607ece1412568327729c31428a85633183f1f82f
[x262.git] / encoder / encoder.c
1 /*****************************************************************************
2  * encoder.c: top-level encoder functions
3  *****************************************************************************
4  * Copyright (C) 2003-2011 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         /* Due to the proliferation of broken players that don't handle dupes properly. */
647         h->param.analyse.i_weighted_pred = X264_MIN( h->param.analyse.i_weighted_pred, X264_WEIGHTP_SIMPLE );
648         if( h->param.b_fake_interlaced )
649             h->param.b_pic_struct = 1;
650     }
651
652     h->param.i_frame_reference = x264_clip3( h->param.i_frame_reference, 1, X264_REF_MAX );
653     h->param.i_dpb_size = x264_clip3( h->param.i_dpb_size, 1, X264_REF_MAX );
654     if( h->param.i_scenecut_threshold < 0 )
655         h->param.i_scenecut_threshold = 0;
656     h->param.analyse.i_direct_mv_pred = x264_clip3( h->param.analyse.i_direct_mv_pred, X264_DIRECT_PRED_NONE, X264_DIRECT_PRED_AUTO );
657     if( !h->param.analyse.i_subpel_refine && h->param.analyse.i_direct_mv_pred > X264_DIRECT_PRED_SPATIAL )
658     {
659         x264_log( h, X264_LOG_WARNING, "subme=0 + direct=temporal is not supported\n" );
660         h->param.analyse.i_direct_mv_pred = X264_DIRECT_PRED_SPATIAL;
661     }
662     h->param.i_bframe = x264_clip3( h->param.i_bframe, 0, X264_MIN( X264_BFRAME_MAX, h->param.i_keyint_max-1 ) );
663     h->param.i_bframe_bias = x264_clip3( h->param.i_bframe_bias, -90, 100 );
664     if( h->param.i_bframe <= 1 )
665         h->param.i_bframe_pyramid = X264_B_PYRAMID_NONE;
666     h->param.i_bframe_pyramid = x264_clip3( h->param.i_bframe_pyramid, X264_B_PYRAMID_NONE, X264_B_PYRAMID_NORMAL );
667     h->param.i_bframe_adaptive = x264_clip3( h->param.i_bframe_adaptive, X264_B_ADAPT_NONE, X264_B_ADAPT_TRELLIS );
668     if( !h->param.i_bframe )
669     {
670         h->param.i_bframe_adaptive = X264_B_ADAPT_NONE;
671         h->param.analyse.i_direct_mv_pred = 0;
672         h->param.analyse.b_weighted_bipred = 0;
673         h->param.b_open_gop = 0;
674     }
675     if( h->param.b_intra_refresh && h->param.i_bframe_pyramid == X264_B_PYRAMID_NORMAL )
676     {
677         x264_log( h, X264_LOG_WARNING, "b-pyramid normal + intra-refresh is not supported\n" );
678         h->param.i_bframe_pyramid = X264_B_PYRAMID_STRICT;
679     }
680     if( h->param.b_intra_refresh && (h->param.i_frame_reference > 1 || h->param.i_dpb_size > 1) )
681     {
682         x264_log( h, X264_LOG_WARNING, "ref > 1 + intra-refresh is not supported\n" );
683         h->param.i_frame_reference = 1;
684         h->param.i_dpb_size = 1;
685     }
686     if( h->param.b_intra_refresh && h->param.b_open_gop )
687     {
688         x264_log( h, X264_LOG_WARNING, "intra-refresh is not compatible with open-gop\n" );
689         h->param.b_open_gop = 0;
690     }
691     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;
692     if( h->param.i_keyint_min == X264_KEYINT_MIN_AUTO )
693         h->param.i_keyint_min = X264_MIN( h->param.i_keyint_max / 10, fps );
694     h->param.i_keyint_min = x264_clip3( h->param.i_keyint_min, 1, h->param.i_keyint_max/2+1 );
695     h->param.rc.i_lookahead = x264_clip3( h->param.rc.i_lookahead, 0, X264_LOOKAHEAD_MAX );
696     {
697         int maxrate = X264_MAX( h->param.rc.i_vbv_max_bitrate, h->param.rc.i_bitrate );
698         float bufsize = maxrate ? (float)h->param.rc.i_vbv_buffer_size / maxrate : 0;
699         h->param.rc.i_lookahead = X264_MIN( h->param.rc.i_lookahead, X264_MAX( h->param.i_keyint_max, bufsize*fps ) );
700     }
701
702     if( !h->param.i_timebase_num || !h->param.i_timebase_den || !(h->param.b_vfr_input || h->param.b_pulldown) )
703     {
704         h->param.i_timebase_num = h->param.i_fps_den;
705         h->param.i_timebase_den = h->param.i_fps_num;
706     }
707
708     h->param.rc.f_qcompress = x264_clip3f( h->param.rc.f_qcompress, 0.0, 1.0 );
709     if( h->param.i_keyint_max == 1 || h->param.rc.f_qcompress == 1 )
710         h->param.rc.b_mb_tree = 0;
711     if( (!h->param.b_intra_refresh && h->param.i_keyint_max != X264_KEYINT_MAX_INFINITE) &&
712         !h->param.rc.i_lookahead && h->param.rc.b_mb_tree )
713     {
714         x264_log( h, X264_LOG_WARNING, "lookaheadless mb-tree requires intra refresh or infinite keyint\n" );
715         h->param.rc.b_mb_tree = 0;
716     }
717     if( h->param.rc.b_stat_read )
718         h->param.rc.i_lookahead = 0;
719 #if HAVE_THREAD
720     if( h->param.i_sync_lookahead < 0 )
721         h->param.i_sync_lookahead = h->param.i_bframe + 1;
722     h->param.i_sync_lookahead = X264_MIN( h->param.i_sync_lookahead, X264_LOOKAHEAD_MAX );
723     if( h->param.rc.b_stat_read || h->i_thread_frames == 1 )
724         h->param.i_sync_lookahead = 0;
725 #else
726     h->param.i_sync_lookahead = 0;
727 #endif
728
729     h->param.i_deblocking_filter_alphac0 = x264_clip3( h->param.i_deblocking_filter_alphac0, -6, 6 );
730     h->param.i_deblocking_filter_beta    = x264_clip3( h->param.i_deblocking_filter_beta, -6, 6 );
731     h->param.analyse.i_luma_deadzone[0] = x264_clip3( h->param.analyse.i_luma_deadzone[0], 0, 32 );
732     h->param.analyse.i_luma_deadzone[1] = x264_clip3( h->param.analyse.i_luma_deadzone[1], 0, 32 );
733
734     h->param.i_cabac_init_idc = x264_clip3( h->param.i_cabac_init_idc, 0, 2 );
735
736     if( h->param.i_cqm_preset < X264_CQM_FLAT || h->param.i_cqm_preset > X264_CQM_CUSTOM )
737         h->param.i_cqm_preset = X264_CQM_FLAT;
738
739     if( h->param.analyse.i_me_method < X264_ME_DIA ||
740         h->param.analyse.i_me_method > X264_ME_TESA )
741         h->param.analyse.i_me_method = X264_ME_HEX;
742     h->param.analyse.i_me_range = x264_clip3( h->param.analyse.i_me_range, 4, 1024 );
743     if( h->param.analyse.i_me_range > 16 && h->param.analyse.i_me_method <= X264_ME_HEX )
744         h->param.analyse.i_me_range = 16;
745     if( h->param.analyse.i_me_method == X264_ME_TESA &&
746         (h->mb.b_lossless || h->param.analyse.i_subpel_refine <= 1) )
747         h->param.analyse.i_me_method = X264_ME_ESA;
748     h->param.analyse.b_mixed_references = h->param.analyse.b_mixed_references && h->param.i_frame_reference > 1;
749     h->param.analyse.inter &= X264_ANALYSE_PSUB16x16|X264_ANALYSE_PSUB8x8|X264_ANALYSE_BSUB16x16|
750                               X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
751     h->param.analyse.intra &= X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
752     if( !(h->param.analyse.inter & X264_ANALYSE_PSUB16x16) )
753         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
754     if( !h->param.analyse.b_transform_8x8 )
755     {
756         h->param.analyse.inter &= ~X264_ANALYSE_I8x8;
757         h->param.analyse.intra &= ~X264_ANALYSE_I8x8;
758     }
759     h->param.analyse.i_trellis = x264_clip3( h->param.analyse.i_trellis, 0, 2 );
760     h->param.rc.i_aq_mode = x264_clip3( h->param.rc.i_aq_mode, 0, 2 );
761     h->param.rc.f_aq_strength = x264_clip3f( h->param.rc.f_aq_strength, 0, 3 );
762     if( h->param.rc.f_aq_strength == 0 )
763         h->param.rc.i_aq_mode = 0;
764
765     if( h->param.i_log_level < X264_LOG_INFO )
766     {
767         h->param.analyse.b_psnr = 0;
768         h->param.analyse.b_ssim = 0;
769     }
770     /* Warn users trying to measure PSNR/SSIM with psy opts on. */
771     if( b_open && (h->param.analyse.b_psnr || h->param.analyse.b_ssim) )
772     {
773         char *s = NULL;
774
775         if( h->param.analyse.b_psy )
776         {
777             s = h->param.analyse.b_psnr ? "psnr" : "ssim";
778             x264_log( h, X264_LOG_WARNING, "--%s used with psy on: results will be invalid!\n", s );
779         }
780         else if( !h->param.rc.i_aq_mode && h->param.analyse.b_ssim )
781         {
782             x264_log( h, X264_LOG_WARNING, "--ssim used with AQ off: results will be invalid!\n" );
783             s = "ssim";
784         }
785         else if(  h->param.rc.i_aq_mode && h->param.analyse.b_psnr )
786         {
787             x264_log( h, X264_LOG_WARNING, "--psnr used with AQ on: results will be invalid!\n" );
788             s = "psnr";
789         }
790         if( s )
791             x264_log( h, X264_LOG_WARNING, "--tune %s should be used if attempting to benchmark %s!\n", s, s );
792     }
793
794     if( !h->param.analyse.b_psy )
795     {
796         h->param.analyse.f_psy_rd = 0;
797         h->param.analyse.f_psy_trellis = 0;
798     }
799     h->param.analyse.f_psy_rd = x264_clip3f( h->param.analyse.f_psy_rd, 0, 10 );
800     h->param.analyse.f_psy_trellis = x264_clip3f( h->param.analyse.f_psy_trellis, 0, 10 );
801     h->mb.i_psy_rd = h->param.analyse.i_subpel_refine >= 6 ? FIX8( h->param.analyse.f_psy_rd ) : 0;
802     h->mb.i_psy_trellis = h->param.analyse.i_trellis ? FIX8( h->param.analyse.f_psy_trellis / 4 ) : 0;
803     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -32, 32);
804     /* In 4:4:4 mode, chroma gets twice as much resolution, so we can halve its quality. */
805     if( b_open && i_csp >= X264_CSP_I444 && i_csp < X264_CSP_BGR && h->param.analyse.b_psy )
806         h->param.analyse.i_chroma_qp_offset += 6;
807     /* Psy RDO increases overall quantizers to improve the quality of luma--this indirectly hurts chroma quality */
808     /* so we lower the chroma QP offset to compensate */
809     if( b_open && h->mb.i_psy_rd )
810         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_rd < 0.25 ? 1 : 2;
811     /* Psy trellis has a similar effect. */
812     if( b_open && h->mb.i_psy_trellis )
813         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_trellis < 0.25 ? 1 : 2;
814     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -12, 12);
815     /* MB-tree requires AQ to be on, even if the strength is zero. */
816     if( !h->param.rc.i_aq_mode && h->param.rc.b_mb_tree )
817     {
818         h->param.rc.i_aq_mode = 1;
819         h->param.rc.f_aq_strength = 0;
820     }
821     h->param.analyse.i_noise_reduction = x264_clip3( h->param.analyse.i_noise_reduction, 0, 1<<16 );
822     if( h->param.analyse.i_subpel_refine >= 10 && (h->param.analyse.i_trellis != 2 || !h->param.rc.i_aq_mode) )
823         h->param.analyse.i_subpel_refine = 9;
824
825     {
826         const x264_level_t *l = x264_levels;
827         if( h->param.i_level_idc < 0 )
828         {
829             int maxrate_bak = h->param.rc.i_vbv_max_bitrate;
830             if( h->param.rc.i_rc_method == X264_RC_ABR && h->param.rc.i_vbv_buffer_size <= 0 )
831                 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate * 2;
832             x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
833             do h->param.i_level_idc = l->level_idc;
834                 while( l[1].level_idc && x264_validate_levels( h, 0 ) && l++ );
835             h->param.rc.i_vbv_max_bitrate = maxrate_bak;
836         }
837         else
838         {
839             while( l->level_idc && l->level_idc != h->param.i_level_idc )
840                 l++;
841             if( l->level_idc == 0 )
842             {
843                 x264_log( h, X264_LOG_ERROR, "invalid level_idc: %d\n", h->param.i_level_idc );
844                 return -1;
845             }
846         }
847         if( h->param.analyse.i_mv_range <= 0 )
848             h->param.analyse.i_mv_range = l->mv_range >> PARAM_INTERLACED;
849         else
850             h->param.analyse.i_mv_range = x264_clip3(h->param.analyse.i_mv_range, 32, 512 >> PARAM_INTERLACED);
851     }
852
853     h->param.analyse.i_weighted_pred = x264_clip3( h->param.analyse.i_weighted_pred, X264_WEIGHTP_NONE, X264_WEIGHTP_SMART );
854
855     if( PARAM_INTERLACED )
856     {
857         if( h->param.analyse.i_me_method >= X264_ME_ESA )
858         {
859             x264_log( h, X264_LOG_WARNING, "interlace + me=esa is not implemented\n" );
860             h->param.analyse.i_me_method = X264_ME_UMH;
861         }
862         if( h->param.analyse.i_weighted_pred > 0 )
863         {
864             x264_log( h, X264_LOG_WARNING, "interlace + weightp is not implemented\n" );
865             h->param.analyse.i_weighted_pred = X264_WEIGHTP_NONE;
866         }
867     }
868
869     if( !h->param.analyse.i_weighted_pred && h->param.rc.b_mb_tree && h->param.analyse.b_psy )
870         h->param.analyse.i_weighted_pred = X264_WEIGHTP_FAKE;
871
872     if( h->i_thread_frames > 1 )
873     {
874         int r = h->param.analyse.i_mv_range_thread;
875         int r2;
876         if( r <= 0 )
877         {
878             // half of the available space is reserved and divided evenly among the threads,
879             // the rest is allocated to whichever thread is far enough ahead to use it.
880             // reserving more space increases quality for some videos, but costs more time
881             // in thread synchronization.
882             int max_range = (h->param.i_height + X264_THREAD_HEIGHT) / h->i_thread_frames - X264_THREAD_HEIGHT;
883             r = max_range / 2;
884         }
885         r = X264_MAX( r, h->param.analyse.i_me_range );
886         r = X264_MIN( r, h->param.analyse.i_mv_range );
887         // round up to use the whole mb row
888         r2 = (r & ~15) + ((-X264_THREAD_HEIGHT) & 15);
889         if( r2 < r )
890             r2 += 16;
891         x264_log( h, X264_LOG_DEBUG, "using mv_range_thread = %d\n", r2 );
892         h->param.analyse.i_mv_range_thread = r2;
893     }
894
895     if( h->param.rc.f_rate_tolerance < 0 )
896         h->param.rc.f_rate_tolerance = 0;
897     if( h->param.rc.f_qblur < 0 )
898         h->param.rc.f_qblur = 0;
899     if( h->param.rc.f_complexity_blur < 0 )
900         h->param.rc.f_complexity_blur = 0;
901
902     h->param.i_sps_id &= 31;
903
904     if( PARAM_INTERLACED )
905         h->param.b_pic_struct = 1;
906
907     h->param.i_nal_hrd = x264_clip3( h->param.i_nal_hrd, X264_NAL_HRD_NONE, X264_NAL_HRD_CBR );
908
909     if( h->param.i_nal_hrd && !h->param.rc.i_vbv_buffer_size )
910     {
911         x264_log( h, X264_LOG_WARNING, "NAL HRD parameters require VBV parameters\n" );
912         h->param.i_nal_hrd = X264_NAL_HRD_NONE;
913     }
914
915     if( h->param.i_nal_hrd == X264_NAL_HRD_CBR &&
916        (h->param.rc.i_bitrate != h->param.rc.i_vbv_max_bitrate || !h->param.rc.i_vbv_max_bitrate) )
917     {
918         x264_log( h, X264_LOG_WARNING, "CBR HRD requires constant bitrate\n" );
919         h->param.i_nal_hrd = X264_NAL_HRD_VBR;
920     }
921
922     /* ensure the booleans are 0 or 1 so they can be used in math */
923 #define BOOLIFY(x) h->param.x = !!h->param.x
924     BOOLIFY( b_cabac );
925     BOOLIFY( b_constrained_intra );
926     BOOLIFY( b_deblocking_filter );
927     BOOLIFY( b_deterministic );
928     BOOLIFY( b_sliced_threads );
929     BOOLIFY( b_interlaced );
930     BOOLIFY( b_intra_refresh );
931     BOOLIFY( b_visualize );
932     BOOLIFY( b_aud );
933     BOOLIFY( b_repeat_headers );
934     BOOLIFY( b_annexb );
935     BOOLIFY( b_vfr_input );
936     BOOLIFY( b_pulldown );
937     BOOLIFY( b_tff );
938     BOOLIFY( b_pic_struct );
939     BOOLIFY( b_fake_interlaced );
940     BOOLIFY( b_open_gop );
941     BOOLIFY( b_bluray_compat );
942     BOOLIFY( analyse.b_transform_8x8 );
943     BOOLIFY( analyse.b_weighted_bipred );
944     BOOLIFY( analyse.b_chroma_me );
945     BOOLIFY( analyse.b_mixed_references );
946     BOOLIFY( analyse.b_fast_pskip );
947     BOOLIFY( analyse.b_dct_decimate );
948     BOOLIFY( analyse.b_psy );
949     BOOLIFY( analyse.b_psnr );
950     BOOLIFY( analyse.b_ssim );
951     BOOLIFY( rc.b_stat_write );
952     BOOLIFY( rc.b_stat_read );
953     BOOLIFY( rc.b_mb_tree );
954 #undef BOOLIFY
955
956     return 0;
957 }
958
959 static void mbcmp_init( x264_t *h )
960 {
961     int satd = !h->mb.b_lossless && h->param.analyse.i_subpel_refine > 1;
962     memcpy( h->pixf.mbcmp, satd ? h->pixf.satd : h->pixf.sad_aligned, sizeof(h->pixf.mbcmp) );
963     memcpy( h->pixf.mbcmp_unaligned, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.mbcmp_unaligned) );
964     h->pixf.intra_mbcmp_x3_16x16 = satd ? h->pixf.intra_satd_x3_16x16 : h->pixf.intra_sad_x3_16x16;
965     h->pixf.intra_mbcmp_x3_8x16c = satd ? h->pixf.intra_satd_x3_8x16c : h->pixf.intra_sad_x3_8x16c;
966     h->pixf.intra_mbcmp_x3_8x8c  = satd ? h->pixf.intra_satd_x3_8x8c  : h->pixf.intra_sad_x3_8x8c;
967     h->pixf.intra_mbcmp_x3_8x8 = satd ? h->pixf.intra_sa8d_x3_8x8 : h->pixf.intra_sad_x3_8x8;
968     h->pixf.intra_mbcmp_x3_4x4 = satd ? h->pixf.intra_satd_x3_4x4 : h->pixf.intra_sad_x3_4x4;
969     h->pixf.intra_mbcmp_x9_4x4 = h->param.b_cpu_independent || h->mb.b_lossless ? NULL
970                                : satd ? h->pixf.intra_satd_x9_4x4 : h->pixf.intra_sad_x9_4x4;
971     h->pixf.intra_mbcmp_x9_8x8 = h->param.b_cpu_independent || h->mb.b_lossless ? NULL
972                                : satd ? h->pixf.intra_sa8d_x9_8x8 : h->pixf.intra_sad_x9_8x8;
973     satd &= h->param.analyse.i_me_method == X264_ME_TESA;
974     memcpy( h->pixf.fpelcmp, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.fpelcmp) );
975     memcpy( h->pixf.fpelcmp_x3, satd ? h->pixf.satd_x3 : h->pixf.sad_x3, sizeof(h->pixf.fpelcmp_x3) );
976     memcpy( h->pixf.fpelcmp_x4, satd ? h->pixf.satd_x4 : h->pixf.sad_x4, sizeof(h->pixf.fpelcmp_x4) );
977 }
978
979 static void chroma_dsp_init( x264_t *h )
980 {
981     memcpy( h->luma2chroma_pixel, x264_luma2chroma_pixel[CHROMA_FORMAT], sizeof(h->luma2chroma_pixel) );
982
983     switch( CHROMA_FORMAT )
984     {
985         case CHROMA_420:
986             memcpy( h->predict_chroma, h->predict_8x8c, sizeof(h->predict_chroma) );
987             h->mc.prefetch_fenc = h->mc.prefetch_fenc_420;
988             h->loopf.deblock_chroma[0] = h->loopf.deblock_h_chroma_420;
989             h->loopf.deblock_chroma_intra[0] = h->loopf.deblock_h_chroma_420_intra;
990             h->loopf.deblock_chroma_mbaff = h->loopf.deblock_chroma_420_mbaff;
991             h->loopf.deblock_chroma_intra_mbaff = h->loopf.deblock_chroma_420_intra_mbaff;
992             h->pixf.intra_mbcmp_x3_chroma = h->pixf.intra_mbcmp_x3_8x8c;
993             h->quantf.coeff_last[DCT_CHROMA_DC] = h->quantf.coeff_last4;
994             h->quantf.coeff_level_run[DCT_CHROMA_DC] = h->quantf.coeff_level_run4;
995             break;
996         case CHROMA_422:
997             memcpy( h->predict_chroma, h->predict_8x16c, sizeof(h->predict_chroma) );
998             h->mc.prefetch_fenc = h->mc.prefetch_fenc_422;
999             h->loopf.deblock_chroma[0] = h->loopf.deblock_h_chroma_422;
1000             h->loopf.deblock_chroma_intra[0] = h->loopf.deblock_h_chroma_422_intra;
1001             h->loopf.deblock_chroma_mbaff = h->loopf.deblock_chroma_422_mbaff;
1002             h->loopf.deblock_chroma_intra_mbaff = h->loopf.deblock_chroma_422_intra_mbaff;
1003             h->pixf.intra_mbcmp_x3_chroma = h->pixf.intra_mbcmp_x3_8x16c;
1004             h->quantf.coeff_last[DCT_CHROMA_DC] = h->quantf.coeff_last8;
1005             h->quantf.coeff_level_run[DCT_CHROMA_DC] = h->quantf.coeff_level_run8;
1006             break;
1007         case CHROMA_444:
1008             h->mc.prefetch_fenc = h->mc.prefetch_fenc_422; /* FIXME: doesn't cover V plane */
1009             h->loopf.deblock_chroma_mbaff = h->loopf.deblock_luma_mbaff;
1010             h->loopf.deblock_chroma_intra_mbaff = h->loopf.deblock_luma_intra_mbaff;
1011             break;
1012     }
1013 }
1014
1015 static void x264_set_aspect_ratio( x264_t *h, x264_param_t *param, int initial )
1016 {
1017     /* VUI */
1018     if( param->vui.i_sar_width > 0 && param->vui.i_sar_height > 0 )
1019     {
1020         uint32_t i_w = param->vui.i_sar_width;
1021         uint32_t i_h = param->vui.i_sar_height;
1022         uint32_t old_w = h->param.vui.i_sar_width;
1023         uint32_t old_h = h->param.vui.i_sar_height;
1024
1025         x264_reduce_fraction( &i_w, &i_h );
1026
1027         while( i_w > 65535 || i_h > 65535 )
1028         {
1029             i_w /= 2;
1030             i_h /= 2;
1031         }
1032
1033         x264_reduce_fraction( &i_w, &i_h );
1034
1035         if( i_w != old_w || i_h != old_h || initial )
1036         {
1037             h->param.vui.i_sar_width = 0;
1038             h->param.vui.i_sar_height = 0;
1039             if( i_w == 0 || i_h == 0 )
1040                 x264_log( h, X264_LOG_WARNING, "cannot create valid sample aspect ratio\n" );
1041             else
1042             {
1043                 x264_log( h, initial?X264_LOG_INFO:X264_LOG_DEBUG, "using SAR=%d/%d\n", i_w, i_h );
1044                 h->param.vui.i_sar_width = i_w;
1045                 h->param.vui.i_sar_height = i_h;
1046             }
1047             x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
1048         }
1049     }
1050 }
1051
1052 /****************************************************************************
1053  * x264_encoder_open:
1054  ****************************************************************************/
1055 x264_t *x264_encoder_open( x264_param_t *param )
1056 {
1057     x264_t *h;
1058     char buf[1000], *p;
1059     int qp, i_slicetype_length;
1060
1061     CHECKED_MALLOCZERO( h, sizeof(x264_t) );
1062
1063     /* Create a copy of param */
1064     memcpy( &h->param, param, sizeof(x264_param_t) );
1065
1066     if( param->param_free )
1067         param->param_free( param );
1068
1069     if( x264_threading_init() )
1070     {
1071         x264_log( h, X264_LOG_ERROR, "unable to initialize threading\n" );
1072         goto fail;
1073     }
1074
1075     if( x264_validate_parameters( h, 1 ) < 0 )
1076         goto fail;
1077
1078     if( h->param.psz_cqm_file )
1079         if( x264_cqm_parse_file( h, h->param.psz_cqm_file ) < 0 )
1080             goto fail;
1081
1082     if( h->param.rc.psz_stat_out )
1083         h->param.rc.psz_stat_out = strdup( h->param.rc.psz_stat_out );
1084     if( h->param.rc.psz_stat_in )
1085         h->param.rc.psz_stat_in = strdup( h->param.rc.psz_stat_in );
1086
1087     x264_reduce_fraction( &h->param.i_fps_num, &h->param.i_fps_den );
1088     x264_reduce_fraction( &h->param.i_timebase_num, &h->param.i_timebase_den );
1089
1090     /* Init x264_t */
1091     h->i_frame = -1;
1092     h->i_frame_num = 0;
1093     h->i_idr_pic_id = 0;
1094
1095     if( (uint64_t)h->param.i_timebase_den * 2 > UINT32_MAX )
1096     {
1097         x264_log( h, X264_LOG_ERROR, "Effective timebase denominator %u exceeds H.264 maximum\n", h->param.i_timebase_den );
1098         goto fail;
1099     }
1100
1101     x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
1102     x264_pps_init( h->pps, h->param.i_sps_id, &h->param, h->sps );
1103
1104     x264_set_aspect_ratio( h, &h->param, 1 );
1105
1106     x264_validate_levels( h, 1 );
1107
1108     h->chroma_qp_table = i_chroma_qp_table + 12 + h->pps->i_chroma_qp_index_offset;
1109
1110     if( x264_cqm_init( h ) < 0 )
1111         goto fail;
1112
1113     h->mb.i_mb_width = h->sps->i_mb_width;
1114     h->mb.i_mb_height = h->sps->i_mb_height;
1115     h->mb.i_mb_count = h->mb.i_mb_width * h->mb.i_mb_height;
1116
1117     h->mb.chroma_h_shift = CHROMA_FORMAT == CHROMA_420 || CHROMA_FORMAT == CHROMA_422;
1118     h->mb.chroma_v_shift = CHROMA_FORMAT == CHROMA_420;
1119
1120     /* Adaptive MBAFF and subme 0 are not supported as we require halving motion
1121      * vectors during prediction, resulting in hpel mvs.
1122      * The chosen solution is to make MBAFF non-adaptive in this case. */
1123     h->mb.b_adaptive_mbaff = PARAM_INTERLACED && h->param.analyse.i_subpel_refine;
1124
1125     /* Init frames. */
1126     if( h->param.i_bframe_adaptive == X264_B_ADAPT_TRELLIS && !h->param.rc.b_stat_read )
1127         h->frames.i_delay = X264_MAX(h->param.i_bframe,3)*4;
1128     else
1129         h->frames.i_delay = h->param.i_bframe;
1130     if( h->param.rc.b_mb_tree || h->param.rc.i_vbv_buffer_size )
1131         h->frames.i_delay = X264_MAX( h->frames.i_delay, h->param.rc.i_lookahead );
1132     i_slicetype_length = h->frames.i_delay;
1133     h->frames.i_delay += h->i_thread_frames - 1;
1134     h->frames.i_delay += h->param.i_sync_lookahead;
1135     h->frames.i_delay += h->param.b_vfr_input;
1136     h->frames.i_bframe_delay = h->param.i_bframe ? (h->param.i_bframe_pyramid ? 2 : 1) : 0;
1137
1138     h->frames.i_max_ref0 = h->param.i_frame_reference;
1139     h->frames.i_max_ref1 = X264_MIN( h->sps->vui.i_num_reorder_frames, h->param.i_frame_reference );
1140     h->frames.i_max_dpb  = h->sps->vui.i_max_dec_frame_buffering;
1141     h->frames.b_have_lowres = !h->param.rc.b_stat_read
1142         && ( h->param.rc.i_rc_method == X264_RC_ABR
1143           || h->param.rc.i_rc_method == X264_RC_CRF
1144           || h->param.i_bframe_adaptive
1145           || h->param.i_scenecut_threshold
1146           || h->param.rc.b_mb_tree
1147           || h->param.analyse.i_weighted_pred );
1148     h->frames.b_have_lowres |= h->param.rc.b_stat_read && h->param.rc.i_vbv_buffer_size > 0;
1149     h->frames.b_have_sub8x8_esa = !!(h->param.analyse.inter & X264_ANALYSE_PSUB8x8);
1150
1151     h->frames.i_last_idr =
1152     h->frames.i_last_keyframe = - h->param.i_keyint_max;
1153     h->frames.i_input    = 0;
1154     h->frames.i_largest_pts = h->frames.i_second_largest_pts = -1;
1155     h->frames.i_poc_last_open_gop = -1;
1156
1157     CHECKED_MALLOCZERO( h->frames.unused[0], (h->frames.i_delay + 3) * sizeof(x264_frame_t *) );
1158     /* Allocate room for max refs plus a few extra just in case. */
1159     CHECKED_MALLOCZERO( h->frames.unused[1], (h->i_thread_frames + X264_REF_MAX + 4) * sizeof(x264_frame_t *) );
1160     CHECKED_MALLOCZERO( h->frames.current, (h->param.i_sync_lookahead + h->param.i_bframe
1161                         + h->i_thread_frames + 3) * sizeof(x264_frame_t *) );
1162     if( h->param.analyse.i_weighted_pred > 0 )
1163         CHECKED_MALLOCZERO( h->frames.blank_unused, h->i_thread_frames * 4 * sizeof(x264_frame_t *) );
1164     h->i_ref[0] = h->i_ref[1] = 0;
1165     h->i_cpb_delay = h->i_coded_fields = h->i_disp_fields = 0;
1166     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);
1167     h->i_disp_fields_last_frame = -1;
1168     x264_rdo_init();
1169
1170     /* init CPU functions */
1171     x264_predict_16x16_init( h->param.cpu, h->predict_16x16 );
1172     x264_predict_8x8c_init( h->param.cpu, h->predict_8x8c );
1173     x264_predict_8x16c_init( h->param.cpu, h->predict_8x16c );
1174     x264_predict_8x8_init( h->param.cpu, h->predict_8x8, &h->predict_8x8_filter );
1175     x264_predict_4x4_init( h->param.cpu, h->predict_4x4 );
1176     if( h->param.b_cabac )
1177         x264_cabac_init( h );
1178     else
1179         x264_cavlc_init();
1180     x264_pixel_init( h->param.cpu, &h->pixf );
1181     x264_dct_init( h->param.cpu, &h->dctf );
1182     x264_zigzag_init( h->param.cpu, &h->zigzagf_progressive, &h->zigzagf_interlaced );
1183     memcpy( &h->zigzagf, PARAM_INTERLACED ? &h->zigzagf_interlaced : &h->zigzagf_progressive, sizeof(h->zigzagf) );
1184     x264_mc_init( h->param.cpu, &h->mc );
1185     x264_quant_init( h, h->param.cpu, &h->quantf );
1186     x264_deblock_init( h->param.cpu, &h->loopf, PARAM_INTERLACED );
1187     x264_bitstream_init( h->param.cpu, &h->bsf );
1188     x264_dct_init_weights();
1189
1190     mbcmp_init( h );
1191     chroma_dsp_init( h );
1192
1193     p = buf + sprintf( buf, "using cpu capabilities:" );
1194     for( int i = 0; x264_cpu_names[i].flags; i++ )
1195     {
1196         if( !strcmp(x264_cpu_names[i].name, "SSE2")
1197             && h->param.cpu & (X264_CPU_SSE2_IS_FAST|X264_CPU_SSE2_IS_SLOW) )
1198             continue;
1199         if( !strcmp(x264_cpu_names[i].name, "SSE3")
1200             && (h->param.cpu & X264_CPU_SSSE3 || !(h->param.cpu & X264_CPU_CACHELINE_64)) )
1201             continue;
1202         if( !strcmp(x264_cpu_names[i].name, "SSE4.1")
1203             && (h->param.cpu & X264_CPU_SSE42) )
1204             continue;
1205         if( (h->param.cpu & x264_cpu_names[i].flags) == x264_cpu_names[i].flags
1206             && (!i || x264_cpu_names[i].flags != x264_cpu_names[i-1].flags) )
1207             p += sprintf( p, " %s", x264_cpu_names[i].name );
1208     }
1209     if( !h->param.cpu )
1210         p += sprintf( p, " none!" );
1211     x264_log( h, X264_LOG_INFO, "%s\n", buf );
1212
1213     float *logs = x264_analyse_prepare_costs( h );
1214     if( !logs )
1215         goto fail;
1216     for( qp = X264_MIN( h->param.rc.i_qp_min, QP_MAX_SPEC ); qp <= h->param.rc.i_qp_max; qp++ )
1217         if( x264_analyse_init_costs( h, logs, qp ) )
1218             goto fail;
1219     if( x264_analyse_init_costs( h, logs, X264_LOOKAHEAD_QP ) )
1220         goto fail;
1221     x264_free( logs );
1222
1223     static const uint16_t cost_mv_correct[7] = { 24, 47, 95, 189, 379, 757, 1515 };
1224     /* Checks for known miscompilation issues. */
1225     if( h->cost_mv[X264_LOOKAHEAD_QP][2013] != cost_mv_correct[BIT_DEPTH-8] )
1226     {
1227         x264_log( h, X264_LOG_ERROR, "MV cost test failed: x264 has been miscompiled!\n" );
1228         goto fail;
1229     }
1230
1231     /* Must be volatile or else GCC will optimize it out. */
1232     volatile int temp = 392;
1233     if( x264_clz( temp ) != 23 )
1234     {
1235         x264_log( h, X264_LOG_ERROR, "CLZ test failed: x264 has been miscompiled!\n" );
1236 #if ARCH_X86 || ARCH_X86_64
1237         x264_log( h, X264_LOG_ERROR, "Are you attempting to run an SSE4a-targeted build on a CPU that\n" );
1238         x264_log( h, X264_LOG_ERROR, "doesn't support it?\n" );
1239 #endif
1240         goto fail;
1241     }
1242
1243     h->out.i_nal = 0;
1244     h->out.i_bitstream = X264_MAX( 1000000, h->param.i_width * h->param.i_height * 4
1245         * ( h->param.rc.i_rc_method == X264_RC_ABR ? pow( 0.95, h->param.rc.i_qp_min )
1246           : pow( 0.95, h->param.rc.i_qp_constant ) * X264_MAX( 1, h->param.rc.f_ip_factor )));
1247
1248     h->nal_buffer_size = h->out.i_bitstream * 3/2 + 4;
1249     CHECKED_MALLOC( h->nal_buffer, h->nal_buffer_size );
1250
1251     if( h->param.i_threads > 1 &&
1252         x264_threadpool_init( &h->threadpool, h->param.i_threads, (void*)x264_encoder_thread_init, h ) )
1253         goto fail;
1254
1255     h->thread[0] = h;
1256     for( int i = 1; i < h->param.i_threads + !!h->param.i_sync_lookahead; i++ )
1257         CHECKED_MALLOC( h->thread[i], sizeof(x264_t) );
1258
1259     for( int i = 0; i < h->param.i_threads; i++ )
1260     {
1261         int init_nal_count = h->param.i_slice_count + 3;
1262         int allocate_threadlocal_data = !h->param.b_sliced_threads || !i;
1263         if( i > 0 )
1264             *h->thread[i] = *h;
1265
1266         if( allocate_threadlocal_data )
1267         {
1268             h->thread[i]->fdec = x264_frame_pop_unused( h, 1 );
1269             if( !h->thread[i]->fdec )
1270                 goto fail;
1271         }
1272         else
1273             h->thread[i]->fdec = h->thread[0]->fdec;
1274
1275         CHECKED_MALLOC( h->thread[i]->out.p_bitstream, h->out.i_bitstream );
1276         /* Start each thread with room for init_nal_count NAL units; it'll realloc later if needed. */
1277         CHECKED_MALLOC( h->thread[i]->out.nal, init_nal_count*sizeof(x264_nal_t) );
1278         h->thread[i]->out.i_nals_allocated = init_nal_count;
1279
1280         if( allocate_threadlocal_data && x264_macroblock_cache_allocate( h->thread[i] ) < 0 )
1281             goto fail;
1282     }
1283
1284     if( x264_lookahead_init( h, i_slicetype_length ) )
1285         goto fail;
1286
1287     for( int i = 0; i < h->param.i_threads; i++ )
1288         if( x264_macroblock_thread_allocate( h->thread[i], 0 ) < 0 )
1289             goto fail;
1290
1291     if( x264_ratecontrol_new( h ) < 0 )
1292         goto fail;
1293
1294     if( h->param.i_nal_hrd )
1295     {
1296         x264_log( h, X264_LOG_DEBUG, "HRD bitrate: %i bits/sec\n", h->sps->vui.hrd.i_bit_rate_unscaled );
1297         x264_log( h, X264_LOG_DEBUG, "CPB size: %i bits\n", h->sps->vui.hrd.i_cpb_size_unscaled );
1298     }
1299
1300     if( h->param.psz_dump_yuv )
1301     {
1302         /* create or truncate the reconstructed video file */
1303         FILE *f = fopen( h->param.psz_dump_yuv, "w" );
1304         if( !f )
1305         {
1306             x264_log( h, X264_LOG_ERROR, "dump_yuv: can't write to %s\n", h->param.psz_dump_yuv );
1307             goto fail;
1308         }
1309         else if( !x264_is_regular_file( f ) )
1310         {
1311             x264_log( h, X264_LOG_ERROR, "dump_yuv: incompatible with non-regular file %s\n", h->param.psz_dump_yuv );
1312             goto fail;
1313         }
1314         fclose( f );
1315     }
1316
1317     const char *profile = h->sps->i_profile_idc == PROFILE_BASELINE ? "Constrained Baseline" :
1318                           h->sps->i_profile_idc == PROFILE_MAIN ? "Main" :
1319                           h->sps->i_profile_idc == PROFILE_HIGH ? "High" :
1320                           h->sps->i_profile_idc == PROFILE_HIGH10 ? (h->sps->b_constraint_set3 == 1 ? "High 10 Intra" : "High 10") :
1321                           h->sps->i_profile_idc == PROFILE_HIGH422 ? (h->sps->b_constraint_set3 == 1 ? "High 4:2:2 Intra" : "High 4:2:2") :
1322                           h->sps->b_constraint_set3 == 1 ? "High 4:4:4 Intra" : "High 4:4:4 Predictive";
1323     char level[4];
1324     snprintf( level, sizeof(level), "%d.%d", h->sps->i_level_idc/10, h->sps->i_level_idc%10 );
1325     if( h->sps->i_level_idc == 9 || ( h->sps->i_level_idc == 11 && h->sps->b_constraint_set3 &&
1326         (h->sps->i_profile_idc >= PROFILE_BASELINE && h->sps->i_profile_idc <= PROFILE_EXTENDED) ) )
1327         strcpy( level, "1b" );
1328
1329     if( h->sps->i_profile_idc < PROFILE_HIGH10 )
1330     {
1331         x264_log( h, X264_LOG_INFO, "profile %s, level %s\n",
1332             profile, level );
1333     }
1334     else
1335     {
1336         static const char * const subsampling[4] = { "4:0:0", "4:2:0", "4:2:2", "4:4:4" };
1337         x264_log( h, X264_LOG_INFO, "profile %s, level %s, %s %d-bit\n",
1338             profile, level, subsampling[CHROMA_FORMAT], BIT_DEPTH );
1339     }
1340
1341     return h;
1342 fail:
1343     x264_free( h );
1344     return NULL;
1345 }
1346
1347 /****************************************************************************
1348  * x264_encoder_reconfig:
1349  ****************************************************************************/
1350 int x264_encoder_reconfig( x264_t *h, x264_param_t *param )
1351 {
1352     int rc_reconfig = 0;
1353     h = h->thread[h->thread[0]->i_thread_phase];
1354     x264_set_aspect_ratio( h, param, 0 );
1355 #define COPY(var) h->param.var = param->var
1356     COPY( i_frame_reference ); // but never uses more refs than initially specified
1357     COPY( i_bframe_bias );
1358     if( h->param.i_scenecut_threshold )
1359         COPY( i_scenecut_threshold ); // can't turn it on or off, only vary the threshold
1360     COPY( b_deblocking_filter );
1361     COPY( i_deblocking_filter_alphac0 );
1362     COPY( i_deblocking_filter_beta );
1363     COPY( i_frame_packing );
1364     COPY( analyse.inter );
1365     COPY( analyse.intra );
1366     COPY( analyse.i_direct_mv_pred );
1367     /* Scratch buffer prevents me_range from being increased for esa/tesa */
1368     if( h->param.analyse.i_me_method < X264_ME_ESA || param->analyse.i_me_range < h->param.analyse.i_me_range )
1369         COPY( analyse.i_me_range );
1370     COPY( analyse.i_noise_reduction );
1371     /* We can't switch out of subme=0 during encoding. */
1372     if( h->param.analyse.i_subpel_refine )
1373         COPY( analyse.i_subpel_refine );
1374     COPY( analyse.i_trellis );
1375     COPY( analyse.b_chroma_me );
1376     COPY( analyse.b_dct_decimate );
1377     COPY( analyse.b_fast_pskip );
1378     COPY( analyse.b_mixed_references );
1379     COPY( analyse.f_psy_rd );
1380     COPY( analyse.f_psy_trellis );
1381     COPY( crop_rect );
1382     // can only twiddle these if they were enabled to begin with:
1383     if( h->param.analyse.i_me_method >= X264_ME_ESA || param->analyse.i_me_method < X264_ME_ESA )
1384         COPY( analyse.i_me_method );
1385     if( h->param.analyse.i_me_method >= X264_ME_ESA && !h->frames.b_have_sub8x8_esa )
1386         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
1387     if( h->pps->b_transform_8x8_mode )
1388         COPY( analyse.b_transform_8x8 );
1389     if( h->frames.i_max_ref1 > 1 )
1390         COPY( i_bframe_pyramid );
1391     COPY( i_slice_max_size );
1392     COPY( i_slice_max_mbs );
1393     COPY( i_slice_count );
1394     COPY( b_tff );
1395
1396     /* VBV can't be turned on if it wasn't on to begin with */
1397     if( h->param.rc.i_vbv_max_bitrate > 0 && h->param.rc.i_vbv_buffer_size > 0 &&
1398           param->rc.i_vbv_max_bitrate > 0 &&   param->rc.i_vbv_buffer_size > 0 )
1399     {
1400         rc_reconfig |= h->param.rc.i_vbv_max_bitrate != param->rc.i_vbv_max_bitrate;
1401         rc_reconfig |= h->param.rc.i_vbv_buffer_size != param->rc.i_vbv_buffer_size;
1402         rc_reconfig |= h->param.rc.i_bitrate != param->rc.i_bitrate;
1403         COPY( rc.i_vbv_max_bitrate );
1404         COPY( rc.i_vbv_buffer_size );
1405         COPY( rc.i_bitrate );
1406     }
1407     rc_reconfig |= h->param.rc.f_rf_constant != param->rc.f_rf_constant;
1408     rc_reconfig |= h->param.rc.f_rf_constant_max != param->rc.f_rf_constant_max;
1409     COPY( rc.f_rf_constant );
1410     COPY( rc.f_rf_constant_max );
1411 #undef COPY
1412
1413     mbcmp_init( h );
1414
1415     int ret = x264_validate_parameters( h, 0 );
1416
1417     /* Supported reconfiguration options (1-pass only):
1418      * vbv-maxrate
1419      * vbv-bufsize
1420      * crf
1421      * bitrate (CBR only) */
1422     if( !ret && rc_reconfig )
1423         x264_ratecontrol_init_reconfigurable( h, 0 );
1424
1425     return ret;
1426 }
1427
1428 /****************************************************************************
1429  * x264_encoder_parameters:
1430  ****************************************************************************/
1431 void x264_encoder_parameters( x264_t *h, x264_param_t *param )
1432 {
1433     memcpy( param, &h->thread[h->i_thread_phase]->param, sizeof(x264_param_t) );
1434 }
1435
1436 /* internal usage */
1437 static void x264_nal_start( x264_t *h, int i_type, int i_ref_idc )
1438 {
1439     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1440
1441     nal->i_ref_idc        = i_ref_idc;
1442     nal->i_type           = i_type;
1443     nal->b_long_startcode = 1;
1444
1445     nal->i_payload= 0;
1446     nal->p_payload= &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8];
1447 }
1448
1449 /* if number of allocated nals is not enough, re-allocate a larger one. */
1450 static int x264_nal_check_buffer( x264_t *h )
1451 {
1452     if( h->out.i_nal >= h->out.i_nals_allocated )
1453     {
1454         x264_nal_t *new_out = x264_malloc( sizeof(x264_nal_t) * (h->out.i_nals_allocated*2) );
1455         if( !new_out )
1456             return -1;
1457         memcpy( new_out, h->out.nal, sizeof(x264_nal_t) * (h->out.i_nals_allocated) );
1458         x264_free( h->out.nal );
1459         h->out.nal = new_out;
1460         h->out.i_nals_allocated *= 2;
1461     }
1462     return 0;
1463 }
1464
1465 static int x264_nal_end( x264_t *h )
1466 {
1467     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1468     uint8_t *end = &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8];
1469     nal->i_payload = end - nal->p_payload;
1470     /* nal_escape_mmx reads past the end of the input.
1471      * While undefined padding wouldn't actually affect the output, it makes valgrind unhappy. */
1472     memset( end, 0xff, 32 );
1473     if( h->param.nalu_process )
1474         h->param.nalu_process( h, nal );
1475     h->out.i_nal++;
1476
1477     return x264_nal_check_buffer( h );
1478 }
1479
1480 static int x264_encoder_encapsulate_nals( x264_t *h, int start )
1481 {
1482     int nal_size = 0, previous_nal_size = 0;
1483
1484     if( h->param.nalu_process )
1485     {
1486         for( int i = start; i < h->out.i_nal; i++ )
1487             nal_size += h->out.nal[i].i_payload;
1488         return nal_size;
1489     }
1490
1491     for( int i = 0; i < start; i++ )
1492         previous_nal_size += h->out.nal[i].i_payload;
1493
1494     for( int i = start; i < h->out.i_nal; i++ )
1495         nal_size += h->out.nal[i].i_payload;
1496
1497     /* Worst-case NAL unit escaping: reallocate the buffer if it's too small. */
1498     int necessary_size = nal_size * 3/2 + h->out.i_nal * 4;
1499     if( h->nal_buffer_size < necessary_size )
1500     {
1501         h->nal_buffer_size = necessary_size * 2;
1502         uint8_t *buf = x264_malloc( h->nal_buffer_size );
1503         if( !buf )
1504             return -1;
1505         if( previous_nal_size )
1506             memcpy( buf, h->nal_buffer, previous_nal_size );
1507         x264_free( h->nal_buffer );
1508         h->nal_buffer = buf;
1509     }
1510
1511     uint8_t *nal_buffer = h->nal_buffer + previous_nal_size;
1512
1513     for( int i = start; i < h->out.i_nal; i++ )
1514     {
1515         h->out.nal[i].b_long_startcode = !i || h->out.nal[i].i_type == NAL_SPS || h->out.nal[i].i_type == NAL_PPS;
1516         x264_nal_encode( h, nal_buffer, &h->out.nal[i] );
1517         nal_buffer += h->out.nal[i].i_payload;
1518     }
1519
1520     x264_emms();
1521
1522     return nal_buffer - (h->nal_buffer + previous_nal_size);
1523 }
1524
1525 /****************************************************************************
1526  * x264_encoder_headers:
1527  ****************************************************************************/
1528 int x264_encoder_headers( x264_t *h, x264_nal_t **pp_nal, int *pi_nal )
1529 {
1530     int frame_size = 0;
1531     /* init bitstream context */
1532     h->out.i_nal = 0;
1533     bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
1534
1535     /* Write SEI, SPS and PPS. */
1536
1537     /* generate sequence parameters */
1538     x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
1539     x264_sps_write( &h->out.bs, h->sps );
1540     if( x264_nal_end( h ) )
1541         return -1;
1542
1543     /* generate picture parameters */
1544     x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
1545     x264_pps_write( &h->out.bs, h->sps, h->pps );
1546     if( x264_nal_end( h ) )
1547         return -1;
1548
1549     /* identify ourselves */
1550     x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
1551     if( x264_sei_version_write( h, &h->out.bs ) )
1552         return -1;
1553     if( x264_nal_end( h ) )
1554         return -1;
1555
1556     frame_size = x264_encoder_encapsulate_nals( h, 0 );
1557     if( frame_size < 0 )
1558         return -1;
1559
1560     /* now set output*/
1561     *pi_nal = h->out.i_nal;
1562     *pp_nal = &h->out.nal[0];
1563     h->out.i_nal = 0;
1564
1565     return frame_size;
1566 }
1567
1568 /* Check to see whether we have chosen a reference list ordering different
1569  * from the standard's default. */
1570 static inline void x264_reference_check_reorder( x264_t *h )
1571 {
1572     /* The reorder check doesn't check for missing frames, so just
1573      * force a reorder if one of the reference list is corrupt. */
1574     for( int i = 0; h->frames.reference[i]; i++ )
1575         if( h->frames.reference[i]->b_corrupt )
1576         {
1577             h->b_ref_reorder[0] = 1;
1578             return;
1579         }
1580     for( int list = 0; list <= (h->sh.i_type == SLICE_TYPE_B); list++ )
1581         for( int i = 0; i < h->i_ref[list] - 1; i++ )
1582         {
1583             int framenum_diff = h->fref[list][i+1]->i_frame_num - h->fref[list][i]->i_frame_num;
1584             int poc_diff = h->fref[list][i+1]->i_poc - h->fref[list][i]->i_poc;
1585             /* P and B-frames use different default orders. */
1586             if( h->sh.i_type == SLICE_TYPE_P ? framenum_diff > 0 : list == 1 ? poc_diff < 0 : poc_diff > 0 )
1587             {
1588                 h->b_ref_reorder[list] = 1;
1589                 return;
1590             }
1591         }
1592 }
1593
1594 /* return -1 on failure, else return the index of the new reference frame */
1595 int x264_weighted_reference_duplicate( x264_t *h, int i_ref, const x264_weight_t *w )
1596 {
1597     int i = h->i_ref[0];
1598     int j = 1;
1599     x264_frame_t *newframe;
1600     if( i <= 1 ) /* empty list, definitely can't duplicate frame */
1601         return -1;
1602
1603     //Duplication is only used in X264_WEIGHTP_SMART
1604     if( h->param.analyse.i_weighted_pred != X264_WEIGHTP_SMART )
1605         return -1;
1606
1607     /* Duplication is a hack to compensate for crappy rounding in motion compensation.
1608      * With high bit depth, it's not worth doing, so turn it off except in the case of
1609      * unweighted dupes. */
1610     if( BIT_DEPTH > 8 && w != x264_weight_none )
1611         return -1;
1612
1613     newframe = x264_frame_pop_blank_unused( h );
1614     if( !newframe )
1615         return -1;
1616
1617     //FIXME: probably don't need to copy everything
1618     *newframe = *h->fref[0][i_ref];
1619     newframe->i_reference_count = 1;
1620     newframe->orig = h->fref[0][i_ref];
1621     newframe->b_duplicate = 1;
1622     memcpy( h->fenc->weight[j], w, sizeof(h->fenc->weight[i]) );
1623
1624     /* shift the frames to make space for the dupe. */
1625     h->b_ref_reorder[0] = 1;
1626     if( h->i_ref[0] < X264_REF_MAX )
1627         ++h->i_ref[0];
1628     h->fref[0][X264_REF_MAX-1] = NULL;
1629     x264_frame_unshift( &h->fref[0][j], newframe );
1630
1631     return j;
1632 }
1633
1634 static void x264_weighted_pred_init( x264_t *h )
1635 {
1636     /* for now no analysis and set all weights to nothing */
1637     for( int i_ref = 0; i_ref < h->i_ref[0]; i_ref++ )
1638         h->fenc->weighted[i_ref] = h->fref[0][i_ref]->filtered[0][0];
1639
1640     // FIXME: This only supports weighting of one reference frame
1641     // and duplicates of that frame.
1642     h->fenc->i_lines_weighted = 0;
1643
1644     for( int i_ref = 0; i_ref < (h->i_ref[0] << SLICE_MBAFF); i_ref++ )
1645         for( int i = 0; i < 3; i++ )
1646             h->sh.weight[i_ref][i].weightfn = NULL;
1647
1648
1649     if( h->sh.i_type != SLICE_TYPE_P || h->param.analyse.i_weighted_pred <= 0 )
1650         return;
1651
1652     int i_padv = PADV << PARAM_INTERLACED;
1653     int denom = -1;
1654     int weightplane[2] = { 0, 0 };
1655     int buffer_next = 0;
1656     for( int i = 0; i < 3; i++ )
1657     {
1658         for( int j = 0; j < h->i_ref[0]; j++ )
1659         {
1660             if( h->fenc->weight[j][i].weightfn )
1661             {
1662                 h->sh.weight[j][i] = h->fenc->weight[j][i];
1663                 // if weight is useless, don't write it to stream
1664                 if( h->sh.weight[j][i].i_scale == 1<<h->sh.weight[j][i].i_denom && h->sh.weight[j][i].i_offset == 0 )
1665                     h->sh.weight[j][i].weightfn = NULL;
1666                 else
1667                 {
1668                     if( !weightplane[!!i] )
1669                     {
1670                         weightplane[!!i] = 1;
1671                         h->sh.weight[0][!!i].i_denom = denom = h->sh.weight[j][i].i_denom;
1672                         assert( x264_clip3( denom, 0, 7 ) == denom );
1673                     }
1674
1675                     assert( h->sh.weight[j][i].i_denom == denom );
1676                     if( !i )
1677                     {
1678                         h->fenc->weighted[j] = h->mb.p_weight_buf[buffer_next++] + h->fenc->i_stride[0] * i_padv + PADH;
1679                         //scale full resolution frame
1680                         if( h->param.i_threads == 1 )
1681                         {
1682                             pixel *src = h->fref[0][j]->filtered[0][0] - h->fref[0][j]->i_stride[0]*i_padv - PADH;
1683                             pixel *dst = h->fenc->weighted[j] - h->fenc->i_stride[0]*i_padv - PADH;
1684                             int stride = h->fenc->i_stride[0];
1685                             int width = h->fenc->i_width[0] + PADH*2;
1686                             int height = h->fenc->i_lines[0] + i_padv*2;
1687                             x264_weight_scale_plane( h, dst, stride, src, stride, width, height, &h->sh.weight[j][0] );
1688                             h->fenc->i_lines_weighted = height;
1689                         }
1690                     }
1691                 }
1692             }
1693         }
1694     }
1695
1696     if( weightplane[1] )
1697         for( int i = 0; i < h->i_ref[0]; i++ )
1698         {
1699             if( h->sh.weight[i][1].weightfn && !h->sh.weight[i][2].weightfn )
1700             {
1701                 h->sh.weight[i][2].i_scale = 1 << h->sh.weight[0][1].i_denom;
1702                 h->sh.weight[i][2].i_offset = 0;
1703             }
1704             else if( h->sh.weight[i][2].weightfn && !h->sh.weight[i][1].weightfn )
1705             {
1706                 h->sh.weight[i][1].i_scale = 1 << h->sh.weight[0][1].i_denom;
1707                 h->sh.weight[i][1].i_offset = 0;
1708             }
1709         }
1710
1711     if( !weightplane[0] )
1712         h->sh.weight[0][0].i_denom = 0;
1713     if( !weightplane[1] )
1714         h->sh.weight[0][1].i_denom = 0;
1715     h->sh.weight[0][2].i_denom = h->sh.weight[0][1].i_denom;
1716 }
1717
1718 static inline int x264_reference_distance( x264_t *h, x264_frame_t *frame )
1719 {
1720     if( h->param.i_frame_packing == 5 )
1721         return abs((h->fenc->i_frame&~1) - (frame->i_frame&~1)) +
1722                   ((h->fenc->i_frame&1) != (frame->i_frame&1));
1723     else
1724         return abs(h->fenc->i_frame - frame->i_frame);
1725 }
1726
1727 static inline void x264_reference_build_list( x264_t *h, int i_poc )
1728 {
1729     int b_ok;
1730
1731     /* build ref list 0/1 */
1732     h->mb.pic.i_fref[0] = h->i_ref[0] = 0;
1733     h->mb.pic.i_fref[1] = h->i_ref[1] = 0;
1734     if( h->sh.i_type == SLICE_TYPE_I )
1735         return;
1736
1737     for( int i = 0; h->frames.reference[i]; i++ )
1738     {
1739         if( h->frames.reference[i]->b_corrupt )
1740             continue;
1741         if( h->frames.reference[i]->i_poc < i_poc )
1742             h->fref[0][h->i_ref[0]++] = h->frames.reference[i];
1743         else if( h->frames.reference[i]->i_poc > i_poc )
1744             h->fref[1][h->i_ref[1]++] = h->frames.reference[i];
1745     }
1746
1747     /* Order reference lists by distance from the current frame. */
1748     for( int list = 0; list < 2; list++ )
1749     {
1750         h->fref_nearest[list] = h->fref[list][0];
1751         do
1752         {
1753             b_ok = 1;
1754             for( int i = 0; i < h->i_ref[list] - 1; i++ )
1755             {
1756                 if( list ? h->fref[list][i+1]->i_poc < h->fref_nearest[list]->i_poc
1757                          : h->fref[list][i+1]->i_poc > h->fref_nearest[list]->i_poc )
1758                     h->fref_nearest[list] = h->fref[list][i+1];
1759                 if( x264_reference_distance( h, h->fref[list][i] ) > x264_reference_distance( h, h->fref[list][i+1] ) )
1760                 {
1761                     XCHG( x264_frame_t*, h->fref[list][i], h->fref[list][i+1] );
1762                     b_ok = 0;
1763                     break;
1764                 }
1765             }
1766         } while( !b_ok );
1767     }
1768
1769     if( h->sh.i_mmco_remove_from_end )
1770         for( int i = h->i_ref[0]-1; i >= h->i_ref[0] - h->sh.i_mmco_remove_from_end; i-- )
1771         {
1772             int diff = h->i_frame_num - h->fref[0][i]->i_frame_num;
1773             h->sh.mmco[h->sh.i_mmco_command_count].i_poc = h->fref[0][i]->i_poc;
1774             h->sh.mmco[h->sh.i_mmco_command_count++].i_difference_of_pic_nums = diff;
1775         }
1776
1777     x264_reference_check_reorder( h );
1778
1779     h->i_ref[1] = X264_MIN( h->i_ref[1], h->frames.i_max_ref1 );
1780     h->i_ref[0] = X264_MIN( h->i_ref[0], h->frames.i_max_ref0 );
1781     h->i_ref[0] = X264_MIN( h->i_ref[0], h->param.i_frame_reference ); // if reconfig() has lowered the limit
1782
1783     /* For Blu-ray compliance, don't reference frames outside of the minigop. */
1784     if( IS_X264_TYPE_B( h->fenc->i_type ) && h->param.b_bluray_compat )
1785         h->i_ref[0] = X264_MIN( h->i_ref[0], IS_X264_TYPE_B( h->fref[0][0]->i_type ) + 1 );
1786
1787     /* add duplicates */
1788     if( h->fenc->i_type == X264_TYPE_P )
1789     {
1790         int idx = -1;
1791         if( h->param.analyse.i_weighted_pred >= X264_WEIGHTP_SIMPLE )
1792         {
1793             x264_weight_t w[3];
1794             w[1].weightfn = w[2].weightfn = NULL;
1795             if( h->param.rc.b_stat_read )
1796                 x264_ratecontrol_set_weights( h, h->fenc );
1797
1798             if( !h->fenc->weight[0][0].weightfn )
1799             {
1800                 h->fenc->weight[0][0].i_denom = 0;
1801                 SET_WEIGHT( w[0], 1, 1, 0, -1 );
1802                 idx = x264_weighted_reference_duplicate( h, 0, w );
1803             }
1804             else
1805             {
1806                 if( h->fenc->weight[0][0].i_scale == 1<<h->fenc->weight[0][0].i_denom )
1807                 {
1808                     SET_WEIGHT( h->fenc->weight[0][0], 1, 1, 0, h->fenc->weight[0][0].i_offset );
1809                 }
1810                 x264_weighted_reference_duplicate( h, 0, x264_weight_none );
1811                 if( h->fenc->weight[0][0].i_offset > -128 )
1812                 {
1813                     w[0] = h->fenc->weight[0][0];
1814                     w[0].i_offset--;
1815                     h->mc.weight_cache( h, &w[0] );
1816                     idx = x264_weighted_reference_duplicate( h, 0, w );
1817                 }
1818             }
1819         }
1820         h->mb.ref_blind_dupe = idx;
1821     }
1822
1823     assert( h->i_ref[0] + h->i_ref[1] <= X264_REF_MAX );
1824     h->mb.pic.i_fref[0] = h->i_ref[0];
1825     h->mb.pic.i_fref[1] = h->i_ref[1];
1826 }
1827
1828 static void x264_fdec_filter_row( x264_t *h, int mb_y, int b_inloop )
1829 {
1830     /* mb_y is the mb to be encoded next, not the mb to be filtered here */
1831     int b_hpel = h->fdec->b_kept_as_ref;
1832     int b_deblock = h->sh.i_disable_deblocking_filter_idc != 1;
1833     int b_end = mb_y == h->i_threadslice_end;
1834     int b_measure_quality = 1;
1835     int min_y = mb_y - (1 << SLICE_MBAFF);
1836     int b_start = min_y == h->i_threadslice_start;
1837     /* Even in interlaced mode, deblocking never modifies more than 4 pixels
1838      * above each MB, as bS=4 doesn't happen for the top of interlaced mbpairs. */
1839     int minpix_y = min_y*16 - 4 * !b_start;
1840     int maxpix_y = mb_y*16 - 4 * !b_end;
1841     b_deblock &= b_hpel || h->param.psz_dump_yuv;
1842     if( h->param.b_sliced_threads && b_start && min_y && !b_inloop )
1843     {
1844         b_deblock = 0;         /* We already deblocked on the inloop pass. */
1845         b_measure_quality = 0; /* We already measured quality on the inloop pass. */
1846     }
1847     if( mb_y & SLICE_MBAFF )
1848         return;
1849     if( min_y < h->i_threadslice_start )
1850         return;
1851
1852     if( b_deblock )
1853         for( int y = min_y; y < mb_y; y += (1 << SLICE_MBAFF) )
1854             x264_frame_deblock_row( h, y );
1855
1856     /* FIXME: Prediction requires different borders for interlaced/progressive mc,
1857      * but the actual image data is equivalent. For now, maintain this
1858      * consistency by copying deblocked pixels between planes. */
1859     if( PARAM_INTERLACED )
1860         for( int p = 0; p < h->fdec->i_plane; p++ )
1861             for( int i = minpix_y>>(CHROMA_V_SHIFT && p); i < maxpix_y>>(CHROMA_V_SHIFT && p); i++ )
1862                 memcpy( h->fdec->plane_fld[p] + i*h->fdec->i_stride[p],
1863                         h->fdec->plane[p]     + i*h->fdec->i_stride[p],
1864                         h->mb.i_mb_width*16*sizeof(pixel) );
1865
1866     if( b_hpel )
1867     {
1868         int end = mb_y == h->mb.i_mb_height;
1869         x264_frame_expand_border( h, h->fdec, min_y, end );
1870         if( h->param.analyse.i_subpel_refine )
1871         {
1872             x264_frame_filter( h, h->fdec, min_y, end );
1873             x264_frame_expand_border_filtered( h, h->fdec, min_y, end );
1874         }
1875     }
1876
1877     if( SLICE_MBAFF )
1878         for( int i = 0; i < 3; i++ )
1879         {
1880             XCHG( pixel *, h->intra_border_backup[0][i], h->intra_border_backup[3][i] );
1881             XCHG( pixel *, h->intra_border_backup[1][i], h->intra_border_backup[4][i] );
1882         }
1883
1884     if( h->i_thread_frames > 1 && h->fdec->b_kept_as_ref )
1885         x264_frame_cond_broadcast( h->fdec, mb_y*16 + (b_end ? 10000 : -(X264_THREAD_HEIGHT << SLICE_MBAFF)) );
1886
1887     if( b_measure_quality )
1888     {
1889         maxpix_y = X264_MIN( maxpix_y, h->param.i_height );
1890         if( h->param.analyse.b_psnr )
1891         {
1892             for( int p = 0; p < (CHROMA444 ? 3 : 1); p++ )
1893                 h->stat.frame.i_ssd[p] += x264_pixel_ssd_wxh( &h->pixf,
1894                     h->fdec->plane[p] + minpix_y * h->fdec->i_stride[p], h->fdec->i_stride[p],
1895                     h->fenc->plane[p] + minpix_y * h->fenc->i_stride[p], h->fenc->i_stride[p],
1896                     h->param.i_width, maxpix_y-minpix_y );
1897             if( !CHROMA444 )
1898             {
1899                 uint64_t ssd_u, ssd_v;
1900                 int v_shift = CHROMA_V_SHIFT;
1901                 x264_pixel_ssd_nv12( &h->pixf,
1902                     h->fdec->plane[1] + (minpix_y>>v_shift) * h->fdec->i_stride[1], h->fdec->i_stride[1],
1903                     h->fenc->plane[1] + (minpix_y>>v_shift) * h->fenc->i_stride[1], h->fenc->i_stride[1],
1904                     h->param.i_width>>1, (maxpix_y-minpix_y)>>v_shift, &ssd_u, &ssd_v );
1905                 h->stat.frame.i_ssd[1] += ssd_u;
1906                 h->stat.frame.i_ssd[2] += ssd_v;
1907             }
1908         }
1909
1910         if( h->param.analyse.b_ssim )
1911         {
1912             int ssim_cnt;
1913             x264_emms();
1914             /* offset by 2 pixels to avoid alignment of ssim blocks with dct blocks,
1915              * and overlap by 4 */
1916             minpix_y += b_start ? 2 : -6;
1917             h->stat.frame.f_ssim +=
1918                 x264_pixel_ssim_wxh( &h->pixf,
1919                     h->fdec->plane[0] + 2+minpix_y*h->fdec->i_stride[0], h->fdec->i_stride[0],
1920                     h->fenc->plane[0] + 2+minpix_y*h->fenc->i_stride[0], h->fenc->i_stride[0],
1921                     h->param.i_width-2, maxpix_y-minpix_y, h->scratch_buffer, &ssim_cnt );
1922             h->stat.frame.i_ssim_cnt += ssim_cnt;
1923         }
1924     }
1925 }
1926
1927 static inline int x264_reference_update( x264_t *h )
1928 {
1929     if( !h->fdec->b_kept_as_ref )
1930     {
1931         if( h->i_thread_frames > 1 )
1932         {
1933             x264_frame_push_unused( h, h->fdec );
1934             h->fdec = x264_frame_pop_unused( h, 1 );
1935             if( !h->fdec )
1936                 return -1;
1937         }
1938         return 0;
1939     }
1940
1941     /* apply mmco from previous frame. */
1942     for( int i = 0; i < h->sh.i_mmco_command_count; i++ )
1943         for( int j = 0; h->frames.reference[j]; j++ )
1944             if( h->frames.reference[j]->i_poc == h->sh.mmco[i].i_poc )
1945                 x264_frame_push_unused( h, x264_frame_shift( &h->frames.reference[j] ) );
1946
1947     /* move frame in the buffer */
1948     x264_frame_push( h->frames.reference, h->fdec );
1949     if( h->frames.reference[h->sps->i_num_ref_frames] )
1950         x264_frame_push_unused( h, x264_frame_shift( h->frames.reference ) );
1951     h->fdec = x264_frame_pop_unused( h, 1 );
1952     if( !h->fdec )
1953         return -1;
1954     return 0;
1955 }
1956
1957 static inline void x264_reference_reset( x264_t *h )
1958 {
1959     while( h->frames.reference[0] )
1960         x264_frame_push_unused( h, x264_frame_pop( h->frames.reference ) );
1961     h->fdec->i_poc =
1962     h->fenc->i_poc = 0;
1963 }
1964
1965 static inline void x264_reference_hierarchy_reset( x264_t *h )
1966 {
1967     int ref;
1968     int b_hasdelayframe = 0;
1969
1970     /* look for delay frames -- chain must only contain frames that are disposable */
1971     for( int i = 0; h->frames.current[i] && IS_DISPOSABLE( h->frames.current[i]->i_type ); i++ )
1972         b_hasdelayframe |= h->frames.current[i]->i_coded
1973                         != h->frames.current[i]->i_frame + h->sps->vui.i_num_reorder_frames;
1974
1975     /* This function must handle b-pyramid and clear frames for open-gop */
1976     if( h->param.i_bframe_pyramid != X264_B_PYRAMID_STRICT && !b_hasdelayframe && h->frames.i_poc_last_open_gop == -1 )
1977         return;
1978
1979     /* Remove last BREF. There will never be old BREFs in the
1980      * dpb during a BREF decode when pyramid == STRICT */
1981     for( ref = 0; h->frames.reference[ref]; ref++ )
1982     {
1983         if( ( h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT
1984             && h->frames.reference[ref]->i_type == X264_TYPE_BREF )
1985             || ( h->frames.reference[ref]->i_poc < h->frames.i_poc_last_open_gop
1986             && h->sh.i_type != SLICE_TYPE_B ) )
1987         {
1988             int diff = h->i_frame_num - h->frames.reference[ref]->i_frame_num;
1989             h->sh.mmco[h->sh.i_mmco_command_count].i_difference_of_pic_nums = diff;
1990             h->sh.mmco[h->sh.i_mmco_command_count++].i_poc = h->frames.reference[ref]->i_poc;
1991             x264_frame_push_unused( h, x264_frame_shift( &h->frames.reference[ref] ) );
1992             h->b_ref_reorder[0] = 1;
1993             ref--;
1994         }
1995     }
1996
1997     /* Prepare room in the dpb for the delayed display time of the later b-frame's */
1998     if( h->param.i_bframe_pyramid )
1999         h->sh.i_mmco_remove_from_end = X264_MAX( ref + 2 - h->frames.i_max_dpb, 0 );
2000 }
2001
2002 static inline void x264_slice_init( x264_t *h, int i_nal_type, int i_global_qp )
2003 {
2004     /* ------------------------ Create slice header  ----------------------- */
2005     if( i_nal_type == NAL_SLICE_IDR )
2006     {
2007         x264_slice_header_init( h, &h->sh, h->sps, h->pps, h->i_idr_pic_id, h->i_frame_num, i_global_qp );
2008
2009         /* alternate id */
2010         h->i_idr_pic_id ^= 1;
2011     }
2012     else
2013     {
2014         x264_slice_header_init( h, &h->sh, h->sps, h->pps, -1, h->i_frame_num, i_global_qp );
2015
2016         h->sh.i_num_ref_idx_l0_active = h->i_ref[0] <= 0 ? 1 : h->i_ref[0];
2017         h->sh.i_num_ref_idx_l1_active = h->i_ref[1] <= 0 ? 1 : h->i_ref[1];
2018         if( h->sh.i_num_ref_idx_l0_active != h->pps->i_num_ref_idx_l0_default_active ||
2019             (h->sh.i_type == SLICE_TYPE_B && h->sh.i_num_ref_idx_l1_active != h->pps->i_num_ref_idx_l1_default_active) )
2020         {
2021             h->sh.b_num_ref_idx_override = 1;
2022         }
2023     }
2024
2025     if( h->fenc->i_type == X264_TYPE_BREF && h->param.b_bluray_compat && h->sh.i_mmco_command_count )
2026     {
2027         h->b_sh_backup = 1;
2028         h->sh_backup = h->sh;
2029     }
2030
2031     h->fdec->i_frame_num = h->sh.i_frame_num;
2032
2033     if( h->sps->i_poc_type == 0 )
2034     {
2035         h->sh.i_poc = h->fdec->i_poc;
2036         if( PARAM_INTERLACED )
2037         {
2038             h->sh.i_delta_poc_bottom = h->param.b_tff ? 1 : -1;
2039             h->sh.i_poc += h->sh.i_delta_poc_bottom == -1;
2040         }
2041         else
2042             h->sh.i_delta_poc_bottom = 0;
2043         h->fdec->i_delta_poc[0] = h->sh.i_delta_poc_bottom == -1;
2044         h->fdec->i_delta_poc[1] = h->sh.i_delta_poc_bottom ==  1;
2045     }
2046     else
2047     {
2048         /* Nothing to do ? */
2049     }
2050
2051     x264_macroblock_slice_init( h );
2052 }
2053
2054 static int x264_slice_write( x264_t *h )
2055 {
2056     int i_skip;
2057     int mb_xy, i_mb_x, i_mb_y;
2058     int i_skip_bak = 0; /* Shut up GCC. */
2059     bs_t UNINIT(bs_bak);
2060     x264_cabac_t cabac_bak;
2061     uint8_t cabac_prevbyte_bak = 0; /* Shut up GCC. */
2062     int mv_bits_bak = 0;
2063     int tex_bits_bak = 0;
2064     /* NALUs other than the first use a 3-byte startcode.
2065      * Add one extra byte for the rbsp, and one more for the final CABAC putbyte.
2066      * Then add an extra 5 bytes just in case, to account for random NAL escapes and
2067      * other inaccuracies. */
2068     int overhead_guess = (NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal)) + 1 + h->param.b_cabac + 5;
2069     int slice_max_size = h->param.i_slice_max_size > 0 ? (h->param.i_slice_max_size-overhead_guess)*8 : 0;
2070     int back_up_bitstream = slice_max_size || (!h->param.b_cabac && h->sps->i_profile_idc < PROFILE_HIGH);
2071     int starting_bits = bs_pos(&h->out.bs);
2072     int b_deblock = h->sh.i_disable_deblocking_filter_idc != 1;
2073     int b_hpel = h->fdec->b_kept_as_ref;
2074     uint8_t *last_emu_check;
2075     b_deblock &= b_hpel || h->param.psz_dump_yuv;
2076     bs_realign( &h->out.bs );
2077
2078     /* Slice */
2079     x264_nal_start( h, h->i_nal_type, h->i_nal_ref_idc );
2080     h->out.nal[h->out.i_nal].i_first_mb = h->sh.i_first_mb;
2081
2082     /* Slice header */
2083     x264_macroblock_thread_init( h );
2084
2085     /* If this isn't the first slice in the threadslice, set the slice QP
2086      * equal to the last QP in the previous slice for more accurate
2087      * CABAC initialization. */
2088     if( h->sh.i_first_mb != h->i_threadslice_start * h->mb.i_mb_width )
2089     {
2090         h->sh.i_qp = h->mb.i_last_qp;
2091         h->sh.i_qp_delta = h->sh.i_qp - h->pps->i_pic_init_qp;
2092     }
2093
2094     x264_slice_header_write( &h->out.bs, &h->sh, h->i_nal_ref_idc );
2095     if( h->param.b_cabac )
2096     {
2097         /* alignment needed */
2098         bs_align_1( &h->out.bs );
2099
2100         /* init cabac */
2101         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 );
2102         x264_cabac_encode_init ( &h->cabac, h->out.bs.p, h->out.bs.p_end );
2103         last_emu_check = h->cabac.p;
2104     }
2105     else
2106         last_emu_check = h->out.bs.p;
2107     h->mb.i_last_qp = h->sh.i_qp;
2108     h->mb.i_last_dqp = 0;
2109     h->mb.field_decoding_flag = 0;
2110
2111     i_mb_y = h->sh.i_first_mb / h->mb.i_mb_width;
2112     i_mb_x = h->sh.i_first_mb % h->mb.i_mb_width;
2113     i_skip = 0;
2114
2115     while( 1 )
2116     {
2117         mb_xy = i_mb_x + i_mb_y * h->mb.i_mb_width;
2118         int mb_spos = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
2119
2120         if( !(i_mb_y & SLICE_MBAFF) )
2121         {
2122             if( x264_bitstream_check_buffer( h ) )
2123                 return -1;
2124
2125             if( back_up_bitstream )
2126             {
2127                 mv_bits_bak = h->stat.frame.i_mv_bits;
2128                 tex_bits_bak = h->stat.frame.i_tex_bits;
2129                 /* We don't need the contexts because flushing the CABAC encoder has no context
2130                  * dependency and macroblocks are only re-encoded in the case where a slice is
2131                  * ended (and thus the content of all contexts are thrown away). */
2132                 if( h->param.b_cabac )
2133                 {
2134                     memcpy( &cabac_bak, &h->cabac, offsetof(x264_cabac_t, f8_bits_encoded) );
2135                     /* x264's CABAC writer modifies the previous byte during carry, so it has to be
2136                      * backed up. */
2137                     cabac_prevbyte_bak = h->cabac.p[-1];
2138                 }
2139                 else
2140                 {
2141                     bs_bak = h->out.bs;
2142                     i_skip_bak = i_skip;
2143                 }
2144             }
2145         }
2146
2147         if( i_mb_x == 0 && !h->mb.b_reencode_mb )
2148             x264_fdec_filter_row( h, i_mb_y, 1 );
2149
2150         if( PARAM_INTERLACED )
2151         {
2152             if( h->mb.b_adaptive_mbaff )
2153             {
2154                 if( !(i_mb_y&1) )
2155                 {
2156                     /* FIXME: VSAD is fast but fairly poor at choosing the best interlace type. */
2157                     h->mb.b_interlaced = x264_field_vsad( h, i_mb_x, i_mb_y );
2158                     memcpy( &h->zigzagf, MB_INTERLACED ? &h->zigzagf_interlaced : &h->zigzagf_progressive, sizeof(h->zigzagf) );
2159                     if( !MB_INTERLACED && (i_mb_y+2) == h->mb.i_mb_height )
2160                         x264_expand_border_mbpair( h, i_mb_x, i_mb_y );
2161                 }
2162             }
2163             h->mb.field[mb_xy] = MB_INTERLACED;
2164         }
2165
2166         /* load cache */
2167         if( SLICE_MBAFF )
2168             x264_macroblock_cache_load_interlaced( h, i_mb_x, i_mb_y );
2169         else
2170             x264_macroblock_cache_load_progressive( h, i_mb_x, i_mb_y );
2171
2172         x264_macroblock_analyse( h );
2173
2174         /* encode this macroblock -> be careful it can change the mb type to P_SKIP if needed */
2175 reencode:
2176         x264_macroblock_encode( h );
2177
2178         if( h->param.b_cabac )
2179         {
2180             if( mb_xy > h->sh.i_first_mb && !(SLICE_MBAFF && (i_mb_y&1)) )
2181                 x264_cabac_encode_terminal( &h->cabac );
2182
2183             if( IS_SKIP( h->mb.i_type ) )
2184                 x264_cabac_mb_skip( h, 1 );
2185             else
2186             {
2187                 if( h->sh.i_type != SLICE_TYPE_I )
2188                     x264_cabac_mb_skip( h, 0 );
2189                 x264_macroblock_write_cabac( h, &h->cabac );
2190             }
2191         }
2192         else
2193         {
2194             if( IS_SKIP( h->mb.i_type ) )
2195                 i_skip++;
2196             else
2197             {
2198                 if( h->sh.i_type != SLICE_TYPE_I )
2199                 {
2200                     bs_write_ue( &h->out.bs, i_skip );  /* skip run */
2201                     i_skip = 0;
2202                 }
2203                 x264_macroblock_write_cavlc( h );
2204                 /* If there was a CAVLC level code overflow, try again at a higher QP. */
2205                 if( h->mb.b_overflow )
2206                 {
2207                     h->mb.i_chroma_qp = h->chroma_qp_table[++h->mb.i_qp];
2208                     h->mb.i_skip_intra = 0;
2209                     h->mb.b_skip_mc = 0;
2210                     h->mb.b_overflow = 0;
2211                     h->out.bs = bs_bak;
2212                     i_skip = i_skip_bak;
2213                     h->stat.frame.i_mv_bits = mv_bits_bak;
2214                     h->stat.frame.i_tex_bits = tex_bits_bak;
2215                     goto reencode;
2216                 }
2217             }
2218         }
2219
2220         int total_bits = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
2221         int mb_size = total_bits - mb_spos;
2222
2223         if( slice_max_size )
2224         {
2225             /* Count the skip run, just in case. */
2226             if( !h->param.b_cabac )
2227                 total_bits += bs_size_ue_big( i_skip );
2228             /* Check for escape bytes. */
2229             uint8_t *end = h->param.b_cabac ? h->cabac.p : h->out.bs.p;
2230             for( ; last_emu_check < end - 2; last_emu_check++ )
2231                 if( last_emu_check[0] == 0 && last_emu_check[1] == 0 && last_emu_check[2] <= 3 )
2232                 {
2233                     slice_max_size -= 8;
2234                     last_emu_check++;
2235                 }
2236             /* We'll just re-encode this last macroblock if we go over the max slice size. */
2237             if( total_bits - starting_bits > slice_max_size && !h->mb.b_reencode_mb )
2238             {
2239                 if( mb_xy != h->sh.i_first_mb )
2240                 {
2241                     h->stat.frame.i_mv_bits = mv_bits_bak;
2242                     h->stat.frame.i_tex_bits = tex_bits_bak;
2243                     if( h->param.b_cabac )
2244                     {
2245                         memcpy( &h->cabac, &cabac_bak, offsetof(x264_cabac_t, f8_bits_encoded) );
2246                         h->cabac.p[-1] = cabac_prevbyte_bak;
2247                     }
2248                     else
2249                     {
2250                         h->out.bs = bs_bak;
2251                         i_skip = i_skip_bak;
2252                     }
2253                     h->mb.b_reencode_mb = 1;
2254                     if( SLICE_MBAFF )
2255                     {
2256                         // set to bottom of previous mbpair
2257                         if( i_mb_x )
2258                             h->sh.i_last_mb = mb_xy-1+h->mb.i_mb_stride*(!(i_mb_y&1));
2259                         else
2260                             h->sh.i_last_mb = (i_mb_y-2+!(i_mb_y&1))*h->mb.i_mb_stride + h->mb.i_mb_width - 1;
2261                     }
2262                     else
2263                         h->sh.i_last_mb = mb_xy-1;
2264                     break;
2265                 }
2266                 else
2267                 {
2268                     h->sh.i_last_mb = mb_xy;
2269                     h->mb.b_reencode_mb = 0;
2270                 }
2271             }
2272             else
2273                 h->mb.b_reencode_mb = 0;
2274         }
2275
2276 #if HAVE_VISUALIZE
2277         if( h->param.b_visualize )
2278             x264_visualize_mb( h );
2279 #endif
2280
2281         /* save cache */
2282         x264_macroblock_cache_save( h );
2283
2284         /* accumulate mb stats */
2285         h->stat.frame.i_mb_count[h->mb.i_type]++;
2286
2287         int b_intra = IS_INTRA( h->mb.i_type );
2288         int b_skip = IS_SKIP( h->mb.i_type );
2289         if( h->param.i_log_level >= X264_LOG_INFO || h->param.rc.b_stat_write )
2290         {
2291             if( !b_intra && !b_skip && !IS_DIRECT( h->mb.i_type ) )
2292             {
2293                 if( h->mb.i_partition != D_8x8 )
2294                         h->stat.frame.i_mb_partition[h->mb.i_partition] += 4;
2295                     else
2296                         for( int i = 0; i < 4; i++ )
2297                             h->stat.frame.i_mb_partition[h->mb.i_sub_partition[i]] ++;
2298                 if( h->param.i_frame_reference > 1 )
2299                     for( int i_list = 0; i_list <= (h->sh.i_type == SLICE_TYPE_B); i_list++ )
2300                         for( int i = 0; i < 4; i++ )
2301                         {
2302                             int i_ref = h->mb.cache.ref[i_list][ x264_scan8[4*i] ];
2303                             if( i_ref >= 0 )
2304                                 h->stat.frame.i_mb_count_ref[i_list][i_ref] ++;
2305                         }
2306             }
2307         }
2308
2309         if( h->param.i_log_level >= X264_LOG_INFO )
2310         {
2311             if( h->mb.i_cbp_luma | h->mb.i_cbp_chroma )
2312             {
2313                 if( CHROMA444 )
2314                 {
2315                     for( int i = 0; i < 4; i++ )
2316                         if( h->mb.i_cbp_luma & (1 << i) )
2317                             for( int p = 0; p < 3; p++ )
2318                             {
2319                                 int s8 = i*4+p*16;
2320                                 int nnz8x8 = M16( &h->mb.cache.non_zero_count[x264_scan8[s8]+0] )
2321                                            | M16( &h->mb.cache.non_zero_count[x264_scan8[s8]+8] );
2322                                 h->stat.frame.i_mb_cbp[!b_intra + p*2] += !!nnz8x8;
2323                             }
2324                 }
2325                 else
2326                 {
2327                     int cbpsum = (h->mb.i_cbp_luma&1) + ((h->mb.i_cbp_luma>>1)&1)
2328                                + ((h->mb.i_cbp_luma>>2)&1) + (h->mb.i_cbp_luma>>3);
2329                     h->stat.frame.i_mb_cbp[!b_intra + 0] += cbpsum;
2330                     h->stat.frame.i_mb_cbp[!b_intra + 2] += !!h->mb.i_cbp_chroma;
2331                     h->stat.frame.i_mb_cbp[!b_intra + 4] += h->mb.i_cbp_chroma >> 1;
2332                 }
2333             }
2334             if( h->mb.i_cbp_luma && !b_intra )
2335             {
2336                 h->stat.frame.i_mb_count_8x8dct[0] ++;
2337                 h->stat.frame.i_mb_count_8x8dct[1] += h->mb.b_transform_8x8;
2338             }
2339             if( b_intra && h->mb.i_type != I_PCM )
2340             {
2341                 if( h->mb.i_type == I_16x16 )
2342                     h->stat.frame.i_mb_pred_mode[0][h->mb.i_intra16x16_pred_mode]++;
2343                 else if( h->mb.i_type == I_8x8 )
2344                     for( int i = 0; i < 16; i += 4 )
2345                         h->stat.frame.i_mb_pred_mode[1][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
2346                 else //if( h->mb.i_type == I_4x4 )
2347                     for( int i = 0; i < 16; i++ )
2348                         h->stat.frame.i_mb_pred_mode[2][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
2349                 h->stat.frame.i_mb_pred_mode[3][x264_mb_chroma_pred_mode_fix[h->mb.i_chroma_pred_mode]]++;
2350             }
2351             h->stat.frame.i_mb_field[b_intra?0:b_skip?2:1] += MB_INTERLACED;
2352         }
2353
2354         /* calculate deblock strength values (actual deblocking is done per-row along with hpel) */
2355         if( b_deblock )
2356             x264_macroblock_deblock_strength( h );
2357
2358         x264_ratecontrol_mb( h, mb_size );
2359
2360         if( mb_xy == h->sh.i_last_mb )
2361             break;
2362
2363         if( SLICE_MBAFF )
2364         {
2365             i_mb_x += i_mb_y & 1;
2366             i_mb_y ^= i_mb_x < h->mb.i_mb_width;
2367         }
2368         else
2369             i_mb_x++;
2370         if( i_mb_x == h->mb.i_mb_width )
2371         {
2372             i_mb_y++;
2373             i_mb_x = 0;
2374         }
2375     }
2376     h->out.nal[h->out.i_nal].i_last_mb = h->sh.i_last_mb;
2377
2378     if( h->param.b_cabac )
2379     {
2380         x264_cabac_encode_flush( h, &h->cabac );
2381         h->out.bs.p = h->cabac.p;
2382     }
2383     else
2384     {
2385         if( i_skip > 0 )
2386             bs_write_ue( &h->out.bs, i_skip );  /* last skip run */
2387         /* rbsp_slice_trailing_bits */
2388         bs_rbsp_trailing( &h->out.bs );
2389         bs_flush( &h->out.bs );
2390     }
2391     if( x264_nal_end( h ) )
2392         return -1;
2393
2394     if( h->sh.i_last_mb == (h->i_threadslice_end * h->mb.i_mb_width - 1) )
2395     {
2396         h->stat.frame.i_misc_bits = bs_pos( &h->out.bs )
2397                                   + (h->out.i_nal*NALU_OVERHEAD * 8)
2398                                   - h->stat.frame.i_tex_bits
2399                                   - h->stat.frame.i_mv_bits;
2400         x264_fdec_filter_row( h, h->i_threadslice_end, 1 );
2401     }
2402
2403     return 0;
2404 }
2405
2406 static void x264_thread_sync_context( x264_t *dst, x264_t *src )
2407 {
2408     if( dst == src )
2409         return;
2410
2411     // reference counting
2412     for( x264_frame_t **f = src->frames.reference; *f; f++ )
2413         (*f)->i_reference_count++;
2414     for( x264_frame_t **f = dst->frames.reference; *f; f++ )
2415         x264_frame_push_unused( src, *f );
2416     src->fdec->i_reference_count++;
2417     x264_frame_push_unused( src, dst->fdec );
2418
2419     // copy everything except the per-thread pointers and the constants.
2420     memcpy( &dst->i_frame, &src->i_frame, offsetof(x264_t, mb.type) - offsetof(x264_t, i_frame) );
2421     dst->param = src->param;
2422     dst->stat = src->stat;
2423     dst->pixf = src->pixf;
2424 }
2425
2426 static void x264_thread_sync_stat( x264_t *dst, x264_t *src )
2427 {
2428     if( dst == src )
2429         return;
2430     memcpy( &dst->stat.i_frame_count, &src->stat.i_frame_count, sizeof(dst->stat) - sizeof(dst->stat.frame) );
2431 }
2432
2433 static void *x264_slices_write( x264_t *h )
2434 {
2435     int i_slice_num = 0;
2436     int last_thread_mb = h->sh.i_last_mb;
2437
2438 #if HAVE_VISUALIZE
2439     if( h->param.b_visualize )
2440         if( x264_visualize_init( h ) )
2441             return (void *)-1;
2442 #endif
2443
2444     /* init stats */
2445     memset( &h->stat.frame, 0, sizeof(h->stat.frame) );
2446     h->mb.b_reencode_mb = 0;
2447     while( h->sh.i_first_mb + SLICE_MBAFF*h->mb.i_mb_stride <= last_thread_mb )
2448     {
2449         h->sh.i_last_mb = last_thread_mb;
2450         if( h->param.i_slice_max_mbs )
2451         {
2452             if( SLICE_MBAFF )
2453             {
2454                 // convert first to mbaff form, add slice-max-mbs, then convert back to normal form
2455                 int last_mbaff = 2*(h->sh.i_first_mb % h->mb.i_mb_width)
2456                     + h->mb.i_mb_width*(h->sh.i_first_mb / h->mb.i_mb_width)
2457                     + h->param.i_slice_max_mbs - 1;
2458                 int last_x = (last_mbaff % (2*h->mb.i_mb_width))/2;
2459                 int last_y = (last_mbaff / (2*h->mb.i_mb_width))*2 + 1;
2460                 h->sh.i_last_mb = last_x + h->mb.i_mb_stride*last_y;
2461             }
2462             else
2463                 h->sh.i_last_mb = h->sh.i_first_mb + h->param.i_slice_max_mbs - 1;
2464         }
2465         else if( h->param.i_slice_count && !h->param.b_sliced_threads )
2466         {
2467             int height = h->mb.i_mb_height >> PARAM_INTERLACED;
2468             int width = h->mb.i_mb_width << PARAM_INTERLACED;
2469             i_slice_num++;
2470             h->sh.i_last_mb = (height * i_slice_num + h->param.i_slice_count/2) / h->param.i_slice_count * width - 1;
2471         }
2472         h->sh.i_last_mb = X264_MIN( h->sh.i_last_mb, last_thread_mb );
2473         if( x264_stack_align( x264_slice_write, h ) )
2474             return (void *)-1;
2475         h->sh.i_first_mb = h->sh.i_last_mb + 1;
2476         // if i_first_mb is not the last mb in a row then go to the next mb in MBAFF order
2477         if( SLICE_MBAFF && h->sh.i_first_mb % h->mb.i_mb_width )
2478             h->sh.i_first_mb -= h->mb.i_mb_stride;
2479     }
2480
2481 #if HAVE_VISUALIZE
2482     if( h->param.b_visualize )
2483     {
2484         x264_visualize_show( h );
2485         x264_visualize_close( h );
2486     }
2487 #endif
2488
2489     return (void *)0;
2490 }
2491
2492 static int x264_threaded_slices_write( x264_t *h )
2493 {
2494     /* set first/last mb and sync contexts */
2495     for( int i = 0; i < h->param.i_threads; i++ )
2496     {
2497         x264_t *t = h->thread[i];
2498         if( i )
2499         {
2500             t->param = h->param;
2501             memcpy( &t->i_frame, &h->i_frame, offsetof(x264_t, rc) - offsetof(x264_t, i_frame) );
2502         }
2503         int height = h->mb.i_mb_height >> PARAM_INTERLACED;
2504         t->i_threadslice_start = ((height *  i    + h->param.i_slice_count/2) / h->param.i_threads) << PARAM_INTERLACED;
2505         t->i_threadslice_end   = ((height * (i+1) + h->param.i_slice_count/2) / h->param.i_threads) << PARAM_INTERLACED;
2506         t->sh.i_first_mb = t->i_threadslice_start * h->mb.i_mb_width;
2507         t->sh.i_last_mb  =   t->i_threadslice_end * h->mb.i_mb_width - 1;
2508     }
2509
2510     x264_stack_align( x264_analyse_weight_frame, h, h->mb.i_mb_height*16 + 16 );
2511
2512     x264_threads_distribute_ratecontrol( h );
2513
2514     /* dispatch */
2515     for( int i = 0; i < h->param.i_threads; i++ )
2516     {
2517         x264_threadpool_run( h->threadpool, (void*)x264_slices_write, h->thread[i] );
2518         h->thread[i]->b_thread_active = 1;
2519     }
2520     for( int i = 0; i < h->param.i_threads; i++ )
2521     {
2522         h->thread[i]->b_thread_active = 0;
2523         if( (intptr_t)x264_threadpool_wait( h->threadpool, h->thread[i] ) )
2524             return -1;
2525     }
2526
2527     /* Go back and fix up the hpel on the borders between slices. */
2528     for( int i = 1; i < h->param.i_threads; i++ )
2529     {
2530         x264_fdec_filter_row( h->thread[i], h->thread[i]->i_threadslice_start + 1, 0 );
2531         if( SLICE_MBAFF )
2532             x264_fdec_filter_row( h->thread[i], h->thread[i]->i_threadslice_start + 2, 0 );
2533     }
2534
2535     x264_threads_merge_ratecontrol( h );
2536
2537     for( int i = 1; i < h->param.i_threads; i++ )
2538     {
2539         x264_t *t = h->thread[i];
2540         for( int j = 0; j < t->out.i_nal; j++ )
2541         {
2542             h->out.nal[h->out.i_nal] = t->out.nal[j];
2543             h->out.i_nal++;
2544             x264_nal_check_buffer( h );
2545         }
2546         /* All entries in stat.frame are ints except for ssd/ssim. */
2547         for( int j = 0; j < (offsetof(x264_t,stat.frame.i_ssd) - offsetof(x264_t,stat.frame.i_mv_bits)) / sizeof(int); j++ )
2548             ((int*)&h->stat.frame)[j] += ((int*)&t->stat.frame)[j];
2549         for( int j = 0; j < 3; j++ )
2550             h->stat.frame.i_ssd[j] += t->stat.frame.i_ssd[j];
2551         h->stat.frame.f_ssim += t->stat.frame.f_ssim;
2552         h->stat.frame.i_ssim_cnt += t->stat.frame.i_ssim_cnt;
2553     }
2554
2555     return 0;
2556 }
2557
2558 void x264_encoder_intra_refresh( x264_t *h )
2559 {
2560     h = h->thread[h->i_thread_phase];
2561     h->b_queued_intra_refresh = 1;
2562 }
2563
2564 int x264_encoder_invalidate_reference( x264_t *h, int64_t pts )
2565 {
2566     if( h->param.i_bframe )
2567     {
2568         x264_log( h, X264_LOG_ERROR, "x264_encoder_invalidate_reference is not supported with B-frames enabled\n" );
2569         return -1;
2570     }
2571     if( h->param.b_intra_refresh )
2572     {
2573         x264_log( h, X264_LOG_ERROR, "x264_encoder_invalidate_reference is not supported with intra refresh enabled\n" );
2574         return -1;
2575     }
2576     h = h->thread[h->i_thread_phase];
2577     if( pts >= h->i_last_idr_pts )
2578     {
2579         for( int i = 0; h->frames.reference[i]; i++ )
2580             if( pts <= h->frames.reference[i]->i_pts )
2581                 h->frames.reference[i]->b_corrupt = 1;
2582         if( pts <= h->fdec->i_pts )
2583             h->fdec->b_corrupt = 1;
2584     }
2585     return 0;
2586 }
2587
2588 /****************************************************************************
2589  * x264_encoder_encode:
2590  *  XXX: i_poc   : is the poc of the current given picture
2591  *       i_frame : is the number of the frame being coded
2592  *  ex:  type frame poc
2593  *       I      0   2*0
2594  *       P      1   2*3
2595  *       B      2   2*1
2596  *       B      3   2*2
2597  *       P      4   2*6
2598  *       B      5   2*4
2599  *       B      6   2*5
2600  ****************************************************************************/
2601 int     x264_encoder_encode( x264_t *h,
2602                              x264_nal_t **pp_nal, int *pi_nal,
2603                              x264_picture_t *pic_in,
2604                              x264_picture_t *pic_out )
2605 {
2606     x264_t *thread_current, *thread_prev, *thread_oldest;
2607     int i_nal_type, i_nal_ref_idc, i_global_qp;
2608     int overhead = NALU_OVERHEAD;
2609
2610     if( h->i_thread_frames > 1 )
2611     {
2612         thread_prev    = h->thread[ h->i_thread_phase ];
2613         h->i_thread_phase = (h->i_thread_phase + 1) % h->i_thread_frames;
2614         thread_current = h->thread[ h->i_thread_phase ];
2615         thread_oldest  = h->thread[ (h->i_thread_phase + 1) % h->i_thread_frames ];
2616         x264_thread_sync_context( thread_current, thread_prev );
2617         x264_thread_sync_ratecontrol( thread_current, thread_prev, thread_oldest );
2618         h = thread_current;
2619     }
2620     else
2621     {
2622         thread_current =
2623         thread_oldest  = h;
2624     }
2625 #if HAVE_MMX
2626     if( h->param.cpu&X264_CPU_SSE_MISALIGN )
2627         x264_cpu_mask_misalign_sse();
2628 #endif
2629
2630     // ok to call this before encoding any frames, since the initial values of fdec have b_kept_as_ref=0
2631     if( x264_reference_update( h ) )
2632         return -1;
2633     h->fdec->i_lines_completed = -1;
2634
2635     /* no data out */
2636     *pi_nal = 0;
2637     *pp_nal = NULL;
2638
2639     /* ------------------- Setup new frame from picture -------------------- */
2640     if( pic_in != NULL )
2641     {
2642         /* 1: Copy the picture to a frame and move it to a buffer */
2643         x264_frame_t *fenc = x264_frame_pop_unused( h, 0 );
2644         if( !fenc )
2645             return -1;
2646
2647         if( x264_frame_copy_picture( h, fenc, pic_in ) < 0 )
2648             return -1;
2649
2650         if( h->param.i_width != 16 * h->mb.i_mb_width ||
2651             h->param.i_height != 16 * h->mb.i_mb_height )
2652             x264_frame_expand_border_mod16( h, fenc );
2653
2654         fenc->i_frame = h->frames.i_input++;
2655
2656         if( fenc->i_frame == 0 )
2657             h->frames.i_first_pts = fenc->i_pts;
2658         if( h->frames.i_bframe_delay && fenc->i_frame == h->frames.i_bframe_delay )
2659             h->frames.i_bframe_delay_time = fenc->i_pts - h->frames.i_first_pts;
2660
2661         if( h->param.b_vfr_input && fenc->i_pts <= h->frames.i_largest_pts )
2662             x264_log( h, X264_LOG_WARNING, "non-strictly-monotonic PTS\n" );
2663
2664         h->frames.i_second_largest_pts = h->frames.i_largest_pts;
2665         h->frames.i_largest_pts = fenc->i_pts;
2666
2667         if( (fenc->i_pic_struct < PIC_STRUCT_AUTO) || (fenc->i_pic_struct > PIC_STRUCT_TRIPLE) )
2668             fenc->i_pic_struct = PIC_STRUCT_AUTO;
2669
2670         if( fenc->i_pic_struct == PIC_STRUCT_AUTO )
2671         {
2672 #if HAVE_INTERLACED
2673             int b_interlaced = fenc->param ? fenc->param->b_interlaced : h->param.b_interlaced;
2674 #else
2675             int b_interlaced = 0;
2676 #endif
2677             if( b_interlaced )
2678             {
2679                 int b_tff = fenc->param ? fenc->param->b_tff : h->param.b_tff;
2680                 fenc->i_pic_struct = b_tff ? PIC_STRUCT_TOP_BOTTOM : PIC_STRUCT_BOTTOM_TOP;
2681             }
2682             else
2683                 fenc->i_pic_struct = PIC_STRUCT_PROGRESSIVE;
2684         }
2685
2686         if( h->param.rc.b_mb_tree && h->param.rc.b_stat_read )
2687         {
2688             if( x264_macroblock_tree_read( h, fenc, pic_in->prop.quant_offsets ) )
2689                 return -1;
2690         }
2691         else
2692             x264_stack_align( x264_adaptive_quant_frame, h, fenc, pic_in->prop.quant_offsets );
2693
2694         if( pic_in->prop.quant_offsets_free )
2695             pic_in->prop.quant_offsets_free( pic_in->prop.quant_offsets );
2696
2697         if( h->frames.b_have_lowres )
2698             x264_frame_init_lowres( h, fenc );
2699
2700         /* 2: Place the frame into the queue for its slice type decision */
2701         x264_lookahead_put_frame( h, fenc );
2702
2703         if( h->frames.i_input <= h->frames.i_delay + 1 - h->i_thread_frames )
2704         {
2705             /* Nothing yet to encode, waiting for filling of buffers */
2706             pic_out->i_type = X264_TYPE_AUTO;
2707             return 0;
2708         }
2709     }
2710     else
2711     {
2712         /* signal kills for lookahead thread */
2713         x264_pthread_mutex_lock( &h->lookahead->ifbuf.mutex );
2714         h->lookahead->b_exit_thread = 1;
2715         x264_pthread_cond_broadcast( &h->lookahead->ifbuf.cv_fill );
2716         x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
2717     }
2718
2719     h->i_frame++;
2720     /* 3: The picture is analyzed in the lookahead */
2721     if( !h->frames.current[0] )
2722         x264_lookahead_get_frames( h );
2723
2724     if( !h->frames.current[0] && x264_lookahead_is_empty( h ) )
2725         return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
2726
2727     /* ------------------- Get frame to be encoded ------------------------- */
2728     /* 4: get picture to encode */
2729     h->fenc = x264_frame_shift( h->frames.current );
2730     if( h->i_frame == h->i_thread_frames - 1 )
2731         h->i_reordered_pts_delay = h->fenc->i_reordered_pts;
2732     if( h->fenc->param )
2733     {
2734         x264_encoder_reconfig( h, h->fenc->param );
2735         if( h->fenc->param->param_free )
2736             h->fenc->param->param_free( h->fenc->param );
2737     }
2738
2739     if( !IS_X264_TYPE_I( h->fenc->i_type ) )
2740     {
2741         int valid_refs_left = 0;
2742         for( int i = 0; h->frames.reference[i]; i++ )
2743             if( !h->frames.reference[i]->b_corrupt )
2744                 valid_refs_left++;
2745         /* No valid reference frames left: force an IDR. */
2746         if( !valid_refs_left )
2747         {
2748             h->fenc->b_keyframe = 1;
2749             h->fenc->i_type = X264_TYPE_IDR;
2750         }
2751     }
2752
2753     if( h->fenc->b_keyframe )
2754     {
2755         h->frames.i_last_keyframe = h->fenc->i_frame;
2756         if( h->fenc->i_type == X264_TYPE_IDR )
2757         {
2758             h->i_frame_num = 0;
2759             h->frames.i_last_idr = h->fenc->i_frame;
2760         }
2761     }
2762     h->sh.i_mmco_command_count =
2763     h->sh.i_mmco_remove_from_end = 0;
2764     h->b_ref_reorder[0] =
2765     h->b_ref_reorder[1] = 0;
2766     h->fdec->i_poc =
2767     h->fenc->i_poc = 2 * ( h->fenc->i_frame - X264_MAX( h->frames.i_last_idr, 0 ) );
2768
2769     /* ------------------- Setup frame context ----------------------------- */
2770     /* 5: Init data dependent of frame type */
2771     if( h->fenc->i_type == X264_TYPE_IDR )
2772     {
2773         /* reset ref pictures */
2774         i_nal_type    = NAL_SLICE_IDR;
2775         i_nal_ref_idc = NAL_PRIORITY_HIGHEST;
2776         h->sh.i_type = SLICE_TYPE_I;
2777         x264_reference_reset( h );
2778         h->frames.i_poc_last_open_gop = -1;
2779     }
2780     else if( h->fenc->i_type == X264_TYPE_I )
2781     {
2782         i_nal_type    = NAL_SLICE;
2783         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
2784         h->sh.i_type = SLICE_TYPE_I;
2785         x264_reference_hierarchy_reset( h );
2786         if( h->param.b_open_gop )
2787             h->frames.i_poc_last_open_gop = h->fenc->b_keyframe ? h->fenc->i_poc : -1;
2788     }
2789     else if( h->fenc->i_type == X264_TYPE_P )
2790     {
2791         i_nal_type    = NAL_SLICE;
2792         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
2793         h->sh.i_type = SLICE_TYPE_P;
2794         x264_reference_hierarchy_reset( h );
2795         h->frames.i_poc_last_open_gop = -1;
2796     }
2797     else if( h->fenc->i_type == X264_TYPE_BREF )
2798     {
2799         i_nal_type    = NAL_SLICE;
2800         i_nal_ref_idc = h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT ? NAL_PRIORITY_LOW : NAL_PRIORITY_HIGH;
2801         h->sh.i_type = SLICE_TYPE_B;
2802         x264_reference_hierarchy_reset( h );
2803     }
2804     else    /* B frame */
2805     {
2806         i_nal_type    = NAL_SLICE;
2807         i_nal_ref_idc = NAL_PRIORITY_DISPOSABLE;
2808         h->sh.i_type = SLICE_TYPE_B;
2809     }
2810
2811     h->fdec->i_type = h->fenc->i_type;
2812     h->fdec->i_frame = h->fenc->i_frame;
2813     h->fenc->b_kept_as_ref =
2814     h->fdec->b_kept_as_ref = i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE && h->param.i_keyint_max > 1;
2815
2816     h->fdec->i_pts = h->fenc->i_pts;
2817     if( h->frames.i_bframe_delay )
2818     {
2819         int64_t *prev_reordered_pts = thread_current->frames.i_prev_reordered_pts;
2820         h->fdec->i_dts = h->i_frame > h->frames.i_bframe_delay
2821                        ? prev_reordered_pts[ (h->i_frame - h->frames.i_bframe_delay) % h->frames.i_bframe_delay ]
2822                        : h->fenc->i_reordered_pts - h->frames.i_bframe_delay_time;
2823         prev_reordered_pts[ h->i_frame % h->frames.i_bframe_delay ] = h->fenc->i_reordered_pts;
2824     }
2825     else
2826         h->fdec->i_dts = h->fenc->i_reordered_pts;
2827     if( h->fenc->i_type == X264_TYPE_IDR )
2828         h->i_last_idr_pts = h->fdec->i_pts;
2829
2830     /* ------------------- Init                ----------------------------- */
2831     /* build ref list 0/1 */
2832     x264_reference_build_list( h, h->fdec->i_poc );
2833
2834     /* ---------------------- Write the bitstream -------------------------- */
2835     /* Init bitstream context */
2836     if( h->param.b_sliced_threads )
2837     {
2838         for( int i = 0; i < h->param.i_threads; i++ )
2839         {
2840             bs_init( &h->thread[i]->out.bs, h->thread[i]->out.p_bitstream, h->thread[i]->out.i_bitstream );
2841             h->thread[i]->out.i_nal = 0;
2842         }
2843     }
2844     else
2845     {
2846         bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
2847         h->out.i_nal = 0;
2848     }
2849
2850     if( h->param.b_aud )
2851     {
2852         int pic_type;
2853
2854         if( h->sh.i_type == SLICE_TYPE_I )
2855             pic_type = 0;
2856         else if( h->sh.i_type == SLICE_TYPE_P )
2857             pic_type = 1;
2858         else if( h->sh.i_type == SLICE_TYPE_B )
2859             pic_type = 2;
2860         else
2861             pic_type = 7;
2862
2863         x264_nal_start( h, NAL_AUD, NAL_PRIORITY_DISPOSABLE );
2864         bs_write( &h->out.bs, 3, pic_type );
2865         bs_rbsp_trailing( &h->out.bs );
2866         if( x264_nal_end( h ) )
2867             return -1;
2868         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2869     }
2870
2871     h->i_nal_type = i_nal_type;
2872     h->i_nal_ref_idc = i_nal_ref_idc;
2873
2874     if( h->param.b_intra_refresh )
2875     {
2876         if( IS_X264_TYPE_I( h->fenc->i_type ) )
2877         {
2878             h->fdec->i_frames_since_pir = 0;
2879             h->b_queued_intra_refresh = 0;
2880             /* PIR is currently only supported with ref == 1, so any intra frame effectively refreshes
2881              * the whole frame and counts as an intra refresh. */
2882             h->fdec->f_pir_position = h->mb.i_mb_width;
2883         }
2884         else if( h->fenc->i_type == X264_TYPE_P )
2885         {
2886             int pocdiff = (h->fdec->i_poc - h->fref[0][0]->i_poc)/2;
2887             float increment = X264_MAX( ((float)h->mb.i_mb_width-1) / h->param.i_keyint_max, 1 );
2888             h->fdec->f_pir_position = h->fref[0][0]->f_pir_position;
2889             h->fdec->i_frames_since_pir = h->fref[0][0]->i_frames_since_pir + pocdiff;
2890             if( h->fdec->i_frames_since_pir >= h->param.i_keyint_max ||
2891                 (h->b_queued_intra_refresh && h->fdec->f_pir_position + 0.5 >= h->mb.i_mb_width) )
2892             {
2893                 h->fdec->f_pir_position = 0;
2894                 h->fdec->i_frames_since_pir = 0;
2895                 h->b_queued_intra_refresh = 0;
2896                 h->fenc->b_keyframe = 1;
2897             }
2898             h->fdec->i_pir_start_col = h->fdec->f_pir_position+0.5;
2899             h->fdec->f_pir_position += increment * pocdiff;
2900             h->fdec->i_pir_end_col = h->fdec->f_pir_position+0.5;
2901             /* If our intra refresh has reached the right side of the frame, we're done. */
2902             if( h->fdec->i_pir_end_col >= h->mb.i_mb_width - 1 )
2903                 h->fdec->f_pir_position = h->mb.i_mb_width;
2904         }
2905     }
2906
2907     if( h->fenc->b_keyframe )
2908     {
2909         /* Write SPS and PPS */
2910         if( h->param.b_repeat_headers )
2911         {
2912             /* generate sequence parameters */
2913             x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
2914             x264_sps_write( &h->out.bs, h->sps );
2915             if( x264_nal_end( h ) )
2916                 return -1;
2917             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
2918
2919             /* generate picture parameters */
2920             x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
2921             x264_pps_write( &h->out.bs, h->sps, h->pps );
2922             if( x264_nal_end( h ) )
2923                 return -1;
2924             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
2925         }
2926
2927         /* when frame threading is used, buffering period sei is written in x264_encoder_frame_end */
2928         if( h->i_thread_frames == 1 && h->sps->vui.b_nal_hrd_parameters_present )
2929         {
2930             x264_hrd_fullness( h );
2931             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2932             x264_sei_buffering_period_write( h, &h->out.bs );
2933             if( x264_nal_end( h ) )
2934                return -1;
2935             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
2936         }
2937     }
2938
2939     /* write extra sei */
2940     for( int i = 0; i < h->fenc->extra_sei.num_payloads; i++ )
2941     {
2942         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2943         x264_sei_write( &h->out.bs, h->fenc->extra_sei.payloads[i].payload, h->fenc->extra_sei.payloads[i].payload_size,
2944                         h->fenc->extra_sei.payloads[i].payload_type );
2945         if( x264_nal_end( h ) )
2946             return -1;
2947         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2948         if( h->fenc->extra_sei.sei_free && h->fenc->extra_sei.payloads[i].payload )
2949             h->fenc->extra_sei.sei_free( h->fenc->extra_sei.payloads[i].payload );
2950     }
2951
2952     if( h->fenc->extra_sei.sei_free && h->fenc->extra_sei.payloads )
2953         h->fenc->extra_sei.sei_free( h->fenc->extra_sei.payloads );
2954
2955     if( h->fenc->b_keyframe )
2956     {
2957         if( h->param.b_repeat_headers && h->fenc->i_frame == 0 )
2958         {
2959             /* identify ourself */
2960             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2961             if( x264_sei_version_write( h, &h->out.bs ) )
2962                 return -1;
2963             if( x264_nal_end( h ) )
2964                 return -1;
2965             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2966         }
2967
2968         if( h->fenc->i_type != X264_TYPE_IDR )
2969         {
2970             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;
2971             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2972             x264_sei_recovery_point_write( h, &h->out.bs, time_to_recovery );
2973             if( x264_nal_end( h ) )
2974                 return -1;
2975             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2976         }
2977
2978         if ( h->param.i_frame_packing >= 0 )
2979         {
2980             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2981             x264_sei_frame_packing_write( h, &h->out.bs );
2982             if( x264_nal_end( h ) )
2983                 return -1;
2984             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2985         }
2986     }
2987
2988     /* generate sei pic timing */
2989     if( h->sps->vui.b_pic_struct_present || h->sps->vui.b_nal_hrd_parameters_present )
2990     {
2991         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2992         x264_sei_pic_timing_write( h, &h->out.bs );
2993         if( x264_nal_end( h ) )
2994             return -1;
2995         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2996     }
2997
2998     /* As required by Blu-ray. */
2999     if( !IS_X264_TYPE_B( h->fenc->i_type ) && h->b_sh_backup )
3000     {
3001         h->b_sh_backup = 0;
3002         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3003         x264_sei_dec_ref_pic_marking_write( h, &h->out.bs );
3004         if( x264_nal_end( h ) )
3005             return -1;
3006         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
3007     }
3008
3009     if( h->fenc->b_keyframe && h->param.b_intra_refresh )
3010         h->i_cpb_delay_pir_offset = h->fenc->i_cpb_delay;
3011
3012     /* Init the rate control */
3013     /* FIXME: Include slice header bit cost. */
3014     x264_ratecontrol_start( h, h->fenc->i_qpplus1, overhead*8 );
3015     i_global_qp = x264_ratecontrol_qp( h );
3016
3017     pic_out->i_qpplus1 =
3018     h->fdec->i_qpplus1 = i_global_qp + 1;
3019
3020     if( h->param.rc.b_stat_read && h->sh.i_type != SLICE_TYPE_I )
3021     {
3022         x264_reference_build_list_optimal( h );
3023         x264_reference_check_reorder( h );
3024     }
3025
3026     if( h->i_ref[0] )
3027         h->fdec->i_poc_l0ref0 = h->fref[0][0]->i_poc;
3028
3029     /* ------------------------ Create slice header  ----------------------- */
3030     x264_slice_init( h, i_nal_type, i_global_qp );
3031
3032     /*------------------------- Weights -------------------------------------*/
3033     if( h->sh.i_type == SLICE_TYPE_B )
3034         x264_macroblock_bipred_init( h );
3035
3036     x264_weighted_pred_init( h );
3037
3038     if( i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE )
3039         h->i_frame_num++;
3040
3041     /* Write frame */
3042     h->i_threadslice_start = 0;
3043     h->i_threadslice_end = h->mb.i_mb_height;
3044     if( h->i_thread_frames > 1 )
3045     {
3046         x264_threadpool_run( h->threadpool, (void*)x264_slices_write, h );
3047         h->b_thread_active = 1;
3048     }
3049     else if( h->param.b_sliced_threads )
3050     {
3051         if( x264_threaded_slices_write( h ) )
3052             return -1;
3053     }
3054     else
3055         if( (intptr_t)x264_slices_write( h ) )
3056             return -1;
3057
3058     return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
3059 }
3060
3061 static int x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
3062                                    x264_nal_t **pp_nal, int *pi_nal,
3063                                    x264_picture_t *pic_out )
3064 {
3065     char psz_message[80];
3066
3067     if( h->b_thread_active )
3068     {
3069         h->b_thread_active = 0;
3070         if( (intptr_t)x264_threadpool_wait( h->threadpool, h ) )
3071             return -1;
3072     }
3073     if( !h->out.i_nal )
3074     {
3075         pic_out->i_type = X264_TYPE_AUTO;
3076         return 0;
3077     }
3078
3079     x264_emms();
3080     /* generate buffering period sei and insert it into place */
3081     if( h->i_thread_frames > 1 && h->fenc->b_keyframe && h->sps->vui.b_nal_hrd_parameters_present )
3082     {
3083         x264_hrd_fullness( h );
3084         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3085         x264_sei_buffering_period_write( h, &h->out.bs );
3086         if( x264_nal_end( h ) )
3087            return -1;
3088         /* buffering period sei must follow AUD, SPS and PPS and precede all other SEIs */
3089         int idx = 0;
3090         while( h->out.nal[idx].i_type == NAL_AUD ||
3091                h->out.nal[idx].i_type == NAL_SPS ||
3092                h->out.nal[idx].i_type == NAL_PPS )
3093             idx++;
3094         x264_nal_t nal_tmp = h->out.nal[h->out.i_nal-1];
3095         memmove( &h->out.nal[idx+1], &h->out.nal[idx], (h->out.i_nal-idx-1)*sizeof(x264_nal_t) );
3096         h->out.nal[idx] = nal_tmp;
3097     }
3098
3099     int frame_size = x264_encoder_encapsulate_nals( h, 0 );
3100     if( frame_size < 0 )
3101         return -1;
3102
3103     /* Set output picture properties */
3104     pic_out->i_type = h->fenc->i_type;
3105
3106     pic_out->b_keyframe = h->fenc->b_keyframe;
3107     pic_out->i_pic_struct = h->fenc->i_pic_struct;
3108
3109     pic_out->i_pts = h->fdec->i_pts;
3110     pic_out->i_dts = h->fdec->i_dts;
3111
3112     if( pic_out->i_pts < pic_out->i_dts )
3113         x264_log( h, X264_LOG_WARNING, "invalid DTS: PTS is less than DTS\n" );
3114
3115     pic_out->opaque = h->fenc->opaque;
3116
3117     pic_out->img.i_csp = h->fdec->i_csp;
3118 #if HIGH_BIT_DEPTH
3119     pic_out->img.i_csp |= X264_CSP_HIGH_DEPTH;
3120 #endif
3121     pic_out->img.i_plane = h->fdec->i_plane;
3122     for( int i = 0; i < pic_out->img.i_plane; i++ )
3123     {
3124         pic_out->img.i_stride[i] = h->fdec->i_stride[i] * sizeof(pixel);
3125         pic_out->img.plane[i] = (uint8_t*)h->fdec->plane[i];
3126     }
3127
3128     x264_frame_push_unused( thread_current, h->fenc );
3129
3130     /* ---------------------- Update encoder state ------------------------- */
3131
3132     /* update rc */
3133     int filler = 0;
3134     if( x264_ratecontrol_end( h, frame_size * 8, &filler ) < 0 )
3135         return -1;
3136
3137     pic_out->hrd_timing = h->fenc->hrd_timing;
3138
3139     while( filler > 0 )
3140     {
3141         int f, overhead;
3142         overhead = (FILLER_OVERHEAD - h->param.b_annexb);
3143         if( h->param.i_slice_max_size && filler > h->param.i_slice_max_size )
3144         {
3145             int next_size = filler - h->param.i_slice_max_size;
3146             int overflow = X264_MAX( overhead - next_size, 0 );
3147             f = h->param.i_slice_max_size - overhead - overflow;
3148         }
3149         else
3150             f = X264_MAX( 0, filler - overhead );
3151
3152         x264_nal_start( h, NAL_FILLER, NAL_PRIORITY_DISPOSABLE );
3153         x264_filler_write( h, &h->out.bs, f );
3154         if( x264_nal_end( h ) )
3155             return -1;
3156         int total_size = x264_encoder_encapsulate_nals( h, h->out.i_nal-1 );
3157         if( total_size < 0 )
3158             return -1;
3159         frame_size += total_size;
3160         filler -= total_size;
3161     }
3162
3163     /* End bitstream, set output  */
3164     *pi_nal = h->out.i_nal;
3165     *pp_nal = h->out.nal;
3166
3167     h->out.i_nal = 0;
3168
3169     x264_noise_reduction_update( h );
3170
3171     /* ---------------------- Compute/Print statistics --------------------- */
3172     x264_thread_sync_stat( h, h->thread[0] );
3173
3174     /* Slice stat */
3175     h->stat.i_frame_count[h->sh.i_type]++;
3176     h->stat.i_frame_size[h->sh.i_type] += frame_size;
3177     h->stat.f_frame_qp[h->sh.i_type] += h->fdec->f_qp_avg_aq;
3178
3179     for( int i = 0; i < X264_MBTYPE_MAX; i++ )
3180         h->stat.i_mb_count[h->sh.i_type][i] += h->stat.frame.i_mb_count[i];
3181     for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
3182         h->stat.i_mb_partition[h->sh.i_type][i] += h->stat.frame.i_mb_partition[i];
3183     for( int i = 0; i < 2; i++ )
3184         h->stat.i_mb_count_8x8dct[i] += h->stat.frame.i_mb_count_8x8dct[i];
3185     for( int i = 0; i < 6; i++ )
3186         h->stat.i_mb_cbp[i] += h->stat.frame.i_mb_cbp[i];
3187     for( int i = 0; i < 4; i++ )
3188         for( int j = 0; j < 13; j++ )
3189             h->stat.i_mb_pred_mode[i][j] += h->stat.frame.i_mb_pred_mode[i][j];
3190     if( h->sh.i_type != SLICE_TYPE_I )
3191         for( int i_list = 0; i_list < 2; i_list++ )
3192             for( int i = 0; i < X264_REF_MAX*2; i++ )
3193                 h->stat.i_mb_count_ref[h->sh.i_type][i_list][i] += h->stat.frame.i_mb_count_ref[i_list][i];
3194     for( int i = 0; i < 3; i++ )
3195         h->stat.i_mb_field[i] += h->stat.frame.i_mb_field[i];
3196     if( h->sh.i_type == SLICE_TYPE_P && h->param.analyse.i_weighted_pred >= X264_WEIGHTP_SIMPLE )
3197     {
3198         h->stat.i_wpred[0] += !!h->sh.weight[0][0].weightfn;
3199         h->stat.i_wpred[1] += !!h->sh.weight[0][1].weightfn || !!h->sh.weight[0][2].weightfn;
3200     }
3201     if( h->sh.i_type == SLICE_TYPE_B )
3202     {
3203         h->stat.i_direct_frames[ h->sh.b_direct_spatial_mv_pred ] ++;
3204         if( h->mb.b_direct_auto_write )
3205         {
3206             //FIXME somewhat arbitrary time constants
3207             if( h->stat.i_direct_score[0] + h->stat.i_direct_score[1] > h->mb.i_mb_count )
3208                 for( int i = 0; i < 2; i++ )
3209                     h->stat.i_direct_score[i] = h->stat.i_direct_score[i] * 9/10;
3210             for( int i = 0; i < 2; i++ )
3211                 h->stat.i_direct_score[i] += h->stat.frame.i_direct_score[i];
3212         }
3213     }
3214     else
3215         h->stat.i_consecutive_bframes[h->fenc->i_bframes]++;
3216
3217     psz_message[0] = '\0';
3218     double dur = h->fenc->f_duration;
3219     h->stat.f_frame_duration[h->sh.i_type] += dur;
3220     if( h->param.analyse.b_psnr )
3221     {
3222         int64_t ssd[3] =
3223         {
3224             h->stat.frame.i_ssd[0],
3225             h->stat.frame.i_ssd[1],
3226             h->stat.frame.i_ssd[2],
3227         };
3228         int luma_size = h->param.i_width * h->param.i_height;
3229         int chroma_size = CHROMA_SIZE( luma_size );
3230         double psnr_y = x264_psnr( ssd[0], luma_size );
3231         double psnr_u = x264_psnr( ssd[1], chroma_size );
3232         double psnr_v = x264_psnr( ssd[2], chroma_size );
3233
3234         h->stat.f_ssd_global[h->sh.i_type]   += dur * (ssd[0] + ssd[1] + ssd[2]);
3235         h->stat.f_psnr_average[h->sh.i_type] += dur * x264_psnr( ssd[0] + ssd[1] + ssd[2], luma_size + chroma_size*2 );
3236         h->stat.f_psnr_mean_y[h->sh.i_type]  += dur * psnr_y;
3237         h->stat.f_psnr_mean_u[h->sh.i_type]  += dur * psnr_u;
3238         h->stat.f_psnr_mean_v[h->sh.i_type]  += dur * psnr_v;
3239
3240         snprintf( psz_message, 80, " PSNR Y:%5.2f U:%5.2f V:%5.2f", psnr_y, psnr_u, psnr_v );
3241     }
3242
3243     if( h->param.analyse.b_ssim )
3244     {
3245         double ssim_y = h->stat.frame.f_ssim
3246                       / h->stat.frame.i_ssim_cnt;
3247         h->stat.f_ssim_mean_y[h->sh.i_type] += ssim_y * dur;
3248         snprintf( psz_message + strlen(psz_message), 80 - strlen(psz_message),
3249                   " SSIM Y:%.5f", ssim_y );
3250     }
3251     psz_message[79] = '\0';
3252
3253     x264_log( h, X264_LOG_DEBUG,
3254                   "frame=%4d QP=%.2f NAL=%d Slice:%c Poc:%-3d I:%-4d P:%-4d SKIP:%-4d size=%d bytes%s\n",
3255               h->i_frame,
3256               h->fdec->f_qp_avg_aq,
3257               h->i_nal_ref_idc,
3258               h->sh.i_type == SLICE_TYPE_I ? 'I' : (h->sh.i_type == SLICE_TYPE_P ? 'P' : 'B' ),
3259               h->fdec->i_poc,
3260               h->stat.frame.i_mb_count_i,
3261               h->stat.frame.i_mb_count_p,
3262               h->stat.frame.i_mb_count_skip,
3263               frame_size,
3264               psz_message );
3265
3266     // keep stats all in one place
3267     x264_thread_sync_stat( h->thread[0], h );
3268     // for the use of the next frame
3269     x264_thread_sync_stat( thread_current, h );
3270
3271 #ifdef DEBUG_MB_TYPE
3272 {
3273     static const char mb_chars[] = { 'i', 'i', 'I', 'C', 'P', '8', 'S',
3274         'D', '<', 'X', 'B', 'X', '>', 'B', 'B', 'B', 'B', '8', 'S' };
3275     for( int mb_xy = 0; mb_xy < h->mb.i_mb_width * h->mb.i_mb_height; mb_xy++ )
3276     {
3277         if( h->mb.type[mb_xy] < X264_MBTYPE_MAX && h->mb.type[mb_xy] >= 0 )
3278             fprintf( stderr, "%c ", mb_chars[ h->mb.type[mb_xy] ] );
3279         else
3280             fprintf( stderr, "? " );
3281
3282         if( (mb_xy+1) % h->mb.i_mb_width == 0 )
3283             fprintf( stderr, "\n" );
3284     }
3285 }
3286 #endif
3287
3288     /* Remove duplicates, must be done near the end as breaks h->fref0 array
3289      * by freeing some of its pointers. */
3290     for( int i = 0; i < h->i_ref[0]; i++ )
3291         if( h->fref[0][i] && h->fref[0][i]->b_duplicate )
3292         {
3293             x264_frame_push_blank_unused( h, h->fref[0][i] );
3294             h->fref[0][i] = 0;
3295         }
3296
3297     if( h->param.psz_dump_yuv )
3298         x264_frame_dump( h );
3299     x264_emms();
3300
3301     return frame_size;
3302 }
3303
3304 static void x264_print_intra( int64_t *i_mb_count, double i_count, int b_print_pcm, char *intra )
3305 {
3306     intra += sprintf( intra, "I16..4%s: %4.1f%% %4.1f%% %4.1f%%",
3307         b_print_pcm ? "..PCM" : "",
3308         i_mb_count[I_16x16]/ i_count,
3309         i_mb_count[I_8x8]  / i_count,
3310         i_mb_count[I_4x4]  / i_count );
3311     if( b_print_pcm )
3312         sprintf( intra, " %4.1f%%", i_mb_count[I_PCM]  / i_count );
3313 }
3314
3315 /****************************************************************************
3316  * x264_encoder_close:
3317  ****************************************************************************/
3318 void    x264_encoder_close  ( x264_t *h )
3319 {
3320     int64_t i_yuv_size = FRAME_SIZE( h->param.i_width * h->param.i_height );
3321     int64_t i_mb_count_size[2][7] = {{0}};
3322     char buf[200];
3323     int b_print_pcm = h->stat.i_mb_count[SLICE_TYPE_I][I_PCM]
3324                    || h->stat.i_mb_count[SLICE_TYPE_P][I_PCM]
3325                    || h->stat.i_mb_count[SLICE_TYPE_B][I_PCM];
3326
3327     x264_lookahead_delete( h );
3328
3329     if( h->param.i_threads > 1 )
3330         x264_threadpool_delete( h->threadpool );
3331     if( h->i_thread_frames > 1 )
3332     {
3333         for( int i = 0; i < h->i_thread_frames; i++ )
3334             if( h->thread[i]->b_thread_active )
3335             {
3336                 assert( h->thread[i]->fenc->i_reference_count == 1 );
3337                 x264_frame_delete( h->thread[i]->fenc );
3338             }
3339
3340         x264_t *thread_prev = h->thread[h->i_thread_phase];
3341         x264_thread_sync_ratecontrol( h, thread_prev, h );
3342         x264_thread_sync_ratecontrol( thread_prev, thread_prev, h );
3343         h->i_frame = thread_prev->i_frame + 1 - h->i_thread_frames;
3344     }
3345     h->i_frame++;
3346
3347     /* Slices used and PSNR */
3348     for( int i = 0; i < 3; i++ )
3349     {
3350         static const uint8_t slice_order[] = { SLICE_TYPE_I, SLICE_TYPE_P, SLICE_TYPE_B };
3351         int i_slice = slice_order[i];
3352
3353         if( h->stat.i_frame_count[i_slice] > 0 )
3354         {
3355             int i_count = h->stat.i_frame_count[i_slice];
3356             double dur =  h->stat.f_frame_duration[i_slice];
3357             if( h->param.analyse.b_psnr )
3358             {
3359                 x264_log( h, X264_LOG_INFO,
3360                           "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",
3361                           slice_type_to_char[i_slice],
3362                           i_count,
3363                           h->stat.f_frame_qp[i_slice] / i_count,
3364                           (double)h->stat.i_frame_size[i_slice] / i_count,
3365                           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,
3366                           h->stat.f_psnr_average[i_slice] / dur,
3367                           x264_psnr( h->stat.f_ssd_global[i_slice], dur * i_yuv_size ) );
3368             }
3369             else
3370             {
3371                 x264_log( h, X264_LOG_INFO,
3372                           "frame %c:%-5d Avg QP:%5.2f  size:%6.0f\n",
3373                           slice_type_to_char[i_slice],
3374                           i_count,
3375                           h->stat.f_frame_qp[i_slice] / i_count,
3376                           (double)h->stat.i_frame_size[i_slice] / i_count );
3377             }
3378         }
3379     }
3380     if( h->param.i_bframe && h->stat.i_frame_count[SLICE_TYPE_B] )
3381     {
3382         char *p = buf;
3383         int den = 0;
3384         // weight by number of frames (including the I/P-frames) that are in a sequence of N B-frames
3385         for( int i = 0; i <= h->param.i_bframe; i++ )
3386             den += (i+1) * h->stat.i_consecutive_bframes[i];
3387         for( int i = 0; i <= h->param.i_bframe; i++ )
3388             p += sprintf( p, " %4.1f%%", 100. * (i+1) * h->stat.i_consecutive_bframes[i] / den );
3389         x264_log( h, X264_LOG_INFO, "consecutive B-frames:%s\n", buf );
3390     }
3391
3392     for( int i_type = 0; i_type < 2; i_type++ )
3393         for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
3394         {
3395             if( i == D_DIRECT_8x8 ) continue; /* direct is counted as its own type */
3396             i_mb_count_size[i_type][x264_mb_partition_pixel_table[i]] += h->stat.i_mb_partition[i_type][i];
3397         }
3398
3399     /* MB types used */
3400     if( h->stat.i_frame_count[SLICE_TYPE_I] > 0 )
3401     {
3402         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_I];
3403         double i_count = h->stat.i_frame_count[SLICE_TYPE_I] * h->mb.i_mb_count / 100.0;
3404         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
3405         x264_log( h, X264_LOG_INFO, "mb I  %s\n", buf );
3406     }
3407     if( h->stat.i_frame_count[SLICE_TYPE_P] > 0 )
3408     {
3409         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_P];
3410         double i_count = h->stat.i_frame_count[SLICE_TYPE_P] * h->mb.i_mb_count / 100.0;
3411         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_P];
3412         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
3413         x264_log( h, X264_LOG_INFO,
3414                   "mb P  %s  P16..4: %4.1f%% %4.1f%% %4.1f%% %4.1f%% %4.1f%%    skip:%4.1f%%\n",
3415                   buf,
3416                   i_mb_size[PIXEL_16x16] / (i_count*4),
3417                   (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
3418                   i_mb_size[PIXEL_8x8] / (i_count*4),
3419                   (i_mb_size[PIXEL_8x4] + i_mb_size[PIXEL_4x8]) / (i_count*4),
3420                   i_mb_size[PIXEL_4x4] / (i_count*4),
3421                   i_mb_count[P_SKIP] / i_count );
3422     }
3423     if( h->stat.i_frame_count[SLICE_TYPE_B] > 0 )
3424     {
3425         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_B];
3426         double i_count = h->stat.i_frame_count[SLICE_TYPE_B] * h->mb.i_mb_count / 100.0;
3427         double i_mb_list_count;
3428         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_B];
3429         int64_t list_count[3] = {0}; /* 0 == L0, 1 == L1, 2 == BI */
3430         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
3431         for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
3432             for( int j = 0; j < 2; j++ )
3433             {
3434                 int l0 = x264_mb_type_list_table[i][0][j];
3435                 int l1 = x264_mb_type_list_table[i][1][j];
3436                 if( l0 || l1 )
3437                     list_count[l1+l0*l1] += h->stat.i_mb_count[SLICE_TYPE_B][i] * 2;
3438             }
3439         list_count[0] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L0_8x8];
3440         list_count[1] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L1_8x8];
3441         list_count[2] += h->stat.i_mb_partition[SLICE_TYPE_B][D_BI_8x8];
3442         i_mb_count[B_DIRECT] += (h->stat.i_mb_partition[SLICE_TYPE_B][D_DIRECT_8x8]+2)/4;
3443         i_mb_list_count = (list_count[0] + list_count[1] + list_count[2]) / 100.0;
3444         sprintf( buf + strlen(buf), "  B16..8: %4.1f%% %4.1f%% %4.1f%%  direct:%4.1f%%  skip:%4.1f%%",
3445                  i_mb_size[PIXEL_16x16] / (i_count*4),
3446                  (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
3447                  i_mb_size[PIXEL_8x8] / (i_count*4),
3448                  i_mb_count[B_DIRECT] / i_count,
3449                  i_mb_count[B_SKIP]   / i_count );
3450         if( i_mb_list_count != 0 )
3451             sprintf( buf + strlen(buf), "  L0:%4.1f%% L1:%4.1f%% BI:%4.1f%%",
3452                      list_count[0] / i_mb_list_count,
3453                      list_count[1] / i_mb_list_count,
3454                      list_count[2] / i_mb_list_count );
3455         x264_log( h, X264_LOG_INFO, "mb B  %s\n", buf );
3456     }
3457
3458     x264_ratecontrol_summary( h );
3459
3460     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 )
3461     {
3462 #define SUM3(p) (p[SLICE_TYPE_I] + p[SLICE_TYPE_P] + p[SLICE_TYPE_B])
3463 #define SUM3b(p,o) (p[SLICE_TYPE_I][o] + p[SLICE_TYPE_P][o] + p[SLICE_TYPE_B][o])
3464         int64_t i_i8x8 = SUM3b( h->stat.i_mb_count, I_8x8 );
3465         int64_t i_intra = i_i8x8 + SUM3b( h->stat.i_mb_count, I_4x4 )
3466                                  + SUM3b( h->stat.i_mb_count, I_16x16 );
3467         int64_t i_all_intra = i_intra + SUM3b( h->stat.i_mb_count, I_PCM);
3468         int64_t i_skip = SUM3b( h->stat.i_mb_count, P_SKIP )
3469                        + SUM3b( h->stat.i_mb_count, B_SKIP );
3470         const int i_count = h->stat.i_frame_count[SLICE_TYPE_I] +
3471                             h->stat.i_frame_count[SLICE_TYPE_P] +
3472                             h->stat.i_frame_count[SLICE_TYPE_B];
3473         int64_t i_mb_count = (int64_t)i_count * h->mb.i_mb_count;
3474         int64_t i_inter = i_mb_count - i_skip - i_intra;
3475         const double duration = h->stat.f_frame_duration[SLICE_TYPE_I] +
3476                                 h->stat.f_frame_duration[SLICE_TYPE_P] +
3477                                 h->stat.f_frame_duration[SLICE_TYPE_B];
3478         float f_bitrate = SUM3(h->stat.i_frame_size) / duration / 125;
3479
3480         if( PARAM_INTERLACED )
3481         {
3482             char *fieldstats = buf;
3483             fieldstats[0] = 0;
3484             if( i_inter )
3485                 fieldstats += sprintf( fieldstats, " inter:%.1f%%", h->stat.i_mb_field[1] * 100.0 / i_inter );
3486             if( i_skip )
3487                 fieldstats += sprintf( fieldstats, " skip:%.1f%%", h->stat.i_mb_field[2] * 100.0 / i_skip );
3488             x264_log( h, X264_LOG_INFO, "field mbs: intra: %.1f%%%s\n",
3489                       h->stat.i_mb_field[0] * 100.0 / i_intra, buf );
3490         }
3491
3492         if( h->pps->b_transform_8x8_mode )
3493         {
3494             buf[0] = 0;
3495             if( h->stat.i_mb_count_8x8dct[0] )
3496                 sprintf( buf, " inter:%.1f%%", 100. * h->stat.i_mb_count_8x8dct[1] / h->stat.i_mb_count_8x8dct[0] );
3497             x264_log( h, X264_LOG_INFO, "8x8 transform intra:%.1f%%%s\n", 100. * i_i8x8 / i_intra, buf );
3498         }
3499
3500         if( (h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO ||
3501             (h->stat.i_direct_frames[0] && h->stat.i_direct_frames[1]))
3502             && h->stat.i_frame_count[SLICE_TYPE_B] )
3503         {
3504             x264_log( h, X264_LOG_INFO, "direct mvs  spatial:%.1f%% temporal:%.1f%%\n",
3505                       h->stat.i_direct_frames[1] * 100. / h->stat.i_frame_count[SLICE_TYPE_B],
3506                       h->stat.i_direct_frames[0] * 100. / h->stat.i_frame_count[SLICE_TYPE_B] );
3507         }
3508
3509         buf[0] = 0;
3510         int csize = CHROMA444 ? 4 : 1;
3511         if( i_mb_count != i_all_intra )
3512             sprintf( buf, " inter: %.1f%% %.1f%% %.1f%%",
3513                      h->stat.i_mb_cbp[1] * 100.0 / ((i_mb_count - i_all_intra)*4),
3514                      h->stat.i_mb_cbp[3] * 100.0 / ((i_mb_count - i_all_intra)*csize),
3515                      h->stat.i_mb_cbp[5] * 100.0 / ((i_mb_count - i_all_intra)*csize) );
3516         x264_log( h, X264_LOG_INFO, "coded y,%s,%s intra: %.1f%% %.1f%% %.1f%%%s\n",
3517                   CHROMA444?"u":"uvDC", CHROMA444?"v":"uvAC",
3518                   h->stat.i_mb_cbp[0] * 100.0 / (i_all_intra*4),
3519                   h->stat.i_mb_cbp[2] * 100.0 / (i_all_intra*csize),
3520                   h->stat.i_mb_cbp[4] * 100.0 / (i_all_intra*csize), buf );
3521
3522         int64_t fixed_pred_modes[4][9] = {{0}};
3523         int64_t sum_pred_modes[4] = {0};
3524         for( int i = 0; i <= I_PRED_16x16_DC_128; i++ )
3525         {
3526             fixed_pred_modes[0][x264_mb_pred_mode16x16_fix[i]] += h->stat.i_mb_pred_mode[0][i];
3527             sum_pred_modes[0] += h->stat.i_mb_pred_mode[0][i];
3528         }
3529         if( sum_pred_modes[0] )
3530             x264_log( h, X264_LOG_INFO, "i16 v,h,dc,p: %2.0f%% %2.0f%% %2.0f%% %2.0f%%\n",
3531                       fixed_pred_modes[0][0] * 100.0 / sum_pred_modes[0],
3532                       fixed_pred_modes[0][1] * 100.0 / sum_pred_modes[0],
3533                       fixed_pred_modes[0][2] * 100.0 / sum_pred_modes[0],
3534                       fixed_pred_modes[0][3] * 100.0 / sum_pred_modes[0] );
3535         for( int i = 1; i <= 2; i++ )
3536         {
3537             for( int j = 0; j <= I_PRED_8x8_DC_128; j++ )
3538             {
3539                 fixed_pred_modes[i][x264_mb_pred_mode4x4_fix(j)] += h->stat.i_mb_pred_mode[i][j];
3540                 sum_pred_modes[i] += h->stat.i_mb_pred_mode[i][j];
3541             }
3542             if( sum_pred_modes[i] )
3543                 x264_log( h, X264_LOG_INFO, "i%d v,h,dc,ddl,ddr,vr,hd,vl,hu: %2.0f%% %2.0f%% %2.0f%% %2.0f%% %2.0f%% %2.0f%% %2.0f%% %2.0f%% %2.0f%%\n", (3-i)*4,
3544                           fixed_pred_modes[i][0] * 100.0 / sum_pred_modes[i],
3545                           fixed_pred_modes[i][1] * 100.0 / sum_pred_modes[i],
3546                           fixed_pred_modes[i][2] * 100.0 / sum_pred_modes[i],
3547                           fixed_pred_modes[i][3] * 100.0 / sum_pred_modes[i],
3548                           fixed_pred_modes[i][4] * 100.0 / sum_pred_modes[i],