ratecontrol_mb_qp( h );
2578     h->sh.i_qp = SPEC_QP( h->sh.i_qp );
2579     h->sh.i_qp_delta = h->sh.i_qp - h->pps->i_pic_init_qp;
2580
2581     x264_slice_header_write( &h->out.bs, &h->sh, h->i_nal_ref_idc );
2582     if( h->param.b_cabac )
2583     {
2584         /* alignment needed */
2585         bs_align_1( &h->out.bs );
2586
2587         /* init cabac */
2588         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 );
2589         x264_cabac_encode_init ( &h->cabac, h->out.bs.p, h->out.bs.p_end );
2590         last_emu_check = h->cabac.p;
2591     }
2592     else
2593         last_emu_check = h->out.bs.p;
2594     h->mb.i_last_qp = h->sh.i_qp;
2595     h->mb.i_last_dqp = 0;
2596     h->mb.field_decoding_flag = 0;
2597
2598     i_mb_y = h->sh.i_first_mb / h->mb.i_mb_width;
2599     i_mb_x = h->sh.i_first_mb % h->mb.i_mb_width;
2600     i_skip = 0;
2601
2602     while( 1 )
2603     {
2604         mb_xy = i_mb_x + i_mb_y * h->mb.i_mb_width;
2605         int mb_spos = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
2606
2607         if( i_mb_x == 0 )
2608         {
2609             if( x264_bitstream_check_buffer( h ) )
2610                 return -1;
2611             if( !(i_mb_y & SLICE_MBAFF) && h->param.rc.i_vbv_buffer_size )
2612                 x264_bitstream_backup( h, &bs_bak[BS_BAK_ROW_VBV], i_skip, 1 );
2613             if( !h->mb.b_reencode_mb )
2614                 x264_fdec_filter_row( h, i_mb_y, 0 );
2615         }
2616
2617         if( !(i_mb_y & SLICE_MBAFF) && back_up_bitstream )
2618         {
2619             x264_bitstream_backup( h, &bs_bak[BS_BAK_SLICE_MAX_SIZE], i_skip, 0 );
2620             if( slice_max_size && (thread_last_mb+1-mb_xy) == h->param.i_slice_min_mbs )
2621                 x264_bitstream_backup( h, &bs_bak[BS_BAK_SLICE_MIN_MBS], i_skip, 0 );
2622         }
2623
2624         if( PARAM_INTERLACED )
2625         {
2626             if( h->mb.b_adaptive_mbaff )
2627             {
2628                 if( !(i_mb_y&1) )
2629                 {
2630                     /* FIXME: VSAD is fast but fairly poor at choosing the best interlace type. */
2631                     h->mb.b_interlaced = x264_field_vsad( h, i_mb_x, i_mb_y );
2632                     memcpy( &h->zigzagf, MB_INTERLACED ? &h->zigzagf_interlaced : &h->zigzagf_progressive, sizeof(h->zigzagf) );
2633                     if( !MB_INTERLACED && (i_mb_y+2) == h->mb.i_mb_height )
2634                         x264_expand_border_mbpair( h, i_mb_x, i_mb_y );
2635                 }
2636             }
2637             h->mb.field[mb_xy] = MB_INTERLACED;
2638         }
2639
2640         /* load cache */
2641         if( SLICE_MBAFF )
2642             x264_macroblock_cache_load_interlaced( h, i_mb_x, i_mb_y );
2643         else
2644             x264_macroblock_cache_load_progressive( h, i_mb_x, i_mb_y );
2645
2646         x264_macroblock_analyse( h );
2647
2648         /* encode this macroblock -> be careful it can change the mb type to P_SKIP if needed */
2649 reencode:
2650         x264_macroblock_encode( h );
2651
2652         if( h->param.b_cabac )
2653         {
2654             if( mb_xy > h->sh.i_first_mb && !(SLICE_MBAFF && (i_mb_y&1)) )
2655                 x264_cabac_encode_terminal( &h->cabac );
2656
2657             if( IS_SKIP( h->mb.i_type ) )
2658                 x264_cabac_mb_skip( h, 1 );
2659             else
2660             {
2661                 if( h->sh.i_type != SLICE_TYPE_I )
2662                     x264_cabac_mb_skip( h, 0 );
2663                 x264_macroblock_write_cabac( h, &h->cabac );
2664             }
2665         }
2666         else
2667         {
2668             if( IS_SKIP( h->mb.i_type ) )
2669                 i_skip++;
2670             else
2671             {
2672                 if( h->sh.i_type != SLICE_TYPE_I )
2673                 {
2674                     bs_write_ue( &h->out.bs, i_skip );  /* skip run */
2675                     i_skip = 0;
2676                 }
2677                 x264_macroblock_write_cavlc( h );
2678                 /* If there was a CAVLC level code overflow, try again at a higher QP. */
2679                 if( h->mb.b_overflow )
2680                 {
2681                     h->mb.i_chroma_qp = h->chroma_qp_table[++h->mb.i_qp];
2682                     h->mb.i_skip_intra = 0;
2683                     h->mb.b_skip_mc = 0;
2684                     h->mb.b_overflow = 0;
2685                     x264_bitstream_restore( h, &bs_bak[BS_BAK_SLICE_MAX_SIZE], &i_skip, 0 );
2686                     goto reencode;
2687                 }
2688             }
2689         }
2690
2691         int total_bits = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
2692         int mb_size = total_bits - mb_spos;
2693
2694         if( slice_max_size && (!SLICE_MBAFF || (i_mb_y&1)) )
2695         {
2696             /* Count the skip run, just in case. */
2697             if( !h->param.b_cabac )
2698                 total_bits += bs_size_ue_big( i_skip );
2699             /* Check for escape bytes. */
2700             uint8_t *end = h->param.b_cabac ? h->cabac.p : h->out.bs.p;
2701             for( ; last_emu_check < end - 2; last_emu_check++ )
2702                 if( last_emu_check[0] == 0 && last_emu_check[1] == 0 && last_emu_check[2] <= 3 )
2703                 {
2704                     slice_max_size -= 8;
2705                     last_emu_check++;
2706                 }
2707             /* We'll just re-encode this last macroblock if we go over the max slice size. */
2708             if( total_bits - starting_bits > slice_max_size && !h->mb.b_reencode_mb )
2709             {
2710                 if( !x264_frame_new_slice( h, h->fdec ) )
2711                 {
2712                     /* Handle the most obnoxious slice-min-mbs edge case: we need to end the slice
2713                      * because it's gone over the maximum size, but doing so would violate slice-min-mbs.
2714                      * If possible, roll back to the last checkpoint and try again.
2715                      * We could try raising QP, but that would break in the case where a slice spans multiple
2716                      * rows, which the re-encoding infrastructure can't currently handle. */
2717                     if( mb_xy <= thread_last_mb && (thread_last_mb+1-mb_xy) < h->param.i_slice_min_mbs )
2718                     {
2719                         if( thread_last_mb-h->param.i_slice_min_mbs < h->sh.i_first_mb+h->param.i_slice_min_mbs )
2720                         {
2721                             x264_log( h, X264_LOG_WARNING, "slice-max-size violated (frame %d, cause: slice-min-mbs)\n", h->i_frame );
2722                             slice_max_size = 0;
2723                             goto cont;
2724                         }
2725                         x264_bitstream_restore( h, &bs_bak[BS_BAK_SLICE_MIN_MBS], &i_skip, 0 );
2726                         h->mb.b_reencode_mb = 1;
2727                         h->sh.i_last_mb = thread_last_mb-h->param.i_slice_min_mbs;
2728                         break;
2729                     }
2730                     if( mb_xy-SLICE_MBAFF*h->mb.i_mb_stride != h->sh.i_first_mb )
2731                     {
2732                         x264_bitstream_restore( h, &bs_bak[BS_BAK_SLICE_MAX_SIZE], &i_skip, 0 );
2733                         h->mb.b_reencode_mb = 1;
2734                         if( SLICE_MBAFF )
2735                         {
2736                             // set to bottom of previous mbpair
2737                             if( i_mb_x )
2738                                 h->sh.i_last_mb = mb_xy-1+h->mb.i_mb_stride*(!(i_mb_y&1));
2739                             else
2740                                 h->sh.i_last_mb = (i_mb_y-2+!(i_mb_y&1))*h->mb.i_mb_stride + h->mb.i_mb_width - 1;
2741                         }
2742                         else
2743                             h->sh.i_last_mb = mb_xy-1;
2744                         break;
2745                     }
2746                     else
2747                         h->sh.i_last_mb = mb_xy;
2748                 }
2749                 else
2750                     slice_max_size = 0;
2751             }
2752         }
2753 cont:
2754         h->mb.b_reencode_mb = 0;
2755
2756 #if HAVE_VISUALIZE
2757         if( h->param.b_visualize )
2758             x264_visualize_mb( h );
2759 #endif
2760
2761         /* save cache */
2762         x264_macroblock_cache_save( h );
2763
2764         if( x264_ratecontrol_mb( h, mb_size ) < 0 )
2765         {
2766             x264_bitstream_restore( h, &bs_bak[BS_BAK_ROW_VBV], &i_skip, 1 );
2767             h->mb.b_reencode_mb = 1;
2768             i_mb_x = 0;
2769             i_mb_y = i_mb_y - SLICE_MBAFF;
2770             h->mb.i_mb_prev_xy = i_mb_y * h->mb.i_mb_stride - 1;
2771             h->sh.i_last_mb = orig_last_mb;
2772             continue;
2773         }
2774
2775         /* accumulate mb stats */
2776         h->stat.frame.i_mb_count[h->mb.i_type]++;
2777
2778         int b_intra = IS_INTRA( h->mb.i_type );
2779         int b_skip = IS_SKIP( h->mb.i_type );
2780         if( h->param.i_log_level >= X264_LOG_INFO || h->param.rc.b_stat_write )
2781         {
2782             if( !b_intra && !b_skip && !IS_DIRECT( h->mb.i_type ) )
2783             {
2784                 if( h->mb.i_partition != D_8x8 )
2785                         h->stat.frame.i_mb_partition[h->mb.i_partition] += 4;
2786                     else
2787                         for( int i = 0; i < 4; i++ )
2788                             h->stat.frame.i_mb_partition[h->mb.i_sub_partition[i]] ++;
2789                 if( h->param.i_frame_reference > 1 )
2790                     for( int i_list = 0; i_list <= (h->sh.i_type == SLICE_TYPE_B); i_list++ )
2791                         for( int i = 0; i < 4; i++ )
2792                         {
2793                             int i_ref = h->mb.cache.ref[i_list][ x264_scan8[4*i] ];
2794                             if( i_ref >= 0 )
2795                                 h->stat.frame.i_mb_count_ref[i_list][i_ref] ++;
2796                         }
2797             }
2798         }
2799
2800         if( h->param.i_log_level >= X264_LOG_INFO )
2801         {
2802             if( h->mb.i_cbp_luma | h->mb.i_cbp_chroma )
2803             {
2804                 if( CHROMA444 )
2805                 {
2806                     for( int i = 0; i < 4; i++ )
2807                         if( h->mb.i_cbp_luma & (1 << i) )
2808                             for( int p = 0; p < 3; p++ )
2809                             {
2810                                 int s8 = i*4+p*16;
2811                                 int nnz8x8 = M16( &h->mb.cache.non_zero_count[x264_scan8[s8]+0] )
2812                                            | M16( &h->mb.cache.non_zero_count[x264_scan8[s8]+8] );
2813                                 h->stat.frame.i_mb_cbp[!b_intra + p*2] += !!nnz8x8;
2814                             }
2815                 }
2816                 else
2817                 {
2818                     int cbpsum = (h->mb.i_cbp_luma&1) + ((h->mb.i_cbp_luma>>1)&1)
2819                                + ((h->mb.i_cbp_luma>>2)&1) + (h->mb.i_cbp_luma>>3);
2820                     h->stat.frame.i_mb_cbp[!b_intra + 0] += cbpsum;
2821                     h->stat.frame.i_mb_cbp[!b_intra + 2] += !!h->mb.i_cbp_chroma;
2822                     h->stat.frame.i_mb_cbp[!b_intra + 4] += h->mb.i_cbp_chroma >> 1;
2823                 }
2824             }
2825             if( h->mb.i_cbp_luma && !b_intra )
2826             {
2827                 h->stat.frame.i_mb_count_8x8dct[0] ++;
2828                 h->stat.frame.i_mb_count_8x8dct[1] += h->mb.b_transform_8x8;
2829             }
2830             if( b_intra && h->mb.i_type != I_PCM )
2831             {
2832                 if( h->mb.i_type == I_16x16 )
2833                     h->stat.frame.i_mb_pred_mode[0][h->mb.i_intra16x16_pred_mode]++;
2834                 else if( h->mb.i_type == I_8x8 )
2835                     for( int i = 0; i < 16; i += 4 )
2836                         h->stat.frame.i_mb_pred_mode[1][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
2837                 else //if( h->mb.i_type == I_4x4 )
2838                     for( int i = 0; i < 16; i++ )
2839                         h->stat.frame.i_mb_pred_mode[2][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
2840                 h->stat.frame.i_mb_pred_mode[3][x264_mb_chroma_pred_mode_fix[h->mb.i_chroma_pred_mode]]++;
2841             }
2842             h->stat.frame.i_mb_field[b_intra?0:b_skip?2:1] += MB_INTERLACED;
2843         }
2844
2845         /* calculate deblock strength values (actual deblocking is done per-row along with hpel) */
2846         if( b_deblock )
2847             x264_macroblock_deblock_strength( h );
2848
2849         if( mb_xy == h->sh.i_last_mb )
2850             break;
2851
2852         if( SLICE_MBAFF )
2853         {
2854             i_mb_x += i_mb_y & 1;
2855             i_mb_y ^= i_mb_x < h->mb.i_mb_width;
2856         }
2857         else
2858             i_mb_x++;
2859         if( i_mb_x == h->mb.i_mb_width )
2860         {
2861             i_mb_y++;
2862             i_mb_x = 0;
2863         }
2864     }
2865     if( h->sh.i_last_mb < h->sh.i_first_mb )
2866         return 0;
2867
2868     h->out.nal[h->out.i_nal].i_last_mb = h->sh.i_last_mb;
2869
2870     if( h->param.b_cabac )
2871     {
2872         x264_cabac_encode_flush( h, &h->cabac );
2873         h->out.bs.p = h->cabac.p;
2874     }
2875     else
2876     {
2877         if( i_skip > 0 )
2878             bs_write_ue( &h->out.bs, i_skip );  /* last skip run */
2879         /* rbsp_slice_trailing_bits */
2880         bs_rbsp_trailing( &h->out.bs );
2881         bs_flush( &h->out.bs );
2882     }
2883     if( x264_nal_end( h ) )
2884         return -1;
2885
2886     if( h->sh.i_last_mb == (h->i_threadslice_end * h->mb.i_mb_width - 1) )
2887     {
2888         h->stat.frame.i_misc_bits = bs_pos( &h->out.bs )
2889                                   + (h->out.i_nal*NALU_OVERHEAD * 8)
2890                                   - h->stat.frame.i_tex_bits
2891                                   - h->stat.frame.i_mv_bits;
2892         x264_fdec_filter_row( h, h->i_threadslice_end, 0 );
2893
2894         if( h->param.b_sliced_threads )
2895         {
2896             /* Tell the main thread we're done. */
2897             x264_threadslice_cond_broadcast( h, 1 );
2898             /* Do hpel now */
2899             for( int mb_y = h->i_threadslice_start; mb_y <= h->i_threadslice_end; mb_y++ )
2900                 x264_fdec_filter_row( h, mb_y, 1 );
2901             x264_threadslice_cond_broadcast( h, 2 );
2902             /* Do the first row of hpel, now that the previous slice is done */
2903             if( h->i_thread_idx > 0 )
2904             {
2905                 x264_threadslice_cond_wait( h->thread[h->i_thread_idx-1], 2 );
2906                 x264_fdec_filter_row( h, h->i_threadslice_start + (1 << SLICE_MBAFF), 2 );
2907             }
2908         }
2909
2910         /* Free mb info after the last thread's done using it */
2911         if( h->fdec->mb_info_free && (!h->param.b_sliced_threads || h->i_thread_idx == (h->param.i_threads-1)) )
2912         {
2913             h->fdec->mb_info_free( h->fdec->mb_info );
2914             h->fdec->mb_info = NULL;
2915             h->fdec->mb_info_free = NULL;
2916         }
2917     }
2918
2919     return 0;
2920 }
2921
2922 static void x264_thread_sync_context( x264_t *dst, x264_t *src )
2923 {
2924     if( dst == src )
2925         return;
2926
2927     // reference counting
2928     for( x264_frame_t **f = src->frames.reference; *f; f++ )
2929         (*f)->i_reference_count++;
2930     for( x264_frame_t **f = dst->frames.reference; *f; f++ )
2931         x264_frame_push_unused( src, *f );
2932     src->fdec->i_reference_count++;
2933     x264_frame_push_unused( src, dst->fdec );
2934
2935     // copy everything except the per-thread pointers and the constants.
2936     memcpy( &dst->i_frame, &src->i_frame, offsetof(x264_t, mb.base) - offsetof(x264_t, i_frame) );
2937     dst->param = src->param;
2938     dst->stat = src->stat;
2939     dst->pixf = src->pixf;
2940     dst->reconfig = src->reconfig;
2941 }
2942
2943 static void x264_thread_sync_stat( x264_t *dst, x264_t *src )
2944 {
2945     if( dst == src )
2946         return;
2947     memcpy( &dst->stat.i_frame_count, &src->stat.i_frame_count, sizeof(dst->stat) - sizeof(dst->stat.frame) );
2948 }
2949
2950 static void *x264_slices_write( x264_t *h )
2951 {
2952     int i_slice_num = 0;
2953     int last_thread_mb = h->sh.i_last_mb;
2954
2955 #if HAVE_VISUALIZE
2956     if( h->param.b_visualize )
2957         if( x264_visualize_init( h ) )
2958             goto fail;
2959 #endif
2960
2961     /* init stats */
2962     memset( &h->stat.frame, 0, sizeof(h->stat.frame) );
2963     h->mb.b_reencode_mb = 0;
2964     while( h->sh.i_first_mb + SLICE_MBAFF*h->mb.i_mb_stride <= last_thread_mb )
2965     {
2966         h->sh.i_last_mb = last_thread_mb;
2967         if( !i_slice_num || !x264_frame_new_slice( h, h->fdec ) )
2968         {
2969             if( h->param.i_slice_max_mbs )
2970             {
2971                 if( SLICE_MBAFF )
2972                 {
2973                     // convert first to mbaff form, add slice-max-mbs, then convert back to normal form
2974                     int last_mbaff = 2*(h->sh.i_first_mb % h->mb.i_mb_width)
2975                         + h->mb.i_mb_width*(h->sh.i_first_mb / h->mb.i_mb_width)
2976                         + h->param.i_slice_max_mbs - 1;
2977                     int last_x = (last_mbaff % (2*h->mb.i_mb_width))/2;
2978                     int last_y = (last_mbaff / (2*h->mb.i_mb_width))*2 + 1;
2979                     h->sh.i_last_mb = last_x + h->mb.i_mb_stride*last_y;
2980                 }
2981                 else
2982                 {
2983                     h->sh.i_last_mb = h->sh.i_first_mb + h->param.i_slice_max_mbs - 1;
2984                     if( h->sh.i_last_mb < last_thread_mb && last_thread_mb - h->sh.i_last_mb < h->param.i_slice_min_mbs )
2985                         h->sh.i_last_mb = last_thread_mb - h->param.i_slice_min_mbs;
2986                 }
2987                 i_slice_num++;
2988             }
2989             else if( h->param.i_slice_count && !h->param.b_sliced_threads )
2990             {
2991                 int height = h->mb.i_mb_height >> PARAM_INTERLACED;
2992                 int width = h->mb.i_mb_width << PARAM_INTERLACED;
2993                 i_slice_num++;
2994                 h->sh.i_last_mb = (height * i_slice_num + h->param.i_slice_count/2) / h->param.i_slice_count * width - 1;
2995             }
2996         }
2997         h->sh.i_last_mb = X264_MIN( h->sh.i_last_mb, last_thread_mb );
2998         if( x264_stack_align( x264_slice_write, h ) )
2999             goto fail;
3000         h->sh.i_first_mb = h->sh.i_last_mb + 1;
3001         // if i_first_mb is not the last mb in a row then go to the next mb in MBAFF order
3002         if( SLICE_MBAFF && h->sh.i_first_mb % h->mb.i_mb_width )
3003             h->sh.i_first_mb -= h->mb.i_mb_stride;
3004     }
3005
3006 #if HAVE_VISUALIZE
3007     if( h->param.b_visualize )
3008     {
3009         x264_visualize_show( h );
3010         x264_visualize_close( h );
3011     }
3012 #endif
3013
3014     return (void *)0;
3015
3016 fail:
3017     /* Tell other threads we're done, so they wouldn't wait for it */
3018     if( h->param.b_sliced_threads )
3019         x264_threadslice_cond_broadcast( h, 2 );
3020     return (void *)-1;
3021 }
3022
3023 static int x264_threaded_slices_write( x264_t *h )
3024 {
3025     /* set first/last mb and sync contexts */
3026     for( int i = 0; i < h->param.i_threads; i++ )
3027     {
3028         x264_t *t = h->thread[i];
3029         if( i )
3030         {
3031             t->param = h->param;
3032             memcpy( &t->i_frame, &h->i_frame, offsetof(x264_t, rc) - offsetof(x264_t, i_frame) );
3033         }
3034         int height = h->mb.i_mb_height >> PARAM_INTERLACED;
3035         t->i_threadslice_start = ((height *  i    + h->param.i_slice_count/2) / h->param.i_threads) << PARAM_INTERLACED;
3036         t->i_threadslice_end   = ((height * (i+1) + h->param.i_slice_count/2) / h->param.i_threads) << PARAM_INTERLACED;
3037         t->sh.i_first_mb = t->i_threadslice_start * h->mb.i_mb_width;
3038         t->sh.i_last_mb  =   t->i_threadslice_end * h->mb.i_mb_width - 1;
3039     }
3040
3041     x264_stack_align( x264_analyse_weight_frame, h, h->mb.i_mb_height*16 + 16 );
3042
3043     x264_threads_distribute_ratecontrol( h );
3044
3045     /* setup */
3046     for( int i = 0; i < h->param.i_threads; i++ )
3047     {
3048         h->thread[i]->i_thread_idx = i;
3049         h->thread[i]->b_thread_active = 1;
3050         x264_threadslice_cond_broadcast( h->thread[i], 0 );
3051     }
3052     /* dispatch */
3053     for( int i = 0; i < h->param.i_threads; i++ )
3054         x264_threadpool_run( h->threadpool, (void*)x264_slices_write, h->thread[i] );
3055     /* wait */
3056     for( int i = 0; i < h->param.i_threads; i++ )
3057         x264_threadslice_cond_wait( h->thread[i], 1 );
3058
3059     x264_threads_merge_ratecontrol( h );
3060
3061     for( int i = 1; i < h->param.i_threads; i++ )
3062     {
3063         x264_t *t = h->thread[i];
3064         for( int j = 0; j < t->out.i_nal; j++ )
3065         {
3066             h->out.nal[h->out.i_nal] = t->out.nal[j];
3067             h->out.i_nal++;
3068             x264_nal_check_buffer( h );
3069         }
3070         /* All entries in stat.frame are ints except for ssd/ssim. */
3071         for( int j = 0; j < (offsetof(x264_t,stat.frame.i_ssd) - offsetof(x264_t,stat.frame.i_mv_bits)) / sizeof(int); j++ )
3072             ((int*)&h->stat.frame)[j] += ((int*)&t->stat.frame)[j];
3073         for( int j = 0; j < 3; j++ )
3074             h->stat.frame.i_ssd[j] += t->stat.frame.i_ssd[j];
3075         h->stat.frame.f_ssim += t->stat.frame.f_ssim;
3076         h->stat.frame.i_ssim_cnt += t->stat.frame.i_ssim_cnt;
3077     }
3078
3079     return 0;
3080 }
3081
3082 void x264_encoder_intra_refresh( x264_t *h )
3083 {
3084     h = h->thread[h->i_thread_phase];
3085     h->b_queued_intra_refresh = 1;
3086 }
3087
3088 int x264_encoder_invalidate_reference( x264_t *h, int64_t pts )
3089 {
3090     if( h->param.i_bframe )
3091     {
3092         x264_log( h, X264_LOG_ERROR, "x264_encoder_invalidate_reference is not supported with B-frames enabled\n" );
3093         return -1;
3094     }
3095     if( h->param.b_intra_refresh )
3096     {
3097         x264_log( h, X264_LOG_ERROR, "x264_encoder_invalidate_reference is not supported with intra refresh enabled\n" );
3098         return -1;
3099     }
3100     h = h->thread[h->i_thread_phase];
3101     if( pts >= h->i_last_idr_pts )
3102     {
3103         for( int i = 0; h->frames.reference[i]; i++ )
3104             if( pts <= h->frames.reference[i]->i_pts )
3105                 h->frames.reference[i]->b_corrupt = 1;
3106         if( pts <= h->fdec->i_pts )
3107             h->fdec->b_corrupt = 1;
3108     }
3109     return 0;
3110 }
3111
3112 /****************************************************************************
3113  * x264_encoder_encode:
3114  *  XXX: i_poc   : is the poc of the current given picture
3115  *       i_frame : is the number of the frame being coded
3116  *  ex:  type frame poc
3117  *       I      0   2*0
3118  *       P      1   2*3
3119  *       B      2   2*1
3120  *       B      3   2*2
3121  *       P      4   2*6
3122  *       B      5   2*4
3123  *       B      6   2*5
3124  ****************************************************************************/
3125 int     x264_encoder_encode( x264_t *h,
3126                              x264_nal_t **pp_nal, int *pi_nal,
3127                              x264_picture_t *pic_in,
3128                              x264_picture_t *pic_out )
3129 {
3130     x264_t *thread_current, *thread_prev, *thread_oldest;
3131     int i_nal_type, i_nal_ref_idc, i_global_qp;
3132     int overhead = NALU_OVERHEAD;
3133
3134 #if HAVE_OPENCL
3135     if( h->opencl.b_fatal_error )
3136         return -1;
3137 #endif
3138
3139     if( h->i_thread_frames > 1 )
3140     {
3141         thread_prev    = h->thread[ h->i_thread_phase ];
3142         h->i_thread_phase = (h->i_thread_phase + 1) % h->i_thread_frames;
3143         thread_current = h->thread[ h->i_thread_phase ];
3144         thread_oldest  = h->thread[ (h->i_thread_phase + 1) % h->i_thread_frames ];
3145         x264_thread_sync_context( thread_current, thread_prev );
3146         x264_thread_sync_ratecontrol( thread_current, thread_prev, thread_oldest );
3147         h = thread_current;
3148     }
3149     else
3150     {
3151         thread_current =
3152         thread_oldest  = h;
3153     }
3154     h->i_cpb_delay_pir_offset = h->i_cpb_delay_pir_offset_next;
3155
3156     /* no data out */
3157     *pi_nal = 0;
3158     *pp_nal = NULL;
3159
3160     /* ------------------- Setup new frame from picture -------------------- */
3161     if( pic_in != NULL )
3162     {
3163         /* 1: Copy the picture to a frame and move it to a buffer */
3164         x264_frame_t *fenc = x264_frame_pop_unused( h, 0 );
3165         if( !fenc )
3166             return -1;
3167
3168         if( x264_frame_copy_picture( h, fenc, pic_in ) < 0 )
3169             return -1;
3170
3171         if( h->param.i_width != 16 * h->mb.i_mb_width ||
3172             h->param.i_height != 16 * h->mb.i_mb_height )
3173             x264_frame_expand_border_mod16( h, fenc );
3174
3175         fenc->i_frame = h->frames.i_input++;
3176
3177         if( fenc->i_frame == 0 )
3178             h->frames.i_first_pts = fenc->i_pts;
3179         if( h->frames.i_bframe_delay && fenc->i_frame == h->frames.i_bframe_delay )
3180             h->frames.i_bframe_delay_time = fenc->i_pts - h->frames.i_first_pts;
3181
3182         if( h->param.b_vfr_input && fenc->i_pts <= h->frames.i_largest_pts )
3183             x264_log( h, X264_LOG_WARNING, "non-strictly-monotonic PTS\n" );
3184
3185         h->frames.i_second_largest_pts = h->frames.i_largest_pts;
3186         h->frames.i_largest_pts = fenc->i_pts;
3187
3188         if( (fenc->i_pic_struct < PIC_STRUCT_AUTO) || (fenc->i_pic_struct > PIC_STRUCT_TRIPLE) )
3189             fenc->i_pic_struct = PIC_STRUCT_AUTO;
3190
3191         if( fenc->i_pic_struct == PIC_STRUCT_AUTO )
3192         {
3193 #if HAVE_INTERLACED
3194             int b_interlaced = fenc->param ? fenc->param->b_interlaced : h->param.b_interlaced;
3195 #else
3196             int b_interlaced = 0;
3197 #endif
3198             if( b_interlaced )
3199             {
3200                 int b_tff = fenc->param ? fenc->param->b_tff : h->param.b_tff;
3201                 fenc->i_pic_struct = b_tff ? PIC_STRUCT_TOP_BOTTOM : PIC_STRUCT_BOTTOM_TOP;
3202             }
3203             else
3204                 fenc->i_pic_struct = PIC_STRUCT_PROGRESSIVE;
3205         }
3206
3207         if( h->param.rc.b_mb_tree && h->param.rc.b_stat_read )
3208         {
3209             if( x264_macroblock_tree_read( h, fenc, pic_in->prop.quant_offsets ) )
3210                 return -1;
3211         }
3212         else
3213             x264_stack_align( x264_adaptive_quant_frame, h, fenc, pic_in->prop.quant_offsets );
3214
3215         if( pic_in->prop.quant_offsets_free )
3216             pic_in->prop.quant_offsets_free( pic_in->prop.quant_offsets );
3217
3218         if( h->frames.b_have_lowres )
3219             x264_frame_init_lowres( h, fenc );
3220
3221         /* 2: Place the frame into the queue for its slice type decision */
3222         x264_lookahead_put_frame( h, fenc );
3223
3224         if( h->frames.i_input <= h->frames.i_delay + 1 - h->i_thread_frames )
3225         {
3226             /* Nothing yet to encode, waiting for filling of buffers */
3227             pic_out->i_type = X264_TYPE_AUTO;
3228             return 0;
3229         }
3230     }
3231     else
3232     {
3233         /* signal kills for lookahead thread */
3234         x264_pthread_mutex_lock( &h->lookahead->ifbuf.mutex );
3235         h->lookahead->b_exit_thread = 1;
3236         x264_pthread_cond_broadcast( &h->lookahead->ifbuf.cv_fill );
3237         x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
3238     }
3239
3240     h->i_frame++;
3241     /* 3: The picture is analyzed in the lookahead */
3242     if( !h->frames.current[0] )
3243         x264_lookahead_get_frames( h );
3244
3245     if( !h->frames.current[0] && x264_lookahead_is_empty( h ) )
3246         return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
3247
3248     /* ------------------- Get frame to be encoded ------------------------- */
3249     /* 4: get picture to encode */
3250     h->fenc = x264_frame_shift( h->frames.current );
3251
3252     /* If applicable, wait for previous frame reconstruction to finish */
3253     if( h->param.b_sliced_threads )
3254         if( x264_threadpool_wait_all( h ) < 0 )
3255             return -1;
3256
3257     if( h->i_frame == h->i_thread_frames - 1 )
3258         h->i_reordered_pts_delay = h->fenc->i_reordered_pts;
3259     if( h->reconfig )
3260     {
3261         x264_encoder_reconfig_apply( h, &h->reconfig_h->param );
3262         h->reconfig = 0;
3263     }
3264     if( h->fenc->param )
3265     {
3266         x264_encoder_reconfig_apply( h, h->fenc->param );
3267         if( h->fenc->param->param_free )
3268         {
3269             h->fenc->param->param_free( h->fenc->param );
3270             h->fenc->param = NULL;
3271         }
3272     }
3273
3274     // ok to call this before encoding any frames, since the initial values of fdec have b_kept_as_ref=0
3275     if( x264_reference_update( h ) )
3276         return -1;
3277     h->fdec->i_lines_completed = -1;
3278
3279     if( !IS_X264_TYPE_I( h->fenc->i_type ) )
3280     {
3281         int valid_refs_left = 0;
3282         for( int i = 0; h->frames.reference[i]; i++ )
3283             if( !h->frames.reference[i]->b_corrupt )
3284                 valid_refs_left++;
3285         /* No valid reference frames left: force an IDR. */
3286         if( !valid_refs_left )
3287         {
3288             h->fenc->b_keyframe = 1;
3289             h->fenc->i_type = X264_TYPE_IDR;
3290         }
3291     }
3292
3293     if( h->fenc->b_keyframe )
3294     {
3295         h->frames.i_last_keyframe = h->fenc->i_frame;
3296         if( h->fenc->i_type == X264_TYPE_IDR )
3297         {
3298             h->i_frame_num = 0;
3299             h->frames.i_last_idr = h->fenc->i_frame;
3300         }
3301     }
3302     h->sh.i_mmco_command_count =
3303     h->sh.i_mmco_remove_from_end = 0;
3304     h->b_ref_reorder[0] =
3305     h->b_ref_reorder[1] = 0;
3306     h->fdec->i_poc =
3307     h->fenc->i_poc = 2 * ( h->fenc->i_frame - X264_MAX( h->frames.i_last_idr, 0 ) );
3308
3309     /* ------------------- Setup frame context ----------------------------- */
3310     /* 5: Init data dependent of frame type */
3311     if( h->fenc->i_type == X264_TYPE_IDR )
3312     {
3313         /* reset ref pictures */
3314         i_nal_type    = NAL_SLICE_IDR;
3315         i_nal_ref_idc = NAL_PRIORITY_HIGHEST;
3316         h->sh.i_type = SLICE_TYPE_I;
3317         x264_reference_reset( h );
3318         h->frames.i_poc_last_open_gop = -1;
3319     }
3320     else if( h->fenc->i_type == X264_TYPE_I )
3321     {
3322         i_nal_type    = NAL_SLICE;
3323         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
3324         h->sh.i_type = SLICE_TYPE_I;
3325         x264_reference_hierarchy_reset( h );
3326         if( h->param.b_open_gop )
3327             h->frames.i_poc_last_open_gop = h->fenc->b_keyframe ? h->fenc->i_poc : -1;
3328     }
3329     else if( h->fenc->i_type == X264_TYPE_P )
3330     {
3331         i_nal_type    = NAL_SLICE;
3332         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
3333         h->sh.i_type = SLICE_TYPE_P;
3334         x264_reference_hierarchy_reset( h );
3335         h->frames.i_poc_last_open_gop = -1;
3336     }
3337     else if( h->fenc->i_type == X264_TYPE_BREF )
3338     {
3339         i_nal_type    = NAL_SLICE;
3340         i_nal_ref_idc = h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT ? NAL_PRIORITY_LOW : NAL_PRIORITY_HIGH;
3341         h->sh.i_type = SLICE_TYPE_B;
3342         x264_reference_hierarchy_reset( h );
3343     }
3344     else    /* B frame */
3345     {
3346         i_nal_type    = NAL_SLICE;
3347         i_nal_ref_idc = NAL_PRIORITY_DISPOSABLE;
3348         h->sh.i_type = SLICE_TYPE_B;
3349     }
3350
3351     h->fdec->i_type = h->fenc->i_type;
3352     h->fdec->i_frame = h->fenc->i_frame;
3353     h->fenc->b_kept_as_ref =
3354     h->fdec->b_kept_as_ref = i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE && h->param.i_keyint_max > 1;
3355
3356     h->fdec->mb_info = h->fenc->mb_info;
3357     h->fdec->mb_info_free = h->fenc->mb_info_free;
3358     h->fenc->mb_info = NULL;
3359     h->fenc->mb_info_free = NULL;
3360
3361     h->fdec->i_pts = h->fenc->i_pts;
3362     if( h->frames.i_bframe_delay )
3363     {
3364         int64_t *prev_reordered_pts = thread_current->frames.i_prev_reordered_pts;
3365         h->fdec->i_dts = h->i_frame > h->frames.i_bframe_delay
3366                        ? prev_reordered_pts[ (h->i_frame - h->frames.i_bframe_delay) % h->frames.i_bframe_delay ]
3367                        : h->fenc->i_reordered_pts - h->frames.i_bframe_delay_time;
3368         prev_reordered_pts[ h->i_frame % h->frames.i_bframe_delay ] = h->fenc->i_reordered_pts;
3369     }
3370     else
3371         h->fdec->i_dts = h->fenc->i_reordered_pts;
3372     if( h->fenc->i_type == X264_TYPE_IDR )
3373         h->i_last_idr_pts = h->fdec->i_pts;
3374
3375     /* ------------------- Init                ----------------------------- */
3376     /* build ref list 0/1 */
3377     x264_reference_build_list( h, h->fdec->i_poc );
3378
3379     /* ---------------------- Write the bitstream -------------------------- */
3380     /* Init bitstream context */
3381     if( h->param.b_sliced_threads )
3382     {
3383         for( int i = 0; i < h->param.i_threads; i++ )
3384         {
3385             bs_init( &h->thread[i]->out.bs, h->thread[i]->out.p_bitstream, h->thread[i]->out.i_bitstream );
3386             h->thread[i]->out.i_nal = 0;
3387         }
3388     }
3389     else
3390     {
3391         bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
3392         h->out.i_nal = 0;
3393     }
3394
3395     if( h->param.b_aud )
3396     {
3397         int pic_type;
3398
3399         if( h->sh.i_type == SLICE_TYPE_I )
3400             pic_type = 0;
3401         else if( h->sh.i_type == SLICE_TYPE_P )
3402             pic_type = 1;
3403         else if( h->sh.i_type == SLICE_TYPE_B )
3404             pic_type = 2;
3405         else
3406             pic_type = 7;
3407
3408         x264_nal_start( h, NAL_AUD, NAL_PRIORITY_DISPOSABLE );
3409         bs_write( &h->out.bs, 3, pic_type );
3410         bs_rbsp_trailing( &h->out.bs );
3411         if( x264_nal_end( h ) )
3412             return -1;
3413         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
3414     }
3415
3416     h->i_nal_type = i_nal_type;
3417     h->i_nal_ref_idc = i_nal_ref_idc;
3418
3419     if( h->param.b_intra_refresh )
3420     {
3421         if( IS_X264_TYPE_I( h->fenc->i_type ) )
3422         {
3423             h->fdec->i_frames_since_pir = 0;
3424             h->b_queued_intra_refresh = 0;
3425             /* PIR is currently only supported with ref == 1, so any intra frame effectively refreshes
3426              * the whole frame and counts as an intra refresh. */
3427             h->fdec->f_pir_position = h->mb.i_mb_width;
3428         }
3429         else if( h->fenc->i_type == X264_TYPE_P )
3430         {
3431             int pocdiff = (h->fdec->i_poc - h->fref[0][0]->i_poc)/2;
3432             float increment = X264_MAX( ((float)h->mb.i_mb_width-1) / h->param.i_keyint_max, 1 );
3433             h->fdec->f_pir_position = h->fref[0][0]->f_pir_position;
3434             h->fdec->i_frames_since_pir = h->fref[0][0]->i_frames_since_pir + pocdiff;
3435             if( h->fdec->i_frames_since_pir >= h->param.i_keyint_max ||
3436                 (h->b_queued_intra_refresh && h->fdec->f_pir_position + 0.5 >= h->mb.i_mb_width) )
3437             {
3438                 h->fdec->f_pir_position = 0;
3439                 h->fdec->i_frames_since_pir = 0;
3440                 h->b_queued_intra_refresh = 0;
3441                 h->fenc->b_keyframe = 1;
3442             }
3443             h->fdec->i_pir_start_col = h->fdec->f_pir_position+0.5;
3444             h->fdec->f_pir_position += increment * pocdiff;
3445             h->fdec->i_pir_end_col = h->fdec->f_pir_position+0.5;
3446             /* If our intra refresh has reached the right side of the frame, we're done. */
3447             if( h->fdec->i_pir_end_col >= h->mb.i_mb_width - 1 )
3448             {
3449                 h->fdec->f_pir_position = h->mb.i_mb_width;
3450                 h->fdec->i_pir_end_col = h->mb.i_mb_width - 1;
3451             }
3452         }
3453     }
3454
3455     if( h->fenc->b_keyframe )
3456     {
3457         /* Write SPS and PPS */
3458         if( h->param.b_repeat_headers )
3459         {
3460             /* generate sequence parameters */
3461             x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
3462             x264_sps_write( &h->out.bs, h->sps );
3463             if( x264_nal_end( h ) )
3464                 return -1;
3465             /* Pad AUD/SPS to 256 bytes like Panasonic */
3466             if( h->param.b_avcintra_compat )
3467                 h->out.nal[h->out.i_nal-1].i_padding = 256 - bs_pos( &h->out.bs ) / 8 - 2*NALU_OVERHEAD;
3468             overhead += h->out.nal[h->out.i_nal-1].i_payload + h->out.nal[h->out.i_nal-1].i_padding + NALU_OVERHEAD;
3469
3470             /* generate picture parameters */
3471             x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
3472             x264_pps_write( &h->out.bs, h->sps, h->pps );
3473             if( x264_nal_end( h ) )
3474                 return -1;
3475             if( h->param.b_avcintra_compat )
3476                 h->out.nal[h->out.i_nal-1].i_padding = 256 - h->out.nal[h->out.i_nal-1].i_payload - NALU_OVERHEAD;
3477             overhead += h->out.nal[h->out.i_nal-1].i_payload + h->out.nal[h->out.i_nal-1].i_padding + NALU_OVERHEAD;
3478         }
3479
3480         /* when frame threading is used, buffering period sei is written in x264_encoder_frame_end */
3481         if( h->i_thread_frames == 1 && h->sps->vui.b_nal_hrd_parameters_present )
3482         {
3483             x264_hrd_fullness( h );
3484             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3485             x264_sei_buffering_period_write( h, &h->out.bs );
3486             if( x264_nal_end( h ) )
3487                return -1;
3488             overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3489         }
3490     }
3491
3492     /* write extra sei */
3493     for( int i = 0; i < h->fenc->extra_sei.num_payloads; i++ )
3494     {
3495         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3496         x264_sei_write( &h->out.bs, h->fenc->extra_sei.payloads[i].payload, h->fenc->extra_sei.payloads[i].payload_size,
3497                         h->fenc->extra_sei.payloads[i].payload_type );
3498         if( x264_nal_end( h ) )
3499             return -1;
3500         overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3501         if( h->fenc->extra_sei.sei_free )
3502         {
3503             h->fenc->extra_sei.sei_free( h->fenc->extra_sei.payloads[i].payload );
3504             h->fenc->extra_sei.payloads[i].payload = NULL;
3505         }
3506     }
3507
3508     if( h->fenc->extra_sei.sei_free )
3509     {
3510         h->fenc->extra_sei.sei_free( h->fenc->extra_sei.payloads );
3511         h->fenc->extra_sei.payloads = NULL;
3512         h->fenc->extra_sei.sei_free = NULL;
3513     }
3514
3515     if( h->fenc->b_keyframe )
3516     {
3517         /* Avid's decoder strictly wants two SEIs for AVC-Intra so we can't insert the x264 SEI */
3518         if( h->param.b_repeat_headers && h->fenc->i_frame == 0 && !h->param.b_avcintra_compat )
3519         {
3520             /* identify ourself */
3521             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3522             if( x264_sei_version_write( h, &h->out.bs ) )
3523                 return -1;
3524             if( x264_nal_end( h ) )
3525                 return -1;
3526             overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3527         }
3528
3529         if( h->fenc->i_type != X264_TYPE_IDR )
3530         {
3531             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;
3532             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3533             x264_sei_recovery_point_write( h, &h->out.bs, time_to_recovery );
3534             if( x264_nal_end( h ) )
3535                 return -1;
3536             overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3537         }
3538
3539         if( h->param.i_frame_packing >= 0 )
3540         {
3541             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3542             x264_sei_frame_packing_write( h, &h->out.bs );
3543             if( x264_nal_end( h ) )
3544                 return -1;
3545             overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3546         }
3547     }
3548
3549     /* generate sei pic timing */
3550     if( h->sps->vui.b_pic_struct_present || h->sps->vui.b_nal_hrd_parameters_present )
3551     {
3552         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3553         x264_sei_pic_timing_write( h, &h->out.bs );
3554         if( x264_nal_end( h ) )
3555             return -1;
3556         overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3557     }
3558
3559     /* As required by Blu-ray. */
3560     if( !IS_X264_TYPE_B( h->fenc->i_type ) && h->b_sh_backup )
3561     {
3562         h->b_sh_backup = 0;
3563         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3564         x264_sei_dec_ref_pic_marking_write( h, &h->out.bs );
3565         if( x264_nal_end( h ) )
3566             return -1;
3567         overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3568     }
3569
3570     if( h->fenc->b_keyframe && h->param.b_intra_refresh )
3571         h->i_cpb_delay_pir_offset_next = h->fenc->i_cpb_delay;
3572
3573     /* Filler space: 10 or 18 SEIs' worth of space, depending on resolution */
3574     if( h->param.b_avcintra_compat )
3575     {
3576         /* Write an empty filler NAL to mimic the AUD in the P2 format*/
3577         if( h->param.b_avcintra_compat )
3578         {
3579             x264_nal_start( h, NAL_FILLER, NAL_PRIORITY_DISPOSABLE );
3580             x264_filler_write( h, &h->out.bs, 0 );
3581             if( x264_nal_end( h ) )
3582                 return -1;
3583             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
3584         }
3585
3586         /* All lengths are magic lengths that decoders expect to see */
3587         /* "UMID" SEI */
3588         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3589         if( x264_sei_avcintra_write( h, &h->out.bs, 497, "UMID" ) < 0 )
3590             return -1;
3591         if( x264_nal_end( h ) )
3592             return -1;
3593         overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3594
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_write( h, &h->out.bs, unpadded_len, "VANC" ) < 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             spr