Add slices-max feature
[x262.git] / encoder / encoder.c
1 /*****************************************************************************
2  * encoder.c: top-level encoder functions
3  *****************************************************************************
4  * Copyright (C) 2003-2013 x264 project
5  *
6  * Authors: Laurent Aimar <fenrir@via.ecp.fr>
7  *          Loren Merritt <lorenm@u.washington.edu>
8  *          Jason Garrett-Glaser <darkshikari@gmail.com>
9  *
10  * This program is free software; you can redistribute it and/or modify
11  * it under the terms of the GNU General Public License as published by
12  * the Free Software Foundation; either version 2 of the License, or
13  * (at your option) any later version.
14  *
15  * This program is distributed in the hope that it will be useful,
16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
18  * GNU General Public License for more details.
19  *
20  * You should have received a copy of the GNU General Public License
21  * along with this program; if not, write to the Free Software
22  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02111, USA.
23  *
24  * This program is also available under a commercial proprietary license.
25  * For more information, contact us at licensing@x264.com.
26  *****************************************************************************/
27
28 #include "common/common.h"
29
30 #include "set.h"
31 #include "analyse.h"
32 #include "ratecontrol.h"
33 #include "macroblock.h"
34 #include "me.h"
35
36 #if HAVE_VISUALIZE
37 #include "common/visualize.h"
38 #endif
39
40 //#define DEBUG_MB_TYPE
41
42 #define bs_write_ue bs_write_ue_big
43
44 static int x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
45                                    x264_nal_t **pp_nal, int *pi_nal,
46                                    x264_picture_t *pic_out );
47
48 /****************************************************************************
49  *
50  ******************************* x264 libs **********************************
51  *
52  ****************************************************************************/
53 static double x264_psnr( double sqe, double size )
54 {
55     double mse = sqe / (PIXEL_MAX*PIXEL_MAX * size);
56     if( mse <= 0.0000000001 ) /* Max 100dB */
57         return 100;
58
59     return -10.0 * log10( mse );
60 }
61
62 static double x264_ssim( double ssim )
63 {
64     double inv_ssim = 1 - ssim;
65     if( inv_ssim <= 0.0000000001 ) /* Max 100dB */
66         return 100;
67
68     return -10.0 * log10( inv_ssim );
69 }
70
71 static int x264_threadpool_wait_all( x264_t *h )
72 {
73     for( int i = 0; i < h->param.i_threads; i++ )
74         if( h->thread[i]->b_thread_active )
75         {
76             h->thread[i]->b_thread_active = 0;
77             if( (intptr_t)x264_threadpool_wait( h->threadpool, h->thread[i] ) < 0 )
78                 return -1;
79         }
80     return 0;
81 }
82
83 static void x264_frame_dump( x264_t *h )
84 {
85     FILE *f = fopen( h->param.psz_dump_yuv, "r+b" );
86     if( !f )
87         return;
88
89     /* Wait for the threads to finish deblocking */
90     if( h->param.b_sliced_threads )
91         x264_threadpool_wait_all( h );
92
93     /* Write the frame in display order */
94     int frame_size = FRAME_SIZE( h->param.i_height * h->param.i_width * sizeof(pixel) );
95     fseek( f, (uint64_t)h->fdec->i_frame * frame_size, SEEK_SET );
96     for( int p = 0; p < (CHROMA444 ? 3 : 1); p++ )
97         for( int y = 0; y < h->param.i_height; y++ )
98             fwrite( &h->fdec->plane[p][y*h->fdec->i_stride[p]], sizeof(pixel), h->param.i_width, f );
99     if( !CHROMA444 )
100     {
101         int cw = h->param.i_width>>1;
102         int ch = h->param.i_height>>CHROMA_V_SHIFT;
103         pixel *planeu = x264_malloc( (cw*ch*2+32)*sizeof(pixel) );
104         pixel *planev = planeu + cw*ch + 16;
105         h->mc.plane_copy_deinterleave( planeu, cw, planev, cw, h->fdec->plane[1], h->fdec->i_stride[1], cw, ch );
106         fwrite( planeu, 1, cw*ch*sizeof(pixel), f );
107         fwrite( planev, 1, cw*ch*sizeof(pixel), f );
108         x264_free( planeu );
109     }
110     fclose( f );
111 }
112
113 /* Fill "default" values */
114 static void x264_slice_header_init( x264_t *h, x264_slice_header_t *sh,
115                                     x264_sps_t *sps, x264_pps_t *pps,
116                                     int i_idr_pic_id, int i_frame, int i_qp )
117 {
118     x264_param_t *param = &h->param;
119
120     /* First we fill all fields */
121     sh->sps = sps;
122     sh->pps = pps;
123
124     sh->i_first_mb  = 0;
125     sh->i_last_mb   = h->mb.i_mb_count - 1;
126     sh->i_pps_id    = pps->i_id;
127
128     sh->i_frame_num = i_frame;
129
130     sh->b_mbaff = PARAM_INTERLACED;
131     sh->b_field_pic = 0;    /* no field support for now */
132     sh->b_bottom_field = 0; /* not yet used */
133
134     sh->i_idr_pic_id = i_idr_pic_id;
135
136     /* poc stuff, fixed later */
137     sh->i_poc = 0;
138     sh->i_delta_poc_bottom = 0;
139     sh->i_delta_poc[0] = 0;
140     sh->i_delta_poc[1] = 0;
141
142     sh->i_redundant_pic_cnt = 0;
143
144     h->mb.b_direct_auto_write = h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO
145                                 && h->param.i_bframe
146                                 && ( h->param.rc.b_stat_write || !h->param.rc.b_stat_read );
147
148     if( !h->mb.b_direct_auto_read && sh->i_type == SLICE_TYPE_B )
149     {
150         if( h->fref[1][0]->i_poc_l0ref0 == h->fref[0][0]->i_poc )
151         {
152             if( h->mb.b_direct_auto_write )
153                 sh->b_direct_spatial_mv_pred = ( h->stat.i_direct_score[1] > h->stat.i_direct_score[0] );
154             else
155                 sh->b_direct_spatial_mv_pred = ( param->analyse.i_direct_mv_pred == X264_DIRECT_PRED_SPATIAL );
156         }
157         else
158         {
159             h->mb.b_direct_auto_write = 0;
160             sh->b_direct_spatial_mv_pred = 1;
161         }
162     }
163     /* else b_direct_spatial_mv_pred was read from the 2pass statsfile */
164
165     sh->b_num_ref_idx_override = 0;
166     sh->i_num_ref_idx_l0_active = 1;
167     sh->i_num_ref_idx_l1_active = 1;
168
169     sh->b_ref_pic_list_reordering[0] = h->b_ref_reorder[0];
170     sh->b_ref_pic_list_reordering[1] = h->b_ref_reorder[1];
171
172     /* If the ref list isn't in the default order, construct reordering header */
173     for( int list = 0; list < 2; list++ )
174     {
175         if( sh->b_ref_pic_list_reordering[list] )
176         {
177             int pred_frame_num = i_frame;
178             for( int i = 0; i < h->i_ref[list]; i++ )
179             {
180                 int diff = h->fref[list][i]->i_frame_num - pred_frame_num;
181                 sh->ref_pic_list_order[list][i].idc = ( diff > 0 );
182                 sh->ref_pic_list_order[list][i].arg = (abs(diff) - 1) & ((1 << sps->i_log2_max_frame_num) - 1);
183                 pred_frame_num = h->fref[list][i]->i_frame_num;
184             }
185         }
186     }
187
188     sh->i_cabac_init_idc = param->i_cabac_init_idc;
189
190     sh->i_qp = SPEC_QP(i_qp);
191     sh->i_qp_delta = sh->i_qp - pps->i_pic_init_qp;
192     sh->b_sp_for_swidth = 0;
193     sh->i_qs_delta = 0;
194
195     int deblock_thresh = i_qp + 2 * X264_MIN(param->i_deblocking_filter_alphac0, param->i_deblocking_filter_beta);
196     /* If effective qp <= 15, deblocking would have no effect anyway */
197     if( param->b_deblocking_filter && (h->mb.b_variable_qp || 15 < deblock_thresh ) )
198         sh->i_disable_deblocking_filter_idc = param->b_sliced_threads ? 2 : 0;
199     else
200         sh->i_disable_deblocking_filter_idc = 1;
201     sh->i_alpha_c0_offset = param->i_deblocking_filter_alphac0 << 1;
202     sh->i_beta_offset = param->i_deblocking_filter_beta << 1;
203 }
204
205 static void x264_slice_header_write( bs_t *s, x264_slice_header_t *sh, int i_nal_ref_idc )
206 {
207     if( sh->b_mbaff )
208     {
209         int first_x = sh->i_first_mb % sh->sps->i_mb_width;
210         int first_y = sh->i_first_mb / sh->sps->i_mb_width;
211         assert( (first_y&1) == 0 );
212         bs_write_ue( s, (2*first_x + sh->sps->i_mb_width*(first_y&~1) + (first_y&1)) >> 1 );
213     }
214     else
215         bs_write_ue( s, sh->i_first_mb );
216
217     bs_write_ue( s, sh->i_type + 5 );   /* same type things */
218     bs_write_ue( s, sh->i_pps_id );
219     bs_write( s, sh->sps->i_log2_max_frame_num, sh->i_frame_num & ((1<<sh->sps->i_log2_max_frame_num)-1) );
220
221     if( !sh->sps->b_frame_mbs_only )
222     {
223         bs_write1( s, sh->b_field_pic );
224         if( sh->b_field_pic )
225             bs_write1( s, sh->b_bottom_field );
226     }
227
228     if( sh->i_idr_pic_id >= 0 ) /* NAL IDR */
229         bs_write_ue( s, sh->i_idr_pic_id );
230
231     if( sh->sps->i_poc_type == 0 )
232     {
233         bs_write( s, sh->sps->i_log2_max_poc_lsb, sh->i_poc & ((1<<sh->sps->i_log2_max_poc_lsb)-1) );
234         if( sh->pps->b_pic_order && !sh->b_field_pic )
235             bs_write_se( s, sh->i_delta_poc_bottom );
236     }
237
238     if( sh->pps->b_redundant_pic_cnt )
239         bs_write_ue( s, sh->i_redundant_pic_cnt );
240
241     if( sh->i_type == SLICE_TYPE_B )
242         bs_write1( s, sh->b_direct_spatial_mv_pred );
243
244     if( sh->i_type == SLICE_TYPE_P || sh->i_type == SLICE_TYPE_B )
245     {
246         bs_write1( s, sh->b_num_ref_idx_override );
247         if( sh->b_num_ref_idx_override )
248         {
249             bs_write_ue( s, sh->i_num_ref_idx_l0_active - 1 );
250             if( sh->i_type == SLICE_TYPE_B )
251                 bs_write_ue( s, sh->i_num_ref_idx_l1_active - 1 );
252         }
253     }
254
255     /* ref pic list reordering */
256     if( sh->i_type != SLICE_TYPE_I )
257     {
258         bs_write1( s, sh->b_ref_pic_list_reordering[0] );
259         if( sh->b_ref_pic_list_reordering[0] )
260         {
261             for( int i = 0; i < sh->i_num_ref_idx_l0_active; i++ )
262             {
263                 bs_write_ue( s, sh->ref_pic_list_order[0][i].idc );
264                 bs_write_ue( s, sh->ref_pic_list_order[0][i].arg );
265             }
266             bs_write_ue( s, 3 );
267         }
268     }
269     if( sh->i_type == SLICE_TYPE_B )
270     {
271         bs_write1( s, sh->b_ref_pic_list_reordering[1] );
272         if( sh->b_ref_pic_list_reordering[1] )
273         {
274             for( int i = 0; i < sh->i_num_ref_idx_l1_active; i++ )
275             {
276                 bs_write_ue( s, sh->ref_pic_list_order[1][i].idc );
277                 bs_write_ue( s, sh->ref_pic_list_order[1][i].arg );
278             }
279             bs_write_ue( s, 3 );
280         }
281     }
282
283     sh->b_weighted_pred = 0;
284     if( sh->pps->b_weighted_pred && sh->i_type == SLICE_TYPE_P )
285     {
286         sh->b_weighted_pred = sh->weight[0][0].weightfn || sh->weight[0][1].weightfn || sh->weight[0][2].weightfn;
287         /* pred_weight_table() */
288         bs_write_ue( s, sh->weight[0][0].i_denom );
289         bs_write_ue( s, sh->weight[0][1].i_denom );
290         for( int i = 0; i < sh->i_num_ref_idx_l0_active; i++ )
291         {
292             int luma_weight_l0_flag = !!sh->weight[i][0].weightfn;
293             int chroma_weight_l0_flag = !!sh->weight[i][1].weightfn || !!sh->weight[i][2].weightfn;
294             bs_write1( s, luma_weight_l0_flag );
295             if( luma_weight_l0_flag )
296             {
297                 bs_write_se( s, sh->weight[i][0].i_scale );
298                 bs_write_se( s, sh->weight[i][0].i_offset );
299             }
300             bs_write1( s, chroma_weight_l0_flag );
301             if( chroma_weight_l0_flag )
302             {
303                 for( int j = 1; j < 3; j++ )
304                 {
305                     bs_write_se( s, sh->weight[i][j].i_scale );
306                     bs_write_se( s, sh->weight[i][j].i_offset );
307                 }
308             }
309         }
310     }
311     else if( sh->pps->b_weighted_bipred == 1 && sh->i_type == SLICE_TYPE_B )
312     {
313       /* TODO */
314     }
315
316     if( i_nal_ref_idc != 0 )
317     {
318         if( sh->i_idr_pic_id >= 0 )
319         {
320             bs_write1( s, 0 );  /* no output of prior pics flag */
321             bs_write1( s, 0 );  /* long term reference flag */
322         }
323         else
324         {
325             bs_write1( s, sh->i_mmco_command_count > 0 ); /* adaptive_ref_pic_marking_mode_flag */
326             if( sh->i_mmco_command_count > 0 )
327             {
328                 for( int i = 0; i < sh->i_mmco_command_count; i++ )
329                 {
330                     bs_write_ue( s, 1 ); /* mark short term ref as unused */
331                     bs_write_ue( s, sh->mmco[i].i_difference_of_pic_nums - 1 );
332                 }
333                 bs_write_ue( s, 0 ); /* end command list */
334             }
335         }
336     }
337
338     if( sh->pps->b_cabac && sh->i_type != SLICE_TYPE_I )
339         bs_write_ue( s, sh->i_cabac_init_idc );
340
341     bs_write_se( s, sh->i_qp_delta );      /* slice qp delta */
342
343     if( sh->pps->b_deblocking_filter_control )
344     {
345         bs_write_ue( s, sh->i_disable_deblocking_filter_idc );
346         if( sh->i_disable_deblocking_filter_idc != 1 )
347         {
348             bs_write_se( s, sh->i_alpha_c0_offset >> 1 );
349             bs_write_se( s, sh->i_beta_offset >> 1 );
350         }
351     }
352 }
353
354 /* If we are within a reasonable distance of the end of the memory allocated for the bitstream, */
355 /* reallocate, adding an arbitrary amount of space. */
356 static int x264_bitstream_check_buffer( x264_t *h )
357 {
358     uint8_t *bs_bak = h->out.p_bitstream;
359     int max_row_size = (2500 << SLICE_MBAFF) * h->mb.i_mb_width;
360     if( (h->param.b_cabac && (h->cabac.p_end - h->cabac.p < max_row_size)) ||
361         (h->out.bs.p_end - h->out.bs.p < max_row_size) )
362     {
363         h->out.i_bitstream += max_row_size;
364         CHECKED_MALLOC( h->out.p_bitstream, h->out.i_bitstream );
365         h->mc.memcpy_aligned( h->out.p_bitstream, bs_bak, (h->out.i_bitstream - max_row_size) & ~15 );
366         intptr_t delta = h->out.p_bitstream - bs_bak;
367
368         h->out.bs.p_start += delta;
369         h->out.bs.p += delta;
370         h->out.bs.p_end = h->out.p_bitstream + h->out.i_bitstream;
371
372         h->cabac.p_start += delta;
373         h->cabac.p += delta;
374         h->cabac.p_end = h->out.p_bitstream + h->out.i_bitstream;
375
376         for( int i = 0; i <= h->out.i_nal; i++ )
377             h->out.nal[i].p_payload += delta;
378         x264_free( bs_bak );
379     }
380     return 0;
381 fail:
382     x264_free( bs_bak );
383     return -1;
384 }
385
386 #if HAVE_THREAD
387 static void x264_encoder_thread_init( x264_t *h )
388 {
389     if( h->param.i_sync_lookahead )
390         x264_lower_thread_priority( 10 );
391
392 #if HAVE_MMX
393     /* Misalign mask has to be set separately for each thread. */
394     if( h->param.cpu&X264_CPU_SSE_MISALIGN )
395         x264_cpu_mask_misalign_sse();
396 #endif
397 }
398
399 static void x264_lookahead_thread_init( x264_t *h )
400 {
401 #if HAVE_MMX
402     /* Misalign mask has to be set separately for each thread. */
403     if( h->param.cpu&X264_CPU_SSE_MISALIGN )
404         x264_cpu_mask_misalign_sse();
405 #endif
406 }
407 #endif
408
409 /****************************************************************************
410  *
411  ****************************************************************************
412  ****************************** External API*********************************
413  ****************************************************************************
414  *
415  ****************************************************************************/
416
417 static int x264_validate_parameters( x264_t *h, int b_open )
418 {
419 #if HAVE_MMX
420     if( b_open )
421     {
422         int cpuflags = x264_cpu_detect();
423         int fail = 0;
424 #ifdef __SSE__
425         if( !(cpuflags & X264_CPU_SSE) )
426         {
427             x264_log( h, X264_LOG_ERROR, "your cpu does not support SSE1, but x264 was compiled with asm\n");
428             fail = 1;
429         }
430 #else
431         if( !(cpuflags & X264_CPU_MMX2) )
432         {
433             x264_log( h, X264_LOG_ERROR, "your cpu does not support MMXEXT, but x264 was compiled with asm\n");
434             fail = 1;
435         }
436 #endif
437         if( !fail && !(cpuflags & X264_CPU_CMOV) )
438         {
439             x264_log( h, X264_LOG_ERROR, "your cpu does not support CMOV, but x264 was compiled with asm\n");
440             fail = 1;
441         }
442         if( fail )
443         {
444             x264_log( h, X264_LOG_ERROR, "to run x264, recompile without asm (configure --disable-asm)\n");
445             return -1;
446         }
447     }
448 #endif
449
450 #if HAVE_INTERLACED
451     h->param.b_interlaced = !!PARAM_INTERLACED;
452 #else
453     if( h->param.b_interlaced )
454     {
455         x264_log( h, X264_LOG_ERROR, "not compiled with interlaced support\n" );
456         return -1;
457     }
458 #endif
459
460     if( h->param.i_width <= 0 || h->param.i_height <= 0 )
461     {
462         x264_log( h, X264_LOG_ERROR, "invalid width x height (%dx%d)\n",
463                   h->param.i_width, h->param.i_height );
464         return -1;
465     }
466
467     int i_csp = h->param.i_csp & X264_CSP_MASK;
468 #if X264_CHROMA_FORMAT
469     if( CHROMA_FORMAT != CHROMA_420 && i_csp >= X264_CSP_I420 && i_csp <= X264_CSP_NV12 )
470     {
471         x264_log( h, X264_LOG_ERROR, "not compiled with 4:2:0 support\n" );
472         return -1;
473     }
474     else if( CHROMA_FORMAT != CHROMA_422 && i_csp >= X264_CSP_I422 && i_csp <= X264_CSP_NV16 )
475     {
476         x264_log( h, X264_LOG_ERROR, "not compiled with 4:2:2 support\n" );
477         return -1;
478     }
479     else if( CHROMA_FORMAT != CHROMA_444 && i_csp >= X264_CSP_I444 && i_csp <= X264_CSP_RGB )
480     {
481         x264_log( h, X264_LOG_ERROR, "not compiled with 4:4:4 support\n" );
482         return -1;
483     }
484 #endif
485     if( i_csp <= X264_CSP_NONE || i_csp >= X264_CSP_MAX )
486     {
487         x264_log( h, X264_LOG_ERROR, "invalid CSP (only I420/YV12/NV12/I422/YV16/NV16/I444/YV24/BGR/BGRA/RGB supported)\n" );
488         return -1;
489     }
490
491     if( i_csp < X264_CSP_I444 && h->param.i_width % 2 )
492     {
493         x264_log( h, X264_LOG_ERROR, "width not divisible by 2 (%dx%d)\n",
494                   h->param.i_width, h->param.i_height );
495         return -1;
496     }
497
498     if( i_csp < X264_CSP_I422 && PARAM_INTERLACED && h->param.i_height % 4 )
499     {
500         x264_log( h, X264_LOG_ERROR, "height not divisible by 4 (%dx%d)\n",
501                   h->param.i_width, h->param.i_height );
502         return -1;
503     }
504
505     if( (i_csp < X264_CSP_I422 || PARAM_INTERLACED) && h->param.i_height % 2 )
506     {
507         x264_log( h, X264_LOG_ERROR, "height not divisible by 2 (%dx%d)\n",
508                   h->param.i_width, h->param.i_height );
509         return -1;
510     }
511
512     if( (h->param.crop_rect.i_left + h->param.crop_rect.i_right ) >= h->param.i_width ||
513         (h->param.crop_rect.i_top  + h->param.crop_rect.i_bottom) >= h->param.i_height )
514     {
515         x264_log( h, X264_LOG_ERROR, "invalid crop-rect %u,%u,%u,%u\n", h->param.crop_rect.i_left,
516                   h->param.crop_rect.i_top, h->param.crop_rect.i_right,  h->param.crop_rect.i_bottom );
517         return -1;
518     }
519
520     if( h->param.i_threads == X264_THREADS_AUTO )
521         h->param.i_threads = x264_cpu_num_processors() * (h->param.b_sliced_threads?2:3)/2;
522     int max_sliced_threads = X264_MAX( 1, (h->param.i_height+15)/16 / 4 );
523     if( h->param.i_threads > 1 )
524     {
525 #if !HAVE_THREAD
526         x264_log( h, X264_LOG_WARNING, "not compiled with thread support!\n");
527         h->param.i_threads = 1;
528 #endif
529         /* Avoid absurdly small thread slices as they can reduce performance
530          * and VBV compliance.  Capped at an arbitrary 4 rows per thread. */
531         if( h->param.b_sliced_threads )
532             h->param.i_threads = X264_MIN( h->param.i_threads, max_sliced_threads );
533     }
534     h->param.i_threads = x264_clip3( h->param.i_threads, 1, X264_THREAD_MAX );
535     if( h->param.i_threads == 1 )
536     {
537         h->param.b_sliced_threads = 0;
538         h->param.i_lookahead_threads = 1;
539     }
540     h->i_thread_frames = h->param.b_sliced_threads ? 1 : h->param.i_threads;
541     if( h->i_thread_frames > 1 )
542         h->param.nalu_process = NULL;
543
544     h->param.i_keyint_max = x264_clip3( h->param.i_keyint_max, 1, X264_KEYINT_MAX_INFINITE );
545     if( h->param.i_keyint_max == 1 )
546     {
547         h->param.b_intra_refresh = 0;
548         h->param.analyse.i_weighted_pred = 0;
549     }
550
551     h->param.i_frame_packing = x264_clip3( h->param.i_frame_packing, -1, 5 );
552
553     /* Detect default ffmpeg settings and terminate with an error. */
554     if( b_open )
555     {
556         int score = 0;
557         score += h->param.analyse.i_me_range == 0;
558         score += h->param.rc.i_qp_step == 3;
559         score += h->param.i_keyint_max == 12;
560         score += h->param.rc.i_qp_min == 2;
561         score += h->param.rc.i_qp_max == 31;
562         score += h->param.rc.f_qcompress == 0.5;
563         score += fabs(h->param.rc.f_ip_factor - 1.25) < 0.01;
564         score += fabs(h->param.rc.f_pb_factor - 1.25) < 0.01;
565         score += h->param.analyse.inter == 0 && h->param.analyse.i_subpel_refine == 8;
566         if( score >= 5 )
567         {
568             x264_log( h, X264_LOG_ERROR, "broken ffmpeg default settings detected\n" );
569             x264_log( h, X264_LOG_ERROR, "use an encoding preset (e.g. -vpre medium)\n" );
570             x264_log( h, X264_LOG_ERROR, "preset usage: -vpre <speed> -vpre <profile>\n" );
571             x264_log( h, X264_LOG_ERROR, "speed presets are listed in x264 --help\n" );
572             x264_log( h, X264_LOG_ERROR, "profile is optional; x264 defaults to high\n" );
573             return -1;
574         }
575     }
576
577     if( h->param.rc.i_rc_method < 0 || h->param.rc.i_rc_method > 2 )
578     {
579         x264_log( h, X264_LOG_ERROR, "no ratecontrol method specified\n" );
580         return -1;
581     }
582     h->param.rc.f_rf_constant = x264_clip3f( h->param.rc.f_rf_constant, -QP_BD_OFFSET, 51 );
583     h->param.rc.f_rf_constant_max = x264_clip3f( h->param.rc.f_rf_constant_max, -QP_BD_OFFSET, 51 );
584     h->param.rc.i_qp_constant = x264_clip3( h->param.rc.i_qp_constant, 0, QP_MAX );
585     h->param.analyse.i_subpel_refine = x264_clip3( h->param.analyse.i_subpel_refine, 0, 11 );
586     h->param.rc.f_ip_factor = X264_MAX( h->param.rc.f_ip_factor, 0.01f );
587     h->param.rc.f_pb_factor = X264_MAX( h->param.rc.f_pb_factor, 0.01f );
588     if( h->param.rc.i_rc_method == X264_RC_CRF )
589     {
590         h->param.rc.i_qp_constant = h->param.rc.f_rf_constant + QP_BD_OFFSET;
591         h->param.rc.i_bitrate = 0;
592     }
593     if( b_open && (h->param.rc.i_rc_method == X264_RC_CQP || h->param.rc.i_rc_method == X264_RC_CRF)
594         && h->param.rc.i_qp_constant == 0 )
595     {
596         h->mb.b_lossless = 1;
597         h->param.i_cqm_preset = X264_CQM_FLAT;
598         h->param.psz_cqm_file = NULL;
599         h->param.rc.i_rc_method = X264_RC_CQP;
600         h->param.rc.f_ip_factor = 1;
601         h->param.rc.f_pb_factor = 1;
602         h->param.analyse.b_psnr = 0;
603         h->param.analyse.b_ssim = 0;
604         h->param.analyse.i_chroma_qp_offset = 0;
605         h->param.analyse.i_trellis = 0;
606         h->param.analyse.b_fast_pskip = 0;
607         h->param.analyse.i_noise_reduction = 0;
608         h->param.analyse.b_psy = 0;
609         h->param.i_bframe = 0;
610         /* 8x8dct is not useful without RD in CAVLC lossless */
611         if( !h->param.b_cabac && h->param.analyse.i_subpel_refine < 6 )
612             h->param.analyse.b_transform_8x8 = 0;
613     }
614     if( h->param.rc.i_rc_method == X264_RC_CQP )
615     {
616         float qp_p = h->param.rc.i_qp_constant;
617         float qp_i = qp_p - 6*log2f( h->param.rc.f_ip_factor );
618         float qp_b = qp_p + 6*log2f( h->param.rc.f_pb_factor );
619         h->param.rc.i_qp_min = x264_clip3( (int)(X264_MIN3( qp_p, qp_i, qp_b )), 0, QP_MAX );
620         h->param.rc.i_qp_max = x264_clip3( (int)(X264_MAX3( qp_p, qp_i, qp_b ) + .999), 0, QP_MAX );
621         h->param.rc.i_aq_mode = 0;
622         h->param.rc.b_mb_tree = 0;
623         h->param.rc.i_bitrate = 0;
624     }
625     h->param.rc.i_qp_max = x264_clip3( h->param.rc.i_qp_max, 0, QP_MAX );
626     h->param.rc.i_qp_min = x264_clip3( h->param.rc.i_qp_min, 0, h->param.rc.i_qp_max );
627     h->param.rc.i_qp_step = x264_clip3( h->param.rc.i_qp_step, 2, QP_MAX );
628     h->param.rc.i_bitrate = x264_clip3( h->param.rc.i_bitrate, 0, 2000000 );
629     if( h->param.rc.i_rc_method == X264_RC_ABR && !h->param.rc.i_bitrate )
630     {
631         x264_log( h, X264_LOG_ERROR, "bitrate not specified\n" );
632         return -1;
633     }
634     h->param.rc.i_vbv_buffer_size = x264_clip3( h->param.rc.i_vbv_buffer_size, 0, 2000000 );
635     h->param.rc.i_vbv_max_bitrate = x264_clip3( h->param.rc.i_vbv_max_bitrate, 0, 2000000 );
636     h->param.rc.f_vbv_buffer_init = x264_clip3f( h->param.rc.f_vbv_buffer_init, 0, 2000000 );
637     if( h->param.rc.i_vbv_buffer_size )
638     {
639         if( h->param.rc.i_rc_method == X264_RC_CQP )
640         {
641             x264_log( h, X264_LOG_WARNING, "VBV is incompatible with constant QP, ignored.\n" );
642             h->param.rc.i_vbv_max_bitrate = 0;
643             h->param.rc.i_vbv_buffer_size = 0;
644         }
645         else if( h->param.rc.i_vbv_max_bitrate == 0 )
646         {
647             if( h->param.rc.i_rc_method == X264_RC_ABR )
648             {
649                 x264_log( h, X264_LOG_WARNING, "VBV maxrate unspecified, assuming CBR\n" );
650                 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate;
651             }
652             else
653             {
654                 x264_log( h, X264_LOG_WARNING, "VBV bufsize set but maxrate unspecified, ignored\n" );
655                 h->param.rc.i_vbv_buffer_size = 0;
656             }
657         }
658         else if( h->param.rc.i_vbv_max_bitrate < h->param.rc.i_bitrate &&
659                  h->param.rc.i_rc_method == X264_RC_ABR )
660         {
661             x264_log( h, X264_LOG_WARNING, "max bitrate less than average bitrate, assuming CBR\n" );
662             h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate;
663         }
664     }
665     else if( h->param.rc.i_vbv_max_bitrate )
666     {
667         x264_log( h, X264_LOG_WARNING, "VBV maxrate specified, but no bufsize, ignored\n" );
668         h->param.rc.i_vbv_max_bitrate = 0;
669     }
670
671     h->param.i_slice_max_size = X264_MAX( h->param.i_slice_max_size, 0 );
672     h->param.i_slice_max_mbs = X264_MAX( h->param.i_slice_max_mbs, 0 );
673     h->param.i_slice_min_mbs = X264_MAX( h->param.i_slice_min_mbs, 0 );
674     if( h->param.i_slice_max_mbs )
675         h->param.i_slice_min_mbs = X264_MIN( h->param.i_slice_min_mbs, h->param.i_slice_max_mbs/2 );
676     else if( !h->param.i_slice_max_size )
677         h->param.i_slice_min_mbs = 0;
678     if( PARAM_INTERLACED && h->param.i_slice_min_mbs )
679     {
680         x264_log( h, X264_LOG_WARNING, "interlace + slice-min-mbs is not implemented\n" );
681         h->param.i_slice_min_mbs = 0;
682     }
683     int mb_width = (h->param.i_width+15)/16;
684     if( h->param.i_slice_min_mbs > mb_width )
685     {
686         x264_log( h, X264_LOG_WARNING, "slice-min-mbs > row mb size (%d) not implemented\n", mb_width );
687         h->param.i_slice_min_mbs = mb_width;
688     }
689
690     int max_slices = (h->param.i_height+((16<<PARAM_INTERLACED)-1))/(16<<PARAM_INTERLACED);
691     if( h->param.b_sliced_threads )
692         h->param.i_slice_count = x264_clip3( h->param.i_threads, 0, max_slices );
693     else
694     {
695         h->param.i_slice_count = x264_clip3( h->param.i_slice_count, 0, max_slices );
696         if( h->param.i_slice_max_mbs || h->param.i_slice_max_size )
697             h->param.i_slice_count = 0;
698     }
699     h->param.i_slice_count_max = X264_MAX( h->param.i_slice_count, h->param.i_slice_count_max );
700
701     if( h->param.b_bluray_compat )
702     {
703         h->param.i_bframe_pyramid = X264_MIN( X264_B_PYRAMID_STRICT, h->param.i_bframe_pyramid );
704         h->param.i_bframe = X264_MIN( h->param.i_bframe, 3 );
705         h->param.b_aud = 1;
706         h->param.i_nal_hrd = X264_MAX( h->param.i_nal_hrd, X264_NAL_HRD_VBR );
707         h->param.i_slice_max_size = 0;
708         h->param.i_slice_max_mbs = 0;
709         h->param.b_intra_refresh = 0;
710         h->param.i_frame_reference = X264_MIN( h->param.i_frame_reference, 6 );
711         h->param.i_dpb_size = X264_MIN( h->param.i_dpb_size, 6 );
712         /* Don't use I-frames, because Blu-ray treats them the same as IDR. */
713         h->param.i_keyint_min = 1;
714         /* Due to the proliferation of broken players that don't handle dupes properly. */
715         h->param.analyse.i_weighted_pred = X264_MIN( h->param.analyse.i_weighted_pred, X264_WEIGHTP_SIMPLE );
716         if( h->param.b_fake_interlaced )
717             h->param.b_pic_struct = 1;
718     }
719
720     h->param.i_frame_reference = x264_clip3( h->param.i_frame_reference, 1, X264_REF_MAX );
721     h->param.i_dpb_size = x264_clip3( h->param.i_dpb_size, 1, X264_REF_MAX );
722     if( h->param.i_scenecut_threshold < 0 )
723         h->param.i_scenecut_threshold = 0;
724     h->param.analyse.i_direct_mv_pred = x264_clip3( h->param.analyse.i_direct_mv_pred, X264_DIRECT_PRED_NONE, X264_DIRECT_PRED_AUTO );
725     if( !h->param.analyse.i_subpel_refine && h->param.analyse.i_direct_mv_pred > X264_DIRECT_PRED_SPATIAL )
726     {
727         x264_log( h, X264_LOG_WARNING, "subme=0 + direct=temporal is not supported\n" );
728         h->param.analyse.i_direct_mv_pred = X264_DIRECT_PRED_SPATIAL;
729     }
730     h->param.i_bframe = x264_clip3( h->param.i_bframe, 0, X264_MIN( X264_BFRAME_MAX, h->param.i_keyint_max-1 ) );
731     h->param.i_bframe_bias = x264_clip3( h->param.i_bframe_bias, -90, 100 );
732     if( h->param.i_bframe <= 1 )
733         h->param.i_bframe_pyramid = X264_B_PYRAMID_NONE;
734     h->param.i_bframe_pyramid = x264_clip3( h->param.i_bframe_pyramid, X264_B_PYRAMID_NONE, X264_B_PYRAMID_NORMAL );
735     h->param.i_bframe_adaptive = x264_clip3( h->param.i_bframe_adaptive, X264_B_ADAPT_NONE, X264_B_ADAPT_TRELLIS );
736     if( !h->param.i_bframe )
737     {
738         h->param.i_bframe_adaptive = X264_B_ADAPT_NONE;
739         h->param.analyse.i_direct_mv_pred = 0;
740         h->param.analyse.b_weighted_bipred = 0;
741         h->param.b_open_gop = 0;
742     }
743     if( h->param.b_intra_refresh && h->param.i_bframe_pyramid == X264_B_PYRAMID_NORMAL )
744     {
745         x264_log( h, X264_LOG_WARNING, "b-pyramid normal + intra-refresh is not supported\n" );
746         h->param.i_bframe_pyramid = X264_B_PYRAMID_STRICT;
747     }
748     if( h->param.b_intra_refresh && (h->param.i_frame_reference > 1 || h->param.i_dpb_size > 1) )
749     {
750         x264_log( h, X264_LOG_WARNING, "ref > 1 + intra-refresh is not supported\n" );
751         h->param.i_frame_reference = 1;
752         h->param.i_dpb_size = 1;
753     }
754     if( h->param.b_intra_refresh && h->param.b_open_gop )
755     {
756         x264_log( h, X264_LOG_WARNING, "intra-refresh is not compatible with open-gop\n" );
757         h->param.b_open_gop = 0;
758     }
759     if( !h->param.i_fps_num || !h->param.i_fps_den )
760     {
761         h->param.i_fps_num = 25;
762         h->param.i_fps_den = 1;
763     }
764     float fps = (float) h->param.i_fps_num / h->param.i_fps_den;
765     if( h->param.i_keyint_min == X264_KEYINT_MIN_AUTO )
766         h->param.i_keyint_min = X264_MIN( h->param.i_keyint_max / 10, fps );
767     h->param.i_keyint_min = x264_clip3( h->param.i_keyint_min, 1, h->param.i_keyint_max/2+1 );
768     h->param.rc.i_lookahead = x264_clip3( h->param.rc.i_lookahead, 0, X264_LOOKAHEAD_MAX );
769     {
770         int maxrate = X264_MAX( h->param.rc.i_vbv_max_bitrate, h->param.rc.i_bitrate );
771         float bufsize = maxrate ? (float)h->param.rc.i_vbv_buffer_size / maxrate : 0;
772         h->param.rc.i_lookahead = X264_MIN( h->param.rc.i_lookahead, X264_MAX( h->param.i_keyint_max, bufsize*fps ) );
773     }
774
775     if( !h->param.i_timebase_num || !h->param.i_timebase_den || !(h->param.b_vfr_input || h->param.b_pulldown) )
776     {
777         h->param.i_timebase_num = h->param.i_fps_den;
778         h->param.i_timebase_den = h->param.i_fps_num;
779     }
780
781     h->param.rc.f_qcompress = x264_clip3f( h->param.rc.f_qcompress, 0.0, 1.0 );
782     if( h->param.i_keyint_max == 1 || h->param.rc.f_qcompress == 1 )
783         h->param.rc.b_mb_tree = 0;
784     if( (!h->param.b_intra_refresh && h->param.i_keyint_max != X264_KEYINT_MAX_INFINITE) &&
785         !h->param.rc.i_lookahead && h->param.rc.b_mb_tree )
786     {
787         x264_log( h, X264_LOG_WARNING, "lookaheadless mb-tree requires intra refresh or infinite keyint\n" );
788         h->param.rc.b_mb_tree = 0;
789     }
790     if( b_open && h->param.rc.b_stat_read )
791         h->param.rc.i_lookahead = 0;
792 #if HAVE_THREAD
793     if( h->param.i_sync_lookahead < 0 )
794         h->param.i_sync_lookahead = h->param.i_bframe + 1;
795     h->param.i_sync_lookahead = X264_MIN( h->param.i_sync_lookahead, X264_LOOKAHEAD_MAX );
796     if( h->param.rc.b_stat_read || h->i_thread_frames == 1 )
797         h->param.i_sync_lookahead = 0;
798 #else
799     h->param.i_sync_lookahead = 0;
800 #endif
801
802     h->param.i_deblocking_filter_alphac0 = x264_clip3( h->param.i_deblocking_filter_alphac0, -6, 6 );
803     h->param.i_deblocking_filter_beta    = x264_clip3( h->param.i_deblocking_filter_beta, -6, 6 );
804     h->param.analyse.i_luma_deadzone[0] = x264_clip3( h->param.analyse.i_luma_deadzone[0], 0, 32 );
805     h->param.analyse.i_luma_deadzone[1] = x264_clip3( h->param.analyse.i_luma_deadzone[1], 0, 32 );
806
807     h->param.i_cabac_init_idc = x264_clip3( h->param.i_cabac_init_idc, 0, 2 );
808
809     if( h->param.i_cqm_preset < X264_CQM_FLAT || h->param.i_cqm_preset > X264_CQM_CUSTOM )
810         h->param.i_cqm_preset = X264_CQM_FLAT;
811
812     if( h->param.analyse.i_me_method < X264_ME_DIA ||
813         h->param.analyse.i_me_method > X264_ME_TESA )
814         h->param.analyse.i_me_method = X264_ME_HEX;
815     h->param.analyse.i_me_range = x264_clip3( h->param.analyse.i_me_range, 4, 1024 );
816     if( h->param.analyse.i_me_range > 16 && h->param.analyse.i_me_method <= X264_ME_HEX )
817         h->param.analyse.i_me_range = 16;
818     if( h->param.analyse.i_me_method == X264_ME_TESA &&
819         (h->mb.b_lossless || h->param.analyse.i_subpel_refine <= 1) )
820         h->param.analyse.i_me_method = X264_ME_ESA;
821     h->param.analyse.b_mixed_references = h->param.analyse.b_mixed_references && h->param.i_frame_reference > 1;
822     h->param.analyse.inter &= X264_ANALYSE_PSUB16x16|X264_ANALYSE_PSUB8x8|X264_ANALYSE_BSUB16x16|
823                               X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
824     h->param.analyse.intra &= X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
825     if( !(h->param.analyse.inter & X264_ANALYSE_PSUB16x16) )
826         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
827     if( !h->param.analyse.b_transform_8x8 )
828     {
829         h->param.analyse.inter &= ~X264_ANALYSE_I8x8;
830         h->param.analyse.intra &= ~X264_ANALYSE_I8x8;
831     }
832     h->param.analyse.i_trellis = x264_clip3( h->param.analyse.i_trellis, 0, 2 );
833     h->param.rc.i_aq_mode = x264_clip3( h->param.rc.i_aq_mode, 0, 2 );
834     h->param.rc.f_aq_strength = x264_clip3f( h->param.rc.f_aq_strength, 0, 3 );
835     if( h->param.rc.f_aq_strength == 0 )
836         h->param.rc.i_aq_mode = 0;
837
838     if( h->param.i_log_level < X264_LOG_INFO )
839     {
840         h->param.analyse.b_psnr = 0;
841         h->param.analyse.b_ssim = 0;
842     }
843     /* Warn users trying to measure PSNR/SSIM with psy opts on. */
844     if( b_open && (h->param.analyse.b_psnr || h->param.analyse.b_ssim) )
845     {
846         char *s = NULL;
847
848         if( h->param.analyse.b_psy )
849         {
850             s = h->param.analyse.b_psnr ? "psnr" : "ssim";
851             x264_log( h, X264_LOG_WARNING, "--%s used with psy on: results will be invalid!\n", s );
852         }
853         else if( !h->param.rc.i_aq_mode && h->param.analyse.b_ssim )
854         {
855             x264_log( h, X264_LOG_WARNING, "--ssim used with AQ off: results will be invalid!\n" );
856             s = "ssim";
857         }
858         else if(  h->param.rc.i_aq_mode && h->param.analyse.b_psnr )
859         {
860             x264_log( h, X264_LOG_WARNING, "--psnr used with AQ on: results will be invalid!\n" );
861             s = "psnr";
862         }
863         if( s )
864             x264_log( h, X264_LOG_WARNING, "--tune %s should be used if attempting to benchmark %s!\n", s, s );
865     }
866
867     if( !h->param.analyse.b_psy )
868     {
869         h->param.analyse.f_psy_rd = 0;
870         h->param.analyse.f_psy_trellis = 0;
871     }
872     h->param.analyse.f_psy_rd = x264_clip3f( h->param.analyse.f_psy_rd, 0, 10 );
873     h->param.analyse.f_psy_trellis = x264_clip3f( h->param.analyse.f_psy_trellis, 0, 10 );
874     h->mb.i_psy_rd = h->param.analyse.i_subpel_refine >= 6 ? FIX8( h->param.analyse.f_psy_rd ) : 0;
875     h->mb.i_psy_trellis = h->param.analyse.i_trellis ? FIX8( h->param.analyse.f_psy_trellis / 4 ) : 0;
876     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -32, 32);
877     /* In 4:4:4 mode, chroma gets twice as much resolution, so we can halve its quality. */
878     if( b_open && i_csp >= X264_CSP_I444 && i_csp < X264_CSP_BGR && h->param.analyse.b_psy )
879         h->param.analyse.i_chroma_qp_offset += 6;
880     /* Psy RDO increases overall quantizers to improve the quality of luma--this indirectly hurts chroma quality */
881     /* so we lower the chroma QP offset to compensate */
882     if( b_open && h->mb.i_psy_rd )
883         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_rd < 0.25 ? 1 : 2;
884     /* Psy trellis has a similar effect. */
885     if( b_open && h->mb.i_psy_trellis )
886         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_trellis < 0.25 ? 1 : 2;
887     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -12, 12);
888     /* MB-tree requires AQ to be on, even if the strength is zero. */
889     if( !h->param.rc.i_aq_mode && h->param.rc.b_mb_tree )
890     {
891         h->param.rc.i_aq_mode = 1;
892         h->param.rc.f_aq_strength = 0;
893     }
894     h->param.analyse.i_noise_reduction = x264_clip3( h->param.analyse.i_noise_reduction, 0, 1<<16 );
895     if( h->param.analyse.i_subpel_refine >= 10 && (h->param.analyse.i_trellis != 2 || !h->param.rc.i_aq_mode) )
896         h->param.analyse.i_subpel_refine = 9;
897
898     {
899         const x264_level_t *l = x264_levels;
900         if( h->param.i_level_idc < 0 )
901         {
902             int maxrate_bak = h->param.rc.i_vbv_max_bitrate;
903             if( h->param.rc.i_rc_method == X264_RC_ABR && h->param.rc.i_vbv_buffer_size <= 0 )
904                 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate * 2;
905             x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
906             do h->param.i_level_idc = l->level_idc;
907                 while( l[1].level_idc && x264_validate_levels( h, 0 ) && l++ );
908             h->param.rc.i_vbv_max_bitrate = maxrate_bak;
909         }
910         else
911         {
912             while( l->level_idc && l->level_idc != h->param.i_level_idc )
913                 l++;
914             if( l->level_idc == 0 )
915             {
916                 x264_log( h, X264_LOG_ERROR, "invalid level_idc: %d\n", h->param.i_level_idc );
917                 return -1;
918             }
919         }
920         if( h->param.analyse.i_mv_range <= 0 )
921             h->param.analyse.i_mv_range = l->mv_range >> PARAM_INTERLACED;
922         else
923             h->param.analyse.i_mv_range = x264_clip3(h->param.analyse.i_mv_range, 32, 512 >> PARAM_INTERLACED);
924     }
925
926     h->param.analyse.i_weighted_pred = x264_clip3( h->param.analyse.i_weighted_pred, X264_WEIGHTP_NONE, X264_WEIGHTP_SMART );
927
928     if( h->param.i_lookahead_threads == X264_THREADS_AUTO )
929     {
930         if( h->param.b_sliced_threads )
931             h->param.i_lookahead_threads = h->param.i_threads;
932         else
933         {
934             /* If we're using much slower lookahead settings than encoding settings, it helps a lot to use
935              * more lookahead threads.  This typically happens in the first pass of a two-pass encode, so
936              * try to guess at this sort of case.
937              *
938              * Tuned by a little bit of real encoding with the various presets. */
939             int badapt = h->param.i_bframe_adaptive == X264_B_ADAPT_TRELLIS;
940             int subme = X264_MIN( h->param.analyse.i_subpel_refine / 3, 3 ) + (h->param.analyse.i_subpel_refine > 1);
941             int bframes = X264_MIN( (h->param.i_bframe - 1) / 3, 3 );
942
943             /* [b-adapt 0/1 vs 2][quantized subme][quantized bframes] */
944             static const uint8_t lookahead_thread_div[2][5][4] =
945             {{{6,6,6,6}, {3,3,3,3}, {4,4,4,4}, {6,6,6,6}, {12,12,12,12}},
946              {{3,2,1,1}, {2,1,1,1}, {4,3,2,1}, {6,4,3,2}, {12, 9, 6, 4}}};
947
948             h->param.i_lookahead_threads = h->param.i_threads / lookahead_thread_div[badapt][subme][bframes];
949             /* Since too many lookahead threads significantly degrades lookahead accuracy, limit auto
950              * lookahead threads to about 8 macroblock rows high each at worst.  This number is chosen
951              * pretty much arbitrarily. */
952             h->param.i_lookahead_threads = X264_MIN( h->param.i_lookahead_threads, h->param.i_height / 128 );
953         }
954     }
955     h->param.i_lookahead_threads = x264_clip3( h->param.i_lookahead_threads, 1, X264_MIN( max_sliced_threads, X264_LOOKAHEAD_THREAD_MAX ) );
956
957     if( PARAM_INTERLACED )
958     {
959         if( h->param.analyse.i_me_method >= X264_ME_ESA )
960         {
961             x264_log( h, X264_LOG_WARNING, "interlace + me=esa is not implemented\n" );
962             h->param.analyse.i_me_method = X264_ME_UMH;
963         }
964         if( h->param.analyse.i_weighted_pred > 0 )
965         {
966             x264_log( h, X264_LOG_WARNING, "interlace + weightp is not implemented\n" );
967             h->param.analyse.i_weighted_pred = X264_WEIGHTP_NONE;
968         }
969     }
970
971     if( !h->param.analyse.i_weighted_pred && h->param.rc.b_mb_tree && h->param.analyse.b_psy )
972         h->param.analyse.i_weighted_pred = X264_WEIGHTP_FAKE;
973
974     if( h->i_thread_frames > 1 )
975     {
976         int r = h->param.analyse.i_mv_range_thread;
977         int r2;
978         if( r <= 0 )
979         {
980             // half of the available space is reserved and divided evenly among the threads,
981             // the rest is allocated to whichever thread is far enough ahead to use it.
982             // reserving more space increases quality for some videos, but costs more time
983             // in thread synchronization.
984             int max_range = (h->param.i_height + X264_THREAD_HEIGHT) / h->i_thread_frames - X264_THREAD_HEIGHT;
985             r = max_range / 2;
986         }
987         r = X264_MAX( r, h->param.analyse.i_me_range );
988         r = X264_MIN( r, h->param.analyse.i_mv_range );
989         // round up to use the whole mb row
990         r2 = (r & ~15) + ((-X264_THREAD_HEIGHT) & 15);
991         if( r2 < r )
992             r2 += 16;
993         x264_log( h, X264_LOG_DEBUG, "using mv_range_thread = %d\n", r2 );
994         h->param.analyse.i_mv_range_thread = r2;
995     }
996
997     if( h->param.rc.f_rate_tolerance < 0 )
998         h->param.rc.f_rate_tolerance = 0;
999     if( h->param.rc.f_qblur < 0 )
1000         h->param.rc.f_qblur = 0;
1001     if( h->param.rc.f_complexity_blur < 0 )
1002         h->param.rc.f_complexity_blur = 0;
1003
1004     h->param.i_sps_id &= 31;
1005
1006     if( PARAM_INTERLACED )
1007         h->param.b_pic_struct = 1;
1008
1009     h->param.i_nal_hrd = x264_clip3( h->param.i_nal_hrd, X264_NAL_HRD_NONE, X264_NAL_HRD_CBR );
1010
1011     if( h->param.i_nal_hrd && !h->param.rc.i_vbv_buffer_size )
1012     {
1013         x264_log( h, X264_LOG_WARNING, "NAL HRD parameters require VBV parameters\n" );
1014         h->param.i_nal_hrd = X264_NAL_HRD_NONE;
1015     }
1016
1017     if( h->param.i_nal_hrd == X264_NAL_HRD_CBR &&
1018        (h->param.rc.i_bitrate != h->param.rc.i_vbv_max_bitrate || !h->param.rc.i_vbv_max_bitrate) )
1019     {
1020         x264_log( h, X264_LOG_WARNING, "CBR HRD requires constant bitrate\n" );
1021         h->param.i_nal_hrd = X264_NAL_HRD_VBR;
1022     }
1023
1024     /* ensure the booleans are 0 or 1 so they can be used in math */
1025 #define BOOLIFY(x) h->param.x = !!h->param.x
1026     BOOLIFY( b_cabac );
1027     BOOLIFY( b_constrained_intra );
1028     BOOLIFY( b_deblocking_filter );
1029     BOOLIFY( b_deterministic );
1030     BOOLIFY( b_sliced_threads );
1031     BOOLIFY( b_interlaced );
1032     BOOLIFY( b_intra_refresh );
1033     BOOLIFY( b_visualize );
1034     BOOLIFY( b_aud );
1035     BOOLIFY( b_repeat_headers );
1036     BOOLIFY( b_annexb );
1037     BOOLIFY( b_vfr_input );
1038     BOOLIFY( b_pulldown );
1039     BOOLIFY( b_tff );
1040     BOOLIFY( b_pic_struct );
1041     BOOLIFY( b_fake_interlaced );
1042     BOOLIFY( b_open_gop );
1043     BOOLIFY( b_bluray_compat );
1044     BOOLIFY( b_full_recon );
1045     BOOLIFY( analyse.b_transform_8x8 );
1046     BOOLIFY( analyse.b_weighted_bipred );
1047     BOOLIFY( analyse.b_chroma_me );
1048     BOOLIFY( analyse.b_mixed_references );
1049     BOOLIFY( analyse.b_fast_pskip );
1050     BOOLIFY( analyse.b_dct_decimate );
1051     BOOLIFY( analyse.b_psy );
1052     BOOLIFY( analyse.b_psnr );
1053     BOOLIFY( analyse.b_ssim );
1054     BOOLIFY( rc.b_stat_write );
1055     BOOLIFY( rc.b_stat_read );
1056     BOOLIFY( rc.b_mb_tree );
1057 #undef BOOLIFY
1058
1059     return 0;
1060 }
1061
1062 static void mbcmp_init( x264_t *h )
1063 {
1064     int satd = !h->mb.b_lossless && h->param.analyse.i_subpel_refine > 1;
1065     memcpy( h->pixf.mbcmp, satd ? h->pixf.satd : h->pixf.sad_aligned, sizeof(h->pixf.mbcmp) );
1066     memcpy( h->pixf.mbcmp_unaligned, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.mbcmp_unaligned) );
1067     h->pixf.intra_mbcmp_x3_16x16 = satd ? h->pixf.intra_satd_x3_16x16 : h->pixf.intra_sad_x3_16x16;
1068     h->pixf.intra_mbcmp_x3_8x16c = satd ? h->pixf.intra_satd_x3_8x16c : h->pixf.intra_sad_x3_8x16c;
1069     h->pixf.intra_mbcmp_x3_8x8c  = satd ? h->pixf.intra_satd_x3_8x8c  : h->pixf.intra_sad_x3_8x8c;
1070     h->pixf.intra_mbcmp_x3_8x8 = satd ? h->pixf.intra_sa8d_x3_8x8 : h->pixf.intra_sad_x3_8x8;
1071     h->pixf.intra_mbcmp_x3_4x4 = satd ? h->pixf.intra_satd_x3_4x4 : h->pixf.intra_sad_x3_4x4;
1072     h->pixf.intra_mbcmp_x9_4x4 = h->param.b_cpu_independent || h->mb.b_lossless ? NULL
1073                                : satd ? h->pixf.intra_satd_x9_4x4 : h->pixf.intra_sad_x9_4x4;
1074     h->pixf.intra_mbcmp_x9_8x8 = h->param.b_cpu_independent || h->mb.b_lossless ? NULL
1075                                : satd ? h->pixf.intra_sa8d_x9_8x8 : h->pixf.intra_sad_x9_8x8;
1076     satd &= h->param.analyse.i_me_method == X264_ME_TESA;
1077     memcpy( h->pixf.fpelcmp, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.fpelcmp) );
1078     memcpy( h->pixf.fpelcmp_x3, satd ? h->pixf.satd_x3 : h->pixf.sad_x3, sizeof(h->pixf.fpelcmp_x3) );
1079     memcpy( h->pixf.fpelcmp_x4, satd ? h->pixf.satd_x4 : h->pixf.sad_x4, sizeof(h->pixf.fpelcmp_x4) );
1080 }
1081
1082 static void chroma_dsp_init( x264_t *h )
1083 {
1084     memcpy( h->luma2chroma_pixel, x264_luma2chroma_pixel[CHROMA_FORMAT], sizeof(h->luma2chroma_pixel) );
1085
1086     switch( CHROMA_FORMAT )
1087     {
1088         case CHROMA_420:
1089             memcpy( h->predict_chroma, h->predict_8x8c, sizeof(h->predict_chroma) );
1090             h->mc.prefetch_fenc = h->mc.prefetch_fenc_420;
1091             h->loopf.deblock_chroma[0] = h->loopf.deblock_h_chroma_420;
1092             h->loopf.deblock_chroma_intra[0] = h->loopf.deblock_h_chroma_420_intra;
1093             h->loopf.deblock_chroma_mbaff = h->loopf.deblock_chroma_420_mbaff;
1094             h->loopf.deblock_chroma_intra_mbaff = h->loopf.deblock_chroma_420_intra_mbaff;
1095             h->pixf.intra_mbcmp_x3_chroma = h->pixf.intra_mbcmp_x3_8x8c;
1096             h->quantf.coeff_last[DCT_CHROMA_DC] = h->quantf.coeff_last4;
1097             h->quantf.coeff_level_run[DCT_CHROMA_DC] = h->quantf.coeff_level_run4;
1098             break;
1099         case CHROMA_422:
1100             memcpy( h->predict_chroma, h->predict_8x16c, sizeof(h->predict_chroma) );
1101             h->mc.prefetch_fenc = h->mc.prefetch_fenc_422;
1102             h->loopf.deblock_chroma[0] = h->loopf.deblock_h_chroma_422;
1103             h->loopf.deblock_chroma_intra[0] = h->loopf.deblock_h_chroma_422_intra;
1104             h->loopf.deblock_chroma_mbaff = h->loopf.deblock_chroma_422_mbaff;
1105             h->loopf.deblock_chroma_intra_mbaff = h->loopf.deblock_chroma_422_intra_mbaff;
1106             h->pixf.intra_mbcmp_x3_chroma = h->pixf.intra_mbcmp_x3_8x16c;
1107             h->quantf.coeff_last[DCT_CHROMA_DC] = h->quantf.coeff_last8;
1108             h->quantf.coeff_level_run[DCT_CHROMA_DC] = h->quantf.coeff_level_run8;
1109             break;
1110         case CHROMA_444:
1111             h->mc.prefetch_fenc = h->mc.prefetch_fenc_422; /* FIXME: doesn't cover V plane */
1112             h->loopf.deblock_chroma_mbaff = h->loopf.deblock_luma_mbaff;
1113             h->loopf.deblock_chroma_intra_mbaff = h->loopf.deblock_luma_intra_mbaff;
1114             break;
1115     }
1116 }
1117
1118 static void x264_set_aspect_ratio( x264_t *h, x264_param_t *param, int initial )
1119 {
1120     /* VUI */
1121     if( param->vui.i_sar_width > 0 && param->vui.i_sar_height > 0 )
1122     {
1123         uint32_t i_w = param->vui.i_sar_width;
1124         uint32_t i_h = param->vui.i_sar_height;
1125         uint32_t old_w = h->param.vui.i_sar_width;
1126         uint32_t old_h = h->param.vui.i_sar_height;
1127
1128         x264_reduce_fraction( &i_w, &i_h );
1129
1130         while( i_w > 65535 || i_h > 65535 )
1131         {
1132             i_w /= 2;
1133             i_h /= 2;
1134         }
1135
1136         x264_reduce_fraction( &i_w, &i_h );
1137
1138         if( i_w != old_w || i_h != old_h || initial )
1139         {
1140             h->param.vui.i_sar_width = 0;
1141             h->param.vui.i_sar_height = 0;
1142             if( i_w == 0 || i_h == 0 )
1143                 x264_log( h, X264_LOG_WARNING, "cannot create valid sample aspect ratio\n" );
1144             else
1145             {
1146                 x264_log( h, initial?X264_LOG_INFO:X264_LOG_DEBUG, "using SAR=%d/%d\n", i_w, i_h );
1147                 h->param.vui.i_sar_width = i_w;
1148                 h->param.vui.i_sar_height = i_h;
1149             }
1150             x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
1151         }
1152     }
1153 }
1154
1155 /****************************************************************************
1156  * x264_encoder_open:
1157  ****************************************************************************/
1158 x264_t *x264_encoder_open( x264_param_t *param )
1159 {
1160     x264_t *h;
1161     char buf[1000], *p;
1162     int qp, i_slicetype_length;
1163
1164     CHECKED_MALLOCZERO( h, sizeof(x264_t) );
1165
1166     /* Create a copy of param */
1167     memcpy( &h->param, param, sizeof(x264_param_t) );
1168
1169     if( param->param_free )
1170         param->param_free( param );
1171
1172     if( x264_threading_init() )
1173     {
1174         x264_log( h, X264_LOG_ERROR, "unable to initialize threading\n" );
1175         goto fail;
1176     }
1177
1178     if( x264_validate_parameters( h, 1 ) < 0 )
1179         goto fail;
1180
1181     if( h->param.psz_cqm_file )
1182         if( x264_cqm_parse_file( h, h->param.psz_cqm_file ) < 0 )
1183             goto fail;
1184
1185     if( h->param.rc.psz_stat_out )
1186         h->param.rc.psz_stat_out = strdup( h->param.rc.psz_stat_out );
1187     if( h->param.rc.psz_stat_in )
1188         h->param.rc.psz_stat_in = strdup( h->param.rc.psz_stat_in );
1189
1190     x264_reduce_fraction( &h->param.i_fps_num, &h->param.i_fps_den );
1191     x264_reduce_fraction( &h->param.i_timebase_num, &h->param.i_timebase_den );
1192
1193     /* Init x264_t */
1194     h->i_frame = -1;
1195     h->i_frame_num = 0;
1196     h->i_idr_pic_id = 0;
1197
1198     if( (uint64_t)h->param.i_timebase_den * 2 > UINT32_MAX )
1199     {
1200         x264_log( h, X264_LOG_ERROR, "Effective timebase denominator %u exceeds H.264 maximum\n", h->param.i_timebase_den );
1201         goto fail;
1202     }
1203
1204     x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
1205     x264_pps_init( h->pps, h->param.i_sps_id, &h->param, h->sps );
1206
1207     x264_set_aspect_ratio( h, &h->param, 1 );
1208
1209     x264_validate_levels( h, 1 );
1210
1211     h->chroma_qp_table = i_chroma_qp_table + 12 + h->pps->i_chroma_qp_index_offset;
1212
1213     if( x264_cqm_init( h ) < 0 )
1214         goto fail;
1215
1216     h->mb.i_mb_width = h->sps->i_mb_width;
1217     h->mb.i_mb_height = h->sps->i_mb_height;
1218     h->mb.i_mb_count = h->mb.i_mb_width * h->mb.i_mb_height;
1219
1220     h->mb.chroma_h_shift = CHROMA_FORMAT == CHROMA_420 || CHROMA_FORMAT == CHROMA_422;
1221     h->mb.chroma_v_shift = CHROMA_FORMAT == CHROMA_420;
1222
1223     /* Adaptive MBAFF and subme 0 are not supported as we require halving motion
1224      * vectors during prediction, resulting in hpel mvs.
1225      * The chosen solution is to make MBAFF non-adaptive in this case. */
1226     h->mb.b_adaptive_mbaff = PARAM_INTERLACED && h->param.analyse.i_subpel_refine;
1227
1228     /* Init frames. */
1229     if( h->param.i_bframe_adaptive == X264_B_ADAPT_TRELLIS && !h->param.rc.b_stat_read )
1230         h->frames.i_delay = X264_MAX(h->param.i_bframe,3)*4;
1231     else
1232         h->frames.i_delay = h->param.i_bframe;
1233     if( h->param.rc.b_mb_tree || h->param.rc.i_vbv_buffer_size )
1234         h->frames.i_delay = X264_MAX( h->frames.i_delay, h->param.rc.i_lookahead );
1235     i_slicetype_length = h->frames.i_delay;
1236     h->frames.i_delay += h->i_thread_frames - 1;
1237     h->frames.i_delay += h->param.i_sync_lookahead;
1238     h->frames.i_delay += h->param.b_vfr_input;
1239     h->frames.i_bframe_delay = h->param.i_bframe ? (h->param.i_bframe_pyramid ? 2 : 1) : 0;
1240
1241     h->frames.i_max_ref0 = h->param.i_frame_reference;
1242     h->frames.i_max_ref1 = X264_MIN( h->sps->vui.i_num_reorder_frames, h->param.i_frame_reference );
1243     h->frames.i_max_dpb  = h->sps->vui.i_max_dec_frame_buffering;
1244     h->frames.b_have_lowres = !h->param.rc.b_stat_read
1245         && ( h->param.rc.i_rc_method == X264_RC_ABR
1246           || h->param.rc.i_rc_method == X264_RC_CRF
1247           || h->param.i_bframe_adaptive
1248           || h->param.i_scenecut_threshold
1249           || h->param.rc.b_mb_tree
1250           || h->param.analyse.i_weighted_pred );
1251     h->frames.b_have_lowres |= h->param.rc.b_stat_read && h->param.rc.i_vbv_buffer_size > 0;
1252     h->frames.b_have_sub8x8_esa = !!(h->param.analyse.inter & X264_ANALYSE_PSUB8x8);
1253
1254     h->frames.i_last_idr =
1255     h->frames.i_last_keyframe = - h->param.i_keyint_max;
1256     h->frames.i_input    = 0;
1257     h->frames.i_largest_pts = h->frames.i_second_largest_pts = -1;
1258     h->frames.i_poc_last_open_gop = -1;
1259
1260     CHECKED_MALLOCZERO( h->frames.unused[0], (h->frames.i_delay + 3) * sizeof(x264_frame_t *) );
1261     /* Allocate room for max refs plus a few extra just in case. */
1262     CHECKED_MALLOCZERO( h->frames.unused[1], (h->i_thread_frames + X264_REF_MAX + 4) * sizeof(x264_frame_t *) );
1263     CHECKED_MALLOCZERO( h->frames.current, (h->param.i_sync_lookahead + h->param.i_bframe
1264                         + h->i_thread_frames + 3) * sizeof(x264_frame_t *) );
1265     if( h->param.analyse.i_weighted_pred > 0 )
1266         CHECKED_MALLOCZERO( h->frames.blank_unused, h->i_thread_frames * 4 * sizeof(x264_frame_t *) );
1267     h->i_ref[0] = h->i_ref[1] = 0;
1268     h->i_cpb_delay = h->i_coded_fields = h->i_disp_fields = 0;
1269     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);
1270     h->i_disp_fields_last_frame = -1;
1271     x264_rdo_init();
1272
1273     /* init CPU functions */
1274     x264_predict_16x16_init( h->param.cpu, h->predict_16x16 );
1275     x264_predict_8x8c_init( h->param.cpu, h->predict_8x8c );
1276     x264_predict_8x16c_init( h->param.cpu, h->predict_8x16c );
1277     x264_predict_8x8_init( h->param.cpu, h->predict_8x8, &h->predict_8x8_filter );
1278     x264_predict_4x4_init( h->param.cpu, h->predict_4x4 );
1279     x264_pixel_init( h->param.cpu, &h->pixf );
1280     x264_dct_init( h->param.cpu, &h->dctf );
1281     x264_zigzag_init( h->param.cpu, &h->zigzagf_progressive, &h->zigzagf_interlaced );
1282     memcpy( &h->zigzagf, PARAM_INTERLACED ? &h->zigzagf_interlaced : &h->zigzagf_progressive, sizeof(h->zigzagf) );
1283     x264_mc_init( h->param.cpu, &h->mc, h->param.b_cpu_independent );
1284     x264_quant_init( h, h->param.cpu, &h->quantf );
1285     x264_deblock_init( h->param.cpu, &h->loopf, PARAM_INTERLACED );
1286     x264_bitstream_init( h->param.cpu, &h->bsf );
1287     if( h->param.b_cabac )
1288         x264_cabac_init( h );
1289     else
1290         x264_stack_align( x264_cavlc_init, h );
1291
1292     mbcmp_init( h );
1293     chroma_dsp_init( h );
1294
1295     p = buf + sprintf( buf, "using cpu capabilities:" );
1296     for( int i = 0; x264_cpu_names[i].flags; i++ )
1297     {
1298         if( !strcmp(x264_cpu_names[i].name, "SSE")
1299             && h->param.cpu & (X264_CPU_SSE2) )
1300             continue;
1301         if( !strcmp(x264_cpu_names[i].name, "SSE2")
1302             && h->param.cpu & (X264_CPU_SSE2_IS_FAST|X264_CPU_SSE2_IS_SLOW) )
1303             continue;
1304         if( !strcmp(x264_cpu_names[i].name, "SSE3")
1305             && (h->param.cpu & X264_CPU_SSSE3 || !(h->param.cpu & X264_CPU_CACHELINE_64)) )
1306             continue;
1307         if( !strcmp(x264_cpu_names[i].name, "SSE4.1")
1308             && (h->param.cpu & X264_CPU_SSE42) )
1309             continue;
1310         if( (h->param.cpu & x264_cpu_names[i].flags) == x264_cpu_names[i].flags
1311             && (!i || x264_cpu_names[i].flags != x264_cpu_names[i-1].flags) )
1312             p += sprintf( p, " %s", x264_cpu_names[i].name );
1313     }
1314     if( !h->param.cpu )
1315         p += sprintf( p, " none!" );
1316     x264_log( h, X264_LOG_INFO, "%s\n", buf );
1317
1318     float *logs = x264_analyse_prepare_costs( h );
1319     if( !logs )
1320         goto fail;
1321     for( qp = X264_MIN( h->param.rc.i_qp_min, QP_MAX_SPEC ); qp <= h->param.rc.i_qp_max; qp++ )
1322         if( x264_analyse_init_costs( h, logs, qp ) )
1323             goto fail;
1324     if( x264_analyse_init_costs( h, logs, X264_LOOKAHEAD_QP ) )
1325         goto fail;
1326     x264_free( logs );
1327
1328     static const uint16_t cost_mv_correct[7] = { 24, 47, 95, 189, 379, 757, 1515 };
1329     /* Checks for known miscompilation issues. */
1330     if( h->cost_mv[X264_LOOKAHEAD_QP][2013] != cost_mv_correct[BIT_DEPTH-8] )
1331     {
1332         x264_log( h, X264_LOG_ERROR, "MV cost test failed: x264 has been miscompiled!\n" );
1333         goto fail;
1334     }
1335
1336     /* Must be volatile or else GCC will optimize it out. */
1337     volatile int temp = 392;
1338     if( x264_clz( temp ) != 23 )
1339     {
1340         x264_log( h, X264_LOG_ERROR, "CLZ test failed: x264 has been miscompiled!\n" );
1341 #if ARCH_X86 || ARCH_X86_64
1342         x264_log( h, X264_LOG_ERROR, "Are you attempting to run an SSE4a/LZCNT-targeted build on a CPU that\n" );
1343         x264_log( h, X264_LOG_ERROR, "doesn't support it?\n" );
1344 #endif
1345         goto fail;
1346     }
1347
1348     h->out.i_nal = 0;
1349     h->out.i_bitstream = X264_MAX( 1000000, h->param.i_width * h->param.i_height * 4
1350         * ( h->param.rc.i_rc_method == X264_RC_ABR ? pow( 0.95, h->param.rc.i_qp_min )
1351           : pow( 0.95, h->param.rc.i_qp_constant ) * X264_MAX( 1, h->param.rc.f_ip_factor )));
1352
1353     h->nal_buffer_size = h->out.i_bitstream * 3/2 + 4;
1354     CHECKED_MALLOC( h->nal_buffer, h->nal_buffer_size );
1355
1356     if( h->param.i_threads > 1 &&
1357         x264_threadpool_init( &h->threadpool, h->param.i_threads, (void*)x264_encoder_thread_init, h ) )
1358         goto fail;
1359     if( h->param.i_lookahead_threads > 1 &&
1360         x264_threadpool_init( &h->lookaheadpool, h->param.i_lookahead_threads, (void*)x264_lookahead_thread_init, h ) )
1361         goto fail;
1362
1363     h->thread[0] = h;
1364     for( int i = 1; i < h->param.i_threads + !!h->param.i_sync_lookahead; i++ )
1365         CHECKED_MALLOC( h->thread[i], sizeof(x264_t) );
1366     if( h->param.i_lookahead_threads > 1 )
1367         for( int i = 0; i < h->param.i_lookahead_threads; i++ )
1368         {
1369             CHECKED_MALLOC( h->lookahead_thread[i], sizeof(x264_t) );
1370             *h->lookahead_thread[i] = *h;
1371         }
1372
1373     for( int i = 0; i < h->param.i_threads; i++ )
1374     {
1375         int init_nal_count = h->param.i_slice_count + 3;
1376         int allocate_threadlocal_data = !h->param.b_sliced_threads || !i;
1377         if( i > 0 )
1378             *h->thread[i] = *h;
1379
1380         if( x264_pthread_mutex_init( &h->thread[i]->mutex, NULL ) )
1381             goto fail;
1382         if( x264_pthread_cond_init( &h->thread[i]->cv, NULL ) )
1383             goto fail;
1384
1385         if( allocate_threadlocal_data )
1386         {
1387             h->thread[i]->fdec = x264_frame_pop_unused( h, 1 );
1388             if( !h->thread[i]->fdec )
1389                 goto fail;
1390         }
1391         else
1392             h->thread[i]->fdec = h->thread[0]->fdec;
1393
1394         CHECKED_MALLOC( h->thread[i]->out.p_bitstream, h->out.i_bitstream );
1395         /* Start each thread with room for init_nal_count NAL units; it'll realloc later if needed. */
1396         CHECKED_MALLOC( h->thread[i]->out.nal, init_nal_count*sizeof(x264_nal_t) );
1397         h->thread[i]->out.i_nals_allocated = init_nal_count;
1398
1399         if( allocate_threadlocal_data && x264_macroblock_cache_allocate( h->thread[i] ) < 0 )
1400             goto fail;
1401     }
1402
1403     if( x264_lookahead_init( h, i_slicetype_length ) )
1404         goto fail;
1405
1406     for( int i = 0; i < h->param.i_threads; i++ )
1407         if( x264_macroblock_thread_allocate( h->thread[i], 0 ) < 0 )
1408             goto fail;
1409
1410     if( x264_ratecontrol_new( h ) < 0 )
1411         goto fail;
1412
1413     if( h->param.i_nal_hrd )
1414     {
1415         x264_log( h, X264_LOG_DEBUG, "HRD bitrate: %i bits/sec\n", h->sps->vui.hrd.i_bit_rate_unscaled );
1416         x264_log( h, X264_LOG_DEBUG, "CPB size: %i bits\n", h->sps->vui.hrd.i_cpb_size_unscaled );
1417     }
1418
1419     if( h->param.psz_dump_yuv )
1420     {
1421         /* create or truncate the reconstructed video file */
1422         FILE *f = fopen( h->param.psz_dump_yuv, "w" );
1423         if( !f )
1424         {
1425             x264_log( h, X264_LOG_ERROR, "dump_yuv: can't write to %s\n", h->param.psz_dump_yuv );
1426             goto fail;
1427         }
1428         else if( !x264_is_regular_file( f ) )
1429         {
1430             x264_log( h, X264_LOG_ERROR, "dump_yuv: incompatible with non-regular file %s\n", h->param.psz_dump_yuv );
1431             goto fail;
1432         }
1433         fclose( f );
1434     }
1435
1436     const char *profile = h->sps->i_profile_idc == PROFILE_BASELINE ? "Constrained Baseline" :
1437                           h->sps->i_profile_idc == PROFILE_MAIN ? "Main" :
1438                           h->sps->i_profile_idc == PROFILE_HIGH ? "High" :
1439                           h->sps->i_profile_idc == PROFILE_HIGH10 ? (h->sps->b_constraint_set3 == 1 ? "High 10 Intra" : "High 10") :
1440                           h->sps->i_profile_idc == PROFILE_HIGH422 ? (h->sps->b_constraint_set3 == 1 ? "High 4:2:2 Intra" : "High 4:2:2") :
1441                           h->sps->b_constraint_set3 == 1 ? "High 4:4:4 Intra" : "High 4:4:4 Predictive";
1442     char level[4];
1443     snprintf( level, sizeof(level), "%d.%d", h->sps->i_level_idc/10, h->sps->i_level_idc%10 );
1444     if( h->sps->i_level_idc == 9 || ( h->sps->i_level_idc == 11 && h->sps->b_constraint_set3 &&
1445         (h->sps->i_profile_idc == PROFILE_BASELINE || h->sps->i_profile_idc == PROFILE_MAIN) ) )
1446         strcpy( level, "1b" );
1447
1448     if( h->sps->i_profile_idc < PROFILE_HIGH10 )
1449     {
1450         x264_log( h, X264_LOG_INFO, "profile %s, level %s\n",
1451             profile, level );
1452     }
1453     else
1454     {
1455         static const char * const subsampling[4] = { "4:0:0", "4:2:0", "4:2:2", "4:4:4" };
1456         x264_log( h, X264_LOG_INFO, "profile %s, level %s, %s %d-bit\n",
1457             profile, level, subsampling[CHROMA_FORMAT], BIT_DEPTH );
1458     }
1459
1460     return h;
1461 fail:
1462     x264_free( h );
1463     return NULL;
1464 }
1465
1466 /****************************************************************************
1467  * x264_encoder_reconfig:
1468  ****************************************************************************/
1469 int x264_encoder_reconfig( x264_t *h, x264_param_t *param )
1470 {
1471     /* If the previous frame isn't done encoding, reconfiguring is probably dangerous. */
1472     if( h->param.b_sliced_threads )
1473         if( x264_threadpool_wait_all( h ) < 0 )
1474             return -1;
1475
1476     int rc_reconfig = 0;
1477     h = h->thread[h->thread[0]->i_thread_phase];
1478     x264_set_aspect_ratio( h, param, 0 );
1479 #define COPY(var) h->param.var = param->var
1480     COPY( i_frame_reference ); // but never uses more refs than initially specified
1481     COPY( i_bframe_bias );
1482     if( h->param.i_scenecut_threshold )
1483         COPY( i_scenecut_threshold ); // can't turn it on or off, only vary the threshold
1484     COPY( b_deblocking_filter );
1485     COPY( i_deblocking_filter_alphac0 );
1486     COPY( i_deblocking_filter_beta );
1487     COPY( i_frame_packing );
1488     COPY( analyse.inter );
1489     COPY( analyse.intra );
1490     COPY( analyse.i_direct_mv_pred );
1491     /* Scratch buffer prevents me_range from being increased for esa/tesa */
1492     if( h->param.analyse.i_me_method < X264_ME_ESA || param->analyse.i_me_range < h->param.analyse.i_me_range )
1493         COPY( analyse.i_me_range );
1494     COPY( analyse.i_noise_reduction );
1495     /* We can't switch out of subme=0 during encoding. */
1496     if( h->param.analyse.i_subpel_refine )
1497         COPY( analyse.i_subpel_refine );
1498     COPY( analyse.i_trellis );
1499     COPY( analyse.b_chroma_me );
1500     COPY( analyse.b_dct_decimate );
1501     COPY( analyse.b_fast_pskip );
1502     COPY( analyse.b_mixed_references );
1503     COPY( analyse.f_psy_rd );
1504     COPY( analyse.f_psy_trellis );
1505     COPY( crop_rect );
1506     // can only twiddle these if they were enabled to begin with:
1507     if( h->param.analyse.i_me_method >= X264_ME_ESA || param->analyse.i_me_method < X264_ME_ESA )
1508         COPY( analyse.i_me_method );
1509     if( h->param.analyse.i_me_method >= X264_ME_ESA && !h->frames.b_have_sub8x8_esa )
1510         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
1511     if( h->pps->b_transform_8x8_mode )
1512         COPY( analyse.b_transform_8x8 );
1513     if( h->frames.i_max_ref1 > 1 )
1514         COPY( i_bframe_pyramid );
1515     COPY( i_slice_max_size );
1516     COPY( i_slice_max_mbs );
1517     COPY( i_slice_min_mbs );
1518     COPY( i_slice_count );
1519     COPY( i_slice_count_max );
1520     COPY( b_tff );
1521
1522     /* VBV can't be turned on if it wasn't on to begin with */
1523     if( h->param.rc.i_vbv_max_bitrate > 0 && h->param.rc.i_vbv_buffer_size > 0 &&
1524           param->rc.i_vbv_max_bitrate > 0 &&   param->rc.i_vbv_buffer_size > 0 )
1525     {
1526         rc_reconfig |= h->param.rc.i_vbv_max_bitrate != param->rc.i_vbv_max_bitrate;
1527         rc_reconfig |= h->param.rc.i_vbv_buffer_size != param->rc.i_vbv_buffer_size;
1528         rc_reconfig |= h->param.rc.i_bitrate != param->rc.i_bitrate;
1529         COPY( rc.i_vbv_max_bitrate );
1530         COPY( rc.i_vbv_buffer_size );
1531         COPY( rc.i_bitrate );
1532     }
1533     rc_reconfig |= h->param.rc.f_rf_constant != param->rc.f_rf_constant;
1534     rc_reconfig |= h->param.rc.f_rf_constant_max != param->rc.f_rf_constant_max;
1535     COPY( rc.f_rf_constant );
1536     COPY( rc.f_rf_constant_max );
1537 #undef COPY
1538
1539     mbcmp_init( h );
1540
1541     int ret = x264_validate_parameters( h, 0 );
1542
1543     /* Supported reconfiguration options (1-pass only):
1544      * vbv-maxrate
1545      * vbv-bufsize
1546      * crf
1547      * bitrate (CBR only) */
1548     if( !ret && rc_reconfig )
1549         x264_ratecontrol_init_reconfigurable( h, 0 );
1550
1551     return ret;
1552 }
1553
1554 /****************************************************************************
1555  * x264_encoder_parameters:
1556  ****************************************************************************/
1557 void x264_encoder_parameters( x264_t *h, x264_param_t *param )
1558 {
1559     memcpy( param, &h->thread[h->i_thread_phase]->param, sizeof(x264_param_t) );
1560 }
1561
1562 /* internal usage */
1563 static void x264_nal_start( x264_t *h, int i_type, int i_ref_idc )
1564 {
1565     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1566
1567     nal->i_ref_idc        = i_ref_idc;
1568     nal->i_type           = i_type;
1569     nal->b_long_startcode = 1;
1570
1571     nal->i_payload= 0;
1572     nal->p_payload= &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8];
1573 }
1574
1575 /* if number of allocated nals is not enough, re-allocate a larger one. */
1576 static int x264_nal_check_buffer( x264_t *h )
1577 {
1578     if( h->out.i_nal >= h->out.i_nals_allocated )
1579     {
1580         x264_nal_t *new_out = x264_malloc( sizeof(x264_nal_t) * (h->out.i_nals_allocated*2) );
1581         if( !new_out )
1582             return -1;
1583         memcpy( new_out, h->out.nal, sizeof(x264_nal_t) * (h->out.i_nals_allocated) );
1584         x264_free( h->out.nal );
1585         h->out.nal = new_out;
1586         h->out.i_nals_allocated *= 2;
1587     }
1588     return 0;
1589 }
1590
1591 static int x264_nal_end( x264_t *h )
1592 {
1593     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1594     uint8_t *end = &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8];
1595     nal->i_payload = end - nal->p_payload;
1596     /* nal_escape_mmx reads past the end of the input.
1597      * While undefined padding wouldn't actually affect the output, it makes valgrind unhappy. */
1598     memset( end, 0xff, 32 );
1599     if( h->param.nalu_process )
1600         h->param.nalu_process( h, nal, h->fenc->opaque );
1601     h->out.i_nal++;
1602
1603     return x264_nal_check_buffer( h );
1604 }
1605
1606 static int x264_encoder_encapsulate_nals( x264_t *h, int start )
1607 {
1608     int nal_size = 0, previous_nal_size = 0;
1609
1610     if( h->param.nalu_process )
1611     {
1612         for( int i = start; i < h->out.i_nal; i++ )
1613             nal_size += h->out.nal[i].i_payload;
1614         return nal_size;
1615     }
1616
1617     for( int i = 0; i < start; i++ )
1618         previous_nal_size += h->out.nal[i].i_payload;
1619
1620     for( int i = start; i < h->out.i_nal; i++ )
1621         nal_size += h->out.nal[i].i_payload;
1622
1623     /* Worst-case NAL unit escaping: reallocate the buffer if it's too small. */
1624     int necessary_size = nal_size * 3/2 + h->out.i_nal * 4;
1625     if( h->nal_buffer_size < necessary_size )
1626     {
1627         h->nal_buffer_size = necessary_size * 2;
1628         uint8_t *buf = x264_malloc( h->nal_buffer_size );
1629         if( !buf )
1630             return -1;
1631         if( previous_nal_size )
1632             memcpy( buf, h->nal_buffer, previous_nal_size );
1633         x264_free( h->nal_buffer );
1634         h->nal_buffer = buf;
1635     }
1636
1637     uint8_t *nal_buffer = h->nal_buffer + previous_nal_size;
1638
1639     for( int i = start; i < h->out.i_nal; i++ )
1640     {
1641         h->out.nal[i].b_long_startcode = !i || h->out.nal[i].i_type == NAL_SPS || h->out.nal[i].i_type == NAL_PPS;
1642         x264_nal_encode( h, nal_buffer, &h->out.nal[i] );
1643         nal_buffer += h->out.nal[i].i_payload;
1644     }
1645
1646     x264_emms();
1647
1648     return nal_buffer - (h->nal_buffer + previous_nal_size);
1649 }
1650
1651 /****************************************************************************
1652  * x264_encoder_headers:
1653  ****************************************************************************/
1654 int x264_encoder_headers( x264_t *h, x264_nal_t **pp_nal, int *pi_nal )
1655 {
1656     int frame_size = 0;
1657     /* init bitstream context */
1658     h->out.i_nal = 0;
1659     bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
1660
1661     /* Write SEI, SPS and PPS. */
1662
1663     /* generate sequence parameters */
1664     x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
1665     x264_sps_write( &h->out.bs, h->sps );
1666     if( x264_nal_end( h ) )
1667         return -1;
1668
1669     /* generate picture parameters */
1670     x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
1671     x264_pps_write( &h->out.bs, h->sps, h->pps );
1672     if( x264_nal_end( h ) )
1673         return -1;
1674
1675     /* identify ourselves */
1676     x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
1677     if( x264_sei_version_write( h, &h->out.bs ) )
1678         return -1;
1679     if( x264_nal_end( h ) )
1680         return -1;
1681
1682     frame_size = x264_encoder_encapsulate_nals( h, 0 );
1683     if( frame_size < 0 )
1684         return -1;
1685
1686     /* now set output*/
1687     *pi_nal = h->out.i_nal;
1688     *pp_nal = &h->out.nal[0];
1689     h->out.i_nal = 0;
1690
1691     return frame_size;
1692 }
1693
1694 /* Check to see whether we have chosen a reference list ordering different
1695  * from the standard's default. */
1696 static inline void x264_reference_check_reorder( x264_t *h )
1697 {
1698     /* The reorder check doesn't check for missing frames, so just
1699      * force a reorder if one of the reference list is corrupt. */
1700     for( int i = 0; h->frames.reference[i]; i++ )
1701         if( h->frames.reference[i]->b_corrupt )
1702         {
1703             h->b_ref_reorder[0] = 1;
1704             return;
1705         }
1706     for( int list = 0; list <= (h->sh.i_type == SLICE_TYPE_B); list++ )
1707         for( int i = 0; i < h->i_ref[list] - 1; i++ )
1708         {
1709             int framenum_diff = h->fref[list][i+1]->i_frame_num - h->fref[list][i]->i_frame_num;
1710             int poc_diff = h->fref[list][i+1]->i_poc - h->fref[list][i]->i_poc;
1711             /* P and B-frames use different default orders. */
1712             if( h->sh.i_type == SLICE_TYPE_P ? framenum_diff > 0 : list == 1 ? poc_diff < 0 : poc_diff > 0 )
1713             {
1714                 h->b_ref_reorder[list] = 1;
1715                 return;
1716             }
1717         }
1718 }
1719
1720 /* return -1 on failure, else return the index of the new reference frame */
1721 int x264_weighted_reference_duplicate( x264_t *h, int i_ref, const x264_weight_t *w )
1722 {
1723     int i = h->i_ref[0];
1724     int j = 1;
1725     x264_frame_t *newframe;
1726     if( i <= 1 ) /* empty list, definitely can't duplicate frame */
1727         return -1;
1728
1729     //Duplication is only used in X264_WEIGHTP_SMART
1730     if( h->param.analyse.i_weighted_pred != X264_WEIGHTP_SMART )
1731         return -1;
1732
1733     /* Duplication is a hack to compensate for crappy rounding in motion compensation.
1734      * With high bit depth, it's not worth doing, so turn it off except in the case of
1735      * unweighted dupes. */
1736     if( BIT_DEPTH > 8 && w != x264_weight_none )
1737         return -1;
1738
1739     newframe = x264_frame_pop_blank_unused( h );
1740     if( !newframe )
1741         return -1;
1742
1743     //FIXME: probably don't need to copy everything
1744     *newframe = *h->fref[0][i_ref];
1745     newframe->i_reference_count = 1;
1746     newframe->orig = h->fref[0][i_ref];
1747     newframe->b_duplicate = 1;
1748     memcpy( h->fenc->weight[j], w, sizeof(h->fenc->weight[i]) );
1749
1750     /* shift the frames to make space for the dupe. */
1751     h->b_ref_reorder[0] = 1;
1752     if( h->i_ref[0] < X264_REF_MAX )
1753         ++h->i_ref[0];
1754     h->fref[0][X264_REF_MAX-1] = NULL;
1755     x264_frame_unshift( &h->fref[0][j], newframe );
1756
1757     return j;
1758 }
1759
1760 static void x264_weighted_pred_init( x264_t *h )
1761 {
1762     /* for now no analysis and set all weights to nothing */
1763     for( int i_ref = 0; i_ref < h->i_ref[0]; i_ref++ )
1764         h->fenc->weighted[i_ref] = h->fref[0][i_ref]->filtered[0][0];
1765
1766     // FIXME: This only supports weighting of one reference frame
1767     // and duplicates of that frame.
1768     h->fenc->i_lines_weighted = 0;
1769
1770     for( int i_ref = 0; i_ref < (h->i_ref[0] << SLICE_MBAFF); i_ref++ )
1771         for( int i = 0; i < 3; i++ )
1772             h->sh.weight[i_ref][i].weightfn = NULL;
1773
1774
1775     if( h->sh.i_type != SLICE_TYPE_P || h->param.analyse.i_weighted_pred <= 0 )
1776         return;
1777
1778     int i_padv = PADV << PARAM_INTERLACED;
1779     int denom = -1;
1780     int weightplane[2] = { 0, 0 };
1781     int buffer_next = 0;
1782     for( int i = 0; i < 3; i++ )
1783     {
1784         for( int j = 0; j < h->i_ref[0]; j++ )
1785         {
1786             if( h->fenc->weight[j][i].weightfn )
1787             {
1788                 h->sh.weight[j][i] = h->fenc->weight[j][i];
1789                 // if weight is useless, don't write it to stream
1790                 if( h->sh.weight[j][i].i_scale == 1<<h->sh.weight[j][i].i_denom && h->sh.weight[j][i].i_offset == 0 )
1791                     h->sh.weight[j][i].weightfn = NULL;
1792                 else
1793                 {
1794                     if( !weightplane[!!i] )
1795                     {
1796                         weightplane[!!i] = 1;
1797                         h->sh.weight[0][!!i].i_denom = denom = h->sh.weight[j][i].i_denom;
1798                         assert( x264_clip3( denom, 0, 7 ) == denom );
1799                     }
1800
1801                     assert( h->sh.weight[j][i].i_denom == denom );
1802                     if( !i )
1803                     {
1804                         h->fenc->weighted[j] = h->mb.p_weight_buf[buffer_next++] + h->fenc->i_stride[0] * i_padv + PADH;
1805                         //scale full resolution frame
1806                         if( h->param.i_threads == 1 )
1807                         {
1808                             pixel *src = h->fref[0][j]->filtered[0][0] - h->fref[0][j]->i_stride[0]*i_padv - PADH;
1809                             pixel *dst = h->fenc->weighted[j] - h->fenc->i_stride[0]*i_padv - PADH;
1810                             int stride = h->fenc->i_stride[0];
1811                             int width = h->fenc->i_width[0] + PADH*2;
1812                             int height = h->fenc->i_lines[0] + i_padv*2;
1813                             x264_weight_scale_plane( h, dst, stride, src, stride, width, height, &h->sh.weight[j][0] );
1814                             h->fenc->i_lines_weighted = height;
1815                         }
1816                     }
1817                 }
1818             }
1819         }
1820     }
1821
1822     if( weightplane[1] )
1823         for( int i = 0; i < h->i_ref[0]; i++ )
1824         {
1825             if( h->sh.weight[i][1].weightfn && !h->sh.weight[i][2].weightfn )
1826             {
1827                 h->sh.weight[i][2].i_scale = 1 << h->sh.weight[0][1].i_denom;
1828                 h->sh.weight[i][2].i_offset = 0;
1829             }
1830             else if( h->sh.weight[i][2].weightfn && !h->sh.weight[i][1].weightfn )
1831             {
1832                 h->sh.weight[i][1].i_scale = 1 << h->sh.weight[0][1].i_denom;
1833                 h->sh.weight[i][1].i_offset = 0;
1834             }
1835         }
1836
1837     if( !weightplane[0] )
1838         h->sh.weight[0][0].i_denom = 0;
1839     if( !weightplane[1] )
1840         h->sh.weight[0][1].i_denom = 0;
1841     h->sh.weight[0][2].i_denom = h->sh.weight[0][1].i_denom;
1842 }
1843
1844 static inline int x264_reference_distance( x264_t *h, x264_frame_t *frame )
1845 {
1846     if( h->param.i_frame_packing == 5 )
1847         return abs((h->fenc->i_frame&~1) - (frame->i_frame&~1)) +
1848                   ((h->fenc->i_frame&1) != (frame->i_frame&1));
1849     else
1850         return abs(h->fenc->i_frame - frame->i_frame);
1851 }
1852
1853 static inline void x264_reference_build_list( x264_t *h, int i_poc )
1854 {
1855     int b_ok;
1856
1857     /* build ref list 0/1 */
1858     h->mb.pic.i_fref[0] = h->i_ref[0] = 0;
1859     h->mb.pic.i_fref[1] = h->i_ref[1] = 0;
1860     if( h->sh.i_type == SLICE_TYPE_I )
1861         return;
1862
1863     for( int i = 0; h->frames.reference[i]; i++ )
1864     {
1865         if( h->frames.reference[i]->b_corrupt )
1866             continue;
1867         if( h->frames.reference[i]->i_poc < i_poc )
1868             h->fref[0][h->i_ref[0]++] = h->frames.reference[i];
1869         else if( h->frames.reference[i]->i_poc > i_poc )
1870             h->fref[1][h->i_ref[1]++] = h->frames.reference[i];
1871     }
1872
1873     /* Order reference lists by distance from the current frame. */
1874     for( int list = 0; list < 2; list++ )
1875     {
1876         h->fref_nearest[list] = h->fref[list][0];
1877         do
1878         {
1879             b_ok = 1;
1880             for( int i = 0; i < h->i_ref[list] - 1; i++ )
1881             {
1882                 if( list ? h->fref[list][i+1]->i_poc < h->fref_nearest[list]->i_poc
1883                          : h->fref[list][i+1]->i_poc > h->fref_nearest[list]->i_poc )
1884                     h->fref_nearest[list] = h->fref[list][i+1];
1885                 if( x264_reference_distance( h, h->fref[list][i] ) > x264_reference_distance( h, h->fref[list][i+1] ) )
1886                 {
1887                     XCHG( x264_frame_t*, h->fref[list][i], h->fref[list][i+1] );
1888                     b_ok = 0;
1889                     break;
1890                 }
1891             }
1892         } while( !b_ok );
1893     }
1894
1895     if( h->sh.i_mmco_remove_from_end )
1896         for( int i = h->i_ref[0]-1; i >= h->i_ref[0] - h->sh.i_mmco_remove_from_end; i-- )
1897         {
1898             int diff = h->i_frame_num - h->fref[0][i]->i_frame_num;
1899             h->sh.mmco[h->sh.i_mmco_command_count].i_poc = h->fref[0][i]->i_poc;
1900             h->sh.mmco[h->sh.i_mmco_command_count++].i_difference_of_pic_nums = diff;
1901         }
1902
1903     x264_reference_check_reorder( h );
1904
1905     h->i_ref[1] = X264_MIN( h->i_ref[1], h->frames.i_max_ref1 );
1906     h->i_ref[0] = X264_MIN( h->i_ref[0], h->frames.i_max_ref0 );
1907     h->i_ref[0] = X264_MIN( h->i_ref[0], h->param.i_frame_reference ); // if reconfig() has lowered the limit
1908
1909     /* For Blu-ray compliance, don't reference frames outside of the minigop. */
1910     if( IS_X264_TYPE_B( h->fenc->i_type ) && h->param.b_bluray_compat )
1911         h->i_ref[0] = X264_MIN( h->i_ref[0], IS_X264_TYPE_B( h->fref[0][0]->i_type ) + 1 );
1912
1913     /* add duplicates */
1914     if( h->fenc->i_type == X264_TYPE_P )
1915     {
1916         int idx = -1;
1917         if( h->param.analyse.i_weighted_pred >= X264_WEIGHTP_SIMPLE )
1918         {
1919             x264_weight_t w[3];
1920             w[1].weightfn = w[2].weightfn = NULL;
1921             if( h->param.rc.b_stat_read )
1922                 x264_ratecontrol_set_weights( h, h->fenc );
1923
1924             if( !h->fenc->weight[0][0].weightfn )
1925             {
1926                 h->fenc->weight[0][0].i_denom = 0;
1927                 SET_WEIGHT( w[0], 1, 1, 0, -1 );
1928                 idx = x264_weighted_reference_duplicate( h, 0, w );
1929             }
1930             else
1931             {
1932                 if( h->fenc->weight[0][0].i_scale == 1<<h->fenc->weight[0][0].i_denom )
1933                 {
1934                     SET_WEIGHT( h->fenc->weight[0][0], 1, 1, 0, h->fenc->weight[0][0].i_offset );
1935                 }
1936                 x264_weighted_reference_duplicate( h, 0, x264_weight_none );
1937                 if( h->fenc->weight[0][0].i_offset > -128 )
1938                 {
1939                     w[0] = h->fenc->weight[0][0];
1940                     w[0].i_offset--;
1941                     h->mc.weight_cache( h, &w[0] );
1942                     idx = x264_weighted_reference_duplicate( h, 0, w );
1943                 }
1944             }
1945         }
1946         h->mb.ref_blind_dupe = idx;
1947     }
1948
1949     assert( h->i_ref[0] + h->i_ref[1] <= X264_REF_MAX );
1950     h->mb.pic.i_fref[0] = h->i_ref[0];
1951     h->mb.pic.i_fref[1] = h->i_ref[1];
1952 }
1953
1954 static void x264_fdec_filter_row( x264_t *h, int mb_y, int pass )
1955 {
1956     /* mb_y is the mb to be encoded next, not the mb to be filtered here */
1957     int b_hpel = h->fdec->b_kept_as_ref;
1958     int b_deblock = h->sh.i_disable_deblocking_filter_idc != 1;
1959     int b_end = mb_y == h->i_threadslice_end;
1960     int b_measure_quality = 1;
1961     int min_y = mb_y - (1 << SLICE_MBAFF);
1962     int b_start = min_y == h->i_threadslice_start;
1963     /* Even in interlaced mode, deblocking never modifies more than 4 pixels
1964      * above each MB, as bS=4 doesn't happen for the top of interlaced mbpairs. */
1965     int minpix_y = min_y*16 - 4 * !b_start;
1966     int maxpix_y = mb_y*16 - 4 * !b_end;
1967     b_deblock &= b_hpel || h->param.b_full_recon || h->param.psz_dump_yuv;
1968     if( h->param.b_sliced_threads )
1969     {
1970         switch( pass )
1971         {
1972             /* During encode: only do deblock if asked for */
1973             default:
1974             case 0:
1975                 b_deblock &= h->param.b_full_recon;
1976                 b_hpel = 0;
1977                 break;
1978             /* During post-encode pass: do deblock if not done yet, do hpel for all
1979              * rows except those between slices. */
1980             case 1:
1981                 b_deblock &= !h->param.b_full_recon;
1982                 b_hpel &= !(b_start && min_y > 0);
1983                 b_measure_quality = 0;
1984                 break;
1985             /* Final pass: do the rows between slices in sequence. */
1986             case 2:
1987                 b_deblock = 0;
1988                 b_measure_quality = 0;
1989                 break;
1990         }
1991     }
1992     if( mb_y & SLICE_MBAFF )
1993         return;
1994     if( min_y < h->i_threadslice_start )
1995         return;
1996
1997     if( b_deblock )
1998         for( int y = min_y; y < mb_y; y += (1 << SLICE_MBAFF) )
1999             x264_frame_deblock_row( h, y );
2000
2001     /* FIXME: Prediction requires different borders for interlaced/progressive mc,
2002      * but the actual image data is equivalent. For now, maintain this
2003      * consistency by copying deblocked pixels between planes. */
2004     if( PARAM_INTERLACED && (!h->param.b_sliced_threads || pass == 1) )
2005         for( int p = 0; p < h->fdec->i_plane; p++ )
2006             for( int i = minpix_y>>(CHROMA_V_SHIFT && p); i < maxpix_y>>(CHROMA_V_SHIFT && p); i++ )
2007                 memcpy( h->fdec->plane_fld[p] + i*h->fdec->i_stride[p],
2008                         h->fdec->plane[p]     + i*h->fdec->i_stride[p],
2009                         h->mb.i_mb_width*16*sizeof(pixel) );
2010
2011     if( h->fdec->b_kept_as_ref && (!h->param.b_sliced_threads || pass == 1) )
2012         x264_frame_expand_border( h, h->fdec, min_y );
2013     if( b_hpel )
2014     {
2015         int end = mb_y == h->mb.i_mb_height;
2016         /* Can't do hpel until the previous slice is done encoding. */
2017         if( h->param.analyse.i_subpel_refine )
2018         {
2019             x264_frame_filter( h, h->fdec, min_y, end );
2020             x264_frame_expand_border_filtered( h, h->fdec, min_y, end );
2021         }
2022     }
2023
2024     if( SLICE_MBAFF && pass == 0 )
2025         for( int i = 0; i < 3; i++ )
2026         {
2027             XCHG( pixel *, h->intra_border_backup[0][i], h->intra_border_backup[3][i] );
2028             XCHG( pixel *, h->intra_border_backup[1][i], h->intra_border_backup[4][i] );
2029         }
2030
2031     if( h->i_thread_frames > 1 && h->fdec->b_kept_as_ref )
2032         x264_frame_cond_broadcast( h->fdec, mb_y*16 + (b_end ? 10000 : -(X264_THREAD_HEIGHT << SLICE_MBAFF)) );
2033
2034     if( b_measure_quality )
2035     {
2036         maxpix_y = X264_MIN( maxpix_y, h->param.i_height );
2037         if( h->param.analyse.b_psnr )
2038         {
2039             for( int p = 0; p < (CHROMA444 ? 3 : 1); p++ )
2040                 h->stat.frame.i_ssd[p] += x264_pixel_ssd_wxh( &h->pixf,
2041                     h->fdec->plane[p] + minpix_y * h->fdec->i_stride[p], h->fdec->i_stride[p],
2042                     h->fenc->plane[p] + minpix_y * h->fenc->i_stride[p], h->fenc->i_stride[p],
2043                     h->param.i_width, maxpix_y-minpix_y );
2044             if( !CHROMA444 )
2045             {
2046                 uint64_t ssd_u, ssd_v;
2047                 int v_shift = CHROMA_V_SHIFT;
2048                 x264_pixel_ssd_nv12( &h->pixf,
2049                     h->fdec->plane[1] + (minpix_y>>v_shift) * h->fdec->i_stride[1], h->fdec->i_stride[1],
2050                     h->fenc->plane[1] + (minpix_y>>v_shift) * h->fenc->i_stride[1], h->fenc->i_stride[1],
2051                     h->param.i_width>>1, (maxpix_y-minpix_y)>>v_shift, &ssd_u, &ssd_v );
2052                 h->stat.frame.i_ssd[1] += ssd_u;
2053                 h->stat.frame.i_ssd[2] += ssd_v;
2054             }
2055         }
2056
2057         if( h->param.analyse.b_ssim )
2058         {
2059             int ssim_cnt;
2060             x264_emms();
2061             /* offset by 2 pixels to avoid alignment of ssim blocks with dct blocks,
2062              * and overlap by 4 */
2063             minpix_y += b_start ? 2 : -6;
2064             h->stat.frame.f_ssim +=
2065                 x264_pixel_ssim_wxh( &h->pixf,
2066                     h->fdec->plane[0] + 2+minpix_y*h->fdec->i_stride[0], h->fdec->i_stride[0],
2067                     h->fenc->plane[0] + 2+minpix_y*h->fenc->i_stride[0], h->fenc->i_stride[0],
2068                     h->param.i_width-2, maxpix_y-minpix_y, h->scratch_buffer, &ssim_cnt );
2069             h->stat.frame.i_ssim_cnt += ssim_cnt;
2070         }
2071     }
2072 }
2073
2074 static inline int x264_reference_update( x264_t *h )
2075 {
2076     if( !h->fdec->b_kept_as_ref )
2077     {
2078         if( h->i_thread_frames > 1 )
2079         {
2080             x264_frame_push_unused( h, h->fdec );
2081             h->fdec = x264_frame_pop_unused( h, 1 );
2082             if( !h->fdec )
2083                 return -1;
2084         }
2085         return 0;
2086     }
2087
2088     /* apply mmco from previous frame. */
2089     for( int i = 0; i < h->sh.i_mmco_command_count; i++ )
2090         for( int j = 0; h->frames.reference[j]; j++ )
2091             if( h->frames.reference[j]->i_poc == h->sh.mmco[i].i_poc )
2092                 x264_frame_push_unused( h, x264_frame_shift( &h->frames.reference[j] ) );
2093
2094     /* move frame in the buffer */
2095     x264_frame_push( h->frames.reference, h->fdec );
2096     if( h->frames.reference[h->sps->i_num_ref_frames] )
2097         x264_frame_push_unused( h, x264_frame_shift( h->frames.reference ) );
2098     h->fdec = x264_frame_pop_unused( h, 1 );
2099     if( !h->fdec )
2100         return -1;
2101     return 0;
2102 }
2103
2104 static inline void x264_reference_reset( x264_t *h )
2105 {
2106     while( h->frames.reference[0] )
2107         x264_frame_push_unused( h, x264_frame_pop( h->frames.reference ) );
2108     h->fdec->i_poc =
2109     h->fenc->i_poc = 0;
2110 }
2111
2112 static inline void x264_reference_hierarchy_reset( x264_t *h )
2113 {
2114     int ref;
2115     int b_hasdelayframe = 0;
2116
2117     /* look for delay frames -- chain must only contain frames that are disposable */
2118     for( int i = 0; h->frames.current[i] && IS_DISPOSABLE( h->frames.current[i]->i_type ); i++ )
2119         b_hasdelayframe |= h->frames.current[i]->i_coded
2120                         != h->frames.current[i]->i_frame + h->sps->vui.i_num_reorder_frames;
2121
2122     /* This function must handle b-pyramid and clear frames for open-gop */
2123     if( h->param.i_bframe_pyramid != X264_B_PYRAMID_STRICT && !b_hasdelayframe && h->frames.i_poc_last_open_gop == -1 )
2124         return;
2125
2126     /* Remove last BREF. There will never be old BREFs in the
2127      * dpb during a BREF decode when pyramid == STRICT */
2128     for( ref = 0; h->frames.reference[ref]; ref++ )
2129     {
2130         if( ( h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT
2131             && h->frames.reference[ref]->i_type == X264_TYPE_BREF )
2132             || ( h->frames.reference[ref]->i_poc < h->frames.i_poc_last_open_gop
2133             && h->sh.i_type != SLICE_TYPE_B ) )
2134         {
2135             int diff = h->i_frame_num - h->frames.reference[ref]->i_frame_num;
2136             h->sh.mmco[h->sh.i_mmco_command_count].i_difference_of_pic_nums = diff;
2137             h->sh.mmco[h->sh.i_mmco_command_count++].i_poc = h->frames.reference[ref]->i_poc;
2138             x264_frame_push_unused( h, x264_frame_shift( &h->frames.reference[ref] ) );
2139             h->b_ref_reorder[0] = 1;
2140             ref--;
2141         }
2142     }
2143
2144     /* Prepare room in the dpb for the delayed display time of the later b-frame's */
2145     if( h->param.i_bframe_pyramid )
2146         h->sh.i_mmco_remove_from_end = X264_MAX( ref + 2 - h->frames.i_max_dpb, 0 );
2147 }
2148
2149 static inline void x264_slice_init( x264_t *h, int i_nal_type, int i_global_qp )
2150 {
2151     /* ------------------------ Create slice header  ----------------------- */
2152     if( i_nal_type == NAL_SLICE_IDR )
2153     {
2154         x264_slice_header_init( h, &h->sh, h->sps, h->pps, h->i_idr_pic_id, h->i_frame_num, i_global_qp );
2155
2156         /* alternate id */
2157         h->i_idr_pic_id ^= 1;
2158     }
2159     else
2160     {
2161         x264_slice_header_init( h, &h->sh, h->sps, h->pps, -1, h->i_frame_num, i_global_qp );
2162
2163         h->sh.i_num_ref_idx_l0_active = h->i_ref[0] <= 0 ? 1 : h->i_ref[0];
2164         h->sh.i_num_ref_idx_l1_active = h->i_ref[1] <= 0 ? 1 : h->i_ref[1];
2165         if( h->sh.i_num_ref_idx_l0_active != h->pps->i_num_ref_idx_l0_default_active ||
2166             (h->sh.i_type == SLICE_TYPE_B && h->sh.i_num_ref_idx_l1_active != h->pps->i_num_ref_idx_l1_default_active) )
2167         {
2168             h->sh.b_num_ref_idx_override = 1;
2169         }
2170     }
2171
2172     if( h->fenc->i_type == X264_TYPE_BREF && h->param.b_bluray_compat && h->sh.i_mmco_command_count )
2173     {
2174         h->b_sh_backup = 1;
2175         h->sh_backup = h->sh;
2176     }
2177
2178     h->fdec->i_frame_num = h->sh.i_frame_num;
2179
2180     if( h->sps->i_poc_type == 0 )
2181     {
2182         h->sh.i_poc = h->fdec->i_poc;
2183         if( PARAM_INTERLACED )
2184         {
2185             h->sh.i_delta_poc_bottom = h->param.b_tff ? 1 : -1;
2186             h->sh.i_poc += h->sh.i_delta_poc_bottom == -1;
2187         }
2188         else
2189             h->sh.i_delta_poc_bottom = 0;
2190         h->fdec->i_delta_poc[0] = h->sh.i_delta_poc_bottom == -1;
2191         h->fdec->i_delta_poc[1] = h->sh.i_delta_poc_bottom ==  1;
2192     }
2193     else
2194     {
2195         /* Nothing to do ? */
2196     }
2197
2198     x264_macroblock_slice_init( h );
2199 }
2200
2201 typedef struct
2202 {
2203     int skip;
2204     uint8_t cabac_prevbyte;
2205     bs_t bs;
2206     x264_cabac_t cabac;
2207     x264_frame_stat_t stat;
2208     int last_qp;
2209     int last_dqp;
2210     int field_decoding_flag;
2211 } x264_bs_bak_t;
2212
2213 static ALWAYS_INLINE void x264_bitstream_backup( x264_t *h, x264_bs_bak_t *bak, int i_skip, int full )
2214 {
2215     if( full )
2216     {
2217         bak->stat = h->stat.frame;
2218         bak->last_qp = h->mb.i_last_qp;
2219         bak->last_dqp = h->mb.i_last_dqp;
2220         bak->field_decoding_flag = h->mb.field_decoding_flag;
2221     }
2222     else
2223     {
2224         bak->stat.i_mv_bits = h->stat.frame.i_mv_bits;
2225         bak->stat.i_tex_bits = h->stat.frame.i_tex_bits;
2226     }
2227     /* In the per-MB backup, we don't need the contexts because flushing the CABAC
2228      * encoder has no context dependency and in this case, a slice is ended (and
2229      * thus the content of all contexts are thrown away). */
2230     if( h->param.b_cabac )
2231     {
2232         if( full )
2233             memcpy( &bak->cabac, &h->cabac, sizeof(x264_cabac_t) );
2234         else
2235             memcpy( &bak->cabac, &h->cabac, offsetof(x264_cabac_t, f8_bits_encoded) );
2236         /* x264's CABAC writer modifies the previous byte during carry, so it has to be
2237          * backed up. */
2238         bak->cabac_prevbyte = h->cabac.p[-1];
2239     }
2240     else
2241     {
2242         bak->bs = h->out.bs;
2243         bak->skip = i_skip;
2244     }
2245 }
2246
2247 static ALWAYS_INLINE void x264_bitstream_restore( x264_t *h, x264_bs_bak_t *bak, int *skip, int full )
2248 {
2249     if( full )
2250     {
2251         h->stat.frame = bak->stat;
2252         h->mb.i_last_qp = bak->last_qp;
2253         h->mb.i_last_dqp = bak->last_dqp;
2254         h->mb.field_decoding_flag = bak->field_decoding_flag;
2255     }
2256     else
2257     {
2258         h->stat.frame.i_mv_bits = bak->stat.i_mv_bits;
2259         h->stat.frame.i_tex_bits = bak->stat.i_tex_bits;
2260     }
2261     if( h->param.b_cabac )
2262     {
2263         if( full )
2264             memcpy( &h->cabac, &bak->cabac, sizeof(x264_cabac_t) );
2265         else
2266             memcpy( &h->cabac, &bak->cabac, offsetof(x264_cabac_t, f8_bits_encoded) );
2267         h->cabac.p[-1] = bak->cabac_prevbyte;
2268     }
2269     else
2270     {
2271         h->out.bs = bak->bs;
2272         *skip = bak->skip;
2273     }
2274 }
2275
2276 static int x264_slice_write( x264_t *h )
2277 {
2278     int i_skip;
2279     int mb_xy, i_mb_x, i_mb_y;
2280     /* NALUs other than the first use a 3-byte startcode.
2281      * Add one extra byte for the rbsp, and one more for the final CABAC putbyte.
2282      * Then add an extra 5 bytes just in case, to account for random NAL escapes and
2283      * other inaccuracies. */
2284     int overhead_guess = (NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal)) + 1 + h->param.b_cabac + 5;
2285     int slice_max_size = h->param.i_slice_max_size > 0 ? (h->param.i_slice_max_size-overhead_guess)*8 : 0;
2286     int back_up_bitstream = slice_max_size || (!h->param.b_cabac && h->sps->i_profile_idc < PROFILE_HIGH);
2287     int starting_bits = bs_pos(&h->out.bs);
2288     int b_deblock = h->sh.i_disable_deblocking_filter_idc != 1;
2289     int b_hpel = h->fdec->b_kept_as_ref;
2290     int orig_last_mb = h->sh.i_last_mb;
2291     int thread_last_mb = h->i_threadslice_end * h->mb.i_mb_width - 1;
2292     uint8_t *last_emu_check;
2293 #define BS_BAK_SLICE_MAX_SIZE 0
2294 #define BS_BAK_SLICE_MIN_MBS  1
2295 #define BS_BAK_ROW_VBV        2
2296     x264_bs_bak_t bs_bak[3];
2297     b_deblock &= b_hpel || h->param.b_full_recon || h->param.psz_dump_yuv;
2298     bs_realign( &h->out.bs );
2299
2300     /* Slice */
2301     x264_nal_start( h, h->i_nal_type, h->i_nal_ref_idc );
2302     h->out.nal[h->out.i_nal].i_first_mb = h->sh.i_first_mb;
2303
2304     /* Slice header */
2305     x264_macroblock_thread_init( h );
2306
2307     /* Set the QP equal to the first QP in the slice for more accurate CABAC initialization. */
2308     h->mb.i_mb_xy = h->sh.i_first_mb;
2309     h->sh.i_qp = x264_ratecontrol_mb_qp( h );
2310     h->sh.i_qp = SPEC_QP( h->sh.i_qp );
2311     h->sh.i_qp_delta = h->sh.i_qp - h->pps->i_pic_init_qp;
2312
2313     x264_slice_header_write( &h->out.bs, &h->sh, h->i_nal_ref_idc );
2314     if( h->param.b_cabac )
2315     {
2316         /* alignment needed */
2317         bs_align_1( &h->out.bs );
2318
2319         /* init cabac */
2320         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 );
2321         x264_cabac_encode_init ( &h->cabac, h->out.bs.p, h->out.bs.p_end );
2322         last_emu_check = h->cabac.p;
2323     }
2324     else
2325         last_emu_check = h->out.bs.p;
2326     h->mb.i_last_qp = h->sh.i_qp;
2327     h->mb.i_last_dqp = 0;
2328     h->mb.field_decoding_flag = 0;
2329
2330     i_mb_y = h->sh.i_first_mb / h->mb.i_mb_width;
2331     i_mb_x = h->sh.i_first_mb % h->mb.i_mb_width;
2332     i_skip = 0;
2333
2334     while( 1 )
2335     {
2336         mb_xy = i_mb_x + i_mb_y * h->mb.i_mb_width;
2337         int mb_spos = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
2338
2339         if( i_mb_x == 0 )
2340         {
2341             if( x264_bitstream_check_buffer( h ) )
2342                 return -1;
2343             if( !(i_mb_y & SLICE_MBAFF) && h->param.rc.i_vbv_buffer_size )
2344                 x264_bitstream_backup( h, &bs_bak[BS_BAK_ROW_VBV], i_skip, 1 );
2345             if( !h->mb.b_reencode_mb )
2346                 x264_fdec_filter_row( h, i_mb_y, 0 );
2347         }
2348
2349         if( !(i_mb_y & SLICE_MBAFF) && back_up_bitstream )
2350         {
2351             x264_bitstream_backup( h, &bs_bak[BS_BAK_SLICE_MAX_SIZE], i_skip, 0 );
2352             if( slice_max_size && (thread_last_mb+1-mb_xy) == h->param.i_slice_min_mbs )
2353                 x264_bitstream_backup( h, &bs_bak[BS_BAK_SLICE_MIN_MBS], i_skip, 0 );
2354         }
2355
2356         if( PARAM_INTERLACED )
2357         {
2358             if( h->mb.b_adaptive_mbaff )
2359             {
2360                 if( !(i_mb_y&1) )
2361                 {
2362                     /* FIXME: VSAD is fast but fairly poor at choosing the best interlace type. */
2363                     h->mb.b_interlaced = x264_field_vsad( h, i_mb_x, i_mb_y );
2364                     memcpy( &h->zigzagf, MB_INTERLACED ? &h->zigzagf_interlaced : &h->zigzagf_progressive, sizeof(h->zigzagf) );
2365                     if( !MB_INTERLACED && (i_mb_y+2) == h->mb.i_mb_height )
2366                         x264_expand_border_mbpair( h, i_mb_x, i_mb_y );
2367                 }
2368             }
2369             h->mb.field[mb_xy] = MB_INTERLACED;
2370         }
2371
2372         /* load cache */
2373         if( SLICE_MBAFF )
2374             x264_macroblock_cache_load_interlaced( h, i_mb_x, i_mb_y );
2375         else
2376             x264_macroblock_cache_load_progressive( h, i_mb_x, i_mb_y );
2377
2378         x264_macroblock_analyse( h );
2379
2380         /* encode this macroblock -> be careful it can change the mb type to P_SKIP if needed */
2381 reencode:
2382         x264_macroblock_encode( h );
2383
2384         if( h->param.b_cabac )
2385         {
2386             if( mb_xy > h->sh.i_first_mb && !(SLICE_MBAFF && (i_mb_y&1)) )
2387                 x264_cabac_encode_terminal( &h->cabac );
2388
2389             if( IS_SKIP( h->mb.i_type ) )
2390                 x264_cabac_mb_skip( h, 1 );
2391             else
2392             {
2393                 if( h->sh.i_type != SLICE_TYPE_I )
2394                     x264_cabac_mb_skip( h, 0 );
2395                 x264_macroblock_write_cabac( h, &h->cabac );
2396             }
2397         }
2398         else
2399         {
2400             if( IS_SKIP( h->mb.i_type ) )
2401                 i_skip++;
2402             else
2403             {
2404                 if( h->sh.i_type != SLICE_TYPE_I )
2405                 {
2406                     bs_write_ue( &h->out.bs, i_skip );  /* skip run */
2407                     i_skip = 0;
2408                 }
2409                 x264_macroblock_write_cavlc( h );
2410                 /* If there was a CAVLC level code overflow, try again at a higher QP. */
2411                 if( h->mb.b_overflow )
2412                 {
2413                     h->mb.i_chroma_qp = h->chroma_qp_table[++h->mb.i_qp];
2414                     h->mb.i_skip_intra = 0;
2415                     h->mb.b_skip_mc = 0;
2416                     h->mb.b_overflow = 0;
2417                     x264_bitstream_restore( h, &bs_bak[BS_BAK_SLICE_MAX_SIZE], &i_skip, 0 );
2418                     goto reencode;
2419                 }
2420             }
2421         }
2422
2423         int total_bits = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
2424         int mb_size = total_bits - mb_spos;
2425
2426         if( slice_max_size && (!SLICE_MBAFF || (i_mb_y&1)) )
2427         {
2428             /* Count the skip run, just in case. */
2429             if( !h->param.b_cabac )
2430                 total_bits += bs_size_ue_big( i_skip );
2431             /* Check for escape bytes. */
2432             uint8_t *end = h->param.b_cabac ? h->cabac.p : h->out.bs.p;
2433             for( ; last_emu_check < end - 2; last_emu_check++ )
2434                 if( last_emu_check[0] == 0 && last_emu_check[1] == 0 && last_emu_check[2] <= 3 )
2435                 {
2436                     slice_max_size -= 8;
2437                     last_emu_check++;
2438                 }
2439             /* We'll just re-encode this last macroblock if we go over the max slice size. */
2440             if( total_bits - starting_bits > slice_max_size && !h->mb.b_reencode_mb )
2441             {
2442                 if( !x264_frame_new_slice( h, h->fdec ) )
2443                 {
2444                     /* Handle the most obnoxious slice-min-mbs edge case: we need to end the slice
2445                      * because it's gone over the maximum size, but doing so would violate slice-min-mbs.
2446                      * If possible, roll back to the last checkpoint and try again.
2447                      * We could try raising QP, but that would break in the case where a slice spans multiple
2448                      * rows, which the re-encoding infrastructure can't currently handle. */
2449                     if( mb_xy < thread_last_mb && (thread_last_mb+1-mb_xy) < h->param.i_slice_min_mbs )
2450                     {
2451                         if( thread_last_mb-h->param.i_slice_min_mbs < h->sh.i_first_mb+h->param.i_slice_min_mbs )
2452                         {
2453                             x264_log( h, X264_LOG_WARNING, "slice-max-size violated (frame %d, cause: slice-min-mbs)\n", h->i_frame );
2454                             slice_max_size = 0;
2455                             goto cont;
2456                         }
2457                         x264_bitstream_restore( h, &bs_bak[BS_BAK_SLICE_MIN_MBS], &i_skip, 0 );
2458                         h->mb.b_reencode_mb = 1;
2459                         h->sh.i_last_mb = thread_last_mb-h->param.i_slice_min_mbs;
2460                         break;
2461                     }
2462                     if( mb_xy-SLICE_MBAFF*h->mb.i_mb_stride != h->sh.i_first_mb )
2463                     {
2464                         x264_bitstream_restore( h, &bs_bak[BS_BAK_SLICE_MAX_SIZE], &i_skip, 0 );
2465                         h->mb.b_reencode_mb = 1;
2466                         if( SLICE_MBAFF )
2467                         {
2468                             // set to bottom of previous mbpair
2469                             if( i_mb_x )
2470                                 h->sh.i_last_mb = mb_xy-1+h->mb.i_mb_stride*(!(i_mb_y&1));
2471                             else
2472                                 h->sh.i_last_mb = (i_mb_y-2+!(i_mb_y&1))*h->mb.i_mb_stride + h->mb.i_mb_width - 1;
2473                         }
2474                         else
2475                             h->sh.i_last_mb = mb_xy-1;
2476                         break;
2477                     }
2478                     else
2479                         h->sh.i_last_mb = mb_xy;
2480                 }
2481                 else
2482                     slice_max_size = 0;
2483             }
2484         }
2485 cont:
2486         h->mb.b_reencode_mb = 0;
2487
2488 #if HAVE_VISUALIZE
2489         if( h->param.b_visualize )
2490             x264_visualize_mb( h );
2491 #endif
2492
2493         /* save cache */
2494         x264_macroblock_cache_save( h );
2495
2496         if( x264_ratecontrol_mb( h, mb_size ) < 0 )
2497         {
2498             x264_bitstream_restore( h, &bs_bak[BS_BAK_ROW_VBV], &i_skip, 1 );
2499             h->mb.b_reencode_mb = 1;
2500             i_mb_x = 0;
2501             i_mb_y = i_mb_y - SLICE_MBAFF;
2502             h->mb.i_mb_prev_xy = i_mb_y * h->mb.i_mb_stride - 1;
2503             h->sh.i_last_mb = orig_last_mb;
2504             continue;
2505         }
2506
2507         /* accumulate mb stats */
2508         h->stat.frame.i_mb_count[h->mb.i_type]++;
2509
2510         int b_intra = IS_INTRA( h->mb.i_type );
2511         int b_skip = IS_SKIP( h->mb.i_type );
2512         if( h->param.i_log_level >= X264_LOG_INFO || h->param.rc.b_stat_write )
2513         {
2514             if( !b_intra && !b_skip && !IS_DIRECT( h->mb.i_type ) )
2515             {
2516                 if( h->mb.i_partition != D_8x8 )
2517                         h->stat.frame.i_mb_partition[h->mb.i_partition] += 4;
2518                     else
2519                         for( int i = 0; i < 4; i++ )
2520                             h->stat.frame.i_mb_partition[h->mb.i_sub_partition[i]] ++;
2521                 if( h->param.i_frame_reference > 1 )
2522                     for( int i_list = 0; i_list <= (h->sh.i_type == SLICE_TYPE_B); i_list++ )
2523                         for( int i = 0; i < 4; i++ )
2524                         {
2525                             int i_ref = h->mb.cache.ref[i_list][ x264_scan8[4*i] ];
2526                             if( i_ref >= 0 )
2527                                 h->stat.frame.i_mb_count_ref[i_list][i_ref] ++;
2528                         }
2529             }
2530         }
2531
2532         if( h->param.i_log_level >= X264_LOG_INFO )
2533         {
2534             if( h->mb.i_cbp_luma | h->mb.i_cbp_chroma )
2535             {
2536                 if( CHROMA444 )
2537                 {
2538                     for( int i = 0; i < 4; i++ )
2539                         if( h->mb.i_cbp_luma & (1 << i) )
2540                             for( int p = 0; p < 3; p++ )
2541                             {
2542                                 int s8 = i*4+p*16;
2543                                 int nnz8x8 = M16( &h->mb.cache.non_zero_count[x264_scan8[s8]+0] )
2544                                            | M16( &h->mb.cache.non_zero_count[x264_scan8[s8]+8] );
2545                                 h->stat.frame.i_mb_cbp[!b_intra + p*2] += !!nnz8x8;
2546                             }
2547                 }
2548                 else
2549                 {
2550                     int cbpsum = (h->mb.i_cbp_luma&1) + ((h->mb.i_cbp_luma>>1)&1)
2551                                + ((h->mb.i_cbp_luma>>2)&1) + (h->mb.i_cbp_luma>>3);
2552                     h->stat.frame.i_mb_cbp[!b_intra + 0] += cbpsum;
2553                     h->stat.frame.i_mb_cbp[!b_intra + 2] += !!h->mb.i_cbp_chroma;
2554                     h->stat.frame.i_mb_cbp[!b_intra + 4] += h->mb.i_cbp_chroma >> 1;
2555                 }
2556             }
2557             if( h->mb.i_cbp_luma && !b_intra )
2558             {
2559                 h->stat.frame.i_mb_count_8x8dct[0] ++;
2560                 h->stat.frame.i_mb_count_8x8dct[1] += h->mb.b_transform_8x8;
2561             }
2562             if( b_intra && h->mb.i_type != I_PCM )
2563             {
2564                 if( h->mb.i_type == I_16x16 )
2565                     h->stat.frame.i_mb_pred_mode[0][h->mb.i_intra16x16_pred_mode]++;
2566                 else if( h->mb.i_type == I_8x8 )
2567                     for( int i = 0; i < 16; i += 4 )
2568                         h->stat.frame.i_mb_pred_mode[1][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
2569                 else //if( h->mb.i_type == I_4x4 )
2570                     for( int i = 0; i < 16; i++ )
2571                         h->stat.frame.i_mb_pred_mode[2][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
2572                 h->stat.frame.i_mb_pred_mode[3][x264_mb_chroma_pred_mode_fix[h->mb.i_chroma_pred_mode]]++;
2573             }
2574             h->stat.frame.i_mb_field[b_intra?0:b_skip?2:1] += MB_INTERLACED;
2575         }
2576
2577         /* calculate deblock strength values (actual deblocking is done per-row along with hpel) */
2578         if( b_deblock )
2579             x264_macroblock_deblock_strength( h );
2580
2581         if( mb_xy == h->sh.i_last_mb )
2582             break;
2583
2584         if( SLICE_MBAFF )
2585         {
2586             i_mb_x += i_mb_y & 1;
2587             i_mb_y ^= i_mb_x < h->mb.i_mb_width;
2588         }
2589         else
2590             i_mb_x++;
2591         if( i_mb_x == h->mb.i_mb_width )
2592         {
2593             i_mb_y++;
2594             i_mb_x = 0;
2595         }
2596     }
2597     if( h->sh.i_last_mb < h->sh.i_first_mb )
2598         return 0;
2599
2600     h->out.nal[h->out.i_nal].i_last_mb = h->sh.i_last_mb;
2601
2602     if( h->param.b_cabac )
2603     {
2604         x264_cabac_encode_flush( h, &h->cabac );
2605         h->out.bs.p = h->cabac.p;
2606     }
2607     else
2608     {
2609         if( i_skip > 0 )
2610             bs_write_ue( &h->out.bs, i_skip );  /* last skip run */
2611         /* rbsp_slice_trailing_bits */
2612         bs_rbsp_trailing( &h->out.bs );
2613         bs_flush( &h->out.bs );
2614     }
2615     if( x264_nal_end( h ) )
2616         return -1;
2617
2618     if( h->sh.i_last_mb == (h->i_threadslice_end * h->mb.i_mb_width - 1) )
2619     {
2620         h->stat.frame.i_misc_bits = bs_pos( &h->out.bs )
2621                                   + (h->out.i_nal*NALU_OVERHEAD * 8)
2622                                   - h->stat.frame.i_tex_bits
2623                                   - h->stat.frame.i_mv_bits;
2624         x264_fdec_filter_row( h, h->i_threadslice_end, 0 );
2625
2626         if( h->param.b_sliced_threads )
2627         {
2628             /* Tell the main thread we're done. */
2629             x264_threadslice_cond_broadcast( h, 1 );
2630             /* Do hpel now */
2631             for( int mb_y = h->i_threadslice_start; mb_y <= h->i_threadslice_end; mb_y++ )
2632                 x264_fdec_filter_row( h, mb_y, 1 );
2633             x264_threadslice_cond_broadcast( h, 2 );
2634             /* Do the first row of hpel, now that the previous slice is done */
2635             if( h->i_thread_idx > 0 )
2636             {
2637                 x264_threadslice_cond_wait( h->thread[h->i_thread_idx-1], 2 );
2638                 x264_fdec_filter_row( h, h->i_threadslice_start + (1 << SLICE_MBAFF), 2 );
2639             }
2640         }
2641
2642         /* Free mb info after the last thread's done using it */
2643         if( h->fdec->mb_info_free && (!h->param.b_sliced_threads || h->i_thread_idx == (h->param.i_threads-1)) )
2644         {
2645             h->fdec->mb_info_free( h->fdec->mb_info );
2646             h->fdec->mb_info = NULL;
2647             h->fdec->mb_info_free = NULL;
2648         }
2649     }
2650
2651     return 0;
2652 }
2653
2654 static void x264_thread_sync_context( x264_t *dst, x264_t *src )
2655 {
2656     if( dst == src )
2657         return;
2658
2659     // reference counting
2660     for( x264_frame_t **f = src->frames.reference; *f; f++ )
2661         (*f)->i_reference_count++;
2662     for( x264_frame_t **f = dst->frames.reference; *f; f++ )
2663         x264_frame_push_unused( src, *f );
2664     src->fdec->i_reference_count++;
2665     x264_frame_push_unused( src, dst->fdec );
2666
2667     // copy everything except the per-thread pointers and the constants.
2668     memcpy( &dst->i_frame, &src->i_frame, offsetof(x264_t, mb.type) - offsetof(x264_t, i_frame) );
2669     dst->param = src->param;
2670     dst->stat = src->stat;
2671     dst->pixf = src->pixf;
2672 }
2673
2674 static void x264_thread_sync_stat( x264_t *dst, x264_t *src )
2675 {
2676     if( dst == src )
2677         return;
2678     memcpy( &dst->stat.i_frame_count, &src->stat.i_frame_count, sizeof(dst->stat) - sizeof(dst->stat.frame) );
2679 }
2680
2681 static void *x264_slices_write( x264_t *h )
2682 {
2683     int i_slice_num = 0;
2684     int last_thread_mb = h->sh.i_last_mb;
2685
2686 #if HAVE_VISUALIZE
2687     if( h->param.b_visualize )
2688         if( x264_visualize_init( h ) )
2689             goto fail;
2690 #endif
2691
2692     /* init stats */
2693     memset( &h->stat.frame, 0, sizeof(h->stat.frame) );
2694     h->mb.b_reencode_mb = 0;
2695     while( h->sh.i_first_mb + SLICE_MBAFF*h->mb.i_mb_stride <= last_thread_mb )
2696     {
2697         h->sh.i_last_mb = last_thread_mb;
2698         if( !i_slice_num || !x264_frame_new_slice( h, h->fdec ) )
2699         {
2700             if( h->param.i_slice_max_mbs )
2701             {
2702                 if( SLICE_MBAFF )
2703                 {
2704                     // convert first to mbaff form, add slice-max-mbs, then convert back to normal form
2705                     int last_mbaff = 2*(h->sh.i_first_mb % h->mb.i_mb_width)
2706                         + h->mb.i_mb_width*(h->sh.i_first_mb / h->mb.i_mb_width)
2707                         + h->param.i_slice_max_mbs - 1;
2708                     int last_x = (last_mbaff % (2*h->mb.i_mb_width))/2;
2709                     int last_y = (last_mbaff / (2*h->mb.i_mb_width))*2 + 1;
2710                     h->sh.i_last_mb = last_x + h->mb.i_mb_stride*last_y;
2711                 }
2712                 else
2713                 {
2714                     h->sh.i_last_mb = h->sh.i_first_mb + h->param.i_slice_max_mbs - 1;
2715                     if( h->sh.i_last_mb < last_thread_mb && last_thread_mb - h->sh.i_last_mb < h->param.i_slice_min_mbs )
2716                         h->sh.i_last_mb = last_thread_mb - h->param.i_slice_min_mbs;
2717                 }
2718                 i_slice_num++;
2719             }
2720             else if( h->param.i_slice_count && !h->param.b_sliced_threads )
2721             {
2722                 int height = h->mb.i_mb_height >> PARAM_INTERLACED;
2723                 int width = h->mb.i_mb_width << PARAM_INTERLACED;
2724                 i_slice_num++;
2725                 h->sh.i_last_mb = (height * i_slice_num + h->param.i_slice_count/2) / h->param.i_slice_count * width - 1;
2726             }
2727         }
2728         h->sh.i_last_mb = X264_MIN( h->sh.i_last_mb, last_thread_mb );
2729         if( x264_stack_align( x264_slice_write, h ) )
2730             goto fail;
2731         h->sh.i_first_mb = h->sh.i_last_mb + 1;
2732         // if i_first_mb is not the last mb in a row then go to the next mb in MBAFF order
2733         if( SLICE_MBAFF && h->sh.i_first_mb % h->mb.i_mb_width )
2734             h->sh.i_first_mb -= h->mb.i_mb_stride;
2735     }
2736
2737 #if HAVE_VISUALIZE
2738     if( h->param.b_visualize )
2739     {
2740         x264_visualize_show( h );
2741         x264_visualize_close( h );
2742     }
2743 #endif
2744
2745     return (void *)0;
2746
2747 fail:
2748     /* Tell other threads we're done, so they wouldn't wait for it */
2749     if( h->param.b_sliced_threads )
2750         x264_threadslice_cond_broadcast( h, 2 );
2751     return (void *)-1;
2752 }
2753
2754 static int x264_threaded_slices_write( x264_t *h )
2755 {
2756     /* set first/last mb and sync contexts */
2757     for( int i = 0; i < h->param.i_threads; i++ )
2758     {
2759         x264_t *t = h->thread[i];
2760         if( i )
2761         {
2762             t->param = h->param;
2763             memcpy( &t->i_frame, &h->i_frame, offsetof(x264_t, rc) - offsetof(x264_t, i_frame) );
2764         }
2765         int height = h->mb.i_mb_height >> PARAM_INTERLACED;
2766         t->i_threadslice_start = ((height *  i    + h->param.i_slice_count/2) / h->param.i_threads) << PARAM_INTERLACED;
2767         t->i_threadslice_end   = ((height * (i+1) + h->param.i_slice_count/2) / h->param.i_threads) << PARAM_INTERLACED;
2768         t->sh.i_first_mb = t->i_threadslice_start * h->mb.i_mb_width;
2769         t->sh.i_last_mb  =   t->i_threadslice_end * h->mb.i_mb_width - 1;
2770     }
2771
2772     x264_stack_align( x264_analyse_weight_frame, h, h->mb.i_mb_height*16 + 16 );
2773
2774     x264_threads_distribute_ratecontrol( h );
2775
2776     /* setup */
2777     for( int i = 0; i < h->param.i_threads; i++ )
2778     {
2779         h->thread[i]->i_thread_idx = i;
2780         h->thread[i]->b_thread_active = 1;
2781         x264_threadslice_cond_broadcast( h->thread[i], 0 );
2782     }
2783     /* dispatch */
2784     for( int i = 0; i < h->param.i_threads; i++ )
2785         x264_threadpool_run( h->threadpool, (void*)x264_slices_write, h->thread[i] );
2786     /* wait */
2787     for( int i = 0; i < h->param.i_threads; i++ )
2788         x264_threadslice_cond_wait( h->thread[i], 1 );
2789
2790     x264_threads_merge_ratecontrol( h );
2791
2792     for( int i = 1; i < h->param.i_threads; i++ )
2793     {
2794         x264_t *t = h->thread[i];
2795         for( int j = 0; j < t->out.i_nal; j++ )
2796         {
2797             h->out.nal[h->out.i_nal] = t->out.nal[j];
2798             h->out.i_nal++;
2799             x264_nal_check_buffer( h );
2800         }
2801         /* All entries in stat.frame are ints except for ssd/ssim. */
2802         for( int j = 0; j < (offsetof(x264_t,stat.frame.i_ssd) - offsetof(x264_t,stat.frame.i_mv_bits)) / sizeof(int); j++ )
2803             ((int*)&h->stat.frame)[j] += ((int*)&t->stat.frame)[j];
2804         for( int j = 0; j < 3; j++ )
2805             h->stat.frame.i_ssd[j] += t->stat.frame.i_ssd[j];
2806         h->stat.frame.f_ssim += t->stat.frame.f_ssim;
2807         h->stat.frame.i_ssim_cnt += t->stat.frame.i_ssim_cnt;
2808     }
2809
2810     return 0;
2811 }
2812
2813 void x264_encoder_intra_refresh( x264_t *h )
2814 {
2815     h = h->thread[h->i_thread_phase];
2816     h->b_queued_intra_refresh = 1;
2817 }
2818
2819 int x264_encoder_invalidate_reference( x264_t *h, int64_t pts )
2820 {
2821     if( h->param.i_bframe )
2822     {
2823         x264_log( h, X264_LOG_ERROR, "x264_encoder_invalidate_reference is not supported with B-frames enabled\n" );
2824         return -1;
2825     }
2826     if( h->param.b_intra_refresh )
2827     {
2828         x264_log( h, X264_LOG_ERROR, "x264_encoder_invalidate_reference is not supported with intra refresh enabled\n" );
2829         return -1;
2830     }
2831     h = h->thread[h->i_thread_phase];
2832     if( pts >= h->i_last_idr_pts )
2833     {
2834         for( int i = 0; h->frames.reference[i]; i++ )
2835             if( pts <= h->frames.reference[i]->i_pts )
2836                 h->frames.reference[i]->b_corrupt = 1;
2837         if( pts <= h->fdec->i_pts )
2838             h->fdec->b_corrupt = 1;
2839     }
2840     return 0;
2841 }
2842
2843 /****************************************************************************
2844  * x264_encoder_encode:
2845  *  XXX: i_poc   : is the poc of the current given picture
2846  *       i_frame : is the number of the frame being coded
2847  *  ex:  type frame poc
2848  *       I      0   2*0
2849  *       P      1   2*3
2850  *       B      2   2*1
2851  *       B      3   2*2
2852  *       P      4   2*6
2853  *       B      5   2*4
2854  *       B      6   2*5
2855  ****************************************************************************/
2856 int     x264_encoder_encode( x264_t *h,
2857                              x264_nal_t **pp_nal, int *pi_nal,
2858                              x264_picture_t *pic_in,
2859                              x264_picture_t *pic_out )
2860 {
2861     x264_t *thread_current, *thread_prev, *thread_oldest;
2862     int i_nal_type, i_nal_ref_idc, i_global_qp;
2863     int overhead = NALU_OVERHEAD;
2864
2865     if( h->i_thread_frames > 1 )
2866     {
2867         thread_prev    = h->thread[ h->i_thread_phase ];
2868         h->i_thread_phase = (h->i_thread_phase + 1) % h->i_thread_frames;
2869         thread_current = h->thread[ h->i_thread_phase ];
2870         thread_oldest  = h->thread[ (h->i_thread_phase + 1) % h->i_thread_frames ];
2871         x264_thread_sync_context( thread_current, thread_prev );
2872         x264_thread_sync_ratecontrol( thread_current, thread_prev, thread_oldest );
2873         h = thread_current;
2874     }
2875     else
2876     {
2877         thread_current =
2878         thread_oldest  = h;
2879     }
2880 #if HAVE_MMX
2881     if( h->param.cpu&X264_CPU_SSE_MISALIGN )
2882         x264_cpu_mask_misalign_sse();
2883 #endif
2884     h->i_cpb_delay_pir_offset = h->i_cpb_delay_pir_offset_next;
2885
2886     /* no data out */
2887     *pi_nal = 0;
2888     *pp_nal = NULL;
2889
2890     /* ------------------- Setup new frame from picture -------------------- */
2891     if( pic_in != NULL )
2892     {
2893         /* 1: Copy the picture to a frame and move it to a buffer */
2894         x264_frame_t *fenc = x264_frame_pop_unused( h, 0 );
2895         if( !fenc )
2896             return -1;
2897
2898         if( x264_frame_copy_picture( h, fenc, pic_in ) < 0 )
2899             return -1;
2900
2901         if( h->param.i_width != 16 * h->mb.i_mb_width ||
2902             h->param.i_height != 16 * h->mb.i_mb_height )
2903             x264_frame_expand_border_mod16( h, fenc );
2904
2905         fenc->i_frame = h->frames.i_input++;
2906
2907         if( fenc->i_frame == 0 )
2908             h->frames.i_first_pts = fenc->i_pts;
2909         if( h->frames.i_bframe_delay && fenc->i_frame == h->frames.i_bframe_delay )
2910             h->frames.i_bframe_delay_time = fenc->i_pts - h->frames.i_first_pts;
2911
2912         if( h->param.b_vfr_input && fenc->i_pts <= h->frames.i_largest_pts )
2913             x264_log( h, X264_LOG_WARNING, "non-strictly-monotonic PTS\n" );
2914
2915         h->frames.i_second_largest_pts = h->frames.i_largest_pts;
2916         h->frames.i_largest_pts = fenc->i_pts;
2917
2918         if( (fenc->i_pic_struct < PIC_STRUCT_AUTO) || (fenc->i_pic_struct > PIC_STRUCT_TRIPLE) )
2919             fenc->i_pic_struct = PIC_STRUCT_AUTO;
2920
2921         if( fenc->i_pic_struct == PIC_STRUCT_AUTO )
2922         {
2923 #if HAVE_INTERLACED
2924             int b_interlaced = fenc->param ? fenc->param->b_interlaced : h->param.b_interlaced;
2925 #else
2926             int b_interlaced = 0;
2927 #endif
2928             if( b_interlaced )
2929             {
2930                 int b_tff = fenc->param ? fenc->param->b_tff : h->param.b_tff;
2931                 fenc->i_pic_struct = b_tff ? PIC_STRUCT_TOP_BOTTOM : PIC_STRUCT_BOTTOM_TOP;
2932             }
2933             else
2934                 fenc->i_pic_struct = PIC_STRUCT_PROGRESSIVE;
2935         }
2936
2937         if( h->param.rc.b_mb_tree && h->param.rc.b_stat_read )
2938         {
2939             if( x264_macroblock_tree_read( h, fenc, pic_in->prop.quant_offsets ) )
2940                 return -1;
2941         }
2942         else
2943             x264_stack_align( x264_adaptive_quant_frame, h, fenc, pic_in->prop.quant_offsets );
2944
2945         if( pic_in->prop.quant_offsets_free )
2946             pic_in->prop.quant_offsets_free( pic_in->prop.quant_offsets );
2947
2948         if( h->frames.b_have_lowres )
2949             x264_frame_init_lowres( h, fenc );
2950
2951         /* 2: Place the frame into the queue for its slice type decision */
2952         x264_lookahead_put_frame( h, fenc );
2953
2954         if( h->frames.i_input <= h->frames.i_delay + 1 - h->i_thread_frames )
2955         {
2956             /* Nothing yet to encode, waiting for filling of buffers */
2957             pic_out->i_type = X264_TYPE_AUTO;
2958             return 0;
2959         }
2960     }
2961     else
2962     {
2963         /* signal kills for lookahead thread */
2964         x264_pthread_mutex_lock( &h->lookahead->ifbuf.mutex );
2965         h->lookahead->b_exit_thread = 1;
2966         x264_pthread_cond_broadcast( &h->lookahead->ifbuf.cv_fill );
2967         x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
2968     }
2969
2970     h->i_frame++;
2971     /* 3: The picture is analyzed in the lookahead */
2972     if( !h->frames.current[0] )
2973         x264_lookahead_get_frames( h );
2974
2975     if( !h->frames.current[0] && x264_lookahead_is_empty( h ) )
2976         return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
2977
2978     /* ------------------- Get frame to be encoded ------------------------- */
2979     /* 4: get picture to encode */
2980     h->fenc = x264_frame_shift( h->frames.current );
2981
2982     /* If applicable, wait for previous frame reconstruction to finish */
2983     if( h->param.b_sliced_threads )
2984         if( x264_threadpool_wait_all( h ) < 0 )
2985             return -1;
2986
2987     if( h->i_frame == h->i_thread_frames - 1 )
2988         h->i_reordered_pts_delay = h->fenc->i_reordered_pts;
2989     if( h->fenc->param )
2990     {
2991         x264_encoder_reconfig( h, h->fenc->param );
2992         if( h->fenc->param->param_free )
2993         {
2994             h->fenc->param->param_free( h->fenc->param );
2995             h->fenc->param = NULL;
2996         }
2997     }
2998
2999     // ok to call this before encoding any frames, since the initial values of fdec have b_kept_as_ref=0
3000     if( x264_reference_update( h ) )
3001         return -1;
3002     h->fdec->i_lines_completed = -1;
3003
3004     if( !IS_X264_TYPE_I( h->fenc->i_type ) )
3005     {
3006         int valid_refs_left = 0;
3007         for( int i = 0; h->frames.reference[i]; i++ )
3008             if( !h->frames.reference[i]->b_corrupt )
3009                 valid_refs_left++;
3010         /* No valid reference frames left: force an IDR. */
3011         if( !valid_refs_left )
3012         {
3013             h->fenc->b_keyframe = 1;
3014             h->fenc->i_type = X264_TYPE_IDR;
3015         }
3016     }
3017
3018     if( h->fenc->b_keyframe )
3019     {
3020         h->frames.i_last_keyframe = h->fenc->i_frame;
3021         if( h->fenc->i_type == X264_TYPE_IDR )
3022         {
3023             h->i_frame_num = 0;
3024             h->frames.i_last_idr = h->fenc->i_frame;
3025         }
3026     }
3027     h->sh.i_mmco_command_count =
3028     h->sh.i_mmco_remove_from_end = 0;
3029     h->b_ref_reorder[0] =
3030     h->b_ref_reorder[1] = 0;
3031     h->fdec->i_poc =
3032     h->fenc->i_poc = 2 * ( h->fenc->i_frame - X264_MAX( h->frames.i_last_idr, 0 ) );
3033
3034     /* ------------------- Setup frame context ----------------------------- */
3035     /* 5: Init data dependent of frame type */
3036     if( h->fenc->i_type == X264_TYPE_IDR )
3037     {
3038         /* reset ref pictures */
3039         i_nal_type    = NAL_SLICE_IDR;
3040         i_nal_ref_idc = NAL_PRIORITY_HIGHEST;
3041         h->sh.i_type = SLICE_TYPE_I;
3042         x264_reference_reset( h );
3043         h->frames.i_poc_last_open_gop = -1;
3044     }
3045     else if( h->fenc->i_type == X264_TYPE_I )
3046     {
3047         i_nal_type    = NAL_SLICE;
3048         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
3049         h->sh.i_type = SLICE_TYPE_I;
3050         x264_reference_hierarchy_reset( h );
3051         if( h->param.b_open_gop )
3052             h->frames.i_poc_last_open_gop = h->fenc->b_keyframe ? h->fenc->i_poc : -1;
3053     }
3054     else if( h->fenc->i_type == X264_TYPE_P )
3055     {
3056         i_nal_type    = NAL_SLICE;
3057         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
3058         h->sh.i_type = SLICE_TYPE_P;
3059         x264_reference_hierarchy_reset( h );
3060         h->frames.i_poc_last_open_gop = -1;
3061     }
3062     else if( h->fenc->i_type == X264_TYPE_BREF )
3063     {
3064         i_nal_type    = NAL_SLICE;
3065         i_nal_ref_idc = h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT ? NAL_PRIORITY_LOW : NAL_PRIORITY_HIGH;
3066         h->sh.i_type = SLICE_TYPE_B;
3067         x264_reference_hierarchy_reset( h );
3068     }
3069     else    /* B frame */
3070     {
3071         i_nal_type    = NAL_SLICE;
3072         i_nal_ref_idc = NAL_PRIORITY_DISPOSABLE;
3073         h->sh.i_type = SLICE_TYPE_B;
3074     }
3075
3076     h->fdec->i_type = h->fenc->i_type;
3077     h->fdec->i_frame = h->fenc->i_frame;
3078     h->fenc->b_kept_as_ref =
3079     h->fdec->b_kept_as_ref = i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE && h->param.i_keyint_max > 1;
3080
3081     h->fdec->mb_info = h->fenc->mb_info;
3082     h->fdec->mb_info_free = h->fenc->mb_info_free;
3083     h->fenc->mb_info = NULL;
3084     h->fenc->mb_info_free = NULL;
3085
3086     h->fdec->i_pts = h->fenc->i_pts;
3087     if( h->frames.i_bframe_delay )
3088     {
3089         int64_t *prev_reordered_pts = thread_current->frames.i_prev_reordered_pts;
3090         h->fdec->i_dts = h->i_frame > h->frames.i_bframe_delay
3091                        ? prev_reordered_pts[ (h->i_frame - h->frames.i_bframe_delay) % h->frames.i_bframe_delay ]
3092                        : h->fenc->i_reordered_pts - h->frames.i_bframe_delay_time;
3093         prev_reordered_pts[ h->i_frame % h->frames.i_bframe_delay ] = h->fenc->i_reordered_pts;
3094     }
3095     else
3096         h->fdec->i_dts = h->fenc->i_reordered_pts;
3097     if( h->fenc->i_type == X264_TYPE_IDR )
3098         h->i_last_idr_pts = h->fdec->i_pts;
3099
3100     /* ------------------- Init                ----------------------------- */
3101     /* build ref list 0/1 */
3102     x264_reference_build_list( h, h->fdec->i_poc );
3103
3104     /* ---------------------- Write the bitstream -------------------------- */
3105     /* Init bitstream context */
3106     if( h->param.b_sliced_threads )
3107     {
3108         for( int i = 0; i < h->param.i_threads; i++ )
3109         {
3110             bs_init( &h->thread[i]->out.bs, h->thread[i]->out.p_bitstream, h->thread[i]->out.i_bitstream );
3111             h->thread[i]->out.i_nal = 0;
3112         }
3113     }
3114     else
3115     {
3116         bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
3117         h->out.i_nal = 0;
3118     }
3119
3120     if( h->param.b_aud )
3121     {
3122         int pic_type;
3123
3124         if( h->sh.i_type == SLICE_TYPE_I )
3125             pic_type = 0;
3126         else if( h->sh.i_type == SLICE_TYPE_P )
3127             pic_type = 1;
3128         else if( h->sh.i_type == SLICE_TYPE_B )
3129             pic_type = 2;
3130         else
3131             pic_type = 7;
3132
3133         x264_nal_start( h, NAL_AUD, NAL_PRIORITY_DISPOSABLE );
3134         bs_write( &h->out.bs, 3, pic_type );
3135         bs_rbsp_trailing( &h->out.bs );
3136         if( x264_nal_end( h ) )
3137             return -1;
3138         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
3139     }
3140
3141     h->i_nal_type = i_nal_type;
3142     h->i_nal_ref_idc = i_nal_ref_idc;
3143
3144     if( h->param.b_intra_refresh )
3145     {
3146         if( IS_X264_TYPE_I( h->fenc->i_type ) )
3147         {
3148             h->fdec->i_frames_since_pir = 0;
3149             h->b_queued_intra_refresh = 0;
3150             /* PIR is currently only supported with ref == 1, so any intra frame effectively refreshes
3151              * the whole frame and counts as an intra refresh. */
3152             h->fdec->f_pir_position = h->mb.i_mb_width;
3153         }
3154         else if( h->fenc->i_type == X264_TYPE_P )
3155         {
3156             int pocdiff = (h->fdec->i_poc - h->fref[0][0]->i_poc)/2;
3157             float increment = X264_MAX( ((float)h->mb.i_mb_width-1) / h->param.i_keyint_max, 1 );
3158             h->fdec->f_pir_position = h->fref[0][0]->f_pir_position;
3159             h->fdec->i_frames_since_pir = h->fref[0][0]->i_frames_since_pir + pocdiff;
3160             if( h->fdec->i_frames_since_pir >= h->param.i_keyint_max ||
3161                 (h->b_queued_intra_refresh && h->fdec->f_pir_position + 0.5 >= h->mb.i_mb_width) )
3162             {
3163                 h->fdec->f_pir_position = 0;
3164                 h->fdec->i_frames_since_pir = 0;
3165                 h->b_queued_intra_refresh = 0;
3166                 h->fenc->b_keyframe = 1;
3167             }
3168             h->fdec->i_pir_start_col = h->fdec->f_pir_position+0.5;
3169             h->fdec->f_pir_position += increment * pocdiff;
3170             h->fdec->i_pir_end_col = h->fdec->f_pir_position+0.5;
3171             /* If our intra refresh has reached the right side of the frame, we're done. */
3172             if( h->fdec->i_pir_end_col >= h->mb.i_mb_width - 1 )
3173             {
3174                 h->fdec->f_pir_position = h->mb.i_mb_width;
3175                 h->fdec->i_pir_end_col = h->mb.i_mb_width - 1;
3176             }
3177         }
3178     }
3179
3180     if( h->fenc->b_keyframe )
3181     {
3182         /* Write SPS and PPS */
3183         if( h->param.b_repeat_headers )
3184         {
3185             /* generate sequence parameters */
3186             x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
3187             x264_sps_write( &h->out.bs, h->sps );
3188             if( x264_nal_end( h ) )
3189                 return -1;
3190             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
3191
3192             /* generate picture parameters */
3193             x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
3194             x264_pps_write( &h->out.bs, h->sps, h->pps );
3195             if( x264_nal_end( h ) )
3196                 return -1;
3197             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
3198         }
3199
3200         /* when frame threading is used, buffering period sei is written in x264_encoder_frame_end */
3201         if( h->i_thread_frames == 1 && h->sps->vui.b_nal_hrd_parameters_present )
3202         {
3203             x264_hrd_fullness( h );
3204             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3205             x264_sei_buffering_period_write( h, &h->out.bs );
3206             if( x264_nal_end( h ) )
3207                return -1;
3208             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
3209         }
3210     }
3211
3212     /* write extra sei */
3213     for( int i = 0; i < h->fenc->extra_sei.num_payloads; i++ )
3214     {
3215         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3216         x264_sei_write( &h->out.bs, h->fenc->extra_sei.payloads[i].payload, h->fenc->extra_sei.payloads[i].payload_size,
3217                         h->fenc->extra_sei.payloads[i].payload_type );
3218         if( x264_nal_end( h ) )
3219             return -1;
3220         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
3221         if( h->fenc->extra_sei.sei_free )
3222         {
3223             h->fenc->extra_sei.sei_free( h->fenc->extra_sei.payloads[i].payload );
3224             h->fenc->extra_sei.payloads[i].payload = NULL;
3225         }
3226     }
3227
3228     if( h->fenc->extra_sei.sei_free )
3229     {
3230         h->fenc->extra_sei.sei_free( h->fenc->extra_sei.payloads );
3231         h->fenc->extra_sei.payloads = NULL;
3232         h->fenc->extra_sei.sei_free = NULL;
3233     }
3234
3235     if( h->fenc->b_keyframe )
3236     {
3237         if( h->param.b_repeat_headers && h->fenc->i_frame == 0 )
3238         {
3239             /* identify ourself */
3240             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3241             if( x264_sei_version_write( h, &h->out.bs ) )
3242                 return -1;
3243             if( x264_nal_end( h ) )
3244                 return -1;
3245             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
3246         }
3247
3248         if( h->fenc->i_type != X264_TYPE_IDR )
3249         {
3250             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;
3251             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3252             x264_sei_recovery_point_write( h, &h->out.bs, time_to_recovery );
3253             if( x264_nal_end( h ) )
3254                 return -1;
3255             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
3256         }
3257
3258         if ( h->param.i_frame_packing >= 0 )
3259         {
3260             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3261             x264_sei_frame_packing_write( h, &h->out.bs );
3262             if( x264_nal_end( h ) )
3263                 return -1;
3264             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
3265         }
3266     }
3267
3268     /* generate sei pic timing */
3269     if( h->sps->vui.b_pic_struct_present || h->sps->vui.b_nal_hrd_parameters_present )
3270     {
3271         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3272         x264_sei_pic_timing_write( h, &h->out.bs );
3273         if( x264_nal_end( h ) )
3274             return -1;
3275         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
3276     }
3277
3278     /* As required by Blu-ray. */
3279     if( !IS_X264_TYPE_B( h->fenc->i_type ) && h->b_sh_backup )
3280     {
3281         h->b_sh_backup = 0;
3282         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3283         x264_sei_dec_ref_pic_marking_write( h, &h->out.bs );
3284         if( x264_nal_end( h ) )
3285             return -1;
3286         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
3287     }
3288
3289     if( h->fenc->b_keyframe && h->param.b_intra_refresh )
3290         h->i_cpb_delay_pir_offset_next = h->fenc->i_cpb_delay;
3291
3292     /* Init the rate control */
3293     /* FIXME: Include slice header bit cost. */
3294     x264_ratecontrol_start( h, h->fenc->i_qpplus1, overhead*8 );
3295     i_global_qp = x264_ratecontrol_qp( h );
3296
3297     pic_out->i_qpplus1 =
3298     h->fdec->i_qpplus1 = i_global_qp + 1;
3299
3300     if( h->param.rc.b_stat_read && h->sh.i_type != SLICE_TYPE_I )
3301     {
3302         x264_reference_build_list_optimal( h );
3303         x264_reference_check_reorder( h );
3304     }
3305
3306     if( h->i_ref[0] )
3307         h->fdec->i_poc_l0ref0 = h->fref[0][0]->i_poc;
3308
3309     /* ------------------------ Create slice header  ----------------------- */
3310     x264_slice_init( h, i_nal_type, i_global_qp );
3311
3312     /*------------------------- Weights -------------------------------------*/
3313     if( h->sh.i_type == SLICE_TYPE_B )
3314         x264_macroblock_bipred_init( h );
3315
3316     x264_weighted_pred_init( h );
3317
3318     if( i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE )
3319         h->i_frame_num++;
3320
3321     /* Write frame */
3322     h->i_threadslice_start = 0;
3323     h->i_threadslice_end = h->mb.i_mb_height;
3324     if( h->i_thread_frames > 1 )
3325     {
3326         x264_threadpool_run( h->threadpool, (void*)x264_slices_write, h );
3327         h->b_thread_active = 1;
3328     }
3329     else if( h->param.b_sliced_threads )
3330     {
3331         if( x264_threaded_slices_write( h ) )
3332             return -1;
3333     }
3334     else
3335         if( (intptr_t)x264_slices_write( h ) )
3336             return -1;
3337
3338     return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
3339 }
3340
3341 static int x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
3342                                    x264_nal_t **pp_nal, int *pi_nal,
3343                                    x264_picture_t *pic_out )
3344 {
3345     char psz_message[80];
3346
3347     if( !h->param.b_sliced_threads && h->b_thread_active )
3348     {
3349         h->b_thread_active = 0;
3350         if( (intptr_t)x264_threadpool_wait( h->threadpool, h ) )
3351             return -1;
3352     }
3353     if( !h->out.i_nal )
3354     {
3355         pic_out->i_type = X264_TYPE_AUTO;
3356         return 0;
3357     }
3358
3359     x264_emms();
3360
3361     /* generate buffering period sei and insert it into place */
3362     if( h->i_thread_frames > 1 && h->fenc->b_keyframe && h->sps->vui.b_nal_hrd_parameters_present )
3363     {
3364         x264_hrd_fullness( h );
3365         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3366         x264_sei_buffering_period_write( h, &h->out.bs );
3367         if( x264_nal_end( h ) )
3368            return -1;
3369         /* buffering period sei must follow AUD, SPS and PPS and precede all other SEIs */
3370         int idx = 0;
3371         while( h->out.nal[idx].i_type == NAL_AUD ||
3372                h->out.nal[idx].i_type == NAL_SPS ||
3373                h->out.nal[idx].i_type == NAL_PPS )
3374             idx++;
3375         x264_nal_t nal_tmp = h->out.nal[h->out.i_nal-1];
3376         memmove( &h->out.nal[idx+1], &h->out.nal[idx], (h->out.i_nal-idx-1)*sizeof(x264_nal_t) );
3377         h->out.nal[idx] = nal_tmp;
3378     }
3379
3380     int frame_size = x264_encoder_encapsulate_nals( h, 0 );
3381     if( frame_size < 0 )
3382         return -1;
3383
3384     /* Set output picture properties */
3385     pic_out->i_type = h->fenc->i_type;
3386
3387     pic_out->b_keyframe = h->fenc->b_keyframe;
3388     pic_out->i_pic_struct = h->fenc->i_pic_struct;
3389
3390     pic_out->i_pts = h->fdec->i_pts;
3391     pic_out->i_dts = h->fdec->i_dts;
3392
3393     if( pic_out->i_pts < pic_out->i_dts )
3394         x264_log( h, X264_LOG_WARNING, "invalid DTS: PTS is less than DTS\n" );
3395
3396     pic_out->opaque = h->fenc->opaque;
3397
3398     pic_out->img.i_csp = h->fdec->i_csp;
3399 #if HIGH_BIT_DEPTH
3400     pic_out->img.i_csp |= X264_CSP_HIGH_DEPTH;
3401 #endif
3402     pic_out->img.i_plane = h->fdec->i_plane;
3403     for( int i = 0; i < pic_out->img.i_plane; i++ )
3404     {
3405         pic_out->img.i_stride[i] = h->fdec->i_stride[i] * sizeof(pixel);
3406         pic_out->img.plane[i] = (uint8_t*)h->fdec->plane[i];
3407     }
3408
3409     x264_frame_push_unused( thread_current, h->fenc );
3410
3411     /* ---------------------- Update encoder state ------------------------- */
3412
3413     /* update rc */
3414     int filler = 0;
3415     if( x264_ratecontrol_end( h, frame_size * 8, &filler ) < 0 )
3416         return -1;
3417
3418     pic_out->hrd_timing = h->fenc->hrd_timing;
3419     pic_out->prop.f_crf_avg = h->fdec->f_crf_avg;
3420
3421     while( filler > 0 )
3422     {
3423         int f, overhead;
3424         overhead = (FILLER_OVERHEAD - h->param.b_annexb);
3425         if( h->param.i_slice_max_size && filler > h->param.i_slice_max_size )
3426         {
3427             int next_size = filler - h->param.i_slice_max_size;
3428             int overflow = X264_MAX( overhead - next_size, 0 );
3429             f = h->param.i_slice_max_size - overhead - overflow;
3430         }
3431         else
3432             f = X264_MAX( 0, filler - overhead );
3433
3434         x264_nal_start( h, NAL_FILLER, NAL_PRIORITY_DISPOSABLE );
3435         x264_filler_write( h, &h->out.bs, f );
3436         if( x264_nal_end( h ) )
3437             return -1;
3438         int total_size = x264_encoder_encapsulate_nals( h, h->out.i_nal-1 );
3439         if( total_size < 0 )
3440             return -1;
3441         frame_size += total_size;
3442         filler -= total_size;
3443     }
3444
3445     /* End bitstream, set output  */
3446     *pi_nal = h->out.i_nal;
3447     *pp_nal = h->out.nal;
3448
3449     h->out.i_nal = 0;
3450
3451     x264_noise_reduction_update( h );
3452
3453     /* ---------------------- Compute/Print statistics --------------------- */
3454     x264_thread_sync_stat( h, h->thread[0] );
3455
3456     /* Slice stat */
3457     h->stat.i_frame_count[h->sh.i_type]++;
3458     h->stat.i_frame_size[h->sh.i_type] += frame_size;
3459     h->stat.f_frame_qp[h->sh.i_type] += h->fdec->f_qp_avg_aq;
3460
3461     for( int i = 0; i < X264_MBTYPE_MAX; i++ )
3462         h->stat.i_mb_count[h->sh.i_type][i] += h->stat.frame.i_mb_count[i];
3463     for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
3464         h->stat.i_mb_partition[h->sh.i_type][i] += h->stat.frame.i_mb_partition[i];
3465     for( int i = 0; i < 2; i++ )
3466         h->stat.i_mb_count_8x8dct[i] += h->stat.frame.i_mb_count_8x8dct[i];
3467     for( int i = 0; i < 6; i++ )
3468         h->stat.i_mb_cbp[i] += h->stat.frame.i_mb_cbp[i];
3469     for( int i = 0; i < 4; i++ )
3470         for( int j = 0; j < 13; j++ )
3471             h->stat.i_mb_pred_mode[i][j] += h->stat.frame.i_mb_pred_mode[i][j];
3472     if( h->sh.i_type != SLICE_TYPE_I )
3473         for( int i_list = 0; i_list < 2; i_list++ )
3474             for( int i = 0; i < X264_REF_MAX*2; i++ )
3475                 h->stat.i_mb_count_ref[h->sh.i_type][i_list][i] += h->stat.frame.i_mb_count_ref[i_list][i];
3476     for( int i = 0; i < 3; i++ )
3477         h->stat.i_mb_field[i] += h->stat.frame.i_mb_field[i];
3478     if( h->sh.i_type == SLICE_TYPE_P && h->param.analyse.i_weighted_pred >= X264_WEIGHTP_SIMPLE )
3479     {
3480         h->stat.i_wpred[0] += !!h->sh.weight[0][0].weightfn;
3481         h->stat.i_wpred[1] += !!h->sh.weight[0][1].weightfn || !!h->sh.weight[0][2].weightfn;
3482     }
3483     if( h->sh.i_type == SLICE_TYPE_B )
3484     {
3485         h->stat.i_direct_frames[ h->sh.b_direct_spatial_mv_pred ] ++;
3486         if( h->mb.b_direct_auto_write )
3487         {
3488             //FIXME somewhat arbitrary time constants
3489             if( h->stat.i_direct_score[0] + h->stat.i_direct_score[1] > h->mb.i_mb_count )
3490                 for( int i = 0; i < 2; i++ )
3491                     h->stat.i_direct_score[i] = h->stat.i_direct_score[i] * 9/10;
3492             for( int i = 0; i < 2; i++ )
3493                 h->stat.i_direct_score[i] += h->stat.frame.i_direct_score[i];
3494         }
3495     }
3496     else
3497         h->stat.i_consecutive_bframes[h->fenc->i_bframes]++;
3498
3499     psz_message[0] = '\0';
3500     double dur = h->fenc->f_duration;
3501     h->stat.f_frame_duration[h->sh.i_type] += dur;
3502     if( h->param.analyse.b_psnr )
3503     {
3504         int64_t ssd[3] =
3505         {
3506             h->stat.frame.i_ssd[0],
3507             h->stat.frame.i_ssd[1],
3508             h->stat.frame.i_ssd[2],
3509         };
3510         int luma_size = h->param.i_width * h->param.i_height;
3511         int chroma_size = CHROMA_SIZE( luma_size );
3512         pic_out->prop.f_psnr[0] = x264_psnr( ssd[0], luma_size );
3513         pic_out->prop.f_psnr[1] = x264_psnr( ssd[1], chroma_size );
3514         pic_out->prop.f_psnr[2] = x264_psnr( ssd[2], chroma_size );
3515         pic_out->prop.f_psnr_avg = x264_psnr( ssd[0] + ssd[1] + ssd[2], luma_size + chroma_size*2 );
3516
3517         h->stat.f_ssd_global[h->sh.i_type]   += dur * (ssd[0] + ssd[1] + ssd[2]);
3518         h->stat.f_psnr_average[h->sh.i_type] += dur * pic_out->prop.f_psnr_avg;
3519         h->stat.f_psnr_mean_y[h->sh.i_type]  += dur * pic_out->prop.f_psnr[0];
3520         h->stat.f_psnr_mean_u[h->sh.i_type]  += dur * pic_out->prop.f_psnr[1];
3521         h->stat.f_psnr_mean_v[h->sh.i_type]  += dur * pic_out->prop.f_psnr[2];
3522
3523         snprintf( psz_message, 80, " PSNR Y:%5.2f U:%5.2f V:%5.2f", pic_out->prop.f_psnr[0],
3524                                                                     pic_out->prop.f_psnr[1],
3525                                                                     pic_out->prop.f_psnr[2] );
3526     }
3527
3528     if( h->param.analyse.b_ssim )
3529     {
3530         pic_out->prop.f_ssim = h->stat.frame.f_ssim / h->stat.frame.i_ssim_cnt;
3531         h->stat.f_ssim_mean_y[h->sh.i_type] += pic_out->prop.f_ssim * dur;
3532         snprintf( psz_message + strlen(psz_message), 80 - strlen(psz_message),
3533                   " SSIM Y:%.5f", pic_out->prop.f_ssim );
3534     }
3535     psz_message[79] = '\0';
3536
3537     x264_log( h, X264_LOG_DEBUG,
3538                   "frame=%4d QP=%.2f NAL=%d Slice:%c Poc:%-3d I:%-4d P:%-4d SKIP:%-4d size=%d bytes%s\n",
3539               h->i_frame,
3540               h->fdec->f_qp_avg_aq,
3541               h->i_nal_ref_idc,
3542               h->sh.i_type == SLICE_TYPE_I ? 'I' : (h->sh.i_type == SLICE_TYPE_P ? 'P' : 'B' ),
3543               h->fdec->i_poc,
3544               h->stat.frame.i_mb_count_i,
3545               h->stat.frame.i_mb_count_p,
3546               h->stat.frame.i_mb_count_skip,
3547               frame_size,
3548               psz_message );
3549
3550     // keep stats all in one place
3551     x264_thread_sync_stat( h->thread[0], h );
3552     // for the use of the next frame
3553     x264_thread_sync_stat( thread_current, h );
3554
3555 #ifdef DEBUG_MB_TYPE
3556 {
3557     static const char mb_chars[] = { 'i', 'i', 'I', 'C', 'P', '8', 'S',
3558         'D', '<', 'X', 'B', 'X', '>', 'B', 'B', 'B', 'B', '8', 'S' };
3559     for( int mb_xy = 0; mb_xy < h->mb.i_mb_width * h->mb.i_mb_height; mb_xy++ )
3560     {
3561         if( h->mb.type[mb_xy] < X264_MBTYPE_MAX && h->mb.type[mb_xy] >= 0 )
3562             fprintf( stderr, "%c ", mb_chars[ h->mb.type[mb_xy] ] );
3563         else
3564             fprintf( stderr, "? " );
3565
3566         if( (mb_xy+1) % h->mb.i_mb_width == 0 )
3567             fprintf( stderr, "\n" );
3568     }
3569 }
3570 #endif
3571
3572     /* Remove duplicates, must be done near the end as breaks h->fref0 array
3573      * by freeing some of its pointers. */
3574     for( int i = 0; i < h->i_ref[0]; i++ )
3575         if( h->fref[0][i] && h->fref[0][i]->b_duplicate )
3576         {
3577             x264_frame_push_blank_unused( h, h->fref[0][i] );
3578             h->fref[0][i] = 0;
3579         }
3580
3581     if( h->param.psz_dump_yuv )
3582         x264_frame_dump( h );
3583     x264_emms();
3584
3585     return frame_size;
3586 }
3587
3588 static void x264_print_intra( int64_t *i_mb_count, double i_count, int b_print_pcm, char *intra )
3589 {
3590     intra += sprintf( intra, "I16..4%s: %4.1f%% %4.1f%% %4.1f%%",
3591         b_print_pcm ? "..PCM" : "",
3592         i_mb_count[I_16x16]/ i_count,
3593         i_mb_count[I_8x8]  / i_count,
3594         i_mb_count[I_4x4]  / i_count );
3595     if( b_print_pcm )
3596         sprintf( intra, " %4.1f%%", i_mb_count[I_PCM]  / i_count );
3597 }
3598
3599 /****************************************************************************
3600  * x264_encoder_close:
3601  ****************************************************************************/
3602 void    x264_encoder_close  ( x264_t *h )
3603 {
3604     int64_t i_yuv_size = FRAME_SIZE( h->param.i_width * h->param.i_height );
3605     int64_t i_mb_count_size[2][7] = {{0}};
3606     char buf[200];
3607     int b_print_pcm = h->stat.i_mb_count[SLICE_TYPE_I][I_PCM]
3608                    || h->stat.i_mb_count[SLICE_TYPE_P][I_PCM]
3609                    || h->stat.i_mb_count[SLICE_TYPE_B][I_PCM];
3610
3611     x264_lookahead_delete( h );
3612
3613     if( h->param.b_sliced_threads )
3614         x264_threadpool_wait_all( h );
3615     if( h->param.i_threads > 1 )
3616         x264_threadpool_delete( h->threadpool );
3617     if( h->param.i_lookahead_threads > 1 )
3618         x264_threadpool_delete( h->lookaheadpool );
3619     if( h->i_thread_frames > 1 )
3620     {
3621         for( int i = 0; i < h->i_thread_frames; i++ )
3622             if( h->thread[i]->b_thread_active )
3623             {
3624                 assert( h->thread[i]->fenc->i_reference_count == 1 );
3625                 x264_frame_delete( h->thread[i]->fenc );
3626             }
3627
3628         x264_t *thread_prev = h->thread[h->i_thread_phase];
3629         x264_thread_sync_ratecontrol( h, thread_prev, h );
3630         x264_thread_sync_ratecontrol( thread_prev, thread_prev, h );
3631         h->i_frame = thread_prev->i_frame + 1 - h->i_thread_frames;
3632     }
3633     h->i_frame++;
3634
3635     /* Slices used and PSNR */
3636     for( int i = 0; i < 3; i++ )
3637     {
3638         static const uint8_t slice_order[] = { SLICE_TYPE_I, SLICE_TYPE_P, SLICE_TYPE_B };
3639         int i_slice = slice_order[i];
3640
3641         if( h->stat.i_frame_count[i_slice] > 0 )
3642         {
3643             int i_count = h->stat.i_frame_count[i_slice];
3644             double dur =  h->stat.f_frame_duration[i_slice];
3645             if( h->param.analyse.b_psnr )
3646             {
3647                 x264_log( h, X264_LOG_INFO,
3648                           "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",
3649                           slice_type_to_char[i_slice],
3650                           i_count,
3651                           h->stat.f_frame_qp[i_slice] / i_count,
3652                           (double)h->stat.i_frame_size[i_slice] / i_count,
3653                           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,
3654                           h->stat.f_psnr_average[i_slice] / dur,
3655                           x264_psnr( h->stat.f_ssd_global[i_slice], dur * i_yuv_size ) );
3656             }
3657             else
3658             {
3659                 x264_log( h, X264_LOG_INFO,
3660                           "frame %c:%-5d Avg QP:%5.2f  size:%6.0f\n",
3661                           slice_type_to_char[i_slice],
3662                           i_count,
3663                           h->stat.f_frame_qp[i_slice] / i_count,
3664                           (double)h->stat.i_frame_size[i_slice] / i_count );
3665             }
3666         }
3667     }
3668     if( h->param.i_bframe && h->stat.i_frame_count[SLICE_TYPE_B] )
3669     {
3670         char *p = buf;
3671         int den = 0;
3672         // weight by number of frames (including the I/P-frames) that are in a sequence of N B-frames
3673         for( int i = 0; i <= h->param.i_bframe; i++ )
3674             den += (i+1) * h->stat.i_consecutive_bframes[i];
3675         for( int i = 0; i <= h->param.i_bframe; i++ )
3676             p += sprintf( p, " %4.1f%%", 100. * (i+1) * h->stat.i_consecutive_bframes[i] / den );
3677         x264_log( h, X264_LOG_INFO, "consecutive B-frames:%s\n", buf );
3678     }
3679
3680     for( int i_type = 0; i_type < 2; i_type++ )
3681         for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
3682         {
3683             if( i == D_DIRECT_8x8 ) continue; /* direct is counted as its own type */
3684             i_mb_count_size[i_type][x264_mb_partition_pixel_table[i]] += h->stat.i_mb_partition[i_type][i];
3685         }
3686
3687     /* MB types used */
3688     if( h->stat.i_frame_count[SLICE_TYPE_I] > 0 )
3689     {
3690         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_I];
3691         double i_count = h->stat.i_frame_count[SLICE_TYPE_I] * h->mb.i_mb_count / 100.0;
3692         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
3693         x264_log( h, X264_LOG_INFO, "mb I  %s\n", buf );
3694     }
3695     if( h->stat.i_frame_count[SLICE_TYPE_P] > 0 )
3696     {
3697         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_P];
3698         double i_count = h->stat.i_frame_count[SLICE_TYPE_P] * h->mb.i_mb_count / 100.0;
3699         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_P];
3700         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
3701         x264_log( h, X264_LOG_INFO,
3702                   "mb P  %s  P16..4: %4.1f%% %4.1f%% %4.1f%% %4.1f%% %4.1f%%    skip:%4.1f%%\n",
3703                   buf,
3704                   i_mb_size[PIXEL_16x16] / (i_count*4),
3705                   (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
3706                   i_mb_size[PIXEL_8x8] / (i_count*4),
3707                   (i_mb_size[PIXEL_8x4] + i_mb_size[PIXEL_4x8]) / (i_count*4),
3708                   i_mb_size[PIXEL_4x4] / (i_count*4),
3709                   i_mb_count[P_SKIP] / i_count );
3710     }
3711     if( h->stat.i_frame_count[SLICE_TYPE_B] > 0 )
3712     {
3713         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_B];
3714         double i_count = h->stat.i_frame_count[SLICE_TYPE_B] * h->mb.i_mb_count / 100.0;
3715         double i_mb_list_count;
3716         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_B];
3717         int64_t list_count[3] = {0}; /* 0 == L0, 1 == L1, 2 == BI */
3718         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
3719         for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
3720             for( int j = 0; j < 2; j++ )
3721             {
3722                 int l0 = x264_mb_type_list_table[i][0][j];
3723                 int l1 = x264_mb_type_list_table[i][1][j];
3724                 if( l0 || l1 )
3725                     list_count[l1+l0*l1] += h->stat.i_mb_count[SLICE_TYPE_B][i] * 2;
3726             }
3727         list_count[0] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L0_8x8];
3728         list_count[1] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L1_8x8];
3729         list_count[2] += h->stat.i_mb_partition[SLICE_TYPE_B][D_BI_8x8];
3730         i_mb_count[B_DIRECT] += (h->stat.i_mb_partition[SLICE_TYPE_B][D_DIRECT_8x8]+2)/4;
3731         i_mb_list_count = (list_count[0] + list_count[1] + list_count[2]) / 100.0;
3732         sprintf( buf + strlen(buf), "  B16..8: %4.1f%% %4.1f%% %4.1f%%  direct:%4.1f%%  skip:%4.1f%%",
3733                  i_mb_size[PIXEL_16x16] / (i_count*4),
3734                  (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
3735                  i_mb_size[PIXEL_8x8] / (i_count*4),
3736                  i_mb_count[B_DIRECT] / i_count,
3737                  i_mb_count[B_SKIP]   / i_count );
3738         if( i_mb_list_count != 0 )
3739             sprintf( buf + strlen(buf), "  L0:%4.1f%% L1:%4.1f%% BI:%4.1f%%",
3740                      list_count[0] / i_mb_list_count,
3741                      list_count[1] / i_mb_list_count,
3742                      list_count[2] / i_mb_list_count );
3743         x264_log( h, X264_LOG_INFO, "mb B  %s\n", buf );
3744     }
3745
3746     x264_ratecontrol_summary( h );
3747
3748     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 )
3749     {
3750 #define SUM3(p) (p[SLICE_TYPE_I] + p[SLICE_TYPE_P] + p[SLICE_TYPE_B])
3751 #define SUM3b(p,o) (p[SLICE_TYPE_I][o] + p[SLICE_TYPE_P][o] + p[SLICE_TYPE_B][o])
3752         int64_t i_i8x8 = SUM3b( h->stat.i_mb_count, I_8x8 );
3753         int64_t i_intra = i_i8x8 + SUM3b( h->stat.i_mb_count, I_4x4 )
3754                                  + SUM3b( h->stat.i_mb_count, I_16x16 );
3755         int64_t i_all_intra = i_intra + SUM3b( h->stat.i_mb_count, I_PCM);
3756         int64_t i_skip = SUM3b( h->stat.i_mb_count, P_SKIP )
3757                        + SUM3b( h->stat.i_mb_count, B_SKIP );
3758         const int i_count = h->stat.i_frame_count[SLICE_TYPE_I] +
3759                             h->stat.i_frame_count[SLICE_TYPE_P] +
3760                             h->stat.i_frame_count[SLICE_TYPE_B];
3761         int64_t i_mb_count = (int64_t)i_count * h->mb.i_mb_count;
3762         int64_t i_inter = i_mb_count - i_skip - i_intra;
3763         const double duration = h->stat.f_frame_duration[SLICE_TYPE_I] +
3764                                 h->stat.f_frame_duration[SLICE_TYPE_P] +
3765                                 h->stat.f_frame_duration[SLICE_TYPE_B];
3766         float f_bitrate = SUM3(h->stat.i_frame_size) / duration / 125;
3767
3768         if( PARAM_INTERLACED )
3769         {
3770             char *fieldstats = buf;
3771             fieldstats[0] = 0;
3772             if( i_inter )
3773                 fieldstats += sprintf( fieldstats, " inter:%.1f%%", h->stat.i_mb_field[1] * 100.0 / i_inter );
3774             if( i_skip )
3775                 fieldstats += sprintf( fieldstats, " skip:%.1f%%", h->stat.i_mb_field[2] * 100.0 / i_skip );
3776             x264_log( h, X264_LOG_INFO, "field mbs: intra: %.1f%%%s\n",
3777                       h->stat.i_mb_field[0] * 100.0 / i_intra, buf );
3778         }
3779
3780         if( h->pps->b_transform_8x8_mode )
3781         {
3782             buf[0] = 0;
3783             if( h->stat.i_mb_count_8x8dct[0] )
3784                 sprintf( buf, " inter:%.1f%%", 100. * h->stat.i_mb_count_8x8dct[1] / h->stat.i_mb_count_8x8dct[0] );
3785             x264_log( h, X264_LOG_INFO, "8x8 transform intra:%.1f%%%s\n", 100. * i_i8x8 / i_intra, buf );
3786         }
3787
3788         if( (h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO ||
3789             (h->stat.i_direct_frames[0] && h->stat.i_direct_frames[1]))
3790             && h->stat.i_frame_count[SLICE_TYPE_B] )
3791         {
3792             x264_log( h, X264_LOG_INFO, "direct mvs  spatial:%.1f%% temporal:%.1f%%\n",
3793                       h->stat.i_direct_frames[1] * 100. / h->stat.i_frame_count[SLICE_TYPE_B],
3794                       h->stat.i_direct_frames[0] * 100. / h->stat.i_frame_count[SLICE_TYPE_B] );
3795         }
3796
3797         buf[0] = 0;
3798         int csize = CHROMA444 ? 4 : 1;
3799         if( i_mb_count != i_all_intra )
3800             sprintf( buf, " inter: %.1f%% %.1f%% %.1f%%",
3801                      h->stat.i_mb_cbp[1] * 100.0 / ((i_mb_count - i_all_intra)*4),
3802                      h->stat.i_mb_cbp[3] * 100.0 / ((i_mb_count - i_all_intra)*csize),
3803                      h->stat.i_mb_cbp[5] * 100.0 / ((i_mb_count - i_all_intra)*csize) );
3804         x264_log( h, X264_LOG_INFO, "coded y,%s,%s intra: %.1f%% %.1f%% %.1f%%%s\n",
3805                   CHROMA444?"u":"uvDC", CHROMA444?"v":"uvAC",
3806                   h->stat.i_mb_cbp[0] * 100.0 / (i_all_intra*4),
3807                   h->stat.i_mb_cbp[2] * 100.0 / (i_all_intra*csize),
3808                   h->stat.i_mb_cbp[4] * 100.0 / (i_all_intra*csize), buf );
3809
3810         int64_t fixed_pred_modes[4][9] = {{0}};
3811         int64_t sum_pred_modes[4] = {0};
3812         for( int i = 0; i <= I_PRED_16x16_DC_128; i++ )
3813         {
3814             fixed_pred_modes[0][x264_mb_pred_mode16x16_fix[i]] += h->stat.i_mb_pred_mode[0][i];
3815             sum_pred_modes[0] += h->stat.i_mb_pred_mode[0][i];
3816         }
3817         if( sum_pred_modes[0] )
3818             x264_log( h, X264_LOG_INFO, "i16 v,h,dc,p: %2.0f%% %2.0f%% %2.0f%% %2.0f%%\n",
3819                       fixed_pred_modes[0][0] * 100.0 / sum_pred_modes[0],
3820                       fixed_pred_modes[0][1] * 100.0 / sum_pred_modes[0],
3821                       fixed_pred_modes[0][2] * 100.0 / sum_pred_modes[0],
3822                       fixed_pred_modes[0][3] * 100.0 / sum_pred_modes[0] );
3823         for( int i = 1; i <= 2; i++ )
3824         {
3825             for( int j = 0; j <= I_PRED_8x8_DC_128; j++ )
3826             {
3827                 fixed_pred_modes[i][x264_mb_pred_mode4x4_fix(j)] += h->stat.i_mb_pred_mode[i][j];
3828                 sum_pred_modes[i] += h->stat.i_mb_pred_mode[i][j];
3829             }
3830             if( sum_pred_modes[i] )
3831                 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,
3832                           fixed_pred_modes[i][0] * 100.0 / sum_pred_modes[i],
3833                           fixed_pred_modes[i][1] * 100.0 / sum_pred_modes[i],
3834                           fixed_pred_modes[i][2] * 100.0 / sum_pred_modes[i],
3835                           fixed_pred_modes[i][3] * 100.0 / sum_pred_modes[i],
3836                           fixed_pred_modes[i][4] * 100.0 / sum_pred_modes[i],
3837                           fixed_pred_modes[i][5] * 100.0 / sum_pred_modes[i],
3838                           fixed_pred_modes[i][6] * 100.0 / sum_pred_modes[i],
3839                           fixed_pred_modes[i][7] * 100.0 / sum_pred_modes[i],
3840                           fixed_pred_modes[i][8] * 100.0 / sum_pred_modes[i] );
3841         }
3842         for( int i = 0; i <= I_PRED_CHROMA_DC_128; i++ )
3843         {
3844             fixed_pred_modes[3][x264_mb_chroma_pred_mode_fix[i]] += h->stat.i_mb_pred_mode[3][i];
3845             sum_pred_modes[3] += h->stat.i_mb_pred_mode[3][i];
3846         }
3847         if( sum_pred_modes[3] && !CHROMA444 )
3848             x264_log( h, X264_LOG_INFO, "i8c dc,h,v,p: %2.0f%% %2.0f%% %2.0f%% %2.0f%%\n",
3849                       fixed_pred_modes[3][0] * 100.0 / sum_pred_modes[3],
3850                       fixed_pred_modes[3][1] * 100.0 / sum_pred_modes[3],
3851                       fixed_pred_modes[3][2] * 100.0 / sum_pred_modes[3],
3852                       fixed_pred_modes[3][3] * 100.0 / sum_pred_modes[3] );
3853
3854         if( h->param.analyse.i_weighted_pred >= X264_WEIGHTP_SIMPLE && h->stat.i_frame_count[SLICE_TYPE_P] > 0 )
3855             x264_log( h, X264_LOG_INFO, "Weighted P-Frames: Y:%.1f%% UV:%.1f%%\n",
3856                       h->stat.i_wpred[0] * 100.0 / h->stat.i_frame_count[SLICE_TYPE_P],
3857                       h->stat.i_wpred[1] * 100.0 / h->stat.i_frame_count[SLICE_TYPE_P] );
3858
3859         for( int i_list = 0; i_list < 2; i_list++ )
3860             for( int i_slice = 0; i_slice < 2; i_slice++ )
3861             {
3862                 char *p = buf;
3863                 int64_t i_den = 0;
3864                 int i_max = 0;
3865                 for( int i = 0; i < X264_REF_MAX*2; i++ )
3866                     if( h->stat.i_mb_count_ref[i_slice][i_list][i] )
3867                     {
3868                         i_den += h->stat.i_mb_count_ref[i_slice][i_list][i];
3869                         i_max = i;
3870                     }
3871                 if( i_max == 0 )
3872                     continue;
3873                 for( int i = 0; i <= i_max; i++ )
3874                     p += sprintf( p, " %4.1f%%", 100. * h->stat.i_mb_count_ref[i_slice][i_list][i] / i_den );
3875                 x264_log( h, X264_LOG_INFO, "ref %c L%d:%s\n", "PB"[i_slice], i_list, buf );
3876             }
3877
3878         if( h->param.analyse.b_ssim )
3879         {
3880             float ssim = SUM3( h->stat.f_ssim_mean_y ) / duration;
3881             x264_log( h, X264_LOG_INFO, "SSIM Mean Y:%.7f (%6.3fdb)\n", ssim, x264_ssim( ssim ) );
3882         }
3883         if( h->param.analyse.b_psnr )
3884         {
3885             x264_log( h, X264_LOG_INFO,
3886                       "PSNR Mean Y:%6.3f U:%6.3f V:%6.3f Avg:%6.3f Global:%6.3f kb/s:%.2f\n",
3887                       SUM3( h->stat.f_psnr_mean_y ) / duration,
3888                       SUM3( h->stat.f_psnr_mean_u ) / duration,
3889                       SUM3( h->stat.f_psnr_mean_v ) / duration,
3890                       SUM3( h->stat.f_psnr_average ) / duration,
3891                       x264_psnr( SUM3( h->stat.f_ssd_global ), duration * i_yuv_size ),
3892                       f_bitrate );
3893         }
3894         else
3895             x264_log( h, X264_LOG_INFO, "kb/s:%.2f\n", f_bitrate );
3896     }
3897
3898     /* rc */
3899     x264_ratecontrol_delete( h );
3900
3901     /* param */
3902     if( h->param.rc.psz_stat_out )
3903         free( h->param.rc.psz_stat_out );
3904     if( h->param.rc.psz_stat_in )
3905         free( h->param.rc.psz_stat_in );
3906
3907     x264_cqm_delete( h );
3908     x264_free( h->nal_buffer );
3909     x264_analyse_free_costs( h );
3910
3911     if( h->i_thread_frames > 1 )
3912         h = h->thread[h->i_thread_phase];
3913
3914     /* frames */
3915     x264_frame_delete_list( h->frames.unused[0] );
3916     x264_frame_delete_list( h->frames.unused[1] );
3917     x264_frame_delete_list( h->frames.current );
3918     x264_frame_delete_list( h->frames.blank_unused );
3919
3920     h = h->thread[0];
3921
3922     for( int i = 0; i < h->i_thread_frames; i++ )
3923         if( h->thread[i]->b_thread_active )
3924             for( int j = 0; j < h->thread[i]->i_ref[0]; j++ )
3925                 if( h->thread[i]->fref[0][j] && h->thread[i]->fref[0][j]->b_duplicate )
3926                     x264_frame_delete( h->thread[i]->fref[0][j] );
3927
3928     if( h->param.i_lookahead_threads > 1 )
3929         for( int i = 0; i < h->param.i_lookahead_threads; i++ )
3930             x264_free( h->lookahead_thread[i] );
3931
3932     for( int i = h->param.i_threads - 1; i >= 0; i-- )
3933     {
3934         x264_frame_t **frame;
3935
3936         if( !h->param.b_sliced_threads || i == 0 )
3937         {
3938             for( frame = h->thread[i]->frames.reference; *frame; frame++ )
3939             {
3940                 assert( (*frame)->i_reference_count > 0 );
3941                 (*frame)->i_reference_count--;
3942                 if( (*frame)->i_reference_count == 0 )
3943                     x264_frame_delete( *frame );
3944             }
3945             frame = &h->thread[i]->fdec;
3946             if( *frame )
3947             {
3948                 assert( (*frame)->i_reference_count > 0 );
3949                 (*frame)->i_reference_count--;
3950                 if( (*frame)->i_reference_count == 0 )
3951                     x264_frame_delete( *frame );
3952             }
3953             x264_macroblock_cache_free( h->thread[i] );
3954         }
3955         x264_macroblock_thread_free( h->thread[i], 0 );
3956         x264_free( h->thread[i]->out.p_bitstream );
3957         x264_free( h->thread[i]->out.nal );
3958         x264_pthread_mutex_destroy( &h->thread[i]->mutex );
3959         x264_pthread_cond_destroy( &h->thread[i]->cv );
3960         x264_free( h->thread[i] );
3961     }
3962 }
3963
3964 int x264_encoder_delayed_frames( x264_t *h )
3965 {
3966     int delayed_frames = 0;
3967     if( h->i_thread_frames > 1 )
3968     {
3969         for( int i = 0; i < h->i_thread_frames; i++ )
3970             delayed_frames += h->thread[i]->b_thread_active;
3971         h = h->thread[h->i_thread_phase];
3972     }
3973     for( int i = 0; h->frames.current[i]; i++ )
3974         delayed_frames++;
3975     x264_pthread_mutex_lock( &h->lookahead->ofbuf.mutex );
3976     x264_pthread_mutex_lock( &h->lookahead->ifbuf.mutex );
3977     x264_pthread_mutex_lock( &h->lookahead->next.mutex );
3978     delayed_frames += h->lookahead->ifbuf.i_size + h->lookahead->next.i_size + h->lookahead->ofbuf.i_size;
3979     x264_pthread_mutex_unlock( &h->lookahead->next.mutex );
3980     x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
3981     x264_pthread_mutex_unlock( &h->lookahead->ofbuf.mutex );
3982     return delayed_frames;
3983 }
3984
3985 int x264_encoder_maximum_delayed_frames( x264_t *h )
3986 {
3987     return h->frames.i_delay;
3988 }