f446583688770b74955219f2e7973dc20fd6b2cb
[x262.git] / encoder / encoder.c
1 /*****************************************************************************
2  * encoder.c: top-level encoder functions
3  *****************************************************************************
4  * Copyright (C) 2003-2012 x264 project
5  *
6  * Authors: Laurent Aimar <fenrir@via.ecp.fr>
7  *          Loren Merritt <lorenm@u.washington.edu>
8  *          Jason Garrett-Glaser <darkshikari@gmail.com>
9  *
10  * This program is free software; you can redistribute it and/or modify
11  * it under the terms of the GNU General Public License as published by
12  * the Free Software Foundation; either version 2 of the License, or
13  * (at your option) any later version.
14  *
15  * This program is distributed in the hope that it will be useful,
16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
18  * GNU General Public License for more details.
19  *
20  * You should have received a copy of the GNU General Public License
21  * along with this program; if not, write to the Free Software
22  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02111, USA.
23  *
24  * This program is also available under a commercial proprietary license.
25  * For more information, contact us at licensing@x264.com.
26  *****************************************************************************/
27
28 #include "common/common.h"
29
30 #include "set.h"
31 #include "analyse.h"
32 #include "ratecontrol.h"
33 #include "macroblock.h"
34 #include "me.h"
35
36 #if HAVE_VISUALIZE
37 #include "common/visualize.h"
38 #endif
39
40 //#define DEBUG_MB_TYPE
41
42 #define bs_write_ue bs_write_ue_big
43
44 static int x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
45                                    x264_nal_t **pp_nal, int *pi_nal,
46                                    x264_picture_t *pic_out );
47
48 /****************************************************************************
49  *
50  ******************************* x264 libs **********************************
51  *
52  ****************************************************************************/
53 static double x264_psnr( double sqe, double size )
54 {
55     double mse = sqe / (PIXEL_MAX*PIXEL_MAX * size);
56     if( mse <= 0.0000000001 ) /* Max 100dB */
57         return 100;
58
59     return -10.0 * log10( mse );
60 }
61
62 static double x264_ssim( double ssim )
63 {
64     double inv_ssim = 1 - ssim;
65     if( inv_ssim <= 0.0000000001 ) /* Max 100dB */
66         return 100;
67
68     return -10.0 * log10( inv_ssim );
69 }
70
71 static void x264_frame_dump( x264_t *h )
72 {
73     FILE *f = fopen( h->param.psz_dump_yuv, "r+b" );
74     if( !f )
75         return;
76
77     /* Write the frame in display order */
78     int frame_size = FRAME_SIZE( h->param.i_height * h->param.i_width * sizeof(pixel) );
79     fseek( f, (uint64_t)h->fdec->i_frame * frame_size, SEEK_SET );
80     for( int p = 0; p < (CHROMA444 ? 3 : 1); p++ )
81         for( int y = 0; y < h->param.i_height; y++ )
82             fwrite( &h->fdec->plane[p][y*h->fdec->i_stride[p]], sizeof(pixel), h->param.i_width, f );
83     if( !CHROMA444 )
84     {
85         int cw = h->param.i_width>>1;
86         int ch = h->param.i_height>>CHROMA_V_SHIFT;
87         pixel *planeu = x264_malloc( (cw*ch*2+32)*sizeof(pixel) );
88         pixel *planev = planeu + cw*ch + 16;
89         h->mc.plane_copy_deinterleave( planeu, cw, planev, cw, h->fdec->plane[1], h->fdec->i_stride[1], cw, ch );
90         fwrite( planeu, 1, cw*ch*sizeof(pixel), f );
91         fwrite( planev, 1, cw*ch*sizeof(pixel), f );
92         x264_free( planeu );
93     }
94     fclose( f );
95 }
96
97 /* Fill "default" values */
98 static void x264_slice_header_init( x264_t *h, x264_slice_header_t *sh,
99                                     x264_sps_t *sps, x264_pps_t *pps,
100                                     int i_idr_pic_id, int i_frame, int i_qp )
101 {
102     x264_param_t *param = &h->param;
103
104     /* First we fill all fields */
105     sh->sps = sps;
106     sh->pps = pps;
107
108     sh->i_first_mb  = 0;
109     sh->i_last_mb   = h->mb.i_mb_count - 1;
110     sh->i_pps_id    = pps->i_id;
111
112     sh->i_frame_num = i_frame;
113
114     sh->b_mbaff = PARAM_INTERLACED;
115     sh->b_field_pic = 0;    /* no field support for now */
116     sh->b_bottom_field = 0; /* not yet used */
117
118     sh->i_idr_pic_id = i_idr_pic_id;
119
120     /* poc stuff, fixed later */
121     sh->i_poc = 0;
122     sh->i_delta_poc_bottom = 0;
123     sh->i_delta_poc[0] = 0;
124     sh->i_delta_poc[1] = 0;
125
126     sh->i_redundant_pic_cnt = 0;
127
128     h->mb.b_direct_auto_write = h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO
129                                 && h->param.i_bframe
130                                 && ( h->param.rc.b_stat_write || !h->param.rc.b_stat_read );
131
132     if( !h->mb.b_direct_auto_read && sh->i_type == SLICE_TYPE_B )
133     {
134         if( h->fref[1][0]->i_poc_l0ref0 == h->fref[0][0]->i_poc )
135         {
136             if( h->mb.b_direct_auto_write )
137                 sh->b_direct_spatial_mv_pred = ( h->stat.i_direct_score[1] > h->stat.i_direct_score[0] );
138             else
139                 sh->b_direct_spatial_mv_pred = ( param->analyse.i_direct_mv_pred == X264_DIRECT_PRED_SPATIAL );
140         }
141         else
142         {
143             h->mb.b_direct_auto_write = 0;
144             sh->b_direct_spatial_mv_pred = 1;
145         }
146     }
147     /* else b_direct_spatial_mv_pred was read from the 2pass statsfile */
148
149     sh->b_num_ref_idx_override = 0;
150     sh->i_num_ref_idx_l0_active = 1;
151     sh->i_num_ref_idx_l1_active = 1;
152
153     sh->b_ref_pic_list_reordering[0] = h->b_ref_reorder[0];
154     sh->b_ref_pic_list_reordering[1] = h->b_ref_reorder[1];
155
156     /* If the ref list isn't in the default order, construct reordering header */
157     for( int list = 0; list < 2; list++ )
158     {
159         if( sh->b_ref_pic_list_reordering[list] )
160         {
161             int pred_frame_num = i_frame;
162             for( int i = 0; i < h->i_ref[list]; i++ )
163             {
164                 int diff = h->fref[list][i]->i_frame_num - pred_frame_num;
165                 sh->ref_pic_list_order[list][i].idc = ( diff > 0 );
166                 sh->ref_pic_list_order[list][i].arg = (abs(diff) - 1) & ((1 << sps->i_log2_max_frame_num) - 1);
167                 pred_frame_num = h->fref[list][i]->i_frame_num;
168             }
169         }
170     }
171
172     sh->i_cabac_init_idc = param->i_cabac_init_idc;
173
174     sh->i_qp = SPEC_QP(i_qp);
175     sh->i_qp_delta = sh->i_qp - pps->i_pic_init_qp;
176     sh->b_sp_for_swidth = 0;
177     sh->i_qs_delta = 0;
178
179     int deblock_thresh = i_qp + 2 * X264_MIN(param->i_deblocking_filter_alphac0, param->i_deblocking_filter_beta);
180     /* If effective qp <= 15, deblocking would have no effect anyway */
181     if( param->b_deblocking_filter && (h->mb.b_variable_qp || 15 < deblock_thresh ) )
182         sh->i_disable_deblocking_filter_idc = param->b_sliced_threads ? 2 : 0;
183     else
184         sh->i_disable_deblocking_filter_idc = 1;
185     sh->i_alpha_c0_offset = param->i_deblocking_filter_alphac0 << 1;
186     sh->i_beta_offset = param->i_deblocking_filter_beta << 1;
187 }
188
189 static void x264_slice_header_write( bs_t *s, x264_slice_header_t *sh, int i_nal_ref_idc )
190 {
191     if( sh->b_mbaff )
192     {
193         int first_x = sh->i_first_mb % sh->sps->i_mb_width;
194         int first_y = sh->i_first_mb / sh->sps->i_mb_width;
195         assert( (first_y&1) == 0 );
196         bs_write_ue( s, (2*first_x + sh->sps->i_mb_width*(first_y&~1) + (first_y&1)) >> 1 );
197     }
198     else
199         bs_write_ue( s, sh->i_first_mb );
200
201     bs_write_ue( s, sh->i_type + 5 );   /* same type things */
202     bs_write_ue( s, sh->i_pps_id );
203     bs_write( s, sh->sps->i_log2_max_frame_num, sh->i_frame_num & ((1<<sh->sps->i_log2_max_frame_num)-1) );
204
205     if( !sh->sps->b_frame_mbs_only )
206     {
207         bs_write1( s, sh->b_field_pic );
208         if( sh->b_field_pic )
209             bs_write1( s, sh->b_bottom_field );
210     }
211
212     if( sh->i_idr_pic_id >= 0 ) /* NAL IDR */
213         bs_write_ue( s, sh->i_idr_pic_id );
214
215     if( sh->sps->i_poc_type == 0 )
216     {
217         bs_write( s, sh->sps->i_log2_max_poc_lsb, sh->i_poc & ((1<<sh->sps->i_log2_max_poc_lsb)-1) );
218         if( sh->pps->b_pic_order && !sh->b_field_pic )
219             bs_write_se( s, sh->i_delta_poc_bottom );
220     }
221
222     if( sh->pps->b_redundant_pic_cnt )
223         bs_write_ue( s, sh->i_redundant_pic_cnt );
224
225     if( sh->i_type == SLICE_TYPE_B )
226         bs_write1( s, sh->b_direct_spatial_mv_pred );
227
228     if( sh->i_type == SLICE_TYPE_P || sh->i_type == SLICE_TYPE_B )
229     {
230         bs_write1( s, sh->b_num_ref_idx_override );
231         if( sh->b_num_ref_idx_override )
232         {
233             bs_write_ue( s, sh->i_num_ref_idx_l0_active - 1 );
234             if( sh->i_type == SLICE_TYPE_B )
235                 bs_write_ue( s, sh->i_num_ref_idx_l1_active - 1 );
236         }
237     }
238
239     /* ref pic list reordering */
240     if( sh->i_type != SLICE_TYPE_I )
241     {
242         bs_write1( s, sh->b_ref_pic_list_reordering[0] );
243         if( sh->b_ref_pic_list_reordering[0] )
244         {
245             for( int i = 0; i < sh->i_num_ref_idx_l0_active; i++ )
246             {
247                 bs_write_ue( s, sh->ref_pic_list_order[0][i].idc );
248                 bs_write_ue( s, sh->ref_pic_list_order[0][i].arg );
249             }
250             bs_write_ue( s, 3 );
251         }
252     }
253     if( sh->i_type == SLICE_TYPE_B )
254     {
255         bs_write1( s, sh->b_ref_pic_list_reordering[1] );
256         if( sh->b_ref_pic_list_reordering[1] )
257         {
258             for( int i = 0; i < sh->i_num_ref_idx_l1_active; i++ )
259             {
260                 bs_write_ue( s, sh->ref_pic_list_order[1][i].idc );
261                 bs_write_ue( s, sh->ref_pic_list_order[1][i].arg );
262             }
263             bs_write_ue( s, 3 );
264         }
265     }
266
267     if( sh->pps->b_weighted_pred && sh->i_type == SLICE_TYPE_P )
268     {
269         /* pred_weight_table() */
270         bs_write_ue( s, sh->weight[0][0].i_denom );
271         bs_write_ue( s, sh->weight[0][1].i_denom );
272         for( int i = 0; i < sh->i_num_ref_idx_l0_active; i++ )
273         {
274             int luma_weight_l0_flag = !!sh->weight[i][0].weightfn;
275             int chroma_weight_l0_flag = !!sh->weight[i][1].weightfn || !!sh->weight[i][2].weightfn;
276             bs_write1( s, luma_weight_l0_flag );
277             if( luma_weight_l0_flag )
278             {
279                 bs_write_se( s, sh->weight[i][0].i_scale );
280                 bs_write_se( s, sh->weight[i][0].i_offset );
281             }
282             bs_write1( s, chroma_weight_l0_flag );
283             if( chroma_weight_l0_flag )
284             {
285                 for( int j = 1; j < 3; j++ )
286                 {
287                     bs_write_se( s, sh->weight[i][j].i_scale );
288                     bs_write_se( s, sh->weight[i][j].i_offset );
289                 }
290             }
291         }
292     }
293     else if( sh->pps->b_weighted_bipred == 1 && sh->i_type == SLICE_TYPE_B )
294     {
295       /* TODO */
296     }
297
298     if( i_nal_ref_idc != 0 )
299     {
300         if( sh->i_idr_pic_id >= 0 )
301         {
302             bs_write1( s, 0 );  /* no output of prior pics flag */
303             bs_write1( s, 0 );  /* long term reference flag */
304         }
305         else
306         {
307             bs_write1( s, sh->i_mmco_command_count > 0 ); /* adaptive_ref_pic_marking_mode_flag */
308             if( sh->i_mmco_command_count > 0 )
309             {
310                 for( int i = 0; i < sh->i_mmco_command_count; i++ )
311                 {
312                     bs_write_ue( s, 1 ); /* mark short term ref as unused */
313                     bs_write_ue( s, sh->mmco[i].i_difference_of_pic_nums - 1 );
314                 }
315                 bs_write_ue( s, 0 ); /* end command list */
316             }
317         }
318     }
319
320     if( sh->pps->b_cabac && sh->i_type != SLICE_TYPE_I )
321         bs_write_ue( s, sh->i_cabac_init_idc );
322
323     bs_write_se( s, sh->i_qp_delta );      /* slice qp delta */
324
325     if( sh->pps->b_deblocking_filter_control )
326     {
327         bs_write_ue( s, sh->i_disable_deblocking_filter_idc );
328         if( sh->i_disable_deblocking_filter_idc != 1 )
329         {
330             bs_write_se( s, sh->i_alpha_c0_offset >> 1 );
331             bs_write_se( s, sh->i_beta_offset >> 1 );
332         }
333     }
334 }
335
336 /* If we are within a reasonable distance of the end of the memory allocated for the bitstream, */
337 /* reallocate, adding an arbitrary amount of space (100 kilobytes). */
338 static int x264_bitstream_check_buffer( x264_t *h )
339 {
340     uint8_t *bs_bak = h->out.p_bitstream;
341     int max_mb_size = 2500 << SLICE_MBAFF;
342     if( (h->param.b_cabac && (h->cabac.p_end - h->cabac.p < max_mb_size)) ||
343         (h->out.bs.p_end - h->out.bs.p < max_mb_size) )
344     {
345         h->out.i_bitstream += 100000;
346         CHECKED_MALLOC( h->out.p_bitstream, h->out.i_bitstream );
347         h->mc.memcpy_aligned( h->out.p_bitstream, bs_bak, (h->out.i_bitstream - 100000) & ~15 );
348         intptr_t delta = h->out.p_bitstream - bs_bak;
349
350         h->out.bs.p_start += delta;
351         h->out.bs.p += delta;
352         h->out.bs.p_end = h->out.p_bitstream + h->out.i_bitstream;
353
354         h->cabac.p_start += delta;
355         h->cabac.p += delta;
356         h->cabac.p_end = h->out.p_bitstream + h->out.i_bitstream;
357
358         for( int i = 0; i <= h->out.i_nal; i++ )
359             h->out.nal[i].p_payload += delta;
360         x264_free( bs_bak );
361     }
362     return 0;
363 fail:
364     x264_free( bs_bak );
365     return -1;
366 }
367
368 #if HAVE_THREAD
369 static void x264_encoder_thread_init( x264_t *h )
370 {
371     if( h->param.i_sync_lookahead )
372         x264_lower_thread_priority( 10 );
373
374 #if HAVE_MMX
375     /* Misalign mask has to be set separately for each thread. */
376     if( h->param.cpu&X264_CPU_SSE_MISALIGN )
377         x264_cpu_mask_misalign_sse();
378 #endif
379 }
380 #endif
381
382 /****************************************************************************
383  *
384  ****************************************************************************
385  ****************************** External API*********************************
386  ****************************************************************************
387  *
388  ****************************************************************************/
389
390 static int x264_validate_parameters( x264_t *h, int b_open )
391 {
392 #if HAVE_MMX
393 #ifdef __SSE__
394     if( b_open && !(x264_cpu_detect() & X264_CPU_SSE) )
395     {
396         x264_log( h, X264_LOG_ERROR, "your cpu does not support SSE1, but x264 was compiled with asm support\n");
397 #else
398     if( b_open && !(x264_cpu_detect() & X264_CPU_MMX2) )
399     {
400         x264_log( h, X264_LOG_ERROR, "your cpu does not support MMXEXT, but x264 was compiled with asm support\n");
401 #endif
402         x264_log( h, X264_LOG_ERROR, "to run x264, recompile without asm support (configure --disable-asm)\n");
403         return -1;
404     }
405 #endif
406
407 #if HAVE_INTERLACED
408     h->param.b_interlaced = !!PARAM_INTERLACED;
409 #else
410     if( h->param.b_interlaced )
411     {
412         x264_log( h, X264_LOG_ERROR, "not compiled with interlaced support\n" );
413         return -1;
414     }
415 #endif
416
417     if( h->param.i_width <= 0 || h->param.i_height <= 0 )
418     {
419         x264_log( h, X264_LOG_ERROR, "invalid width x height (%dx%d)\n",
420                   h->param.i_width, h->param.i_height );
421         return -1;
422     }
423
424     int i_csp = h->param.i_csp & X264_CSP_MASK;
425 #if X264_CHROMA_FORMAT
426     if( CHROMA_FORMAT != CHROMA_420 && i_csp >= X264_CSP_I420 && i_csp <= X264_CSP_NV12 )
427     {
428         x264_log( h, X264_LOG_ERROR, "not compiled with 4:2:0 support\n" );
429         return -1;
430     }
431     else if( CHROMA_FORMAT != CHROMA_422 && i_csp >= X264_CSP_I422 && i_csp <= X264_CSP_NV16 )
432     {
433         x264_log( h, X264_LOG_ERROR, "not compiled with 4:2:2 support\n" );
434         return -1;
435     }
436     else if( CHROMA_FORMAT != CHROMA_444 && i_csp >= X264_CSP_I444 && i_csp <= X264_CSP_RGB )
437     {
438         x264_log( h, X264_LOG_ERROR, "not compiled with 4:4:4 support\n" );
439         return -1;
440     }
441 #endif
442     if( i_csp <= X264_CSP_NONE || i_csp >= X264_CSP_MAX )
443     {
444         x264_log( h, X264_LOG_ERROR, "invalid CSP (only I420/YV12/NV12/I422/YV16/NV16/I444/YV24/BGR/BGRA/RGB supported)\n" );
445         return -1;
446     }
447
448     if( i_csp < X264_CSP_I444 && h->param.i_width % 2 )
449     {
450         x264_log( h, X264_LOG_ERROR, "width not divisible by 2 (%dx%d)\n",
451                   h->param.i_width, h->param.i_height );
452         return -1;
453     }
454
455     if( i_csp < X264_CSP_I422 && PARAM_INTERLACED && h->param.i_height % 4 )
456     {
457         x264_log( h, X264_LOG_ERROR, "height not divisible by 4 (%dx%d)\n",
458                   h->param.i_width, h->param.i_height );
459         return -1;
460     }
461
462     if( (i_csp < X264_CSP_I422 || PARAM_INTERLACED) && h->param.i_height % 2 )
463     {
464         x264_log( h, X264_LOG_ERROR, "height not divisible by 2 (%dx%d)\n",
465                   h->param.i_width, h->param.i_height );
466         return -1;
467     }
468
469     if( (h->param.crop_rect.i_left + h->param.crop_rect.i_right ) >= h->param.i_width ||
470         (h->param.crop_rect.i_top  + h->param.crop_rect.i_bottom) >= h->param.i_height )
471     {
472         x264_log( h, X264_LOG_ERROR, "invalid crop-rect %u,%u,%u,%u\n", h->param.crop_rect.i_left,
473                   h->param.crop_rect.i_top, h->param.crop_rect.i_right,  h->param.crop_rect.i_bottom );
474         return -1;
475     }
476
477     if( h->param.i_threads == X264_THREADS_AUTO )
478         h->param.i_threads = x264_cpu_num_processors() * (h->param.b_sliced_threads?2:3)/2;
479     if( h->param.i_threads > 1 )
480     {
481 #if !HAVE_THREAD
482         x264_log( h, X264_LOG_WARNING, "not compiled with thread support!\n");
483         h->param.i_threads = 1;
484 #endif
485         /* Avoid absurdly small thread slices as they can reduce performance
486          * and VBV compliance.  Capped at an arbitrary 4 rows per thread. */
487         if( h->param.b_sliced_threads )
488         {
489             int max_threads = (h->param.i_height+15)/16 / 4;
490             h->param.i_threads = X264_MIN( h->param.i_threads, max_threads );
491         }
492     }
493     h->param.i_threads = x264_clip3( h->param.i_threads, 1, X264_THREAD_MAX );
494     if( h->param.i_threads == 1 )
495         h->param.b_sliced_threads = 0;
496     h->i_thread_frames = h->param.b_sliced_threads ? 1 : h->param.i_threads;
497     if( h->i_thread_frames > 1 )
498         h->param.nalu_process = NULL;
499
500     h->param.i_keyint_max = x264_clip3( h->param.i_keyint_max, 1, X264_KEYINT_MAX_INFINITE );
501     if( h->param.i_keyint_max == 1 )
502     {
503         h->param.b_intra_refresh = 0;
504         h->param.analyse.i_weighted_pred = 0;
505     }
506
507     h->param.i_frame_packing = x264_clip3( h->param.i_frame_packing, -1, 5 );
508
509     /* Detect default ffmpeg settings and terminate with an error. */
510     if( b_open )
511     {
512         int score = 0;
513         score += h->param.analyse.i_me_range == 0;
514         score += h->param.rc.i_qp_step == 3;
515         score += h->param.i_keyint_max == 12;
516         score += h->param.rc.i_qp_min == 2;
517         score += h->param.rc.i_qp_max == 31;
518         score += h->param.rc.f_qcompress == 0.5;
519         score += fabs(h->param.rc.f_ip_factor - 1.25) < 0.01;
520         score += fabs(h->param.rc.f_pb_factor - 1.25) < 0.01;
521         score += h->param.analyse.inter == 0 && h->param.analyse.i_subpel_refine == 8;
522         if( score >= 5 )
523         {
524             x264_log( h, X264_LOG_ERROR, "broken ffmpeg default settings detected\n" );
525             x264_log( h, X264_LOG_ERROR, "use an encoding preset (e.g. -vpre medium)\n" );
526             x264_log( h, X264_LOG_ERROR, "preset usage: -vpre <speed> -vpre <profile>\n" );
527             x264_log( h, X264_LOG_ERROR, "speed presets are listed in x264 --help\n" );
528             x264_log( h, X264_LOG_ERROR, "profile is optional; x264 defaults to high\n" );
529             return -1;
530         }
531     }
532
533     if( h->param.rc.i_rc_method < 0 || h->param.rc.i_rc_method > 2 )
534     {
535         x264_log( h, X264_LOG_ERROR, "no ratecontrol method specified\n" );
536         return -1;
537     }
538     h->param.rc.f_rf_constant = x264_clip3f( h->param.rc.f_rf_constant, -QP_BD_OFFSET, 51 );
539     h->param.rc.f_rf_constant_max = x264_clip3f( h->param.rc.f_rf_constant_max, -QP_BD_OFFSET, 51 );
540     h->param.rc.i_qp_constant = x264_clip3( h->param.rc.i_qp_constant, 0, QP_MAX );
541     h->param.analyse.i_subpel_refine = x264_clip3( h->param.analyse.i_subpel_refine, 0, 11 );
542     h->param.rc.f_ip_factor = X264_MAX( h->param.rc.f_ip_factor, 0.01f );
543     h->param.rc.f_pb_factor = X264_MAX( h->param.rc.f_pb_factor, 0.01f );
544     if( h->param.rc.i_rc_method == X264_RC_CRF )
545     {
546         h->param.rc.i_qp_constant = h->param.rc.f_rf_constant + QP_BD_OFFSET;
547         h->param.rc.i_bitrate = 0;
548     }
549     if( (h->param.rc.i_rc_method == X264_RC_CQP || h->param.rc.i_rc_method == X264_RC_CRF)
550         && h->param.rc.i_qp_constant == 0 )
551     {
552         h->mb.b_lossless = 1;
553         h->param.i_cqm_preset = X264_CQM_FLAT;
554         h->param.psz_cqm_file = NULL;
555         h->param.rc.i_rc_method = X264_RC_CQP;
556         h->param.rc.f_ip_factor = 1;
557         h->param.rc.f_pb_factor = 1;
558         h->param.analyse.b_psnr = 0;
559         h->param.analyse.b_ssim = 0;
560         h->param.analyse.i_chroma_qp_offset = 0;
561         h->param.analyse.i_trellis = 0;
562         h->param.analyse.b_fast_pskip = 0;
563         h->param.analyse.i_noise_reduction = 0;
564         h->param.analyse.b_psy = 0;
565         h->param.i_bframe = 0;
566         /* 8x8dct is not useful without RD in CAVLC lossless */
567         if( !h->param.b_cabac && h->param.analyse.i_subpel_refine < 6 )
568             h->param.analyse.b_transform_8x8 = 0;
569     }
570     if( h->param.rc.i_rc_method == X264_RC_CQP )
571     {
572         float qp_p = h->param.rc.i_qp_constant;
573         float qp_i = qp_p - 6*log2f( h->param.rc.f_ip_factor );
574         float qp_b = qp_p + 6*log2f( h->param.rc.f_pb_factor );
575         h->param.rc.i_qp_min = x264_clip3( (int)(X264_MIN3( qp_p, qp_i, qp_b )), 0, QP_MAX );
576         h->param.rc.i_qp_max = x264_clip3( (int)(X264_MAX3( qp_p, qp_i, qp_b ) + .999), 0, QP_MAX );
577         h->param.rc.i_aq_mode = 0;
578         h->param.rc.b_mb_tree = 0;
579         h->param.rc.i_bitrate = 0;
580     }
581     h->param.rc.i_qp_max = x264_clip3( h->param.rc.i_qp_max, 0, QP_MAX );
582     h->param.rc.i_qp_min = x264_clip3( h->param.rc.i_qp_min, 0, h->param.rc.i_qp_max );
583     h->param.rc.i_qp_step = x264_clip3( h->param.rc.i_qp_step, 0, QP_MAX );
584     h->param.rc.i_bitrate = x264_clip3( h->param.rc.i_bitrate, 0, 2000000 );
585     h->param.rc.i_vbv_buffer_size = x264_clip3( h->param.rc.i_vbv_buffer_size, 0, 2000000 );
586     h->param.rc.i_vbv_max_bitrate = x264_clip3( h->param.rc.i_vbv_max_bitrate, 0, 2000000 );
587     h->param.rc.f_vbv_buffer_init = x264_clip3f( h->param.rc.f_vbv_buffer_init, 0, 2000000 );
588     if( h->param.rc.i_vbv_buffer_size )
589     {
590         if( h->param.rc.i_rc_method == X264_RC_CQP )
591         {
592             x264_log( h, X264_LOG_WARNING, "VBV is incompatible with constant QP, ignored.\n" );
593             h->param.rc.i_vbv_max_bitrate = 0;
594             h->param.rc.i_vbv_buffer_size = 0;
595         }
596         else if( h->param.rc.i_vbv_max_bitrate == 0 )
597         {
598             if( h->param.rc.i_rc_method == X264_RC_ABR )
599             {
600                 x264_log( h, X264_LOG_WARNING, "VBV maxrate unspecified, assuming CBR\n" );
601                 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate;
602             }
603             else
604             {
605                 x264_log( h, X264_LOG_WARNING, "VBV bufsize set but maxrate unspecified, ignored\n" );
606                 h->param.rc.i_vbv_buffer_size = 0;
607             }
608         }
609         else if( h->param.rc.i_vbv_max_bitrate < h->param.rc.i_bitrate &&
610                  h->param.rc.i_rc_method == X264_RC_ABR )
611         {
612             x264_log( h, X264_LOG_WARNING, "max bitrate less than average bitrate, assuming CBR\n" );
613             h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate;
614         }
615     }
616     else if( h->param.rc.i_vbv_max_bitrate )
617     {
618         x264_log( h, X264_LOG_WARNING, "VBV maxrate specified, but no bufsize, ignored\n" );
619         h->param.rc.i_vbv_max_bitrate = 0;
620     }
621
622     h->param.i_slice_max_size = X264_MAX( h->param.i_slice_max_size, 0 );
623     h->param.i_slice_max_mbs = X264_MAX( h->param.i_slice_max_mbs, 0 );
624
625     int max_slices = (h->param.i_height+((16<<PARAM_INTERLACED)-1))/(16<<PARAM_INTERLACED);
626     if( h->param.b_sliced_threads )
627         h->param.i_slice_count = x264_clip3( h->param.i_threads, 0, max_slices );
628     else
629     {
630         h->param.i_slice_count = x264_clip3( h->param.i_slice_count, 0, max_slices );
631         if( h->param.i_slice_max_mbs || h->param.i_slice_max_size )
632             h->param.i_slice_count = 0;
633     }
634
635     if( h->param.b_bluray_compat )
636     {
637         h->param.i_bframe_pyramid = X264_MIN( X264_B_PYRAMID_STRICT, h->param.i_bframe_pyramid );
638         h->param.i_bframe = X264_MIN( h->param.i_bframe, 3 );
639         h->param.b_aud = 1;
640         h->param.i_nal_hrd = X264_MAX( h->param.i_nal_hrd, X264_NAL_HRD_VBR );
641         h->param.i_slice_max_size = 0;
642         h->param.i_slice_max_mbs = 0;
643         h->param.b_intra_refresh = 0;
644         h->param.i_frame_reference = X264_MIN( h->param.i_frame_reference, 6 );
645         h->param.i_dpb_size = X264_MIN( h->param.i_dpb_size, 6 );
646         /* Don't use I-frames, because Blu-ray treats them the same as IDR. */
647         h->param.i_keyint_min = 1;
648         /* Due to the proliferation of broken players that don't handle dupes properly. */
649         h->param.analyse.i_weighted_pred = X264_MIN( h->param.analyse.i_weighted_pred, X264_WEIGHTP_SIMPLE );
650         if( h->param.b_fake_interlaced )
651             h->param.b_pic_struct = 1;
652     }
653
654     h->param.i_frame_reference = x264_clip3( h->param.i_frame_reference, 1, X264_REF_MAX );
655     h->param.i_dpb_size = x264_clip3( h->param.i_dpb_size, 1, X264_REF_MAX );
656     if( h->param.i_scenecut_threshold < 0 )
657         h->param.i_scenecut_threshold = 0;
658     h->param.analyse.i_direct_mv_pred = x264_clip3( h->param.analyse.i_direct_mv_pred, X264_DIRECT_PRED_NONE, X264_DIRECT_PRED_AUTO );
659     if( !h->param.analyse.i_subpel_refine && h->param.analyse.i_direct_mv_pred > X264_DIRECT_PRED_SPATIAL )
660     {
661         x264_log( h, X264_LOG_WARNING, "subme=0 + direct=temporal is not supported\n" );
662         h->param.analyse.i_direct_mv_pred = X264_DIRECT_PRED_SPATIAL;
663     }
664     h->param.i_bframe = x264_clip3( h->param.i_bframe, 0, X264_MIN( X264_BFRAME_MAX, h->param.i_keyint_max-1 ) );
665     h->param.i_bframe_bias = x264_clip3( h->param.i_bframe_bias, -90, 100 );
666     if( h->param.i_bframe <= 1 )
667         h->param.i_bframe_pyramid = X264_B_PYRAMID_NONE;
668     h->param.i_bframe_pyramid = x264_clip3( h->param.i_bframe_pyramid, X264_B_PYRAMID_NONE, X264_B_PYRAMID_NORMAL );
669     h->param.i_bframe_adaptive = x264_clip3( h->param.i_bframe_adaptive, X264_B_ADAPT_NONE, X264_B_ADAPT_TRELLIS );
670     if( !h->param.i_bframe )
671     {
672         h->param.i_bframe_adaptive = X264_B_ADAPT_NONE;
673         h->param.analyse.i_direct_mv_pred = 0;
674         h->param.analyse.b_weighted_bipred = 0;
675         h->param.b_open_gop = 0;
676     }
677     if( h->param.b_intra_refresh && h->param.i_bframe_pyramid == X264_B_PYRAMID_NORMAL )
678     {
679         x264_log( h, X264_LOG_WARNING, "b-pyramid normal + intra-refresh is not supported\n" );
680         h->param.i_bframe_pyramid = X264_B_PYRAMID_STRICT;
681     }
682     if( h->param.b_intra_refresh && (h->param.i_frame_reference > 1 || h->param.i_dpb_size > 1) )
683     {
684         x264_log( h, X264_LOG_WARNING, "ref > 1 + intra-refresh is not supported\n" );
685         h->param.i_frame_reference = 1;
686         h->param.i_dpb_size = 1;
687     }
688     if( h->param.b_intra_refresh && h->param.b_open_gop )
689     {
690         x264_log( h, X264_LOG_WARNING, "intra-refresh is not compatible with open-gop\n" );
691         h->param.b_open_gop = 0;
692     }
693     float fps = h->param.i_fps_num > 0 && h->param.i_fps_den > 0 ? (float) h->param.i_fps_num / h->param.i_fps_den : 25.0;
694     if( h->param.i_keyint_min == X264_KEYINT_MIN_AUTO )
695         h->param.i_keyint_min = X264_MIN( h->param.i_keyint_max / 10, fps );
696     h->param.i_keyint_min = x264_clip3( h->param.i_keyint_min, 1, h->param.i_keyint_max/2+1 );
697     h->param.rc.i_lookahead = x264_clip3( h->param.rc.i_lookahead, 0, X264_LOOKAHEAD_MAX );
698     {
699         int maxrate = X264_MAX( h->param.rc.i_vbv_max_bitrate, h->param.rc.i_bitrate );
700         float bufsize = maxrate ? (float)h->param.rc.i_vbv_buffer_size / maxrate : 0;
701         h->param.rc.i_lookahead = X264_MIN( h->param.rc.i_lookahead, X264_MAX( h->param.i_keyint_max, bufsize*fps ) );
702     }
703
704     if( !h->param.i_timebase_num || !h->param.i_timebase_den || !(h->param.b_vfr_input || h->param.b_pulldown) )
705     {
706         h->param.i_timebase_num = h->param.i_fps_den;
707         h->param.i_timebase_den = h->param.i_fps_num;
708     }
709
710     h->param.rc.f_qcompress = x264_clip3f( h->param.rc.f_qcompress, 0.0, 1.0 );
711     if( h->param.i_keyint_max == 1 || h->param.rc.f_qcompress == 1 )
712         h->param.rc.b_mb_tree = 0;
713     if( (!h->param.b_intra_refresh && h->param.i_keyint_max != X264_KEYINT_MAX_INFINITE) &&
714         !h->param.rc.i_lookahead && h->param.rc.b_mb_tree )
715     {
716         x264_log( h, X264_LOG_WARNING, "lookaheadless mb-tree requires intra refresh or infinite keyint\n" );
717         h->param.rc.b_mb_tree = 0;
718     }
719     if( h->param.rc.b_stat_read )
720         h->param.rc.i_lookahead = 0;
721 #if HAVE_THREAD
722     if( h->param.i_sync_lookahead < 0 )
723         h->param.i_sync_lookahead = h->param.i_bframe + 1;
724     h->param.i_sync_lookahead = X264_MIN( h->param.i_sync_lookahead, X264_LOOKAHEAD_MAX );
725     if( h->param.rc.b_stat_read || h->i_thread_frames == 1 )
726         h->param.i_sync_lookahead = 0;
727 #else
728     h->param.i_sync_lookahead = 0;
729 #endif
730
731     h->param.i_deblocking_filter_alphac0 = x264_clip3( h->param.i_deblocking_filter_alphac0, -6, 6 );
732     h->param.i_deblocking_filter_beta    = x264_clip3( h->param.i_deblocking_filter_beta, -6, 6 );
733     h->param.analyse.i_luma_deadzone[0] = x264_clip3( h->param.analyse.i_luma_deadzone[0], 0, 32 );
734     h->param.analyse.i_luma_deadzone[1] = x264_clip3( h->param.analyse.i_luma_deadzone[1], 0, 32 );
735
736     h->param.i_cabac_init_idc = x264_clip3( h->param.i_cabac_init_idc, 0, 2 );
737
738     if( h->param.i_cqm_preset < X264_CQM_FLAT || h->param.i_cqm_preset > X264_CQM_CUSTOM )
739         h->param.i_cqm_preset = X264_CQM_FLAT;
740
741     if( h->param.analyse.i_me_method < X264_ME_DIA ||
742         h->param.analyse.i_me_method > X264_ME_TESA )
743         h->param.analyse.i_me_method = X264_ME_HEX;
744     h->param.analyse.i_me_range = x264_clip3( h->param.analyse.i_me_range, 4, 1024 );
745     if( h->param.analyse.i_me_range > 16 && h->param.analyse.i_me_method <= X264_ME_HEX )
746         h->param.analyse.i_me_range = 16;
747     if( h->param.analyse.i_me_method == X264_ME_TESA &&
748         (h->mb.b_lossless || h->param.analyse.i_subpel_refine <= 1) )
749         h->param.analyse.i_me_method = X264_ME_ESA;
750     h->param.analyse.b_mixed_references = h->param.analyse.b_mixed_references && h->param.i_frame_reference > 1;
751     h->param.analyse.inter &= X264_ANALYSE_PSUB16x16|X264_ANALYSE_PSUB8x8|X264_ANALYSE_BSUB16x16|
752                               X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
753     h->param.analyse.intra &= X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
754     if( !(h->param.analyse.inter & X264_ANALYSE_PSUB16x16) )
755         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
756     if( !h->param.analyse.b_transform_8x8 )
757     {
758         h->param.analyse.inter &= ~X264_ANALYSE_I8x8;
759         h->param.analyse.intra &= ~X264_ANALYSE_I8x8;
760     }
761     h->param.analyse.i_trellis = x264_clip3( h->param.analyse.i_trellis, 0, 2 );
762     h->param.rc.i_aq_mode = x264_clip3( h->param.rc.i_aq_mode, 0, 2 );
763     h->param.rc.f_aq_strength = x264_clip3f( h->param.rc.f_aq_strength, 0, 3 );
764     if( h->param.rc.f_aq_strength == 0 )
765         h->param.rc.i_aq_mode = 0;
766
767     if( h->param.i_log_level < X264_LOG_INFO )
768     {
769         h->param.analyse.b_psnr = 0;
770         h->param.analyse.b_ssim = 0;
771     }
772     /* Warn users trying to measure PSNR/SSIM with psy opts on. */
773     if( b_open && (h->param.analyse.b_psnr || h->param.analyse.b_ssim) )
774     {
775         char *s = NULL;
776
777         if( h->param.analyse.b_psy )
778         {
779             s = h->param.analyse.b_psnr ? "psnr" : "ssim";
780             x264_log( h, X264_LOG_WARNING, "--%s used with psy on: results will be invalid!\n", s );
781         }
782         else if( !h->param.rc.i_aq_mode && h->param.analyse.b_ssim )
783         {
784             x264_log( h, X264_LOG_WARNING, "--ssim used with AQ off: results will be invalid!\n" );
785             s = "ssim";
786         }
787         else if(  h->param.rc.i_aq_mode && h->param.analyse.b_psnr )
788         {
789             x264_log( h, X264_LOG_WARNING, "--psnr used with AQ on: results will be invalid!\n" );
790             s = "psnr";
791         }
792         if( s )
793             x264_log( h, X264_LOG_WARNING, "--tune %s should be used if attempting to benchmark %s!\n", s, s );
794     }
795
796     if( !h->param.analyse.b_psy )
797     {
798         h->param.analyse.f_psy_rd = 0;
799         h->param.analyse.f_psy_trellis = 0;
800     }
801     h->param.analyse.f_psy_rd = x264_clip3f( h->param.analyse.f_psy_rd, 0, 10 );
802     h->param.analyse.f_psy_trellis = x264_clip3f( h->param.analyse.f_psy_trellis, 0, 10 );
803     h->mb.i_psy_rd = h->param.analyse.i_subpel_refine >= 6 ? FIX8( h->param.analyse.f_psy_rd ) : 0;
804     h->mb.i_psy_trellis = h->param.analyse.i_trellis ? FIX8( h->param.analyse.f_psy_trellis / 4 ) : 0;
805     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -32, 32);
806     /* In 4:4:4 mode, chroma gets twice as much resolution, so we can halve its quality. */
807     if( b_open && i_csp >= X264_CSP_I444 && i_csp < X264_CSP_BGR && h->param.analyse.b_psy )
808         h->param.analyse.i_chroma_qp_offset += 6;
809     /* Psy RDO increases overall quantizers to improve the quality of luma--this indirectly hurts chroma quality */
810     /* so we lower the chroma QP offset to compensate */
811     if( b_open && h->mb.i_psy_rd )
812         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_rd < 0.25 ? 1 : 2;
813     /* Psy trellis has a similar effect. */
814     if( b_open && h->mb.i_psy_trellis )
815         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_trellis < 0.25 ? 1 : 2;
816     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -12, 12);
817     /* MB-tree requires AQ to be on, even if the strength is zero. */
818     if( !h->param.rc.i_aq_mode && h->param.rc.b_mb_tree )
819     {
820         h->param.rc.i_aq_mode = 1;
821         h->param.rc.f_aq_strength = 0;
822     }
823     h->param.analyse.i_noise_reduction = x264_clip3( h->param.analyse.i_noise_reduction, 0, 1<<16 );
824     if( h->param.analyse.i_subpel_refine >= 10 && (h->param.analyse.i_trellis != 2 || !h->param.rc.i_aq_mode) )
825         h->param.analyse.i_subpel_refine = 9;
826
827     {
828         const x264_level_t *l = x264_levels;
829         if( h->param.i_level_idc < 0 )
830         {
831             int maxrate_bak = h->param.rc.i_vbv_max_bitrate;
832             if( h->param.rc.i_rc_method == X264_RC_ABR && h->param.rc.i_vbv_buffer_size <= 0 )
833                 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate * 2;
834             x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
835             do h->param.i_level_idc = l->level_idc;
836                 while( l[1].level_idc && x264_validate_levels( h, 0 ) && l++ );
837             h->param.rc.i_vbv_max_bitrate = maxrate_bak;
838         }
839         else
840         {
841             while( l->level_idc && l->level_idc != h->param.i_level_idc )
842                 l++;
843             if( l->level_idc == 0 )
844             {
845                 x264_log( h, X264_LOG_ERROR, "invalid level_idc: %d\n", h->param.i_level_idc );
846                 return -1;
847             }
848         }
849         if( h->param.analyse.i_mv_range <= 0 )
850             h->param.analyse.i_mv_range = l->mv_range >> PARAM_INTERLACED;
851         else
852             h->param.analyse.i_mv_range = x264_clip3(h->param.analyse.i_mv_range, 32, 512 >> PARAM_INTERLACED);
853     }
854
855     h->param.analyse.i_weighted_pred = x264_clip3( h->param.analyse.i_weighted_pred, X264_WEIGHTP_NONE, X264_WEIGHTP_SMART );
856
857     if( PARAM_INTERLACED )
858     {
859         if( h->param.analyse.i_me_method >= X264_ME_ESA )
860         {
861             x264_log( h, X264_LOG_WARNING, "interlace + me=esa is not implemented\n" );
862             h->param.analyse.i_me_method = X264_ME_UMH;
863         }
864         if( h->param.analyse.i_weighted_pred > 0 )
865         {
866             x264_log( h, X264_LOG_WARNING, "interlace + weightp is not implemented\n" );
867             h->param.analyse.i_weighted_pred = X264_WEIGHTP_NONE;
868         }
869     }
870
871     if( !h->param.analyse.i_weighted_pred && h->param.rc.b_mb_tree && h->param.analyse.b_psy )
872         h->param.analyse.i_weighted_pred = X264_WEIGHTP_FAKE;
873
874     if( h->i_thread_frames > 1 )
875     {
876         int r = h->param.analyse.i_mv_range_thread;
877         int r2;
878         if( r <= 0 )
879         {
880             // half of the available space is reserved and divided evenly among the threads,
881             // the rest is allocated to whichever thread is far enough ahead to use it.
882             // reserving more space increases quality for some videos, but costs more time
883             // in thread synchronization.
884             int max_range = (h->param.i_height + X264_THREAD_HEIGHT) / h->i_thread_frames - X264_THREAD_HEIGHT;
885             r = max_range / 2;
886         }
887         r = X264_MAX( r, h->param.analyse.i_me_range );
888         r = X264_MIN( r, h->param.analyse.i_mv_range );
889         // round up to use the whole mb row
890         r2 = (r & ~15) + ((-X264_THREAD_HEIGHT) & 15);
891         if( r2 < r )
892             r2 += 16;
893         x264_log( h, X264_LOG_DEBUG, "using mv_range_thread = %d\n", r2 );
894         h->param.analyse.i_mv_range_thread = r2;
895     }
896
897     if( h->param.rc.f_rate_tolerance < 0 )
898         h->param.rc.f_rate_tolerance = 0;
899     if( h->param.rc.f_qblur < 0 )
900         h->param.rc.f_qblur = 0;
901     if( h->param.rc.f_complexity_blur < 0 )
902         h->param.rc.f_complexity_blur = 0;
903
904     h->param.i_sps_id &= 31;
905
906     if( PARAM_INTERLACED )
907         h->param.b_pic_struct = 1;
908
909     h->param.i_nal_hrd = x264_clip3( h->param.i_nal_hrd, X264_NAL_HRD_NONE, X264_NAL_HRD_CBR );
910
911     if( h->param.i_nal_hrd && !h->param.rc.i_vbv_buffer_size )
912     {
913         x264_log( h, X264_LOG_WARNING, "NAL HRD parameters require VBV parameters\n" );
914         h->param.i_nal_hrd = X264_NAL_HRD_NONE;
915     }
916
917     if( h->param.i_nal_hrd == X264_NAL_HRD_CBR &&
918        (h->param.rc.i_bitrate != h->param.rc.i_vbv_max_bitrate || !h->param.rc.i_vbv_max_bitrate) )
919     {
920         x264_log( h, X264_LOG_WARNING, "CBR HRD requires constant bitrate\n" );
921         h->param.i_nal_hrd = X264_NAL_HRD_VBR;
922     }
923
924     /* ensure the booleans are 0 or 1 so they can be used in math */
925 #define BOOLIFY(x) h->param.x = !!h->param.x
926     BOOLIFY( b_cabac );
927     BOOLIFY( b_constrained_intra );
928     BOOLIFY( b_deblocking_filter );
929     BOOLIFY( b_deterministic );
930     BOOLIFY( b_sliced_threads );
931     BOOLIFY( b_interlaced );
932     BOOLIFY( b_intra_refresh );
933     BOOLIFY( b_visualize );
934     BOOLIFY( b_aud );
935     BOOLIFY( b_repeat_headers );
936     BOOLIFY( b_annexb );
937     BOOLIFY( b_vfr_input );
938     BOOLIFY( b_pulldown );
939     BOOLIFY( b_tff );
940     BOOLIFY( b_pic_struct );
941     BOOLIFY( b_fake_interlaced );
942     BOOLIFY( b_open_gop );
943     BOOLIFY( b_bluray_compat );
944     BOOLIFY( analyse.b_transform_8x8 );
945     BOOLIFY( analyse.b_weighted_bipred );
946     BOOLIFY( analyse.b_chroma_me );
947     BOOLIFY( analyse.b_mixed_references );
948     BOOLIFY( analyse.b_fast_pskip );
949     BOOLIFY( analyse.b_dct_decimate );
950     BOOLIFY( analyse.b_psy );
951     BOOLIFY( analyse.b_psnr );
952     BOOLIFY( analyse.b_ssim );
953     BOOLIFY( rc.b_stat_write );
954     BOOLIFY( rc.b_stat_read );
955     BOOLIFY( rc.b_mb_tree );
956 #undef BOOLIFY
957
958     return 0;
959 }
960
961 static void mbcmp_init( x264_t *h )
962 {
963     int satd = !h->mb.b_lossless && h->param.analyse.i_subpel_refine > 1;
964     memcpy( h->pixf.mbcmp, satd ? h->pixf.satd : h->pixf.sad_aligned, sizeof(h->pixf.mbcmp) );
965     memcpy( h->pixf.mbcmp_unaligned, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.mbcmp_unaligned) );
966     h->pixf.intra_mbcmp_x3_16x16 = satd ? h->pixf.intra_satd_x3_16x16 : h->pixf.intra_sad_x3_16x16;
967     h->pixf.intra_mbcmp_x3_8x16c = satd ? h->pixf.intra_satd_x3_8x16c : h->pixf.intra_sad_x3_8x16c;
968     h->pixf.intra_mbcmp_x3_8x8c  = satd ? h->pixf.intra_satd_x3_8x8c  : h->pixf.intra_sad_x3_8x8c;
969     h->pixf.intra_mbcmp_x3_8x8 = satd ? h->pixf.intra_sa8d_x3_8x8 : h->pixf.intra_sad_x3_8x8;
970     h->pixf.intra_mbcmp_x3_4x4 = satd ? h->pixf.intra_satd_x3_4x4 : h->pixf.intra_sad_x3_4x4;
971     h->pixf.intra_mbcmp_x9_4x4 = h->param.b_cpu_independent || h->mb.b_lossless ? NULL
972                                : satd ? h->pixf.intra_satd_x9_4x4 : h->pixf.intra_sad_x9_4x4;
973     h->pixf.intra_mbcmp_x9_8x8 = h->param.b_cpu_independent || h->mb.b_lossless ? NULL
974                                : satd ? h->pixf.intra_sa8d_x9_8x8 : h->pixf.intra_sad_x9_8x8;
975     satd &= h->param.analyse.i_me_method == X264_ME_TESA;
976     memcpy( h->pixf.fpelcmp, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.fpelcmp) );
977     memcpy( h->pixf.fpelcmp_x3, satd ? h->pixf.satd_x3 : h->pixf.sad_x3, sizeof(h->pixf.fpelcmp_x3) );
978     memcpy( h->pixf.fpelcmp_x4, satd ? h->pixf.satd_x4 : h->pixf.sad_x4, sizeof(h->pixf.fpelcmp_x4) );
979 }
980
981 static void chroma_dsp_init( x264_t *h )
982 {
983     memcpy( h->luma2chroma_pixel, x264_luma2chroma_pixel[CHROMA_FORMAT], sizeof(h->luma2chroma_pixel) );
984
985     switch( CHROMA_FORMAT )
986     {
987         case CHROMA_420:
988             memcpy( h->predict_chroma, h->predict_8x8c, sizeof(h->predict_chroma) );
989             h->mc.prefetch_fenc = h->mc.prefetch_fenc_420;
990             h->loopf.deblock_chroma[0] = h->loopf.deblock_h_chroma_420;
991             h->loopf.deblock_chroma_intra[0] = h->loopf.deblock_h_chroma_420_intra;
992             h->loopf.deblock_chroma_mbaff = h->loopf.deblock_chroma_420_mbaff;
993             h->loopf.deblock_chroma_intra_mbaff = h->loopf.deblock_chroma_420_intra_mbaff;
994             h->pixf.intra_mbcmp_x3_chroma = h->pixf.intra_mbcmp_x3_8x8c;
995             h->quantf.coeff_last[DCT_CHROMA_DC] = h->quantf.coeff_last4;
996             h->quantf.coeff_level_run[DCT_CHROMA_DC] = h->quantf.coeff_level_run4;
997             break;
998         case CHROMA_422:
999             memcpy( h->predict_chroma, h->predict_8x16c, sizeof(h->predict_chroma) );
1000             h->mc.prefetch_fenc = h->mc.prefetch_fenc_422;
1001             h->loopf.deblock_chroma[0] = h->loopf.deblock_h_chroma_422;
1002             h->loopf.deblock_chroma_intra[0] = h->loopf.deblock_h_chroma_422_intra;
1003             h->loopf.deblock_chroma_mbaff = h->loopf.deblock_chroma_422_mbaff;
1004             h->loopf.deblock_chroma_intra_mbaff = h->loopf.deblock_chroma_422_intra_mbaff;
1005             h->pixf.intra_mbcmp_x3_chroma = h->pixf.intra_mbcmp_x3_8x16c;
1006             h->quantf.coeff_last[DCT_CHROMA_DC] = h->quantf.coeff_last8;
1007             h->quantf.coeff_level_run[DCT_CHROMA_DC] = h->quantf.coeff_level_run8;
1008             break;
1009         case CHROMA_444:
1010             h->mc.prefetch_fenc = h->mc.prefetch_fenc_422; /* FIXME: doesn't cover V plane */
1011             h->loopf.deblock_chroma_mbaff = h->loopf.deblock_luma_mbaff;
1012             h->loopf.deblock_chroma_intra_mbaff = h->loopf.deblock_luma_intra_mbaff;
1013             break;
1014     }
1015 }
1016
1017 static void x264_set_aspect_ratio( x264_t *h, x264_param_t *param, int initial )
1018 {
1019     /* VUI */
1020     if( param->vui.i_sar_width > 0 && param->vui.i_sar_height > 0 )
1021     {
1022         uint32_t i_w = param->vui.i_sar_width;
1023         uint32_t i_h = param->vui.i_sar_height;
1024         uint32_t old_w = h->param.vui.i_sar_width;
1025         uint32_t old_h = h->param.vui.i_sar_height;
1026
1027         x264_reduce_fraction( &i_w, &i_h );
1028
1029         while( i_w > 65535 || i_h > 65535 )
1030         {
1031             i_w /= 2;
1032             i_h /= 2;
1033         }
1034
1035         x264_reduce_fraction( &i_w, &i_h );
1036
1037         if( i_w != old_w || i_h != old_h || initial )
1038         {
1039             h->param.vui.i_sar_width = 0;
1040             h->param.vui.i_sar_height = 0;
1041             if( i_w == 0 || i_h == 0 )
1042                 x264_log( h, X264_LOG_WARNING, "cannot create valid sample aspect ratio\n" );
1043             else
1044             {
1045                 x264_log( h, initial?X264_LOG_INFO:X264_LOG_DEBUG, "using SAR=%d/%d\n", i_w, i_h );
1046                 h->param.vui.i_sar_width = i_w;
1047                 h->param.vui.i_sar_height = i_h;
1048             }
1049             x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
1050         }
1051     }
1052 }
1053
1054 /****************************************************************************
1055  * x264_encoder_open:
1056  ****************************************************************************/
1057 x264_t *x264_encoder_open( x264_param_t *param )
1058 {
1059     x264_t *h;
1060     char buf[1000], *p;
1061     int qp, i_slicetype_length;
1062
1063     CHECKED_MALLOCZERO( h, sizeof(x264_t) );
1064
1065     /* Create a copy of param */
1066     memcpy( &h->param, param, sizeof(x264_param_t) );
1067
1068     if( param->param_free )
1069         param->param_free( param );
1070
1071     if( x264_threading_init() )
1072     {
1073         x264_log( h, X264_LOG_ERROR, "unable to initialize threading\n" );
1074         goto fail;
1075     }
1076
1077     if( x264_validate_parameters( h, 1 ) < 0 )
1078         goto fail;
1079
1080     if( h->param.psz_cqm_file )
1081         if( x264_cqm_parse_file( h, h->param.psz_cqm_file ) < 0 )
1082             goto fail;
1083
1084     if( h->param.rc.psz_stat_out )
1085         h->param.rc.psz_stat_out = strdup( h->param.rc.psz_stat_out );
1086     if( h->param.rc.psz_stat_in )
1087         h->param.rc.psz_stat_in = strdup( h->param.rc.psz_stat_in );
1088
1089     x264_reduce_fraction( &h->param.i_fps_num, &h->param.i_fps_den );
1090     x264_reduce_fraction( &h->param.i_timebase_num, &h->param.i_timebase_den );
1091
1092     /* Init x264_t */
1093     h->i_frame = -1;
1094     h->i_frame_num = 0;
1095     h->i_idr_pic_id = 0;
1096
1097     if( (uint64_t)h->param.i_timebase_den * 2 > UINT32_MAX )
1098     {
1099         x264_log( h, X264_LOG_ERROR, "Effective timebase denominator %u exceeds H.264 maximum\n", h->param.i_timebase_den );
1100         goto fail;
1101     }
1102
1103     x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
1104     x264_pps_init( h->pps, h->param.i_sps_id, &h->param, h->sps );
1105
1106     x264_set_aspect_ratio( h, &h->param, 1 );
1107
1108     x264_validate_levels( h, 1 );
1109
1110     h->chroma_qp_table = i_chroma_qp_table + 12 + h->pps->i_chroma_qp_index_offset;
1111
1112     if( x264_cqm_init( h ) < 0 )
1113         goto fail;
1114
1115     h->mb.i_mb_width = h->sps->i_mb_width;
1116     h->mb.i_mb_height = h->sps->i_mb_height;
1117     h->mb.i_mb_count = h->mb.i_mb_width * h->mb.i_mb_height;
1118
1119     h->mb.chroma_h_shift = CHROMA_FORMAT == CHROMA_420 || CHROMA_FORMAT == CHROMA_422;
1120     h->mb.chroma_v_shift = CHROMA_FORMAT == CHROMA_420;
1121
1122     /* Adaptive MBAFF and subme 0 are not supported as we require halving motion
1123      * vectors during prediction, resulting in hpel mvs.
1124      * The chosen solution is to make MBAFF non-adaptive in this case. */
1125     h->mb.b_adaptive_mbaff = PARAM_INTERLACED && h->param.analyse.i_subpel_refine;
1126
1127     /* Init frames. */
1128     if( h->param.i_bframe_adaptive == X264_B_ADAPT_TRELLIS && !h->param.rc.b_stat_read )
1129         h->frames.i_delay = X264_MAX(h->param.i_bframe,3)*4;
1130     else
1131         h->frames.i_delay = h->param.i_bframe;
1132     if( h->param.rc.b_mb_tree || h->param.rc.i_vbv_buffer_size )
1133         h->frames.i_delay = X264_MAX( h->frames.i_delay, h->param.rc.i_lookahead );
1134     i_slicetype_length = h->frames.i_delay;
1135     h->frames.i_delay += h->i_thread_frames - 1;
1136     h->frames.i_delay += h->param.i_sync_lookahead;
1137     h->frames.i_delay += h->param.b_vfr_input;
1138     h->frames.i_bframe_delay = h->param.i_bframe ? (h->param.i_bframe_pyramid ? 2 : 1) : 0;
1139
1140     h->frames.i_max_ref0 = h->param.i_frame_reference;
1141     h->frames.i_max_ref1 = X264_MIN( h->sps->vui.i_num_reorder_frames, h->param.i_frame_reference );
1142     h->frames.i_max_dpb  = h->sps->vui.i_max_dec_frame_buffering;
1143     h->frames.b_have_lowres = !h->param.rc.b_stat_read
1144         && ( h->param.rc.i_rc_method == X264_RC_ABR
1145           || h->param.rc.i_rc_method == X264_RC_CRF
1146           || h->param.i_bframe_adaptive
1147           || h->param.i_scenecut_threshold
1148           || h->param.rc.b_mb_tree
1149           || h->param.analyse.i_weighted_pred );
1150     h->frames.b_have_lowres |= h->param.rc.b_stat_read && h->param.rc.i_vbv_buffer_size > 0;
1151     h->frames.b_have_sub8x8_esa = !!(h->param.analyse.inter & X264_ANALYSE_PSUB8x8);
1152
1153     h->frames.i_last_idr =
1154     h->frames.i_last_keyframe = - h->param.i_keyint_max;
1155     h->frames.i_input    = 0;
1156     h->frames.i_largest_pts = h->frames.i_second_largest_pts = -1;
1157     h->frames.i_poc_last_open_gop = -1;
1158
1159     CHECKED_MALLOCZERO( h->frames.unused[0], (h->frames.i_delay + 3) * sizeof(x264_frame_t *) );
1160     /* Allocate room for max refs plus a few extra just in case. */
1161     CHECKED_MALLOCZERO( h->frames.unused[1], (h->i_thread_frames + X264_REF_MAX + 4) * sizeof(x264_frame_t *) );
1162     CHECKED_MALLOCZERO( h->frames.current, (h->param.i_sync_lookahead + h->param.i_bframe
1163                         + h->i_thread_frames + 3) * sizeof(x264_frame_t *) );
1164     if( h->param.analyse.i_weighted_pred > 0 )
1165         CHECKED_MALLOCZERO( h->frames.blank_unused, h->i_thread_frames * 4 * sizeof(x264_frame_t *) );
1166     h->i_ref[0] = h->i_ref[1] = 0;
1167     h->i_cpb_delay = h->i_coded_fields = h->i_disp_fields = 0;
1168     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);
1169     h->i_disp_fields_last_frame = -1;
1170     x264_rdo_init();
1171
1172     /* init CPU functions */
1173     x264_predict_16x16_init( h->param.cpu, h->predict_16x16 );
1174     x264_predict_8x8c_init( h->param.cpu, h->predict_8x8c );
1175     x264_predict_8x16c_init( h->param.cpu, h->predict_8x16c );
1176     x264_predict_8x8_init( h->param.cpu, h->predict_8x8, &h->predict_8x8_filter );
1177     x264_predict_4x4_init( h->param.cpu, h->predict_4x4 );
1178     x264_pixel_init( h->param.cpu, &h->pixf );
1179     x264_dct_init( h->param.cpu, &h->dctf );
1180     x264_zigzag_init( h->param.cpu, &h->zigzagf_progressive, &h->zigzagf_interlaced );
1181     memcpy( &h->zigzagf, PARAM_INTERLACED ? &h->zigzagf_interlaced : &h->zigzagf_progressive, sizeof(h->zigzagf) );
1182     x264_mc_init( h->param.cpu, &h->mc );
1183     x264_quant_init( h, h->param.cpu, &h->quantf );
1184     x264_deblock_init( h->param.cpu, &h->loopf, PARAM_INTERLACED );
1185     x264_bitstream_init( h->param.cpu, &h->bsf );
1186     if( h->param.b_cabac )
1187         x264_cabac_init( h );
1188     else
1189         x264_stack_align( x264_cavlc_init, h );
1190
1191     mbcmp_init( h );
1192     chroma_dsp_init( h );
1193
1194     p = buf + sprintf( buf, "using cpu capabilities:" );
1195     for( int i = 0; x264_cpu_names[i].flags; i++ )
1196     {
1197         if( !strcmp(x264_cpu_names[i].name, "SSE2")
1198             && h->param.cpu & (X264_CPU_SSE2_IS_FAST|X264_CPU_SSE2_IS_SLOW) )
1199             continue;
1200         if( !strcmp(x264_cpu_names[i].name, "SSE3")
1201             && (h->param.cpu & X264_CPU_SSSE3 || !(h->param.cpu & X264_CPU_CACHELINE_64)) )
1202             continue;
1203         if( !strcmp(x264_cpu_names[i].name, "SSE4.1")
1204             && (h->param.cpu & X264_CPU_SSE42) )
1205             continue;
1206         if( (h->param.cpu & x264_cpu_names[i].flags) == x264_cpu_names[i].flags
1207             && (!i || x264_cpu_names[i].flags != x264_cpu_names[i-1].flags) )
1208             p += sprintf( p, " %s", x264_cpu_names[i].name );
1209     }
1210     if( !h->param.cpu )
1211         p += sprintf( p, " none!" );
1212     x264_log( h, X264_LOG_INFO, "%s\n", buf );
1213
1214     float *logs = x264_analyse_prepare_costs( h );
1215     if( !logs )
1216         goto fail;
1217     for( qp = X264_MIN( h->param.rc.i_qp_min, QP_MAX_SPEC ); qp <= h->param.rc.i_qp_max; qp++ )
1218         if( x264_analyse_init_costs( h, logs, qp ) )
1219             goto fail;
1220     if( x264_analyse_init_costs( h, logs, X264_LOOKAHEAD_QP ) )
1221         goto fail;
1222     x264_free( logs );
1223
1224     static const uint16_t cost_mv_correct[7] = { 24, 47, 95, 189, 379, 757, 1515 };
1225     /* Checks for known miscompilation issues. */
1226     if( h->cost_mv[X264_LOOKAHEAD_QP][2013] != cost_mv_correct[BIT_DEPTH-8] )
1227     {
1228         x264_log( h, X264_LOG_ERROR, "MV cost test failed: x264 has been miscompiled!\n" );
1229         goto fail;
1230     }
1231
1232     /* Must be volatile or else GCC will optimize it out. */
1233     volatile int temp = 392;
1234     if( x264_clz( temp ) != 23 )
1235     {
1236         x264_log( h, X264_LOG_ERROR, "CLZ test failed: x264 has been miscompiled!\n" );
1237 #if ARCH_X86 || ARCH_X86_64
1238         x264_log( h, X264_LOG_ERROR, "Are you attempting to run an SSE4a-targeted build on a CPU that\n" );
1239         x264_log( h, X264_LOG_ERROR, "doesn't support it?\n" );
1240 #endif
1241         goto fail;
1242     }
1243
1244     h->out.i_nal = 0;
1245     h->out.i_bitstream = X264_MAX( 1000000, h->param.i_width * h->param.i_height * 4
1246         * ( h->param.rc.i_rc_method == X264_RC_ABR ? pow( 0.95, h->param.rc.i_qp_min )
1247           : pow( 0.95, h->param.rc.i_qp_constant ) * X264_MAX( 1, h->param.rc.f_ip_factor )));
1248
1249     h->nal_buffer_size = h->out.i_bitstream * 3/2 + 4;
1250     CHECKED_MALLOC( h->nal_buffer, h->nal_buffer_size );
1251
1252     if( h->param.i_threads > 1 &&
1253         x264_threadpool_init( &h->threadpool, h->param.i_threads, (void*)x264_encoder_thread_init, h ) )
1254         goto fail;
1255
1256     h->thread[0] = h;
1257     for( int i = 1; i < h->param.i_threads + !!h->param.i_sync_lookahead; i++ )
1258         CHECKED_MALLOC( h->thread[i], sizeof(x264_t) );
1259
1260     for( int i = 0; i < h->param.i_threads; i++ )
1261     {
1262         int init_nal_count = h->param.i_slice_count + 3;
1263         int allocate_threadlocal_data = !h->param.b_sliced_threads || !i;
1264         if( i > 0 )
1265             *h->thread[i] = *h;
1266
1267         if( allocate_threadlocal_data )
1268         {
1269             h->thread[i]->fdec = x264_frame_pop_unused( h, 1 );
1270             if( !h->thread[i]->fdec )
1271                 goto fail;
1272         }
1273         else
1274             h->thread[i]->fdec = h->thread[0]->fdec;
1275
1276         CHECKED_MALLOC( h->thread[i]->out.p_bitstream, h->out.i_bitstream );
1277         /* Start each thread with room for init_nal_count NAL units; it'll realloc later if needed. */
1278         CHECKED_MALLOC( h->thread[i]->out.nal, init_nal_count*sizeof(x264_nal_t) );
1279         h->thread[i]->out.i_nals_allocated = init_nal_count;
1280
1281         if( allocate_threadlocal_data && x264_macroblock_cache_allocate( h->thread[i] ) < 0 )
1282             goto fail;
1283     }
1284
1285     if( x264_lookahead_init( h, i_slicetype_length ) )
1286         goto fail;
1287
1288     for( int i = 0; i < h->param.i_threads; i++ )
1289         if( x264_macroblock_thread_allocate( h->thread[i], 0 ) < 0 )
1290             goto fail;
1291
1292     if( x264_ratecontrol_new( h ) < 0 )
1293         goto fail;
1294
1295     if( h->param.i_nal_hrd )
1296     {
1297         x264_log( h, X264_LOG_DEBUG, "HRD bitrate: %i bits/sec\n", h->sps->vui.hrd.i_bit_rate_unscaled );
1298         x264_log( h, X264_LOG_DEBUG, "CPB size: %i bits\n", h->sps->vui.hrd.i_cpb_size_unscaled );
1299     }
1300
1301     if( h->param.psz_dump_yuv )
1302     {
1303         /* create or truncate the reconstructed video file */
1304         FILE *f = fopen( h->param.psz_dump_yuv, "w" );
1305         if( !f )
1306         {
1307             x264_log( h, X264_LOG_ERROR, "dump_yuv: can't write to %s\n", h->param.psz_dump_yuv );
1308             goto fail;
1309         }
1310         else if( !x264_is_regular_file( f ) )
1311         {
1312             x264_log( h, X264_LOG_ERROR, "dump_yuv: incompatible with non-regular file %s\n", h->param.psz_dump_yuv );
1313             goto fail;
1314         }
1315         fclose( f );
1316     }
1317
1318     const char *profile = h->sps->i_profile_idc == PROFILE_BASELINE ? "Constrained Baseline" :
1319                           h->sps->i_profile_idc == PROFILE_MAIN ? "Main" :
1320                           h->sps->i_profile_idc == PROFILE_HIGH ? "High" :
1321                           h->sps->i_profile_idc == PROFILE_HIGH10 ? (h->sps->b_constraint_set3 == 1 ? "High 10 Intra" : "High 10") :
1322                           h->sps->i_profile_idc == PROFILE_HIGH422 ? (h->sps->b_constraint_set3 == 1 ? "High 4:2:2 Intra" : "High 4:2:2") :
1323                           h->sps->b_constraint_set3 == 1 ? "High 4:4:4 Intra" : "High 4:4:4 Predictive";
1324     char level[4];
1325     snprintf( level, sizeof(level), "%d.%d", h->sps->i_level_idc/10, h->sps->i_level_idc%10 );
1326     if( h->sps->i_level_idc == 9 || ( h->sps->i_level_idc == 11 && h->sps->b_constraint_set3 &&
1327         (h->sps->i_profile_idc >= PROFILE_BASELINE && h->sps->i_profile_idc <= PROFILE_EXTENDED) ) )
1328         strcpy( level, "1b" );
1329
1330     if( h->sps->i_profile_idc < PROFILE_HIGH10 )
1331     {
1332         x264_log( h, X264_LOG_INFO, "profile %s, level %s\n",
1333             profile, level );
1334     }
1335     else
1336     {
1337         static const char * const subsampling[4] = { "4:0:0", "4:2:0", "4:2:2", "4:4:4" };
1338         x264_log( h, X264_LOG_INFO, "profile %s, level %s, %s %d-bit\n",
1339             profile, level, subsampling[CHROMA_FORMAT], BIT_DEPTH );
1340     }
1341
1342     return h;
1343 fail:
1344     x264_free( h );
1345     return NULL;
1346 }
1347
1348 /****************************************************************************
1349  * x264_encoder_reconfig:
1350  ****************************************************************************/
1351 int x264_encoder_reconfig( x264_t *h, x264_param_t *param )
1352 {
1353     int rc_reconfig = 0;
1354     h = h->thread[h->thread[0]->i_thread_phase];
1355     x264_set_aspect_ratio( h, param, 0 );
1356 #define COPY(var) h->param.var = param->var
1357     COPY( i_frame_reference ); // but never uses more refs than initially specified
1358     COPY( i_bframe_bias );
1359     if( h->param.i_scenecut_threshold )
1360         COPY( i_scenecut_threshold ); // can't turn it on or off, only vary the threshold
1361     COPY( b_deblocking_filter );
1362     COPY( i_deblocking_filter_alphac0 );
1363     COPY( i_deblocking_filter_beta );
1364     COPY( i_frame_packing );
1365     COPY( analyse.inter );
1366     COPY( analyse.intra );
1367     COPY( analyse.i_direct_mv_pred );
1368     /* Scratch buffer prevents me_range from being increased for esa/tesa */
1369     if( h->param.analyse.i_me_method < X264_ME_ESA || param->analyse.i_me_range < h->param.analyse.i_me_range )
1370         COPY( analyse.i_me_range );
1371     COPY( analyse.i_noise_reduction );
1372     /* We can't switch out of subme=0 during encoding. */
1373     if( h->param.analyse.i_subpel_refine )
1374         COPY( analyse.i_subpel_refine );
1375     COPY( analyse.i_trellis );
1376     COPY( analyse.b_chroma_me );
1377     COPY( analyse.b_dct_decimate );
1378     COPY( analyse.b_fast_pskip );
1379     COPY( analyse.b_mixed_references );
1380     COPY( analyse.f_psy_rd );
1381     COPY( analyse.f_psy_trellis );
1382     COPY( crop_rect );
1383     // can only twiddle these if they were enabled to begin with:
1384     if( h->param.analyse.i_me_method >= X264_ME_ESA || param->analyse.i_me_method < X264_ME_ESA )
1385         COPY( analyse.i_me_method );
1386     if( h->param.analyse.i_me_method >= X264_ME_ESA && !h->frames.b_have_sub8x8_esa )
1387         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
1388     if( h->pps->b_transform_8x8_mode )
1389         COPY( analyse.b_transform_8x8 );
1390     if( h->frames.i_max_ref1 > 1 )
1391         COPY( i_bframe_pyramid );
1392     COPY( i_slice_max_size );
1393     COPY( i_slice_max_mbs );
1394     COPY( i_slice_count );
1395     COPY( b_tff );
1396
1397     /* VBV can't be turned on if it wasn't on to begin with */
1398     if( h->param.rc.i_vbv_max_bitrate > 0 && h->param.rc.i_vbv_buffer_size > 0 &&
1399           param->rc.i_vbv_max_bitrate > 0 &&   param->rc.i_vbv_buffer_size > 0 )
1400     {
1401         rc_reconfig |= h->param.rc.i_vbv_max_bitrate != param->rc.i_vbv_max_bitrate;
1402         rc_reconfig |= h->param.rc.i_vbv_buffer_size != param->rc.i_vbv_buffer_size;
1403         rc_reconfig |= h->param.rc.i_bitrate != param->rc.i_bitrate;
1404         COPY( rc.i_vbv_max_bitrate );
1405         COPY( rc.i_vbv_buffer_size );
1406         COPY( rc.i_bitrate );
1407     }
1408     rc_reconfig |= h->param.rc.f_rf_constant != param->rc.f_rf_constant;
1409     rc_reconfig |= h->param.rc.f_rf_constant_max != param->rc.f_rf_constant_max;
1410     COPY( rc.f_rf_constant );
1411     COPY( rc.f_rf_constant_max );
1412 #undef COPY
1413
1414     mbcmp_init( h );
1415
1416     int ret = x264_validate_parameters( h, 0 );
1417
1418     /* Supported reconfiguration options (1-pass only):
1419      * vbv-maxrate
1420      * vbv-bufsize
1421      * crf
1422      * bitrate (CBR only) */
1423     if( !ret && rc_reconfig )
1424         x264_ratecontrol_init_reconfigurable( h, 0 );
1425
1426     return ret;
1427 }
1428
1429 /****************************************************************************
1430  * x264_encoder_parameters:
1431  ****************************************************************************/
1432 void x264_encoder_parameters( x264_t *h, x264_param_t *param )
1433 {
1434     memcpy( param, &h->thread[h->i_thread_phase]->param, sizeof(x264_param_t) );
1435 }
1436
1437 /* internal usage */
1438 static void x264_nal_start( x264_t *h, int i_type, int i_ref_idc )
1439 {
1440     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1441
1442     nal->i_ref_idc        = i_ref_idc;
1443     nal->i_type           = i_type;
1444     nal->b_long_startcode = 1;
1445
1446     nal->i_payload= 0;
1447     nal->p_payload= &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8];
1448 }
1449
1450 /* if number of allocated nals is not enough, re-allocate a larger one. */
1451 static int x264_nal_check_buffer( x264_t *h )
1452 {
1453     if( h->out.i_nal >= h->out.i_nals_allocated )
1454     {
1455         x264_nal_t *new_out = x264_malloc( sizeof(x264_nal_t) * (h->out.i_nals_allocated*2) );
1456         if( !new_out )
1457             return -1;
1458         memcpy( new_out, h->out.nal, sizeof(x264_nal_t) * (h->out.i_nals_allocated) );
1459         x264_free( h->out.nal );
1460         h->out.nal = new_out;
1461         h->out.i_nals_allocated *= 2;
1462     }
1463     return 0;
1464 }
1465
1466 static int x264_nal_end( x264_t *h )
1467 {
1468     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1469     uint8_t *end = &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8];
1470     nal->i_payload = end - nal->p_payload;
1471     /* nal_escape_mmx reads past the end of the input.
1472      * While undefined padding wouldn't actually affect the output, it makes valgrind unhappy. */
1473     memset( end, 0xff, 32 );
1474     if( h->param.nalu_process )
1475         h->param.nalu_process( h, nal );
1476     h->out.i_nal++;
1477
1478     return x264_nal_check_buffer( h );
1479 }
1480
1481 static int x264_encoder_encapsulate_nals( x264_t *h, int start )
1482 {
1483     int nal_size = 0, previous_nal_size = 0;
1484
1485     if( h->param.nalu_process )
1486     {
1487         for( int i = start; i < h->out.i_nal; i++ )
1488             nal_size += h->out.nal[i].i_payload;
1489         return nal_size;
1490     }
1491
1492     for( int i = 0; i < start; i++ )
1493         previous_nal_size += h->out.nal[i].i_payload;
1494
1495     for( int i = start; i < h->out.i_nal; i++ )
1496         nal_size += h->out.nal[i].i_payload;
1497
1498     /* Worst-case NAL unit escaping: reallocate the buffer if it's too small. */
1499     int necessary_size = nal_size * 3/2 + h->out.i_nal * 4;
1500     if( h->nal_buffer_size < necessary_size )
1501     {
1502         h->nal_buffer_size = necessary_size * 2;
1503         uint8_t *buf = x264_malloc( h->nal_buffer_size );
1504         if( !buf )
1505             return -1;
1506         if( previous_nal_size )
1507             memcpy( buf, h->nal_buffer, previous_nal_size );
1508         x264_free( h->nal_buffer );
1509         h->nal_buffer = buf;
1510     }
1511
1512     uint8_t *nal_buffer = h->nal_buffer + previous_nal_size;
1513
1514     for( int i = start; i < h->out.i_nal; i++ )
1515     {
1516         h->out.nal[i].b_long_startcode = !i || h->out.nal[i].i_type == NAL_SPS || h->out.nal[i].i_type == NAL_PPS;
1517         x264_nal_encode( h, nal_buffer, &h->out.nal[i] );
1518         nal_buffer += h->out.nal[i].i_payload;
1519     }
1520
1521     x264_emms();
1522
1523     return nal_buffer - (h->nal_buffer + previous_nal_size);
1524 }
1525
1526 /****************************************************************************
1527  * x264_encoder_headers:
1528  ****************************************************************************/
1529 int x264_encoder_headers( x264_t *h, x264_nal_t **pp_nal, int *pi_nal )
1530 {
1531     int frame_size = 0;
1532     /* init bitstream context */
1533     h->out.i_nal = 0;
1534     bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
1535
1536     /* Write SEI, SPS and PPS. */
1537
1538     /* generate sequence parameters */
1539     x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
1540     x264_sps_write( &h->out.bs, h->sps );
1541     if( x264_nal_end( h ) )
1542         return -1;
1543
1544     /* generate picture parameters */
1545     x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
1546     x264_pps_write( &h->out.bs, h->sps, h->pps );
1547     if( x264_nal_end( h ) )
1548         return -1;
1549
1550     /* identify ourselves */
1551     x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
1552     if( x264_sei_version_write( h, &h->out.bs ) )
1553         return -1;
1554     if( x264_nal_end( h ) )
1555         return -1;
1556
1557     frame_size = x264_encoder_encapsulate_nals( h, 0 );
1558     if( frame_size < 0 )
1559         return -1;
1560
1561     /* now set output*/
1562     *pi_nal = h->out.i_nal;
1563     *pp_nal = &h->out.nal[0];
1564     h->out.i_nal = 0;
1565
1566     return frame_size;
1567 }
1568
1569 /* Check to see whether we have chosen a reference list ordering different
1570  * from the standard's default. */
1571 static inline void x264_reference_check_reorder( x264_t *h )
1572 {
1573     /* The reorder check doesn't check for missing frames, so just
1574      * force a reorder if one of the reference list is corrupt. */
1575     for( int i = 0; h->frames.reference[i]; i++ )
1576         if( h->frames.reference[i]->b_corrupt )
1577         {
1578             h->b_ref_reorder[0] = 1;
1579             return;
1580         }
1581     for( int list = 0; list <= (h->sh.i_type == SLICE_TYPE_B); list++ )
1582         for( int i = 0; i < h->i_ref[list] - 1; i++ )
1583         {
1584             int framenum_diff = h->fref[list][i+1]->i_frame_num - h->fref[list][i]->i_frame_num;
1585             int poc_diff = h->fref[list][i+1]->i_poc - h->fref[list][i]->i_poc;
1586             /* P and B-frames use different default orders. */
1587             if( h->sh.i_type == SLICE_TYPE_P ? framenum_diff > 0 : list == 1 ? poc_diff < 0 : poc_diff > 0 )
1588             {
1589                 h->b_ref_reorder[list] = 1;
1590                 return;
1591             }
1592         }
1593 }
1594
1595 /* return -1 on failure, else return the index of the new reference frame */
1596 int x264_weighted_reference_duplicate( x264_t *h, int i_ref, const x264_weight_t *w )
1597 {
1598     int i = h->i_ref[0];
1599     int j = 1;
1600     x264_frame_t *newframe;
1601     if( i <= 1 ) /* empty list, definitely can't duplicate frame */
1602         return -1;
1603
1604     //Duplication is only used in X264_WEIGHTP_SMART
1605     if( h->param.analyse.i_weighted_pred != X264_WEIGHTP_SMART )
1606         return -1;
1607
1608     /* Duplication is a hack to compensate for crappy rounding in motion compensation.
1609      * With high bit depth, it's not worth doing, so turn it off except in the case of
1610      * unweighted dupes. */
1611     if( BIT_DEPTH > 8 && w != x264_weight_none )
1612         return -1;
1613
1614     newframe = x264_frame_pop_blank_unused( h );
1615     if( !newframe )
1616         return -1;
1617
1618     //FIXME: probably don't need to copy everything
1619     *newframe = *h->fref[0][i_ref];
1620     newframe->i_reference_count = 1;
1621     newframe->orig = h->fref[0][i_ref];
1622     newframe->b_duplicate = 1;
1623     memcpy( h->fenc->weight[j], w, sizeof(h->fenc->weight[i]) );
1624
1625     /* shift the frames to make space for the dupe. */
1626     h->b_ref_reorder[0] = 1;
1627     if( h->i_ref[0] < X264_REF_MAX )
1628         ++h->i_ref[0];
1629     h->fref[0][X264_REF_MAX-1] = NULL;
1630     x264_frame_unshift( &h->fref[0][j], newframe );
1631
1632     return j;
1633 }
1634
1635 static void x264_weighted_pred_init( x264_t *h )
1636 {
1637     /* for now no analysis and set all weights to nothing */
1638     for( int i_ref = 0; i_ref < h->i_ref[0]; i_ref++ )
1639         h->fenc->weighted[i_ref] = h->fref[0][i_ref]->filtered[0][0];
1640
1641     // FIXME: This only supports weighting of one reference frame
1642     // and duplicates of that frame.
1643     h->fenc->i_lines_weighted = 0;
1644
1645     for( int i_ref = 0; i_ref < (h->i_ref[0] << SLICE_MBAFF); i_ref++ )
1646         for( int i = 0; i < 3; i++ )
1647             h->sh.weight[i_ref][i].weightfn = NULL;
1648
1649
1650     if( h->sh.i_type != SLICE_TYPE_P || h->param.analyse.i_weighted_pred <= 0 )
1651         return;
1652
1653     int i_padv = PADV << PARAM_INTERLACED;
1654     int denom = -1;
1655     int weightplane[2] = { 0, 0 };
1656     int buffer_next = 0;
1657     for( int i = 0; i < 3; i++ )
1658     {
1659         for( int j = 0; j < h->i_ref[0]; j++ )
1660         {
1661             if( h->fenc->weight[j][i].weightfn )
1662             {
1663                 h->sh.weight[j][i] = h->fenc->weight[j][i];
1664                 // if weight is useless, don't write it to stream
1665                 if( h->sh.weight[j][i].i_scale == 1<<h->sh.weight[j][i].i_denom && h->sh.weight[j][i].i_offset == 0 )
1666                     h->sh.weight[j][i].weightfn = NULL;
1667                 else
1668                 {
1669                     if( !weightplane[!!i] )
1670                     {
1671                         weightplane[!!i] = 1;
1672                         h->sh.weight[0][!!i].i_denom = denom = h->sh.weight[j][i].i_denom;
1673                         assert( x264_clip3( denom, 0, 7 ) == denom );
1674                     }
1675
1676                     assert( h->sh.weight[j][i].i_denom == denom );
1677                     if( !i )
1678                     {
1679                         h->fenc->weighted[j] = h->mb.p_weight_buf[buffer_next++] + h->fenc->i_stride[0] * i_padv + PADH;
1680                         //scale full resolution frame
1681                         if( h->param.i_threads == 1 )
1682                         {
1683                             pixel *src = h->fref[0][j]->filtered[0][0] - h->fref[0][j]->i_stride[0]*i_padv - PADH;
1684                             pixel *dst = h->fenc->weighted[j] - h->fenc->i_stride[0]*i_padv - PADH;
1685                             int stride = h->fenc->i_stride[0];
1686                             int width = h->fenc->i_width[0] + PADH*2;
1687                             int height = h->fenc->i_lines[0] + i_padv*2;
1688                             x264_weight_scale_plane( h, dst, stride, src, stride, width, height, &h->sh.weight[j][0] );
1689                             h->fenc->i_lines_weighted = height;
1690                         }
1691                     }
1692                 }
1693             }
1694         }
1695     }
1696
1697     if( weightplane[1] )
1698         for( int i = 0; i < h->i_ref[0]; i++ )
1699         {
1700             if( h->sh.weight[i][1].weightfn && !h->sh.weight[i][2].weightfn )
1701             {
1702                 h->sh.weight[i][2].i_scale = 1 << h->sh.weight[0][1].i_denom;
1703                 h->sh.weight[i][2].i_offset = 0;
1704             }
1705             else if( h->sh.weight[i][2].weightfn && !h->sh.weight[i][1].weightfn )
1706             {
1707                 h->sh.weight[i][1].i_scale = 1 << h->sh.weight[0][1].i_denom;
1708                 h->sh.weight[i][1].i_offset = 0;
1709             }
1710         }
1711
1712     if( !weightplane[0] )
1713         h->sh.weight[0][0].i_denom = 0;
1714     if( !weightplane[1] )
1715         h->sh.weight[0][1].i_denom = 0;
1716     h->sh.weight[0][2].i_denom = h->sh.weight[0][1].i_denom;
1717 }
1718
1719 static inline int x264_reference_distance( x264_t *h, x264_frame_t *frame )
1720 {
1721     if( h->param.i_frame_packing == 5 )
1722         return abs((h->fenc->i_frame&~1) - (frame->i_frame&~1)) +
1723                   ((h->fenc->i_frame&1) != (frame->i_frame&1));
1724     else
1725         return abs(h->fenc->i_frame - frame->i_frame);
1726 }
1727
1728 static inline void x264_reference_build_list( x264_t *h, int i_poc )
1729 {
1730     int b_ok;
1731
1732     /* build ref list 0/1 */
1733     h->mb.pic.i_fref[0] = h->i_ref[0] = 0;
1734     h->mb.pic.i_fref[1] = h->i_ref[1] = 0;
1735     if( h->sh.i_type == SLICE_TYPE_I )
1736         return;
1737
1738     for( int i = 0; h->frames.reference[i]; i++ )
1739     {
1740         if( h->frames.reference[i]->b_corrupt )
1741             continue;
1742         if( h->frames.reference[i]->i_poc < i_poc )
1743             h->fref[0][h->i_ref[0]++] = h->frames.reference[i];
1744         else if( h->frames.reference[i]->i_poc > i_poc )
1745             h->fref[1][h->i_ref[1]++] = h->frames.reference[i];
1746     }
1747
1748     /* Order reference lists by distance from the current frame. */
1749     for( int list = 0; list < 2; list++ )
1750     {
1751         h->fref_nearest[list] = h->fref[list][0];
1752         do
1753         {
1754             b_ok = 1;
1755             for( int i = 0; i < h->i_ref[list] - 1; i++ )
1756             {
1757                 if( list ? h->fref[list][i+1]->i_poc < h->fref_nearest[list]->i_poc
1758                          : h->fref[list][i+1]->i_poc > h->fref_nearest[list]->i_poc )
1759                     h->fref_nearest[list] = h->fref[list][i+1];
1760                 if( x264_reference_distance( h, h->fref[list][i] ) > x264_reference_distance( h, h->fref[list][i+1] ) )
1761                 {
1762                     XCHG( x264_frame_t*, h->fref[list][i], h->fref[list][i+1] );
1763                     b_ok = 0;
1764                     break;
1765                 }
1766             }
1767         } while( !b_ok );
1768     }
1769
1770     if( h->sh.i_mmco_remove_from_end )
1771         for( int i = h->i_ref[0]-1; i >= h->i_ref[0] - h->sh.i_mmco_remove_from_end; i-- )
1772         {
1773             int diff = h->i_frame_num - h->fref[0][i]->i_frame_num;
1774             h->sh.mmco[h->sh.i_mmco_command_count].i_poc = h->fref[0][i]->i_poc;
1775             h->sh.mmco[h->sh.i_mmco_command_count++].i_difference_of_pic_nums = diff;
1776         }
1777
1778     x264_reference_check_reorder( h );
1779
1780     h->i_ref[1] = X264_MIN( h->i_ref[1], h->frames.i_max_ref1 );
1781     h->i_ref[0] = X264_MIN( h->i_ref[0], h->frames.i_max_ref0 );
1782     h->i_ref[0] = X264_MIN( h->i_ref[0], h->param.i_frame_reference ); // if reconfig() has lowered the limit
1783
1784     /* For Blu-ray compliance, don't reference frames outside of the minigop. */
1785     if( IS_X264_TYPE_B( h->fenc->i_type ) && h->param.b_bluray_compat )
1786         h->i_ref[0] = X264_MIN( h->i_ref[0], IS_X264_TYPE_B( h->fref[0][0]->i_type ) + 1 );
1787
1788     /* add duplicates */
1789     if( h->fenc->i_type == X264_TYPE_P )
1790     {
1791         int idx = -1;
1792         if( h->param.analyse.i_weighted_pred >= X264_WEIGHTP_SIMPLE )
1793         {
1794             x264_weight_t w[3];
1795             w[1].weightfn = w[2].weightfn = NULL;
1796             if( h->param.rc.b_stat_read )
1797                 x264_ratecontrol_set_weights( h, h->fenc );
1798
1799             if( !h->fenc->weight[0][0].weightfn )
1800             {
1801                 h->fenc->weight[0][0].i_denom = 0;
1802                 SET_WEIGHT( w[0], 1, 1, 0, -1 );
1803                 idx = x264_weighted_reference_duplicate( h, 0, w );
1804             }
1805             else
1806             {
1807                 if( h->fenc->weight[0][0].i_scale == 1<<h->fenc->weight[0][0].i_denom )
1808                 {
1809                     SET_WEIGHT( h->fenc->weight[0][0], 1, 1, 0, h->fenc->weight[0][0].i_offset );
1810                 }
1811                 x264_weighted_reference_duplicate( h, 0, x264_weight_none );
1812                 if( h->fenc->weight[0][0].i_offset > -128 )
1813                 {
1814                     w[0] = h->fenc->weight[0][0];
1815                     w[0].i_offset--;
1816                     h->mc.weight_cache( h, &w[0] );
1817                     idx = x264_weighted_reference_duplicate( h, 0, w );
1818                 }
1819             }
1820         }
1821         h->mb.ref_blind_dupe = idx;
1822     }
1823
1824     assert( h->i_ref[0] + h->i_ref[1] <= X264_REF_MAX );
1825     h->mb.pic.i_fref[0] = h->i_ref[0];
1826     h->mb.pic.i_fref[1] = h->i_ref[1];
1827 }
1828
1829 static void x264_fdec_filter_row( x264_t *h, int mb_y, int b_inloop )
1830 {
1831     /* mb_y is the mb to be encoded next, not the mb to be filtered here */
1832     int b_hpel = h->fdec->b_kept_as_ref;
1833     int b_deblock = h->sh.i_disable_deblocking_filter_idc != 1;
1834     int b_end = mb_y == h->i_threadslice_end;
1835     int b_measure_quality = 1;
1836     int min_y = mb_y - (1 << SLICE_MBAFF);
1837     int b_start = min_y == h->i_threadslice_start;
1838     /* Even in interlaced mode, deblocking never modifies more than 4 pixels
1839      * above each MB, as bS=4 doesn't happen for the top of interlaced mbpairs. */
1840     int minpix_y = min_y*16 - 4 * !b_start;
1841     int maxpix_y = mb_y*16 - 4 * !b_end;
1842     b_deblock &= b_hpel || h->param.psz_dump_yuv;
1843     if( h->param.b_sliced_threads && b_start && min_y && !b_inloop )
1844     {
1845         b_deblock = 0;         /* We already deblocked on the inloop pass. */
1846         b_measure_quality = 0; /* We already measured quality on the inloop pass. */
1847     }
1848     if( mb_y & SLICE_MBAFF )
1849         return;
1850     if( min_y < h->i_threadslice_start )
1851         return;
1852
1853     if( b_deblock )
1854         for( int y = min_y; y < mb_y; y += (1 << SLICE_MBAFF) )
1855             x264_frame_deblock_row( h, y );
1856
1857     /* FIXME: Prediction requires different borders for interlaced/progressive mc,
1858      * but the actual image data is equivalent. For now, maintain this
1859      * consistency by copying deblocked pixels between planes. */
1860     if( PARAM_INTERLACED )
1861         for( int p = 0; p < h->fdec->i_plane; p++ )
1862             for( int i = minpix_y>>(CHROMA_V_SHIFT && p); i < maxpix_y>>(CHROMA_V_SHIFT && p); i++ )
1863                 memcpy( h->fdec->plane_fld[p] + i*h->fdec->i_stride[p],
1864                         h->fdec->plane[p]     + i*h->fdec->i_stride[p],
1865                         h->mb.i_mb_width*16*sizeof(pixel) );
1866
1867     if( b_hpel )
1868     {
1869         int end = mb_y == h->mb.i_mb_height;
1870         x264_frame_expand_border( h, h->fdec, min_y, end );
1871         if( h->param.analyse.i_subpel_refine )
1872         {
1873             x264_frame_filter( h, h->fdec, min_y, end );
1874             x264_frame_expand_border_filtered( h, h->fdec, min_y, end );
1875         }
1876     }
1877
1878     if( SLICE_MBAFF )
1879         for( int i = 0; i < 3; i++ )
1880         {
1881             XCHG( pixel *, h->intra_border_backup[0][i], h->intra_border_backup[3][i] );
1882             XCHG( pixel *, h->intra_border_backup[1][i], h->intra_border_backup[4][i] );
1883         }
1884
1885     if( h->i_thread_frames > 1 && h->fdec->b_kept_as_ref )
1886         x264_frame_cond_broadcast( h->fdec, mb_y*16 + (b_end ? 10000 : -(X264_THREAD_HEIGHT << SLICE_MBAFF)) );
1887
1888     if( b_measure_quality )
1889     {
1890         maxpix_y = X264_MIN( maxpix_y, h->param.i_height );
1891         if( h->param.analyse.b_psnr )
1892         {
1893             for( int p = 0; p < (CHROMA444 ? 3 : 1); p++ )
1894                 h->stat.frame.i_ssd[p] += x264_pixel_ssd_wxh( &h->pixf,
1895                     h->fdec->plane[p] + minpix_y * h->fdec->i_stride[p], h->fdec->i_stride[p],
1896                     h->fenc->plane[p] + minpix_y * h->fenc->i_stride[p], h->fenc->i_stride[p],
1897                     h->param.i_width, maxpix_y-minpix_y );
1898             if( !CHROMA444 )
1899             {
1900                 uint64_t ssd_u, ssd_v;
1901                 int v_shift = CHROMA_V_SHIFT;
1902                 x264_pixel_ssd_nv12( &h->pixf,
1903                     h->fdec->plane[1] + (minpix_y>>v_shift) * h->fdec->i_stride[1], h->fdec->i_stride[1],
1904                     h->fenc->plane[1] + (minpix_y>>v_shift) * h->fenc->i_stride[1], h->fenc->i_stride[1],
1905                     h->param.i_width>>1, (maxpix_y-minpix_y)>>v_shift, &ssd_u, &ssd_v );
1906                 h->stat.frame.i_ssd[1] += ssd_u;
1907                 h->stat.frame.i_ssd[2] += ssd_v;
1908             }
1909         }
1910
1911         if( h->param.analyse.b_ssim )
1912         {
1913             int ssim_cnt;
1914             x264_emms();
1915             /* offset by 2 pixels to avoid alignment of ssim blocks with dct blocks,
1916              * and overlap by 4 */
1917             minpix_y += b_start ? 2 : -6;
1918             h->stat.frame.f_ssim +=
1919                 x264_pixel_ssim_wxh( &h->pixf,
1920                     h->fdec->plane[0] + 2+minpix_y*h->fdec->i_stride[0], h->fdec->i_stride[0],
1921                     h->fenc->plane[0] + 2+minpix_y*h->fenc->i_stride[0], h->fenc->i_stride[0],
1922                     h->param.i_width-2, maxpix_y-minpix_y, h->scratch_buffer, &ssim_cnt );
1923             h->stat.frame.i_ssim_cnt += ssim_cnt;
1924         }
1925     }
1926 }
1927
1928 static inline int x264_reference_update( x264_t *h )
1929 {
1930     if( !h->fdec->b_kept_as_ref )
1931     {
1932         if( h->i_thread_frames > 1 )
1933         {
1934             x264_frame_push_unused( h, h->fdec );
1935             h->fdec = x264_frame_pop_unused( h, 1 );
1936             if( !h->fdec )
1937                 return -1;
1938         }
1939         return 0;
1940     }
1941
1942     /* apply mmco from previous frame. */
1943     for( int i = 0; i < h->sh.i_mmco_command_count; i++ )
1944         for( int j = 0; h->frames.reference[j]; j++ )
1945             if( h->frames.reference[j]->i_poc == h->sh.mmco[i].i_poc )
1946                 x264_frame_push_unused( h, x264_frame_shift( &h->frames.reference[j] ) );
1947
1948     /* move frame in the buffer */
1949     x264_frame_push( h->frames.reference, h->fdec );
1950     if( h->frames.reference[h->sps->i_num_ref_frames] )
1951         x264_frame_push_unused( h, x264_frame_shift( h->frames.reference ) );
1952     h->fdec = x264_frame_pop_unused( h, 1 );
1953     if( !h->fdec )
1954         return -1;
1955     return 0;
1956 }
1957
1958 static inline void x264_reference_reset( x264_t *h )
1959 {
1960     while( h->frames.reference[0] )
1961         x264_frame_push_unused( h, x264_frame_pop( h->frames.reference ) );
1962     h->fdec->i_poc =
1963     h->fenc->i_poc = 0;
1964 }
1965
1966 static inline void x264_reference_hierarchy_reset( x264_t *h )
1967 {
1968     int ref;
1969     int b_hasdelayframe = 0;
1970
1971     /* look for delay frames -- chain must only contain frames that are disposable */
1972     for( int i = 0; h->frames.current[i] && IS_DISPOSABLE( h->frames.current[i]->i_type ); i++ )
1973         b_hasdelayframe |= h->frames.current[i]->i_coded
1974                         != h->frames.current[i]->i_frame + h->sps->vui.i_num_reorder_frames;
1975
1976     /* This function must handle b-pyramid and clear frames for open-gop */
1977     if( h->param.i_bframe_pyramid != X264_B_PYRAMID_STRICT && !b_hasdelayframe && h->frames.i_poc_last_open_gop == -1 )
1978         return;
1979
1980     /* Remove last BREF. There will never be old BREFs in the
1981      * dpb during a BREF decode when pyramid == STRICT */
1982     for( ref = 0; h->frames.reference[ref]; ref++ )
1983     {
1984         if( ( h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT
1985             && h->frames.reference[ref]->i_type == X264_TYPE_BREF )
1986             || ( h->frames.reference[ref]->i_poc < h->frames.i_poc_last_open_gop
1987             && h->sh.i_type != SLICE_TYPE_B ) )
1988         {
1989             int diff = h->i_frame_num - h->frames.reference[ref]->i_frame_num;
1990             h->sh.mmco[h->sh.i_mmco_command_count].i_difference_of_pic_nums = diff;
1991             h->sh.mmco[h->sh.i_mmco_command_count++].i_poc = h->frames.reference[ref]->i_poc;
1992             x264_frame_push_unused( h, x264_frame_shift( &h->frames.reference[ref] ) );
1993             h->b_ref_reorder[0] = 1;
1994             ref--;
1995         }
1996     }
1997
1998     /* Prepare room in the dpb for the delayed display time of the later b-frame's */
1999     if( h->param.i_bframe_pyramid )
2000         h->sh.i_mmco_remove_from_end = X264_MAX( ref + 2 - h->frames.i_max_dpb, 0 );
2001 }
2002
2003 static inline void x264_slice_init( x264_t *h, int i_nal_type, int i_global_qp )
2004 {
2005     /* ------------------------ Create slice header  ----------------------- */
2006     if( i_nal_type == NAL_SLICE_IDR )
2007     {
2008         x264_slice_header_init( h, &h->sh, h->sps, h->pps, h->i_idr_pic_id, h->i_frame_num, i_global_qp );
2009
2010         /* alternate id */
2011         h->i_idr_pic_id ^= 1;
2012     }
2013     else
2014     {
2015         x264_slice_header_init( h, &h->sh, h->sps, h->pps, -1, h->i_frame_num, i_global_qp );
2016
2017         h->sh.i_num_ref_idx_l0_active = h->i_ref[0] <= 0 ? 1 : h->i_ref[0];
2018         h->sh.i_num_ref_idx_l1_active = h->i_ref[1] <= 0 ? 1 : h->i_ref[1];
2019         if( h->sh.i_num_ref_idx_l0_active != h->pps->i_num_ref_idx_l0_default_active ||
2020             (h->sh.i_type == SLICE_TYPE_B && h->sh.i_num_ref_idx_l1_active != h->pps->i_num_ref_idx_l1_default_active) )
2021         {
2022             h->sh.b_num_ref_idx_override = 1;
2023         }
2024     }
2025
2026     if( h->fenc->i_type == X264_TYPE_BREF && h->param.b_bluray_compat && h->sh.i_mmco_command_count )
2027     {
2028         h->b_sh_backup = 1;
2029         h->sh_backup = h->sh;
2030     }
2031
2032     h->fdec->i_frame_num = h->sh.i_frame_num;
2033
2034     if( h->sps->i_poc_type == 0 )
2035     {
2036         h->sh.i_poc = h->fdec->i_poc;
2037         if( PARAM_INTERLACED )
2038         {
2039             h->sh.i_delta_poc_bottom = h->param.b_tff ? 1 : -1;
2040             h->sh.i_poc += h->sh.i_delta_poc_bottom == -1;
2041         }
2042         else
2043             h->sh.i_delta_poc_bottom = 0;
2044         h->fdec->i_delta_poc[0] = h->sh.i_delta_poc_bottom == -1;
2045         h->fdec->i_delta_poc[1] = h->sh.i_delta_poc_bottom ==  1;
2046     }
2047     else
2048     {
2049         /* Nothing to do ? */
2050     }
2051
2052     x264_macroblock_slice_init( h );
2053 }
2054
2055 static int x264_slice_write( x264_t *h )
2056 {
2057     int i_skip;
2058     int mb_xy, i_mb_x, i_mb_y;
2059     int i_skip_bak = 0; /* Shut up GCC. */
2060     bs_t UNINIT(bs_bak);
2061     x264_cabac_t cabac_bak;
2062     uint8_t cabac_prevbyte_bak = 0; /* Shut up GCC. */
2063     int mv_bits_bak = 0;
2064     int tex_bits_bak = 0;
2065     /* NALUs other than the first use a 3-byte startcode.
2066      * Add one extra byte for the rbsp, and one more for the final CABAC putbyte.
2067      * Then add an extra 5 bytes just in case, to account for random NAL escapes and
2068      * other inaccuracies. */
2069     int overhead_guess = (NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal)) + 1 + h->param.b_cabac + 5;
2070     int slice_max_size = h->param.i_slice_max_size > 0 ? (h->param.i_slice_max_size-overhead_guess)*8 : 0;
2071     int back_up_bitstream = slice_max_size || (!h->param.b_cabac && h->sps->i_profile_idc < PROFILE_HIGH);
2072     int starting_bits = bs_pos(&h->out.bs);
2073     int b_deblock = h->sh.i_disable_deblocking_filter_idc != 1;
2074     int b_hpel = h->fdec->b_kept_as_ref;
2075     uint8_t *last_emu_check;
2076     b_deblock &= b_hpel || h->param.psz_dump_yuv;
2077     bs_realign( &h->out.bs );
2078
2079     /* Slice */
2080     x264_nal_start( h, h->i_nal_type, h->i_nal_ref_idc );
2081     h->out.nal[h->out.i_nal].i_first_mb = h->sh.i_first_mb;
2082
2083     /* Slice header */
2084     x264_macroblock_thread_init( h );
2085
2086     /* If this isn't the first slice in the threadslice, set the slice QP
2087      * equal to the last QP in the previous slice for more accurate
2088      * CABAC initialization. */
2089     if( h->sh.i_first_mb != h->i_threadslice_start * h->mb.i_mb_width )
2090     {
2091         h->sh.i_qp = h->mb.i_last_qp;
2092         h->sh.i_qp_delta = h->sh.i_qp - h->pps->i_pic_init_qp;
2093     }
2094
2095     x264_slice_header_write( &h->out.bs, &h->sh, h->i_nal_ref_idc );
2096     if( h->param.b_cabac )
2097     {
2098         /* alignment needed */
2099         bs_align_1( &h->out.bs );
2100
2101         /* init cabac */
2102         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 );
2103         x264_cabac_encode_init ( &h->cabac, h->out.bs.p, h->out.bs.p_end );
2104         last_emu_check = h->cabac.p;
2105     }
2106     else
2107         last_emu_check = h->out.bs.p;
2108     h->mb.i_last_qp = h->sh.i_qp;
2109     h->mb.i_last_dqp = 0;
2110     h->mb.field_decoding_flag = 0;
2111
2112     i_mb_y = h->sh.i_first_mb / h->mb.i_mb_width;
2113     i_mb_x = h->sh.i_first_mb % h->mb.i_mb_width;
2114     i_skip = 0;
2115
2116     while( 1 )
2117     {
2118         mb_xy = i_mb_x + i_mb_y * h->mb.i_mb_width;
2119         int mb_spos = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
2120
2121         if( !(i_mb_y & SLICE_MBAFF) )
2122         {
2123             if( x264_bitstream_check_buffer( h ) )
2124                 return -1;
2125
2126             if( back_up_bitstream )
2127             {
2128                 mv_bits_bak = h->stat.frame.i_mv_bits;
2129                 tex_bits_bak = h->stat.frame.i_tex_bits;
2130                 /* We don't need the contexts because flushing the CABAC encoder has no context
2131                  * dependency and macroblocks are only re-encoded in the case where a slice is
2132                  * ended (and thus the content of all contexts are thrown away). */
2133                 if( h->param.b_cabac )
2134                 {
2135                     memcpy( &cabac_bak, &h->cabac, offsetof(x264_cabac_t, f8_bits_encoded) );
2136                     /* x264's CABAC writer modifies the previous byte during carry, so it has to be
2137                      * backed up. */
2138                     cabac_prevbyte_bak = h->cabac.p[-1];
2139                 }
2140                 else
2141                 {
2142                     bs_bak = h->out.bs;
2143                     i_skip_bak = i_skip;
2144                 }
2145             }
2146         }
2147
2148         if( i_mb_x == 0 && !h->mb.b_reencode_mb )
2149             x264_fdec_filter_row( h, i_mb_y, 1 );
2150
2151         if( PARAM_INTERLACED )
2152         {
2153             if( h->mb.b_adaptive_mbaff )
2154             {
2155                 if( !(i_mb_y&1) )
2156                 {
2157                     /* FIXME: VSAD is fast but fairly poor at choosing the best interlace type. */
2158                     h->mb.b_interlaced = x264_field_vsad( h, i_mb_x, i_mb_y );
2159                     memcpy( &h->zigzagf, MB_INTERLACED ? &h->zigzagf_interlaced : &h->zigzagf_progressive, sizeof(h->zigzagf) );
2160                     if( !MB_INTERLACED && (i_mb_y+2) == h->mb.i_mb_height )
2161                         x264_expand_border_mbpair( h, i_mb_x, i_mb_y );
2162                 }
2163             }
2164             h->mb.field[mb_xy] = MB_INTERLACED;
2165         }
2166
2167         /* load cache */
2168         if( SLICE_MBAFF )
2169             x264_macroblock_cache_load_interlaced( h, i_mb_x, i_mb_y );
2170         else
2171             x264_macroblock_cache_load_progressive( h, i_mb_x, i_mb_y );
2172
2173         x264_macroblock_analyse( h );
2174
2175         /* encode this macroblock -> be careful it can change the mb type to P_SKIP if needed */
2176 reencode:
2177         x264_macroblock_encode( h );
2178
2179         if( h->param.b_cabac )
2180         {
2181             if( mb_xy > h->sh.i_first_mb && !(SLICE_MBAFF && (i_mb_y&1)) )
2182                 x264_cabac_encode_terminal( &h->cabac );
2183
2184             if( IS_SKIP( h->mb.i_type ) )
2185                 x264_cabac_mb_skip( h, 1 );
2186             else
2187             {
2188                 if( h->sh.i_type != SLICE_TYPE_I )
2189                     x264_cabac_mb_skip( h, 0 );
2190                 x264_macroblock_write_cabac( h, &h->cabac );
2191             }
2192         }
2193         else
2194         {
2195             if( IS_SKIP( h->mb.i_type ) )
2196                 i_skip++;
2197             else
2198             {
2199                 if( h->sh.i_type != SLICE_TYPE_I )
2200                 {
2201                     bs_write_ue( &h->out.bs, i_skip );  /* skip run */
2202                     i_skip = 0;
2203                 }
2204                 x264_macroblock_write_cavlc( h );
2205                 /* If there was a CAVLC level code overflow, try again at a higher QP. */
2206                 if( h->mb.b_overflow )
2207                 {
2208                     h->mb.i_chroma_qp = h->chroma_qp_table[++h->mb.i_qp];
2209                     h->mb.i_skip_intra = 0;
2210                     h->mb.b_skip_mc = 0;
2211                     h->mb.b_overflow = 0;
2212                     h->out.bs = bs_bak;
2213                     i_skip = i_skip_bak;
2214                     h->stat.frame.i_mv_bits = mv_bits_bak;
2215                     h->stat.frame.i_tex_bits = tex_bits_bak;
2216                     goto reencode;
2217                 }
2218             }
2219         }
2220
2221         int total_bits = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
2222         int mb_size = total_bits - mb_spos;
2223
2224         if( slice_max_size && (!SLICE_MBAFF || (i_mb_y&1)) )
2225         {
2226             /* Count the skip run, just in case. */
2227             if( !h->param.b_cabac )
2228                 total_bits += bs_size_ue_big( i_skip );
2229             /* Check for escape bytes. */
2230             uint8_t *end = h->param.b_cabac ? h->cabac.p : h->out.bs.p;
2231             for( ; last_emu_check < end - 2; last_emu_check++ )
2232                 if( last_emu_check[0] == 0 && last_emu_check[1] == 0 && last_emu_check[2] <= 3 )
2233                 {
2234                     slice_max_size -= 8;
2235                     last_emu_check++;
2236                 }
2237             /* We'll just re-encode this last macroblock if we go over the max slice size. */
2238             if( total_bits - starting_bits > slice_max_size && !h->mb.b_reencode_mb )
2239             {
2240                 if( mb_xy-SLICE_MBAFF*h->mb.i_mb_stride != h->sh.i_first_mb )
2241                 {
2242                     h->stat.frame.i_mv_bits = mv_bits_bak;
2243                     h->stat.frame.i_tex_bits = tex_bits_bak;
2244                     if( h->param.b_cabac )
2245                     {
2246                         memcpy( &h->cabac, &cabac_bak, offsetof(x264_cabac_t, f8_bits_encoded) );
2247                         h->cabac.p[-1] = cabac_prevbyte_bak;
2248                     }
2249                     else
2250                     {
2251                         h->out.bs = bs_bak;
2252                         i_skip = i_skip_bak;
2253                     }
2254                     h->mb.b_reencode_mb = 1;
2255                     if( SLICE_MBAFF )
2256                     {
2257                         // set to bottom of previous mbpair
2258                         if( i_mb_x )
2259                             h->sh.i_last_mb = mb_xy-1+h->mb.i_mb_stride*(!(i_mb_y&1));
2260                         else
2261                             h->sh.i_last_mb = (i_mb_y-2+!(i_mb_y&1))*h->mb.i_mb_stride + h->mb.i_mb_width - 1;
2262                     }
2263                     else
2264                         h->sh.i_last_mb = mb_xy-1;
2265                     break;
2266                 }
2267                 else
2268                 {
2269                     h->sh.i_last_mb = mb_xy;
2270                     h->mb.b_reencode_mb = 0;
2271                 }
2272             }
2273             else
2274                 h->mb.b_reencode_mb = 0;
2275         }
2276
2277 #if HAVE_VISUALIZE
2278         if( h->param.b_visualize )
2279             x264_visualize_mb( h );
2280 #endif
2281
2282         /* save cache */
2283         x264_macroblock_cache_save( h );
2284
2285         /* accumulate mb stats */
2286         h->stat.frame.i_mb_count[h->mb.i_type]++;
2287
2288         int b_intra = IS_INTRA( h->mb.i_type );
2289         int b_skip = IS_SKIP( h->mb.i_type );
2290         if( h->param.i_log_level >= X264_LOG_INFO || h->param.rc.b_stat_write )
2291         {
2292             if( !b_intra && !b_skip && !IS_DIRECT( h->mb.i_type ) )
2293             {
2294                 if( h->mb.i_partition != D_8x8 )
2295                         h->stat.frame.i_mb_partition[h->mb.i_partition] += 4;
2296                     else
2297                         for( int i = 0; i < 4; i++ )
2298                             h->stat.frame.i_mb_partition[h->mb.i_sub_partition[i]] ++;
2299                 if( h->param.i_frame_reference > 1 )
2300                     for( int i_list = 0; i_list <= (h->sh.i_type == SLICE_TYPE_B); i_list++ )
2301                         for( int i = 0; i < 4; i++ )
2302                         {
2303                             int i_ref = h->mb.cache.ref[i_list][ x264_scan8[4*i] ];
2304                             if( i_ref >= 0 )
2305                                 h->stat.frame.i_mb_count_ref[i_list][i_ref] ++;
2306                         }
2307             }
2308         }
2309
2310         if( h->param.i_log_level >= X264_LOG_INFO )
2311         {
2312             if( h->mb.i_cbp_luma | h->mb.i_cbp_chroma )
2313             {
2314                 if( CHROMA444 )
2315                 {
2316                     for( int i = 0; i < 4; i++ )
2317                         if( h->mb.i_cbp_luma & (1 << i) )
2318                             for( int p = 0; p < 3; p++ )
2319                             {
2320                                 int s8 = i*4+p*16;
2321                                 int nnz8x8 = M16( &h->mb.cache.non_zero_count[x264_scan8[s8]+0] )
2322                                            | M16( &h->mb.cache.non_zero_count[x264_scan8[s8]+8] );
2323                                 h->stat.frame.i_mb_cbp[!b_intra + p*2] += !!nnz8x8;
2324                             }
2325                 }
2326                 else
2327                 {
2328                     int cbpsum = (h->mb.i_cbp_luma&1) + ((h->mb.i_cbp_luma>>1)&1)
2329                                + ((h->mb.i_cbp_luma>>2)&1) + (h->mb.i_cbp_luma>>3);
2330                     h->stat.frame.i_mb_cbp[!b_intra + 0] += cbpsum;
2331                     h->stat.frame.i_mb_cbp[!b_intra + 2] += !!h->mb.i_cbp_chroma;
2332                     h->stat.frame.i_mb_cbp[!b_intra + 4] += h->mb.i_cbp_chroma >> 1;
2333                 }
2334             }
2335             if( h->mb.i_cbp_luma && !b_intra )
2336             {
2337                 h->stat.frame.i_mb_count_8x8dct[0] ++;
2338                 h->stat.frame.i_mb_count_8x8dct[1] += h->mb.b_transform_8x8;
2339             }
2340             if( b_intra && h->mb.i_type != I_PCM )
2341             {
2342                 if( h->mb.i_type == I_16x16 )
2343                     h->stat.frame.i_mb_pred_mode[0][h->mb.i_intra16x16_pred_mode]++;
2344                 else if( h->mb.i_type == I_8x8 )
2345                     for( int i = 0; i < 16; i += 4 )
2346                         h->stat.frame.i_mb_pred_mode[1][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
2347                 else //if( h->mb.i_type == I_4x4 )
2348                     for( int i = 0; i < 16; i++ )
2349                         h->stat.frame.i_mb_pred_mode[2][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
2350                 h->stat.frame.i_mb_pred_mode[3][x264_mb_chroma_pred_mode_fix[h->mb.i_chroma_pred_mode]]++;
2351             }
2352             h->stat.frame.i_mb_field[b_intra?0:b_skip?2:1] += MB_INTERLACED;
2353         }
2354
2355         /* calculate deblock strength values (actual deblocking is done per-row along with hpel) */
2356         if( b_deblock )
2357             x264_macroblock_deblock_strength( h );
2358
2359         x264_ratecontrol_mb( h, mb_size );
2360
2361         if( mb_xy == h->sh.i_last_mb )
2362             break;
2363
2364         if( SLICE_MBAFF )
2365         {
2366             i_mb_x += i_mb_y & 1;
2367             i_mb_y ^= i_mb_x < h->mb.i_mb_width;
2368         }
2369         else
2370             i_mb_x++;
2371         if( i_mb_x == h->mb.i_mb_width )
2372         {
2373             i_mb_y++;
2374             i_mb_x = 0;
2375         }
2376     }
2377     h->out.nal[h->out.i_nal].i_last_mb = h->sh.i_last_mb;
2378
2379     if( h->param.b_cabac )
2380     {
2381         x264_cabac_encode_flush( h, &h->cabac );
2382         h->out.bs.p = h->cabac.p;
2383     }
2384     else
2385     {
2386         if( i_skip > 0 )
2387             bs_write_ue( &h->out.bs, i_skip );  /* last skip run */
2388         /* rbsp_slice_trailing_bits */
2389         bs_rbsp_trailing( &h->out.bs );
2390         bs_flush( &h->out.bs );
2391     }
2392     if( x264_nal_end( h ) )
2393         return -1;
2394
2395     if( h->sh.i_last_mb == (h->i_threadslice_end * h->mb.i_mb_width - 1) )
2396     {
2397         h->stat.frame.i_misc_bits = bs_pos( &h->out.bs )
2398                                   + (h->out.i_nal*NALU_OVERHEAD * 8)
2399                                   - h->stat.frame.i_tex_bits
2400                                   - h->stat.frame.i_mv_bits;
2401         x264_fdec_filter_row( h, h->i_threadslice_end, 1 );
2402     }
2403
2404     return 0;
2405 }
2406
2407 static void x264_thread_sync_context( x264_t *dst, x264_t *src )
2408 {
2409     if( dst == src )
2410         return;
2411
2412     // reference counting
2413     for( x264_frame_t **f = src->frames.reference; *f; f++ )
2414         (*f)->i_reference_count++;
2415     for( x264_frame_t **f = dst->frames.reference; *f; f++ )
2416         x264_frame_push_unused( src, *f );
2417     src->fdec->i_reference_count++;
2418     x264_frame_push_unused( src, dst->fdec );
2419
2420     // copy everything except the per-thread pointers and the constants.
2421     memcpy( &dst->i_frame, &src->i_frame, offsetof(x264_t, mb.type) - offsetof(x264_t, i_frame) );
2422     dst->param = src->param;
2423     dst->stat = src->stat;
2424     dst->pixf = src->pixf;
2425 }
2426
2427 static void x264_thread_sync_stat( x264_t *dst, x264_t *src )
2428 {
2429     if( dst == src )
2430         return;
2431     memcpy( &dst->stat.i_frame_count, &src->stat.i_frame_count, sizeof(dst->stat) - sizeof(dst->stat.frame) );
2432 }
2433
2434 static void *x264_slices_write( x264_t *h )
2435 {
2436     int i_slice_num = 0;
2437     int last_thread_mb = h->sh.i_last_mb;
2438
2439 #if HAVE_VISUALIZE
2440     if( h->param.b_visualize )
2441         if( x264_visualize_init( h ) )
2442             return (void *)-1;
2443 #endif
2444
2445     /* init stats */
2446     memset( &h->stat.frame, 0, sizeof(h->stat.frame) );
2447     h->mb.b_reencode_mb = 0;
2448     while( h->sh.i_first_mb + SLICE_MBAFF*h->mb.i_mb_stride <= last_thread_mb )
2449     {
2450         h->sh.i_last_mb = last_thread_mb;
2451         if( h->param.i_slice_max_mbs )
2452         {
2453             if( SLICE_MBAFF )
2454             {
2455                 // convert first to mbaff form, add slice-max-mbs, then convert back to normal form
2456                 int last_mbaff = 2*(h->sh.i_first_mb % h->mb.i_mb_width)
2457                     + h->mb.i_mb_width*(h->sh.i_first_mb / h->mb.i_mb_width)
2458                     + h->param.i_slice_max_mbs - 1;
2459                 int last_x = (last_mbaff % (2*h->mb.i_mb_width))/2;
2460                 int last_y = (last_mbaff / (2*h->mb.i_mb_width))*2 + 1;
2461                 h->sh.i_last_mb = last_x + h->mb.i_mb_stride*last_y;
2462             }
2463             else
2464                 h->sh.i_last_mb = h->sh.i_first_mb + h->param.i_slice_max_mbs - 1;
2465         }
2466         else if( h->param.i_slice_count && !h->param.b_sliced_threads )
2467         {
2468             int height = h->mb.i_mb_height >> PARAM_INTERLACED;
2469             int width = h->mb.i_mb_width << PARAM_INTERLACED;
2470             i_slice_num++;
2471             h->sh.i_last_mb = (height * i_slice_num + h->param.i_slice_count/2) / h->param.i_slice_count * width - 1;
2472         }
2473         h->sh.i_last_mb = X264_MIN( h->sh.i_last_mb, last_thread_mb );
2474         if( x264_stack_align( x264_slice_write, h ) )
2475             return (void *)-1;
2476         h->sh.i_first_mb = h->sh.i_last_mb + 1;
2477         // if i_first_mb is not the last mb in a row then go to the next mb in MBAFF order
2478         if( SLICE_MBAFF && h->sh.i_first_mb % h->mb.i_mb_width )
2479             h->sh.i_first_mb -= h->mb.i_mb_stride;
2480     }
2481
2482 #if HAVE_VISUALIZE
2483     if( h->param.b_visualize )
2484     {
2485         x264_visualize_show( h );
2486         x264_visualize_close( h );
2487     }
2488 #endif
2489
2490     return (void *)0;
2491 }
2492
2493 static int x264_threaded_slices_write( x264_t *h )
2494 {
2495     /* set first/last mb and sync contexts */
2496     for( int i = 0; i < h->param.i_threads; i++ )
2497     {
2498         x264_t *t = h->thread[i];
2499         if( i )
2500         {
2501             t->param = h->param;
2502             memcpy( &t->i_frame, &h->i_frame, offsetof(x264_t, rc) - offsetof(x264_t, i_frame) );
2503         }
2504         int height = h->mb.i_mb_height >> PARAM_INTERLACED;
2505         t->i_threadslice_start = ((height *  i    + h->param.i_slice_count/2) / h->param.i_threads) << PARAM_INTERLACED;
2506         t->i_threadslice_end   = ((height * (i+1) + h->param.i_slice_count/2) / h->param.i_threads) << PARAM_INTERLACED;
2507         t->sh.i_first_mb = t->i_threadslice_start * h->mb.i_mb_width;
2508         t->sh.i_last_mb  =   t->i_threadslice_end * h->mb.i_mb_width - 1;
2509     }
2510
2511     x264_stack_align( x264_analyse_weight_frame, h, h->mb.i_mb_height*16 + 16 );
2512
2513     x264_threads_distribute_ratecontrol( h );
2514
2515     /* dispatch */
2516     for( int i = 0; i < h->param.i_threads; i++ )
2517     {
2518         x264_threadpool_run( h->threadpool, (void*)x264_slices_write, h->thread[i] );
2519         h->thread[i]->b_thread_active = 1;
2520     }
2521     for( int i = 0; i < h->param.i_threads; i++ )
2522     {
2523         h->thread[i]->b_thread_active = 0;
2524         if( (intptr_t)x264_threadpool_wait( h->threadpool, h->thread[i] ) )
2525             return -1;
2526     }
2527
2528     /* Go back and fix up the hpel on the borders between slices. */
2529     for( int i = 1; i < h->param.i_threads; i++ )
2530     {
2531         x264_fdec_filter_row( h->thread[i], h->thread[i]->i_threadslice_start + 1, 0 );
2532         if( SLICE_MBAFF )
2533             x264_fdec_filter_row( h->thread[i], h->thread[i]->i_threadslice_start + 2, 0 );
2534     }
2535
2536     x264_threads_merge_ratecontrol( h );
2537
2538     for( int i = 1; i < h->param.i_threads; i++ )
2539     {
2540         x264_t *t = h->thread[i];
2541         for( int j = 0; j < t->out.i_nal; j++ )
2542         {
2543             h->out.nal[h->out.i_nal] = t->out.nal[j];
2544             h->out.i_nal++;
2545             x264_nal_check_buffer( h );
2546         }
2547         /* All entries in stat.frame are ints except for ssd/ssim. */
2548         for( int j = 0; j < (offsetof(x264_t,stat.frame.i_ssd) - offsetof(x264_t,stat.frame.i_mv_bits)) / sizeof(int); j++ )
2549             ((int*)&h->stat.frame)[j] += ((int*)&t->stat.frame)[j];
2550         for( int j = 0; j < 3; j++ )
2551             h->stat.frame.i_ssd[j] += t->stat.frame.i_ssd[j];
2552         h->stat.frame.f_ssim += t->stat.frame.f_ssim;
2553         h->stat.frame.i_ssim_cnt += t->stat.frame.i_ssim_cnt;
2554     }
2555
2556     return 0;
2557 }
2558
2559 void x264_encoder_intra_refresh( x264_t *h )
2560 {
2561     h = h->thread[h->i_thread_phase];
2562     h->b_queued_intra_refresh = 1;
2563 }
2564
2565 int x264_encoder_invalidate_reference( x264_t *h, int64_t pts )
2566 {
2567     if( h->param.i_bframe )
2568     {
2569         x264_log( h, X264_LOG_ERROR, "x264_encoder_invalidate_reference is not supported with B-frames enabled\n" );
2570         return -1;
2571     }
2572     if( h->param.b_intra_refresh )
2573     {
2574         x264_log( h, X264_LOG_ERROR, "x264_encoder_invalidate_reference is not supported with intra refresh enabled\n" );
2575         return -1;
2576     }
2577     h = h->thread[h->i_thread_phase];
2578     if( pts >= h->i_last_idr_pts )
2579     {
2580         for( int i = 0; h->frames.reference[i]; i++ )
2581             if( pts <= h->frames.reference[i]->i_pts )
2582                 h->frames.reference[i]->b_corrupt = 1;
2583         if( pts <= h->fdec->i_pts )
2584             h->fdec->b_corrupt = 1;
2585     }
2586     return 0;
2587 }
2588
2589 /****************************************************************************
2590  * x264_encoder_encode:
2591  *  XXX: i_poc   : is the poc of the current given picture
2592  *       i_frame : is the number of the frame being coded
2593  *  ex:  type frame poc
2594  *       I      0   2*0
2595  *       P      1   2*3
2596  *       B      2   2*1
2597  *       B      3   2*2
2598  *       P      4   2*6
2599  *       B      5   2*4
2600  *       B      6   2*5
2601  ****************************************************************************/
2602 int     x264_encoder_encode( x264_t *h,
2603                              x264_nal_t **pp_nal, int *pi_nal,
2604                              x264_picture_t *pic_in,
2605                              x264_picture_t *pic_out )
2606 {
2607     x264_t *thread_current, *thread_prev, *thread_oldest;
2608     int i_nal_type, i_nal_ref_idc, i_global_qp;
2609     int overhead = NALU_OVERHEAD;
2610
2611     if( h->i_thread_frames > 1 )
2612     {
2613         thread_prev    = h->thread[ h->i_thread_phase ];
2614         h->i_thread_phase = (h->i_thread_phase + 1) % h->i_thread_frames;
2615         thread_current = h->thread[ h->i_thread_phase ];
2616         thread_oldest  = h->thread[ (h->i_thread_phase + 1) % h->i_thread_frames ];
2617         x264_thread_sync_context( thread_current, thread_prev );
2618         x264_thread_sync_ratecontrol( thread_current, thread_prev, thread_oldest );
2619         h = thread_current;
2620     }
2621     else
2622     {
2623         thread_current =
2624         thread_oldest  = h;
2625     }
2626 #if HAVE_MMX
2627     if( h->param.cpu&X264_CPU_SSE_MISALIGN )
2628         x264_cpu_mask_misalign_sse();
2629 #endif
2630
2631     // ok to call this before encoding any frames, since the initial values of fdec have b_kept_as_ref=0
2632     if( x264_reference_update( h ) )
2633         return -1;
2634     h->fdec->i_lines_completed = -1;
2635
2636     /* no data out */
2637     *pi_nal = 0;
2638     *pp_nal = NULL;
2639
2640     /* ------------------- Setup new frame from picture -------------------- */
2641     if( pic_in != NULL )
2642     {
2643         /* 1: Copy the picture to a frame and move it to a buffer */
2644         x264_frame_t *fenc = x264_frame_pop_unused( h, 0 );
2645         if( !fenc )
2646             return -1;
2647
2648         if( x264_frame_copy_picture( h, fenc, pic_in ) < 0 )
2649             return -1;
2650
2651         if( h->param.i_width != 16 * h->mb.i_mb_width ||
2652             h->param.i_height != 16 * h->mb.i_mb_height )
2653             x264_frame_expand_border_mod16( h, fenc );
2654
2655         fenc->i_frame = h->frames.i_input++;
2656
2657         if( fenc->i_frame == 0 )
2658             h->frames.i_first_pts = fenc->i_pts;
2659         if( h->frames.i_bframe_delay && fenc->i_frame == h->frames.i_bframe_delay )
2660             h->frames.i_bframe_delay_time = fenc->i_pts - h->frames.i_first_pts;
2661
2662         if( h->param.b_vfr_input && fenc->i_pts <= h->frames.i_largest_pts )
2663             x264_log( h, X264_LOG_WARNING, "non-strictly-monotonic PTS\n" );
2664
2665         h->frames.i_second_largest_pts = h->frames.i_largest_pts;
2666         h->frames.i_largest_pts = fenc->i_pts;
2667
2668         if( (fenc->i_pic_struct < PIC_STRUCT_AUTO) || (fenc->i_pic_struct > PIC_STRUCT_TRIPLE) )
2669             fenc->i_pic_struct = PIC_STRUCT_AUTO;
2670
2671         if( fenc->i_pic_struct == PIC_STRUCT_AUTO )
2672         {
2673 #if HAVE_INTERLACED
2674             int b_interlaced = fenc->param ? fenc->param->b_interlaced : h->param.b_interlaced;
2675 #else
2676             int b_interlaced = 0;
2677 #endif
2678             if( b_interlaced )
2679             {
2680                 int b_tff = fenc->param ? fenc->param->b_tff : h->param.b_tff;
2681                 fenc->i_pic_struct = b_tff ? PIC_STRUCT_TOP_BOTTOM : PIC_STRUCT_BOTTOM_TOP;
2682             }
2683             else
2684                 fenc->i_pic_struct = PIC_STRUCT_PROGRESSIVE;
2685         }
2686
2687         if( h->param.rc.b_mb_tree && h->param.rc.b_stat_read )
2688         {
2689             if( x264_macroblock_tree_read( h, fenc, pic_in->prop.quant_offsets ) )
2690                 return -1;
2691         }
2692         else
2693             x264_stack_align( x264_adaptive_quant_frame, h, fenc, pic_in->prop.quant_offsets );
2694
2695         if( pic_in->prop.quant_offsets_free )
2696             pic_in->prop.quant_offsets_free( pic_in->prop.quant_offsets );
2697
2698         if( h->frames.b_have_lowres )
2699             x264_frame_init_lowres( h, fenc );
2700
2701         /* 2: Place the frame into the queue for its slice type decision */
2702         x264_lookahead_put_frame( h, fenc );
2703
2704         if( h->frames.i_input <= h->frames.i_delay + 1 - h->i_thread_frames )
2705         {
2706             /* Nothing yet to encode, waiting for filling of buffers */
2707             pic_out->i_type = X264_TYPE_AUTO;
2708             return 0;
2709         }
2710     }
2711     else
2712     {
2713         /* signal kills for lookahead thread */
2714         x264_pthread_mutex_lock( &h->lookahead->ifbuf.mutex );
2715         h->lookahead->b_exit_thread = 1;
2716         x264_pthread_cond_broadcast( &h->lookahead->ifbuf.cv_fill );
2717         x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
2718     }
2719
2720     h->i_frame++;
2721     /* 3: The picture is analyzed in the lookahead */
2722     if( !h->frames.current[0] )
2723         x264_lookahead_get_frames( h );
2724
2725     if( !h->frames.current[0] && x264_lookahead_is_empty( h ) )
2726         return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
2727
2728     /* ------------------- Get frame to be encoded ------------------------- */
2729     /* 4: get picture to encode */
2730     h->fenc = x264_frame_shift( h->frames.current );
2731     if( h->i_frame == h->i_thread_frames - 1 )
2732         h->i_reordered_pts_delay = h->fenc->i_reordered_pts;
2733     if( h->fenc->param )
2734     {
2735         x264_encoder_reconfig( h, h->fenc->param );
2736         if( h->fenc->param->param_free )
2737             h->fenc->param->param_free( h->fenc->param );
2738     }
2739
2740     if( !IS_X264_TYPE_I( h->fenc->i_type ) )
2741     {
2742         int valid_refs_left = 0;
2743         for( int i = 0; h->frames.reference[i]; i++ )
2744             if( !h->frames.reference[i]->b_corrupt )
2745                 valid_refs_left++;
2746         /* No valid reference frames left: force an IDR. */
2747         if( !valid_refs_left )
2748         {
2749             h->fenc->b_keyframe = 1;
2750             h->fenc->i_type = X264_TYPE_IDR;
2751         }
2752     }
2753
2754     if( h->fenc->b_keyframe )
2755     {
2756         h->frames.i_last_keyframe = h->fenc->i_frame;
2757         if( h->fenc->i_type == X264_TYPE_IDR )
2758         {
2759             h->i_frame_num = 0;
2760             h->frames.i_last_idr = h->fenc->i_frame;
2761         }
2762     }
2763     h->sh.i_mmco_command_count =
2764     h->sh.i_mmco_remove_from_end = 0;
2765     h->b_ref_reorder[0] =
2766     h->b_ref_reorder[1] = 0;
2767     h->fdec->i_poc =
2768     h->fenc->i_poc = 2 * ( h->fenc->i_frame - X264_MAX( h->frames.i_last_idr, 0 ) );
2769
2770     /* ------------------- Setup frame context ----------------------------- */
2771     /* 5: Init data dependent of frame type */
2772     if( h->fenc->i_type == X264_TYPE_IDR )
2773     {
2774         /* reset ref pictures */
2775         i_nal_type    = NAL_SLICE_IDR;
2776         i_nal_ref_idc = NAL_PRIORITY_HIGHEST;
2777         h->sh.i_type = SLICE_TYPE_I;
2778         x264_reference_reset( h );
2779         h->frames.i_poc_last_open_gop = -1;
2780     }
2781     else if( h->fenc->i_type == X264_TYPE_I )
2782     {
2783         i_nal_type    = NAL_SLICE;
2784         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
2785         h->sh.i_type = SLICE_TYPE_I;
2786         x264_reference_hierarchy_reset( h );
2787         if( h->param.b_open_gop )
2788             h->frames.i_poc_last_open_gop = h->fenc->b_keyframe ? h->fenc->i_poc : -1;
2789     }
2790     else if( h->fenc->i_type == X264_TYPE_P )
2791     {
2792         i_nal_type    = NAL_SLICE;
2793         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
2794         h->sh.i_type = SLICE_TYPE_P;
2795         x264_reference_hierarchy_reset( h );
2796         h->frames.i_poc_last_open_gop = -1;
2797     }
2798     else if( h->fenc->i_type == X264_TYPE_BREF )
2799     {
2800         i_nal_type    = NAL_SLICE;
2801         i_nal_ref_idc = h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT ? NAL_PRIORITY_LOW : NAL_PRIORITY_HIGH;
2802         h->sh.i_type = SLICE_TYPE_B;
2803         x264_reference_hierarchy_reset( h );
2804     }
2805     else    /* B frame */
2806     {
2807         i_nal_type    = NAL_SLICE;
2808         i_nal_ref_idc = NAL_PRIORITY_DISPOSABLE;
2809         h->sh.i_type = SLICE_TYPE_B;
2810     }
2811
2812     h->fdec->i_type = h->fenc->i_type;
2813     h->fdec->i_frame = h->fenc->i_frame;
2814     h->fenc->b_kept_as_ref =
2815     h->fdec->b_kept_as_ref = i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE && h->param.i_keyint_max > 1;
2816
2817     h->fdec->i_pts = h->fenc->i_pts;
2818     if( h->frames.i_bframe_delay )
2819     {
2820         int64_t *prev_reordered_pts = thread_current->frames.i_prev_reordered_pts;
2821         h->fdec->i_dts = h->i_frame > h->frames.i_bframe_delay
2822                        ? prev_reordered_pts[ (h->i_frame - h->frames.i_bframe_delay) % h->frames.i_bframe_delay ]
2823                        : h->fenc->i_reordered_pts - h->frames.i_bframe_delay_time;
2824         prev_reordered_pts[ h->i_frame % h->frames.i_bframe_delay ] = h->fenc->i_reordered_pts;
2825     }
2826     else
2827         h->fdec->i_dts = h->fenc->i_reordered_pts;
2828     if( h->fenc->i_type == X264_TYPE_IDR )
2829         h->i_last_idr_pts = h->fdec->i_pts;
2830
2831     /* ------------------- Init                ----------------------------- */
2832     /* build ref list 0/1 */
2833     x264_reference_build_list( h, h->fdec->i_poc );
2834
2835     /* ---------------------- Write the bitstream -------------------------- */
2836     /* Init bitstream context */
2837     if( h->param.b_sliced_threads )
2838     {
2839         for( int i = 0; i < h->param.i_threads; i++ )
2840         {
2841             bs_init( &h->thread[i]->out.bs, h->thread[i]->out.p_bitstream, h->thread[i]->out.i_bitstream );
2842             h->thread[i]->out.i_nal = 0;
2843         }
2844     }
2845     else
2846     {
2847         bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
2848         h->out.i_nal = 0;
2849     }
2850
2851     if( h->param.b_aud )
2852     {
2853         int pic_type;
2854
2855         if( h->sh.i_type == SLICE_TYPE_I )
2856             pic_type = 0;
2857         else if( h->sh.i_type == SLICE_TYPE_P )
2858             pic_type = 1;
2859         else if( h->sh.i_type == SLICE_TYPE_B )
2860             pic_type = 2;
2861         else
2862             pic_type = 7;
2863
2864         x264_nal_start( h, NAL_AUD, NAL_PRIORITY_DISPOSABLE );
2865         bs_write( &h->out.bs, 3, pic_type );
2866         bs_rbsp_trailing( &h->out.bs );
2867         if( x264_nal_end( h ) )
2868             return -1;
2869         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2870     }
2871
2872     h->i_nal_type = i_nal_type;
2873     h->i_nal_ref_idc = i_nal_ref_idc;
2874
2875     if( h->param.b_intra_refresh )
2876     {
2877         if( IS_X264_TYPE_I( h->fenc->i_type ) )
2878         {
2879             h->fdec->i_frames_since_pir = 0;
2880             h->b_queued_intra_refresh = 0;
2881             /* PIR is currently only supported with ref == 1, so any intra frame effectively refreshes
2882              * the whole frame and counts as an intra refresh. */
2883             h->fdec->f_pir_position = h->mb.i_mb_width;
2884         }
2885         else if( h->fenc->i_type == X264_TYPE_P )
2886         {
2887             int pocdiff = (h->fdec->i_poc - h->fref[0][0]->i_poc)/2;
2888             float increment = X264_MAX( ((float)h->mb.i_mb_width-1) / h->param.i_keyint_max, 1 );
2889             h->fdec->f_pir_position = h->fref[0][0]->f_pir_position;
2890             h->fdec->i_frames_since_pir = h->fref[0][0]->i_frames_since_pir + pocdiff;
2891             if( h->fdec->i_frames_since_pir >= h->param.i_keyint_max ||
2892                 (h->b_queued_intra_refresh && h->fdec->f_pir_position + 0.5 >= h->mb.i_mb_width) )
2893             {
2894                 h->fdec->f_pir_position = 0;
2895                 h->fdec->i_frames_since_pir = 0;
2896                 h->b_queued_intra_refresh = 0;
2897                 h->fenc->b_keyframe = 1;
2898             }
2899             h->fdec->i_pir_start_col = h->fdec->f_pir_position+0.5;
2900             h->fdec->f_pir_position += increment * pocdiff;
2901             h->fdec->i_pir_end_col = h->fdec->f_pir_position+0.5;
2902             /* If our intra refresh has reached the right side of the frame, we're done. */
2903             if( h->fdec->i_pir_end_col >= h->mb.i_mb_width - 1 )
2904             {
2905                 h->fdec->f_pir_position = h->mb.i_mb_width;
2906                 h->fdec->i_pir_end_col = h->mb.i_mb_width - 1;
2907             }
2908         }
2909     }
2910
2911     if( h->fenc->b_keyframe )
2912     {
2913         /* Write SPS and PPS */
2914         if( h->param.b_repeat_headers )
2915         {
2916             /* generate sequence parameters */
2917             x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
2918             x264_sps_write( &h->out.bs, h->sps );
2919             if( x264_nal_end( h ) )
2920                 return -1;
2921             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
2922
2923             /* generate picture parameters */
2924             x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
2925             x264_pps_write( &h->out.bs, h->sps, h->pps );
2926             if( x264_nal_end( h ) )
2927                 return -1;
2928             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
2929         }
2930
2931         /* when frame threading is used, buffering period sei is written in x264_encoder_frame_end */
2932         if( h->i_thread_frames == 1 && h->sps->vui.b_nal_hrd_parameters_present )
2933         {
2934             x264_hrd_fullness( h );
2935             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2936             x264_sei_buffering_period_write( h, &h->out.bs );
2937             if( x264_nal_end( h ) )
2938                return -1;
2939             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
2940         }
2941     }
2942
2943     /* write extra sei */
2944     for( int i = 0; i < h->fenc->extra_sei.num_payloads; i++ )
2945     {
2946         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2947         x264_sei_write( &h->out.bs, h->fenc->extra_sei.payloads[i].payload, h->fenc->extra_sei.payloads[i].payload_size,
2948                         h->fenc->extra_sei.payloads[i].payload_type );
2949         if( x264_nal_end( h ) )
2950             return -1;
2951         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2952         if( h->fenc->extra_sei.sei_free && h->fenc->extra_sei.payloads[i].payload )
2953             h->fenc->extra_sei.sei_free( h->fenc->extra_sei.payloads[i].payload );
2954     }
2955
2956     if( h->fenc->extra_sei.sei_free && h->fenc->extra_sei.payloads )
2957         h->fenc->extra_sei.sei_free( h->fenc->extra_sei.payloads );
2958
2959     if( h->fenc->b_keyframe )
2960     {
2961         if( h->param.b_repeat_headers && h->fenc->i_frame == 0 )
2962         {
2963             /* identify ourself */
2964             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2965             if( x264_sei_version_write( h, &h->out.bs ) )
2966                 return -1;
2967             if( x264_nal_end( h ) )
2968                 return -1;
2969             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2970         }
2971
2972         if( h->fenc->i_type != X264_TYPE_IDR )
2973         {
2974             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;
2975             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2976             x264_sei_recovery_point_write( h, &h->out.bs, time_to_recovery );
2977             if( x264_nal_end( h ) )
2978                 return -1;
2979             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2980         }
2981
2982         if ( h->param.i_frame_packing >= 0 )
2983         {
2984             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2985             x264_sei_frame_packing_write( h, &h->out.bs );
2986             if( x264_nal_end( h ) )
2987                 return -1;
2988             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2989         }
2990     }
2991
2992     /* generate sei pic timing */
2993     if( h->sps->vui.b_pic_struct_present || h->sps->vui.b_nal_hrd_parameters_present )
2994     {
2995         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2996         x264_sei_pic_timing_write( h, &h->out.bs );
2997         if( x264_nal_end( h ) )
2998             return -1;
2999         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
3000     }
3001
3002     /* As required by Blu-ray. */
3003     if( !IS_X264_TYPE_B( h->fenc->i_type ) && h->b_sh_backup )
3004     {
3005         h->b_sh_backup = 0;
3006         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3007         x264_sei_dec_ref_pic_marking_write( h, &h->out.bs );
3008         if( x264_nal_end( h ) )
3009             return -1;
3010         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
3011     }
3012
3013     if( h->fenc->b_keyframe && h->param.b_intra_refresh )
3014         h->i_cpb_delay_pir_offset = h->fenc->i_cpb_delay;
3015
3016     /* Init the rate control */
3017     /* FIXME: Include slice header bit cost. */
3018     x264_ratecontrol_start( h, h->fenc->i_qpplus1, overhead*8 );
3019     i_global_qp = x264_ratecontrol_qp( h );
3020
3021     pic_out->i_qpplus1 =
3022     h->fdec->i_qpplus1 = i_global_qp + 1;
3023
3024     if( h->param.rc.b_stat_read && h->sh.i_type != SLICE_TYPE_I )
3025     {
3026         x264_reference_build_list_optimal( h );
3027         x264_reference_check_reorder( h );
3028     }
3029
3030     if( h->i_ref[0] )
3031         h->fdec->i_poc_l0ref0 = h->fref[0][0]->i_poc;
3032
3033     /* ------------------------ Create slice header  ----------------------- */
3034     x264_slice_init( h, i_nal_type, i_global_qp );
3035
3036     /*------------------------- Weights -------------------------------------*/
3037     if( h->sh.i_type == SLICE_TYPE_B )
3038         x264_macroblock_bipred_init( h );
3039
3040     x264_weighted_pred_init( h );
3041
3042     if( i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE )
3043         h->i_frame_num++;
3044
3045     /* Write frame */
3046     h->i_threadslice_start = 0;
3047     h->i_threadslice_end = h->mb.i_mb_height;
3048     if( h->i_thread_frames > 1 )
3049     {
3050         x264_threadpool_run( h->threadpool, (void*)x264_slices_write, h );
3051         h->b_thread_active = 1;
3052     }
3053     else if( h->param.b_sliced_threads )
3054     {
3055         if( x264_threaded_slices_write( h ) )
3056             return -1;
3057     }
3058     else
3059         if( (intptr_t)x264_slices_write( h ) )
3060             return -1;
3061
3062     return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
3063 }
3064
3065 static int x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
3066                                    x264_nal_t **pp_nal, int *pi_nal,
3067                                    x264_picture_t *pic_out )
3068 {
3069     char psz_message[80];
3070
3071     if( h->b_thread_active )
3072     {
3073         h->b_thread_active = 0;
3074         if( (intptr_t)x264_threadpool_wait( h->threadpool, h ) )
3075             return -1;
3076     }
3077     if( !h->out.i_nal )
3078     {
3079         pic_out->i_type = X264_TYPE_AUTO;
3080         return 0;
3081     }
3082
3083     x264_emms();
3084     /* generate buffering period sei and insert it into place */
3085     if( h->i_thread_frames > 1 && h->fenc->b_keyframe && h->sps->vui.b_nal_hrd_parameters_present )
3086     {
3087         x264_hrd_fullness( h );
3088         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3089         x264_sei_buffering_period_write( h, &h->out.bs );
3090         if( x264_nal_end( h ) )
3091            return -1;
3092         /* buffering period sei must follow AUD, SPS and PPS and precede all other SEIs */
3093         int idx = 0;
3094         while( h->out.nal[idx].i_type == NAL_AUD ||
3095                h->out.nal[idx].i_type == NAL_SPS ||
3096                h->out.nal[idx].i_type == NAL_PPS )
3097             idx++;
3098         x264_nal_t nal_tmp = h->out.nal[h->out.i_nal-1];
3099         memmove( &h->out.nal[idx+1], &h->out.nal[idx], (h->out.i_nal-idx-1)*sizeof(x264_nal_t) );
3100         h->out.nal[idx] = nal_tmp;
3101     }
3102
3103     int frame_size = x264_encoder_encapsulate_nals( h, 0 );
3104     if( frame_size < 0 )
3105         return -1;
3106
3107     /* Set output picture properties */
3108     pic_out->i_type = h->fenc->i_type;
3109
3110     pic_out->b_keyframe = h->fenc->b_keyframe;
3111     pic_out->i_pic_struct = h->fenc->i_pic_struct;
3112
3113     pic_out->i_pts = h->fdec->i_pts;
3114     pic_out->i_dts = h->fdec->i_dts;
3115
3116     if( pic_out->i_pts < pic_out->i_dts )
3117         x264_log( h, X264_LOG_WARNING, "invalid DTS: PTS is less than DTS\n" );
3118
3119     pic_out->opaque = h->fenc->opaque;
3120
3121     pic_out->img.i_csp = h->fdec->i_csp;
3122 #if HIGH_BIT_DEPTH
3123     pic_out->img.i_csp |= X264_CSP_HIGH_DEPTH;
3124 #endif
3125     pic_out->img.i_plane = h->fdec->i_plane;
3126     for( int i = 0; i < pic_out->img.i_plane; i++ )
3127     {
3128         pic_out->img.i_stride[i] = h->fdec->i_stride[i] * sizeof(pixel);
3129         pic_out->img.plane[i] = (uint8_t*)h->fdec->plane[i];
3130     }
3131
3132     x264_frame_push_unused( thread_current, h->fenc );
3133
3134     /* ---------------------- Update encoder state ------------------------- */
3135
3136     /* update rc */
3137     int filler = 0;
3138     if( x264_ratecontrol_end( h, frame_size * 8, &filler ) < 0 )
3139         return -1;
3140
3141     pic_out->hrd_timing = h->fenc->hrd_timing;
3142
3143     while( filler > 0 )
3144     {
3145         int f, overhead;
3146         overhead = (FILLER_OVERHEAD - h->param.b_annexb);
3147         if( h->param.i_slice_max_size && filler > h->param.i_slice_max_size )
3148         {
3149             int next_size = filler - h->param.i_slice_max_size;
3150             int overflow = X264_MAX( overhead - next_size, 0 );
3151             f = h->param.i_slice_max_size - overhead - overflow;
3152         }
3153         else
3154             f = X264_MAX( 0, filler - overhead );
3155
3156         x264_nal_start( h, NAL_FILLER, NAL_PRIORITY_DISPOSABLE );
3157         x264_filler_write( h, &h->out.bs, f );
3158         if( x264_nal_end( h ) )
3159             return -1;
3160         int total_size = x264_encoder_encapsulate_nals( h, h->out.i_nal-1 );
3161         if( total_size < 0 )
3162             return -1;
3163         frame_size += total_size;
3164         filler -= total_size;
3165     }
3166
3167     /* End bitstream, set output  */
3168     *pi_nal = h->out.i_nal;
3169     *pp_nal = h->out.nal;
3170
3171     h->out.i_nal = 0;
3172
3173     x264_noise_reduction_update( h );
3174
3175     /* ---------------------- Compute/Print statistics --------------------- */
3176     x264_thread_sync_stat( h, h->thread[0] );
3177
3178     /* Slice stat */
3179     h->stat.i_frame_count[h->sh.i_type]++;
3180     h->stat.i_frame_size[h->sh.i_type] += frame_size;
3181     h->stat.f_frame_qp[h->sh.i_type] += h->fdec->f_qp_avg_aq;
3182
3183     for( int i = 0; i < X264_MBTYPE_MAX; i++ )
3184         h->stat.i_mb_count[h->sh.i_type][i] += h->stat.frame.i_mb_count[i];
3185     for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
3186         h->stat.i_mb_partition[h->sh.i_type][i] += h->stat.frame.i_mb_partition[i];
3187     for( int i = 0; i < 2; i++ )
3188         h->stat.i_mb_count_8x8dct[i] += h->stat.frame.i_mb_count_8x8dct[i];
3189     for( int i = 0; i < 6; i++ )
3190         h->stat.i_mb_cbp[i] += h->stat.frame.i_mb_cbp[i];
3191     for( int i = 0; i < 4; i++ )
3192         for( int j = 0; j < 13; j++ )
3193             h->stat.i_mb_pred_mode[i][j] += h->stat.frame.i_mb_pred_mode[i][j];
3194     if( h->sh.i_type != SLICE_TYPE_I )
3195         for( int i_list = 0; i_list < 2; i_list++ )
3196             for( int i = 0; i < X264_REF_MAX*2; i++ )
3197                 h->stat.i_mb_count_ref[h->sh.i_type][i_list][i] += h->stat.frame.i_mb_count_ref[i_list][i];
3198     for( int i = 0; i < 3; i++ )
3199         h->stat.i_mb_field[i] += h->stat.frame.i_mb_field[i];
3200     if( h->sh.i_type == SLICE_TYPE_P && h->param.analyse.i_weighted_pred >= X264_WEIGHTP_SIMPLE )
3201     {
3202         h->stat.i_wpred[0] += !!h->sh.weight[0][0].weightfn;
3203         h->stat.i_wpred[1] += !!h->sh.weight[0][1].weightfn || !!h->sh.weight[0][2].weightfn;
3204     }
3205     if( h->sh.i_type == SLICE_TYPE_B )
3206     {
3207         h->stat.i_direct_frames[ h->sh.b_direct_spatial_mv_pred ] ++;
3208         if( h->mb.b_direct_auto_write )
3209         {
3210             //FIXME somewhat arbitrary time constants
3211             if( h->stat.i_direct_score[0] + h->stat.i_direct_score[1] > h->mb.i_mb_count )
3212                 for( int i = 0; i < 2; i++ )
3213                     h->stat.i_direct_score[i] = h->stat.i_direct_score[i] * 9/10;
3214             for( int i = 0; i < 2; i++ )
3215                 h->stat.i_direct_score[i] += h->stat.frame.i_direct_score[i];
3216         }
3217     }
3218     else
3219         h->stat.i_consecutive_bframes[h->fenc->i_bframes]++;
3220
3221     psz_message[0] = '\0';
3222     double dur = h->fenc->f_duration;
3223     h->stat.f_frame_duration[h->sh.i_type] += dur;
3224     if( h->param.analyse.b_psnr )
3225     {
3226         int64_t ssd[3] =
3227         {
3228             h->stat.frame.i_ssd[0],
3229             h->stat.frame.i_ssd[1],
3230             h->stat.frame.i_ssd[2],
3231         };
3232         int luma_size = h->param.i_width * h->param.i_height;
3233         int chroma_size = CHROMA_SIZE( luma_size );
3234         pic_out->prop.f_psnr[0] = x264_psnr( ssd[0], luma_size );
3235         pic_out->prop.f_psnr[1] = x264_psnr( ssd[1], chroma_size );
3236         pic_out->prop.f_psnr[2] = x264_psnr( ssd[2], chroma_size );
3237         pic_out->prop.f_psnr_avg = x264_psnr( ssd[0] + ssd[1] + ssd[2], luma_size + chroma_size*2 );
3238
3239         h->stat.f_ssd_global[h->sh.i_type]   += dur * (ssd[0] + ssd[1] + ssd[2]);
3240         h->stat.f_psnr_average[h->sh.i_type] += dur * pic_out->prop.f_psnr_avg;
3241         h->stat.f_psnr_mean_y[h->sh.i_type]  += dur * pic_out->prop.f_psnr[0];
3242         h->stat.f_psnr_mean_u[h->sh.i_type]  += dur * pic_out->prop.f_psnr[1];
3243         h->stat.f_psnr_mean_v[h->sh.i_type]  += dur * pic_out->prop.f_psnr[2];
3244
3245         snprintf( psz_message, 80, " PSNR Y:%5.2f U:%5.2f V:%5.2f", pic_out->prop.f_psnr[0],
3246                                                                     pic_out->prop.f_psnr[1],
3247                                                                     pic_out->prop.f_psnr[2] );
3248     }
3249
3250     if( h->param.analyse.b_ssim )
3251     {
3252         pic_out->prop.f_ssim = h->stat.frame.f_ssim / h->stat.frame.i_ssim_cnt;
3253         h->stat.f_ssim_mean_y[h->sh.i_type] += pic_out->prop.f_ssim * dur;
3254         snprintf( psz_message + strlen(psz_message), 80 - strlen(psz_message),
3255                   " SSIM Y:%.5f", pic_out->prop.f_ssim );
3256     }
3257     psz_message[79] = '\0';
3258
3259     x264_log( h, X264_LOG_DEBUG,
3260                   "frame=%4d QP=%.2f NAL=%d Slice:%c Poc:%-3d I:%-4d P:%-4d SKIP:%-4d size=%d bytes%s\n",
3261               h->i_frame,
3262               h->fdec->f_qp_avg_aq,
3263               h->i_nal_ref_idc,
3264               h->sh.i_type == SLICE_TYPE_I ? 'I' : (h->sh.i_type == SLICE_TYPE_P ? 'P' : 'B' ),
3265               h->fdec->i_poc,
3266               h->stat.frame.i_mb_count_i,
3267               h->stat.frame.i_mb_count_p,
3268               h->stat.frame.i_mb_count_skip,
3269               frame_size,
3270               psz_message );
3271
3272     // keep stats all in one place
3273     x264_thread_sync_stat( h->thread[0], h );
3274     // for the use of the next frame
3275     x264_thread_sync_stat( thread_current, h );
3276
3277 #ifdef DEBUG_MB_TYPE
3278 {
3279     static const char mb_chars[] = { 'i', 'i', 'I', 'C', 'P', '8', 'S',
3280         'D', '<', 'X', 'B', 'X', '>', 'B', 'B', 'B', 'B', '8', 'S' };
3281     for( int mb_xy = 0; mb_xy < h->mb.i_mb_width * h->mb.i_mb_height; mb_xy++ )
3282     {
3283         if( h->mb.type[mb_xy] < X264_MBTYPE_MAX && h->mb.type[mb_xy] >= 0 )
3284             fprintf( stderr, "%c ", mb_chars[ h->mb.type[mb_xy] ] );
3285         else
3286             fprintf( stderr, "? " );
3287
3288         if( (mb_xy+1) % h->mb.i_mb_width == 0 )
3289             fprintf( stderr, "\n" );
3290     }
3291 }
3292 #endif
3293
3294     /* Remove duplicates, must be done near the end as breaks h->fref0 array
3295      * by freeing some of its pointers. */
3296     for( int i = 0; i < h->i_ref[0]; i++ )
3297         if( h->fref[0][i] && h->fref[0][i]->b_duplicate )
3298         {
3299             x264_frame_push_blank_unused( h, h->fref[0][i] );
3300             h->fref[0][i] = 0;
3301         }
3302
3303     if( h->param.psz_dump_yuv )
3304         x264_frame_dump( h );
3305     x264_emms();
3306
3307     return frame_size;
3308 }
3309
3310 static void x264_print_intra( int64_t *i_mb_count, double i_count, int b_print_pcm, char *intra )
3311 {
3312     intra += sprintf( intra, "I16..4%s: %4.1f%% %4.1f%% %4.1f%%",
3313         b_print_pcm ? "..PCM" : "",
3314         i_mb_count[I_16x16]/ i_count,
3315         i_mb_count[I_8x8]  / i_count,
3316         i_mb_count[I_4x4]  / i_count );
3317     if( b_print_pcm )
3318         sprintf( intra, " %4.1f%%", i_mb_count[I_PCM]  / i_count );
3319 }
3320
3321 /****************************************************************************
3322  * x264_encoder_close:
3323  ****************************************************************************/
3324 void    x264_encoder_close  ( x264_t *h )
3325 {
3326     int64_t i_yuv_size = FRAME_SIZE( h->param.i_width * h->param.i_height );
3327     int64_t i_mb_count_size[2][7] = {{0}};
3328     char buf[200];
3329     int b_print_pcm = h->stat.i_mb_count[SLICE_TYPE_I][I_PCM]
3330                    || h->stat.i_mb_count[SLICE_TYPE_P][I_PCM]
3331                    || h->stat.i_mb_count[SLICE_TYPE_B][I_PCM];
3332
3333     x264_lookahead_delete( h );
3334
3335     if( h->param.i_threads > 1 )
3336         x264_threadpool_delete( h->threadpool );
3337     if( h->i_thread_frames > 1 )
3338     {
3339         for( int i = 0; i < h->i_thread_frames; i++ )
3340             if( h->thread[i]->b_thread_active )
3341             {
3342                 assert( h->thread[i]->fenc->i_reference_count == 1 );
3343                 x264_frame_delete( h->thread[i]->fenc );
3344             }
3345
3346         x264_t *thread_prev = h->thread[h->i_thread_phase];
3347         x264_thread_sync_ratecontrol( h, thread_prev, h );
3348         x264_thread_sync_ratecontrol( thread_prev, thread_prev, h );
3349         h->i_frame = thread_prev->i_frame + 1 - h->i_thread_frames;
3350     }
3351     h->i_frame++;
3352
3353     /* Slices used and PSNR */
3354     for( int i = 0; i < 3; i++ )
3355     {
3356         static const uint8_t slice_order[] = { SLICE_TYPE_I, SLICE_TYPE_P, SLICE_TYPE_B };
3357         int i_slice = slice_order[i];
3358
3359         if( h->stat.i_frame_count[i_slice] > 0 )
3360         {
3361             int i_count = h->stat.i_frame_count[i_slice];
3362             double dur =  h->stat.f_frame_duration[i_slice];
3363             if( h->param.analyse.b_psnr )
3364             {
3365                 x264_log( h, X264_LOG_INFO,
3366                           "frame %c:%-5d Avg QP:%5.2f  size:%6.0f  PSNR Mean Y:%5.2f U:%5.2f V:%5.2f Avg:%5.2f Global:%5.2f\n",
3367                           slice_type_to_char[i_slice],
3368                           i_count,
3369                           h->stat.f_frame_qp[i_slice] / i_count,
3370                           (double)h->stat.i_frame_size[i_slice] / i_count,
3371                           h->stat.f_psnr_mean_y[i_slice] / dur, h->stat.f_psnr_mean_u[i_slice] / dur, h->stat.f_psnr_mean_v[i_slice] / dur,
3372                           h->stat.f_psnr_average[i_slice] / dur,
3373                           x264_psnr( h->stat.f_ssd_global[i_slice], dur * i_yuv_size ) );
3374             }
3375             else
3376             {
3377                 x264_log( h, X264_LOG_INFO,
3378                           "frame %c:%-5d Avg QP:%5.2f  size:%6.0f\n",
3379                           slice_type_to_char[i_slice],
3380                           i_count,
3381                           h->stat.f_frame_qp[i_slice] / i_count,
3382                           (double)h->stat.i_frame_size[i_slice] / i_count );
3383             }
3384         }
3385     }
3386     if( h->param.i_bframe && h->stat.i_frame_count[SLICE_TYPE_B] )
3387     {
3388         char *p = buf;
3389         int den = 0;
3390         // weight by number of frames (including the I/P-frames) that are in a sequence of N B-frames
3391         for( int i = 0; i <= h->param.i_bframe; i++ )
3392             den += (i+1) * h->stat.i_consecutive_bframes[i];
3393         for( int i = 0; i <= h->param.i_bframe; i++ )
3394             p += sprintf( p, " %4.1f%%", 100. * (i+1) * h->stat.i_consecutive_bframes[i] / den );
3395         x264_log( h, X264_LOG_INFO, "consecutive B-frames:%s\n", buf );
3396     }
3397
3398     for( int i_type = 0; i_type < 2; i_type++ )
3399         for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
3400         {
3401             if( i == D_DIRECT_8x8 ) continue; /* direct is counted as its own type */
3402             i_mb_count_size[i_type][x264_mb_partition_pixel_table[i]] += h->stat.i_mb_partition[i_type][i];
3403         }
3404
3405     /* MB types used */
3406     if( h->stat.i_frame_count[SLICE_TYPE_I] > 0 )
3407     {
3408         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_I];
3409         double i_count = h->stat.i_frame_count[SLICE_TYPE_I] * h->mb.i_mb_count / 100.0;
3410         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
3411         x264_log( h, X264_LOG_INFO, "mb I  %s\n", buf );
3412     }
3413     if( h->stat.i_frame_count[SLICE_TYPE_P] > 0 )
3414     {
3415         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_P];
3416         double i_count = h->stat.i_frame_count[SLICE_TYPE_P] * h->mb.i_mb_count / 100.0;
3417         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_P];
3418         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
3419         x264_log( h, X264_LOG_INFO,
3420                   "mb P  %s  P16..4: %4.1f%% %4.1f%% %4.1f%% %4.1f%% %4.1f%%    skip:%4.1f%%\n",
3421                   buf,
3422                   i_mb_size[PIXEL_16x16] / (i_count*4),
3423                   (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
3424                   i_mb_size[PIXEL_8x8] / (i_count*4),
3425                   (i_mb_size[PIXEL_8x4] + i_mb_size[PIXEL_4x8]) / (i_count*4),
3426                   i_mb_size[PIXEL_4x4] / (i_count*4),
3427                   i_mb_count[P_SKIP] / i_count );
3428     }
3429     if( h->stat.i_frame_count[SLICE_TYPE_B] > 0 )
3430     {
3431         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_B];
3432         double i_count = h->stat.i_frame_count[SLICE_TYPE_B] * h->mb.i_mb_count / 100.0;
3433         double i_mb_list_count;
3434         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_B];
3435         int64_t list_count[3] = {0}; /* 0 == L0, 1 == L1, 2 == BI */
3436         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
3437         for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
3438             for( int j = 0; j < 2; j++ )
3439             {
3440                 int l0 = x264_mb_type_list_table[i][0][j];
3441                 int l1 = x264_mb_type_list_table[i][1][j];
3442                 if( l0 || l1 )
3443                     list_count[l1+l0*l1] += h->stat.i_mb_count[SLICE_TYPE_B][i] * 2;
3444             }
3445         list_count[0] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L0_8x8];
3446         list_count[1] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L1_8x8];
3447         list_count[2] += h->stat.i_mb_partition[SLICE_TYPE_B][D_BI_8x8];
3448         i_mb_count[B_DIRECT] += (h->stat.i_mb_partition[SLICE_TYPE_B][D_DIRECT_8x8]+2)/4;
3449         i_mb_list_count = (list_count[0] + list_count[1] + list_count[2]) / 100.0;
3450         sprintf( buf + strlen(buf), "  B16..8: %4.1f%% %4.1f%% %4.1f%%  direct:%4.1f%%  skip:%4.1f%%",
3451                  i_mb_size[PIXEL_16x16] / (i_count*4),
3452                  (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
3453                  i_mb_size[PIXEL_8x8] / (i_count*4),
3454                  i_mb_count[B_DIRECT] / i_count,
3455                  i_mb_count[B_SKIP]   / i_count );
3456         if( i_mb_list_count != 0 )
3457             sprintf( buf + strlen(buf), "  L0:%4.1f%% L1:%4.1f%% BI:%4.1f%%",
3458                      list_count[0] / i_mb_list_count,
3459                      list_count[1] / i_mb_list_count,
3460                      list_count[2] / i_mb_list_count );
3461         x264_log( h, X264_LOG_INFO, "mb B  %s\n", buf );
3462     }
3463
3464     x264_ratecontrol_summary( h );
3465
3466     if( h->stat.i_frame_count[SLICE_TYPE_I] + h->stat.i_frame_count[SLICE_TYPE_P] + h->stat.i_frame_count[SLICE_TYPE_B] > 0 )
3467     {
3468 #define SUM3(p) (p[SLICE_TYPE_I] + p[SLICE_TYPE_P] + p[SLICE_TYPE_B])
3469 #define SUM3b(p,o) (p[SLICE_TYPE_I][o] + p[SLICE_TYPE_P][o] + p[SLICE_TYPE_B][o])
3470         int64_t i_i8x8 = SUM3b( h->stat.i_mb_count, I_8x8 );
3471         int64_t i_intra = i_i8x8 + SUM3b( h->stat.i_mb_count, I_4x4 )
3472                                  + SUM3b( h->stat.i_mb_count, I_16x16 );
3473         int64_t i_all_intra = i_intra + SUM3b( h->stat.i_mb_count, I_PCM);
3474         int64_t i_skip = SUM3b( h->stat.i_mb_count, P_SKIP )
3475                        + SUM3b( h->stat.i_mb_count, B_SKIP );
3476         const int i_count = h->stat.i_frame_count[SLICE_TYPE_I] +
3477                             h->stat.i_frame_count[SLICE_TYPE_P] +
3478                             h->stat.i_frame_count[SLICE_TYPE_B];
3479         int64_t i_mb_count = (int64_t)i_count * h->mb.i_mb_count;
3480         int64_t i_inter = i_mb_count - i_skip - i_intra;
3481         const double duration = h->stat.f_frame_duration[SLICE_TYPE_I] +
3482                                 h->stat.f_frame_duration[SLICE_TYPE_P] +
3483                                 h->stat.f_frame_duration[SLICE_TYPE_B];
3484         float f_bitrate = SUM3(h->stat.i_frame_size) / duration / 125;
3485
3486         if( PARAM_INTERLACED )
3487         {
3488             char *fieldstats = buf;
3489             fieldstats[0] = 0;
3490             if( i_inter )
3491                 fieldstats += sprintf( fieldstats, " inter:%.1f%%", h->stat.i_mb_field[1] * 100.0 / i_inter );
3492             if( i_skip )
3493                 fieldstats += sprintf( fieldstats, " skip:%.1f%%", h->stat.i_mb_field[2] * 100.0 / i_skip );
3494             x264_log( h, X264_LOG_INFO, "field mbs: intra: %.1f%%%s\n",
3495                       h->stat.i_mb_field[0] * 100.0 / i_intra, buf );
3496         }
3497
3498         if( h->pps->b_transform_8x8_mode )
3499         {
3500             buf[0] = 0;
3501             if( h->stat.i_mb_count_8x8dct[0] )
3502                 sprintf( buf, " inter:%.1f%%", 100. * h->stat.i_mb_count_8x8dct[1] / h->stat.i_mb_count_8x8dct[0] );
3503             x264_log( h, X264_LOG_INFO, "8x8 transform intra:%.1f%%%s\n", 100. * i_i8x8 / i_intra, buf );
3504         }
3505
3506         if( (h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO ||
3507             (h->stat.i_direct_frames[0] && h->stat.i_direct_frames[1]))
3508             && h->stat.i_frame_count[SLICE_TYPE_B] )
3509         {
3510             x264_log( h, X264_LOG_INFO, "direct mvs  spatial:%.1f%% temporal:%.1f%%\n",
3511                       h->stat.i_direct_frames[1] * 100. / h->stat.i_frame_count[SLICE_TYPE_B],
3512                       h->stat.i_direct_frames[0] * 100. / h->stat.i_frame_count[SLICE_TYPE_B] );
3513         }
3514
3515         buf[0] = 0;
3516         int csize = CHROMA444 ? 4 : 1;
3517         if( i_mb_count != i_all_intra )
3518             sprintf( buf, " inter: %.1f%% %.1f%% %.1f%%",
3519                      h->stat.i_mb_cbp[1] * 100.0 / ((i_mb_count - i_all_intra)*4),
3520                      h->stat.i_mb_cbp[3] * 100.0 / ((i_mb_count - i_all_intra)*csize),
3521                      h->stat.i_mb_cbp[5] * 100.0 / ((i_mb_count - i_all_intra)*csize) );
3522         x264_log( h, X264_LOG_INFO, "coded y,%s,%s intra: %.1f%% %.1f%% %.1f%%%s\n",
3523                   CHROMA444?"u":"uvDC", CHROMA444?"v":"uvAC",
3524                   h->stat.i_mb_cbp[0] * 100.0 / (i_all_intra*4),
3525                   h->stat.i_mb_cbp[2] * 100.0 / (i_all_intra*csize),
3526                   h->stat.i_mb_cbp[4] * 100.0 / (i_all_intra*csize), buf );
3527
3528         int64_t fixed_pred_modes[4][9] = {{0}};
3529         int64_t sum_pred_modes[4] = {0};
3530         for( int i = 0; i <= I_PRED_16x16_DC_128; i++ )
3531         {
3532             fixed_pred_modes[0][x264_mb_pred_mode16x16_fix[i]] += h->stat.i_mb_pred_mode[0][i];
3533             sum_pred_modes[0] += h->stat.i_mb_pred_mode[0][i];
3534         }
3535         if( sum_pred_modes[0] )
3536             x264_log( h, X264_LOG_INFO, "i16 v,h,dc,p: %2.0f%% %2.0f%% %2.0f%% %2.0f%%\n",
3537                       fixed_pred_modes[0][0] * 100.0 / sum_pred_modes[0],
3538                       fixed_pred_modes[0][1] * 100.0 / sum_pred_modes[0],
3539                       fixed_pred_modes[0][2] * 100.0 / sum_pred_modes[0],
3540                       fixed_pred_modes[0][3] * 100.0 / sum_pred_modes[0] );
3541         for( int i = 1; i <= 2; i++ )
3542         {
3543             for( int j = 0; j <= I_PRED_8x8_DC_128; j++ )
3544             {
3545                 fixed_pred_modes[i][x264_mb_pred_mode4x4_fix(j)] += h->stat.i_mb_pred_mode[i][j];
3546                 sum_pred_modes[i] += h->stat.i_mb_pred_mode[i][j];
3547             }
3548             if( sum_pred_mo