c87d79de2436776e5ac866570ece87f6be5296fb
[x264/x264-sandbox.git] / encoder / encoder.c
1 /*****************************************************************************
2  * encoder.c: top-level encoder functions
3  *****************************************************************************
4  * Copyright (C) 2003-2015 x264 project
5  *
6  * Authors: Laurent Aimar <fenrir@via.ecp.fr>
7  *          Loren Merritt <lorenm@u.washington.edu>
8  *          Fiona Glaser <fiona@x264.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 #if HAVE_INTEL_DISPATCHER
36 #include "extras/intel_dispatcher.h"
37 #endif
38
39 //#define DEBUG_MB_TYPE
40
41 #define bs_write_ue bs_write_ue_big
42
43 static int x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
44                                    x264_nal_t **pp_nal, int *pi_nal,
45                                    x264_picture_t *pic_out );
46
47 /****************************************************************************
48  *
49  ******************************* x264 libs **********************************
50  *
51  ****************************************************************************/
52 static double x264_psnr( double sqe, double size )
53 {
54     double mse = sqe / (PIXEL_MAX*PIXEL_MAX * size);
55     if( mse <= 0.0000000001 ) /* Max 100dB */
56         return 100;
57
58     return -10.0 * log10( mse );
59 }
60
61 static double x264_ssim( double ssim )
62 {
63     double inv_ssim = 1 - ssim;
64     if( inv_ssim <= 0.0000000001 ) /* Max 100dB */
65         return 100;
66
67     return -10.0 * log10( inv_ssim );
68 }
69
70 static int x264_threadpool_wait_all( x264_t *h )
71 {
72     for( int i = 0; i < h->param.i_threads; i++ )
73         if( h->thread[i]->b_thread_active )
74         {
75             h->thread[i]->b_thread_active = 0;
76             if( (intptr_t)x264_threadpool_wait( h->threadpool, h->thread[i] ) < 0 )
77                 return -1;
78         }
79     return 0;
80 }
81
82 static void x264_frame_dump( x264_t *h )
83 {
84     FILE *f = x264_fopen( h->param.psz_dump_yuv, "r+b" );
85     if( !f )
86         return;
87
88     /* Wait for the threads to finish deblocking */
89     if( h->param.b_sliced_threads )
90         x264_threadpool_wait_all( h );
91
92     /* Write the frame in display order */
93     int frame_size = FRAME_SIZE( h->param.i_height * h->param.i_width * sizeof(pixel) );
94     if( !fseek( f, (int64_t)h->fdec->i_frame * frame_size, SEEK_SET ) )
95     {
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             if( planeu )
105             {
106                 pixel *planev = planeu + cw*ch + 16;
107                 h->mc.plane_copy_deinterleave( planeu, cw, planev, cw, h->fdec->plane[1], h->fdec->i_stride[1], cw, ch );
108                 fwrite( planeu, 1, cw*ch*sizeof(pixel), f );
109                 fwrite( planev, 1, cw*ch*sizeof(pixel), f );
110                 x264_free( planeu );
111             }
112         }
113     }
114     fclose( f );
115 }
116
117 /* Fill "default" values */
118 static void x264_slice_header_init( x264_t *h, x264_slice_header_t *sh,
119                                     x264_sps_t *sps, x264_pps_t *pps,
120                                     int i_idr_pic_id, int i_frame, int i_qp )
121 {
122     x264_param_t *param = &h->param;
123
124     /* First we fill all fields */
125     sh->sps = sps;
126     sh->pps = pps;
127
128     sh->i_first_mb  = 0;
129     sh->i_last_mb   = h->mb.i_mb_count - 1;
130     sh->i_pps_id    = pps->i_id;
131
132     sh->i_frame_num = i_frame;
133
134     sh->b_mbaff = PARAM_INTERLACED;
135     sh->b_field_pic = 0;    /* no field support for now */
136     sh->b_bottom_field = 0; /* not yet used */
137
138     sh->i_idr_pic_id = i_idr_pic_id;
139
140     /* poc stuff, fixed later */
141     sh->i_poc = 0;
142     sh->i_delta_poc_bottom = 0;
143     sh->i_delta_poc[0] = 0;
144     sh->i_delta_poc[1] = 0;
145
146     sh->i_redundant_pic_cnt = 0;
147
148     h->mb.b_direct_auto_write = h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO
149                                 && h->param.i_bframe
150                                 && ( h->param.rc.b_stat_write || !h->param.rc.b_stat_read );
151
152     if( !h->mb.b_direct_auto_read && sh->i_type == SLICE_TYPE_B )
153     {
154         if( h->fref[1][0]->i_poc_l0ref0 == h->fref[0][0]->i_poc )
155         {
156             if( h->mb.b_direct_auto_write )
157                 sh->b_direct_spatial_mv_pred = ( h->stat.i_direct_score[1] > h->stat.i_direct_score[0] );
158             else
159                 sh->b_direct_spatial_mv_pred = ( param->analyse.i_direct_mv_pred == X264_DIRECT_PRED_SPATIAL );
160         }
161         else
162         {
163             h->mb.b_direct_auto_write = 0;
164             sh->b_direct_spatial_mv_pred = 1;
165         }
166     }
167     /* else b_direct_spatial_mv_pred was read from the 2pass statsfile */
168
169     sh->b_num_ref_idx_override = 0;
170     sh->i_num_ref_idx_l0_active = 1;
171     sh->i_num_ref_idx_l1_active = 1;
172
173     sh->b_ref_pic_list_reordering[0] = h->b_ref_reorder[0];
174     sh->b_ref_pic_list_reordering[1] = h->b_ref_reorder[1];
175
176     /* If the ref list isn't in the default order, construct reordering header */
177     for( int list = 0; list < 2; list++ )
178     {
179         if( sh->b_ref_pic_list_reordering[list] )
180         {
181             int pred_frame_num = i_frame;
182             for( int i = 0; i < h->i_ref[list]; i++ )
183             {
184                 int diff = h->fref[list][i]->i_frame_num - pred_frame_num;
185                 sh->ref_pic_list_order[list][i].idc = ( diff > 0 );
186                 sh->ref_pic_list_order[list][i].arg = (abs(diff) - 1) & ((1 << sps->i_log2_max_frame_num) - 1);
187                 pred_frame_num = h->fref[list][i]->i_frame_num;
188             }
189         }
190     }
191
192     sh->i_cabac_init_idc = param->i_cabac_init_idc;
193
194     sh->i_qp = SPEC_QP(i_qp);
195     sh->i_qp_delta = sh->i_qp - pps->i_pic_init_qp;
196     sh->b_sp_for_swidth = 0;
197     sh->i_qs_delta = 0;
198
199     int deblock_thresh = i_qp + 2 * X264_MIN(param->i_deblocking_filter_alphac0, param->i_deblocking_filter_beta);
200     /* If effective qp <= 15, deblocking would have no effect anyway */
201     if( param->b_deblocking_filter && (h->mb.b_variable_qp || 15 < deblock_thresh ) )
202         sh->i_disable_deblocking_filter_idc = param->b_sliced_threads ? 2 : 0;
203     else
204         sh->i_disable_deblocking_filter_idc = 1;
205     sh->i_alpha_c0_offset = param->i_deblocking_filter_alphac0 << 1;
206     sh->i_beta_offset = param->i_deblocking_filter_beta << 1;
207 }
208
209 static void x264_slice_header_write( bs_t *s, x264_slice_header_t *sh, int i_nal_ref_idc )
210 {
211     if( sh->b_mbaff )
212     {
213         int first_x = sh->i_first_mb % sh->sps->i_mb_width;
214         int first_y = sh->i_first_mb / sh->sps->i_mb_width;
215         assert( (first_y&1) == 0 );
216         bs_write_ue( s, (2*first_x + sh->sps->i_mb_width*(first_y&~1) + (first_y&1)) >> 1 );
217     }
218     else
219         bs_write_ue( s, sh->i_first_mb );
220
221     bs_write_ue( s, sh->i_type + 5 );   /* same type things */
222     bs_write_ue( s, sh->i_pps_id );
223     bs_write( s, sh->sps->i_log2_max_frame_num, sh->i_frame_num & ((1<<sh->sps->i_log2_max_frame_num)-1) );
224
225     if( !sh->sps->b_frame_mbs_only )
226     {
227         bs_write1( s, sh->b_field_pic );
228         if( sh->b_field_pic )
229             bs_write1( s, sh->b_bottom_field );
230     }
231
232     if( sh->i_idr_pic_id >= 0 ) /* NAL IDR */
233         bs_write_ue( s, sh->i_idr_pic_id );
234
235     if( sh->sps->i_poc_type == 0 )
236     {
237         bs_write( s, sh->sps->i_log2_max_poc_lsb, sh->i_poc & ((1<<sh->sps->i_log2_max_poc_lsb)-1) );
238         if( sh->pps->b_pic_order && !sh->b_field_pic )
239             bs_write_se( s, sh->i_delta_poc_bottom );
240     }
241
242     if( sh->pps->b_redundant_pic_cnt )
243         bs_write_ue( s, sh->i_redundant_pic_cnt );
244
245     if( sh->i_type == SLICE_TYPE_B )
246         bs_write1( s, sh->b_direct_spatial_mv_pred );
247
248     if( sh->i_type == SLICE_TYPE_P || sh->i_type == SLICE_TYPE_B )
249     {
250         bs_write1( s, sh->b_num_ref_idx_override );
251         if( sh->b_num_ref_idx_override )
252         {
253             bs_write_ue( s, sh->i_num_ref_idx_l0_active - 1 );
254             if( sh->i_type == SLICE_TYPE_B )
255                 bs_write_ue( s, sh->i_num_ref_idx_l1_active - 1 );
256         }
257     }
258
259     /* ref pic list reordering */
260     if( sh->i_type != SLICE_TYPE_I )
261     {
262         bs_write1( s, sh->b_ref_pic_list_reordering[0] );
263         if( sh->b_ref_pic_list_reordering[0] )
264         {
265             for( int i = 0; i < sh->i_num_ref_idx_l0_active; i++ )
266             {
267                 bs_write_ue( s, sh->ref_pic_list_order[0][i].idc );
268                 bs_write_ue( s, sh->ref_pic_list_order[0][i].arg );
269             }
270             bs_write_ue( s, 3 );
271         }
272     }
273     if( sh->i_type == SLICE_TYPE_B )
274     {
275         bs_write1( s, sh->b_ref_pic_list_reordering[1] );
276         if( sh->b_ref_pic_list_reordering[1] )
277         {
278             for( int i = 0; i < sh->i_num_ref_idx_l1_active; i++ )
279             {
280                 bs_write_ue( s, sh->ref_pic_list_order[1][i].idc );
281                 bs_write_ue( s, sh->ref_pic_list_order[1][i].arg );
282             }
283             bs_write_ue( s, 3 );
284         }
285     }
286
287     sh->b_weighted_pred = 0;
288     if( sh->pps->b_weighted_pred && sh->i_type == SLICE_TYPE_P )
289     {
290         sh->b_weighted_pred = sh->weight[0][0].weightfn || sh->weight[0][1].weightfn || sh->weight[0][2].weightfn;
291         /* pred_weight_table() */
292         bs_write_ue( s, sh->weight[0][0].i_denom );
293         bs_write_ue( s, sh->weight[0][1].i_denom );
294         for( int i = 0; i < sh->i_num_ref_idx_l0_active; i++ )
295         {
296             int luma_weight_l0_flag = !!sh->weight[i][0].weightfn;
297             int chroma_weight_l0_flag = !!sh->weight[i][1].weightfn || !!sh->weight[i][2].weightfn;
298             bs_write1( s, luma_weight_l0_flag );
299             if( luma_weight_l0_flag )
300             {
301                 bs_write_se( s, sh->weight[i][0].i_scale );
302                 bs_write_se( s, sh->weight[i][0].i_offset );
303             }
304             bs_write1( s, chroma_weight_l0_flag );
305             if( chroma_weight_l0_flag )
306             {
307                 for( int j = 1; j < 3; j++ )
308                 {
309                     bs_write_se( s, sh->weight[i][j].i_scale );
310                     bs_write_se( s, sh->weight[i][j].i_offset );
311                 }
312             }
313         }
314     }
315     else if( sh->pps->b_weighted_bipred == 1 && sh->i_type == SLICE_TYPE_B )
316     {
317       /* TODO */
318     }
319
320     if( i_nal_ref_idc != 0 )
321     {
322         if( sh->i_idr_pic_id >= 0 )
323         {
324             bs_write1( s, 0 );  /* no output of prior pics flag */
325             bs_write1( s, 0 );  /* long term reference flag */
326         }
327         else
328         {
329             bs_write1( s, sh->i_mmco_command_count > 0 ); /* adaptive_ref_pic_marking_mode_flag */
330             if( sh->i_mmco_command_count > 0 )
331             {
332                 for( int i = 0; i < sh->i_mmco_command_count; i++ )
333                 {
334                     bs_write_ue( s, 1 ); /* mark short term ref as unused */
335                     bs_write_ue( s, sh->mmco[i].i_difference_of_pic_nums - 1 );
336                 }
337                 bs_write_ue( s, 0 ); /* end command list */
338             }
339         }
340     }
341
342     if( sh->pps->b_cabac && sh->i_type != SLICE_TYPE_I )
343         bs_write_ue( s, sh->i_cabac_init_idc );
344
345     bs_write_se( s, sh->i_qp_delta );      /* slice qp delta */
346
347     if( sh->pps->b_deblocking_filter_control )
348     {
349         bs_write_ue( s, sh->i_disable_deblocking_filter_idc );
350         if( sh->i_disable_deblocking_filter_idc != 1 )
351         {
352             bs_write_se( s, sh->i_alpha_c0_offset >> 1 );
353             bs_write_se( s, sh->i_beta_offset >> 1 );
354         }
355     }
356 }
357
358 /* If we are within a reasonable distance of the end of the memory allocated for the bitstream, */
359 /* reallocate, adding an arbitrary amount of space. */
360 static int x264_bitstream_check_buffer_internal( x264_t *h, int size, int b_cabac, int i_nal )
361 {
362     if( (b_cabac && (h->cabac.p_end - h->cabac.p < size)) ||
363         (h->out.bs.p_end - h->out.bs.p < size) )
364     {
365         int buf_size = h->out.i_bitstream + size;
366         uint8_t *buf = x264_malloc( buf_size );
367         if( !buf )
368             return -1;
369         int aligned_size = h->out.i_bitstream & ~15;
370         h->mc.memcpy_aligned( buf, h->out.p_bitstream, aligned_size );
371         memcpy( buf + aligned_size, h->out.p_bitstream + aligned_size, h->out.i_bitstream - aligned_size );
372
373         intptr_t delta = buf - h->out.p_bitstream;
374
375         h->out.bs.p_start += delta;
376         h->out.bs.p += delta;
377         h->out.bs.p_end = buf + buf_size;
378
379         h->cabac.p_start += delta;
380         h->cabac.p += delta;
381         h->cabac.p_end = buf + buf_size;
382
383         for( int i = 0; i <= i_nal; i++ )
384             h->out.nal[i].p_payload += delta;
385
386         x264_free( h->out.p_bitstream );
387         h->out.p_bitstream = buf;
388         h->out.i_bitstream = buf_size;
389     }
390     return 0;
391 }
392
393 static int x264_bitstream_check_buffer( x264_t *h )
394 {
395     int max_row_size = (2500 << SLICE_MBAFF) * h->mb.i_mb_width;
396     return x264_bitstream_check_buffer_internal( h, max_row_size, h->param.b_cabac, h->out.i_nal );
397 }
398
399 static int x264_bitstream_check_buffer_filler( x264_t *h, int filler )
400 {
401     filler += 32; // add padding for safety
402     return x264_bitstream_check_buffer_internal( h, filler, 0, -1 );
403 }
404
405 #if HAVE_THREAD
406 static void x264_encoder_thread_init( x264_t *h )
407 {
408     if( h->param.i_sync_lookahead )
409         x264_lower_thread_priority( 10 );
410 }
411 #endif
412
413 /****************************************************************************
414  *
415  ****************************************************************************
416  ****************************** External API*********************************
417  ****************************************************************************
418  *
419  ****************************************************************************/
420
421 static int x264_validate_parameters( x264_t *h, int b_open )
422 {
423     if( !h->param.pf_log )
424     {
425         x264_log( NULL, X264_LOG_ERROR, "pf_log not set! did you forget to call x264_param_default?\n" );
426         return -1;
427     }
428
429 #if HAVE_MMX
430     if( b_open )
431     {
432         int cpuflags = x264_cpu_detect();
433         int fail = 0;
434 #ifdef __SSE__
435         if( !(cpuflags & X264_CPU_SSE) )
436         {
437             x264_log( h, X264_LOG_ERROR, "your cpu does not support SSE1, but x264 was compiled with asm\n");
438             fail = 1;
439         }
440 #else
441         if( !(cpuflags & X264_CPU_MMX2) )
442         {
443             x264_log( h, X264_LOG_ERROR, "your cpu does not support MMXEXT, but x264 was compiled with asm\n");
444             fail = 1;
445         }
446 #endif
447         if( !fail && !(cpuflags & X264_CPU_CMOV) )
448         {
449             x264_log( h, X264_LOG_ERROR, "your cpu does not support CMOV, but x264 was compiled with asm\n");
450             fail = 1;
451         }
452         if( fail )
453         {
454             x264_log( h, X264_LOG_ERROR, "to run x264, recompile without asm (configure --disable-asm)\n");
455             return -1;
456         }
457     }
458 #endif
459
460 #if HAVE_INTERLACED
461     h->param.b_interlaced = !!PARAM_INTERLACED;
462 #else
463     if( h->param.b_interlaced )
464     {
465         x264_log( h, X264_LOG_ERROR, "not compiled with interlaced support\n" );
466         return -1;
467     }
468 #endif
469
470     if( h->param.i_width <= 0 || h->param.i_height <= 0 )
471     {
472         x264_log( h, X264_LOG_ERROR, "invalid width x height (%dx%d)\n",
473                   h->param.i_width, h->param.i_height );
474         return -1;
475     }
476
477     int i_csp = h->param.i_csp & X264_CSP_MASK;
478 #if X264_CHROMA_FORMAT
479     if( CHROMA_FORMAT != CHROMA_420 && i_csp >= X264_CSP_I420 && i_csp < X264_CSP_I422 )
480     {
481         x264_log( h, X264_LOG_ERROR, "not compiled with 4:2:0 support\n" );
482         return -1;
483     }
484     else if( CHROMA_FORMAT != CHROMA_422 && i_csp >= X264_CSP_I422 && i_csp < X264_CSP_I444 )
485     {
486         x264_log( h, X264_LOG_ERROR, "not compiled with 4:2:2 support\n" );
487         return -1;
488     }
489     else if( CHROMA_FORMAT != CHROMA_444 && i_csp >= X264_CSP_I444 && i_csp <= X264_CSP_RGB )
490     {
491         x264_log( h, X264_LOG_ERROR, "not compiled with 4:4:4 support\n" );
492         return -1;
493     }
494 #endif
495     if( i_csp <= X264_CSP_NONE || i_csp >= X264_CSP_MAX )
496     {
497         x264_log( h, X264_LOG_ERROR, "invalid CSP (only I420/YV12/NV12/NV21/I422/YV16/NV16/I444/YV24/BGR/BGRA/RGB supported)\n" );
498         return -1;
499     }
500
501     int w_mod = i_csp < X264_CSP_I444 ? 2 : 1;
502     int h_mod = (i_csp < X264_CSP_I422 ? 2 : 1) << PARAM_INTERLACED;
503     if( h->param.i_width % w_mod )
504     {
505         x264_log( h, X264_LOG_ERROR, "width not divisible by %d (%dx%d)\n",
506                   w_mod, h->param.i_width, h->param.i_height );
507         return -1;
508     }
509     if( h->param.i_height % h_mod )
510     {
511         x264_log( h, X264_LOG_ERROR, "height not divisible by %d (%dx%d)\n",
512                   h_mod, h->param.i_width, h->param.i_height );
513         return -1;
514     }
515
516     if( h->param.crop_rect.i_left   >= h->param.i_width ||
517         h->param.crop_rect.i_right  >= h->param.i_width ||
518         h->param.crop_rect.i_top    >= h->param.i_height ||
519         h->param.crop_rect.i_bottom >= h->param.i_height ||
520         h->param.crop_rect.i_left + h->param.crop_rect.i_right  >= h->param.i_width ||
521         h->param.crop_rect.i_top  + h->param.crop_rect.i_bottom >= h->param.i_height )
522     {
523         x264_log( h, X264_LOG_ERROR, "invalid crop-rect %u,%u,%u,%u\n", h->param.crop_rect.i_left,
524                   h->param.crop_rect.i_top, h->param.crop_rect.i_right,  h->param.crop_rect.i_bottom );
525         return -1;
526     }
527     if( h->param.crop_rect.i_left % w_mod || h->param.crop_rect.i_right  % w_mod ||
528         h->param.crop_rect.i_top  % h_mod || h->param.crop_rect.i_bottom % h_mod )
529     {
530         x264_log( h, X264_LOG_ERROR, "crop-rect %u,%u,%u,%u not divisible by %dx%d\n", h->param.crop_rect.i_left,
531                   h->param.crop_rect.i_top, h->param.crop_rect.i_right,  h->param.crop_rect.i_bottom, w_mod, h_mod );
532         return -1;
533     }
534
535     if( h->param.vui.i_sar_width <= 0 || h->param.vui.i_sar_height <= 0 )
536     {
537         h->param.vui.i_sar_width = 0;
538         h->param.vui.i_sar_height = 0;
539     }
540
541     if( h->param.i_threads == X264_THREADS_AUTO )
542     {
543         h->param.i_threads = x264_cpu_num_processors() * (h->param.b_sliced_threads?2:3)/2;
544         /* Avoid too many threads as they don't improve performance and
545          * complicate VBV. Capped at an arbitrary 2 rows per thread. */
546         int max_threads = X264_MAX( 1, (h->param.i_height+15)/16 / 2 );
547         h->param.i_threads = X264_MIN( h->param.i_threads, max_threads );
548     }
549     int max_sliced_threads = X264_MAX( 1, (h->param.i_height+15)/16 / 4 );
550     if( h->param.i_threads > 1 )
551     {
552 #if !HAVE_THREAD
553         x264_log( h, X264_LOG_WARNING, "not compiled with thread support!\n");
554         h->param.i_threads = 1;
555 #endif
556         /* Avoid absurdly small thread slices as they can reduce performance
557          * and VBV compliance.  Capped at an arbitrary 4 rows per thread. */
558         if( h->param.b_sliced_threads )
559             h->param.i_threads = X264_MIN( h->param.i_threads, max_sliced_threads );
560     }
561     h->param.i_threads = x264_clip3( h->param.i_threads, 1, X264_THREAD_MAX );
562     if( h->param.i_threads == 1 )
563     {
564         h->param.b_sliced_threads = 0;
565         h->param.i_lookahead_threads = 1;
566     }
567     h->i_thread_frames = h->param.b_sliced_threads ? 1 : h->param.i_threads;
568     if( h->i_thread_frames > 1 )
569         h->param.nalu_process = NULL;
570
571     if( h->param.b_opencl )
572     {
573 #if !HAVE_OPENCL
574         x264_log( h, X264_LOG_WARNING, "OpenCL: not compiled with OpenCL support, disabling\n" );
575         h->param.b_opencl = 0;
576 #elif BIT_DEPTH > 8
577         x264_log( h, X264_LOG_WARNING, "OpenCL lookahead does not support high bit depth, disabling opencl\n" );
578         h->param.b_opencl = 0;
579 #else
580         if( h->param.i_width < 32 || h->param.i_height < 32 )
581         {
582             x264_log( h, X264_LOG_WARNING, "OpenCL: frame size is too small, disabling opencl\n" );
583             h->param.b_opencl = 0;
584         }
585 #endif
586         if( h->param.opencl_device_id && h->param.i_opencl_device )
587         {
588             x264_log( h, X264_LOG_WARNING, "OpenCL: device id and device skip count configured; dropping skip\n" );
589             h->param.i_opencl_device = 0;
590         }
591     }
592
593     h->param.i_keyint_max = x264_clip3( h->param.i_keyint_max, 1, X264_KEYINT_MAX_INFINITE );
594     if( h->param.i_keyint_max == 1 )
595     {
596         h->param.b_intra_refresh = 0;
597         h->param.analyse.i_weighted_pred = 0;
598         h->param.i_frame_reference = 1;
599         h->param.i_dpb_size = 1;
600     }
601
602     if( h->param.i_frame_packing < -1 || h->param.i_frame_packing > 7 )
603     {
604         x264_log( h, X264_LOG_WARNING, "ignoring unknown frame packing value\n" );
605         h->param.i_frame_packing = -1;
606     }
607     if( h->param.i_frame_packing == 7 &&
608         ((h->param.i_width - h->param.crop_rect.i_left - h->param.crop_rect.i_right)  % 3 ||
609          (h->param.i_height - h->param.crop_rect.i_top - h->param.crop_rect.i_bottom) % 3) )
610     {
611         x264_log( h, X264_LOG_ERROR, "cropped resolution %dx%d not compatible with tile format frame packing\n",
612                   h->param.i_width - h->param.crop_rect.i_left - h->param.crop_rect.i_right,
613                   h->param.i_height - h->param.crop_rect.i_top - h->param.crop_rect.i_bottom );
614         return -1;
615     }
616
617     /* Detect default ffmpeg settings and terminate with an error. */
618     if( b_open )
619     {
620         int score = 0;
621         score += h->param.analyse.i_me_range == 0;
622         score += h->param.rc.i_qp_step == 3;
623         score += h->param.i_keyint_max == 12;
624         score += h->param.rc.i_qp_min == 2;
625         score += h->param.rc.i_qp_max == 31;
626         score += h->param.rc.f_qcompress == 0.5;
627         score += fabs(h->param.rc.f_ip_factor - 1.25) < 0.01;
628         score += fabs(h->param.rc.f_pb_factor - 1.25) < 0.01;
629         score += h->param.analyse.inter == 0 && h->param.analyse.i_subpel_refine == 8;
630         if( score >= 5 )
631         {
632             x264_log( h, X264_LOG_ERROR, "broken ffmpeg default settings detected\n" );
633             x264_log( h, X264_LOG_ERROR, "use an encoding preset (e.g. -vpre medium)\n" );
634             x264_log( h, X264_LOG_ERROR, "preset usage: -vpre <speed> -vpre <profile>\n" );
635             x264_log( h, X264_LOG_ERROR, "speed presets are listed in x264 --help\n" );
636             x264_log( h, X264_LOG_ERROR, "profile is optional; x264 defaults to high\n" );
637             return -1;
638         }
639     }
640
641     if( h->param.rc.i_rc_method < 0 || h->param.rc.i_rc_method > 2 )
642     {
643         x264_log( h, X264_LOG_ERROR, "no ratecontrol method specified\n" );
644         return -1;
645     }
646
647     if( PARAM_INTERLACED )
648         h->param.b_pic_struct = 1;
649
650     if( h->param.i_avcintra_class )
651     {
652         if( BIT_DEPTH != 10 )
653         {
654             x264_log( h, X264_LOG_ERROR, "%2d-bit AVC-Intra is not widely compatible\n", BIT_DEPTH );
655             x264_log( h, X264_LOG_ERROR, "10-bit x264 is required to encode AVC-Intra\n" );
656             return -1;
657         }
658
659         int type = h->param.i_avcintra_class == 200 ? 2 :
660                    h->param.i_avcintra_class == 100 ? 1 :
661                    h->param.i_avcintra_class == 50 ? 0 : -1;
662         if( type < 0 )
663         {
664             x264_log( h, X264_LOG_ERROR, "Invalid AVC-Intra class\n" );
665             return -1;
666         }
667
668         /* [50/100/200][res][fps] */
669         static const struct
670         {
671             uint16_t fps_num;
672             uint16_t fps_den;
673             uint8_t interlaced;
674             uint16_t frame_size;
675             const uint8_t *cqm_4ic;
676             const uint8_t *cqm_8iy;
677         } avcintra_lut[3][2][7] =
678         {
679             {{{ 60000, 1001, 0,  912, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy },
680               {    50,    1, 0, 1100, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy },
681               { 30000, 1001, 0,  912, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy },
682               {    25,    1, 0, 1100, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy },
683               { 24000, 1001, 0,  912, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy }},
684              {{ 30000, 1001, 1, 1820, x264_cqm_avci50_4ic, x264_cqm_avci50_1080i_8iy },
685               {    25,    1, 1, 2196, x264_cqm_avci50_4ic, x264_cqm_avci50_1080i_8iy },
686               { 60000, 1001, 0, 1820, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy },
687               { 30000, 1001, 0, 1820, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy },
688               {    50,    1, 0, 2196, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy },
689               {    25,    1, 0, 2196, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy },
690               { 24000, 1001, 0, 1820, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy }}},
691             {{{ 60000, 1001, 0, 1848, x264_cqm_avci100_720p_4ic, x264_cqm_avci100_720p_8iy },
692               {    50,    1, 0, 2224, x264_cqm_avci100_720p_4ic, x264_cqm_avci100_720p_8iy },
693               { 30000, 1001, 0, 1848, x264_cqm_avci100_720p_4ic, x264_cqm_avci100_720p_8iy },
694               {    25,    1, 0, 2224, x264_cqm_avci100_720p_4ic, x264_cqm_avci100_720p_8iy },
695               { 24000, 1001, 0, 1848, x264_cqm_avci100_720p_4ic, x264_cqm_avci100_720p_8iy }},
696              {{ 30000, 1001, 1, 3692, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080i_8iy },
697               {    25,    1, 1, 4444, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080i_8iy },
698               { 60000, 1001, 0, 3692, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080p_8iy },
699               { 30000, 1001, 0, 3692, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080p_8iy },
700               {    50,    1, 0, 4444, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080p_8iy },
701               {    25,    1, 0, 4444, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080p_8iy },
702               { 24000, 1001, 0, 3692, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080p_8iy }}},
703             {{{ 60000, 1001, 0, 3724, x264_cqm_avci100_720p_4ic, x264_cqm_avci100_720p_8iy },
704               {    50,    1, 0, 4472, x264_cqm_avci100_720p_4ic, x264_cqm_avci100_720p_8iy }},
705              {{ 30000, 1001, 1, 7444, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080i_8iy },
706               {    25,    1, 1, 8940, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080i_8iy },
707               { 60000, 1001, 0, 7444, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080p_8iy },
708               { 30000, 1001, 0, 7444, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080p_8iy },
709               {    50,    1, 0, 8940, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080p_8iy },
710               {    25,    1, 0, 8940, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080p_8iy },
711               { 24000, 1001, 0, 7444, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080p_8iy }}}
712         };
713
714         int res = -1;
715         if( i_csp >= X264_CSP_I420 && i_csp < X264_CSP_I422 && !type )
716         {
717             if(      h->param.i_width == 1440 && h->param.i_height == 1080 ) res =  1;
718             else if( h->param.i_width ==  960 && h->param.i_height ==  720 ) res =  0;
719         }
720         else if( i_csp >= X264_CSP_I422 && i_csp < X264_CSP_I444 && type )
721         {
722             if(      h->param.i_width == 1920 && h->param.i_height == 1080 ) res =  1;
723             else if( h->param.i_width == 1280 && h->param.i_height ==  720 ) res =  0;
724         }
725         else
726         {
727             x264_log( h, X264_LOG_ERROR, "Invalid colorspace for AVC-Intra %d\n", h->param.i_avcintra_class );
728             return -1;
729         }
730
731         if( res < 0 )
732         {
733             x264_log( h, X264_LOG_ERROR, "Resolution %dx%d invalid for AVC-Intra %d\n",
734                       h->param.i_width, h->param.i_height, h->param.i_avcintra_class );
735             return -1;
736         }
737
738         if( h->param.nalu_process )
739         {
740             x264_log( h, X264_LOG_ERROR, "nalu_process is not supported in AVC-Intra mode\n" );
741             return -1;
742         }
743
744         if( !h->param.b_repeat_headers )
745         {
746             x264_log( h, X264_LOG_ERROR, "Separate headers not supported in AVC-Intra mode\n" );
747             return -1;
748         }
749
750         int i;
751         uint32_t fps_num = h->param.i_fps_num, fps_den = h->param.i_fps_den;
752         x264_reduce_fraction( &fps_num, &fps_den );
753         for( i = 0; i < 7; i++ )
754         {
755             if( avcintra_lut[type][res][i].fps_num == fps_num &&
756                 avcintra_lut[type][res][i].fps_den == fps_den &&
757                 avcintra_lut[type][res][i].interlaced == PARAM_INTERLACED )
758             {
759                 break;
760             }
761         }
762         if( i == 7 )
763         {
764             x264_log( h, X264_LOG_ERROR, "FPS %d/%d%c not compatible with AVC-Intra\n",
765                       h->param.i_fps_num, h->param.i_fps_den, PARAM_INTERLACED ? 'i' : 'p' );
766             return -1;
767         }
768
769         h->param.i_keyint_max = 1;
770         h->param.b_intra_refresh = 0;
771         h->param.analyse.i_weighted_pred = 0;
772         h->param.i_frame_reference = 1;
773         h->param.i_dpb_size = 1;
774
775         h->param.b_bluray_compat = 0;
776         h->param.b_vfr_input = 0;
777         h->param.b_aud = 1;
778         h->param.vui.i_chroma_loc = 0;
779         h->param.i_nal_hrd = X264_NAL_HRD_NONE;
780         h->param.b_deblocking_filter = 0;
781         h->param.b_stitchable = 1;
782         h->param.b_pic_struct = 0;
783         h->param.analyse.b_transform_8x8 = 1;
784         h->param.analyse.intra = X264_ANALYSE_I8x8;
785         h->param.analyse.i_chroma_qp_offset = res && type ? 3 : 4;
786         h->param.b_cabac = !type;
787         h->param.rc.i_vbv_buffer_size = avcintra_lut[type][res][i].frame_size;
788         h->param.rc.i_vbv_max_bitrate =
789         h->param.rc.i_bitrate = h->param.rc.i_vbv_buffer_size * fps_num / fps_den;
790         h->param.rc.i_rc_method = X264_RC_ABR;
791         h->param.rc.f_vbv_buffer_init = 1.0;
792         h->param.rc.b_filler = 1;
793         h->param.i_cqm_preset = X264_CQM_CUSTOM;
794         memcpy( h->param.cqm_4iy, x264_cqm_jvt4i, sizeof(h->param.cqm_4iy) );
795         memcpy( h->param.cqm_4ic, avcintra_lut[type][res][i].cqm_4ic, sizeof(h->param.cqm_4ic) );
796         memcpy( h->param.cqm_8iy, avcintra_lut[type][res][i].cqm_8iy, sizeof(h->param.cqm_8iy) );
797
798         /* Need exactly 10 slices of equal MB count... why?  $deity knows... */
799         h->param.i_slice_max_mbs = ((h->param.i_width + 15) / 16) * ((h->param.i_height + 15) / 16) / 10;
800         h->param.i_slice_max_size = 0;
801         /* The slice structure only allows a maximum of 2 threads for 1080i/p
802          * and 1 or 5 threads for 720p */
803         if( h->param.b_sliced_threads )
804         {
805             if( res )
806                 h->param.i_threads = X264_MIN( 2, h->param.i_threads );
807             else
808             {
809                 h->param.i_threads = X264_MIN( 5, h->param.i_threads );
810                 if( h->param.i_threads < 5 )
811                     h->param.i_threads = 1;
812             }
813         }
814
815         if( type )
816             h->param.vui.i_sar_width = h->param.vui.i_sar_height = 1;
817         else
818         {
819             h->param.vui.i_sar_width  = 4;
820             h->param.vui.i_sar_height = 3;
821         }
822
823         /* Official encoder doesn't appear to go under 13
824          * and Avid cannot handle negative QPs */
825         h->param.rc.i_qp_min = X264_MAX( h->param.rc.i_qp_min, QP_BD_OFFSET + 1 );
826     }
827
828     h->param.rc.f_rf_constant = x264_clip3f( h->param.rc.f_rf_constant, -QP_BD_OFFSET, 51 );
829     h->param.rc.f_rf_constant_max = x264_clip3f( h->param.rc.f_rf_constant_max, -QP_BD_OFFSET, 51 );
830     h->param.rc.i_qp_constant = x264_clip3( h->param.rc.i_qp_constant, 0, QP_MAX );
831     h->param.analyse.i_subpel_refine = x264_clip3( h->param.analyse.i_subpel_refine, 0, 11 );
832     h->param.rc.f_ip_factor = X264_MAX( h->param.rc.f_ip_factor, 0.01f );
833     h->param.rc.f_pb_factor = X264_MAX( h->param.rc.f_pb_factor, 0.01f );
834     if( h->param.rc.i_rc_method == X264_RC_CRF )
835     {
836         h->param.rc.i_qp_constant = h->param.rc.f_rf_constant + QP_BD_OFFSET;
837         h->param.rc.i_bitrate = 0;
838     }
839     if( b_open && (h->param.rc.i_rc_method == X264_RC_CQP || h->param.rc.i_rc_method == X264_RC_CRF)
840         && h->param.rc.i_qp_constant == 0 )
841     {
842         h->mb.b_lossless = 1;
843         h->param.i_cqm_preset = X264_CQM_FLAT;
844         h->param.psz_cqm_file = NULL;
845         h->param.rc.i_rc_method = X264_RC_CQP;
846         h->param.rc.f_ip_factor = 1;
847         h->param.rc.f_pb_factor = 1;
848         h->param.analyse.b_psnr = 0;
849         h->param.analyse.b_ssim = 0;
850         h->param.analyse.i_chroma_qp_offset = 0;
851         h->param.analyse.i_trellis = 0;
852         h->param.analyse.b_fast_pskip = 0;
853         h->param.analyse.i_noise_reduction = 0;
854         h->param.analyse.b_psy = 0;
855         h->param.i_bframe = 0;
856         /* 8x8dct is not useful without RD in CAVLC lossless */
857         if( !h->param.b_cabac && h->param.analyse.i_subpel_refine < 6 )
858             h->param.analyse.b_transform_8x8 = 0;
859         h->param.analyse.inter &= ~X264_ANALYSE_I8x8;
860         h->param.analyse.intra &= ~X264_ANALYSE_I8x8;
861     }
862     if( h->param.rc.i_rc_method == X264_RC_CQP )
863     {
864         float qp_p = h->param.rc.i_qp_constant;
865         float qp_i = qp_p - 6*log2f( h->param.rc.f_ip_factor );
866         float qp_b = qp_p + 6*log2f( h->param.rc.f_pb_factor );
867         h->param.rc.i_qp_min = x264_clip3( (int)(X264_MIN3( qp_p, qp_i, qp_b )), 0, QP_MAX );
868         h->param.rc.i_qp_max = x264_clip3( (int)(X264_MAX3( qp_p, qp_i, qp_b ) + .999), 0, QP_MAX );
869         h->param.rc.i_aq_mode = 0;
870         h->param.rc.b_mb_tree = 0;
871         h->param.rc.i_bitrate = 0;
872     }
873     h->param.rc.i_qp_max = x264_clip3( h->param.rc.i_qp_max, 0, QP_MAX );
874     h->param.rc.i_qp_min = x264_clip3( h->param.rc.i_qp_min, 0, h->param.rc.i_qp_max );
875     h->param.rc.i_qp_step = x264_clip3( h->param.rc.i_qp_step, 2, QP_MAX );
876     h->param.rc.i_bitrate = x264_clip3( h->param.rc.i_bitrate, 0, 2000000 );
877     if( h->param.rc.i_rc_method == X264_RC_ABR && !h->param.rc.i_bitrate )
878     {
879         x264_log( h, X264_LOG_ERROR, "bitrate not specified\n" );
880         return -1;
881     }
882     h->param.rc.i_vbv_buffer_size = x264_clip3( h->param.rc.i_vbv_buffer_size, 0, 2000000 );
883     h->param.rc.i_vbv_max_bitrate = x264_clip3( h->param.rc.i_vbv_max_bitrate, 0, 2000000 );
884     h->param.rc.f_vbv_buffer_init = x264_clip3f( h->param.rc.f_vbv_buffer_init, 0, 2000000 );
885     if( h->param.rc.i_vbv_buffer_size )
886     {
887         if( h->param.rc.i_rc_method == X264_RC_CQP )
888         {
889             x264_log( h, X264_LOG_WARNING, "VBV is incompatible with constant QP, ignored.\n" );
890             h->param.rc.i_vbv_max_bitrate = 0;
891             h->param.rc.i_vbv_buffer_size = 0;
892         }
893         else if( h->param.rc.i_vbv_max_bitrate == 0 )
894         {
895             if( h->param.rc.i_rc_method == X264_RC_ABR )
896             {
897                 x264_log( h, X264_LOG_WARNING, "VBV maxrate unspecified, assuming CBR\n" );
898                 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate;
899             }
900             else
901             {
902                 x264_log( h, X264_LOG_WARNING, "VBV bufsize set but maxrate unspecified, ignored\n" );
903                 h->param.rc.i_vbv_buffer_size = 0;
904             }
905         }
906         else if( h->param.rc.i_vbv_max_bitrate < h->param.rc.i_bitrate &&
907                  h->param.rc.i_rc_method == X264_RC_ABR )
908         {
909             x264_log( h, X264_LOG_WARNING, "max bitrate less than average bitrate, assuming CBR\n" );
910             h->param.rc.i_bitrate = h->param.rc.i_vbv_max_bitrate;
911         }
912     }
913     else if( h->param.rc.i_vbv_max_bitrate )
914     {
915         x264_log( h, X264_LOG_WARNING, "VBV maxrate specified, but no bufsize, ignored\n" );
916         h->param.rc.i_vbv_max_bitrate = 0;
917     }
918
919     h->param.i_slice_max_size = X264_MAX( h->param.i_slice_max_size, 0 );
920     h->param.i_slice_max_mbs = X264_MAX( h->param.i_slice_max_mbs, 0 );
921     h->param.i_slice_min_mbs = X264_MAX( h->param.i_slice_min_mbs, 0 );
922     if( h->param.i_slice_max_mbs )
923         h->param.i_slice_min_mbs = X264_MIN( h->param.i_slice_min_mbs, h->param.i_slice_max_mbs/2 );
924     else if( !h->param.i_slice_max_size )
925         h->param.i_slice_min_mbs = 0;
926     if( PARAM_INTERLACED && h->param.i_slice_min_mbs )
927     {
928         x264_log( h, X264_LOG_WARNING, "interlace + slice-min-mbs is not implemented\n" );
929         h->param.i_slice_min_mbs = 0;
930     }
931     int mb_width = (h->param.i_width+15)/16;
932     if( h->param.i_slice_min_mbs > mb_width )
933     {
934         x264_log( h, X264_LOG_WARNING, "slice-min-mbs > row mb size (%d) not implemented\n", mb_width );
935         h->param.i_slice_min_mbs = mb_width;
936     }
937
938     int max_slices = (h->param.i_height+((16<<PARAM_INTERLACED)-1))/(16<<PARAM_INTERLACED);
939     if( h->param.b_sliced_threads )
940         h->param.i_slice_count = x264_clip3( h->param.i_threads, 0, max_slices );
941     else
942     {
943         h->param.i_slice_count = x264_clip3( h->param.i_slice_count, 0, max_slices );
944         if( h->param.i_slice_max_mbs || h->param.i_slice_max_size )
945             h->param.i_slice_count = 0;
946     }
947     if( h->param.i_slice_count_max > 0 )
948         h->param.i_slice_count_max = X264_MAX( h->param.i_slice_count, h->param.i_slice_count_max );
949
950     if( h->param.b_bluray_compat )
951     {
952         h->param.i_bframe_pyramid = X264_MIN( X264_B_PYRAMID_STRICT, h->param.i_bframe_pyramid );
953         h->param.i_bframe = X264_MIN( h->param.i_bframe, 3 );
954         h->param.b_aud = 1;
955         h->param.i_nal_hrd = X264_MAX( h->param.i_nal_hrd, X264_NAL_HRD_VBR );
956         h->param.i_slice_max_size = 0;
957         h->param.i_slice_max_mbs = 0;
958         h->param.b_intra_refresh = 0;
959         h->param.i_frame_reference = X264_MIN( h->param.i_frame_reference, 6 );
960         h->param.i_dpb_size = X264_MIN( h->param.i_dpb_size, 6 );
961         /* Don't use I-frames, because Blu-ray treats them the same as IDR. */
962         h->param.i_keyint_min = 1;
963         /* Due to the proliferation of broken players that don't handle dupes properly. */
964         h->param.analyse.i_weighted_pred = X264_MIN( h->param.analyse.i_weighted_pred, X264_WEIGHTP_SIMPLE );
965         if( h->param.b_fake_interlaced )
966             h->param.b_pic_struct = 1;
967     }
968
969     h->param.i_frame_reference = x264_clip3( h->param.i_frame_reference, 1, X264_REF_MAX );
970     h->param.i_dpb_size = x264_clip3( h->param.i_dpb_size, 1, X264_REF_MAX );
971     if( h->param.i_scenecut_threshold < 0 )
972         h->param.i_scenecut_threshold = 0;
973     h->param.analyse.i_direct_mv_pred = x264_clip3( h->param.analyse.i_direct_mv_pred, X264_DIRECT_PRED_NONE, X264_DIRECT_PRED_AUTO );
974     if( !h->param.analyse.i_subpel_refine && h->param.analyse.i_direct_mv_pred > X264_DIRECT_PRED_SPATIAL )
975     {
976         x264_log( h, X264_LOG_WARNING, "subme=0 + direct=temporal is not supported\n" );
977         h->param.analyse.i_direct_mv_pred = X264_DIRECT_PRED_SPATIAL;
978     }
979     h->param.i_bframe = x264_clip3( h->param.i_bframe, 0, X264_MIN( X264_BFRAME_MAX, h->param.i_keyint_max-1 ) );
980     h->param.i_bframe_bias = x264_clip3( h->param.i_bframe_bias, -90, 100 );
981     if( h->param.i_bframe <= 1 )
982         h->param.i_bframe_pyramid = X264_B_PYRAMID_NONE;
983     h->param.i_bframe_pyramid = x264_clip3( h->param.i_bframe_pyramid, X264_B_PYRAMID_NONE, X264_B_PYRAMID_NORMAL );
984     h->param.i_bframe_adaptive = x264_clip3( h->param.i_bframe_adaptive, X264_B_ADAPT_NONE, X264_B_ADAPT_TRELLIS );
985     if( !h->param.i_bframe )
986     {
987         h->param.i_bframe_adaptive = X264_B_ADAPT_NONE;
988         h->param.analyse.i_direct_mv_pred = 0;
989         h->param.analyse.b_weighted_bipred = 0;
990         h->param.b_open_gop = 0;
991     }
992     if( h->param.b_intra_refresh && h->param.i_bframe_pyramid == X264_B_PYRAMID_NORMAL )
993     {
994         x264_log( h, X264_LOG_WARNING, "b-pyramid normal + intra-refresh is not supported\n" );
995         h->param.i_bframe_pyramid = X264_B_PYRAMID_STRICT;
996     }
997     if( h->param.b_intra_refresh && (h->param.i_frame_reference > 1 || h->param.i_dpb_size > 1) )
998     {
999         x264_log( h, X264_LOG_WARNING, "ref > 1 + intra-refresh is not supported\n" );
1000         h->param.i_frame_reference = 1;
1001         h->param.i_dpb_size = 1;
1002     }
1003     if( h->param.b_intra_refresh && h->param.b_open_gop )
1004     {
1005         x264_log( h, X264_LOG_WARNING, "intra-refresh is not compatible with open-gop\n" );
1006         h->param.b_open_gop = 0;
1007     }
1008     if( !h->param.i_fps_num || !h->param.i_fps_den )
1009     {
1010         h->param.i_fps_num = 25;
1011         h->param.i_fps_den = 1;
1012     }
1013     float fps = (float)h->param.i_fps_num / h->param.i_fps_den;
1014     if( h->param.i_keyint_min == X264_KEYINT_MIN_AUTO )
1015         h->param.i_keyint_min = X264_MIN( h->param.i_keyint_max / 10, (int)fps );
1016     h->param.i_keyint_min = x264_clip3( h->param.i_keyint_min, 1, h->param.i_keyint_max/2+1 );
1017     h->param.rc.i_lookahead = x264_clip3( h->param.rc.i_lookahead, 0, X264_LOOKAHEAD_MAX );
1018     {
1019         int maxrate = X264_MAX( h->param.rc.i_vbv_max_bitrate, h->param.rc.i_bitrate );
1020         float bufsize = maxrate ? (float)h->param.rc.i_vbv_buffer_size / maxrate : 0;
1021         h->param.rc.i_lookahead = X264_MIN( h->param.rc.i_lookahead, X264_MAX( h->param.i_keyint_max, bufsize*fps ) );
1022     }
1023
1024     if( !h->param.i_timebase_num || !h->param.i_timebase_den || !(h->param.b_vfr_input || h->param.b_pulldown) )
1025     {
1026         h->param.i_timebase_num = h->param.i_fps_den;
1027         h->param.i_timebase_den = h->param.i_fps_num;
1028     }
1029
1030     h->param.rc.f_qcompress = x264_clip3f( h->param.rc.f_qcompress, 0.0, 1.0 );
1031     if( h->param.i_keyint_max == 1 || h->param.rc.f_qcompress == 1 )
1032         h->param.rc.b_mb_tree = 0;
1033     if( (!h->param.b_intra_refresh && h->param.i_keyint_max != X264_KEYINT_MAX_INFINITE) &&
1034         !h->param.rc.i_lookahead && h->param.rc.b_mb_tree )
1035     {
1036         x264_log( h, X264_LOG_WARNING, "lookaheadless mb-tree requires intra refresh or infinite keyint\n" );
1037         h->param.rc.b_mb_tree = 0;
1038     }
1039     if( b_open && h->param.rc.b_stat_read )
1040         h->param.rc.i_lookahead = 0;
1041 #if HAVE_THREAD
1042     if( h->param.i_sync_lookahead < 0 )
1043         h->param.i_sync_lookahead = h->param.i_bframe + 1;
1044     h->param.i_sync_lookahead = X264_MIN( h->param.i_sync_lookahead, X264_LOOKAHEAD_MAX );
1045     if( h->param.rc.b_stat_read || h->i_thread_frames == 1 )
1046         h->param.i_sync_lookahead = 0;
1047 #else
1048     h->param.i_sync_lookahead = 0;
1049 #endif
1050
1051     h->param.i_deblocking_filter_alphac0 = x264_clip3( h->param.i_deblocking_filter_alphac0, -6, 6 );
1052     h->param.i_deblocking_filter_beta    = x264_clip3( h->param.i_deblocking_filter_beta, -6, 6 );
1053     h->param.analyse.i_luma_deadzone[0] = x264_clip3( h->param.analyse.i_luma_deadzone[0], 0, 32 );
1054     h->param.analyse.i_luma_deadzone[1] = x264_clip3( h->param.analyse.i_luma_deadzone[1], 0, 32 );
1055
1056     h->param.i_cabac_init_idc = x264_clip3( h->param.i_cabac_init_idc, 0, 2 );
1057
1058     if( h->param.i_cqm_preset < X264_CQM_FLAT || h->param.i_cqm_preset > X264_CQM_CUSTOM )
1059         h->param.i_cqm_preset = X264_CQM_FLAT;
1060
1061     if( h->param.analyse.i_me_method < X264_ME_DIA ||
1062         h->param.analyse.i_me_method > X264_ME_TESA )
1063         h->param.analyse.i_me_method = X264_ME_HEX;
1064     h->param.analyse.i_me_range = x264_clip3( h->param.analyse.i_me_range, 4, 1024 );
1065     if( h->param.analyse.i_me_range > 16 && h->param.analyse.i_me_method <= X264_ME_HEX )
1066         h->param.analyse.i_me_range = 16;
1067     if( h->param.analyse.i_me_method == X264_ME_TESA &&
1068         (h->mb.b_lossless || h->param.analyse.i_subpel_refine <= 1) )
1069         h->param.analyse.i_me_method = X264_ME_ESA;
1070     h->param.analyse.b_mixed_references = h->param.analyse.b_mixed_references && h->param.i_frame_reference > 1;
1071     h->param.analyse.inter &= X264_ANALYSE_PSUB16x16|X264_ANALYSE_PSUB8x8|X264_ANALYSE_BSUB16x16|
1072                               X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
1073     h->param.analyse.intra &= X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
1074     if( !(h->param.analyse.inter & X264_ANALYSE_PSUB16x16) )
1075         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
1076     if( !h->param.analyse.b_transform_8x8 )
1077     {
1078         h->param.analyse.inter &= ~X264_ANALYSE_I8x8;
1079         h->param.analyse.intra &= ~X264_ANALYSE_I8x8;
1080     }
1081     h->param.analyse.i_trellis = x264_clip3( h->param.analyse.i_trellis, 0, 2 );
1082     h->param.rc.i_aq_mode = x264_clip3( h->param.rc.i_aq_mode, 0, 3 );
1083     h->param.rc.f_aq_strength = x264_clip3f( h->param.rc.f_aq_strength, 0, 3 );
1084     if( h->param.rc.f_aq_strength == 0 )
1085         h->param.rc.i_aq_mode = 0;
1086
1087     if( h->param.i_log_level < X264_LOG_INFO )
1088     {
1089         h->param.analyse.b_psnr = 0;
1090         h->param.analyse.b_ssim = 0;
1091     }
1092     /* Warn users trying to measure PSNR/SSIM with psy opts on. */
1093     if( b_open && (h->param.analyse.b_psnr || h->param.analyse.b_ssim) )
1094     {
1095         char *s = NULL;
1096
1097         if( h->param.analyse.b_psy )
1098         {
1099             s = h->param.analyse.b_psnr ? "psnr" : "ssim";
1100             x264_log( h, X264_LOG_WARNING, "--%s used with psy on: results will be invalid!\n", s );
1101         }
1102         else if( !h->param.rc.i_aq_mode && h->param.analyse.b_ssim )
1103         {
1104             x264_log( h, X264_LOG_WARNING, "--ssim used with AQ off: results will be invalid!\n" );
1105             s = "ssim";
1106         }
1107         else if(  h->param.rc.i_aq_mode && h->param.analyse.b_psnr )
1108         {
1109             x264_log( h, X264_LOG_WARNING, "--psnr used with AQ on: results will be invalid!\n" );
1110             s = "psnr";
1111         }
1112         if( s )
1113             x264_log( h, X264_LOG_WARNING, "--tune %s should be used if attempting to benchmark %s!\n", s, s );
1114     }
1115
1116     if( !h->param.analyse.b_psy )
1117     {
1118         h->param.analyse.f_psy_rd = 0;
1119         h->param.analyse.f_psy_trellis = 0;
1120     }
1121     h->param.analyse.f_psy_rd = x264_clip3f( h->param.analyse.f_psy_rd, 0, 10 );
1122     h->param.analyse.f_psy_trellis = x264_clip3f( h->param.analyse.f_psy_trellis, 0, 10 );
1123     h->mb.i_psy_rd = h->param.analyse.i_subpel_refine >= 6 ? FIX8( h->param.analyse.f_psy_rd ) : 0;
1124     h->mb.i_psy_trellis = h->param.analyse.i_trellis ? FIX8( h->param.analyse.f_psy_trellis / 4 ) : 0;
1125     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -32, 32);
1126     /* In 4:4:4 mode, chroma gets twice as much resolution, so we can halve its quality. */
1127     if( b_open && i_csp >= X264_CSP_I444 && i_csp < X264_CSP_BGR && h->param.analyse.b_psy )
1128         h->param.analyse.i_chroma_qp_offset += 6;
1129     /* Psy RDO increases overall quantizers to improve the quality of luma--this indirectly hurts chroma quality */
1130     /* so we lower the chroma QP offset to compensate */
1131     if( b_open && h->mb.i_psy_rd && !h->param.i_avcintra_class )
1132         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_rd < 0.25 ? 1 : 2;
1133     /* Psy trellis has a similar effect. */
1134     if( b_open && h->mb.i_psy_trellis && !h->param.i_avcintra_class )
1135         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_trellis < 0.25 ? 1 : 2;
1136     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -12, 12);
1137     /* MB-tree requires AQ to be on, even if the strength is zero. */
1138     if( !h->param.rc.i_aq_mode && h->param.rc.b_mb_tree )
1139     {
1140         h->param.rc.i_aq_mode = 1;
1141         h->param.rc.f_aq_strength = 0;
1142     }
1143     h->param.analyse.i_noise_reduction = x264_clip3( h->param.analyse.i_noise_reduction, 0, 1<<16 );
1144     if( h->param.analyse.i_subpel_refine >= 10 && (h->param.analyse.i_trellis != 2 || !h->param.rc.i_aq_mode) )
1145         h->param.analyse.i_subpel_refine = 9;
1146
1147     {
1148         const x264_level_t *l = x264_levels;
1149         if( h->param.i_level_idc < 0 )
1150         {
1151             int maxrate_bak = h->param.rc.i_vbv_max_bitrate;
1152             if( h->param.rc.i_rc_method == X264_RC_ABR && h->param.rc.i_vbv_buffer_size <= 0 )
1153                 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate * 2;
1154             x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
1155             do h->param.i_level_idc = l->level_idc;
1156                 while( l[1].level_idc && x264_validate_levels( h, 0 ) && l++ );
1157             h->param.rc.i_vbv_max_bitrate = maxrate_bak;
1158         }
1159         else
1160         {
1161             while( l->level_idc && l->level_idc != h->param.i_level_idc )
1162                 l++;
1163             if( l->level_idc == 0 )
1164             {
1165                 x264_log( h, X264_LOG_ERROR, "invalid level_idc: %d\n", h->param.i_level_idc );
1166                 return -1;
1167             }
1168         }
1169         if( h->param.analyse.i_mv_range <= 0 )
1170             h->param.analyse.i_mv_range = l->mv_range >> PARAM_INTERLACED;
1171         else
1172             h->param.analyse.i_mv_range = x264_clip3(h->param.analyse.i_mv_range, 32, 512 >> PARAM_INTERLACED);
1173     }
1174
1175     h->param.analyse.i_weighted_pred = x264_clip3( h->param.analyse.i_weighted_pred, X264_WEIGHTP_NONE, X264_WEIGHTP_SMART );
1176
1177     if( h->param.i_lookahead_threads == X264_THREADS_AUTO )
1178     {
1179         if( h->param.b_sliced_threads )
1180             h->param.i_lookahead_threads = h->param.i_threads;
1181         else
1182         {
1183             /* If we're using much slower lookahead settings than encoding settings, it helps a lot to use
1184              * more lookahead threads.  This typically happens in the first pass of a two-pass encode, so
1185              * try to guess at this sort of case.
1186              *
1187              * Tuned by a little bit of real encoding with the various presets. */
1188             int badapt = h->param.i_bframe_adaptive == X264_B_ADAPT_TRELLIS;
1189             int subme = X264_MIN( h->param.analyse.i_subpel_refine / 3, 3 ) + (h->param.analyse.i_subpel_refine > 1);
1190             int bframes = X264_MIN( (h->param.i_bframe - 1) / 3, 3 );
1191
1192             /* [b-adapt 0/1 vs 2][quantized subme][quantized bframes] */
1193             static const uint8_t lookahead_thread_div[2][5][4] =
1194             {{{6,6,6,6}, {3,3,3,3}, {4,4,4,4}, {6,6,6,6}, {12,12,12,12}},
1195              {{3,2,1,1}, {2,1,1,1}, {4,3,2,1}, {6,4,3,2}, {12, 9, 6, 4}}};
1196
1197             h->param.i_lookahead_threads = h->param.i_threads / lookahead_thread_div[badapt][subme][bframes];
1198             /* Since too many lookahead threads significantly degrades lookahead accuracy, limit auto
1199              * lookahead threads to about 8 macroblock rows high each at worst.  This number is chosen
1200              * pretty much arbitrarily. */
1201             h->param.i_lookahead_threads = X264_MIN( h->param.i_lookahead_threads, h->param.i_height / 128 );
1202         }
1203     }
1204     h->param.i_lookahead_threads = x264_clip3( h->param.i_lookahead_threads, 1, X264_MIN( max_sliced_threads, X264_LOOKAHEAD_THREAD_MAX ) );
1205
1206     if( PARAM_INTERLACED )
1207     {
1208         if( h->param.analyse.i_me_method >= X264_ME_ESA )
1209         {
1210             x264_log( h, X264_LOG_WARNING, "interlace + me=esa is not implemented\n" );
1211             h->param.analyse.i_me_method = X264_ME_UMH;
1212         }
1213         if( h->param.analyse.i_weighted_pred > 0 )
1214         {
1215             x264_log( h, X264_LOG_WARNING, "interlace + weightp is not implemented\n" );
1216             h->param.analyse.i_weighted_pred = X264_WEIGHTP_NONE;
1217         }
1218     }
1219
1220     if( !h->param.analyse.i_weighted_pred && h->param.rc.b_mb_tree && h->param.analyse.b_psy )
1221         h->param.analyse.i_weighted_pred = X264_WEIGHTP_FAKE;
1222
1223     if( h->i_thread_frames > 1 )
1224     {
1225         int r = h->param.analyse.i_mv_range_thread;
1226         int r2;
1227         if( r <= 0 )
1228         {
1229             // half of the available space is reserved and divided evenly among the threads,
1230             // the rest is allocated to whichever thread is far enough ahead to use it.
1231             // reserving more space increases quality for some videos, but costs more time
1232             // in thread synchronization.
1233             int max_range = (h->param.i_height + X264_THREAD_HEIGHT) / h->i_thread_frames - X264_THREAD_HEIGHT;
1234             r = max_range / 2;
1235         }
1236         r = X264_MAX( r, h->param.analyse.i_me_range );
1237         r = X264_MIN( r, h->param.analyse.i_mv_range );
1238         // round up to use the whole mb row
1239         r2 = (r & ~15) + ((-X264_THREAD_HEIGHT) & 15);
1240         if( r2 < r )
1241             r2 += 16;
1242         x264_log( h, X264_LOG_DEBUG, "using mv_range_thread = %d\n", r2 );
1243         h->param.analyse.i_mv_range_thread = r2;
1244     }
1245
1246     if( h->param.rc.f_rate_tolerance < 0 )
1247         h->param.rc.f_rate_tolerance = 0;
1248     if( h->param.rc.f_qblur < 0 )
1249         h->param.rc.f_qblur = 0;
1250     if( h->param.rc.f_complexity_blur < 0 )
1251         h->param.rc.f_complexity_blur = 0;
1252
1253     h->param.i_sps_id &= 31;
1254
1255     h->param.i_nal_hrd = x264_clip3( h->param.i_nal_hrd, X264_NAL_HRD_NONE, X264_NAL_HRD_CBR );
1256
1257     if( h->param.i_nal_hrd && !h->param.rc.i_vbv_buffer_size )
1258     {
1259         x264_log( h, X264_LOG_WARNING, "NAL HRD parameters require VBV parameters\n" );
1260         h->param.i_nal_hrd = X264_NAL_HRD_NONE;
1261     }
1262
1263     if( h->param.i_nal_hrd == X264_NAL_HRD_CBR &&
1264        (h->param.rc.i_bitrate != h->param.rc.i_vbv_max_bitrate || !h->param.rc.i_vbv_max_bitrate) )
1265     {
1266         x264_log( h, X264_LOG_WARNING, "CBR HRD requires constant bitrate\n" );
1267         h->param.i_nal_hrd = X264_NAL_HRD_VBR;
1268     }
1269
1270     if( h->param.i_nal_hrd == X264_NAL_HRD_CBR )
1271         h->param.rc.b_filler = 1;
1272
1273     /* ensure the booleans are 0 or 1 so they can be used in math */
1274 #define BOOLIFY(x) h->param.x = !!h->param.x
1275     BOOLIFY( b_cabac );
1276     BOOLIFY( b_constrained_intra );
1277     BOOLIFY( b_deblocking_filter );
1278     BOOLIFY( b_deterministic );
1279     BOOLIFY( b_sliced_threads );
1280     BOOLIFY( b_interlaced );
1281     BOOLIFY( b_intra_refresh );
1282     BOOLIFY( b_aud );
1283     BOOLIFY( b_repeat_headers );
1284     BOOLIFY( b_annexb );
1285     BOOLIFY( b_vfr_input );
1286     BOOLIFY( b_pulldown );
1287     BOOLIFY( b_tff );
1288     BOOLIFY( b_pic_struct );
1289     BOOLIFY( b_fake_interlaced );
1290     BOOLIFY( b_open_gop );
1291     BOOLIFY( b_bluray_compat );
1292     BOOLIFY( b_stitchable );
1293     BOOLIFY( b_full_recon );
1294     BOOLIFY( b_opencl );
1295     BOOLIFY( analyse.b_transform_8x8 );
1296     BOOLIFY( analyse.b_weighted_bipred );
1297     BOOLIFY( analyse.b_chroma_me );
1298     BOOLIFY( analyse.b_mixed_references );
1299     BOOLIFY( analyse.b_fast_pskip );
1300     BOOLIFY( analyse.b_dct_decimate );
1301     BOOLIFY( analyse.b_psy );
1302     BOOLIFY( analyse.b_psnr );
1303     BOOLIFY( analyse.b_ssim );
1304     BOOLIFY( rc.b_stat_write );
1305     BOOLIFY( rc.b_stat_read );
1306     BOOLIFY( rc.b_mb_tree );
1307     BOOLIFY( rc.b_filler );
1308 #undef BOOLIFY
1309
1310     return 0;
1311 }
1312
1313 static void mbcmp_init( x264_t *h )
1314 {
1315     int satd = !h->mb.b_lossless && h->param.analyse.i_subpel_refine > 1;
1316     memcpy( h->pixf.mbcmp, satd ? h->pixf.satd : h->pixf.sad_aligned, sizeof(h->pixf.mbcmp) );
1317     memcpy( h->pixf.mbcmp_unaligned, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.mbcmp_unaligned) );
1318     h->pixf.intra_mbcmp_x3_16x16 = satd ? h->pixf.intra_satd_x3_16x16 : h->pixf.intra_sad_x3_16x16;
1319     h->pixf.intra_mbcmp_x3_8x16c = satd ? h->pixf.intra_satd_x3_8x16c : h->pixf.intra_sad_x3_8x16c;
1320     h->pixf.intra_mbcmp_x3_8x8c  = satd ? h->pixf.intra_satd_x3_8x8c  : h->pixf.intra_sad_x3_8x8c;
1321     h->pixf.intra_mbcmp_x3_8x8 = satd ? h->pixf.intra_sa8d_x3_8x8 : h->pixf.intra_sad_x3_8x8;
1322     h->pixf.intra_mbcmp_x3_4x4 = satd ? h->pixf.intra_satd_x3_4x4 : h->pixf.intra_sad_x3_4x4;
1323     h->pixf.intra_mbcmp_x9_4x4 = h->param.b_cpu_independent || h->mb.b_lossless ? NULL
1324                                : satd ? h->pixf.intra_satd_x9_4x4 : h->pixf.intra_sad_x9_4x4;
1325     h->pixf.intra_mbcmp_x9_8x8 = h->param.b_cpu_independent || h->mb.b_lossless ? NULL
1326                                : satd ? h->pixf.intra_sa8d_x9_8x8 : h->pixf.intra_sad_x9_8x8;
1327     satd &= h->param.analyse.i_me_method == X264_ME_TESA;
1328     memcpy( h->pixf.fpelcmp, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.fpelcmp) );
1329     memcpy( h->pixf.fpelcmp_x3, satd ? h->pixf.satd_x3 : h->pixf.sad_x3, sizeof(h->pixf.fpelcmp_x3) );
1330     memcpy( h->pixf.fpelcmp_x4, satd ? h->pixf.satd_x4 : h->pixf.sad_x4, sizeof(h->pixf.fpelcmp_x4) );
1331 }
1332
1333 static void chroma_dsp_init( x264_t *h )
1334 {
1335     memcpy( h->luma2chroma_pixel, x264_luma2chroma_pixel[CHROMA_FORMAT], sizeof(h->luma2chroma_pixel) );
1336
1337     switch( CHROMA_FORMAT )
1338     {
1339         case CHROMA_420:
1340             memcpy( h->predict_chroma, h->predict_8x8c, sizeof(h->predict_chroma) );
1341             h->mc.prefetch_fenc = h->mc.prefetch_fenc_420;
1342             h->loopf.deblock_chroma[0] = h->loopf.deblock_h_chroma_420;
1343             h->loopf.deblock_chroma_intra[0] = h->loopf.deblock_h_chroma_420_intra;
1344             h->loopf.deblock_chroma_mbaff = h->loopf.deblock_chroma_420_mbaff;
1345             h->loopf.deblock_chroma_intra_mbaff = h->loopf.deblock_chroma_420_intra_mbaff;
1346             h->pixf.intra_mbcmp_x3_chroma = h->pixf.intra_mbcmp_x3_8x8c;
1347             h->quantf.coeff_last[DCT_CHROMA_DC] = h->quantf.coeff_last4;
1348             h->quantf.coeff_level_run[DCT_CHROMA_DC] = h->quantf.coeff_level_run4;
1349             break;
1350         case CHROMA_422:
1351             memcpy( h->predict_chroma, h->predict_8x16c, sizeof(h->predict_chroma) );
1352             h->mc.prefetch_fenc = h->mc.prefetch_fenc_422;
1353             h->loopf.deblock_chroma[0] = h->loopf.deblock_h_chroma_422;
1354             h->loopf.deblock_chroma_intra[0] = h->loopf.deblock_h_chroma_422_intra;
1355             h->loopf.deblock_chroma_mbaff = h->loopf.deblock_chroma_422_mbaff;
1356             h->loopf.deblock_chroma_intra_mbaff = h->loopf.deblock_chroma_422_intra_mbaff;
1357             h->pixf.intra_mbcmp_x3_chroma = h->pixf.intra_mbcmp_x3_8x16c;
1358             h->quantf.coeff_last[DCT_CHROMA_DC] = h->quantf.coeff_last8;
1359             h->quantf.coeff_level_run[DCT_CHROMA_DC] = h->quantf.coeff_level_run8;
1360             break;
1361         case CHROMA_444:
1362             h->mc.prefetch_fenc = h->mc.prefetch_fenc_422; /* FIXME: doesn't cover V plane */
1363             h->loopf.deblock_chroma_mbaff = h->loopf.deblock_luma_mbaff;
1364             h->loopf.deblock_chroma_intra_mbaff = h->loopf.deblock_luma_intra_mbaff;
1365             break;
1366     }
1367 }
1368
1369 static void x264_set_aspect_ratio( x264_t *h, x264_param_t *param, int initial )
1370 {
1371     /* VUI */
1372     if( param->vui.i_sar_width > 0 && param->vui.i_sar_height > 0 )
1373     {
1374         uint32_t i_w = param->vui.i_sar_width;
1375         uint32_t i_h = param->vui.i_sar_height;
1376         uint32_t old_w = h->param.vui.i_sar_width;
1377         uint32_t old_h = h->param.vui.i_sar_height;
1378
1379         x264_reduce_fraction( &i_w, &i_h );
1380
1381         while( i_w > 65535 || i_h > 65535 )
1382         {
1383             i_w /= 2;
1384             i_h /= 2;
1385         }
1386
1387         x264_reduce_fraction( &i_w, &i_h );
1388
1389         if( i_w != old_w || i_h != old_h || initial )
1390         {
1391             h->param.vui.i_sar_width = 0;
1392             h->param.vui.i_sar_height = 0;
1393             if( i_w == 0 || i_h == 0 )
1394                 x264_log( h, X264_LOG_WARNING, "cannot create valid sample aspect ratio\n" );
1395             else
1396             {
1397                 x264_log( h, initial?X264_LOG_INFO:X264_LOG_DEBUG, "using SAR=%d/%d\n", i_w, i_h );
1398                 h->param.vui.i_sar_width = i_w;
1399                 h->param.vui.i_sar_height = i_h;
1400             }
1401         }
1402     }
1403 }
1404
1405 /****************************************************************************
1406  * x264_encoder_open:
1407  ****************************************************************************/
1408 x264_t *x264_encoder_open( x264_param_t *param )
1409 {
1410     x264_t *h;
1411     char buf[1000], *p;
1412     int qp, i_slicetype_length;
1413
1414     CHECKED_MALLOCZERO( h, sizeof(x264_t) );
1415
1416     /* Create a copy of param */
1417     memcpy( &h->param, param, sizeof(x264_param_t) );
1418
1419     if( param->param_free )
1420         param->param_free( param );
1421
1422 #if HAVE_INTEL_DISPATCHER
1423     x264_intel_dispatcher_override();
1424 #endif
1425
1426     if( x264_threading_init() )
1427     {
1428         x264_log( h, X264_LOG_ERROR, "unable to initialize threading\n" );
1429         goto fail;
1430     }
1431
1432     if( x264_validate_parameters( h, 1 ) < 0 )
1433         goto fail;
1434
1435     if( h->param.psz_cqm_file )
1436         if( x264_cqm_parse_file( h, h->param.psz_cqm_file ) < 0 )
1437             goto fail;
1438
1439     if( h->param.rc.psz_stat_out )
1440         h->param.rc.psz_stat_out = strdup( h->param.rc.psz_stat_out );
1441     if( h->param.rc.psz_stat_in )
1442         h->param.rc.psz_stat_in = strdup( h->param.rc.psz_stat_in );
1443
1444     x264_reduce_fraction( &h->param.i_fps_num, &h->param.i_fps_den );
1445     x264_reduce_fraction( &h->param.i_timebase_num, &h->param.i_timebase_den );
1446
1447     /* Init x264_t */
1448     h->i_frame = -1;
1449     h->i_frame_num = 0;
1450
1451     if( h->param.i_avcintra_class )
1452         h->i_idr_pic_id = 5;
1453     else
1454         h->i_idr_pic_id = 0;
1455
1456     if( (uint64_t)h->param.i_timebase_den * 2 > UINT32_MAX )
1457     {
1458         x264_log( h, X264_LOG_ERROR, "Effective timebase denominator %u exceeds H.264 maximum\n", h->param.i_timebase_den );
1459         goto fail;
1460     }
1461
1462     x264_set_aspect_ratio( h, &h->param, 1 );
1463
1464     x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
1465     x264_pps_init( h->pps, h->param.i_sps_id, &h->param, h->sps );
1466
1467     x264_validate_levels( h, 1 );
1468
1469     h->chroma_qp_table = i_chroma_qp_table + 12 + h->pps->i_chroma_qp_index_offset;
1470
1471     if( x264_cqm_init( h ) < 0 )
1472         goto fail;
1473
1474     h->mb.i_mb_width = h->sps->i_mb_width;
1475     h->mb.i_mb_height = h->sps->i_mb_height;
1476     h->mb.i_mb_count = h->mb.i_mb_width * h->mb.i_mb_height;
1477
1478     h->mb.chroma_h_shift = CHROMA_FORMAT == CHROMA_420 || CHROMA_FORMAT == CHROMA_422;
1479     h->mb.chroma_v_shift = CHROMA_FORMAT == CHROMA_420;
1480
1481     /* Adaptive MBAFF and subme 0 are not supported as we require halving motion
1482      * vectors during prediction, resulting in hpel mvs.
1483      * The chosen solution is to make MBAFF non-adaptive in this case. */
1484     h->mb.b_adaptive_mbaff = PARAM_INTERLACED && h->param.analyse.i_subpel_refine;
1485
1486     /* Init frames. */
1487     if( h->param.i_bframe_adaptive == X264_B_ADAPT_TRELLIS && !h->param.rc.b_stat_read )
1488         h->frames.i_delay = X264_MAX(h->param.i_bframe,3)*4;
1489     else
1490         h->frames.i_delay = h->param.i_bframe;
1491     if( h->param.rc.b_mb_tree || h->param.rc.i_vbv_buffer_size )
1492         h->frames.i_delay = X264_MAX( h->frames.i_delay, h->param.rc.i_lookahead );
1493     i_slicetype_length = h->frames.i_delay;
1494     h->frames.i_delay += h->i_thread_frames - 1;
1495     h->frames.i_delay += h->param.i_sync_lookahead;
1496     h->frames.i_delay += h->param.b_vfr_input;
1497     h->frames.i_bframe_delay = h->param.i_bframe ? (h->param.i_bframe_pyramid ? 2 : 1) : 0;
1498
1499     h->frames.i_max_ref0 = h->param.i_frame_reference;
1500     h->frames.i_max_ref1 = X264_MIN( h->sps->vui.i_num_reorder_frames, h->param.i_frame_reference );
1501     h->frames.i_max_dpb  = h->sps->vui.i_max_dec_frame_buffering;
1502     h->frames.b_have_lowres = !h->param.rc.b_stat_read
1503         && ( h->param.rc.i_rc_method == X264_RC_ABR
1504           || h->param.rc.i_rc_method == X264_RC_CRF
1505           || h->param.i_bframe_adaptive
1506           || h->param.i_scenecut_threshold
1507           || h->param.rc.b_mb_tree
1508           || h->param.analyse.i_weighted_pred );
1509     h->frames.b_have_lowres |= h->param.rc.b_stat_read && h->param.rc.i_vbv_buffer_size > 0;
1510     h->frames.b_have_sub8x8_esa = !!(h->param.analyse.inter & X264_ANALYSE_PSUB8x8);
1511
1512     h->frames.i_last_idr =
1513     h->frames.i_last_keyframe = - h->param.i_keyint_max;
1514     h->frames.i_input    = 0;
1515     h->frames.i_largest_pts = h->frames.i_second_largest_pts = -1;
1516     h->frames.i_poc_last_open_gop = -1;
1517
1518     CHECKED_MALLOCZERO( h->frames.unused[0], (h->frames.i_delay + 3) * sizeof(x264_frame_t *) );
1519     /* Allocate room for max refs plus a few extra just in case. */
1520     CHECKED_MALLOCZERO( h->frames.unused[1], (h->i_thread_frames + X264_REF_MAX + 4) * sizeof(x264_frame_t *) );
1521     CHECKED_MALLOCZERO( h->frames.current, (h->param.i_sync_lookahead + h->param.i_bframe
1522                         + h->i_thread_frames + 3) * sizeof(x264_frame_t *) );
1523     if( h->param.analyse.i_weighted_pred > 0 )
1524         CHECKED_MALLOCZERO( h->frames.blank_unused, h->i_thread_frames * 4 * sizeof(x264_frame_t *) );
1525     h->i_ref[0] = h->i_ref[1] = 0;
1526     h->i_cpb_delay = h->i_coded_fields = h->i_disp_fields = 0;
1527     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);
1528     h->i_disp_fields_last_frame = -1;
1529     x264_rdo_init();
1530
1531     /* init CPU functions */
1532     x264_predict_16x16_init( h->param.cpu, h->predict_16x16 );
1533     x264_predict_8x8c_init( h->param.cpu, h->predict_8x8c );
1534     x264_predict_8x16c_init( h->param.cpu, h->predict_8x16c );
1535     x264_predict_8x8_init( h->param.cpu, h->predict_8x8, &h->predict_8x8_filter );
1536     x264_predict_4x4_init( h->param.cpu, h->predict_4x4 );
1537     x264_pixel_init( h->param.cpu, &h->pixf );
1538     x264_dct_init( h->param.cpu, &h->dctf );
1539     x264_zigzag_init( h->param.cpu, &h->zigzagf_progressive, &h->zigzagf_interlaced );
1540     memcpy( &h->zigzagf, PARAM_INTERLACED ? &h->zigzagf_interlaced : &h->zigzagf_progressive, sizeof(h->zigzagf) );
1541     x264_mc_init( h->param.cpu, &h->mc, h->param.b_cpu_independent );
1542     x264_quant_init( h, h->param.cpu, &h->quantf );
1543     x264_deblock_init( h->param.cpu, &h->loopf, PARAM_INTERLACED );
1544     x264_bitstream_init( h->param.cpu, &h->bsf );
1545     if( h->param.b_cabac )
1546         x264_cabac_init( h );
1547     else
1548         x264_stack_align( x264_cavlc_init, h );
1549
1550     mbcmp_init( h );
1551     chroma_dsp_init( h );
1552
1553     p = buf + sprintf( buf, "using cpu capabilities:" );
1554     for( int i = 0; x264_cpu_names[i].flags; i++ )
1555     {
1556         if( !strcmp(x264_cpu_names[i].name, "SSE")
1557             && h->param.cpu & (X264_CPU_SSE2) )
1558             continue;
1559         if( !strcmp(x264_cpu_names[i].name, "SSE2")
1560             && h->param.cpu & (X264_CPU_SSE2_IS_FAST|X264_CPU_SSE2_IS_SLOW) )
1561             continue;
1562         if( !strcmp(x264_cpu_names[i].name, "SSE3")
1563             && (h->param.cpu & X264_CPU_SSSE3 || !(h->param.cpu & X264_CPU_CACHELINE_64)) )
1564             continue;
1565         if( !strcmp(x264_cpu_names[i].name, "SSE4.1")
1566             && (h->param.cpu & X264_CPU_SSE42) )
1567             continue;
1568         if( !strcmp(x264_cpu_names[i].name, "BMI1")
1569             && (h->param.cpu & X264_CPU_BMI2) )
1570             continue;
1571         if( (h->param.cpu & x264_cpu_names[i].flags) == x264_cpu_names[i].flags
1572             && (!i || x264_cpu_names[i].flags != x264_cpu_names[i-1].flags) )
1573             p += sprintf( p, " %s", x264_cpu_names[i].name );
1574     }
1575     if( !h->param.cpu )
1576         p += sprintf( p, " none!" );
1577     x264_log( h, X264_LOG_INFO, "%s\n", buf );
1578
1579     float *logs = x264_analyse_prepare_costs( h );
1580     if( !logs )
1581         goto fail;
1582     for( qp = X264_MIN( h->param.rc.i_qp_min, QP_MAX_SPEC ); qp <= h->param.rc.i_qp_max; qp++ )
1583         if( x264_analyse_init_costs( h, logs, qp ) )
1584             goto fail;
1585     if( x264_analyse_init_costs( h, logs, X264_LOOKAHEAD_QP ) )
1586         goto fail;
1587     x264_free( logs );
1588
1589     static const uint16_t cost_mv_correct[7] = { 24, 47, 95, 189, 379, 757, 1515 };
1590     /* Checks for known miscompilation issues. */
1591     if( h->cost_mv[X264_LOOKAHEAD_QP][2013] != cost_mv_correct[BIT_DEPTH-8] )
1592     {
1593         x264_log( h, X264_LOG_ERROR, "MV cost test failed: x264 has been miscompiled!\n" );
1594         goto fail;
1595     }
1596
1597     /* Must be volatile or else GCC will optimize it out. */
1598     volatile int temp = 392;
1599     if( x264_clz( temp ) != 23 )
1600     {
1601         x264_log( h, X264_LOG_ERROR, "CLZ test failed: x264 has been miscompiled!\n" );
1602 #if ARCH_X86 || ARCH_X86_64
1603         x264_log( h, X264_LOG_ERROR, "Are you attempting to run an SSE4a/LZCNT-targeted build on a CPU that\n" );
1604         x264_log( h, X264_LOG_ERROR, "doesn't support it?\n" );
1605 #endif
1606         goto fail;
1607     }
1608
1609     h->out.i_nal = 0;
1610     h->out.i_bitstream = X264_MAX( 1000000, h->param.i_width * h->param.i_height * 4
1611         * ( h->param.rc.i_rc_method == X264_RC_ABR ? pow( 0.95, h->param.rc.i_qp_min )
1612           : pow( 0.95, h->param.rc.i_qp_constant ) * X264_MAX( 1, h->param.rc.f_ip_factor )));
1613
1614     h->nal_buffer_size = h->out.i_bitstream * 3/2 + 4 + 64; /* +4 for startcode, +64 for nal_escape assembly padding */
1615     CHECKED_MALLOC( h->nal_buffer, h->nal_buffer_size );
1616
1617     CHECKED_MALLOC( h->reconfig_h, sizeof(x264_t) );
1618
1619     if( h->param.i_threads > 1 &&
1620         x264_threadpool_init( &h->threadpool, h->param.i_threads, (void*)x264_encoder_thread_init, h ) )
1621         goto fail;
1622     if( h->param.i_lookahead_threads > 1 &&
1623         x264_threadpool_init( &h->lookaheadpool, h->param.i_lookahead_threads, NULL, NULL ) )
1624         goto fail;
1625
1626 #if HAVE_OPENCL
1627     if( h->param.b_opencl )
1628     {
1629         h->opencl.ocl = x264_opencl_load_library();
1630         if( !h->opencl.ocl )
1631         {
1632             x264_log( h, X264_LOG_WARNING, "failed to load OpenCL\n" );
1633             h->param.b_opencl = 0;
1634         }
1635     }
1636 #endif
1637
1638     h->thread[0] = h;
1639     for( int i = 1; i < h->param.i_threads + !!h->param.i_sync_lookahead; i++ )
1640         CHECKED_MALLOC( h->thread[i], sizeof(x264_t) );
1641     if( h->param.i_lookahead_threads > 1 )
1642         for( int i = 0; i < h->param.i_lookahead_threads; i++ )
1643         {
1644             CHECKED_MALLOC( h->lookahead_thread[i], sizeof(x264_t) );
1645             *h->lookahead_thread[i] = *h;
1646         }
1647     *h->reconfig_h = *h;
1648
1649     for( int i = 0; i < h->param.i_threads; i++ )
1650     {
1651         int init_nal_count = h->param.i_slice_count + 3;
1652         int allocate_threadlocal_data = !h->param.b_sliced_threads || !i;
1653         if( i > 0 )
1654             *h->thread[i] = *h;
1655
1656         if( x264_pthread_mutex_init( &h->thread[i]->mutex, NULL ) )
1657             goto fail;
1658         if( x264_pthread_cond_init( &h->thread[i]->cv, NULL ) )
1659             goto fail;
1660
1661         if( allocate_threadlocal_data )
1662         {
1663             h->thread[i]->fdec = x264_frame_pop_unused( h, 1 );
1664             if( !h->thread[i]->fdec )
1665                 goto fail;
1666         }
1667         else
1668             h->thread[i]->fdec = h->thread[0]->fdec;
1669
1670         CHECKED_MALLOC( h->thread[i]->out.p_bitstream, h->out.i_bitstream );
1671         /* Start each thread with room for init_nal_count NAL units; it'll realloc later if needed. */
1672         CHECKED_MALLOC( h->thread[i]->out.nal, init_nal_count*sizeof(x264_nal_t) );
1673         h->thread[i]->out.i_nals_allocated = init_nal_count;
1674
1675         if( allocate_threadlocal_data && x264_macroblock_cache_allocate( h->thread[i] ) < 0 )
1676             goto fail;
1677     }
1678
1679 #if HAVE_OPENCL
1680     if( h->param.b_opencl && x264_opencl_lookahead_init( h ) < 0 )
1681         h->param.b_opencl = 0;
1682 #endif
1683
1684     if( x264_lookahead_init( h, i_slicetype_length ) )
1685         goto fail;
1686
1687     for( int i = 0; i < h->param.i_threads; i++ )
1688         if( x264_macroblock_thread_allocate( h->thread[i], 0 ) < 0 )
1689             goto fail;
1690
1691     if( x264_ratecontrol_new( h ) < 0 )
1692         goto fail;
1693
1694     if( h->param.i_nal_hrd )
1695     {
1696         x264_log( h, X264_LOG_DEBUG, "HRD bitrate: %i bits/sec\n", h->sps->vui.hrd.i_bit_rate_unscaled );
1697         x264_log( h, X264_LOG_DEBUG, "CPB size: %i bits\n", h->sps->vui.hrd.i_cpb_size_unscaled );
1698     }
1699
1700     if( h->param.psz_dump_yuv )
1701     {
1702         /* create or truncate the reconstructed video file */
1703         FILE *f = x264_fopen( h->param.psz_dump_yuv, "w" );
1704         if( !f )
1705         {
1706             x264_log( h, X264_LOG_ERROR, "dump_yuv: can't write to %s\n", h->param.psz_dump_yuv );
1707             goto fail;
1708         }
1709         else if( !x264_is_regular_file( f ) )
1710         {
1711             x264_log( h, X264_LOG_ERROR, "dump_yuv: incompatible with non-regular file %s\n", h->param.psz_dump_yuv );
1712             fclose( f );
1713             goto fail;
1714         }
1715         fclose( f );
1716     }
1717
1718     const char *profile = h->sps->i_profile_idc == PROFILE_BASELINE ? "Constrained Baseline" :
1719                           h->sps->i_profile_idc == PROFILE_MAIN ? "Main" :
1720                           h->sps->i_profile_idc == PROFILE_HIGH ? "High" :
1721                           h->sps->i_profile_idc == PROFILE_HIGH10 ? (h->sps->b_constraint_set3 == 1 ? "High 10 Intra" : "High 10") :
1722                           h->sps->i_profile_idc == PROFILE_HIGH422 ? (h->sps->b_constraint_set3 == 1 ? "High 4:2:2 Intra" : "High 4:2:2") :
1723                           h->sps->b_constraint_set3 == 1 ? "High 4:4:4 Intra" : "High 4:4:4 Predictive";
1724     char level[4];
1725     snprintf( level, sizeof(level), "%d.%d", h->sps->i_level_idc/10, h->sps->i_level_idc%10 );
1726     if( h->sps->i_level_idc == 9 || ( h->sps->i_level_idc == 11 && h->sps->b_constraint_set3 &&
1727         (h->sps->i_profile_idc == PROFILE_BASELINE || h->sps->i_profile_idc == PROFILE_MAIN) ) )
1728         strcpy( level, "1b" );
1729
1730     if( h->sps->i_profile_idc < PROFILE_HIGH10 )
1731     {
1732         x264_log( h, X264_LOG_INFO, "profile %s, level %s\n",
1733             profile, level );
1734     }
1735     else
1736     {
1737         static const char * const subsampling[4] = { "4:0:0", "4:2:0", "4:2:2", "4:4:4" };
1738         x264_log( h, X264_LOG_INFO, "profile %s, level %s, %s %d-bit\n",
1739             profile, level, subsampling[CHROMA_FORMAT], BIT_DEPTH );
1740     }
1741
1742     return h;
1743 fail:
1744     x264_free( h );
1745     return NULL;
1746 }
1747
1748 /****************************************************************************/
1749 static int x264_encoder_try_reconfig( x264_t *h, x264_param_t *param, int *rc_reconfig )
1750 {
1751     *rc_reconfig = 0;
1752     x264_set_aspect_ratio( h, param, 0 );
1753 #define COPY(var) h->param.var = param->var
1754     COPY( i_frame_reference ); // but never uses more refs than initially specified
1755     COPY( i_bframe_bias );
1756     if( h->param.i_scenecut_threshold )
1757         COPY( i_scenecut_threshold ); // can't turn it on or off, only vary the threshold
1758     COPY( b_deblocking_filter );
1759     COPY( i_deblocking_filter_alphac0 );
1760     COPY( i_deblocking_filter_beta );
1761     COPY( i_frame_packing );
1762     COPY( analyse.inter );
1763     COPY( analyse.intra );
1764     COPY( analyse.i_direct_mv_pred );
1765     /* Scratch buffer prevents me_range from being increased for esa/tesa */
1766     if( h->param.analyse.i_me_method < X264_ME_ESA || param->analyse.i_me_range < h->param.analyse.i_me_range )
1767         COPY( analyse.i_me_range );
1768     COPY( analyse.i_noise_reduction );
1769     /* We can't switch out of subme=0 during encoding. */
1770     if( h->param.analyse.i_subpel_refine )
1771         COPY( analyse.i_subpel_refine );
1772     COPY( analyse.i_trellis );
1773     COPY( analyse.b_chroma_me );
1774     COPY( analyse.b_dct_decimate );
1775     COPY( analyse.b_fast_pskip );
1776     COPY( analyse.b_mixed_references );
1777     COPY( analyse.f_psy_rd );
1778     COPY( analyse.f_psy_trellis );
1779     COPY( crop_rect );
1780     // can only twiddle these if they were enabled to begin with:
1781     if( h->param.analyse.i_me_method >= X264_ME_ESA || param->analyse.i_me_method < X264_ME_ESA )
1782         COPY( analyse.i_me_method );
1783     if( h->param.analyse.i_me_method >= X264_ME_ESA && !h->frames.b_have_sub8x8_esa )
1784         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
1785     if( h->pps->b_transform_8x8_mode )
1786         COPY( analyse.b_transform_8x8 );
1787     if( h->frames.i_max_ref1 > 1 )
1788         COPY( i_bframe_pyramid );
1789     COPY( i_slice_max_size );
1790     COPY( i_slice_max_mbs );
1791     COPY( i_slice_min_mbs );
1792     COPY( i_slice_count );
1793     COPY( i_slice_count_max );
1794     COPY( b_tff );
1795
1796     /* VBV can't be turned on if it wasn't on to begin with */
1797     if( h->param.rc.i_vbv_max_bitrate > 0 && h->param.rc.i_vbv_buffer_size > 0 &&
1798           param->rc.i_vbv_max_bitrate > 0 &&   param->rc.i_vbv_buffer_size > 0 )
1799     {
1800         *rc_reconfig |= h->param.rc.i_vbv_max_bitrate != param->rc.i_vbv_max_bitrate;
1801         *rc_reconfig |= h->param.rc.i_vbv_buffer_size != param->rc.i_vbv_buffer_size;
1802         *rc_reconfig |= h->param.rc.i_bitrate != param->rc.i_bitrate;
1803         COPY( rc.i_vbv_max_bitrate );
1804         COPY( rc.i_vbv_buffer_size );
1805         COPY( rc.i_bitrate );
1806     }
1807     *rc_reconfig |= h->param.rc.f_rf_constant != param->rc.f_rf_constant;
1808     *rc_reconfig |= h->param.rc.f_rf_constant_max != param->rc.f_rf_constant_max;
1809     COPY( rc.f_rf_constant );
1810     COPY( rc.f_rf_constant_max );
1811 #undef COPY
1812
1813     return x264_validate_parameters( h, 0 );
1814 }
1815
1816 int x264_encoder_reconfig_apply( x264_t *h, x264_param_t *param )
1817 {
1818     int rc_reconfig;
1819     int ret = x264_encoder_try_reconfig( h, param, &rc_reconfig );
1820
1821     mbcmp_init( h );
1822     if( !ret )
1823         x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
1824
1825     /* Supported reconfiguration options (1-pass only):
1826      * vbv-maxrate
1827      * vbv-bufsize
1828      * crf
1829      * bitrate (CBR only) */
1830     if( !ret && rc_reconfig )
1831         x264_ratecontrol_init_reconfigurable( h, 0 );
1832
1833     return ret;
1834 }
1835
1836 /****************************************************************************
1837  * x264_encoder_reconfig:
1838  ****************************************************************************/
1839 int x264_encoder_reconfig( x264_t *h, x264_param_t *param )
1840 {
1841     h = h->thread[h->thread[0]->i_thread_phase];
1842     x264_param_t param_save = h->reconfig_h->param;
1843     h->reconfig_h->param = h->param;
1844
1845     int rc_reconfig;
1846     int ret = x264_encoder_try_reconfig( h->reconfig_h, param, &rc_reconfig );
1847     if( !ret )
1848         h->reconfig = 1;
1849     else
1850         h->reconfig_h->param = param_save;
1851
1852     return ret;
1853 }
1854
1855 /****************************************************************************
1856  * x264_encoder_parameters:
1857  ****************************************************************************/
1858 void x264_encoder_parameters( x264_t *h, x264_param_t *param )
1859 {
1860     memcpy( param, &h->thread[h->i_thread_phase]->param, sizeof(x264_param_t) );
1861 }
1862
1863 /* internal usage */
1864 static void x264_nal_start( x264_t *h, int i_type, int i_ref_idc )
1865 {
1866     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1867
1868     nal->i_ref_idc        = i_ref_idc;
1869     nal->i_type           = i_type;
1870     nal->b_long_startcode = 1;
1871
1872     nal->i_payload= 0;
1873     nal->p_payload= &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8];
1874     nal->i_padding= 0;
1875 }
1876
1877 /* if number of allocated nals is not enough, re-allocate a larger one. */
1878 static int x264_nal_check_buffer( x264_t *h )
1879 {
1880     if( h->out.i_nal >= h->out.i_nals_allocated )
1881     {
1882         x264_nal_t *new_out = x264_malloc( sizeof(x264_nal_t) * (h->out.i_nals_allocated*2) );
1883         if( !new_out )
1884             return -1;
1885         memcpy( new_out, h->out.nal, sizeof(x264_nal_t) * (h->out.i_nals_allocated) );
1886         x264_free( h->out.nal );
1887         h->out.nal = new_out;
1888         h->out.i_nals_allocated *= 2;
1889     }
1890     return 0;
1891 }
1892
1893 static int x264_nal_end( x264_t *h )
1894 {
1895     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1896     uint8_t *end = &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8];
1897     nal->i_payload = end - nal->p_payload;
1898     /* Assembly implementation of nal_escape reads past the end of the input.
1899      * While undefined padding wouldn't actually affect the output, it makes valgrind unhappy. */
1900     memset( end, 0xff, 64 );
1901     if( h->param.nalu_process )
1902         h->param.nalu_process( h, nal, h->fenc->opaque );
1903     h->out.i_nal++;
1904
1905     return x264_nal_check_buffer( h );
1906 }
1907
1908 static int x264_check_encapsulated_buffer( x264_t *h, x264_t *h0, int start,
1909                                            int previous_nal_size, int necessary_size )
1910 {
1911     if( h0->nal_buffer_size < necessary_size )
1912     {
1913         necessary_size *= 2;
1914         uint8_t *buf = x264_malloc( necessary_size );
1915         if( !buf )
1916             return -1;
1917         if( previous_nal_size )
1918             memcpy( buf, h0->nal_buffer, previous_nal_size );
1919
1920         intptr_t delta = buf - h0->nal_buffer;
1921         for( int i = 0; i < start; i++ )
1922             h->out.nal[i].p_payload += delta;
1923
1924         x264_free( h0->nal_buffer );
1925         h0->nal_buffer = buf;
1926         h0->nal_buffer_size = necessary_size;
1927     }
1928
1929     return 0;
1930 }
1931
1932 static int x264_encoder_encapsulate_nals( x264_t *h, int start )
1933 {
1934     x264_t *h0 = h->thread[0];
1935     int nal_size = 0, previous_nal_size = 0;
1936
1937     if( h->param.nalu_process )
1938     {
1939         for( int i = start; i < h->out.i_nal; i++ )
1940             nal_size += h->out.nal[i].i_payload;
1941         return nal_size;
1942     }
1943
1944     for( int i = 0; i < start; i++ )
1945         previous_nal_size += h->out.nal[i].i_payload;
1946
1947     for( int i = start; i < h->out.i_nal; i++ )
1948         nal_size += h->out.nal[i].i_payload;
1949
1950     /* Worst-case NAL unit escaping: reallocate the buffer if it's too small. */
1951     int necessary_size = previous_nal_size + nal_size * 3/2 + h->out.i_nal * 4 + 4 + 64;
1952     for( int i = start; i < h->out.i_nal; i++ )
1953         necessary_size += h->out.nal[i].i_padding;
1954     if( x264_check_encapsulated_buffer( h, h0, start, previous_nal_size, necessary_size ) )
1955         return -1;
1956
1957     uint8_t *nal_buffer = h0->nal_buffer + previous_nal_size;
1958
1959     for( int i = start; i < h->out.i_nal; i++ )
1960     {
1961         int old_payload_len = h->out.nal[i].i_payload;
1962         h->out.nal[i].b_long_startcode = !i || h->out.nal[i].i_type == NAL_SPS || h->out.nal[i].i_type == NAL_PPS ||
1963                                          h->param.i_avcintra_class;
1964         x264_nal_encode( h, nal_buffer, &h->out.nal[i] );
1965         nal_buffer += h->out.nal[i].i_payload;
1966         if( h->param.i_avcintra_class )
1967         {
1968             h->out.nal[i].i_padding -= h->out.nal[i].i_payload - (old_payload_len + NALU_OVERHEAD);
1969             if( h->out.nal[i].i_padding > 0 )
1970             {
1971                 memset( nal_buffer, 0, h->out.nal[i].i_padding );
1972                 nal_buffer += h->out.nal[i].i_padding;
1973                 h->out.nal[i].i_payload += h->out.nal[i].i_padding;
1974             }
1975             h->out.nal[i].i_padding = X264_MAX( h->out.nal[i].i_padding, 0 );
1976         }
1977     }
1978
1979     x264_emms();
1980
1981     return nal_buffer - (h0->nal_buffer + previous_nal_size);
1982 }
1983
1984 /****************************************************************************
1985  * x264_encoder_headers:
1986  ****************************************************************************/
1987 int x264_encoder_headers( x264_t *h, x264_nal_t **pp_nal, int *pi_nal )
1988 {
1989     int frame_size = 0;
1990     /* init bitstream context */
1991     h->out.i_nal = 0;
1992     bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
1993
1994     /* Write SEI, SPS and PPS. */
1995
1996     /* generate sequence parameters */
1997     x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
1998     x264_sps_write( &h->out.bs, h->sps );
1999     if( x264_nal_end( h ) )
2000         return -1;
2001
2002     /* generate picture parameters */
2003     x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
2004     x264_pps_write( &h->out.bs, h->sps, h->pps );
2005     if( x264_nal_end( h ) )
2006         return -1;
2007
2008     /* identify ourselves */
2009     x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2010     if( x264_sei_version_write( h, &h->out.bs ) )
2011         return -1;
2012     if( x264_nal_end( h ) )
2013         return -1;
2014
2015     frame_size = x264_encoder_encapsulate_nals( h, 0 );
2016     if( frame_size < 0 )
2017         return -1;
2018
2019     /* now set output*/
2020     *pi_nal = h->out.i_nal;
2021     *pp_nal = &h->out.nal[0];
2022     h->out.i_nal = 0;
2023
2024     return frame_size;
2025 }
2026
2027 /* Check to see whether we have chosen a reference list ordering different
2028  * from the standard's default. */
2029 static inline void x264_reference_check_reorder( x264_t *h )
2030 {
2031     /* The reorder check doesn't check for missing frames, so just
2032      * force a reorder if one of the reference list is corrupt. */
2033     for( int i = 0; h->frames.reference[i]; i++ )
2034         if( h->frames.reference[i]->b_corrupt )
2035         {
2036             h->b_ref_reorder[0] = 1;
2037             return;
2038         }
2039     for( int list = 0; list <= (h->sh.i_type == SLICE_TYPE_B); list++ )
2040         for( int i = 0; i < h->i_ref[list] - 1; i++ )
2041         {
2042             int framenum_diff = h->fref[list][i+1]->i_frame_num - h->fref[list][i]->i_frame_num;
2043             int poc_diff = h->fref[list][i+1]->i_poc - h->fref[list][i]->i_poc;
2044             /* P and B-frames use different default orders. */
2045             if( h->sh.i_type == SLICE_TYPE_P ? framenum_diff > 0 : list == 1 ? poc_diff < 0 : poc_diff > 0 )
2046             {
2047                 h->b_ref_reorder[list] = 1;
2048                 return;
2049             }
2050         }
2051 }
2052
2053 /* return -1 on failure, else return the index of the new reference frame */
2054 int x264_weighted_reference_duplicate( x264_t *h, int i_ref, const x264_weight_t *w )
2055 {
2056     int i = h->i_ref[0];
2057     int j = 1;
2058     x264_frame_t *newframe;
2059     if( i <= 1 ) /* empty list, definitely can't duplicate frame */
2060         return -1;
2061
2062     //Duplication is only used in X264_WEIGHTP_SMART
2063     if( h->param.analyse.i_weighted_pred != X264_WEIGHTP_SMART )
2064         return -1;
2065
2066     /* Duplication is a hack to compensate for crappy rounding in motion compensation.
2067      * With high bit depth, it's not worth doing, so turn it off except in the case of
2068      * unweighted dupes. */
2069     if( BIT_DEPTH > 8 && w != x264_weight_none )
2070         return -1;
2071
2072     newframe = x264_frame_pop_blank_unused( h );
2073     if( !newframe )
2074         return -1;
2075
2076     //FIXME: probably don't need to copy everything
2077     *newframe = *h->fref[0][i_ref];
2078     newframe->i_reference_count = 1;
2079     newframe->orig = h->fref[0][i_ref];
2080     newframe->b_duplicate = 1;
2081     memcpy( h->fenc->weight[j], w, sizeof(h->fenc->weight[i]) );
2082
2083     /* shift the frames to make space for the dupe. */
2084     h->b_ref_reorder[0] = 1;
2085     if( h->i_ref[0] < X264_REF_MAX )
2086         ++h->i_ref[0];
2087     h->fref[0][X264_REF_MAX-1] = NULL;
2088     x264_frame_unshift( &h->fref[0][j], newframe );
2089
2090     return j;
2091 }
2092
2093 static void x264_weighted_pred_init( x264_t *h )
2094 {
2095     /* for now no analysis and set all weights to nothing */
2096     for( int i_ref = 0; i_ref < h->i_ref[0]; i_ref++ )
2097         h->fenc->weighted[i_ref] = h->fref[0][i_ref]->filtered[0][0];
2098
2099     // FIXME: This only supports weighting of one reference frame
2100     // and duplicates of that frame.
2101     h->fenc->i_lines_weighted = 0;
2102
2103     for( int i_ref = 0; i_ref < (h->i_ref[0] << SLICE_MBAFF); i_ref++ )
2104         for( int i = 0; i < 3; i++ )
2105             h->sh.weight[i_ref][i].weightfn = NULL;
2106
2107
2108     if( h->sh.i_type != SLICE_TYPE_P || h->param.analyse.i_weighted_pred <= 0 )
2109         return;
2110
2111     int i_padv = PADV << PARAM_INTERLACED;
2112     int denom = -1;
2113     int weightplane[2] = { 0, 0 };
2114     int buffer_next = 0;
2115     for( int i = 0; i < 3; i++ )
2116     {
2117         for( int j = 0; j < h->i_ref[0]; j++ )
2118         {
2119             if( h->fenc->weight[j][i].weightfn )
2120             {
2121                 h->sh.weight[j][i] = h->fenc->weight[j][i];
2122                 // if weight is useless, don't write it to stream
2123                 if( h->sh.weight[j][i].i_scale == 1<<h->sh.weight[j][i].i_denom && h->sh.weight[j][i].i_offset == 0 )
2124                     h->sh.weight[j][i].weightfn = NULL;
2125                 else
2126                 {
2127                     if( !weightplane[!!i] )
2128                     {
2129                         weightplane[!!i] = 1;
2130                         h->sh.weight[0][!!i].i_denom = denom = h->sh.weight[j][i].i_denom;
2131                         assert( x264_clip3( denom, 0, 7 ) == denom );
2132                     }
2133
2134                     assert( h->sh.weight[j][i].i_denom == denom );
2135                     if( !i )
2136                     {
2137                         h->fenc->weighted[j] = h->mb.p_weight_buf[buffer_next++] + h->fenc->i_stride[0] * i_padv + PADH;
2138                         //scale full resolution frame
2139                         if( h->param.i_threads == 1 )
2140                         {
2141                             pixel *src = h->fref[0][j]->filtered[0][0] - h->fref[0][j]->i_stride[0]*i_padv - PADH;
2142                             pixel *dst = h->fenc->weighted[j] - h->fenc->i_stride[0]*i_padv - PADH;
2143                             int stride = h->fenc->i_stride[0];
2144                             int width = h->fenc->i_width[0] + PADH*2;
2145                             int height = h->fenc->i_lines[0] + i_padv*2;
2146                             x264_weight_scale_plane( h, dst, stride, src, stride, width, height, &h->sh.weight[j][0] );
2147                             h->fenc->i_lines_weighted = height;
2148                         }
2149                     }
2150                 }
2151             }
2152         }
2153     }
2154
2155     if( weightplane[1] )
2156         for( int i = 0; i < h->i_ref[0]; i++ )
2157         {
2158             if( h->sh.weight[i][1].weightfn && !h->sh.weight[i][2].weightfn )
2159             {
2160                 h->sh.weight[i][2].i_scale = 1 << h->sh.weight[0][1].i_denom;
2161                 h->sh.weight[i][2].i_offset = 0;
2162             }
2163             else if( h->sh.weight[i][2].weightfn && !h->sh.weight[i][1].weightfn )
2164             {
2165                 h->sh.weight[i][1].i_scale = 1 << h->sh.weight[0][1].i_denom;
2166                 h->sh.weight[i][1].i_offset = 0;
2167             }
2168         }
2169
2170     if( !weightplane[0] )
2171         h->sh.weight[0][0].i_denom = 0;
2172     if( !weightplane[1] )
2173         h->sh.weight[0][1].i_denom = 0;
2174     h->sh.weight[0][2].i_denom = h->sh.weight[0][1].i_denom;
2175 }
2176
2177 static inline int x264_reference_distance( x264_t *h, x264_frame_t *frame )
2178 {
2179     if( h->param.i_frame_packing == 5 )
2180         return abs((h->fenc->i_frame&~1) - (frame->i_frame&~1)) +
2181                   ((h->fenc->i_frame&1) != (frame->i_frame&1));
2182     else
2183         return abs(h->fenc->i_frame - frame->i_frame);
2184 }
2185
2186 static inline void x264_reference_build_list( x264_t *h, int i_poc )
2187 {
2188     int b_ok;
2189
2190     /* build ref list 0/1 */
2191     h->mb.pic.i_fref[0] = h->i_ref[0] = 0;
2192     h->mb.pic.i_fref[1] = h->i_ref[1] = 0;
2193     if( h->sh.i_type == SLICE_TYPE_I )
2194         return;
2195
2196     for( int i = 0; h->frames.reference[i]; i++ )
2197     {
2198         if( h->frames.reference[i]->b_corrupt )
2199             continue;
2200         if( h->frames.reference[i]->i_poc < i_poc )
2201             h->fref[0][h->i_ref[0]++] = h->frames.reference[i];
2202         else if( h->frames.reference[i]->i_poc > i_poc )
2203             h->fref[1][h->i_ref[1]++] = h->frames.reference[i];
2204     }
2205
2206     if( h->sh.i_mmco_remove_from_end )
2207     {
2208         /* Order ref0 for MMCO remove */
2209         do
2210         {
2211             b_ok = 1;
2212             for( int i = 0; i < h->i_ref[0] - 1; i++ )
2213             {
2214                 if( h->fref[0][i]->i_frame < h->fref[0][i+1]->i_frame )
2215                 {
2216                     XCHG( x264_frame_t*, h->fref[0][i], h->fref[0][i+1] );
2217                     b_ok = 0;
2218                     break;
2219                 }
2220             }
2221         } while( !b_ok );
2222
2223         for( int i = h->i_ref[0]-1; i >= h->i_ref[0] - h->sh.i_mmco_remove_from_end; i-- )
2224         {
2225             int diff = h->i_frame_num - h->fref[0][i]->i_frame_num;
2226             h->sh.mmco[h->sh.i_mmco_command_count].i_poc = h->fref[0][i]->i_poc;
2227             h->sh.mmco[h->sh.i_mmco_command_count++].i_difference_of_pic_nums = diff;
2228         }
2229     }
2230
2231     /* Order reference lists by distance from the current frame. */
2232     for( int list = 0; list < 2; list++ )
2233     {
2234         h->fref_nearest[list] = h->fref[list][0];
2235         do
2236         {
2237             b_ok = 1;
2238             for( int i = 0; i < h->i_ref[list] - 1; i++ )
2239             {
2240                 if( list ? h->fref[list][i+1]->i_poc < h->fref_nearest[list]->i_poc
2241                          : h->fref[list][i+1]->i_poc > h->fref_nearest[list]->i_poc )
2242                     h->fref_nearest[list] = h->fref[list][i+1];
2243                 if( x264_reference_distance( h, h->fref[list][i] ) > x264_reference_distance( h, h->fref[list][i+1] ) )
2244                 {
2245                     XCHG( x264_frame_t*, h->fref[list][i], h->fref[list][i+1] );
2246                     b_ok = 0;
2247                     break;
2248                 }
2249             }
2250         } while( !b_ok );
2251     }
2252
2253     x264_reference_check_reorder( h );
2254
2255     h->i_ref[1] = X264_MIN( h->i_ref[1], h->frames.i_max_ref1 );
2256     h->i_ref[0] = X264_MIN( h->i_ref[0], h->frames.i_max_ref0 );
2257     h->i_ref[0] = X264_MIN( h->i_ref[0], h->param.i_frame_reference ); // if reconfig() has lowered the limit
2258
2259     /* For Blu-ray compliance, don't reference frames outside of the minigop. */
2260     if( IS_X264_TYPE_B( h->fenc->i_type ) && h->param.b_bluray_compat )
2261         h->i_ref[0] = X264_MIN( h->i_ref[0], IS_X264_TYPE_B( h->fref[0][0]->i_type ) + 1 );
2262
2263     /* add duplicates */
2264     if( h->fenc->i_type == X264_TYPE_P )
2265     {
2266         int idx = -1;
2267         if( h->param.analyse.i_weighted_pred >= X264_WEIGHTP_SIMPLE )
2268         {
2269             x264_weight_t w[3];
2270             w[1].weightfn = w[2].weightfn = NULL;
2271             if( h->param.rc.b_stat_read )
2272                 x264_ratecontrol_set_weights( h, h->fenc );
2273
2274             if( !h->fenc->weight[0][0].weightfn )
2275             {
2276                 h->fenc->weight[0][0].i_denom = 0;
2277                 SET_WEIGHT( w[0], 1, 1, 0, -1 );
2278                 idx = x264_weighted_reference_duplicate( h, 0, w );
2279             }
2280             else
2281             {
2282                 if( h->fenc->weight[0][0].i_scale == 1<<h->fenc->weight[0][0].i_denom )
2283                 {
2284                     SET_WEIGHT( h->fenc->weight[0][0], 1, 1, 0, h->fenc->weight[0][0].i_offset );
2285                 }
2286                 x264_weighted_reference_duplicate( h, 0, x264_weight_none );
2287                 if( h->fenc->weight[0][0].i_offset > -128 )
2288                 {
2289                     w[0] = h->fenc->weight[0][0];
2290                     w[0].i_offset--;
2291                     h->mc.weight_cache( h, &w[0] );
2292                     idx = x264_weighted_reference_duplicate( h, 0, w );
2293                 }
2294             }
2295         }
2296         h->mb.ref_blind_dupe = idx;
2297     }
2298
2299     assert( h->i_ref[0] + h->i_ref[1] <= X264_REF_MAX );
2300     h->mb.pic.i_fref[0] = h->i_ref[0];
2301     h->mb.pic.i_fref[1] = h->i_ref[1];
2302 }
2303
2304 static void x264_fdec_filter_row( x264_t *h, int mb_y, int pass )
2305 {
2306     /* mb_y is the mb to be encoded next, not the mb to be filtered here */
2307     int b_hpel = h->fdec->b_kept_as_ref;
2308     int b_deblock = h->sh.i_disable_deblocking_filter_idc != 1;
2309     int b_end = mb_y == h->i_threadslice_end;
2310     int b_measure_quality = 1;
2311     int min_y = mb_y - (1 << SLICE_MBAFF);
2312     int b_start = min_y == h->i_threadslice_start;
2313     /* Even in interlaced mode, deblocking never modifies more than 4 pixels
2314      * above each MB, as bS=4 doesn't happen for the top of interlaced mbpairs. */
2315     int minpix_y = min_y*16 - 4 * !b_start;
2316     int maxpix_y = mb_y*16 - 4 * !b_end;
2317     b_deblock &= b_hpel || h->param.b_full_recon || h->param.psz_dump_yuv;
2318     if( h->param.b_sliced_threads )
2319     {
2320         switch( pass )
2321         {
2322             /* During encode: only do deblock if asked for */
2323             default:
2324             case 0:
2325                 b_deblock &= h->param.b_full_recon;
2326                 b_hpel = 0;
2327                 break;
2328             /* During post-encode pass: do deblock if not done yet, do hpel for all
2329              * rows except those between slices. */
2330             case 1:
2331                 b_deblock &= !h->param.b_full_recon;
2332                 b_hpel &= !(b_start && min_y > 0);
2333                 b_measure_quality = 0;
2334                 break;
2335             /* Final pass: do the rows between slices in sequence. */
2336             case 2:
2337                 b_deblock = 0;
2338                 b_measure_quality = 0;
2339                 break;
2340         }
2341     }
2342     if( mb_y & SLICE_MBAFF )
2343         return;
2344     if( min_y < h->i_threadslice_start )
2345         return;
2346
2347     if( b_deblock )
2348         for( int y = min_y; y < mb_y; y += (1 << SLICE_MBAFF) )
2349             x264_frame_deblock_row( h, y );
2350
2351     /* FIXME: Prediction requires different borders for interlaced/progressive mc,
2352      * but the actual image data is equivalent. For now, maintain this
2353      * consistency by copying deblocked pixels between planes. */
2354     if( PARAM_INTERLACED && (!h->param.b_sliced_threads || pass == 1) )
2355         for( int p = 0; p < h->fdec->i_plane; p++ )
2356             for( int i = minpix_y>>(CHROMA_V_SHIFT && p); i < maxpix_y>>(CHROMA_V_SHIFT && p); i++ )
2357                 memcpy( h->fdec->plane_fld[p] + i*h->fdec->i_stride[p],
2358                         h->fdec->plane[p]     + i*h->fdec->i_stride[p],
2359                         h->mb.i_mb_width*16*sizeof(pixel) );
2360
2361     if( h->fdec->b_kept_as_ref && (!h->param.b_sliced_threads || pass == 1) )
2362         x264_frame_expand_border( h, h->fdec, min_y );
2363     if( b_hpel )
2364     {
2365         int end = mb_y == h->mb.i_mb_height;
2366         /* Can't do hpel until the previous slice is done encoding. */
2367         if( h->param.analyse.i_subpel_refine )
2368         {
2369             x264_frame_filter( h, h->fdec, min_y, end );
2370             x264_frame_expand_border_filtered( h, h->fdec, min_y, end );
2371         }
2372     }
2373
2374     if( SLICE_MBAFF && pass == 0 )
2375         for( int i = 0; i < 3; i++ )
2376         {
2377             XCHG( pixel *, h->intra_border_backup[0][i], h->intra_border_backup[3][i] );
2378             XCHG( pixel *, h->intra_border_backup[1][i], h->intra_border_backup[4][i] );
2379         }
2380
2381     if( h->i_thread_frames > 1 && h->fdec->b_kept_as_ref )
2382         x264_frame_cond_broadcast( h->fdec, mb_y*16 + (b_end ? 10000 : -(X264_THREAD_HEIGHT << SLICE_MBAFF)) );
2383
2384     if( b_measure_quality )
2385     {
2386         maxpix_y = X264_MIN( maxpix_y, h->param.i_height );
2387         if( h->param.analyse.b_psnr )
2388         {
2389             for( int p = 0; p < (CHROMA444 ? 3 : 1); p++ )
2390                 h->stat.frame.i_ssd[p] += x264_pixel_ssd_wxh( &h->pixf,
2391                     h->fdec->plane[p] + minpix_y * h->fdec->i_stride[p], h->fdec->i_stride[p],
2392                     h->fenc->plane[p] + minpix_y * h->fenc->i_stride[p], h->fenc->i_stride[p],
2393                     h->param.i_width, maxpix_y-minpix_y );
2394             if( !CHROMA444 )
2395             {
2396                 uint64_t ssd_u, ssd_v;
2397                 int v_shift = CHROMA_V_SHIFT;
2398                 x264_pixel_ssd_nv12( &h->pixf,
2399                     h->fdec->plane[1] + (minpix_y>>v_shift) * h->fdec->i_stride[1], h->fdec->i_stride[1],
2400                     h->fenc->plane[1] + (minpix_y>>v_shift) * h->fenc->i_stride[1], h->fenc->i_stride[1],
2401                     h->param.i_width>>1, (maxpix_y-minpix_y)>>v_shift, &ssd_u, &ssd_v );
2402                 h->stat.frame.i_ssd[1] += ssd_u;
2403                 h->stat.frame.i_ssd[2] += ssd_v;
2404             }
2405         }
2406
2407         if( h->param.analyse.b_ssim )
2408         {
2409             int ssim_cnt;
2410             x264_emms();
2411             /* offset by 2 pixels to avoid alignment of ssim blocks with dct blocks,
2412              * and overlap by 4 */
2413             minpix_y += b_start ? 2 : -6;
2414             h->stat.frame.f_ssim +=
2415                 x264_pixel_ssim_wxh( &h->pixf,
2416                     h->fdec->plane[0] + 2+minpix_y*h->fdec->i_stride[0], h->fdec->i_stride[0],
2417                     h->fenc->plane[0] + 2+minpix_y*h->fenc->i_stride[0], h->fenc->i_stride[0],
2418                     h->param.i_width-2, maxpix_y-minpix_y, h->scratch_buffer, &ssim_cnt );
2419             h->stat.frame.i_ssim_cnt += ssim_cnt;
2420         }
2421     }
2422 }
2423
2424 static inline int x264_reference_update( x264_t *h )
2425 {
2426     if( !h->fdec->b_kept_as_ref )
2427     {
2428         if( h->i_thread_frames > 1 )
2429         {
2430             x264_frame_push_unused( h, h->fdec );
2431             h->fdec = x264_frame_pop_unused( h, 1 );
2432             if( !h->fdec )
2433                 return -1;
2434         }
2435         return 0;
2436     }
2437
2438     /* apply mmco from previous frame. */
2439     for( int i = 0; i < h->sh.i_mmco_command_count; i++ )
2440         for( int j = 0; h->frames.reference[j]; j++ )
2441             if( h->frames.reference[j]->i_poc == h->sh.mmco[i].i_poc )
2442                 x264_frame_push_unused( h, x264_frame_shift( &h->frames.reference[j] ) );
2443
2444     /* move frame in the buffer */
2445     x264_frame_push( h->frames.reference, h->fdec );
2446     if( h->frames.reference[h->sps->i_num_ref_frames] )
2447         x264_frame_push_unused( h, x264_frame_shift( h->frames.reference ) );
2448     h->fdec = x264_frame_pop_unused( h, 1 );
2449     if( !h->fdec )
2450         return -1;
2451     return 0;
2452 }
2453
2454 static inline void x264_reference_reset( x264_t *h )
2455 {
2456     while( h->frames.reference[0] )
2457         x264_frame_push_unused( h, x264_frame_pop( h->frames.reference ) );
2458     h->fdec->i_poc =
2459     h->fenc->i_poc = 0;
2460 }
2461
2462 static inline void x264_reference_hierarchy_reset( x264_t *h )
2463 {
2464     int ref;
2465     int b_hasdelayframe = 0;
2466
2467     /* look for delay frames -- chain must only contain frames that are disposable */
2468     for( int i = 0; h->frames.current[i] && IS_DISPOSABLE( h->frames.current[i]->i_type ); i++ )
2469         b_hasdelayframe |= h->frames.current[i]->i_coded
2470                         != h->frames.current[i]->i_frame + h->sps->vui.i_num_reorder_frames;
2471
2472     /* This function must handle b-pyramid and clear frames for open-gop */
2473     if( h->param.i_bframe_pyramid != X264_B_PYRAMID_STRICT && !b_hasdelayframe && h->frames.i_poc_last_open_gop == -1 )
2474         return;
2475
2476     /* Remove last BREF. There will never be old BREFs in the
2477      * dpb during a BREF decode when pyramid == STRICT */
2478     for( ref = 0; h->frames.reference[ref]; ref++ )
2479     {
2480         if( ( h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT
2481             && h->frames.reference[ref]->i_type == X264_TYPE_BREF )
2482             || ( h->frames.reference[ref]->i_poc < h->frames.i_poc_last_open_gop
2483             && h->sh.i_type != SLICE_TYPE_B ) )
2484         {
2485             int diff = h->i_frame_num - h->frames.reference[ref]->i_frame_num;
2486             h->sh.mmco[h->sh.i_mmco_command_count].i_difference_of_pic_nums = diff;
2487             h->sh.mmco[h->sh.i_mmco_command_count++].i_poc = h->frames.reference[ref]->i_poc;
2488             x264_frame_push_unused( h, x264_frame_shift( &h->frames.reference[ref] ) );
2489             h->b_ref_reorder[0] = 1;
2490             ref--;
2491         }
2492     }
2493
2494     /* Prepare room in the dpb for the delayed display time of the later b-frame's */
2495     if( h->param.i_bframe_pyramid )
2496         h->sh.i_mmco_remove_from_end = X264_MAX( ref + 2 - h->frames.i_max_dpb, 0 );
2497 }
2498
2499 static inline void x264_slice_init( x264_t *h, int i_nal_type, int i_global_qp )
2500 {
2501     /* ------------------------ Create slice header  ----------------------- */
2502     if( i_nal_type == NAL_SLICE_IDR )
2503     {
2504         x264_slice_header_init( h, &h->sh, h->sps, h->pps, h->i_idr_pic_id, h->i_frame_num, i_global_qp );
2505
2506         /* alternate id */
2507         if( h->param.i_avcintra_class )
2508         {
2509             switch( h->i_idr_pic_id )
2510             {
2511                 case 5:
2512                     h->i_idr_pic_id = 3;
2513                     break;
2514                 case 3:
2515                     h->i_idr_pic_id = 4;
2516                     break;
2517                 case 4:
2518                 default:
2519                     h->i_idr_pic_id = 5;
2520                     break;
2521             }
2522         }
2523         else
2524             h->i_idr_pic_id ^= 1;
2525     }
2526     else
2527     {
2528         x264_slice_header_init( h, &h->sh, h->sps, h->pps, -1, h->i_frame_num, i_global_qp );
2529
2530         h->sh.i_num_ref_idx_l0_active = h->i_ref[0] <= 0 ? 1 : h->i_ref[0];
2531         h->sh.i_num_ref_idx_l1_active = h->i_ref[1] <= 0 ? 1 : h->i_ref[1];
2532         if( h->sh.i_num_ref_idx_l0_active != h->pps->i_num_ref_idx_l0_default_active ||
2533             (h->sh.i_type == SLICE_TYPE_B && h->sh.i_num_ref_idx_l1_active != h->pps->i_num_ref_idx_l1_default_active) )
2534         {
2535             h->sh.b_num_ref_idx_override = 1;
2536         }
2537     }
2538
2539     if( h->fenc->i_type == X264_TYPE_BREF && h->param.b_bluray_compat && h->sh.i_mmco_command_count )
2540     {
2541         h->b_sh_backup = 1;
2542         h->sh_backup = h->sh;
2543     }
2544
2545     h->fdec->i_frame_num = h->sh.i_frame_num;
2546
2547     if( h->sps->i_poc_type == 0 )
2548     {
2549         h->sh.i_poc = h->fdec->i_poc;
2550         if( PARAM_INTERLACED )
2551         {
2552             h->sh.i_delta_poc_bottom = h->param.b_tff ? 1 : -1;
2553             h->sh.i_poc += h->sh.i_delta_poc_bottom == -1;
2554         }
2555         else
2556             h->sh.i_delta_poc_bottom = 0;
2557         h->fdec->i_delta_poc[0] = h->sh.i_delta_poc_bottom == -1;
2558         h->fdec->i_delta_poc[1] = h->sh.i_delta_poc_bottom ==  1;
2559     }
2560     else
2561     {
2562         /* Nothing to do ? */
2563     }
2564
2565     x264_macroblock_slice_init( h );
2566 }
2567
2568 typedef struct
2569 {
2570     int skip;
2571     uint8_t cabac_prevbyte;
2572     bs_t bs;
2573     x264_cabac_t cabac;
2574     x264_frame_stat_t stat;
2575     int last_qp;
2576     int last_dqp;
2577     int field_decoding_flag;
2578 } x264_bs_bak_t;
2579
2580 static ALWAYS_INLINE void x264_bitstream_backup( x264_t *h, x264_bs_bak_t *bak, int i_skip, int full )
2581 {
2582     if( full )
2583     {
2584         bak->stat = h->stat.frame;
2585         bak->last_qp = h->mb.i_last_qp;
2586         bak->last_dqp = h->mb.i_last_dqp;
2587         bak->field_decoding_flag = h->mb.field_decoding_flag;
2588     }
2589     else
2590     {
2591         bak->stat.i_mv_bits = h->stat.frame.i_mv_bits;
2592         bak->stat.i_tex_bits = h->stat.frame.i_tex_bits;
2593     }
2594     /* In the per-MB backup, we don't need the contexts because flushing the CABAC
2595      * encoder has no context dependency and in this case, a slice is ended (and
2596      * thus the content of all contexts are thrown away). */
2597     if( h->param.b_cabac )
2598     {
2599         if( full )
2600             memcpy( &bak->cabac, &h->cabac, sizeof(x264_cabac_t) );
2601         else
2602             memcpy( &bak->cabac, &h->cabac, offsetof(x264_cabac_t, f8_bits_encoded) );
2603         /* x264's CABAC writer modifies the previous byte during carry, so it has to be
2604          * backed up. */
2605         bak->cabac_prevbyte = h->cabac.p[-1];
2606     }
2607     else
2608     {
2609         bak->bs = h->out.bs;
2610         bak->skip = i_skip;
2611     }
2612 }
2613
2614 static ALWAYS_INLINE void x264_bitstream_restore( x264_t *h, x264_bs_bak_t *bak, int *skip, int full )
2615 {
2616     if( full )
2617     {
2618         h->stat.frame = bak->stat;
2619         h->mb.i_last_qp = bak->last_qp;
2620         h->mb.i_last_dqp = bak->last_dqp;
2621         h->mb.field_decoding_flag = bak->field_decoding_flag;
2622     }
2623     else
2624     {
2625         h->stat.frame.i_mv_bits = bak->stat.i_mv_bits;
2626         h->stat.frame.i_tex_bits = bak->stat.i_tex_bits;
2627     }
2628     if( h->param.b_cabac )
2629     {
2630         if( full )
2631             memcpy( &h->cabac, &bak->cabac, sizeof(x264_cabac_t) );
2632         else
2633             memcpy( &h->cabac, &bak->cabac, offsetof(x264_cabac_t, f8_bits_encoded) );
2634         h->cabac.p[-1] = bak->cabac_prevbyte;
2635     }
2636     else
2637     {
2638         h->out.bs = bak->bs;
2639         *skip = bak->skip;
2640     }
2641 }
2642
2643 static intptr_t x264_slice_write( x264_t *h )
2644 {
2645     int i_skip;
2646     int mb_xy, i_mb_x, i_mb_y;
2647     /* NALUs other than the first use a 3-byte startcode.
2648      * Add one extra byte for the rbsp, and one more for the final CABAC putbyte.
2649      * Then add an extra 5 bytes just in case, to account for random NAL escapes and
2650      * other inaccuracies. */
2651     int overhead_guess = (NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal)) + 1 + h->param.b_cabac + 5;
2652     int slice_max_size = h->param.i_slice_max_size > 0 ? (h->param.i_slice_max_size-overhead_guess)*8 : 0;
2653     int back_up_bitstream_cavlc = !h->param.b_cabac && h->sps->i_profile_idc < PROFILE_HIGH;
2654     int back_up_bitstream = slice_max_size || back_up_bitstream_cavlc;
2655     int starting_bits = bs_pos(&h->out.bs);
2656     int b_deblock = h->sh.i_disable_deblocking_filter_idc != 1;
2657     int b_hpel = h->fdec->b_kept_as_ref;
2658     int orig_last_mb = h->sh.i_last_mb;
2659     int thread_last_mb = h->i_threadslice_end * h->mb.i_mb_width - 1;
2660     uint8_t *last_emu_check;
2661 #define BS_BAK_SLICE_MAX_SIZE 0
2662 #define BS_BAK_CAVLC_OVERFLOW 1
2663 #define BS_BAK_SLICE_MIN_MBS  2
2664 #define BS_BAK_ROW_VBV        3
2665     x264_bs_bak_t bs_bak[4];
2666     b_deblock &= b_hpel || h->param.b_full_recon || h->param.psz_dump_yuv;
2667     bs_realign( &h->out.bs );
2668
2669     /* Slice */
2670     x264_nal_start( h, h->i_nal_type, h->i_nal_ref_idc );
2671     h->out.nal[h->out.i_nal].i_first_mb = h->sh.i_first_mb;
2672
2673     /* Slice header */
2674     x264_macroblock_thread_init( h );
2675
2676     /* Set the QP equal to the first QP in the slice for more accurate CABAC initialization. */
2677     h->mb.i_mb_xy = h->sh.i_first_mb;
2678     h->sh.i_qp = x264_ratecontrol_mb_qp( h );
2679     h->sh.i_qp = SPEC_QP( h->sh.i_qp );
2680     h->sh.i_qp_delta = h->sh.i_qp - h->pps->i_pic_init_qp;
2681
2682     x264_slice_header_write( &h->out.bs, &h->sh, h->i_nal_ref_idc );
2683     if( h->param.b_cabac )
2684     {
2685         /* alignment needed */
2686         bs_align_1( &h->out.bs );
2687
2688         /* init cabac */
2689         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 );
2690         x264_cabac_encode_init ( &h->cabac, h->out.bs.p, h->out.bs.p_end );
2691         last_emu_check = h->cabac.p;
2692     }
2693     else
2694         last_emu_check = h->out.bs.p;
2695     h->mb.i_last_qp = h->sh.i_qp;
2696     h->mb.i_last_dqp = 0;
2697     h->mb.field_decoding_flag = 0;
2698
2699     i_mb_y = h->sh.i_first_mb / h->mb.i_mb_width;
2700     i_mb_x = h->sh.i_first_mb % h->mb.i_mb_width;
2701     i_skip = 0;
2702
2703     while( 1 )
2704     {
2705         mb_xy = i_mb_x + i_mb_y * h->mb.i_mb_width;
2706         int mb_spos = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
2707
2708         if( i_mb_x == 0 )
2709         {
2710             if( x264_bitstream_check_buffer( h ) )
2711                 return -1;
2712             if( !(i_mb_y & SLICE_MBAFF) && h->param.rc.i_vbv_buffer_size )
2713                 x264_bitstream_backup( h, &bs_bak[BS_BAK_ROW_VBV], i_skip, 1 );
2714             if( !h->mb.b_reencode_mb )
2715                 x264_fdec_filter_row( h, i_mb_y, 0 );
2716         }
2717
2718         if( back_up_bitstream )
2719         {
2720             if( back_up_bitstream_cavlc )
2721                 x264_bitstream_backup( h, &bs_bak[BS_BAK_CAVLC_OVERFLOW], i_skip, 0 );
2722             if( slice_max_size && !(i_mb_y & SLICE_MBAFF) )
2723             {
2724                 x264_bitstream_backup( h, &bs_bak[BS_BAK_SLICE_MAX_SIZE], i_skip, 0 );
2725                 if( (thread_last_mb+1-mb_xy) == h->param.i_slice_min_mbs )
2726                     x264_bitstream_backup( h, &bs_bak[BS_BAK_SLICE_MIN_MBS], i_skip, 0 );
2727             }
2728         }
2729
2730         if( PARAM_INTERLACED )
2731         {
2732             if( h->mb.b_adaptive_mbaff )
2733             {
2734                 if( !(i_mb_y&1) )
2735                 {
2736                     /* FIXME: VSAD is fast but fairly poor at choosing the best interlace type. */
2737                     h->mb.b_interlaced = x264_field_vsad( h, i_mb_x, i_mb_y );
2738                     memcpy( &h->zigzagf, MB_INTERLACED ? &h->zigzagf_interlaced : &h->zigzagf_progressive, sizeof(h->zigzagf) );
2739                     if( !MB_INTERLACED && (i_mb_y+2) == h->mb.i_mb_height )
2740                         x264_expand_border_mbpair( h, i_mb_x, i_mb_y );
2741                 }
2742             }
2743             h->mb.field[mb_xy] = MB_INTERLACED;
2744         }
2745
2746         /* load cache */
2747         if( SLICE_MBAFF )
2748             x264_macroblock_cache_load_interlaced( h, i_mb_x, i_mb_y );
2749         else
2750             x264_macroblock_cache_load_progressive( h, i_mb_x, i_mb_y );
2751
2752         x264_macroblock_analyse( h );
2753
2754         /* encode this macroblock -> be careful it can change the mb type to P_SKIP if needed */
2755 reencode:
2756         x264_macroblock_encode( h );
2757
2758         if( h->param.b_cabac )
2759         {
2760             if( mb_xy > h->sh.i_first_mb && !(SLICE_MBAFF && (i_mb_y&1)) )
2761                 x264_cabac_encode_terminal( &h->cabac );
2762
2763             if( IS_SKIP( h->mb.i_type ) )
2764                 x264_cabac_mb_skip( h, 1 );
2765             else
2766             {
2767                 if( h->sh.i_type != SLICE_TYPE_I )
2768                     x264_cabac_mb_skip( h, 0 );
2769                 x264_macroblock_write_cabac( h, &h->cabac );
2770             }
2771         }
2772         else
2773         {
2774             if( IS_SKIP( h->mb.i_type ) )
2775                 i_skip++;
2776             else
2777             {
2778                 if( h->sh.i_type != SLICE_TYPE_I )
2779                 {
2780                     bs_write_ue( &h->out.bs, i_skip );  /* skip run */
2781                     i_skip = 0;
2782                 }
2783                 x264_macroblock_write_cavlc( h );
2784                 /* If there was a CAVLC level code overflow, try again at a higher QP. */
2785                 if( h->mb.b_overflow )
2786                 {
2787                     h->mb.i_chroma_qp = h->chroma_qp_table[++h->mb.i_qp];
2788                     h->mb.i_skip_intra = 0;
2789                     h->mb.b_skip_mc = 0;
2790                     h->mb.b_overflow = 0;
2791                     x264_bitstream_restore( h, &bs_bak[BS_BAK_CAVLC_OVERFLOW], &i_skip, 0 );
2792                     goto reencode;
2793                 }
2794             }
2795         }
2796
2797         int total_bits = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
2798         int mb_size = total_bits - mb_spos;
2799
2800         if( slice_max_size && (!SLICE_MBAFF || (i_mb_y&1)) )
2801         {
2802             /* Count the skip run, just in case. */
2803             if( !h->param.b_cabac )
2804                 total_bits += bs_size_ue_big( i_skip );
2805             /* Check for escape bytes. */
2806             uint8_t *end = h->param.b_cabac ? h->cabac.p : h->out.bs.p;
2807             for( ; last_emu_check < end - 2; last_emu_check++ )
2808                 if( last_emu_check[0] == 0 && last_emu_check[1] == 0 && last_emu_check[2] <= 3 )
2809                 {
2810                     slice_max_size -= 8;
2811                     last_emu_check++;
2812                 }
2813             /* We'll just re-encode this last macroblock if we go over the max slice size. */
2814             if( total_bits - starting_bits > slice_max_size && !h->mb.b_reencode_mb )
2815             {
2816                 if( !x264_frame_new_slice( h, h->fdec ) )
2817                 {
2818                     /* Handle the most obnoxious slice-min-mbs edge case: we need to end the slice
2819                      * because it's gone over the maximum size, but doing so would violate slice-min-mbs.
2820                      * If possible, roll back to the last checkpoint and try again.
2821                      * We could try raising QP, but that would break in the case where a slice spans multiple
2822                      * rows, which the re-encoding infrastructure can't currently handle. */
2823                     if( mb_xy <= thread_last_mb && (thread_last_mb+1-mb_xy) < h->param.i_slice_min_mbs )
2824                     {
2825                         if( thread_last_mb-h->param.i_slice_min_mbs < h->sh.i_first_mb+h->param.i_slice_min_mbs )
2826                         {
2827                             x264_log( h, X264_LOG_WARNING, "slice-max-size violated (frame %d, cause: slice-min-mbs)\n", h->i_frame );
2828                             slice_max_size = 0;
2829                             goto cont;
2830                         }
2831                         x264_bitstream_restore( h, &bs_bak[BS_BAK_SLICE_MIN_MBS], &i_skip, 0 );
2832                         h->mb.b_reencode_mb = 1;
2833                         h->sh.i_last_mb = thread_last_mb-h->param.i_slice_min_mbs;
2834                         break;
2835                     }
2836                     if( mb_xy-SLICE_MBAFF*h->mb.i_mb_stride != h->sh.i_first_mb )
2837                     {
2838                         x264_bitstream_restore( h, &bs_bak[BS_BAK_SLICE_MAX_SIZE], &i_skip, 0 );
2839                         h->mb.b_reencode_mb = 1;
2840                         if( SLICE_MBAFF )
2841                         {
2842                             // set to bottom of previous mbpair
2843                             if( i_mb_x )
2844                                 h->sh.i_last_mb = mb_xy-1+h->mb.i_mb_stride*(!(i_mb_y&1));
2845                             else
2846                                 h->sh.i_last_mb = (i_mb_y-2+!(i_mb_y&1))*h->mb.i_mb_stride + h->mb.i_mb_width - 1;
2847                         }
2848                         else
2849                             h->sh.i_last_mb = mb_xy-1;
2850                         break;
2851                     }
2852                     else
2853                         h->sh.i_last_mb = mb_xy;
2854                 }
2855                 else
2856                     slice_max_size = 0;
2857             }
2858         }
2859 cont:
2860         h->mb.b_reencode_mb = 0;
2861
2862         /* save cache */
2863         x264_macroblock_cache_save( h );
2864
2865         if( x264_ratecontrol_mb( h, mb_size ) < 0 )
2866         {
2867             x264_bitstream_restore( h, &bs_bak[BS_BAK_ROW_VBV], &i_skip, 1 );
2868             h->mb.b_reencode_mb = 1;
2869             i_mb_x = 0;
2870             i_mb_y = i_mb_y - SLICE_MBAFF;
2871             h->mb.i_mb_prev_xy = i_mb_y * h->mb.i_mb_stride - 1;
2872             h->sh.i_last_mb = orig_last_mb;
2873             continue;
2874         }
2875
2876         /* accumulate mb stats */
2877         h->stat.frame.i_mb_count[h->mb.i_type]++;
2878
2879         int b_intra = IS_INTRA( h->mb.i_type );
2880         int b_skip = IS_SKIP( h->mb.i_type );
2881         if( h->param.i_log_level >= X264_LOG_INFO || h->param.rc.b_stat_write )
2882         {
2883             if( !b_intra && !b_skip && !IS_DIRECT( h->mb.i_type ) )
2884             {
2885                 if( h->mb.i_partition != D_8x8 )
2886                         h->stat.frame.i_mb_partition[h->mb.i_partition] += 4;
2887                     else
2888                         for( int i = 0; i < 4; i++ )
2889                             h->stat.frame.i_mb_partition[h->mb.i_sub_partition[i]] ++;
2890                 if( h->param.i_frame_reference > 1 )
2891                     for( int i_list = 0; i_list <= (h->sh.i_type == SLICE_TYPE_B); i_list++ )
2892                         for( int i = 0; i < 4; i++ )
2893                         {
2894                             int i_ref = h->mb.cache.ref[i_list][ x264_scan8[4*i] ];
2895                             if( i_ref >= 0 )
2896                                 h->stat.frame.i_mb_count_ref[i_list][i_ref] ++;
2897                         }
2898             }
2899         }
2900
2901         if( h->param.i_log_level >= X264_LOG_INFO )
2902         {
2903             if( h->mb.i_cbp_luma | h->mb.i_cbp_chroma )
2904             {
2905                 if( CHROMA444 )
2906                 {
2907                     for( int i = 0; i < 4; i++ )
2908                         if( h->mb.i_cbp_luma & (1 << i) )
2909                             for( int p = 0; p < 3; p++ )
2910                             {
2911                                 int s8 = i*4+p*16;
2912                                 int nnz8x8 = M16( &h->mb.cache.non_zero_count[x264_scan8[s8]+0] )
2913                                            | M16( &h->mb.cache.non_zero_count[x264_scan8[s8]+8] );
2914                                 h->stat.frame.i_mb_cbp[!b_intra + p*2] += !!nnz8x8;
2915                             }
2916                 }
2917                 else
2918                 {
2919                     int cbpsum = (h->mb.i_cbp_luma&1) + ((h->mb.i_cbp_luma>>1)&1)
2920                                + ((h->mb.i_cbp_luma>>2)&1) + (h->mb.i_cbp_luma>>3);
2921                     h->stat.frame.i_mb_cbp[!b_intra + 0] += cbpsum;
2922                     h->stat.frame.i_mb_cbp[!b_intra + 2] += !!h->mb.i_cbp_chroma;
2923                     h->stat.frame.i_mb_cbp[!b_intra + 4] += h->mb.i_cbp_chroma >> 1;
2924                 }
2925             }
2926             if( h->mb.i_cbp_luma && !b_intra )
2927             {
2928                 h->stat.frame.i_mb_count_8x8dct[0] ++;
2929                 h->stat.frame.i_mb_count_8x8dct[1] += h->mb.b_transform_8x8;
2930             }
2931             if( b_intra && h->mb.i_type != I_PCM )
2932             {
2933                 if( h->mb.i_type == I_16x16 )
2934                     h->stat.frame.i_mb_pred_mode[0][h->mb.i_intra16x16_pred_mode]++;
2935                 else if( h->mb.i_type == I_8x8 )
2936                     for( int i = 0; i < 16; i += 4 )
2937                         h->stat.frame.i_mb_pred_mode[1][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
2938                 else //if( h->mb.i_type == I_4x4 )
2939                     for( int i = 0; i < 16; i++ )
2940                         h->stat.frame.i_mb_pred_mode[2][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
2941                 h->stat.frame.i_mb_pred_mode[3][x264_mb_chroma_pred_mode_fix[h->mb.i_chroma_pred_mode]]++;
2942             }
2943             h->stat.frame.i_mb_field[b_intra?0:b_skip?2:1] += MB_INTERLACED;
2944         }
2945
2946         /* calculate deblock strength values (actual deblocking is done per-row along with hpel) */
2947         if( b_deblock )
2948             x264_macroblock_deblock_strength( h );
2949
2950         if( mb_xy == h->sh.i_last_mb )
2951             break;
2952
2953         if( SLICE_MBAFF )
2954         {
2955             i_mb_x += i_mb_y & 1;
2956             i_mb_y ^= i_mb_x < h->mb.i_mb_width;
2957         }
2958         else
2959             i_mb_x++;
2960         if( i_mb_x == h->mb.i_mb_width )
2961         {
2962             i_mb_y++;
2963             i_mb_x = 0;
2964         }
2965     }
2966     if( h->sh.i_last_mb < h->sh.i_first_mb )
2967         return 0;
2968
2969     h->out.nal[h->out.i_nal].i_last_mb = h->sh.i_last_mb;
2970
2971     if( h->param.b_cabac )
2972     {
2973         x264_cabac_encode_flush( h, &h->cabac );
2974         h->out.bs.p = h->cabac.p;
2975     }
2976     else
2977     {
2978         if( i_skip > 0 )
2979             bs_write_ue( &h->out.bs, i_skip );  /* last skip run */
2980         /* rbsp_slice_trailing_bits */
2981         bs_rbsp_trailing( &h->out.bs );
2982         bs_flush( &h->out.bs );
2983     }
2984     if( x264_nal_end( h ) )
2985         return -1;
2986
2987     if( h->sh.i_last_mb == (h->i_threadslice_end * h->mb.i_mb_width - 1) )
2988     {
2989         h->stat.frame.i_misc_bits = bs_pos( &h->out.bs )
2990                                   + (h->out.i_nal*NALU_OVERHEAD * 8)
2991                                   - h->stat.frame.i_tex_bits
2992                                   - h->stat.frame.i_mv_bits;
2993         x264_fdec_filter_row( h, h->i_threadslice_end, 0 );
2994
2995         if( h->param.b_sliced_threads )
2996         {
2997             /* Tell the main thread we're done. */
2998             x264_threadslice_cond_broadcast( h, 1 );
2999             /* Do hpel now */
3000             for( int mb_y = h->i_threadslice_start; mb_y <= h->i_threadslice_end; mb_y++ )
3001                 x264_fdec_filter_row( h, mb_y, 1 );
3002             x264_threadslice_cond_broadcast( h, 2 );
3003             /* Do the first row of hpel, now that the previous slice is done */
3004             if( h->i_thread_idx > 0 )
3005             {
3006                 x264_threadslice_cond_wait( h->thread[h->i_thread_idx-1], 2 );
3007                 x264_fdec_filter_row( h, h->i_threadslice_start + (1 << SLICE_MBAFF), 2 );
3008             }
3009         }
3010
3011         /* Free mb info after the last thread's done using it */
3012         if( h->fdec->mb_info_free && (!h->param.b_sliced_threads || h->i_thread_idx == (h->param.i_threads-1)) )
3013         {
3014             h->fdec->mb_info_free( h->fdec->mb_info );
3015             h->fdec->mb_info = NULL;
3016             h->fdec->mb_info_free = NULL;
3017         }
3018     }
3019
3020     return 0;
3021 }
3022
3023 static void x264_thread_sync_context( x264_t *dst, x264_t *src )
3024 {
3025     if( dst == src )
3026         return;
3027
3028     // reference counting
3029     for( x264_frame_t **f = src->frames.reference; *f; f++ )
3030         (*f)->i_reference_count++;
3031     for( x264_frame_t **f = dst->frames.reference; *f; f++ )
3032         x264_frame_push_unused( src, *f );
3033     src->fdec->i_reference_count++;
3034     x264_frame_push_unused( src, dst->fdec );
3035
3036     // copy everything except the per-thread pointers and the constants.
3037     memcpy( &dst->i_frame, &src->i_frame, offsetof(x264_t, mb.base) - offsetof(x264_t, i_frame) );
3038     dst->param = src->param;
3039     dst->stat = src->stat;
3040     dst->pixf = src->pixf;
3041     dst->reconfig = src->reconfig;
3042 }
3043
3044 static void x264_thread_sync_stat( x264_t *dst, x264_t *src )
3045 {
3046     if( dst != src )
3047         memcpy( &dst->stat, &src->stat, offsetof(x264_t, stat.frame) - offsetof(x264_t, stat) );
3048 }
3049
3050 static void *x264_slices_write( x264_t *h )
3051 {
3052     int i_slice_num = 0;
3053     int last_thread_mb = h->sh.i_last_mb;
3054
3055     /* init stats */
3056     memset( &h->stat.frame, 0, sizeof(h->stat.frame) );
3057     h->mb.b_reencode_mb = 0;
3058     while( h->sh.i_first_mb + SLICE_MBAFF*h->mb.i_mb_stride <= last_thread_mb )
3059     {
3060         h->sh.i_last_mb = last_thread_mb;
3061         if( !i_slice_num || !x264_frame_new_slice( h, h->fdec ) )
3062         {
3063             if( h->param.i_slice_max_mbs )
3064             {
3065                 if( SLICE_MBAFF )
3066                 {
3067                     // convert first to mbaff form, add slice-max-mbs, then convert back to normal form
3068                     int last_mbaff = 2*(h->sh.i_first_mb % h->mb.i_mb_width)
3069                         + h->mb.i_mb_width*(h->sh.i_first_mb / h->mb.i_mb_width)
3070                         + h->param.i_slice_max_mbs - 1;
3071                     int last_x = (last_mbaff % (2*h->mb.i_mb_width))/2;
3072                     int last_y = (last_mbaff / (2*h->mb.i_mb_width))*2 + 1;
3073                     h->sh.i_last_mb = last_x + h->mb.i_mb_stride*last_y;
3074                 }
3075                 else
3076                 {
3077                     h->sh.i_last_mb = h->sh.i_first_mb + h->param.i_slice_max_mbs - 1;
3078                     if( h->sh.i_last_mb < last_thread_mb && last_thread_mb - h->sh.i_last_mb < h->param.i_slice_min_mbs )
3079                         h->sh.i_last_mb = last_thread_mb - h->param.i_slice_min_mbs;
3080                 }
3081                 i_slice_num++;
3082             }
3083             else if( h->param.i_slice_count && !h->param.b_sliced_threads )
3084             {
3085                 int height = h->mb.i_mb_height >> PARAM_INTERLACED;
3086                 int width = h->mb.i_mb_width << PARAM_INTERLACED;
3087                 i_slice_num++;
3088                 h->sh.i_last_mb = (height * i_slice_num + h->param.i_slice_count/2) / h->param.i_slice_count * width - 1;
3089             }
3090         }
3091         h->sh.i_last_mb = X264_MIN( h->sh.i_last_mb, last_thread_mb );
3092         if( x264_stack_align( x264_slice_write, h ) )
3093             goto fail;
3094         h->sh.i_first_mb = h->sh.i_last_mb + 1;
3095         // if i_first_mb is not the last mb in a row then go to the next mb in MBAFF order
3096         if( SLICE_MBAFF && h->sh.i_first_mb % h->mb.i_mb_width )
3097             h->sh.i_first_mb -= h->mb.i_mb_stride;
3098     }
3099
3100     return (void *)0;
3101
3102 fail:
3103     /* Tell other threads we're done, so they wouldn't wait for it */
3104     if( h->param.b_sliced_threads )
3105         x264_threadslice_cond_broadcast( h, 2 );
3106     return (void *)-1;
3107 }
3108
3109 static int x264_threaded_slices_write( x264_t *h )
3110 {
3111     /* set first/last mb and sync contexts */
3112     for( int i = 0; i < h->param.i_threads; i++ )
3113     {
3114         x264_t *t = h->thread[i];
3115         if( i )
3116         {
3117             t->param = h->param;
3118             memcpy( &t->i_frame, &h->i_frame, offsetof(x264_t, rc) - offsetof(x264_t, i_frame) );
3119         }
3120         int height = h->mb.i_mb_height >> PARAM_INTERLACED;
3121         t->i_threadslice_start = ((height *  i    + h->param.i_slice_count/2) / h->param.i_threads) << PARAM_INTERLACED;
3122         t->i_threadslice_end   = ((height * (i+1) + h->param.i_slice_count/2) / h->param.i_threads) << PARAM_INTERLACED;
3123         t->sh.i_first_mb = t->i_threadslice_start * h->mb.i_mb_width;
3124         t->sh.i_last_mb  =   t->i_threadslice_end * h->mb.i_mb_width - 1;
3125     }
3126
3127     x264_stack_align( x264_analyse_weight_frame, h, h->mb.i_mb_height*16 + 16 );
3128
3129     x264_threads_distribute_ratecontrol( h );
3130
3131     /* setup */
3132     for( int i = 0; i < h->param.i_threads; i++ )
3133     {
3134         h->thread[i]->i_thread_idx = i;
3135         h->thread[i]->b_thread_active = 1;
3136         x264_threadslice_cond_broadcast( h->thread[i], 0 );
3137     }
3138     /* dispatch */
3139     for( int i = 0; i < h->param.i_threads; i++ )
3140         x264_threadpool_run( h->threadpool, (void*)x264_slices_write, h->thread[i] );
3141     /* wait */
3142     for( int i = 0; i < h->param.i_threads; i++ )
3143         x264_threadslice_cond_wait( h->thread[i], 1 );
3144
3145     x264_threads_merge_ratecontrol( h );
3146
3147     for( int i = 1; i < h->param.i_threads; i++ )
3148     {
3149         x264_t *t = h->thread[i];
3150         for( int j = 0; j < t->out.i_nal; j++ )
3151         {
3152             h->out.nal[h->out.i_nal] = t->out.nal[j];
3153             h->out.i_nal++;
3154             x264_nal_check_buffer( h );
3155         }
3156         /* All entries in stat.frame are ints except for ssd/ssim. */
3157         for( int j = 0; j < (offsetof(x264_t,stat.frame.i_ssd) - offsetof(x264_t,stat.frame.i_mv_bits)) / sizeof(int); j++ )
3158             ((int*)&h->stat.frame)[j] += ((int*)&t->stat.frame)[j];
3159         for( int j = 0; j < 3; j++ )
3160             h->stat.frame.i_ssd[j] += t->stat.frame.i_ssd[j];
3161         h->stat.frame.f_ssim += t->stat.frame.f_ssim;
3162         h->stat.frame.i_ssim_cnt += t->stat.frame.i_ssim_cnt;
3163     }
3164
3165     return 0;
3166 }
3167
3168 void x264_encoder_intra_refresh( x264_t *h )
3169 {
3170     h = h->thread[h->i_thread_phase];
3171     h->b_queued_intra_refresh = 1;
3172 }
3173
3174 int x264_encoder_invalidate_reference( x264_t *h, int64_t pts )
3175 {
3176     if( h->param.i_bframe )
3177     {
3178         x264_log( h, X264_LOG_ERROR, "x264_encoder_invalidate_reference is not supported with B-frames enabled\n" );
3179         return -1;
3180     }
3181     if( h->param.b_intra_refresh )
3182     {
3183         x264_log( h, X264_LOG_ERROR, "x264_encoder_invalidate_reference is not supported with intra refresh enabled\n" );
3184         return -1;
3185     }
3186     h = h->thread[h->i_thread_phase];
3187     if( pts >= h->i_last_idr_pts )
3188     {
3189         for( int i = 0; h->frames.reference[i]; i++ )
3190             if( pts <= h->frames.reference[i]->i_pts )
3191                 h->frames.reference[i]->b_corrupt = 1;
3192         if( pts <= h->fdec->i_pts )
3193             h->fdec->b_corrupt = 1;
3194     }
3195     return 0;
3196 }
3197
3198 /****************************************************************************
3199  * x264_encoder_encode:
3200  *  XXX: i_poc   : is the poc of the current given picture
3201  *       i_frame : is the number of the frame being coded
3202  *  ex:  type frame poc
3203  *       I      0   2*0
3204  *       P      1   2*3
3205  *       B      2   2*1
3206  *       B      3   2*2
3207  *       P      4   2*6
3208  *       B      5   2*4
3209  *       B      6   2*5
3210  ****************************************************************************/
3211 int     x264_encoder_encode( x264_t *h,
3212                              x264_nal_t **pp_nal, int *pi_nal,
3213                              x264_picture_t *pic_in,
3214                              x264_picture_t *pic_out )
3215 {
3216     x264_t *thread_current, *thread_prev, *thread_oldest;
3217     int i_nal_type, i_nal_ref_idc, i_global_qp;
3218     int overhead = NALU_OVERHEAD;
3219
3220 #if HAVE_OPENCL
3221     if( h->opencl.b_fatal_error )
3222         return -1;
3223 #endif
3224
3225     if( h->i_thread_frames > 1 )
3226     {
3227         thread_prev    = h->thread[ h->i_thread_phase ];
3228         h->i_thread_phase = (h->i_thread_phase + 1) % h->i_thread_frames;
3229         thread_current = h->thread[ h->i_thread_phase ];
3230         thread_oldest  = h->thread[ (h->i_thread_phase + 1) % h->i_thread_frames ];
3231         x264_thread_sync_context( thread_current, thread_prev );
3232         x264_thread_sync_ratecontrol( thread_current, thread_prev, thread_oldest );
3233         h = thread_current;
3234     }
3235     else
3236     {
3237         thread_current =
3238         thread_oldest  = h;
3239     }
3240     h->i_cpb_delay_pir_offset = h->i_cpb_delay_pir_offset_next;
3241
3242     /* no data out */
3243     *pi_nal = 0;
3244     *pp_nal = NULL;
3245
3246     /* ------------------- Setup new frame from picture -------------------- */
3247     if( pic_in != NULL )
3248     {
3249         if( h->lookahead->b_exit_thread )
3250         {
3251             x264_log( h, X264_LOG_ERROR, "lookahead thread is already stopped\n" );
3252             return -1;
3253         }
3254
3255         /* 1: Copy the picture to a frame and move it to a buffer */
3256         x264_frame_t *fenc = x264_frame_pop_unused( h, 0 );
3257         if( !fenc )
3258             return -1;
3259
3260         if( x264_frame_copy_picture( h, fenc, pic_in ) < 0 )
3261             return -1;
3262
3263         if( h->param.i_width != 16 * h->mb.i_mb_width ||
3264             h->param.i_height != 16 * h->mb.i_mb_height )
3265             x264_frame_expand_border_mod16( h, fenc );
3266
3267         fenc->i_frame = h->frames.i_input++;
3268
3269         if( fenc->i_frame == 0 )
3270             h->frames.i_first_pts = fenc->i_pts;
3271         if( h->frames.i_bframe_delay && fenc->i_frame == h->frames.i_bframe_delay )
3272             h->frames.i_bframe_delay_time = fenc->i_pts - h->frames.i_first_pts;
3273
3274         if( h->param.b_vfr_input && fenc->i_pts <= h->frames.i_largest_pts )
3275             x264_log( h, X264_LOG_WARNING, "non-strictly-monotonic PTS\n" );
3276
3277         h->frames.i_second_largest_pts = h->frames.i_largest_pts;
3278         h->frames.i_largest_pts = fenc->i_pts;
3279
3280         if( (fenc->i_pic_struct < PIC_STRUCT_AUTO) || (fenc->i_pic_struct > PIC_STRUCT_TRIPLE) )
3281             fenc->i_pic_struct = PIC_STRUCT_AUTO;
3282
3283         if( fenc->i_pic_struct == PIC_STRUCT_AUTO )
3284         {
3285 #if HAVE_INTERLACED
3286             int b_interlaced = fenc->param ? fenc->param->b_interlaced : h->param.b_interlaced;
3287 #else
3288             int b_interlaced = 0;
3289 #endif
3290             if( b_interlaced )
3291             {
3292                 int b_tff = fenc->param ? fenc->param->b_tff : h->param.b_tff;
3293                 fenc->i_pic_struct = b_tff ? PIC_STRUCT_TOP_BOTTOM : PIC_STRUCT_BOTTOM_TOP;
3294             }
3295             else
3296                 fenc->i_pic_struct = PIC_STRUCT_PROGRESSIVE;
3297         }
3298
3299         if( h->param.rc.b_mb_tree && h->param.rc.b_stat_read )
3300         {
3301             if( x264_macroblock_tree_read( h, fenc, pic_in->prop.quant_offsets ) )
3302                 return -1;
3303         }
3304         else
3305             x264_stack_align( x264_adaptive_quant_frame, h, fenc, pic_in->prop.quant_offsets );
3306
3307         if( pic_in->prop.quant_offsets_free )
3308             pic_in->prop.quant_offsets_free( pic_in->prop.quant_offsets );
3309
3310         if( h->frames.b_have_lowres )
3311             x264_frame_init_lowres( h, fenc );
3312
3313         /* 2: Place the frame into the queue for its slice type decision */
3314         x264_lookahead_put_frame( h, fenc );
3315
3316         if( h->frames.i_input <= h->frames.i_delay + 1 - h->i_thread_frames )
3317         {
3318             /* Nothing yet to encode, waiting for filling of buffers */
3319             pic_out->i_type = X264_TYPE_AUTO;
3320             return 0;
3321         }
3322     }
3323     else
3324     {
3325         /* signal kills for lookahead thread */
3326         x264_pthread_mutex_lock( &h->lookahead->ifbuf.mutex );
3327         h->lookahead->b_exit_thread = 1;
3328         x264_pthread_cond_broadcast( &h->lookahead->ifbuf.cv_fill );
3329         x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
3330     }
3331
3332     h->i_frame++;
3333     /* 3: The picture is analyzed in the lookahead */
3334     if( !h->frames.current[0] )
3335         x264_lookahead_get_frames( h );
3336
3337     if( !h->frames.current[0] && x264_lookahead_is_empty( h ) )
3338         return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
3339
3340     /* ------------------- Get frame to be encoded ------------------------- */
3341     /* 4: get picture to encode */
3342     h->fenc = x264_frame_shift( h->frames.current );
3343
3344     /* If applicable, wait for previous frame reconstruction to finish */
3345     if( h->param.b_sliced_threads )
3346         if( x264_threadpool_wait_all( h ) < 0 )
3347             return -1;
3348
3349     if( h->i_frame == h->i_thread_frames - 1 )
3350         h->i_reordered_pts_delay = h->fenc->i_reordered_pts;
3351     if( h->reconfig )
3352     {
3353         x264_encoder_reconfig_apply( h, &h->reconfig_h->param );
3354         h->reconfig = 0;
3355     }
3356     if( h->fenc->param )
3357     {
3358         x264_encoder_reconfig_apply( h, h->fenc->param );
3359         if( h->fenc->param->param_free )
3360         {
3361             h->fenc->param->param_free( h->fenc->param );
3362             h->fenc->param = NULL;
3363         }
3364     }
3365
3366     // ok to call this before encoding any frames, since the initial values of fdec have b_kept_as_ref=0
3367     if( x264_reference_update( h ) )
3368         return -1;
3369     h->fdec->i_lines_completed = -1;
3370
3371     if( !IS_X264_TYPE_I( h->fenc->i_type ) )
3372     {
3373         int valid_refs_left = 0;
3374         for( int i = 0; h->frames.reference[i]; i++ )
3375             if( !h->frames.reference[i]->b_corrupt )
3376                 valid_refs_left++;
3377         /* No valid reference frames left: force an IDR. */
3378         if( !valid_refs_left )
3379         {
3380             h->fenc->b_keyframe = 1;
3381             h->fenc->i_type = X264_TYPE_IDR;
3382         }
3383     }
3384
3385     if( h->fenc->b_keyframe )
3386     {
3387         h->frames.i_last_keyframe = h->fenc->i_frame;
3388         if( h->fenc->i_type == X264_TYPE_IDR )
3389         {
3390             h->i_frame_num = 0;
3391             h->frames.i_last_idr = h->fenc->i_frame;
3392         }
3393     }
3394     h->sh.i_mmco_command_count =
3395     h->sh.i_mmco_remove_from_end = 0;
3396     h->b_ref_reorder[0] =
3397     h->b_ref_reorder[1] = 0;
3398     h->fdec->i_poc =
3399     h->fenc->i_poc = 2 * ( h->fenc->i_frame - X264_MAX( h->frames.i_last_idr, 0 ) );
3400
3401     /* ------------------- Setup frame context ----------------------------- */
3402     /* 5: Init data dependent of frame type */
3403     if( h->fenc->i_type == X264_TYPE_IDR )
3404     {
3405         /* reset ref pictures */
3406         i_nal_type    = NAL_SLICE_IDR;
3407         i_nal_ref_idc = NAL_PRIORITY_HIGHEST;
3408         h->sh.i_type = SLICE_TYPE_I;
3409         x264_reference_reset( h );
3410         h->frames.i_poc_last_open_gop = -1;
3411     }
3412     else if( h->fenc->i_type == X264_TYPE_I )
3413     {
3414         i_nal_type    = NAL_SLICE;
3415         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
3416         h->sh.i_type = SLICE_TYPE_I;
3417         x264_reference_hierarchy_reset( h );
3418         if( h->param.b_open_gop )
3419             h->frames.i_poc_last_open_gop = h->fenc->b_keyframe ? h->fenc->i_poc : -1;
3420     }
3421     else if( h->fenc->i_type == X264_TYPE_P )
3422     {
3423         i_nal_type    = NAL_SLICE;
3424         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
3425         h->sh.i_type = SLICE_TYPE_P;
3426         x264_reference_hierarchy_reset( h );
3427         h->frames.i_poc_last_open_gop = -1;
3428     }
3429     else if( h->fenc->i_type == X264_TYPE_BREF )
3430     {
3431         i_nal_type    = NAL_SLICE;
3432         i_nal_ref_idc = h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT ? NAL_PRIORITY_LOW : NAL_PRIORITY_HIGH;
3433         h->sh.i_type = SLICE_TYPE_B;
3434         x264_reference_hierarchy_reset( h );
3435     }
3436     else    /* B frame */
3437     {
3438         i_nal_type    = NAL_SLICE;
3439         i_nal_ref_idc = NAL_PRIORITY_DISPOSABLE;
3440         h->sh.i_type = SLICE_TYPE_B;
3441     }
3442
3443     h->fdec->i_type = h->fenc->i_type;
3444     h->fdec->i_frame = h->fenc->i_frame;
3445     h->fenc->b_kept_as_ref =
3446     h->fdec->b_kept_as_ref = i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE && h->param.i_keyint_max > 1;
3447
3448     h->fdec->mb_info = h->fenc->mb_info;
3449     h->fdec->mb_info_free = h->fenc->mb_info_free;
3450     h->fenc->mb_info = NULL;
3451     h->fenc->mb_info_free = NULL;
3452
3453     h->fdec->i_pts = h->fenc->i_pts;
3454     if( h->frames.i_bframe_delay )
3455     {
3456         int64_t *prev_reordered_pts = thread_current->frames.i_prev_reordered_pts;
3457         h->fdec->i_dts = h->i_frame > h->frames.i_bframe_delay
3458                        ? prev_reordered_pts[ (h->i_frame - h->frames.i_bframe_delay) % h->frames.i_bframe_delay ]
3459                        : h->fenc->i_reordered_pts - h->frames.i_bframe_delay_time;
3460         prev_reordered_pts[ h->i_frame % h->frames.i_bframe_delay ] = h->fenc->i_reordered_pts;
3461     }
3462     else
3463         h->fdec->i_dts = h->fenc->i_reordered_pts;
3464     if( h->fenc->i_type == X264_TYPE_IDR )
3465         h->i_last_idr_pts = h->fdec->i_pts;
3466
3467     /* ------------------- Init                ----------------------------- */
3468     /* build ref list 0/1 */
3469     x264_reference_build_list( h, h->fdec->i_poc );
3470
3471     /* ---------------------- Write the bitstream -------------------------- */
3472     /* Init bitstream context */
3473     if( h->param.b_sliced_threads )
3474     {
3475         for( int i = 0; i < h->param.i_threads; i++ )
3476         {
3477             bs_init( &h->thread[i]->out.bs, h->thread[i]->out.p_bitstream, h->thread[i]->out.i_bitstream );
3478             h->thread[i]->out.i_nal = 0;
3479         }
3480     }
3481     else
3482     {
3483         bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
3484         h->out.i_nal = 0;
3485     }
3486
3487     if( h->param.b_aud )
3488     {
3489         int pic_type;
3490
3491         if( h->sh.i_type == SLICE_TYPE_I )
3492             pic_type = 0;
3493         else if( h->sh.i_type == SLICE_TYPE_P )
3494             pic_type = 1;
3495         else if( h->sh.i_type == SLICE_TYPE_B )
3496             pic_type = 2;
3497         else
3498             pic_type = 7;
3499
3500         x264_nal_start( h, NAL_AUD, NAL_PRIORITY_DISPOSABLE );
3501         bs_write( &h->out.bs, 3, pic_type );
3502         bs_rbsp_trailing( &h->out.bs );
3503         if( x264_nal_end( h ) )
3504             return -1;
3505         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
3506     }
3507
3508     h->i_nal_type = i_nal_type;
3509     h->i_nal_ref_idc = i_nal_ref_idc;
3510
3511     if( h->param.b_intra_refresh )
3512     {
3513         if( IS_X264_TYPE_I( h->fenc->i_type ) )
3514         {
3515             h->fdec->i_frames_since_pir = 0;
3516             h->b_queued_intra_refresh = 0;
3517             /* PIR is currently only supported with ref == 1, so any intra frame effectively refreshes
3518              * the whole frame and counts as an intra refresh. */
3519             h->fdec->f_pir_position = h->mb.i_mb_width;
3520         }
3521         else if( h->fenc->i_type == X264_TYPE_P )
3522         {
3523             int pocdiff = (h->fdec->i_poc - h->fref[0][0]->i_poc)/2;
3524             float increment = X264_MAX( ((float)h->mb.i_mb_width-1) / h->param.i_keyint_max, 1 );
3525             h->fdec->f_pir_position = h->fref[0][0]->f_pir_position;
3526             h->fdec->i_frames_since_pir = h->fref[0][0]->i_frames_since_pir + pocdiff;
3527             if( h->fdec->i_frames_since_pir >= h->param.i_keyint_max ||
3528                 (h->b_queued_intra_refresh && h->fdec->f_pir_position + 0.5 >= h->mb.i_mb_width) )
3529             {
3530                 h->fdec->f_pir_position = 0;
3531                 h->fdec->i_frames_since_pir = 0;
3532                 h->b_queued_intra_refresh = 0;
3533                 h->fenc->b_keyframe = 1;
3534             }
3535             h->fdec->i_pir_start_col = h->fdec->f_pir_position+0.5;
3536             h->fdec->f_pir_position += increment * pocdiff;
3537             h->fdec->i_pir_end_col = h->fdec->f_pir_position+0.5;
3538             /* If our intra refresh has reached the right side of the frame, we're done. */
3539             if( h->fdec->i_pir_end_col >= h->mb.i_mb_width - 1 )
3540             {
3541                 h->fdec->f_pir_position = h->mb.i_mb_width;
3542                 h->fdec->i_pir_end_col = h->mb.i_mb_width - 1;
3543             }
3544         }
3545     }
3546
3547     if( h->fenc->b_keyframe )
3548     {
3549         /* Write SPS and PPS */
3550         if( h->param.b_repeat_headers )
3551         {
3552             /* generate sequence parameters */
3553             x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
3554             x264_sps_write( &h->out.bs, h->sps );
3555             if( x264_nal_end( h ) )
3556                 return -1;
3557             /* Pad AUD/SPS to 256 bytes like Panasonic */
3558             if( h->param.i_avcintra_class )
3559                 h->out.nal[h->out.i_nal-1].i_padding = 256 - bs_pos( &h->out.bs ) / 8 - 2*NALU_OVERHEAD;
3560             overhead += h->out.nal[h->out.i_nal-1].i_payload + h->out.nal[h->out.i_nal-1].i_padding + NALU_OVERHEAD;
3561
3562             /* generate picture parameters */
3563             x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
3564             x264_pps_write( &h->out.bs, h->sps, h->pps );
3565             if( x264_nal_end( h ) )
3566                 return -1;
3567             if( h->param.i_avcintra_class )
3568                 h->out.nal[h->out.i_nal-1].i_padding = 256 - h->out.nal[h->out.i_nal-1].i_payload - NALU_OVERHEAD;
3569             overhead += h->out.nal[h->out.i_nal-1].i_payload + h->out.nal[h->out.i_nal-1].i_padding + NALU_OVERHEAD;
3570         }
3571
3572         /* when frame threading is used, buffering period sei is written in x264_encoder_frame_end */
3573         if( h->i_thread_frames == 1 && h->sps->vui.b_nal_hrd_parameters_present )
3574         {
3575             x264_hrd_fullness( h );
3576             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3577             x264_sei_buffering_period_write( h, &h->out.bs );
3578             if( x264_nal_end( h ) )
3579                return -1;
3580             overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3581         }
3582     }
3583
3584     /* write extra sei */
3585     for( int i = 0; i < h->fenc->extra_sei.num_payloads; i++ )
3586     {
3587         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3588         x264_sei_write( &h->out.bs, h->fenc->extra_sei.payloads[i].payload, h->fenc->extra_sei.payloads[i].payload_size,
3589                         h->fenc->extra_sei.payloads[i].payload_type );
3590         if( x264_nal_end( h ) )
3591             return -1;
3592         overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3593         if( h->fenc->extra_sei.sei_free )
3594         {
3595             h->fenc->extra_sei.sei_free( h->fenc->extra_sei.payloads[i].payload );
3596             h->fenc->extra_sei.payloads[i].payload = NULL;
3597         }
3598     }
3599
3600     if( h->fenc->extra_sei.sei_free )
3601     {
3602         h->fenc->extra_sei.sei_free( h->fenc->extra_sei.payloads );
3603         h->fenc->extra_sei.payloads = NULL;
3604         h->fenc->extra_sei.sei_free = NULL;
3605     }
3606
3607     if( h->fenc->b_keyframe )
3608     {
3609         /* Avid's decoder strictly wants two SEIs for AVC-Intra so we can't insert the x264 SEI */
3610         if( h->param.b_repeat_headers && h->fenc->i_frame == 0 && !h->param.i_avcintra_class )
3611         {
3612             /* identify ourself */
3613             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3614             if( x264_sei_version_write( h, &h->out.bs ) )
3615                 return -1;
3616             if( x264_nal_end( h ) )
3617                 return -1;
3618             overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3619         }
3620
3621         if( h->fenc->i_type != X264_TYPE_IDR )
3622         {
3623             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;
3624             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3625             x264_sei_recovery_point_write( h, &h->out.bs, time_to_recovery );
3626             if( x264_nal_end( h ) )
3627                 return -1;
3628             overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3629         }
3630     }
3631
3632     if( h->param.i_frame_packing >= 0 && (h->fenc->b_keyframe || h->param.i_frame_packing == 5) )
3633     {
3634         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3635         x264_sei_frame_packing_write( h, &h->out.bs );
3636         if( x264_nal_end( h ) )
3637             return -1;
3638         overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3639     }
3640
3641     /* generate sei pic timing */
3642     if( h->sps->vui.b_pic_struct_present || h->sps->vui.b_nal_hrd_parameters_present )
3643     {
3644         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3645         x264_sei_pic_timing_write( h, &h->out.bs );
3646         if( x264_nal_end( h ) )
3647             return -1;
3648         overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3649     }
3650
3651     /* As required by Blu-ray. */
3652     if( !IS_X264_TYPE_B( h->fenc->i_type ) && h->b_sh_backup )
3653     {
3654         h->b_sh_backup = 0;
3655         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3656         x264_sei_dec_ref_pic_marking_write( h, &h->out.bs );
3657         if( x264_nal_end( h ) )
3658             return -1;
3659         overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3660     }
3661
3662     if( h->fenc->b_keyframe && h->param.b_intra_refresh )
3663         h->i_cpb_delay_pir_offset_next = h->fenc->i_cpb_delay;
3664
3665     /* Filler space: 10 or 18 SEIs' worth of space, depending on resolution */
3666     if( h->param.i_avcintra_class )
3667     {
3668         /* Write an empty filler NAL to mimic the AUD in the P2 format*/
3669         x264_nal_start( h, NAL_FILLER, NAL_PRIORITY_DISPOSABLE );
3670         x264_filler_write( h, &h->out.bs, 0 );
3671         if( x264_nal_end( h ) )
3672             return -1;
3673         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
3674
3675         /* All lengths are magic lengths that decoders expect to see */
3676         /* "UMID" SEI */
3677         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3678         if( x264_sei_avcintra_umid_write( h, &h->out.bs ) < 0 )
3679             return -1;
3680         if( x264_nal_end( h ) )
3681             return -1;
3682         overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3683
3684         int unpadded_len;
3685         int total_len;
3686         if( h->param.i_height == 1080 )
3687         {
3688             unpadded_len = 5780;
3689             total_len = 17*512;
3690         }
3691         else
3692         {
3693             unpadded_len = 2900;
3694             total_len = 9*512;
3695         }
3696         /* "VANC" SEI */
3697         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3698         if( x264_sei_avcintra_vanc_write( h, &h->out.bs, unpadded_len ) < 0 )
3699             return -1;
3700         if( x264_nal_end( h ) )
3701             return -1;
3702
3703         h->out.nal[h->out.i_nal-1].i_padding = total_len - h->out.nal[h->out.i_nal-1].i_payload - SEI_OVERHEAD;
3704         overhead += h->out.nal[h->out.i_nal-1].i_payload + h->out.nal[h->out.i_nal-1].i_padding + SEI_OVERHEAD;
3705     }
3706
3707     /* Init the rate control */
3708     /* FIXME: Include slice header bit cost. */
3709     x264_ratecontrol_start( h, h->fenc->i_qpplus1, overhead*8 );
3710     i_global_qp = x264_ratecontrol_qp( h );
3711
3712     pic_out->i_qpplus1 =
3713     h->fdec->i_qpplus1 = i_global_qp + 1;
3714
3715     if( h->param.rc.b_stat_read && h->sh.i_type != SLICE_TYPE_I )
3716     {
3717         x264_reference_build_list_optimal( h );
3718         x264_reference_check_reorder( h );
3719     }
3720
3721     if( h->i_ref[0] )
3722         h->fdec->i_poc_l0ref0 = h->fref[0][0]->i_poc;
3723
3724     /* ------------------------ Create slice header  ----------------------- */
3725     x264_slice_init( h, i_nal_type, i_global_qp );
3726
3727     /*------------------------- Weights -------------------------------------*/
3728     if( h->sh.i_type == SLICE_TYPE_B )
3729         x264_macroblock_bipred_init( h );
3730
3731     x264_weighted_pred_init( h );
3732
3733     if( i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE )
3734         h->i_frame_num++;
3735
3736     /* Write frame */
3737     h->i_threadslice_start = 0;
3738     h->i_threadslice_end = h->mb.i_mb_height;
3739     if( h->i_thread_frames > 1 )
3740     {
3741         x264_threadpool_run( h->threadpool, (void*)x264_slices_write, h );
3742         h->b_thread_active = 1;
3743     }
3744     else if( h->param.b_sliced_threads )
3745     {
3746         if( x264_threaded_slices_write( h ) )
3747             return -1;
3748     }
3749     else
3750         if( (intptr_t)x264_slices_write( h ) )
3751             return -1;
3752
3753     return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
3754 }
3755
3756 static int x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
3757                                    x264_nal_t **pp_nal, int *pi_nal,
3758                                    x264_picture_t *pic_out )
3759 {
3760     char psz_message[80];
3761
3762     if( !h->param.b_sliced_threads && h->b_thread_active )
3763     {
3764         h->b_thread_active = 0;
3765         if( (intptr_t)x264_threadpool_wait( h->threadpool, h ) )
3766             return -1;
3767     }
3768     if( !h->out.i_nal )
3769     {
3770         pic_out->i_type = X264_TYPE_AUTO;
3771         return 0;
3772     }
3773
3774     x264_emms();
3775
3776     /* generate buffering period sei and insert it into place */
3777     if( h->i_thread_frames > 1 && h->fenc->b_keyframe && h->sps->vui.b_nal_hrd_parameters_present )
3778     {
3779         x264_hrd_fullness( h );
3780         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3781         x264_sei_buffering_period_write( h, &h->out.bs );
3782         if( x264_nal_end( h ) )
3783            return -1;
3784         /* buffering period sei must follow AUD, SPS and PPS and precede all other SEIs */
3785         int idx = 0;
3786         while( h->out.nal[idx].i_type == NAL_AUD ||
3787                h->out.nal[idx].i_type == NAL_SPS ||
3788                h->out.nal[idx].i_type == NAL_PPS )
3789             idx++;
3790         x264_nal_t nal_tmp = h->out.nal[h->out.i_nal-1];
3791         memmove( &h->out.nal[idx+1], &h->out.nal[idx], (h->out.i_nal-idx-1)*sizeof(x264_nal_t) );
3792         h->out.nal[idx] = nal_tmp;
3793     }
3794
3795     int frame_size = x264_encoder_encapsulate_nals( h, 0 );
3796     if( frame_size < 0 )
3797         return -1;
3798
3799     /* Set output picture properties */
3800     pic_out->i_type = h->fenc->i_type;
3801
3802     pic_out->b_keyframe = h->fenc->b_keyframe;
3803     pic_out->i_pic_struct = h->fenc->i_pic_struct;
3804
3805     pic_out->i_pts = h->fdec->i_pts;
3806     pic_out->i_dts = h->fdec->i_dts;
3807
3808     if( pic_out->i_pts < pic_out->i_dts )
3809         x264_log( h, X264_LOG_WARNING, "invalid DTS: PTS is less than DTS\n" );
3810
3811     pic_out->opaque = h->fenc->opaque;
3812
3813     pic_out->img.i_csp = h->fdec->i_csp;
3814 #if HIGH_BIT_DEPTH
3815     pic_out->img.i_csp |= X264_CSP_HIGH_DEPTH;
3816 #endif
3817     pic_out->img.i_plane = h->fdec->i_plane;
3818     for( int i = 0; i < pic_out->img.i_plane; i++ )
3819     {
3820         pic_out->img.i_stride[i] = h->fdec->i_stride[i] * sizeof(pixel);
3821         pic_out->img.plane[i] = (uint8_t*)h->fdec->plane[i];
3822     }
3823
3824     x264_frame_push_unused( thread_current, h->fenc );
3825
3826     /* ---------------------- Update encoder state ------------------------- */
3827
3828     /* update rc */
3829     int filler = 0;
3830     if( x264_ratecontrol_end( h, frame_size * 8, &filler ) < 0 )
3831         return -1;
3832
3833     pic_out->hrd_timing = h->fenc->hrd_timing;
3834     pic_out->prop.f_crf_avg = h->fdec->f_crf_avg;
3835
3836     /* Filler in AVC-Intra mode is written as zero bytes to the last slice
3837      * We don't know the size of the last slice until encapsulation so we add filler to the encapsulated NAL */
3838     if( h->param.i_avcintra_class )
3839     {
3840         x264_t *h0 = h->thread[0];
3841         int ret = x264_check_encapsulated_buffer( h, h0, h->out.i_nal, frame_size, frame_size + filler );
3842         if( ret < 0 )
3843             return -1;
3844         memset( h->out.nal[0].p_payload + frame_size, 0, filler );
3845         h->out.nal[h->out.i_nal-1].i_payload += filler;
3846         h->out.nal[h->out.i_nal-1].i_padding = filler;
3847         frame_size += filler;
3848     }
3849     else
3850     {
3851         while( filler > 0 )
3852         {
3853             int f, overhead;
3854             overhead = (FILLER_OVERHEAD - h->param.b_annexb);
3855             if( h->param.i_slice_max_size && filler > h->param.i_slice_max_size )
3856             {
3857                 int next_size = filler - h->param.i_slice_max_size;
3858                 int overflow = X264_MAX( overhead - next_size, 0 );
3859                 f = h->param.i_slice_max_size - overhead - overflow;
3860             }
3861             else
3862                 f = X264_MAX( 0, filler - overhead );
3863
3864             if( x264_bitstream_check_buffer_filler( h, f ) )
3865                 return -1;
3866             x264_nal_start( h, NAL_FILLER, NAL_PRIORITY_DISPOSABLE );
3867             x264_filler_write( h, &h->out.bs, f );
3868             if( x264_nal_end( h ) )
3869                 return -1;
3870             int total_size = x264_encoder_encapsulate_nals( h, h->out.i_nal-1 );
3871             if( total_size < 0 )
3872                 return -1;
3873             frame_size += total_size;
3874             filler -= total_size;
3875         }
3876     }
3877
3878     /* End bitstream, set output  */
3879     *pi_nal = h->out.i_nal;
3880     *pp_nal = h->out.nal;
3881
3882     h->out.i_nal = 0;
3883
3884     x264_noise_reduction_update( h );
3885
3886     /* ---------------------- Compute/Print statistics --------------------- */
3887     x264_thread_sync_stat( h, h->thread[0] );
3888
3889     /* Slice stat */
3890     h->stat.i_frame_count[h->sh.i_type]++;
3891     h->stat.i_frame_size[h->sh.i_type] += frame_size;
3892     h->stat.f_frame_qp[h->sh.i_type] += h->fdec->f_qp_avg_aq;
3893
3894     for( int i = 0; i < X264_MBTYPE_MAX; i++ )
3895         h->stat.i_mb_count[h->sh.i_type][i] += h->stat.frame.i_mb_count[i];
3896     for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
3897         h->stat.i_mb_partition[h->sh.i_type][i] += h->stat.frame.i_mb_partition[i];
3898     for( int i = 0; i < 2; i++ )
3899         h->stat.i_mb_count_8x8dct[i] += h->stat.frame.i_mb_count_8x8dct[i];
3900     for( int i = 0; i < 6; i++ )
3901         h->stat.i_mb_cbp[i] += h->stat.frame.i_mb_cbp[i];
3902     for( int i = 0; i < 4; i++ )
3903         for( int j = 0; j < 13; j++ )
3904             h->stat.i_mb_pred_mode[i][j] += h->stat.frame.i_mb_pred_mode[i][j];
3905     if( h->sh.i_type != SLICE_TYPE_I )
3906         for( int i_list = 0; i_list < 2; i_list++ )
3907             for( int i = 0; i < X264_REF_MAX*2; i++ )
3908                 h->stat.i_mb_count_ref[h->sh.i_type][i_list][i] += h->stat.frame.i_mb_count_ref[i_list][i];
3909     for( int i = 0; i < 3; i++ )
3910         h->stat.i_mb_field[i] += h->stat.frame.i_mb_field[i];
3911     if( h->sh.i_type == SLICE_TYPE_P && h->param.analyse.i_weighted_pred >= X264_WEIGHTP_SIMPLE )
3912     {
3913         h->stat.i_wpred[0] += !!h->sh.weight[0][0].weightfn;
3914         h->stat.i_wpred[1] += !!h->sh.weight[0][1].weightfn || !!h->sh.weight[0][2].weightfn;
3915     }
3916     if( h->sh.i_type == SLICE_TYPE_B )
3917     {
3918         h->stat.i_direct_frames[ h->sh.b_direct_spatial_mv_pred ] ++;
3919         if( h->mb.b_direct_auto_write )
3920         {
3921             //FIXME somewhat arbitrary time constants
3922             if( h->stat.i_direct_score[0] + h->stat.i_direct_score[1] > h->mb.i_mb_count )
3923                 for( int i = 0; i < 2; i++ )
3924                     h->stat.i_direct_score[i] = h->stat.i_direct_score[i] * 9/10;
3925             for( int i = 0; i < 2; i++ )
3926                 h->stat.i_direct_score[i] += h->stat.frame.i_direct_score[i];
3927         }
3928     }
3929     else
3930         h->stat.i_consecutive_bframes[h->fenc->i_bframes]++;
3931
3932     psz_message[0] = '\0';
3933     double dur = h->fenc->f_duration;
3934     h->stat.f_frame_duration[h->sh.i_type] += dur;
3935     if( h->param.analyse.b_psnr )
3936     {
3937         int64_t ssd[3] =
3938         {
3939             h->stat.frame.i_ssd[0],
3940             h->stat.frame.i_ssd[1],
3941             h->stat.frame.i_ssd[2],
3942         };
3943         int luma_size = h->param.i_width * h->param.i_height;
3944         int chroma_size = CHROMA_SIZE( luma_size );
3945         pic_out->prop.f_psnr[0] = x264_psnr( ssd[0], luma_size );
3946         pic_out->prop.f_psnr[1] = x264_psnr( ssd[1], chroma_size );
3947         pic_out->prop.f_psnr[2] = x264_psnr( ssd[2], chroma_size );
3948         pic_out->prop.f_psnr_avg = x264_psnr( ssd[0] + ssd[1] + ssd[2], luma_size + chroma_size*2 );
3949
3950         h->stat.f_ssd_global[h->sh.i_type]   += dur * (ssd[0] + ssd[1] + ssd[2]);
3951         h->stat.f_psnr_average[h->sh.i_type] += dur * pic_out->prop.f_psnr_avg;
3952         h->stat.f_psnr_mean_y[h->sh.i_type]  += dur * pic_out->prop.f_psnr[0];
3953         h->stat.f_psnr_mean_u[h->sh.i_type]  += dur * pic_out->prop.f_psnr[1];
3954         h->stat.f_psnr_mean_v[h->sh.i_type]  += dur * pic_out->prop.f_psnr[2];
3955
3956         snprintf( psz_message, 80, " PSNR Y:%5.2f U:%5.2f V:%5.2f", pic_out->prop.f_psnr[0],
3957                                                                     pic_out->prop.f_psnr[1],
3958                                                                     pic_out->prop.f_psnr[2] );
3959     }
3960
3961     if( h->param.analyse.b_ssim )
3962     {
3963         pic_out->prop.f_ssim = h->stat.frame.f_ssim / h->stat.frame.i_ssim_cnt;
3964         h->stat.f_ssim_mean_y[h->sh.i_type] += pic_out->prop.f_ssim * dur;
3965         int msg_len = strlen(psz_message);
3966         snprintf( psz_message + msg_len, 80 - msg_len, " SSIM Y:%.5f", pic_out->prop.f_ssim );
3967