Sliced-threads: do hpel and deblock after returning
[x262.git] / encoder / encoder.c
1 /*****************************************************************************
2  * encoder.c: top-level encoder functions
3  *****************************************************************************
4  * Copyright (C) 2003-2012 x264 project
5  *
6  * Authors: Laurent Aimar <fenrir@via.ecp.fr>
7  *          Loren Merritt <lorenm@u.washington.edu>
8  *          Jason Garrett-Glaser <darkshikari@gmail.com>
9  *
10  * This program is free software; you can redistribute it and/or modify
11  * it under the terms of the GNU General Public License as published by
12  * the Free Software Foundation; either version 2 of the License, or
13  * (at your option) any later version.
14  *
15  * This program is distributed in the hope that it will be useful,
16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
18  * GNU General Public License for more details.
19  *
20  * You should have received a copy of the GNU General Public License
21  * along with this program; if not, write to the Free Software
22  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02111, USA.
23  *
24  * This program is also available under a commercial proprietary license.
25  * For more information, contact us at licensing@x264.com.
26  *****************************************************************************/
27
28 #include "common/common.h"
29
30 #include "set.h"
31 #include "analyse.h"
32 #include "ratecontrol.h"
33 #include "macroblock.h"
34 #include "me.h"
35
36 #if HAVE_VISUALIZE
37 #include "common/visualize.h"
38 #endif
39
40 //#define DEBUG_MB_TYPE
41
42 #define bs_write_ue bs_write_ue_big
43
44 static int x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
45                                    x264_nal_t **pp_nal, int *pi_nal,
46                                    x264_picture_t *pic_out );
47
48 /****************************************************************************
49  *
50  ******************************* x264 libs **********************************
51  *
52  ****************************************************************************/
53 static double x264_psnr( double sqe, double size )
54 {
55     double mse = sqe / (PIXEL_MAX*PIXEL_MAX * size);
56     if( mse <= 0.0000000001 ) /* Max 100dB */
57         return 100;
58
59     return -10.0 * log10( mse );
60 }
61
62 static double x264_ssim( double ssim )
63 {
64     double inv_ssim = 1 - ssim;
65     if( inv_ssim <= 0.0000000001 ) /* Max 100dB */
66         return 100;
67
68     return -10.0 * log10( inv_ssim );
69 }
70
71 static 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     if( sh->pps->b_weighted_pred && sh->i_type == SLICE_TYPE_P )
284     {
285         /* pred_weight_table() */
286         bs_write_ue( s, sh->weight[0][0].i_denom );
287         bs_write_ue( s, sh->weight[0][1].i_denom );
288         for( int i = 0; i < sh->i_num_ref_idx_l0_active; i++ )
289         {
290             int luma_weight_l0_flag = !!sh->weight[i][0].weightfn;
291             int chroma_weight_l0_flag = !!sh->weight[i][1].weightfn || !!sh->weight[i][2].weightfn;
292             bs_write1( s, luma_weight_l0_flag );
293             if( luma_weight_l0_flag )
294             {
295                 bs_write_se( s, sh->weight[i][0].i_scale );
296                 bs_write_se( s, sh->weight[i][0].i_offset );
297             }
298             bs_write1( s, chroma_weight_l0_flag );
299             if( chroma_weight_l0_flag )
300             {
301                 for( int j = 1; j < 3; j++ )
302                 {
303                     bs_write_se( s, sh->weight[i][j].i_scale );
304                     bs_write_se( s, sh->weight[i][j].i_offset );
305                 }
306             }
307         }
308     }
309     else if( sh->pps->b_weighted_bipred == 1 && sh->i_type == SLICE_TYPE_B )
310     {
311       /* TODO */
312     }
313
314     if( i_nal_ref_idc != 0 )
315     {
316         if( sh->i_idr_pic_id >= 0 )
317         {
318             bs_write1( s, 0 );  /* no output of prior pics flag */
319             bs_write1( s, 0 );  /* long term reference flag */
320         }
321         else
322         {
323             bs_write1( s, sh->i_mmco_command_count > 0 ); /* adaptive_ref_pic_marking_mode_flag */
324             if( sh->i_mmco_command_count > 0 )
325             {
326                 for( int i = 0; i < sh->i_mmco_command_count; i++ )
327                 {
328                     bs_write_ue( s, 1 ); /* mark short term ref as unused */
329                     bs_write_ue( s, sh->mmco[i].i_difference_of_pic_nums - 1 );
330                 }
331                 bs_write_ue( s, 0 ); /* end command list */
332             }
333         }
334     }
335
336     if( sh->pps->b_cabac && sh->i_type != SLICE_TYPE_I )
337         bs_write_ue( s, sh->i_cabac_init_idc );
338
339     bs_write_se( s, sh->i_qp_delta );      /* slice qp delta */
340
341     if( sh->pps->b_deblocking_filter_control )
342     {
343         bs_write_ue( s, sh->i_disable_deblocking_filter_idc );
344         if( sh->i_disable_deblocking_filter_idc != 1 )
345         {
346             bs_write_se( s, sh->i_alpha_c0_offset >> 1 );
347             bs_write_se( s, sh->i_beta_offset >> 1 );
348         }
349     }
350 }
351
352 /* If we are within a reasonable distance of the end of the memory allocated for the bitstream, */
353 /* reallocate, adding an arbitrary amount of space. */
354 static int x264_bitstream_check_buffer( x264_t *h )
355 {
356     uint8_t *bs_bak = h->out.p_bitstream;
357     int max_row_size = (2500 << SLICE_MBAFF) * h->mb.i_mb_width;
358     if( (h->param.b_cabac && (h->cabac.p_end - h->cabac.p < max_row_size)) ||
359         (h->out.bs.p_end - h->out.bs.p < max_row_size) )
360     {
361         h->out.i_bitstream += max_row_size;
362         CHECKED_MALLOC( h->out.p_bitstream, h->out.i_bitstream );
363         h->mc.memcpy_aligned( h->out.p_bitstream, bs_bak, (h->out.i_bitstream - max_row_size) & ~15 );
364         intptr_t delta = h->out.p_bitstream - bs_bak;
365
366         h->out.bs.p_start += delta;
367         h->out.bs.p += delta;
368         h->out.bs.p_end = h->out.p_bitstream + h->out.i_bitstream;
369
370         h->cabac.p_start += delta;
371         h->cabac.p += delta;
372         h->cabac.p_end = h->out.p_bitstream + h->out.i_bitstream;
373
374         for( int i = 0; i <= h->out.i_nal; i++ )
375             h->out.nal[i].p_payload += delta;
376         x264_free( bs_bak );
377     }
378     return 0;
379 fail:
380     x264_free( bs_bak );
381     return -1;
382 }
383
384 #if HAVE_THREAD
385 static void x264_encoder_thread_init( x264_t *h )
386 {
387     if( h->param.i_sync_lookahead )
388         x264_lower_thread_priority( 10 );
389
390 #if HAVE_MMX
391     /* Misalign mask has to be set separately for each thread. */
392     if( h->param.cpu&X264_CPU_SSE_MISALIGN )
393         x264_cpu_mask_misalign_sse();
394 #endif
395 }
396 #endif
397
398 /****************************************************************************
399  *
400  ****************************************************************************
401  ****************************** External API*********************************
402  ****************************************************************************
403  *
404  ****************************************************************************/
405
406 static int x264_validate_parameters( x264_t *h, int b_open )
407 {
408 #if HAVE_MMX
409 #ifdef __SSE__
410     if( b_open && !(x264_cpu_detect() & X264_CPU_SSE) )
411     {
412         x264_log( h, X264_LOG_ERROR, "your cpu does not support SSE1, but x264 was compiled with asm support\n");
413 #else
414     if( b_open && !(x264_cpu_detect() & X264_CPU_MMX2) )
415     {
416         x264_log( h, X264_LOG_ERROR, "your cpu does not support MMXEXT, but x264 was compiled with asm support\n");
417 #endif
418         x264_log( h, X264_LOG_ERROR, "to run x264, recompile without asm support (configure --disable-asm)\n");
419         return -1;
420     }
421 #endif
422
423 #if HAVE_INTERLACED
424     h->param.b_interlaced = !!PARAM_INTERLACED;
425 #else
426     if( h->param.b_interlaced )
427     {
428         x264_log( h, X264_LOG_ERROR, "not compiled with interlaced support\n" );
429         return -1;
430     }
431 #endif
432
433     if( h->param.i_width <= 0 || h->param.i_height <= 0 )
434     {
435         x264_log( h, X264_LOG_ERROR, "invalid width x height (%dx%d)\n",
436                   h->param.i_width, h->param.i_height );
437         return -1;
438     }
439
440     int i_csp = h->param.i_csp & X264_CSP_MASK;
441 #if X264_CHROMA_FORMAT
442     if( CHROMA_FORMAT != CHROMA_420 && i_csp >= X264_CSP_I420 && i_csp <= X264_CSP_NV12 )
443     {
444         x264_log( h, X264_LOG_ERROR, "not compiled with 4:2:0 support\n" );
445         return -1;
446     }
447     else if( CHROMA_FORMAT != CHROMA_422 && i_csp >= X264_CSP_I422 && i_csp <= X264_CSP_NV16 )
448     {
449         x264_log( h, X264_LOG_ERROR, "not compiled with 4:2:2 support\n" );
450         return -1;
451     }
452     else if( CHROMA_FORMAT != CHROMA_444 && i_csp >= X264_CSP_I444 && i_csp <= X264_CSP_RGB )
453     {
454         x264_log( h, X264_LOG_ERROR, "not compiled with 4:4:4 support\n" );
455         return -1;
456     }
457 #endif
458     if( i_csp <= X264_CSP_NONE || i_csp >= X264_CSP_MAX )
459     {
460         x264_log( h, X264_LOG_ERROR, "invalid CSP (only I420/YV12/NV12/I422/YV16/NV16/I444/YV24/BGR/BGRA/RGB supported)\n" );
461         return -1;
462     }
463
464     if( i_csp < X264_CSP_I444 && h->param.i_width % 2 )
465     {
466         x264_log( h, X264_LOG_ERROR, "width not divisible by 2 (%dx%d)\n",
467                   h->param.i_width, h->param.i_height );
468         return -1;
469     }
470
471     if( i_csp < X264_CSP_I422 && PARAM_INTERLACED && h->param.i_height % 4 )
472     {
473         x264_log( h, X264_LOG_ERROR, "height not divisible by 4 (%dx%d)\n",
474                   h->param.i_width, h->param.i_height );
475         return -1;
476     }
477
478     if( (i_csp < X264_CSP_I422 || PARAM_INTERLACED) && h->param.i_height % 2 )
479     {
480         x264_log( h, X264_LOG_ERROR, "height not divisible by 2 (%dx%d)\n",
481                   h->param.i_width, h->param.i_height );
482         return -1;
483     }
484
485     if( (h->param.crop_rect.i_left + h->param.crop_rect.i_right ) >= h->param.i_width ||
486         (h->param.crop_rect.i_top  + h->param.crop_rect.i_bottom) >= h->param.i_height )
487     {
488         x264_log( h, X264_LOG_ERROR, "invalid crop-rect %u,%u,%u,%u\n", h->param.crop_rect.i_left,
489                   h->param.crop_rect.i_top, h->param.crop_rect.i_right,  h->param.crop_rect.i_bottom );
490         return -1;
491     }
492
493     if( h->param.i_threads == X264_THREADS_AUTO )
494         h->param.i_threads = x264_cpu_num_processors() * (h->param.b_sliced_threads?2:3)/2;
495     if( h->param.i_threads > 1 )
496     {
497 #if !HAVE_THREAD
498         x264_log( h, X264_LOG_WARNING, "not compiled with thread support!\n");
499         h->param.i_threads = 1;
500 #endif
501         /* Avoid absurdly small thread slices as they can reduce performance
502          * and VBV compliance.  Capped at an arbitrary 4 rows per thread. */
503         if( h->param.b_sliced_threads )
504         {
505             int max_threads = (h->param.i_height+15)/16 / 4;
506             h->param.i_threads = X264_MIN( h->param.i_threads, max_threads );
507         }
508     }
509     h->param.i_threads = x264_clip3( h->param.i_threads, 1, X264_THREAD_MAX );
510     if( h->param.i_threads == 1 )
511         h->param.b_sliced_threads = 0;
512     h->i_thread_frames = h->param.b_sliced_threads ? 1 : h->param.i_threads;
513     if( h->i_thread_frames > 1 )
514         h->param.nalu_process = NULL;
515
516     h->param.i_keyint_max = x264_clip3( h->param.i_keyint_max, 1, X264_KEYINT_MAX_INFINITE );
517     if( h->param.i_keyint_max == 1 )
518     {
519         h->param.b_intra_refresh = 0;
520         h->param.analyse.i_weighted_pred = 0;
521     }
522
523     h->param.i_frame_packing = x264_clip3( h->param.i_frame_packing, -1, 5 );
524
525     /* Detect default ffmpeg settings and terminate with an error. */
526     if( b_open )
527     {
528         int score = 0;
529         score += h->param.analyse.i_me_range == 0;
530         score += h->param.rc.i_qp_step == 3;
531         score += h->param.i_keyint_max == 12;
532         score += h->param.rc.i_qp_min == 2;
533         score += h->param.rc.i_qp_max == 31;
534         score += h->param.rc.f_qcompress == 0.5;
535         score += fabs(h->param.rc.f_ip_factor - 1.25) < 0.01;
536         score += fabs(h->param.rc.f_pb_factor - 1.25) < 0.01;
537         score += h->param.analyse.inter == 0 && h->param.analyse.i_subpel_refine == 8;
538         if( score >= 5 )
539         {
540             x264_log( h, X264_LOG_ERROR, "broken ffmpeg default settings detected\n" );
541             x264_log( h, X264_LOG_ERROR, "use an encoding preset (e.g. -vpre medium)\n" );
542             x264_log( h, X264_LOG_ERROR, "preset usage: -vpre <speed> -vpre <profile>\n" );
543             x264_log( h, X264_LOG_ERROR, "speed presets are listed in x264 --help\n" );
544             x264_log( h, X264_LOG_ERROR, "profile is optional; x264 defaults to high\n" );
545             return -1;
546         }
547     }
548
549     if( h->param.rc.i_rc_method < 0 || h->param.rc.i_rc_method > 2 )
550     {
551         x264_log( h, X264_LOG_ERROR, "no ratecontrol method specified\n" );
552         return -1;
553     }
554     h->param.rc.f_rf_constant = x264_clip3f( h->param.rc.f_rf_constant, -QP_BD_OFFSET, 51 );
555     h->param.rc.f_rf_constant_max = x264_clip3f( h->param.rc.f_rf_constant_max, -QP_BD_OFFSET, 51 );
556     h->param.rc.i_qp_constant = x264_clip3( h->param.rc.i_qp_constant, 0, QP_MAX );
557     h->param.analyse.i_subpel_refine = x264_clip3( h->param.analyse.i_subpel_refine, 0, 11 );
558     h->param.rc.f_ip_factor = X264_MAX( h->param.rc.f_ip_factor, 0.01f );
559     h->param.rc.f_pb_factor = X264_MAX( h->param.rc.f_pb_factor, 0.01f );
560     if( h->param.rc.i_rc_method == X264_RC_CRF )
561     {
562         h->param.rc.i_qp_constant = h->param.rc.f_rf_constant + QP_BD_OFFSET;
563         h->param.rc.i_bitrate = 0;
564     }
565     if( (h->param.rc.i_rc_method == X264_RC_CQP || h->param.rc.i_rc_method == X264_RC_CRF)
566         && h->param.rc.i_qp_constant == 0 )
567     {
568         h->mb.b_lossless = 1;
569         h->param.i_cqm_preset = X264_CQM_FLAT;
570         h->param.psz_cqm_file = NULL;
571         h->param.rc.i_rc_method = X264_RC_CQP;
572         h->param.rc.f_ip_factor = 1;
573         h->param.rc.f_pb_factor = 1;
574         h->param.analyse.b_psnr = 0;
575         h->param.analyse.b_ssim = 0;
576         h->param.analyse.i_chroma_qp_offset = 0;
577         h->param.analyse.i_trellis = 0;
578         h->param.analyse.b_fast_pskip = 0;
579         h->param.analyse.i_noise_reduction = 0;
580         h->param.analyse.b_psy = 0;
581         h->param.i_bframe = 0;
582         /* 8x8dct is not useful without RD in CAVLC lossless */
583         if( !h->param.b_cabac && h->param.analyse.i_subpel_refine < 6 )
584             h->param.analyse.b_transform_8x8 = 0;
585     }
586     if( h->param.rc.i_rc_method == X264_RC_CQP )
587     {
588         float qp_p = h->param.rc.i_qp_constant;
589         float qp_i = qp_p - 6*log2f( h->param.rc.f_ip_factor );
590         float qp_b = qp_p + 6*log2f( h->param.rc.f_pb_factor );
591         h->param.rc.i_qp_min = x264_clip3( (int)(X264_MIN3( qp_p, qp_i, qp_b )), 0, QP_MAX );
592         h->param.rc.i_qp_max = x264_clip3( (int)(X264_MAX3( qp_p, qp_i, qp_b ) + .999), 0, QP_MAX );
593         h->param.rc.i_aq_mode = 0;
594         h->param.rc.b_mb_tree = 0;
595         h->param.rc.i_bitrate = 0;
596     }
597     h->param.rc.i_qp_max = x264_clip3( h->param.rc.i_qp_max, 0, QP_MAX );
598     h->param.rc.i_qp_min = x264_clip3( h->param.rc.i_qp_min, 0, h->param.rc.i_qp_max );
599     h->param.rc.i_qp_step = x264_clip3( h->param.rc.i_qp_step, 2, QP_MAX );
600     h->param.rc.i_bitrate = x264_clip3( h->param.rc.i_bitrate, 0, 2000000 );
601     h->param.rc.i_vbv_buffer_size = x264_clip3( h->param.rc.i_vbv_buffer_size, 0, 2000000 );
602     h->param.rc.i_vbv_max_bitrate = x264_clip3( h->param.rc.i_vbv_max_bitrate, 0, 2000000 );
603     h->param.rc.f_vbv_buffer_init = x264_clip3f( h->param.rc.f_vbv_buffer_init, 0, 2000000 );
604     if( h->param.rc.i_vbv_buffer_size )
605     {
606         if( h->param.rc.i_rc_method == X264_RC_CQP )
607         {
608             x264_log( h, X264_LOG_WARNING, "VBV is incompatible with constant QP, ignored.\n" );
609             h->param.rc.i_vbv_max_bitrate = 0;
610             h->param.rc.i_vbv_buffer_size = 0;
611         }
612         else if( h->param.rc.i_vbv_max_bitrate == 0 )
613         {
614             if( h->param.rc.i_rc_method == X264_RC_ABR )
615             {
616                 x264_log( h, X264_LOG_WARNING, "VBV maxrate unspecified, assuming CBR\n" );
617                 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate;
618             }
619             else
620             {
621                 x264_log( h, X264_LOG_WARNING, "VBV bufsize set but maxrate unspecified, ignored\n" );
622                 h->param.rc.i_vbv_buffer_size = 0;
623             }
624         }
625         else if( h->param.rc.i_vbv_max_bitrate < h->param.rc.i_bitrate &&
626                  h->param.rc.i_rc_method == X264_RC_ABR )
627         {
628             x264_log( h, X264_LOG_WARNING, "max bitrate less than average bitrate, assuming CBR\n" );
629             h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate;
630         }
631     }
632     else if( h->param.rc.i_vbv_max_bitrate )
633     {
634         x264_log( h, X264_LOG_WARNING, "VBV maxrate specified, but no bufsize, ignored\n" );
635         h->param.rc.i_vbv_max_bitrate = 0;
636     }
637
638     h->param.i_slice_max_size = X264_MAX( h->param.i_slice_max_size, 0 );
639     h->param.i_slice_max_mbs = X264_MAX( h->param.i_slice_max_mbs, 0 );
640
641     int max_slices = (h->param.i_height+((16<<PARAM_INTERLACED)-1))/(16<<PARAM_INTERLACED);
642     if( h->param.b_sliced_threads )
643         h->param.i_slice_count = x264_clip3( h->param.i_threads, 0, max_slices );
644     else
645     {
646         h->param.i_slice_count = x264_clip3( h->param.i_slice_count, 0, max_slices );
647         if( h->param.i_slice_max_mbs || h->param.i_slice_max_size )
648             h->param.i_slice_count = 0;
649     }
650
651     if( h->param.b_bluray_compat )
652     {
653         h->param.i_bframe_pyramid = X264_MIN( X264_B_PYRAMID_STRICT, h->param.i_bframe_pyramid );
654         h->param.i_bframe = X264_MIN( h->param.i_bframe, 3 );
655         h->param.b_aud = 1;
656         h->param.i_nal_hrd = X264_MAX( h->param.i_nal_hrd, X264_NAL_HRD_VBR );
657         h->param.i_slice_max_size = 0;
658         h->param.i_slice_max_mbs = 0;
659         h->param.b_intra_refresh = 0;
660         h->param.i_frame_reference = X264_MIN( h->param.i_frame_reference, 6 );
661         h->param.i_dpb_size = X264_MIN( h->param.i_dpb_size, 6 );
662         /* Don't use I-frames, because Blu-ray treats them the same as IDR. */
663         h->param.i_keyint_min = 1;
664         /* Due to the proliferation of broken players that don't handle dupes properly. */
665         h->param.analyse.i_weighted_pred = X264_MIN( h->param.analyse.i_weighted_pred, X264_WEIGHTP_SIMPLE );
666         if( h->param.b_fake_interlaced )
667             h->param.b_pic_struct = 1;
668     }
669
670     h->param.i_frame_reference = x264_clip3( h->param.i_frame_reference, 1, X264_REF_MAX );
671     h->param.i_dpb_size = x264_clip3( h->param.i_dpb_size, 1, X264_REF_MAX );
672     if( h->param.i_scenecut_threshold < 0 )
673         h->param.i_scenecut_threshold = 0;
674     h->param.analyse.i_direct_mv_pred = x264_clip3( h->param.analyse.i_direct_mv_pred, X264_DIRECT_PRED_NONE, X264_DIRECT_PRED_AUTO );
675     if( !h->param.analyse.i_subpel_refine && h->param.analyse.i_direct_mv_pred > X264_DIRECT_PRED_SPATIAL )
676     {
677         x264_log( h, X264_LOG_WARNING, "subme=0 + direct=temporal is not supported\n" );
678         h->param.analyse.i_direct_mv_pred = X264_DIRECT_PRED_SPATIAL;
679     }
680     h->param.i_bframe = x264_clip3( h->param.i_bframe, 0, X264_MIN( X264_BFRAME_MAX, h->param.i_keyint_max-1 ) );
681     h->param.i_bframe_bias = x264_clip3( h->param.i_bframe_bias, -90, 100 );
682     if( h->param.i_bframe <= 1 )
683         h->param.i_bframe_pyramid = X264_B_PYRAMID_NONE;
684     h->param.i_bframe_pyramid = x264_clip3( h->param.i_bframe_pyramid, X264_B_PYRAMID_NONE, X264_B_PYRAMID_NORMAL );
685     h->param.i_bframe_adaptive = x264_clip3( h->param.i_bframe_adaptive, X264_B_ADAPT_NONE, X264_B_ADAPT_TRELLIS );
686     if( !h->param.i_bframe )
687     {
688         h->param.i_bframe_adaptive = X264_B_ADAPT_NONE;
689         h->param.analyse.i_direct_mv_pred = 0;
690         h->param.analyse.b_weighted_bipred = 0;
691         h->param.b_open_gop = 0;
692     }
693     if( h->param.b_intra_refresh && h->param.i_bframe_pyramid == X264_B_PYRAMID_NORMAL )
694     {
695         x264_log( h, X264_LOG_WARNING, "b-pyramid normal + intra-refresh is not supported\n" );
696         h->param.i_bframe_pyramid = X264_B_PYRAMID_STRICT;
697     }
698     if( h->param.b_intra_refresh && (h->param.i_frame_reference > 1 || h->param.i_dpb_size > 1) )
699     {
700         x264_log( h, X264_LOG_WARNING, "ref > 1 + intra-refresh is not supported\n" );
701         h->param.i_frame_reference = 1;
702         h->param.i_dpb_size = 1;
703     }
704     if( h->param.b_intra_refresh && h->param.b_open_gop )
705     {
706         x264_log( h, X264_LOG_WARNING, "intra-refresh is not compatible with open-gop\n" );
707         h->param.b_open_gop = 0;
708     }
709     float fps = h->param.i_fps_num > 0 && h->param.i_fps_den > 0 ? (float) h->param.i_fps_num / h->param.i_fps_den : 25.0;
710     if( h->param.i_keyint_min == X264_KEYINT_MIN_AUTO )
711         h->param.i_keyint_min = X264_MIN( h->param.i_keyint_max / 10, fps );
712     h->param.i_keyint_min = x264_clip3( h->param.i_keyint_min, 1, h->param.i_keyint_max/2+1 );
713     h->param.rc.i_lookahead = x264_clip3( h->param.rc.i_lookahead, 0, X264_LOOKAHEAD_MAX );
714     {
715         int maxrate = X264_MAX( h->param.rc.i_vbv_max_bitrate, h->param.rc.i_bitrate );
716         float bufsize = maxrate ? (float)h->param.rc.i_vbv_buffer_size / maxrate : 0;
717         h->param.rc.i_lookahead = X264_MIN( h->param.rc.i_lookahead, X264_MAX( h->param.i_keyint_max, bufsize*fps ) );
718     }
719
720     if( !h->param.i_timebase_num || !h->param.i_timebase_den || !(h->param.b_vfr_input || h->param.b_pulldown) )
721     {
722         h->param.i_timebase_num = h->param.i_fps_den;
723         h->param.i_timebase_den = h->param.i_fps_num;
724     }
725
726     h->param.rc.f_qcompress = x264_clip3f( h->param.rc.f_qcompress, 0.0, 1.0 );
727     if( h->param.i_keyint_max == 1 || h->param.rc.f_qcompress == 1 )
728         h->param.rc.b_mb_tree = 0;
729     if( (!h->param.b_intra_refresh && h->param.i_keyint_max != X264_KEYINT_MAX_INFINITE) &&
730         !h->param.rc.i_lookahead && h->param.rc.b_mb_tree )
731     {
732         x264_log( h, X264_LOG_WARNING, "lookaheadless mb-tree requires intra refresh or infinite keyint\n" );
733         h->param.rc.b_mb_tree = 0;
734     }
735     if( h->param.rc.b_stat_read )
736         h->param.rc.i_lookahead = 0;
737 #if HAVE_THREAD
738     if( h->param.i_sync_lookahead < 0 )
739         h->param.i_sync_lookahead = h->param.i_bframe + 1;
740     h->param.i_sync_lookahead = X264_MIN( h->param.i_sync_lookahead, X264_LOOKAHEAD_MAX );
741     if( h->param.rc.b_stat_read || h->i_thread_frames == 1 )
742         h->param.i_sync_lookahead = 0;
743 #else
744     h->param.i_sync_lookahead = 0;
745 #endif
746
747     h->param.i_deblocking_filter_alphac0 = x264_clip3( h->param.i_deblocking_filter_alphac0, -6, 6 );
748     h->param.i_deblocking_filter_beta    = x264_clip3( h->param.i_deblocking_filter_beta, -6, 6 );
749     h->param.analyse.i_luma_deadzone[0] = x264_clip3( h->param.analyse.i_luma_deadzone[0], 0, 32 );
750     h->param.analyse.i_luma_deadzone[1] = x264_clip3( h->param.analyse.i_luma_deadzone[1], 0, 32 );
751
752     h->param.i_cabac_init_idc = x264_clip3( h->param.i_cabac_init_idc, 0, 2 );
753
754     if( h->param.i_cqm_preset < X264_CQM_FLAT || h->param.i_cqm_preset > X264_CQM_CUSTOM )
755         h->param.i_cqm_preset = X264_CQM_FLAT;
756
757     if( h->param.analyse.i_me_method < X264_ME_DIA ||
758         h->param.analyse.i_me_method > X264_ME_TESA )
759         h->param.analyse.i_me_method = X264_ME_HEX;
760     h->param.analyse.i_me_range = x264_clip3( h->param.analyse.i_me_range, 4, 1024 );
761     if( h->param.analyse.i_me_range > 16 && h->param.analyse.i_me_method <= X264_ME_HEX )
762         h->param.analyse.i_me_range = 16;
763     if( h->param.analyse.i_me_method == X264_ME_TESA &&
764         (h->mb.b_lossless || h->param.analyse.i_subpel_refine <= 1) )
765         h->param.analyse.i_me_method = X264_ME_ESA;
766     h->param.analyse.b_mixed_references = h->param.analyse.b_mixed_references && h->param.i_frame_reference > 1;
767     h->param.analyse.inter &= X264_ANALYSE_PSUB16x16|X264_ANALYSE_PSUB8x8|X264_ANALYSE_BSUB16x16|
768                               X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
769     h->param.analyse.intra &= X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
770     if( !(h->param.analyse.inter & X264_ANALYSE_PSUB16x16) )
771         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
772     if( !h->param.analyse.b_transform_8x8 )
773     {
774         h->param.analyse.inter &= ~X264_ANALYSE_I8x8;
775         h->param.analyse.intra &= ~X264_ANALYSE_I8x8;
776     }
777     h->param.analyse.i_trellis = x264_clip3( h->param.analyse.i_trellis, 0, 2 );
778     h->param.rc.i_aq_mode = x264_clip3( h->param.rc.i_aq_mode, 0, 2 );
779     h->param.rc.f_aq_strength = x264_clip3f( h->param.rc.f_aq_strength, 0, 3 );
780     if( h->param.rc.f_aq_strength == 0 )
781         h->param.rc.i_aq_mode = 0;
782
783     if( h->param.i_log_level < X264_LOG_INFO )
784     {
785         h->param.analyse.b_psnr = 0;
786         h->param.analyse.b_ssim = 0;
787     }
788     /* Warn users trying to measure PSNR/SSIM with psy opts on. */
789     if( b_open && (h->param.analyse.b_psnr || h->param.analyse.b_ssim) )
790     {
791         char *s = NULL;
792
793         if( h->param.analyse.b_psy )
794         {
795             s = h->param.analyse.b_psnr ? "psnr" : "ssim";
796             x264_log( h, X264_LOG_WARNING, "--%s used with psy on: results will be invalid!\n", s );
797         }
798         else if( !h->param.rc.i_aq_mode && h->param.analyse.b_ssim )
799         {
800             x264_log( h, X264_LOG_WARNING, "--ssim used with AQ off: results will be invalid!\n" );
801             s = "ssim";
802         }
803         else if(  h->param.rc.i_aq_mode && h->param.analyse.b_psnr )
804         {
805             x264_log( h, X264_LOG_WARNING, "--psnr used with AQ on: results will be invalid!\n" );
806             s = "psnr";
807         }
808         if( s )
809             x264_log( h, X264_LOG_WARNING, "--tune %s should be used if attempting to benchmark %s!\n", s, s );
810     }
811
812     if( !h->param.analyse.b_psy )
813     {
814         h->param.analyse.f_psy_rd = 0;
815         h->param.analyse.f_psy_trellis = 0;
816     }
817     h->param.analyse.f_psy_rd = x264_clip3f( h->param.analyse.f_psy_rd, 0, 10 );
818     h->param.analyse.f_psy_trellis = x264_clip3f( h->param.analyse.f_psy_trellis, 0, 10 );
819     h->mb.i_psy_rd = h->param.analyse.i_subpel_refine >= 6 ? FIX8( h->param.analyse.f_psy_rd ) : 0;
820     h->mb.i_psy_trellis = h->param.analyse.i_trellis ? FIX8( h->param.analyse.f_psy_trellis / 4 ) : 0;
821     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -32, 32);
822     /* In 4:4:4 mode, chroma gets twice as much resolution, so we can halve its quality. */
823     if( b_open && i_csp >= X264_CSP_I444 && i_csp < X264_CSP_BGR && h->param.analyse.b_psy )
824         h->param.analyse.i_chroma_qp_offset += 6;
825     /* Psy RDO increases overall quantizers to improve the quality of luma--this indirectly hurts chroma quality */
826     /* so we lower the chroma QP offset to compensate */
827     if( b_open && h->mb.i_psy_rd )
828         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_rd < 0.25 ? 1 : 2;
829     /* Psy trellis has a similar effect. */
830     if( b_open && h->mb.i_psy_trellis )
831         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_trellis < 0.25 ? 1 : 2;
832     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -12, 12);
833     /* MB-tree requires AQ to be on, even if the strength is zero. */
834     if( !h->param.rc.i_aq_mode && h->param.rc.b_mb_tree )
835     {
836         h->param.rc.i_aq_mode = 1;
837         h->param.rc.f_aq_strength = 0;
838     }
839     h->param.analyse.i_noise_reduction = x264_clip3( h->param.analyse.i_noise_reduction, 0, 1<<16 );
840     if( h->param.analyse.i_subpel_refine >= 10 && (h->param.analyse.i_trellis != 2 || !h->param.rc.i_aq_mode) )
841         h->param.analyse.i_subpel_refine = 9;
842
843     {
844         const x264_level_t *l = x264_levels;
845         if( h->param.i_level_idc < 0 )
846         {
847             int maxrate_bak = h->param.rc.i_vbv_max_bitrate;
848             if( h->param.rc.i_rc_method == X264_RC_ABR && h->param.rc.i_vbv_buffer_size <= 0 )
849                 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate * 2;
850             x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
851             do h->param.i_level_idc = l->level_idc;
852                 while( l[1].level_idc && x264_validate_levels( h, 0 ) && l++ );
853             h->param.rc.i_vbv_max_bitrate = maxrate_bak;
854         }
855         else
856         {
857             while( l->level_idc && l->level_idc != h->param.i_level_idc )
858                 l++;
859             if( l->level_idc == 0 )
860             {
861                 x264_log( h, X264_LOG_ERROR, "invalid level_idc: %d\n", h->param.i_level_idc );
862                 return -1;
863             }
864         }
865         if( h->param.analyse.i_mv_range <= 0 )
866             h->param.analyse.i_mv_range = l->mv_range >> PARAM_INTERLACED;
867         else
868             h->param.analyse.i_mv_range = x264_clip3(h->param.analyse.i_mv_range, 32, 512 >> PARAM_INTERLACED);
869     }
870
871     h->param.analyse.i_weighted_pred = x264_clip3( h->param.analyse.i_weighted_pred, X264_WEIGHTP_NONE, X264_WEIGHTP_SMART );
872
873     if( PARAM_INTERLACED )
874     {
875         if( h->param.analyse.i_me_method >= X264_ME_ESA )
876         {
877             x264_log( h, X264_LOG_WARNING, "interlace + me=esa is not implemented\n" );
878             h->param.analyse.i_me_method = X264_ME_UMH;
879         }
880         if( h->param.analyse.i_weighted_pred > 0 )
881         {
882             x264_log( h, X264_LOG_WARNING, "interlace + weightp is not implemented\n" );
883             h->param.analyse.i_weighted_pred = X264_WEIGHTP_NONE;
884         }
885     }
886
887     if( !h->param.analyse.i_weighted_pred && h->param.rc.b_mb_tree && h->param.analyse.b_psy )
888         h->param.analyse.i_weighted_pred = X264_WEIGHTP_FAKE;
889
890     if( h->i_thread_frames > 1 )
891     {
892         int r = h->param.analyse.i_mv_range_thread;
893         int r2;
894         if( r <= 0 )
895         {
896             // half of the available space is reserved and divided evenly among the threads,
897             // the rest is allocated to whichever thread is far enough ahead to use it.
898             // reserving more space increases quality for some videos, but costs more time
899             // in thread synchronization.
900             int max_range = (h->param.i_height + X264_THREAD_HEIGHT) / h->i_thread_frames - X264_THREAD_HEIGHT;
901             r = max_range / 2;
902         }
903         r = X264_MAX( r, h->param.analyse.i_me_range );
904         r = X264_MIN( r, h->param.analyse.i_mv_range );
905         // round up to use the whole mb row
906         r2 = (r & ~15) + ((-X264_THREAD_HEIGHT) & 15);
907         if( r2 < r )
908             r2 += 16;
909         x264_log( h, X264_LOG_DEBUG, "using mv_range_thread = %d\n", r2 );
910         h->param.analyse.i_mv_range_thread = r2;
911     }
912
913     if( h->param.rc.f_rate_tolerance < 0 )
914         h->param.rc.f_rate_tolerance = 0;
915     if( h->param.rc.f_qblur < 0 )
916         h->param.rc.f_qblur = 0;
917     if( h->param.rc.f_complexity_blur < 0 )
918         h->param.rc.f_complexity_blur = 0;
919
920     h->param.i_sps_id &= 31;
921
922     if( PARAM_INTERLACED )
923         h->param.b_pic_struct = 1;
924
925     h->param.i_nal_hrd = x264_clip3( h->param.i_nal_hrd, X264_NAL_HRD_NONE, X264_NAL_HRD_CBR );
926
927     if( h->param.i_nal_hrd && !h->param.rc.i_vbv_buffer_size )
928     {
929         x264_log( h, X264_LOG_WARNING, "NAL HRD parameters require VBV parameters\n" );
930         h->param.i_nal_hrd = X264_NAL_HRD_NONE;
931     }
932
933     if( h->param.i_nal_hrd == X264_NAL_HRD_CBR &&
934        (h->param.rc.i_bitrate != h->param.rc.i_vbv_max_bitrate || !h->param.rc.i_vbv_max_bitrate) )
935     {
936         x264_log( h, X264_LOG_WARNING, "CBR HRD requires constant bitrate\n" );
937         h->param.i_nal_hrd = X264_NAL_HRD_VBR;
938     }
939
940     /* ensure the booleans are 0 or 1 so they can be used in math */
941 #define BOOLIFY(x) h->param.x = !!h->param.x
942     BOOLIFY( b_cabac );
943     BOOLIFY( b_constrained_intra );
944     BOOLIFY( b_deblocking_filter );
945     BOOLIFY( b_deterministic );
946     BOOLIFY( b_sliced_threads );
947     BOOLIFY( b_interlaced );
948     BOOLIFY( b_intra_refresh );
949     BOOLIFY( b_visualize );
950     BOOLIFY( b_aud );
951     BOOLIFY( b_repeat_headers );
952     BOOLIFY( b_annexb );
953     BOOLIFY( b_vfr_input );
954     BOOLIFY( b_pulldown );
955     BOOLIFY( b_tff );
956     BOOLIFY( b_pic_struct );
957     BOOLIFY( b_fake_interlaced );
958     BOOLIFY( b_open_gop );
959     BOOLIFY( b_bluray_compat );
960     BOOLIFY( b_full_recon );
961     BOOLIFY( analyse.b_transform_8x8 );
962     BOOLIFY( analyse.b_weighted_bipred );
963     BOOLIFY( analyse.b_chroma_me );
964     BOOLIFY( analyse.b_mixed_references );
965     BOOLIFY( analyse.b_fast_pskip );
966     BOOLIFY( analyse.b_dct_decimate );
967     BOOLIFY( analyse.b_psy );
968     BOOLIFY( analyse.b_psnr );
969     BOOLIFY( analyse.b_ssim );
970     BOOLIFY( rc.b_stat_write );
971     BOOLIFY( rc.b_stat_read );
972     BOOLIFY( rc.b_mb_tree );
973 #undef BOOLIFY
974
975     return 0;
976 }
977
978 static void mbcmp_init( x264_t *h )
979 {
980     int satd = !h->mb.b_lossless && h->param.analyse.i_subpel_refine > 1;
981     memcpy( h->pixf.mbcmp, satd ? h->pixf.satd : h->pixf.sad_aligned, sizeof(h->pixf.mbcmp) );
982     memcpy( h->pixf.mbcmp_unaligned, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.mbcmp_unaligned) );
983     h->pixf.intra_mbcmp_x3_16x16 = satd ? h->pixf.intra_satd_x3_16x16 : h->pixf.intra_sad_x3_16x16;
984     h->pixf.intra_mbcmp_x3_8x16c = satd ? h->pixf.intra_satd_x3_8x16c : h->pixf.intra_sad_x3_8x16c;
985     h->pixf.intra_mbcmp_x3_8x8c  = satd ? h->pixf.intra_satd_x3_8x8c  : h->pixf.intra_sad_x3_8x8c;
986     h->pixf.intra_mbcmp_x3_8x8 = satd ? h->pixf.intra_sa8d_x3_8x8 : h->pixf.intra_sad_x3_8x8;
987     h->pixf.intra_mbcmp_x3_4x4 = satd ? h->pixf.intra_satd_x3_4x4 : h->pixf.intra_sad_x3_4x4;
988     h->pixf.intra_mbcmp_x9_4x4 = h->param.b_cpu_independent || h->mb.b_lossless ? NULL
989                                : satd ? h->pixf.intra_satd_x9_4x4 : h->pixf.intra_sad_x9_4x4;
990     h->pixf.intra_mbcmp_x9_8x8 = h->param.b_cpu_independent || h->mb.b_lossless ? NULL
991                                : satd ? h->pixf.intra_sa8d_x9_8x8 : h->pixf.intra_sad_x9_8x8;
992     satd &= h->param.analyse.i_me_method == X264_ME_TESA;
993     memcpy( h->pixf.fpelcmp, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.fpelcmp) );
994     memcpy( h->pixf.fpelcmp_x3, satd ? h->pixf.satd_x3 : h->pixf.sad_x3, sizeof(h->pixf.fpelcmp_x3) );
995     memcpy( h->pixf.fpelcmp_x4, satd ? h->pixf.satd_x4 : h->pixf.sad_x4, sizeof(h->pixf.fpelcmp_x4) );
996 }
997
998 static void chroma_dsp_init( x264_t *h )
999 {
1000     memcpy( h->luma2chroma_pixel, x264_luma2chroma_pixel[CHROMA_FORMAT], sizeof(h->luma2chroma_pixel) );
1001
1002     switch( CHROMA_FORMAT )
1003     {
1004         case CHROMA_420:
1005             memcpy( h->predict_chroma, h->predict_8x8c, sizeof(h->predict_chroma) );
1006             h->mc.prefetch_fenc = h->mc.prefetch_fenc_420;
1007             h->loopf.deblock_chroma[0] = h->loopf.deblock_h_chroma_420;
1008             h->loopf.deblock_chroma_intra[0] = h->loopf.deblock_h_chroma_420_intra;
1009             h->loopf.deblock_chroma_mbaff = h->loopf.deblock_chroma_420_mbaff;
1010             h->loopf.deblock_chroma_intra_mbaff = h->loopf.deblock_chroma_420_intra_mbaff;
1011             h->pixf.intra_mbcmp_x3_chroma = h->pixf.intra_mbcmp_x3_8x8c;
1012             h->quantf.coeff_last[DCT_CHROMA_DC] = h->quantf.coeff_last4;
1013             h->quantf.coeff_level_run[DCT_CHROMA_DC] = h->quantf.coeff_level_run4;
1014             break;
1015         case CHROMA_422:
1016             memcpy( h->predict_chroma, h->predict_8x16c, sizeof(h->predict_chroma) );
1017             h->mc.prefetch_fenc = h->mc.prefetch_fenc_422;
1018             h->loopf.deblock_chroma[0] = h->loopf.deblock_h_chroma_422;
1019             h->loopf.deblock_chroma_intra[0] = h->loopf.deblock_h_chroma_422_intra;
1020             h->loopf.deblock_chroma_mbaff = h->loopf.deblock_chroma_422_mbaff;
1021             h->loopf.deblock_chroma_intra_mbaff = h->loopf.deblock_chroma_422_intra_mbaff;
1022             h->pixf.intra_mbcmp_x3_chroma = h->pixf.intra_mbcmp_x3_8x16c;
1023             h->quantf.coeff_last[DCT_CHROMA_DC] = h->quantf.coeff_last8;
1024             h->quantf.coeff_level_run[DCT_CHROMA_DC] = h->quantf.coeff_level_run8;
1025             break;
1026         case CHROMA_444:
1027             h->mc.prefetch_fenc = h->mc.prefetch_fenc_422; /* FIXME: doesn't cover V plane */
1028             h->loopf.deblock_chroma_mbaff = h->loopf.deblock_luma_mbaff;
1029             h->loopf.deblock_chroma_intra_mbaff = h->loopf.deblock_luma_intra_mbaff;
1030             break;
1031     }
1032 }
1033
1034 static void x264_set_aspect_ratio( x264_t *h, x264_param_t *param, int initial )
1035 {
1036     /* VUI */
1037     if( param->vui.i_sar_width > 0 && param->vui.i_sar_height > 0 )
1038     {
1039         uint32_t i_w = param->vui.i_sar_width;
1040         uint32_t i_h = param->vui.i_sar_height;
1041         uint32_t old_w = h->param.vui.i_sar_width;
1042         uint32_t old_h = h->param.vui.i_sar_height;
1043
1044         x264_reduce_fraction( &i_w, &i_h );
1045
1046         while( i_w > 65535 || i_h > 65535 )
1047         {
1048             i_w /= 2;
1049             i_h /= 2;
1050         }
1051
1052         x264_reduce_fraction( &i_w, &i_h );
1053
1054         if( i_w != old_w || i_h != old_h || initial )
1055         {
1056             h->param.vui.i_sar_width = 0;
1057             h->param.vui.i_sar_height = 0;
1058             if( i_w == 0 || i_h == 0 )
1059                 x264_log( h, X264_LOG_WARNING, "cannot create valid sample aspect ratio\n" );
1060             else
1061             {
1062                 x264_log( h, initial?X264_LOG_INFO:X264_LOG_DEBUG, "using SAR=%d/%d\n", i_w, i_h );
1063                 h->param.vui.i_sar_width = i_w;
1064                 h->param.vui.i_sar_height = i_h;
1065             }
1066             x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
1067         }
1068     }
1069 }
1070
1071 /****************************************************************************
1072  * x264_encoder_open:
1073  ****************************************************************************/
1074 x264_t *x264_encoder_open( x264_param_t *param )
1075 {
1076     x264_t *h;
1077     char buf[1000], *p;
1078     int qp, i_slicetype_length;
1079
1080     CHECKED_MALLOCZERO( h, sizeof(x264_t) );
1081
1082     /* Create a copy of param */
1083     memcpy( &h->param, param, sizeof(x264_param_t) );
1084
1085     if( param->param_free )
1086         param->param_free( param );
1087
1088     if( x264_threading_init() )
1089     {
1090         x264_log( h, X264_LOG_ERROR, "unable to initialize threading\n" );
1091         goto fail;
1092     }
1093
1094     if( x264_validate_parameters( h, 1 ) < 0 )
1095         goto fail;
1096
1097     if( h->param.psz_cqm_file )
1098         if( x264_cqm_parse_file( h, h->param.psz_cqm_file ) < 0 )
1099             goto fail;
1100
1101     if( h->param.rc.psz_stat_out )
1102         h->param.rc.psz_stat_out = strdup( h->param.rc.psz_stat_out );
1103     if( h->param.rc.psz_stat_in )
1104         h->param.rc.psz_stat_in = strdup( h->param.rc.psz_stat_in );
1105
1106     x264_reduce_fraction( &h->param.i_fps_num, &h->param.i_fps_den );
1107     x264_reduce_fraction( &h->param.i_timebase_num, &h->param.i_timebase_den );
1108
1109     /* Init x264_t */
1110     h->i_frame = -1;
1111     h->i_frame_num = 0;
1112     h->i_idr_pic_id = 0;
1113
1114     if( (uint64_t)h->param.i_timebase_den * 2 > UINT32_MAX )
1115     {
1116         x264_log( h, X264_LOG_ERROR, "Effective timebase denominator %u exceeds H.264 maximum\n", h->param.i_timebase_den );
1117         goto fail;
1118     }
1119
1120     x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
1121     x264_pps_init( h->pps, h->param.i_sps_id, &h->param, h->sps );
1122
1123     x264_set_aspect_ratio( h, &h->param, 1 );
1124
1125     x264_validate_levels( h, 1 );
1126
1127     h->chroma_qp_table = i_chroma_qp_table + 12 + h->pps->i_chroma_qp_index_offset;
1128
1129     if( x264_cqm_init( h ) < 0 )
1130         goto fail;
1131
1132     h->mb.i_mb_width = h->sps->i_mb_width;
1133     h->mb.i_mb_height = h->sps->i_mb_height;
1134     h->mb.i_mb_count = h->mb.i_mb_width * h->mb.i_mb_height;
1135
1136     h->mb.chroma_h_shift = CHROMA_FORMAT == CHROMA_420 || CHROMA_FORMAT == CHROMA_422;
1137     h->mb.chroma_v_shift = CHROMA_FORMAT == CHROMA_420;
1138
1139     /* Adaptive MBAFF and subme 0 are not supported as we require halving motion
1140      * vectors during prediction, resulting in hpel mvs.
1141      * The chosen solution is to make MBAFF non-adaptive in this case. */
1142     h->mb.b_adaptive_mbaff = PARAM_INTERLACED && h->param.analyse.i_subpel_refine;
1143
1144     /* Init frames. */
1145     if( h->param.i_bframe_adaptive == X264_B_ADAPT_TRELLIS && !h->param.rc.b_stat_read )
1146         h->frames.i_delay = X264_MAX(h->param.i_bframe,3)*4;
1147     else
1148         h->frames.i_delay = h->param.i_bframe;
1149     if( h->param.rc.b_mb_tree || h->param.rc.i_vbv_buffer_size )
1150         h->frames.i_delay = X264_MAX( h->frames.i_delay, h->param.rc.i_lookahead );
1151     i_slicetype_length = h->frames.i_delay;
1152     h->frames.i_delay += h->i_thread_frames - 1;
1153     h->frames.i_delay += h->param.i_sync_lookahead;
1154     h->frames.i_delay += h->param.b_vfr_input;
1155     h->frames.i_bframe_delay = h->param.i_bframe ? (h->param.i_bframe_pyramid ? 2 : 1) : 0;
1156
1157     h->frames.i_max_ref0 = h->param.i_frame_reference;
1158     h->frames.i_max_ref1 = X264_MIN( h->sps->vui.i_num_reorder_frames, h->param.i_frame_reference );
1159     h->frames.i_max_dpb  = h->sps->vui.i_max_dec_frame_buffering;
1160     h->frames.b_have_lowres = !h->param.rc.b_stat_read
1161         && ( h->param.rc.i_rc_method == X264_RC_ABR
1162           || h->param.rc.i_rc_method == X264_RC_CRF
1163           || h->param.i_bframe_adaptive
1164           || h->param.i_scenecut_threshold
1165           || h->param.rc.b_mb_tree
1166           || h->param.analyse.i_weighted_pred );
1167     h->frames.b_have_lowres |= h->param.rc.b_stat_read && h->param.rc.i_vbv_buffer_size > 0;
1168     h->frames.b_have_sub8x8_esa = !!(h->param.analyse.inter & X264_ANALYSE_PSUB8x8);
1169
1170     h->frames.i_last_idr =
1171     h->frames.i_last_keyframe = - h->param.i_keyint_max;
1172     h->frames.i_input    = 0;
1173     h->frames.i_largest_pts = h->frames.i_second_largest_pts = -1;
1174     h->frames.i_poc_last_open_gop = -1;
1175
1176     CHECKED_MALLOCZERO( h->frames.unused[0], (h->frames.i_delay + 3) * sizeof(x264_frame_t *) );
1177     /* Allocate room for max refs plus a few extra just in case. */
1178     CHECKED_MALLOCZERO( h->frames.unused[1], (h->i_thread_frames + X264_REF_MAX + 4) * sizeof(x264_frame_t *) );
1179     CHECKED_MALLOCZERO( h->frames.current, (h->param.i_sync_lookahead + h->param.i_bframe
1180                         + h->i_thread_frames + 3) * sizeof(x264_frame_t *) );
1181     if( h->param.analyse.i_weighted_pred > 0 )
1182         CHECKED_MALLOCZERO( h->frames.blank_unused, h->i_thread_frames * 4 * sizeof(x264_frame_t *) );
1183     h->i_ref[0] = h->i_ref[1] = 0;
1184     h->i_cpb_delay = h->i_coded_fields = h->i_disp_fields = 0;
1185     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);
1186     h->i_disp_fields_last_frame = -1;
1187     x264_rdo_init();
1188
1189     /* init CPU functions */
1190     x264_predict_16x16_init( h->param.cpu, h->predict_16x16 );
1191     x264_predict_8x8c_init( h->param.cpu, h->predict_8x8c );
1192     x264_predict_8x16c_init( h->param.cpu, h->predict_8x16c );
1193     x264_predict_8x8_init( h->param.cpu, h->predict_8x8, &h->predict_8x8_filter );
1194     x264_predict_4x4_init( h->param.cpu, h->predict_4x4 );
1195     x264_pixel_init( h->param.cpu, &h->pixf );
1196     x264_dct_init( h->param.cpu, &h->dctf );
1197     x264_zigzag_init( h->param.cpu, &h->zigzagf_progressive, &h->zigzagf_interlaced );
1198     memcpy( &h->zigzagf, PARAM_INTERLACED ? &h->zigzagf_interlaced : &h->zigzagf_progressive, sizeof(h->zigzagf) );
1199     x264_mc_init( h->param.cpu, &h->mc );
1200     x264_quant_init( h, h->param.cpu, &h->quantf );
1201     x264_deblock_init( h->param.cpu, &h->loopf, PARAM_INTERLACED );
1202     x264_bitstream_init( h->param.cpu, &h->bsf );
1203     if( h->param.b_cabac )
1204         x264_cabac_init( h );
1205     else
1206         x264_stack_align( x264_cavlc_init, h );
1207
1208     mbcmp_init( h );
1209     chroma_dsp_init( h );
1210
1211     p = buf + sprintf( buf, "using cpu capabilities:" );
1212     for( int i = 0; x264_cpu_names[i].flags; i++ )
1213     {
1214         if( !strcmp(x264_cpu_names[i].name, "SSE2")
1215             && h->param.cpu & (X264_CPU_SSE2_IS_FAST|X264_CPU_SSE2_IS_SLOW) )
1216             continue;
1217         if( !strcmp(x264_cpu_names[i].name, "SSE3")
1218             && (h->param.cpu & X264_CPU_SSSE3 || !(h->param.cpu & X264_CPU_CACHELINE_64)) )
1219             continue;
1220         if( !strcmp(x264_cpu_names[i].name, "SSE4.1")
1221             && (h->param.cpu & X264_CPU_SSE42) )
1222             continue;
1223         if( (h->param.cpu & x264_cpu_names[i].flags) == x264_cpu_names[i].flags
1224             && (!i || x264_cpu_names[i].flags != x264_cpu_names[i-1].flags) )
1225             p += sprintf( p, " %s", x264_cpu_names[i].name );
1226     }
1227     if( !h->param.cpu )
1228         p += sprintf( p, " none!" );
1229     x264_log( h, X264_LOG_INFO, "%s\n", buf );
1230
1231     float *logs = x264_analyse_prepare_costs( h );
1232     if( !logs )
1233         goto fail;
1234     for( qp = X264_MIN( h->param.rc.i_qp_min, QP_MAX_SPEC ); qp <= h->param.rc.i_qp_max; qp++ )
1235         if( x264_analyse_init_costs( h, logs, qp ) )
1236             goto fail;
1237     if( x264_analyse_init_costs( h, logs, X264_LOOKAHEAD_QP ) )
1238         goto fail;
1239     x264_free( logs );
1240
1241     static const uint16_t cost_mv_correct[7] = { 24, 47, 95, 189, 379, 757, 1515 };
1242     /* Checks for known miscompilation issues. */
1243     if( h->cost_mv[X264_LOOKAHEAD_QP][2013] != cost_mv_correct[BIT_DEPTH-8] )
1244     {
1245         x264_log( h, X264_LOG_ERROR, "MV cost test failed: x264 has been miscompiled!\n" );
1246         goto fail;
1247     }
1248
1249     /* Must be volatile or else GCC will optimize it out. */
1250     volatile int temp = 392;
1251     if( x264_clz( temp ) != 23 )
1252     {
1253         x264_log( h, X264_LOG_ERROR, "CLZ test failed: x264 has been miscompiled!\n" );
1254 #if ARCH_X86 || ARCH_X86_64
1255         x264_log( h, X264_LOG_ERROR, "Are you attempting to run an SSE4a-targeted build on a CPU that\n" );
1256         x264_log( h, X264_LOG_ERROR, "doesn't support it?\n" );
1257 #endif
1258         goto fail;
1259     }
1260
1261     h->out.i_nal = 0;
1262     h->out.i_bitstream = X264_MAX( 1000000, h->param.i_width * h->param.i_height * 4
1263         * ( h->param.rc.i_rc_method == X264_RC_ABR ? pow( 0.95, h->param.rc.i_qp_min )
1264           : pow( 0.95, h->param.rc.i_qp_constant ) * X264_MAX( 1, h->param.rc.f_ip_factor )));
1265
1266     h->nal_buffer_size = h->out.i_bitstream * 3/2 + 4;
1267     CHECKED_MALLOC( h->nal_buffer, h->nal_buffer_size );
1268
1269     if( h->param.i_threads > 1 &&
1270         x264_threadpool_init( &h->threadpool, h->param.i_threads, (void*)x264_encoder_thread_init, h ) )
1271         goto fail;
1272
1273     h->thread[0] = h;
1274     for( int i = 0; i < h->param.i_threads + !!h->param.i_sync_lookahead; i++ )
1275     {
1276         if( i )
1277             CHECKED_MALLOC( h->thread[i], sizeof(x264_t) );
1278         if( i < h->param.i_threads )
1279         {
1280             if( x264_pthread_mutex_init( &h->thread[i]->mutex, NULL ) )
1281                 goto fail;
1282             if( x264_pthread_cond_init( &h->thread[i]->cv, NULL ) )
1283                 goto fail;
1284         }
1285     }
1286
1287     for( int i = 0; i < h->param.i_threads; i++ )
1288     {
1289         int init_nal_count = h->param.i_slice_count + 3;
1290         int allocate_threadlocal_data = !h->param.b_sliced_threads || !i;
1291         if( i > 0 )
1292             *h->thread[i] = *h;
1293
1294         if( allocate_threadlocal_data )
1295         {
1296             h->thread[i]->fdec = x264_frame_pop_unused( h, 1 );
1297             if( !h->thread[i]->fdec )
1298                 goto fail;
1299         }
1300         else
1301             h->thread[i]->fdec = h->thread[0]->fdec;
1302
1303         CHECKED_MALLOC( h->thread[i]->out.p_bitstream, h->out.i_bitstream );
1304         /* Start each thread with room for init_nal_count NAL units; it'll realloc later if needed. */
1305         CHECKED_MALLOC( h->thread[i]->out.nal, init_nal_count*sizeof(x264_nal_t) );
1306         h->thread[i]->out.i_nals_allocated = init_nal_count;
1307
1308         if( allocate_threadlocal_data && x264_macroblock_cache_allocate( h->thread[i] ) < 0 )
1309             goto fail;
1310     }
1311
1312     if( x264_lookahead_init( h, i_slicetype_length ) )
1313         goto fail;
1314
1315     for( int i = 0; i < h->param.i_threads; i++ )
1316         if( x264_macroblock_thread_allocate( h->thread[i], 0 ) < 0 )
1317             goto fail;
1318
1319     if( x264_ratecontrol_new( h ) < 0 )
1320         goto fail;
1321
1322     if( h->param.i_nal_hrd )
1323     {
1324         x264_log( h, X264_LOG_DEBUG, "HRD bitrate: %i bits/sec\n", h->sps->vui.hrd.i_bit_rate_unscaled );
1325         x264_log( h, X264_LOG_DEBUG, "CPB size: %i bits\n", h->sps->vui.hrd.i_cpb_size_unscaled );
1326     }
1327
1328     if( h->param.psz_dump_yuv )
1329     {
1330         /* create or truncate the reconstructed video file */
1331         FILE *f = fopen( h->param.psz_dump_yuv, "w" );
1332         if( !f )
1333         {
1334             x264_log( h, X264_LOG_ERROR, "dump_yuv: can't write to %s\n", h->param.psz_dump_yuv );
1335             goto fail;
1336         }
1337         else if( !x264_is_regular_file( f ) )
1338         {
1339             x264_log( h, X264_LOG_ERROR, "dump_yuv: incompatible with non-regular file %s\n", h->param.psz_dump_yuv );
1340             goto fail;
1341         }
1342         fclose( f );
1343     }
1344
1345     const char *profile = h->sps->i_profile_idc == PROFILE_BASELINE ? "Constrained Baseline" :
1346                           h->sps->i_profile_idc == PROFILE_MAIN ? "Main" :
1347                           h->sps->i_profile_idc == PROFILE_HIGH ? "High" :
1348                           h->sps->i_profile_idc == PROFILE_HIGH10 ? (h->sps->b_constraint_set3 == 1 ? "High 10 Intra" : "High 10") :
1349                           h->sps->i_profile_idc == PROFILE_HIGH422 ? (h->sps->b_constraint_set3 == 1 ? "High 4:2:2 Intra" : "High 4:2:2") :
1350                           h->sps->b_constraint_set3 == 1 ? "High 4:4:4 Intra" : "High 4:4:4 Predictive";
1351     char level[4];
1352     snprintf( level, sizeof(level), "%d.%d", h->sps->i_level_idc/10, h->sps->i_level_idc%10 );
1353     if( h->sps->i_level_idc == 9 || ( h->sps->i_level_idc == 11 && h->sps->b_constraint_set3 &&
1354         (h->sps->i_profile_idc >= PROFILE_BASELINE && h->sps->i_profile_idc <= PROFILE_EXTENDED) ) )
1355         strcpy( level, "1b" );
1356
1357     if( h->sps->i_profile_idc < PROFILE_HIGH10 )
1358     {
1359         x264_log( h, X264_LOG_INFO, "profile %s, level %s\n",
1360             profile, level );
1361     }
1362     else
1363     {
1364         static const char * const subsampling[4] = { "4:0:0", "4:2:0", "4:2:2", "4:4:4" };
1365         x264_log( h, X264_LOG_INFO, "profile %s, level %s, %s %d-bit\n",
1366             profile, level, subsampling[CHROMA_FORMAT], BIT_DEPTH );
1367     }
1368
1369     return h;
1370 fail:
1371     x264_free( h );
1372     return NULL;
1373 }
1374
1375 /****************************************************************************
1376  * x264_encoder_reconfig:
1377  ****************************************************************************/
1378 int x264_encoder_reconfig( x264_t *h, x264_param_t *param )
1379 {
1380     /* If the previous frame isn't done encoding, reconfiguring is probably dangerous. */
1381     if( h->param.b_sliced_threads )
1382         if( x264_threadpool_wait_all( h ) < 0 )
1383             return -1;
1384
1385     int rc_reconfig = 0;
1386     h = h->thread[h->thread[0]->i_thread_phase];
1387     x264_set_aspect_ratio( h, param, 0 );
1388 #define COPY(var) h->param.var = param->var
1389     COPY( i_frame_reference ); // but never uses more refs than initially specified
1390     COPY( i_bframe_bias );
1391     if( h->param.i_scenecut_threshold )
1392         COPY( i_scenecut_threshold ); // can't turn it on or off, only vary the threshold
1393     COPY( b_deblocking_filter );
1394     COPY( i_deblocking_filter_alphac0 );
1395     COPY( i_deblocking_filter_beta );
1396     COPY( i_frame_packing );
1397     COPY( analyse.inter );
1398     COPY( analyse.intra );
1399     COPY( analyse.i_direct_mv_pred );
1400     /* Scratch buffer prevents me_range from being increased for esa/tesa */
1401     if( h->param.analyse.i_me_method < X264_ME_ESA || param->analyse.i_me_range < h->param.analyse.i_me_range )
1402         COPY( analyse.i_me_range );
1403     COPY( analyse.i_noise_reduction );
1404     /* We can't switch out of subme=0 during encoding. */
1405     if( h->param.analyse.i_subpel_refine )
1406         COPY( analyse.i_subpel_refine );
1407     COPY( analyse.i_trellis );
1408     COPY( analyse.b_chroma_me );
1409     COPY( analyse.b_dct_decimate );
1410     COPY( analyse.b_fast_pskip );
1411     COPY( analyse.b_mixed_references );
1412     COPY( analyse.f_psy_rd );
1413     COPY( analyse.f_psy_trellis );
1414     COPY( crop_rect );
1415     // can only twiddle these if they were enabled to begin with:
1416     if( h->param.analyse.i_me_method >= X264_ME_ESA || param->analyse.i_me_method < X264_ME_ESA )
1417         COPY( analyse.i_me_method );
1418     if( h->param.analyse.i_me_method >= X264_ME_ESA && !h->frames.b_have_sub8x8_esa )
1419         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
1420     if( h->pps->b_transform_8x8_mode )
1421         COPY( analyse.b_transform_8x8 );
1422     if( h->frames.i_max_ref1 > 1 )
1423         COPY( i_bframe_pyramid );
1424     COPY( i_slice_max_size );
1425     COPY( i_slice_max_mbs );
1426     COPY( i_slice_count );
1427     COPY( b_tff );
1428
1429     /* VBV can't be turned on if it wasn't on to begin with */
1430     if( h->param.rc.i_vbv_max_bitrate > 0 && h->param.rc.i_vbv_buffer_size > 0 &&
1431           param->rc.i_vbv_max_bitrate > 0 &&   param->rc.i_vbv_buffer_size > 0 )
1432     {
1433         rc_reconfig |= h->param.rc.i_vbv_max_bitrate != param->rc.i_vbv_max_bitrate;
1434         rc_reconfig |= h->param.rc.i_vbv_buffer_size != param->rc.i_vbv_buffer_size;
1435         rc_reconfig |= h->param.rc.i_bitrate != param->rc.i_bitrate;
1436         COPY( rc.i_vbv_max_bitrate );
1437         COPY( rc.i_vbv_buffer_size );
1438         COPY( rc.i_bitrate );
1439     }
1440     rc_reconfig |= h->param.rc.f_rf_constant != param->rc.f_rf_constant;
1441     rc_reconfig |= h->param.rc.f_rf_constant_max != param->rc.f_rf_constant_max;
1442     COPY( rc.f_rf_constant );
1443     COPY( rc.f_rf_constant_max );
1444 #undef COPY
1445
1446     mbcmp_init( h );
1447
1448     int ret = x264_validate_parameters( h, 0 );
1449
1450     /* Supported reconfiguration options (1-pass only):
1451      * vbv-maxrate
1452      * vbv-bufsize
1453      * crf
1454      * bitrate (CBR only) */
1455     if( !ret && rc_reconfig )
1456         x264_ratecontrol_init_reconfigurable( h, 0 );
1457
1458     return ret;
1459 }
1460
1461 /****************************************************************************
1462  * x264_encoder_parameters:
1463  ****************************************************************************/
1464 void x264_encoder_parameters( x264_t *h, x264_param_t *param )
1465 {
1466     memcpy( param, &h->thread[h->i_thread_phase]->param, sizeof(x264_param_t) );
1467 }
1468
1469 /* internal usage */
1470 static void x264_nal_start( x264_t *h, int i_type, int i_ref_idc )
1471 {
1472     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1473
1474     nal->i_ref_idc        = i_ref_idc;
1475     nal->i_type           = i_type;
1476     nal->b_long_startcode = 1;
1477
1478     nal->i_payload= 0;
1479     nal->p_payload= &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8];
1480 }
1481
1482 /* if number of allocated nals is not enough, re-allocate a larger one. */
1483 static int x264_nal_check_buffer( x264_t *h )
1484 {
1485     if( h->out.i_nal >= h->out.i_nals_allocated )
1486     {
1487         x264_nal_t *new_out = x264_malloc( sizeof(x264_nal_t) * (h->out.i_nals_allocated*2) );
1488         if( !new_out )
1489             return -1;
1490         memcpy( new_out, h->out.nal, sizeof(x264_nal_t) * (h->out.i_nals_allocated) );
1491         x264_free( h->out.nal );
1492         h->out.nal = new_out;
1493         h->out.i_nals_allocated *= 2;
1494     }
1495     return 0;
1496 }
1497
1498 static int x264_nal_end( x264_t *h )
1499 {
1500     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1501     uint8_t *end = &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8];
1502     nal->i_payload = end - nal->p_payload;
1503     /* nal_escape_mmx reads past the end of the input.
1504      * While undefined padding wouldn't actually affect the output, it makes valgrind unhappy. */
1505     memset( end, 0xff, 32 );
1506     if( h->param.nalu_process )
1507         h->param.nalu_process( h, nal );
1508     h->out.i_nal++;
1509
1510     return x264_nal_check_buffer( h );
1511 }
1512
1513 static int x264_encoder_encapsulate_nals( x264_t *h, int start )
1514 {
1515     int nal_size = 0, previous_nal_size = 0;
1516
1517     if( h->param.nalu_process )
1518     {
1519         for( int i = start; i < h->out.i_nal; i++ )
1520             nal_size += h->out.nal[i].i_payload;
1521         return nal_size;
1522     }
1523
1524     for( int i = 0; i < start; i++ )
1525         previous_nal_size += h->out.nal[i].i_payload;
1526
1527     for( int i = start; i < h->out.i_nal; i++ )
1528         nal_size += h->out.nal[i].i_payload;
1529
1530     /* Worst-case NAL unit escaping: reallocate the buffer if it's too small. */
1531     int necessary_size = nal_size * 3/2 + h->out.i_nal * 4;
1532     if( h->nal_buffer_size < necessary_size )
1533     {
1534         h->nal_buffer_size = necessary_size * 2;
1535         uint8_t *buf = x264_malloc( h->nal_buffer_size );
1536         if( !buf )
1537             return -1;
1538         if( previous_nal_size )
1539             memcpy( buf, h->nal_buffer, previous_nal_size );
1540         x264_free( h->nal_buffer );
1541         h->nal_buffer = buf;
1542     }
1543
1544     uint8_t *nal_buffer = h->nal_buffer + previous_nal_size;
1545
1546     for( int i = start; i < h->out.i_nal; i++ )
1547     {
1548         h->out.nal[i].b_long_startcode = !i || h->out.nal[i].i_type == NAL_SPS || h->out.nal[i].i_type == NAL_PPS;
1549         x264_nal_encode( h, nal_buffer, &h->out.nal[i] );
1550         nal_buffer += h->out.nal[i].i_payload;
1551     }
1552
1553     x264_emms();
1554
1555     return nal_buffer - (h->nal_buffer + previous_nal_size);
1556 }
1557
1558 /****************************************************************************
1559  * x264_encoder_headers:
1560  ****************************************************************************/
1561 int x264_encoder_headers( x264_t *h, x264_nal_t **pp_nal, int *pi_nal )
1562 {
1563     int frame_size = 0;
1564     /* init bitstream context */
1565     h->out.i_nal = 0;
1566     bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
1567
1568     /* Write SEI, SPS and PPS. */
1569
1570     /* generate sequence parameters */
1571     x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
1572     x264_sps_write( &h->out.bs, h->sps );
1573     if( x264_nal_end( h ) )
1574         return -1;
1575
1576     /* generate picture parameters */
1577     x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
1578     x264_pps_write( &h->out.bs, h->sps, h->pps );
1579     if( x264_nal_end( h ) )
1580         return -1;
1581
1582     /* identify ourselves */
1583     x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
1584     if( x264_sei_version_write( h, &h->out.bs ) )
1585         return -1;
1586     if( x264_nal_end( h ) )
1587         return -1;
1588
1589     frame_size = x264_encoder_encapsulate_nals( h, 0 );
1590     if( frame_size < 0 )
1591         return -1;
1592
1593     /* now set output*/
1594     *pi_nal = h->out.i_nal;
1595     *pp_nal = &h->out.nal[0];
1596     h->out.i_nal = 0;
1597
1598     return frame_size;
1599 }
1600
1601 /* Check to see whether we have chosen a reference list ordering different
1602  * from the standard's default. */
1603 static inline void x264_reference_check_reorder( x264_t *h )
1604 {
1605     /* The reorder check doesn't check for missing frames, so just
1606      * force a reorder if one of the reference list is corrupt. */
1607     for( int i = 0; h->frames.reference[i]; i++ )
1608         if( h->frames.reference[i]->b_corrupt )
1609         {
1610             h->b_ref_reorder[0] = 1;
1611             return;
1612         }
1613     for( int list = 0; list <= (h->sh.i_type == SLICE_TYPE_B); list++ )
1614         for( int i = 0; i < h->i_ref[list] - 1; i++ )
1615         {
1616             int framenum_diff = h->fref[list][i+1]->i_frame_num - h->fref[list][i]->i_frame_num;
1617             int poc_diff = h->fref[list][i+1]->i_poc - h->fref[list][i]->i_poc;
1618             /* P and B-frames use different default orders. */
1619             if( h->sh.i_type == SLICE_TYPE_P ? framenum_diff > 0 : list == 1 ? poc_diff < 0 : poc_diff > 0 )
1620             {
1621                 h->b_ref_reorder[list] = 1;
1622                 return;
1623             }
1624         }
1625 }
1626
1627 /* return -1 on failure, else return the index of the new reference frame */
1628 int x264_weighted_reference_duplicate( x264_t *h, int i_ref, const x264_weight_t *w )
1629 {
1630     int i = h->i_ref[0];
1631     int j = 1;
1632     x264_frame_t *newframe;
1633     if( i <= 1 ) /* empty list, definitely can't duplicate frame */
1634         return -1;
1635
1636     //Duplication is only used in X264_WEIGHTP_SMART
1637     if( h->param.analyse.i_weighted_pred != X264_WEIGHTP_SMART )
1638         return -1;
1639
1640     /* Duplication is a hack to compensate for crappy rounding in motion compensation.
1641      * With high bit depth, it's not worth doing, so turn it off except in the case of
1642      * unweighted dupes. */
1643     if( BIT_DEPTH > 8 && w != x264_weight_none )
1644         return -1;
1645
1646     newframe = x264_frame_pop_blank_unused( h );
1647     if( !newframe )
1648         return -1;
1649
1650     //FIXME: probably don't need to copy everything
1651     *newframe = *h->fref[0][i_ref];
1652     newframe->i_reference_count = 1;
1653     newframe->orig = h->fref[0][i_ref];
1654     newframe->b_duplicate = 1;
1655     memcpy( h->fenc->weight[j], w, sizeof(h->fenc->weight[i]) );
1656
1657     /* shift the frames to make space for the dupe. */
1658     h->b_ref_reorder[0] = 1;
1659     if( h->i_ref[0] < X264_REF_MAX )
1660         ++h->i_ref[0];
1661     h->fref[0][X264_REF_MAX-1] = NULL;
1662     x264_frame_unshift( &h->fref[0][j], newframe );
1663
1664     return j;
1665 }
1666
1667 static void x264_weighted_pred_init( x264_t *h )
1668 {
1669     /* for now no analysis and set all weights to nothing */
1670     for( int i_ref = 0; i_ref < h->i_ref[0]; i_ref++ )
1671         h->fenc->weighted[i_ref] = h->fref[0][i_ref]->filtered[0][0];
1672
1673     // FIXME: This only supports weighting of one reference frame
1674     // and duplicates of that frame.
1675     h->fenc->i_lines_weighted = 0;
1676
1677     for( int i_ref = 0; i_ref < (h->i_ref[0] << SLICE_MBAFF); i_ref++ )
1678         for( int i = 0; i < 3; i++ )
1679             h->sh.weight[i_ref][i].weightfn = NULL;
1680
1681
1682     if( h->sh.i_type != SLICE_TYPE_P || h->param.analyse.i_weighted_pred <= 0 )
1683         return;
1684
1685     int i_padv = PADV << PARAM_INTERLACED;
1686     int denom = -1;
1687     int weightplane[2] = { 0, 0 };
1688     int buffer_next = 0;
1689     for( int i = 0; i < 3; i++ )
1690     {
1691         for( int j = 0; j < h->i_ref[0]; j++ )
1692         {
1693             if( h->fenc->weight[j][i].weightfn )
1694             {
1695                 h->sh.weight[j][i] = h->fenc->weight[j][i];
1696                 // if weight is useless, don't write it to stream
1697                 if( h->sh.weight[j][i].i_scale == 1<<h->sh.weight[j][i].i_denom && h->sh.weight[j][i].i_offset == 0 )
1698                     h->sh.weight[j][i].weightfn = NULL;
1699                 else
1700                 {
1701                     if( !weightplane[!!i] )
1702                     {
1703                         weightplane[!!i] = 1;
1704                         h->sh.weight[0][!!i].i_denom = denom = h->sh.weight[j][i].i_denom;
1705                         assert( x264_clip3( denom, 0, 7 ) == denom );
1706                     }
1707
1708                     assert( h->sh.weight[j][i].i_denom == denom );
1709                     if( !i )
1710                     {
1711                         h->fenc->weighted[j] = h->mb.p_weight_buf[buffer_next++] + h->fenc->i_stride[0] * i_padv + PADH;
1712                         //scale full resolution frame
1713                         if( h->param.i_threads == 1 )
1714                         {
1715                             pixel *src = h->fref[0][j]->filtered[0][0] - h->fref[0][j]->i_stride[0]*i_padv - PADH;
1716                             pixel *dst = h->fenc->weighted[j] - h->fenc->i_stride[0]*i_padv - PADH;
1717                             int stride = h->fenc->i_stride[0];
1718                             int width = h->fenc->i_width[0] + PADH*2;
1719                             int height = h->fenc->i_lines[0] + i_padv*2;
1720                             x264_weight_scale_plane( h, dst, stride, src, stride, width, height, &h->sh.weight[j][0] );
1721                             h->fenc->i_lines_weighted = height;
1722                         }
1723                     }
1724                 }
1725             }
1726         }
1727     }
1728
1729     if( weightplane[1] )
1730         for( int i = 0; i < h->i_ref[0]; i++ )
1731         {
1732             if( h->sh.weight[i][1].weightfn && !h->sh.weight[i][2].weightfn )
1733             {
1734                 h->sh.weight[i][2].i_scale = 1 << h->sh.weight[0][1].i_denom;
1735                 h->sh.weight[i][2].i_offset = 0;
1736             }
1737             else if( h->sh.weight[i][2].weightfn && !h->sh.weight[i][1].weightfn )
1738             {
1739                 h->sh.weight[i][1].i_scale = 1 << h->sh.weight[0][1].i_denom;
1740                 h->sh.weight[i][1].i_offset = 0;
1741             }
1742         }
1743
1744     if( !weightplane[0] )
1745         h->sh.weight[0][0].i_denom = 0;
1746     if( !weightplane[1] )
1747         h->sh.weight[0][1].i_denom = 0;
1748     h->sh.weight[0][2].i_denom = h->sh.weight[0][1].i_denom;
1749 }
1750
1751 static inline int x264_reference_distance( x264_t *h, x264_frame_t *frame )
1752 {
1753     if( h->param.i_frame_packing == 5 )
1754         return abs((h->fenc->i_frame&~1) - (frame->i_frame&~1)) +
1755                   ((h->fenc->i_frame&1) != (frame->i_frame&1));
1756     else
1757         return abs(h->fenc->i_frame - frame->i_frame);
1758 }
1759
1760 static inline void x264_reference_build_list( x264_t *h, int i_poc )
1761 {
1762     int b_ok;
1763
1764     /* build ref list 0/1 */
1765     h->mb.pic.i_fref[0] = h->i_ref[0] = 0;
1766     h->mb.pic.i_fref[1] = h->i_ref[1] = 0;
1767     if( h->sh.i_type == SLICE_TYPE_I )
1768         return;
1769
1770     for( int i = 0; h->frames.reference[i]; i++ )
1771     {
1772         if( h->frames.reference[i]->b_corrupt )
1773             continue;
1774         if( h->frames.reference[i]->i_poc < i_poc )
1775             h->fref[0][h->i_ref[0]++] = h->frames.reference[i];
1776         else if( h->frames.reference[i]->i_poc > i_poc )
1777             h->fref[1][h->i_ref[1]++] = h->frames.reference[i];
1778     }
1779
1780     /* Order reference lists by distance from the current frame. */
1781     for( int list = 0; list < 2; list++ )
1782     {
1783         h->fref_nearest[list] = h->fref[list][0];
1784         do
1785         {
1786             b_ok = 1;
1787             for( int i = 0; i < h->i_ref[list] - 1; i++ )
1788             {
1789                 if( list ? h->fref[list][i+1]->i_poc < h->fref_nearest[list]->i_poc
1790                          : h->fref[list][i+1]->i_poc > h->fref_nearest[list]->i_poc )
1791                     h->fref_nearest[list] = h->fref[list][i+1];
1792                 if( x264_reference_distance( h, h->fref[list][i] ) > x264_reference_distance( h, h->fref[list][i+1] ) )
1793                 {
1794                     XCHG( x264_frame_t*, h->fref[list][i], h->fref[list][i+1] );
1795                     b_ok = 0;
1796                     break;
1797                 }
1798             }
1799         } while( !b_ok );
1800     }
1801
1802     if( h->sh.i_mmco_remove_from_end )
1803         for( int i = h->i_ref[0]-1; i >= h->i_ref[0] - h->sh.i_mmco_remove_from_end; i-- )
1804         {
1805             int diff = h->i_frame_num - h->fref[0][i]->i_frame_num;
1806             h->sh.mmco[h->sh.i_mmco_command_count].i_poc = h->fref[0][i]->i_poc;
1807             h->sh.mmco[h->sh.i_mmco_command_count++].i_difference_of_pic_nums = diff;
1808         }
1809
1810     x264_reference_check_reorder( h );
1811
1812     h->i_ref[1] = X264_MIN( h->i_ref[1], h->frames.i_max_ref1 );
1813     h->i_ref[0] = X264_MIN( h->i_ref[0], h->frames.i_max_ref0 );
1814     h->i_ref[0] = X264_MIN( h->i_ref[0], h->param.i_frame_reference ); // if reconfig() has lowered the limit
1815
1816     /* For Blu-ray compliance, don't reference frames outside of the minigop. */
1817     if( IS_X264_TYPE_B( h->fenc->i_type ) && h->param.b_bluray_compat )
1818         h->i_ref[0] = X264_MIN( h->i_ref[0], IS_X264_TYPE_B( h->fref[0][0]->i_type ) + 1 );
1819
1820     /* add duplicates */
1821     if( h->fenc->i_type == X264_TYPE_P )
1822     {
1823         int idx = -1;
1824         if( h->param.analyse.i_weighted_pred >= X264_WEIGHTP_SIMPLE )
1825         {
1826             x264_weight_t w[3];
1827             w[1].weightfn = w[2].weightfn = NULL;
1828             if( h->param.rc.b_stat_read )
1829                 x264_ratecontrol_set_weights( h, h->fenc );
1830
1831             if( !h->fenc->weight[0][0].weightfn )
1832             {
1833                 h->fenc->weight[0][0].i_denom = 0;
1834                 SET_WEIGHT( w[0], 1, 1, 0, -1 );
1835                 idx = x264_weighted_reference_duplicate( h, 0, w );
1836             }
1837             else
1838             {
1839                 if( h->fenc->weight[0][0].i_scale == 1<<h->fenc->weight[0][0].i_denom )
1840                 {
1841                     SET_WEIGHT( h->fenc->weight[0][0], 1, 1, 0, h->fenc->weight[0][0].i_offset );
1842                 }
1843                 x264_weighted_reference_duplicate( h, 0, x264_weight_none );
1844                 if( h->fenc->weight[0][0].i_offset > -128 )
1845                 {
1846                     w[0] = h->fenc->weight[0][0];
1847                     w[0].i_offset--;
1848                     h->mc.weight_cache( h, &w[0] );
1849                     idx = x264_weighted_reference_duplicate( h, 0, w );
1850                 }
1851             }
1852         }
1853         h->mb.ref_blind_dupe = idx;
1854     }
1855
1856     assert( h->i_ref[0] + h->i_ref[1] <= X264_REF_MAX );
1857     h->mb.pic.i_fref[0] = h->i_ref[0];
1858     h->mb.pic.i_fref[1] = h->i_ref[1];
1859 }
1860
1861 static void x264_fdec_filter_row( x264_t *h, int mb_y, int pass )
1862 {
1863     /* mb_y is the mb to be encoded next, not the mb to be filtered here */
1864     int b_hpel = h->fdec->b_kept_as_ref;
1865     int b_deblock = h->sh.i_disable_deblocking_filter_idc != 1;
1866     int b_end = mb_y == h->i_threadslice_end;
1867     int b_measure_quality = 1;
1868     int min_y = mb_y - (1 << SLICE_MBAFF);
1869     int b_start = min_y == h->i_threadslice_start;
1870     /* Even in interlaced mode, deblocking never modifies more than 4 pixels
1871      * above each MB, as bS=4 doesn't happen for the top of interlaced mbpairs. */
1872     int minpix_y = min_y*16 - 4 * !b_start;
1873     int maxpix_y = mb_y*16 - 4 * !b_end;
1874     b_deblock &= b_hpel || h->param.b_full_recon || h->param.psz_dump_yuv;
1875     if( h->param.b_sliced_threads )
1876     {
1877         switch( pass )
1878         {
1879             /* During encode: only do deblock if asked for */
1880             default:
1881             case 0:
1882                 b_deblock &= h->param.b_full_recon;
1883                 b_hpel = 0;
1884                 break;
1885             /* During post-encode pass: do deblock if not done yet, do hpel for all
1886              * rows except those between slices. */
1887             case 1:
1888                 b_deblock &= !h->param.b_full_recon;
1889                 b_hpel &= !(b_start && min_y > 0);
1890                 b_measure_quality = 0;
1891                 break;
1892             /* Final pass: do the rows between slices in sequence. */
1893             case 2:
1894                 b_deblock = 0;
1895                 b_measure_quality = 0;
1896                 break;
1897         }
1898     }
1899     if( mb_y & SLICE_MBAFF )
1900         return;
1901     if( min_y < h->i_threadslice_start )
1902         return;
1903
1904     if( b_deblock )
1905         for( int y = min_y; y < mb_y; y += (1 << SLICE_MBAFF) )
1906             x264_frame_deblock_row( h, y );
1907
1908     /* FIXME: Prediction requires different borders for interlaced/progressive mc,
1909      * but the actual image data is equivalent. For now, maintain this
1910      * consistency by copying deblocked pixels between planes. */
1911     if( PARAM_INTERLACED && (!h->param.b_sliced_threads || pass == 1) )
1912         for( int p = 0; p < h->fdec->i_plane; p++ )
1913             for( int i = minpix_y>>(CHROMA_V_SHIFT && p); i < maxpix_y>>(CHROMA_V_SHIFT && p); i++ )
1914                 memcpy( h->fdec->plane_fld[p] + i*h->fdec->i_stride[p],
1915                         h->fdec->plane[p]     + i*h->fdec->i_stride[p],
1916                         h->mb.i_mb_width*16*sizeof(pixel) );
1917
1918     if( h->fdec->b_kept_as_ref && (!h->param.b_sliced_threads || pass == 1) )
1919         x264_frame_expand_border( h, h->fdec, min_y );
1920     if( b_hpel )
1921     {
1922         int end = mb_y == h->mb.i_mb_height;
1923         /* Can't do hpel until the previous slice is done encoding. */
1924         if( h->param.analyse.i_subpel_refine )
1925         {
1926             x264_frame_filter( h, h->fdec, min_y, end );
1927             x264_frame_expand_border_filtered( h, h->fdec, min_y, end );
1928         }
1929     }
1930
1931     if( SLICE_MBAFF && pass == 0 )
1932         for( int i = 0; i < 3; i++ )
1933         {
1934             XCHG( pixel *, h->intra_border_backup[0][i], h->intra_border_backup[3][i] );
1935             XCHG( pixel *, h->intra_border_backup[1][i], h->intra_border_backup[4][i] );
1936         }
1937
1938     if( h->i_thread_frames > 1 && h->fdec->b_kept_as_ref )
1939         x264_frame_cond_broadcast( h->fdec, mb_y*16 + (b_end ? 10000 : -(X264_THREAD_HEIGHT << SLICE_MBAFF)) );
1940
1941     if( b_measure_quality )
1942     {
1943         maxpix_y = X264_MIN( maxpix_y, h->param.i_height );
1944         if( h->param.analyse.b_psnr )
1945         {
1946             for( int p = 0; p < (CHROMA444 ? 3 : 1); p++ )
1947                 h->stat.frame.i_ssd[p] += x264_pixel_ssd_wxh( &h->pixf,
1948                     h->fdec->plane[p] + minpix_y * h->fdec->i_stride[p], h->fdec->i_stride[p],
1949                     h->fenc->plane[p] + minpix_y * h->fenc->i_stride[p], h->fenc->i_stride[p],
1950                     h->param.i_width, maxpix_y-minpix_y );
1951             if( !CHROMA444 )
1952             {
1953                 uint64_t ssd_u, ssd_v;
1954                 int v_shift = CHROMA_V_SHIFT;
1955                 x264_pixel_ssd_nv12( &h->pixf,
1956                     h->fdec->plane[1] + (minpix_y>>v_shift) * h->fdec->i_stride[1], h->fdec->i_stride[1],
1957                     h->fenc->plane[1] + (minpix_y>>v_shift) * h->fenc->i_stride[1], h->fenc->i_stride[1],
1958                     h->param.i_width>>1, (maxpix_y-minpix_y)>>v_shift, &ssd_u, &ssd_v );
1959                 h->stat.frame.i_ssd[1] += ssd_u;
1960                 h->stat.frame.i_ssd[2] += ssd_v;
1961             }
1962         }
1963
1964         if( h->param.analyse.b_ssim )
1965         {
1966             int ssim_cnt;
1967             x264_emms();
1968             /* offset by 2 pixels to avoid alignment of ssim blocks with dct blocks,
1969              * and overlap by 4 */
1970             minpix_y += b_start ? 2 : -6;
1971             h->stat.frame.f_ssim +=
1972                 x264_pixel_ssim_wxh( &h->pixf,
1973                     h->fdec->plane[0] + 2+minpix_y*h->fdec->i_stride[0], h->fdec->i_stride[0],
1974                     h->fenc->plane[0] + 2+minpix_y*h->fenc->i_stride[0], h->fenc->i_stride[0],
1975                     h->param.i_width-2, maxpix_y-minpix_y, h->scratch_buffer, &ssim_cnt );
1976             h->stat.frame.i_ssim_cnt += ssim_cnt;
1977         }
1978     }
1979 }
1980
1981 static inline int x264_reference_update( x264_t *h )
1982 {
1983     if( !h->fdec->b_kept_as_ref )
1984     {
1985         if( h->i_thread_frames > 1 )
1986         {
1987             x264_frame_push_unused( h, h->fdec );
1988             h->fdec = x264_frame_pop_unused( h, 1 );
1989             if( !h->fdec )
1990                 return -1;
1991         }
1992         return 0;
1993     }
1994
1995     /* apply mmco from previous frame. */
1996     for( int i = 0; i < h->sh.i_mmco_command_count; i++ )
1997         for( int j = 0; h->frames.reference[j]; j++ )
1998             if( h->frames.reference[j]->i_poc == h->sh.mmco[i].i_poc )
1999                 x264_frame_push_unused( h, x264_frame_shift( &h->frames.reference[j] ) );
2000
2001     /* move frame in the buffer */
2002     x264_frame_push( h->frames.reference, h->fdec );
2003     if( h->frames.reference[h->sps->i_num_ref_frames] )
2004         x264_frame_push_unused( h, x264_frame_shift( h->frames.reference ) );
2005     h->fdec = x264_frame_pop_unused( h, 1 );
2006     if( !h->fdec )
2007         return -1;
2008     return 0;
2009 }
2010
2011 static inline void x264_reference_reset( x264_t *h )
2012 {
2013     while( h->frames.reference[0] )
2014         x264_frame_push_unused( h, x264_frame_pop( h->frames.reference ) );
2015     h->fdec->i_poc =
2016     h->fenc->i_poc = 0;
2017 }
2018
2019 static inline void x264_reference_hierarchy_reset( x264_t *h )
2020 {
2021     int ref;
2022     int b_hasdelayframe = 0;
2023
2024     /* look for delay frames -- chain must only contain frames that are disposable */
2025     for( int i = 0; h->frames.current[i] && IS_DISPOSABLE( h->frames.current[i]->i_type ); i++ )
2026         b_hasdelayframe |= h->frames.current[i]->i_coded
2027                         != h->frames.current[i]->i_frame + h->sps->vui.i_num_reorder_frames;
2028
2029     /* This function must handle b-pyramid and clear frames for open-gop */
2030     if( h->param.i_bframe_pyramid != X264_B_PYRAMID_STRICT && !b_hasdelayframe && h->frames.i_poc_last_open_gop == -1 )
2031         return;
2032
2033     /* Remove last BREF. There will never be old BREFs in the
2034      * dpb during a BREF decode when pyramid == STRICT */
2035     for( ref = 0; h->frames.reference[ref]; ref++ )
2036     {
2037         if( ( h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT
2038             && h->frames.reference[ref]->i_type == X264_TYPE_BREF )
2039             || ( h->frames.reference[ref]->i_poc < h->frames.i_poc_last_open_gop
2040             && h->sh.i_type != SLICE_TYPE_B ) )
2041         {
2042             int diff = h->i_frame_num - h->frames.reference[ref]->i_frame_num;
2043             h->sh.mmco[h->sh.i_mmco_command_count].i_difference_of_pic_nums = diff;
2044             h->sh.mmco[h->sh.i_mmco_command_count++].i_poc = h->frames.reference[ref]->i_poc;
2045             x264_frame_push_unused( h, x264_frame_shift( &h->frames.reference[ref] ) );
2046             h->b_ref_reorder[0] = 1;
2047             ref--;
2048         }
2049     }
2050
2051     /* Prepare room in the dpb for the delayed display time of the later b-frame's */
2052     if( h->param.i_bframe_pyramid )
2053         h->sh.i_mmco_remove_from_end = X264_MAX( ref + 2 - h->frames.i_max_dpb, 0 );
2054 }
2055
2056 static inline void x264_slice_init( x264_t *h, int i_nal_type, int i_global_qp )
2057 {
2058     /* ------------------------ Create slice header  ----------------------- */
2059     if( i_nal_type == NAL_SLICE_IDR )
2060     {
2061         x264_slice_header_init( h, &h->sh, h->sps, h->pps, h->i_idr_pic_id, h->i_frame_num, i_global_qp );
2062
2063         /* alternate id */
2064         h->i_idr_pic_id ^= 1;
2065     }
2066     else
2067     {
2068         x264_slice_header_init( h, &h->sh, h->sps, h->pps, -1, h->i_frame_num, i_global_qp );
2069
2070         h->sh.i_num_ref_idx_l0_active = h->i_ref[0] <= 0 ? 1 : h->i_ref[0];
2071         h->sh.i_num_ref_idx_l1_active = h->i_ref[1] <= 0 ? 1 : h->i_ref[1];
2072         if( h->sh.i_num_ref_idx_l0_active != h->pps->i_num_ref_idx_l0_default_active ||
2073             (h->sh.i_type == SLICE_TYPE_B && h->sh.i_num_ref_idx_l1_active != h->pps->i_num_ref_idx_l1_default_active) )
2074         {
2075             h->sh.b_num_ref_idx_override = 1;
2076         }
2077     }
2078
2079     if( h->fenc->i_type == X264_TYPE_BREF && h->param.b_bluray_compat && h->sh.i_mmco_command_count )
2080     {
2081         h->b_sh_backup = 1;
2082         h->sh_backup = h->sh;
2083     }
2084
2085     h->fdec->i_frame_num = h->sh.i_frame_num;
2086
2087     if( h->sps->i_poc_type == 0 )
2088     {
2089         h->sh.i_poc = h->fdec->i_poc;
2090         if( PARAM_INTERLACED )
2091         {
2092             h->sh.i_delta_poc_bottom = h->param.b_tff ? 1 : -1;
2093             h->sh.i_poc += h->sh.i_delta_poc_bottom == -1;
2094         }
2095         else
2096             h->sh.i_delta_poc_bottom = 0;
2097         h->fdec->i_delta_poc[0] = h->sh.i_delta_poc_bottom == -1;
2098         h->fdec->i_delta_poc[1] = h->sh.i_delta_poc_bottom ==  1;
2099     }
2100     else
2101     {
2102         /* Nothing to do ? */
2103     }
2104
2105     x264_macroblock_slice_init( h );
2106 }
2107
2108 typedef struct
2109 {
2110     int skip;
2111     uint8_t cabac_prevbyte;
2112     bs_t bs;
2113     x264_cabac_t cabac;
2114     x264_frame_stat_t stat;
2115     int last_qp;
2116     int last_dqp;
2117     int field_decoding_flag;
2118 } x264_bs_bak_t;
2119
2120 static ALWAYS_INLINE void x264_bitstream_backup( x264_t *h, x264_bs_bak_t *bak, int i_skip, int full )
2121 {
2122     if( full )
2123     {
2124         bak->stat = h->stat.frame;
2125         bak->last_qp = h->mb.i_last_qp;
2126         bak->last_dqp = h->mb.i_last_dqp;
2127         bak->field_decoding_flag = h->mb.field_decoding_flag;
2128     }
2129     else
2130     {
2131         bak->stat.i_mv_bits = h->stat.frame.i_mv_bits;
2132         bak->stat.i_tex_bits = h->stat.frame.i_tex_bits;
2133     }
2134     /* In the per-MB backup, we don't need the contexts because flushing the CABAC
2135      * encoder has no context dependency and in this case, a slice is ended (and
2136      * thus the content of all contexts are thrown away). */
2137     if( h->param.b_cabac )
2138     {
2139         if( full )
2140             memcpy( &bak->cabac, &h->cabac, sizeof(x264_cabac_t) );
2141         else
2142             memcpy( &bak->cabac, &h->cabac, offsetof(x264_cabac_t, f8_bits_encoded) );
2143         /* x264's CABAC writer modifies the previous byte during carry, so it has to be
2144          * backed up. */
2145         bak->cabac_prevbyte = h->cabac.p[-1];
2146     }
2147     else
2148     {
2149         bak->bs = h->out.bs;
2150         bak->skip = i_skip;
2151     }
2152 }
2153
2154 static ALWAYS_INLINE void x264_bitstream_restore( x264_t *h, x264_bs_bak_t *bak, int *skip, int full )
2155 {
2156     if( full )
2157     {
2158         h->stat.frame = bak->stat;
2159         h->mb.i_last_qp = bak->last_qp;
2160         h->mb.i_last_dqp = bak->last_dqp;
2161         h->mb.field_decoding_flag = bak->field_decoding_flag;
2162     }
2163     else
2164     {
2165         h->stat.frame.i_mv_bits = bak->stat.i_mv_bits;
2166         h->stat.frame.i_tex_bits = bak->stat.i_tex_bits;
2167     }
2168     if( h->param.b_cabac )
2169     {
2170         if( full )
2171             memcpy( &h->cabac, &bak->cabac, sizeof(x264_cabac_t) );
2172         else
2173             memcpy( &h->cabac, &bak->cabac, offsetof(x264_cabac_t, f8_bits_encoded) );
2174         h->cabac.p[-1] = bak->cabac_prevbyte;
2175     }
2176     else
2177     {
2178         h->out.bs = bak->bs;
2179         *skip = bak->skip;
2180     }
2181 }
2182
2183 static int x264_slice_write( x264_t *h )
2184 {
2185     int i_skip;
2186     int mb_xy, i_mb_x, i_mb_y;
2187     /* NALUs other than the first use a 3-byte startcode.
2188      * Add one extra byte for the rbsp, and one more for the final CABAC putbyte.
2189      * Then add an extra 5 bytes just in case, to account for random NAL escapes and
2190      * other inaccuracies. */
2191     int overhead_guess = (NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal)) + 1 + h->param.b_cabac + 5;
2192     int slice_max_size = h->param.i_slice_max_size > 0 ? (h->param.i_slice_max_size-overhead_guess)*8 : 0;
2193     int back_up_bitstream = slice_max_size || (!h->param.b_cabac && h->sps->i_profile_idc < PROFILE_HIGH);
2194     int starting_bits = bs_pos(&h->out.bs);
2195     int b_deblock = h->sh.i_disable_deblocking_filter_idc != 1;
2196     int b_hpel = h->fdec->b_kept_as_ref;
2197     int orig_last_mb = h->sh.i_last_mb;
2198     uint8_t *last_emu_check;
2199     x264_bs_bak_t bs_bak[2];
2200     b_deblock &= b_hpel || h->param.b_full_recon || h->param.psz_dump_yuv;
2201     bs_realign( &h->out.bs );
2202
2203     /* Slice */
2204     x264_nal_start( h, h->i_nal_type, h->i_nal_ref_idc );
2205     h->out.nal[h->out.i_nal].i_first_mb = h->sh.i_first_mb;
2206
2207     /* Slice header */
2208     x264_macroblock_thread_init( h );
2209
2210     /* If this isn't the first slice in the threadslice, set the slice QP
2211      * equal to the last QP in the previous slice for more accurate
2212      * CABAC initialization. */
2213     if( h->sh.i_first_mb != h->i_threadslice_start * h->mb.i_mb_width )
2214     {
2215         h->sh.i_qp = h->mb.i_last_qp;
2216         h->sh.i_qp_delta = h->sh.i_qp - h->pps->i_pic_init_qp;
2217     }
2218
2219     x264_slice_header_write( &h->out.bs, &h->sh, h->i_nal_ref_idc );
2220     if( h->param.b_cabac )
2221     {
2222         /* alignment needed */
2223         bs_align_1( &h->out.bs );
2224
2225         /* init cabac */
2226         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 );
2227         x264_cabac_encode_init ( &h->cabac, h->out.bs.p, h->out.bs.p_end );
2228         last_emu_check = h->cabac.p;
2229     }
2230     else
2231         last_emu_check = h->out.bs.p;
2232     h->mb.i_last_qp = h->sh.i_qp;
2233     h->mb.i_last_dqp = 0;
2234     h->mb.field_decoding_flag = 0;
2235
2236     i_mb_y = h->sh.i_first_mb / h->mb.i_mb_width;
2237     i_mb_x = h->sh.i_first_mb % h->mb.i_mb_width;
2238     i_skip = 0;
2239
2240     while( 1 )
2241     {
2242         mb_xy = i_mb_x + i_mb_y * h->mb.i_mb_width;
2243         int mb_spos = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
2244
2245         if( i_mb_x == 0 )
2246         {
2247             if( x264_bitstream_check_buffer( h ) )
2248                 return -1;
2249             if( !(i_mb_y & SLICE_MBAFF) && h->param.rc.i_vbv_buffer_size )
2250                 x264_bitstream_backup( h, &bs_bak[1], i_skip, 1 );
2251             if( !h->mb.b_reencode_mb )
2252                 x264_fdec_filter_row( h, i_mb_y, 0 );
2253         }
2254
2255         if( !(i_mb_y & SLICE_MBAFF) && back_up_bitstream )
2256             x264_bitstream_backup( h, &bs_bak[0], i_skip, 0 );
2257
2258         if( PARAM_INTERLACED )
2259         {
2260             if( h->mb.b_adaptive_mbaff )
2261             {
2262                 if( !(i_mb_y&1) )
2263                 {
2264                     /* FIXME: VSAD is fast but fairly poor at choosing the best interlace type. */
2265                     h->mb.b_interlaced = x264_field_vsad( h, i_mb_x, i_mb_y );
2266                     memcpy( &h->zigzagf, MB_INTERLACED ? &h->zigzagf_interlaced : &h->zigzagf_progressive, sizeof(h->zigzagf) );
2267                     if( !MB_INTERLACED && (i_mb_y+2) == h->mb.i_mb_height )
2268                         x264_expand_border_mbpair( h, i_mb_x, i_mb_y );
2269                 }
2270             }
2271             h->mb.field[mb_xy] = MB_INTERLACED;
2272         }
2273
2274         /* load cache */
2275         if( SLICE_MBAFF )
2276             x264_macroblock_cache_load_interlaced( h, i_mb_x, i_mb_y );
2277         else
2278             x264_macroblock_cache_load_progressive( h, i_mb_x, i_mb_y );
2279
2280         x264_macroblock_analyse( h );
2281
2282         /* encode this macroblock -> be careful it can change the mb type to P_SKIP if needed */
2283 reencode:
2284         x264_macroblock_encode( h );
2285
2286         if( h->param.b_cabac )
2287         {
2288             if( mb_xy > h->sh.i_first_mb && !(SLICE_MBAFF && (i_mb_y&1)) )
2289                 x264_cabac_encode_terminal( &h->cabac );
2290
2291             if( IS_SKIP( h->mb.i_type ) )
2292                 x264_cabac_mb_skip( h, 1 );
2293             else
2294             {
2295                 if( h->sh.i_type != SLICE_TYPE_I )
2296                     x264_cabac_mb_skip( h, 0 );
2297                 x264_macroblock_write_cabac( h, &h->cabac );
2298             }
2299         }
2300         else
2301         {
2302             if( IS_SKIP( h->mb.i_type ) )
2303                 i_skip++;
2304             else
2305             {
2306                 if( h->sh.i_type != SLICE_TYPE_I )
2307                 {
2308                     bs_write_ue( &h->out.bs, i_skip );  /* skip run */
2309                     i_skip = 0;
2310                 }
2311                 x264_macroblock_write_cavlc( h );
2312                 /* If there was a CAVLC level code overflow, try again at a higher QP. */
2313                 if( h->mb.b_overflow )
2314                 {
2315                     h->mb.i_chroma_qp = h->chroma_qp_table[++h->mb.i_qp];
2316                     h->mb.i_skip_intra = 0;
2317                     h->mb.b_skip_mc = 0;
2318                     h->mb.b_overflow = 0;
2319                     x264_bitstream_restore( h, &bs_bak[0], &i_skip, 0 );
2320                     goto reencode;
2321                 }
2322             }
2323         }
2324
2325         int total_bits = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
2326         int mb_size = total_bits - mb_spos;
2327
2328         if( slice_max_size && (!SLICE_MBAFF || (i_mb_y&1)) )
2329         {
2330             /* Count the skip run, just in case. */
2331             if( !h->param.b_cabac )
2332                 total_bits += bs_size_ue_big( i_skip );
2333             /* Check for escape bytes. */
2334             uint8_t *end = h->param.b_cabac ? h->cabac.p : h->out.bs.p;
2335             for( ; last_emu_check < end - 2; last_emu_check++ )
2336                 if( last_emu_check[0] == 0 && last_emu_check[1] == 0 && last_emu_check[2] <= 3 )
2337                 {
2338                     slice_max_size -= 8;
2339                     last_emu_check++;
2340                 }
2341             /* We'll just re-encode this last macroblock if we go over the max slice size. */
2342             if( total_bits - starting_bits > slice_max_size && !h->mb.b_reencode_mb )
2343             {
2344                 if( mb_xy-SLICE_MBAFF*h->mb.i_mb_stride != h->sh.i_first_mb )
2345                 {
2346                     x264_bitstream_restore( h, &bs_bak[0], &i_skip, 0 );
2347                     h->mb.b_reencode_mb = 1;
2348                     if( SLICE_MBAFF )
2349                     {
2350                         // set to bottom of previous mbpair
2351                         if( i_mb_x )
2352                             h->sh.i_last_mb = mb_xy-1+h->mb.i_mb_stride*(!(i_mb_y&1));
2353                         else
2354                             h->sh.i_last_mb = (i_mb_y-2+!(i_mb_y&1))*h->mb.i_mb_stride + h->mb.i_mb_width - 1;
2355                     }
2356                     else
2357                         h->sh.i_last_mb = mb_xy-1;
2358                     break;
2359                 }
2360                 else
2361                     h->sh.i_last_mb = mb_xy;
2362             }
2363         }
2364         h->mb.b_reencode_mb = 0;
2365
2366 #if HAVE_VISUALIZE
2367         if( h->param.b_visualize )
2368             x264_visualize_mb( h );
2369 #endif
2370
2371         /* save cache */
2372         x264_macroblock_cache_save( h );
2373
2374         if( x264_ratecontrol_mb( h, mb_size ) < 0 )
2375         {
2376             x264_bitstream_restore( h, &bs_bak[1], &i_skip, 1 );
2377             h->mb.b_reencode_mb = 1;
2378             i_mb_x = 0;
2379             i_mb_y = i_mb_y - SLICE_MBAFF;
2380             h->mb.i_mb_prev_xy = i_mb_y * h->mb.i_mb_stride - 1;
2381             h->sh.i_last_mb = orig_last_mb;
2382             continue;
2383         }
2384
2385         /* accumulate mb stats */
2386         h->stat.frame.i_mb_count[h->mb.i_type]++;
2387
2388         int b_intra = IS_INTRA( h->mb.i_type );
2389         int b_skip = IS_SKIP( h->mb.i_type );
2390         if( h->param.i_log_level >= X264_LOG_INFO || h->param.rc.b_stat_write )
2391         {
2392             if( !b_intra && !b_skip && !IS_DIRECT( h->mb.i_type ) )
2393             {
2394                 if( h->mb.i_partition != D_8x8 )
2395                         h->stat.frame.i_mb_partition[h->mb.i_partition] += 4;
2396                     else
2397                         for( int i = 0; i < 4; i++ )
2398                             h->stat.frame.i_mb_partition[h->mb.i_sub_partition[i]] ++;
2399                 if( h->param.i_frame_reference > 1 )
2400                     for( int i_list = 0; i_list <= (h->sh.i_type == SLICE_TYPE_B); i_list++ )
2401                         for( int i = 0; i < 4; i++ )
2402                         {
2403                             int i_ref = h->mb.cache.ref[i_list][ x264_scan8[4*i] ];
2404                             if( i_ref >= 0 )
2405                                 h->stat.frame.i_mb_count_ref[i_list][i_ref] ++;
2406                         }
2407             }
2408         }
2409
2410         if( h->param.i_log_level >= X264_LOG_INFO )
2411         {
2412             if( h->mb.i_cbp_luma | h->mb.i_cbp_chroma )
2413             {
2414                 if( CHROMA444 )
2415                 {
2416                     for( int i = 0; i < 4; i++ )
2417                         if( h->mb.i_cbp_luma & (1 << i) )
2418                             for( int p = 0; p < 3; p++ )
2419                             {
2420                                 int s8 = i*4+p*16;
2421                                 int nnz8x8 = M16( &h->mb.cache.non_zero_count[x264_scan8[s8]+0] )
2422                                            | M16( &h->mb.cache.non_zero_count[x264_scan8[s8]+8] );
2423                                 h->stat.frame.i_mb_cbp[!b_intra + p*2] += !!nnz8x8;
2424                             }
2425                 }
2426                 else
2427                 {
2428                     int cbpsum = (h->mb.i_cbp_luma&1) + ((h->mb.i_cbp_luma>>1)&1)
2429                                + ((h->mb.i_cbp_luma>>2)&1) + (h->mb.i_cbp_luma>>3);
2430                     h->stat.frame.i_mb_cbp[!b_intra + 0] += cbpsum;
2431                     h->stat.frame.i_mb_cbp[!b_intra + 2] += !!h->mb.i_cbp_chroma;
2432                     h->stat.frame.i_mb_cbp[!b_intra + 4] += h->mb.i_cbp_chroma >> 1;
2433                 }
2434             }
2435             if( h->mb.i_cbp_luma && !b_intra )
2436             {
2437                 h->stat.frame.i_mb_count_8x8dct[0] ++;
2438                 h->stat.frame.i_mb_count_8x8dct[1] += h->mb.b_transform_8x8;
2439             }
2440             if( b_intra && h->mb.i_type != I_PCM )
2441             {
2442                 if( h->mb.i_type == I_16x16 )
2443                     h->stat.frame.i_mb_pred_mode[0][h->mb.i_intra16x16_pred_mode]++;
2444                 else if( h->mb.i_type == I_8x8 )
2445                     for( int i = 0; i < 16; i += 4 )
2446                         h->stat.frame.i_mb_pred_mode[1][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
2447                 else //if( h->mb.i_type == I_4x4 )
2448                     for( int i = 0; i < 16; i++ )
2449                         h->stat.frame.i_mb_pred_mode[2][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
2450                 h->stat.frame.i_mb_pred_mode[3][x264_mb_chroma_pred_mode_fix[h->mb.i_chroma_pred_mode]]++;
2451             }
2452             h->stat.frame.i_mb_field[b_intra?0:b_skip?2:1] += MB_INTERLACED;
2453         }
2454
2455         /* calculate deblock strength values (actual deblocking is done per-row along with hpel) */
2456         if( b_deblock )
2457             x264_macroblock_deblock_strength( h );
2458
2459         if( mb_xy == h->sh.i_last_mb )
2460             break;
2461
2462         if( SLICE_MBAFF )
2463         {
2464             i_mb_x += i_mb_y & 1;
2465             i_mb_y ^= i_mb_x < h->mb.i_mb_width;
2466         }
2467         else
2468             i_mb_x++;
2469         if( i_mb_x == h->mb.i_mb_width )
2470         {
2471             i_mb_y++;
2472             i_mb_x = 0;
2473         }
2474     }
2475     h->out.nal[h->out.i_nal].i_last_mb = h->sh.i_last_mb;
2476
2477     if( h->param.b_cabac )
2478     {
2479         x264_cabac_encode_flush( h, &h->cabac );
2480         h->out.bs.p = h->cabac.p;
2481     }
2482     else
2483     {
2484         if( i_skip > 0 )
2485             bs_write_ue( &h->out.bs, i_skip );  /* last skip run */
2486         /* rbsp_slice_trailing_bits */
2487         bs_rbsp_trailing( &h->out.bs );
2488         bs_flush( &h->out.bs );
2489     }
2490     if( x264_nal_end( h ) )
2491         return -1;
2492
2493     if( h->sh.i_last_mb == (h->i_threadslice_end * h->mb.i_mb_width - 1) )
2494     {
2495         h->stat.frame.i_misc_bits = bs_pos( &h->out.bs )
2496                                   + (h->out.i_nal*NALU_OVERHEAD * 8)
2497                                   - h->stat.frame.i_tex_bits
2498                                   - h->stat.frame.i_mv_bits;
2499         x264_fdec_filter_row( h, h->i_threadslice_end, 0 );
2500
2501         if( h->param.b_sliced_threads )
2502         {
2503             /* Tell the main thread we're done. */
2504             x264_threadslice_cond_broadcast( h, 1 );
2505             /* Do hpel now */
2506             for( int mb_y = h->i_threadslice_start; mb_y <= h->i_threadslice_end; mb_y++ )
2507                 x264_fdec_filter_row( h, mb_y, 1 );
2508             x264_threadslice_cond_broadcast( h, 2 );
2509             /* Do the first row of hpel, now that the previous slice is done */
2510             if( h->i_thread_idx > 0 )
2511             {
2512                 x264_threadslice_cond_wait( h->thread[h->i_thread_idx-1], 2 );
2513                 x264_fdec_filter_row( h, h->i_threadslice_start + (1 << SLICE_MBAFF), 2 );
2514             }
2515         }
2516     }
2517
2518     return 0;
2519 }
2520
2521 static void x264_thread_sync_context( x264_t *dst, x264_t *src )
2522 {
2523     if( dst == src )
2524         return;
2525
2526     // reference counting
2527     for( x264_frame_t **f = src->frames.reference; *f; f++ )
2528         (*f)->i_reference_count++;
2529     for( x264_frame_t **f = dst->frames.reference; *f; f++ )
2530         x264_frame_push_unused( src, *f );
2531     src->fdec->i_reference_count++;
2532     x264_frame_push_unused( src, dst->fdec );
2533
2534     // copy everything except the per-thread pointers and the constants.
2535     memcpy( &dst->i_frame, &src->i_frame, offsetof(x264_t, mb.type) - offsetof(x264_t, i_frame) );
2536     dst->param = src->param;
2537     dst->stat = src->stat;
2538     dst->pixf = src->pixf;
2539 }
2540
2541 static void x264_thread_sync_stat( x264_t *dst, x264_t *src )
2542 {
2543     if( dst == src )
2544         return;
2545     memcpy( &dst->stat.i_frame_count, &src->stat.i_frame_count, sizeof(dst->stat) - sizeof(dst->stat.frame) );
2546 }
2547
2548 static void *x264_slices_write( x264_t *h )
2549 {
2550     int i_slice_num = 0;
2551     int last_thread_mb = h->sh.i_last_mb;
2552
2553 #if HAVE_VISUALIZE
2554     if( h->param.b_visualize )
2555         if( x264_visualize_init( h ) )
2556             goto fail;
2557 #endif
2558
2559     /* init stats */
2560     memset( &h->stat.frame, 0, sizeof(h->stat.frame) );
2561     h->mb.b_reencode_mb = 0;
2562     while( h->sh.i_first_mb + SLICE_MBAFF*h->mb.i_mb_stride <= last_thread_mb )
2563     {
2564         h->sh.i_last_mb = last_thread_mb;
2565         if( h->param.i_slice_max_mbs )
2566         {
2567             if( SLICE_MBAFF )
2568             {
2569                 // convert first to mbaff form, add slice-max-mbs, then convert back to normal form
2570                 int last_mbaff = 2*(h->sh.i_first_mb % h->mb.i_mb_width)
2571                     + h->mb.i_mb_width*(h->sh.i_first_mb / h->mb.i_mb_width)
2572                     + h->param.i_slice_max_mbs - 1;
2573                 int last_x = (last_mbaff % (2*h->mb.i_mb_width))/2;
2574                 int last_y = (last_mbaff / (2*h->mb.i_mb_width))*2 + 1;
2575                 h->sh.i_last_mb = last_x + h->mb.i_mb_stride*last_y;
2576             }
2577             else
2578                 h->sh.i_last_mb = h->sh.i_first_mb + h->param.i_slice_max_mbs - 1;
2579         }
2580         else if( h->param.i_slice_count && !h->param.b_sliced_threads )
2581         {
2582             int height = h->mb.i_mb_height >> PARAM_INTERLACED;
2583             int width = h->mb.i_mb_width << PARAM_INTERLACED;
2584             i_slice_num++;
2585             h->sh.i_last_mb = (height * i_slice_num + h->param.i_slice_count/2) / h->param.i_slice_count * width - 1;
2586         }
2587         h->sh.i_last_mb = X264_MIN( h->sh.i_last_mb, last_thread_mb );
2588         if( x264_stack_align( x264_slice_write, h ) )
2589             goto fail;
2590         h->sh.i_first_mb = h->sh.i_last_mb + 1;
2591         // if i_first_mb is not the last mb in a row then go to the next mb in MBAFF order
2592         if( SLICE_MBAFF && h->sh.i_first_mb % h->mb.i_mb_width )
2593             h->sh.i_first_mb -= h->mb.i_mb_stride;
2594     }
2595
2596 #if HAVE_VISUALIZE
2597     if( h->param.b_visualize )
2598     {
2599         x264_visualize_show( h );
2600         x264_visualize_close( h );
2601     }
2602 #endif
2603
2604     return (void *)0;
2605
2606 fail:
2607     /* Tell other threads we're done, so they wouldn't wait for it */
2608     if( h->param.b_sliced_threads )
2609         x264_threadslice_cond_broadcast( h, 2 );
2610     return (void *)-1;
2611 }
2612
2613 static int x264_threaded_slices_write( x264_t *h )
2614 {
2615     /* set first/last mb and sync contexts */
2616     for( int i = 0; i < h->param.i_threads; i++ )
2617     {
2618         x264_t *t = h->thread[i];
2619         if( i )
2620         {
2621             t->param = h->param;
2622             memcpy( &t->i_frame, &h->i_frame, offsetof(x264_t, rc) - offsetof(x264_t, i_frame) );
2623         }
2624         int height = h->mb.i_mb_height >> PARAM_INTERLACED;
2625         t->i_threadslice_start = ((height *  i    + h->param.i_slice_count/2) / h->param.i_threads) << PARAM_INTERLACED;
2626         t->i_threadslice_end   = ((height * (i+1) + h->param.i_slice_count/2) / h->param.i_threads) << PARAM_INTERLACED;
2627         t->sh.i_first_mb = t->i_threadslice_start * h->mb.i_mb_width;
2628         t->sh.i_last_mb  =   t->i_threadslice_end * h->mb.i_mb_width - 1;
2629     }
2630
2631     x264_stack_align( x264_analyse_weight_frame, h, h->mb.i_mb_height*16 + 16 );
2632
2633     x264_threads_distribute_ratecontrol( h );
2634
2635     /* setup */
2636     for( int i = 0; i < h->param.i_threads; i++ )
2637     {
2638         h->thread[i]->i_thread_idx = i;
2639         h->thread[i]->b_thread_active = 1;
2640         x264_threadslice_cond_broadcast( h->thread[i], 0 );
2641     }
2642     /* dispatch */
2643     for( int i = 0; i < h->param.i_threads; i++ )
2644         x264_threadpool_run( h->threadpool, (void*)x264_slices_write, h->thread[i] );
2645     /* wait */
2646     for( int i = 0; i < h->param.i_threads; i++ )
2647         x264_threadslice_cond_wait( h->thread[i], 1 );
2648
2649     x264_threads_merge_ratecontrol( h );
2650
2651     for( int i = 1; i < h->param.i_threads; i++ )
2652     {
2653         x264_t *t = h->thread[i];
2654         for( int j = 0; j < t->out.i_nal; j++ )
2655         {
2656             h->out.nal[h->out.i_nal] = t->out.nal[j];
2657             h->out.i_nal++;
2658             x264_nal_check_buffer( h );
2659         }
2660         /* All entries in stat.frame are ints except for ssd/ssim. */
2661         for( int j = 0; j < (offsetof(x264_t,stat.frame.i_ssd) - offsetof(x264_t,stat.frame.i_mv_bits)) / sizeof(int); j++ )
2662             ((int*)&h->stat.frame)[j] += ((int*)&t->stat.frame)[j];
2663         for( int j = 0; j < 3; j++ )
2664             h->stat.frame.i_ssd[j] += t->stat.frame.i_ssd[j];
2665         h->stat.frame.f_ssim += t->stat.frame.f_ssim;
2666         h->stat.frame.i_ssim_cnt += t->stat.frame.i_ssim_cnt;
2667     }
2668
2669     return 0;
2670 }
2671
2672 void x264_encoder_intra_refresh( x264_t *h )
2673 {
2674     h = h->thread[h->i_thread_phase];
2675     h->b_queued_intra_refresh = 1;
2676 }
2677
2678 int x264_encoder_invalidate_reference( x264_t *h, int64_t pts )
2679 {
2680     if( h->param.i_bframe )
2681     {
2682         x264_log( h, X264_LOG_ERROR, "x264_encoder_invalidate_reference is not supported with B-frames enabled\n" );
2683         return -1;
2684     }
2685     if( h->param.b_intra_refresh )
2686     {
2687         x264_log( h, X264_LOG_ERROR, "x264_encoder_invalidate_reference is not supported with intra refresh enabled\n" );
2688         return -1;
2689     }
2690     h = h->thread[h->i_thread_phase];
2691     if( pts >= h->i_last_idr_pts )
2692     {
2693         for( int i = 0; h->frames.reference[i]; i++ )
2694             if( pts <= h->frames.reference[i]->i_pts )
2695                 h->frames.reference[i]->b_corrupt = 1;
2696         if( pts <= h->fdec->i_pts )
2697             h->fdec->b_corrupt = 1;
2698     }
2699     return 0;
2700 }
2701
2702 /****************************************************************************
2703  * x264_encoder_encode:
2704  *  XXX: i_poc   : is the poc of the current given picture
2705  *       i_frame : is the number of the frame being coded
2706  *  ex:  type frame poc
2707  *       I      0   2*0
2708  *       P      1   2*3
2709  *       B      2   2*1
2710  *       B      3   2*2
2711  *       P      4   2*6
2712  *       B      5   2*4
2713  *       B      6   2*5
2714  ****************************************************************************/
2715 int     x264_encoder_encode( x264_t *h,
2716                              x264_nal_t **pp_nal, int *pi_nal,
2717                              x264_picture_t *pic_in,
2718                              x264_picture_t *pic_out )
2719 {
2720     x264_t *thread_current, *thread_prev, *thread_oldest;
2721     int i_nal_type, i_nal_ref_idc, i_global_qp;
2722     int overhead = NALU_OVERHEAD;
2723
2724     if( h->i_thread_frames > 1 )
2725     {
2726         thread_prev    = h->thread[ h->i_thread_phase ];
2727         h->i_thread_phase = (h->i_thread_phase + 1) % h->i_thread_frames;
2728         thread_current = h->thread[ h->i_thread_phase ];
2729         thread_oldest  = h->thread[ (h->i_thread_phase + 1) % h->i_thread_frames ];
2730         x264_thread_sync_context( thread_current, thread_prev );
2731         x264_thread_sync_ratecontrol( thread_current, thread_prev, thread_oldest );
2732         h = thread_current;
2733     }
2734     else
2735     {
2736         thread_current =
2737         thread_oldest  = h;
2738     }
2739 #if HAVE_MMX
2740     if( h->param.cpu&X264_CPU_SSE_MISALIGN )
2741         x264_cpu_mask_misalign_sse();
2742 #endif
2743
2744     /* no data out */
2745     *pi_nal = 0;
2746     *pp_nal = NULL;
2747
2748     /* ------------------- Setup new frame from picture -------------------- */
2749     if( pic_in != NULL )
2750     {
2751         /* 1: Copy the picture to a frame and move it to a buffer */
2752         x264_frame_t *fenc = x264_frame_pop_unused( h, 0 );
2753         if( !fenc )
2754             return -1;
2755
2756         if( x264_frame_copy_picture( h, fenc, pic_in ) < 0 )
2757             return -1;
2758
2759         if( h->param.i_width != 16 * h->mb.i_mb_width ||
2760             h->param.i_height != 16 * h->mb.i_mb_height )
2761             x264_frame_expand_border_mod16( h, fenc );
2762
2763         fenc->i_frame = h->frames.i_input++;
2764
2765         if( fenc->i_frame == 0 )
2766             h->frames.i_first_pts = fenc->i_pts;
2767         if( h->frames.i_bframe_delay && fenc->i_frame == h->frames.i_bframe_delay )
2768             h->frames.i_bframe_delay_time = fenc->i_pts - h->frames.i_first_pts;
2769
2770         if( h->param.b_vfr_input && fenc->i_pts <= h->frames.i_largest_pts )
2771             x264_log( h, X264_LOG_WARNING, "non-strictly-monotonic PTS\n" );
2772
2773         h->frames.i_second_largest_pts = h->frames.i_largest_pts;
2774         h->frames.i_largest_pts = fenc->i_pts;
2775
2776         if( (fenc->i_pic_struct < PIC_STRUCT_AUTO) || (fenc->i_pic_struct > PIC_STRUCT_TRIPLE) )
2777             fenc->i_pic_struct = PIC_STRUCT_AUTO;
2778
2779         if( fenc->i_pic_struct == PIC_STRUCT_AUTO )
2780         {
2781 #if HAVE_INTERLACED
2782             int b_interlaced = fenc->param ? fenc->param->b_interlaced : h->param.b_interlaced;
2783 #else
2784             int b_interlaced = 0;
2785 #endif
2786             if( b_interlaced )
2787             {
2788                 int b_tff = fenc->param ? fenc->param->b_tff : h->param.b_tff;
2789                 fenc->i_pic_struct = b_tff ? PIC_STRUCT_TOP_BOTTOM : PIC_STRUCT_BOTTOM_TOP;
2790             }
2791             else
2792                 fenc->i_pic_struct = PIC_STRUCT_PROGRESSIVE;
2793         }
2794
2795         if( h->param.rc.b_mb_tree && h->param.rc.b_stat_read )
2796         {
2797             if( x264_macroblock_tree_read( h, fenc, pic_in->prop.quant_offsets ) )
2798                 return -1;
2799         }
2800         else
2801             x264_stack_align( x264_adaptive_quant_frame, h, fenc, pic_in->prop.quant_offsets );
2802
2803         if( pic_in->prop.quant_offsets_free )
2804             pic_in->prop.quant_offsets_free( pic_in->prop.quant_offsets );
2805
2806         if( h->frames.b_have_lowres )
2807             x264_frame_init_lowres( h, fenc );
2808
2809         /* 2: Place the frame into the queue for its slice type decision */
2810         x264_lookahead_put_frame( h, fenc );
2811
2812         if( h->frames.i_input <= h->frames.i_delay + 1 - h->i_thread_frames )
2813         {
2814             /* Nothing yet to encode, waiting for filling of buffers */
2815             pic_out->i_type = X264_TYPE_AUTO;
2816             return 0;
2817         }
2818     }
2819     else
2820     {
2821         /* signal kills for lookahead thread */
2822         x264_pthread_mutex_lock( &h->lookahead->ifbuf.mutex );
2823         h->lookahead->b_exit_thread = 1;
2824         x264_pthread_cond_broadcast( &h->lookahead->ifbuf.cv_fill );
2825         x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
2826     }
2827
2828     h->i_frame++;
2829     /* 3: The picture is analyzed in the lookahead */
2830     if( !h->frames.current[0] )
2831         x264_lookahead_get_frames( h );
2832
2833     if( !h->frames.current[0] && x264_lookahead_is_empty( h ) )
2834         return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
2835
2836     /* ------------------- Get frame to be encoded ------------------------- */
2837     /* 4: get picture to encode */
2838     h->fenc = x264_frame_shift( h->frames.current );
2839
2840     /* If applicable, wait for previous frame reconstruction to finish */
2841     if( h->param.b_sliced_threads )
2842         if( x264_threadpool_wait_all( h ) < 0 )
2843             return -1;
2844
2845     if( h->i_frame == h->i_thread_frames - 1 )
2846         h->i_reordered_pts_delay = h->fenc->i_reordered_pts;
2847     if( h->fenc->param )
2848     {
2849         x264_encoder_reconfig( h, h->fenc->param );
2850         if( h->fenc->param->param_free )
2851             h->fenc->param->param_free( h->fenc->param );
2852     }
2853
2854     // ok to call this before encoding any frames, since the initial values of fdec have b_kept_as_ref=0
2855     if( x264_reference_update( h ) )
2856         return -1;
2857     h->fdec->i_lines_completed = -1;
2858
2859     if( !IS_X264_TYPE_I( h->fenc->i_type ) )
2860     {
2861         int valid_refs_left = 0;
2862         for( int i = 0; h->frames.reference[i]; i++ )
2863             if( !h->frames.reference[i]->b_corrupt )
2864                 valid_refs_left++;
2865         /* No valid reference frames left: force an IDR. */
2866         if( !valid_refs_left )
2867         {
2868             h->fenc->b_keyframe = 1;
2869             h->fenc->i_type = X264_TYPE_IDR;
2870         }
2871     }
2872
2873     if( h->fenc->b_keyframe )
2874     {
2875         h->frames.i_last_keyframe = h->fenc->i_frame;
2876         if( h->fenc->i_type == X264_TYPE_IDR )
2877         {
2878             h->i_frame_num = 0;
2879             h->frames.i_last_idr = h->fenc->i_frame;
2880         }
2881     }
2882     h->sh.i_mmco_command_count =
2883     h->sh.i_mmco_remove_from_end = 0;
2884     h->b_ref_reorder[0] =
2885     h->b_ref_reorder[1] = 0;
2886     h->fdec->i_poc =
2887     h->fenc->i_poc = 2 * ( h->fenc->i_frame - X264_MAX( h->frames.i_last_idr, 0 ) );
2888
2889     /* ------------------- Setup frame context ----------------------------- */
2890     /* 5: Init data dependent of frame type */
2891     if( h->fenc->i_type == X264_TYPE_IDR )
2892     {
2893         /* reset ref pictures */
2894         i_nal_type    = NAL_SLICE_IDR;
2895         i_nal_ref_idc = NAL_PRIORITY_HIGHEST;
2896         h->sh.i_type = SLICE_TYPE_I;
2897         x264_reference_reset( h );
2898         h->frames.i_poc_last_open_gop = -1;
2899     }
2900     else if( h->fenc->i_type == X264_TYPE_I )
2901     {
2902         i_nal_type    = NAL_SLICE;
2903         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
2904         h->sh.i_type = SLICE_TYPE_I;
2905         x264_reference_hierarchy_reset( h );
2906         if( h->param.b_open_gop )
2907             h->frames.i_poc_last_open_gop = h->fenc->b_keyframe ? h->fenc->i_poc : -1;
2908     }
2909     else if( h->fenc->i_type == X264_TYPE_P )
2910     {
2911         i_nal_type    = NAL_SLICE;
2912         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
2913         h->sh.i_type = SLICE_TYPE_P;
2914         x264_reference_hierarchy_reset( h );
2915         h->frames.i_poc_last_open_gop = -1;
2916     }
2917     else if( h->fenc->i_type == X264_TYPE_BREF )
2918     {
2919         i_nal_type    = NAL_SLICE;
2920         i_nal_ref_idc = h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT ? NAL_PRIORITY_LOW : NAL_PRIORITY_HIGH;
2921         h->sh.i_type = SLICE_TYPE_B;
2922         x264_reference_hierarchy_reset( h );
2923     }
2924     else    /* B frame */
2925     {
2926         i_nal_type    = NAL_SLICE;
2927         i_nal_ref_idc = NAL_PRIORITY_DISPOSABLE;
2928         h->sh.i_type = SLICE_TYPE_B;
2929     }
2930
2931     h->fdec->i_type = h->fenc->i_type;
2932     h->fdec->i_frame = h->fenc->i_frame;
2933     h->fenc->b_kept_as_ref =
2934     h->fdec->b_kept_as_ref = i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE && h->param.i_keyint_max > 1;
2935
2936     h->fdec->i_pts = h->fenc->i_pts;
2937     if( h->frames.i_bframe_delay )
2938     {
2939         int64_t *prev_reordered_pts = thread_current->frames.i_prev_reordered_pts;
2940         h->fdec->i_dts = h->i_frame > h->frames.i_bframe_delay
2941                        ? prev_reordered_pts[ (h->i_frame - h->frames.i_bframe_delay) % h->frames.i_bframe_delay ]
2942                        : h->fenc->i_reordered_pts - h->frames.i_bframe_delay_time;
2943         prev_reordered_pts[ h->i_frame % h->frames.i_bframe_delay ] = h->fenc->i_reordered_pts;
2944     }
2945     else
2946         h->fdec->i_dts = h->fenc->i_reordered_pts;
2947     if( h->fenc->i_type == X264_TYPE_IDR )
2948         h->i_last_idr_pts = h->fdec->i_pts;
2949
2950     /* ------------------- Init                ----------------------------- */
2951     /* build ref list 0/1 */
2952     x264_reference_build_list( h, h->fdec->i_poc );
2953
2954     /* ---------------------- Write the bitstream -------------------------- */
2955     /* Init bitstream context */
2956     if( h->param.b_sliced_threads )
2957     {
2958         for( int i = 0; i < h->param.i_threads; i++ )
2959         {
2960             bs_init( &h->thread[i]->out.bs, h->thread[i]->out.p_bitstream, h->thread[i]->out.i_bitstream );
2961             h->thread[i]->out.i_nal = 0;
2962         }
2963     }
2964     else
2965     {
2966         bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
2967         h->out.i_nal = 0;
2968     }
2969
2970     if( h->param.b_aud )
2971     {
2972         int pic_type;
2973
2974         if( h->sh.i_type == SLICE_TYPE_I )
2975             pic_type = 0;
2976         else if( h->sh.i_type == SLICE_TYPE_P )
2977             pic_type = 1;
2978         else if( h->sh.i_type == SLICE_TYPE_B )
2979             pic_type = 2;
2980         else
2981             pic_type = 7;
2982
2983         x264_nal_start( h, NAL_AUD, NAL_PRIORITY_DISPOSABLE );
2984         bs_write( &h->out.bs, 3, pic_type );
2985         bs_rbsp_trailing( &h->out.bs );
2986         if( x264_nal_end( h ) )
2987             return -1;
2988         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2989     }
2990
2991     h->i_nal_type = i_nal_type;
2992     h->i_nal_ref_idc = i_nal_ref_idc;
2993
2994     if( h->param.b_intra_refresh )
2995     {
2996         if( IS_X264_TYPE_I( h->fenc->i_type ) )
2997         {
2998             h->fdec->i_frames_since_pir = 0;
2999             h->b_queued_intra_refresh = 0;
3000             /* PIR is currently only supported with ref == 1, so any intra frame effectively refreshes
3001              * the whole frame and counts as an intra refresh. */
3002             h->fdec->f_pir_position = h->mb.i_mb_width;
3003         }
3004         else if( h->fenc->i_type == X264_TYPE_P )
3005         {
3006             int pocdiff = (h->fdec->i_poc - h->fref[0][0]->i_poc)/2;
3007             float increment = X264_MAX( ((float)h->mb.i_mb_width-1) / h->param.i_keyint_max, 1 );
3008             h->fdec->f_pir_position = h->fref[0][0]->f_pir_position;
3009             h->fdec->i_frames_since_pir = h->fref[0][0]->i_frames_since_pir + pocdiff;
3010             if( h->fdec->i_frames_since_pir >= h->param.i_keyint_max ||
3011                 (h->b_queued_intra_refresh && h->fdec->f_pir_position + 0.5 >= h->mb.i_mb_width) )
3012             {
3013                 h->fdec->f_pir_position = 0;
3014                 h->fdec->i_frames_since_pir = 0;
3015                 h->b_queued_intra_refresh = 0;
3016                 h->fenc->b_keyframe = 1;
3017             }
3018             h->fdec->i_pir_start_col = h->fdec->f_pir_position+0.5;
3019             h->fdec->f_pir_position += increment * pocdiff;
3020             h->fdec->i_pir_end_col = h->fdec->f_pir_position+0.5;
3021             /* If our intra refresh has reached the right side of the frame, we're done. */
3022             if( h->fdec->i_pir_end_col >= h->mb.i_mb_width - 1 )
3023             {
3024                 h->fdec->f_pir_position = h->mb.i_mb_width;
3025                 h->fdec->i_pir_end_col = h->mb.i_mb_width - 1;
3026             }
3027         }
3028     }
3029
3030     if( h->fenc->b_keyframe )
3031     {
3032         /* Write SPS and PPS */
3033         if( h->param.b_repeat_headers )
3034         {
3035             /* generate sequence parameters */
3036             x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
3037             x264_sps_write( &h->out.bs, h->sps );
3038             if( x264_nal_end( h ) )
3039                 return -1;
3040             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
3041
3042             /* generate picture parameters */
3043             x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
3044             x264_pps_write( &h->out.bs, h->sps, h->pps );
3045             if( x264_nal_end( h ) )
3046                 return -1;
3047             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
3048         }
3049
3050         /* when frame threading is used, buffering period sei is written in x264_encoder_frame_end */
3051         if( h->i_thread_frames == 1 && h->sps->vui.b_nal_hrd_parameters_present )
3052         {
3053             x264_hrd_fullness( h );
3054             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3055             x264_sei_buffering_period_write( h, &h->out.bs );
3056             if( x264_nal_end( h ) )
3057                return -1;
3058             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
3059         }
3060     }
3061
3062     /* write extra sei */
3063     for( int i = 0; i < h->fenc->extra_sei.num_payloads; i++ )
3064     {
3065         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3066         x264_sei_write( &h->out.bs, h->fenc->extra_sei.payloads[i].payload, h->fenc->extra_sei.payloads[i].payload_size,
3067                         h->fenc->extra_sei.payloads[i].payload_type );
3068         if( x264_nal_end( h ) )
3069             return -1;
3070         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
3071         if( h->fenc->extra_sei.sei_free && h->fenc->extra_sei.payloads[i].payload )
3072             h->fenc->extra_sei.sei_free( h->fenc->extra_sei.payloads[i].payload );
3073     }
3074
3075     if( h->fenc->extra_sei.sei_free && h->fenc->extra_sei.payloads )
3076         h->fenc->extra_sei.sei_free( h->fenc->extra_sei.payloads );
3077
3078     if( h->fenc->b_keyframe )
3079     {
3080         if( h->param.b_repeat_headers && h->fenc->i_frame == 0 )
3081         {
3082             /* identify ourself */
3083             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3084             if( x264_sei_version_write( h, &h->out.bs ) )
3085                 return -1;
3086             if( x264_nal_end( h ) )
3087                 return -1;
3088             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
3089         }
3090
3091         if( h->fenc->i_type != X264_TYPE_IDR )
3092         {
3093             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;
3094             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3095             x264_sei_recovery_point_write( h, &h->out.bs, time_to_recovery );
3096             if( x264_nal_end( h ) )
3097                 return -1;
3098             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
3099         }
3100
3101         if ( h->param.i_frame_packing >= 0 )
3102         {
3103             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3104             x264_sei_frame_packing_write( h, &h->out.bs );
3105             if( x264_nal_end( h ) )
3106                 return -1;
3107             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
3108         }
3109     }
3110
3111     /* generate sei pic timing */
3112     if( h->sps->vui.b_pic_struct_present || h->sps->vui.b_nal_hrd_parameters_present )
3113     {
3114         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3115         x264_sei_pic_timing_write( h, &h->out.bs );
3116         if( x264_nal_end( h ) )
3117             return -1;
3118         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
3119     }
3120
3121     /* As required by Blu-ray. */
3122     if( !IS_X264_TYPE_B( h->fenc->i_type ) && h->b_sh_backup )
3123     {
3124         h->b_sh_backup = 0;
3125         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3126         x264_sei_dec_ref_pic_marking_write( h, &h->out.bs );
3127         if( x264_nal_end( h ) )
3128             return -1;
3129         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
3130     }
3131
3132     if( h->fenc->b_keyframe && h->param.b_intra_refresh )
3133         h->i_cpb_delay_pir_offset = h->fenc->i_cpb_delay;
3134
3135     /* Init the rate control */
3136     /* FIXME: Include slice header bit cost. */
3137     x264_ratecontrol_start( h, h->fenc->i_qpplus1, overhead*8 );
3138     i_global_qp = x264_ratecontrol_qp( h );
3139
3140     pic_out->i_qpplus1 =
3141     h->fdec->i_qpplus1 = i_global_qp + 1;
3142
3143     if( h->param.rc.b_stat_read && h->sh.i_type != SLICE_TYPE_I )
3144     {
3145         x264_reference_build_list_optimal( h );
3146         x264_reference_check_reorder( h );
3147     }
3148
3149     if( h->i_ref[0] )
3150         h->fdec->i_poc_l0ref0 = h->fref[0][0]->i_poc;
3151
3152     /* ------------------------ Create slice header  ----------------------- */
3153     x264_slice_init( h, i_nal_type, i_global_qp );
3154
3155     /*------------------------- Weights -------------------------------------*/
3156     if( h->sh.i_type == SLICE_TYPE_B )
3157         x264_macroblock_bipred_init( h );
3158
3159     x264_weighted_pred_init( h );
3160
3161     if( i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE )
3162         h->i_frame_num++;
3163
3164     /* Write frame */
3165     h->i_threadslice_start = 0;
3166     h->i_threadslice_end = h->mb.i_mb_height;
3167     if( h->i_thread_frames > 1 )
3168     {
3169         x264_threadpool_run( h->threadpool, (void*)x264_slices_write, h );
3170         h->b_thread_active = 1;
3171     }
3172     else if( h->param.b_sliced_threads )
3173     {
3174         if( x264_threaded_slices_write( h ) )
3175             return -1;
3176     }
3177     else
3178         if( (intptr_t)x264_slices_write( h ) )
3179             return -1;
3180
3181     return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
3182 }
3183
3184 static int x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
3185                                    x264_nal_t **pp_nal, int *pi_nal,
3186                                    x264_picture_t *pic_out )
3187 {
3188     char psz_message[80];
3189
3190     if( !h->param.b_sliced_threads && h->b_thread_active )
3191     {
3192         h->b_thread_active = 0;
3193         if( (intptr_t)x264_threadpool_wait( h->threadpool, h ) )
3194             return -1;
3195     }
3196     if( !h->out.i_nal )
3197     {
3198         pic_out->i_type = X264_TYPE_AUTO;
3199         return 0;
3200     }
3201
3202     x264_emms();
3203     /* generate buffering period sei and insert it into place */
3204     if( h->i_thread_frames > 1 && h->fenc->b_keyframe && h->sps->vui.b_nal_hrd_parameters_present )
3205     {
3206         x264_hrd_fullness( h );
3207         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3208         x264_sei_buffering_period_write( h, &h->out.bs );
3209         if( x264_nal_end( h ) )
3210            return -1;
3211         /* buffering period sei must follow AUD, SPS and PPS and precede all other SEIs */
3212         int idx = 0;
3213         while( h->out.nal[idx].i_type == NAL_AUD ||
3214                h->out.nal[idx].i_type == NAL_SPS ||
3215                h->out.nal[idx].i_type == NAL_PPS )
3216             idx++;
3217         x264_nal_t nal_tmp = h->out.nal[h->out.i_nal-1];
3218         memmove( &h->out.nal[idx+1], &h->out.nal[idx], (h->out.i_nal-idx-1)*sizeof(x264_nal_t) );
3219         h->out.nal[idx] = nal_tmp;
3220     }
3221
3222     int frame_size = x264_encoder_encapsulate_nals( h, 0 );
3223     if( frame_size < 0 )
3224         return -1;
3225
3226     /* Set output picture properties */
3227     pic_out->i_type = h->fenc->i_type;
3228
3229     pic_out->b_keyframe = h->fenc->b_keyframe;
3230     pic_out->i_pic_struct = h->fenc->i_pic_struct;
3231
3232     pic_out->i_pts = h->fdec->i_pts;
3233     pic_out->i_dts = h->fdec->i_dts;
3234
3235     if( pic_out->i_pts < pic_out->i_dts )
3236         x264_log( h, X264_LOG_WARNING, "invalid DTS: PTS is less than DTS\n" );
3237
3238     pic_out->opaque = h->fenc->opaque;
3239
3240     pic_out->img.i_csp = h->fdec->i_csp;
3241 #if HIGH_BIT_DEPTH
3242     pic_out->img.i_csp |= X264_CSP_HIGH_DEPTH;
3243 #endif
3244     pic_out->img.i_plane = h->fdec->i_plane;
3245     for( int i = 0; i < pic_out->img.i_plane; i++ )
3246     {
3247         pic_out->img.i_stride[i] = h->fdec->i_stride[i] * sizeof(pixel);
3248         pic_out->img.plane[i] = (uint8_t*)h->fdec->plane[i];
3249     }
3250
3251     x264_frame_push_unused( thread_current, h->fenc );
3252
3253     /* ---------------------- Update encoder state ------------------------- */
3254
3255     /* update rc */
3256     int filler = 0;
3257     if( x264_ratecontrol_end( h, frame_size * 8, &filler ) < 0 )
3258         return -1;
3259
3260     pic_out->hrd_timing = h->fenc->hrd_timing;
3261
3262     while( filler > 0 )
3263     {
3264         int f, overhead;
3265         overhead = (FILLER_OVERHEAD - h->param.b_annexb);
3266         if( h->param.i_slice_max_size && filler > h->param.i_slice_max_size )
3267         {
3268             int next_size = filler - h->param.i_slice_max_size;
3269             int overflow = X264_MAX( overhead - next_size, 0 );
3270             f = h->param.i_slice_max_size - overhead - overflow;
3271         }
3272         else
3273             f = X264_MAX( 0, filler - overhead );
3274
3275         x264_nal_start( h, NAL_FILLER, NAL_PRIORITY_DISPOSABLE );
3276         x264_filler_write( h, &h->out.bs, f );
3277         if( x264_nal_end( h ) )
3278             return -1;
3279         int total_size = x264_encoder_encapsulate_nals( h, h->out.i_nal-1 );
3280         if( total_size < 0 )
3281             return -1;
3282         frame_size += total_size;
3283         filler -= total_size;
3284     }
3285
3286     /* End bitstream, set output  */
3287     *pi_nal = h->out.i_nal;
3288     *pp_nal = h->out.nal;
3289
3290     h->out.i_nal = 0;
3291
3292     x264_noise_reduction_update( h );
3293
3294     /* ---------------------- Compute/Print statistics --------------------- */
3295     x264_thread_sync_stat( h, h->thread[0] );
3296
3297     /* Slice stat */
3298     h->stat.i_frame_count[h->sh.i_type]++;
3299     h->stat.i_frame_size[h->sh.i_type] += frame_size;
3300     h->stat.f_frame_qp[h->sh.i_type] += h->fdec->f_qp_avg_aq;
3301
3302     for( int i = 0; i < X264_MBTYPE_MAX; i++ )
3303         h->stat.i_mb_count[h->sh.i_type][i] += h->stat.frame.i_mb_count[i];
3304     for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
3305         h->stat.i_mb_partition[h->sh.i_type][i] += h->stat.frame.i_mb_partition[i];
3306     for( int i = 0; i < 2; i++ )
3307         h->stat.i_mb_count_8x8dct[i] += h->stat.frame.i_mb_count_8x8dct[i];
3308     for( int i = 0; i < 6; i++ )
3309         h->stat.i_mb_cbp[i] += h->stat.frame.i_mb_cbp[i];
3310     for( int i = 0; i < 4; i++ )
3311         for( int j = 0; j < 13; j++ )
3312             h->stat.i_mb_pred_mode[i][j] += h->stat.frame.i_mb_pred_mode[i][j];
3313     if( h->sh.i_type != SLICE_TYPE_I )
3314         for( int i_list = 0; i_list < 2; i_list++ )
3315             for( int i = 0; i < X264_REF_MAX*2; i++ )
3316                 h->stat.i_mb_count_ref[h->sh.i_type][i_list][i] += h->stat.frame.i_mb_count_ref[i_list][i];
3317     for( int i = 0; i < 3; i++ )
3318         h->stat.i_mb_field[i] += h->stat.frame.i_mb_field[i];
3319     if( h->sh.i_type == SLICE_TYPE_P && h->param.analyse.i_weighted_pred >= X264_WEIGHTP_SIMPLE )
3320     {
3321         h->stat.i_wpred[0] += !!h->sh.weight[0][0].weightfn;
3322         h->stat.i_wpred[1] += !!h->sh.weight[0][1].weightfn || !!h->sh.weight[0][2].weightfn;
3323     }
3324     if( h->sh.i_type == SLICE_TYPE_B )
3325     {
3326         h->stat.i_direct_frames[ h->sh.b_direct_spatial_mv_pred ] ++;
3327         if( h->mb.b_direct_auto_write )
3328         {
3329             //FIXME somewhat arbitrary time constants
3330             if( h->stat.i_direct_score[0] + h->stat.i_direct_score[1] > h->mb.i_mb_count )
3331                 for( int i = 0; i < 2; i++ )
3332                     h->stat.i_direct_score[i] = h->stat.i_direct_score[i] * 9/10;
3333             for( int i = 0; i < 2; i++ )
3334                 h->stat.i_direct_score[i] += h->stat.frame.i_direct_score[i];
3335         }
3336     }
3337     else
3338         h->stat.i_consecutive_bframes[h->fenc->i_bframes]++;
3339
3340     psz_message[0] = '\0';
3341     double dur = h->fenc->f_duration;
3342     h->stat.f_frame_duration[h->sh.i_type] += dur;
3343     if( h->param.analyse.b_psnr )
3344     {
3345         int64_t ssd[3] =
3346         {
3347             h->stat.frame.i_ssd[0],
3348             h->stat.frame.i_ssd[1],
3349             h->stat.frame.i_ssd[2],
3350         };
3351         int luma_size = h->param.i_width * h->param.i_height;
3352         int chroma_size = CHROMA_SIZE( luma_size );
3353         pic_out->prop.f_psnr[0] = x264_psnr( ssd[0], luma_size );
3354         pic_out->prop.f_psnr[1] = x264_psnr( ssd[1], chroma_size );
3355         pic_out->prop.f_psnr[2] = x264_psnr( ssd[2], chroma_size );
3356         pic_out->prop.f_psnr_avg = x264_psnr( ssd[0] + ssd[1] + ssd[2], luma_size + chroma_size*2 );
3357
3358         h->stat.f_ssd_global[h->sh.i_type]   += dur * (ssd[0] + ssd[1] + ssd[2]);
3359         h->stat.f_psnr_average[h->sh.i_type] += dur * pic_out->prop.f_psnr_avg;
3360         h->stat.f_psnr_mean_y[h->sh.i_type]  += dur * pic_out->prop.f_psnr[0];
3361         h->stat.f_psnr_mean_u[h->sh.i_type]  += dur * pic_out->prop.f_psnr[1];
3362         h->stat.f_psnr_mean_v[h->sh.i_type]  += dur * pic_out->prop.f_psnr[2];
3363
3364         snprintf( psz_message, 80, " PSNR Y:%5.2f U:%5.2f V:%5.2f", pic_out->prop.f_psnr[0],
3365                                                                     pic_out->prop.f_psnr[1],
3366                                                                     pic_out->prop.f_psnr[2] );
3367     }
3368
3369     if( h->param.analyse.b_ssim )
3370     {
3371         pic_out->prop.f_ssim = h->stat.frame.f_ssim / h->stat.frame.i_ssim_cnt;
3372         h->stat.f_ssim_mean_y[h->sh.i_type] += pic_out->prop.f_ssim * dur;
3373         snprintf( psz_message + strlen(psz_message), 80 - strlen(psz_message),
3374                   " SSIM Y:%.5f", pic_out->prop.f_ssim );
3375     }
3376     psz_message[79] = '\0';
3377
3378     x264_log( h, X264_LOG_DEBUG,
3379                   "frame=%4d QP=%.2f NAL=%d Slice:%c Poc:%-3d I:%-4d P:%-4d SKIP:%-4d size=%d bytes%s\n",
3380               h->i_frame,
3381               h->fdec->f_qp_avg_aq,
3382               h->i_nal_ref_idc,
3383               h->sh.i_type == SLICE_TYPE_I ? 'I' : (h->sh.i_type == SLICE_TYPE_P ? 'P' : 'B' ),
3384               h->fdec->i_poc,
3385               h->stat.frame.i_mb_count_i,
3386               h->stat.frame.i_mb_count_p,
3387               h->stat.frame.i_mb_count_skip,
3388               frame_size,
3389               psz_message );
3390
3391     // keep stats all in one place
3392     x264_thread_sync_stat( h->thread[0], h );
3393     // for the use of the next frame
3394     x264_thread_sync_stat( thread_current, h );
3395
3396 #ifdef DEBUG_MB_TYPE
3397 {
3398     static const char mb_chars[] = { 'i', 'i', 'I', 'C', 'P', '8', 'S',
3399         'D', '<', 'X', 'B', 'X', '>', 'B', 'B', 'B', 'B', '8', 'S' };
3400     for( int mb_xy = 0; mb_xy < h->mb.i_mb_width * h->mb.i_mb_height; mb_xy++ )
3401     {
3402         if( h->mb.type[mb_xy] < X264_MBTYPE_MAX && h->mb.type[mb_xy] >= 0 )
3403             fprintf( stderr, "%c ", mb_chars[ h->mb.type[mb_xy] ] );
3404         else
3405             fprintf( stderr, "? " );
3406
3407         if( (mb_xy+1) % h->mb.i_mb_width == 0 )
3408             fprintf( stderr, "\n" );
3409     }
3410 }
3411 #endif
3412
3413     /* Remove duplicates, must be done near the end as breaks h->fref0 array
3414      * by freeing some of its pointers. */
3415     for( int i = 0; i < h->i_ref[0]; i++ )
3416         if( h->fref[0][i] && h->fref[0][i]->b_duplicate )
3417         {
3418             x264_frame_push_blank_unused( h, h->fref[0][i] );
3419             h->fref[0][i] = 0;
3420         }
3421
3422     if( h->param.psz_dump_yuv )
3423         x264_frame_dump( h );
3424     x264_emms();
3425
3426     return frame_size;
3427 }
3428
3429 static void x264_print_intra( int64_t *i_mb_count, double i_count, int b_print_pcm, char *intra )
3430 {
3431     intra += sprintf( intra, "I16..4%s: %4.1f%% %4.1f%% %4.1f%%",
3432         b_print_pcm ? "..PCM" : "",
3433         i_mb_count[I_16x16]/ i_count,
3434         i_mb_count[I_8x8]  / i_count,
3435         i_mb_count[I_4x4]  / i_count );
3436     if( b_print_pcm )
3437         sprintf( intra, " %4.1f%%", i_mb_count[I_PCM]  / i_count );
3438 }
3439
3440 /****************************************************************************
3441  * x264_encoder_close:
3442  ****************************************************************************/
3443 void    x264_encoder_close  ( x264_t *h )
3444 {
3445     int64_t i_yuv_size = FRAME_SIZE( h->param.i_width * h->param.i_height );
3446     int64_t i_mb_count_size[2][7] = {{0}};
3447     char buf[200];
3448     int b_print_pcm = h->stat.i_mb_count[SLICE_TYPE_I][I_PCM]
3449                    || h->stat.i_mb_count[SLICE_TYPE_P][I_PCM]
3450                    || h->stat.i_mb_count[SLICE_TYPE_B][I_PCM];
3451
3452     x264_lookahead_delete( h );
3453
3454     if( h->param.b_sliced_threads )
3455         x264_threadpool_wait_all( h );
3456     if( h->param.i_threads > 1 )
3457         x264_threadpool_delete( h->threadpool );
3458     if( h->i_thread_frames > 1 )
3459     {
3460         for( int i = 0; i < h->i_thread_frames; i++ )
3461             if( h->thread[i]->b_thread_active )
3462             {
3463                 assert( h->thread[i]->fenc->i_reference_count == 1 );
3464                 x264_frame_delete( h->thread[i]->fenc );
3465             }
3466
3467         x264_t *thread_prev = h->thread[h->i_thread_phase];
3468         x264_thread_sync_ratecontrol( h, thread_prev, h );
3469         x264_thread_sync_ratecontrol( thread_prev, thread_prev, h );
3470         h->i_frame = thread_prev->i_frame + 1 - h->i_thread_frames;
3471     }
3472     h->i_frame++;
3473
3474     /* Slices used and PSNR */
3475     for( int i = 0; i < 3; i++ )
3476     {
3477         static const uint8_t slice_order[] = { SLICE_TYPE_I, SLICE_TYPE_P, SLICE_TYPE_B };
3478         int i_slice = slice_order[i];
3479
3480         if( h->stat.i_frame_count[i_slice] > 0 )
3481         {
3482             int i_count = h->stat.i_frame_count[i_slice];
3483             double dur =  h->stat.f_frame_duration[i_slice];
3484             if( h->param.analyse.b_psnr )
3485             {
3486                 x264_log( h, X264_LOG_INFO,
3487                           "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",
3488                           slice_type_to_char[i_slice],
3489                           i_count,
3490                           h->stat.f_frame_qp[i_slice] / i_count,
3491                           (double)h->stat.i_frame_size[i_slice] / i_count,
3492                           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,
3493                           h->stat.f_psnr_average[i_slice] / dur,
3494                           x264_psnr( h->stat.f_ssd_global[i_slice], dur * i_yuv_size ) );
3495             }
3496             else
3497             {
3498                 x264_log( h, X264_LOG_INFO,
3499                           "frame %c:%-5d Avg QP:%5.2f  size:%6.0f\n",
3500                           slice_type_to_char[i_slice],
3501                           i_count,
3502                           h->stat.f_frame_qp[i_slice] / i_count,
3503                           (double)h->stat.i_frame_size[i_slice] / i_count );
3504             }
3505         }
3506     }
3507     if( h->param.i_bframe && h->stat.i_frame_count[SLICE_TYPE_B] )
3508     {
3509         char *p = buf;
3510         int den = 0;
3511         // weight by number of frames (including the I/P-frames) that are in a sequence of N B-frames
3512         for( int i = 0; i <= h->param.i_bframe; i++ )
3513             den += (i+1) * h->stat.i_consecutive_bframes[i];
3514         for( int i = 0; i <= h->param.i_bframe; i++ )
3515             p += sprintf( p, " %4.1f%%", 100. * (i+1) * h->stat.i_consecutive_bframes[i] / den );
3516         x264_log( h, X264_LOG_INFO, "consecutive B-frames:%s\n", buf );
3517     }
3518
3519     for( int i_type = 0; i_type < 2; i_type++ )
3520         for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
3521         {
3522             if( i == D_DIRECT_8x8 ) continue; /* direct is counted as its own type */
3523             i_mb_count_size[i_type][x264_mb_partition_pixel_table[i]] += h->stat.i_mb_partition[i_type][i];
3524         }
3525
3526     /* MB types used */
3527     if( h->stat.i_frame_count[SLICE_TYPE_I] > 0 )
3528     {
3529         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_I];
3530         double i_count = h->stat.i_frame_count[SLICE_TYPE_I] * h->mb.i_mb_count / 100.0;
3531         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
3532         x264_log( h, X264_LOG_INFO, "mb I  %s\n", buf );
3533     }
3534     if( h->stat.i_frame_count[SLICE_TYPE_P] > 0 )
3535     {
3536         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_P];
3537         double i_count = h->stat.i_frame_count[SLICE_TYPE_P] * h->mb.i_mb_count / 100.0;
3538         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_P];
3539         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
3540         x264_log( h, X264_LOG_INFO,
3541                   "mb P  %s  P16..4: %4.1f%% %4.1f%% %4.1f%% %4.1f%% %4.1f%%    skip:%4.1f%%\n",
3542                   buf,
3543                   i_mb_size[PIXEL_16x16] / (i_count*4),
3544                   (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
3545                   i_mb_size[PIXEL_8x8] / (i_count*4),
3546                   (i_mb_size[PIXEL_8x4] + i_mb_size[PIXEL_4x8]) / (i_count*4),
3547                   i_mb_size[PIXEL_4x4] / (i_count*4),
3548                   i_mb_count[P_SKIP] / i_count );
3549     }
3550     if( h->stat.i_frame_count[SLICE_TYPE_B] > 0 )
3551     {
3552         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_B];
3553         double i_count = h->stat.i_frame_count[SLICE_TYPE_B] * h->mb.i_mb_count / 100.0;
3554         double i_mb_list_count;
3555         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_B];
3556         int64_t list_count[3] = {0}; /* 0 == L0, 1 == L1, 2 == BI */
3557         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
3558         for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
3559             for( int j = 0; j < 2; j++ )
3560             {
3561                 int l0 = x264_mb_type_list_table[i][0][j];
3562                 int l1 = x264_mb_type_list_table[i][1][j];
3563                 if( l0 || l1 )
3564                     list_count[l1+l0*l1] += h->stat.i_mb_count[SLICE_TYPE_B][i] * 2;
3565             }
3566         list_count[0] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L0_8x8];
3567         list_count[1] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L1_8x8];
3568         list_count[2] += h->stat.i_mb_partition[SLICE_TYPE_B][D_BI_8x8];
3569         i_mb_count[B_DIRECT] += (h->stat.i_mb_partition[SLICE_TYPE_B][D_DIRECT_8x8]+2)/4;
3570         i_mb_list_count = (list_count[0] + list_count[1] + list_count[2]) / 100.0;
3571         sprintf( buf + strlen(buf), "  B16..8: %4.1f%% %4.1f%% %4.1f%%  direct:%4.1f%%  skip:%4.1f%%",
3572                  i_mb_size[PIXEL_16x16] / (i_count*4),
3573                  (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
3574                  i_mb_size[PIXEL_8x8] / (i_count*4),
3575                  i_mb_count[B_DIRECT] / i_count,
3576                  i_mb_count[B_SKIP]   / i_count );
3577         if( i_mb_list_count != 0 )
3578             sprintf( buf + strlen(buf), "  L0:%4.1f%% L1:%4.1f%% BI:%4.1f%%",
3579                      list_count[0] / i_mb_list_count,
3580                      list_count[1] / i_mb_list_count,
3581                      list_count[2] / i_mb_list_count );
3582         x264_log( h, X264_LOG_INFO, "mb B  %s\n", buf );
3583     }
3584
3585     x264_ratecontrol_summary( h );
3586
3587     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 )
3588     {
3589 #define SUM3(p) (p[SLICE_TYPE_I] + p[SLICE_TYPE_P] + p[SLICE_TYPE_B])
3590 #define SUM3b(p,o) (p[SLICE_TYPE_I][o] + p[SLICE_TYPE_P][o] + p[SLICE_TYPE_B][o])
3591         int64_t i_i8x8 = SUM3b( h->stat.i_mb_count, I_8x8 );
3592         int64_t i_intra = i_i8x8 + SUM3b( h->stat.i_mb_count, I_4x4 )
3593                                  + SUM3b( h->stat.i_mb_count, I_16x16 );
3594         int64_t i_all_intra = i_intra + SUM3b( h->stat.i_mb_count, I_PCM);
3595         int64_t i_skip = SUM3b( h->stat.i_mb_count, P_SKIP )
3596                        + SUM3b( h->stat.i_mb_count, B_SKIP );
3597         const int i_count = h->stat.i_frame_count[SLICE_TYPE_I] +
3598                             h->stat.i_frame_count[SLICE_TYPE_P] +
3599                             h->stat.i_frame_count[SLICE_TYPE_B];
3600         int64_t i_mb_count = (int64_t)i_count * h->mb.i_mb_count;
3601         int64_t i_inter = i_mb_count - i_skip - i_intra;
3602         const double duration = h->stat.f_frame_duration[SLICE_TYPE_I] +
3603                                 h->stat.f_frame_duration[SLICE_TYPE_P] +
3604                                 h->stat.f_frame_duration[SLICE_TYPE_B];
3605         float f_bitrate = SUM3(h->stat.i_frame_size) / duration / 125;
3606
3607         if( PARAM_INTERLACED )
3608         {
3609             char *fieldstats = buf;
3610             fieldstats[0] = 0;
3611             if( i_inter )
3612                 fieldstats += sprintf( fieldstats, " inter:%.1f%%", h->stat.i_mb_field[1] * 100.0 / i_inter );
3613             if( i_skip )
3614                 fieldstats += sprintf( fieldstats, " skip:%.1f%%", h->stat.i_mb_field[2] * 100.0 / i_skip );
3615             x264_log( h, X264_LOG_INFO, "field mbs: intra: %.1f%%%s\n",
3616                       h->stat.i_mb_field[0] * 100.0 / i_intra, buf );
3617         }
3618
3619         if( h->pps->b_transform_8x8_mode )
3620         {
3621             buf[0] = 0;
3622             if( h->stat.i_mb_count_8x8dct[0] )
3623                 sprintf( buf, " inter:%.1f%%", 100. * h->stat.i_mb_count_8x8dct[1] / h->stat.i_mb_count_8x8dct[0] );
3624             x264_log( h, X264_LOG_INFO, "8x8 transform intra:%.1f%%%s\n", 100. * i_i8x8 / i_intra, buf );
3625         }
3626
3627         if( (h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO ||
3628             (h->stat.i_direct_frames[0] && h->stat.i_direct_frames[1]))
3629             && h->stat.i_frame_count[SLICE_TYPE_B] )
3630         {
3631             x264_log( h, X264_LOG_INFO, "direct mvs  spatial:%.1f%% temporal:%.1f%%\n",
3632                       h->stat.i_direct_frames[1] * 100. / h->stat.i_frame_count[SLICE_TYPE_B],
3633                       h->stat.i_direct_frames[0] * 100. / h->stat.i_frame_count[SLICE_TYPE_B] );
3634         }
3635
3636         buf[0] = 0;
3637         int csize = CHROMA444 ? 4 : 1;
3638         if( i_mb_count != i_all_intra )
3639             sprintf( buf, " inter: %.1f%% %.1f%% %.1f%%",
3640                      h->stat.i_mb_cbp[1] * 100.0 / ((i_mb_count - i_all_intra)*4),
3641                      h->stat.i_mb_cbp[3] * 100.0 / ((i_mb_count - i_all_intra)*csize),
3642                      h->stat.i_mb_cbp[5] * 100.0 / ((i_mb_count - i_all_intra)*csize) );
3643         x264_log( h, X264_LOG_INFO, "coded y,%s,%s intra: %.1f%% %.1f%% %.1f%%%s\n",
3644                   CHROMA444?"u":"uvDC", CHROMA444?"v":"uvAC",
3645                   h->stat.i_mb_cbp[0] * 100.0 / (i_all_intra*4),
3646                   h->stat.i_mb_cbp[2] * 100.0 / (i_all_intra*csize),
3647                   h->stat.i_mb_cbp[4] * 100.0 / (i_all_intra*csize), buf );
3648
3649         int64_t fixed_pred_modes[4][9] = {{0}};
3650         int64_t sum_pred_modes[4] = {0};
3651         for( int i = 0; i <= I_PRED_16x16_DC_128; i++ )
3652         {
3653             fixed_pred_modes[0][x264_mb_pred_mode16x16_fix[i]] += h->stat.i_mb_pred_mode[0][i];
3654             sum_pred_modes[0] += h->stat.i_mb_pred_mode[0][i];
3655         }
3656         if( sum_pred_modes[0] )
3657             x264_log( h, X264_LOG_INFO, "i16 v,h,dc,p: %2.0f%% %2.0f%% %2.0f%% %2.0f%%\n",
3658                       fixed_pred_modes[0][0] * 100.0 / sum_pred_modes[0],
3659                       fixed_pred_modes[0][1] * 100.0 / sum_pred_modes[0],
3660                       fixed_pred_modes[0][2] * 100.0 / sum_pred_modes[0],
3661                       fixed_pred_modes[0][3] * 100.0 / sum_pred_modes[0] );
3662         for( int i = 1; i <= 2; i++ )
3663         {
3664             for( int j = 0; j <= I_PRED_8x8_DC_128; j++ )
3665             {
3666                 fixed_pred_modes[i][x264_mb_pred_mode4x4_fix(j)] += h->stat.i_mb_pred_mode[i][j];
3667                 sum_pred_modes[i] += h->stat.i_mb_pred_mode[i][j];
3668             }
3669             if( sum_pred_modes[i] )
3670                 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,
3671                           fixed_pred_modes[i][0] * 100.0 / sum_pred_modes[i],
3672                           fixed_pred_modes[i][1] * 100.0 / sum_pred_modes[i],
3673                           fixed_pred_modes[i][2] * 100.0 / sum_pred_modes[i],
3674                           fixed_pred_modes[i][3] * 100.0 / sum_pred_modes[i],
3675                           fixed_pred_modes[i][4] * 100.0 / sum_pred_modes[i],
3676                           fixed_pred_modes[i][5] * 100.0 / sum_pred_modes[i],
3677                           fixed_pred_modes[i][6] * 100.0 / sum_pred_modes[i],
3678                           fixed_pred_modes[i][7] * 100.0 / sum_pred_modes[i],
3679                           fixed_pred_modes[i][8] * 100.0 / sum_pred_modes[i] );
3680         }
3681         for( int i = 0; i <= I_PRED_CHROMA_DC_128; i++ )
3682         {
3683             fixed_pred_modes[3][x264_mb_chroma_pred_mode_fix[i]] += h->stat.i_mb_pred_mode[3][i];
3684             sum_pred_modes[3] += h->stat.i_mb_pred_mode[3][i];
3685         }
3686         if( sum_pred_modes[3] && !CHROMA444 )
3687             x264_log( h, X264_LOG_INFO, "i8c dc,h,v,p: %2.0f%% %2.0f%% %2.0f%% %2.0f%%\n",
3688                       fixed_pred_modes[3][0] * 100.0 / sum_pred_modes[3],
3689                       fixed_pred_modes[3][1] * 100.0 / sum_pred_modes[3],
3690                       fixed_pred_modes[3][2] * 100.0 / sum_pred_modes[3],
3691                       fixed_pred_modes[3][3] * 100.0 / sum_pred_modes[3] );
3692
3693         if( h->param.analyse.i_weighted_pred >= X264_WEIGHTP_SIMPLE && h->stat.i_frame_count[SLICE_TYPE_P] > 0 )
3694             x264_log( h, X264_LOG_INFO, "Weighted P-Frames: Y:%.1f%% UV:%.1f%%\n",
3695                       h->stat.i_wpred[0] * 100.0 / h->stat.i_frame_count[SLICE_TYPE_P],
3696                       h->stat.i_wpred[1] * 100.0 / h->stat.i_frame_count[SLICE_TYPE_P] );
3697
3698         for( int i_list = 0; i_list < 2; i_list++ )
3699             for( int i_slice = 0; i_slice < 2; i_slice++ )
3700             {
3701                 char *p = buf;
3702                 int64_t i_den = 0;
3703                 int i_max = 0;
3704                 for( int i = 0; i < X264_REF_MAX*2; i++ )
3705                     if( h->stat.i_mb_count_ref[i_slice][i_list][i] )
3706                     {
3707                         i_den += h->stat.i_mb_count_ref[i_slice][i_list][i];
3708                         i_max = i;
3709                     }
3710                 if( i_max == 0 )
3711                     continue;
3712                 for( int i = 0; i <= i_max; i++ )
3713                     p += sprintf( p, " %4.1f%%", 100. * h->stat.i_mb_count_ref[i_slice][i_list][i] / i_den );
3714                 x264_log( h, X264_LOG_INFO, "ref %c L%d:%s\n", "PB"[i_slice], i_list, buf );
3715             }
3716
3717         if( h->param.analyse.b_ssim )
3718         {
3719             float ssim = SUM3( h->stat.f_ssim_mean_y ) / duration;
3720             x264_log( h, X264_LOG_INFO, "SSIM Mean Y:%.7f (%6.3fdb)\n", ssim, x264_ssim( ssim ) );
3721         }
3722         if( h->param.analyse.b_psnr )
3723         {
3724             x264_log( h, X264_LOG_INFO,
3725                       "PSNR Mean Y:%6.3f U:%6.3f V:%6.3f Avg:%6.3f Global:%6.3f kb/s:%.2f\n",
3726                       SUM3( h->stat.f_psnr_mean_y ) / duration,
3727                       SUM3( h->stat.f_psnr_mean_u ) / duration,
3728                       SUM3( h->stat.f_psnr_mean_v ) / duration,
3729                       SUM3( h->stat.f_psnr_average ) / duration,
3730                       x264_psnr( SUM3( h->stat.f_ssd_global ), duration * i_yuv_size ),
3731                       f_bitrate );
3732         }
3733         else
3734             x264_log( h, X264_LOG_INFO, "kb/s:%.2f\n", f_bitrate );
3735     }
3736
3737     /* rc */
3738     x264_ratecontrol_delete( h );
3739
3740     /* param */
3741     if( h->param.rc.psz_stat_out )
3742         free( h->param.rc.psz_stat_out );
3743     if( h->param.rc.psz_stat_in )
3744         free( h->param.rc.psz_stat_in );
3745
3746     x264_cqm_delete( h );
3747     x264_free( h->nal_buffer );
3748     x264_analyse_free_costs( h );
3749
3750     if( h->i_thread_frames > 1 )
3751         h = h->thread[h->i_thread_phase];
3752
3753     /* frames */
3754     x264_frame_delete_list( h->frames.unused[0] );
3755     x264_frame_delete_list( h->frames.unused[1] );
3756     x264_frame_delete_list( h->frames.current );
3757     x264_frame_delete_list( h->frames.blank_unused );
3758
3759     h = h->thread[0];
3760
3761     for( int i = 0; i < h->i_thread_frames; i++ )
3762         if( h->thread[i]->b_thread_active )
3763             for( int j = 0; j < h->thread[i]->i_ref[0]; j++ )
3764                 if( h->thread[i]->fref[0][j] && h->thread[i]->fref[0][j]->b_duplicate )
3765                     x264_frame_delete( h->thread[i]->fref[0][j] );
3766
3767     for( int i = h->param.i_threads - 1; i >= 0; i-- )
3768     {
3769         x264_frame_t **frame;
3770
3771         if( !h->param.b_sliced_threads || i == 0 )
3772         {
3773             for( frame = h->thread[i]->frames.reference; *frame; frame++ )
3774             {
3775                 assert( (*frame)->i_reference_count > 0 );
3776                 (*frame)->i_reference_count--;
3777                 if( (*frame)->i_reference_count == 0 )
3778                     x264_frame_delete( *frame );
3779             }
3780             frame = &h->thread[i]->fdec;
3781             if( *frame )
3782             {
3783                 assert( (*frame)->i_reference_count > 0 );
3784                 (*frame)->i_reference_count--;
3785                 if( (*frame)->i_reference_count == 0 )
3786                     x264_frame_delete( *frame );
3787             }
3788             x264_macroblock_cache_free( h->thread[i] );
3789         }
3790         x264_macroblock_thread_free( h->thread[i], 0 );
3791         x264_free( h->thread[i]->out.p_bitstream );
3792         x264_free( h->thread[i]->out.nal );
3793         x264_pthread_mutex_destroy( &h->thread[i]->mutex );
3794         x264_pthread_cond_destroy( &h->thread[i]->cv );
3795         x264_free( h->thread[i] );
3796     }
3797 }
3798
3799 int x264_encoder_delayed_frames( x264_t *h )
3800 {
3801     int delayed_frames = 0;
3802     if( h->i_thread_frames > 1 )
3803     {
3804         for( int i = 0; i < h->i_thread_frames; i++ )
3805             delayed_frames += h->thread[i]->b_thread_active;
3806         h = h->thread[h->i_thread_phase];
3807     }
3808     for( int i = 0; h->frames.current[i]; i++ )
3809         delayed_frames++;
3810     x264_pthread_mutex_lock( &h->lookahead->ofbuf.mutex );
3811     x264_pthread_mutex_lock( &h->lookahead->ifbuf.mutex );
3812     x264_pthread_mutex_lock( &h->lookahead->next.mutex );
3813     delayed_frames += h->lookahead->ifbuf.i_size + h->lookahead->next.i_size + h->lookahead->ofbuf.i_size;
3814     x264_pthread_mutex_unlock( &h->lookahead->next.mutex );
3815     x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
3816     x264_pthread_mutex_unlock( &h->lookahead->ofbuf.mutex );
3817     return delayed_frames;
3818 }
3819
3820 int x264_encoder_maximum_delayed_frames( x264_t *h )
3821 {
3822     return h->frames.i_delay;
3823 }