OpenCL support improvement/refactoring
[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( x264_t *h )
357 {
358     uint8_t *bs_bak = h->out.p_bitstream;
359     int max_row_size = (2500 << SLICE_MBAFF) * h->mb.i_mb_width;
360     if( (h->param.b_cabac && (h->cabac.p_end - h->cabac.p < max_row_size)) ||
361         (h->out.bs.p_end - h->out.bs.p < max_row_size) )
362     {
363         h->out.i_bitstream += max_row_size;
364         CHECKED_MALLOC( h->out.p_bitstream, h->out.i_bitstream );
365         h->mc.memcpy_aligned( h->out.p_bitstream, bs_bak, (h->out.i_bitstream - max_row_size) & ~15 );
366         intptr_t delta = h->out.p_bitstream - bs_bak;
367
368         h->out.bs.p_start += delta;
369         h->out.bs.p += delta;
370         h->out.bs.p_end = h->out.p_bitstream + h->out.i_bitstream;
371
372         h->cabac.p_start += delta;
373         h->cabac.p += delta;
374         h->cabac.p_end = h->out.p_bitstream + h->out.i_bitstream;
375
376         for( int i = 0; i <= h->out.i_nal; i++ )
377             h->out.nal[i].p_payload += delta;
378         x264_free( bs_bak );
379     }
380     return 0;
381 fail:
382     x264_free( bs_bak );
383     return -1;
384 }
385
386 #if HAVE_THREAD
387 static void x264_encoder_thread_init( x264_t *h )
388 {
389     if( h->param.i_sync_lookahead )
390         x264_lower_thread_priority( 10 );
391
392 #if HAVE_MMX
393     /* Misalign mask has to be set separately for each thread. */
394     if( h->param.cpu&X264_CPU_SSE_MISALIGN )
395         x264_cpu_mask_misalign_sse();
396 #endif
397 }
398
399 static void x264_lookahead_thread_init( x264_t *h )
400 {
401 #if HAVE_MMX
402     /* Misalign mask has to be set separately for each thread. */
403     if( h->param.cpu&X264_CPU_SSE_MISALIGN )
404         x264_cpu_mask_misalign_sse();
405 #endif
406 }
407 #endif
408
409 /****************************************************************************
410  *
411  ****************************************************************************
412  ****************************** External API*********************************
413  ****************************************************************************
414  *
415  ****************************************************************************/
416
417 static int x264_validate_parameters( x264_t *h, int b_open )
418 {
419 #if HAVE_MMX
420     if( b_open )
421     {
422         int cpuflags = x264_cpu_detect();
423         int fail = 0;
424 #ifdef __SSE__
425         if( !(cpuflags & X264_CPU_SSE) )
426         {
427             x264_log( h, X264_LOG_ERROR, "your cpu does not support SSE1, but x264 was compiled with asm\n");
428             fail = 1;
429         }
430 #else
431         if( !(cpuflags & X264_CPU_MMX2) )
432         {
433             x264_log( h, X264_LOG_ERROR, "your cpu does not support MMXEXT, but x264 was compiled with asm\n");
434             fail = 1;
435         }
436 #endif
437         if( !fail && !(cpuflags & X264_CPU_CMOV) )
438         {
439             x264_log( h, X264_LOG_ERROR, "your cpu does not support CMOV, but x264 was compiled with asm\n");
440             fail = 1;
441         }
442         if( fail )
443         {
444             x264_log( h, X264_LOG_ERROR, "to run x264, recompile without asm (configure --disable-asm)\n");
445             return -1;
446         }
447     }
448 #endif
449
450 #if HAVE_INTERLACED
451     h->param.b_interlaced = !!PARAM_INTERLACED;
452 #else
453     if( h->param.b_interlaced )
454     {
455         x264_log( h, X264_LOG_ERROR, "not compiled with interlaced support\n" );
456         return -1;
457     }
458 #endif
459
460     if( h->param.i_width <= 0 || h->param.i_height <= 0 )
461     {
462         x264_log( h, X264_LOG_ERROR, "invalid width x height (%dx%d)\n",
463                   h->param.i_width, h->param.i_height );
464         return -1;
465     }
466
467     int i_csp = h->param.i_csp & X264_CSP_MASK;
468 #if X264_CHROMA_FORMAT
469     if( CHROMA_FORMAT != CHROMA_420 && i_csp >= X264_CSP_I420 && i_csp <= X264_CSP_NV12 )
470     {
471         x264_log( h, X264_LOG_ERROR, "not compiled with 4:2:0 support\n" );
472         return -1;
473     }
474     else if( CHROMA_FORMAT != CHROMA_422 && i_csp >= X264_CSP_I422 && i_csp <= X264_CSP_NV16 )
475     {
476         x264_log( h, X264_LOG_ERROR, "not compiled with 4:2:2 support\n" );
477         return -1;
478     }
479     else if( CHROMA_FORMAT != CHROMA_444 && i_csp >= X264_CSP_I444 && i_csp <= X264_CSP_RGB )
480     {
481         x264_log( h, X264_LOG_ERROR, "not compiled with 4:4:4 support\n" );
482         return -1;
483     }
484 #endif
485     if( i_csp <= X264_CSP_NONE || i_csp >= X264_CSP_MAX )
486     {
487         x264_log( h, X264_LOG_ERROR, "invalid CSP (only I420/YV12/NV12/I422/YV16/NV16/I444/YV24/BGR/BGRA/RGB supported)\n" );
488         return -1;
489     }
490
491     if( i_csp < X264_CSP_I444 && h->param.i_width % 2 )
492     {
493         x264_log( h, X264_LOG_ERROR, "width not divisible by 2 (%dx%d)\n",
494                   h->param.i_width, h->param.i_height );
495         return -1;
496     }
497
498     if( i_csp < X264_CSP_I422 && PARAM_INTERLACED && h->param.i_height % 4 )
499     {
500         x264_log( h, X264_LOG_ERROR, "height not divisible by 4 (%dx%d)\n",
501                   h->param.i_width, h->param.i_height );
502         return -1;
503     }
504
505     if( (i_csp < X264_CSP_I422 || PARAM_INTERLACED) && h->param.i_height % 2 )
506     {
507         x264_log( h, X264_LOG_ERROR, "height not divisible by 2 (%dx%d)\n",
508                   h->param.i_width, h->param.i_height );
509         return -1;
510     }
511
512     if( (h->param.crop_rect.i_left + h->param.crop_rect.i_right ) >= h->param.i_width ||
513         (h->param.crop_rect.i_top  + h->param.crop_rect.i_bottom) >= h->param.i_height )
514     {
515         x264_log( h, X264_LOG_ERROR, "invalid crop-rect %u,%u,%u,%u\n", h->param.crop_rect.i_left,
516                   h->param.crop_rect.i_top, h->param.crop_rect.i_right,  h->param.crop_rect.i_bottom );
517         return -1;
518     }
519
520     if( h->param.i_threads == X264_THREADS_AUTO )
521         h->param.i_threads = x264_cpu_num_processors() * (h->param.b_sliced_threads?2:3)/2;
522     int max_sliced_threads = X264_MAX( 1, (h->param.i_height+15)/16 / 4 );
523     if( h->param.i_threads > 1 )
524     {
525 #if !HAVE_THREAD
526         x264_log( h, X264_LOG_WARNING, "not compiled with thread support!\n");
527         h->param.i_threads = 1;
528 #endif
529         /* Avoid absurdly small thread slices as they can reduce performance
530          * and VBV compliance.  Capped at an arbitrary 4 rows per thread. */
531         if( h->param.b_sliced_threads )
532             h->param.i_threads = X264_MIN( h->param.i_threads, max_sliced_threads );
533     }
534     h->param.i_threads = x264_clip3( h->param.i_threads, 1, X264_THREAD_MAX );
535     if( h->param.i_threads == 1 )
536     {
537         h->param.b_sliced_threads = 0;
538         h->param.i_lookahead_threads = 1;
539     }
540     h->i_thread_frames = h->param.b_sliced_threads ? 1 : h->param.i_threads;
541     if( h->i_thread_frames > 1 )
542         h->param.nalu_process = NULL;
543
544     if( h->param.b_opencl )
545     {
546 #if !HAVE_OPENCL
547         x264_log( h, X264_LOG_WARNING, "OpenCL: not compiled with OpenCL support, disabling\n" );
548         h->param.b_opencl = 0;
549 #elif BIT_DEPTH > 8
550         x264_log( h, X264_LOG_WARNING, "OpenCL lookahead does not support high bit depth, disabling opencl\n" );
551         h->param.b_opencl = 0;
552 #else
553         if( h->param.i_width < 32 || h->param.i_height < 32 )
554         {
555             x264_log( h, X264_LOG_WARNING, "OpenCL: frame size is too small, disabling opencl\n" );
556             h->param.b_opencl = 0;
557         }
558 #endif
559         if( h->param.opencl_device_id && h->param.i_opencl_device )
560         {
561             x264_log( h, X264_LOG_WARNING, "OpenCL: device id and device skip count configured; dropping skip\n" );
562             h->param.i_opencl_device = 0;
563         }
564     }
565
566     h->param.i_keyint_max = x264_clip3( h->param.i_keyint_max, 1, X264_KEYINT_MAX_INFINITE );
567     if( h->param.i_keyint_max == 1 )
568     {
569         h->param.b_intra_refresh = 0;
570         h->param.analyse.i_weighted_pred = 0;
571     }
572
573     h->param.i_frame_packing = x264_clip3( h->param.i_frame_packing, -1, 5 );
574
575     /* Detect default ffmpeg settings and terminate with an error. */
576     if( b_open )
577     {
578         int score = 0;
579         score += h->param.analyse.i_me_range == 0;
580         score += h->param.rc.i_qp_step == 3;
581         score += h->param.i_keyint_max == 12;
582         score += h->param.rc.i_qp_min == 2;
583         score += h->param.rc.i_qp_max == 31;
584         score += h->param.rc.f_qcompress == 0.5;
585         score += fabs(h->param.rc.f_ip_factor - 1.25) < 0.01;
586         score += fabs(h->param.rc.f_pb_factor - 1.25) < 0.01;
587         score += h->param.analyse.inter == 0 && h->param.analyse.i_subpel_refine == 8;
588         if( score >= 5 )
589         {
590             x264_log( h, X264_LOG_ERROR, "broken ffmpeg default settings detected\n" );
591             x264_log( h, X264_LOG_ERROR, "use an encoding preset (e.g. -vpre medium)\n" );
592             x264_log( h, X264_LOG_ERROR, "preset usage: -vpre <speed> -vpre <profile>\n" );
593             x264_log( h, X264_LOG_ERROR, "speed presets are listed in x264 --help\n" );
594             x264_log( h, X264_LOG_ERROR, "profile is optional; x264 defaults to high\n" );
595             return -1;
596         }
597     }
598
599     if( h->param.rc.i_rc_method < 0 || h->param.rc.i_rc_method > 2 )
600     {
601         x264_log( h, X264_LOG_ERROR, "no ratecontrol method specified\n" );
602         return -1;
603     }
604     h->param.rc.f_rf_constant = x264_clip3f( h->param.rc.f_rf_constant, -QP_BD_OFFSET, 51 );
605     h->param.rc.f_rf_constant_max = x264_clip3f( h->param.rc.f_rf_constant_max, -QP_BD_OFFSET, 51 );
606     h->param.rc.i_qp_constant = x264_clip3( h->param.rc.i_qp_constant, 0, QP_MAX );
607     h->param.analyse.i_subpel_refine = x264_clip3( h->param.analyse.i_subpel_refine, 0, 11 );
608     h->param.rc.f_ip_factor = X264_MAX( h->param.rc.f_ip_factor, 0.01f );
609     h->param.rc.f_pb_factor = X264_MAX( h->param.rc.f_pb_factor, 0.01f );
610     if( h->param.rc.i_rc_method == X264_RC_CRF )
611     {
612         h->param.rc.i_qp_constant = h->param.rc.f_rf_constant + QP_BD_OFFSET;
613         h->param.rc.i_bitrate = 0;
614     }
615     if( b_open && (h->param.rc.i_rc_method == X264_RC_CQP || h->param.rc.i_rc_method == X264_RC_CRF)
616         && h->param.rc.i_qp_constant == 0 )
617     {
618         h->mb.b_lossless = 1;
619         h->param.i_cqm_preset = X264_CQM_FLAT;
620         h->param.psz_cqm_file = NULL;
621         h->param.rc.i_rc_method = X264_RC_CQP;
622         h->param.rc.f_ip_factor = 1;
623         h->param.rc.f_pb_factor = 1;
624         h->param.analyse.b_psnr = 0;
625         h->param.analyse.b_ssim = 0;
626         h->param.analyse.i_chroma_qp_offset = 0;
627         h->param.analyse.i_trellis = 0;
628         h->param.analyse.b_fast_pskip = 0;
629         h->param.analyse.i_noise_reduction = 0;
630         h->param.analyse.b_psy = 0;
631         h->param.i_bframe = 0;
632         /* 8x8dct is not useful without RD in CAVLC lossless */
633         if( !h->param.b_cabac && h->param.analyse.i_subpel_refine < 6 )
634             h->param.analyse.b_transform_8x8 = 0;
635     }
636     if( h->param.rc.i_rc_method == X264_RC_CQP )
637     {
638         float qp_p = h->param.rc.i_qp_constant;
639         float qp_i = qp_p - 6*log2f( h->param.rc.f_ip_factor );
640         float qp_b = qp_p + 6*log2f( h->param.rc.f_pb_factor );
641         h->param.rc.i_qp_min = x264_clip3( (int)(X264_MIN3( qp_p, qp_i, qp_b )), 0, QP_MAX );
642         h->param.rc.i_qp_max = x264_clip3( (int)(X264_MAX3( qp_p, qp_i, qp_b ) + .999), 0, QP_MAX );
643         h->param.rc.i_aq_mode = 0;
644         h->param.rc.b_mb_tree = 0;
645         h->param.rc.i_bitrate = 0;
646     }
647     h->param.rc.i_qp_max = x264_clip3( h->param.rc.i_qp_max, 0, QP_MAX );
648     h->param.rc.i_qp_min = x264_clip3( h->param.rc.i_qp_min, 0, h->param.rc.i_qp_max );
649     h->param.rc.i_qp_step = x264_clip3( h->param.rc.i_qp_step, 2, QP_MAX );
650     h->param.rc.i_bitrate = x264_clip3( h->param.rc.i_bitrate, 0, 2000000 );
651     if( h->param.rc.i_rc_method == X264_RC_ABR && !h->param.rc.i_bitrate )
652     {
653         x264_log( h, X264_LOG_ERROR, "bitrate not specified\n" );
654         return -1;
655     }
656     h->param.rc.i_vbv_buffer_size = x264_clip3( h->param.rc.i_vbv_buffer_size, 0, 2000000 );
657     h->param.rc.i_vbv_max_bitrate = x264_clip3( h->param.rc.i_vbv_max_bitrate, 0, 2000000 );
658     h->param.rc.f_vbv_buffer_init = x264_clip3f( h->param.rc.f_vbv_buffer_init, 0, 2000000 );
659     if( h->param.rc.i_vbv_buffer_size )
660     {
661         if( h->param.rc.i_rc_method == X264_RC_CQP )
662         {
663             x264_log( h, X264_LOG_WARNING, "VBV is incompatible with constant QP, ignored.\n" );
664             h->param.rc.i_vbv_max_bitrate = 0;
665             h->param.rc.i_vbv_buffer_size = 0;
666         }
667         else if( h->param.rc.i_vbv_max_bitrate == 0 )
668         {
669             if( h->param.rc.i_rc_method == X264_RC_ABR )
670             {
671                 x264_log( h, X264_LOG_WARNING, "VBV maxrate unspecified, assuming CBR\n" );
672                 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate;
673             }
674             else
675             {
676                 x264_log( h, X264_LOG_WARNING, "VBV bufsize set but maxrate unspecified, ignored\n" );
677                 h->param.rc.i_vbv_buffer_size = 0;
678             }
679         }
680         else if( h->param.rc.i_vbv_max_bitrate < h->param.rc.i_bitrate &&
681                  h->param.rc.i_rc_method == X264_RC_ABR )
682         {
683             x264_log( h, X264_LOG_WARNING, "max bitrate less than average bitrate, assuming CBR\n" );
684             h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate;
685         }
686     }
687     else if( h->param.rc.i_vbv_max_bitrate )
688     {
689         x264_log( h, X264_LOG_WARNING, "VBV maxrate specified, but no bufsize, ignored\n" );
690         h->param.rc.i_vbv_max_bitrate = 0;
691     }
692
693     h->param.i_slice_max_size = X264_MAX( h->param.i_slice_max_size, 0 );
694     h->param.i_slice_max_mbs = X264_MAX( h->param.i_slice_max_mbs, 0 );
695     h->param.i_slice_min_mbs = X264_MAX( h->param.i_slice_min_mbs, 0 );
696     if( h->param.i_slice_max_mbs )
697         h->param.i_slice_min_mbs = X264_MIN( h->param.i_slice_min_mbs, h->param.i_slice_max_mbs/2 );
698     else if( !h->param.i_slice_max_size )
699         h->param.i_slice_min_mbs = 0;
700     if( PARAM_INTERLACED && h->param.i_slice_min_mbs )
701     {
702         x264_log( h, X264_LOG_WARNING, "interlace + slice-min-mbs is not implemented\n" );
703         h->param.i_slice_min_mbs = 0;
704     }
705     int mb_width = (h->param.i_width+15)/16;
706     if( h->param.i_slice_min_mbs > mb_width )
707     {
708         x264_log( h, X264_LOG_WARNING, "slice-min-mbs > row mb size (%d) not implemented\n", mb_width );
709         h->param.i_slice_min_mbs = mb_width;
710     }
711
712     int max_slices = (h->param.i_height+((16<<PARAM_INTERLACED)-1))/(16<<PARAM_INTERLACED);
713     if( h->param.b_sliced_threads )
714         h->param.i_slice_count = x264_clip3( h->param.i_threads, 0, max_slices );
715     else
716     {
717         h->param.i_slice_count = x264_clip3( h->param.i_slice_count, 0, max_slices );
718         if( h->param.i_slice_max_mbs || h->param.i_slice_max_size )
719             h->param.i_slice_count = 0;
720     }
721     if( h->param.i_slice_count_max > 0 )
722         h->param.i_slice_count_max = X264_MAX( h->param.i_slice_count, h->param.i_slice_count_max );
723
724     if( h->param.b_bluray_compat )
725     {
726         h->param.i_bframe_pyramid = X264_MIN( X264_B_PYRAMID_STRICT, h->param.i_bframe_pyramid );
727         h->param.i_bframe = X264_MIN( h->param.i_bframe, 3 );
728         h->param.b_aud = 1;
729         h->param.i_nal_hrd = X264_MAX( h->param.i_nal_hrd, X264_NAL_HRD_VBR );
730         h->param.i_slice_max_size = 0;
731         h->param.i_slice_max_mbs = 0;
732         h->param.b_intra_refresh = 0;
733         h->param.i_frame_reference = X264_MIN( h->param.i_frame_reference, 6 );
734         h->param.i_dpb_size = X264_MIN( h->param.i_dpb_size, 6 );
735         /* Don't use I-frames, because Blu-ray treats them the same as IDR. */
736         h->param.i_keyint_min = 1;
737         /* Due to the proliferation of broken players that don't handle dupes properly. */
738         h->param.analyse.i_weighted_pred = X264_MIN( h->param.analyse.i_weighted_pred, X264_WEIGHTP_SIMPLE );
739         if( h->param.b_fake_interlaced )
740             h->param.b_pic_struct = 1;
741     }
742
743     h->param.i_frame_reference = x264_clip3( h->param.i_frame_reference, 1, X264_REF_MAX );
744     h->param.i_dpb_size = x264_clip3( h->param.i_dpb_size, 1, X264_REF_MAX );
745     if( h->param.i_scenecut_threshold < 0 )
746         h->param.i_scenecut_threshold = 0;
747     h->param.analyse.i_direct_mv_pred = x264_clip3( h->param.analyse.i_direct_mv_pred, X264_DIRECT_PRED_NONE, X264_DIRECT_PRED_AUTO );
748     if( !h->param.analyse.i_subpel_refine && h->param.analyse.i_direct_mv_pred > X264_DIRECT_PRED_SPATIAL )
749     {
750         x264_log( h, X264_LOG_WARNING, "subme=0 + direct=temporal is not supported\n" );
751         h->param.analyse.i_direct_mv_pred = X264_DIRECT_PRED_SPATIAL;
752     }
753     h->param.i_bframe = x264_clip3( h->param.i_bframe, 0, X264_MIN( X264_BFRAME_MAX, h->param.i_keyint_max-1 ) );
754     h->param.i_bframe_bias = x264_clip3( h->param.i_bframe_bias, -90, 100 );
755     if( h->param.i_bframe <= 1 )
756         h->param.i_bframe_pyramid = X264_B_PYRAMID_NONE;
757     h->param.i_bframe_pyramid = x264_clip3( h->param.i_bframe_pyramid, X264_B_PYRAMID_NONE, X264_B_PYRAMID_NORMAL );
758     h->param.i_bframe_adaptive = x264_clip3( h->param.i_bframe_adaptive, X264_B_ADAPT_NONE, X264_B_ADAPT_TRELLIS );
759     if( !h->param.i_bframe )
760     {
761         h->param.i_bframe_adaptive = X264_B_ADAPT_NONE;
762         h->param.analyse.i_direct_mv_pred = 0;
763         h->param.analyse.b_weighted_bipred = 0;
764         h->param.b_open_gop = 0;
765     }
766     if( h->param.b_intra_refresh && h->param.i_bframe_pyramid == X264_B_PYRAMID_NORMAL )
767     {
768         x264_log( h, X264_LOG_WARNING, "b-pyramid normal + intra-refresh is not supported\n" );
769         h->param.i_bframe_pyramid = X264_B_PYRAMID_STRICT;
770     }
771     if( h->param.b_intra_refresh && (h->param.i_frame_reference > 1 || h->param.i_dpb_size > 1) )
772     {
773         x264_log( h, X264_LOG_WARNING, "ref > 1 + intra-refresh is not supported\n" );
774         h->param.i_frame_reference = 1;
775         h->param.i_dpb_size = 1;
776     }
777     if( h->param.b_intra_refresh && h->param.b_open_gop )
778     {
779         x264_log( h, X264_LOG_WARNING, "intra-refresh is not compatible with open-gop\n" );
780         h->param.b_open_gop = 0;
781     }
782     if( !h->param.i_fps_num || !h->param.i_fps_den )
783     {
784         h->param.i_fps_num = 25;
785         h->param.i_fps_den = 1;
786     }
787     float fps = (float) h->param.i_fps_num / h->param.i_fps_den;
788     if( h->param.i_keyint_min == X264_KEYINT_MIN_AUTO )
789         h->param.i_keyint_min = X264_MIN( h->param.i_keyint_max / 10, fps );
790     h->param.i_keyint_min = x264_clip3( h->param.i_keyint_min, 1, h->param.i_keyint_max/2+1 );
791     h->param.rc.i_lookahead = x264_clip3( h->param.rc.i_lookahead, 0, X264_LOOKAHEAD_MAX );
792     {
793         int maxrate = X264_MAX( h->param.rc.i_vbv_max_bitrate, h->param.rc.i_bitrate );
794         float bufsize = maxrate ? (float)h->param.rc.i_vbv_buffer_size / maxrate : 0;
795         h->param.rc.i_lookahead = X264_MIN( h->param.rc.i_lookahead, X264_MAX( h->param.i_keyint_max, bufsize*fps ) );
796     }
797
798     if( !h->param.i_timebase_num || !h->param.i_timebase_den || !(h->param.b_vfr_input || h->param.b_pulldown) )
799     {
800         h->param.i_timebase_num = h->param.i_fps_den;
801         h->param.i_timebase_den = h->param.i_fps_num;
802     }
803
804     h->param.rc.f_qcompress = x264_clip3f( h->param.rc.f_qcompress, 0.0, 1.0 );
805     if( h->param.i_keyint_max == 1 || h->param.rc.f_qcompress == 1 )
806         h->param.rc.b_mb_tree = 0;
807     if( (!h->param.b_intra_refresh && h->param.i_keyint_max != X264_KEYINT_MAX_INFINITE) &&
808         !h->param.rc.i_lookahead && h->param.rc.b_mb_tree )
809     {
810         x264_log( h, X264_LOG_WARNING, "lookaheadless mb-tree requires intra refresh or infinite keyint\n" );
811         h->param.rc.b_mb_tree = 0;
812     }
813     if( b_open && h->param.rc.b_stat_read )
814         h->param.rc.i_lookahead = 0;
815 #if HAVE_THREAD
816     if( h->param.i_sync_lookahead < 0 )
817         h->param.i_sync_lookahead = h->param.i_bframe + 1;
818     h->param.i_sync_lookahead = X264_MIN( h->param.i_sync_lookahead, X264_LOOKAHEAD_MAX );
819     if( h->param.rc.b_stat_read || h->i_thread_frames == 1 )
820         h->param.i_sync_lookahead = 0;
821 #else
822     h->param.i_sync_lookahead = 0;
823 #endif
824
825     h->param.i_deblocking_filter_alphac0 = x264_clip3( h->param.i_deblocking_filter_alphac0, -6, 6 );
826     h->param.i_deblocking_filter_beta    = x264_clip3( h->param.i_deblocking_filter_beta, -6, 6 );
827     h->param.analyse.i_luma_deadzone[0] = x264_clip3( h->param.analyse.i_luma_deadzone[0], 0, 32 );
828     h->param.analyse.i_luma_deadzone[1] = x264_clip3( h->param.analyse.i_luma_deadzone[1], 0, 32 );
829
830     h->param.i_cabac_init_idc = x264_clip3( h->param.i_cabac_init_idc, 0, 2 );
831
832     if( h->param.i_cqm_preset < X264_CQM_FLAT || h->param.i_cqm_preset > X264_CQM_CUSTOM )
833         h->param.i_cqm_preset = X264_CQM_FLAT;
834
835     if( h->param.analyse.i_me_method < X264_ME_DIA ||
836         h->param.analyse.i_me_method > X264_ME_TESA )
837         h->param.analyse.i_me_method = X264_ME_HEX;
838     h->param.analyse.i_me_range = x264_clip3( h->param.analyse.i_me_range, 4, 1024 );
839     if( h->param.analyse.i_me_range > 16 && h->param.analyse.i_me_method <= X264_ME_HEX )
840         h->param.analyse.i_me_range = 16;
841     if( h->param.analyse.i_me_method == X264_ME_TESA &&
842         (h->mb.b_lossless || h->param.analyse.i_subpel_refine <= 1) )
843         h->param.analyse.i_me_method = X264_ME_ESA;
844     h->param.analyse.b_mixed_references = h->param.analyse.b_mixed_references && h->param.i_frame_reference > 1;
845     h->param.analyse.inter &= X264_ANALYSE_PSUB16x16|X264_ANALYSE_PSUB8x8|X264_ANALYSE_BSUB16x16|
846                               X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
847     h->param.analyse.intra &= X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
848     if( !(h->param.analyse.inter & X264_ANALYSE_PSUB16x16) )
849         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
850     if( !h->param.analyse.b_transform_8x8 )
851     {
852         h->param.analyse.inter &= ~X264_ANALYSE_I8x8;
853         h->param.analyse.intra &= ~X264_ANALYSE_I8x8;
854     }
855     h->param.analyse.i_trellis = x264_clip3( h->param.analyse.i_trellis, 0, 2 );
856     h->param.rc.i_aq_mode = x264_clip3( h->param.rc.i_aq_mode, 0, 2 );
857     h->param.rc.f_aq_strength = x264_clip3f( h->param.rc.f_aq_strength, 0, 3 );
858     if( h->param.rc.f_aq_strength == 0 )
859         h->param.rc.i_aq_mode = 0;
860
861     if( h->param.i_log_level < X264_LOG_INFO )
862     {
863         h->param.analyse.b_psnr = 0;
864         h->param.analyse.b_ssim = 0;
865     }
866     /* Warn users trying to measure PSNR/SSIM with psy opts on. */
867     if( b_open && (h->param.analyse.b_psnr || h->param.analyse.b_ssim) )
868     {
869         char *s = NULL;
870
871         if( h->param.analyse.b_psy )
872         {
873             s = h->param.analyse.b_psnr ? "psnr" : "ssim";
874             x264_log( h, X264_LOG_WARNING, "--%s used with psy on: results will be invalid!\n", s );
875         }
876         else if( !h->param.rc.i_aq_mode && h->param.analyse.b_ssim )
877         {
878             x264_log( h, X264_LOG_WARNING, "--ssim used with AQ off: results will be invalid!\n" );
879             s = "ssim";
880         }
881         else if(  h->param.rc.i_aq_mode && h->param.analyse.b_psnr )
882         {
883             x264_log( h, X264_LOG_WARNING, "--psnr used with AQ on: results will be invalid!\n" );
884             s = "psnr";
885         }
886         if( s )
887             x264_log( h, X264_LOG_WARNING, "--tune %s should be used if attempting to benchmark %s!\n", s, s );
888     }
889
890     if( !h->param.analyse.b_psy )
891     {
892         h->param.analyse.f_psy_rd = 0;
893         h->param.analyse.f_psy_trellis = 0;
894     }
895     h->param.analyse.f_psy_rd = x264_clip3f( h->param.analyse.f_psy_rd, 0, 10 );
896     h->param.analyse.f_psy_trellis = x264_clip3f( h->param.analyse.f_psy_trellis, 0, 10 );
897     h->mb.i_psy_rd = h->param.analyse.i_subpel_refine >= 6 ? FIX8( h->param.analyse.f_psy_rd ) : 0;
898     h->mb.i_psy_trellis = h->param.analyse.i_trellis ? FIX8( h->param.analyse.f_psy_trellis / 4 ) : 0;
899     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -32, 32);
900     /* In 4:4:4 mode, chroma gets twice as much resolution, so we can halve its quality. */
901     if( b_open && i_csp >= X264_CSP_I444 && i_csp < X264_CSP_BGR && h->param.analyse.b_psy )
902         h->param.analyse.i_chroma_qp_offset += 6;
903     /* Psy RDO increases overall quantizers to improve the quality of luma--this indirectly hurts chroma quality */
904     /* so we lower the chroma QP offset to compensate */
905     if( b_open && h->mb.i_psy_rd )
906         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_rd < 0.25 ? 1 : 2;
907     /* Psy trellis has a similar effect. */
908     if( b_open && h->mb.i_psy_trellis )
909         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_trellis < 0.25 ? 1 : 2;
910     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -12, 12);
911     /* MB-tree requires AQ to be on, even if the strength is zero. */
912     if( !h->param.rc.i_aq_mode && h->param.rc.b_mb_tree )
913     {
914         h->param.rc.i_aq_mode = 1;
915         h->param.rc.f_aq_strength = 0;
916     }
917     h->param.analyse.i_noise_reduction = x264_clip3( h->param.analyse.i_noise_reduction, 0, 1<<16 );
918     if( h->param.analyse.i_subpel_refine >= 10 && (h->param.analyse.i_trellis != 2 || !h->param.rc.i_aq_mode) )
919         h->param.analyse.i_subpel_refine = 9;
920
921     {
922         const x264_level_t *l = x264_levels;
923         if( h->param.i_level_idc < 0 )
924         {
925             int maxrate_bak = h->param.rc.i_vbv_max_bitrate;
926             if( h->param.rc.i_rc_method == X264_RC_ABR && h->param.rc.i_vbv_buffer_size <= 0 )
927                 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate * 2;
928             x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
929             do h->param.i_level_idc = l->level_idc;
930                 while( l[1].level_idc && x264_validate_levels( h, 0 ) && l++ );
931             h->param.rc.i_vbv_max_bitrate = maxrate_bak;
932         }
933         else
934         {
935             while( l->level_idc && l->level_idc != h->param.i_level_idc )
936                 l++;
937             if( l->level_idc == 0 )
938             {
939                 x264_log( h, X264_LOG_ERROR, "invalid level_idc: %d\n", h->param.i_level_idc );
940                 return -1;
941             }
942         }
943         if( h->param.analyse.i_mv_range <= 0 )
944             h->param.analyse.i_mv_range = l->mv_range >> PARAM_INTERLACED;
945         else
946             h->param.analyse.i_mv_range = x264_clip3(h->param.analyse.i_mv_range, 32, 512 >> PARAM_INTERLACED);
947     }
948
949     h->param.analyse.i_weighted_pred = x264_clip3( h->param.analyse.i_weighted_pred, X264_WEIGHTP_NONE, X264_WEIGHTP_SMART );
950
951     if( h->param.i_lookahead_threads == X264_THREADS_AUTO )
952     {
953         if( h->param.b_sliced_threads )
954             h->param.i_lookahead_threads = h->param.i_threads;
955         else
956         {
957             /* If we're using much slower lookahead settings than encoding settings, it helps a lot to use
958              * more lookahead threads.  This typically happens in the first pass of a two-pass encode, so
959              * try to guess at this sort of case.
960              *
961              * Tuned by a little bit of real encoding with the various presets. */
962             int badapt = h->param.i_bframe_adaptive == X264_B_ADAPT_TRELLIS;
963             int subme = X264_MIN( h->param.analyse.i_subpel_refine / 3, 3 ) + (h->param.analyse.i_subpel_refine > 1);
964             int bframes = X264_MIN( (h->param.i_bframe - 1) / 3, 3 );
965
966             /* [b-adapt 0/1 vs 2][quantized subme][quantized bframes] */
967             static const uint8_t lookahead_thread_div[2][5][4] =
968             {{{6,6,6,6}, {3,3,3,3}, {4,4,4,4}, {6,6,6,6}, {12,12,12,12}},
969              {{3,2,1,1}, {2,1,1,1}, {4,3,2,1}, {6,4,3,2}, {12, 9, 6, 4}}};
970
971             h->param.i_lookahead_threads = h->param.i_threads / lookahead_thread_div[badapt][subme][bframes];
972             /* Since too many lookahead threads significantly degrades lookahead accuracy, limit auto
973              * lookahead threads to about 8 macroblock rows high each at worst.  This number is chosen
974              * pretty much arbitrarily. */
975             h->param.i_lookahead_threads = X264_MIN( h->param.i_lookahead_threads, h->param.i_height / 128 );
976         }
977     }
978     h->param.i_lookahead_threads = x264_clip3( h->param.i_lookahead_threads, 1, X264_MIN( max_sliced_threads, X264_LOOKAHEAD_THREAD_MAX ) );
979
980     if( PARAM_INTERLACED )
981     {
982         if( h->param.analyse.i_me_method >= X264_ME_ESA )
983         {
984             x264_log( h, X264_LOG_WARNING, "interlace + me=esa is not implemented\n" );
985             h->param.analyse.i_me_method = X264_ME_UMH;
986         }
987         if( h->param.analyse.i_weighted_pred > 0 )
988         {
989             x264_log( h, X264_LOG_WARNING, "interlace + weightp is not implemented\n" );
990             h->param.analyse.i_weighted_pred = X264_WEIGHTP_NONE;
991         }
992     }
993
994     if( !h->param.analyse.i_weighted_pred && h->param.rc.b_mb_tree && h->param.analyse.b_psy )
995         h->param.analyse.i_weighted_pred = X264_WEIGHTP_FAKE;
996
997     if( h->i_thread_frames > 1 )
998     {
999         int r = h->param.analyse.i_mv_range_thread;
1000         int r2;
1001         if( r <= 0 )
1002         {
1003             // half of the available space is reserved and divided evenly among the threads,
1004             // the rest is allocated to whichever thread is far enough ahead to use it.
1005             // reserving more space increases quality for some videos, but costs more time
1006             // in thread synchronization.
1007             int max_range = (h->param.i_height + X264_THREAD_HEIGHT) / h->i_thread_frames - X264_THREAD_HEIGHT;
1008             r = max_range / 2;
1009         }
1010         r = X264_MAX( r, h->param.analyse.i_me_range );
1011         r = X264_MIN( r, h->param.analyse.i_mv_range );
1012         // round up to use the whole mb row
1013         r2 = (r & ~15) + ((-X264_THREAD_HEIGHT) & 15);
1014         if( r2 < r )
1015             r2 += 16;
1016         x264_log( h, X264_LOG_DEBUG, "using mv_range_thread = %d\n", r2 );
1017         h->param.analyse.i_mv_range_thread = r2;
1018     }
1019
1020     if( h->param.rc.f_rate_tolerance < 0 )
1021         h->param.rc.f_rate_tolerance = 0;
1022     if( h->param.rc.f_qblur < 0 )
1023         h->param.rc.f_qblur = 0;
1024     if( h->param.rc.f_complexity_blur < 0 )
1025         h->param.rc.f_complexity_blur = 0;
1026
1027     h->param.i_sps_id &= 31;
1028
1029     if( PARAM_INTERLACED )
1030         h->param.b_pic_struct = 1;
1031
1032     h->param.i_nal_hrd = x264_clip3( h->param.i_nal_hrd, X264_NAL_HRD_NONE, X264_NAL_HRD_CBR );
1033
1034     if( h->param.i_nal_hrd && !h->param.rc.i_vbv_buffer_size )
1035     {
1036         x264_log( h, X264_LOG_WARNING, "NAL HRD parameters require VBV parameters\n" );
1037         h->param.i_nal_hrd = X264_NAL_HRD_NONE;
1038     }
1039
1040     if( h->param.i_nal_hrd == X264_NAL_HRD_CBR &&
1041        (h->param.rc.i_bitrate != h->param.rc.i_vbv_max_bitrate || !h->param.rc.i_vbv_max_bitrate) )
1042     {
1043         x264_log( h, X264_LOG_WARNING, "CBR HRD requires constant bitrate\n" );
1044         h->param.i_nal_hrd = X264_NAL_HRD_VBR;
1045     }
1046
1047     /* ensure the booleans are 0 or 1 so they can be used in math */
1048 #define BOOLIFY(x) h->param.x = !!h->param.x
1049     BOOLIFY( b_cabac );
1050     BOOLIFY( b_constrained_intra );
1051     BOOLIFY( b_deblocking_filter );
1052     BOOLIFY( b_deterministic );
1053     BOOLIFY( b_sliced_threads );
1054     BOOLIFY( b_interlaced );
1055     BOOLIFY( b_intra_refresh );
1056     BOOLIFY( b_visualize );
1057     BOOLIFY( b_aud );
1058     BOOLIFY( b_repeat_headers );
1059     BOOLIFY( b_annexb );
1060     BOOLIFY( b_vfr_input );
1061     BOOLIFY( b_pulldown );
1062     BOOLIFY( b_tff );
1063     BOOLIFY( b_pic_struct );
1064     BOOLIFY( b_fake_interlaced );
1065     BOOLIFY( b_open_gop );
1066     BOOLIFY( b_bluray_compat );
1067     BOOLIFY( b_full_recon );
1068     BOOLIFY( b_opencl );
1069     BOOLIFY( analyse.b_transform_8x8 );
1070     BOOLIFY( analyse.b_weighted_bipred );
1071     BOOLIFY( analyse.b_chroma_me );
1072     BOOLIFY( analyse.b_mixed_references );
1073     BOOLIFY( analyse.b_fast_pskip );
1074     BOOLIFY( analyse.b_dct_decimate );
1075     BOOLIFY( analyse.b_psy );
1076     BOOLIFY( analyse.b_psnr );
1077     BOOLIFY( analyse.b_ssim );
1078     BOOLIFY( rc.b_stat_write );
1079     BOOLIFY( rc.b_stat_read );
1080     BOOLIFY( rc.b_mb_tree );
1081 #undef BOOLIFY
1082
1083     return 0;
1084 }
1085
1086 static void mbcmp_init( x264_t *h )
1087 {
1088     int satd = !h->mb.b_lossless && h->param.analyse.i_subpel_refine > 1;
1089     memcpy( h->pixf.mbcmp, satd ? h->pixf.satd : h->pixf.sad_aligned, sizeof(h->pixf.mbcmp) );
1090     memcpy( h->pixf.mbcmp_unaligned, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.mbcmp_unaligned) );
1091     h->pixf.intra_mbcmp_x3_16x16 = satd ? h->pixf.intra_satd_x3_16x16 : h->pixf.intra_sad_x3_16x16;
1092     h->pixf.intra_mbcmp_x3_8x16c = satd ? h->pixf.intra_satd_x3_8x16c : h->pixf.intra_sad_x3_8x16c;
1093     h->pixf.intra_mbcmp_x3_8x8c  = satd ? h->pixf.intra_satd_x3_8x8c  : h->pixf.intra_sad_x3_8x8c;
1094     h->pixf.intra_mbcmp_x3_8x8 = satd ? h->pixf.intra_sa8d_x3_8x8 : h->pixf.intra_sad_x3_8x8;
1095     h->pixf.intra_mbcmp_x3_4x4 = satd ? h->pixf.intra_satd_x3_4x4 : h->pixf.intra_sad_x3_4x4;
1096     h->pixf.intra_mbcmp_x9_4x4 = h->param.b_cpu_independent || h->mb.b_lossless ? NULL
1097                                : satd ? h->pixf.intra_satd_x9_4x4 : h->pixf.intra_sad_x9_4x4;
1098     h->pixf.intra_mbcmp_x9_8x8 = h->param.b_cpu_independent || h->mb.b_lossless ? NULL
1099                                : satd ? h->pixf.intra_sa8d_x9_8x8 : h->pixf.intra_sad_x9_8x8;
1100     satd &= h->param.analyse.i_me_method == X264_ME_TESA;
1101     memcpy( h->pixf.fpelcmp, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.fpelcmp) );
1102     memcpy( h->pixf.fpelcmp_x3, satd ? h->pixf.satd_x3 : h->pixf.sad_x3, sizeof(h->pixf.fpelcmp_x3) );
1103     memcpy( h->pixf.fpelcmp_x4, satd ? h->pixf.satd_x4 : h->pixf.sad_x4, sizeof(h->pixf.fpelcmp_x4) );
1104 }
1105
1106 static void chroma_dsp_init( x264_t *h )
1107 {
1108     memcpy( h->luma2chroma_pixel, x264_luma2chroma_pixel[CHROMA_FORMAT], sizeof(h->luma2chroma_pixel) );
1109
1110     switch( CHROMA_FORMAT )
1111     {
1112         case CHROMA_420:
1113             memcpy( h->predict_chroma, h->predict_8x8c, sizeof(h->predict_chroma) );
1114             h->mc.prefetch_fenc = h->mc.prefetch_fenc_420;
1115             h->loopf.deblock_chroma[0] = h->loopf.deblock_h_chroma_420;
1116             h->loopf.deblock_chroma_intra[0] = h->loopf.deblock_h_chroma_420_intra;
1117             h->loopf.deblock_chroma_mbaff = h->loopf.deblock_chroma_420_mbaff;
1118             h->loopf.deblock_chroma_intra_mbaff = h->loopf.deblock_chroma_420_intra_mbaff;
1119             h->pixf.intra_mbcmp_x3_chroma = h->pixf.intra_mbcmp_x3_8x8c;
1120             h->quantf.coeff_last[DCT_CHROMA_DC] = h->quantf.coeff_last4;
1121             h->quantf.coeff_level_run[DCT_CHROMA_DC] = h->quantf.coeff_level_run4;
1122             break;
1123         case CHROMA_422:
1124             memcpy( h->predict_chroma, h->predict_8x16c, sizeof(h->predict_chroma) );
1125             h->mc.prefetch_fenc = h->mc.prefetch_fenc_422;
1126             h->loopf.deblock_chroma[0] = h->loopf.deblock_h_chroma_422;
1127             h->loopf.deblock_chroma_intra[0] = h->loopf.deblock_h_chroma_422_intra;
1128             h->loopf.deblock_chroma_mbaff = h->loopf.deblock_chroma_422_mbaff;
1129             h->loopf.deblock_chroma_intra_mbaff = h->loopf.deblock_chroma_422_intra_mbaff;
1130             h->pixf.intra_mbcmp_x3_chroma = h->pixf.intra_mbcmp_x3_8x16c;
1131             h->quantf.coeff_last[DCT_CHROMA_DC] = h->quantf.coeff_last8;
1132             h->quantf.coeff_level_run[DCT_CHROMA_DC] = h->quantf.coeff_level_run8;
1133             break;
1134         case CHROMA_444:
1135             h->mc.prefetch_fenc = h->mc.prefetch_fenc_422; /* FIXME: doesn't cover V plane */
1136             h->loopf.deblock_chroma_mbaff = h->loopf.deblock_luma_mbaff;
1137             h->loopf.deblock_chroma_intra_mbaff = h->loopf.deblock_luma_intra_mbaff;
1138             break;
1139     }
1140 }
1141
1142 static void x264_set_aspect_ratio( x264_t *h, x264_param_t *param, int initial )
1143 {
1144     /* VUI */
1145     if( param->vui.i_sar_width > 0 && param->vui.i_sar_height > 0 )
1146     {
1147         uint32_t i_w = param->vui.i_sar_width;
1148         uint32_t i_h = param->vui.i_sar_height;
1149         uint32_t old_w = h->param.vui.i_sar_width;
1150         uint32_t old_h = h->param.vui.i_sar_height;
1151
1152         x264_reduce_fraction( &i_w, &i_h );
1153
1154         while( i_w > 65535 || i_h > 65535 )
1155         {
1156             i_w /= 2;
1157             i_h /= 2;
1158         }
1159
1160         x264_reduce_fraction( &i_w, &i_h );
1161
1162         if( i_w != old_w || i_h != old_h || initial )
1163         {
1164             h->param.vui.i_sar_width = 0;
1165             h->param.vui.i_sar_height = 0;
1166             if( i_w == 0 || i_h == 0 )
1167                 x264_log( h, X264_LOG_WARNING, "cannot create valid sample aspect ratio\n" );
1168             else
1169             {
1170                 x264_log( h, initial?X264_LOG_INFO:X264_LOG_DEBUG, "using SAR=%d/%d\n", i_w, i_h );
1171                 h->param.vui.i_sar_width = i_w;
1172                 h->param.vui.i_sar_height = i_h;
1173             }
1174             x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
1175         }
1176     }
1177 }
1178
1179 /****************************************************************************
1180  * x264_encoder_open:
1181  ****************************************************************************/
1182 x264_t *x264_encoder_open( x264_param_t *param )
1183 {
1184     x264_t *h;
1185     char buf[1000], *p;
1186     int qp, i_slicetype_length;
1187
1188     CHECKED_MALLOCZERO( h, sizeof(x264_t) );
1189
1190     /* Create a copy of param */
1191     memcpy( &h->param, param, sizeof(x264_param_t) );
1192
1193     if( param->param_free )
1194         param->param_free( param );
1195
1196     if( x264_threading_init() )
1197     {
1198         x264_log( h, X264_LOG_ERROR, "unable to initialize threading\n" );
1199         goto fail;
1200     }
1201
1202     if( x264_validate_parameters( h, 1 ) < 0 )
1203         goto fail;
1204
1205     if( h->param.psz_cqm_file )
1206         if( x264_cqm_parse_file( h, h->param.psz_cqm_file ) < 0 )
1207             goto fail;
1208
1209     if( h->param.rc.psz_stat_out )
1210         h->param.rc.psz_stat_out = strdup( h->param.rc.psz_stat_out );
1211     if( h->param.rc.psz_stat_in )
1212         h->param.rc.psz_stat_in = strdup( h->param.rc.psz_stat_in );
1213
1214     x264_reduce_fraction( &h->param.i_fps_num, &h->param.i_fps_den );
1215     x264_reduce_fraction( &h->param.i_timebase_num, &h->param.i_timebase_den );
1216
1217     /* Init x264_t */
1218     h->i_frame = -1;
1219     h->i_frame_num = 0;
1220     h->i_idr_pic_id = 0;
1221
1222     if( (uint64_t)h->param.i_timebase_den * 2 > UINT32_MAX )
1223     {
1224         x264_log( h, X264_LOG_ERROR, "Effective timebase denominator %u exceeds H.264 maximum\n", h->param.i_timebase_den );
1225         goto fail;
1226     }
1227
1228     x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
1229     x264_pps_init( h->pps, h->param.i_sps_id, &h->param, h->sps );
1230
1231     x264_set_aspect_ratio( h, &h->param, 1 );
1232
1233     x264_validate_levels( h, 1 );
1234
1235     h->chroma_qp_table = i_chroma_qp_table + 12 + h->pps->i_chroma_qp_index_offset;
1236
1237     if( x264_cqm_init( h ) < 0 )
1238         goto fail;
1239
1240     h->mb.i_mb_width = h->sps->i_mb_width;
1241     h->mb.i_mb_height = h->sps->i_mb_height;
1242     h->mb.i_mb_count = h->mb.i_mb_width * h->mb.i_mb_height;
1243
1244     h->mb.chroma_h_shift = CHROMA_FORMAT == CHROMA_420 || CHROMA_FORMAT == CHROMA_422;
1245     h->mb.chroma_v_shift = CHROMA_FORMAT == CHROMA_420;
1246
1247     /* Adaptive MBAFF and subme 0 are not supported as we require halving motion
1248      * vectors during prediction, resulting in hpel mvs.
1249      * The chosen solution is to make MBAFF non-adaptive in this case. */
1250     h->mb.b_adaptive_mbaff = PARAM_INTERLACED && h->param.analyse.i_subpel_refine;
1251
1252     /* Init frames. */
1253     if( h->param.i_bframe_adaptive == X264_B_ADAPT_TRELLIS && !h->param.rc.b_stat_read )
1254         h->frames.i_delay = X264_MAX(h->param.i_bframe,3)*4;
1255     else
1256         h->frames.i_delay = h->param.i_bframe;
1257     if( h->param.rc.b_mb_tree || h->param.rc.i_vbv_buffer_size )
1258         h->frames.i_delay = X264_MAX( h->frames.i_delay, h->param.rc.i_lookahead );
1259     i_slicetype_length = h->frames.i_delay;
1260     h->frames.i_delay += h->i_thread_frames - 1;
1261     h->frames.i_delay += h->param.i_sync_lookahead;
1262     h->frames.i_delay += h->param.b_vfr_input;
1263     h->frames.i_bframe_delay = h->param.i_bframe ? (h->param.i_bframe_pyramid ? 2 : 1) : 0;
1264
1265     h->frames.i_max_ref0 = h->param.i_frame_reference;
1266     h->frames.i_max_ref1 = X264_MIN( h->sps->vui.i_num_reorder_frames, h->param.i_frame_reference );
1267     h->frames.i_max_dpb  = h->sps->vui.i_max_dec_frame_buffering;
1268     h->frames.b_have_lowres = !h->param.rc.b_stat_read
1269         && ( h->param.rc.i_rc_method == X264_RC_ABR
1270           || h->param.rc.i_rc_method == X264_RC_CRF
1271           || h->param.i_bframe_adaptive
1272           || h->param.i_scenecut_threshold
1273           || h->param.rc.b_mb_tree
1274           || h->param.analyse.i_weighted_pred );
1275     h->frames.b_have_lowres |= h->param.rc.b_stat_read && h->param.rc.i_vbv_buffer_size > 0;
1276     h->frames.b_have_sub8x8_esa = !!(h->param.analyse.inter & X264_ANALYSE_PSUB8x8);
1277
1278     h->frames.i_last_idr =
1279     h->frames.i_last_keyframe = - h->param.i_keyint_max;
1280     h->frames.i_input    = 0;
1281     h->frames.i_largest_pts = h->frames.i_second_largest_pts = -1;
1282     h->frames.i_poc_last_open_gop = -1;
1283
1284     CHECKED_MALLOCZERO( h->frames.unused[0], (h->frames.i_delay + 3) * sizeof(x264_frame_t *) );
1285     /* Allocate room for max refs plus a few extra just in case. */
1286     CHECKED_MALLOCZERO( h->frames.unused[1], (h->i_thread_frames + X264_REF_MAX + 4) * sizeof(x264_frame_t *) );
1287     CHECKED_MALLOCZERO( h->frames.current, (h->param.i_sync_lookahead + h->param.i_bframe
1288                         + h->i_thread_frames + 3) * sizeof(x264_frame_t *) );
1289     if( h->param.analyse.i_weighted_pred > 0 )
1290         CHECKED_MALLOCZERO( h->frames.blank_unused, h->i_thread_frames * 4 * sizeof(x264_frame_t *) );
1291     h->i_ref[0] = h->i_ref[1] = 0;
1292     h->i_cpb_delay = h->i_coded_fields = h->i_disp_fields = 0;
1293     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);
1294     h->i_disp_fields_last_frame = -1;
1295     x264_rdo_init();
1296
1297     /* init CPU functions */
1298     x264_predict_16x16_init( h->param.cpu, h->predict_16x16 );
1299     x264_predict_8x8c_init( h->param.cpu, h->predict_8x8c );
1300     x264_predict_8x16c_init( h->param.cpu, h->predict_8x16c );
1301     x264_predict_8x8_init( h->param.cpu, h->predict_8x8, &h->predict_8x8_filter );
1302     x264_predict_4x4_init( h->param.cpu, h->predict_4x4 );
1303     x264_pixel_init( h->param.cpu, &h->pixf );
1304     x264_dct_init( h->param.cpu, &h->dctf );
1305     x264_zigzag_init( h->param.cpu, &h->zigzagf_progressive, &h->zigzagf_interlaced );
1306     memcpy( &h->zigzagf, PARAM_INTERLACED ? &h->zigzagf_interlaced : &h->zigzagf_progressive, sizeof(h->zigzagf) );
1307     x264_mc_init( h->param.cpu, &h->mc, h->param.b_cpu_independent );
1308     x264_quant_init( h, h->param.cpu, &h->quantf );
1309     x264_deblock_init( h->param.cpu, &h->loopf, PARAM_INTERLACED );
1310     x264_bitstream_init( h->param.cpu, &h->bsf );
1311     if( h->param.b_cabac )
1312         x264_cabac_init( h );
1313     else
1314         x264_stack_align( x264_cavlc_init, h );
1315
1316     mbcmp_init( h );
1317     chroma_dsp_init( h );
1318
1319     p = buf + sprintf( buf, "using cpu capabilities:" );
1320     for( int i = 0; x264_cpu_names[i].flags; i++ )
1321     {
1322         if( !strcmp(x264_cpu_names[i].name, "SSE")
1323             && h->param.cpu & (X264_CPU_SSE2) )
1324             continue;
1325         if( !strcmp(x264_cpu_names[i].name, "SSE2")
1326             && h->param.cpu & (X264_CPU_SSE2_IS_FAST|X264_CPU_SSE2_IS_SLOW) )
1327             continue;
1328         if( !strcmp(x264_cpu_names[i].name, "SSE3")
1329             && (h->param.cpu & X264_CPU_SSSE3 || !(h->param.cpu & X264_CPU_CACHELINE_64)) )
1330             continue;
1331         if( !strcmp(x264_cpu_names[i].name, "SSE4.1")
1332             && (h->param.cpu & X264_CPU_SSE42) )
1333             continue;
1334         if( !strcmp(x264_cpu_names[i].name, "BMI1")
1335             && (h->param.cpu & X264_CPU_BMI2) )
1336             continue;
1337         if( (h->param.cpu & x264_cpu_names[i].flags) == x264_cpu_names[i].flags
1338             && (!i || x264_cpu_names[i].flags != x264_cpu_names[i-1].flags) )
1339             p += sprintf( p, " %s", x264_cpu_names[i].name );
1340     }
1341     if( !h->param.cpu )
1342         p += sprintf( p, " none!" );
1343     x264_log( h, X264_LOG_INFO, "%s\n", buf );
1344
1345     float *logs = x264_analyse_prepare_costs( h );
1346     if( !logs )
1347         goto fail;
1348     for( qp = X264_MIN( h->param.rc.i_qp_min, QP_MAX_SPEC ); qp <= h->param.rc.i_qp_max; qp++ )
1349         if( x264_analyse_init_costs( h, logs, qp ) )
1350             goto fail;
1351     if( x264_analyse_init_costs( h, logs, X264_LOOKAHEAD_QP ) )
1352         goto fail;
1353     x264_free( logs );
1354
1355     static const uint16_t cost_mv_correct[7] = { 24, 47, 95, 189, 379, 757, 1515 };
1356     /* Checks for known miscompilation issues. */
1357     if( h->cost_mv[X264_LOOKAHEAD_QP][2013] != cost_mv_correct[BIT_DEPTH-8] )
1358     {
1359         x264_log( h, X264_LOG_ERROR, "MV cost test failed: x264 has been miscompiled!\n" );
1360         goto fail;
1361     }
1362
1363     /* Must be volatile or else GCC will optimize it out. */
1364     volatile int temp = 392;
1365     if( x264_clz( temp ) != 23 )
1366     {
1367         x264_log( h, X264_LOG_ERROR, "CLZ test failed: x264 has been miscompiled!\n" );
1368 #if ARCH_X86 || ARCH_X86_64
1369         x264_log( h, X264_LOG_ERROR, "Are you attempting to run an SSE4a/LZCNT-targeted build on a CPU that\n" );
1370         x264_log( h, X264_LOG_ERROR, "doesn't support it?\n" );
1371 #endif
1372         goto fail;
1373     }
1374
1375     h->out.i_nal = 0;
1376     h->out.i_bitstream = X264_MAX( 1000000, h->param.i_width * h->param.i_height * 4
1377         * ( h->param.rc.i_rc_method == X264_RC_ABR ? pow( 0.95, h->param.rc.i_qp_min )
1378           : pow( 0.95, h->param.rc.i_qp_constant ) * X264_MAX( 1, h->param.rc.f_ip_factor )));
1379
1380     h->nal_buffer_size = h->out.i_bitstream * 3/2 + 4 + 64; /* +4 for startcode, +64 for nal_escape assembly padding */
1381     CHECKED_MALLOC( h->nal_buffer, h->nal_buffer_size );
1382
1383     if( h->param.i_threads > 1 &&
1384         x264_threadpool_init( &h->threadpool, h->param.i_threads, (void*)x264_encoder_thread_init, h ) )
1385         goto fail;
1386     if( h->param.i_lookahead_threads > 1 &&
1387         x264_threadpool_init( &h->lookaheadpool, h->param.i_lookahead_threads, (void*)x264_lookahead_thread_init, h ) )
1388         goto fail;
1389
1390 #if HAVE_OPENCL
1391     if( h->param.b_opencl )
1392     {
1393         h->opencl.ocl = x264_opencl_load_library();
1394         if( !h->opencl.ocl )
1395         {
1396             x264_log( h, X264_LOG_WARNING, "failed to load OpenCL\n" );
1397             h->param.b_opencl = 0;
1398         }
1399     }
1400 #endif
1401
1402     h->thread[0] = h;
1403     for( int i = 1; i < h->param.i_threads + !!h->param.i_sync_lookahead; i++ )
1404         CHECKED_MALLOC( h->thread[i], sizeof(x264_t) );
1405     if( h->param.i_lookahead_threads > 1 )
1406         for( int i = 0; i < h->param.i_lookahead_threads; i++ )
1407         {
1408             CHECKED_MALLOC( h->lookahead_thread[i], sizeof(x264_t) );
1409             *h->lookahead_thread[i] = *h;
1410         }
1411
1412     for( int i = 0; i < h->param.i_threads; i++ )
1413     {
1414         int init_nal_count = h->param.i_slice_count + 3;
1415         int allocate_threadlocal_data = !h->param.b_sliced_threads || !i;
1416         if( i > 0 )
1417             *h->thread[i] = *h;
1418
1419         if( x264_pthread_mutex_init( &h->thread[i]->mutex, NULL ) )
1420             goto fail;
1421         if( x264_pthread_cond_init( &h->thread[i]->cv, NULL ) )
1422             goto fail;
1423
1424         if( allocate_threadlocal_data )
1425         {
1426             h->thread[i]->fdec = x264_frame_pop_unused( h, 1 );
1427             if( !h->thread[i]->fdec )
1428                 goto fail;
1429         }
1430         else
1431             h->thread[i]->fdec = h->thread[0]->fdec;
1432
1433         CHECKED_MALLOC( h->thread[i]->out.p_bitstream, h->out.i_bitstream );
1434         /* Start each thread with room for init_nal_count NAL units; it'll realloc later if needed. */
1435         CHECKED_MALLOC( h->thread[i]->out.nal, init_nal_count*sizeof(x264_nal_t) );
1436         h->thread[i]->out.i_nals_allocated = init_nal_count;
1437
1438         if( allocate_threadlocal_data && x264_macroblock_cache_allocate( h->thread[i] ) < 0 )
1439             goto fail;
1440     }
1441
1442 #if HAVE_OPENCL
1443     if( h->param.b_opencl && x264_opencl_lookahead_init( h ) < 0 )
1444         h->param.b_opencl = 0;
1445 #endif
1446
1447     if( x264_lookahead_init( h, i_slicetype_length ) )
1448         goto fail;
1449
1450     for( int i = 0; i < h->param.i_threads; i++ )
1451         if( x264_macroblock_thread_allocate( h->thread[i], 0 ) < 0 )
1452             goto fail;
1453
1454     if( x264_ratecontrol_new( h ) < 0 )
1455         goto fail;
1456
1457     if( h->param.i_nal_hrd )
1458     {
1459         x264_log( h, X264_LOG_DEBUG, "HRD bitrate: %i bits/sec\n", h->sps->vui.hrd.i_bit_rate_unscaled );
1460         x264_log( h, X264_LOG_DEBUG, "CPB size: %i bits\n", h->sps->vui.hrd.i_cpb_size_unscaled );
1461     }
1462
1463     if( h->param.psz_dump_yuv )
1464     {
1465         /* create or truncate the reconstructed video file */
1466         FILE *f = fopen( h->param.psz_dump_yuv, "w" );
1467         if( !f )
1468         {
1469             x264_log( h, X264_LOG_ERROR, "dump_yuv: can't write to %s\n", h->param.psz_dump_yuv );
1470             goto fail;
1471         }
1472         else if( !x264_is_regular_file( f ) )
1473         {
1474             x264_log( h, X264_LOG_ERROR, "dump_yuv: incompatible with non-regular file %s\n", h->param.psz_dump_yuv );
1475             goto fail;
1476         }
1477         fclose( f );
1478     }
1479
1480     const char *profile = h->sps->i_profile_idc == PROFILE_BASELINE ? "Constrained Baseline" :
1481                           h->sps->i_profile_idc == PROFILE_MAIN ? "Main" :
1482                           h->sps->i_profile_idc == PROFILE_HIGH ? "High" :
1483                           h->sps->i_profile_idc == PROFILE_HIGH10 ? (h->sps->b_constraint_set3 == 1 ? "High 10 Intra" : "High 10") :
1484                           h->sps->i_profile_idc == PROFILE_HIGH422 ? (h->sps->b_constraint_set3 == 1 ? "High 4:2:2 Intra" : "High 4:2:2") :
1485                           h->sps->b_constraint_set3 == 1 ? "High 4:4:4 Intra" : "High 4:4:4 Predictive";
1486     char level[4];
1487     snprintf( level, sizeof(level), "%d.%d", h->sps->i_level_idc/10, h->sps->i_level_idc%10 );
1488     if( h->sps->i_level_idc == 9 || ( h->sps->i_level_idc == 11 && h->sps->b_constraint_set3 &&
1489         (h->sps->i_profile_idc == PROFILE_BASELINE || h->sps->i_profile_idc == PROFILE_MAIN) ) )
1490         strcpy( level, "1b" );
1491
1492     if( h->sps->i_profile_idc < PROFILE_HIGH10 )
1493     {
1494         x264_log( h, X264_LOG_INFO, "profile %s, level %s\n",
1495             profile, level );
1496     }
1497     else
1498     {
1499         static const char * const subsampling[4] = { "4:0:0", "4:2:0", "4:2:2", "4:4:4" };
1500         x264_log( h, X264_LOG_INFO, "profile %s, level %s, %s %d-bit\n",
1501             profile, level, subsampling[CHROMA_FORMAT], BIT_DEPTH );
1502     }
1503
1504     return h;
1505 fail:
1506     x264_free( h );
1507     return NULL;
1508 }
1509
1510 /****************************************************************************
1511  * x264_encoder_reconfig:
1512  ****************************************************************************/
1513 int x264_encoder_reconfig( x264_t *h, x264_param_t *param )
1514 {
1515     /* If the previous frame isn't done encoding, reconfiguring is probably dangerous. */
1516     if( h->param.b_sliced_threads )
1517         if( x264_threadpool_wait_all( h ) < 0 )
1518             return -1;
1519
1520     int rc_reconfig = 0;
1521     h = h->thread[h->thread[0]->i_thread_phase];
1522     x264_set_aspect_ratio( h, param, 0 );
1523 #define COPY(var) h->param.var = param->var
1524     COPY( i_frame_reference ); // but never uses more refs than initially specified
1525     COPY( i_bframe_bias );
1526     if( h->param.i_scenecut_threshold )
1527         COPY( i_scenecut_threshold ); // can't turn it on or off, only vary the threshold
1528     COPY( b_deblocking_filter );
1529     COPY( i_deblocking_filter_alphac0 );
1530     COPY( i_deblocking_filter_beta );
1531     COPY( i_frame_packing );
1532     COPY( analyse.inter );
1533     COPY( analyse.intra );
1534     COPY( analyse.i_direct_mv_pred );
1535     /* Scratch buffer prevents me_range from being increased for esa/tesa */
1536     if( h->param.analyse.i_me_method < X264_ME_ESA || param->analyse.i_me_range < h->param.analyse.i_me_range )
1537         COPY( analyse.i_me_range );
1538     COPY( analyse.i_noise_reduction );
1539     /* We can't switch out of subme=0 during encoding. */
1540     if( h->param.analyse.i_subpel_refine )
1541         COPY( analyse.i_subpel_refine );
1542     COPY( analyse.i_trellis );
1543     COPY( analyse.b_chroma_me );
1544     COPY( analyse.b_dct_decimate );
1545     COPY( analyse.b_fast_pskip );
1546     COPY( analyse.b_mixed_references );
1547     COPY( analyse.f_psy_rd );
1548     COPY( analyse.f_psy_trellis );
1549     COPY( crop_rect );
1550     // can only twiddle these if they were enabled to begin with:
1551     if( h->param.analyse.i_me_method >= X264_ME_ESA || param->analyse.i_me_method < X264_ME_ESA )
1552         COPY( analyse.i_me_method );
1553     if( h->param.analyse.i_me_method >= X264_ME_ESA && !h->frames.b_have_sub8x8_esa )
1554         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
1555     if( h->pps->b_transform_8x8_mode )
1556         COPY( analyse.b_transform_8x8 );
1557     if( h->frames.i_max_ref1 > 1 )
1558         COPY( i_bframe_pyramid );
1559     COPY( i_slice_max_size );
1560     COPY( i_slice_max_mbs );
1561     COPY( i_slice_min_mbs );
1562     COPY( i_slice_count );
1563     COPY( i_slice_count_max );
1564     COPY( b_tff );
1565
1566     /* VBV can't be turned on if it wasn't on to begin with */
1567     if( h->param.rc.i_vbv_max_bitrate > 0 && h->param.rc.i_vbv_buffer_size > 0 &&
1568           param->rc.i_vbv_max_bitrate > 0 &&   param->rc.i_vbv_buffer_size > 0 )
1569     {
1570         rc_reconfig |= h->param.rc.i_vbv_max_bitrate != param->rc.i_vbv_max_bitrate;
1571         rc_reconfig |= h->param.rc.i_vbv_buffer_size != param->rc.i_vbv_buffer_size;
1572         rc_reconfig |= h->param.rc.i_bitrate != param->rc.i_bitrate;
1573         COPY( rc.i_vbv_max_bitrate );
1574         COPY( rc.i_vbv_buffer_size );
1575         COPY( rc.i_bitrate );
1576     }
1577     rc_reconfig |= h->param.rc.f_rf_constant != param->rc.f_rf_constant;
1578     rc_reconfig |= h->param.rc.f_rf_constant_max != param->rc.f_rf_constant_max;
1579     COPY( rc.f_rf_constant );
1580     COPY( rc.f_rf_constant_max );
1581 #undef COPY
1582
1583     mbcmp_init( h );
1584
1585     int ret = x264_validate_parameters( h, 0 );
1586
1587     /* Supported reconfiguration options (1-pass only):
1588      * vbv-maxrate
1589      * vbv-bufsize
1590      * crf
1591      * bitrate (CBR only) */
1592     if( !ret && rc_reconfig )
1593         x264_ratecontrol_init_reconfigurable( h, 0 );
1594
1595     return ret;
1596 }
1597
1598 /****************************************************************************
1599  * x264_encoder_parameters:
1600  ****************************************************************************/
1601 void x264_encoder_parameters( x264_t *h, x264_param_t *param )
1602 {
1603     memcpy( param, &h->thread[h->i_thread_phase]->param, sizeof(x264_param_t) );
1604 }
1605
1606 /* internal usage */
1607 static void x264_nal_start( x264_t *h, int i_type, int i_ref_idc )
1608 {
1609     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1610
1611     nal->i_ref_idc        = i_ref_idc;
1612     nal->i_type           = i_type;
1613     nal->b_long_startcode = 1;
1614
1615     nal->i_payload= 0;
1616     nal->p_payload= &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8];
1617 }
1618
1619 /* if number of allocated nals is not enough, re-allocate a larger one. */
1620 static int x264_nal_check_buffer( x264_t *h )
1621 {
1622     if( h->out.i_nal >= h->out.i_nals_allocated )
1623     {
1624         x264_nal_t *new_out = x264_malloc( sizeof(x264_nal_t) * (h->out.i_nals_allocated*2) );
1625         if( !new_out )
1626             return -1;
1627         memcpy( new_out, h->out.nal, sizeof(x264_nal_t) * (h->out.i_nals_allocated) );
1628         x264_free( h->out.nal );
1629         h->out.nal = new_out;
1630         h->out.i_nals_allocated *= 2;
1631     }
1632     return 0;
1633 }
1634
1635 static int x264_nal_end( x264_t *h )
1636 {
1637     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1638     uint8_t *end = &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8];
1639     nal->i_payload = end - nal->p_payload;
1640     /* Assembly implementation of nal_escape reads past the end of the input.
1641      * While undefined padding wouldn't actually affect the output, it makes valgrind unhappy. */
1642     memset( end, 0xff, 64 );
1643     if( h->param.nalu_process )
1644         h->param.nalu_process( h, nal, h->fenc->opaque );
1645     h->out.i_nal++;
1646
1647     return x264_nal_check_buffer( h );
1648 }
1649
1650 static int x264_encoder_encapsulate_nals( x264_t *h, int start )
1651 {
1652     int nal_size = 0, previous_nal_size = 0;
1653
1654     if( h->param.nalu_process )
1655     {
1656         for( int i = start; i < h->out.i_nal; i++ )
1657             nal_size += h->out.nal[i].i_payload;
1658         return nal_size;
1659     }
1660
1661     for( int i = 0; i < start; i++ )
1662         previous_nal_size += h->out.nal[i].i_payload;
1663
1664     for( int i = start; i < h->out.i_nal; i++ )
1665         nal_size += h->out.nal[i].i_payload;
1666
1667     /* Worst-case NAL unit escaping: reallocate the buffer if it's too small. */
1668     int necessary_size = nal_size * 3/2 + h->out.i_nal * 4 + 4 + 64;
1669     if( h->nal_buffer_size < necessary_size )
1670     {
1671         h->nal_buffer_size = necessary_size * 2;
1672         uint8_t *buf = x264_malloc( h->nal_buffer_size );
1673         if( !buf )
1674             return -1;
1675         if( previous_nal_size )
1676             memcpy( buf, h->nal_buffer, previous_nal_size );
1677         x264_free( h->nal_buffer );
1678         h->nal_buffer = buf;
1679     }
1680
1681     uint8_t *nal_buffer = h->nal_buffer + previous_nal_size;
1682
1683     for( int i = start; i < h->out.i_nal; i++ )
1684     {
1685         h->out.nal[i].b_long_startcode = !i || h->out.nal[i].i_type == NAL_SPS || h->out.nal[i].i_type == NAL_PPS;
1686         x264_nal_encode( h, nal_buffer, &h->out.nal[i] );
1687         nal_buffer += h->out.nal[i].i_payload;
1688     }
1689
1690     x264_emms();
1691
1692     return nal_buffer - (h->nal_buffer + previous_nal_size);
1693 }
1694
1695 /****************************************************************************
1696  * x264_encoder_headers:
1697  ****************************************************************************/
1698 int x264_encoder_headers( x264_t *h, x264_nal_t **pp_nal, int *pi_nal )
1699 {
1700     int frame_size = 0;
1701     /* init bitstream context */
1702     h->out.i_nal = 0;
1703     bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
1704
1705     /* Write SEI, SPS and PPS. */
1706
1707     /* generate sequence parameters */
1708     x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
1709     x264_sps_write( &h->out.bs, h->sps );
1710     if( x264_nal_end( h ) )
1711         return -1;
1712
1713     /* generate picture parameters */
1714     x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
1715     x264_pps_write( &h->out.bs, h->sps, h->pps );
1716     if( x264_nal_end( h ) )
1717         return -1;
1718
1719     /* identify ourselves */
1720     x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
1721     if( x264_sei_version_write( h, &h->out.bs ) )
1722         return -1;
1723     if( x264_nal_end( h ) )
1724         return -1;
1725
1726     frame_size = x264_encoder_encapsulate_nals( h, 0 );
1727     if( frame_size < 0 )
1728         return -1;
1729
1730     /* now set output*/
1731     *pi_nal = h->out.i_nal;
1732     *pp_nal = &h->out.nal[0];
1733     h->out.i_nal = 0;
1734
1735     return frame_size;
1736 }
1737
1738 /* Check to see whether we have chosen a reference list ordering different
1739  * from the standard's default. */
1740 static inline void x264_reference_check_reorder( x264_t *h )
1741 {
1742     /* The reorder check doesn't check for missing frames, so just
1743      * force a reorder if one of the reference list is corrupt. */
1744     for( int i = 0; h->frames.reference[i]; i++ )
1745         if( h->frames.reference[i]->b_corrupt )
1746         {
1747             h->b_ref_reorder[0] = 1;
1748             return;
1749         }
1750     for( int list = 0; list <= (h->sh.i_type == SLICE_TYPE_B); list++ )
1751         for( int i = 0; i < h->i_ref[list] - 1; i++ )
1752         {
1753             int framenum_diff = h->fref[list][i+1]->i_frame_num - h->fref[list][i]->i_frame_num;
1754             int poc_diff = h->fref[list][i+1]->i_poc - h->fref[list][i]->i_poc;
1755             /* P and B-frames use different default orders. */
1756             if( h->sh.i_type == SLICE_TYPE_P ? framenum_diff > 0 : list == 1 ? poc_diff < 0 : poc_diff > 0 )
1757             {
1758                 h->b_ref_reorder[list] = 1;
1759                 return;
1760             }
1761         }
1762 }
1763
1764 /* return -1 on failure, else return the index of the new reference frame */
1765 int x264_weighted_reference_duplicate( x264_t *h, int i_ref, const x264_weight_t *w )
1766 {
1767     int i = h->i_ref[0];
1768     int j = 1;
1769     x264_frame_t *newframe;
1770     if( i <= 1 ) /* empty list, definitely can't duplicate frame */
1771         return -1;
1772
1773     //Duplication is only used in X264_WEIGHTP_SMART
1774     if( h->param.analyse.i_weighted_pred != X264_WEIGHTP_SMART )
1775         return -1;
1776
1777     /* Duplication is a hack to compensate for crappy rounding in motion compensation.
1778      * With high bit depth, it's not worth doing, so turn it off except in the case of
1779      * unweighted dupes. */
1780     if( BIT_DEPTH > 8 && w != x264_weight_none )
1781         return -1;
1782
1783     newframe = x264_frame_pop_blank_unused( h );
1784     if( !newframe )
1785         return -1;
1786
1787     //FIXME: probably don't need to copy everything
1788     *newframe = *h->fref[0][i_ref];
1789     newframe->i_reference_count = 1;
1790     newframe->orig = h->fref[0][i_ref];
1791     newframe->b_duplicate = 1;
1792     memcpy( h->fenc->weight[j], w, sizeof(h->fenc->weight[i]) );
1793
1794     /* shift the frames to make space for the dupe. */
1795     h->b_ref_reorder[0] = 1;
1796     if( h->i_ref[0] < X264_REF_MAX )
1797         ++h->i_ref[0];
1798     h->fref[0][X264_REF_MAX-1] = NULL;
1799     x264_frame_unshift( &h->fref[0][j], newframe );
1800
1801     return j;
1802 }
1803
1804 static void x264_weighted_pred_init( x264_t *h )
1805 {
1806     /* for now no analysis and set all weights to nothing */
1807     for( int i_ref = 0; i_ref < h->i_ref[0]; i_ref++ )
1808         h->fenc->weighted[i_ref] = h->fref[0][i_ref]->filtered[0][0];
1809
1810     // FIXME: This only supports weighting of one reference frame
1811     // and duplicates of that frame.
1812     h->fenc->i_lines_weighted = 0;
1813
1814     for( int i_ref = 0; i_ref < (h->i_ref[0] << SLICE_MBAFF); i_ref++ )
1815         for( int i = 0; i < 3; i++ )
1816             h->sh.weight[i_ref][i].weightfn = NULL;
1817
1818
1819     if( h->sh.i_type != SLICE_TYPE_P || h->param.analyse.i_weighted_pred <= 0 )
1820         return;
1821
1822     int i_padv = PADV << PARAM_INTERLACED;
1823     int denom = -1;
1824     int weightplane[2] = { 0, 0 };
1825     int buffer_next = 0;
1826     for( int i = 0; i < 3; i++ )
1827     {
1828         for( int j = 0; j < h->i_ref[0]; j++ )
1829         {
1830             if( h->fenc->weight[j][i].weightfn )
1831             {
1832                 h->sh.weight[j][i] = h->fenc->weight[j][i];
1833                 // if weight is useless, don't write it to stream
1834                 if( h->sh.weight[j][i].i_scale == 1<<h->sh.weight[j][i].i_denom && h->sh.weight[j][i].i_offset == 0 )
1835                     h->sh.weight[j][i].weightfn = NULL;
1836                 else
1837                 {
1838                     if( !weightplane[!!i] )
1839                     {
1840                         weightplane[!!i] = 1;
1841                         h->sh.weight[0][!!i].i_denom = denom = h->sh.weight[j][i].i_denom;
1842                         assert( x264_clip3( denom, 0, 7 ) == denom );
1843                     }
1844
1845                     assert( h->sh.weight[j][i].i_denom == denom );
1846                     if( !i )
1847                     {
1848                         h->fenc->weighted[j] = h->mb.p_weight_buf[buffer_next++] + h->fenc->i_stride[0] * i_padv + PADH;
1849                         //scale full resolution frame
1850                         if( h->param.i_threads == 1 )
1851                         {
1852                             pixel *src = h->fref[0][j]->filtered[0][0] - h->fref[0][j]->i_stride[0]*i_padv - PADH;
1853                             pixel *dst = h->fenc->weighted[j] - h->fenc->i_stride[0]*i_padv - PADH;
1854                             int stride = h->fenc->i_stride[0];
1855                             int width = h->fenc->i_width[0] + PADH*2;
1856                             int height = h->fenc->i_lines[0] + i_padv*2;
1857                             x264_weight_scale_plane( h, dst, stride, src, stride, width, height, &h->sh.weight[j][0] );
1858                             h->fenc->i_lines_weighted = height;
1859                         }
1860                     }
1861                 }
1862             }
1863         }
1864     }
1865
1866     if( weightplane[1] )
1867         for( int i = 0; i < h->i_ref[0]; i++ )
1868         {
1869             if( h->sh.weight[i][1].weightfn && !h->sh.weight[i][2].weightfn )
1870             {
1871                 h->sh.weight[i][2].i_scale = 1 << h->sh.weight[0][1].i_denom;
1872                 h->sh.weight[i][2].i_offset = 0;
1873             }
1874             else if( h->sh.weight[i][2].weightfn && !h->sh.weight[i][1].weightfn )
1875             {
1876                 h->sh.weight[i][1].i_scale = 1 << h->sh.weight[0][1].i_denom;
1877                 h->sh.weight[i][1].i_offset = 0;
1878             }
1879         }
1880
1881     if( !weightplane[0] )
1882         h->sh.weight[0][0].i_denom = 0;
1883     if( !weightplane[1] )
1884         h->sh.weight[0][1].i_denom = 0;
1885     h->sh.weight[0][2].i_denom = h->sh.weight[0][1].i_denom;
1886 }
1887
1888 static inline int x264_reference_distance( x264_t *h, x264_frame_t *frame )
1889 {
1890     if( h->param.i_frame_packing == 5 )
1891         return abs((h->fenc->i_frame&~1) - (frame->i_frame&~1)) +
1892                   ((h->fenc->i_frame&1) != (frame->i_frame&1));
1893     else
1894         return abs(h->fenc->i_frame - frame->i_frame);
1895 }
1896
1897 static inline void x264_reference_build_list( x264_t *h, int i_poc )
1898 {
1899     int b_ok;
1900
1901     /* build ref list 0/1 */
1902     h->mb.pic.i_fref[0] = h->i_ref[0] = 0;
1903     h->mb.pic.i_fref[1] = h->i_ref[1] = 0;
1904     if( h->sh.i_type == SLICE_TYPE_I )
1905         return;
1906
1907     for( int i = 0; h->frames.reference[i]; i++ )
1908     {
1909         if( h->frames.reference[i]->b_corrupt )
1910             continue;
1911         if( h->frames.reference[i]->i_poc < i_poc )
1912             h->fref[0][h->i_ref[0]++] = h->frames.reference[i];
1913         else if( h->frames.reference[i]->i_poc > i_poc )
1914             h->fref[1][h->i_ref[1]++] = h->frames.reference[i];
1915     }
1916
1917     /* Order reference lists by distance from the current frame. */
1918     for( int list = 0; list < 2; list++ )
1919     {
1920         h->fref_nearest[list] = h->fref[list][0];
1921         do
1922         {
1923             b_ok = 1;
1924             for( int i = 0; i < h->i_ref[list] - 1; i++ )
1925             {
1926                 if( list ? h->fref[list][i+1]->i_poc < h->fref_nearest[list]->i_poc
1927                          : h->fref[list][i+1]->i_poc > h->fref_nearest[list]->i_poc )
1928                     h->fref_nearest[list] = h->fref[list][i+1];
1929                 if( x264_reference_distance( h, h->fref[list][i] ) > x264_reference_distance( h, h->fref[list][i+1] ) )
1930                 {
1931                     XCHG( x264_frame_t*, h->fref[list][i], h->fref[list][i+1] );
1932                     b_ok = 0;
1933                     break;
1934                 }
1935             }
1936         } while( !b_ok );
1937     }
1938
1939     if( h->sh.i_mmco_remove_from_end )
1940         for( int i = h->i_ref[0]-1; i >= h->i_ref[0] - h->sh.i_mmco_remove_from_end; i-- )
1941         {
1942             int diff = h->i_frame_num - h->fref[0][i]->i_frame_num;
1943             h->sh.mmco[h->sh.i_mmco_command_count].i_poc = h->fref[0][i]->i_poc;
1944             h->sh.mmco[h->sh.i_mmco_command_count++].i_difference_of_pic_nums = diff;
1945         }
1946
1947     x264_reference_check_reorder( h );
1948
1949     h->i_ref[1] = X264_MIN( h->i_ref[1], h->frames.i_max_ref1 );
1950     h->i_ref[0] = X264_MIN( h->i_ref[0], h->frames.i_max_ref0 );
1951     h->i_ref[0] = X264_MIN( h->i_ref[0], h->param.i_frame_reference ); // if reconfig() has lowered the limit
1952
1953     /* For Blu-ray compliance, don't reference frames outside of the minigop. */
1954     if( IS_X264_TYPE_B( h->fenc->i_type ) && h->param.b_bluray_compat )
1955         h->i_ref[0] = X264_MIN( h->i_ref[0], IS_X264_TYPE_B( h->fref[0][0]->i_type ) + 1 );
1956
1957     /* add duplicates */
1958     if( h->fenc->i_type == X264_TYPE_P )
1959     {
1960         int idx = -1;
1961         if( h->param.analyse.i_weighted_pred >= X264_WEIGHTP_SIMPLE )
1962         {
1963             x264_weight_t w[3];
1964             w[1].weightfn = w[2].weightfn = NULL;
1965             if( h->param.rc.b_stat_read )
1966                 x264_ratecontrol_set_weights( h, h->fenc );
1967
1968             if( !h->fenc->weight[0][0].weightfn )
1969             {
1970                 h->fenc->weight[0][0].i_denom = 0;
1971                 SET_WEIGHT( w[0], 1, 1, 0, -1 );
1972                 idx = x264_weighted_reference_duplicate( h, 0, w );
1973             }
1974             else
1975             {
1976                 if( h->fenc->weight[0][0].i_scale == 1<<h->fenc->weight[0][0].i_denom )
1977                 {
1978                     SET_WEIGHT( h->fenc->weight[0][0], 1, 1, 0, h->fenc->weight[0][0].i_offset );
1979                 }
1980                 x264_weighted_reference_duplicate( h, 0, x264_weight_none );
1981                 if( h->fenc->weight[0][0].i_offset > -128 )
1982                 {
1983                     w[0] = h->fenc->weight[0][0];
1984                     w[0].i_offset--;
1985                     h->mc.weight_cache( h, &w[0] );
1986                     idx = x264_weighted_reference_duplicate( h, 0, w );
1987                 }
1988             }
1989         }
1990         h->mb.ref_blind_dupe = idx;
1991     }
1992
1993     assert( h->i_ref[0] + h->i_ref[1] <= X264_REF_MAX );
1994     h->mb.pic.i_fref[0] = h->i_ref[0];
1995     h->mb.pic.i_fref[1] = h->i_ref[1];
1996 }
1997
1998 static void x264_fdec_filter_row( x264_t *h, int mb_y, int pass )
1999 {
2000     /* mb_y is the mb to be encoded next, not the mb to be filtered here */
2001     int b_hpel = h->fdec->b_kept_as_ref;
2002     int b_deblock = h->sh.i_disable_deblocking_filter_idc != 1;
2003     int b_end = mb_y == h->i_threadslice_end;
2004     int b_measure_quality = 1;
2005     int min_y = mb_y - (1 << SLICE_MBAFF);
2006     int b_start = min_y == h->i_threadslice_start;
2007     /* Even in interlaced mode, deblocking never modifies more than 4 pixels
2008      * above each MB, as bS=4 doesn't happen for the top of interlaced mbpairs. */
2009     int minpix_y = min_y*16 - 4 * !b_start;
2010     int maxpix_y = mb_y*16 - 4 * !b_end;
2011     b_deblock &= b_hpel || h->param.b_full_recon || h->param.psz_dump_yuv;
2012     if( h->param.b_sliced_threads )
2013     {
2014         switch( pass )
2015         {
2016             /* During encode: only do deblock if asked for */
2017             default:
2018             case 0:
2019                 b_deblock &= h->param.b_full_recon;
2020                 b_hpel = 0;
2021                 break;
2022             /* During post-encode pass: do deblock if not done yet, do hpel for all
2023              * rows except those between slices. */
2024             case 1:
2025                 b_deblock &= !h->param.b_full_recon;
2026                 b_hpel &= !(b_start && min_y > 0);
2027                 b_measure_quality = 0;
2028                 break;
2029             /* Final pass: do the rows between slices in sequence. */
2030             case 2:
2031                 b_deblock = 0;
2032                 b_measure_quality = 0;
2033                 break;
2034         }
2035     }
2036     if( mb_y & SLICE_MBAFF )
2037         return;
2038     if( min_y < h->i_threadslice_start )
2039         return;
2040
2041     if( b_deblock )
2042         for( int y = min_y; y < mb_y; y += (1 << SLICE_MBAFF) )
2043             x264_frame_deblock_row( h, y );
2044
2045     /* FIXME: Prediction requires different borders for interlaced/progressive mc,
2046      * but the actual image data is equivalent. For now, maintain this
2047      * consistency by copying deblocked pixels between planes. */
2048     if( PARAM_INTERLACED && (!h->param.b_sliced_threads || pass == 1) )
2049         for( int p = 0; p < h->fdec->i_plane; p++ )
2050             for( int i = minpix_y>>(CHROMA_V_SHIFT && p); i < maxpix_y>>(CHROMA_V_SHIFT && p); i++ )
2051                 memcpy( h->fdec->plane_fld[p] + i*h->fdec->i_stride[p],
2052                         h->fdec->plane[p]     + i*h->fdec->i_stride[p],
2053                         h->mb.i_mb_width*16*sizeof(pixel) );
2054
2055     if( h->fdec->b_kept_as_ref && (!h->param.b_sliced_threads || pass == 1) )
2056         x264_frame_expand_border( h, h->fdec, min_y );
2057     if( b_hpel )
2058     {
2059         int end = mb_y == h->mb.i_mb_height;
2060         /* Can't do hpel until the previous slice is done encoding. */
2061         if( h->param.analyse.i_subpel_refine )
2062         {
2063             x264_frame_filter( h, h->fdec, min_y, end );
2064             x264_frame_expand_border_filtered( h, h->fdec, min_y, end );
2065         }
2066     }
2067
2068     if( SLICE_MBAFF && pass == 0 )
2069         for( int i = 0; i < 3; i++ )
2070         {
2071             XCHG( pixel *, h->intra_border_backup[0][i], h->intra_border_backup[3][i] );
2072             XCHG( pixel *, h->intra_border_backup[1][i], h->intra_border_backup[4][i] );
2073         }
2074
2075     if( h->i_thread_frames > 1 && h->fdec->b_kept_as_ref )
2076         x264_frame_cond_broadcast( h->fdec, mb_y*16 + (b_end ? 10000 : -(X264_THREAD_HEIGHT << SLICE_MBAFF)) );
2077
2078     if( b_measure_quality )
2079     {
2080         maxpix_y = X264_MIN( maxpix_y, h->param.i_height );
2081         if( h->param.analyse.b_psnr )
2082         {
2083             for( int p = 0; p < (CHROMA444 ? 3 : 1); p++ )
2084                 h->stat.frame.i_ssd[p] += x264_pixel_ssd_wxh( &h->pixf,
2085                     h->fdec->plane[p] + minpix_y * h->fdec->i_stride[p], h->fdec->i_stride[p],
2086                     h->fenc->plane[p] + minpix_y * h->fenc->i_stride[p], h->fenc->i_stride[p],
2087                     h->param.i_width, maxpix_y-minpix_y );
2088             if( !CHROMA444 )
2089             {
2090                 uint64_t ssd_u, ssd_v;
2091                 int v_shift = CHROMA_V_SHIFT;
2092                 x264_pixel_ssd_nv12( &h->pixf,
2093                     h->fdec->plane[1] + (minpix_y>>v_shift) * h->fdec->i_stride[1], h->fdec->i_stride[1],
2094                     h->fenc->plane[1] + (minpix_y>>v_shift) * h->fenc->i_stride[1], h->fenc->i_stride[1],
2095                     h->param.i_width>>1, (maxpix_y-minpix_y)>>v_shift, &ssd_u, &ssd_v );
2096                 h->stat.frame.i_ssd[1] += ssd_u;
2097                 h->stat.frame.i_ssd[2] += ssd_v;
2098             }
2099         }
2100
2101         if( h->param.analyse.b_ssim )
2102         {
2103             int ssim_cnt;
2104             x264_emms();
2105             /* offset by 2 pixels to avoid alignment of ssim blocks with dct blocks,
2106              * and overlap by 4 */
2107             minpix_y += b_start ? 2 : -6;
2108             h->stat.frame.f_ssim +=
2109                 x264_pixel_ssim_wxh( &h->pixf,
2110                     h->fdec->plane[0] + 2+minpix_y*h->fdec->i_stride[0], h->fdec->i_stride[0],
2111                     h->fenc->plane[0] + 2+minpix_y*h->fenc->i_stride[0], h->fenc->i_stride[0],
2112                     h->param.i_width-2, maxpix_y-minpix_y, h->scratch_buffer, &ssim_cnt );
2113             h->stat.frame.i_ssim_cnt += ssim_cnt;
2114         }
2115     }
2116 }
2117
2118 static inline int x264_reference_update( x264_t *h )
2119 {
2120     if( !h->fdec->b_kept_as_ref )
2121     {
2122         if( h->i_thread_frames > 1 )
2123         {
2124             x264_frame_push_unused( h, h->fdec );
2125             h->fdec = x264_frame_pop_unused( h, 1 );
2126             if( !h->fdec )
2127                 return -1;
2128         }
2129         return 0;
2130     }
2131
2132     /* apply mmco from previous frame. */
2133     for( int i = 0; i < h->sh.i_mmco_command_count; i++ )
2134         for( int j = 0; h->frames.reference[j]; j++ )
2135             if( h->frames.reference[j]->i_poc == h->sh.mmco[i].i_poc )
2136                 x264_frame_push_unused( h, x264_frame_shift( &h->frames.reference[j] ) );
2137
2138     /* move frame in the buffer */
2139     x264_frame_push( h->frames.reference, h->fdec );
2140     if( h->frames.reference[h->sps->i_num_ref_frames] )
2141         x264_frame_push_unused( h, x264_frame_shift( h->frames.reference ) );
2142     h->fdec = x264_frame_pop_unused( h, 1 );
2143     if( !h->fdec )
2144         return -1;
2145     return 0;
2146 }
2147
2148 static inline void x264_reference_reset( x264_t *h )
2149 {
2150     while( h->frames.reference[0] )
2151         x264_frame_push_unused( h, x264_frame_pop( h->frames.reference ) );
2152     h->fdec->i_poc =
2153     h->fenc->i_poc = 0;
2154 }
2155
2156 static inline void x264_reference_hierarchy_reset( x264_t *h )
2157 {
2158     int ref;
2159     int b_hasdelayframe = 0;
2160
2161     /* look for delay frames -- chain must only contain frames that are disposable */
2162     for( int i = 0; h->frames.current[i] && IS_DISPOSABLE( h->frames.current[i]->i_type ); i++ )
2163         b_hasdelayframe |= h->frames.current[i]->i_coded
2164                         != h->frames.current[i]->i_frame + h->sps->vui.i_num_reorder_frames;
2165
2166     /* This function must handle b-pyramid and clear frames for open-gop */
2167     if( h->param.i_bframe_pyramid != X264_B_PYRAMID_STRICT && !b_hasdelayframe && h->frames.i_poc_last_open_gop == -1 )
2168         return;
2169
2170     /* Remove last BREF. There will never be old BREFs in the
2171      * dpb during a BREF decode when pyramid == STRICT */
2172     for( ref = 0; h->frames.reference[ref]; ref++ )
2173     {
2174         if( ( h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT
2175             && h->frames.reference[ref]->i_type == X264_TYPE_BREF )
2176             || ( h->frames.reference[ref]->i_poc < h->frames.i_poc_last_open_gop
2177             && h->sh.i_type != SLICE_TYPE_B ) )
2178         {
2179             int diff = h->i_frame_num - h->frames.reference[ref]->i_frame_num;
2180             h->sh.mmco[h->sh.i_mmco_command_count].i_difference_of_pic_nums = diff;
2181             h->sh.mmco[h->sh.i_mmco_command_count++].i_poc = h->frames.reference[ref]->i_poc;
2182             x264_frame_push_unused( h, x264_frame_shift( &h->frames.reference[ref] ) );
2183             h->b_ref_reorder[0] = 1;
2184             ref--;
2185         }
2186     }
2187
2188     /* Prepare room in the dpb for the delayed display time of the later b-frame's */
2189     if( h->param.i_bframe_pyramid )
2190         h->sh.i_mmco_remove_from_end = X264_MAX( ref + 2 - h->frames.i_max_dpb, 0 );
2191 }
2192
2193 static inline void x264_slice_init( x264_t *h, int i_nal_type, int i_global_qp )
2194 {
2195     /* ------------------------ Create slice header  ----------------------- */
2196     if( i_nal_type == NAL_SLICE_IDR )
2197     {
2198         x264_slice_header_init( h, &h->sh, h->sps, h->pps, h->i_idr_pic_id, h->i_frame_num, i_global_qp );
2199
2200         /* alternate id */
2201         h->i_idr_pic_id ^= 1;
2202     }
2203     else
2204     {
2205         x264_slice_header_init( h, &h->sh, h->sps, h->pps, -1, h->i_frame_num, i_global_qp );
2206
2207         h->sh.i_num_ref_idx_l0_active = h->i_ref[0] <= 0 ? 1 : h->i_ref[0];
2208         h->sh.i_num_ref_idx_l1_active = h->i_ref[1] <= 0 ? 1 : h->i_ref[1];
2209         if( h->sh.i_num_ref_idx_l0_active != h->pps->i_num_ref_idx_l0_default_active ||
2210             (h->sh.i_type == SLICE_TYPE_B && h->sh.i_num_ref_idx_l1_active != h->pps->i_num_ref_idx_l1_default_active) )
2211         {
2212             h->sh.b_num_ref_idx_override = 1;
2213         }
2214     }
2215
2216     if( h->fenc->i_type == X264_TYPE_BREF && h->param.b_bluray_compat && h->sh.i_mmco_command_count )
2217     {
2218         h->b_sh_backup = 1;
2219         h->sh_backup = h->sh;
2220     }
2221
2222     h->fdec->i_frame_num = h->sh.i_frame_num;
2223
2224     if( h->sps->i_poc_type == 0 )
2225     {
2226         h->sh.i_poc = h->fdec->i_poc;
2227         if( PARAM_INTERLACED )
2228         {
2229             h->sh.i_delta_poc_bottom = h->param.b_tff ? 1 : -1;
2230             h->sh.i_poc += h->sh.i_delta_poc_bottom == -1;
2231         }
2232         else
2233             h->sh.i_delta_poc_bottom = 0;
2234         h->fdec->i_delta_poc[0] = h->sh.i_delta_poc_bottom == -1;
2235         h->fdec->i_delta_poc[1] = h->sh.i_delta_poc_bottom ==  1;
2236     }
2237     else
2238     {
2239         /* Nothing to do ? */
2240     }
2241
2242     x264_macroblock_slice_init( h );
2243 }
2244
2245 typedef struct
2246 {
2247     int skip;
2248     uint8_t cabac_prevbyte;
2249     bs_t bs;
2250     x264_cabac_t cabac;
2251     x264_frame_stat_t stat;
2252     int last_qp;
2253     int last_dqp;
2254     int field_decoding_flag;
2255 } x264_bs_bak_t;
2256
2257 static ALWAYS_INLINE void x264_bitstream_backup( x264_t *h, x264_bs_bak_t *bak, int i_skip, int full )
2258 {
2259     if( full )
2260     {
2261         bak->stat = h->stat.frame;
2262         bak->last_qp = h->mb.i_last_qp;
2263         bak->last_dqp = h->mb.i_last_dqp;
2264         bak->field_decoding_flag = h->mb.field_decoding_flag;
2265     }
2266     else
2267     {
2268         bak->stat.i_mv_bits = h->stat.frame.i_mv_bits;
2269         bak->stat.i_tex_bits = h->stat.frame.i_tex_bits;
2270     }
2271     /* In the per-MB backup, we don't need the contexts because flushing the CABAC
2272      * encoder has no context dependency and in this case, a slice is ended (and
2273      * thus the content of all contexts are thrown away). */
2274     if( h->param.b_cabac )
2275     {
2276         if( full )
2277             memcpy( &bak->cabac, &h->cabac, sizeof(x264_cabac_t) );
2278         else
2279             memcpy( &bak->cabac, &h->cabac, offsetof(x264_cabac_t, f8_bits_encoded) );
2280         /* x264's CABAC writer modifies the previous byte during carry, so it has to be
2281          * backed up. */
2282         bak->cabac_prevbyte = h->cabac.p[-1];
2283     }
2284     else
2285     {
2286         bak->bs = h->out.bs;
2287         bak->skip = i_skip;
2288     }
2289 }
2290
2291 static ALWAYS_INLINE void x264_bitstream_restore( x264_t *h, x264_bs_bak_t *bak, int *skip, int full )
2292 {
2293     if( full )
2294     {
2295         h->stat.frame = bak->stat;
2296         h->mb.i_last_qp = bak->last_qp;
2297         h->mb.i_last_dqp = bak->last_dqp;
2298         h->mb.field_decoding_flag = bak->field_decoding_flag;
2299     }
2300     else
2301     {
2302         h->stat.frame.i_mv_bits = bak->stat.i_mv_bits;
2303         h->stat.frame.i_tex_bits = bak->stat.i_tex_bits;
2304     }
2305     if( h->param.b_cabac )
2306     {
2307         if( full )
2308             memcpy( &h->cabac, &bak->cabac, sizeof(x264_cabac_t) );
2309         else
2310             memcpy( &h->cabac, &bak->cabac, offsetof(x264_cabac_t, f8_bits_encoded) );
2311         h->cabac.p[-1] = bak->cabac_prevbyte;
2312     }
2313     else
2314     {
2315         h->out.bs = bak->bs;
2316         *skip = bak->skip;
2317     }
2318 }
2319
2320 static int x264_slice_write( x264_t *h )
2321 {
2322     int i_skip;
2323     int mb_xy, i_mb_x, i_mb_y;
2324     /* NALUs other than the first use a 3-byte startcode.
2325      * Add one extra byte for the rbsp, and one more for the final CABAC putbyte.
2326      * Then add an extra 5 bytes just in case, to account for random NAL escapes and
2327      * other inaccuracies. */
2328     int overhead_guess = (NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal)) + 1 + h->param.b_cabac + 5;
2329     int slice_max_size = h->param.i_slice_max_size > 0 ? (h->param.i_slice_max_size-overhead_guess)*8 : 0;
2330     int back_up_bitstream = slice_max_size || (!h->param.b_cabac && h->sps->i_profile_idc < PROFILE_HIGH);
2331     int starting_bits = bs_pos(&h->out.bs);
2332     int b_deblock = h->sh.i_disable_deblocking_filter_idc != 1;
2333     int b_hpel = h->fdec->b_kept_as_ref;
2334     int orig_last_mb = h->sh.i_last_mb;
2335     int thread_last_mb = h->i_threadslice_end * h->mb.i_mb_width - 1;
2336     uint8_t *last_emu_check;
2337 #define BS_BAK_SLICE_MAX_SIZE 0
2338 #define BS_BAK_SLICE_MIN_MBS  1
2339 #define BS_BAK_ROW_VBV        2
2340     x264_bs_bak_t bs_bak[3];
2341     b_deblock &= b_hpel || h->param.b_full_recon || h->param.psz_dump_yuv;
2342     bs_realign( &h->out.bs );
2343
2344     /* Slice */
2345     x264_nal_start( h, h->i_nal_type, h->i_nal_ref_idc );
2346     h->out.nal[h->out.i_nal].i_first_mb = h->sh.i_first_mb;
2347
2348     /* Slice header */
2349     x264_macroblock_thread_init( h );
2350
2351     /* Set the QP equal to the first QP in the slice for more accurate CABAC initialization. */
2352     h->mb.i_mb_xy = h->sh.i_first_mb;
2353     h->sh.i_qp = x264_ratecontrol_mb_qp( h );
2354     h->sh.i_qp = SPEC_QP( h->sh.i_qp );
2355     h->sh.i_qp_delta = h->sh.i_qp - h->pps->i_pic_init_qp;
2356
2357     x264_slice_header_write( &h->out.bs, &h->sh, h->i_nal_ref_idc );
2358     if( h->param.b_cabac )
2359     {
2360         /* alignment needed */
2361         bs_align_1( &h->out.bs );
2362
2363         /* init cabac */
2364         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 );
2365         x264_cabac_encode_init ( &h->cabac, h->out.bs.p, h->out.bs.p_end );
2366         last_emu_check = h->cabac.p;
2367     }
2368     else
2369         last_emu_check = h->out.bs.p;
2370     h->mb.i_last_qp = h->sh.i_qp;
2371     h->mb.i_last_dqp = 0;
2372     h->mb.field_decoding_flag = 0;
2373
2374     i_mb_y = h->sh.i_first_mb / h->mb.i_mb_width;
2375     i_mb_x = h->sh.i_first_mb % h->mb.i_mb_width;
2376     i_skip = 0;
2377
2378     while( 1 )
2379     {
2380         mb_xy = i_mb_x + i_mb_y * h->mb.i_mb_width;
2381         int mb_spos = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
2382
2383         if( i_mb_x == 0 )
2384         {
2385             if( x264_bitstream_check_buffer( h ) )
2386                 return -1;
2387             if( !(i_mb_y & SLICE_MBAFF) && h->param.rc.i_vbv_buffer_size )
2388                 x264_bitstream_backup( h, &bs_bak[BS_BAK_ROW_VBV], i_skip, 1 );
2389             if( !h->mb.b_reencode_mb )
2390                 x264_fdec_filter_row( h, i_mb_y, 0 );
2391         }
2392
2393         if( !(i_mb_y & SLICE_MBAFF) && back_up_bitstream )
2394         {
2395             x264_bitstream_backup( h, &bs_bak[BS_BAK_SLICE_MAX_SIZE], i_skip, 0 );
2396             if( slice_max_size && (thread_last_mb+1-mb_xy) == h->param.i_slice_min_mbs )
2397                 x264_bitstream_backup( h, &bs_bak[BS_BAK_SLICE_MIN_MBS], i_skip, 0 );
2398         }
2399
2400         if( PARAM_INTERLACED )
2401         {
2402             if( h->mb.b_adaptive_mbaff )
2403             {
2404                 if( !(i_mb_y&1) )
2405                 {
2406                     /* FIXME: VSAD is fast but fairly poor at choosing the best interlace type. */
2407                     h->mb.b_interlaced = x264_field_vsad( h, i_mb_x, i_mb_y );
2408                     memcpy( &h->zigzagf, MB_INTERLACED ? &h->zigzagf_interlaced : &h->zigzagf_progressive, sizeof(h->zigzagf) );
2409                     if( !MB_INTERLACED && (i_mb_y+2) == h->mb.i_mb_height )
2410                         x264_expand_border_mbpair( h, i_mb_x, i_mb_y );
2411                 }
2412             }
2413             h->mb.field[mb_xy] = MB_INTERLACED;
2414         }
2415
2416         /* load cache */
2417         if( SLICE_MBAFF )
2418             x264_macroblock_cache_load_interlaced( h, i_mb_x, i_mb_y );
2419         else
2420             x264_macroblock_cache_load_progressive( h, i_mb_x, i_mb_y );
2421
2422         x264_macroblock_analyse( h );
2423
2424         /* encode this macroblock -> be careful it can change the mb type to P_SKIP if needed */
2425 reencode:
2426         x264_macroblock_encode( h );
2427
2428         if( h->param.b_cabac )
2429         {
2430             if( mb_xy > h->sh.i_first_mb && !(SLICE_MBAFF && (i_mb_y&1)) )
2431                 x264_cabac_encode_terminal( &h->cabac );
2432
2433             if( IS_SKIP( h->mb.i_type ) )
2434                 x264_cabac_mb_skip( h, 1 );
2435             else
2436             {
2437                 if( h->sh.i_type != SLICE_TYPE_I )
2438                     x264_cabac_mb_skip( h, 0 );
2439                 x264_macroblock_write_cabac( h, &h->cabac );
2440             }
2441         }
2442         else
2443         {
2444             if( IS_SKIP( h->mb.i_type ) )
2445                 i_skip++;
2446             else
2447             {
2448                 if( h->sh.i_type != SLICE_TYPE_I )
2449                 {
2450                     bs_write_ue( &h->out.bs, i_skip );  /* skip run */
2451                     i_skip = 0;
2452                 }
2453                 x264_macroblock_write_cavlc( h );
2454                 /* If there was a CAVLC level code overflow, try again at a higher QP. */
2455                 if( h->mb.b_overflow )
2456                 {
2457                     h->mb.i_chroma_qp = h->chroma_qp_table[++h->mb.i_qp];
2458                     h->mb.i_skip_intra = 0;
2459                     h->mb.b_skip_mc = 0;
2460                     h->mb.b_overflow = 0;
2461                     x264_bitstream_restore( h, &bs_bak[BS_BAK_SLICE_MAX_SIZE], &i_skip, 0 );
2462                     goto reencode;
2463                 }
2464             }
2465         }
2466
2467         int total_bits = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
2468         int mb_size = total_bits - mb_spos;
2469
2470         if( slice_max_size && (!SLICE_MBAFF || (i_mb_y&1)) )
2471         {
2472             /* Count the skip run, just in case. */
2473             if( !h->param.b_cabac )
2474                 total_bits += bs_size_ue_big( i_skip );
2475             /* Check for escape bytes. */
2476             uint8_t *end = h->param.b_cabac ? h->cabac.p : h->out.bs.p;
2477             for( ; last_emu_check < end - 2; last_emu_check++ )
2478                 if( last_emu_check[0] == 0 && last_emu_check[1] == 0 && last_emu_check[2] <= 3 )
2479                 {
2480                     slice_max_size -= 8;
2481                     last_emu_check++;
2482                 }
2483             /* We'll just re-encode this last macroblock if we go over the max slice size. */
2484             if( total_bits - starting_bits > slice_max_size && !h->mb.b_reencode_mb )
2485             {
2486                 if( !x264_frame_new_slice( h, h->fdec ) )
2487                 {
2488                     /* Handle the most obnoxious slice-min-mbs edge case: we need to end the slice
2489                      * because it's gone over the maximum size, but doing so would violate slice-min-mbs.
2490                      * If possible, roll back to the last checkpoint and try again.
2491                      * We could try raising QP, but that would break in the case where a slice spans multiple
2492                      * rows, which the re-encoding infrastructure can't currently handle. */
2493                     if( mb_xy <= thread_last_mb && (thread_last_mb+1-mb_xy) < h->param.i_slice_min_mbs )
2494                     {
2495                         if( thread_last_mb-h->param.i_slice_min_mbs < h->sh.i_first_mb+h->param.i_slice_min_mbs )
2496                         {
2497                             x264_log( h, X264_LOG_WARNING, "slice-max-size violated (frame %d, cause: slice-min-mbs)\n", h->i_frame );
2498                             slice_max_size = 0;
2499                             goto cont;
2500                         }
2501                         x264_bitstream_restore( h, &bs_bak[BS_BAK_SLICE_MIN_MBS], &i_skip, 0 );
2502                         h->mb.b_reencode_mb = 1;
2503                         h->sh.i_last_mb = thread_last_mb-h->param.i_slice_min_mbs;
2504                         break;
2505                     }
2506                     if( mb_xy-SLICE_MBAFF*h->mb.i_mb_stride != h->sh.i_first_mb )
2507                     {
2508                         x264_bitstream_restore( h, &bs_bak[BS_BAK_SLICE_MAX_SIZE], &i_skip, 0 );
2509                         h->mb.b_reencode_mb = 1;
2510                         if( SLICE_MBAFF )
2511                         {
2512                             // set to bottom of previous mbpair
2513                             if( i_mb_x )
2514                                 h->sh.i_last_mb = mb_xy-1+h->mb.i_mb_stride*(!(i_mb_y&1));
2515                             else
2516                                 h->sh.i_last_mb = (i_mb_y-2+!(i_mb_y&1))*h->mb.i_mb_stride + h->mb.i_mb_width - 1;
2517                         }
2518                         else
2519                             h->sh.i_last_mb = mb_xy-1;
2520                         break;
2521                     }
2522                     else
2523                         h->sh.i_last_mb = mb_xy;
2524                 }
2525                 else
2526                     slice_max_size = 0;
2527             }
2528         }
2529 cont:
2530         h->mb.b_reencode_mb = 0;
2531
2532 #if HAVE_VISUALIZE
2533         if( h->param.b_visualize )
2534             x264_visualize_mb( h );
2535 #endif
2536
2537         /* save cache */
2538         x264_macroblock_cache_save( h );
2539
2540         if( x264_ratecontrol_mb( h, mb_size ) < 0 )
2541         {
2542             x264_bitstream_restore( h, &bs_bak[BS_BAK_ROW_VBV], &i_skip, 1 );
2543             h->mb.b_reencode_mb = 1;
2544             i_mb_x = 0;
2545             i_mb_y = i_mb_y - SLICE_MBAFF;
2546             h->mb.i_mb_prev_xy = i_mb_y * h->mb.i_mb_stride - 1;
2547             h->sh.i_last_mb = orig_last_mb;
2548             continue;
2549         }
2550
2551         /* accumulate mb stats */
2552         h->stat.frame.i_mb_count[h->mb.i_type]++;
2553
2554         int b_intra = IS_INTRA( h->mb.i_type );
2555         int b_skip = IS_SKIP( h->mb.i_type );
2556         if( h->param.i_log_level >= X264_LOG_INFO || h->param.rc.b_stat_write )
2557         {
2558             if( !b_intra && !b_skip && !IS_DIRECT( h->mb.i_type ) )
2559             {
2560                 if( h->mb.i_partition != D_8x8 )
2561                         h->stat.frame.i_mb_partition[h->mb.i_partition] += 4;
2562                     else
2563                         for( int i = 0; i < 4; i++ )
2564                             h->stat.frame.i_mb_partition[h->mb.i_sub_partition[i]] ++;
2565                 if( h->param.i_frame_reference > 1 )
2566                     for( int i_list = 0; i_list <= (h->sh.i_type == SLICE_TYPE_B); i_list++ )
2567                         for( int i = 0; i < 4; i++ )
2568                         {
2569                             int i_ref = h->mb.cache.ref[i_list][ x264_scan8[4*i] ];
2570                             if( i_ref >= 0 )
2571                                 h->stat.frame.i_mb_count_ref[i_list][i_ref] ++;
2572                         }
2573             }
2574         }
2575
2576         if( h->param.i_log_level >= X264_LOG_INFO )
2577         {
2578             if( h->mb.i_cbp_luma | h->mb.i_cbp_chroma )
2579             {
2580                 if( CHROMA444 )
2581                 {
2582                     for( int i = 0; i < 4; i++ )
2583                         if( h->mb.i_cbp_luma & (1 << i) )
2584                             for( int p = 0; p < 3; p++ )
2585                             {
2586                                 int s8 = i*4+p*16;
2587                                 int nnz8x8 = M16( &h->mb.cache.non_zero_count[x264_scan8[s8]+0] )
2588                                            | M16( &h->mb.cache.non_zero_count[x264_scan8[s8]+8] );
2589                                 h->stat.frame.i_mb_cbp[!b_intra + p*2] += !!nnz8x8;
2590                             }
2591                 }
2592                 else
2593                 {
2594                     int cbpsum = (h->mb.i_cbp_luma&1) + ((h->mb.i_cbp_luma>>1)&1)
2595                                + ((h->mb.i_cbp_luma>>2)&1) + (h->mb.i_cbp_luma>>3);
2596                     h->stat.frame.i_mb_cbp[!b_intra + 0] += cbpsum;
2597                     h->stat.frame.i_mb_cbp[!b_intra + 2] += !!h->mb.i_cbp_chroma;
2598                     h->stat.frame.i_mb_cbp[!b_intra + 4] += h->mb.i_cbp_chroma >> 1;
2599                 }
2600             }
2601             if( h->mb.i_cbp_luma && !b_intra )
2602             {
2603                 h->stat.frame.i_mb_count_8x8dct[0] ++;
2604                 h->stat.frame.i_mb_count_8x8dct[1] += h->mb.b_transform_8x8;
2605             }
2606             if( b_intra && h->mb.i_type != I_PCM )
2607             {
2608                 if( h->mb.i_type == I_16x16 )
2609                     h->stat.frame.i_mb_pred_mode[0][h->mb.i_intra16x16_pred_mode]++;
2610                 else if( h->mb.i_type == I_8x8 )
2611                     for( int i = 0; i < 16; i += 4 )
2612                         h->stat.frame.i_mb_pred_mode[1][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
2613                 else //if( h->mb.i_type == I_4x4 )
2614                     for( int i = 0; i < 16; i++ )
2615                         h->stat.frame.i_mb_pred_mode[2][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
2616                 h->stat.frame.i_mb_pred_mode[3][x264_mb_chroma_pred_mode_fix[h->mb.i_chroma_pred_mode]]++;
2617             }
2618             h->stat.frame.i_mb_field[b_intra?0:b_skip?2:1] += MB_INTERLACED;
2619         }
2620
2621         /* calculate deblock strength values (actual deblocking is done per-row along with hpel) */
2622         if( b_deblock )
2623             x264_macroblock_deblock_strength( h );
2624
2625         if( mb_xy == h->sh.i_last_mb )
2626             break;
2627
2628         if( SLICE_MBAFF )
2629         {
2630             i_mb_x += i_mb_y & 1;
2631             i_mb_y ^= i_mb_x < h->mb.i_mb_width;
2632         }
2633         else
2634             i_mb_x++;
2635         if( i_mb_x == h->mb.i_mb_width )
2636         {
2637             i_mb_y++;
2638             i_mb_x = 0;
2639         }
2640     }
2641     if( h->sh.i_last_mb < h->sh.i_first_mb )
2642         return 0;
2643
2644     h->out.nal[h->out.i_nal].i_last_mb = h->sh.i_last_mb;
2645
2646     if( h->param.b_cabac )
2647     {
2648         x264_cabac_encode_flush( h, &h->cabac );
2649         h->out.bs.p = h->cabac.p;
2650     }
2651     else
2652     {
2653         if( i_skip > 0 )
2654             bs_write_ue( &h->out.bs, i_skip );  /* last skip run */
2655         /* rbsp_slice_trailing_bits */
2656         bs_rbsp_trailing( &h->out.bs );
2657         bs_flush( &h->out.bs );
2658     }
2659     if( x264_nal_end( h ) )
2660         return -1;
2661
2662     if( h->sh.i_last_mb == (h->i_threadslice_end * h->mb.i_mb_width - 1) )
2663     {
2664         h->stat.frame.i_misc_bits = bs_pos( &h->out.bs )
2665                                   + (h->out.i_nal*NALU_OVERHEAD * 8)
2666                                   - h->stat.frame.i_tex_bits
2667                                   - h->stat.frame.i_mv_bits;
2668         x264_fdec_filter_row( h, h->i_threadslice_end, 0 );
2669
2670         if( h->param.b_sliced_threads )
2671         {
2672             /* Tell the main thread we're done. */
2673             x264_threadslice_cond_broadcast( h, 1 );
2674             /* Do hpel now */
2675             for( int mb_y = h->i_threadslice_start; mb_y <= h->i_threadslice_end; mb_y++ )
2676                 x264_fdec_filter_row( h, mb_y, 1 );
2677             x264_threadslice_cond_broadcast( h, 2 );
2678             /* Do the first row of hpel, now that the previous slice is done */
2679             if( h->i_thread_idx > 0 )
2680             {
2681                 x264_threadslice_cond_wait( h->thread[h->i_thread_idx-1], 2 );
2682                 x264_fdec_filter_row( h, h->i_threadslice_start + (1 << SLICE_MBAFF), 2 );
2683             }
2684         }
2685
2686         /* Free mb info after the last thread's done using it */
2687         if( h->fdec->mb_info_free && (!h->param.b_sliced_threads || h->i_thread_idx == (h->param.i_threads-1)) )
2688         {
2689             h->fdec->mb_info_free( h->fdec->mb_info );
2690             h->fdec->mb_info = NULL;
2691             h->fdec->mb_info_free = NULL;
2692         }
2693     }
2694
2695     return 0;
2696 }
2697
2698 static void x264_thread_sync_context( x264_t *dst, x264_t *src )
2699 {
2700     if( dst == src )
2701         return;
2702
2703     // reference counting
2704     for( x264_frame_t **f = src->frames.reference; *f; f++ )
2705         (*f)->i_reference_count++;
2706     for( x264_frame_t **f = dst->frames.reference; *f; f++ )
2707         x264_frame_push_unused( src, *f );
2708     src->fdec->i_reference_count++;
2709     x264_frame_push_unused( src, dst->fdec );
2710
2711     // copy everything except the per-thread pointers and the constants.
2712     memcpy( &dst->i_frame, &src->i_frame, offsetof(x264_t, mb.type) - offsetof(x264_t, i_frame) );
2713     dst->param = src->param;
2714     dst->stat = src->stat;
2715     dst->pixf = src->pixf;
2716 }
2717
2718 static void x264_thread_sync_stat( x264_t *dst, x264_t *src )
2719 {
2720     if( dst == src )
2721         return;
2722     memcpy( &dst->stat.i_frame_count, &src->stat.i_frame_count, sizeof(dst->stat) - sizeof(dst->stat.frame) );
2723 }
2724
2725 static void *x264_slices_write( x264_t *h )
2726 {
2727     int i_slice_num = 0;
2728     int last_thread_mb = h->sh.i_last_mb;
2729
2730 #if HAVE_VISUALIZE
2731     if( h->param.b_visualize )
2732         if( x264_visualize_init( h ) )
2733             goto fail;
2734 #endif
2735
2736     /* init stats */
2737     memset( &h->stat.frame, 0, sizeof(h->stat.frame) );
2738     h->mb.b_reencode_mb = 0;
2739     while( h->sh.i_first_mb + SLICE_MBAFF*h->mb.i_mb_stride <= last_thread_mb )
2740     {
2741         h->sh.i_last_mb = last_thread_mb;
2742         if( !i_slice_num || !x264_frame_new_slice( h, h->fdec ) )
2743         {
2744             if( h->param.i_slice_max_mbs )
2745             {
2746                 if( SLICE_MBAFF )
2747                 {
2748                     // convert first to mbaff form, add slice-max-mbs, then convert back to normal form
2749                     int last_mbaff = 2*(h->sh.i_first_mb % h->mb.i_mb_width)
2750                         + h->mb.i_mb_width*(h->sh.i_first_mb / h->mb.i_mb_width)
2751                         + h->param.i_slice_max_mbs - 1;
2752                     int last_x = (last_mbaff % (2*h->mb.i_mb_width))/2;
2753                     int last_y = (last_mbaff / (2*h->mb.i_mb_width))*2 + 1;
2754                     h->sh.i_last_mb = last_x + h->mb.i_mb_stride*last_y;
2755                 }
2756                 else
2757                 {
2758                     h->sh.i_last_mb = h->sh.i_first_mb + h->param.i_slice_max_mbs - 1;
2759                     if( h->sh.i_last_mb < last_thread_mb && last_thread_mb - h->sh.i_last_mb < h->param.i_slice_min_mbs )
2760                         h->sh.i_last_mb = last_thread_mb - h->param.i_slice_min_mbs;
2761                 }
2762                 i_slice_num++;
2763             }
2764             else if( h->param.i_slice_count && !h->param.b_sliced_threads )
2765             {
2766                 int height = h->mb.i_mb_height >> PARAM_INTERLACED;
2767                 int width = h->mb.i_mb_width << PARAM_INTERLACED;
2768                 i_slice_num++;
2769                 h->sh.i_last_mb = (height * i_slice_num + h->param.i_slice_count/2) / h->param.i_slice_count * width - 1;
2770             }
2771         }
2772         h->sh.i_last_mb = X264_MIN( h->sh.i_last_mb, last_thread_mb );
2773         if( x264_stack_align( x264_slice_write, h ) )
2774             goto fail;
2775         h->sh.i_first_mb = h->sh.i_last_mb + 1;
2776         // if i_first_mb is not the last mb in a row then go to the next mb in MBAFF order
2777         if( SLICE_MBAFF && h->sh.i_first_mb % h->mb.i_mb_width )
2778             h->sh.i_first_mb -= h->mb.i_mb_stride;
2779     }
2780
2781 #if HAVE_VISUALIZE
2782     if( h->param.b_visualize )
2783     {
2784         x264_visualize_show( h );
2785         x264_visualize_close( h );
2786     }
2787 #endif
2788
2789     return (void *)0;
2790
2791 fail:
2792     /* Tell other threads we're done, so they wouldn't wait for it */
2793     if( h->param.b_sliced_threads )
2794         x264_threadslice_cond_broadcast( h, 2 );
2795     return (void *)-1;
2796 }
2797
2798 static int x264_threaded_slices_write( x264_t *h )
2799 {
2800     /* set first/last mb and sync contexts */
2801     for( int i = 0; i < h->param.i_threads; i++ )
2802     {
2803         x264_t *t = h->thread[i];
2804         if( i )
2805         {
2806             t->param = h->param;
2807             memcpy( &t->i_frame, &h->i_frame, offsetof(x264_t, rc) - offsetof(x264_t, i_frame) );
2808         }
2809         int height = h->mb.i_mb_height >> PARAM_INTERLACED;
2810         t->i_threadslice_start = ((height *  i    + h->param.i_slice_count/2) / h->param.i_threads) << PARAM_INTERLACED;
2811         t->i_threadslice_end   = ((height * (i+1) + h->param.i_slice_count/2) / h->param.i_threads) << PARAM_INTERLACED;
2812         t->sh.i_first_mb = t->i_threadslice_start * h->mb.i_mb_width;
2813         t->sh.i_last_mb  =   t->i_threadslice_end * h->mb.i_mb_width - 1;
2814     }
2815
2816     x264_stack_align( x264_analyse_weight_frame, h, h->mb.i_mb_height*16 + 16 );
2817
2818     x264_threads_distribute_ratecontrol( h );
2819
2820     /* setup */
2821     for( int i = 0; i < h->param.i_threads; i++ )
2822     {
2823         h->thread[i]->i_thread_idx = i;
2824         h->thread[i]->b_thread_active = 1;
2825         x264_threadslice_cond_broadcast( h->thread[i], 0 );
2826     }
2827     /* dispatch */
2828     for( int i = 0; i < h->param.i_threads; i++ )
2829         x264_threadpool_run( h->threadpool, (void*)x264_slices_write, h->thread[i] );
2830     /* wait */
2831     for( int i = 0; i < h->param.i_threads; i++ )
2832         x264_threadslice_cond_wait( h->thread[i], 1 );
2833
2834     x264_threads_merge_ratecontrol( h );
2835
2836     for( int i = 1; i < h->param.i_threads; i++ )
2837     {
2838         x264_t *t = h->thread[i];
2839         for( int j = 0; j < t->out.i_nal; j++ )
2840         {
2841             h->out.nal[h->out.i_nal] = t->out.nal[j];
2842             h->out.i_nal++;
2843             x264_nal_check_buffer( h );
2844         }
2845         /* All entries in stat.frame are ints except for ssd/ssim. */
2846         for( int j = 0; j < (offsetof(x264_t,stat.frame.i_ssd) - offsetof(x264_t,stat.frame.i_mv_bits)) / sizeof(int); j++ )
2847             ((int*)&h->stat.frame)[j] += ((int*)&t->stat.frame)[j];
2848         for( int j = 0; j < 3; j++ )
2849             h->stat.frame.i_ssd[j] += t->stat.frame.i_ssd[j];
2850         h->stat.frame.f_ssim += t->stat.frame.f_ssim;
2851         h->stat.frame.i_ssim_cnt += t->stat.frame.i_ssim_cnt;
2852     }
2853
2854     return 0;
2855 }
2856
2857 void x264_encoder_intra_refresh( x264_t *h )
2858 {
2859     h = h->thread[h->i_thread_phase];
2860     h->b_queued_intra_refresh = 1;
2861 }
2862
2863 int x264_encoder_invalidate_reference( x264_t *h, int64_t pts )
2864 {
2865     if( h->param.i_bframe )
2866     {
2867         x264_log( h, X264_LOG_ERROR, "x264_encoder_invalidate_reference is not supported with B-frames enabled\n" );
2868         return -1;
2869     }
2870     if( h->param.b_intra_refresh )
2871     {
2872         x264_log( h, X264_LOG_ERROR, "x264_encoder_invalidate_reference is not supported with intra refresh enabled\n" );
2873         return -1;
2874     }
2875     h = h->thread[h->i_thread_phase];
2876     if( pts >= h->i_last_idr_pts )
2877     {
2878         for( int i = 0; h->frames.reference[i]; i++ )
2879             if( pts <= h->frames.reference[i]->i_pts )
2880                 h->frames.reference[i]->b_corrupt = 1;
2881         if( pts <= h->fdec->i_pts )
2882             h->fdec->b_corrupt = 1;
2883     }
2884     return 0;
2885 }
2886
2887 /****************************************************************************
2888  * x264_encoder_encode:
2889  *  XXX: i_poc   : is the poc of the current given picture
2890  *       i_frame : is the number of the frame being coded
2891  *  ex:  type frame poc
2892  *       I      0   2*0
2893  *       P      1   2*3
2894  *       B      2   2*1
2895  *       B      3   2*2
2896  *       P      4   2*6
2897  *       B      5   2*4
2898  *       B      6   2*5
2899  ****************************************************************************/
2900 int     x264_encoder_encode( x264_t *h,
2901                              x264_nal_t **pp_nal, int *pi_nal,
2902                              x264_picture_t *pic_in,
2903                              x264_picture_t *pic_out )
2904 {
2905     x264_t *thread_current, *thread_prev, *thread_oldest;
2906     int i_nal_type, i_nal_ref_idc, i_global_qp;
2907     int overhead = NALU_OVERHEAD;
2908
2909 #if HAVE_OPENCL
2910     if( h->opencl.b_fatal_error )
2911         return -1;
2912 #endif
2913
2914     if( h->i_thread_frames > 1 )
2915     {
2916         thread_prev    = h->thread[ h->i_thread_phase ];
2917         h->i_thread_phase = (h->i_thread_phase + 1) % h->i_thread_frames;
2918         thread_current = h->thread[ h->i_thread_phase ];
2919         thread_oldest  = h->thread[ (h->i_thread_phase + 1) % h->i_thread_frames ];
2920         x264_thread_sync_context( thread_current, thread_prev );
2921         x264_thread_sync_ratecontrol( thread_current, thread_prev, thread_oldest );
2922         h = thread_current;
2923     }
2924     else
2925     {
2926         thread_current =
2927         thread_oldest  = h;
2928     }
2929 #if HAVE_MMX
2930     if( h->param.cpu&X264_CPU_SSE_MISALIGN )
2931         x264_cpu_mask_misalign_sse();
2932 #endif
2933     h->i_cpb_delay_pir_offset = h->i_cpb_delay_pir_offset_next;
2934
2935     /* no data out */
2936     *pi_nal = 0;
2937     *pp_nal = NULL;
2938
2939     /* ------------------- Setup new frame from picture -------------------- */
2940     if( pic_in != NULL )
2941     {
2942         /* 1: Copy the picture to a frame and move it to a buffer */
2943         x264_frame_t *fenc = x264_frame_pop_unused( h, 0 );
2944         if( !fenc )
2945             return -1;
2946
2947         if( x264_frame_copy_picture( h, fenc, pic_in ) < 0 )
2948             return -1;
2949
2950         if( h->param.i_width != 16 * h->mb.i_mb_width ||
2951             h->param.i_height != 16 * h->mb.i_mb_height )
2952             x264_frame_expand_border_mod16( h, fenc );
2953
2954         fenc->i_frame = h->frames.i_input++;
2955
2956         if( fenc->i_frame == 0 )
2957             h->frames.i_first_pts = fenc->i_pts;
2958         if( h->frames.i_bframe_delay && fenc->i_frame == h->frames.i_bframe_delay )
2959             h->frames.i_bframe_delay_time = fenc->i_pts - h->frames.i_first_pts;
2960
2961         if( h->param.b_vfr_input && fenc->i_pts <= h->frames.i_largest_pts )
2962             x264_log( h, X264_LOG_WARNING, "non-strictly-monotonic PTS\n" );
2963
2964         h->frames.i_second_largest_pts = h->frames.i_largest_pts;
2965         h->frames.i_largest_pts = fenc->i_pts;
2966
2967         if( (fenc->i_pic_struct < PIC_STRUCT_AUTO) || (fenc->i_pic_struct > PIC_STRUCT_TRIPLE) )
2968             fenc->i_pic_struct = PIC_STRUCT_AUTO;
2969
2970         if( fenc->i_pic_struct == PIC_STRUCT_AUTO )
2971         {
2972 #if HAVE_INTERLACED
2973             int b_interlaced = fenc->param ? fenc->param->b_interlaced : h->param.b_interlaced;
2974 #else
2975             int b_interlaced = 0;
2976 #endif
2977             if( b_interlaced )
2978             {
2979                 int b_tff = fenc->param ? fenc->param->b_tff : h->param.b_tff;
2980                 fenc->i_pic_struct = b_tff ? PIC_STRUCT_TOP_BOTTOM : PIC_STRUCT_BOTTOM_TOP;
2981             }
2982             else
2983                 fenc->i_pic_struct = PIC_STRUCT_PROGRESSIVE;
2984         }
2985
2986         if( h->param.rc.b_mb_tree && h->param.rc.b_stat_read )
2987         {
2988             if( x264_macroblock_tree_read( h, fenc, pic_in->prop.quant_offsets ) )
2989                 return -1;
2990         }
2991         else
2992             x264_stack_align( x264_adaptive_quant_frame, h, fenc, pic_in->prop.quant_offsets );
2993
2994         if( pic_in->prop.quant_offsets_free )
2995             pic_in->prop.quant_offsets_free( pic_in->prop.quant_offsets );
2996
2997         if( h->frames.b_have_lowres )
2998             x264_frame_init_lowres( h, fenc );
2999
3000         /* 2: Place the frame into the queue for its slice type decision */
3001         x264_lookahead_put_frame( h, fenc );
3002
3003         if( h->frames.i_input <= h->frames.i_delay + 1 - h->i_thread_frames )
3004         {
3005             /* Nothing yet to encode, waiting for filling of buffers */
3006             pic_out->i_type = X264_TYPE_AUTO;
3007             return 0;
3008         }
3009     }
3010     else
3011     {
3012         /* signal kills for lookahead thread */
3013         x264_pthread_mutex_lock( &h->lookahead->ifbuf.mutex );
3014         h->lookahead->b_exit_thread = 1;
3015         x264_pthread_cond_broadcast( &h->lookahead->ifbuf.cv_fill );
3016         x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
3017     }
3018
3019     h->i_frame++;
3020     /* 3: The picture is analyzed in the lookahead */
3021     if( !h->frames.current[0] )
3022         x264_lookahead_get_frames( h );
3023
3024     if( !h->frames.current[0] && x264_lookahead_is_empty( h ) )
3025         return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
3026
3027     /* ------------------- Get frame to be encoded ------------------------- */
3028     /* 4: get picture to encode */
3029     h->fenc = x264_frame_shift( h->frames.current );
3030
3031     /* If applicable, wait for previous frame reconstruction to finish */
3032     if( h->param.b_sliced_threads )
3033         if( x264_threadpool_wait_all( h ) < 0 )
3034             return -1;
3035
3036     if( h->i_frame == h->i_thread_frames - 1 )
3037         h->i_reordered_pts_delay = h->fenc->i_reordered_pts;
3038     if( h->fenc->param )
3039     {
3040         x264_encoder_reconfig( h, h->fenc->param );
3041         if( h->fenc->param->param_free )
3042         {
3043             h->fenc->param->param_free( h->fenc->param );
3044             h->fenc->param = NULL;
3045         }
3046     }
3047
3048     // ok to call this before encoding any frames, since the initial values of fdec have b_kept_as_ref=0
3049     if( x264_reference_update( h ) )
3050         return -1;
3051     h->fdec->i_lines_completed = -1;
3052
3053     if( !IS_X264_TYPE_I( h->fenc->i_type ) )
3054     {
3055         int valid_refs_left = 0;
3056         for( int i = 0; h->frames.reference[i]; i++ )
3057             if( !h->frames.reference[i]->b_corrupt )
3058                 valid_refs_left++;
3059         /* No valid reference frames left: force an IDR. */
3060         if( !valid_refs_left )
3061         {
3062             h->fenc->b_keyframe = 1;
3063             h->fenc->i_type = X264_TYPE_IDR;
3064         }
3065     }
3066
3067     if( h->fenc->b_keyframe )
3068     {
3069         h->frames.i_last_keyframe = h->fenc->i_frame;
3070         if( h->fenc->i_type == X264_TYPE_IDR )
3071         {
3072             h->i_frame_num = 0;
3073             h->frames.i_last_idr = h->fenc->i_frame;
3074         }
3075     }
3076     h->sh.i_mmco_command_count =
3077     h->sh.i_mmco_remove_from_end = 0;
3078     h->b_ref_reorder[0] =
3079     h->b_ref_reorder[1] = 0;
3080     h->fdec->i_poc =
3081     h->fenc->i_poc = 2 * ( h->fenc->i_frame - X264_MAX( h->frames.i_last_idr, 0 ) );
3082
3083     /* ------------------- Setup frame context ----------------------------- */
3084     /* 5: Init data dependent of frame type */
3085     if( h->fenc->i_type == X264_TYPE_IDR )
3086     {
3087         /* reset ref pictures */
3088         i_nal_type    = NAL_SLICE_IDR;
3089         i_nal_ref_idc = NAL_PRIORITY_HIGHEST;
3090         h->sh.i_type = SLICE_TYPE_I;
3091         x264_reference_reset( h );
3092         h->frames.i_poc_last_open_gop = -1;
3093     }
3094     else if( h->fenc->i_type == X264_TYPE_I )
3095     {
3096         i_nal_type    = NAL_SLICE;
3097         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
3098         h->sh.i_type = SLICE_TYPE_I;
3099         x264_reference_hierarchy_reset( h );
3100         if( h->param.b_open_gop )
3101             h->frames.i_poc_last_open_gop = h->fenc->b_keyframe ? h->fenc->i_poc : -1;
3102     }
3103     else if( h->fenc->i_type == X264_TYPE_P )
3104     {
3105         i_nal_type    = NAL_SLICE;
3106         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
3107         h->sh.i_type = SLICE_TYPE_P;
3108         x264_reference_hierarchy_reset( h );
3109         h->frames.i_poc_last_open_gop = -1;
3110     }
3111     else if( h->fenc->i_type == X264_TYPE_BREF )
3112     {
3113         i_nal_type    = NAL_SLICE;
3114         i_nal_ref_idc = h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT ? NAL_PRIORITY_LOW : NAL_PRIORITY_HIGH;
3115         h->sh.i_type = SLICE_TYPE_B;
3116         x264_reference_hierarchy_reset( h );
3117     }
3118     else    /* B frame */
3119     {
3120         i_nal_type    = NAL_SLICE;
3121         i_nal_ref_idc = NAL_PRIORITY_DISPOSABLE;
3122         h->sh.i_type = SLICE_TYPE_B;
3123     }
3124
3125     h->fdec->i_type = h->fenc->i_type;
3126     h->fdec->i_frame = h->fenc->i_frame;
3127     h->fenc->b_kept_as_ref =
3128     h->fdec->b_kept_as_ref = i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE && h->param.i_keyint_max > 1;
3129
3130     h->fdec->mb_info = h->fenc->mb_info;
3131     h->fdec->mb_info_free = h->fenc->mb_info_free;
3132     h->fenc->mb_info = NULL;
3133     h->fenc->mb_info_free = NULL;
3134
3135     h->fdec->i_pts = h->fenc->i_pts;
3136     if( h->frames.i_bframe_delay )
3137     {
3138         int64_t *prev_reordered_pts = thread_current->frames.i_prev_reordered_pts;
3139         h->fdec->i_dts = h->i_frame > h->frames.i_bframe_delay
3140                        ? prev_reordered_pts[ (h->i_frame - h->frames.i_bframe_delay) % h->frames.i_bframe_delay ]
3141                        : h->fenc->i_reordered_pts - h->frames.i_bframe_delay_time;
3142         prev_reordered_pts[ h->i_frame % h->frames.i_bframe_delay ] = h->fenc->i_reordered_pts;
3143     }
3144     else
3145         h->fdec->i_dts = h->fenc->i_reordered_pts;
3146     if( h->fenc->i_type == X264_TYPE_IDR )
3147         h->i_last_idr_pts = h->fdec->i_pts;
3148
3149     /* ------------------- Init                ----------------------------- */
3150     /* build ref list 0/1 */
3151     x264_reference_build_list( h, h->fdec->i_poc );
3152
3153     /* ---------------------- Write the bitstream -------------------------- */
3154     /* Init bitstream context */
3155     if( h->param.b_sliced_threads )
3156     {
3157         for( int i = 0; i < h->param.i_threads; i++ )
3158         {
3159             bs_init( &h->thread[i]->out.bs, h->thread[i]->out.p_bitstream, h->thread[i]->out.i_bitstream );
3160             h->thread[i]->out.i_nal = 0;
3161         }
3162     }
3163     else
3164     {
3165         bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
3166         h->out.i_nal = 0;
3167     }
3168
3169     if( h->param.b_aud )
3170     {
3171         int pic_type;
3172
3173         if( h->sh.i_type == SLICE_TYPE_I )
3174             pic_type = 0;
3175         else if( h->sh.i_type == SLICE_TYPE_P )
3176             pic_type = 1;
3177         else if( h->sh.i_type == SLICE_TYPE_B )
3178             pic_type = 2;
3179         else
3180             pic_type = 7;
3181
3182         x264_nal_start( h, NAL_AUD, NAL_PRIORITY_DISPOSABLE );
3183         bs_write( &h->out.bs, 3, pic_type );
3184         bs_rbsp_trailing( &h->out.bs );
3185         if( x264_nal_end( h ) )
3186             return -1;
3187         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
3188     }
3189
3190     h->i_nal_type = i_nal_type;
3191     h->i_nal_ref_idc = i_nal_ref_idc;
3192
3193     if( h->param.b_intra_refresh )
3194     {
3195         if( IS_X264_TYPE_I( h->fenc->i_type ) )
3196         {
3197             h->fdec->i_frames_since_pir = 0;
3198             h->b_queued_intra_refresh = 0;
3199             /* PIR is currently only supported with ref == 1, so any intra frame effectively refreshes
3200              * the whole frame and counts as an intra refresh. */
3201             h->fdec->f_pir_position = h->mb.i_mb_width;
3202         }
3203         else if( h->fenc->i_type == X264_TYPE_P )
3204         {
3205             int pocdiff = (h->fdec->i_poc - h->fref[0][0]->i_poc)/2;
3206             float increment = X264_MAX( ((float)h->mb.i_mb_width-1) / h->param.i_keyint_max, 1 );
3207             h->fdec->f_pir_position = h->fref[0][0]->f_pir_position;
3208             h->fdec->i_frames_since_pir = h->fref[0][0]->i_frames_since_pir + pocdiff;
3209             if( h->fdec->i_frames_since_pir >= h->param.i_keyint_max ||
3210                 (h->b_queued_intra_refresh && h->fdec->f_pir_position + 0.5 >= h->mb.i_mb_width) )
3211             {
3212                 h->fdec->f_pir_position = 0;
3213                 h->fdec->i_frames_since_pir = 0;
3214                 h->b_queued_intra_refresh = 0;
3215                 h->fenc->b_keyframe = 1;
3216             }
3217             h->fdec->i_pir_start_col = h->fdec->f_pir_position+0.5;
3218             h->fdec->f_pir_position += increment * pocdiff;
3219             h->fdec->i_pir_end_col = h->fdec->f_pir_position+0.5;
3220             /* If our intra refresh has reached the right side of the frame, we're done. */
3221             if( h->fdec->i_pir_end_col >= h->mb.i_mb_width - 1 )
3222             {
3223                 h->fdec->f_pir_position = h->mb.i_mb_width;
3224                 h->fdec->i_pir_end_col = h->mb.i_mb_width - 1;
3225             }
3226         }
3227     }
3228
3229     if( h->fenc->b_keyframe )
3230     {
3231         /* Write SPS and PPS */
3232         if( h->param.b_repeat_headers )
3233         {
3234             /* generate sequence parameters */
3235             x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
3236             x264_sps_write( &h->out.bs, h->sps );
3237             if( x264_nal_end( h ) )
3238                 return -1;
3239             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
3240
3241             /* generate picture parameters */
3242             x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
3243             x264_pps_write( &h->out.bs, h->sps, h->pps );
3244             if( x264_nal_end( h ) )
3245                 return -1;
3246             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
3247         }
3248
3249         /* when frame threading is used, buffering period sei is written in x264_encoder_frame_end */
3250         if( h->i_thread_frames == 1 && h->sps->vui.b_nal_hrd_parameters_present )
3251         {
3252             x264_hrd_fullness( h );
3253             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3254             x264_sei_buffering_period_write( h, &h->out.bs );
3255             if( x264_nal_end( h ) )
3256                return -1;
3257             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
3258         }
3259     }
3260
3261     /* write extra sei */
3262     for( int i = 0; i < h->fenc->extra_sei.num_payloads; i++ )
3263     {
3264         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3265         x264_sei_write( &h->out.bs, h->fenc->extra_sei.payloads[i].payload, h->fenc->extra_sei.payloads[i].payload_size,
3266                         h->fenc->extra_sei.payloads[i].payload_type );
3267         if( x264_nal_end( h ) )
3268             return -1;
3269         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
3270         if( h->fenc->extra_sei.sei_free )
3271         {
3272             h->fenc->extra_sei.sei_free( h->fenc->extra_sei.payloads[i].payload );
3273             h->fenc->extra_sei.payloads[i].payload = NULL;
3274         }
3275     }
3276
3277     if( h->fenc->extra_sei.sei_free )
3278     {
3279         h->fenc->extra_sei.sei_free( h->fenc->extra_sei.payloads );
3280         h->fenc->extra_sei.payloads = NULL;
3281         h->fenc->extra_sei.sei_free = NULL;
3282     }
3283
3284     if( h->fenc->b_keyframe )
3285     {
3286         if( h->param.b_repeat_headers && h->fenc->i_frame == 0 )
3287         {
3288             /* identify ourself */
3289             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3290             if( x264_sei_version_write( h, &h->out.bs ) )
3291                 return -1;
3292             if( x264_nal_end( h ) )
3293                 return -1;
3294             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
3295         }
3296
3297         if( h->fenc->i_type != X264_TYPE_IDR )
3298         {
3299             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;
3300             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3301             x264_sei_recovery_point_write( h, &h->out.bs, time_to_recovery );
3302             if( x264_nal_end( h ) )
3303                 return -1;
3304             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
3305         }
3306
3307         if ( h->param.i_frame_packing >= 0 )
3308         {
3309             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3310             x264_sei_frame_packing_write( h, &h->out.bs );
3311             if( x264_nal_end( h ) )
3312                 return -1;
3313             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
3314         }
3315     }
3316
3317     /* generate sei pic timing */
3318     if( h->sps->vui.b_pic_struct_present || h->sps->vui.b_nal_hrd_parameters_present )
3319     {
3320         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3321         x264_sei_pic_timing_write( h, &h->out.bs );
3322         if( x264_nal_end( h ) )
3323             return -1;
3324         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
3325     }
3326
3327     /* As required by Blu-ray. */
3328     if( !IS_X264_TYPE_B( h->fenc->i_type ) && h->b_sh_backup )
3329     {
3330         h->b_sh_backup = 0;
3331         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3332         x264_sei_dec_ref_pic_marking_write( h, &h->out.bs );
3333         if( x264_nal_end( h ) )
3334             return -1;
3335         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
3336     }
3337
3338     if( h->fenc->b_keyframe && h->param.b_intra_refresh )
3339         h->i_cpb_delay_pir_offset_next = h->fenc->i_cpb_delay;
3340
3341     /* Init the rate control */
3342     /* FIXME: Include slice header bit cost. */
3343     x264_ratecontrol_start( h, h->fenc->i_qpplus1, overhead*8 );
3344     i_global_qp = x264_ratecontrol_qp( h );
3345
3346     pic_out->i_qpplus1 =
3347     h->fdec->i_qpplus1 = i_global_qp + 1;
3348
3349     if( h->param.rc.b_stat_read && h->sh.i_type != SLICE_TYPE_I )
3350     {
3351         x264_reference_build_list_optimal( h );
3352         x264_reference_check_reorder( h );
3353     }
3354
3355     if( h->i_ref[0] )
3356         h->fdec->i_poc_l0ref0 = h->fref[0][0]->i_poc;
3357
3358     /* ------------------------ Create slice header  ----------------------- */
3359     x264_slice_init( h, i_nal_type, i_global_qp );
3360
3361     /*------------------------- Weights -------------------------------------*/
3362     if( h->sh.i_type == SLICE_TYPE_B )
3363         x264_macroblock_bipred_init( h );
3364
3365     x264_weighted_pred_init( h );
3366
3367     if( i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE )
3368         h->i_frame_num++;
3369
3370     /* Write frame */
3371     h->i_threadslice_start = 0;
3372     h->i_threadslice_end = h->mb.i_mb_height;
3373     if( h->i_thread_frames > 1 )
3374     {
3375         x264_threadpool_run( h->threadpool, (void*)x264_slices_write, h );
3376         h->b_thread_active = 1;
3377     }
3378     else if( h->param.b_sliced_threads )
3379     {
3380         if( x264_threaded_slices_write( h ) )
3381             return -1;
3382     }
3383     else
3384         if( (intptr_t)x264_slices_write( h ) )
3385             return -1;
3386
3387     return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
3388 }
3389
3390 static int x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
3391                                    x264_nal_t **pp_nal, int *pi_nal,
3392                                    x264_picture_t *pic_out )
3393 {
3394     char psz_message[80];
3395
3396     if( !h->param.b_sliced_threads && h->b_thread_active )
3397     {
3398         h->b_thread_active = 0;
3399         if( (intptr_t)x264_threadpool_wait( h->threadpool, h ) )
3400             return -1;
3401     }
3402     if( !h->out.i_nal )
3403     {
3404         pic_out->i_type = X264_TYPE_AUTO;
3405         return 0;
3406     }
3407
3408     x264_emms();
3409
3410     /* generate buffering period sei and insert it into place */
3411     if( h->i_thread_frames > 1 && h->fenc->b_keyframe && h->sps->vui.b_nal_hrd_parameters_present )
3412     {
3413         x264_hrd_fullness( h );
3414         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3415         x264_sei_buffering_period_write( h, &h->out.bs );
3416         if( x264_nal_end( h ) )
3417            return -1;
3418         /* buffering period sei must follow AUD, SPS and PPS and precede all other SEIs */
3419         int idx = 0;
3420         while( h->out.nal[idx].i_type == NAL_AUD ||
3421                h->out.nal[idx].i_type == NAL_SPS ||
3422                h->out.nal[idx].i_type == NAL_PPS )
3423             idx++;
3424         x264_nal_t nal_tmp = h->out.nal[h->out.i_nal-1];
3425         memmove( &h->out.nal[idx+1], &h->out.nal[idx], (h->out.i_nal-idx-1)*sizeof(x264_nal_t) );
3426         h->out.nal[idx] = nal_tmp;
3427     }
3428
3429     int frame_size = x264_encoder_encapsulate_nals( h, 0 );
3430     if( frame_size < 0 )
3431         return -1;
3432
3433     /* Set output picture properties */
3434     pic_out->i_type = h->fenc->i_type;
3435
3436     pic_out->b_keyframe = h->fenc->b_keyframe;
3437     pic_out->i_pic_struct = h->fenc->i_pic_struct;
3438
3439     pic_out->i_pts = h->fdec->i_pts;
3440     pic_out->i_dts = h->fdec->i_dts;
3441
3442     if( pic_out->i_pts < pic_out->i_dts )
3443         x264_log( h, X264_LOG_WARNING, "invalid DTS: PTS is less than DTS\n" );
3444
3445     pic_out->opaque = h->fenc->opaque;
3446
3447     pic_out->img.i_csp = h->fdec->i_csp;
3448 #if HIGH_BIT_DEPTH
3449     pic_out->img.i_csp |= X264_CSP_HIGH_DEPTH;
3450 #endif
3451     pic_out->img.i_plane = h->fdec->i_plane;
3452     for( int i = 0; i < pic_out->img.i_plane; i++ )
3453     {
3454         pic_out->img.i_stride[i] = h->fdec->i_stride[i] * sizeof(pixel);
3455         pic_out->img.plane[i] = (uint8_t*)h->fdec->plane[i];
3456     }
3457
3458     x264_frame_push_unused( thread_current, h->fenc );
3459
3460     /* ---------------------- Update encoder state ------------------------- */
3461
3462     /* update rc */
3463     int filler = 0;
3464     if( x264_ratecontrol_end( h, frame_size * 8, &filler ) < 0 )
3465         return -1;
3466
3467     pic_out->hrd_timing = h->fenc->hrd_timing;
3468     pic_out->prop.f_crf_avg = h->fdec->f_crf_avg;
3469
3470     while( filler > 0 )
3471     {
3472         int f, overhead;
3473         overhead = (FILLER_OVERHEAD - h->param.b_annexb);
3474         if( h->param.i_slice_max_size && filler > h->param.i_slice_max_size )
3475         {
3476             int next_size = filler - h->param.i_slice_max_size;
3477             int overflow = X264_MAX( overhead - next_size, 0 );
3478             f = h->param.i_slice_max_size - overhead - overflow;
3479         }
3480         else
3481             f = X264_MAX( 0, filler - overhead );
3482
3483         x264_nal_start( h, NAL_FILLER, NAL_PRIORITY_DISPOSABLE );
3484         x264_filler_write( h, &h->out.bs, f );
3485         if( x264_nal_end( h ) )
3486             return -1;
3487         int total_size = x264_encoder_encapsulate_nals( h, h->out.i_nal-1 );
3488         if( total_size < 0 )
3489             return -1;
3490         frame_size += total_size;
3491         filler -= total_size;
3492     }
3493
3494     /* End bitstream, set output  */
3495     *pi_nal = h->out.i_nal;
3496     *pp_nal = h->out.nal;
3497
3498     h->out.i_nal = 0;
3499
3500     x264_noise_reduction_update( h );
3501
3502     /* ---------------------- Compute/Print statistics --------------------- */
3503     x264_thread_sync_stat( h, h->thread[0] );
3504
3505     /* Slice stat */
3506     h->stat.i_frame_count[h->sh.i_type]++;
3507     h->stat.i_frame_size[h->sh.i_type] += frame_size;
3508     h->stat.f_frame_qp[h->sh.i_type] += h->fdec->f_qp_avg_aq;
3509
3510     for( int i = 0; i < X264_MBTYPE_MAX; i++ )
3511         h->stat.i_mb_count[h->sh.i_type][i] += h->stat.frame.i_mb_count[i];
3512     for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
3513         h->stat.i_mb_partition[h->sh.i_type][i] += h->stat.frame.i_mb_partition[i];
3514     for( int i = 0; i < 2; i++ )
3515         h->stat.i_mb_count_8x8dct[i] += h->stat.frame.i_mb_count_8x8dct[i];
3516     for( int i = 0; i < 6; i++ )
3517         h->stat.i_mb_cbp[i] += h->stat.frame.i_mb_cbp[i];
3518     for( int i = 0; i < 4; i++ )
3519         for( int j = 0; j < 13; j++ )
3520             h->stat.i_mb_pred_mode[i][j] += h->stat.frame.i_mb_pred_mode[i][j];
3521     if( h->sh.i_type != SLICE_TYPE_I )
3522         for( int i_list = 0; i_list < 2; i_list++ )
3523             for( int i = 0; i < X264_REF_MAX*2; i++ )
3524                 h->stat.i_mb_count_ref[h->sh.i_type][i_list][i] += h->stat.frame.i_mb_count_ref[i_list][i];
3525     for( int i = 0; i < 3; i++ )
3526         h->stat.i_mb_field[i] += h->stat.frame.i_mb_field[i];
3527     if( h->sh.i_type == SLICE_TYPE_P && h->param.analyse.i_weighted_pred >= X264_WEIGHTP_SIMPLE )
3528     {
3529         h->stat.i_wpred[0] += !!h->sh.weight[0][0].weightfn;
3530         h->stat.i_wpred[1] += !!h->sh.weight[0][1].weightfn || !!h->sh.weight[0][2].weightfn;
3531     }
3532     if( h->sh.i_type == SLICE_TYPE_B )
3533     {
3534         h->stat.i_direct_frames[ h->sh.b_direct_spatial_mv_pred ] ++;
3535         if( h->mb.b_direct_auto_write )
3536         {
3537             //FIXME somewhat arbitrary time constants
3538             if( h->stat.i_direct_score[0] + h->stat.i_direct_score[1] > h->mb.i_mb_count )
3539                 for( int i = 0; i < 2; i++ )
3540                     h->stat.i_direct_score[i] = h->stat.i_direct_score[i] * 9/10;
3541             for( int i = 0; i < 2; i++ )
3542                 h->stat.i_direct_score[i] += h->stat.frame.i_direct_score[i];
3543         }
3544     }
3545     else
3546         h->stat.i_consecutive_bframes[h->fenc->i_bframes]++;
3547
3548     psz_message[0] = '\0';
3549     double dur = h->fenc->f_duration;
3550     h->stat.f_frame_duration[h->sh.i_type] += dur;
3551     if( h->param.analyse.b_psnr )
3552     {
3553         int64_t ssd[3] =
3554         {
3555             h->stat.frame.i_ssd[0],
3556             h->stat.frame.i_ssd[1],
3557             h->stat.frame.i_ssd[2],
3558         };
3559         int luma_size = h->param.i_width * h->param.i_height;
3560         int chroma_size = CHROMA_SIZE( luma_size );
3561         pic_out->prop.f_psnr[0] = x264_psnr( ssd[0], luma_size );
3562         pic_out->prop.f_psnr[1] = x264_psnr( ssd[1], chroma_size );
3563         pic_out->prop.f_psnr[2] = x264_psnr( ssd[2], chroma_size );
3564         pic_out->prop.f_psnr_avg = x264_psnr( ssd[0] + ssd[1] + ssd[2], luma_size + chroma_size*2 );
3565
3566         h->stat.f_ssd_global[h->sh.i_type]   += dur * (ssd[0] + ssd[1] + ssd[2]);
3567         h->stat.f_psnr_average[h->sh.i_type] += dur * pic_out->prop.f_psnr_avg;
3568         h->stat.f_psnr_mean_y[h->sh.i_type]  += dur * pic_out->prop.f_psnr[0];
3569         h->stat.f_psnr_mean_u[h->sh.i_type]  += dur * pic_out->prop.f_psnr[1];
3570         h->stat.f_psnr_mean_v[h->sh.i_type]  += dur * pic_out->prop.f_psnr[2];