Add "--stitchable" option for segmented encoding
[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 = 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 #if HAVE_MMX
408     /* Misalign mask has to be set separately for each thread. */
409     if( h->param.cpu&X264_CPU_SSE_MISALIGN )
410         x264_cpu_mask_misalign_sse();
411 #endif
412 }
413
414 static void x264_lookahead_thread_init( x264_t *h )
415 {
416 #if HAVE_MMX
417     /* Misalign mask has to be set separately for each thread. */
418     if( h->param.cpu&X264_CPU_SSE_MISALIGN )
419         x264_cpu_mask_misalign_sse();
420 #endif
421 }
422 #endif
423
424 /****************************************************************************
425  *
426  ****************************************************************************
427  ****************************** External API*********************************
428  ****************************************************************************
429  *
430  ****************************************************************************/
431
432 static int x264_validate_parameters( x264_t *h, int b_open )
433 {
434 #if HAVE_MMX
435     if( b_open )
436     {
437         int cpuflags = x264_cpu_detect();
438         int fail = 0;
439 #ifdef __SSE__
440         if( !(cpuflags & X264_CPU_SSE) )
441         {
442             x264_log( h, X264_LOG_ERROR, "your cpu does not support SSE1, but x264 was compiled with asm\n");
443             fail = 1;
444         }
445 #else
446         if( !(cpuflags & X264_CPU_MMX2) )
447         {
448             x264_log( h, X264_LOG_ERROR, "your cpu does not support MMXEXT, but x264 was compiled with asm\n");
449             fail = 1;
450         }
451 #endif
452         if( !fail && !(cpuflags & X264_CPU_CMOV) )
453         {
454             x264_log( h, X264_LOG_ERROR, "your cpu does not support CMOV, but x264 was compiled with asm\n");
455             fail = 1;
456         }
457         if( fail )
458         {
459             x264_log( h, X264_LOG_ERROR, "to run x264, recompile without asm (configure --disable-asm)\n");
460             return -1;
461         }
462     }
463 #endif
464
465 #if HAVE_INTERLACED
466     h->param.b_interlaced = !!PARAM_INTERLACED;
467 #else
468     if( h->param.b_interlaced )
469     {
470         x264_log( h, X264_LOG_ERROR, "not compiled with interlaced support\n" );
471         return -1;
472     }
473 #endif
474
475     if( h->param.i_width <= 0 || h->param.i_height <= 0 )
476     {
477         x264_log( h, X264_LOG_ERROR, "invalid width x height (%dx%d)\n",
478                   h->param.i_width, h->param.i_height );
479         return -1;
480     }
481
482     int i_csp = h->param.i_csp & X264_CSP_MASK;
483 #if X264_CHROMA_FORMAT
484     if( CHROMA_FORMAT != CHROMA_420 && i_csp >= X264_CSP_I420 && i_csp <= X264_CSP_NV12 )
485     {
486         x264_log( h, X264_LOG_ERROR, "not compiled with 4:2:0 support\n" );
487         return -1;
488     }
489     else if( CHROMA_FORMAT != CHROMA_422 && i_csp >= X264_CSP_I422 && i_csp <= X264_CSP_NV16 )
490     {
491         x264_log( h, X264_LOG_ERROR, "not compiled with 4:2:2 support\n" );
492         return -1;
493     }
494     else if( CHROMA_FORMAT != CHROMA_444 && i_csp >= X264_CSP_I444 && i_csp <= X264_CSP_RGB )
495     {
496         x264_log( h, X264_LOG_ERROR, "not compiled with 4:4:4 support\n" );
497         return -1;
498     }
499 #endif
500     if( i_csp <= X264_CSP_NONE || i_csp >= X264_CSP_MAX )
501     {
502         x264_log( h, X264_LOG_ERROR, "invalid CSP (only I420/YV12/NV12/I422/YV16/NV16/I444/YV24/BGR/BGRA/RGB supported)\n" );
503         return -1;
504     }
505
506     if( i_csp < X264_CSP_I444 && h->param.i_width % 2 )
507     {
508         x264_log( h, X264_LOG_ERROR, "width not divisible by 2 (%dx%d)\n",
509                   h->param.i_width, h->param.i_height );
510         return -1;
511     }
512
513     if( i_csp < X264_CSP_I422 && PARAM_INTERLACED && h->param.i_height % 4 )
514     {
515         x264_log( h, X264_LOG_ERROR, "height not divisible by 4 (%dx%d)\n",
516                   h->param.i_width, h->param.i_height );
517         return -1;
518     }
519
520     if( (i_csp < X264_CSP_I422 || PARAM_INTERLACED) && h->param.i_height % 2 )
521     {
522         x264_log( h, X264_LOG_ERROR, "height not divisible by 2 (%dx%d)\n",
523                   h->param.i_width, h->param.i_height );
524         return -1;
525     }
526
527     if( (h->param.crop_rect.i_left + h->param.crop_rect.i_right ) >= h->param.i_width ||
528         (h->param.crop_rect.i_top  + h->param.crop_rect.i_bottom) >= h->param.i_height )
529     {
530         x264_log( h, X264_LOG_ERROR, "invalid crop-rect %u,%u,%u,%u\n", h->param.crop_rect.i_left,
531                   h->param.crop_rect.i_top, h->param.crop_rect.i_right,  h->param.crop_rect.i_bottom );
532         return -1;
533     }
534
535     if( h->param.i_threads == X264_THREADS_AUTO )
536         h->param.i_threads = x264_cpu_num_processors() * (h->param.b_sliced_threads?2:3)/2;
537     int max_sliced_threads = X264_MAX( 1, (h->param.i_height+15)/16 / 4 );
538     if( h->param.i_threads > 1 )
539     {
540 #if !HAVE_THREAD
541         x264_log( h, X264_LOG_WARNING, "not compiled with thread support!\n");
542         h->param.i_threads = 1;
543 #endif
544         /* Avoid absurdly small thread slices as they can reduce performance
545          * and VBV compliance.  Capped at an arbitrary 4 rows per thread. */
546         if( h->param.b_sliced_threads )
547             h->param.i_threads = X264_MIN( h->param.i_threads, max_sliced_threads );
548     }
549     h->param.i_threads = x264_clip3( h->param.i_threads, 1, X264_THREAD_MAX );
550     if( h->param.i_threads == 1 )
551     {
552         h->param.b_sliced_threads = 0;
553         h->param.i_lookahead_threads = 1;
554     }
555     h->i_thread_frames = h->param.b_sliced_threads ? 1 : h->param.i_threads;
556     if( h->i_thread_frames > 1 )
557         h->param.nalu_process = NULL;
558
559     if( h->param.b_opencl )
560     {
561 #if !HAVE_OPENCL
562         x264_log( h, X264_LOG_WARNING, "OpenCL: not compiled with OpenCL support, disabling\n" );
563         h->param.b_opencl = 0;
564 #elif BIT_DEPTH > 8
565         x264_log( h, X264_LOG_WARNING, "OpenCL lookahead does not support high bit depth, disabling opencl\n" );
566         h->param.b_opencl = 0;
567 #else
568         if( h->param.i_width < 32 || h->param.i_height < 32 )
569         {
570             x264_log( h, X264_LOG_WARNING, "OpenCL: frame size is too small, disabling opencl\n" );
571             h->param.b_opencl = 0;
572         }
573 #endif
574         if( h->param.opencl_device_id && h->param.i_opencl_device )
575         {
576             x264_log( h, X264_LOG_WARNING, "OpenCL: device id and device skip count configured; dropping skip\n" );
577             h->param.i_opencl_device = 0;
578         }
579     }
580
581     h->param.i_keyint_max = x264_clip3( h->param.i_keyint_max, 1, X264_KEYINT_MAX_INFINITE );
582     if( h->param.i_keyint_max == 1 )
583     {
584         h->param.b_intra_refresh = 0;
585         h->param.analyse.i_weighted_pred = 0;
586     }
587
588     h->param.i_frame_packing = x264_clip3( h->param.i_frame_packing, -1, 5 );
589
590     /* Detect default ffmpeg settings and terminate with an error. */
591     if( b_open )
592     {
593         int score = 0;
594         score += h->param.analyse.i_me_range == 0;
595         score += h->param.rc.i_qp_step == 3;
596         score += h->param.i_keyint_max == 12;
597         score += h->param.rc.i_qp_min == 2;
598         score += h->param.rc.i_qp_max == 31;
599         score += h->param.rc.f_qcompress == 0.5;
600         score += fabs(h->param.rc.f_ip_factor - 1.25) < 0.01;
601         score += fabs(h->param.rc.f_pb_factor - 1.25) < 0.01;
602         score += h->param.analyse.inter == 0 && h->param.analyse.i_subpel_refine == 8;
603         if( score >= 5 )
604         {
605             x264_log( h, X264_LOG_ERROR, "broken ffmpeg default settings detected\n" );
606             x264_log( h, X264_LOG_ERROR, "use an encoding preset (e.g. -vpre medium)\n" );
607             x264_log( h, X264_LOG_ERROR, "preset usage: -vpre <speed> -vpre <profile>\n" );
608             x264_log( h, X264_LOG_ERROR, "speed presets are listed in x264 --help\n" );
609             x264_log( h, X264_LOG_ERROR, "profile is optional; x264 defaults to high\n" );
610             return -1;
611         }
612     }
613
614     if( h->param.rc.i_rc_method < 0 || h->param.rc.i_rc_method > 2 )
615     {
616         x264_log( h, X264_LOG_ERROR, "no ratecontrol method specified\n" );
617         return -1;
618     }
619     h->param.rc.f_rf_constant = x264_clip3f( h->param.rc.f_rf_constant, -QP_BD_OFFSET, 51 );
620     h->param.rc.f_rf_constant_max = x264_clip3f( h->param.rc.f_rf_constant_max, -QP_BD_OFFSET, 51 );
621     h->param.rc.i_qp_constant = x264_clip3( h->param.rc.i_qp_constant, 0, QP_MAX );
622     h->param.analyse.i_subpel_refine = x264_clip3( h->param.analyse.i_subpel_refine, 0, 11 );
623     h->param.rc.f_ip_factor = X264_MAX( h->param.rc.f_ip_factor, 0.01f );
624     h->param.rc.f_pb_factor = X264_MAX( h->param.rc.f_pb_factor, 0.01f );
625     if( h->param.rc.i_rc_method == X264_RC_CRF )
626     {
627         h->param.rc.i_qp_constant = h->param.rc.f_rf_constant + QP_BD_OFFSET;
628         h->param.rc.i_bitrate = 0;
629     }
630     if( b_open && (h->param.rc.i_rc_method == X264_RC_CQP || h->param.rc.i_rc_method == X264_RC_CRF)
631         && h->param.rc.i_qp_constant == 0 )
632     {
633         h->mb.b_lossless = 1;
634         h->param.i_cqm_preset = X264_CQM_FLAT;
635         h->param.psz_cqm_file = NULL;
636         h->param.rc.i_rc_method = X264_RC_CQP;
637         h->param.rc.f_ip_factor = 1;
638         h->param.rc.f_pb_factor = 1;
639         h->param.analyse.b_psnr = 0;
640         h->param.analyse.b_ssim = 0;
641         h->param.analyse.i_chroma_qp_offset = 0;
642         h->param.analyse.i_trellis = 0;
643         h->param.analyse.b_fast_pskip = 0;
644         h->param.analyse.i_noise_reduction = 0;
645         h->param.analyse.b_psy = 0;
646         h->param.i_bframe = 0;
647         /* 8x8dct is not useful without RD in CAVLC lossless */
648         if( !h->param.b_cabac && h->param.analyse.i_subpel_refine < 6 )
649             h->param.analyse.b_transform_8x8 = 0;
650     }
651     if( h->param.rc.i_rc_method == X264_RC_CQP )
652     {
653         float qp_p = h->param.rc.i_qp_constant;
654         float qp_i = qp_p - 6*log2f( h->param.rc.f_ip_factor );
655         float qp_b = qp_p + 6*log2f( h->param.rc.f_pb_factor );
656         h->param.rc.i_qp_min = x264_clip3( (int)(X264_MIN3( qp_p, qp_i, qp_b )), 0, QP_MAX );
657         h->param.rc.i_qp_max = x264_clip3( (int)(X264_MAX3( qp_p, qp_i, qp_b ) + .999), 0, QP_MAX );
658         h->param.rc.i_aq_mode = 0;
659         h->param.rc.b_mb_tree = 0;
660         h->param.rc.i_bitrate = 0;
661     }
662     h->param.rc.i_qp_max = x264_clip3( h->param.rc.i_qp_max, 0, QP_MAX );
663     h->param.rc.i_qp_min = x264_clip3( h->param.rc.i_qp_min, 0, h->param.rc.i_qp_max );
664     h->param.rc.i_qp_step = x264_clip3( h->param.rc.i_qp_step, 2, QP_MAX );
665     h->param.rc.i_bitrate = x264_clip3( h->param.rc.i_bitrate, 0, 2000000 );
666     if( h->param.rc.i_rc_method == X264_RC_ABR && !h->param.rc.i_bitrate )
667     {
668         x264_log( h, X264_LOG_ERROR, "bitrate not specified\n" );
669         return -1;
670     }
671     h->param.rc.i_vbv_buffer_size = x264_clip3( h->param.rc.i_vbv_buffer_size, 0, 2000000 );
672     h->param.rc.i_vbv_max_bitrate = x264_clip3( h->param.rc.i_vbv_max_bitrate, 0, 2000000 );
673     h->param.rc.f_vbv_buffer_init = x264_clip3f( h->param.rc.f_vbv_buffer_init, 0, 2000000 );
674     if( h->param.rc.i_vbv_buffer_size )
675     {
676         if( h->param.rc.i_rc_method == X264_RC_CQP )
677         {
678             x264_log( h, X264_LOG_WARNING, "VBV is incompatible with constant QP, ignored.\n" );
679             h->param.rc.i_vbv_max_bitrate = 0;
680             h->param.rc.i_vbv_buffer_size = 0;
681         }
682         else if( h->param.rc.i_vbv_max_bitrate == 0 )
683         {
684             if( h->param.rc.i_rc_method == X264_RC_ABR )
685             {
686                 x264_log( h, X264_LOG_WARNING, "VBV maxrate unspecified, assuming CBR\n" );
687                 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate;
688             }
689             else
690             {
691                 x264_log( h, X264_LOG_WARNING, "VBV bufsize set but maxrate unspecified, ignored\n" );
692                 h->param.rc.i_vbv_buffer_size = 0;
693             }
694         }
695         else if( h->param.rc.i_vbv_max_bitrate < h->param.rc.i_bitrate &&
696                  h->param.rc.i_rc_method == X264_RC_ABR )
697         {
698             x264_log( h, X264_LOG_WARNING, "max bitrate less than average bitrate, assuming CBR\n" );
699             h->param.rc.i_bitrate = h->param.rc.i_vbv_max_bitrate;
700         }
701     }
702     else if( h->param.rc.i_vbv_max_bitrate )
703     {
704         x264_log( h, X264_LOG_WARNING, "VBV maxrate specified, but no bufsize, ignored\n" );
705         h->param.rc.i_vbv_max_bitrate = 0;
706     }
707
708     h->param.i_slice_max_size = X264_MAX( h->param.i_slice_max_size, 0 );
709     h->param.i_slice_max_mbs = X264_MAX( h->param.i_slice_max_mbs, 0 );
710     h->param.i_slice_min_mbs = X264_MAX( h->param.i_slice_min_mbs, 0 );
711     if( h->param.i_slice_max_mbs )
712         h->param.i_slice_min_mbs = X264_MIN( h->param.i_slice_min_mbs, h->param.i_slice_max_mbs/2 );
713     else if( !h->param.i_slice_max_size )
714         h->param.i_slice_min_mbs = 0;
715     if( PARAM_INTERLACED && h->param.i_slice_min_mbs )
716     {
717         x264_log( h, X264_LOG_WARNING, "interlace + slice-min-mbs is not implemented\n" );
718         h->param.i_slice_min_mbs = 0;
719     }
720     int mb_width = (h->param.i_width+15)/16;
721     if( h->param.i_slice_min_mbs > mb_width )
722     {
723         x264_log( h, X264_LOG_WARNING, "slice-min-mbs > row mb size (%d) not implemented\n", mb_width );
724         h->param.i_slice_min_mbs = mb_width;
725     }
726
727     int max_slices = (h->param.i_height+((16<<PARAM_INTERLACED)-1))/(16<<PARAM_INTERLACED);
728     if( h->param.b_sliced_threads )
729         h->param.i_slice_count = x264_clip3( h->param.i_threads, 0, max_slices );
730     else
731     {
732         h->param.i_slice_count = x264_clip3( h->param.i_slice_count, 0, max_slices );
733         if( h->param.i_slice_max_mbs || h->param.i_slice_max_size )
734             h->param.i_slice_count = 0;
735     }
736     if( h->param.i_slice_count_max > 0 )
737         h->param.i_slice_count_max = X264_MAX( h->param.i_slice_count, h->param.i_slice_count_max );
738
739     if( h->param.b_bluray_compat )
740     {
741         h->param.i_bframe_pyramid = X264_MIN( X264_B_PYRAMID_STRICT, h->param.i_bframe_pyramid );
742         h->param.i_bframe = X264_MIN( h->param.i_bframe, 3 );
743         h->param.b_aud = 1;
744         h->param.i_nal_hrd = X264_MAX( h->param.i_nal_hrd, X264_NAL_HRD_VBR );
745         h->param.i_slice_max_size = 0;
746         h->param.i_slice_max_mbs = 0;
747         h->param.b_intra_refresh = 0;
748         h->param.i_frame_reference = X264_MIN( h->param.i_frame_reference, 6 );
749         h->param.i_dpb_size = X264_MIN( h->param.i_dpb_size, 6 );
750         /* Don't use I-frames, because Blu-ray treats them the same as IDR. */
751         h->param.i_keyint_min = 1;
752         /* Due to the proliferation of broken players that don't handle dupes properly. */
753         h->param.analyse.i_weighted_pred = X264_MIN( h->param.analyse.i_weighted_pred, X264_WEIGHTP_SIMPLE );
754         if( h->param.b_fake_interlaced )
755             h->param.b_pic_struct = 1;
756     }
757
758     h->param.i_frame_reference = x264_clip3( h->param.i_frame_reference, 1, X264_REF_MAX );
759     h->param.i_dpb_size = x264_clip3( h->param.i_dpb_size, 1, X264_REF_MAX );
760     if( h->param.i_scenecut_threshold < 0 )
761         h->param.i_scenecut_threshold = 0;
762     h->param.analyse.i_direct_mv_pred = x264_clip3( h->param.analyse.i_direct_mv_pred, X264_DIRECT_PRED_NONE, X264_DIRECT_PRED_AUTO );
763     if( !h->param.analyse.i_subpel_refine && h->param.analyse.i_direct_mv_pred > X264_DIRECT_PRED_SPATIAL )
764     {
765         x264_log( h, X264_LOG_WARNING, "subme=0 + direct=temporal is not supported\n" );
766         h->param.analyse.i_direct_mv_pred = X264_DIRECT_PRED_SPATIAL;
767     }
768     h->param.i_bframe = x264_clip3( h->param.i_bframe, 0, X264_MIN( X264_BFRAME_MAX, h->param.i_keyint_max-1 ) );
769     h->param.i_bframe_bias = x264_clip3( h->param.i_bframe_bias, -90, 100 );
770     if( h->param.i_bframe <= 1 )
771         h->param.i_bframe_pyramid = X264_B_PYRAMID_NONE;
772     h->param.i_bframe_pyramid = x264_clip3( h->param.i_bframe_pyramid, X264_B_PYRAMID_NONE, X264_B_PYRAMID_NORMAL );
773     h->param.i_bframe_adaptive = x264_clip3( h->param.i_bframe_adaptive, X264_B_ADAPT_NONE, X264_B_ADAPT_TRELLIS );
774     if( !h->param.i_bframe )
775     {
776         h->param.i_bframe_adaptive = X264_B_ADAPT_NONE;
777         h->param.analyse.i_direct_mv_pred = 0;
778         h->param.analyse.b_weighted_bipred = 0;
779         h->param.b_open_gop = 0;
780     }
781     if( h->param.b_intra_refresh && h->param.i_bframe_pyramid == X264_B_PYRAMID_NORMAL )
782     {
783         x264_log( h, X264_LOG_WARNING, "b-pyramid normal + intra-refresh is not supported\n" );
784         h->param.i_bframe_pyramid = X264_B_PYRAMID_STRICT;
785     }
786     if( h->param.b_intra_refresh && (h->param.i_frame_reference > 1 || h->param.i_dpb_size > 1) )
787     {
788         x264_log( h, X264_LOG_WARNING, "ref > 1 + intra-refresh is not supported\n" );
789         h->param.i_frame_reference = 1;
790         h->param.i_dpb_size = 1;
791     }
792     if( h->param.b_intra_refresh && h->param.b_open_gop )
793     {
794         x264_log( h, X264_LOG_WARNING, "intra-refresh is not compatible with open-gop\n" );
795         h->param.b_open_gop = 0;
796     }
797     if( !h->param.i_fps_num || !h->param.i_fps_den )
798     {
799         h->param.i_fps_num = 25;
800         h->param.i_fps_den = 1;
801     }
802     float fps = (float) h->param.i_fps_num / h->param.i_fps_den;
803     if( h->param.i_keyint_min == X264_KEYINT_MIN_AUTO )
804         h->param.i_keyint_min = X264_MIN( h->param.i_keyint_max / 10, fps );
805     h->param.i_keyint_min = x264_clip3( h->param.i_keyint_min, 1, h->param.i_keyint_max/2+1 );
806     h->param.rc.i_lookahead = x264_clip3( h->param.rc.i_lookahead, 0, X264_LOOKAHEAD_MAX );
807     {
808         int maxrate = X264_MAX( h->param.rc.i_vbv_max_bitrate, h->param.rc.i_bitrate );
809         float bufsize = maxrate ? (float)h->param.rc.i_vbv_buffer_size / maxrate : 0;
810         h->param.rc.i_lookahead = X264_MIN( h->param.rc.i_lookahead, X264_MAX( h->param.i_keyint_max, bufsize*fps ) );
811     }
812
813     if( !h->param.i_timebase_num || !h->param.i_timebase_den || !(h->param.b_vfr_input || h->param.b_pulldown) )
814     {
815         h->param.i_timebase_num = h->param.i_fps_den;
816         h->param.i_timebase_den = h->param.i_fps_num;
817     }
818
819     h->param.rc.f_qcompress = x264_clip3f( h->param.rc.f_qcompress, 0.0, 1.0 );
820     if( h->param.i_keyint_max == 1 || h->param.rc.f_qcompress == 1 )
821         h->param.rc.b_mb_tree = 0;
822     if( (!h->param.b_intra_refresh && h->param.i_keyint_max != X264_KEYINT_MAX_INFINITE) &&
823         !h->param.rc.i_lookahead && h->param.rc.b_mb_tree )
824     {
825         x264_log( h, X264_LOG_WARNING, "lookaheadless mb-tree requires intra refresh or infinite keyint\n" );
826         h->param.rc.b_mb_tree = 0;
827     }
828     if( b_open && h->param.rc.b_stat_read )
829         h->param.rc.i_lookahead = 0;
830 #if HAVE_THREAD
831     if( h->param.i_sync_lookahead < 0 )
832         h->param.i_sync_lookahead = h->param.i_bframe + 1;
833     h->param.i_sync_lookahead = X264_MIN( h->param.i_sync_lookahead, X264_LOOKAHEAD_MAX );
834     if( h->param.rc.b_stat_read || h->i_thread_frames == 1 )
835         h->param.i_sync_lookahead = 0;
836 #else
837     h->param.i_sync_lookahead = 0;
838 #endif
839
840     h->param.i_deblocking_filter_alphac0 = x264_clip3( h->param.i_deblocking_filter_alphac0, -6, 6 );
841     h->param.i_deblocking_filter_beta    = x264_clip3( h->param.i_deblocking_filter_beta, -6, 6 );
842     h->param.analyse.i_luma_deadzone[0] = x264_clip3( h->param.analyse.i_luma_deadzone[0], 0, 32 );
843     h->param.analyse.i_luma_deadzone[1] = x264_clip3( h->param.analyse.i_luma_deadzone[1], 0, 32 );
844
845     h->param.i_cabac_init_idc = x264_clip3( h->param.i_cabac_init_idc, 0, 2 );
846
847     if( h->param.i_cqm_preset < X264_CQM_FLAT || h->param.i_cqm_preset > X264_CQM_CUSTOM )
848         h->param.i_cqm_preset = X264_CQM_FLAT;
849
850     if( h->param.analyse.i_me_method < X264_ME_DIA ||
851         h->param.analyse.i_me_method > X264_ME_TESA )
852         h->param.analyse.i_me_method = X264_ME_HEX;
853     h->param.analyse.i_me_range = x264_clip3( h->param.analyse.i_me_range, 4, 1024 );
854     if( h->param.analyse.i_me_range > 16 && h->param.analyse.i_me_method <= X264_ME_HEX )
855         h->param.analyse.i_me_range = 16;
856     if( h->param.analyse.i_me_method == X264_ME_TESA &&
857         (h->mb.b_lossless || h->param.analyse.i_subpel_refine <= 1) )
858         h->param.analyse.i_me_method = X264_ME_ESA;
859     h->param.analyse.b_mixed_references = h->param.analyse.b_mixed_references && h->param.i_frame_reference > 1;
860     h->param.analyse.inter &= X264_ANALYSE_PSUB16x16|X264_ANALYSE_PSUB8x8|X264_ANALYSE_BSUB16x16|
861                               X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
862     h->param.analyse.intra &= X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
863     if( !(h->param.analyse.inter & X264_ANALYSE_PSUB16x16) )
864         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
865     if( !h->param.analyse.b_transform_8x8 )
866     {
867         h->param.analyse.inter &= ~X264_ANALYSE_I8x8;
868         h->param.analyse.intra &= ~X264_ANALYSE_I8x8;
869     }
870     h->param.analyse.i_trellis = x264_clip3( h->param.analyse.i_trellis, 0, 2 );
871     h->param.rc.i_aq_mode = x264_clip3( h->param.rc.i_aq_mode, 0, 2 );
872     h->param.rc.f_aq_strength = x264_clip3f( h->param.rc.f_aq_strength, 0, 3 );
873     if( h->param.rc.f_aq_strength == 0 )
874         h->param.rc.i_aq_mode = 0;
875
876     if( h->param.i_log_level < X264_LOG_INFO )
877     {
878         h->param.analyse.b_psnr = 0;
879         h->param.analyse.b_ssim = 0;
880     }
881     /* Warn users trying to measure PSNR/SSIM with psy opts on. */
882     if( b_open && (h->param.analyse.b_psnr || h->param.analyse.b_ssim) )
883     {
884         char *s = NULL;
885
886         if( h->param.analyse.b_psy )
887         {
888             s = h->param.analyse.b_psnr ? "psnr" : "ssim";
889             x264_log( h, X264_LOG_WARNING, "--%s used with psy on: results will be invalid!\n", s );
890         }
891         else if( !h->param.rc.i_aq_mode && h->param.analyse.b_ssim )
892         {
893             x264_log( h, X264_LOG_WARNING, "--ssim used with AQ off: results will be invalid!\n" );
894             s = "ssim";
895         }
896         else if(  h->param.rc.i_aq_mode && h->param.analyse.b_psnr )
897         {
898             x264_log( h, X264_LOG_WARNING, "--psnr used with AQ on: results will be invalid!\n" );
899             s = "psnr";
900         }
901         if( s )
902             x264_log( h, X264_LOG_WARNING, "--tune %s should be used if attempting to benchmark %s!\n", s, s );
903     }
904
905     if( !h->param.analyse.b_psy )
906     {
907         h->param.analyse.f_psy_rd = 0;
908         h->param.analyse.f_psy_trellis = 0;
909     }
910     h->param.analyse.f_psy_rd = x264_clip3f( h->param.analyse.f_psy_rd, 0, 10 );
911     h->param.analyse.f_psy_trellis = x264_clip3f( h->param.analyse.f_psy_trellis, 0, 10 );
912     h->mb.i_psy_rd = h->param.analyse.i_subpel_refine >= 6 ? FIX8( h->param.analyse.f_psy_rd ) : 0;
913     h->mb.i_psy_trellis = h->param.analyse.i_trellis ? FIX8( h->param.analyse.f_psy_trellis / 4 ) : 0;
914     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -32, 32);
915     /* In 4:4:4 mode, chroma gets twice as much resolution, so we can halve its quality. */
916     if( b_open && i_csp >= X264_CSP_I444 && i_csp < X264_CSP_BGR && h->param.analyse.b_psy )
917         h->param.analyse.i_chroma_qp_offset += 6;
918     /* Psy RDO increases overall quantizers to improve the quality of luma--this indirectly hurts chroma quality */
919     /* so we lower the chroma QP offset to compensate */
920     if( b_open && h->mb.i_psy_rd )
921         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_rd < 0.25 ? 1 : 2;
922     /* Psy trellis has a similar effect. */
923     if( b_open && h->mb.i_psy_trellis )
924         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_trellis < 0.25 ? 1 : 2;
925     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -12, 12);
926     /* MB-tree requires AQ to be on, even if the strength is zero. */
927     if( !h->param.rc.i_aq_mode && h->param.rc.b_mb_tree )
928     {
929         h->param.rc.i_aq_mode = 1;
930         h->param.rc.f_aq_strength = 0;
931     }
932     h->param.analyse.i_noise_reduction = x264_clip3( h->param.analyse.i_noise_reduction, 0, 1<<16 );
933     if( h->param.analyse.i_subpel_refine >= 10 && (h->param.analyse.i_trellis != 2 || !h->param.rc.i_aq_mode) )
934         h->param.analyse.i_subpel_refine = 9;
935
936     {
937         const x264_level_t *l = x264_levels;
938         if( h->param.i_level_idc < 0 )
939         {
940             int maxrate_bak = h->param.rc.i_vbv_max_bitrate;
941             if( h->param.rc.i_rc_method == X264_RC_ABR && h->param.rc.i_vbv_buffer_size <= 0 )
942                 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate * 2;
943             x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
944             do h->param.i_level_idc = l->level_idc;
945                 while( l[1].level_idc && x264_validate_levels( h, 0 ) && l++ );
946             h->param.rc.i_vbv_max_bitrate = maxrate_bak;
947         }
948         else
949         {
950             while( l->level_idc && l->level_idc != h->param.i_level_idc )
951                 l++;
952             if( l->level_idc == 0 )
953             {
954                 x264_log( h, X264_LOG_ERROR, "invalid level_idc: %d\n", h->param.i_level_idc );
955                 return -1;
956             }
957         }
958         if( h->param.analyse.i_mv_range <= 0 )
959             h->param.analyse.i_mv_range = l->mv_range >> PARAM_INTERLACED;
960         else
961             h->param.analyse.i_mv_range = x264_clip3(h->param.analyse.i_mv_range, 32, 512 >> PARAM_INTERLACED);
962     }
963
964     h->param.analyse.i_weighted_pred = x264_clip3( h->param.analyse.i_weighted_pred, X264_WEIGHTP_NONE, X264_WEIGHTP_SMART );
965
966     if( h->param.i_lookahead_threads == X264_THREADS_AUTO )
967     {
968         if( h->param.b_sliced_threads )
969             h->param.i_lookahead_threads = h->param.i_threads;
970         else
971         {
972             /* If we're using much slower lookahead settings than encoding settings, it helps a lot to use
973              * more lookahead threads.  This typically happens in the first pass of a two-pass encode, so
974              * try to guess at this sort of case.
975              *
976              * Tuned by a little bit of real encoding with the various presets. */
977             int badapt = h->param.i_bframe_adaptive == X264_B_ADAPT_TRELLIS;
978             int subme = X264_MIN( h->param.analyse.i_subpel_refine / 3, 3 ) + (h->param.analyse.i_subpel_refine > 1);
979             int bframes = X264_MIN( (h->param.i_bframe - 1) / 3, 3 );
980
981             /* [b-adapt 0/1 vs 2][quantized subme][quantized bframes] */
982             static const uint8_t lookahead_thread_div[2][5][4] =
983             {{{6,6,6,6}, {3,3,3,3}, {4,4,4,4}, {6,6,6,6}, {12,12,12,12}},
984              {{3,2,1,1}, {2,1,1,1}, {4,3,2,1}, {6,4,3,2}, {12, 9, 6, 4}}};
985
986             h->param.i_lookahead_threads = h->param.i_threads / lookahead_thread_div[badapt][subme][bframes];
987             /* Since too many lookahead threads significantly degrades lookahead accuracy, limit auto
988              * lookahead threads to about 8 macroblock rows high each at worst.  This number is chosen
989              * pretty much arbitrarily. */
990             h->param.i_lookahead_threads = X264_MIN( h->param.i_lookahead_threads, h->param.i_height / 128 );
991         }
992     }
993     h->param.i_lookahead_threads = x264_clip3( h->param.i_lookahead_threads, 1, X264_MIN( max_sliced_threads, X264_LOOKAHEAD_THREAD_MAX ) );
994
995     if( PARAM_INTERLACED )
996     {
997         if( h->param.analyse.i_me_method >= X264_ME_ESA )
998         {
999             x264_log( h, X264_LOG_WARNING, "interlace + me=esa is not implemented\n" );
1000             h->param.analyse.i_me_method = X264_ME_UMH;
1001         }
1002         if( h->param.analyse.i_weighted_pred > 0 )
1003         {
1004             x264_log( h, X264_LOG_WARNING, "interlace + weightp is not implemented\n" );
1005             h->param.analyse.i_weighted_pred = X264_WEIGHTP_NONE;
1006         }
1007     }
1008
1009     if( !h->param.analyse.i_weighted_pred && h->param.rc.b_mb_tree && h->param.analyse.b_psy )
1010         h->param.analyse.i_weighted_pred = X264_WEIGHTP_FAKE;
1011
1012     if( h->i_thread_frames > 1 )
1013     {
1014         int r = h->param.analyse.i_mv_range_thread;
1015         int r2;
1016         if( r <= 0 )
1017         {
1018             // half of the available space is reserved and divided evenly among the threads,
1019             // the rest is allocated to whichever thread is far enough ahead to use it.
1020             // reserving more space increases quality for some videos, but costs more time
1021             // in thread synchronization.
1022             int max_range = (h->param.i_height + X264_THREAD_HEIGHT) / h->i_thread_frames - X264_THREAD_HEIGHT;
1023             r = max_range / 2;
1024         }
1025         r = X264_MAX( r, h->param.analyse.i_me_range );
1026         r = X264_MIN( r, h->param.analyse.i_mv_range );
1027         // round up to use the whole mb row
1028         r2 = (r & ~15) + ((-X264_THREAD_HEIGHT) & 15);
1029         if( r2 < r )
1030             r2 += 16;
1031         x264_log( h, X264_LOG_DEBUG, "using mv_range_thread = %d\n", r2 );
1032         h->param.analyse.i_mv_range_thread = r2;
1033     }
1034
1035     if( h->param.rc.f_rate_tolerance < 0 )
1036         h->param.rc.f_rate_tolerance = 0;
1037     if( h->param.rc.f_qblur < 0 )
1038         h->param.rc.f_qblur = 0;
1039     if( h->param.rc.f_complexity_blur < 0 )
1040         h->param.rc.f_complexity_blur = 0;
1041
1042     h->param.i_sps_id &= 31;
1043
1044     if( PARAM_INTERLACED )
1045         h->param.b_pic_struct = 1;
1046
1047     h->param.i_nal_hrd = x264_clip3( h->param.i_nal_hrd, X264_NAL_HRD_NONE, X264_NAL_HRD_CBR );
1048
1049     if( h->param.i_nal_hrd && !h->param.rc.i_vbv_buffer_size )
1050     {
1051         x264_log( h, X264_LOG_WARNING, "NAL HRD parameters require VBV parameters\n" );
1052         h->param.i_nal_hrd = X264_NAL_HRD_NONE;
1053     }
1054
1055     if( h->param.i_nal_hrd == X264_NAL_HRD_CBR &&
1056        (h->param.rc.i_bitrate != h->param.rc.i_vbv_max_bitrate || !h->param.rc.i_vbv_max_bitrate) )
1057     {
1058         x264_log( h, X264_LOG_WARNING, "CBR HRD requires constant bitrate\n" );
1059         h->param.i_nal_hrd = X264_NAL_HRD_VBR;
1060     }
1061
1062     /* ensure the booleans are 0 or 1 so they can be used in math */
1063 #define BOOLIFY(x) h->param.x = !!h->param.x
1064     BOOLIFY( b_cabac );
1065     BOOLIFY( b_constrained_intra );
1066     BOOLIFY( b_deblocking_filter );
1067     BOOLIFY( b_deterministic );
1068     BOOLIFY( b_sliced_threads );
1069     BOOLIFY( b_interlaced );
1070     BOOLIFY( b_intra_refresh );
1071     BOOLIFY( b_visualize );
1072     BOOLIFY( b_aud );
1073     BOOLIFY( b_repeat_headers );
1074     BOOLIFY( b_annexb );
1075     BOOLIFY( b_vfr_input );
1076     BOOLIFY( b_pulldown );
1077     BOOLIFY( b_tff );
1078     BOOLIFY( b_pic_struct );
1079     BOOLIFY( b_fake_interlaced );
1080     BOOLIFY( b_open_gop );
1081     BOOLIFY( b_bluray_compat );
1082     BOOLIFY( b_stitchable );
1083     BOOLIFY( b_full_recon );
1084     BOOLIFY( b_opencl );
1085     BOOLIFY( analyse.b_transform_8x8 );
1086     BOOLIFY( analyse.b_weighted_bipred );
1087     BOOLIFY( analyse.b_chroma_me );
1088     BOOLIFY( analyse.b_mixed_references );
1089     BOOLIFY( analyse.b_fast_pskip );
1090     BOOLIFY( analyse.b_dct_decimate );
1091     BOOLIFY( analyse.b_psy );
1092     BOOLIFY( analyse.b_psnr );
1093     BOOLIFY( analyse.b_ssim );
1094     BOOLIFY( rc.b_stat_write );
1095     BOOLIFY( rc.b_stat_read );
1096     BOOLIFY( rc.b_mb_tree );
1097 #undef BOOLIFY
1098
1099     return 0;
1100 }
1101
1102 static void mbcmp_init( x264_t *h )
1103 {
1104     int satd = !h->mb.b_lossless && h->param.analyse.i_subpel_refine > 1;
1105     memcpy( h->pixf.mbcmp, satd ? h->pixf.satd : h->pixf.sad_aligned, sizeof(h->pixf.mbcmp) );
1106     memcpy( h->pixf.mbcmp_unaligned, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.mbcmp_unaligned) );
1107     h->pixf.intra_mbcmp_x3_16x16 = satd ? h->pixf.intra_satd_x3_16x16 : h->pixf.intra_sad_x3_16x16;
1108     h->pixf.intra_mbcmp_x3_8x16c = satd ? h->pixf.intra_satd_x3_8x16c : h->pixf.intra_sad_x3_8x16c;
1109     h->pixf.intra_mbcmp_x3_8x8c  = satd ? h->pixf.intra_satd_x3_8x8c  : h->pixf.intra_sad_x3_8x8c;
1110     h->pixf.intra_mbcmp_x3_8x8 = satd ? h->pixf.intra_sa8d_x3_8x8 : h->pixf.intra_sad_x3_8x8;
1111     h->pixf.intra_mbcmp_x3_4x4 = satd ? h->pixf.intra_satd_x3_4x4 : h->pixf.intra_sad_x3_4x4;
1112     h->pixf.intra_mbcmp_x9_4x4 = h->param.b_cpu_independent || h->mb.b_lossless ? NULL
1113                                : satd ? h->pixf.intra_satd_x9_4x4 : h->pixf.intra_sad_x9_4x4;
1114     h->pixf.intra_mbcmp_x9_8x8 = h->param.b_cpu_independent || h->mb.b_lossless ? NULL
1115                                : satd ? h->pixf.intra_sa8d_x9_8x8 : h->pixf.intra_sad_x9_8x8;
1116     satd &= h->param.analyse.i_me_method == X264_ME_TESA;
1117     memcpy( h->pixf.fpelcmp, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.fpelcmp) );
1118     memcpy( h->pixf.fpelcmp_x3, satd ? h->pixf.satd_x3 : h->pixf.sad_x3, sizeof(h->pixf.fpelcmp_x3) );
1119     memcpy( h->pixf.fpelcmp_x4, satd ? h->pixf.satd_x4 : h->pixf.sad_x4, sizeof(h->pixf.fpelcmp_x4) );
1120 }
1121
1122 static void chroma_dsp_init( x264_t *h )
1123 {
1124     memcpy( h->luma2chroma_pixel, x264_luma2chroma_pixel[CHROMA_FORMAT], sizeof(h->luma2chroma_pixel) );
1125
1126     switch( CHROMA_FORMAT )
1127     {
1128         case CHROMA_420:
1129             memcpy( h->predict_chroma, h->predict_8x8c, sizeof(h->predict_chroma) );
1130             h->mc.prefetch_fenc = h->mc.prefetch_fenc_420;
1131             h->loopf.deblock_chroma[0] = h->loopf.deblock_h_chroma_420;
1132             h->loopf.deblock_chroma_intra[0] = h->loopf.deblock_h_chroma_420_intra;
1133             h->loopf.deblock_chroma_mbaff = h->loopf.deblock_chroma_420_mbaff;
1134             h->loopf.deblock_chroma_intra_mbaff = h->loopf.deblock_chroma_420_intra_mbaff;
1135             h->pixf.intra_mbcmp_x3_chroma = h->pixf.intra_mbcmp_x3_8x8c;
1136             h->quantf.coeff_last[DCT_CHROMA_DC] = h->quantf.coeff_last4;
1137             h->quantf.coeff_level_run[DCT_CHROMA_DC] = h->quantf.coeff_level_run4;
1138             break;
1139         case CHROMA_422:
1140             memcpy( h->predict_chroma, h->predict_8x16c, sizeof(h->predict_chroma) );
1141             h->mc.prefetch_fenc = h->mc.prefetch_fenc_422;
1142             h->loopf.deblock_chroma[0] = h->loopf.deblock_h_chroma_422;
1143             h->loopf.deblock_chroma_intra[0] = h->loopf.deblock_h_chroma_422_intra;
1144             h->loopf.deblock_chroma_mbaff = h->loopf.deblock_chroma_422_mbaff;
1145             h->loopf.deblock_chroma_intra_mbaff = h->loopf.deblock_chroma_422_intra_mbaff;
1146             h->pixf.intra_mbcmp_x3_chroma = h->pixf.intra_mbcmp_x3_8x16c;
1147             h->quantf.coeff_last[DCT_CHROMA_DC] = h->quantf.coeff_last8;
1148             h->quantf.coeff_level_run[DCT_CHROMA_DC] = h->quantf.coeff_level_run8;
1149             break;
1150         case CHROMA_444:
1151             h->mc.prefetch_fenc = h->mc.prefetch_fenc_422; /* FIXME: doesn't cover V plane */
1152             h->loopf.deblock_chroma_mbaff = h->loopf.deblock_luma_mbaff;
1153             h->loopf.deblock_chroma_intra_mbaff = h->loopf.deblock_luma_intra_mbaff;
1154             break;
1155     }
1156 }
1157
1158 static void x264_set_aspect_ratio( x264_t *h, x264_param_t *param, int initial )
1159 {
1160     /* VUI */
1161     if( param->vui.i_sar_width > 0 && param->vui.i_sar_height > 0 )
1162     {
1163         uint32_t i_w = param->vui.i_sar_width;
1164         uint32_t i_h = param->vui.i_sar_height;
1165         uint32_t old_w = h->param.vui.i_sar_width;
1166         uint32_t old_h = h->param.vui.i_sar_height;
1167
1168         x264_reduce_fraction( &i_w, &i_h );
1169
1170         while( i_w > 65535 || i_h > 65535 )
1171         {
1172             i_w /= 2;
1173             i_h /= 2;
1174         }
1175
1176         x264_reduce_fraction( &i_w, &i_h );
1177
1178         if( i_w != old_w || i_h != old_h || initial )
1179         {
1180             h->param.vui.i_sar_width = 0;
1181             h->param.vui.i_sar_height = 0;
1182             if( i_w == 0 || i_h == 0 )
1183                 x264_log( h, X264_LOG_WARNING, "cannot create valid sample aspect ratio\n" );
1184             else
1185             {
1186                 x264_log( h, initial?X264_LOG_INFO:X264_LOG_DEBUG, "using SAR=%d/%d\n", i_w, i_h );
1187                 h->param.vui.i_sar_width = i_w;
1188                 h->param.vui.i_sar_height = i_h;
1189             }
1190             x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
1191         }
1192     }
1193 }
1194
1195 /****************************************************************************
1196  * x264_encoder_open:
1197  ****************************************************************************/
1198 x264_t *x264_encoder_open( x264_param_t *param )
1199 {
1200     x264_t *h;
1201     char buf[1000], *p;
1202     int qp, i_slicetype_length;
1203
1204     CHECKED_MALLOCZERO( h, sizeof(x264_t) );
1205
1206     /* Create a copy of param */
1207     memcpy( &h->param, param, sizeof(x264_param_t) );
1208
1209     if( param->param_free )
1210         param->param_free( param );
1211
1212     if( x264_threading_init() )
1213     {
1214         x264_log( h, X264_LOG_ERROR, "unable to initialize threading\n" );
1215         goto fail;
1216     }
1217
1218     if( x264_validate_parameters( h, 1 ) < 0 )
1219         goto fail;
1220
1221     if( h->param.psz_cqm_file )
1222         if( x264_cqm_parse_file( h, h->param.psz_cqm_file ) < 0 )
1223             goto fail;
1224
1225     if( h->param.rc.psz_stat_out )
1226         h->param.rc.psz_stat_out = strdup( h->param.rc.psz_stat_out );
1227     if( h->param.rc.psz_stat_in )
1228         h->param.rc.psz_stat_in = strdup( h->param.rc.psz_stat_in );
1229
1230     x264_reduce_fraction( &h->param.i_fps_num, &h->param.i_fps_den );
1231     x264_reduce_fraction( &h->param.i_timebase_num, &h->param.i_timebase_den );
1232
1233     /* Init x264_t */
1234     h->i_frame = -1;
1235     h->i_frame_num = 0;
1236     h->i_idr_pic_id = 0;
1237
1238     if( (uint64_t)h->param.i_timebase_den * 2 > UINT32_MAX )
1239     {
1240         x264_log( h, X264_LOG_ERROR, "Effective timebase denominator %u exceeds H.264 maximum\n", h->param.i_timebase_den );
1241         goto fail;
1242     }
1243
1244     x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
1245     x264_pps_init( h->pps, h->param.i_sps_id, &h->param, h->sps );
1246
1247     x264_set_aspect_ratio( h, &h->param, 1 );
1248
1249     x264_validate_levels( h, 1 );
1250
1251     h->chroma_qp_table = i_chroma_qp_table + 12 + h->pps->i_chroma_qp_index_offset;
1252
1253     if( x264_cqm_init( h ) < 0 )
1254         goto fail;
1255
1256     h->mb.i_mb_width = h->sps->i_mb_width;
1257     h->mb.i_mb_height = h->sps->i_mb_height;
1258     h->mb.i_mb_count = h->mb.i_mb_width * h->mb.i_mb_height;
1259
1260     h->mb.chroma_h_shift = CHROMA_FORMAT == CHROMA_420 || CHROMA_FORMAT == CHROMA_422;
1261     h->mb.chroma_v_shift = CHROMA_FORMAT == CHROMA_420;
1262
1263     /* Adaptive MBAFF and subme 0 are not supported as we require halving motion
1264      * vectors during prediction, resulting in hpel mvs.
1265      * The chosen solution is to make MBAFF non-adaptive in this case. */
1266     h->mb.b_adaptive_mbaff = PARAM_INTERLACED && h->param.analyse.i_subpel_refine;
1267
1268     /* Init frames. */
1269     if( h->param.i_bframe_adaptive == X264_B_ADAPT_TRELLIS && !h->param.rc.b_stat_read )
1270         h->frames.i_delay = X264_MAX(h->param.i_bframe,3)*4;
1271     else
1272         h->frames.i_delay = h->param.i_bframe;
1273     if( h->param.rc.b_mb_tree || h->param.rc.i_vbv_buffer_size )
1274         h->frames.i_delay = X264_MAX( h->frames.i_delay, h->param.rc.i_lookahead );
1275     i_slicetype_length = h->frames.i_delay;
1276     h->frames.i_delay += h->i_thread_frames - 1;
1277     h->frames.i_delay += h->param.i_sync_lookahead;
1278     h->frames.i_delay += h->param.b_vfr_input;
1279     h->frames.i_bframe_delay = h->param.i_bframe ? (h->param.i_bframe_pyramid ? 2 : 1) : 0;
1280
1281     h->frames.i_max_ref0 = h->param.i_frame_reference;
1282     h->frames.i_max_ref1 = X264_MIN( h->sps->vui.i_num_reorder_frames, h->param.i_frame_reference );
1283     h->frames.i_max_dpb  = h->sps->vui.i_max_dec_frame_buffering;
1284     h->frames.b_have_lowres = !h->param.rc.b_stat_read
1285         && ( h->param.rc.i_rc_method == X264_RC_ABR
1286           || h->param.rc.i_rc_method == X264_RC_CRF
1287           || h->param.i_bframe_adaptive
1288           || h->param.i_scenecut_threshold
1289           || h->param.rc.b_mb_tree
1290           || h->param.analyse.i_weighted_pred );
1291     h->frames.b_have_lowres |= h->param.rc.b_stat_read && h->param.rc.i_vbv_buffer_size > 0;
1292     h->frames.b_have_sub8x8_esa = !!(h->param.analyse.inter & X264_ANALYSE_PSUB8x8);
1293
1294     h->frames.i_last_idr =
1295     h->frames.i_last_keyframe = - h->param.i_keyint_max;
1296     h->frames.i_input    = 0;
1297     h->frames.i_largest_pts = h->frames.i_second_largest_pts = -1;
1298     h->frames.i_poc_last_open_gop = -1;
1299
1300     CHECKED_MALLOCZERO( h->frames.unused[0], (h->frames.i_delay + 3) * sizeof(x264_frame_t *) );
1301     /* Allocate room for max refs plus a few extra just in case. */
1302     CHECKED_MALLOCZERO( h->frames.unused[1], (h->i_thread_frames + X264_REF_MAX + 4) * sizeof(x264_frame_t *) );
1303     CHECKED_MALLOCZERO( h->frames.current, (h->param.i_sync_lookahead + h->param.i_bframe
1304                         + h->i_thread_frames + 3) * sizeof(x264_frame_t *) );
1305     if( h->param.analyse.i_weighted_pred > 0 )
1306         CHECKED_MALLOCZERO( h->frames.blank_unused, h->i_thread_frames * 4 * sizeof(x264_frame_t *) );
1307     h->i_ref[0] = h->i_ref[1] = 0;
1308     h->i_cpb_delay = h->i_coded_fields = h->i_disp_fields = 0;
1309     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);
1310     h->i_disp_fields_last_frame = -1;
1311     x264_rdo_init();
1312
1313     /* init CPU functions */
1314     x264_predict_16x16_init( h->param.cpu, h->predict_16x16 );
1315     x264_predict_8x8c_init( h->param.cpu, h->predict_8x8c );
1316     x264_predict_8x16c_init( h->param.cpu, h->predict_8x16c );
1317     x264_predict_8x8_init( h->param.cpu, h->predict_8x8, &h->predict_8x8_filter );
1318     x264_predict_4x4_init( h->param.cpu, h->predict_4x4 );
1319     x264_pixel_init( h->param.cpu, &h->pixf );
1320     x264_dct_init( h->param.cpu, &h->dctf );
1321     x264_zigzag_init( h->param.cpu, &h->zigzagf_progressive, &h->zigzagf_interlaced );
1322     memcpy( &h->zigzagf, PARAM_INTERLACED ? &h->zigzagf_interlaced : &h->zigzagf_progressive, sizeof(h->zigzagf) );
1323     x264_mc_init( h->param.cpu, &h->mc, h->param.b_cpu_independent );
1324     x264_quant_init( h, h->param.cpu, &h->quantf );
1325     x264_deblock_init( h->param.cpu, &h->loopf, PARAM_INTERLACED );
1326     x264_bitstream_init( h->param.cpu, &h->bsf );
1327     if( h->param.b_cabac )
1328         x264_cabac_init( h );
1329     else
1330         x264_stack_align( x264_cavlc_init, h );
1331
1332     mbcmp_init( h );
1333     chroma_dsp_init( h );
1334
1335     p = buf + sprintf( buf, "using cpu capabilities:" );
1336     for( int i = 0; x264_cpu_names[i].flags; i++ )
1337     {
1338         if( !strcmp(x264_cpu_names[i].name, "SSE")
1339             && h->param.cpu & (X264_CPU_SSE2) )
1340             continue;
1341         if( !strcmp(x264_cpu_names[i].name, "SSE2")
1342             && h->param.cpu & (X264_CPU_SSE2_IS_FAST|X264_CPU_SSE2_IS_SLOW) )
1343             continue;
1344         if( !strcmp(x264_cpu_names[i].name, "SSE3")
1345             && (h->param.cpu & X264_CPU_SSSE3 || !(h->param.cpu & X264_CPU_CACHELINE_64)) )
1346             continue;
1347         if( !strcmp(x264_cpu_names[i].name, "SSE4.1")
1348             && (h->param.cpu & X264_CPU_SSE42) )
1349             continue;
1350         if( !strcmp(x264_cpu_names[i].name, "BMI1")
1351             && (h->param.cpu & X264_CPU_BMI2) )
1352             continue;
1353         if( (h->param.cpu & x264_cpu_names[i].flags) == x264_cpu_names[i].flags
1354             && (!i || x264_cpu_names[i].flags != x264_cpu_names[i-1].flags) )
1355             p += sprintf( p, " %s", x264_cpu_names[i].name );
1356     }
1357     if( !h->param.cpu )
1358         p += sprintf( p, " none!" );
1359     x264_log( h, X264_LOG_INFO, "%s\n", buf );
1360
1361     float *logs = x264_analyse_prepare_costs( h );
1362     if( !logs )
1363         goto fail;
1364     for( qp = X264_MIN( h->param.rc.i_qp_min, QP_MAX_SPEC ); qp <= h->param.rc.i_qp_max; qp++ )
1365         if( x264_analyse_init_costs( h, logs, qp ) )
1366             goto fail;
1367     if( x264_analyse_init_costs( h, logs, X264_LOOKAHEAD_QP ) )
1368         goto fail;
1369     x264_free( logs );
1370
1371     static const uint16_t cost_mv_correct[7] = { 24, 47, 95, 189, 379, 757, 1515 };
1372     /* Checks for known miscompilation issues. */
1373     if( h->cost_mv[X264_LOOKAHEAD_QP][2013] != cost_mv_correct[BIT_DEPTH-8] )
1374     {
1375         x264_log( h, X264_LOG_ERROR, "MV cost test failed: x264 has been miscompiled!\n" );
1376         goto fail;
1377     }
1378
1379     /* Must be volatile or else GCC will optimize it out. */
1380     volatile int temp = 392;
1381     if( x264_clz( temp ) != 23 )
1382     {
1383         x264_log( h, X264_LOG_ERROR, "CLZ test failed: x264 has been miscompiled!\n" );
1384 #if ARCH_X86 || ARCH_X86_64
1385         x264_log( h, X264_LOG_ERROR, "Are you attempting to run an SSE4a/LZCNT-targeted build on a CPU that\n" );
1386         x264_log( h, X264_LOG_ERROR, "doesn't support it?\n" );
1387 #endif
1388         goto fail;
1389     }
1390
1391     h->out.i_nal = 0;
1392     h->out.i_bitstream = X264_MAX( 1000000, h->param.i_width * h->param.i_height * 4
1393         * ( h->param.rc.i_rc_method == X264_RC_ABR ? pow( 0.95, h->param.rc.i_qp_min )
1394           : pow( 0.95, h->param.rc.i_qp_constant ) * X264_MAX( 1, h->param.rc.f_ip_factor )));
1395
1396     h->nal_buffer_size = h->out.i_bitstream * 3/2 + 4 + 64; /* +4 for startcode, +64 for nal_escape assembly padding */
1397     CHECKED_MALLOC( h->nal_buffer, h->nal_buffer_size );
1398
1399     if( h->param.i_threads > 1 &&
1400         x264_threadpool_init( &h->threadpool, h->param.i_threads, (void*)x264_encoder_thread_init, h ) )
1401         goto fail;
1402     if( h->param.i_lookahead_threads > 1 &&
1403         x264_threadpool_init( &h->lookaheadpool, h->param.i_lookahead_threads, (void*)x264_lookahead_thread_init, h ) )
1404         goto fail;
1405
1406 #if HAVE_OPENCL
1407     if( h->param.b_opencl )
1408     {
1409         h->opencl.ocl = x264_opencl_load_library();
1410         if( !h->opencl.ocl )
1411         {
1412             x264_log( h, X264_LOG_WARNING, "failed to load OpenCL\n" );
1413             h->param.b_opencl = 0;
1414         }
1415     }
1416 #endif
1417
1418     h->thread[0] = h;
1419     for( int i = 1; i < h->param.i_threads + !!h->param.i_sync_lookahead; i++ )
1420         CHECKED_MALLOC( h->thread[i], sizeof(x264_t) );
1421     if( h->param.i_lookahead_threads > 1 )
1422         for( int i = 0; i < h->param.i_lookahead_threads; i++ )
1423         {
1424             CHECKED_MALLOC( h->lookahead_thread[i], sizeof(x264_t) );
1425             *h->lookahead_thread[i] = *h;
1426         }
1427
1428     for( int i = 0; i < h->param.i_threads; i++ )
1429     {
1430         int init_nal_count = h->param.i_slice_count + 3;
1431         int allocate_threadlocal_data = !h->param.b_sliced_threads || !i;
1432         if( i > 0 )
1433             *h->thread[i] = *h;
1434
1435         if( x264_pthread_mutex_init( &h->thread[i]->mutex, NULL ) )
1436             goto fail;
1437         if( x264_pthread_cond_init( &h->thread[i]->cv, NULL ) )
1438             goto fail;
1439
1440         if( allocate_threadlocal_data )
1441         {
1442             h->thread[i]->fdec = x264_frame_pop_unused( h, 1 );
1443             if( !h->thread[i]->fdec )
1444                 goto fail;
1445         }
1446         else
1447             h->thread[i]->fdec = h->thread[0]->fdec;
1448
1449         CHECKED_MALLOC( h->thread[i]->out.p_bitstream, h->out.i_bitstream );
1450         /* Start each thread with room for init_nal_count NAL units; it'll realloc later if needed. */
1451         CHECKED_MALLOC( h->thread[i]->out.nal, init_nal_count*sizeof(x264_nal_t) );
1452         h->thread[i]->out.i_nals_allocated = init_nal_count;
1453
1454         if( allocate_threadlocal_data && x264_macroblock_cache_allocate( h->thread[i] ) < 0 )
1455             goto fail;
1456     }
1457
1458 #if HAVE_OPENCL
1459     if( h->param.b_opencl && x264_opencl_lookahead_init( h ) < 0 )
1460         h->param.b_opencl = 0;
1461 #endif
1462
1463     if( x264_lookahead_init( h, i_slicetype_length ) )
1464         goto fail;
1465
1466     for( int i = 0; i < h->param.i_threads; i++ )
1467         if( x264_macroblock_thread_allocate( h->thread[i], 0 ) < 0 )
1468             goto fail;
1469
1470     if( x264_ratecontrol_new( h ) < 0 )
1471         goto fail;
1472
1473     if( h->param.i_nal_hrd )
1474     {
1475         x264_log( h, X264_LOG_DEBUG, "HRD bitrate: %i bits/sec\n", h->sps->vui.hrd.i_bit_rate_unscaled );
1476         x264_log( h, X264_LOG_DEBUG, "CPB size: %i bits\n", h->sps->vui.hrd.i_cpb_size_unscaled );
1477     }
1478
1479     if( h->param.psz_dump_yuv )
1480     {
1481         /* create or truncate the reconstructed video file */
1482         FILE *f = fopen( h->param.psz_dump_yuv, "w" );
1483         if( !f )
1484         {
1485             x264_log( h, X264_LOG_ERROR, "dump_yuv: can't write to %s\n", h->param.psz_dump_yuv );
1486             goto fail;
1487         }
1488         else if( !x264_is_regular_file( f ) )
1489         {
1490             x264_log( h, X264_LOG_ERROR, "dump_yuv: incompatible with non-regular file %s\n", h->param.psz_dump_yuv );
1491             goto fail;
1492         }
1493         fclose( f );
1494     }
1495
1496     const char *profile = h->sps->i_profile_idc == PROFILE_BASELINE ? "Constrained Baseline" :
1497                           h->sps->i_profile_idc == PROFILE_MAIN ? "Main" :
1498                           h->sps->i_profile_idc == PROFILE_HIGH ? "High" :
1499                           h->sps->i_profile_idc == PROFILE_HIGH10 ? (h->sps->b_constraint_set3 == 1 ? "High 10 Intra" : "High 10") :
1500                           h->sps->i_profile_idc == PROFILE_HIGH422 ? (h->sps->b_constraint_set3 == 1 ? "High 4:2:2 Intra" : "High 4:2:2") :
1501                           h->sps->b_constraint_set3 == 1 ? "High 4:4:4 Intra" : "High 4:4:4 Predictive";
1502     char level[4];
1503     snprintf( level, sizeof(level), "%d.%d", h->sps->i_level_idc/10, h->sps->i_level_idc%10 );
1504     if( h->sps->i_level_idc == 9 || ( h->sps->i_level_idc == 11 && h->sps->b_constraint_set3 &&
1505         (h->sps->i_profile_idc == PROFILE_BASELINE || h->sps->i_profile_idc == PROFILE_MAIN) ) )
1506         strcpy( level, "1b" );
1507
1508     if( h->sps->i_profile_idc < PROFILE_HIGH10 )
1509     {
1510         x264_log( h, X264_LOG_INFO, "profile %s, level %s\n",
1511             profile, level );
1512     }
1513     else
1514     {
1515         static const char * const subsampling[4] = { "4:0:0", "4:2:0", "4:2:2", "4:4:4" };
1516         x264_log( h, X264_LOG_INFO, "profile %s, level %s, %s %d-bit\n",
1517             profile, level, subsampling[CHROMA_FORMAT], BIT_DEPTH );
1518     }
1519
1520     return h;
1521 fail:
1522     x264_free( h );
1523     return NULL;
1524 }
1525
1526 /****************************************************************************
1527  * x264_encoder_reconfig:
1528  ****************************************************************************/
1529 int x264_encoder_reconfig( x264_t *h, x264_param_t *param )
1530 {
1531     /* If the previous frame isn't done encoding, reconfiguring is probably dangerous. */
1532     if( h->param.b_sliced_threads )
1533         if( x264_threadpool_wait_all( h ) < 0 )
1534             return -1;
1535
1536     int rc_reconfig = 0;
1537     h = h->thread[h->thread[0]->i_thread_phase];
1538     x264_set_aspect_ratio( h, param, 0 );
1539 #define COPY(var) h->param.var = param->var
1540     COPY( i_frame_reference ); // but never uses more refs than initially specified
1541     COPY( i_bframe_bias );
1542     if( h->param.i_scenecut_threshold )
1543         COPY( i_scenecut_threshold ); // can't turn it on or off, only vary the threshold
1544     COPY( b_deblocking_filter );
1545     COPY( i_deblocking_filter_alphac0 );
1546     COPY( i_deblocking_filter_beta );
1547     COPY( i_frame_packing );
1548     COPY( analyse.inter );
1549     COPY( analyse.intra );
1550     COPY( analyse.i_direct_mv_pred );
1551     /* Scratch buffer prevents me_range from being increased for esa/tesa */
1552     if( h->param.analyse.i_me_method < X264_ME_ESA || param->analyse.i_me_range < h->param.analyse.i_me_range )
1553         COPY( analyse.i_me_range );
1554     COPY( analyse.i_noise_reduction );
1555     /* We can't switch out of subme=0 during encoding. */
1556     if( h->param.analyse.i_subpel_refine )
1557         COPY( analyse.i_subpel_refine );
1558     COPY( analyse.i_trellis );
1559     COPY( analyse.b_chroma_me );
1560     COPY( analyse.b_dct_decimate );
1561     COPY( analyse.b_fast_pskip );
1562     COPY( analyse.b_mixed_references );
1563     COPY( analyse.f_psy_rd );
1564     COPY( analyse.f_psy_trellis );
1565     COPY( crop_rect );
1566     // can only twiddle these if they were enabled to begin with:
1567     if( h->param.analyse.i_me_method >= X264_ME_ESA || param->analyse.i_me_method < X264_ME_ESA )
1568         COPY( analyse.i_me_method );
1569     if( h->param.analyse.i_me_method >= X264_ME_ESA && !h->frames.b_have_sub8x8_esa )
1570         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
1571     if( h->pps->b_transform_8x8_mode )
1572         COPY( analyse.b_transform_8x8 );
1573     if( h->frames.i_max_ref1 > 1 )
1574         COPY( i_bframe_pyramid );
1575     COPY( i_slice_max_size );
1576     COPY( i_slice_max_mbs );
1577     COPY( i_slice_min_mbs );
1578     COPY( i_slice_count );
1579     COPY( i_slice_count_max );
1580     COPY( b_tff );
1581
1582     /* VBV can't be turned on if it wasn't on to begin with */
1583     if( h->param.rc.i_vbv_max_bitrate > 0 && h->param.rc.i_vbv_buffer_size > 0 &&
1584           param->rc.i_vbv_max_bitrate > 0 &&   param->rc.i_vbv_buffer_size > 0 )
1585     {
1586         rc_reconfig |= h->param.rc.i_vbv_max_bitrate != param->rc.i_vbv_max_bitrate;
1587         rc_reconfig |= h->param.rc.i_vbv_buffer_size != param->rc.i_vbv_buffer_size;
1588         rc_reconfig |= h->param.rc.i_bitrate != param->rc.i_bitrate;
1589         COPY( rc.i_vbv_max_bitrate );
1590         COPY( rc.i_vbv_buffer_size );
1591         COPY( rc.i_bitrate );
1592     }
1593     rc_reconfig |= h->param.rc.f_rf_constant != param->rc.f_rf_constant;
1594     rc_reconfig |= h->param.rc.f_rf_constant_max != param->rc.f_rf_constant_max;
1595     COPY( rc.f_rf_constant );
1596     COPY( rc.f_rf_constant_max );
1597 #undef COPY
1598
1599     mbcmp_init( h );
1600
1601     int ret = x264_validate_parameters( h, 0 );
1602
1603     /* Supported reconfiguration options (1-pass only):
1604      * vbv-maxrate
1605      * vbv-bufsize
1606      * crf
1607      * bitrate (CBR only) */
1608     if( !ret && rc_reconfig )
1609         x264_ratecontrol_init_reconfigurable( h, 0 );
1610
1611     return ret;
1612 }
1613
1614 /****************************************************************************
1615  * x264_encoder_parameters:
1616  ****************************************************************************/
1617 void x264_encoder_parameters( x264_t *h, x264_param_t *param )
1618 {
1619     memcpy( param, &h->thread[h->i_thread_phase]->param, sizeof(x264_param_t) );
1620 }
1621
1622 /* internal usage */
1623 static void x264_nal_start( x264_t *h, int i_type, int i_ref_idc )
1624 {
1625     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1626
1627     nal->i_ref_idc        = i_ref_idc;
1628     nal->i_type           = i_type;
1629     nal->b_long_startcode = 1;
1630
1631     nal->i_payload= 0;
1632     nal->p_payload= &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8];
1633 }
1634
1635 /* if number of allocated nals is not enough, re-allocate a larger one. */
1636 static int x264_nal_check_buffer( x264_t *h )
1637 {
1638     if( h->out.i_nal >= h->out.i_nals_allocated )
1639     {
1640         x264_nal_t *new_out = x264_malloc( sizeof(x264_nal_t) * (h->out.i_nals_allocated*2) );
1641         if( !new_out )
1642             return -1;
1643         memcpy( new_out, h->out.nal, sizeof(x264_nal_t) * (h->out.i_nals_allocated) );
1644         x264_free( h->out.nal );
1645         h->out.nal = new_out;
1646         h->out.i_nals_allocated *= 2;
1647     }
1648     return 0;
1649 }
1650
1651 static int x264_nal_end( x264_t *h )
1652 {
1653     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1654     uint8_t *end = &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8];
1655     nal->i_payload = end - nal->p_payload;
1656     /* Assembly implementation of nal_escape reads past the end of the input.
1657      * While undefined padding wouldn't actually affect the output, it makes valgrind unhappy. */
1658     memset( end, 0xff, 64 );
1659     if( h->param.nalu_process )
1660         h->param.nalu_process( h, nal, h->fenc->opaque );
1661     h->out.i_nal++;
1662
1663     return x264_nal_check_buffer( h );
1664 }
1665
1666 static int x264_encoder_encapsulate_nals( x264_t *h, int start )
1667 {
1668     x264_t *h0 = h->thread[0];
1669     int nal_size = 0, previous_nal_size = 0;
1670
1671     if( h->param.nalu_process )
1672     {
1673         for( int i = start; i < h->out.i_nal; i++ )
1674             nal_size += h->out.nal[i].i_payload;
1675         return nal_size;
1676     }
1677
1678     for( int i = 0; i < start; i++ )
1679         previous_nal_size += h->out.nal[i].i_payload;
1680
1681     for( int i = start; i < h->out.i_nal; i++ )
1682         nal_size += h->out.nal[i].i_payload;
1683
1684     /* Worst-case NAL unit escaping: reallocate the buffer if it's too small. */
1685     int necessary_size = previous_nal_size + nal_size * 3/2 + h->out.i_nal * 4 + 4 + 64;
1686     if( h0->nal_buffer_size < necessary_size )
1687     {
1688         necessary_size *= 2;
1689         uint8_t *buf = x264_malloc( necessary_size );
1690         if( !buf )
1691             return -1;
1692         if( previous_nal_size )
1693             memcpy( buf, h0->nal_buffer, previous_nal_size );
1694
1695         intptr_t delta = buf - h0->nal_buffer;
1696         for( int i = 0; i < start; i++ )
1697             h->out.nal[i].p_payload += delta;
1698
1699         x264_free( h0->nal_buffer );
1700         h0->nal_buffer = buf;
1701         h0->nal_buffer_size = necessary_size;
1702     }
1703
1704     uint8_t *nal_buffer = h0->nal_buffer + previous_nal_size;
1705
1706     for( int i = start; i < h->out.i_nal; i++ )
1707     {
1708         h->out.nal[i].b_long_startcode = !i || h->out.nal[i].i_type == NAL_SPS || h->out.nal[i].i_type == NAL_PPS;
1709         x264_nal_encode( h, nal_buffer, &h->out.nal[i] );
1710         nal_buffer += h->out.nal[i].i_payload;
1711     }
1712
1713     x264_emms();
1714
1715     return nal_buffer - (h0->nal_buffer + previous_nal_size);
1716 }
1717
1718 /****************************************************************************
1719  * x264_encoder_headers:
1720  ****************************************************************************/
1721 int x264_encoder_headers( x264_t *h, x264_nal_t **pp_nal, int *pi_nal )
1722 {
1723     int frame_size = 0;
1724     /* init bitstream context */
1725     h->out.i_nal = 0;
1726     bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
1727
1728     /* Write SEI, SPS and PPS. */
1729
1730     /* generate sequence parameters */
1731     x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
1732     x264_sps_write( &h->out.bs, h->sps );
1733     if( x264_nal_end( h ) )
1734         return -1;
1735
1736     /* generate picture parameters */
1737     x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
1738     x264_pps_write( &h->out.bs, h->sps, h->pps );
1739     if( x264_nal_end( h ) )
1740         return -1;
1741
1742     /* identify ourselves */
1743     x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
1744     if( x264_sei_version_write( h, &h->out.bs ) )
1745         return -1;
1746     if( x264_nal_end( h ) )
1747         return -1;
1748
1749     frame_size = x264_encoder_encapsulate_nals( h, 0 );
1750     if( frame_size < 0 )
1751         return -1;
1752
1753     /* now set output*/
1754     *pi_nal = h->out.i_nal;
1755     *pp_nal = &h->out.nal[0];
1756     h->out.i_nal = 0;
1757
1758     return frame_size;
1759 }
1760
1761 /* Check to see whether we have chosen a reference list ordering different
1762  * from the standard's default. */
1763 static inline void x264_reference_check_reorder( x264_t *h )
1764 {
1765     /* The reorder check doesn't check for missing frames, so just
1766      * force a reorder if one of the reference list is corrupt. */
1767     for( int i = 0; h->frames.reference[i]; i++ )
1768         if( h->frames.reference[i]->b_corrupt )
1769         {
1770             h->b_ref_reorder[0] = 1;
1771             return;
1772         }
1773     for( int list = 0; list <= (h->sh.i_type == SLICE_TYPE_B); list++ )
1774         for( int i = 0; i < h->i_ref[list] - 1; i++ )
1775         {
1776             int framenum_diff = h->fref[list][i+1]->i_frame_num - h->fref[list][i]->i_frame_num;
1777             int poc_diff = h->fref[list][i+1]->i_poc - h->fref[list][i]->i_poc;
1778             /* P and B-frames use different default orders. */
1779             if( h->sh.i_type == SLICE_TYPE_P ? framenum_diff > 0 : list == 1 ? poc_diff < 0 : poc_diff > 0 )
1780             {
1781                 h->b_ref_reorder[list] = 1;
1782                 return;
1783             }
1784         }
1785 }
1786
1787 /* return -1 on failure, else return the index of the new reference frame */
1788 int x264_weighted_reference_duplicate( x264_t *h, int i_ref, const x264_weight_t *w )
1789 {
1790     int i = h->i_ref[0];
1791     int j = 1;
1792     x264_frame_t *newframe;
1793     if( i <= 1 ) /* empty list, definitely can't duplicate frame */
1794         return -1;
1795
1796     //Duplication is only used in X264_WEIGHTP_SMART
1797     if( h->param.analyse.i_weighted_pred != X264_WEIGHTP_SMART )
1798         return -1;
1799
1800     /* Duplication is a hack to compensate for crappy rounding in motion compensation.
1801      * With high bit depth, it's not worth doing, so turn it off except in the case of
1802      * unweighted dupes. */
1803     if( BIT_DEPTH > 8 && w != x264_weight_none )
1804         return -1;
1805
1806     newframe = x264_frame_pop_blank_unused( h );
1807     if( !newframe )
1808         return -1;
1809
1810     //FIXME: probably don't need to copy everything
1811     *newframe = *h->fref[0][i_ref];
1812     newframe->i_reference_count = 1;
1813     newframe->orig = h->fref[0][i_ref];
1814     newframe->b_duplicate = 1;
1815     memcpy( h->fenc->weight[j], w, sizeof(h->fenc->weight[i]) );
1816
1817     /* shift the frames to make space for the dupe. */
1818     h->b_ref_reorder[0] = 1;
1819     if( h->i_ref[0] < X264_REF_MAX )
1820         ++h->i_ref[0];
1821     h->fref[0][X264_REF_MAX-1] = NULL;
1822     x264_frame_unshift( &h->fref[0][j], newframe );
1823
1824     return j;
1825 }
1826
1827 static void x264_weighted_pred_init( x264_t *h )
1828 {
1829     /* for now no analysis and set all weights to nothing */
1830     for( int i_ref = 0; i_ref < h->i_ref[0]; i_ref++ )
1831         h->fenc->weighted[i_ref] = h->fref[0][i_ref]->filtered[0][0];
1832
1833     // FIXME: This only supports weighting of one reference frame
1834     // and duplicates of that frame.
1835     h->fenc->i_lines_weighted = 0;
1836
1837     for( int i_ref = 0; i_ref < (h->i_ref[0] << SLICE_MBAFF); i_ref++ )
1838         for( int i = 0; i < 3; i++ )
1839             h->sh.weight[i_ref][i].weightfn = NULL;
1840
1841
1842     if( h->sh.i_type != SLICE_TYPE_P || h->param.analyse.i_weighted_pred <= 0 )
1843         return;
1844
1845     int i_padv = PADV << PARAM_INTERLACED;
1846     int denom = -1;
1847     int weightplane[2] = { 0, 0 };
1848     int buffer_next = 0;
1849     for( int i = 0; i < 3; i++ )
1850     {
1851         for( int j = 0; j < h->i_ref[0]; j++ )
1852         {
1853             if( h->fenc->weight[j][i].weightfn )
1854             {
1855                 h->sh.weight[j][i] = h->fenc->weight[j][i];
1856                 // if weight is useless, don't write it to stream
1857                 if( h->sh.weight[j][i].i_scale == 1<<h->sh.weight[j][i].i_denom && h->sh.weight[j][i].i_offset == 0 )
1858                     h->sh.weight[j][i].weightfn = NULL;
1859                 else
1860                 {
1861                     if( !weightplane[!!i] )
1862                     {
1863                         weightplane[!!i] = 1;
1864                         h->sh.weight[0][!!i].i_denom = denom = h->sh.weight[j][i].i_denom;
1865                         assert( x264_clip3( denom, 0, 7 ) == denom );
1866                     }
1867
1868                     assert( h->sh.weight[j][i].i_denom == denom );
1869                     if( !i )
1870                     {
1871                         h->fenc->weighted[j] = h->mb.p_weight_buf[buffer_next++] + h->fenc->i_stride[0] * i_padv + PADH;
1872                         //scale full resolution frame
1873                         if( h->param.i_threads == 1 )
1874                         {
1875                             pixel *src = h->fref[0][j]->filtered[0][0] - h->fref[0][j]->i_stride[0]*i_padv - PADH;
1876                             pixel *dst = h->fenc->weighted[j] - h->fenc->i_stride[0]*i_padv - PADH;
1877                             int stride = h->fenc->i_stride[0];
1878                             int width = h->fenc->i_width[0] + PADH*2;
1879                             int height = h->fenc->i_lines[0] + i_padv*2;
1880                             x264_weight_scale_plane( h, dst, stride, src, stride, width, height, &h->sh.weight[j][0] );
1881                             h->fenc->i_lines_weighted = height;
1882                         }
1883                     }
1884                 }
1885             }
1886         }
1887     }
1888
1889     if( weightplane[1] )
1890         for( int i = 0; i < h->i_ref[0]; i++ )
1891         {
1892             if( h->sh.weight[i][1].weightfn && !h->sh.weight[i][2].weightfn )
1893             {
1894                 h->sh.weight[i][2].i_scale = 1 << h->sh.weight[0][1].i_denom;
1895                 h->sh.weight[i][2].i_offset = 0;
1896             }
1897             else if( h->sh.weight[i][2].weightfn && !h->sh.weight[i][1].weightfn )
1898             {
1899                 h->sh.weight[i][1].i_scale = 1 << h->sh.weight[0][1].i_denom;
1900                 h->sh.weight[i][1].i_offset = 0;
1901             }
1902         }
1903
1904     if( !weightplane[0] )
1905         h->sh.weight[0][0].i_denom = 0;
1906     if( !weightplane[1] )
1907         h->sh.weight[0][1].i_denom = 0;
1908     h->sh.weight[0][2].i_denom = h->sh.weight[0][1].i_denom;
1909 }
1910
1911 static inline int x264_reference_distance( x264_t *h, x264_frame_t *frame )
1912 {
1913     if( h->param.i_frame_packing == 5 )
1914         return abs((h->fenc->i_frame&~1) - (frame->i_frame&~1)) +
1915                   ((h->fenc->i_frame&1) != (frame->i_frame&1));
1916     else
1917         return abs(h->fenc->i_frame - frame->i_frame);
1918 }
1919
1920 static inline void x264_reference_build_list( x264_t *h, int i_poc )
1921 {
1922     int b_ok;
1923
1924     /* build ref list 0/1 */
1925     h->mb.pic.i_fref[0] = h->i_ref[0] = 0;
1926     h->mb.pic.i_fref[1] = h->i_ref[1] = 0;
1927     if( h->sh.i_type == SLICE_TYPE_I )
1928         return;
1929
1930     for( int i = 0; h->frames.reference[i]; i++ )
1931     {
1932         if( h->frames.reference[i]->b_corrupt )
1933             continue;
1934         if( h->frames.reference[i]->i_poc < i_poc )
1935             h->fref[0][h->i_ref[0]++] = h->frames.reference[i];
1936         else if( h->frames.reference[i]->i_poc > i_poc )
1937             h->fref[1][h->i_ref[1]++] = h->frames.reference[i];
1938     }
1939
1940     /* Order reference lists by distance from the current frame. */
1941     for( int list = 0; list < 2; list++ )
1942     {
1943         h->fref_nearest[list] = h->fref[list][0];
1944         do
1945         {
1946             b_ok = 1;
1947             for( int i = 0; i < h->i_ref[list] - 1; i++ )
1948             {
1949                 if( list ? h->fref[list][i+1]->i_poc < h->fref_nearest[list]->i_poc
1950                          : h->fref[list][i+1]->i_poc > h->fref_nearest[list]->i_poc )
1951                     h->fref_nearest[list] = h->fref[list][i+1];
1952                 if( x264_reference_distance( h, h->fref[list][i] ) > x264_reference_distance( h, h->fref[list][i+1] ) )
1953                 {
1954                     XCHG( x264_frame_t*, h->fref[list][i], h->fref[list][i+1] );
1955                     b_ok = 0;
1956                     break;
1957                 }
1958             }
1959         } while( !b_ok );
1960     }
1961
1962     if( h->sh.i_mmco_remove_from_end )
1963         for( int i = h->i_ref[0]-1; i >= h->i_ref[0] - h->sh.i_mmco_remove_from_end; i-- )
1964         {
1965             int diff = h->i_frame_num - h->fref[0][i]->i_frame_num;
1966             h->sh.mmco[h->sh.i_mmco_command_count].i_poc = h->fref[0][i]->i_poc;
1967             h->sh.mmco[h->sh.i_mmco_command_count++].i_difference_of_pic_nums = diff;
1968         }
1969
1970     x264_reference_check_reorder( h );
1971
1972     h->i_ref[1] = X264_MIN( h->i_ref[1], h->frames.i_max_ref1 );
1973     h->i_ref[0] = X264_MIN( h->i_ref[0], h->frames.i_max_ref0 );
1974     h->i_ref[0] = X264_MIN( h->i_ref[0], h->param.i_frame_reference ); // if reconfig() has lowered the limit
1975
1976     /* For Blu-ray compliance, don't reference frames outside of the minigop. */
1977     if( IS_X264_TYPE_B( h->fenc->i_type ) && h->param.b_bluray_compat )
1978         h->i_ref[0] = X264_MIN( h->i_ref[0], IS_X264_TYPE_B( h->fref[0][0]->i_type ) + 1 );
1979
1980     /* add duplicates */
1981     if( h->fenc->i_type == X264_TYPE_P )
1982     {
1983         int idx = -1;
1984         if( h->param.analyse.i_weighted_pred >= X264_WEIGHTP_SIMPLE )
1985         {
1986             x264_weight_t w[3];
1987             w[1].weightfn = w[2].weightfn = NULL;
1988             if( h->param.rc.b_stat_read )
1989                 x264_ratecontrol_set_weights( h, h->fenc );
1990
1991             if( !h->fenc->weight[0][0].weightfn )
1992             {
1993                 h->fenc->weight[0][0].i_denom = 0;
1994                 SET_WEIGHT( w[0], 1, 1, 0, -1 );
1995                 idx = x264_weighted_reference_duplicate( h, 0, w );
1996             }
1997             else
1998             {
1999                 if( h->fenc->weight[0][0].i_scale == 1<<h->fenc->weight[0][0].i_denom )
2000                 {
2001                     SET_WEIGHT( h->fenc->weight[0][0], 1, 1, 0, h->fenc->weight[0][0].i_offset );
2002                 }
2003                 x264_weighted_reference_duplicate( h, 0, x264_weight_none );
2004                 if( h->fenc->weight[0][0].i_offset > -128 )
2005                 {
2006                     w[0] = h->fenc->weight[0][0];
2007                     w[0].i_offset--;
2008                     h->mc.weight_cache( h, &w[0] );
2009                     idx = x264_weighted_reference_duplicate( h, 0, w );
2010                 }
2011             }
2012         }
2013         h->mb.ref_blind_dupe = idx;
2014     }
2015
2016     assert( h->i_ref[0] + h->i_ref[1] <= X264_REF_MAX );
2017     h->mb.pic.i_fref[0] = h->i_ref[0];
2018     h->mb.pic.i_fref[1] = h->i_ref[1];
2019 }
2020
2021 static void x264_fdec_filter_row( x264_t *h, int mb_y, int pass )
2022 {
2023     /* mb_y is the mb to be encoded next, not the mb to be filtered here */
2024     int b_hpel = h->fdec->b_kept_as_ref;
2025     int b_deblock = h->sh.i_disable_deblocking_filter_idc != 1;
2026     int b_end = mb_y == h->i_threadslice_end;
2027     int b_measure_quality = 1;
2028     int min_y = mb_y - (1 << SLICE_MBAFF);
2029     int b_start = min_y == h->i_threadslice_start;
2030     /* Even in interlaced mode, deblocking never modifies more than 4 pixels
2031      * above each MB, as bS=4 doesn't happen for the top of interlaced mbpairs. */
2032     int minpix_y = min_y*16 - 4 * !b_start;
2033     int maxpix_y = mb_y*16 - 4 * !b_end;
2034     b_deblock &= b_hpel || h->param.b_full_recon || h->param.psz_dump_yuv;
2035     if( h->param.b_sliced_threads )
2036     {
2037         switch( pass )
2038         {
2039             /* During encode: only do deblock if asked for */
2040             default:
2041             case 0:
2042                 b_deblock &= h->param.b_full_recon;
2043                 b_hpel = 0;
2044                 break;
2045             /* During post-encode pass: do deblock if not done yet, do hpel for all
2046              * rows except those between slices. */
2047             case 1:
2048                 b_deblock &= !h->param.b_full_recon;
2049                 b_hpel &= !(b_start && min_y > 0);
2050                 b_measure_quality = 0;
2051                 break;
2052             /* Final pass: do the rows between slices in sequence. */
2053             case 2:
2054                 b_deblock = 0;
2055                 b_measure_quality = 0;
2056                 break;
2057         }
2058     }
2059     if( mb_y & SLICE_MBAFF )
2060         return;
2061     if( min_y < h->i_threadslice_start )
2062         return;
2063
2064     if( b_deblock )
2065         for( int y = min_y; y < mb_y; y += (1 << SLICE_MBAFF) )
2066             x264_frame_deblock_row( h, y );
2067
2068     /* FIXME: Prediction requires different borders for interlaced/progressive mc,
2069      * but the actual image data is equivalent. For now, maintain this
2070      * consistency by copying deblocked pixels between planes. */
2071     if( PARAM_INTERLACED && (!h->param.b_sliced_threads || pass == 1) )
2072         for( int p = 0; p < h->fdec->i_plane; p++ )
2073             for( int i = minpix_y>>(CHROMA_V_SHIFT && p); i < maxpix_y>>(CHROMA_V_SHIFT && p); i++ )
2074                 memcpy( h->fdec->plane_fld[p] + i*h->fdec->i_stride[p],
2075                         h->fdec->plane[p]     + i*h->fdec->i_stride[p],
2076                         h->mb.i_mb_width*16*sizeof(pixel) );
2077
2078     if( h->fdec->b_kept_as_ref && (!h->param.b_sliced_threads || pass == 1) )
2079         x264_frame_expand_border( h, h->fdec, min_y );
2080     if( b_hpel )
2081     {
2082         int end = mb_y == h->mb.i_mb_height;
2083         /* Can't do hpel until the previous slice is done encoding. */
2084         if( h->param.analyse.i_subpel_refine )
2085         {
2086             x264_frame_filter( h, h->fdec, min_y, end );
2087             x264_frame_expand_border_filtered( h, h->fdec, min_y, end );
2088         }
2089     }
2090
2091     if( SLICE_MBAFF && pass == 0 )
2092         for( int i = 0; i < 3; i++ )
2093         {
2094             XCHG( pixel *, h->intra_border_backup[0][i], h->intra_border_backup[3][i] );
2095             XCHG( pixel *, h->intra_border_backup[1][i], h->intra_border_backup[4][i] );
2096         }
2097
2098     if( h->i_thread_frames > 1 && h->fdec->b_kept_as_ref )
2099         x264_frame_cond_broadcast( h->fdec, mb_y*16 + (b_end ? 10000 : -(X264_THREAD_HEIGHT << SLICE_MBAFF)) );
2100
2101     if( b_measure_quality )
2102     {
2103         maxpix_y = X264_MIN( maxpix_y, h->param.i_height );
2104         if( h->param.analyse.b_psnr )
2105         {
2106             for( int p = 0; p < (CHROMA444 ? 3 : 1); p++ )
2107                 h->stat.frame.i_ssd[p] += x264_pixel_ssd_wxh( &h->pixf,
2108                     h->fdec->plane[p] + minpix_y * h->fdec->i_stride[p], h->fdec->i_stride[p],
2109                     h->fenc->plane[p] + minpix_y * h->fenc->i_stride[p], h->fenc->i_stride[p],
2110                     h->param.i_width, maxpix_y-minpix_y );
2111             if( !CHROMA444 )
2112             {
2113                 uint64_t ssd_u, ssd_v;
2114                 int v_shift = CHROMA_V_SHIFT;
2115                 x264_pixel_ssd_nv12( &h->pixf,
2116                     h->fdec->plane[1] + (minpix_y>>v_shift) * h->fdec->i_stride[1], h->fdec->i_stride[1],
2117                     h->fenc->plane[1] + (minpix_y>>v_shift) * h->fenc->i_stride[1], h->fenc->i_stride[1],
2118                     h->param.i_width>>1, (maxpix_y-minpix_y)>>v_shift, &ssd_u, &ssd_v );
2119                 h->stat.frame.i_ssd[1] += ssd_u;
2120                 h->stat.frame.i_ssd[2] += ssd_v;
2121             }
2122         }
2123
2124         if( h->param.analyse.b_ssim )
2125         {
2126             int ssim_cnt;
2127             x264_emms();
2128             /* offset by 2 pixels to avoid alignment of ssim blocks with dct blocks,
2129              * and overlap by 4 */
2130             minpix_y += b_start ? 2 : -6;
2131             h->stat.frame.f_ssim +=
2132                 x264_pixel_ssim_wxh( &h->pixf,
2133                     h->fdec->plane[0] + 2+minpix_y*h->fdec->i_stride[0], h->fdec->i_stride[0],
2134                     h->fenc->plane[0] + 2+minpix_y*h->fenc->i_stride[0], h->fenc->i_stride[0],
2135                     h->param.i_width-2, maxpix_y-minpix_y, h->scratch_buffer, &ssim_cnt );
2136             h->stat.frame.i_ssim_cnt += ssim_cnt;
2137         }
2138     }
2139 }
2140
2141 static inline int x264_reference_update( x264_t *h )
2142 {
2143     if( !h->fdec->b_kept_as_ref )
2144     {
2145         if( h->i_thread_frames > 1 )
2146         {
2147             x264_frame_push_unused( h, h->fdec );
2148             h->fdec = x264_frame_pop_unused( h, 1 );
2149             if( !h->fdec )
2150                 return -1;
2151         }
2152         return 0;
2153     }
2154
2155     /* apply mmco from previous frame. */
2156     for( int i = 0; i < h->sh.i_mmco_command_count; i++ )
2157         for( int j = 0; h->frames.reference[j]; j++ )
2158             if( h->frames.reference[j]->i_poc == h->sh.mmco[i].i_poc )
2159                 x264_frame_push_unused( h, x264_frame_shift( &h->frames.reference[j] ) );
2160
2161     /* move frame in the buffer */
2162     x264_frame_push( h->frames.reference, h->fdec );
2163     if( h->frames.reference[h->sps->i_num_ref_frames] )
2164         x264_frame_push_unused( h, x264_frame_shift( h->frames.reference ) );
2165     h->fdec = x264_frame_pop_unused( h, 1 );
2166     if( !h->fdec )
2167         return -1;
2168     return 0;
2169 }
2170
2171 static inline void x264_reference_reset( x264_t *h )
2172 {
2173     while( h->frames.reference[0] )
2174         x264_frame_push_unused( h, x264_frame_pop( h->frames.reference ) );
2175     h->fdec->i_poc =
2176     h->fenc->i_poc = 0;
2177 }
2178
2179 static inline void x264_reference_hierarchy_reset( x264_t *h )
2180 {
2181     int ref;
2182     int b_hasdelayframe = 0;
2183
2184     /* look for delay frames -- chain must only contain frames that are disposable */
2185     for( int i = 0; h->frames.current[i] && IS_DISPOSABLE( h->frames.current[i]->i_type ); i++ )
2186         b_hasdelayframe |= h->frames.current[i]->i_coded
2187                         != h->frames.current[i]->i_frame + h->sps->vui.i_num_reorder_frames;
2188
2189     /* This function must handle b-pyramid and clear frames for open-gop */
2190     if( h->param.i_bframe_pyramid != X264_B_PYRAMID_STRICT && !b_hasdelayframe && h->frames.i_poc_last_open_gop == -1 )
2191         return;
2192
2193     /* Remove last BREF. There will never be old BREFs in the
2194      * dpb during a BREF decode when pyramid == STRICT */
2195     for( ref = 0; h->frames.reference[ref]; ref++ )
2196     {
2197         if( ( h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT
2198             && h->frames.reference[ref]->i_type == X264_TYPE_BREF )
2199             || ( h->frames.reference[ref]->i_poc < h->frames.i_poc_last_open_gop
2200             && h->sh.i_type != SLICE_TYPE_B ) )
2201         {
2202             int diff = h->i_frame_num - h->frames.reference[ref]->i_frame_num;
2203             h->sh.mmco[h->sh.i_mmco_command_count].i_difference_of_pic_nums = diff;
2204             h->sh.mmco[h->sh.i_mmco_command_count++].i_poc = h->frames.reference[ref]->i_poc;
2205             x264_frame_push_unused( h, x264_frame_shift( &h->frames.reference[ref] ) );
2206             h->b_ref_reorder[0] = 1;
2207             ref--;
2208         }
2209     }
2210
2211     /* Prepare room in the dpb for the delayed display time of the later b-frame's */
2212     if( h->param.i_bframe_pyramid )
2213         h->sh.i_mmco_remove_from_end = X264_MAX( ref + 2 - h->frames.i_max_dpb, 0 );
2214 }
2215
2216 static inline void x264_slice_init( x264_t *h, int i_nal_type, int i_global_qp )
2217 {
2218     /* ------------------------ Create slice header  ----------------------- */
2219     if( i_nal_type == NAL_SLICE_IDR )
2220     {
2221         x264_slice_header_init( h, &h->sh, h->sps, h->pps, h->i_idr_pic_id, h->i_frame_num, i_global_qp );
2222
2223         /* alternate id */
2224         h->i_idr_pic_id ^= 1;
2225     }
2226     else
2227     {
2228         x264_slice_header_init( h, &h->sh, h->sps, h->pps, -1, h->i_frame_num, i_global_qp );
2229
2230         h->sh.i_num_ref_idx_l0_active = h->i_ref[0] <= 0 ? 1 : h->i_ref[0];
2231         h->sh.i_num_ref_idx_l1_active = h->i_ref[1] <= 0 ? 1 : h->i_ref[1];
2232         if( h->sh.i_num_ref_idx_l0_active != h->pps->i_num_ref_idx_l0_default_active ||
2233             (h->sh.i_type == SLICE_TYPE_B && h->sh.i_num_ref_idx_l1_active != h->pps->i_num_ref_idx_l1_default_active) )
2234         {
2235             h->sh.b_num_ref_idx_override = 1;
2236         }
2237     }
2238
2239     if( h->fenc->i_type == X264_TYPE_BREF && h->param.b_bluray_compat && h->sh.i_mmco_command_count )
2240     {
2241         h->b_sh_backup = 1;
2242         h->sh_backup = h->sh;
2243     }
2244
2245     h->fdec->i_frame_num = h->sh.i_frame_num;
2246
2247     if( h->sps->i_poc_type == 0 )
2248     {
2249         h->sh.i_poc = h->fdec->i_poc;
2250         if( PARAM_INTERLACED )
2251         {
2252             h->sh.i_delta_poc_bottom = h->param.b_tff ? 1 : -1;
2253             h->sh.i_poc += h->sh.i_delta_poc_bottom == -1;
2254         }
2255         else
2256             h->sh.i_delta_poc_bottom = 0;
2257         h->fdec->i_delta_poc[0] = h->sh.i_delta_poc_bottom == -1;
2258         h->fdec->i_delta_poc[1] = h->sh.i_delta_poc_bottom ==  1;
2259     }
2260     else
2261     {
2262         /* Nothing to do ? */
2263     }
2264
2265     x264_macroblock_slice_init( h );
2266 }
2267
2268 typedef struct
2269 {
2270     int skip;
2271     uint8_t cabac_prevbyte;
2272     bs_t bs;
2273     x264_cabac_t cabac;
2274     x264_frame_stat_t stat;
2275     int last_qp;
2276     int last_dqp;
2277     int field_decoding_flag;
2278 } x264_bs_bak_t;
2279
2280 static ALWAYS_INLINE void x264_bitstream_backup( x264_t *h, x264_bs_bak_t *bak, int i_skip, int full )
2281 {
2282     if( full )
2283     {
2284         bak->stat = h->stat.frame;
2285         bak->last_qp = h->mb.i_last_qp;
2286         bak->last_dqp = h->mb.i_last_dqp;
2287         bak->field_decoding_flag = h->mb.field_decoding_flag;
2288     }
2289     else
2290     {
2291         bak->stat.i_mv_bits = h->stat.frame.i_mv_bits;
2292         bak->stat.i_tex_bits = h->stat.frame.i_tex_bits;
2293     }
2294     /* In the per-MB backup, we don't need the contexts because flushing the CABAC
2295      * encoder has no context dependency and in this case, a slice is ended (and
2296      * thus the content of all contexts are thrown away). */
2297     if( h->param.b_cabac )
2298     {
2299         if( full )
2300             memcpy( &bak->cabac, &h->cabac, sizeof(x264_cabac_t) );
2301         else
2302             memcpy( &bak->cabac, &h->cabac, offsetof(x264_cabac_t, f8_bits_encoded) );
2303         /* x264's CABAC writer modifies the previous byte during carry, so it has to be
2304          * backed up. */
2305         bak->cabac_prevbyte = h->cabac.p[-1];
2306     }
2307     else
2308     {
2309         bak->bs = h->out.bs;
2310         bak->skip = i_skip;
2311     }
2312 }
2313
2314 static ALWAYS_INLINE void x264_bitstream_restore( x264_t *h, x264_bs_bak_t *bak, int *skip, int full )
2315 {
2316     if( full )
2317     {
2318         h->stat.frame = bak->stat;
2319         h->mb.i_last_qp = bak->last_qp;
2320         h->mb.i_last_dqp = bak->last_dqp;
2321         h->mb.field_decoding_flag = bak->field_decoding_flag;
2322     }
2323     else
2324     {
2325         h->stat.frame.i_mv_bits = bak->stat.i_mv_bits;
2326         h->stat.frame.i_tex_bits = bak->stat.i_tex_bits;
2327     }
2328     if( h->param.b_cabac )
2329     {
2330         if( full )
2331             memcpy( &h->cabac, &bak->cabac, sizeof(x264_cabac_t) );
2332         else
2333             memcpy( &h->cabac, &bak->cabac, offsetof(x264_cabac_t, f8_bits_encoded) );
2334         h->cabac.p[-1] = bak->cabac_prevbyte;
2335     }
2336     else
2337     {
2338         h->out.bs = bak->bs;
2339         *skip = bak->skip;
2340     }
2341 }
2342
2343 static int x264_slice_write( x264_t *h )
2344 {
2345     int i_skip;
2346     int mb_xy, i_mb_x, i_mb_y;
2347     /* NALUs other than the first use a 3-byte startcode.
2348      * Add one extra byte for the rbsp, and one more for the final CABAC putbyte.
2349      * Then add an extra 5 bytes just in case, to account for random NAL escapes and
2350      * other inaccuracies. */
2351     int overhead_guess = (NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal)) + 1 + h->param.b_cabac + 5;
2352     int slice_max_size = h->param.i_slice_max_size > 0 ? (h->param.i_slice_max_size-overhead_guess)*8 : 0;
2353     int back_up_bitstream = slice_max_size || (!h->param.b_cabac && h->sps->i_profile_idc < PROFILE_HIGH);
2354     int starting_bits = bs_pos(&h->out.bs);
2355     int b_deblock = h->sh.i_disable_deblocking_filter_idc != 1;
2356     int b_hpel = h->fdec->b_kept_as_ref;
2357     int orig_last_mb = h->sh.i_last_mb;
2358     int thread_last_mb = h->i_threadslice_end * h->mb.i_mb_width - 1;
2359     uint8_t *last_emu_check;
2360 #define BS_BAK_SLICE_MAX_SIZE 0
2361 #define BS_BAK_SLICE_MIN_MBS  1
2362 #define BS_BAK_ROW_VBV        2
2363     x264_bs_bak_t bs_bak[3];
2364     b_deblock &= b_hpel || h->param.b_full_recon || h->param.psz_dump_yuv;
2365     bs_realign( &h->out.bs );
2366
2367     /* Slice */
2368     x264_nal_start( h, h->i_nal_type, h->i_nal_ref_idc );
2369     h->out.nal[h->out.i_nal].i_first_mb = h->sh.i_first_mb;
2370
2371     /* Slice header */
2372     x264_macroblock_thread_init( h );
2373
2374     /* Set the QP equal to the first QP in the slice for more accurate CABAC initialization. */
2375     h->mb.i_mb_xy = h->sh.i_first_mb;
2376     h->sh.i_qp = x264_ratecontrol_mb_qp( h );
2377     h->sh.i_qp = SPEC_QP( h->sh.i_qp );
2378     h->sh.i_qp_delta = h->sh.i_qp - h->pps->i_pic_init_qp;
2379
2380     x264_slice_header_write( &h->out.bs, &h->sh, h->i_nal_ref_idc );
2381     if( h->param.b_cabac )
2382     {
2383         /* alignment needed */
2384         bs_align_1( &h->out.bs );
2385
2386         /* init cabac */
2387         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 );
2388         x264_cabac_encode_init ( &h->cabac, h->out.bs.p, h->out.bs.p_end );
2389         last_emu_check = h->cabac.p;
2390     }
2391     else
2392         last_emu_check = h->out.bs.p;
2393     h->mb.i_last_qp = h->sh.i_qp;
2394     h->mb.i_last_dqp = 0;
2395     h->mb.field_decoding_flag = 0;
2396
2397     i_mb_y = h->sh.i_first_mb / h->mb.i_mb_width;
2398     i_mb_x = h->sh.i_first_mb % h->mb.i_mb_width;
2399     i_skip = 0;
2400
2401     while( 1 )
2402     {
2403         mb_xy = i_mb_x + i_mb_y * h->mb.i_mb_width;
2404         int mb_spos = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
2405
2406         if( i_mb_x == 0 )
2407         {
2408             if( x264_bitstream_check_buffer( h ) )
2409                 return -1;
2410             if( !(i_mb_y & SLICE_MBAFF) && h->param.rc.i_vbv_buffer_size )
2411                 x264_bitstream_backup( h, &bs_bak[BS_BAK_ROW_VBV], i_skip, 1 );
2412             if( !h->mb.b_reencode_mb )
2413                 x264_fdec_filter_row( h, i_mb_y, 0 );
2414         }
2415
2416         if( !(i_mb_y & SLICE_MBAFF) && back_up_bitstream )
2417         {
2418             x264_bitstream_backup( h, &bs_bak[BS_BAK_SLICE_MAX_SIZE], i_skip, 0 );
2419             if( slice_max_size && (thread_last_mb+1-mb_xy) == h->param.i_slice_min_mbs )
2420                 x264_bitstream_backup( h, &bs_bak[BS_BAK_SLICE_MIN_MBS], i_skip, 0 );
2421         }
2422
2423         if( PARAM_INTERLACED )
2424         {
2425             if( h->mb.b_adaptive_mbaff )
2426             {
2427                 if( !(i_mb_y&1) )
2428                 {
2429                     /* FIXME: VSAD is fast but fairly poor at choosing the best interlace type. */
2430                     h->mb.b_interlaced = x264_field_vsad( h, i_mb_x, i_mb_y );
2431                     memcpy( &h->zigzagf, MB_INTERLACED ? &h->zigzagf_interlaced : &h->zigzagf_progressive, sizeof(h->zigzagf) );
2432                     if( !MB_INTERLACED && (i_mb_y+2) == h->mb.i_mb_height )
2433                         x264_expand_border_mbpair( h, i_mb_x, i_mb_y );
2434                 }
2435             }
2436             h->mb.field[mb_xy] = MB_INTERLACED;
2437         }
2438
2439         /* load cache */
2440         if( SLICE_MBAFF )
2441             x264_macroblock_cache_load_interlaced( h, i_mb_x, i_mb_y );
2442         else
2443             x264_macroblock_cache_load_progressive( h, i_mb_x, i_mb_y );
2444
2445         x264_macroblock_analyse( h );
2446
2447         /* encode this macroblock -> be careful it can change the mb type to P_SKIP if needed */
2448 reencode:
2449         x264_macroblock_encode( h );
2450
2451         if( h->param.b_cabac )
2452         {
2453             if( mb_xy > h->sh.i_first_mb && !(SLICE_MBAFF && (i_mb_y&1)) )
2454                 x264_cabac_encode_terminal( &h->cabac );
2455
2456             if( IS_SKIP( h->mb.i_type ) )
2457                 x264_cabac_mb_skip( h, 1 );
2458             else
2459             {
2460                 if( h->sh.i_type != SLICE_TYPE_I )
2461                     x264_cabac_mb_skip( h, 0 );
2462                 x264_macroblock_write_cabac( h, &h->cabac );
2463             }
2464         }
2465         else
2466         {
2467             if( IS_SKIP( h->mb.i_type ) )
2468                 i_skip++;
2469             else
2470             {
2471                 if( h->sh.i_type != SLICE_TYPE_I )
2472                 {
2473                     bs_write_ue( &h->out.bs, i_skip );  /* skip run */
2474                     i_skip = 0;
2475                 }
2476                 x264_macroblock_write_cavlc( h );
2477                 /* If there was a CAVLC level code overflow, try again at a higher QP. */
2478                 if( h->mb.b_overflow )
2479                 {
2480                     h->mb.i_chroma_qp = h->chroma_qp_table[++h->mb.i_qp];
2481                     h->mb.i_skip_intra = 0;
2482                     h->mb.b_skip_mc = 0;
2483                     h->mb.b_overflow = 0;
2484                     x264_bitstream_restore( h, &bs_bak[BS_BAK_SLICE_MAX_SIZE], &i_skip, 0 );
2485                     goto reencode;
2486                 }
2487             }
2488         }
2489
2490         int total_bits = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
2491         int mb_size = total_bits - mb_spos;
2492
2493         if( slice_max_size && (!SLICE_MBAFF || (i_mb_y&1)) )
2494         {
2495             /* Count the skip run, just in case. */
2496             if( !h->param.b_cabac )
2497                 total_bits += bs_size_ue_big( i_skip );
2498             /* Check for escape bytes. */
2499             uint8_t *end = h->param.b_cabac ? h->cabac.p : h->out.bs.p;
2500             for( ; last_emu_check < end - 2; last_emu_check++ )
2501                 if( last_emu_check[0] == 0 && last_emu_check[1] == 0 && last_emu_check[2] <= 3 )
2502                 {
2503                     slice_max_size -= 8;
2504                     last_emu_check++;
2505                 }
2506             /* We'll just re-encode this last macroblock if we go over the max slice size. */
2507             if( total_bits - starting_bits > slice_max_size && !h->mb.b_reencode_mb )
2508             {
2509                 if( !x264_frame_new_slice( h, h->fdec ) )
2510                 {
2511                     /* Handle the most obnoxious slice-min-mbs edge case: we need to end the slice
2512                      * because it's gone over the maximum size, but doing so would violate slice-min-mbs.
2513                      * If possible, roll back to the last checkpoint and try again.
2514                      * We could try raising QP, but that would break in the case where a slice spans multiple
2515                      * rows, which the re-encoding infrastructure can't currently handle. */
2516                     if( mb_xy <= thread_last_mb && (thread_last_mb+1-mb_xy) < h->param.i_slice_min_mbs )
2517                     {
2518                         if( thread_last_mb-h->param.i_slice_min_mbs < h->sh.i_first_mb+h->param.i_slice_min_mbs )
2519                         {
2520                             x264_log( h, X264_LOG_WARNING, "slice-max-size violated (frame %d, cause: slice-min-mbs)\n", h->i_frame );
2521                             slice_max_size = 0;
2522                             goto cont;
2523                         }
2524                         x264_bitstream_restore( h, &bs_bak[BS_BAK_SLICE_MIN_MBS], &i_skip, 0 );
2525                         h->mb.b_reencode_mb = 1;
2526                         h->sh.i_last_mb = thread_last_mb-h->param.i_slice_min_mbs;
2527                         break;
2528                     }
2529                     if( mb_xy-SLICE_MBAFF*h->mb.i_mb_stride != h->sh.i_first_mb )
2530                     {
2531                         x264_bitstream_restore( h, &bs_bak[BS_BAK_SLICE_MAX_SIZE], &i_skip, 0 );
2532                         h->mb.b_reencode_mb = 1;
2533                         if( SLICE_MBAFF )
2534                         {
2535                             // set to bottom of previous mbpair
2536                             if( i_mb_x )
2537                                 h->sh.i_last_mb = mb_xy-1+h->mb.i_mb_stride*(!(i_mb_y&1));
2538                             else
2539                                 h->sh.i_last_mb = (i_mb_y-2+!(i_mb_y&1))*h->mb.i_mb_stride + h->mb.i_mb_width - 1;
2540                         }
2541                         else
2542                             h->sh.i_last_mb = mb_xy-1;
2543                         break;
2544                     }
2545                     else
2546                         h->sh.i_last_mb = mb_xy;
2547                 }
2548                 else
2549                     slice_max_size = 0;
2550             }
2551         }
2552 cont:
2553         h->mb.b_reencode_mb = 0;
2554
2555 #if HAVE_VISUALIZE
2556         if( h->param.b_visualize )
2557             x264_visualize_mb( h );
2558 #endif
2559
2560         /* save cache */
2561         x264_macroblock_cache_save( h );
2562
2563         if( x264_ratecontrol_mb( h, mb_size ) < 0 )
2564         {
2565             x264_bitstream_restore( h, &bs_bak[BS_BAK_ROW_VBV], &i_skip, 1 );
2566             h->mb.b_reencode_mb = 1;
2567             i_mb_x = 0;
2568             i_mb_y = i_mb_y - SLICE_MBAFF;
2569             h->mb.i_mb_prev_xy = i_mb_y * h->mb.i_mb_stride - 1;
2570             h->sh.i_last_mb = orig_last_mb;
2571             continue;
2572         }
2573
2574         /* accumulate mb stats */
2575         h->stat.frame.i_mb_count[h->mb.i_type]++;
2576
2577         int b_intra = IS_INTRA( h->mb.i_type );
2578         int b_skip = IS_SKIP( h->mb.i_type );
2579         if( h->param.i_log_level >= X264_LOG_INFO || h->param.rc.b_stat_write )
2580         {
2581             if( !b_intra && !b_skip && !IS_DIRECT( h->mb.i_type ) )
2582             {
2583                 if( h->mb.i_partition != D_8x8 )
2584                         h->stat.frame.i_mb_partition[h->mb.i_partition] += 4;
2585                     else
2586                         for( int i = 0; i < 4; i++ )
2587                             h->stat.frame.i_mb_partition[h->mb.i_sub_partition[i]] ++;
2588                 if( h->param.i_frame_reference > 1 )
2589                     for( int i_list = 0; i_list <= (h->sh.i_type == SLICE_TYPE_B); i_list++ )
2590                         for( int i = 0; i < 4; i++ )
2591                         {
2592                             int i_ref = h->mb.cache.ref[i_list][ x264_scan8[4*i] ];
2593                             if( i_ref >= 0 )
2594                                 h->stat.frame.i_mb_count_ref[i_list][i_ref] ++;
2595                         }
2596             }
2597         }
2598
2599         if( h->param.i_log_level >= X264_LOG_INFO )
2600         {
2601             if( h->mb.i_cbp_luma | h->mb.i_cbp_chroma )
2602             {
2603                 if( CHROMA444 )
2604                 {
2605                     for( int i = 0; i < 4; i++ )
2606                         if( h->mb.i_cbp_luma & (1 << i) )
2607                             for( int p = 0; p < 3; p++ )
2608                             {
2609                                 int s8 = i*4+p*16;
2610                                 int nnz8x8 = M16( &h->mb.cache.non_zero_count[x264_scan8[s8]+0] )
2611                                            | M16( &h->mb.cache.non_zero_count[x264_scan8[s8]+8] );
2612                                 h->stat.frame.i_mb_cbp[!b_intra + p*2] += !!nnz8x8;
2613                             }
2614                 }
2615                 else
2616                 {
2617                     int cbpsum = (h->mb.i_cbp_luma&1) + ((h->mb.i_cbp_luma>>1)&1)
2618                                + ((h->mb.i_cbp_luma>>2)&1) + (h->mb.i_cbp_luma>>3);
2619                     h->stat.frame.i_mb_cbp[!b_intra + 0] += cbpsum;
2620                     h->stat.frame.i_mb_cbp[!b_intra + 2] += !!h->mb.i_cbp_chroma;
2621                     h->stat.frame.i_mb_cbp[!b_intra + 4] += h->mb.i_cbp_chroma >> 1;
2622                 }
2623             }
2624             if( h->mb.i_cbp_luma && !b_intra )
2625             {
2626                 h->stat.frame.i_mb_count_8x8dct[0] ++;
2627                 h->stat.frame.i_mb_count_8x8dct[1] += h->mb.b_transform_8x8;
2628             }
2629             if( b_intra && h->mb.i_type != I_PCM )
2630             {
2631                 if( h->mb.i_type == I_16x16 )
2632                     h->stat.frame.i_mb_pred_mode[0][h->mb.i_intra16x16_pred_mode]++;
2633                 else if( h->mb.i_type == I_8x8 )
2634                     for( int i = 0; i < 16; i += 4 )
2635                         h->stat.frame.i_mb_pred_mode[1][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
2636                 else //if( h->mb.i_type == I_4x4 )
2637                     for( int i = 0; i < 16; i++ )
2638                         h->stat.frame.i_mb_pred_mode[2][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
2639                 h->stat.frame.i_mb_pred_mode[3][x264_mb_chroma_pred_mode_fix[h->mb.i_chroma_pred_mode]]++;
2640             }
2641             h->stat.frame.i_mb_field[b_intra?0:b_skip?2:1] += MB_INTERLACED;
2642         }
2643
2644         /* calculate deblock strength values (actual deblocking is done per-row along with hpel) */
2645         if( b_deblock )
2646             x264_macroblock_deblock_strength( h );
2647
2648         if( mb_xy == h->sh.i_last_mb )
2649             break;
2650
2651         if( SLICE_MBAFF )
2652         {
2653             i_mb_x += i_mb_y & 1;
2654             i_mb_y ^= i_mb_x < h->mb.i_mb_width;
2655         }
2656         else
2657             i_mb_x++;
2658         if( i_mb_x == h->mb.i_mb_width )
2659         {
2660             i_mb_y++;
2661             i_mb_x = 0;
2662         }
2663     }
2664     if( h->sh.i_last_mb < h->sh.i_first_mb )
2665         return 0;
2666
2667     h->out.nal[h->out.i_nal].i_last_mb = h->sh.i_last_mb;
2668
2669     if( h->param.b_cabac )
2670     {
2671         x264_cabac_encode_flush( h, &h->cabac );
2672         h->out.bs.p = h->cabac.p;
2673     }
2674     else
2675     {
2676         if( i_skip > 0 )
2677             bs_write_ue( &h->out.bs, i_skip );  /* last skip run */
2678         /* rbsp_slice_trailing_bits */
2679         bs_rbsp_trailing( &h->out.bs );
2680         bs_flush( &h->out.bs );
2681     }
2682     if( x264_nal_end( h ) )
2683         return -1;
2684
2685     if( h->sh.i_last_mb == (h->i_threadslice_end * h->mb.i_mb_width - 1) )
2686     {
2687         h->stat.frame.i_misc_bits = bs_pos( &h->out.bs )
2688                                   + (h->out.i_nal*NALU_OVERHEAD * 8)
2689                                   - h->stat.frame.i_tex_bits
2690                                   - h->stat.frame.i_mv_bits;
2691         x264_fdec_filter_row( h, h->i_threadslice_end, 0 );
2692
2693         if( h->param.b_sliced_threads )
2694         {
2695             /* Tell the main thread we're done. */
2696             x264_threadslice_cond_broadcast( h, 1 );
2697             /* Do hpel now */
2698             for( int mb_y = h->i_threadslice_start; mb_y <= h->i_threadslice_end; mb_y++ )
2699                 x264_fdec_filter_row( h, mb_y, 1 );
2700             x264_threadslice_cond_broadcast( h, 2 );
2701             /* Do the first row of hpel, now that the previous slice is done */
2702             if( h->i_thread_idx > 0 )
2703             {
2704                 x264_threadslice_cond_wait( h->thread[h->i_thread_idx-1], 2 );
2705                 x264_fdec_filter_row( h, h->i_threadslice_start + (1 << SLICE_MBAFF), 2 );
2706             }
2707         }
2708
2709         /* Free mb info after the last thread's done using it */
2710         if( h->fdec->mb_info_free && (!h->param.b_sliced_threads || h->i_thread_idx == (h->param.i_threads-1)) )
2711         {
2712             h->fdec->mb_info_free( h->fdec->mb_info );
2713             h->fdec->mb_info = NULL;
2714             h->fdec->mb_info_free = NULL;
2715         }
2716     }
2717
2718     return 0;
2719 }
2720
2721 static void x264_thread_sync_context( x264_t *dst, x264_t *src )
2722 {
2723     if( dst == src )
2724         return;
2725
2726     // reference counting
2727     for( x264_frame_t **f = src->frames.reference; *f; f++ )
2728         (*f)->i_reference_count++;
2729     for( x264_frame_t **f = dst->frames.reference; *f; f++ )
2730         x264_frame_push_unused( src, *f );
2731     src->fdec->i_reference_count++;
2732     x264_frame_push_unused( src, dst->fdec );
2733
2734     // copy everything except the per-thread pointers and the constants.
2735     memcpy( &dst->i_frame, &src->i_frame, offsetof(x264_t, mb.type) - offsetof(x264_t, i_frame) );
2736     dst->param = src->param;
2737     dst->stat = src->stat;
2738     dst->pixf = src->pixf;
2739 }
2740
2741 static void x264_thread_sync_stat( x264_t *dst, x264_t *src )
2742 {
2743     if( dst == src )
2744         return;
2745     memcpy( &dst->stat.i_frame_count, &src->stat.i_frame_count, sizeof(dst->stat) - sizeof(dst->stat.frame) );
2746 }
2747
2748 static void *x264_slices_write( x264_t *h )
2749 {
2750     int i_slice_num = 0;
2751     int last_thread_mb = h->sh.i_last_mb;
2752
2753 #if HAVE_VISUALIZE
2754     if( h->param.b_visualize )
2755         if( x264_visualize_init( h ) )
2756             goto fail;
2757 #endif
2758
2759     /* init stats */
2760     memset( &h->stat.frame, 0, sizeof(h->stat.frame) );
2761     h->mb.b_reencode_mb = 0;
2762     while( h->sh.i_first_mb + SLICE_MBAFF*h->mb.i_mb_stride <= last_thread_mb )
2763     {
2764         h->sh.i_last_mb = last_thread_mb;
2765         if( !i_slice_num || !x264_frame_new_slice( h, h->fdec ) )
2766         {
2767             if( h->param.i_slice_max_mbs )
2768             {
2769                 if( SLICE_MBAFF )
2770                 {
2771                     // convert first to mbaff form, add slice-max-mbs, then convert back to normal form
2772                     int last_mbaff = 2*(h->sh.i_first_mb % h->mb.i_mb_width)
2773                         + h->mb.i_mb_width*(h->sh.i_first_mb / h->mb.i_mb_width)
2774                         + h->param.i_slice_max_mbs - 1;
2775                     int last_x = (last_mbaff % (2*h->mb.i_mb_width))/2;
2776                     int last_y = (last_mbaff / (2*h->mb.i_mb_width))*2 + 1;
2777                     h->sh.i_last_mb = last_x + h->mb.i_mb_stride*last_y;
2778                 }
2779                 else
2780                 {
2781                     h->sh.i_last_mb = h->sh.i_first_mb + h->param.i_slice_max_mbs - 1;
2782                     if( h->sh.i_last_mb < last_thread_mb && last_thread_mb - h->sh.i_last_mb < h->param.i_slice_min_mbs )
2783                         h->sh.i_last_mb = last_thread_mb - h->param.i_slice_min_mbs;
2784                 }
2785                 i_slice_num++;
2786             }
2787             else if( h->param.i_slice_count && !h->param.b_sliced_threads )
2788             {
2789                 int height = h->mb.i_mb_height >> PARAM_INTERLACED;
2790                 int width = h->mb.i_mb_width << PARAM_INTERLACED;
2791                 i_slice_num++;
2792                 h->sh.i_last_mb = (height * i_slice_num + h->param.i_slice_count/2) / h->param.i_slice_count * width - 1;
2793             }
2794         }
2795         h->sh.i_last_mb = X264_MIN( h->sh.i_last_mb, last_thread_mb );
2796         if( x264_stack_align( x264_slice_write, h ) )
2797             goto fail;
2798         h->sh.i_first_mb = h->sh.i_last_mb + 1;
2799         // if i_first_mb is not the last mb in a row then go to the next mb in MBAFF order
2800         if( SLICE_MBAFF && h->sh.i_first_mb % h->mb.i_mb_width )
2801             h->sh.i_first_mb -= h->mb.i_mb_stride;
2802     }
2803
2804 #if HAVE_VISUALIZE
2805     if( h->param.b_visualize )
2806     {
2807         x264_visualize_show( h );
2808         x264_visualize_close( h );
2809     }
2810 #endif
2811
2812     return (void *)0;
2813
2814 fail:
2815     /* Tell other threads we're done, so they wouldn't wait for it */
2816     if( h->param.b_sliced_threads )
2817         x264_threadslice_cond_broadcast( h, 2 );
2818     return (void *)-1;
2819 }
2820
2821 static int x264_threaded_slices_write( x264_t *h )
2822 {
2823     /* set first/last mb and sync contexts */
2824     for( int i = 0; i < h->param.i_threads; i++ )
2825     {
2826         x264_t *t = h->thread[i];
2827         if( i )
2828         {
2829             t->param = h->param;
2830             memcpy( &t->i_frame, &h->i_frame, offsetof(x264_t, rc) - offsetof(x264_t, i_frame) );
2831         }
2832         int height = h->mb.i_mb_height >> PARAM_INTERLACED;
2833         t->i_threadslice_start = ((height *  i    + h->param.i_slice_count/2) / h->param.i_threads) << PARAM_INTERLACED;
2834         t->i_threadslice_end   = ((height * (i+1) + h->param.i_slice_count/2) / h->param.i_threads) << PARAM_INTERLACED;
2835         t->sh.i_first_mb = t->i_threadslice_start * h->mb.i_mb_width;
2836         t->sh.i_last_mb  =   t->i_threadslice_end * h->mb.i_mb_width - 1;
2837     }
2838
2839     x264_stack_align( x264_analyse_weight_frame, h, h->mb.i_mb_height*16 + 16 );
2840
2841     x264_threads_distribute_ratecontrol( h );
2842
2843     /* setup */
2844     for( int i = 0; i < h->param.i_threads; i++ )
2845     {
2846         h->thread[i]->i_thread_idx = i;
2847         h->thread[i]->b_thread_active = 1;
2848         x264_threadslice_cond_broadcast( h->thread[i], 0 );
2849     }
2850     /* dispatch */
2851     for( int i = 0; i < h->param.i_threads; i++ )
2852         x264_threadpool_run( h->threadpool, (void*)x264_slices_write, h->thread[i] );
2853     /* wait */
2854     for( int i = 0; i < h->param.i_threads; i++ )
2855         x264_threadslice_cond_wait( h->thread[i], 1 );
2856
2857     x264_threads_merge_ratecontrol( h );
2858
2859     for( int i = 1; i < h->param.i_threads; i++ )
2860     {
2861         x264_t *t = h->thread[i];
2862         for( int j = 0; j < t->out.i_nal; j++ )
2863         {
2864             h->out.nal[h->out.i_nal] = t->out.nal[j];
2865             h->out.i_nal++;
2866             x264_nal_check_buffer( h );
2867         }
2868         /* All entries in stat.frame are ints except for ssd/ssim. */
2869         for( int j = 0; j < (offsetof(x264_t,stat.frame.i_ssd) - offsetof(x264_t,stat.frame.i_mv_bits)) / sizeof(int); j++ )
2870             ((int*)&h->stat.frame)[j] += ((int*)&t->stat.frame)[j];
2871         for( int j = 0; j < 3; j++ )
2872             h->stat.frame.i_ssd[j] += t->stat.frame.i_ssd[j];
2873         h->stat.frame.f_ssim += t->stat.frame.f_ssim;
2874         h->stat.frame.i_ssim_cnt += t->stat.frame.i_ssim_cnt;
2875     }
2876
2877     return 0;
2878 }
2879
2880 void x264_encoder_intra_refresh( x264_t *h )
2881 {
2882     h = h->thread[h->i_thread_phase];
2883     h->b_queued_intra_refresh = 1;
2884 }
2885
2886 int x264_encoder_invalidate_reference( x264_t *h, int64_t pts )
2887 {
2888     if( h->param.i_bframe )
2889     {
2890         x264_log( h, X264_LOG_ERROR, "x264_encoder_invalidate_reference is not supported with B-frames enabled\n" );
2891         return -1;
2892     }
2893     if( h->param.b_intra_refresh )
2894     {
2895         x264_log( h, X264_LOG_ERROR, "x264_encoder_invalidate_reference is not supported with intra refresh enabled\n" );
2896         return -1;
2897     }
2898     h = h->thread[h->i_thread_phase];
2899     if( pts >= h->i_last_idr_pts )
2900     {
2901         for( int i = 0; h->frames.reference[i]; i++ )
2902             if( pts <= h->frames.reference[i]->i_pts )
2903                 h->frames.reference[i]->b_corrupt = 1;
2904         if( pts <= h->fdec->i_pts )
2905             h->fdec->b_corrupt = 1;
2906     }
2907     return 0;
2908 }
2909
2910 /****************************************************************************
2911  * x264_encoder_encode:
2912  *  XXX: i_poc   : is the poc of the current given picture
2913  *       i_frame : is the number of the frame being coded
2914  *  ex:  type frame poc
2915  *       I      0   2*0
2916  *       P      1   2*3
2917  *       B      2   2*1
2918  *       B      3   2*2
2919  *       P      4   2*6
2920  *       B      5   2*4
2921  *       B      6   2*5
2922  ****************************************************************************/
2923 int     x264_encoder_encode( x264_t *h,
2924                              x264_nal_t **pp_nal, int *pi_nal,
2925                              x264_picture_t *pic_in,
2926                              x264_picture_t *pic_out )
2927 {
2928     x264_t *thread_current, *thread_prev, *thread_oldest;
2929     int i_nal_type, i_nal_ref_idc, i_global_qp;
2930     int overhead = NALU_OVERHEAD;
2931
2932 #if HAVE_OPENCL
2933     if( h->opencl.b_fatal_error )
2934         return -1;
2935 #endif
2936
2937     if( h->i_thread_frames > 1 )
2938     {
2939         thread_prev    = h->thread[ h->i_thread_phase ];
2940         h->i_thread_phase = (h->i_thread_phase + 1) % h->i_thread_frames;
2941         thread_current = h->thread[ h->i_thread_phase ];
2942         thread_oldest  = h->thread[ (h->i_thread_phase + 1) % h->i_thread_frames ];
2943         x264_thread_sync_context( thread_current, thread_prev );
2944         x264_thread_sync_ratecontrol( thread_current, thread_prev, thread_oldest );
2945         h = thread_current;
2946     }
2947     else
2948     {
2949         thread_current =
2950         thread_oldest  = h;
2951     }
2952 #if HAVE_MMX
2953     if( h->param.cpu&X264_CPU_SSE_MISALIGN )
2954         x264_cpu_mask_misalign_sse();
2955 #endif
2956     h->i_cpb_delay_pir_offset = h->i_cpb_delay_pir_offset_next;
2957
2958     /* no data out */
2959     *pi_nal = 0;
2960     *pp_nal = NULL;
2961
2962     /* ------------------- Setup new frame from picture -------------------- */
2963     if( pic_in != NULL )
2964     {
2965         /* 1: Copy the picture to a frame and move it to a buffer */
2966         x264_frame_t *fenc = x264_frame_pop_unused( h, 0 );
2967         if( !fenc )
2968             return -1;
2969
2970         if( x264_frame_copy_picture( h, fenc, pic_in ) < 0 )
2971             return -1;
2972
2973         if( h->param.i_width != 16 * h->mb.i_mb_width ||
2974             h->param.i_height != 16 * h->mb.i_mb_height )
2975             x264_frame_expand_border_mod16( h, fenc );
2976
2977         fenc->i_frame = h->frames.i_input++;
2978
2979         if( fenc->i_frame == 0 )
2980             h->frames.i_first_pts = fenc->i_pts;
2981         if( h->frames.i_bframe_delay && fenc->i_frame == h->frames.i_bframe_delay )
2982             h->frames.i_bframe_delay_time = fenc->i_pts - h->frames.i_first_pts;
2983
2984         if( h->param.b_vfr_input && fenc->i_pts <= h->frames.i_largest_pts )
2985             x264_log( h, X264_LOG_WARNING, "non-strictly-monotonic PTS\n" );
2986
2987         h->frames.i_second_largest_pts = h->frames.i_largest_pts;
2988         h->frames.i_largest_pts = fenc->i_pts;
2989
2990         if( (fenc->i_pic_struct < PIC_STRUCT_AUTO) || (fenc->i_pic_struct > PIC_STRUCT_TRIPLE) )
2991             fenc->i_pic_struct = PIC_STRUCT_AUTO;
2992
2993         if( fenc->i_pic_struct == PIC_STRUCT_AUTO )
2994         {
2995 #if HAVE_INTERLACED
2996             int b_interlaced = fenc->param ? fenc->param->b_interlaced : h->param.b_interlaced;
2997 #else
2998             int b_interlaced = 0;
2999 #endif
3000             if( b_interlaced )
3001             {
3002                 int b_tff = fenc->param ? fenc->param->b_tff : h->param.b_tff;
3003                 fenc->i_pic_struct = b_tff ? PIC_STRUCT_TOP_BOTTOM : PIC_STRUCT_BOTTOM_TOP;
3004             }
3005             else
3006                 fenc->i_pic_struct = PIC_STRUCT_PROGRESSIVE;
3007         }
3008
3009         if( h->param.rc.b_mb_tree && h->param.rc.b_stat_read )
3010         {
3011             if( x264_macroblock_tree_read( h, fenc, pic_in->prop.quant_offsets ) )
3012                 return -1;
3013         }
3014         else
3015             x264_stack_align( x264_adaptive_quant_frame, h, fenc, pic_in->prop.quant_offsets );
3016
3017         if( pic_in->prop.quant_offsets_free )
3018             pic_in->prop.quant_offsets_free( pic_in->prop.quant_offsets );
3019
3020         if( h->frames.b_have_lowres )
3021             x264_frame_init_lowres( h, fenc );
3022
3023         /* 2: Place the frame into the queue for its slice type decision */
3024         x264_lookahead_put_frame( h, fenc );
3025
3026         if( h->frames.i_input <= h->frames.i_delay + 1 - h->i_thread_frames )
3027         {
3028             /* Nothing yet to encode, waiting for filling of buffers */
3029             pic_out->i_type = X264_TYPE_AUTO;
3030             return 0;
3031         }
3032     }
3033     else
3034     {
3035         /* signal kills for lookahead thread */
3036         x264_pthread_mutex_lock( &h->lookahead->ifbuf.mutex );
3037         h->lookahead->b_exit_thread = 1;
3038         x264_pthread_cond_broadcast( &h->lookahead->ifbuf.cv_fill );
3039         x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
3040     }
3041
3042     h->i_frame++;
3043     /* 3: The picture is analyzed in the lookahead */
3044     if( !h->frames.current[0] )
3045         x264_lookahead_get_frames( h );
3046
3047     if( !h->frames.current[0] && x264_lookahead_is_empty( h ) )
3048         return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
3049
3050     /* ------------------- Get frame to be encoded ------------------------- */
3051     /* 4: get picture to encode */
3052     h->fenc = x264_frame_shift( h->frames.current );
3053
3054     /* If applicable, wait for previous frame reconstruction to finish */
3055     if( h->param.b_sliced_threads )
3056         if( x264_threadpool_wait_all( h ) < 0 )
3057             return -1;
3058
3059     if( h->i_frame == h->i_thread_frames - 1 )
3060         h->i_reordered_pts_delay = h->fenc->i_reordered_pts;
3061     if( h->fenc->param )
3062     {
3063         x264_encoder_reconfig( h, h->fenc->param );
3064         if( h->fenc->param->param_free )
3065         {
3066             h->fenc->param->param_free( h->fenc->param );
3067             h->fenc->param = NULL;
3068         }
3069     }
3070
3071     // ok to call this before encoding any frames, since the initial values of fdec have b_kept_as_ref=0
3072     if( x264_reference_update( h ) )
3073         return -1;
3074     h->fdec->i_lines_completed = -1;
3075
3076     if( !IS_X264_TYPE_I( h->fenc->i_type ) )
3077     {
3078         int valid_refs_left = 0;
3079         for( int i = 0; h->frames.reference[i]; i++ )
3080             if( !h->frames.reference[i]->b_corrupt )
3081                 valid_refs_left++;
3082         /* No valid reference frames left: force an IDR. */
3083         if( !valid_refs_left )
3084         {
3085             h->fenc->b_keyframe = 1;
3086             h->fenc->i_type = X264_TYPE_IDR;
3087         }
3088     }
3089
3090     if( h->fenc->b_keyframe )
3091     {
3092         h->frames.i_last_keyframe = h->fenc->i_frame;
3093         if( h->fenc->i_type == X264_TYPE_IDR )
3094         {
3095             h->i_frame_num = 0;
3096             h->frames.i_last_idr = h->fenc->i_frame;
3097         }
3098     }
3099     h->sh.i_mmco_command_count =
3100     h->sh.i_mmco_remove_from_end = 0;
3101     h->b_ref_reorder[0] =
3102     h->b_ref_reorder[1] = 0;
3103     h->fdec->i_poc =
3104     h->fenc->i_poc = 2 * ( h->fenc->i_frame - X264_MAX( h->frames.i_last_idr, 0 ) );
3105
3106     /* ------------------- Setup frame context ----------------------------- */
3107     /* 5: Init data dependent of frame type */
3108     if( h->fenc->i_type == X264_TYPE_IDR )
3109     {
3110         /* reset ref pictures */
3111         i_nal_type    = NAL_SLICE_IDR;
3112         i_nal_ref_idc = NAL_PRIORITY_HIGHEST;
3113         h->sh.i_type = SLICE_TYPE_I;
3114         x264_reference_reset( h );
3115         h->frames.i_poc_last_open_gop = -1;
3116     }
3117     else if( h->fenc->i_type == X264_TYPE_I )
3118     {
3119         i_nal_type    = NAL_SLICE;
3120         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
3121         h->sh.i_type = SLICE_TYPE_I;
3122         x264_reference_hierarchy_reset( h );
3123         if( h->param.b_open_gop )
3124             h->frames.i_poc_last_open_gop = h->fenc->b_keyframe ? h->fenc->i_poc : -1;
3125     }
3126     else if( h->fenc->i_type == X264_TYPE_P )
3127     {
3128         i_nal_type    = NAL_SLICE;
3129         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
3130         h->sh.i_type = SLICE_TYPE_P;
3131         x264_reference_hierarchy_reset( h );
3132         h->frames.i_poc_last_open_gop = -1;
3133     }
3134     else if( h->fenc->i_type == X264_TYPE_BREF )
3135     {
3136         i_nal_type    = NAL_SLICE;
3137         i_nal_ref_idc = h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT ? NAL_PRIORITY_LOW : NAL_PRIORITY_HIGH;
3138         h->sh.i_type = SLICE_TYPE_B;
3139         x264_reference_hierarchy_reset( h );
3140     }
3141     else    /* B frame */
3142     {
3143         i_nal_type    = NAL_SLICE;
3144         i_nal_ref_idc = NAL_PRIORITY_DISPOSABLE;
3145         h->sh.i_type = SLICE_TYPE_B;
3146     }
3147
3148     h->fdec->i_type = h->fenc->i_type;
3149     h->fdec->i_frame = h->fenc->i_frame;
3150     h->fenc->b_kept_as_ref =
3151     h->fdec->b_kept_as_ref = i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE && h->param.i_keyint_max > 1;
3152
3153     h->fdec->mb_info = h->fenc->mb_info;
3154     h->fdec->mb_info_free = h->fenc->mb_info_free;
3155     h->fenc->mb_info = NULL;
3156     h->fenc->mb_info_free = NULL;
3157
3158     h->fdec->i_pts = h->fenc->i_pts;
3159     if( h->frames.i_bframe_delay )
3160     {
3161         int64_t *prev_reordered_pts = thread_current->frames.i_prev_reordered_pts;
3162         h->fdec->i_dts = h->i_frame > h->frames.i_bframe_delay
3163                        ? prev_reordered_pts[ (h->i_frame - h->frames.i_bframe_delay) % h->frames.i_bframe_delay ]
3164                        : h->fenc->i_reordered_pts - h->frames.i_bframe_delay_time;
3165         prev_reordered_pts[ h->i_frame % h->frames.i_bframe_delay ] = h->fenc->i_reordered_pts;
3166     }
3167     else
3168         h->fdec->i_dts = h->fenc->i_reordered_pts;
3169     if( h->fenc->i_type == X264_TYPE_IDR )
3170         h->i_last_idr_pts = h->fdec->i_pts;
3171
3172     /* ------------------- Init                ----------------------------- */
3173     /* build ref list 0/1 */
3174     x264_reference_build_list( h, h->fdec->i_poc );
3175
3176     /* ---------------------- Write the bitstream -------------------------- */
3177     /* Init bitstream context */
3178     if( h->param.b_sliced_threads )
3179     {
3180         for( int i = 0; i < h->param.i_threads; i++ )
3181         {
3182             bs_init( &h->thread[i]->out.bs, h->thread[i]->out.p_bitstream, h->thread[i]->out.i_bitstream );
3183             h->thread[i]->out.i_nal = 0;
3184         }
3185     }
3186     else
3187     {
3188         bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
3189         h->out.i_nal = 0;
3190     }
3191
3192     if( h->param.b_aud )
3193     {
3194         int pic_type;
3195
3196         if( h->sh.i_type == SLICE_TYPE_I )
3197             pic_type = 0;
3198         else if( h->sh.i_type == SLICE_TYPE_P )
3199             pic_type = 1;
3200         else if( h->sh.i_type == SLICE_TYPE_B )
3201             pic_type = 2;
3202         else
3203             pic_type = 7;
3204
3205         x264_nal_start( h, NAL_AUD, NAL_PRIORITY_DISPOSABLE );
3206         bs_write( &h->out.bs, 3, pic_type );
3207         bs_rbsp_trailing( &h->out.bs );
3208         if( x264_nal_end( h ) )
3209             return -1;
3210         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
3211     }
3212
3213     h->i_nal_type = i_nal_type;
3214     h->i_nal_ref_idc = i_nal_ref_idc;
3215
3216     if( h->param.b_intra_refresh )
3217     {
3218         if( IS_X264_TYPE_I( h->fenc->i_type ) )
3219         {
3220             h->fdec->i_frames_since_pir = 0;
3221             h->b_queued_intra_refresh = 0;
3222             /* PIR is currently only supported with ref == 1, so any intra frame effectively refreshes
3223              * the whole frame and counts as an intra refresh. */
3224             h->fdec->f_pir_position = h->mb.i_mb_width;
3225         }
3226         else if( h->fenc->i_type == X264_TYPE_P )
3227         {
3228             int pocdiff = (h->fdec->i_poc - h->fref[0][0]->i_poc)/2;
3229             float increment = X264_MAX( ((float)h->mb.i_mb_width-1) / h->param.i_keyint_max, 1 );
3230             h->fdec->f_pir_position = h->fref[0][0]->f_pir_position;
3231             h->fdec->i_frames_since_pir = h->fref[0][0]->i_frames_since_pir + pocdiff;
3232             if( h->fdec->i_frames_since_pir >= h->param.i_keyint_max ||
3233                 (h->b_queued_intra_refresh && h->fdec->f_pir_position + 0.5 >= h->mb.i_mb_width) )
3234             {
3235                 h->fdec->f_pir_position = 0;
3236                 h->fdec->i_frames_since_pir = 0;
3237                 h->b_queued_intra_refresh = 0;
3238                 h->fenc->b_keyframe = 1;
3239             }
3240             h->fdec->i_pir_start_col = h->fdec->f_pir_position+0.5;
3241             h->fdec->f_pir_position += increment * pocdiff;
3242             h->fdec->i_pir_end_col = h->fdec->f_pir_position+0.5;
3243             /* If our intra refresh has reached the right side of the frame, we're done. */
3244             if( h->fdec->i_pir_end_col >= h->mb.i_mb_width - 1 )
3245             {
3246                 h->fdec->f_pir_position = h->mb.i_mb_width;
3247                 h->fdec->i_pir_end_col = h->mb.i_mb_width - 1;
3248             }
3249         }
3250     }
3251
3252     if( h->fenc->b_keyframe )
3253     {
3254         /* Write SPS and PPS */
3255         if( h->param.b_repeat_headers )
3256         {
3257             /* generate sequence parameters */
3258             x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
3259             x264_sps_write( &h->out.bs, h->sps );
3260             if( x264_nal_end( h ) )
3261                 return -1;
3262             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
3263
3264             /* generate picture parameters */
3265             x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
3266             x264_pps_write( &h->out.bs, h->sps, h->pps );
3267             if( x264_nal_end( h ) )
3268                 return -1;
3269             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
3270         }
3271
3272         /* when frame threading is used, buffering period sei is written in x264_encoder_frame_end */
3273         if( h->i_thread_frames == 1 && h->sps->vui.b_nal_hrd_parameters_present )
3274         {
3275             x264_hrd_fullness( h );
3276             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3277             x264_sei_buffering_period_write( h, &h->out.bs );
3278             if( x264_nal_end( h ) )
3279                return -1;
3280             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
3281         }
3282     }
3283
3284     /* write extra sei */
3285     for( int i = 0; i < h->fenc->extra_sei.num_payloads; i++ )
3286     {
3287         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3288         x264_sei_write( &h->out.bs, h->fenc->extra_sei.payloads[i].payload, h->fenc->extra_sei.payloads[i].payload_size,
3289                         h->fenc->extra_sei.payloads[i].payload_type );
3290         if( x264_nal_end( h ) )
3291             return -1;
3292         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
3293         if( h->fenc->extra_sei.sei_free )
3294         {
3295             h->fenc->extra_sei.sei_free( h->fenc->extra_sei.payloads[i].payload );
3296             h->fenc->extra_sei.payloads[i].payload = NULL;
3297         }
3298     }
3299
3300     if( h->fenc->extra_sei.sei_free )
3301     {
3302         h->fenc->extra_sei.sei_free( h->fenc->extra_sei.payloads );
3303         h->fenc->extra_sei.payloads = NULL;
3304         h->fenc->extra_sei.sei_free = NULL;
3305     }
3306
3307     if( h->fenc->b_keyframe )
3308     {
3309         if( h->param.b_repeat_headers && h->fenc->i_frame == 0 )
3310         {
3311             /* identify ourself */
3312             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3313             if( x264_sei_version_write( h, &h->out.bs ) )
3314                 return -1;
3315             if( x264_nal_end( h ) )
3316                 return -1;
3317             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
3318         }
3319
3320         if( h->fenc->i_type != X264_TYPE_IDR )
3321         {
3322             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;
3323             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3324             x264_sei_recovery_point_write( h, &h->out.bs, time_to_recovery );
3325             if( x264_nal_end( h ) )
3326                 return -1;
3327             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
3328         }
3329
3330         if ( h->param.i_frame_packing >= 0 )
3331         {
3332             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3333             x264_sei_frame_packing_write( h, &h->out.bs );
3334             if( x264_nal_end( h ) )
3335                 return -1;
3336             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
3337         }
3338     }
3339
3340     /* generate sei pic timing */
3341     if( h->sps->vui.b_pic_struct_present || h->sps->vui.b_nal_hrd_parameters_present )
3342     {
3343         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3344         x264_sei_pic_timing_write( h, &h->out.bs );
3345         if( x264_nal_end( h ) )
3346             return -1;
3347         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
3348     }
3349
3350     /* As required by Blu-ray. */
3351     if( !IS_X264_TYPE_B( h->fenc->i_type ) && h->b_sh_backup )
3352     {
3353         h->b_sh_backup = 0;
3354         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3355         x264_sei_dec_ref_pic_marking_write( h, &h->out.bs );
3356         if( x264_nal_end( h ) )
3357             return -1;
3358         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
3359     }
3360
3361     if( h->fenc->b_keyframe && h->param.b_intra_refresh )
3362         h->i_cpb_delay_pir_offset_next = h->fenc->i_cpb_delay;
3363
3364     /* Init the rate control */
3365     /* FIXME: Include slice header bit cost. */
3366     x264_ratecontrol_start( h, h->fenc->i_qpplus1, overhead*8 );
3367     i_global_qp = x264_ratecontrol_qp( h );
3368
3369     pic_out->i_qpplus1 =
3370     h->fdec->i_qpplus1 = i_global_qp + 1;
3371
3372     if( h->param.rc.b_stat_read && h->sh.i_type != SLICE_TYPE_I )
3373     {
3374         x264_reference_build_list_optimal( h );
3375         x264_reference_check_reorder( h );
3376     }
3377
3378     if( h->i_ref[0] )
3379         h->fdec->i_poc_l0ref0 = h->fref[0][0]->i_poc;
3380
3381     /* ------------------------ Create slice header  ----------------------- */
3382     x264_slice_init( h, i_nal_type, i_global_qp );
3383
3384     /*------------------------- Weights -------------------------------------*/
3385     if( h->sh.i_type == SLICE_TYPE_B )
3386         x264_macroblock_bipred_init( h );
3387
3388     x264_weighted_pred_init( h );
3389
3390     if( i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE )
3391         h->i_frame_num++;
3392
3393     /* Write frame */
3394     h->i_threadslice_start = 0;
3395     h->i_threadslice_end = h->mb.i_mb_height;
3396     if( h->i_thread_frames > 1 )
3397     {
3398         x264_threadpool_run( h->threadpool, (void*)x264_slices_write, h );
3399         h->b_thread_active = 1;
3400     }
3401     else if( h->param.b_sliced_threads )
3402     {
3403         if( x264_threaded_slices_write( h ) )
3404             return -1;
3405     }
3406     else
3407         if( (intptr_t)x264_slices_write( h ) )
3408             return -1;
3409
3410     return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
3411 }
3412
3413 static int x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
3414                                    x264_nal_t **pp_nal, int *pi_nal,
3415                                    x264_picture_t *pic_out )
3416 {
3417     char psz_message[80];
3418
3419     if( !h->param.b_sliced_threads && h->b_thread_active )
3420     {
3421         h->b_thread_active = 0;
3422         if( (intptr_t)x264_threadpool_wait( h->threadpool, h ) )
3423             return -1;
3424     }
3425     if( !h->out.i_nal )
3426     {
3427         pic_out->i_type = X264_TYPE_AUTO;
3428         return 0;
3429     }
3430
3431     x264_emms();
3432
3433     /* generate buffering period sei and insert it into place */
3434     if( h->i_thread_frames > 1 && h->fenc->b_keyframe && h->sps->vui.b_nal_hrd_parameters_present )
3435     {
3436         x264_hrd_fullness( h );
3437         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3438         x264_sei_buffering_period_write( h, &h->out.bs );
3439         if( x264_nal_end( h ) )
3440            return -1;
3441         /* buffering period sei must follow AUD, SPS and PPS and precede all other SEIs */
3442         int idx = 0;
3443         while( h->out.nal[idx].i_type == NAL_AUD ||
3444                h->out.nal[idx].i_type == NAL_SPS ||
3445                h->out.nal[idx].i_type == NAL_PPS )
3446             idx++;
3447         x264_nal_t nal_tmp = h->out.nal[h->out.i_nal-1];
3448         memmove( &h->out.nal[idx+1], &h->out.nal[idx], (h->out.i_nal-idx-1)*sizeof(x264_nal_t) );
3449         h->out.nal[idx] = nal_tmp;
3450     }
3451
3452     int frame_size = x264_encoder_encapsulate_nals( h, 0 );
3453     if( frame_size < 0 )
3454         return -1;
3455
3456     /* Set output picture properties */
3457     pic_out->i_type = h->fenc->i_type;
3458
3459     pic_out->b_keyframe = h->fenc->b_keyframe;
3460     pic_out->i_pic_struct = h->fenc->i_pic_struct;
3461
3462     pic_out->i_pts = h->fdec->i_pts;
3463     pic_out->i_dts = h->fdec->i_dts;
3464
3465     if( pic_out->i_pts < pic_out->i_dts )
3466         x264_log( h, X264_LOG_WARNING, "invalid DTS: PTS is less than DTS\n" );
3467
3468     pic_out->opaque = h->fenc->opaque;
3469
3470     pic_out->img.i_csp = h->fdec->i_csp;
3471 #if HIGH_BIT_DEPTH
3472     pic_out->img.i_csp |= X264_CSP_HIGH_DEPTH;
3473 #endif
3474     pic_out->img.i_plane = h->fdec->i_plane;
3475     for( int i = 0; i < pic_out->img.i_plane; i++ )
3476     {
3477         pic_out->img.i_stride[i] = h->fdec->i_stride[i] * sizeof(pixel);
3478         pic_out->img.plane[i] = (uint8_t*)h->fdec->plane[i];
3479     }
3480
3481     x264_frame_push_unused( thread_current, h->fenc );
3482
3483     /* ---------------------- Update encoder state ------------------------- */
3484
3485     /* update rc */
3486     int filler = 0;
3487     if( x264_ratecontrol_end( h, frame_size * 8, &filler ) < 0 )
3488         return -1;
3489
3490     pic_out->hrd_timing = h->fenc->hrd_timing;
3491     pic_out->prop.f_crf_avg = h->fdec->f_crf_avg;
3492
3493     while( filler > 0 )
3494     {
3495         int f, overhead;
3496         overhead = (FILLER_OVERHEAD - h->param.b_annexb);
3497         if( h->param.i_slice_max_size && filler > h->param.i_slice_max_size )
3498         {
3499             int next_size = filler - h->param.i_slice_max_size;
3500             int overflow = X264_MAX( overhead - next_size, 0 );
3501             f = h->param.i_slice_max_size - overhead - overflow;
3502         }
3503         else
3504             f = X264_MAX( 0, filler - overhead );
3505
3506         if( x264_bitstream_check_buffer_filler( h, f ) )
3507             return -1;
3508         x264_nal_start( h, NAL_FILLER, NAL_PRIORITY_DISPOSABLE );
3509         x264_filler_write( h, &h->out.bs, f );
3510         if( x264_nal_end( h ) )
3511             return -1;
3512         int total_size = x264_encoder_encapsulate_nals( h, h->out.i_nal-1 );
3513         if( total_size < 0 )
3514             return -1;
3515         frame_size += total_size;
3516         filler -= total_size;
3517     }
3518
3519     /* End bitstream, set output  */
3520     *pi_nal = h->out.i_nal;
3521     *pp_nal = h->out.nal;
3522
3523     h->out.i_nal = 0;
3524
3525     x264_noise_reduction_update( h );
3526
3527     /* ---------------------- Compute/Print statistics --------------------- */
3528     x264_thread_sync_stat( h, h->thread[0] );
3529
3530     /* Slice stat */
3531     h->stat.i_frame_count[h->sh.i_type]++;
3532     h->stat.i_frame_size[h->sh.i_type] += frame_size;
3533     h->stat.f_frame_qp[h->sh.i_type] += h->fdec->f_qp_avg_aq;
3534
3535     for( int i = 0; i < X264_MBTYPE_MAX; i++ )
3536         h->stat.i_mb_count[h->sh.i_type][i] += h->stat.frame.i_mb_count[i];
3537     for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
3538         h->stat.i_mb_partition[h->sh.i_type][i] += h->stat.frame.i_mb_partition[i];
3539     for( int i = 0; i < 2; i++ )
3540         h->stat.i_mb_count_8x8dct[i] += h->stat.frame.i_mb_count_8x8dct[i];
3541     for( int i = 0; i < 6; i++ )
3542         h->stat.i_mb_cbp[i] += h->stat.frame.i_mb_cbp[i];
3543     for( int i = 0; i < 4; i++ )
3544         for( int j = 0; j < 13; j++ )
3545             h->stat.i_mb_pred_mode[i][j] += h->stat.frame.i_mb_pred_mode[i][j];
3546     if( h->sh.i_type != SLICE_TYPE_I )
3547         for( int i_list = 0; i_list < 2; i_list++ )
3548             for( int i = 0; i < X264_REF_MAX*2; i++ )
3549                 h->stat.i_mb_count_ref[h->sh.i_type][i_list][i] += h->stat.frame.i_mb_count_ref[i_list][i];
3550     for( int i = 0; i < 3; i++ )
3551         h->stat.i_mb_field[i] += h->stat.frame.i_mb_field[i];
3552     if( h->sh.i_type == SLICE_TYPE_P && h->param.analyse.i_weighted_pred >= X264_WEIGHTP_SIMPLE )
3553     {
3554         h->stat.i_wpred[0] += !!h->sh.weight[0][0].weightfn;
3555         h->stat.i_wpred[1] += !!h->sh.weight[0][1].weightfn || !!h->sh.weight[0][2].weightfn;
3556     }
3557     if( h->sh.i_type == SLICE_TYPE_B )
3558     {
3559         h->stat.i_direct_frames[ h->sh.b_direct_spatial_mv_pred ] ++;
3560         if( h->mb.b_direct_auto_write )
3561         {
3562             //FIXME somewhat arbitrary time constants
3563             if( h->stat.i_direct_score[0] + h->stat.i_direct_score[1] > h->mb.i_mb_count )
3564                 for( int i = 0; i < 2; i++ )
3565                     h->stat.i_direct_score[i] = h->stat.i_direct_score[i] * 9/10;
3566             for( int i = 0; i < 2; i++ )
3567                 h->stat.i_direct_score[i] += h->stat.frame.i_direct_score[i];
3568         }
3569     }
3570     else
3571         h->stat.i_consecutive_bframes[h->fenc->i_bframes]++;
3572
3573     psz_message[0] = '\0';
3574     double dur = h->fenc->f_duration;
3575     h->stat.f_frame_duration[h->sh.i_type] += dur;
3576     if( h->param.analyse.b_psnr )
3577     {
3578         int64_t ssd[3] =
3579         {
3580             h->stat.frame.i_ssd[0],
3581             h->stat.frame.i_ssd[1],
3582             h->stat.frame.i_ssd[2],
3583         };
3584         int luma_size = h->param.i_width * h->param.i_height;
3585         int chroma_size = CHROMA_SIZE( luma_size );
3586         pic_out->prop.f_psnr[0] = x264_psnr( ssd[0], luma_size );
3587         pic_out->prop.f_psnr[1] = x264_psnr( ssd[1], chroma_size );
3588         pic_out->prop.f_psnr[2] = x264_psnr( ssd[2], chroma_size );
3589         pic_out->prop.f_psnr_avg = x264_psnr( ssd[0] + ssd[1] + ssd[2], luma_size + chroma_size*2 );
3590
3591         h->stat.f_ssd_global[h->sh.i_type]   += dur * (ssd[0] + ssd[1] + ssd[2]);
3592         h->stat.f_psnr_average[h->sh.i_type] += dur * pic_out->prop.f_psnr_avg;
3593         h->stat.f_psnr_mean_y[h->sh.i_type]  += dur * pic_out->prop.f_psnr[0];
3594         h->stat.f_psnr_mean_u[h->sh.i_type]  += dur * pic_out->prop.f_psnr[1];
3595         h->stat.f_psnr_mean_v[h->sh.i_type]  += dur * pic_out->prop.f_psnr[2];
3596
3597         snprintf( psz_message, 80, " PSNR Y:%5.2f U:%5.2f V:%5.2f", pic_out->prop.f_psnr[0],
3598                                                                     pic_out->prop.f_psnr[1],
3599                                                                     pic_out->prop.f_psnr[2] );
3600     }
3601
3602     if( h->param.analyse.b_ssim )
3603     {
3604         pic_out->prop.f_ssim = h->stat.frame.f_ssim / h->stat.frame.i_ssim_cnt;
3605         h->stat.f_ssim_mean_y[h->sh.i_type] += pic_out->prop.f_ssim * dur;
3606         snprintf( psz_message + strlen(psz_message), 80 - strlen(psz_message),
3607                   " SSIM Y:%.5f", pic_out->prop.f_ssim );
3608     }
3609     psz_message[79] = '\0';
3610
3611     x264_log( h, X264_LOG_DEBUG,
3612                   "frame=%4d QP=%.2f NAL=%d Slice:%c Poc:%-3d I:%-4d P:%-4d SKIP:%-4d size=%d bytes%s\n",
3613               h->i_frame,
3614               h->fdec->f_qp_avg_aq,
3615               h->i_nal_ref_idc,
3616               h->sh.i_type == SLICE_TYPE_I ? 'I' : (h->sh.i_type == SLICE_TYPE_P ? 'P' : 'B' ),
3617               h->fdec->i_poc,
3618               h->stat.frame.i_mb_count_i,
3619               h->stat.frame.i_mb_count_p,
3620               h->stat.frame.i_mb_count_skip,
3621               frame_size,
3622               psz_message );
3623
3624     // keep stats all in one place
3625     x264_thread_sync_stat( h->thread[0], h );
3626     // for the use of the next frame
3627     x264_thread_sync_stat( thread_current, h );
3628
3629 #ifdef DEBUG_MB_TYPE
3630 {
3631     static const char mb_chars[] = { 'i', 'i', 'I', 'C', 'P', '8', 'S',
3632         'D', '<', 'X', 'B', 'X', '>', 'B', 'B', 'B', 'B', '8', 'S' };
3633     for( int mb_xy = 0; mb_xy < h->mb.i_mb_width * h->mb.i_mb_height; mb_xy++ )
3634     {
3635         if( h->mb.type[mb_xy] < X264_MBTYPE_MAX && h->mb.type[mb_xy] >= 0 )
3636             fprintf( stderr, "%c ", mb_chars[ h->mb.type[mb_xy] ] );
3637         else
3638             fprintf( stderr, "? " );
3639
3640         if( (mb_xy+1) % h->mb.i_mb_width == 0 )
3641             fprintf( stderr, "\n" );
3642     }
3643 }
3644 #endif
3645
3646     /* Remove duplicates, must be done near the end as breaks h->fref0 array
3647      * by freeing some of its pointers. */
3648     for( int i = 0; i < h->i_ref[0]; i++ )
3649         if( h->fref[0][i] && h->fref[0][i]->b_duplicate )
3650         {
3651             x264_frame_push_blank_unused( h, h->fref[0][i] );
3652             h->fref[0][i] = 0;
3653         }
3654
3655     if( h->param.psz_dump_yuv )
3656         x264_frame_dump( h );
3657     x264_emms();
3658
3659     return frame_size;
3660 }
3661
3662 static void x264_print_intra( int64_t *i_mb_count, double i_count, int b_print_pcm, char *intra )
3663 {
3664     intra += sprintf( intra, "I16..4%s: %4.1f%% %4.1f%% %4.1f%%",
3665         b_print_pcm ? "..PCM" : "",
3666         i_mb_count[I_16x16]/ i_count,
3667         i_mb_count[I_8x8]  / i_count,
3668         i_mb_count[I_4x4]  / i_count );
3669     if( b_print_pcm )
3670         sprintf( intra, " %4.1f%%", i_mb_count[I_PCM]  / i_count );
3671 }
3672
3673 /****************************************************************************
3674  * x264_encoder_close:
3675  ****************************************************************************/
3676 void    x264_encoder_close  ( x264_t *h )
3677 {
3678     int64_t i_yuv_size = FRAME_SIZE( h->param.i_width * h->param.i_height );
3679     int64_t i_mb_count_size[2][7] = {{0}};
3680     char buf[200];
3681     int b_print_pcm = h->stat.i_mb_count[SLICE_TYPE_I][I_PCM]
3682                    || h->stat.i_mb_count[SLICE_TYPE_P][I_PCM]
3683                    || h->stat.i_mb_count[SLICE_TYPE_B][I_PCM];
3684
3685     x264_lookahead_delete( h );
3686
3687 #if HAVE_OPENCL
3688     x264_opencl_lookahead_delete( h );
3689     x264_opencl_function_t *ocl = h->opencl.ocl;
3690 #endif
3691
3692     if( h->param.b_sliced_threads )
3693         x264_threadpool_wait_all( h );
3694     if( h->param.i_threads > 1 )
3695         x264_threadpool_delete( h->threadpool );
3696     if( h->param.i_lookahead_threads > 1 )
3697         x264_threadpool_delete( h->lookaheadpool );
3698     if( h->i_thread_frames > 1 )
3699     {
3700         for( int i = 0; i < h->i_thread_frames; i++ )
3701             if( h->thread[i]->b_thread_active )
3702             {
3703                 assert( h->thread[i]->fenc->i_reference_count == 1 );
3704                 x264_frame_delete( h->thread[i]->fenc );
3705             }
3706
3707         x264_t *thread_prev = h->thread[h->i_thread_phase];
3708         x264_thread_sync_ratecontrol( h, thread_prev, h );
3709         x264_thread_sync_ratecontrol( thread_prev, thread_prev, h );
3710         h->i_frame = thread_prev->i_frame + 1 - h->i_thread_frames;
3711     }
3712     h->i_frame++;
3713
3714     /* Slices used and PSNR */
3715     for( int i = 0; i < 3; i++ )
3716     {
3717         static const uint8_t slice_order[] = { SLICE_TYPE_I, SLICE_TYPE_P, SLICE_TYPE_B };
3718         int i_slice = slice_order[i];
3719
3720         if( h->stat.i_frame_count[i_slice] > 0 )
3721         {
3722             int i_count = h->stat.i_frame_count[i_slice];
3723             double dur =  h->stat.f_frame_duration[i_slice];
3724             if( h->param.analyse.b_psnr )
3725             {
3726                 x264_log( h, X264_LOG_INFO,
3727                           "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",
3728                           slice_type_to_char[i_slice],
3729                           i_count,
3730                           h->stat.f_frame_qp[i_slice] / i_count,
3731                           (double)h->stat.i_frame_size[i_slice] / i_count,
3732                           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,
3733                           h->stat.f_psnr_average[i_slice] / dur,
3734                           x264_psnr( h->stat.f_ssd_global[i_slice], dur * i_yuv_size ) );
3735             }
3736             else
3737             {
3738                 x264_log( h, X264_LOG_INFO,
3739                           "frame %c:%-5d Avg QP:%5.2f  size:%6.0f\n",
3740                           slice_type_to_char[i_slice],
3741                           i_count,
3742                           h->stat.f_frame_qp[i_slice] / i_count,
3743                           (double)h->stat.i_frame_size[i_slice] / i_count );
3744             }
3745         }
3746     }
3747     if( h->param.i_bframe && h->stat.i_frame_count[SLICE_TYPE_B] )
3748     {
3749         char *p = buf;
3750         int den = 0;
3751         // weight by number of frames (including the I/P-frames) that are in a sequence of N B-frames
3752         for( int i = 0; i <= h->param.i_bframe; i++ )
3753             den += (i+1) * h->stat.i_consecutive_bframes[i];
3754         for( int i = 0; i <= h->param.i_bframe; i++ )
3755             p += sprintf( p, " %4.1f%%", 100. * (i+1) * h->stat.i_consecutive_bframes[i] / den );
3756         x264_log( h, X264_LOG_INFO, "consecutive B-frames:%s\n", buf );
3757     }
3758
3759     for( int i_type = 0; i_type < 2; i_type++ )
3760         for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
3761         {
3762             if( i == D_DIRECT_8x8 ) continue; /* direct is counted as its own type */
3763             i_mb_count_size[i_type][x264_mb_partition_pixel_table[i]] += h->stat.i_mb_partition[i_type][i];
3764         }
3765
3766     /* MB types used */
3767     if( h->stat.i_frame_count[SLICE_TYPE_I] > 0 )
3768     {
3769         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_I];
3770         double i_count = h->stat.i_frame_count[SLICE_TYPE_I] * h->mb.i_mb_count / 100.0;
3771         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
3772         x264_log( h, X264_LOG_INFO, "mb I  %s\n", buf );
3773     }
3774     if( h->stat.i_frame_count[SLICE_TYPE_P] > 0 )
3775     {
3776         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_P];
3777         double i_count = h->stat.i_frame_count[SLICE_TYPE_P] * h->mb.i_mb_count / 100.0;
3778         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_P];
3779         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
3780         x264_log( h, X264_LOG_INFO,
3781                   "mb P  %s  P16..4: %4.1f%% %4.1f%% %4.1f%% %4.1f%% %4.1f%%    skip:%4.1f%%\n",
3782                   buf,
3783                   i_mb_size[PIXEL_16x16] / (i_count*4),
3784                   (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
3785                   i_mb_size[PIXEL_8x8] / (i_count*4),
3786                   (i_mb_size[PIXEL_8x4] + i_mb_size[PIXEL_4x8]) / (i_count*4),
3787                   i_mb_size[PIXEL_4x4] / (i_count*4),
3788                   i_mb_count[P_SKIP] / i_count );
3789     }
3790     if( h->stat.i_frame_count[SLICE_TYPE_B] > 0 )
3791     {
3792         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_B];
3793         double i_count = h->stat.i_frame_count[SLICE_TYPE_B] * h->mb.i_mb_count / 100.0;
3794         double i_mb_list_count;
3795         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_B];
3796         int64_t list_count[3] = {0}; /* 0 == L0, 1 == L1, 2 == BI */
3797         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
3798         for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
3799             for( int j = 0; j < 2; j++ )
3800             {
3801                 int l0 = x264_mb_type_list_table[i][0][j];
3802                 int l1 = x264_mb_type_list_table[i][1][j];
3803                 if( l0 || l1 )
3804                     list_count[l1+l0*l1] += h->stat.i_mb_count[SLICE_TYPE_B][i] * 2;
3805             }
3806         list_count[0] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L0_8x8];
3807         list_count[1] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L1_8x8];
3808         list_count[2] += h->stat.i_mb_partition[SLICE_TYPE_B][D_BI_8x8];
3809         i_mb_count[B_DIRECT] += (h->stat.i_mb_partition[SLICE_TYPE_B][D_DIRECT_8x8]+2)/4;
3810         i_mb_list_count = (list_count[0] + list_count[1] + list_count[2]) / 100.0;
3811         sprintf( buf + strlen(buf), "  B16..8: %4.1f%% %4.1f%% %4.1f%%  direct:%4.1f%%  skip:%4.1f%%",
3812                  i_mb_size[PIXEL_16x16] / (i_count*4),
3813                  (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
3814                  i_mb_size[PIXEL_8x8] / (i_count*4),
3815                  i_mb_count[B_DIRECT] / i_count,
3816                  i_mb_count[B_SKIP]   / i_count );
3817         if( i_mb_list_count != 0 )
3818             sprintf( buf + strlen(buf), "  L0:%4.1f%% L1:%4.1f%% BI:%4.1f%%",
3819                      list_count[0] / i_mb_list_count,
3820                      list_count[1] / i_mb_list_count,
3821                      list_count[2] / i_mb_list_count );
3822         x264_log( h, X264_LOG_INFO, "mb B  %s\n", buf );
3823     }
3824
3825     x264_ratecontrol_summary( h );
3826
3827     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 )
3828     {
3829 #define SUM3(p) (p[SLICE_TYPE_I] + p[SLICE_TYPE_P] + p[SLICE_TYPE_B])
3830 #define SUM3b(p,o) (p[SLICE_TYPE_I][o] + p[SLICE_TYPE_P][o] + p[SLICE_TYPE_B][o])
3831         int64_t i_i8x8 = SUM3b( h->stat.i_mb_count, I_8x8 );
3832         int64_t i_intra = i_i8x8 + SUM3b( h->stat.i_mb_count, I_4x4 )
3833                                  + SUM3b( h->stat.i_mb_count, I_16x16 );
3834         int64_t i_all_intra = i_intra + SUM3b( h->stat.i_mb_count, I_PCM);
3835         int64_t i_skip = SUM3b( h->stat.i_mb_count, P_SKIP )
3836                        + SUM3b( h->stat.i_mb_count, B_SKIP );
3837         const int i_count = h->stat.i_frame_count[SLICE_TYPE_I] +
3838                             h->stat.i_frame_count[SLICE_TYPE_P] +
3839                             h->stat.i_frame_count[SLICE_TYPE_B];
3840         int64_t i_mb_count = (int64_t)i_count * h->mb.i_mb_count;
3841         int64_t i_inter = i_mb_count - i_skip - i_intra;
3842         const double duration = h->stat.f_frame_duration[SLICE_TYPE_I] +
3843                                 h->stat.f_frame_duration[SLICE_TYPE_P] +
3844                                 h->stat.f_frame_duration[SLICE_TYPE_B];
3845         float f_bitrate = SUM3(h->stat.i_frame_size) / duration / 125;
3846
3847         if( PARAM_INTERLACED )
3848         {
3849             char *fieldstats = buf;
3850             fieldstats[0] = 0;
3851             if( i_inter )
3852                 fieldstats += sprintf( fieldstats, " inter:%.1f%%", h->stat.i_mb_field[1] * 100.0 / i_inter );
3853             if( i_skip )
3854                 fieldstats += sprintf( fieldstats, " skip:%.1f%%", h->stat.i_mb_field[2] * 100.0 / i_skip );
3855             x264_log( h, X264_LOG_INFO, "field mbs: intra: %.1f%%%s\n",
3856                       h->stat.i_mb_field[0] * 100.0 / i_intra, buf );
3857         }
3858
3859         if( h->pps->b_transform_8x8_mode )
3860         {
3861             buf[0] = 0;
3862             if( h->stat.i_mb_count_8x8dct[0] )
3863                 sprintf( buf, " inter:%.1f%%", 100. * h->stat.i_mb_count_8x8dct[1] / h->stat.i_mb_count_8x8dct[0] );
3864             x264_log( h, X264_LOG_INFO, "8x8 transform intra:%.1f%%%s\n", 100. * i_i8x8 / i_intra, buf );
3865         }
3866
3867         if( (h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO ||
3868             (h->stat.i_direct_frames[0] && h->stat.i_direct_frames[1]))
3869             && h->stat.i_frame_count[SLICE_TYPE_B] )
3870         {
3871             x264_log( h, X264_LOG_INFO, "direct mvs  spatial:%.1f%% temporal:%.1f%%\n",
3872                       h->stat.i_direct_frames[1] * 100. / h->stat.i_frame_count[SLICE_TYPE_B],
3873                       h->stat.i_direct_frames[0] * 100. / h->stat.i_frame_count[SLICE_TYPE_B] );
3874         }
3875
3876         buf[0] = 0;
3877         int csize = CHROMA444 ? 4 : 1;
3878         if( i_mb_count != i_all_intra )
3879             sprintf( buf, " inter: %.1f%% %.1f%% %.1f%%",
3880                      h->stat.i_mb_cbp[1] * 100.0 / ((i_mb_count - i_all_intra)*4),
3881                      h->stat.i_mb_cbp[3] * 100.0 / ((i_mb_count - i_all_intra)*csize),
3882                      h->stat.i_mb_cbp[5] * 100.0 / ((i_mb_count - i_all_intra)*csize) );
3883         x264_log( h, X264_LOG_INFO, "coded y,%s,%s intra: %.1f%% %.1f%% %.1f%%%s\n",
3884                   CHROMA444?"u":"uvDC", CHROMA444?"v":"uvAC",
3885                   h->stat.i_mb_cbp[0] * 100.0 / (i_all_intra*4),
3886                   h->stat.i_mb_cbp[2] * 100.0 / (i_all_intra*csize),
3887                   h->stat.i_mb_cbp[4] * 100.0 / (i_all_intra*csize), buf );
3888
3889         int64_t fixed_pred_modes[4][9] = {{0}};
3890         int64_t sum_pred_modes[4] = {0};
3891         for( int i = 0; i <= I_PRED_16x16_DC_128; i++ )
3892         {
3893             fixed_pred_modes[0][x264_mb_pred_mode16x16_fix[i]] += h->stat.i_mb_pred_mode[0][i];
3894             sum_pred_modes[0] += h->stat.i_mb_pred_mode[0][i];
3895         }
3896         if( sum_pred_modes[0] )
3897             x264_log( h, X264_LOG_INFO, "i16 v,h,dc,p: %2.0f%% %2.0f%% %2.0f%% %2.0f%%\n",
3898                       fixed_pred_modes[0][0] * 100.0 / sum_pred_modes[0],
3899                       fixed_pred_modes[0][1] * 100.0 / sum_pred_modes[0],
3900                       fixed_pred_modes[0][2] * 100.0 / sum_pred_modes[0],
3901                       fixed_pred_modes[0][3] * 100.0 / sum_pred_modes[0] );
3902         for( int i = 1; i <= 2; i++ )
3903         {
3904             for( int j = 0; j <= I_PRED_8x8_DC_128; j++ )
3905             {
3906                 fixed_pred_modes[i][x264_mb_pred_mode4x4_fix(j)] += h->stat.i_mb_pred_mode[i][j];
3907                 sum_pred_modes[i] += h->stat.i_mb_pred_mode[i][j];
3908             }
3909             if( sum_pred_modes[i] )
3910                 x264_log( h, X264_LOG_INFO, "i%d v,h,dc,ddl,ddr,vr,hd,vl,hu: %2.0f%% %2.0f%% %2.0f%% %2.0f%% %2.0f%% %2.0f%% %2.0f%% %2.0f%% %2.0f%%\n", (3-i)*4,
3911                           fixed_pred_modes[i][0] * 100.0 / sum_pred_modes[i],
3912                           fixed_pred_modes[i][1] * 100.0 / sum_pred_modes[i],
3913                           fixed_pred_modes[i][2] * 100.0 / sum_pred_modes[i],
3914                           fixed_pred_modes[i][3] * 100.0 / sum_pred_modes[i],
3915                           fixed_pred_modes[i][4] * 100.0 / sum_pred_modes[i],
3916                           fixed_pred_modes[i][5] * 100.0 / sum_pred_modes[i],
3917                           fixed_pred_modes[i][6] * 100.0 / sum_pred_modes[i],
3918                           fixed_pred_modes[i][7] * 100.0 / sum_pred_modes[i],
3919                           fixed_pred_modes[i][8] * 100.0 / sum_pred_modes[i] );
3920         }
3921         for( int i = 0; i <= I_PRED_CHROMA_DC_128; i++ )
3922         {
3923             fixed_pred_modes[3][x264_mb_chroma_pred_mode_fix[i]] += h->stat.i_mb_pred_mode[3][i];
3924             sum_pred_modes[3] += h->stat.i_mb_pred_mode[3][i];
3925         }
3926         if( sum_pred_modes[3] && !CHROMA444 )
3927             x264_log( h, X264_LOG_INFO, "i8c dc,h,v,p: %2.0f%% %2.0f%% %2.0f%% %2.0f%%\n",
3928                       fixed_pred_modes[3][0] * 100.0 / sum_pred_modes[3],
3929                       fixed_pred_modes[3][1] * 100.0 / sum_pred_modes[3],
3930                       fixed_pred_modes[3][2] * 100.0 / sum_pred_modes[3],
3931                       fixed_pred_modes[3][3] * 100.0 / sum_pred_modes[3] );
3932
3933         if( h->param.analyse.i_weighted_pred >= X264_WEIGHTP_SIMPLE && h->stat.i_frame_count[SLICE_TYPE_P] > 0 )
3934             x264_log( h, X264_LOG_INFO, "Weighted P-Frames: Y:%.1f%% UV:%.1f%%\n",
3935                       h->stat.i_wpred[0] * 100.0 / h->stat.i_frame_count[SLICE_TYPE_P],
3936                       h->stat.i_wpred[1] * 100.0 / h->stat.i_frame_count[SLICE_TYPE_P] );
3937
3938         for( int i_list = 0; i_list < 2; i_list++ )
3939             for( int i_slice = 0; i_slice < 2; i_slice++ )
3940             {
3941                 char *p = buf;
3942                 int64_t i_den = 0;
3943                 int i_max = 0;
3944                 for( int i = 0; i < X264_REF_MAX*2; i++ )
3945                     if( h->stat.i_mb_count_ref[i_slice][i_list][i] )
3946                     {
3947                         i_den += h->stat.i_mb_count_ref[i_slice][i_list][i];
3948                         i_max = i;
3949                     }
3950                 if( i_max == 0 )
3951                     continue;
3952                 for( int i = 0; i <= i_max; i++ )
3953                     p += sprintf( p, " %4.1f%%", 100. * h->stat.i_mb_count_ref[i_slice][i_list][i] / i_den );
3954                 x264_log( h, X264_LOG_INFO, "ref %c L%d:%s\n", "PB"[i_slice], i_list, buf );
3955             }
3956
3957         if( h->param.analyse.b_ssim )
3958         {
3959             float ssim = SUM3( h->stat.f_ssim_mean_y ) / duration;
3960             x264_log( h, X264_LOG_INFO, "SSIM Mean Y:%.7f (%6.3fdb)\n", ssim, x264_ssim( ssim ) );
3961         }
3962         if( h->param.analyse.b_psnr )
3963         {
3964             x264_log( h, X264_LOG_INFO,
3965                       "PSNR Mean Y:%6.3f U:%6.3f V:%6.3f Avg:%6.3f Global:%6.3f kb/s:%.2f\n",
3966                       SUM3( h->stat.f_psnr_mean_y ) / duration,
3967                       SUM3( h->stat.f_psnr_mean_u ) / duration,
3968                       SUM3( h->stat.f_psnr_mean_v ) / duration,
3969                       SUM3( h->stat.f_psnr_average ) / duration,
3970                       x264_psnr( SUM3( h->stat.f_ssd_global ), duration * i_yuv_size ),
3971                       f_bitrate );
3972         }
3973         else
3974             x264_log( h, X264_LOG_INFO, "kb/s:%.2f\n", f_bitrate );
3975     }
3976
3977     /* rc */
3978     x264_ratecontrol_delete( h );
3979
3980     /* param */
3981     if( h->param.rc.psz_stat_out )
3982         free( h->param.rc.psz_stat_out );
3983     if( h->param.rc.psz_stat_in )
3984         free( h->param.rc.psz_stat_in );
3985
3986     x264_cqm_delete( h );
3987     x264_free( h->nal_buffer );
3988     x264_analyse_free_costs( h );
3989
3990     if( h->i_thread_frames > 1 )
3991         h = h->thread[h->i_thread_phase];
3992
3993     /* frames */
3994     x264_frame_delete_list( h->frames.unused[0] );
3995     x264_frame_delete_list( h->frames.unused[1] );
3996     x264_frame_delete_list( h->frames.current );
3997     x264_frame_delete_list( h->frames.blank_unused );
3998
3999     h = h->thread[0];
4000
4001     for( int i = 0; i < h->i_thread_frames; i++ )
4002         if( h->thread[i]->b_thread_active )
4003             for( int j = 0; j < h->thread[i]->i_ref[0]; j++ )
4004                 if( h->thread[i]->fref[0][j] && h->thread[i]->fref[0][j]->b_duplicate )
4005                     x264_frame_delete( h->thread[i]->fref[0][j] );
4006
4007     if( h->param.i_lookahead_threads > 1 )
4008         for( int i = 0; i < h->param.i_lookahead_threads; i++ )
4009             x264_free( h->lookahead_thread[i] );
4010
4011     for( int i = h->param.i_threads - 1; i >= 0; i-- )
4012     {
4013         x264_frame_t **frame;
4014
4015         if( !h->param.b_sliced_threads || i == 0 )
4016         {
4017             for( frame = h->thread[i]->frames.reference; *frame; frame++ )
4018             {
4019                 assert( (*frame)->i_reference_count > 0 );
4020                 (*frame)->i_reference_count--;
4021                 if( (*frame)->i_reference_count == 0 )
4022                     x264_frame_delete( *frame );
4023             }
4024             frame = &h->thread[i]->fdec;
4025             if( *frame )
4026             {
4027                 assert( (*frame)->i_reference_count > 0 );
4028                 (*frame)->i_reference_count--;
4029                 if( (*frame)->i_reference_count == 0 )
4030                     x264_frame_delete( *frame );
4031             }
4032             x264_macroblock_cache_free( h->thread[i] );
4033         }
4034         x264_macroblock_thread_free( h->thread[i], 0 );
4035         x264_free( h->thread[i]->out.p_bitstream );
4036         x264_free( h->thread[i]->out.nal );
4037         x264_pthread_mutex_destroy( &h->thread[i]->mutex );
4038         x264_pthread_cond_destroy( &h->thread[i]->cv );
4039         x264_free( h->thread[i] );
4040     }
4041 #if HAVE_OPENCL
4042     x264_opencl_close_library( ocl );
4043 #endif
4044 }
4045
4046 int x264_encoder_delayed_frames( x264_t *h )
4047 {
4048     int delayed_frames = 0;
4049     if( h->i_thread_frames > 1 )
4050     {
4051         for( int i = 0; i < h->i_thread_frames; i++ )
4052             delayed_frames += h->thread[i]->b_thread_active;
4053         h = h->thread[h->i_thread_phase];
4054     }
4055     for( int i = 0; h->frames.current[i]; i++ )
4056         delayed_frames++;
4057     x264_pthread_mutex_lock( &h->lookahead->ofbuf.mutex );
4058     x264_pthread_mutex_lock( &h->lookahead->ifbuf.mutex );
4059     x264_pthread_mutex_lock( &h->lookahead->next.mutex );
4060     delayed_frames += h->lookahead->ifbuf.i_size + h->lookahead->next.i_size + h->lookahead->ofbuf.i_size;
4061     x264_pthread_mutex_unlock( &h->lookahead->next.mutex );
4062     x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
4063     x264_pthread_mutex_unlock( &h->lookahead->ofbuf.mutex );
4064     return delayed_frames;
4065 }
4066
4067 int x264_encoder_maximum_delayed_frames( x264_t *h )
4068 {
4069     return h->frames.i_delay;
4070 }