Pass through user data
[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     return -10.0 * log10( 1 - ssim );
65 }
66
67 static void x264_frame_dump( x264_t *h )
68 {
69     FILE *f = fopen( h->param.psz_dump_yuv, "r+b" );
70     if( !f )
71         return;
72
73     /* Write the frame in display order */
74     int frame_size = FRAME_SIZE( h->param.i_height * h->param.i_width * sizeof(pixel) );
75     fseek( f, (uint64_t)h->fdec->i_frame * frame_size, SEEK_SET );
76     for( int p = 0; p < (CHROMA444 ? 3 : 1); p++ )
77         for( int y = 0; y < h->param.i_height; y++ )
78             fwrite( &h->fdec->plane[p][y*h->fdec->i_stride[p]], sizeof(pixel), h->param.i_width, f );
79     if( !CHROMA444 )
80     {
81         int cw = h->param.i_width>>1;
82         int ch = h->param.i_height>>CHROMA_V_SHIFT;
83         pixel *planeu = x264_malloc( (cw*ch*2+32)*sizeof(pixel) );
84         pixel *planev = planeu + cw*ch + 16;
85         h->mc.plane_copy_deinterleave( planeu, cw, planev, cw, h->fdec->plane[1], h->fdec->i_stride[1], cw, ch );
86         fwrite( planeu, 1, cw*ch*sizeof(pixel), f );
87         fwrite( planev, 1, cw*ch*sizeof(pixel), f );
88         x264_free( planeu );
89     }
90     fclose( f );
91 }
92
93 /* Fill "default" values */
94 static void x264_slice_header_init( x264_t *h, x264_slice_header_t *sh,
95                                     x264_sps_t *sps, x264_pps_t *pps,
96                                     int i_idr_pic_id, int i_frame, int i_qp )
97 {
98     x264_param_t *param = &h->param;
99
100     /* First we fill all fields */
101     sh->sps = sps;
102     sh->pps = pps;
103
104     sh->i_first_mb  = 0;
105     sh->i_last_mb   = h->mb.i_mb_count - 1;
106     sh->i_pps_id    = pps->i_id;
107
108     sh->i_frame_num = i_frame;
109
110     sh->b_mbaff = PARAM_INTERLACED;
111     sh->b_field_pic = 0;    /* no field support for now */
112     sh->b_bottom_field = 0; /* not yet used */
113
114     sh->i_idr_pic_id = i_idr_pic_id;
115
116     /* poc stuff, fixed later */
117     sh->i_poc = 0;
118     sh->i_delta_poc_bottom = 0;
119     sh->i_delta_poc[0] = 0;
120     sh->i_delta_poc[1] = 0;
121
122     sh->i_redundant_pic_cnt = 0;
123
124     h->mb.b_direct_auto_write = h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO
125                                 && h->param.i_bframe
126                                 && ( h->param.rc.b_stat_write || !h->param.rc.b_stat_read );
127
128     if( !h->mb.b_direct_auto_read && sh->i_type == SLICE_TYPE_B )
129     {
130         if( h->fref[1][0]->i_poc_l0ref0 == h->fref[0][0]->i_poc )
131         {
132             if( h->mb.b_direct_auto_write )
133                 sh->b_direct_spatial_mv_pred = ( h->stat.i_direct_score[1] > h->stat.i_direct_score[0] );
134             else
135                 sh->b_direct_spatial_mv_pred = ( param->analyse.i_direct_mv_pred == X264_DIRECT_PRED_SPATIAL );
136         }
137         else
138         {
139             h->mb.b_direct_auto_write = 0;
140             sh->b_direct_spatial_mv_pred = 1;
141         }
142     }
143     /* else b_direct_spatial_mv_pred was read from the 2pass statsfile */
144
145     sh->b_num_ref_idx_override = 0;
146     sh->i_num_ref_idx_l0_active = 1;
147     sh->i_num_ref_idx_l1_active = 1;
148
149     sh->b_ref_pic_list_reordering[0] = h->b_ref_reorder[0];
150     sh->b_ref_pic_list_reordering[1] = h->b_ref_reorder[1];
151
152     /* If the ref list isn't in the default order, construct reordering header */
153     for( int list = 0; list < 2; list++ )
154     {
155         if( sh->b_ref_pic_list_reordering[list] )
156         {
157             int pred_frame_num = i_frame;
158             for( int i = 0; i < h->i_ref[list]; i++ )
159             {
160                 int diff = h->fref[list][i]->i_frame_num - pred_frame_num;
161                 sh->ref_pic_list_order[list][i].idc = ( diff > 0 );
162                 sh->ref_pic_list_order[list][i].arg = (abs(diff) - 1) & ((1 << sps->i_log2_max_frame_num) - 1);
163                 pred_frame_num = h->fref[list][i]->i_frame_num;
164             }
165         }
166     }
167
168     sh->i_cabac_init_idc = param->i_cabac_init_idc;
169
170     sh->i_qp = SPEC_QP(i_qp);
171     sh->i_qp_delta = sh->i_qp - pps->i_pic_init_qp;
172     sh->b_sp_for_swidth = 0;
173     sh->i_qs_delta = 0;
174
175     int deblock_thresh = i_qp + 2 * X264_MIN(param->i_deblocking_filter_alphac0, param->i_deblocking_filter_beta);
176     /* If effective qp <= 15, deblocking would have no effect anyway */
177     if( param->b_deblocking_filter && (h->mb.b_variable_qp || 15 < deblock_thresh ) )
178         sh->i_disable_deblocking_filter_idc = param->b_sliced_threads ? 2 : 0;
179     else
180         sh->i_disable_deblocking_filter_idc = 1;
181     sh->i_alpha_c0_offset = param->i_deblocking_filter_alphac0 << 1;
182     sh->i_beta_offset = param->i_deblocking_filter_beta << 1;
183 }
184
185 static void x264_slice_header_write( bs_t *s, x264_slice_header_t *sh, int i_nal_ref_idc )
186 {
187     if( sh->b_mbaff )
188     {
189         int first_x = sh->i_first_mb % sh->sps->i_mb_width;
190         int first_y = sh->i_first_mb / sh->sps->i_mb_width;
191         assert( (first_y&1) == 0 );
192         bs_write_ue( s, (2*first_x + sh->sps->i_mb_width*(first_y&~1) + (first_y&1)) >> 1 );
193     }
194     else
195         bs_write_ue( s, sh->i_first_mb );
196
197     bs_write_ue( s, sh->i_type + 5 );   /* same type things */
198     bs_write_ue( s, sh->i_pps_id );
199     bs_write( s, sh->sps->i_log2_max_frame_num, sh->i_frame_num & ((1<<sh->sps->i_log2_max_frame_num)-1) );
200
201     if( !sh->sps->b_frame_mbs_only )
202     {
203         bs_write1( s, sh->b_field_pic );
204         if( sh->b_field_pic )
205             bs_write1( s, sh->b_bottom_field );
206     }
207
208     if( sh->i_idr_pic_id >= 0 ) /* NAL IDR */
209         bs_write_ue( s, sh->i_idr_pic_id );
210
211     if( sh->sps->i_poc_type == 0 )
212     {
213         bs_write( s, sh->sps->i_log2_max_poc_lsb, sh->i_poc & ((1<<sh->sps->i_log2_max_poc_lsb)-1) );
214         if( sh->pps->b_pic_order && !sh->b_field_pic )
215             bs_write_se( s, sh->i_delta_poc_bottom );
216     }
217
218     if( sh->pps->b_redundant_pic_cnt )
219         bs_write_ue( s, sh->i_redundant_pic_cnt );
220
221     if( sh->i_type == SLICE_TYPE_B )
222         bs_write1( s, sh->b_direct_spatial_mv_pred );
223
224     if( sh->i_type == SLICE_TYPE_P || sh->i_type == SLICE_TYPE_B )
225     {
226         bs_write1( s, sh->b_num_ref_idx_override );
227         if( sh->b_num_ref_idx_override )
228         {
229             bs_write_ue( s, sh->i_num_ref_idx_l0_active - 1 );
230             if( sh->i_type == SLICE_TYPE_B )
231                 bs_write_ue( s, sh->i_num_ref_idx_l1_active - 1 );
232         }
233     }
234
235     /* ref pic list reordering */
236     if( sh->i_type != SLICE_TYPE_I )
237     {
238         bs_write1( s, sh->b_ref_pic_list_reordering[0] );
239         if( sh->b_ref_pic_list_reordering[0] )
240         {
241             for( int i = 0; i < sh->i_num_ref_idx_l0_active; i++ )
242             {
243                 bs_write_ue( s, sh->ref_pic_list_order[0][i].idc );
244                 bs_write_ue( s, sh->ref_pic_list_order[0][i].arg );
245             }
246             bs_write_ue( s, 3 );
247         }
248     }
249     if( sh->i_type == SLICE_TYPE_B )
250     {
251         bs_write1( s, sh->b_ref_pic_list_reordering[1] );
252         if( sh->b_ref_pic_list_reordering[1] )
253         {
254             for( int i = 0; i < sh->i_num_ref_idx_l1_active; i++ )
255             {
256                 bs_write_ue( s, sh->ref_pic_list_order[1][i].idc );
257                 bs_write_ue( s, sh->ref_pic_list_order[1][i].arg );
258             }
259             bs_write_ue( s, 3 );
260         }
261     }
262
263     if( sh->pps->b_weighted_pred && sh->i_type == SLICE_TYPE_P )
264     {
265         /* pred_weight_table() */
266         bs_write_ue( s, sh->weight[0][0].i_denom );
267         bs_write_ue( s, sh->weight[0][1].i_denom );
268         for( int i = 0; i < sh->i_num_ref_idx_l0_active; i++ )
269         {
270             int luma_weight_l0_flag = !!sh->weight[i][0].weightfn;
271             int chroma_weight_l0_flag = !!sh->weight[i][1].weightfn || !!sh->weight[i][2].weightfn;
272             bs_write1( s, luma_weight_l0_flag );
273             if( luma_weight_l0_flag )
274             {
275                 bs_write_se( s, sh->weight[i][0].i_scale );
276                 bs_write_se( s, sh->weight[i][0].i_offset );
277             }
278             bs_write1( s, chroma_weight_l0_flag );
279             if( chroma_weight_l0_flag )
280             {
281                 for( int j = 1; j < 3; j++ )
282                 {
283                     bs_write_se( s, sh->weight[i][j].i_scale );
284                     bs_write_se( s, sh->weight[i][j].i_offset );
285                 }
286             }
287         }
288     }
289     else if( sh->pps->b_weighted_bipred == 1 && sh->i_type == SLICE_TYPE_B )
290     {
291       /* TODO */
292     }
293
294     if( i_nal_ref_idc != 0 )
295     {
296         if( sh->i_idr_pic_id >= 0 )
297         {
298             bs_write1( s, 0 );  /* no output of prior pics flag */
299             bs_write1( s, 0 );  /* long term reference flag */
300         }
301         else
302         {
303             bs_write1( s, sh->i_mmco_command_count > 0 ); /* adaptive_ref_pic_marking_mode_flag */
304             if( sh->i_mmco_command_count > 0 )
305             {
306                 for( int i = 0; i < sh->i_mmco_command_count; i++ )
307                 {
308                     bs_write_ue( s, 1 ); /* mark short term ref as unused */
309                     bs_write_ue( s, sh->mmco[i].i_difference_of_pic_nums - 1 );
310                 }
311                 bs_write_ue( s, 0 ); /* end command list */
312             }
313         }
314     }
315
316     if( sh->pps->b_cabac && sh->i_type != SLICE_TYPE_I )
317         bs_write_ue( s, sh->i_cabac_init_idc );
318
319     bs_write_se( s, sh->i_qp_delta );      /* slice qp delta */
320
321     if( sh->pps->b_deblocking_filter_control )
322     {
323         bs_write_ue( s, sh->i_disable_deblocking_filter_idc );
324         if( sh->i_disable_deblocking_filter_idc != 1 )
325         {
326             bs_write_se( s, sh->i_alpha_c0_offset >> 1 );
327             bs_write_se( s, sh->i_beta_offset >> 1 );
328         }
329     }
330 }
331
332 /* If we are within a reasonable distance of the end of the memory allocated for the bitstream, */
333 /* reallocate, adding an arbitrary amount of space (100 kilobytes). */
334 static int x264_bitstream_check_buffer( x264_t *h )
335 {
336     uint8_t *bs_bak = h->out.p_bitstream;
337     int max_mb_size = 2500 << SLICE_MBAFF;
338     if( (h->param.b_cabac && (h->cabac.p_end - h->cabac.p < max_mb_size)) ||
339         (h->out.bs.p_end - h->out.bs.p < max_mb_size) )
340     {
341         h->out.i_bitstream += 100000;
342         CHECKED_MALLOC( h->out.p_bitstream, h->out.i_bitstream );
343         h->mc.memcpy_aligned( h->out.p_bitstream, bs_bak, (h->out.i_bitstream - 100000) & ~15 );
344         intptr_t delta = h->out.p_bitstream - bs_bak;
345
346         h->out.bs.p_start += delta;
347         h->out.bs.p += delta;
348         h->out.bs.p_end = h->out.p_bitstream + h->out.i_bitstream;
349
350         h->cabac.p_start += delta;
351         h->cabac.p += delta;
352         h->cabac.p_end = h->out.p_bitstream + h->out.i_bitstream;
353
354         for( int i = 0; i <= h->out.i_nal; i++ )
355             h->out.nal[i].p_payload += delta;
356         x264_free( bs_bak );
357     }
358     return 0;
359 fail:
360     x264_free( bs_bak );
361     return -1;
362 }
363
364 #if HAVE_THREAD
365 static void x264_encoder_thread_init( x264_t *h )
366 {
367     if( h->param.i_sync_lookahead )
368         x264_lower_thread_priority( 10 );
369
370 #if HAVE_MMX
371     /* Misalign mask has to be set separately for each thread. */
372     if( h->param.cpu&X264_CPU_SSE_MISALIGN )
373         x264_cpu_mask_misalign_sse();
374 #endif
375 }
376 #endif
377
378 /****************************************************************************
379  *
380  ****************************************************************************
381  ****************************** External API*********************************
382  ****************************************************************************
383  *
384  ****************************************************************************/
385
386 static int x264_validate_parameters( x264_t *h, int b_open )
387 {
388 #if HAVE_MMX
389 #ifdef __SSE__
390     if( b_open && !(x264_cpu_detect() & X264_CPU_SSE) )
391     {
392         x264_log( h, X264_LOG_ERROR, "your cpu does not support SSE1, but x264 was compiled with asm support\n");
393 #else
394     if( b_open && !(x264_cpu_detect() & X264_CPU_MMX2) )
395     {
396         x264_log( h, X264_LOG_ERROR, "your cpu does not support MMXEXT, but x264 was compiled with asm support\n");
397 #endif
398         x264_log( h, X264_LOG_ERROR, "to run x264, recompile without asm support (configure --disable-asm)\n");
399         return -1;
400     }
401 #endif
402
403 #if HAVE_INTERLACED
404     h->param.b_interlaced = !!PARAM_INTERLACED;
405 #else
406     if( h->param.b_interlaced )
407     {
408         x264_log( h, X264_LOG_ERROR, "not compiled with interlaced support\n" );
409         return -1;
410     }
411 #endif
412
413     if( h->param.i_width <= 0 || h->param.i_height <= 0 )
414     {
415         x264_log( h, X264_LOG_ERROR, "invalid width x height (%dx%d)\n",
416                   h->param.i_width, h->param.i_height );
417         return -1;
418     }
419
420     int i_csp = h->param.i_csp & X264_CSP_MASK;
421 #if X264_CHROMA_FORMAT
422     if( CHROMA_FORMAT != CHROMA_420 && i_csp >= X264_CSP_I420 && i_csp <= X264_CSP_NV12 )
423     {
424         x264_log( h, X264_LOG_ERROR, "not compiled with 4:2:0 support\n" );
425         return -1;
426     }
427     else if( CHROMA_FORMAT != CHROMA_422 && i_csp >= X264_CSP_I422 && i_csp <= X264_CSP_NV16 )
428     {
429         x264_log( h, X264_LOG_ERROR, "not compiled with 4:2:2 support\n" );
430         return -1;
431     }
432     else if( CHROMA_FORMAT != CHROMA_444 && i_csp >= X264_CSP_I444 && i_csp <= X264_CSP_RGB )
433     {
434         x264_log( h, X264_LOG_ERROR, "not compiled with 4:4:4 support\n" );
435         return -1;
436     }
437 #endif
438     if( i_csp <= X264_CSP_NONE || i_csp >= X264_CSP_MAX )
439     {
440         x264_log( h, X264_LOG_ERROR, "invalid CSP (only I420/YV12/NV12/I422/YV16/NV16/I444/YV24/BGR/BGRA/RGB supported)\n" );
441         return -1;
442     }
443
444     if( i_csp < X264_CSP_I444 && h->param.i_width % 2 )
445     {
446         x264_log( h, X264_LOG_ERROR, "width not divisible by 2 (%dx%d)\n",
447                   h->param.i_width, h->param.i_height );
448         return -1;
449     }
450
451     if( i_csp < X264_CSP_I422 && PARAM_INTERLACED && h->param.i_height % 4 )
452     {
453         x264_log( h, X264_LOG_ERROR, "height not divisible by 4 (%dx%d)\n",
454                   h->param.i_width, h->param.i_height );
455         return -1;
456     }
457
458     if( (i_csp < X264_CSP_I422 || PARAM_INTERLACED) && h->param.i_height % 2 )
459     {
460         x264_log( h, X264_LOG_ERROR, "height not divisible by 2 (%dx%d)\n",
461                   h->param.i_width, h->param.i_height );
462         return -1;
463     }
464
465     if( (h->param.crop_rect.i_left + h->param.crop_rect.i_right ) >= h->param.i_width ||
466         (h->param.crop_rect.i_top  + h->param.crop_rect.i_bottom) >= h->param.i_height )
467     {
468         x264_log( h, X264_LOG_ERROR, "invalid crop-rect %u,%u,%u,%u\n", h->param.crop_rect.i_left,
469                   h->param.crop_rect.i_top, h->param.crop_rect.i_right,  h->param.crop_rect.i_bottom );
470         return -1;
471     }
472
473     if( h->param.i_threads == X264_THREADS_AUTO )
474         h->param.i_threads = x264_cpu_num_processors() * (h->param.b_sliced_threads?2:3)/2;
475     h->param.i_threads = x264_clip3( h->param.i_threads, 1, X264_THREAD_MAX );
476     if( h->param.i_threads > 1 )
477     {
478 #if !HAVE_THREAD
479         x264_log( h, X264_LOG_WARNING, "not compiled with thread support!\n");
480         h->param.i_threads = 1;
481 #endif
482         /* Avoid absurdly small thread slices as they can reduce performance
483          * and VBV compliance.  Capped at an arbitrary 4 rows per thread. */
484         if( h->param.b_sliced_threads )
485         {
486             int max_threads = (h->param.i_height+15)/16 / 4;
487             h->param.i_threads = X264_MIN( h->param.i_threads, max_threads );
488         }
489     }
490     else
491         h->param.b_sliced_threads = 0;
492     h->i_thread_frames = h->param.b_sliced_threads ? 1 : h->param.i_threads;
493     if( h->i_thread_frames > 1 )
494         h->param.nalu_process = NULL;
495
496     h->param.i_keyint_max = x264_clip3( h->param.i_keyint_max, 1, X264_KEYINT_MAX_INFINITE );
497     if( h->param.i_keyint_max == 1 )
498     {
499         h->param.b_intra_refresh = 0;
500         h->param.analyse.i_weighted_pred = 0;
501     }
502
503     h->param.i_frame_packing = x264_clip3( h->param.i_frame_packing, -1, 5 );
504
505     /* Detect default ffmpeg settings and terminate with an error. */
506     if( b_open )
507     {
508         int score = 0;
509         score += h->param.analyse.i_me_range == 0;
510         score += h->param.rc.i_qp_step == 3;
511         score += h->param.i_keyint_max == 12;
512         score += h->param.rc.i_qp_min == 2;
513         score += h->param.rc.i_qp_max == 31;
514         score += h->param.rc.f_qcompress == 0.5;
515         score += fabs(h->param.rc.f_ip_factor - 1.25) < 0.01;
516         score += fabs(h->param.rc.f_pb_factor - 1.25) < 0.01;
517         score += h->param.analyse.inter == 0 && h->param.analyse.i_subpel_refine == 8;
518         if( score >= 5 )
519         {
520             x264_log( h, X264_LOG_ERROR, "broken ffmpeg default settings detected\n" );
521             x264_log( h, X264_LOG_ERROR, "use an encoding preset (e.g. -vpre medium)\n" );
522             x264_log( h, X264_LOG_ERROR, "preset usage: -vpre <speed> -vpre <profile>\n" );
523             x264_log( h, X264_LOG_ERROR, "speed presets are listed in x264 --help\n" );
524             x264_log( h, X264_LOG_ERROR, "profile is optional; x264 defaults to high\n" );
525             return -1;
526         }
527     }
528
529     if( h->param.rc.i_rc_method < 0 || h->param.rc.i_rc_method > 2 )
530     {
531         x264_log( h, X264_LOG_ERROR, "no ratecontrol method specified\n" );
532         return -1;
533     }
534     h->param.rc.f_rf_constant = x264_clip3f( h->param.rc.f_rf_constant, -QP_BD_OFFSET, 51 );
535     h->param.rc.f_rf_constant_max = x264_clip3f( h->param.rc.f_rf_constant_max, -QP_BD_OFFSET, 51 );
536     h->param.rc.i_qp_constant = x264_clip3( h->param.rc.i_qp_constant, 0, QP_MAX );
537     h->param.analyse.i_subpel_refine = x264_clip3( h->param.analyse.i_subpel_refine, 0, 11 );
538     h->param.rc.f_ip_factor = X264_MAX( h->param.rc.f_ip_factor, 0.01f );
539     h->param.rc.f_pb_factor = X264_MAX( h->param.rc.f_pb_factor, 0.01f );
540     if( h->param.rc.i_rc_method == X264_RC_CRF )
541     {
542         h->param.rc.i_qp_constant = h->param.rc.f_rf_constant + QP_BD_OFFSET;
543         h->param.rc.i_bitrate = 0;
544     }
545     if( (h->param.rc.i_rc_method == X264_RC_CQP || h->param.rc.i_rc_method == X264_RC_CRF)
546         && h->param.rc.i_qp_constant == 0 )
547     {
548         h->mb.b_lossless = 1;
549         h->param.i_cqm_preset = X264_CQM_FLAT;
550         h->param.psz_cqm_file = NULL;
551         h->param.rc.i_rc_method = X264_RC_CQP;
552         h->param.rc.f_ip_factor = 1;
553         h->param.rc.f_pb_factor = 1;
554         h->param.analyse.b_psnr = 0;
555         h->param.analyse.b_ssim = 0;
556         h->param.analyse.i_chroma_qp_offset = 0;
557         h->param.analyse.i_trellis = 0;
558         h->param.analyse.b_fast_pskip = 0;
559         h->param.analyse.i_noise_reduction = 0;
560         h->param.analyse.b_psy = 0;
561         h->param.i_bframe = 0;
562         /* 8x8dct is not useful without RD in CAVLC lossless */
563         if( !h->param.b_cabac && h->param.analyse.i_subpel_refine < 6 )
564             h->param.analyse.b_transform_8x8 = 0;
565     }
566     if( h->param.rc.i_rc_method == X264_RC_CQP )
567     {
568         float qp_p = h->param.rc.i_qp_constant;
569         float qp_i = qp_p - 6*log2f( h->param.rc.f_ip_factor );
570         float qp_b = qp_p + 6*log2f( h->param.rc.f_pb_factor );
571         h->param.rc.i_qp_min = x264_clip3( (int)(X264_MIN3( qp_p, qp_i, qp_b )), 0, QP_MAX );
572         h->param.rc.i_qp_max = x264_clip3( (int)(X264_MAX3( qp_p, qp_i, qp_b ) + .999), 0, QP_MAX );
573         h->param.rc.i_aq_mode = 0;
574         h->param.rc.b_mb_tree = 0;
575         h->param.rc.i_bitrate = 0;
576     }
577     h->param.rc.i_qp_max = x264_clip3( h->param.rc.i_qp_max, 0, QP_MAX );
578     h->param.rc.i_qp_min = x264_clip3( h->param.rc.i_qp_min, 0, h->param.rc.i_qp_max );
579     h->param.rc.i_qp_step = x264_clip3( h->param.rc.i_qp_step, 0, QP_MAX );
580     h->param.rc.i_bitrate = x264_clip3( h->param.rc.i_bitrate, 0, 2000000 );
581     h->param.rc.i_vbv_buffer_size = x264_clip3( h->param.rc.i_vbv_buffer_size, 0, 2000000 );
582     h->param.rc.i_vbv_max_bitrate = x264_clip3( h->param.rc.i_vbv_max_bitrate, 0, 2000000 );
583     h->param.rc.f_vbv_buffer_init = x264_clip3f( h->param.rc.f_vbv_buffer_init, 0, 2000000 );
584     if( h->param.rc.i_vbv_buffer_size )
585     {
586         if( h->param.rc.i_rc_method == X264_RC_CQP )
587         {
588             x264_log( h, X264_LOG_WARNING, "VBV is incompatible with constant QP, ignored.\n" );
589             h->param.rc.i_vbv_max_bitrate = 0;
590             h->param.rc.i_vbv_buffer_size = 0;
591         }
592         else if( h->param.rc.i_vbv_max_bitrate == 0 )
593         {
594             if( h->param.rc.i_rc_method == X264_RC_ABR )
595             {
596                 x264_log( h, X264_LOG_WARNING, "VBV maxrate unspecified, assuming CBR\n" );
597                 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate;
598             }
599             else
600             {
601                 x264_log( h, X264_LOG_WARNING, "VBV bufsize set but maxrate unspecified, ignored\n" );
602                 h->param.rc.i_vbv_buffer_size = 0;
603             }
604         }
605         else if( h->param.rc.i_vbv_max_bitrate < h->param.rc.i_bitrate &&
606                  h->param.rc.i_rc_method == X264_RC_ABR )
607         {
608             x264_log( h, X264_LOG_WARNING, "max bitrate less than average bitrate, assuming CBR\n" );
609             h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate;
610         }
611     }
612     else if( h->param.rc.i_vbv_max_bitrate )
613     {
614         x264_log( h, X264_LOG_WARNING, "VBV maxrate specified, but no bufsize, ignored\n" );
615         h->param.rc.i_vbv_max_bitrate = 0;
616     }
617
618     h->param.i_slice_max_size = X264_MAX( h->param.i_slice_max_size, 0 );
619     h->param.i_slice_max_mbs = X264_MAX( h->param.i_slice_max_mbs, 0 );
620
621     int max_slices = (h->param.i_height+((16<<PARAM_INTERLACED)-1))/(16<<PARAM_INTERLACED);
622     if( h->param.b_sliced_threads )
623         h->param.i_slice_count = x264_clip3( h->param.i_threads, 0, max_slices );
624     else
625     {
626         h->param.i_slice_count = x264_clip3( h->param.i_slice_count, 0, max_slices );
627         if( h->param.i_slice_max_mbs || h->param.i_slice_max_size )
628             h->param.i_slice_count = 0;
629     }
630
631     if( h->param.b_bluray_compat )
632     {
633         h->param.i_bframe_pyramid = X264_MIN( X264_B_PYRAMID_STRICT, h->param.i_bframe_pyramid );
634         h->param.i_bframe = X264_MIN( h->param.i_bframe, 3 );
635         h->param.b_aud = 1;
636         h->param.i_nal_hrd = X264_MAX( h->param.i_nal_hrd, X264_NAL_HRD_VBR );
637         h->param.i_slice_max_size = 0;
638         h->param.i_slice_max_mbs = 0;
639         h->param.b_intra_refresh = 0;
640         h->param.i_frame_reference = X264_MIN( h->param.i_frame_reference, 6 );
641         h->param.i_dpb_size = X264_MIN( h->param.i_dpb_size, 6 );
642         /* Due to the proliferation of broken players that don't handle dupes properly. */
643         h->param.analyse.i_weighted_pred = X264_MIN( h->param.analyse.i_weighted_pred, X264_WEIGHTP_SIMPLE );
644         if( h->param.b_fake_interlaced )
645             h->param.b_pic_struct = 1;
646     }
647
648     h->param.i_frame_reference = x264_clip3( h->param.i_frame_reference, 1, X264_REF_MAX );
649     h->param.i_dpb_size = x264_clip3( h->param.i_dpb_size, 1, X264_REF_MAX );
650     if( h->param.i_scenecut_threshold < 0 )
651         h->param.i_scenecut_threshold = 0;
652     h->param.analyse.i_direct_mv_pred = x264_clip3( h->param.analyse.i_direct_mv_pred, X264_DIRECT_PRED_NONE, X264_DIRECT_PRED_AUTO );
653     if( !h->param.analyse.i_subpel_refine && h->param.analyse.i_direct_mv_pred > X264_DIRECT_PRED_SPATIAL )
654     {
655         x264_log( h, X264_LOG_WARNING, "subme=0 + direct=temporal is not supported\n" );
656         h->param.analyse.i_direct_mv_pred = X264_DIRECT_PRED_SPATIAL;
657     }
658     h->param.i_bframe = x264_clip3( h->param.i_bframe, 0, X264_MIN( X264_BFRAME_MAX, h->param.i_keyint_max-1 ) );
659     h->param.i_bframe_bias = x264_clip3( h->param.i_bframe_bias, -90, 100 );
660     if( h->param.i_bframe <= 1 )
661         h->param.i_bframe_pyramid = X264_B_PYRAMID_NONE;
662     h->param.i_bframe_pyramid = x264_clip3( h->param.i_bframe_pyramid, X264_B_PYRAMID_NONE, X264_B_PYRAMID_NORMAL );
663     h->param.i_bframe_adaptive = x264_clip3( h->param.i_bframe_adaptive, X264_B_ADAPT_NONE, X264_B_ADAPT_TRELLIS );
664     if( !h->param.i_bframe )
665     {
666         h->param.i_bframe_adaptive = X264_B_ADAPT_NONE;
667         h->param.analyse.i_direct_mv_pred = 0;
668         h->param.analyse.b_weighted_bipred = 0;
669         h->param.b_open_gop = 0;
670     }
671     if( h->param.b_intra_refresh && h->param.i_bframe_pyramid == X264_B_PYRAMID_NORMAL )
672     {
673         x264_log( h, X264_LOG_WARNING, "b-pyramid normal + intra-refresh is not supported\n" );
674         h->param.i_bframe_pyramid = X264_B_PYRAMID_STRICT;
675     }
676     if( h->param.b_intra_refresh && (h->param.i_frame_reference > 1 || h->param.i_dpb_size > 1) )
677     {
678         x264_log( h, X264_LOG_WARNING, "ref > 1 + intra-refresh is not supported\n" );
679         h->param.i_frame_reference = 1;
680         h->param.i_dpb_size = 1;
681     }
682     if( h->param.b_intra_refresh && h->param.b_open_gop )
683     {
684         x264_log( h, X264_LOG_WARNING, "intra-refresh is not compatible with open-gop\n" );
685         h->param.b_open_gop = 0;
686     }
687     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;
688     if( h->param.i_keyint_min == X264_KEYINT_MIN_AUTO )
689         h->param.i_keyint_min = X264_MIN( h->param.i_keyint_max / 10, fps );
690     h->param.i_keyint_min = x264_clip3( h->param.i_keyint_min, 1, h->param.i_keyint_max/2+1 );
691     h->param.rc.i_lookahead = x264_clip3( h->param.rc.i_lookahead, 0, X264_LOOKAHEAD_MAX );
692     {
693         int maxrate = X264_MAX( h->param.rc.i_vbv_max_bitrate, h->param.rc.i_bitrate );
694         float bufsize = maxrate ? (float)h->param.rc.i_vbv_buffer_size / maxrate : 0;
695         h->param.rc.i_lookahead = X264_MIN( h->param.rc.i_lookahead, X264_MAX( h->param.i_keyint_max, bufsize*fps ) );
696     }
697
698     if( !h->param.i_timebase_num || !h->param.i_timebase_den || !(h->param.b_vfr_input || h->param.b_pulldown) )
699     {
700         h->param.i_timebase_num = h->param.i_fps_den;
701         h->param.i_timebase_den = h->param.i_fps_num;
702     }
703
704     h->param.rc.f_qcompress = x264_clip3f( h->param.rc.f_qcompress, 0.0, 1.0 );
705     if( h->param.i_keyint_max == 1 || h->param.rc.f_qcompress == 1 )
706         h->param.rc.b_mb_tree = 0;
707     if( (!h->param.b_intra_refresh && h->param.i_keyint_max != X264_KEYINT_MAX_INFINITE) &&
708         !h->param.rc.i_lookahead && h->param.rc.b_mb_tree )
709     {
710         x264_log( h, X264_LOG_WARNING, "lookaheadless mb-tree requires intra refresh or infinite keyint\n" );
711         h->param.rc.b_mb_tree = 0;
712     }
713     if( h->param.rc.b_stat_read )
714         h->param.rc.i_lookahead = 0;
715 #if HAVE_THREAD
716     if( h->param.i_sync_lookahead < 0 )
717         h->param.i_sync_lookahead = h->param.i_bframe + 1;
718     h->param.i_sync_lookahead = X264_MIN( h->param.i_sync_lookahead, X264_LOOKAHEAD_MAX );
719     if( h->param.rc.b_stat_read || h->i_thread_frames == 1 )
720         h->param.i_sync_lookahead = 0;
721 #else
722     h->param.i_sync_lookahead = 0;
723 #endif
724
725     h->param.i_deblocking_filter_alphac0 = x264_clip3( h->param.i_deblocking_filter_alphac0, -6, 6 );
726     h->param.i_deblocking_filter_beta    = x264_clip3( h->param.i_deblocking_filter_beta, -6, 6 );
727     h->param.analyse.i_luma_deadzone[0] = x264_clip3( h->param.analyse.i_luma_deadzone[0], 0, 32 );
728     h->param.analyse.i_luma_deadzone[1] = x264_clip3( h->param.analyse.i_luma_deadzone[1], 0, 32 );
729
730     h->param.i_cabac_init_idc = x264_clip3( h->param.i_cabac_init_idc, 0, 2 );
731
732     if( h->param.i_cqm_preset < X264_CQM_FLAT || h->param.i_cqm_preset > X264_CQM_CUSTOM )
733         h->param.i_cqm_preset = X264_CQM_FLAT;
734
735     if( h->param.analyse.i_me_method < X264_ME_DIA ||
736         h->param.analyse.i_me_method > X264_ME_TESA )
737         h->param.analyse.i_me_method = X264_ME_HEX;
738     h->param.analyse.i_me_range = x264_clip3( h->param.analyse.i_me_range, 4, 1024 );
739     if( h->param.analyse.i_me_range > 16 && h->param.analyse.i_me_method <= X264_ME_HEX )
740         h->param.analyse.i_me_range = 16;
741     if( h->param.analyse.i_me_method == X264_ME_TESA &&
742         (h->mb.b_lossless || h->param.analyse.i_subpel_refine <= 1) )
743         h->param.analyse.i_me_method = X264_ME_ESA;
744     h->param.analyse.b_mixed_references = h->param.analyse.b_mixed_references && h->param.i_frame_reference > 1;
745     h->param.analyse.inter &= X264_ANALYSE_PSUB16x16|X264_ANALYSE_PSUB8x8|X264_ANALYSE_BSUB16x16|
746                               X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
747     h->param.analyse.intra &= X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
748     if( !(h->param.analyse.inter & X264_ANALYSE_PSUB16x16) )
749         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
750     if( !h->param.analyse.b_transform_8x8 )
751     {
752         h->param.analyse.inter &= ~X264_ANALYSE_I8x8;
753         h->param.analyse.intra &= ~X264_ANALYSE_I8x8;
754     }
755     h->param.analyse.i_trellis = x264_clip3( h->param.analyse.i_trellis, 0, 2 );
756     h->param.rc.i_aq_mode = x264_clip3( h->param.rc.i_aq_mode, 0, 2 );
757     h->param.rc.f_aq_strength = x264_clip3f( h->param.rc.f_aq_strength, 0, 3 );
758     if( h->param.rc.f_aq_strength == 0 )
759         h->param.rc.i_aq_mode = 0;
760
761     if( h->param.i_log_level < X264_LOG_INFO )
762     {
763         h->param.analyse.b_psnr = 0;
764         h->param.analyse.b_ssim = 0;
765     }
766     /* Warn users trying to measure PSNR/SSIM with psy opts on. */
767     if( b_open && (h->param.analyse.b_psnr || h->param.analyse.b_ssim) )
768     {
769         char *s = NULL;
770
771         if( h->param.analyse.b_psy )
772         {
773             s = h->param.analyse.b_psnr ? "psnr" : "ssim";
774             x264_log( h, X264_LOG_WARNING, "--%s used with psy on: results will be invalid!\n", s );
775         }
776         else if( !h->param.rc.i_aq_mode && h->param.analyse.b_ssim )
777         {
778             x264_log( h, X264_LOG_WARNING, "--ssim used with AQ off: results will be invalid!\n" );
779             s = "ssim";
780         }
781         else if(  h->param.rc.i_aq_mode && h->param.analyse.b_psnr )
782         {
783             x264_log( h, X264_LOG_WARNING, "--psnr used with AQ on: results will be invalid!\n" );
784             s = "psnr";
785         }
786         if( s )
787             x264_log( h, X264_LOG_WARNING, "--tune %s should be used if attempting to benchmark %s!\n", s, s );
788     }
789
790     if( !h->param.analyse.b_psy )
791     {
792         h->param.analyse.f_psy_rd = 0;
793         h->param.analyse.f_psy_trellis = 0;
794     }
795     h->param.analyse.f_psy_rd = x264_clip3f( h->param.analyse.f_psy_rd, 0, 10 );
796     h->param.analyse.f_psy_trellis = x264_clip3f( h->param.analyse.f_psy_trellis, 0, 10 );
797     h->mb.i_psy_rd = h->param.analyse.i_subpel_refine >= 6 ? FIX8( h->param.analyse.f_psy_rd ) : 0;
798     h->mb.i_psy_trellis = h->param.analyse.i_trellis ? FIX8( h->param.analyse.f_psy_trellis / 4 ) : 0;
799     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -32, 32);
800     /* In 4:4:4 mode, chroma gets twice as much resolution, so we can halve its quality. */
801     if( b_open && i_csp >= X264_CSP_I444 && i_csp < X264_CSP_BGR && h->param.analyse.b_psy )
802         h->param.analyse.i_chroma_qp_offset += 6;
803     /* Psy RDO increases overall quantizers to improve the quality of luma--this indirectly hurts chroma quality */
804     /* so we lower the chroma QP offset to compensate */
805     if( b_open && h->mb.i_psy_rd )
806         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_rd < 0.25 ? 1 : 2;
807     /* Psy trellis has a similar effect. */
808     if( b_open && h->mb.i_psy_trellis )
809         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_trellis < 0.25 ? 1 : 2;
810     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -12, 12);
811     /* MB-tree requires AQ to be on, even if the strength is zero. */
812     if( !h->param.rc.i_aq_mode && h->param.rc.b_mb_tree )
813     {
814         h->param.rc.i_aq_mode = 1;
815         h->param.rc.f_aq_strength = 0;
816     }
817     h->param.analyse.i_noise_reduction = x264_clip3( h->param.analyse.i_noise_reduction, 0, 1<<16 );
818     if( h->param.analyse.i_subpel_refine >= 10 && (h->param.analyse.i_trellis != 2 || !h->param.rc.i_aq_mode) )
819         h->param.analyse.i_subpel_refine = 9;
820
821     {
822         const x264_level_t *l = x264_levels;
823         if( h->param.i_level_idc < 0 )
824         {
825             int maxrate_bak = h->param.rc.i_vbv_max_bitrate;
826             if( h->param.rc.i_rc_method == X264_RC_ABR && h->param.rc.i_vbv_buffer_size <= 0 )
827                 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate * 2;
828             x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
829             do h->param.i_level_idc = l->level_idc;
830                 while( l[1].level_idc && x264_validate_levels( h, 0 ) && l++ );
831             h->param.rc.i_vbv_max_bitrate = maxrate_bak;
832         }
833         else
834         {
835             while( l->level_idc && l->level_idc != h->param.i_level_idc )
836                 l++;
837             if( l->level_idc == 0 )
838             {
839                 x264_log( h, X264_LOG_ERROR, "invalid level_idc: %d\n", h->param.i_level_idc );
840                 return -1;
841             }
842         }
843         if( h->param.analyse.i_mv_range <= 0 )
844             h->param.analyse.i_mv_range = l->mv_range >> PARAM_INTERLACED;
845         else
846             h->param.analyse.i_mv_range = x264_clip3(h->param.analyse.i_mv_range, 32, 512 >> PARAM_INTERLACED);
847     }
848
849     h->param.analyse.i_weighted_pred = x264_clip3( h->param.analyse.i_weighted_pred, X264_WEIGHTP_NONE, X264_WEIGHTP_SMART );
850
851     if( PARAM_INTERLACED )
852     {
853         if( h->param.analyse.i_me_method >= X264_ME_ESA )
854         {
855             x264_log( h, X264_LOG_WARNING, "interlace + me=esa is not implemented\n" );
856             h->param.analyse.i_me_method = X264_ME_UMH;
857         }
858         if( h->param.analyse.i_weighted_pred > 0 )
859         {
860             x264_log( h, X264_LOG_WARNING, "interlace + weightp is not implemented\n" );
861             h->param.analyse.i_weighted_pred = X264_WEIGHTP_NONE;
862         }
863     }
864
865     if( !h->param.analyse.i_weighted_pred && h->param.rc.b_mb_tree && h->param.analyse.b_psy )
866         h->param.analyse.i_weighted_pred = X264_WEIGHTP_FAKE;
867
868     if( h->i_thread_frames > 1 )
869     {
870         int r = h->param.analyse.i_mv_range_thread;
871         int r2;
872         if( r <= 0 )
873         {
874             // half of the available space is reserved and divided evenly among the threads,
875             // the rest is allocated to whichever thread is far enough ahead to use it.
876             // reserving more space increases quality for some videos, but costs more time
877             // in thread synchronization.
878             int max_range = (h->param.i_height + X264_THREAD_HEIGHT) / h->i_thread_frames - X264_THREAD_HEIGHT;
879             r = max_range / 2;
880         }
881         r = X264_MAX( r, h->param.analyse.i_me_range );
882         r = X264_MIN( r, h->param.analyse.i_mv_range );
883         // round up to use the whole mb row
884         r2 = (r & ~15) + ((-X264_THREAD_HEIGHT) & 15);
885         if( r2 < r )
886             r2 += 16;
887         x264_log( h, X264_LOG_DEBUG, "using mv_range_thread = %d\n", r2 );
888         h->param.analyse.i_mv_range_thread = r2;
889     }
890
891     if( h->param.rc.f_rate_tolerance < 0 )
892         h->param.rc.f_rate_tolerance = 0;
893     if( h->param.rc.f_qblur < 0 )
894         h->param.rc.f_qblur = 0;
895     if( h->param.rc.f_complexity_blur < 0 )
896         h->param.rc.f_complexity_blur = 0;
897
898     h->param.i_sps_id &= 31;
899
900     if( PARAM_INTERLACED )
901         h->param.b_pic_struct = 1;
902
903     h->param.i_nal_hrd = x264_clip3( h->param.i_nal_hrd, X264_NAL_HRD_NONE, X264_NAL_HRD_CBR );
904
905     if( h->param.i_nal_hrd && !h->param.rc.i_vbv_buffer_size )
906     {
907         x264_log( h, X264_LOG_WARNING, "NAL HRD parameters require VBV parameters\n" );
908         h->param.i_nal_hrd = X264_NAL_HRD_NONE;
909     }
910
911     if( h->param.i_nal_hrd == X264_NAL_HRD_CBR &&
912        (h->param.rc.i_bitrate != h->param.rc.i_vbv_max_bitrate || !h->param.rc.i_vbv_max_bitrate) )
913     {
914         x264_log( h, X264_LOG_WARNING, "CBR HRD requires constant bitrate\n" );
915         h->param.i_nal_hrd = X264_NAL_HRD_VBR;
916     }
917
918     /* ensure the booleans are 0 or 1 so they can be used in math */
919 #define BOOLIFY(x) h->param.x = !!h->param.x
920     BOOLIFY( b_cabac );
921     BOOLIFY( b_constrained_intra );
922     BOOLIFY( b_deblocking_filter );
923     BOOLIFY( b_deterministic );
924     BOOLIFY( b_sliced_threads );
925     BOOLIFY( b_interlaced );
926     BOOLIFY( b_intra_refresh );
927     BOOLIFY( b_visualize );
928     BOOLIFY( b_aud );
929     BOOLIFY( b_repeat_headers );
930     BOOLIFY( b_annexb );
931     BOOLIFY( b_vfr_input );
932     BOOLIFY( b_pulldown );
933     BOOLIFY( b_tff );
934     BOOLIFY( b_pic_struct );
935     BOOLIFY( b_fake_interlaced );
936     BOOLIFY( b_open_gop );
937     BOOLIFY( b_bluray_compat );
938     BOOLIFY( analyse.b_transform_8x8 );
939     BOOLIFY( analyse.b_weighted_bipred );
940     BOOLIFY( analyse.b_chroma_me );
941     BOOLIFY( analyse.b_mixed_references );
942     BOOLIFY( analyse.b_fast_pskip );
943     BOOLIFY( analyse.b_dct_decimate );
944     BOOLIFY( analyse.b_psy );
945     BOOLIFY( analyse.b_psnr );
946     BOOLIFY( analyse.b_ssim );
947     BOOLIFY( rc.b_stat_write );
948     BOOLIFY( rc.b_stat_read );
949     BOOLIFY( rc.b_mb_tree );
950 #undef BOOLIFY
951
952     return 0;
953 }
954
955 static void mbcmp_init( x264_t *h )
956 {
957     int satd = !h->mb.b_lossless && h->param.analyse.i_subpel_refine > 1;
958     memcpy( h->pixf.mbcmp, satd ? h->pixf.satd : h->pixf.sad_aligned, sizeof(h->pixf.mbcmp) );
959     memcpy( h->pixf.mbcmp_unaligned, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.mbcmp_unaligned) );
960     h->pixf.intra_mbcmp_x3_16x16 = satd ? h->pixf.intra_satd_x3_16x16 : h->pixf.intra_sad_x3_16x16;
961     h->pixf.intra_mbcmp_x3_8x16c = satd ? h->pixf.intra_satd_x3_8x16c : h->pixf.intra_sad_x3_8x16c;
962     h->pixf.intra_mbcmp_x3_8x8c  = satd ? h->pixf.intra_satd_x3_8x8c  : h->pixf.intra_sad_x3_8x8c;
963     h->pixf.intra_mbcmp_x3_8x8 = satd ? h->pixf.intra_sa8d_x3_8x8 : h->pixf.intra_sad_x3_8x8;
964     h->pixf.intra_mbcmp_x3_4x4 = satd ? h->pixf.intra_satd_x3_4x4 : h->pixf.intra_sad_x3_4x4;
965     h->pixf.intra_mbcmp_x9_4x4 = h->param.b_cpu_independent || h->mb.b_lossless ? NULL
966                                : satd ? h->pixf.intra_satd_x9_4x4 : h->pixf.intra_sad_x9_4x4;
967     h->pixf.intra_mbcmp_x9_8x8 = h->param.b_cpu_independent || h->mb.b_lossless ? NULL
968                                : satd ? h->pixf.intra_sa8d_x9_8x8 : h->pixf.intra_sad_x9_8x8;
969     satd &= h->param.analyse.i_me_method == X264_ME_TESA;
970     memcpy( h->pixf.fpelcmp, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.fpelcmp) );
971     memcpy( h->pixf.fpelcmp_x3, satd ? h->pixf.satd_x3 : h->pixf.sad_x3, sizeof(h->pixf.fpelcmp_x3) );
972     memcpy( h->pixf.fpelcmp_x4, satd ? h->pixf.satd_x4 : h->pixf.sad_x4, sizeof(h->pixf.fpelcmp_x4) );
973 }
974
975 static void chroma_dsp_init( x264_t *h )
976 {
977     memcpy( h->luma2chroma_pixel, x264_luma2chroma_pixel[CHROMA_FORMAT], sizeof(h->luma2chroma_pixel) );
978
979     switch( CHROMA_FORMAT )
980     {
981         case CHROMA_420:
982             memcpy( h->predict_chroma, h->predict_8x8c, sizeof(h->predict_chroma) );
983             h->mc.prefetch_fenc = h->mc.prefetch_fenc_420;
984             h->loopf.deblock_chroma[0] = h->loopf.deblock_h_chroma_420;
985             h->loopf.deblock_chroma_intra[0] = h->loopf.deblock_h_chroma_420_intra;
986             h->loopf.deblock_chroma_mbaff = h->loopf.deblock_chroma_420_mbaff;
987             h->loopf.deblock_chroma_intra_mbaff = h->loopf.deblock_chroma_420_intra_mbaff;
988             h->pixf.intra_mbcmp_x3_chroma = h->pixf.intra_mbcmp_x3_8x8c;
989             h->quantf.coeff_last[DCT_CHROMA_DC] = h->quantf.coeff_last4;
990             h->quantf.coeff_level_run[DCT_CHROMA_DC] = h->quantf.coeff_level_run4;
991             break;
992         case CHROMA_422:
993             memcpy( h->predict_chroma, h->predict_8x16c, sizeof(h->predict_chroma) );
994             h->mc.prefetch_fenc = h->mc.prefetch_fenc_422;
995             h->loopf.deblock_chroma[0] = h->loopf.deblock_h_chroma_422;
996             h->loopf.deblock_chroma_intra[0] = h->loopf.deblock_h_chroma_422_intra;
997             h->loopf.deblock_chroma_mbaff = h->loopf.deblock_chroma_422_mbaff;
998             h->loopf.deblock_chroma_intra_mbaff = h->loopf.deblock_chroma_422_intra_mbaff;
999             h->pixf.intra_mbcmp_x3_chroma = h->pixf.intra_mbcmp_x3_8x16c;
1000             h->quantf.coeff_last[DCT_CHROMA_DC] = h->quantf.coeff_last8;
1001             h->quantf.coeff_level_run[DCT_CHROMA_DC] = h->quantf.coeff_level_run8;
1002             break;
1003         case CHROMA_444:
1004             h->mc.prefetch_fenc = h->mc.prefetch_fenc_422; /* FIXME: doesn't cover V plane */
1005             h->loopf.deblock_chroma_mbaff = h->loopf.deblock_luma_mbaff;
1006             h->loopf.deblock_chroma_intra_mbaff = h->loopf.deblock_luma_intra_mbaff;
1007             break;
1008     }
1009 }
1010
1011 static void x264_set_aspect_ratio( x264_t *h, x264_param_t *param, int initial )
1012 {
1013     /* VUI */
1014     if( param->vui.i_sar_width > 0 && param->vui.i_sar_height > 0 )
1015     {
1016         uint32_t i_w = param->vui.i_sar_width;
1017         uint32_t i_h = param->vui.i_sar_height;
1018         uint32_t old_w = h->param.vui.i_sar_width;
1019         uint32_t old_h = h->param.vui.i_sar_height;
1020
1021         x264_reduce_fraction( &i_w, &i_h );
1022
1023         while( i_w > 65535 || i_h > 65535 )
1024         {
1025             i_w /= 2;
1026             i_h /= 2;
1027         }
1028
1029         x264_reduce_fraction( &i_w, &i_h );
1030
1031         if( i_w != old_w || i_h != old_h || initial )
1032         {
1033             h->param.vui.i_sar_width = 0;
1034             h->param.vui.i_sar_height = 0;
1035             if( i_w == 0 || i_h == 0 )
1036                 x264_log( h, X264_LOG_WARNING, "cannot create valid sample aspect ratio\n" );
1037             else
1038             {
1039                 x264_log( h, initial?X264_LOG_INFO:X264_LOG_DEBUG, "using SAR=%d/%d\n", i_w, i_h );
1040                 h->param.vui.i_sar_width = i_w;
1041                 h->param.vui.i_sar_height = i_h;
1042             }
1043             x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
1044         }
1045     }
1046 }
1047
1048 /****************************************************************************
1049  * x264_encoder_open:
1050  ****************************************************************************/
1051 x264_t *x264_encoder_open( x264_param_t *param )
1052 {
1053     x264_t *h;
1054     char buf[1000], *p;
1055     int qp, i_slicetype_length;
1056
1057     CHECKED_MALLOCZERO( h, sizeof(x264_t) );
1058
1059     /* Create a copy of param */
1060     memcpy( &h->param, param, sizeof(x264_param_t) );
1061
1062     if( param->param_free )
1063         param->param_free( param );
1064
1065     if( x264_threading_init() )
1066     {
1067         x264_log( h, X264_LOG_ERROR, "unable to initialize threading\n" );
1068         goto fail;
1069     }
1070
1071     if( x264_validate_parameters( h, 1 ) < 0 )
1072         goto fail;
1073
1074     if( h->param.psz_cqm_file )
1075         if( x264_cqm_parse_file( h, h->param.psz_cqm_file ) < 0 )
1076             goto fail;
1077
1078     if( h->param.rc.psz_stat_out )
1079         h->param.rc.psz_stat_out = strdup( h->param.rc.psz_stat_out );
1080     if( h->param.rc.psz_stat_in )
1081         h->param.rc.psz_stat_in = strdup( h->param.rc.psz_stat_in );
1082
1083     x264_reduce_fraction( &h->param.i_fps_num, &h->param.i_fps_den );
1084     x264_reduce_fraction( &h->param.i_timebase_num, &h->param.i_timebase_den );
1085
1086     /* Init x264_t */
1087     h->i_frame = -1;
1088     h->i_frame_num = 0;
1089     h->i_idr_pic_id = 0;
1090
1091     if( (uint64_t)h->param.i_timebase_den * 2 > UINT32_MAX )
1092     {
1093         x264_log( h, X264_LOG_ERROR, "Effective timebase denominator %u exceeds H.264 maximum\n", h->param.i_timebase_den );
1094         goto fail;
1095     }
1096
1097     x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
1098     x264_pps_init( h->pps, h->param.i_sps_id, &h->param, h->sps );
1099
1100     x264_set_aspect_ratio( h, &h->param, 1 );
1101
1102     x264_validate_levels( h, 1 );
1103
1104     h->chroma_qp_table = i_chroma_qp_table + 12 + h->pps->i_chroma_qp_index_offset;
1105
1106     if( x264_cqm_init( h ) < 0 )
1107         goto fail;
1108
1109     h->mb.i_mb_width = h->sps->i_mb_width;
1110     h->mb.i_mb_height = h->sps->i_mb_height;
1111     h->mb.i_mb_count = h->mb.i_mb_width * h->mb.i_mb_height;
1112
1113     h->mb.chroma_h_shift = CHROMA_FORMAT == CHROMA_420 || CHROMA_FORMAT == CHROMA_422;
1114     h->mb.chroma_v_shift = CHROMA_FORMAT == CHROMA_420;
1115
1116     /* Adaptive MBAFF and subme 0 are not supported as we require halving motion
1117      * vectors during prediction, resulting in hpel mvs.
1118      * The chosen solution is to make MBAFF non-adaptive in this case. */
1119     h->mb.b_adaptive_mbaff = PARAM_INTERLACED && h->param.analyse.i_subpel_refine;
1120
1121     /* Init frames. */
1122     if( h->param.i_bframe_adaptive == X264_B_ADAPT_TRELLIS && !h->param.rc.b_stat_read )
1123         h->frames.i_delay = X264_MAX(h->param.i_bframe,3)*4;
1124     else
1125         h->frames.i_delay = h->param.i_bframe;
1126     if( h->param.rc.b_mb_tree || h->param.rc.i_vbv_buffer_size )
1127         h->frames.i_delay = X264_MAX( h->frames.i_delay, h->param.rc.i_lookahead );
1128     i_slicetype_length = h->frames.i_delay;
1129     h->frames.i_delay += h->i_thread_frames - 1;
1130     h->frames.i_delay += h->param.i_sync_lookahead;
1131     h->frames.i_delay += h->param.b_vfr_input;
1132     h->frames.i_bframe_delay = h->param.i_bframe ? (h->param.i_bframe_pyramid ? 2 : 1) : 0;
1133
1134     h->frames.i_max_ref0 = h->param.i_frame_reference;
1135     h->frames.i_max_ref1 = X264_MIN( h->sps->vui.i_num_reorder_frames, h->param.i_frame_reference );
1136     h->frames.i_max_dpb  = h->sps->vui.i_max_dec_frame_buffering;
1137     h->frames.b_have_lowres = !h->param.rc.b_stat_read
1138         && ( h->param.rc.i_rc_method == X264_RC_ABR
1139           || h->param.rc.i_rc_method == X264_RC_CRF
1140           || h->param.i_bframe_adaptive
1141           || h->param.i_scenecut_threshold
1142           || h->param.rc.b_mb_tree
1143           || h->param.analyse.i_weighted_pred );
1144     h->frames.b_have_lowres |= h->param.rc.b_stat_read && h->param.rc.i_vbv_buffer_size > 0;
1145     h->frames.b_have_sub8x8_esa = !!(h->param.analyse.inter & X264_ANALYSE_PSUB8x8);
1146
1147     h->frames.i_last_idr =
1148     h->frames.i_last_keyframe = - h->param.i_keyint_max;
1149     h->frames.i_input    = 0;
1150     h->frames.i_largest_pts = h->frames.i_second_largest_pts = -1;
1151     h->frames.i_poc_last_open_gop = -1;
1152
1153     CHECKED_MALLOCZERO( h->frames.unused[0], (h->frames.i_delay + 3) * sizeof(x264_frame_t *) );
1154     /* Allocate room for max refs plus a few extra just in case. */
1155     CHECKED_MALLOCZERO( h->frames.unused[1], (h->i_thread_frames + X264_REF_MAX + 4) * sizeof(x264_frame_t *) );
1156     CHECKED_MALLOCZERO( h->frames.current, (h->param.i_sync_lookahead + h->param.i_bframe
1157                         + h->i_thread_frames + 3) * sizeof(x264_frame_t *) );
1158     if( h->param.analyse.i_weighted_pred > 0 )
1159         CHECKED_MALLOCZERO( h->frames.blank_unused, h->i_thread_frames * 4 * sizeof(x264_frame_t *) );
1160     h->i_ref[0] = h->i_ref[1] = 0;
1161     h->i_cpb_delay = h->i_coded_fields = h->i_disp_fields = 0;
1162     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);
1163     h->i_disp_fields_last_frame = -1;
1164     x264_rdo_init();
1165
1166     /* init CPU functions */
1167     x264_predict_16x16_init( h->param.cpu, h->predict_16x16 );
1168     x264_predict_8x8c_init( h->param.cpu, h->predict_8x8c );
1169     x264_predict_8x16c_init( h->param.cpu, h->predict_8x16c );
1170     x264_predict_8x8_init( h->param.cpu, h->predict_8x8, &h->predict_8x8_filter );
1171     x264_predict_4x4_init( h->param.cpu, h->predict_4x4 );
1172     if( h->param.b_cabac )
1173         x264_cabac_init( h );
1174     else
1175         x264_cavlc_init();
1176     x264_pixel_init( h->param.cpu, &h->pixf );
1177     x264_dct_init( h->param.cpu, &h->dctf );
1178     x264_zigzag_init( h->param.cpu, &h->zigzagf_progressive, &h->zigzagf_interlaced );
1179     memcpy( &h->zigzagf, PARAM_INTERLACED ? &h->zigzagf_interlaced : &h->zigzagf_progressive, sizeof(h->zigzagf) );
1180     x264_mc_init( h->param.cpu, &h->mc );
1181     x264_quant_init( h, h->param.cpu, &h->quantf );
1182     x264_deblock_init( h->param.cpu, &h->loopf, PARAM_INTERLACED );
1183     x264_bitstream_init( h->param.cpu, &h->bsf );
1184     x264_dct_init_weights();
1185
1186     mbcmp_init( h );
1187     chroma_dsp_init( h );
1188
1189     p = buf + sprintf( buf, "using cpu capabilities:" );
1190     for( int i = 0; x264_cpu_names[i].flags; i++ )
1191     {
1192         if( !strcmp(x264_cpu_names[i].name, "SSE2")
1193             && h->param.cpu & (X264_CPU_SSE2_IS_FAST|X264_CPU_SSE2_IS_SLOW) )
1194             continue;
1195         if( !strcmp(x264_cpu_names[i].name, "SSE3")
1196             && (h->param.cpu & X264_CPU_SSSE3 || !(h->param.cpu & X264_CPU_CACHELINE_64)) )
1197             continue;
1198         if( !strcmp(x264_cpu_names[i].name, "SSE4.1")
1199             && (h->param.cpu & X264_CPU_SSE42) )
1200             continue;
1201         if( (h->param.cpu & x264_cpu_names[i].flags) == x264_cpu_names[i].flags
1202             && (!i || x264_cpu_names[i].flags != x264_cpu_names[i-1].flags) )
1203             p += sprintf( p, " %s", x264_cpu_names[i].name );
1204     }
1205     if( !h->param.cpu )
1206         p += sprintf( p, " none!" );
1207     x264_log( h, X264_LOG_INFO, "%s\n", buf );
1208
1209     float *logs = x264_analyse_prepare_costs( h );
1210     if( !logs )
1211         goto fail;
1212     for( qp = X264_MIN( h->param.rc.i_qp_min, QP_MAX_SPEC ); qp <= h->param.rc.i_qp_max; qp++ )
1213         if( x264_analyse_init_costs( h, logs, qp ) )
1214             goto fail;
1215     if( x264_analyse_init_costs( h, logs, X264_LOOKAHEAD_QP ) )
1216         goto fail;
1217     x264_free( logs );
1218
1219     static const uint16_t cost_mv_correct[7] = { 24, 47, 95, 189, 379, 757, 1515 };
1220     /* Checks for known miscompilation issues. */
1221     if( h->cost_mv[X264_LOOKAHEAD_QP][2013] != cost_mv_correct[BIT_DEPTH-8] )
1222     {
1223         x264_log( h, X264_LOG_ERROR, "MV cost test failed: x264 has been miscompiled!\n" );
1224         goto fail;
1225     }
1226
1227     /* Must be volatile or else GCC will optimize it out. */
1228     volatile int temp = 392;
1229     if( x264_clz( temp ) != 23 )
1230     {
1231         x264_log( h, X264_LOG_ERROR, "CLZ test failed: x264 has been miscompiled!\n" );
1232 #if ARCH_X86 || ARCH_X86_64
1233         x264_log( h, X264_LOG_ERROR, "Are you attempting to run an SSE4a-targeted build on a CPU that\n" );
1234         x264_log( h, X264_LOG_ERROR, "doesn't support it?\n" );
1235 #endif
1236         goto fail;
1237     }
1238
1239     h->out.i_nal = 0;
1240     h->out.i_bitstream = X264_MAX( 1000000, h->param.i_width * h->param.i_height * 4
1241         * ( h->param.rc.i_rc_method == X264_RC_ABR ? pow( 0.95, h->param.rc.i_qp_min )
1242           : pow( 0.95, h->param.rc.i_qp_constant ) * X264_MAX( 1, h->param.rc.f_ip_factor )));
1243
1244     h->nal_buffer_size = h->out.i_bitstream * 3/2 + 4;
1245     CHECKED_MALLOC( h->nal_buffer, h->nal_buffer_size );
1246
1247     if( h->param.i_threads > 1 &&
1248         x264_threadpool_init( &h->threadpool, h->param.i_threads, (void*)x264_encoder_thread_init, h ) )
1249         goto fail;
1250
1251     h->thread[0] = h;
1252     for( int i = 1; i < h->param.i_threads + !!h->param.i_sync_lookahead; i++ )
1253         CHECKED_MALLOC( h->thread[i], sizeof(x264_t) );
1254
1255     for( int i = 0; i < h->param.i_threads; i++ )
1256     {
1257         int init_nal_count = h->param.i_slice_count + 3;
1258         int allocate_threadlocal_data = !h->param.b_sliced_threads || !i;
1259         if( i > 0 )
1260             *h->thread[i] = *h;
1261
1262         if( allocate_threadlocal_data )
1263         {
1264             h->thread[i]->fdec = x264_frame_pop_unused( h, 1 );
1265             if( !h->thread[i]->fdec )
1266                 goto fail;
1267         }
1268         else
1269             h->thread[i]->fdec = h->thread[0]->fdec;
1270
1271         CHECKED_MALLOC( h->thread[i]->out.p_bitstream, h->out.i_bitstream );
1272         /* Start each thread with room for init_nal_count NAL units; it'll realloc later if needed. */
1273         CHECKED_MALLOC( h->thread[i]->out.nal, init_nal_count*sizeof(x264_nal_t) );
1274         h->thread[i]->out.i_nals_allocated = init_nal_count;
1275
1276         if( allocate_threadlocal_data && x264_macroblock_cache_allocate( h->thread[i] ) < 0 )
1277             goto fail;
1278     }
1279
1280     if( x264_lookahead_init( h, i_slicetype_length ) )
1281         goto fail;
1282
1283     for( int i = 0; i < h->param.i_threads; i++ )
1284         if( x264_macroblock_thread_allocate( h->thread[i], 0 ) < 0 )
1285             goto fail;
1286
1287     if( x264_ratecontrol_new( h ) < 0 )
1288         goto fail;
1289
1290     if( h->param.i_nal_hrd )
1291     {
1292         x264_log( h, X264_LOG_DEBUG, "HRD bitrate: %i bits/sec\n", h->sps->vui.hrd.i_bit_rate_unscaled );
1293         x264_log( h, X264_LOG_DEBUG, "CPB size: %i bits\n", h->sps->vui.hrd.i_cpb_size_unscaled );
1294     }
1295
1296     if( h->param.psz_dump_yuv )
1297     {
1298         /* create or truncate the reconstructed video file */
1299         FILE *f = fopen( h->param.psz_dump_yuv, "w" );
1300         if( !f )
1301         {
1302             x264_log( h, X264_LOG_ERROR, "dump_yuv: can't write to %s\n", h->param.psz_dump_yuv );
1303             goto fail;
1304         }
1305         else if( !x264_is_regular_file( f ) )
1306         {
1307             x264_log( h, X264_LOG_ERROR, "dump_yuv: incompatible with non-regular file %s\n", h->param.psz_dump_yuv );
1308             goto fail;
1309         }
1310         fclose( f );
1311     }
1312
1313     const char *profile = h->sps->i_profile_idc == PROFILE_BASELINE ? "Constrained Baseline" :
1314                           h->sps->i_profile_idc == PROFILE_MAIN ? "Main" :
1315                           h->sps->i_profile_idc == PROFILE_HIGH ? "High" :
1316                           h->sps->i_profile_idc == PROFILE_HIGH10 ? (h->sps->b_constraint_set3 == 1 ? "High 10 Intra" : "High 10") :
1317                           h->sps->i_profile_idc == PROFILE_HIGH422 ? (h->sps->b_constraint_set3 == 1 ? "High 4:2:2 Intra" : "High 4:2:2") :
1318                           h->sps->b_constraint_set3 == 1 ? "High 4:4:4 Intra" : "High 4:4:4 Predictive";
1319     char level[4];
1320     snprintf( level, sizeof(level), "%d.%d", h->sps->i_level_idc/10, h->sps->i_level_idc%10 );
1321     if( h->sps->i_level_idc == 9 || ( h->sps->i_level_idc == 11 && h->sps->b_constraint_set3 &&
1322         (h->sps->i_profile_idc >= PROFILE_BASELINE && h->sps->i_profile_idc <= PROFILE_EXTENDED) ) )
1323         strcpy( level, "1b" );
1324
1325     if( h->sps->i_profile_idc < PROFILE_HIGH10 )
1326     {
1327         x264_log( h, X264_LOG_INFO, "profile %s, level %s\n",
1328             profile, level );
1329     }
1330     else
1331     {
1332         static const char * const subsampling[4] = { "4:0:0", "4:2:0", "4:2:2", "4:4:4" };
1333         x264_log( h, X264_LOG_INFO, "profile %s, level %s, %s %d-bit\n",
1334             profile, level, subsampling[CHROMA_FORMAT], BIT_DEPTH );
1335     }
1336
1337     return h;
1338 fail:
1339     x264_free( h );
1340     return NULL;
1341 }
1342
1343 /****************************************************************************
1344  * x264_encoder_reconfig:
1345  ****************************************************************************/
1346 int x264_encoder_reconfig( x264_t *h, x264_param_t *param )
1347 {
1348     int rc_reconfig = 0;
1349     h = h->thread[h->thread[0]->i_thread_phase];
1350     x264_set_aspect_ratio( h, param, 0 );
1351 #define COPY(var) h->param.var = param->var
1352     COPY( i_frame_reference ); // but never uses more refs than initially specified
1353     COPY( i_bframe_bias );
1354     if( h->param.i_scenecut_threshold )
1355         COPY( i_scenecut_threshold ); // can't turn it on or off, only vary the threshold
1356     COPY( b_deblocking_filter );
1357     COPY( i_deblocking_filter_alphac0 );
1358     COPY( i_deblocking_filter_beta );
1359     COPY( i_frame_packing );
1360     COPY( analyse.inter );
1361     COPY( analyse.intra );
1362     COPY( analyse.i_direct_mv_pred );
1363     /* Scratch buffer prevents me_range from being increased for esa/tesa */
1364     if( h->param.analyse.i_me_method < X264_ME_ESA || param->analyse.i_me_range < h->param.analyse.i_me_range )
1365         COPY( analyse.i_me_range );
1366     COPY( analyse.i_noise_reduction );
1367     /* We can't switch out of subme=0 during encoding. */
1368     if( h->param.analyse.i_subpel_refine )
1369         COPY( analyse.i_subpel_refine );
1370     COPY( analyse.i_trellis );
1371     COPY( analyse.b_chroma_me );
1372     COPY( analyse.b_dct_decimate );
1373     COPY( analyse.b_fast_pskip );
1374     COPY( analyse.b_mixed_references );
1375     COPY( analyse.f_psy_rd );
1376     COPY( analyse.f_psy_trellis );
1377     COPY( crop_rect );
1378     // can only twiddle these if they were enabled to begin with:
1379     if( h->param.analyse.i_me_method >= X264_ME_ESA || param->analyse.i_me_method < X264_ME_ESA )
1380         COPY( analyse.i_me_method );
1381     if( h->param.analyse.i_me_method >= X264_ME_ESA && !h->frames.b_have_sub8x8_esa )
1382         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
1383     if( h->pps->b_transform_8x8_mode )
1384         COPY( analyse.b_transform_8x8 );
1385     if( h->frames.i_max_ref1 > 1 )
1386         COPY( i_bframe_pyramid );
1387     COPY( i_slice_max_size );
1388     COPY( i_slice_max_mbs );
1389     COPY( i_slice_count );
1390     COPY( b_tff );
1391
1392     /* VBV can't be turned on if it wasn't on to begin with */
1393     if( h->param.rc.i_vbv_max_bitrate > 0 && h->param.rc.i_vbv_buffer_size > 0 &&
1394           param->rc.i_vbv_max_bitrate > 0 &&   param->rc.i_vbv_buffer_size > 0 )
1395     {
1396         rc_reconfig |= h->param.rc.i_vbv_max_bitrate != param->rc.i_vbv_max_bitrate;
1397         rc_reconfig |= h->param.rc.i_vbv_buffer_size != param->rc.i_vbv_buffer_size;
1398         rc_reconfig |= h->param.rc.i_bitrate != param->rc.i_bitrate;
1399         COPY( rc.i_vbv_max_bitrate );
1400         COPY( rc.i_vbv_buffer_size );
1401         COPY( rc.i_bitrate );
1402     }
1403     rc_reconfig |= h->param.rc.f_rf_constant != param->rc.f_rf_constant;
1404     rc_reconfig |= h->param.rc.f_rf_constant_max != param->rc.f_rf_constant_max;
1405     COPY( rc.f_rf_constant );
1406     COPY( rc.f_rf_constant_max );
1407 #undef COPY
1408
1409     mbcmp_init( h );
1410
1411     int ret = x264_validate_parameters( h, 0 );
1412
1413     /* Supported reconfiguration options (1-pass only):
1414      * vbv-maxrate
1415      * vbv-bufsize
1416      * crf
1417      * bitrate (CBR only) */
1418     if( !ret && rc_reconfig )
1419         x264_ratecontrol_init_reconfigurable( h, 0 );
1420
1421     return ret;
1422 }
1423
1424 /****************************************************************************
1425  * x264_encoder_parameters:
1426  ****************************************************************************/
1427 void x264_encoder_parameters( x264_t *h, x264_param_t *param )
1428 {
1429     memcpy( param, &h->thread[h->i_thread_phase]->param, sizeof(x264_param_t) );
1430 }
1431
1432 /* internal usage */
1433 static void x264_nal_start( x264_t *h, int i_type, int i_ref_idc )
1434 {
1435     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1436
1437     nal->i_ref_idc        = i_ref_idc;
1438     nal->i_type           = i_type;
1439     nal->b_long_startcode = 1;
1440
1441     nal->i_payload= 0;
1442     nal->p_payload= &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8];
1443 }
1444
1445 /* if number of allocated nals is not enough, re-allocate a larger one. */
1446 static int x264_nal_check_buffer( x264_t *h )
1447 {
1448     if( h->out.i_nal >= h->out.i_nals_allocated )
1449     {
1450         x264_nal_t *new_out = x264_malloc( sizeof(x264_nal_t) * (h->out.i_nals_allocated*2) );
1451         if( !new_out )
1452             return -1;
1453         memcpy( new_out, h->out.nal, sizeof(x264_nal_t) * (h->out.i_nals_allocated) );
1454         x264_free( h->out.nal );
1455         h->out.nal = new_out;
1456         h->out.i_nals_allocated *= 2;
1457     }
1458     return 0;
1459 }
1460
1461 static int x264_nal_end( x264_t *h )
1462 {
1463     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1464     uint8_t *end = &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8];
1465     nal->i_payload = end - nal->p_payload;
1466     /* nal_escape_mmx reads past the end of the input.
1467      * While undefined padding wouldn't actually affect the output, it makes valgrind unhappy. */
1468     memset( end, 0xff, 32 );
1469     if( h->param.nalu_process )
1470         h->param.nalu_process( h, nal );
1471     h->out.i_nal++;
1472
1473     return x264_nal_check_buffer( h );
1474 }
1475
1476 static int x264_encoder_encapsulate_nals( x264_t *h, int start )
1477 {
1478     int nal_size = 0, previous_nal_size = 0;
1479
1480     if( h->param.nalu_process )
1481     {
1482         for( int i = start; i < h->out.i_nal; i++ )
1483             nal_size += h->out.nal[i].i_payload;
1484         return nal_size;
1485     }
1486
1487     for( int i = 0; i < start; i++ )
1488         previous_nal_size += h->out.nal[i].i_payload;
1489
1490     for( int i = start; i < h->out.i_nal; i++ )
1491         nal_size += h->out.nal[i].i_payload;
1492
1493     /* Worst-case NAL unit escaping: reallocate the buffer if it's too small. */
1494     int necessary_size = nal_size * 3/2 + h->out.i_nal * 4;
1495     if( h->nal_buffer_size < necessary_size )
1496     {
1497         h->nal_buffer_size = necessary_size * 2;
1498         uint8_t *buf = x264_malloc( h->nal_buffer_size );
1499         if( !buf )
1500             return -1;
1501         if( previous_nal_size )
1502             memcpy( buf, h->nal_buffer, previous_nal_size );
1503         x264_free( h->nal_buffer );
1504         h->nal_buffer = buf;
1505     }
1506
1507     uint8_t *nal_buffer = h->nal_buffer + previous_nal_size;
1508
1509     for( int i = start; i < h->out.i_nal; i++ )
1510     {
1511         h->out.nal[i].b_long_startcode = !i || h->out.nal[i].i_type == NAL_SPS || h->out.nal[i].i_type == NAL_PPS;
1512         x264_nal_encode( h, nal_buffer, &h->out.nal[i] );
1513         nal_buffer += h->out.nal[i].i_payload;
1514     }
1515
1516     x264_emms();
1517
1518     return nal_buffer - (h->nal_buffer + previous_nal_size);
1519 }
1520
1521 /****************************************************************************
1522  * x264_encoder_headers:
1523  ****************************************************************************/
1524 int x264_encoder_headers( x264_t *h, x264_nal_t **pp_nal, int *pi_nal )
1525 {
1526     int frame_size = 0;
1527     /* init bitstream context */
1528     h->out.i_nal = 0;
1529     bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
1530
1531     /* Write SEI, SPS and PPS. */
1532
1533     /* generate sequence parameters */
1534     x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
1535     x264_sps_write( &h->out.bs, h->sps );
1536     if( x264_nal_end( h ) )
1537         return -1;
1538
1539     /* generate picture parameters */
1540     x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
1541     x264_pps_write( &h->out.bs, h->sps, h->pps );
1542     if( x264_nal_end( h ) )
1543         return -1;
1544
1545     /* identify ourselves */
1546     x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
1547     if( x264_sei_version_write( h, &h->out.bs ) )
1548         return -1;
1549     if( x264_nal_end( h ) )
1550         return -1;
1551
1552     frame_size = x264_encoder_encapsulate_nals( h, 0 );
1553     if( frame_size < 0 )
1554         return -1;
1555
1556     /* now set output*/
1557     *pi_nal = h->out.i_nal;
1558     *pp_nal = &h->out.nal[0];
1559     h->out.i_nal = 0;
1560
1561     return frame_size;
1562 }
1563
1564 /* Check to see whether we have chosen a reference list ordering different
1565  * from the standard's default. */
1566 static inline void x264_reference_check_reorder( x264_t *h )
1567 {
1568     /* The reorder check doesn't check for missing frames, so just
1569      * force a reorder if one of the reference list is corrupt. */
1570     for( int i = 0; h->frames.reference[i]; i++ )
1571         if( h->frames.reference[i]->b_corrupt )
1572         {
1573             h->b_ref_reorder[0] = 1;
1574             return;
1575         }
1576     for( int list = 0; list <= (h->sh.i_type == SLICE_TYPE_B); list++ )
1577         for( int i = 0; i < h->i_ref[list] - 1; i++ )
1578         {
1579             int framenum_diff = h->fref[list][i+1]->i_frame_num - h->fref[list][i]->i_frame_num;
1580             int poc_diff = h->fref[list][i+1]->i_poc - h->fref[list][i]->i_poc;
1581             /* P and B-frames use different default orders. */
1582             if( h->sh.i_type == SLICE_TYPE_P ? framenum_diff > 0 : list == 1 ? poc_diff < 0 : poc_diff > 0 )
1583             {
1584                 h->b_ref_reorder[list] = 1;
1585                 return;
1586             }
1587         }
1588 }
1589
1590 /* return -1 on failure, else return the index of the new reference frame */
1591 int x264_weighted_reference_duplicate( x264_t *h, int i_ref, const x264_weight_t *w )
1592 {
1593     int i = h->i_ref[0];
1594     int j = 1;
1595     x264_frame_t *newframe;
1596     if( i <= 1 ) /* empty list, definitely can't duplicate frame */
1597         return -1;
1598
1599     //Duplication is only used in X264_WEIGHTP_SMART
1600     if( h->param.analyse.i_weighted_pred != X264_WEIGHTP_SMART )
1601         return -1;
1602
1603     /* Duplication is a hack to compensate for crappy rounding in motion compensation.
1604      * With high bit depth, it's not worth doing, so turn it off except in the case of
1605      * unweighted dupes. */
1606     if( BIT_DEPTH > 8 && w != x264_weight_none )
1607         return -1;
1608
1609     newframe = x264_frame_pop_blank_unused( h );
1610     if( !newframe )
1611         return -1;
1612
1613     //FIXME: probably don't need to copy everything
1614     *newframe = *h->fref[0][i_ref];
1615     newframe->i_reference_count = 1;
1616     newframe->orig = h->fref[0][i_ref];
1617     newframe->b_duplicate = 1;
1618     memcpy( h->fenc->weight[j], w, sizeof(h->fenc->weight[i]) );
1619
1620     /* shift the frames to make space for the dupe. */
1621     h->b_ref_reorder[0] = 1;
1622     if( h->i_ref[0] < X264_REF_MAX )
1623         ++h->i_ref[0];
1624     h->fref[0][X264_REF_MAX-1] = NULL;
1625     x264_frame_unshift( &h->fref[0][j], newframe );
1626
1627     return j;
1628 }
1629
1630 static void x264_weighted_pred_init( x264_t *h )
1631 {
1632     /* for now no analysis and set all weights to nothing */
1633     for( int i_ref = 0; i_ref < h->i_ref[0]; i_ref++ )
1634         h->fenc->weighted[i_ref] = h->fref[0][i_ref]->filtered[0][0];
1635
1636     // FIXME: This only supports weighting of one reference frame
1637     // and duplicates of that frame.
1638     h->fenc->i_lines_weighted = 0;
1639
1640     for( int i_ref = 0; i_ref < (h->i_ref[0] << SLICE_MBAFF); i_ref++ )
1641         for( int i = 0; i < 3; i++ )
1642             h->sh.weight[i_ref][i].weightfn = NULL;
1643
1644
1645     if( h->sh.i_type != SLICE_TYPE_P || h->param.analyse.i_weighted_pred <= 0 )
1646         return;
1647
1648     int i_padv = PADV << PARAM_INTERLACED;
1649     int denom = -1;
1650     int weightplane[2] = { 0, 0 };
1651     int buffer_next = 0;
1652     for( int i = 0; i < 3; i++ )
1653     {
1654         for( int j = 0; j < h->i_ref[0]; j++ )
1655         {
1656             if( h->fenc->weight[j][i].weightfn )
1657             {
1658                 h->sh.weight[j][i] = h->fenc->weight[j][i];
1659                 // if weight is useless, don't write it to stream
1660                 if( h->sh.weight[j][i].i_scale == 1<<h->sh.weight[j][i].i_denom && h->sh.weight[j][i].i_offset == 0 )
1661                     h->sh.weight[j][i].weightfn = NULL;
1662                 else
1663                 {
1664                     if( !weightplane[!!i] )
1665                     {
1666                         weightplane[!!i] = 1;
1667                         h->sh.weight[0][!!i].i_denom = denom = h->sh.weight[j][i].i_denom;
1668                         assert( x264_clip3( denom, 0, 7 ) == denom );
1669                     }
1670
1671                     assert( h->sh.weight[j][i].i_denom == denom );
1672                     if( !i )
1673                     {
1674                         h->fenc->weighted[j] = h->mb.p_weight_buf[buffer_next++] + h->fenc->i_stride[0] * i_padv + PADH;
1675                         //scale full resolution frame
1676                         if( h->param.i_threads == 1 )
1677                         {
1678                             pixel *src = h->fref[0][j]->filtered[0][0] - h->fref[0][j]->i_stride[0]*i_padv - PADH;
1679                             pixel *dst = h->fenc->weighted[j] - h->fenc->i_stride[0]*i_padv - PADH;
1680                             int stride = h->fenc->i_stride[0];
1681                             int width = h->fenc->i_width[0] + PADH*2;
1682                             int height = h->fenc->i_lines[0] + i_padv*2;
1683                             x264_weight_scale_plane( h, dst, stride, src, stride, width, height, &h->sh.weight[j][0] );
1684                             h->fenc->i_lines_weighted = height;
1685                         }
1686                     }
1687                 }
1688             }
1689         }
1690     }
1691
1692     if( weightplane[1] )
1693         for( int i = 0; i < h->i_ref[0]; i++ )
1694         {
1695             if( h->sh.weight[i][1].weightfn && !h->sh.weight[i][2].weightfn )
1696             {
1697                 h->sh.weight[i][2].i_scale = 1 << h->sh.weight[0][1].i_denom;
1698                 h->sh.weight[i][2].i_offset = 0;
1699             }
1700             else if( h->sh.weight[i][2].weightfn && !h->sh.weight[i][1].weightfn )
1701             {
1702                 h->sh.weight[i][1].i_scale = 1 << h->sh.weight[0][1].i_denom;
1703                 h->sh.weight[i][1].i_offset = 0;
1704             }
1705         }
1706
1707     if( !weightplane[0] )
1708         h->sh.weight[0][0].i_denom = 0;
1709     if( !weightplane[1] )
1710         h->sh.weight[0][1].i_denom = 0;
1711     h->sh.weight[0][2].i_denom = h->sh.weight[0][1].i_denom;
1712 }
1713
1714 static inline int x264_reference_distance( x264_t *h, x264_frame_t *frame )
1715 {
1716     if( h->param.i_frame_packing == 5 )
1717         return abs((h->fenc->i_frame&~1) - (frame->i_frame&~1)) +
1718                   ((h->fenc->i_frame&1) != (frame->i_frame&1));
1719     else
1720         return abs(h->fenc->i_frame - frame->i_frame);
1721 }
1722
1723 static inline void x264_reference_build_list( x264_t *h, int i_poc )
1724 {
1725     int b_ok;
1726
1727     /* build ref list 0/1 */
1728     h->mb.pic.i_fref[0] = h->i_ref[0] = 0;
1729     h->mb.pic.i_fref[1] = h->i_ref[1] = 0;
1730     if( h->sh.i_type == SLICE_TYPE_I )
1731         return;
1732
1733     for( int i = 0; h->frames.reference[i]; i++ )
1734     {
1735         if( h->frames.reference[i]->b_corrupt )
1736             continue;
1737         if( h->frames.reference[i]->i_poc < i_poc )
1738             h->fref[0][h->i_ref[0]++] = h->frames.reference[i];
1739         else if( h->frames.reference[i]->i_poc > i_poc )
1740             h->fref[1][h->i_ref[1]++] = h->frames.reference[i];
1741     }
1742
1743     /* Order reference lists by distance from the current frame. */
1744     for( int list = 0; list < 2; list++ )
1745     {
1746         h->fref_nearest[list] = h->fref[list][0];
1747         do
1748         {
1749             b_ok = 1;
1750             for( int i = 0; i < h->i_ref[list] - 1; i++ )
1751             {
1752                 if( list ? h->fref[list][i+1]->i_poc < h->fref_nearest[list]->i_poc
1753                          : h->fref[list][i+1]->i_poc > h->fref_nearest[list]->i_poc )
1754                     h->fref_nearest[list] = h->fref[list][i+1];
1755                 if( x264_reference_distance( h, h->fref[list][i] ) > x264_reference_distance( h, h->fref[list][i+1] ) )
1756                 {
1757                     XCHG( x264_frame_t*, h->fref[list][i], h->fref[list][i+1] );
1758                     b_ok = 0;
1759                     break;
1760                 }
1761             }
1762         } while( !b_ok );
1763     }
1764
1765     if( h->sh.i_mmco_remove_from_end )
1766         for( int i = h->i_ref[0]-1; i >= h->i_ref[0] - h->sh.i_mmco_remove_from_end; i-- )
1767         {
1768             int diff = h->i_frame_num - h->fref[0][i]->i_frame_num;
1769             h->sh.mmco[h->sh.i_mmco_command_count].i_poc = h->fref[0][i]->i_poc;
1770             h->sh.mmco[h->sh.i_mmco_command_count++].i_difference_of_pic_nums = diff;
1771         }
1772
1773     x264_reference_check_reorder( h );
1774
1775     h->i_ref[1] = X264_MIN( h->i_ref[1], h->frames.i_max_ref1 );
1776     h->i_ref[0] = X264_MIN( h->i_ref[0], h->frames.i_max_ref0 );
1777     h->i_ref[0] = X264_MIN( h->i_ref[0], h->param.i_frame_reference ); // if reconfig() has lowered the limit
1778
1779     /* For Blu-ray compliance, don't reference frames outside of the minigop. */
1780     if( IS_X264_TYPE_B( h->fenc->i_type ) && h->param.b_bluray_compat )
1781         h->i_ref[0] = X264_MIN( h->i_ref[0], IS_X264_TYPE_B( h->fref[0][0]->i_type ) + 1 );
1782
1783     /* add duplicates */
1784     if( h->fenc->i_type == X264_TYPE_P )
1785     {
1786         int idx = -1;
1787         if( h->param.analyse.i_weighted_pred >= X264_WEIGHTP_SIMPLE )
1788         {
1789             x264_weight_t w[3];
1790             w[1].weightfn = w[2].weightfn = NULL;
1791             if( h->param.rc.b_stat_read )
1792                 x264_ratecontrol_set_weights( h, h->fenc );
1793
1794             if( !h->fenc->weight[0][0].weightfn )
1795             {
1796                 h->fenc->weight[0][0].i_denom = 0;
1797                 SET_WEIGHT( w[0], 1, 1, 0, -1 );
1798                 idx = x264_weighted_reference_duplicate( h, 0, w );
1799             }
1800             else
1801             {
1802                 if( h->fenc->weight[0][0].i_scale == 1<<h->fenc->weight[0][0].i_denom )
1803                 {
1804                     SET_WEIGHT( h->fenc->weight[0][0], 1, 1, 0, h->fenc->weight[0][0].i_offset );
1805                 }
1806                 x264_weighted_reference_duplicate( h, 0, x264_weight_none );
1807                 if( h->fenc->weight[0][0].i_offset > -128 )
1808                 {
1809                     w[0] = h->fenc->weight[0][0];
1810                     w[0].i_offset--;
1811                     h->mc.weight_cache( h, &w[0] );
1812                     idx = x264_weighted_reference_duplicate( h, 0, w );
1813                 }
1814             }
1815         }
1816         h->mb.ref_blind_dupe = idx;
1817     }
1818
1819     assert( h->i_ref[0] + h->i_ref[1] <= X264_REF_MAX );
1820     h->mb.pic.i_fref[0] = h->i_ref[0];
1821     h->mb.pic.i_fref[1] = h->i_ref[1];
1822 }
1823
1824 static void x264_fdec_filter_row( x264_t *h, int mb_y, int b_inloop )
1825 {
1826     /* mb_y is the mb to be encoded next, not the mb to be filtered here */
1827     int b_hpel = h->fdec->b_kept_as_ref;
1828     int b_deblock = h->sh.i_disable_deblocking_filter_idc != 1;
1829     int b_end = mb_y == h->i_threadslice_end;
1830     int b_measure_quality = 1;
1831     int min_y = mb_y - (1 << SLICE_MBAFF);
1832     int b_start = min_y == h->i_threadslice_start;
1833     /* Even in interlaced mode, deblocking never modifies more than 4 pixels
1834      * above each MB, as bS=4 doesn't happen for the top of interlaced mbpairs. */
1835     int minpix_y = min_y*16 - 4 * !b_start;
1836     int maxpix_y = mb_y*16 - 4 * !b_end;
1837     b_deblock &= b_hpel || h->param.psz_dump_yuv;
1838     if( h->param.b_sliced_threads && b_start && min_y && !b_inloop )
1839     {
1840         b_deblock = 0;         /* We already deblocked on the inloop pass. */
1841         b_measure_quality = 0; /* We already measured quality on the inloop pass. */
1842     }
1843     if( mb_y & SLICE_MBAFF )
1844         return;
1845     if( min_y < h->i_threadslice_start )
1846         return;
1847
1848     if( b_deblock )
1849         for( int y = min_y; y < mb_y; y += (1 << SLICE_MBAFF) )
1850             x264_frame_deblock_row( h, y );
1851
1852     /* FIXME: Prediction requires different borders for interlaced/progressive mc,
1853      * but the actual image data is equivalent. For now, maintain this
1854      * consistency by copying deblocked pixels between planes. */
1855     if( PARAM_INTERLACED )
1856         for( int p = 0; p < h->fdec->i_plane; p++ )
1857             for( int i = minpix_y>>(CHROMA_V_SHIFT && p); i < maxpix_y>>(CHROMA_V_SHIFT && p); i++ )
1858                 memcpy( h->fdec->plane_fld[p] + i*h->fdec->i_stride[p],
1859                         h->fdec->plane[p]     + i*h->fdec->i_stride[p],
1860                         h->mb.i_mb_width*16*sizeof(pixel) );
1861
1862     if( b_hpel )
1863     {
1864         int end = mb_y == h->mb.i_mb_height;
1865         x264_frame_expand_border( h, h->fdec, min_y, end );
1866         if( h->param.analyse.i_subpel_refine )
1867         {
1868             x264_frame_filter( h, h->fdec, min_y, end );
1869             x264_frame_expand_border_filtered( h, h->fdec, min_y, end );
1870         }
1871     }
1872
1873     if( SLICE_MBAFF )
1874         for( int i = 0; i < 3; i++ )
1875         {
1876             XCHG( pixel *, h->intra_border_backup[0][i], h->intra_border_backup[3][i] );
1877             XCHG( pixel *, h->intra_border_backup[1][i], h->intra_border_backup[4][i] );
1878         }
1879
1880     if( h->i_thread_frames > 1 && h->fdec->b_kept_as_ref )
1881         x264_frame_cond_broadcast( h->fdec, mb_y*16 + (b_end ? 10000 : -(X264_THREAD_HEIGHT << SLICE_MBAFF)) );
1882
1883     if( b_measure_quality )
1884     {
1885         maxpix_y = X264_MIN( maxpix_y, h->param.i_height );
1886         if( h->param.analyse.b_psnr )
1887         {
1888             for( int p = 0; p < (CHROMA444 ? 3 : 1); p++ )
1889                 h->stat.frame.i_ssd[p] += x264_pixel_ssd_wxh( &h->pixf,
1890                     h->fdec->plane[p] + minpix_y * h->fdec->i_stride[p], h->fdec->i_stride[p],
1891                     h->fenc->plane[p] + minpix_y * h->fenc->i_stride[p], h->fenc->i_stride[p],
1892                     h->param.i_width, maxpix_y-minpix_y );
1893             if( !CHROMA444 )
1894             {
1895                 uint64_t ssd_u, ssd_v;
1896                 int v_shift = CHROMA_V_SHIFT;
1897                 x264_pixel_ssd_nv12( &h->pixf,
1898                     h->fdec->plane[1] + (minpix_y>>v_shift) * h->fdec->i_stride[1], h->fdec->i_stride[1],
1899                     h->fenc->plane[1] + (minpix_y>>v_shift) * h->fenc->i_stride[1], h->fenc->i_stride[1],
1900                     h->param.i_width>>1, (maxpix_y-minpix_y)>>v_shift, &ssd_u, &ssd_v );
1901                 h->stat.frame.i_ssd[1] += ssd_u;
1902                 h->stat.frame.i_ssd[2] += ssd_v;
1903             }
1904         }
1905
1906         if( h->param.analyse.b_ssim )
1907         {
1908             int ssim_cnt;
1909             x264_emms();
1910             /* offset by 2 pixels to avoid alignment of ssim blocks with dct blocks,
1911              * and overlap by 4 */
1912             minpix_y += b_start ? 2 : -6;
1913             h->stat.frame.f_ssim +=
1914                 x264_pixel_ssim_wxh( &h->pixf,
1915                     h->fdec->plane[0] + 2+minpix_y*h->fdec->i_stride[0], h->fdec->i_stride[0],
1916                     h->fenc->plane[0] + 2+minpix_y*h->fenc->i_stride[0], h->fenc->i_stride[0],
1917                     h->param.i_width-2, maxpix_y-minpix_y, h->scratch_buffer, &ssim_cnt );
1918             h->stat.frame.i_ssim_cnt += ssim_cnt;
1919         }
1920     }
1921 }
1922
1923 static inline int x264_reference_update( x264_t *h )
1924 {
1925     if( !h->fdec->b_kept_as_ref )
1926     {
1927         if( h->i_thread_frames > 1 )
1928         {
1929             x264_frame_push_unused( h, h->fdec );
1930             h->fdec = x264_frame_pop_unused( h, 1 );
1931             if( !h->fdec )
1932                 return -1;
1933         }
1934         return 0;
1935     }
1936
1937     /* apply mmco from previous frame. */
1938     for( int i = 0; i < h->sh.i_mmco_command_count; i++ )
1939         for( int j = 0; h->frames.reference[j]; j++ )
1940             if( h->frames.reference[j]->i_poc == h->sh.mmco[i].i_poc )
1941                 x264_frame_push_unused( h, x264_frame_shift( &h->frames.reference[j] ) );
1942
1943     /* move frame in the buffer */
1944     x264_frame_push( h->frames.reference, h->fdec );
1945     if( h->frames.reference[h->sps->i_num_ref_frames] )
1946         x264_frame_push_unused( h, x264_frame_shift( h->frames.reference ) );
1947     h->fdec = x264_frame_pop_unused( h, 1 );
1948     if( !h->fdec )
1949         return -1;
1950     return 0;
1951 }
1952
1953 static inline void x264_reference_reset( x264_t *h )
1954 {
1955     while( h->frames.reference[0] )
1956         x264_frame_push_unused( h, x264_frame_pop( h->frames.reference ) );
1957     h->fdec->i_poc =
1958     h->fenc->i_poc = 0;
1959 }
1960
1961 static inline void x264_reference_hierarchy_reset( x264_t *h )
1962 {
1963     int ref;
1964     int b_hasdelayframe = 0;
1965
1966     /* look for delay frames -- chain must only contain frames that are disposable */
1967     for( int i = 0; h->frames.current[i] && IS_DISPOSABLE( h->frames.current[i]->i_type ); i++ )
1968         b_hasdelayframe |= h->frames.current[i]->i_coded
1969                         != h->frames.current[i]->i_frame + h->sps->vui.i_num_reorder_frames;
1970
1971     /* This function must handle b-pyramid and clear frames for open-gop */
1972     if( h->param.i_bframe_pyramid != X264_B_PYRAMID_STRICT && !b_hasdelayframe && h->frames.i_poc_last_open_gop == -1 )
1973         return;
1974
1975     /* Remove last BREF. There will never be old BREFs in the
1976      * dpb during a BREF decode when pyramid == STRICT */
1977     for( ref = 0; h->frames.reference[ref]; ref++ )
1978     {
1979         if( ( h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT
1980             && h->frames.reference[ref]->i_type == X264_TYPE_BREF )
1981             || ( h->frames.reference[ref]->i_poc < h->frames.i_poc_last_open_gop
1982             && h->sh.i_type != SLICE_TYPE_B ) )
1983         {
1984             int diff = h->i_frame_num - h->frames.reference[ref]->i_frame_num;
1985             h->sh.mmco[h->sh.i_mmco_command_count].i_difference_of_pic_nums = diff;
1986             h->sh.mmco[h->sh.i_mmco_command_count++].i_poc = h->frames.reference[ref]->i_poc;
1987             x264_frame_push_unused( h, x264_frame_shift( &h->frames.reference[ref] ) );
1988             h->b_ref_reorder[0] = 1;
1989             ref--;
1990         }
1991     }
1992
1993     /* Prepare room in the dpb for the delayed display time of the later b-frame's */
1994     if( h->param.i_bframe_pyramid )
1995         h->sh.i_mmco_remove_from_end = X264_MAX( ref + 2 - h->frames.i_max_dpb, 0 );
1996 }
1997
1998 static inline void x264_slice_init( x264_t *h, int i_nal_type, int i_global_qp )
1999 {
2000     /* ------------------------ Create slice header  ----------------------- */
2001     if( i_nal_type == NAL_SLICE_IDR )
2002     {
2003         x264_slice_header_init( h, &h->sh, h->sps, h->pps, h->i_idr_pic_id, h->i_frame_num, i_global_qp );
2004
2005         /* alternate id */
2006         h->i_idr_pic_id ^= 1;
2007     }
2008     else
2009     {
2010         x264_slice_header_init( h, &h->sh, h->sps, h->pps, -1, h->i_frame_num, i_global_qp );
2011
2012         h->sh.i_num_ref_idx_l0_active = h->i_ref[0] <= 0 ? 1 : h->i_ref[0];
2013         h->sh.i_num_ref_idx_l1_active = h->i_ref[1] <= 0 ? 1 : h->i_ref[1];
2014         if( h->sh.i_num_ref_idx_l0_active != h->pps->i_num_ref_idx_l0_default_active ||
2015             (h->sh.i_type == SLICE_TYPE_B && h->sh.i_num_ref_idx_l1_active != h->pps->i_num_ref_idx_l1_default_active) )
2016         {
2017             h->sh.b_num_ref_idx_override = 1;
2018         }
2019     }
2020
2021     if( h->fenc->i_type == X264_TYPE_BREF && h->param.b_bluray_compat && h->sh.i_mmco_command_count )
2022     {
2023         h->b_sh_backup = 1;
2024         h->sh_backup = h->sh;
2025     }
2026
2027     h->fdec->i_frame_num = h->sh.i_frame_num;
2028
2029     if( h->sps->i_poc_type == 0 )
2030     {
2031         h->sh.i_poc = h->fdec->i_poc;
2032         if( PARAM_INTERLACED )
2033         {
2034             h->sh.i_delta_poc_bottom = h->param.b_tff ? 1 : -1;
2035             h->sh.i_poc += h->sh.i_delta_poc_bottom == -1;
2036         }
2037         else
2038             h->sh.i_delta_poc_bottom = 0;
2039         h->fdec->i_delta_poc[0] = h->sh.i_delta_poc_bottom == -1;
2040         h->fdec->i_delta_poc[1] = h->sh.i_delta_poc_bottom ==  1;
2041     }
2042     else
2043     {
2044         /* Nothing to do ? */
2045     }
2046
2047     x264_macroblock_slice_init( h );
2048 }
2049
2050 static int x264_slice_write( x264_t *h )
2051 {
2052     int i_skip;
2053     int mb_xy, i_mb_x, i_mb_y;
2054     int i_skip_bak = 0; /* Shut up GCC. */
2055     bs_t UNINIT(bs_bak);
2056     x264_cabac_t cabac_bak;
2057     uint8_t cabac_prevbyte_bak = 0; /* Shut up GCC. */
2058     int mv_bits_bak = 0;
2059     int tex_bits_bak = 0;
2060     /* NALUs other than the first use a 3-byte startcode.
2061      * Add one extra byte for the rbsp, and one more for the final CABAC putbyte.
2062      * Then add an extra 5 bytes just in case, to account for random NAL escapes and
2063      * other inaccuracies. */
2064     int overhead_guess = (NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal)) + 1 + h->param.b_cabac + 5;
2065     int slice_max_size = h->param.i_slice_max_size > 0 ? (h->param.i_slice_max_size-overhead_guess)*8 : 0;
2066     int back_up_bitstream = slice_max_size || (!h->param.b_cabac && h->sps->i_profile_idc < PROFILE_HIGH);
2067     int starting_bits = bs_pos(&h->out.bs);
2068     int b_deblock = h->sh.i_disable_deblocking_filter_idc != 1;
2069     int b_hpel = h->fdec->b_kept_as_ref;
2070     uint8_t *last_emu_check;
2071     b_deblock &= b_hpel || h->param.psz_dump_yuv;
2072     bs_realign( &h->out.bs );
2073
2074     /* Slice */
2075     x264_nal_start( h, h->i_nal_type, h->i_nal_ref_idc );
2076     h->out.nal[h->out.i_nal].i_first_mb = h->sh.i_first_mb;
2077
2078     /* Slice header */
2079     x264_macroblock_thread_init( h );
2080
2081     /* If this isn't the first slice in the threadslice, set the slice QP
2082      * equal to the last QP in the previous slice for more accurate
2083      * CABAC initialization. */
2084     if( h->sh.i_first_mb != h->i_threadslice_start * h->mb.i_mb_width )
2085     {
2086         h->sh.i_qp = h->mb.i_last_qp;
2087         h->sh.i_qp_delta = h->sh.i_qp - h->pps->i_pic_init_qp;
2088     }
2089
2090     x264_slice_header_write( &h->out.bs, &h->sh, h->i_nal_ref_idc );
2091     if( h->param.b_cabac )
2092     {
2093         /* alignment needed */
2094         bs_align_1( &h->out.bs );
2095
2096         /* init cabac */
2097         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 );
2098         x264_cabac_encode_init ( &h->cabac, h->out.bs.p, h->out.bs.p_end );
2099         last_emu_check = h->cabac.p;
2100     }
2101     else
2102         last_emu_check = h->out.bs.p;
2103     h->mb.i_last_qp = h->sh.i_qp;
2104     h->mb.i_last_dqp = 0;
2105     h->mb.field_decoding_flag = 0;
2106
2107     i_mb_y = h->sh.i_first_mb / h->mb.i_mb_width;
2108     i_mb_x = h->sh.i_first_mb % h->mb.i_mb_width;
2109     i_skip = 0;
2110
2111     while( 1 )
2112     {
2113         mb_xy = i_mb_x + i_mb_y * h->mb.i_mb_width;
2114         int mb_spos = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
2115
2116         if( !(i_mb_y & SLICE_MBAFF) )
2117         {
2118             if( x264_bitstream_check_buffer( h ) )
2119                 return -1;
2120
2121             if( back_up_bitstream )
2122             {
2123                 mv_bits_bak = h->stat.frame.i_mv_bits;
2124                 tex_bits_bak = h->stat.frame.i_tex_bits;
2125                 /* We don't need the contexts because flushing the CABAC encoder has no context
2126                  * dependency and macroblocks are only re-encoded in the case where a slice is
2127                  * ended (and thus the content of all contexts are thrown away). */
2128                 if( h->param.b_cabac )
2129                 {
2130                     memcpy( &cabac_bak, &h->cabac, offsetof(x264_cabac_t, f8_bits_encoded) );
2131                     /* x264's CABAC writer modifies the previous byte during carry, so it has to be
2132                      * backed up. */
2133                     cabac_prevbyte_bak = h->cabac.p[-1];
2134                 }
2135                 else
2136                 {
2137                     bs_bak = h->out.bs;
2138                     i_skip_bak = i_skip;
2139                 }
2140             }
2141         }
2142
2143         if( i_mb_x == 0 && !h->mb.b_reencode_mb )
2144             x264_fdec_filter_row( h, i_mb_y, 1 );
2145
2146         if( PARAM_INTERLACED )
2147         {
2148             if( h->mb.b_adaptive_mbaff )
2149             {
2150                 if( !(i_mb_y&1) )
2151                 {
2152                     /* FIXME: VSAD is fast but fairly poor at choosing the best interlace type. */
2153                     h->mb.b_interlaced = x264_field_vsad( h, i_mb_x, i_mb_y );
2154                     memcpy( &h->zigzagf, MB_INTERLACED ? &h->zigzagf_interlaced : &h->zigzagf_progressive, sizeof(h->zigzagf) );
2155                     if( !MB_INTERLACED && (i_mb_y+2) == h->mb.i_mb_height )
2156                         x264_expand_border_mbpair( h, i_mb_x, i_mb_y );
2157                 }
2158             }
2159             h->mb.field[mb_xy] = MB_INTERLACED;
2160         }
2161
2162         /* load cache */
2163         if( SLICE_MBAFF )
2164             x264_macroblock_cache_load_interlaced( h, i_mb_x, i_mb_y );
2165         else
2166             x264_macroblock_cache_load_progressive( h, i_mb_x, i_mb_y );
2167
2168         x264_macroblock_analyse( h );
2169
2170         /* encode this macroblock -> be careful it can change the mb type to P_SKIP if needed */
2171 reencode:
2172         x264_macroblock_encode( h );
2173
2174         if( h->param.b_cabac )
2175         {
2176             if( mb_xy > h->sh.i_first_mb && !(SLICE_MBAFF && (i_mb_y&1)) )
2177                 x264_cabac_encode_terminal( &h->cabac );
2178
2179             if( IS_SKIP( h->mb.i_type ) )
2180                 x264_cabac_mb_skip( h, 1 );
2181             else
2182             {
2183                 if( h->sh.i_type != SLICE_TYPE_I )
2184                     x264_cabac_mb_skip( h, 0 );
2185                 x264_macroblock_write_cabac( h, &h->cabac );
2186             }
2187         }
2188         else
2189         {
2190             if( IS_SKIP( h->mb.i_type ) )
2191                 i_skip++;
2192             else
2193             {
2194                 if( h->sh.i_type != SLICE_TYPE_I )
2195                 {
2196                     bs_write_ue( &h->out.bs, i_skip );  /* skip run */
2197                     i_skip = 0;
2198                 }
2199                 x264_macroblock_write_cavlc( h );
2200                 /* If there was a CAVLC level code overflow, try again at a higher QP. */
2201                 if( h->mb.b_overflow )
2202                 {
2203                     h->mb.i_chroma_qp = h->chroma_qp_table[++h->mb.i_qp];
2204                     h->mb.i_skip_intra = 0;
2205                     h->mb.b_skip_mc = 0;
2206                     h->mb.b_overflow = 0;
2207                     h->out.bs = bs_bak;
2208                     i_skip = i_skip_bak;
2209                     h->stat.frame.i_mv_bits = mv_bits_bak;
2210                     h->stat.frame.i_tex_bits = tex_bits_bak;
2211                     goto reencode;
2212                 }
2213             }
2214         }
2215
2216         int total_bits = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
2217         int mb_size = total_bits - mb_spos;
2218
2219         if( slice_max_size )
2220         {
2221             /* Count the skip run, just in case. */
2222             if( !h->param.b_cabac )
2223                 total_bits += bs_size_ue_big( i_skip );
2224             /* Check for escape bytes. */
2225             uint8_t *end = h->param.b_cabac ? h->cabac.p : h->out.bs.p;
2226             for( ; last_emu_check < end - 2; last_emu_check++ )
2227                 if( last_emu_check[0] == 0 && last_emu_check[1] == 0 && last_emu_check[2] <= 3 )
2228                 {
2229                     slice_max_size -= 8;
2230                     last_emu_check++;
2231                 }
2232             /* We'll just re-encode this last macroblock if we go over the max slice size. */
2233             if( total_bits - starting_bits > slice_max_size && !h->mb.b_reencode_mb )
2234             {
2235                 if( mb_xy != h->sh.i_first_mb )
2236                 {
2237                     h->stat.frame.i_mv_bits = mv_bits_bak;
2238                     h->stat.frame.i_tex_bits = tex_bits_bak;
2239                     if( h->param.b_cabac )
2240                     {
2241                         memcpy( &h->cabac, &cabac_bak, offsetof(x264_cabac_t, f8_bits_encoded) );
2242                         h->cabac.p[-1] = cabac_prevbyte_bak;
2243                     }
2244                     else
2245                     {
2246                         h->out.bs = bs_bak;
2247                         i_skip = i_skip_bak;
2248                     }
2249                     h->mb.b_reencode_mb = 1;
2250                     if( SLICE_MBAFF )
2251                     {
2252                         // set to bottom of previous mbpair
2253                         if( i_mb_x )
2254                             h->sh.i_last_mb = mb_xy-1+h->mb.i_mb_stride*(!(i_mb_y&1));
2255                         else
2256                             h->sh.i_last_mb = (i_mb_y-2+!(i_mb_y&1))*h->mb.i_mb_stride + h->mb.i_mb_width - 1;
2257                     }
2258                     else
2259                         h->sh.i_last_mb = mb_xy-1;
2260                     break;
2261                 }
2262                 else
2263                 {
2264                     h->sh.i_last_mb = mb_xy;
2265                     h->mb.b_reencode_mb = 0;
2266                 }
2267             }
2268             else
2269                 h->mb.b_reencode_mb = 0;
2270         }
2271
2272 #if HAVE_VISUALIZE
2273         if( h->param.b_visualize )
2274             x264_visualize_mb( h );
2275 #endif
2276
2277         /* save cache */
2278         x264_macroblock_cache_save( h );
2279
2280         /* accumulate mb stats */
2281         h->stat.frame.i_mb_count[h->mb.i_type]++;
2282
2283         int b_intra = IS_INTRA( h->mb.i_type );
2284         int b_skip = IS_SKIP( h->mb.i_type );
2285         if( h->param.i_log_level >= X264_LOG_INFO || h->param.rc.b_stat_write )
2286         {
2287             if( !b_intra && !b_skip && !IS_DIRECT( h->mb.i_type ) )
2288             {
2289                 if( h->mb.i_partition != D_8x8 )
2290                         h->stat.frame.i_mb_partition[h->mb.i_partition] += 4;
2291                     else
2292                         for( int i = 0; i < 4; i++ )
2293                             h->stat.frame.i_mb_partition[h->mb.i_sub_partition[i]] ++;
2294                 if( h->param.i_frame_reference > 1 )
2295                     for( int i_list = 0; i_list <= (h->sh.i_type == SLICE_TYPE_B); i_list++ )
2296                         for( int i = 0; i < 4; i++ )
2297                         {
2298                             int i_ref = h->mb.cache.ref[i_list][ x264_scan8[4*i] ];
2299                             if( i_ref >= 0 )
2300                                 h->stat.frame.i_mb_count_ref[i_list][i_ref] ++;
2301                         }
2302             }
2303         }
2304
2305         if( h->param.i_log_level >= X264_LOG_INFO )
2306         {
2307             if( h->mb.i_cbp_luma | h->mb.i_cbp_chroma )
2308             {
2309                 if( CHROMA444 )
2310                 {
2311                     for( int i = 0; i < 4; i++ )
2312                         if( h->mb.i_cbp_luma & (1 << i) )
2313                             for( int p = 0; p < 3; p++ )
2314                             {
2315                                 int s8 = i*4+p*16;
2316                                 int nnz8x8 = M16( &h->mb.cache.non_zero_count[x264_scan8[s8]+0] )
2317                                            | M16( &h->mb.cache.non_zero_count[x264_scan8[s8]+8] );
2318                                 h->stat.frame.i_mb_cbp[!b_intra + p*2] += !!nnz8x8;
2319                             }
2320                 }
2321                 else
2322                 {
2323                     int cbpsum = (h->mb.i_cbp_luma&1) + ((h->mb.i_cbp_luma>>1)&1)
2324                                + ((h->mb.i_cbp_luma>>2)&1) + (h->mb.i_cbp_luma>>3);
2325                     h->stat.frame.i_mb_cbp[!b_intra + 0] += cbpsum;
2326                     h->stat.frame.i_mb_cbp[!b_intra + 2] += !!h->mb.i_cbp_chroma;
2327                     h->stat.frame.i_mb_cbp[!b_intra + 4] += h->mb.i_cbp_chroma >> 1;
2328                 }
2329             }
2330             if( h->mb.i_cbp_luma && !b_intra )
2331             {
2332                 h->stat.frame.i_mb_count_8x8dct[0] ++;
2333                 h->stat.frame.i_mb_count_8x8dct[1] += h->mb.b_transform_8x8;
2334             }
2335             if( b_intra && h->mb.i_type != I_PCM )
2336             {
2337                 if( h->mb.i_type == I_16x16 )
2338                     h->stat.frame.i_mb_pred_mode[0][h->mb.i_intra16x16_pred_mode]++;
2339                 else if( h->mb.i_type == I_8x8 )
2340                     for( int i = 0; i < 16; i += 4 )
2341                         h->stat.frame.i_mb_pred_mode[1][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
2342                 else //if( h->mb.i_type == I_4x4 )
2343                     for( int i = 0; i < 16; i++ )
2344                         h->stat.frame.i_mb_pred_mode[2][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
2345                 h->stat.frame.i_mb_pred_mode[3][x264_mb_chroma_pred_mode_fix[h->mb.i_chroma_pred_mode]]++;
2346             }
2347             h->stat.frame.i_mb_field[b_intra?0:b_skip?2:1] += MB_INTERLACED;
2348         }
2349
2350         /* calculate deblock strength values (actual deblocking is done per-row along with hpel) */
2351         if( b_deblock )
2352             x264_macroblock_deblock_strength( h );
2353
2354         x264_ratecontrol_mb( h, mb_size );
2355
2356         if( mb_xy == h->sh.i_last_mb )
2357             break;
2358
2359         if( SLICE_MBAFF )
2360         {
2361             i_mb_x += i_mb_y & 1;
2362             i_mb_y ^= i_mb_x < h->mb.i_mb_width;
2363         }
2364         else
2365             i_mb_x++;
2366         if( i_mb_x == h->mb.i_mb_width )
2367         {
2368             i_mb_y++;
2369             i_mb_x = 0;
2370         }
2371     }
2372     h->out.nal[h->out.i_nal].i_last_mb = h->sh.i_last_mb;
2373
2374     if( h->param.b_cabac )
2375     {
2376         x264_cabac_encode_flush( h, &h->cabac );
2377         h->out.bs.p = h->cabac.p;
2378     }
2379     else
2380     {
2381         if( i_skip > 0 )
2382             bs_write_ue( &h->out.bs, i_skip );  /* last skip run */
2383         /* rbsp_slice_trailing_bits */
2384         bs_rbsp_trailing( &h->out.bs );
2385         bs_flush( &h->out.bs );
2386     }
2387     if( x264_nal_end( h ) )
2388         return -1;
2389
2390     if( h->sh.i_last_mb == (h->i_threadslice_end * h->mb.i_mb_width - 1) )
2391     {
2392         h->stat.frame.i_misc_bits = bs_pos( &h->out.bs )
2393                                   + (h->out.i_nal*NALU_OVERHEAD * 8)
2394                                   - h->stat.frame.i_tex_bits
2395                                   - h->stat.frame.i_mv_bits;
2396         x264_fdec_filter_row( h, h->i_threadslice_end, 1 );
2397     }
2398
2399     return 0;
2400 }
2401
2402 static void x264_thread_sync_context( x264_t *dst, x264_t *src )
2403 {
2404     if( dst == src )
2405         return;
2406
2407     // reference counting
2408     for( x264_frame_t **f = src->frames.reference; *f; f++ )
2409         (*f)->i_reference_count++;
2410     for( x264_frame_t **f = dst->frames.reference; *f; f++ )
2411         x264_frame_push_unused( src, *f );
2412     src->fdec->i_reference_count++;
2413     x264_frame_push_unused( src, dst->fdec );
2414
2415     // copy everything except the per-thread pointers and the constants.
2416     memcpy( &dst->i_frame, &src->i_frame, offsetof(x264_t, mb.type) - offsetof(x264_t, i_frame) );
2417     dst->param = src->param;
2418     dst->stat = src->stat;
2419     dst->pixf = src->pixf;
2420 }
2421
2422 static void x264_thread_sync_stat( x264_t *dst, x264_t *src )
2423 {
2424     if( dst == src )
2425         return;
2426     memcpy( &dst->stat.i_frame_count, &src->stat.i_frame_count, sizeof(dst->stat) - sizeof(dst->stat.frame) );
2427 }
2428
2429 static void *x264_slices_write( x264_t *h )
2430 {
2431     int i_slice_num = 0;
2432     int last_thread_mb = h->sh.i_last_mb;
2433
2434 #if HAVE_VISUALIZE
2435     if( h->param.b_visualize )
2436         if( x264_visualize_init( h ) )
2437             return (void *)-1;
2438 #endif
2439
2440     /* init stats */
2441     memset( &h->stat.frame, 0, sizeof(h->stat.frame) );
2442     h->mb.b_reencode_mb = 0;
2443     while( h->sh.i_first_mb + SLICE_MBAFF*h->mb.i_mb_stride <= last_thread_mb )
2444     {
2445         h->sh.i_last_mb = last_thread_mb;
2446         if( h->param.i_slice_max_mbs )
2447         {
2448             if( SLICE_MBAFF )
2449             {
2450                 // convert first to mbaff form, add slice-max-mbs, then convert back to normal form
2451                 int last_mbaff = 2*(h->sh.i_first_mb % h->mb.i_mb_width)
2452                     + h->mb.i_mb_width*(h->sh.i_first_mb / h->mb.i_mb_width)
2453                     + h->param.i_slice_max_mbs - 1;
2454                 int last_x = (last_mbaff % (2*h->mb.i_mb_width))/2;
2455                 int last_y = (last_mbaff / (2*h->mb.i_mb_width))*2 + 1;
2456                 h->sh.i_last_mb = last_x + h->mb.i_mb_stride*last_y;
2457             }
2458             else
2459                 h->sh.i_last_mb = h->sh.i_first_mb + h->param.i_slice_max_mbs - 1;
2460         }
2461         else if( h->param.i_slice_count && !h->param.b_sliced_threads )
2462         {
2463             int height = h->mb.i_mb_height >> PARAM_INTERLACED;
2464             int width = h->mb.i_mb_width << PARAM_INTERLACED;
2465             i_slice_num++;
2466             h->sh.i_last_mb = (height * i_slice_num + h->param.i_slice_count/2) / h->param.i_slice_count * width - 1;
2467         }
2468         h->sh.i_last_mb = X264_MIN( h->sh.i_last_mb, last_thread_mb );
2469         if( x264_stack_align( x264_slice_write, h ) )
2470             return (void *)-1;
2471         h->sh.i_first_mb = h->sh.i_last_mb + 1;
2472         // if i_first_mb is not the last mb in a row then go to the next mb in MBAFF order
2473         if( SLICE_MBAFF && h->sh.i_first_mb % h->mb.i_mb_width )
2474             h->sh.i_first_mb -= h->mb.i_mb_stride;
2475     }
2476
2477 #if HAVE_VISUALIZE
2478     if( h->param.b_visualize )
2479     {
2480         x264_visualize_show( h );
2481         x264_visualize_close( h );
2482     }
2483 #endif
2484
2485     return (void *)0;
2486 }
2487
2488 static int x264_threaded_slices_write( x264_t *h )
2489 {
2490     /* set first/last mb and sync contexts */
2491     for( int i = 0; i < h->param.i_threads; i++ )
2492     {
2493         x264_t *t = h->thread[i];
2494         if( i )
2495         {
2496             t->param = h->param;
2497             memcpy( &t->i_frame, &h->i_frame, offsetof(x264_t, rc) - offsetof(x264_t, i_frame) );
2498         }
2499         int height = h->mb.i_mb_height >> PARAM_INTERLACED;
2500         t->i_threadslice_start = ((height *  i    + h->param.i_slice_count/2) / h->param.i_threads) << PARAM_INTERLACED;
2501         t->i_threadslice_end   = ((height * (i+1) + h->param.i_slice_count/2) / h->param.i_threads) << PARAM_INTERLACED;
2502         t->sh.i_first_mb = t->i_threadslice_start * h->mb.i_mb_width;
2503         t->sh.i_last_mb  =   t->i_threadslice_end * h->mb.i_mb_width - 1;
2504     }
2505
2506     x264_stack_align( x264_analyse_weight_frame, h, h->mb.i_mb_height*16 + 16 );
2507
2508     x264_threads_distribute_ratecontrol( h );
2509
2510     /* dispatch */
2511     for( int i = 0; i < h->param.i_threads; i++ )
2512     {
2513         x264_threadpool_run( h->threadpool, (void*)x264_slices_write, h->thread[i] );
2514         h->thread[i]->b_thread_active = 1;
2515     }
2516     for( int i = 0; i < h->param.i_threads; i++ )
2517     {
2518         h->thread[i]->b_thread_active = 0;
2519         if( (intptr_t)x264_threadpool_wait( h->threadpool, h->thread[i] ) )
2520             return -1;
2521     }
2522
2523     /* Go back and fix up the hpel on the borders between slices. */
2524     for( int i = 1; i < h->param.i_threads; i++ )
2525     {
2526         x264_fdec_filter_row( h->thread[i], h->thread[i]->i_threadslice_start + 1, 0 );
2527         if( SLICE_MBAFF )
2528             x264_fdec_filter_row( h->thread[i], h->thread[i]->i_threadslice_start + 2, 0 );
2529     }
2530
2531     x264_threads_merge_ratecontrol( h );
2532
2533     for( int i = 1; i < h->param.i_threads; i++ )
2534     {
2535         x264_t *t = h->thread[i];
2536         for( int j = 0; j < t->out.i_nal; j++ )
2537         {
2538             h->out.nal[h->out.i_nal] = t->out.nal[j];
2539             h->out.i_nal++;
2540             x264_nal_check_buffer( h );
2541         }
2542         /* All entries in stat.frame are ints except for ssd/ssim. */
2543         for( int j = 0; j < (offsetof(x264_t,stat.frame.i_ssd) - offsetof(x264_t,stat.frame.i_mv_bits)) / sizeof(int); j++ )
2544             ((int*)&h->stat.frame)[j] += ((int*)&t->stat.frame)[j];
2545         for( int j = 0; j < 3; j++ )
2546             h->stat.frame.i_ssd[j] += t->stat.frame.i_ssd[j];
2547         h->stat.frame.f_ssim += t->stat.frame.f_ssim;
2548         h->stat.frame.i_ssim_cnt += t->stat.frame.i_ssim_cnt;
2549     }
2550
2551     return 0;
2552 }
2553
2554 void x264_encoder_intra_refresh( x264_t *h )
2555 {
2556     h = h->thread[h->i_thread_phase];
2557     h->b_queued_intra_refresh = 1;
2558 }
2559
2560 int x264_encoder_invalidate_reference( x264_t *h, int64_t pts )
2561 {
2562     if( h->param.i_bframe )
2563     {
2564         x264_log( h, X264_LOG_ERROR, "x264_encoder_invalidate_reference is not supported with B-frames enabled\n" );
2565         return -1;
2566     }
2567     if( h->param.b_intra_refresh )
2568     {
2569         x264_log( h, X264_LOG_ERROR, "x264_encoder_invalidate_reference is not supported with intra refresh enabled\n" );
2570         return -1;
2571     }
2572     h = h->thread[h->i_thread_phase];
2573     if( pts >= h->i_last_idr_pts )
2574     {
2575         for( int i = 0; h->frames.reference[i]; i++ )
2576             if( pts <= h->frames.reference[i]->i_pts )
2577                 h->frames.reference[i]->b_corrupt = 1;
2578         if( pts <= h->fdec->i_pts )
2579             h->fdec->b_corrupt = 1;
2580     }
2581     return 0;
2582 }
2583
2584 /****************************************************************************
2585  * x264_encoder_encode:
2586  *  XXX: i_poc   : is the poc of the current given picture
2587  *       i_frame : is the number of the frame being coded
2588  *  ex:  type frame poc
2589  *       I      0   2*0
2590  *       P      1   2*3
2591  *       B      2   2*1
2592  *       B      3   2*2
2593  *       P      4   2*6
2594  *       B      5   2*4
2595  *       B      6   2*5
2596  ****************************************************************************/
2597 int     x264_encoder_encode( x264_t *h,
2598                              x264_nal_t **pp_nal, int *pi_nal,
2599                              x264_picture_t *pic_in,
2600                              x264_picture_t *pic_out )
2601 {
2602     x264_t *thread_current, *thread_prev, *thread_oldest;
2603     int i_nal_type, i_nal_ref_idc, i_global_qp;
2604     int overhead = NALU_OVERHEAD;
2605
2606     if( h->i_thread_frames > 1 )
2607     {
2608         thread_prev    = h->thread[ h->i_thread_phase ];
2609         h->i_thread_phase = (h->i_thread_phase + 1) % h->i_thread_frames;
2610         thread_current = h->thread[ h->i_thread_phase ];
2611         thread_oldest  = h->thread[ (h->i_thread_phase + 1) % h->i_thread_frames ];
2612         x264_thread_sync_context( thread_current, thread_prev );
2613         x264_thread_sync_ratecontrol( thread_current, thread_prev, thread_oldest );
2614         h = thread_current;
2615     }
2616     else
2617     {
2618         thread_current =
2619         thread_oldest  = h;
2620     }
2621 #if HAVE_MMX
2622     if( h->param.cpu&X264_CPU_SSE_MISALIGN )
2623         x264_cpu_mask_misalign_sse();
2624 #endif
2625
2626     // ok to call this before encoding any frames, since the initial values of fdec have b_kept_as_ref=0
2627     if( x264_reference_update( h ) )
2628         return -1;
2629     h->fdec->i_lines_completed = -1;
2630
2631     /* no data out */
2632     *pi_nal = 0;
2633     *pp_nal = NULL;
2634
2635     /* ------------------- Setup new frame from picture -------------------- */
2636     if( pic_in != NULL )
2637     {
2638         /* 1: Copy the picture to a frame and move it to a buffer */
2639         x264_frame_t *fenc = x264_frame_pop_unused( h, 0 );
2640         if( !fenc )
2641             return -1;
2642
2643         if( x264_frame_copy_picture( h, fenc, pic_in ) < 0 )
2644             return -1;
2645
2646         if( h->param.i_width != 16 * h->mb.i_mb_width ||
2647             h->param.i_height != 16 * h->mb.i_mb_height )
2648             x264_frame_expand_border_mod16( h, fenc );
2649
2650         fenc->i_frame = h->frames.i_input++;
2651
2652         if( fenc->i_frame == 0 )
2653             h->frames.i_first_pts = fenc->i_pts;
2654         if( h->frames.i_bframe_delay && fenc->i_frame == h->frames.i_bframe_delay )
2655             h->frames.i_bframe_delay_time = fenc->i_pts - h->frames.i_first_pts;
2656
2657         if( h->param.b_vfr_input && fenc->i_pts <= h->frames.i_largest_pts )
2658             x264_log( h, X264_LOG_WARNING, "non-strictly-monotonic PTS\n" );
2659
2660         h->frames.i_second_largest_pts = h->frames.i_largest_pts;
2661         h->frames.i_largest_pts = fenc->i_pts;
2662
2663         if( (fenc->i_pic_struct < PIC_STRUCT_AUTO) || (fenc->i_pic_struct > PIC_STRUCT_TRIPLE) )
2664             fenc->i_pic_struct = PIC_STRUCT_AUTO;
2665
2666         if( fenc->i_pic_struct == PIC_STRUCT_AUTO )
2667         {
2668 #if HAVE_INTERLACED
2669             int b_interlaced = fenc->param ? fenc->param->b_interlaced : h->param.b_interlaced;
2670 #else
2671             int b_interlaced = 0;
2672 #endif
2673             if( b_interlaced )
2674             {
2675                 int b_tff = fenc->param ? fenc->param->b_tff : h->param.b_tff;
2676                 fenc->i_pic_struct = b_tff ? PIC_STRUCT_TOP_BOTTOM : PIC_STRUCT_BOTTOM_TOP;
2677             }
2678             else
2679                 fenc->i_pic_struct = PIC_STRUCT_PROGRESSIVE;
2680         }
2681
2682         if( h->param.rc.b_mb_tree && h->param.rc.b_stat_read )
2683         {
2684             if( x264_macroblock_tree_read( h, fenc, pic_in->prop.quant_offsets ) )
2685                 return -1;
2686         }
2687         else
2688             x264_stack_align( x264_adaptive_quant_frame, h, fenc, pic_in->prop.quant_offsets );
2689
2690         if( pic_in->prop.quant_offsets_free )
2691             pic_in->prop.quant_offsets_free( pic_in->prop.quant_offsets );
2692
2693         if( h->frames.b_have_lowres )
2694             x264_frame_init_lowres( h, fenc );
2695
2696         /* 2: Place the frame into the queue for its slice type decision */
2697         x264_lookahead_put_frame( h, fenc );
2698
2699         if( h->frames.i_input <= h->frames.i_delay + 1 - h->i_thread_frames )
2700         {
2701             /* Nothing yet to encode, waiting for filling of buffers */
2702             pic_out->i_type = X264_TYPE_AUTO;
2703             return 0;
2704         }
2705     }
2706     else
2707     {
2708         /* signal kills for lookahead thread */
2709         x264_pthread_mutex_lock( &h->lookahead->ifbuf.mutex );
2710         h->lookahead->b_exit_thread = 1;
2711         x264_pthread_cond_broadcast( &h->lookahead->ifbuf.cv_fill );
2712         x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
2713     }
2714
2715     h->i_frame++;
2716     /* 3: The picture is analyzed in the lookahead */
2717     if( !h->frames.current[0] )
2718         x264_lookahead_get_frames( h );
2719
2720     if( !h->frames.current[0] && x264_lookahead_is_empty( h ) )
2721         return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
2722
2723     /* ------------------- Get frame to be encoded ------------------------- */
2724     /* 4: get picture to encode */
2725     h->fenc = x264_frame_shift( h->frames.current );
2726     if( h->i_frame == h->i_thread_frames - 1 )
2727         h->i_reordered_pts_delay = h->fenc->i_reordered_pts;
2728     if( h->fenc->param )
2729     {
2730         x264_encoder_reconfig( h, h->fenc->param );
2731         if( h->fenc->param->param_free )
2732             h->fenc->param->param_free( h->fenc->param );
2733     }
2734
2735     if( !IS_X264_TYPE_I( h->fenc->i_type ) )
2736     {
2737         int valid_refs_left = 0;
2738         for( int i = 0; h->frames.reference[i]; i++ )
2739             if( !h->frames.reference[i]->b_corrupt )
2740                 valid_refs_left++;
2741         /* No valid reference frames left: force an IDR. */
2742         if( !valid_refs_left )
2743         {
2744             h->fenc->b_keyframe = 1;
2745             h->fenc->i_type = X264_TYPE_IDR;
2746         }
2747     }
2748
2749     if( h->fenc->b_keyframe )
2750     {
2751         h->frames.i_last_keyframe = h->fenc->i_frame;
2752         if( h->fenc->i_type == X264_TYPE_IDR )
2753         {
2754             h->i_frame_num = 0;
2755             h->frames.i_last_idr = h->fenc->i_frame;
2756         }
2757     }
2758     h->sh.i_mmco_command_count =
2759     h->sh.i_mmco_remove_from_end = 0;
2760     h->b_ref_reorder[0] =
2761     h->b_ref_reorder[1] = 0;
2762     h->fdec->i_poc =
2763     h->fenc->i_poc = 2 * ( h->fenc->i_frame - X264_MAX( h->frames.i_last_idr, 0 ) );
2764
2765     /* ------------------- Setup frame context ----------------------------- */
2766     /* 5: Init data dependent of frame type */
2767     if( h->fenc->i_type == X264_TYPE_IDR )
2768     {
2769         /* reset ref pictures */
2770         i_nal_type    = NAL_SLICE_IDR;
2771         i_nal_ref_idc = NAL_PRIORITY_HIGHEST;
2772         h->sh.i_type = SLICE_TYPE_I;
2773         x264_reference_reset( h );
2774         h->frames.i_poc_last_open_gop = -1;
2775     }
2776     else if( h->fenc->i_type == X264_TYPE_I )
2777     {
2778         i_nal_type    = NAL_SLICE;
2779         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
2780         h->sh.i_type = SLICE_TYPE_I;
2781         x264_reference_hierarchy_reset( h );
2782         if( h->param.b_open_gop )
2783             h->frames.i_poc_last_open_gop = h->fenc->b_keyframe ? h->fenc->i_poc : -1;
2784     }
2785     else if( h->fenc->i_type == X264_TYPE_P )
2786     {
2787         i_nal_type    = NAL_SLICE;
2788         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
2789         h->sh.i_type = SLICE_TYPE_P;
2790         x264_reference_hierarchy_reset( h );
2791         h->frames.i_poc_last_open_gop = -1;
2792     }
2793     else if( h->fenc->i_type == X264_TYPE_BREF )
2794     {
2795         i_nal_type    = NAL_SLICE;
2796         i_nal_ref_idc = h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT ? NAL_PRIORITY_LOW : NAL_PRIORITY_HIGH;
2797         h->sh.i_type = SLICE_TYPE_B;
2798         x264_reference_hierarchy_reset( h );
2799     }
2800     else    /* B frame */
2801     {
2802         i_nal_type    = NAL_SLICE;
2803         i_nal_ref_idc = NAL_PRIORITY_DISPOSABLE;
2804         h->sh.i_type = SLICE_TYPE_B;
2805     }
2806
2807     h->fdec->i_type = h->fenc->i_type;
2808     h->fdec->i_frame = h->fenc->i_frame;
2809     h->fenc->b_kept_as_ref =
2810     h->fdec->b_kept_as_ref = i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE && h->param.i_keyint_max > 1;
2811
2812     h->fdec->i_pts = h->fenc->i_pts;
2813     if( h->frames.i_bframe_delay )
2814     {
2815         int64_t *prev_reordered_pts = thread_current->frames.i_prev_reordered_pts;
2816         h->fdec->i_dts = h->i_frame > h->frames.i_bframe_delay
2817                        ? prev_reordered_pts[ (h->i_frame - h->frames.i_bframe_delay) % h->frames.i_bframe_delay ]
2818                        : h->fenc->i_reordered_pts - h->frames.i_bframe_delay_time;
2819         prev_reordered_pts[ h->i_frame % h->frames.i_bframe_delay ] = h->fenc->i_reordered_pts;
2820     }
2821     else
2822         h->fdec->i_dts = h->fenc->i_reordered_pts;
2823     if( h->fenc->i_type == X264_TYPE_IDR )
2824         h->i_last_idr_pts = h->fdec->i_pts;
2825
2826     /* ------------------- Init                ----------------------------- */
2827     /* build ref list 0/1 */
2828     x264_reference_build_list( h, h->fdec->i_poc );
2829
2830     /* ---------------------- Write the bitstream -------------------------- */
2831     /* Init bitstream context */
2832     if( h->param.b_sliced_threads )
2833     {
2834         for( int i = 0; i < h->param.i_threads; i++ )
2835         {
2836             bs_init( &h->thread[i]->out.bs, h->thread[i]->out.p_bitstream, h->thread[i]->out.i_bitstream );
2837             h->thread[i]->out.i_nal = 0;
2838         }
2839     }
2840     else
2841     {
2842         bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
2843         h->out.i_nal = 0;
2844     }
2845
2846     if( h->param.b_aud )
2847     {
2848         int pic_type;
2849
2850         if( h->sh.i_type == SLICE_TYPE_I )
2851             pic_type = 0;
2852         else if( h->sh.i_type == SLICE_TYPE_P )
2853             pic_type = 1;
2854         else if( h->sh.i_type == SLICE_TYPE_B )
2855             pic_type = 2;
2856         else
2857             pic_type = 7;
2858
2859         x264_nal_start( h, NAL_AUD, NAL_PRIORITY_DISPOSABLE );
2860         bs_write( &h->out.bs, 3, pic_type );
2861         bs_rbsp_trailing( &h->out.bs );
2862         if( x264_nal_end( h ) )
2863             return -1;
2864         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2865     }
2866
2867     h->i_nal_type = i_nal_type;
2868     h->i_nal_ref_idc = i_nal_ref_idc;
2869
2870     if( h->param.b_intra_refresh )
2871     {
2872         if( IS_X264_TYPE_I( h->fenc->i_type ) )
2873         {
2874             h->fdec->i_frames_since_pir = 0;
2875             h->b_queued_intra_refresh = 0;
2876             /* PIR is currently only supported with ref == 1, so any intra frame effectively refreshes
2877              * the whole frame and counts as an intra refresh. */
2878             h->fdec->f_pir_position = h->mb.i_mb_width;
2879         }
2880         else if( h->fenc->i_type == X264_TYPE_P )
2881         {
2882             int pocdiff = (h->fdec->i_poc - h->fref[0][0]->i_poc)/2;
2883             float increment = X264_MAX( ((float)h->mb.i_mb_width-1) / h->param.i_keyint_max, 1 );
2884             h->fdec->f_pir_position = h->fref[0][0]->f_pir_position;
2885             h->fdec->i_frames_since_pir = h->fref[0][0]->i_frames_since_pir + pocdiff;
2886             if( h->fdec->i_frames_since_pir >= h->param.i_keyint_max ||
2887                 (h->b_queued_intra_refresh && h->fdec->f_pir_position + 0.5 >= h->mb.i_mb_width) )
2888             {
2889                 h->fdec->f_pir_position = 0;
2890                 h->fdec->i_frames_since_pir = 0;
2891                 h->b_queued_intra_refresh = 0;
2892                 h->fenc->b_keyframe = 1;
2893             }
2894             h->fdec->i_pir_start_col = h->fdec->f_pir_position+0.5;
2895             h->fdec->f_pir_position += increment * pocdiff;
2896             h->fdec->i_pir_end_col = h->fdec->f_pir_position+0.5;
2897             /* If our intra refresh has reached the right side of the frame, we're done. */
2898             if( h->fdec->i_pir_end_col >= h->mb.i_mb_width - 1 )
2899                 h->fdec->f_pir_position = h->mb.i_mb_width;
2900         }
2901     }
2902
2903     if( h->fenc->b_keyframe )
2904     {
2905         /* Write SPS and PPS */
2906         if( h->param.b_repeat_headers )
2907         {
2908             /* generate sequence parameters */
2909             x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
2910             x264_sps_write( &h->out.bs, h->sps );
2911             if( x264_nal_end( h ) )
2912                 return -1;
2913             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
2914
2915             /* generate picture parameters */
2916             x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
2917             x264_pps_write( &h->out.bs, h->sps, h->pps );
2918             if( x264_nal_end( h ) )
2919                 return -1;
2920             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
2921         }
2922
2923         /* when frame threading is used, buffering period sei is written in x264_encoder_frame_end */
2924         if( h->i_thread_frames == 1 && h->sps->vui.b_nal_hrd_parameters_present )
2925         {
2926             x264_hrd_fullness( h );
2927             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2928             x264_sei_buffering_period_write( h, &h->out.bs );
2929             if( x264_nal_end( h ) )
2930                return -1;
2931             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
2932         }
2933     }
2934
2935     /* write extra sei */
2936     for( int i = 0; i < h->fenc->extra_sei.num_payloads; i++ )
2937     {
2938         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2939         x264_sei_write( &h->out.bs, h->fenc->extra_sei.payloads[i].payload, h->fenc->extra_sei.payloads[i].payload_size,
2940                         h->fenc->extra_sei.payloads[i].payload_type );
2941         if( x264_nal_end( h ) )
2942             return -1;
2943         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2944         if( h->fenc->extra_sei.sei_free && h->fenc->extra_sei.payloads[i].payload )
2945             h->fenc->extra_sei.sei_free( h->fenc->extra_sei.payloads[i].payload );
2946     }
2947
2948     if( h->fenc->extra_sei.sei_free && h->fenc->extra_sei.payloads )
2949         h->fenc->extra_sei.sei_free( h->fenc->extra_sei.payloads );
2950
2951     if( h->fenc->b_keyframe )
2952     {
2953         if( h->param.b_repeat_headers && h->fenc->i_frame == 0 )
2954         {
2955             /* identify ourself */
2956             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2957             if( x264_sei_version_write( h, &h->out.bs ) )
2958                 return -1;
2959             if( x264_nal_end( h ) )
2960                 return -1;
2961             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2962         }
2963
2964         if( h->fenc->i_type != X264_TYPE_IDR )
2965         {
2966             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;
2967             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2968             x264_sei_recovery_point_write( h, &h->out.bs, time_to_recovery );
2969             if( x264_nal_end( h ) )
2970                 return -1;
2971             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2972         }
2973
2974         if ( h->param.i_frame_packing >= 0 )
2975         {
2976             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2977             x264_sei_frame_packing_write( h, &h->out.bs );
2978             if( x264_nal_end( h ) )
2979                 return -1;
2980             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2981         }
2982     }
2983
2984     /* generate sei pic timing */
2985     if( h->sps->vui.b_pic_struct_present || h->sps->vui.b_nal_hrd_parameters_present )
2986     {
2987         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2988         x264_sei_pic_timing_write( h, &h->out.bs );
2989         if( x264_nal_end( h ) )
2990             return -1;
2991         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2992     }
2993
2994     /* As required by Blu-ray. */
2995     if( !IS_X264_TYPE_B( h->fenc->i_type ) && h->b_sh_backup )
2996     {
2997         h->b_sh_backup = 0;
2998         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2999         x264_sei_dec_ref_pic_marking_write( h, &h->out.bs );
3000         if( x264_nal_end( h ) )
3001             return -1;
3002         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
3003     }
3004
3005     if( h->fenc->b_keyframe && h->param.b_intra_refresh )
3006         h->i_cpb_delay_pir_offset = h->fenc->i_cpb_delay;
3007
3008     /* Init the rate control */
3009     /* FIXME: Include slice header bit cost. */
3010     x264_ratecontrol_start( h, h->fenc->i_qpplus1, overhead*8 );
3011     i_global_qp = x264_ratecontrol_qp( h );
3012
3013     pic_out->i_qpplus1 =
3014     h->fdec->i_qpplus1 = i_global_qp + 1;
3015
3016     if( h->param.rc.b_stat_read && h->sh.i_type != SLICE_TYPE_I )
3017     {
3018         x264_reference_build_list_optimal( h );
3019         x264_reference_check_reorder( h );
3020     }
3021
3022     if( h->i_ref[0] )
3023         h->fdec->i_poc_l0ref0 = h->fref[0][0]->i_poc;
3024
3025     /* ------------------------ Create slice header  ----------------------- */
3026     x264_slice_init( h, i_nal_type, i_global_qp );
3027
3028     /*------------------------- Weights -------------------------------------*/
3029     if( h->sh.i_type == SLICE_TYPE_B )
3030         x264_macroblock_bipred_init( h );
3031
3032     x264_weighted_pred_init( h );
3033
3034     if( i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE )
3035         h->i_frame_num++;
3036
3037     /* Write frame */
3038     h->i_threadslice_start = 0;
3039     h->i_threadslice_end = h->mb.i_mb_height;
3040     if( h->i_thread_frames > 1 )
3041     {
3042         x264_threadpool_run( h->threadpool, (void*)x264_slices_write, h );
3043         h->b_thread_active = 1;
3044     }
3045     else if( h->param.b_sliced_threads )
3046     {
3047         if( x264_threaded_slices_write( h ) )
3048             return -1;
3049     }
3050     else
3051         if( (intptr_t)x264_slices_write( h ) )
3052             return -1;
3053
3054     return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
3055 }
3056
3057 static int x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
3058                                    x264_nal_t **pp_nal, int *pi_nal,
3059                                    x264_picture_t *pic_out )
3060 {
3061     char psz_message[80];
3062
3063     if( h->b_thread_active )
3064     {
3065         h->b_thread_active = 0;
3066         if( (intptr_t)x264_threadpool_wait( h->threadpool, h ) )
3067             return -1;
3068     }
3069     if( !h->out.i_nal )
3070     {
3071         pic_out->i_type = X264_TYPE_AUTO;
3072         return 0;
3073     }
3074
3075     x264_emms();
3076     /* generate buffering period sei and insert it into place */
3077     if( h->i_thread_frames > 1 && h->fenc->b_keyframe && h->sps->vui.b_nal_hrd_parameters_present )
3078     {
3079         x264_hrd_fullness( h );
3080         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3081         x264_sei_buffering_period_write( h, &h->out.bs );
3082         if( x264_nal_end( h ) )
3083            return -1;
3084         /* buffering period sei must follow AUD, SPS and PPS and precede all other SEIs */
3085         int idx = 0;
3086         while( h->out.nal[idx].i_type == NAL_AUD ||
3087                h->out.nal[idx].i_type == NAL_SPS ||
3088                h->out.nal[idx].i_type == NAL_PPS )
3089             idx++;
3090         x264_nal_t nal_tmp = h->out.nal[h->out.i_nal-1];
3091         memmove( &h->out.nal[idx+1], &h->out.nal[idx], (h->out.i_nal-idx-1)*sizeof(x264_nal_t) );
3092         h->out.nal[idx] = nal_tmp;
3093     }
3094
3095     int frame_size = x264_encoder_encapsulate_nals( h, 0 );
3096     if( frame_size < 0 )
3097         return -1;
3098
3099     /* Set output picture properties */
3100     pic_out->i_type = h->fenc->i_type;
3101
3102     pic_out->b_keyframe = h->fenc->b_keyframe;
3103     pic_out->i_pic_struct = h->fenc->i_pic_struct;
3104
3105     pic_out->i_pts = h->fdec->i_pts;
3106     pic_out->i_dts = h->fdec->i_dts;
3107
3108     if( pic_out->i_pts < pic_out->i_dts )
3109         x264_log( h, X264_LOG_WARNING, "invalid DTS: PTS is less than DTS\n" );
3110
3111     pic_out->opaque = h->fenc->opaque;
3112
3113     pic_out->img.i_csp = h->fdec->i_csp;
3114 #if HIGH_BIT_DEPTH
3115     pic_out->img.i_csp |= X264_CSP_HIGH_DEPTH;
3116 #endif
3117     pic_out->img.i_plane = h->fdec->i_plane;
3118     for( int i = 0; i < pic_out->img.i_plane; i++ )
3119     {
3120         pic_out->img.i_stride[i] = h->fdec->i_stride[i] * sizeof(pixel);
3121         pic_out->img.plane[i] = (uint8_t*)h->fdec->plane[i];
3122     }
3123
3124     x264_frame_push_unused( thread_current, h->fenc );
3125
3126     /* ---------------------- Update encoder state ------------------------- */
3127
3128     /* update rc */
3129     int filler = 0;
3130     if( x264_ratecontrol_end( h, frame_size * 8, &filler ) < 0 )
3131         return -1;
3132
3133     pic_out->hrd_timing = h->fenc->hrd_timing;
3134
3135     while( filler > 0 )
3136     {
3137         int f, overhead;
3138         overhead = (FILLER_OVERHEAD - h->param.b_annexb);
3139         if( h->param.i_slice_max_size && filler > h->param.i_slice_max_size )
3140         {
3141             int next_size = filler - h->param.i_slice_max_size;
3142             int overflow = X264_MAX( overhead - next_size, 0 );
3143             f = h->param.i_slice_max_size - overhead - overflow;
3144         }
3145         else
3146             f = X264_MAX( 0, filler - overhead );
3147
3148         x264_nal_start( h, NAL_FILLER, NAL_PRIORITY_DISPOSABLE );
3149         x264_filler_write( h, &h->out.bs, f );
3150         if( x264_nal_end( h ) )
3151             return -1;
3152         int total_size = x264_encoder_encapsulate_nals( h, h->out.i_nal-1 );
3153         if( total_size < 0 )
3154             return -1;
3155         frame_size += total_size;
3156         filler -= total_size;
3157     }
3158
3159     /* End bitstream, set output  */
3160     *pi_nal = h->out.i_nal;
3161     *pp_nal = h->out.nal;
3162
3163     h->out.i_nal = 0;
3164
3165     x264_noise_reduction_update( h );
3166
3167     /* ---------------------- Compute/Print statistics --------------------- */
3168     x264_thread_sync_stat( h, h->thread[0] );
3169
3170     /* Slice stat */
3171     h->stat.i_frame_count[h->sh.i_type]++;
3172     h->stat.i_frame_size[h->sh.i_type] += frame_size;
3173     h->stat.f_frame_qp[h->sh.i_type] += h->fdec->f_qp_avg_aq;
3174
3175     for( int i = 0; i < X264_MBTYPE_MAX; i++ )
3176         h->stat.i_mb_count[h->sh.i_type][i] += h->stat.frame.i_mb_count[i];
3177     for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
3178         h->stat.i_mb_partition[h->sh.i_type][i] += h->stat.frame.i_mb_partition[i];
3179     for( int i = 0; i < 2; i++ )
3180         h->stat.i_mb_count_8x8dct[i] += h->stat.frame.i_mb_count_8x8dct[i];
3181     for( int i = 0; i < 6; i++ )
3182         h->stat.i_mb_cbp[i] += h->stat.frame.i_mb_cbp[i];
3183     for( int i = 0; i < 4; i++ )
3184         for( int j = 0; j < 13; j++ )
3185             h->stat.i_mb_pred_mode[i][j] += h->stat.frame.i_mb_pred_mode[i][j];
3186     if( h->sh.i_type != SLICE_TYPE_I )
3187         for( int i_list = 0; i_list < 2; i_list++ )
3188             for( int i = 0; i < X264_REF_MAX*2; i++ )
3189                 h->stat.i_mb_count_ref[h->sh.i_type][i_list][i] += h->stat.frame.i_mb_count_ref[i_list][i];
3190     for( int i = 0; i < 3; i++ )
3191         h->stat.i_mb_field[i] += h->stat.frame.i_mb_field[i];
3192     if( h->sh.i_type == SLICE_TYPE_P && h->param.analyse.i_weighted_pred >= X264_WEIGHTP_SIMPLE )
3193     {
3194         h->stat.i_wpred[0] += !!h->sh.weight[0][0].weightfn;
3195         h->stat.i_wpred[1] += !!h->sh.weight[0][1].weightfn || !!h->sh.weight[0][2].weightfn;
3196     }
3197     if( h->sh.i_type == SLICE_TYPE_B )
3198     {
3199         h->stat.i_direct_frames[ h->sh.b_direct_spatial_mv_pred ] ++;
3200         if( h->mb.b_direct_auto_write )
3201         {
3202             //FIXME somewhat arbitrary time constants
3203             if( h->stat.i_direct_score[0] + h->stat.i_direct_score[1] > h->mb.i_mb_count )
3204                 for( int i = 0; i < 2; i++ )
3205                     h->stat.i_direct_score[i] = h->stat.i_direct_score[i] * 9/10;
3206             for( int i = 0; i < 2; i++ )
3207                 h->stat.i_direct_score[i] += h->stat.frame.i_direct_score[i];
3208         }
3209     }
3210     else
3211         h->stat.i_consecutive_bframes[h->fenc->i_bframes]++;
3212
3213     psz_message[0] = '\0';
3214     double dur = h->fenc->f_duration;
3215     h->stat.f_frame_duration[h->sh.i_type] += dur;
3216     if( h->param.analyse.b_psnr )
3217     {
3218         int64_t ssd[3] =
3219         {
3220             h->stat.frame.i_ssd[0],
3221             h->stat.frame.i_ssd[1],
3222             h->stat.frame.i_ssd[2],
3223         };
3224         int luma_size = h->param.i_width * h->param.i_height;
3225         int chroma_size = CHROMA_SIZE( luma_size );
3226         double psnr_y = x264_psnr( ssd[0], luma_size );
3227         double psnr_u = x264_psnr( ssd[1], chroma_size );
3228         double psnr_v = x264_psnr( ssd[2], chroma_size );
3229
3230         h->stat.f_ssd_global[h->sh.i_type]   += dur * (ssd[0] + ssd[1] + ssd[2]);
3231         h->stat.f_psnr_average[h->sh.i_type] += dur * x264_psnr( ssd[0] + ssd[1] + ssd[2], luma_size + chroma_size*2 );
3232         h->stat.f_psnr_mean_y[h->sh.i_type]  += dur * psnr_y;
3233         h->stat.f_psnr_mean_u[h->sh.i_type]  += dur * psnr_u;
3234         h->stat.f_psnr_mean_v[h->sh.i_type]  += dur * psnr_v;
3235
3236         snprintf( psz_message, 80, " PSNR Y:%5.2f U:%5.2f V:%5.2f", psnr_y, psnr_u, psnr_v );
3237     }
3238
3239     if( h->param.analyse.b_ssim )
3240     {
3241         double ssim_y = h->stat.frame.f_ssim
3242                       / h->stat.frame.i_ssim_cnt;
3243         h->stat.f_ssim_mean_y[h->sh.i_type] += ssim_y * dur;
3244         snprintf( psz_message + strlen(psz_message), 80 - strlen(psz_message),
3245                   " SSIM Y:%.5f", ssim_y );
3246     }
3247     psz_message[79] = '\0';
3248
3249     x264_log( h, X264_LOG_DEBUG,
3250                   "frame=%4d QP=%.2f NAL=%d Slice:%c Poc:%-3d I:%-4d P:%-4d SKIP:%-4d size=%d bytes%s\n",
3251               h->i_frame,
3252               h->fdec->f_qp_avg_aq,
3253               h->i_nal_ref_idc,
3254               h->sh.i_type == SLICE_TYPE_I ? 'I' : (h->sh.i_type == SLICE_TYPE_P ? 'P' : 'B' ),
3255               h->fdec->i_poc,
3256               h->stat.frame.i_mb_count_i,
3257               h->stat.frame.i_mb_count_p,
3258               h->stat.frame.i_mb_count_skip,
3259               frame_size,
3260               psz_message );
3261
3262     // keep stats all in one place
3263     x264_thread_sync_stat( h->thread[0], h );
3264     // for the use of the next frame
3265     x264_thread_sync_stat( thread_current, h );
3266
3267 #ifdef DEBUG_MB_TYPE
3268 {
3269     static const char mb_chars[] = { 'i', 'i', 'I', 'C', 'P', '8', 'S',
3270         'D', '<', 'X', 'B', 'X', '>', 'B', 'B', 'B', 'B', '8', 'S' };
3271     for( int mb_xy = 0; mb_xy < h->mb.i_mb_width * h->mb.i_mb_height; mb_xy++ )
3272     {
3273         if( h->mb.type[mb_xy] < X264_MBTYPE_MAX && h->mb.type[mb_xy] >= 0 )
3274             fprintf( stderr, "%c ", mb_chars[ h->mb.type[mb_xy] ] );
3275         else
3276             fprintf( stderr, "? " );
3277
3278         if( (mb_xy+1) % h->mb.i_mb_width == 0 )
3279             fprintf( stderr, "\n" );
3280     }
3281 }
3282 #endif
3283
3284     /* Remove duplicates, must be done near the end as breaks h->fref0 array
3285      * by freeing some of its pointers. */
3286     for( int i = 0; i < h->i_ref[0]; i++ )
3287         if( h->fref[0][i] && h->fref[0][i]->b_duplicate )
3288         {
3289             x264_frame_push_blank_unused( h, h->fref[0][i] );
3290             h->fref[0][i] = 0;
3291         }
3292
3293     if( h->param.psz_dump_yuv )
3294         x264_frame_dump( h );
3295     x264_emms();
3296
3297     return frame_size;
3298 }
3299
3300 static void x264_print_intra( int64_t *i_mb_count, double i_count, int b_print_pcm, char *intra )
3301 {
3302     intra += sprintf( intra, "I16..4%s: %4.1f%% %4.1f%% %4.1f%%",
3303         b_print_pcm ? "..PCM" : "",
3304         i_mb_count[I_16x16]/ i_count,
3305         i_mb_count[I_8x8]  / i_count,
3306         i_mb_count[I_4x4]  / i_count );
3307     if( b_print_pcm )
3308         sprintf( intra, " %4.1f%%", i_mb_count[I_PCM]  / i_count );
3309 }
3310
3311 /****************************************************************************
3312  * x264_encoder_close:
3313  ****************************************************************************/
3314 void    x264_encoder_close  ( x264_t *h )
3315 {
3316     int64_t i_yuv_size = FRAME_SIZE( h->param.i_width * h->param.i_height );
3317     int64_t i_mb_count_size[2][7] = {{0}};
3318     char buf[200];
3319     int b_print_pcm = h->stat.i_mb_count[SLICE_TYPE_I][I_PCM]
3320                    || h->stat.i_mb_count[SLICE_TYPE_P][I_PCM]
3321                    || h->stat.i_mb_count[SLICE_TYPE_B][I_PCM];
3322
3323     x264_lookahead_delete( h );
3324
3325     if( h->param.i_threads > 1 )
3326         x264_threadpool_delete( h->threadpool );
3327     if( h->i_thread_frames > 1 )
3328     {
3329         for( int i = 0; i < h->i_thread_frames; i++ )
3330             if( h->thread[i]->b_thread_active )
3331             {
3332                 assert( h->thread[i]->fenc->i_reference_count == 1 );
3333                 x264_frame_delete( h->thread[i]->fenc );
3334             }
3335
3336         x264_t *thread_prev = h->thread[h->i_thread_phase];
3337         x264_thread_sync_ratecontrol( h, thread_prev, h );
3338         x264_thread_sync_ratecontrol( thread_prev, thread_prev, h );
3339         h->i_frame = thread_prev->i_frame + 1 - h->i_thread_frames;
3340     }
3341     h->i_frame++;
3342
3343     /* Slices used and PSNR */
3344     for( int i = 0; i < 3; i++ )
3345     {
3346         static const uint8_t slice_order[] = { SLICE_TYPE_I, SLICE_TYPE_P, SLICE_TYPE_B };
3347         int i_slice = slice_order[i];
3348
3349         if( h->stat.i_frame_count[i_slice] > 0 )
3350         {
3351             int i_count = h->stat.i_frame_count[i_slice];
3352             double dur =  h->stat.f_frame_duration[i_slice];
3353             if( h->param.analyse.b_psnr )
3354             {
3355                 x264_log( h, X264_LOG_INFO,
3356                           "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",
3357                           slice_type_to_char[i_slice],
3358                           i_count,
3359                           h->stat.f_frame_qp[i_slice] / i_count,
3360                           (double)h->stat.i_frame_size[i_slice] / i_count,
3361                           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,
3362                           h->stat.f_psnr_average[i_slice] / dur,
3363                           x264_psnr( h->stat.f_ssd_global[i_slice], dur * i_yuv_size ) );
3364             }
3365             else
3366             {
3367                 x264_log( h, X264_LOG_INFO,
3368                           "frame %c:%-5d Avg QP:%5.2f  size:%6.0f\n",
3369                           slice_type_to_char[i_slice],
3370                           i_count,
3371                           h->stat.f_frame_qp[i_slice] / i_count,
3372                           (double)h->stat.i_frame_size[i_slice] / i_count );
3373             }
3374         }
3375     }
3376     if( h->param.i_bframe && h->stat.i_frame_count[SLICE_TYPE_B] )
3377     {
3378         char *p = buf;
3379         int den = 0;
3380         // weight by number of frames (including the I/P-frames) that are in a sequence of N B-frames
3381         for( int i = 0; i <= h->param.i_bframe; i++ )
3382             den += (i+1) * h->stat.i_consecutive_bframes[i];
3383         for( int i = 0; i <= h->param.i_bframe; i++ )
3384             p += sprintf( p, " %4.1f%%", 100. * (i+1) * h->stat.i_consecutive_bframes[i] / den );
3385         x264_log( h, X264_LOG_INFO, "consecutive B-frames:%s\n", buf );
3386     }
3387
3388     for( int i_type = 0; i_type < 2; i_type++ )
3389         for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
3390         {
3391             if( i == D_DIRECT_8x8 ) continue; /* direct is counted as its own type */
3392             i_mb_count_size[i_type][x264_mb_partition_pixel_table[i]] += h->stat.i_mb_partition[i_type][i];
3393         }
3394
3395     /* MB types used */
3396     if( h->stat.i_frame_count[SLICE_TYPE_I] > 0 )
3397     {
3398         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_I];
3399         double i_count = h->stat.i_frame_count[SLICE_TYPE_I] * h->mb.i_mb_count / 100.0;
3400         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
3401         x264_log( h, X264_LOG_INFO, "mb I  %s\n", buf );
3402     }
3403     if( h->stat.i_frame_count[SLICE_TYPE_P] > 0 )
3404     {
3405         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_P];
3406         double i_count = h->stat.i_frame_count[SLICE_TYPE_P] * h->mb.i_mb_count / 100.0;
3407         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_P];
3408         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
3409         x264_log( h, X264_LOG_INFO,
3410                   "mb P  %s  P16..4: %4.1f%% %4.1f%% %4.1f%% %4.1f%% %4.1f%%    skip:%4.1f%%\n",
3411                   buf,
3412                   i_mb_size[PIXEL_16x16] / (i_count*4),
3413                   (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
3414                   i_mb_size[PIXEL_8x8] / (i_count*4),
3415                   (i_mb_size[PIXEL_8x4] + i_mb_size[PIXEL_4x8]) / (i_count*4),
3416                   i_mb_size[PIXEL_4x4] / (i_count*4),
3417                   i_mb_count[P_SKIP] / i_count );
3418     }
3419     if( h->stat.i_frame_count[SLICE_TYPE_B] > 0 )
3420     {
3421         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_B];
3422         double i_count = h->stat.i_frame_count[SLICE_TYPE_B] * h->mb.i_mb_count / 100.0;
3423         double i_mb_list_count;
3424         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_B];
3425         int64_t list_count[3] = {0}; /* 0 == L0, 1 == L1, 2 == BI */
3426         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
3427         for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
3428             for( int j = 0; j < 2; j++ )
3429             {
3430                 int l0 = x264_mb_type_list_table[i][0][j];
3431                 int l1 = x264_mb_type_list_table[i][1][j];
3432                 if( l0 || l1 )
3433                     list_count[l1+l0*l1] += h->stat.i_mb_count[SLICE_TYPE_B][i] * 2;
3434             }
3435         list_count[0] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L0_8x8];
3436         list_count[1] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L1_8x8];
3437         list_count[2] += h->stat.i_mb_partition[SLICE_TYPE_B][D_BI_8x8];
3438         i_mb_count[B_DIRECT] += (h->stat.i_mb_partition[SLICE_TYPE_B][D_DIRECT_8x8]+2)/4;
3439         i_mb_list_count = (list_count[0] + list_count[1] + list_count[2]) / 100.0;
3440         sprintf( buf + strlen(buf), "  B16..8: %4.1f%% %4.1f%% %4.1f%%  direct:%4.1f%%  skip:%4.1f%%",
3441                  i_mb_size[PIXEL_16x16] / (i_count*4),
3442                  (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
3443                  i_mb_size[PIXEL_8x8] / (i_count*4),
3444                  i_mb_count[B_DIRECT] / i_count,
3445                  i_mb_count[B_SKIP]   / i_count );
3446         if( i_mb_list_count != 0 )
3447             sprintf( buf + strlen(buf), "  L0:%4.1f%% L1:%4.1f%% BI:%4.1f%%",
3448                      list_count[0] / i_mb_list_count,
3449                      list_count[1] / i_mb_list_count,
3450                      list_count[2] / i_mb_list_count );
3451         x264_log( h, X264_LOG_INFO, "mb B  %s\n", buf );
3452     }
3453
3454     x264_ratecontrol_summary( h );
3455
3456     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 )
3457     {
3458 #define SUM3(p) (p[SLICE_TYPE_I] + p[SLICE_TYPE_P] + p[SLICE_TYPE_B])
3459 #define SUM3b(p,o) (p[SLICE_TYPE_I][o] + p[SLICE_TYPE_P][o] + p[SLICE_TYPE_B][o])
3460         int64_t i_i8x8 = SUM3b( h->stat.i_mb_count, I_8x8 );
3461         int64_t i_intra = i_i8x8 + SUM3b( h->stat.i_mb_count, I_4x4 )
3462                                  + SUM3b( h->stat.i_mb_count, I_16x16 );
3463         int64_t i_all_intra = i_intra + SUM3b( h->stat.i_mb_count, I_PCM);
3464         int64_t i_skip = SUM3b( h->stat.i_mb_count, P_SKIP )
3465                        + SUM3b( h->stat.i_mb_count, B_SKIP );
3466         const int i_count = h->stat.i_frame_count[SLICE_TYPE_I] +
3467                             h->stat.i_frame_count[SLICE_TYPE_P] +
3468                             h->stat.i_frame_count[SLICE_TYPE_B];
3469         int64_t i_mb_count = (int64_t)i_count * h->mb.i_mb_count;
3470         int64_t i_inter = i_mb_count - i_skip - i_intra;
3471         const double duration = h->stat.f_frame_duration[SLICE_TYPE_I] +
3472                                 h->stat.f_frame_duration[SLICE_TYPE_P] +
3473                                 h->stat.f_frame_duration[SLICE_TYPE_B];
3474         float f_bitrate = SUM3(h->stat.i_frame_size) / duration / 125;
3475
3476         if( PARAM_INTERLACED )
3477         {
3478             char *fieldstats = buf;
3479             fieldstats[0] = 0;
3480             if( i_inter )
3481                 fieldstats += sprintf( fieldstats, " inter:%.1f%%", h->stat.i_mb_field[1] * 100.0 / i_inter );
3482             if( i_skip )
3483                 fieldstats += sprintf( fieldstats, " skip:%.1f%%", h->stat.i_mb_field[2] * 100.0 / i_skip );
3484             x264_log( h, X264_LOG_INFO, "field mbs: intra: %.1f%%%s\n",
3485                       h->stat.i_mb_field[0] * 100.0 / i_intra, buf );
3486         }
3487
3488         if( h->pps->b_transform_8x8_mode )
3489         {
3490             buf[0] = 0;
3491             if( h->stat.i_mb_count_8x8dct[0] )
3492                 sprintf( buf, " inter:%.1f%%", 100. * h->stat.i_mb_count_8x8dct[1] / h->stat.i_mb_count_8x8dct[0] );
3493             x264_log( h, X264_LOG_INFO, "8x8 transform intra:%.1f%%%s\n", 100. * i_i8x8 / i_intra, buf );
3494         }
3495
3496         if( (h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO ||
3497             (h->stat.i_direct_frames[0] && h->stat.i_direct_frames[1]))
3498             && h->stat.i_frame_count[SLICE_TYPE_B] )
3499         {
3500             x264_log( h, X264_LOG_INFO, "direct mvs  spatial:%.1f%% temporal:%.1f%%\n",
3501                       h->stat.i_direct_frames[1] * 100. / h->stat.i_frame_count[SLICE_TYPE_B],
3502                       h->stat.i_direct_frames[0] * 100. / h->stat.i_frame_count[SLICE_TYPE_B] );
3503         }
3504
3505         buf[0] = 0;
3506         int csize = CHROMA444 ? 4 : 1;
3507         if( i_mb_count != i_all_intra )
3508             sprintf( buf, " inter: %.1f%% %.1f%% %.1f%%",
3509                      h->stat.i_mb_cbp[1] * 100.0 / ((i_mb_count - i_all_intra)*4),
3510                      h->stat.i_mb_cbp[3] * 100.0 / ((i_mb_count - i_all_intra)*csize),
3511                      h->stat.i_mb_cbp[5] * 100.0 / ((i_mb_count - i_all_intra)*csize) );
3512         x264_log( h, X264_LOG_INFO, "coded y,%s,%s intra: %.1f%% %.1f%% %.1f%%%s\n",
3513                   CHROMA444?"u":"uvDC", CHROMA444?"v":"uvAC",
3514                   h->stat.i_mb_cbp[0] * 100.0 / (i_all_intra*4),
3515                   h->stat.i_mb_cbp[2] * 100.0 / (i_all_intra*csize),
3516                   h->stat.i_mb_cbp[4] * 100.0 / (i_all_intra*csize), buf );
3517
3518         int64_t fixed_pred_modes[4][9] = {{0}};
3519         int64_t sum_pred_modes[4] = {0};
3520         for( int i = 0; i <= I_PRED_16x16_DC_128; i++ )
3521         {
3522             fixed_pred_modes[0][x264_mb_pred_mode16x16_fix[i]] += h->stat.i_mb_pred_mode[0][i];
3523             sum_pred_modes[0] += h->stat.i_mb_pred_mode[0][i];
3524         }
3525         if( sum_pred_modes[0] )
3526             x264_log( h, X264_LOG_INFO, "i16 v,h,dc,p: %2.0f%% %2.0f%% %2.0f%% %2.0f%%\n",
3527                       fixed_pred_modes[0][0] * 100.0 / sum_pred_modes[0],
3528                       fixed_pred_modes[0][1] * 100.0 / sum_pred_modes[0],
3529                       fixed_pred_modes[0][2] * 100.0 / sum_pred_modes[0],
3530                       fixed_pred_modes[0][3] * 100.0 / sum_pred_modes[0] );
3531         for( int i = 1; i <= 2; i++ )
3532         {
3533             for( int j = 0; j <= I_PRED_8x8_DC_128; j++ )
3534             {
3535                 fixed_pred_modes[i][x264_mb_pred_mode4x4_fix(j)] += h->stat.i_mb_pred_mode[i][j];
3536                 sum_pred_modes[i] += h->stat.i_mb_pred_mode[i][j];
3537             }
3538             if( sum_pred_modes[i] )
3539                 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,
3540                           fixed_pred_modes[i][0] * 100.0 / sum_pred_modes[i],
3541                           fixed_pred_modes[i][1] * 100.0 / sum_pred_modes[i],
3542                           fixed_pred_modes[i][2] * 100.0 / sum_pred_modes[i],
3543                           fixed_pred_modes[i][3] * 100.0 / sum_pred_modes[i],
3544                           fixed_pred_modes[i][4] * 100.0 / sum_pred_modes[i],
3545                           fixed_pred_modes[i][5] * 100.0 / sum_pred_modes[i],
3546                           fixed_pred_modes[i][6] * 100.0 / sum_pred_modes[i],
3547                           fixed_pred_modes[i][7] * 100.0 / sum_pred_modes[i],
3548                           fixed_pred_modes[i][8] * 100.0 / sum_pred_modes[i] );
3549         }
3550         for( int i = 0; i <= I_PRED_CHROMA_DC_128; i++ )
3551         {
3552             fixed_pred_modes[3][x264_mb_chroma_pred_mode_fix[i]] += h->stat.i_mb_pred_mode[3][i];
3553             sum_pred_modes[3] += h->stat.i_mb_pred_mode[3][i];
3554         }
3555         if( sum_pred_modes[3] && !CHROMA444 )
3556             x264_log( h, X264_LOG_INFO, "i8c dc,h,v,p: %2.0f%% %2.0f%% %2.0f%% %2.0f%%\n",
3557                       fixed_pred_modes[3][0] * 100.0 / sum_pred_modes[3],
3558                       fixed_pred_modes[3][1] * 100.0 / sum_pred_modes[3],
3559                       fixed_pred_modes[3][2] * 100.0 / sum_pred_modes[3],
3560                       fixed_pred_modes[3][3] * 100.0 / sum_pred_modes[3] );
3561
3562         if( h->param.analyse.i_weighted_pred >= X264_WEIGHTP_SIMPLE && h->stat.i_frame_count[SLICE_TYPE_P] > 0 )
3563             x264_log( h, X264_LOG_INFO, "Weighted P-Frames: Y:%.1f%% UV:%.1f%%\n",
3564                       h->stat.i_wpred[0] * 100.0 / h->stat.i_frame_count[SLICE_TYPE_P],
3565                       h->stat.i_wpred[1] * 100.0 / h->stat.i_frame_count[SLICE_TYPE_P] );
3566
3567         for( int i_list = 0; i_list < 2; i_list++ )
3568             for( int i_slice = 0; i_slice < 2; i_slice++ )
3569             {
3570                 char *p = buf;
3571                 int64_t i_den = 0;
3572                 int i_max = 0;
3573                 for( int i = 0; i < X264_REF_MAX*2; i++ )
3574                     if( h->stat.i_mb_count_ref[i_slice][i_list][i] )
3575                     {
3576                         i_den += h->stat.i_mb_count_ref[i_slice][i_list][i];
3577                         i_max = i;
3578                     }
3579                 if( i_max == 0 )
3580                     continue;
3581                 for( int i = 0; i <= i_max; i++ )
3582                     p += sprintf( p, " %4.1f%%", 100. * h->stat.i_mb_count_ref[i_slice][i_list][i] / i_den );
3583                 x264_log( h, X264_LOG_INFO, "ref %c L%d:%s\n", "PB"[i_slice], i_list, buf );
3584             }
3585
3586         if( h->param.analyse.b_ssim )
3587         {
3588             float ssim = SUM3( h->stat.f_ssim_mean_y ) / duration;
3589             x264_log( h, X264_LOG_INFO, "SSIM Mean Y:%.7f (%6.3fdb)\n", ssim, x264_ssim( ssim ) );
3590         }
3591         if( h->param.analyse.b_psnr )
3592         {
3593             x264_log( h, X264_LOG_INFO,
3594                       "PSNR Mean Y:%6.3f U:%6.3f V:%6.3f Avg:%6.3f Global:%6.3f kb/s:%.2f\n",
3595                       SUM3( h->stat.f_psnr_mean_y ) / duration,
3596                       SUM3( h->stat.f_psnr_mean_u ) / duration,
3597                       SUM3( h->stat.f_psnr_mean_v ) / duration,
3598                       SUM3( h->stat.f_psnr_average ) / duration,
3599                       x264_psnr( SUM3( h->stat.f_ssd_global ), duration * i_yuv_size ),
3600                       f_bitrate );
3601         }
3602         else
3603             x264_log( h, X264_LOG_INFO, "kb/s:%.2f\n", f_bitrate );
3604     }
3605
3606     /* rc */
3607     x264_ratecontrol_delete( h );
3608
3609     /* param */
3610     if( h->param.rc.psz_stat_out )
3611         free( h->param.rc.psz_stat_out );
3612     if( h->param.rc.psz_stat_in )
3613         free( h->param.rc.psz_stat_in );
3614
3615     x264_cqm_delete( h );
3616     x264_free( h->nal_buffer );
3617     x264_analyse_free_costs( h );
3618
3619     if( h->i_thread_frames > 1)
3620         h = h->thread[h->i_thread_phase];
3621
3622     /* frames */
3623     x264_frame_delete_list( h->frames.unused[0] );
3624     x264_frame_delete_list( h->frames.unused[1] );
3625     x264_frame_delete_list( h->frames.current );
3626     x264_frame_delete_list( h->frames.blank_unused );
3627
3628     h = h->thread[0];
3629
3630     for( int i = 0; i < h->i_thread_frames; i++ )
3631         if( h->thread[i]->b_thread_active )
3632             for( int j = 0; j < h->thread[i]->i_ref[0]; j++ )
3633                 if( h->thread[i]->fref[0][j] && h->thread[i]->fref[0][j]->b_duplicate )
3634                     x264_frame_delete( h->thread[i]->fref[0][j] );
3635
3636     for( int i = h->param.i_threads - 1; i >= 0; i-- )
3637     {
3638         x264_frame_t **frame;
3639
3640         if( !h->param.b_sliced_threads || i == 0 )
3641         {
3642             for( frame = h->thread[i]->frames.reference; *frame; frame++ )
3643             {
3644                 assert( (*frame)->i_reference_count > 0 );
3645                 (*frame)->i_reference_count--;
3646                 if( (*frame)->i_reference_count == 0 )
3647                     x264_frame_delete( *frame );
3648             }
3649             frame = &h->thread[i]->fdec;
3650             if( *frame )
3651             {
3652                 assert( (*frame)->i_reference_count > 0 );
3653                 (*frame)->i_reference_count--;
3654                 if( (*frame)->i_reference_count == 0 )
3655                     x264_frame_delete( *frame );
3656             }
3657             x264_macroblock_cache_free( h->thread[i] );
3658         }
3659         x264_macroblock_thread_free( h->thread[i], 0 );
3660         x264_free( h->thread[i]->out.p_bitstream );
3661         x264_free( h->thread[i]->out.nal);
3662         x264_free( h->thread[i] );
3663     }
3664 }
3665
3666 int x264_encoder_delayed_frames( x264_t *h )
3667 {
3668     int delayed_frames = 0;
3669     if( h->i_thread_frames > 1 )
3670     {
3671         for( int i = 0; i < h->i_thread_frames; i++ )
3672             delayed_frames += h->thread[i]->b_thread_active;
3673         h = h->thread[h->i_thread_phase];
3674     }
3675     for( int i = 0; h->frames.current[i]; i++ )
3676         delayed_frames++;
3677     x264_pthread_mutex_lock( &h->lookahead->ofbuf.mutex );
3678     x264_pthread_mutex_lock( &h->lookahead->ifbuf.mutex );
3679     x264_pthread_mutex_lock( &h->lookahead->next.mutex );
3680     delayed_frames += h->lookahead->ifbuf.i_size + h->lookahead->next.i_size + h->lookahead->ofbuf.i_size;
3681     x264_pthread_mutex_unlock( &h->lookahead->next.mutex );
3682     x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
3683     x264_pthread_mutex_unlock( &h->lookahead->ofbuf.mutex );
3684     return delayed_frames;
3685 }
3686
3687 int x264_encoder_maximum_delayed_frames( x264_t *h )
3688 {
3689     return h->frames.i_delay;
3690 }