AVC-Intra support
[x262.git] / encoder / encoder.c
1 /*****************************************************************************
2  * encoder.c: top-level encoder functions
3  *****************************************************************************
4  * Copyright (C) 2003-2013 x264 project
5  *
6  * Authors: Laurent Aimar <fenrir@via.ecp.fr>
7  *          Loren Merritt <lorenm@u.washington.edu>
8  *          Jason Garrett-Glaser <darkshikari@gmail.com>
9  *
10  * This program is free software; you can redistribute it and/or modify
11  * it under the terms of the GNU General Public License as published by
12  * the Free Software Foundation; either version 2 of the License, or
13  * (at your option) any later version.
14  *
15  * This program is distributed in the hope that it will be useful,
16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
18  * GNU General Public License for more details.
19  *
20  * You should have received a copy of the GNU General Public License
21  * along with this program; if not, write to the Free Software
22  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02111, USA.
23  *
24  * This program is also available under a commercial proprietary license.
25  * For more information, contact us at licensing@x264.com.
26  *****************************************************************************/
27
28 #include "common/common.h"
29
30 #include "set.h"
31 #include "analyse.h"
32 #include "ratecontrol.h"
33 #include "macroblock.h"
34 #include "me.h"
35
36 #if HAVE_VISUALIZE
37 #include "common/visualize.h"
38 #endif
39
40 //#define DEBUG_MB_TYPE
41
42 #define bs_write_ue bs_write_ue_big
43
44 static int x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
45                                    x264_nal_t **pp_nal, int *pi_nal,
46                                    x264_picture_t *pic_out );
47
48 /****************************************************************************
49  *
50  ******************************* x264 libs **********************************
51  *
52  ****************************************************************************/
53 static double x264_psnr( double sqe, double size )
54 {
55     double mse = sqe / (PIXEL_MAX*PIXEL_MAX * size);
56     if( mse <= 0.0000000001 ) /* Max 100dB */
57         return 100;
58
59     return -10.0 * log10( mse );
60 }
61
62 static double x264_ssim( double ssim )
63 {
64     double inv_ssim = 1 - ssim;
65     if( inv_ssim <= 0.0000000001 ) /* Max 100dB */
66         return 100;
67
68     return -10.0 * log10( inv_ssim );
69 }
70
71 static int x264_threadpool_wait_all( x264_t *h )
72 {
73     for( int i = 0; i < h->param.i_threads; i++ )
74         if( h->thread[i]->b_thread_active )
75         {
76             h->thread[i]->b_thread_active = 0;
77             if( (intptr_t)x264_threadpool_wait( h->threadpool, h->thread[i] ) < 0 )
78                 return -1;
79         }
80     return 0;
81 }
82
83 static void x264_frame_dump( x264_t *h )
84 {
85     FILE *f = fopen( h->param.psz_dump_yuv, "r+b" );
86     if( !f )
87         return;
88
89     /* Wait for the threads to finish deblocking */
90     if( h->param.b_sliced_threads )
91         x264_threadpool_wait_all( h );
92
93     /* Write the frame in display order */
94     int frame_size = FRAME_SIZE( h->param.i_height * h->param.i_width * sizeof(pixel) );
95     fseek( f, (uint64_t)h->fdec->i_frame * frame_size, SEEK_SET );
96     for( int p = 0; p < (CHROMA444 ? 3 : 1); p++ )
97         for( int y = 0; y < h->param.i_height; y++ )
98             fwrite( &h->fdec->plane[p][y*h->fdec->i_stride[p]], sizeof(pixel), h->param.i_width, f );
99     if( !CHROMA444 )
100     {
101         int cw = h->param.i_width>>1;
102         int ch = h->param.i_height>>CHROMA_V_SHIFT;
103         pixel *planeu = x264_malloc( (cw*ch*2+32)*sizeof(pixel) );
104         pixel *planev = planeu + cw*ch + 16;
105         h->mc.plane_copy_deinterleave( planeu, cw, planev, cw, h->fdec->plane[1], h->fdec->i_stride[1], cw, ch );
106         fwrite( planeu, 1, cw*ch*sizeof(pixel), f );
107         fwrite( planev, 1, cw*ch*sizeof(pixel), f );
108         x264_free( planeu );
109     }
110     fclose( f );
111 }
112
113 /* Fill "default" values */
114 static void x264_slice_header_init( x264_t *h, x264_slice_header_t *sh,
115                                     x264_sps_t *sps, x264_pps_t *pps,
116                                     int i_idr_pic_id, int i_frame, int i_qp )
117 {
118     x264_param_t *param = &h->param;
119
120     /* First we fill all fields */
121     sh->sps = sps;
122     sh->pps = pps;
123
124     sh->i_first_mb  = 0;
125     sh->i_last_mb   = h->mb.i_mb_count - 1;
126     sh->i_pps_id    = pps->i_id;
127
128     sh->i_frame_num = i_frame;
129
130     sh->b_mbaff = PARAM_INTERLACED;
131     sh->b_field_pic = 0;    /* no field support for now */
132     sh->b_bottom_field = 0; /* not yet used */
133
134     sh->i_idr_pic_id = i_idr_pic_id;
135
136     /* poc stuff, fixed later */
137     sh->i_poc = 0;
138     sh->i_delta_poc_bottom = 0;
139     sh->i_delta_poc[0] = 0;
140     sh->i_delta_poc[1] = 0;
141
142     sh->i_redundant_pic_cnt = 0;
143
144     h->mb.b_direct_auto_write = h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO
145                                 && h->param.i_bframe
146                                 && ( h->param.rc.b_stat_write || !h->param.rc.b_stat_read );
147
148     if( !h->mb.b_direct_auto_read && sh->i_type == SLICE_TYPE_B )
149     {
150         if( h->fref[1][0]->i_poc_l0ref0 == h->fref[0][0]->i_poc )
151         {
152             if( h->mb.b_direct_auto_write )
153                 sh->b_direct_spatial_mv_pred = ( h->stat.i_direct_score[1] > h->stat.i_direct_score[0] );
154             else
155                 sh->b_direct_spatial_mv_pred = ( param->analyse.i_direct_mv_pred == X264_DIRECT_PRED_SPATIAL );
156         }
157         else
158         {
159             h->mb.b_direct_auto_write = 0;
160             sh->b_direct_spatial_mv_pred = 1;
161         }
162     }
163     /* else b_direct_spatial_mv_pred was read from the 2pass statsfile */
164
165     sh->b_num_ref_idx_override = 0;
166     sh->i_num_ref_idx_l0_active = 1;
167     sh->i_num_ref_idx_l1_active = 1;
168
169     sh->b_ref_pic_list_reordering[0] = h->b_ref_reorder[0];
170     sh->b_ref_pic_list_reordering[1] = h->b_ref_reorder[1];
171
172     /* If the ref list isn't in the default order, construct reordering header */
173     for( int list = 0; list < 2; list++ )
174     {
175         if( sh->b_ref_pic_list_reordering[list] )
176         {
177             int pred_frame_num = i_frame;
178             for( int i = 0; i < h->i_ref[list]; i++ )
179             {
180                 int diff = h->fref[list][i]->i_frame_num - pred_frame_num;
181                 sh->ref_pic_list_order[list][i].idc = ( diff > 0 );
182                 sh->ref_pic_list_order[list][i].arg = (abs(diff) - 1) & ((1 << sps->i_log2_max_frame_num) - 1);
183                 pred_frame_num = h->fref[list][i]->i_frame_num;
184             }
185         }
186     }
187
188     sh->i_cabac_init_idc = param->i_cabac_init_idc;
189
190     sh->i_qp = SPEC_QP(i_qp);
191     sh->i_qp_delta = sh->i_qp - pps->i_pic_init_qp;
192     sh->b_sp_for_swidth = 0;
193     sh->i_qs_delta = 0;
194
195     int deblock_thresh = i_qp + 2 * X264_MIN(param->i_deblocking_filter_alphac0, param->i_deblocking_filter_beta);
196     /* If effective qp <= 15, deblocking would have no effect anyway */
197     if( param->b_deblocking_filter && (h->mb.b_variable_qp || 15 < deblock_thresh ) )
198         sh->i_disable_deblocking_filter_idc = param->b_sliced_threads ? 2 : 0;
199     else
200         sh->i_disable_deblocking_filter_idc = 1;
201     sh->i_alpha_c0_offset = param->i_deblocking_filter_alphac0 << 1;
202     sh->i_beta_offset = param->i_deblocking_filter_beta << 1;
203 }
204
205 static void x264_slice_header_write( bs_t *s, x264_slice_header_t *sh, int i_nal_ref_idc )
206 {
207     if( sh->b_mbaff )
208     {
209         int first_x = sh->i_first_mb % sh->sps->i_mb_width;
210         int first_y = sh->i_first_mb / sh->sps->i_mb_width;
211         assert( (first_y&1) == 0 );
212         bs_write_ue( s, (2*first_x + sh->sps->i_mb_width*(first_y&~1) + (first_y&1)) >> 1 );
213     }
214     else
215         bs_write_ue( s, sh->i_first_mb );
216
217     bs_write_ue( s, sh->i_type + 5 );   /* same type things */
218     bs_write_ue( s, sh->i_pps_id );
219     bs_write( s, sh->sps->i_log2_max_frame_num, sh->i_frame_num & ((1<<sh->sps->i_log2_max_frame_num)-1) );
220
221     if( !sh->sps->b_frame_mbs_only )
222     {
223         bs_write1( s, sh->b_field_pic );
224         if( sh->b_field_pic )
225             bs_write1( s, sh->b_bottom_field );
226     }
227
228     if( sh->i_idr_pic_id >= 0 ) /* NAL IDR */
229         bs_write_ue( s, sh->i_idr_pic_id );
230
231     if( sh->sps->i_poc_type == 0 )
232     {
233         bs_write( s, sh->sps->i_log2_max_poc_lsb, sh->i_poc & ((1<<sh->sps->i_log2_max_poc_lsb)-1) );
234         if( sh->pps->b_pic_order && !sh->b_field_pic )
235             bs_write_se( s, sh->i_delta_poc_bottom );
236     }
237
238     if( sh->pps->b_redundant_pic_cnt )
239         bs_write_ue( s, sh->i_redundant_pic_cnt );
240
241     if( sh->i_type == SLICE_TYPE_B )
242         bs_write1( s, sh->b_direct_spatial_mv_pred );
243
244     if( sh->i_type == SLICE_TYPE_P || sh->i_type == SLICE_TYPE_B )
245     {
246         bs_write1( s, sh->b_num_ref_idx_override );
247         if( sh->b_num_ref_idx_override )
248         {
249             bs_write_ue( s, sh->i_num_ref_idx_l0_active - 1 );
250             if( sh->i_type == SLICE_TYPE_B )
251                 bs_write_ue( s, sh->i_num_ref_idx_l1_active - 1 );
252         }
253     }
254
255     /* ref pic list reordering */
256     if( sh->i_type != SLICE_TYPE_I )
257     {
258         bs_write1( s, sh->b_ref_pic_list_reordering[0] );
259         if( sh->b_ref_pic_list_reordering[0] )
260         {
261             for( int i = 0; i < sh->i_num_ref_idx_l0_active; i++ )
262             {
263                 bs_write_ue( s, sh->ref_pic_list_order[0][i].idc );
264                 bs_write_ue( s, sh->ref_pic_list_order[0][i].arg );
265             }
266             bs_write_ue( s, 3 );
267         }
268     }
269     if( sh->i_type == SLICE_TYPE_B )
270     {
271         bs_write1( s, sh->b_ref_pic_list_reordering[1] );
272         if( sh->b_ref_pic_list_reordering[1] )
273         {
274             for( int i = 0; i < sh->i_num_ref_idx_l1_active; i++ )
275             {
276                 bs_write_ue( s, sh->ref_pic_list_order[1][i].idc );
277                 bs_write_ue( s, sh->ref_pic_list_order[1][i].arg );
278             }
279             bs_write_ue( s, 3 );
280         }
281     }
282
283     sh->b_weighted_pred = 0;
284     if( sh->pps->b_weighted_pred && sh->i_type == SLICE_TYPE_P )
285     {
286         sh->b_weighted_pred = sh->weight[0][0].weightfn || sh->weight[0][1].weightfn || sh->weight[0][2].weightfn;
287         /* pred_weight_table() */
288         bs_write_ue( s, sh->weight[0][0].i_denom );
289         bs_write_ue( s, sh->weight[0][1].i_denom );
290         for( int i = 0; i < sh->i_num_ref_idx_l0_active; i++ )
291         {
292             int luma_weight_l0_flag = !!sh->weight[i][0].weightfn;
293             int chroma_weight_l0_flag = !!sh->weight[i][1].weightfn || !!sh->weight[i][2].weightfn;
294             bs_write1( s, luma_weight_l0_flag );
295             if( luma_weight_l0_flag )
296             {
297                 bs_write_se( s, sh->weight[i][0].i_scale );
298                 bs_write_se( s, sh->weight[i][0].i_offset );
299             }
300             bs_write1( s, chroma_weight_l0_flag );
301             if( chroma_weight_l0_flag )
302             {
303                 for( int j = 1; j < 3; j++ )
304                 {
305                     bs_write_se( s, sh->weight[i][j].i_scale );
306                     bs_write_se( s, sh->weight[i][j].i_offset );
307                 }
308             }
309         }
310     }
311     else if( sh->pps->b_weighted_bipred == 1 && sh->i_type == SLICE_TYPE_B )
312     {
313       /* TODO */
314     }
315
316     if( i_nal_ref_idc != 0 )
317     {
318         if( sh->i_idr_pic_id >= 0 )
319         {
320             bs_write1( s, 0 );  /* no output of prior pics flag */
321             bs_write1( s, 0 );  /* long term reference flag */
322         }
323         else
324         {
325             bs_write1( s, sh->i_mmco_command_count > 0 ); /* adaptive_ref_pic_marking_mode_flag */
326             if( sh->i_mmco_command_count > 0 )
327             {
328                 for( int i = 0; i < sh->i_mmco_command_count; i++ )
329                 {
330                     bs_write_ue( s, 1 ); /* mark short term ref as unused */
331                     bs_write_ue( s, sh->mmco[i].i_difference_of_pic_nums - 1 );
332                 }
333                 bs_write_ue( s, 0 ); /* end command list */
334             }
335         }
336     }
337
338     if( sh->pps->b_cabac && sh->i_type != SLICE_TYPE_I )
339         bs_write_ue( s, sh->i_cabac_init_idc );
340
341     bs_write_se( s, sh->i_qp_delta );      /* slice qp delta */
342
343     if( sh->pps->b_deblocking_filter_control )
344     {
345         bs_write_ue( s, sh->i_disable_deblocking_filter_idc );
346         if( sh->i_disable_deblocking_filter_idc != 1 )
347         {
348             bs_write_se( s, sh->i_alpha_c0_offset >> 1 );
349             bs_write_se( s, sh->i_beta_offset >> 1 );
350         }
351     }
352 }
353
354 /* If we are within a reasonable distance of the end of the memory allocated for the bitstream, */
355 /* reallocate, adding an arbitrary amount of space. */
356 static int x264_bitstream_check_buffer_internal( x264_t *h, int size, int b_cabac, int i_nal )
357 {
358     if( (b_cabac && (h->cabac.p_end - h->cabac.p < size)) ||
359         (h->out.bs.p_end - h->out.bs.p < size) )
360     {
361         int buf_size = h->out.i_bitstream + size;
362         uint8_t *buf = x264_malloc( buf_size );
363         if( !buf )
364             return -1;
365         int aligned_size = h->out.i_bitstream & ~15;
366         h->mc.memcpy_aligned( buf, h->out.p_bitstream, aligned_size );
367         memcpy( buf + aligned_size, h->out.p_bitstream + aligned_size, h->out.i_bitstream - aligned_size );
368
369         intptr_t delta = buf - h->out.p_bitstream;
370
371         h->out.bs.p_start += delta;
372         h->out.bs.p += delta;
373         h->out.bs.p_end = buf + buf_size;
374
375         h->cabac.p_start += delta;
376         h->cabac.p += delta;
377         h->cabac.p_end = buf + buf_size;
378
379         for( int i = 0; i <= i_nal; i++ )
380             h->out.nal[i].p_payload += delta;
381
382         x264_free( h->out.p_bitstream );
383         h->out.p_bitstream = buf;
384         h->out.i_bitstream = buf_size;
385     }
386     return 0;
387 }
388
389 static int x264_bitstream_check_buffer( x264_t *h )
390 {
391     int max_row_size = (2500 << SLICE_MBAFF) * h->mb.i_mb_width;
392     return x264_bitstream_check_buffer_internal( h, max_row_size, h->param.b_cabac, h->out.i_nal );
393 }
394
395 static int x264_bitstream_check_buffer_filler( x264_t *h, int filler )
396 {
397     filler += 32; // add padding for safety
398     return x264_bitstream_check_buffer_internal( h, filler, 0, -1 );
399 }
400
401 #if HAVE_THREAD
402 static void x264_encoder_thread_init( x264_t *h )
403 {
404     if( h->param.i_sync_lookahead )
405         x264_lower_thread_priority( 10 );
406 }
407 #endif
408
409 /****************************************************************************
410  *
411  ****************************************************************************
412  ****************************** External API*********************************
413  ****************************************************************************
414  *
415  ****************************************************************************/
416
417 static int x264_validate_parameters( x264_t *h, int b_open )
418 {
419 #if HAVE_MMX
420     if( b_open )
421     {
422         int cpuflags = x264_cpu_detect();
423         int fail = 0;
424 #ifdef __SSE__
425         if( !(cpuflags & X264_CPU_SSE) )
426         {
427             x264_log( h, X264_LOG_ERROR, "your cpu does not support SSE1, but x264 was compiled with asm\n");
428             fail = 1;
429         }
430 #else
431         if( !(cpuflags & X264_CPU_MMX2) )
432         {
433             x264_log( h, X264_LOG_ERROR, "your cpu does not support MMXEXT, but x264 was compiled with asm\n");
434             fail = 1;
435         }
436 #endif
437         if( !fail && !(cpuflags & X264_CPU_CMOV) )
438         {
439             x264_log( h, X264_LOG_ERROR, "your cpu does not support CMOV, but x264 was compiled with asm\n");
440             fail = 1;
441         }
442         if( fail )
443         {
444             x264_log( h, X264_LOG_ERROR, "to run x264, recompile without asm (configure --disable-asm)\n");
445             return -1;
446         }
447     }
448 #endif
449
450 #if HAVE_INTERLACED
451     h->param.b_interlaced = !!PARAM_INTERLACED;
452 #else
453     if( h->param.b_interlaced )
454     {
455         x264_log( h, X264_LOG_ERROR, "not compiled with interlaced support\n" );
456         return -1;
457     }
458 #endif
459
460     if( h->param.i_width <= 0 || h->param.i_height <= 0 )
461     {
462         x264_log( h, X264_LOG_ERROR, "invalid width x height (%dx%d)\n",
463                   h->param.i_width, h->param.i_height );
464         return -1;
465     }
466
467     int i_csp = h->param.i_csp & X264_CSP_MASK;
468 #if X264_CHROMA_FORMAT
469     if( CHROMA_FORMAT != CHROMA_420 && i_csp >= X264_CSP_I420 && i_csp <= X264_CSP_NV12 )
470     {
471         x264_log( h, X264_LOG_ERROR, "not compiled with 4:2:0 support\n" );
472         return -1;
473     }
474     else if( CHROMA_FORMAT != CHROMA_422 && i_csp >= X264_CSP_I422 && i_csp <= X264_CSP_NV16 )
475     {
476         x264_log( h, X264_LOG_ERROR, "not compiled with 4:2:2 support\n" );
477         return -1;
478     }
479     else if( CHROMA_FORMAT != CHROMA_444 && i_csp >= X264_CSP_I444 && i_csp <= X264_CSP_RGB )
480     {
481         x264_log( h, X264_LOG_ERROR, "not compiled with 4:4:4 support\n" );
482         return -1;
483     }
484 #endif
485     if( i_csp <= X264_CSP_NONE || i_csp >= X264_CSP_MAX )
486     {
487         x264_log( h, X264_LOG_ERROR, "invalid CSP (only I420/YV12/NV12/I422/YV16/NV16/I444/YV24/BGR/BGRA/RGB supported)\n" );
488         return -1;
489     }
490
491     if( i_csp < X264_CSP_I444 && h->param.i_width % 2 )
492     {
493         x264_log( h, X264_LOG_ERROR, "width not divisible by 2 (%dx%d)\n",
494                   h->param.i_width, h->param.i_height );
495         return -1;
496     }
497
498     if( i_csp < X264_CSP_I422 && PARAM_INTERLACED && h->param.i_height % 4 )
499     {
500         x264_log( h, X264_LOG_ERROR, "height not divisible by 4 (%dx%d)\n",
501                   h->param.i_width, h->param.i_height );
502         return -1;
503     }
504
505     if( (i_csp < X264_CSP_I422 || PARAM_INTERLACED) && h->param.i_height % 2 )
506     {
507         x264_log( h, X264_LOG_ERROR, "height not divisible by 2 (%dx%d)\n",
508                   h->param.i_width, h->param.i_height );
509         return -1;
510     }
511
512     if( (h->param.crop_rect.i_left + h->param.crop_rect.i_right ) >= h->param.i_width ||
513         (h->param.crop_rect.i_top  + h->param.crop_rect.i_bottom) >= h->param.i_height )
514     {
515         x264_log( h, X264_LOG_ERROR, "invalid crop-rect %u,%u,%u,%u\n", h->param.crop_rect.i_left,
516                   h->param.crop_rect.i_top, h->param.crop_rect.i_right,  h->param.crop_rect.i_bottom );
517         return -1;
518     }
519
520     if( h->param.i_threads == X264_THREADS_AUTO )
521         h->param.i_threads = x264_cpu_num_processors() * (h->param.b_sliced_threads?2:3)/2;
522     int max_sliced_threads = X264_MAX( 1, (h->param.i_height+15)/16 / 4 );
523     if( h->param.i_threads > 1 )
524     {
525 #if !HAVE_THREAD
526         x264_log( h, X264_LOG_WARNING, "not compiled with thread support!\n");
527         h->param.i_threads = 1;
528 #endif
529         /* Avoid absurdly small thread slices as they can reduce performance
530          * and VBV compliance.  Capped at an arbitrary 4 rows per thread. */
531         if( h->param.b_sliced_threads )
532             h->param.i_threads = X264_MIN( h->param.i_threads, max_sliced_threads );
533     }
534     h->param.i_threads = x264_clip3( h->param.i_threads, 1, X264_THREAD_MAX );
535     if( h->param.i_threads == 1 )
536     {
537         h->param.b_sliced_threads = 0;
538         h->param.i_lookahead_threads = 1;
539     }
540     h->i_thread_frames = h->param.b_sliced_threads ? 1 : h->param.i_threads;
541     if( h->i_thread_frames > 1 )
542         h->param.nalu_process = NULL;
543
544     if( h->param.b_opencl )
545     {
546 #if !HAVE_OPENCL
547         x264_log( h, X264_LOG_WARNING, "OpenCL: not compiled with OpenCL support, disabling\n" );
548         h->param.b_opencl = 0;
549 #elif BIT_DEPTH > 8
550         x264_log( h, X264_LOG_WARNING, "OpenCL lookahead does not support high bit depth, disabling opencl\n" );
551         h->param.b_opencl = 0;
552 #else
553         if( h->param.i_width < 32 || h->param.i_height < 32 )
554         {
555             x264_log( h, X264_LOG_WARNING, "OpenCL: frame size is too small, disabling opencl\n" );
556             h->param.b_opencl = 0;
557         }
558 #endif
559         if( h->param.opencl_device_id && h->param.i_opencl_device )
560         {
561             x264_log( h, X264_LOG_WARNING, "OpenCL: device id and device skip count configured; dropping skip\n" );
562             h->param.i_opencl_device = 0;
563         }
564     }
565
566     h->param.i_keyint_max = x264_clip3( h->param.i_keyint_max, 1, X264_KEYINT_MAX_INFINITE );
567     if( h->param.i_keyint_max == 1 )
568     {
569         h->param.b_intra_refresh = 0;
570         h->param.analyse.i_weighted_pred = 0;
571         h->param.i_frame_reference = 1;
572         h->param.i_dpb_size = 1;
573     }
574
575     h->param.i_frame_packing = x264_clip3( h->param.i_frame_packing, -1, 5 );
576
577     /* Detect default ffmpeg settings and terminate with an error. */
578     if( b_open )
579     {
580         int score = 0;
581         score += h->param.analyse.i_me_range == 0;
582         score += h->param.rc.i_qp_step == 3;
583         score += h->param.i_keyint_max == 12;
584         score += h->param.rc.i_qp_min == 2;
585         score += h->param.rc.i_qp_max == 31;
586         score += h->param.rc.f_qcompress == 0.5;
587         score += fabs(h->param.rc.f_ip_factor - 1.25) < 0.01;
588         score += fabs(h->param.rc.f_pb_factor - 1.25) < 0.01;
589         score += h->param.analyse.inter == 0 && h->param.analyse.i_subpel_refine == 8;
590         if( score >= 5 )
591         {
592             x264_log( h, X264_LOG_ERROR, "broken ffmpeg default settings detected\n" );
593             x264_log( h, X264_LOG_ERROR, "use an encoding preset (e.g. -vpre medium)\n" );
594             x264_log( h, X264_LOG_ERROR, "preset usage: -vpre <speed> -vpre <profile>\n" );
595             x264_log( h, X264_LOG_ERROR, "speed presets are listed in x264 --help\n" );
596             x264_log( h, X264_LOG_ERROR, "profile is optional; x264 defaults to high\n" );
597             return -1;
598         }
599     }
600
601     if( h->param.rc.i_rc_method < 0 || h->param.rc.i_rc_method > 2 )
602     {
603         x264_log( h, X264_LOG_ERROR, "no ratecontrol method specified\n" );
604         return -1;
605     }
606
607     if( PARAM_INTERLACED )
608         h->param.b_pic_struct = 1;
609
610     if( h->param.b_avcintra_compat )
611     {
612         if( BIT_DEPTH != 10 )
613         {
614             x264_log( h, X264_LOG_ERROR, "%2d-bit AVC-Intra is not widely compatible\n", BIT_DEPTH );
615             x264_log( h, X264_LOG_ERROR, "10-bit x264 is required to encode AVC-Intra\n" );
616             return -1;
617         }
618
619         /* [50/100][res][fps] */
620         static const struct
621         {
622             uint16_t fps_num;
623             uint16_t fps_den;
624             uint8_t interlaced;
625             uint16_t frame_size;
626             const uint8_t *cqm_4ic;
627             const uint8_t *cqm_8iy;
628         } avcintra_lut[2][2][5] =
629         {
630             {{{ 60000, 1001, 0,  912, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy },
631               {    50,    1, 0, 1100, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy },
632               { 30000, 1001, 0,  912, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy },
633               {    25,    1, 0, 1100, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy },
634               { 24000, 1001, 0,  912, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy }},
635              {{ 30000, 1001, 1, 1820, x264_cqm_avci50_4ic, x264_cqm_avci50_1080i_8iy },
636               {    25,    1, 1, 2196, x264_cqm_avci50_4ic, x264_cqm_avci50_1080i_8iy },
637               { 30000, 1001, 0, 1820, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy },
638               {    25,    1, 0, 2196, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy },
639               { 24000, 1001, 0, 1820, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy }}},
640             {{{ 60000, 1001, 0, 1848, x264_cqm_avci100_720p_4ic, x264_cqm_avci100_720p_8iy },
641               {    50,    1, 0, 2224, x264_cqm_avci100_720p_4ic, x264_cqm_avci100_720p_8iy },
642               { 30000, 1001, 0, 1848, x264_cqm_avci100_720p_4ic, x264_cqm_avci100_720p_8iy },
643               {    25,    1, 0, 2224, x264_cqm_avci100_720p_4ic, x264_cqm_avci100_720p_8iy },
644               { 24000, 1001, 0, 1848, x264_cqm_avci100_720p_4ic, x264_cqm_avci100_720p_8iy }},
645              {{ 30000, 1001, 1, 3692, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080i_8iy },
646               {    25,    1, 1, 4444, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080i_8iy },
647               { 30000, 1001, 0, 3692, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080p_8iy },
648               {    25,    1, 0, 4444, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080p_8iy },
649               { 24000, 1001, 0, 3692, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080p_8iy }}}
650         };
651
652         int res = -1;
653         int type = i_csp == X264_CSP_I422;
654         if( type )
655         {
656             if(      h->param.i_width == 1920 && h->param.i_height == 1080 ) res =  1;
657             else if( h->param.i_width == 1280 && h->param.i_height ==  720 ) res =  0;
658         }
659         else if( i_csp == X264_CSP_I420 )
660         {
661             if(      h->param.i_width == 1440 && h->param.i_height == 1080 ) res =  1;
662             else if( h->param.i_width ==  960 && h->param.i_height ==  720 ) res =  0;
663         }
664         else
665         {
666             x264_log( h, X264_LOG_ERROR, "Invalid colorspace for AVC-Intra\n" );
667             return -1;
668         }
669
670         if( res < 0 )
671         {
672             x264_log( h, X264_LOG_ERROR, "Resolution %dx%d invalid for AVC-Intra %s\n",
673                       h->param.i_width, h->param.i_height, type ? "100" : "50" );
674             return -1;
675         }
676
677         if( h->param.nalu_process )
678         {
679             x264_log( h, X264_LOG_ERROR, "nalu_process is not supported in AVC-Intra mode\n" );
680             return -1;
681         }
682
683         if( !h->param.b_repeat_headers )
684         {
685             x264_log( h, X264_LOG_ERROR, "Separate headers not supported in AVC-Intra mode\n" );
686             return -1;
687         }
688
689         int i;
690         uint32_t fps_num = h->param.i_fps_num, fps_den = h->param.i_fps_den;
691         x264_reduce_fraction( &fps_num, &fps_den );
692         for( i = 0; i < 5; i++ )
693         {
694             if( avcintra_lut[type][res][i].fps_num == fps_num &&
695                 avcintra_lut[type][res][i].fps_den == fps_den &&
696                 avcintra_lut[type][res][i].interlaced == PARAM_INTERLACED )
697             {
698                 break;
699             }
700         }
701         if( i == 5 )
702         {
703             x264_log( h, X264_LOG_ERROR, "FPS %d/%d%c not compatible with AVC-Intra\n",
704                       h->param.i_fps_num, h->param.i_fps_den, PARAM_INTERLACED ? 'i' : 'p' );
705             return -1;
706         }
707
708         h->param.i_keyint_max = 1;
709         h->param.b_intra_refresh = 0;
710         h->param.analyse.i_weighted_pred = 0;
711         h->param.i_frame_reference = 1;
712         h->param.i_dpb_size = 1;
713
714         h->param.b_bluray_compat = 0;
715         h->param.b_vfr_input = 0;
716         h->param.b_aud = 1;
717         h->param.vui.i_chroma_loc = 0;
718         h->param.i_nal_hrd = X264_NAL_HRD_NONE;
719         h->param.b_deblocking_filter = 0;
720         h->param.b_stitchable = 1;
721         h->param.b_pic_struct = 0;
722         h->param.analyse.b_transform_8x8 = 1;
723         h->param.analyse.intra = X264_ANALYSE_I8x8;
724         h->param.analyse.i_chroma_qp_offset = res && type ? 3 : 4;
725         h->param.b_cabac = !type;
726         h->param.rc.i_vbv_buffer_size = avcintra_lut[type][res][i].frame_size;
727         h->param.rc.i_vbv_max_bitrate =
728         h->param.rc.i_bitrate = h->param.rc.i_vbv_buffer_size * fps_num / fps_den;
729         h->param.rc.i_rc_method = X264_RC_ABR;
730         h->param.rc.f_vbv_buffer_init = 1.0;
731         h->param.i_cqm_preset = X264_CQM_CUSTOM;
732         memcpy( h->param.cqm_4iy, x264_cqm_jvt4i, sizeof(h->param.cqm_4iy) );
733         memcpy( h->param.cqm_4ic, avcintra_lut[type][res][i].cqm_4ic, sizeof(h->param.cqm_4ic) );
734         memcpy( h->param.cqm_8iy, avcintra_lut[type][res][i].cqm_8iy, sizeof(h->param.cqm_8iy) );
735
736         /* Need exactly 10 slices of equal MB count... why?  $deity knows... */
737         h->param.i_slice_max_mbs = ((h->param.i_width + 15) / 16) * ((h->param.i_height + 15) / 16) / 10;
738         h->param.i_slice_max_size = 0;
739         /* The slice structure only allows a maximum of 2 threads for 1080i/p
740          * and 1 or 5 threads for 720p */
741         if( h->param.b_sliced_threads )
742         {
743             if( res )
744                 h->param.i_threads = X264_MIN( 2, h->param.i_threads );
745             else
746             {
747                 h->param.i_threads = X264_MIN( 5, h->param.i_threads );
748                 if( h->param.i_threads < 5 )
749                     h->param.i_threads = 1;
750             }
751         }
752
753         if( type )
754             h->param.vui.i_sar_width = h->param.vui.i_sar_height = 1;
755         else
756         {
757             h->param.vui.i_sar_width  = 4;
758             h->param.vui.i_sar_height = 3;
759         }
760
761         /* Avid cannot handle negative QPs */
762         h->param.rc.i_qp_min = X264_MAX( h->param.rc.i_qp_min, QP_BD_OFFSET );
763     }
764
765     h->param.rc.f_rf_constant = x264_clip3f( h->param.rc.f_rf_constant, -QP_BD_OFFSET, 51 );
766     h->param.rc.f_rf_constant_max = x264_clip3f( h->param.rc.f_rf_constant_max, -QP_BD_OFFSET, 51 );
767     h->param.rc.i_qp_constant = x264_clip3( h->param.rc.i_qp_constant, 0, QP_MAX );
768     h->param.analyse.i_subpel_refine = x264_clip3( h->param.analyse.i_subpel_refine, 0, 11 );
769     h->param.rc.f_ip_factor = X264_MAX( h->param.rc.f_ip_factor, 0.01f );
770     h->param.rc.f_pb_factor = X264_MAX( h->param.rc.f_pb_factor, 0.01f );
771     if( h->param.rc.i_rc_method == X264_RC_CRF )
772     {
773         h->param.rc.i_qp_constant = h->param.rc.f_rf_constant + QP_BD_OFFSET;
774         h->param.rc.i_bitrate = 0;
775     }
776     if( b_open && (h->param.rc.i_rc_method == X264_RC_CQP || h->param.rc.i_rc_method == X264_RC_CRF)
777         && h->param.rc.i_qp_constant == 0 )
778     {
779         h->mb.b_lossless = 1;
780         h->param.i_cqm_preset = X264_CQM_FLAT;
781         h->param.psz_cqm_file = NULL;
782         h->param.rc.i_rc_method = X264_RC_CQP;
783         h->param.rc.f_ip_factor = 1;
784         h->param.rc.f_pb_factor = 1;
785         h->param.analyse.b_psnr = 0;
786         h->param.analyse.b_ssim = 0;
787         h->param.analyse.i_chroma_qp_offset = 0;
788         h->param.analyse.i_trellis = 0;
789         h->param.analyse.b_fast_pskip = 0;
790         h->param.analyse.i_noise_reduction = 0;
791         h->param.analyse.b_psy = 0;
792         h->param.i_bframe = 0;
793         /* 8x8dct is not useful without RD in CAVLC lossless */
794         if( !h->param.b_cabac && h->param.analyse.i_subpel_refine < 6 )
795             h->param.analyse.b_transform_8x8 = 0;
796     }
797     if( h->param.rc.i_rc_method == X264_RC_CQP )
798     {
799         float qp_p = h->param.rc.i_qp_constant;
800         float qp_i = qp_p - 6*log2f( h->param.rc.f_ip_factor );
801         float qp_b = qp_p + 6*log2f( h->param.rc.f_pb_factor );
802         h->param.rc.i_qp_min = x264_clip3( (int)(X264_MIN3( qp_p, qp_i, qp_b )), 0, QP_MAX );
803         h->param.rc.i_qp_max = x264_clip3( (int)(X264_MAX3( qp_p, qp_i, qp_b ) + .999), 0, QP_MAX );
804         h->param.rc.i_aq_mode = 0;
805         h->param.rc.b_mb_tree = 0;
806         h->param.rc.i_bitrate = 0;
807     }
808     h->param.rc.i_qp_max = x264_clip3( h->param.rc.i_qp_max, 0, QP_MAX );
809     h->param.rc.i_qp_min = x264_clip3( h->param.rc.i_qp_min, 0, h->param.rc.i_qp_max );
810     h->param.rc.i_qp_step = x264_clip3( h->param.rc.i_qp_step, 2, QP_MAX );
811     h->param.rc.i_bitrate = x264_clip3( h->param.rc.i_bitrate, 0, 2000000 );
812     if( h->param.rc.i_rc_method == X264_RC_ABR && !h->param.rc.i_bitrate )
813     {
814         x264_log( h, X264_LOG_ERROR, "bitrate not specified\n" );
815         return -1;
816     }
817     h->param.rc.i_vbv_buffer_size = x264_clip3( h->param.rc.i_vbv_buffer_size, 0, 2000000 );
818     h->param.rc.i_vbv_max_bitrate = x264_clip3( h->param.rc.i_vbv_max_bitrate, 0, 2000000 );
819     h->param.rc.f_vbv_buffer_init = x264_clip3f( h->param.rc.f_vbv_buffer_init, 0, 2000000 );
820     if( h->param.rc.i_vbv_buffer_size )
821     {
822         if( h->param.rc.i_rc_method == X264_RC_CQP )
823         {
824             x264_log( h, X264_LOG_WARNING, "VBV is incompatible with constant QP, ignored.\n" );
825             h->param.rc.i_vbv_max_bitrate = 0;
826             h->param.rc.i_vbv_buffer_size = 0;
827         }
828         else if( h->param.rc.i_vbv_max_bitrate == 0 )
829         {
830             if( h->param.rc.i_rc_method == X264_RC_ABR )
831             {
832                 x264_log( h, X264_LOG_WARNING, "VBV maxrate unspecified, assuming CBR\n" );
833                 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate;
834             }
835             else
836             {
837                 x264_log( h, X264_LOG_WARNING, "VBV bufsize set but maxrate unspecified, ignored\n" );
838                 h->param.rc.i_vbv_buffer_size = 0;
839             }
840         }
841         else if( h->param.rc.i_vbv_max_bitrate < h->param.rc.i_bitrate &&
842                  h->param.rc.i_rc_method == X264_RC_ABR )
843         {
844             x264_log( h, X264_LOG_WARNING, "max bitrate less than average bitrate, assuming CBR\n" );
845             h->param.rc.i_bitrate = h->param.rc.i_vbv_max_bitrate;
846         }
847     }
848     else if( h->param.rc.i_vbv_max_bitrate )
849     {
850         x264_log( h, X264_LOG_WARNING, "VBV maxrate specified, but no bufsize, ignored\n" );
851         h->param.rc.i_vbv_max_bitrate = 0;
852     }
853
854     h->param.i_slice_max_size = X264_MAX( h->param.i_slice_max_size, 0 );
855     h->param.i_slice_max_mbs = X264_MAX( h->param.i_slice_max_mbs, 0 );
856     h->param.i_slice_min_mbs = X264_MAX( h->param.i_slice_min_mbs, 0 );
857     if( h->param.i_slice_max_mbs )
858         h->param.i_slice_min_mbs = X264_MIN( h->param.i_slice_min_mbs, h->param.i_slice_max_mbs/2 );
859     else if( !h->param.i_slice_max_size )
860         h->param.i_slice_min_mbs = 0;
861     if( PARAM_INTERLACED && h->param.i_slice_min_mbs )
862     {
863         x264_log( h, X264_LOG_WARNING, "interlace + slice-min-mbs is not implemented\n" );
864         h->param.i_slice_min_mbs = 0;
865     }
866     int mb_width = (h->param.i_width+15)/16;
867     if( h->param.i_slice_min_mbs > mb_width )
868     {
869         x264_log( h, X264_LOG_WARNING, "slice-min-mbs > row mb size (%d) not implemented\n", mb_width );
870         h->param.i_slice_min_mbs = mb_width;
871     }
872
873     int max_slices = (h->param.i_height+((16<<PARAM_INTERLACED)-1))/(16<<PARAM_INTERLACED);
874     if( h->param.b_sliced_threads )
875         h->param.i_slice_count = x264_clip3( h->param.i_threads, 0, max_slices );
876     else
877     {
878         h->param.i_slice_count = x264_clip3( h->param.i_slice_count, 0, max_slices );
879         if( h->param.i_slice_max_mbs || h->param.i_slice_max_size )
880             h->param.i_slice_count = 0;
881     }
882     if( h->param.i_slice_count_max > 0 )
883         h->param.i_slice_count_max = X264_MAX( h->param.i_slice_count, h->param.i_slice_count_max );
884
885     if( h->param.b_bluray_compat )
886     {
887         h->param.i_bframe_pyramid = X264_MIN( X264_B_PYRAMID_STRICT, h->param.i_bframe_pyramid );
888         h->param.i_bframe = X264_MIN( h->param.i_bframe, 3 );
889         h->param.b_aud = 1;
890         h->param.i_nal_hrd = X264_MAX( h->param.i_nal_hrd, X264_NAL_HRD_VBR );
891         h->param.i_slice_max_size = 0;
892         h->param.i_slice_max_mbs = 0;
893         h->param.b_intra_refresh = 0;
894         h->param.i_frame_reference = X264_MIN( h->param.i_frame_reference, 6 );
895         h->param.i_dpb_size = X264_MIN( h->param.i_dpb_size, 6 );
896         /* Don't use I-frames, because Blu-ray treats them the same as IDR. */
897         h->param.i_keyint_min = 1;
898         /* Due to the proliferation of broken players that don't handle dupes properly. */
899         h->param.analyse.i_weighted_pred = X264_MIN( h->param.analyse.i_weighted_pred, X264_WEIGHTP_SIMPLE );
900         if( h->param.b_fake_interlaced )
901             h->param.b_pic_struct = 1;
902     }
903
904     h->param.i_frame_reference = x264_clip3( h->param.i_frame_reference, 1, X264_REF_MAX );
905     h->param.i_dpb_size = x264_clip3( h->param.i_dpb_size, 1, X264_REF_MAX );
906     if( h->param.i_scenecut_threshold < 0 )
907         h->param.i_scenecut_threshold = 0;
908     h->param.analyse.i_direct_mv_pred = x264_clip3( h->param.analyse.i_direct_mv_pred, X264_DIRECT_PRED_NONE, X264_DIRECT_PRED_AUTO );
909     if( !h->param.analyse.i_subpel_refine && h->param.analyse.i_direct_mv_pred > X264_DIRECT_PRED_SPATIAL )
910     {
911         x264_log( h, X264_LOG_WARNING, "subme=0 + direct=temporal is not supported\n" );
912         h->param.analyse.i_direct_mv_pred = X264_DIRECT_PRED_SPATIAL;
913     }
914     h->param.i_bframe = x264_clip3( h->param.i_bframe, 0, X264_MIN( X264_BFRAME_MAX, h->param.i_keyint_max-1 ) );
915     h->param.i_bframe_bias = x264_clip3( h->param.i_bframe_bias, -90, 100 );
916     if( h->param.i_bframe <= 1 )
917         h->param.i_bframe_pyramid = X264_B_PYRAMID_NONE;
918     h->param.i_bframe_pyramid = x264_clip3( h->param.i_bframe_pyramid, X264_B_PYRAMID_NONE, X264_B_PYRAMID_NORMAL );
919     h->param.i_bframe_adaptive = x264_clip3( h->param.i_bframe_adaptive, X264_B_ADAPT_NONE, X264_B_ADAPT_TRELLIS );
920     if( !h->param.i_bframe )
921     {
922         h->param.i_bframe_adaptive = X264_B_ADAPT_NONE;
923         h->param.analyse.i_direct_mv_pred = 0;
924         h->param.analyse.b_weighted_bipred = 0;
925         h->param.b_open_gop = 0;
926     }
927     if( h->param.b_intra_refresh && h->param.i_bframe_pyramid == X264_B_PYRAMID_NORMAL )
928     {
929         x264_log( h, X264_LOG_WARNING, "b-pyramid normal + intra-refresh is not supported\n" );
930         h->param.i_bframe_pyramid = X264_B_PYRAMID_STRICT;
931     }
932     if( h->param.b_intra_refresh && (h->param.i_frame_reference > 1 || h->param.i_dpb_size > 1) )
933     {
934         x264_log( h, X264_LOG_WARNING, "ref > 1 + intra-refresh is not supported\n" );
935         h->param.i_frame_reference = 1;
936         h->param.i_dpb_size = 1;
937     }
938     if( h->param.b_intra_refresh && h->param.b_open_gop )
939     {
940         x264_log( h, X264_LOG_WARNING, "intra-refresh is not compatible with open-gop\n" );
941         h->param.b_open_gop = 0;
942     }
943     if( !h->param.i_fps_num || !h->param.i_fps_den )
944     {
945         h->param.i_fps_num = 25;
946         h->param.i_fps_den = 1;
947     }
948     float fps = (float) h->param.i_fps_num / h->param.i_fps_den;
949     if( h->param.i_keyint_min == X264_KEYINT_MIN_AUTO )
950         h->param.i_keyint_min = X264_MIN( h->param.i_keyint_max / 10, fps );
951     h->param.i_keyint_min = x264_clip3( h->param.i_keyint_min, 1, h->param.i_keyint_max/2+1 );
952     h->param.rc.i_lookahead = x264_clip3( h->param.rc.i_lookahead, 0, X264_LOOKAHEAD_MAX );
953     {
954         int maxrate = X264_MAX( h->param.rc.i_vbv_max_bitrate, h->param.rc.i_bitrate );
955         float bufsize = maxrate ? (float)h->param.rc.i_vbv_buffer_size / maxrate : 0;
956         h->param.rc.i_lookahead = X264_MIN( h->param.rc.i_lookahead, X264_MAX( h->param.i_keyint_max, bufsize*fps ) );
957     }
958
959     if( !h->param.i_timebase_num || !h->param.i_timebase_den || !(h->param.b_vfr_input || h->param.b_pulldown) )
960     {
961         h->param.i_timebase_num = h->param.i_fps_den;
962         h->param.i_timebase_den = h->param.i_fps_num;
963     }
964
965     h->param.rc.f_qcompress = x264_clip3f( h->param.rc.f_qcompress, 0.0, 1.0 );
966     if( h->param.i_keyint_max == 1 || h->param.rc.f_qcompress == 1 )
967         h->param.rc.b_mb_tree = 0;
968     if( (!h->param.b_intra_refresh && h->param.i_keyint_max != X264_KEYINT_MAX_INFINITE) &&
969         !h->param.rc.i_lookahead && h->param.rc.b_mb_tree )
970     {
971         x264_log( h, X264_LOG_WARNING, "lookaheadless mb-tree requires intra refresh or infinite keyint\n" );
972         h->param.rc.b_mb_tree = 0;
973     }
974     if( b_open && h->param.rc.b_stat_read )
975         h->param.rc.i_lookahead = 0;
976 #if HAVE_THREAD
977     if( h->param.i_sync_lookahead < 0 )
978         h->param.i_sync_lookahead = h->param.i_bframe + 1;
979     h->param.i_sync_lookahead = X264_MIN( h->param.i_sync_lookahead, X264_LOOKAHEAD_MAX );
980     if( h->param.rc.b_stat_read || h->i_thread_frames == 1 )
981         h->param.i_sync_lookahead = 0;
982 #else
983     h->param.i_sync_lookahead = 0;
984 #endif
985
986     h->param.i_deblocking_filter_alphac0 = x264_clip3( h->param.i_deblocking_filter_alphac0, -6, 6 );
987     h->param.i_deblocking_filter_beta    = x264_clip3( h->param.i_deblocking_filter_beta, -6, 6 );
988     h->param.analyse.i_luma_deadzone[0] = x264_clip3( h->param.analyse.i_luma_deadzone[0], 0, 32 );
989     h->param.analyse.i_luma_deadzone[1] = x264_clip3( h->param.analyse.i_luma_deadzone[1], 0, 32 );
990
991     h->param.i_cabac_init_idc = x264_clip3( h->param.i_cabac_init_idc, 0, 2 );
992
993     if( h->param.i_cqm_preset < X264_CQM_FLAT || h->param.i_cqm_preset > X264_CQM_CUSTOM )
994         h->param.i_cqm_preset = X264_CQM_FLAT;
995
996     if( h->param.analyse.i_me_method < X264_ME_DIA ||
997         h->param.analyse.i_me_method > X264_ME_TESA )
998         h->param.analyse.i_me_method = X264_ME_HEX;
999     h->param.analyse.i_me_range = x264_clip3( h->param.analyse.i_me_range, 4, 1024 );
1000     if( h->param.analyse.i_me_range > 16 && h->param.analyse.i_me_method <= X264_ME_HEX )
1001         h->param.analyse.i_me_range = 16;
1002     if( h->param.analyse.i_me_method == X264_ME_TESA &&
1003         (h->mb.b_lossless || h->param.analyse.i_subpel_refine <= 1) )
1004         h->param.analyse.i_me_method = X264_ME_ESA;
1005     h->param.analyse.b_mixed_references = h->param.analyse.b_mixed_references && h->param.i_frame_reference > 1;
1006     h->param.analyse.inter &= X264_ANALYSE_PSUB16x16|X264_ANALYSE_PSUB8x8|X264_ANALYSE_BSUB16x16|
1007                               X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
1008     h->param.analyse.intra &= X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
1009     if( !(h->param.analyse.inter & X264_ANALYSE_PSUB16x16) )
1010         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
1011     if( !h->param.analyse.b_transform_8x8 )
1012     {
1013         h->param.analyse.inter &= ~X264_ANALYSE_I8x8;
1014         h->param.analyse.intra &= ~X264_ANALYSE_I8x8;
1015     }
1016     h->param.analyse.i_trellis = x264_clip3( h->param.analyse.i_trellis, 0, 2 );
1017     h->param.rc.i_aq_mode = x264_clip3( h->param.rc.i_aq_mode, 0, 2 );
1018     h->param.rc.f_aq_strength = x264_clip3f( h->param.rc.f_aq_strength, 0, 3 );
1019     if( h->param.rc.f_aq_strength == 0 )
1020         h->param.rc.i_aq_mode = 0;
1021
1022     if( h->param.i_log_level < X264_LOG_INFO )
1023     {
1024         h->param.analyse.b_psnr = 0;
1025         h->param.analyse.b_ssim = 0;
1026     }
1027     /* Warn users trying to measure PSNR/SSIM with psy opts on. */
1028     if( b_open && (h->param.analyse.b_psnr || h->param.analyse.b_ssim) )
1029     {
1030         char *s = NULL;
1031
1032         if( h->param.analyse.b_psy )
1033         {
1034             s = h->param.analyse.b_psnr ? "psnr" : "ssim";
1035             x264_log( h, X264_LOG_WARNING, "--%s used with psy on: results will be invalid!\n", s );
1036         }
1037         else if( !h->param.rc.i_aq_mode && h->param.analyse.b_ssim )
1038         {
1039             x264_log( h, X264_LOG_WARNING, "--ssim used with AQ off: results will be invalid!\n" );
1040             s = "ssim";
1041         }
1042         else if(  h->param.rc.i_aq_mode && h->param.analyse.b_psnr )
1043         {
1044             x264_log( h, X264_LOG_WARNING, "--psnr used with AQ on: results will be invalid!\n" );
1045             s = "psnr";
1046         }
1047         if( s )
1048             x264_log( h, X264_LOG_WARNING, "--tune %s should be used if attempting to benchmark %s!\n", s, s );
1049     }
1050
1051     if( !h->param.analyse.b_psy )
1052     {
1053         h->param.analyse.f_psy_rd = 0;
1054         h->param.analyse.f_psy_trellis = 0;
1055     }
1056     h->param.analyse.f_psy_rd = x264_clip3f( h->param.analyse.f_psy_rd, 0, 10 );
1057     h->param.analyse.f_psy_trellis = x264_clip3f( h->param.analyse.f_psy_trellis, 0, 10 );
1058     h->mb.i_psy_rd = h->param.analyse.i_subpel_refine >= 6 ? FIX8( h->param.analyse.f_psy_rd ) : 0;
1059     h->mb.i_psy_trellis = h->param.analyse.i_trellis ? FIX8( h->param.analyse.f_psy_trellis / 4 ) : 0;
1060     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -32, 32);
1061     /* In 4:4:4 mode, chroma gets twice as much resolution, so we can halve its quality. */
1062     if( b_open && i_csp >= X264_CSP_I444 && i_csp < X264_CSP_BGR && h->param.analyse.b_psy )
1063         h->param.analyse.i_chroma_qp_offset += 6;
1064     /* Psy RDO increases overall quantizers to improve the quality of luma--this indirectly hurts chroma quality */
1065     /* so we lower the chroma QP offset to compensate */
1066     if( b_open && h->mb.i_psy_rd && !h->param.b_avcintra_compat )
1067         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_rd < 0.25 ? 1 : 2;
1068     /* Psy trellis has a similar effect. */
1069     if( b_open && h->mb.i_psy_trellis && !h->param.b_avcintra_compat )
1070         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_trellis < 0.25 ? 1 : 2;
1071     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -12, 12);
1072     /* MB-tree requires AQ to be on, even if the strength is zero. */
1073     if( !h->param.rc.i_aq_mode && h->param.rc.b_mb_tree )
1074     {
1075         h->param.rc.i_aq_mode = 1;
1076         h->param.rc.f_aq_strength = 0;
1077     }
1078     h->param.analyse.i_noise_reduction = x264_clip3( h->param.analyse.i_noise_reduction, 0, 1<<16 );
1079     if( h->param.analyse.i_subpel_refine >= 10 && (h->param.analyse.i_trellis != 2 || !h->param.rc.i_aq_mode) )
1080         h->param.analyse.i_subpel_refine = 9;
1081
1082     {
1083         const x264_level_t *l = x264_levels;
1084         if( h->param.i_level_idc < 0 )
1085         {
1086             int maxrate_bak = h->param.rc.i_vbv_max_bitrate;
1087             if( h->param.rc.i_rc_method == X264_RC_ABR && h->param.rc.i_vbv_buffer_size <= 0 )
1088                 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate * 2;
1089             x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
1090             do h->param.i_level_idc = l->level_idc;
1091                 while( l[1].level_idc && x264_validate_levels( h, 0 ) && l++ );
1092             h->param.rc.i_vbv_max_bitrate = maxrate_bak;
1093         }
1094         else
1095         {
1096             while( l->level_idc && l->level_idc != h->param.i_level_idc )
1097                 l++;
1098             if( l->level_idc == 0 )
1099             {
1100                 x264_log( h, X264_LOG_ERROR, "invalid level_idc: %d\n", h->param.i_level_idc );
1101                 return -1;
1102             }
1103         }
1104         if( h->param.analyse.i_mv_range <= 0 )
1105             h->param.analyse.i_mv_range = l->mv_range >> PARAM_INTERLACED;
1106         else
1107             h->param.analyse.i_mv_range = x264_clip3(h->param.analyse.i_mv_range, 32, 512 >> PARAM_INTERLACED);
1108     }
1109
1110     h->param.analyse.i_weighted_pred = x264_clip3( h->param.analyse.i_weighted_pred, X264_WEIGHTP_NONE, X264_WEIGHTP_SMART );
1111
1112     if( h->param.i_lookahead_threads == X264_THREADS_AUTO )
1113     {
1114         if( h->param.b_sliced_threads )
1115             h->param.i_lookahead_threads = h->param.i_threads;
1116         else
1117         {
1118             /* If we're using much slower lookahead settings than encoding settings, it helps a lot to use
1119              * more lookahead threads.  This typically happens in the first pass of a two-pass encode, so
1120              * try to guess at this sort of case.
1121              *
1122              * Tuned by a little bit of real encoding with the various presets. */
1123             int badapt = h->param.i_bframe_adaptive == X264_B_ADAPT_TRELLIS;
1124             int subme = X264_MIN( h->param.analyse.i_subpel_refine / 3, 3 ) + (h->param.analyse.i_subpel_refine > 1);
1125             int bframes = X264_MIN( (h->param.i_bframe - 1) / 3, 3 );
1126
1127             /* [b-adapt 0/1 vs 2][quantized subme][quantized bframes] */
1128             static const uint8_t lookahead_thread_div[2][5][4] =
1129             {{{6,6,6,6}, {3,3,3,3}, {4,4,4,4}, {6,6,6,6}, {12,12,12,12}},
1130              {{3,2,1,1}, {2,1,1,1}, {4,3,2,1}, {6,4,3,2}, {12, 9, 6, 4}}};
1131
1132             h->param.i_lookahead_threads = h->param.i_threads / lookahead_thread_div[badapt][subme][bframes];
1133             /* Since too many lookahead threads significantly degrades lookahead accuracy, limit auto
1134              * lookahead threads to about 8 macroblock rows high each at worst.  This number is chosen
1135              * pretty much arbitrarily. */
1136             h->param.i_lookahead_threads = X264_MIN( h->param.i_lookahead_threads, h->param.i_height / 128 );
1137         }
1138     }
1139     h->param.i_lookahead_threads = x264_clip3( h->param.i_lookahead_threads, 1, X264_MIN( max_sliced_threads, X264_LOOKAHEAD_THREAD_MAX ) );
1140
1141     if( PARAM_INTERLACED )
1142     {
1143         if( h->param.analyse.i_me_method >= X264_ME_ESA )
1144         {
1145             x264_log( h, X264_LOG_WARNING, "interlace + me=esa is not implemented\n" );
1146             h->param.analyse.i_me_method = X264_ME_UMH;
1147         }
1148         if( h->param.analyse.i_weighted_pred > 0 )
1149         {
1150             x264_log( h, X264_LOG_WARNING, "interlace + weightp is not implemented\n" );
1151             h->param.analyse.i_weighted_pred = X264_WEIGHTP_NONE;
1152         }
1153     }
1154
1155     if( !h->param.analyse.i_weighted_pred && h->param.rc.b_mb_tree && h->param.analyse.b_psy )
1156         h->param.analyse.i_weighted_pred = X264_WEIGHTP_FAKE;
1157
1158     if( h->i_thread_frames > 1 )
1159     {
1160         int r = h->param.analyse.i_mv_range_thread;
1161         int r2;
1162         if( r <= 0 )
1163         {
1164             // half of the available space is reserved and divided evenly among the threads,
1165             // the rest is allocated to whichever thread is far enough ahead to use it.
1166             // reserving more space increases quality for some videos, but costs more time
1167             // in thread synchronization.
1168             int max_range = (h->param.i_height + X264_THREAD_HEIGHT) / h->i_thread_frames - X264_THREAD_HEIGHT;
1169             r = max_range / 2;
1170         }
1171         r = X264_MAX( r, h->param.analyse.i_me_range );
1172         r = X264_MIN( r, h->param.analyse.i_mv_range );
1173         // round up to use the whole mb row
1174         r2 = (r & ~15) + ((-X264_THREAD_HEIGHT) & 15);
1175         if( r2 < r )
1176             r2 += 16;
1177         x264_log( h, X264_LOG_DEBUG, "using mv_range_thread = %d\n", r2 );
1178         h->param.analyse.i_mv_range_thread = r2;
1179     }
1180
1181     if( h->param.rc.f_rate_tolerance < 0 )
1182         h->param.rc.f_rate_tolerance = 0;
1183     if( h->param.rc.f_qblur < 0 )
1184         h->param.rc.f_qblur = 0;
1185     if( h->param.rc.f_complexity_blur < 0 )
1186         h->param.rc.f_complexity_blur = 0;
1187
1188     h->param.i_sps_id &= 31;
1189
1190     h->param.i_nal_hrd = x264_clip3( h->param.i_nal_hrd, X264_NAL_HRD_NONE, X264_NAL_HRD_CBR );
1191
1192     if( h->param.i_nal_hrd && !h->param.rc.i_vbv_buffer_size )
1193     {
1194         x264_log( h, X264_LOG_WARNING, "NAL HRD parameters require VBV parameters\n" );
1195         h->param.i_nal_hrd = X264_NAL_HRD_NONE;
1196     }
1197
1198     if( h->param.i_nal_hrd == X264_NAL_HRD_CBR &&
1199        (h->param.rc.i_bitrate != h->param.rc.i_vbv_max_bitrate || !h->param.rc.i_vbv_max_bitrate) )
1200     {
1201         x264_log( h, X264_LOG_WARNING, "CBR HRD requires constant bitrate\n" );
1202         h->param.i_nal_hrd = X264_NAL_HRD_VBR;
1203     }
1204
1205     /* ensure the booleans are 0 or 1 so they can be used in math */
1206 #define BOOLIFY(x) h->param.x = !!h->param.x
1207     BOOLIFY( b_cabac );
1208     BOOLIFY( b_constrained_intra );
1209     BOOLIFY( b_deblocking_filter );
1210     BOOLIFY( b_deterministic );
1211     BOOLIFY( b_sliced_threads );
1212     BOOLIFY( b_interlaced );
1213     BOOLIFY( b_intra_refresh );
1214     BOOLIFY( b_visualize );
1215     BOOLIFY( b_aud );
1216     BOOLIFY( b_repeat_headers );
1217     BOOLIFY( b_annexb );
1218     BOOLIFY( b_vfr_input );
1219     BOOLIFY( b_pulldown );
1220     BOOLIFY( b_tff );
1221     BOOLIFY( b_pic_struct );
1222     BOOLIFY( b_fake_interlaced );
1223     BOOLIFY( b_open_gop );
1224     BOOLIFY( b_bluray_compat );
1225     BOOLIFY( b_stitchable );
1226     BOOLIFY( b_full_recon );
1227     BOOLIFY( b_opencl );
1228     BOOLIFY( b_avcintra_compat );
1229     BOOLIFY( analyse.b_transform_8x8 );
1230     BOOLIFY( analyse.b_weighted_bipred );
1231     BOOLIFY( analyse.b_chroma_me );
1232     BOOLIFY( analyse.b_mixed_references );
1233     BOOLIFY( analyse.b_fast_pskip );
1234     BOOLIFY( analyse.b_dct_decimate );
1235     BOOLIFY( analyse.b_psy );
1236     BOOLIFY( analyse.b_psnr );
1237     BOOLIFY( analyse.b_ssim );
1238     BOOLIFY( rc.b_stat_write );
1239     BOOLIFY( rc.b_stat_read );
1240     BOOLIFY( rc.b_mb_tree );
1241 #undef BOOLIFY
1242
1243     return 0;
1244 }
1245
1246 static void mbcmp_init( x264_t *h )
1247 {
1248     int satd = !h->mb.b_lossless && h->param.analyse.i_subpel_refine > 1;
1249     memcpy( h->pixf.mbcmp, satd ? h->pixf.satd : h->pixf.sad_aligned, sizeof(h->pixf.mbcmp) );
1250     memcpy( h->pixf.mbcmp_unaligned, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.mbcmp_unaligned) );
1251     h->pixf.intra_mbcmp_x3_16x16 = satd ? h->pixf.intra_satd_x3_16x16 : h->pixf.intra_sad_x3_16x16;
1252     h->pixf.intra_mbcmp_x3_8x16c = satd ? h->pixf.intra_satd_x3_8x16c : h->pixf.intra_sad_x3_8x16c;
1253     h->pixf.intra_mbcmp_x3_8x8c  = satd ? h->pixf.intra_satd_x3_8x8c  : h->pixf.intra_sad_x3_8x8c;
1254     h->pixf.intra_mbcmp_x3_8x8 = satd ? h->pixf.intra_sa8d_x3_8x8 : h->pixf.intra_sad_x3_8x8;
1255     h->pixf.intra_mbcmp_x3_4x4 = satd ? h->pixf.intra_satd_x3_4x4 : h->pixf.intra_sad_x3_4x4;
1256     h->pixf.intra_mbcmp_x9_4x4 = h->param.b_cpu_independent || h->mb.b_lossless ? NULL
1257                                : satd ? h->pixf.intra_satd_x9_4x4 : h->pixf.intra_sad_x9_4x4;
1258     h->pixf.intra_mbcmp_x9_8x8 = h->param.b_cpu_independent || h->mb.b_lossless ? NULL
1259                                : satd ? h->pixf.intra_sa8d_x9_8x8 : h->pixf.intra_sad_x9_8x8;
1260     satd &= h->param.analyse.i_me_method == X264_ME_TESA;
1261     memcpy( h->pixf.fpelcmp, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.fpelcmp) );
1262     memcpy( h->pixf.fpelcmp_x3, satd ? h->pixf.satd_x3 : h->pixf.sad_x3, sizeof(h->pixf.fpelcmp_x3) );
1263     memcpy( h->pixf.fpelcmp_x4, satd ? h->pixf.satd_x4 : h->pixf.sad_x4, sizeof(h->pixf.fpelcmp_x4) );
1264 }
1265
1266 static void chroma_dsp_init( x264_t *h )
1267 {
1268     memcpy( h->luma2chroma_pixel, x264_luma2chroma_pixel[CHROMA_FORMAT], sizeof(h->luma2chroma_pixel) );
1269
1270     switch( CHROMA_FORMAT )
1271     {
1272         case CHROMA_420:
1273             memcpy( h->predict_chroma, h->predict_8x8c, sizeof(h->predict_chroma) );
1274             h->mc.prefetch_fenc = h->mc.prefetch_fenc_420;
1275             h->loopf.deblock_chroma[0] = h->loopf.deblock_h_chroma_420;
1276             h->loopf.deblock_chroma_intra[0] = h->loopf.deblock_h_chroma_420_intra;
1277             h->loopf.deblock_chroma_mbaff = h->loopf.deblock_chroma_420_mbaff;
1278             h->loopf.deblock_chroma_intra_mbaff = h->loopf.deblock_chroma_420_intra_mbaff;
1279             h->pixf.intra_mbcmp_x3_chroma = h->pixf.intra_mbcmp_x3_8x8c;
1280             h->quantf.coeff_last[DCT_CHROMA_DC] = h->quantf.coeff_last4;
1281             h->quantf.coeff_level_run[DCT_CHROMA_DC] = h->quantf.coeff_level_run4;
1282             break;
1283         case CHROMA_422:
1284             memcpy( h->predict_chroma, h->predict_8x16c, sizeof(h->predict_chroma) );
1285             h->mc.prefetch_fenc = h->mc.prefetch_fenc_422;
1286             h->loopf.deblock_chroma[0] = h->loopf.deblock_h_chroma_422;
1287             h->loopf.deblock_chroma_intra[0] = h->loopf.deblock_h_chroma_422_intra;
1288             h->loopf.deblock_chroma_mbaff = h->loopf.deblock_chroma_422_mbaff;
1289             h->loopf.deblock_chroma_intra_mbaff = h->loopf.deblock_chroma_422_intra_mbaff;
1290             h->pixf.intra_mbcmp_x3_chroma = h->pixf.intra_mbcmp_x3_8x16c;
1291             h->quantf.coeff_last[DCT_CHROMA_DC] = h->quantf.coeff_last8;
1292             h->quantf.coeff_level_run[DCT_CHROMA_DC] = h->quantf.coeff_level_run8;
1293             break;
1294         case CHROMA_444:
1295             h->mc.prefetch_fenc = h->mc.prefetch_fenc_422; /* FIXME: doesn't cover V plane */
1296             h->loopf.deblock_chroma_mbaff = h->loopf.deblock_luma_mbaff;
1297             h->loopf.deblock_chroma_intra_mbaff = h->loopf.deblock_luma_intra_mbaff;
1298             break;
1299     }
1300 }
1301
1302 static void x264_set_aspect_ratio( x264_t *h, x264_param_t *param, int initial )
1303 {
1304     /* VUI */
1305     if( param->vui.i_sar_width > 0 && param->vui.i_sar_height > 0 )
1306     {
1307         uint32_t i_w = param->vui.i_sar_width;
1308         uint32_t i_h = param->vui.i_sar_height;
1309         uint32_t old_w = h->param.vui.i_sar_width;
1310         uint32_t old_h = h->param.vui.i_sar_height;
1311
1312         x264_reduce_fraction( &i_w, &i_h );
1313
1314         while( i_w > 65535 || i_h > 65535 )
1315         {
1316             i_w /= 2;
1317             i_h /= 2;
1318         }
1319
1320         x264_reduce_fraction( &i_w, &i_h );
1321
1322         if( i_w != old_w || i_h != old_h || initial )
1323         {
1324             h->param.vui.i_sar_width = 0;
1325             h->param.vui.i_sar_height = 0;
1326             if( i_w == 0 || i_h == 0 )
1327                 x264_log( h, X264_LOG_WARNING, "cannot create valid sample aspect ratio\n" );
1328             else
1329             {
1330                 x264_log( h, initial?X264_LOG_INFO:X264_LOG_DEBUG, "using SAR=%d/%d\n", i_w, i_h );
1331                 h->param.vui.i_sar_width = i_w;
1332                 h->param.vui.i_sar_height = i_h;
1333             }
1334             x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
1335         }
1336     }
1337 }
1338
1339 /****************************************************************************
1340  * x264_encoder_open:
1341  ****************************************************************************/
1342 x264_t *x264_encoder_open( x264_param_t *param )
1343 {
1344     x264_t *h;
1345     char buf[1000], *p;
1346     int qp, i_slicetype_length;
1347
1348     CHECKED_MALLOCZERO( h, sizeof(x264_t) );
1349
1350     /* Create a copy of param */
1351     memcpy( &h->param, param, sizeof(x264_param_t) );
1352
1353     if( param->param_free )
1354         param->param_free( param );
1355
1356     if( x264_threading_init() )
1357     {
1358         x264_log( h, X264_LOG_ERROR, "unable to initialize threading\n" );
1359         goto fail;
1360     }
1361
1362     if( x264_validate_parameters( h, 1 ) < 0 )
1363         goto fail;
1364
1365     if( h->param.psz_cqm_file )
1366         if( x264_cqm_parse_file( h, h->param.psz_cqm_file ) < 0 )
1367             goto fail;
1368
1369     if( h->param.rc.psz_stat_out )
1370         h->param.rc.psz_stat_out = strdup( h->param.rc.psz_stat_out );
1371     if( h->param.rc.psz_stat_in )
1372         h->param.rc.psz_stat_in = strdup( h->param.rc.psz_stat_in );
1373
1374     x264_reduce_fraction( &h->param.i_fps_num, &h->param.i_fps_den );
1375     x264_reduce_fraction( &h->param.i_timebase_num, &h->param.i_timebase_den );
1376
1377     /* Init x264_t */
1378     h->i_frame = -1;
1379     h->i_frame_num = 0;
1380     h->i_idr_pic_id = 0;
1381
1382     if( (uint64_t)h->param.i_timebase_den * 2 > UINT32_MAX )
1383     {
1384         x264_log( h, X264_LOG_ERROR, "Effective timebase denominator %u exceeds H.264 maximum\n", h->param.i_timebase_den );
1385         goto fail;
1386     }
1387
1388     x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
1389     x264_pps_init( h->pps, h->param.i_sps_id, &h->param, h->sps );
1390
1391     x264_set_aspect_ratio( h, &h->param, 1 );
1392
1393     x264_validate_levels( h, 1 );
1394
1395     h->chroma_qp_table = i_chroma_qp_table + 12 + h->pps->i_chroma_qp_index_offset;
1396
1397     if( x264_cqm_init( h ) < 0 )
1398         goto fail;
1399
1400     h->mb.i_mb_width = h->sps->i_mb_width;
1401     h->mb.i_mb_height = h->sps->i_mb_height;
1402     h->mb.i_mb_count = h->mb.i_mb_width * h->mb.i_mb_height;
1403
1404     h->mb.chroma_h_shift = CHROMA_FORMAT == CHROMA_420 || CHROMA_FORMAT == CHROMA_422;
1405     h->mb.chroma_v_shift = CHROMA_FORMAT == CHROMA_420;
1406
1407     /* Adaptive MBAFF and subme 0 are not supported as we require halving motion
1408      * vectors during prediction, resulting in hpel mvs.
1409      * The chosen solution is to make MBAFF non-adaptive in this case. */
1410     h->mb.b_adaptive_mbaff = PARAM_INTERLACED && h->param.analyse.i_subpel_refine;
1411
1412     /* Init frames. */
1413     if( h->param.i_bframe_adaptive == X264_B_ADAPT_TRELLIS && !h->param.rc.b_stat_read )
1414         h->frames.i_delay = X264_MAX(h->param.i_bframe,3)*4;
1415     else
1416         h->frames.i_delay = h->param.i_bframe;
1417     if( h->param.rc.b_mb_tree || h->param.rc.i_vbv_buffer_size )
1418         h->frames.i_delay = X264_MAX( h->frames.i_delay, h->param.rc.i_lookahead );
1419     i_slicetype_length = h->frames.i_delay;
1420     h->frames.i_delay += h->i_thread_frames - 1;
1421     h->frames.i_delay += h->param.i_sync_lookahead;
1422     h->frames.i_delay += h->param.b_vfr_input;
1423     h->frames.i_bframe_delay = h->param.i_bframe ? (h->param.i_bframe_pyramid ? 2 : 1) : 0;
1424
1425     h->frames.i_max_ref0 = h->param.i_frame_reference;
1426     h->frames.i_max_ref1 = X264_MIN( h->sps->vui.i_num_reorder_frames, h->param.i_frame_reference );
1427     h->frames.i_max_dpb  = h->sps->vui.i_max_dec_frame_buffering;
1428     h->frames.b_have_lowres = !h->param.rc.b_stat_read
1429         && ( h->param.rc.i_rc_method == X264_RC_ABR
1430           || h->param.rc.i_rc_method == X264_RC_CRF
1431           || h->param.i_bframe_adaptive
1432           || h->param.i_scenecut_threshold
1433           || h->param.rc.b_mb_tree
1434           || h->param.analyse.i_weighted_pred );
1435     h->frames.b_have_lowres |= h->param.rc.b_stat_read && h->param.rc.i_vbv_buffer_size > 0;
1436     h->frames.b_have_sub8x8_esa = !!(h->param.analyse.inter & X264_ANALYSE_PSUB8x8);
1437
1438     h->frames.i_last_idr =
1439     h->frames.i_last_keyframe = - h->param.i_keyint_max;
1440     h->frames.i_input    = 0;
1441     h->frames.i_largest_pts = h->frames.i_second_largest_pts = -1;
1442     h->frames.i_poc_last_open_gop = -1;
1443
1444     CHECKED_MALLOCZERO( h->frames.unused[0], (h->frames.i_delay + 3) * sizeof(x264_frame_t *) );
1445     /* Allocate room for max refs plus a few extra just in case. */
1446     CHECKED_MALLOCZERO( h->frames.unused[1], (h->i_thread_frames + X264_REF_MAX + 4) * sizeof(x264_frame_t *) );
1447     CHECKED_MALLOCZERO( h->frames.current, (h->param.i_sync_lookahead + h->param.i_bframe
1448                         + h->i_thread_frames + 3) * sizeof(x264_frame_t *) );
1449     if( h->param.analyse.i_weighted_pred > 0 )
1450         CHECKED_MALLOCZERO( h->frames.blank_unused, h->i_thread_frames * 4 * sizeof(x264_frame_t *) );
1451     h->i_ref[0] = h->i_ref[1] = 0;
1452     h->i_cpb_delay = h->i_coded_fields = h->i_disp_fields = 0;
1453     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);
1454     h->i_disp_fields_last_frame = -1;
1455     x264_rdo_init();
1456
1457     /* init CPU functions */
1458     x264_predict_16x16_init( h->param.cpu, h->predict_16x16 );
1459     x264_predict_8x8c_init( h->param.cpu, h->predict_8x8c );
1460     x264_predict_8x16c_init( h->param.cpu, h->predict_8x16c );
1461     x264_predict_8x8_init( h->param.cpu, h->predict_8x8, &h->predict_8x8_filter );
1462     x264_predict_4x4_init( h->param.cpu, h->predict_4x4 );
1463     x264_pixel_init( h->param.cpu, &h->pixf );
1464     x264_dct_init( h->param.cpu, &h->dctf );
1465     x264_zigzag_init( h->param.cpu, &h->zigzagf_progressive, &h->zigzagf_interlaced );
1466     memcpy( &h->zigzagf, PARAM_INTERLACED ? &h->zigzagf_interlaced : &h->zigzagf_progressive, sizeof(h->zigzagf) );
1467     x264_mc_init( h->param.cpu, &h->mc, h->param.b_cpu_independent );
1468     x264_quant_init( h, h->param.cpu, &h->quantf );
1469     x264_deblock_init( h->param.cpu, &h->loopf, PARAM_INTERLACED );
1470     x264_bitstream_init( h->param.cpu, &h->bsf );
1471     if( h->param.b_cabac )
1472         x264_cabac_init( h );
1473     else
1474         x264_stack_align( x264_cavlc_init, h );
1475
1476     mbcmp_init( h );
1477     chroma_dsp_init( h );
1478
1479     p = buf + sprintf( buf, "using cpu capabilities:" );
1480     for( int i = 0; x264_cpu_names[i].flags; i++ )
1481     {
1482         if( !strcmp(x264_cpu_names[i].name, "SSE")
1483             && h->param.cpu & (X264_CPU_SSE2) )
1484             continue;
1485         if( !strcmp(x264_cpu_names[i].name, "SSE2")
1486             && h->param.cpu & (X264_CPU_SSE2_IS_FAST|X264_CPU_SSE2_IS_SLOW) )
1487             continue;
1488         if( !strcmp(x264_cpu_names[i].name, "SSE3")
1489             && (h->param.cpu & X264_CPU_SSSE3 || !(h->param.cpu & X264_CPU_CACHELINE_64)) )
1490             continue;
1491         if( !strcmp(x264_cpu_names[i].name, "SSE4.1")
1492             && (h->param.cpu & X264_CPU_SSE42) )
1493             continue;
1494         if( !strcmp(x264_cpu_names[i].name, "BMI1")
1495             && (h->param.cpu & X264_CPU_BMI2) )
1496             continue;
1497         if( (h->param.cpu & x264_cpu_names[i].flags) == x264_cpu_names[i].flags
1498             && (!i || x264_cpu_names[i].flags != x264_cpu_names[i-1].flags) )
1499             p += sprintf( p, " %s", x264_cpu_names[i].name );
1500     }
1501     if( !h->param.cpu )
1502         p += sprintf( p, " none!" );
1503     x264_log( h, X264_LOG_INFO, "%s\n", buf );
1504
1505     float *logs = x264_analyse_prepare_costs( h );
1506     if( !logs )
1507         goto fail;
1508     for( qp = X264_MIN( h->param.rc.i_qp_min, QP_MAX_SPEC ); qp <= h->param.rc.i_qp_max; qp++ )
1509         if( x264_analyse_init_costs( h, logs, qp ) )
1510             goto fail;
1511     if( x264_analyse_init_costs( h, logs, X264_LOOKAHEAD_QP ) )
1512         goto fail;
1513     x264_free( logs );
1514
1515     static const uint16_t cost_mv_correct[7] = { 24, 47, 95, 189, 379, 757, 1515 };
1516     /* Checks for known miscompilation issues. */
1517     if( h->cost_mv[X264_LOOKAHEAD_QP][2013] != cost_mv_correct[BIT_DEPTH-8] )
1518     {
1519         x264_log( h, X264_LOG_ERROR, "MV cost test failed: x264 has been miscompiled!\n" );
1520         goto fail;
1521     }
1522
1523     /* Must be volatile or else GCC will optimize it out. */
1524     volatile int temp = 392;
1525     if( x264_clz( temp ) != 23 )
1526     {
1527         x264_log( h, X264_LOG_ERROR, "CLZ test failed: x264 has been miscompiled!\n" );
1528 #if ARCH_X86 || ARCH_X86_64
1529         x264_log( h, X264_LOG_ERROR, "Are you attempting to run an SSE4a/LZCNT-targeted build on a CPU that\n" );
1530         x264_log( h, X264_LOG_ERROR, "doesn't support it?\n" );
1531 #endif
1532         goto fail;
1533     }
1534
1535     h->out.i_nal = 0;
1536     h->out.i_bitstream = X264_MAX( 1000000, h->param.i_width * h->param.i_height * 4
1537         * ( h->param.rc.i_rc_method == X264_RC_ABR ? pow( 0.95, h->param.rc.i_qp_min )
1538           : pow( 0.95, h->param.rc.i_qp_constant ) * X264_MAX( 1, h->param.rc.f_ip_factor )));
1539
1540     h->nal_buffer_size = h->out.i_bitstream * 3/2 + 4 + 64; /* +4 for startcode, +64 for nal_escape assembly padding */
1541     CHECKED_MALLOC( h->nal_buffer, h->nal_buffer_size );
1542
1543     if( h->param.i_threads > 1 &&
1544         x264_threadpool_init( &h->threadpool, h->param.i_threads, (void*)x264_encoder_thread_init, h ) )
1545         goto fail;
1546     if( h->param.i_lookahead_threads > 1 &&
1547         x264_threadpool_init( &h->lookaheadpool, h->param.i_lookahead_threads, NULL, NULL ) )
1548         goto fail;
1549
1550 #if HAVE_OPENCL
1551     if( h->param.b_opencl )
1552     {
1553         h->opencl.ocl = x264_opencl_load_library();
1554         if( !h->opencl.ocl )
1555         {
1556             x264_log( h, X264_LOG_WARNING, "failed to load OpenCL\n" );
1557             h->param.b_opencl = 0;
1558         }
1559     }
1560 #endif
1561
1562     h->thread[0] = h;
1563     for( int i = 1; i < h->param.i_threads + !!h->param.i_sync_lookahead; i++ )
1564         CHECKED_MALLOC( h->thread[i], sizeof(x264_t) );
1565     if( h->param.i_lookahead_threads > 1 )
1566         for( int i = 0; i < h->param.i_lookahead_threads; i++ )
1567         {
1568             CHECKED_MALLOC( h->lookahead_thread[i], sizeof(x264_t) );
1569             *h->lookahead_thread[i] = *h;
1570         }
1571
1572     for( int i = 0; i < h->param.i_threads; i++ )
1573     {
1574         int init_nal_count = h->param.i_slice_count + 3;
1575         int allocate_threadlocal_data = !h->param.b_sliced_threads || !i;
1576         if( i > 0 )
1577             *h->thread[i] = *h;
1578
1579         if( x264_pthread_mutex_init( &h->thread[i]->mutex, NULL ) )
1580             goto fail;
1581         if( x264_pthread_cond_init( &h->thread[i]->cv, NULL ) )
1582             goto fail;
1583
1584         if( allocate_threadlocal_data )
1585         {
1586             h->thread[i]->fdec = x264_frame_pop_unused( h, 1 );
1587             if( !h->thread[i]->fdec )
1588                 goto fail;
1589         }
1590         else
1591             h->thread[i]->fdec = h->thread[0]->fdec;
1592
1593         CHECKED_MALLOC( h->thread[i]->out.p_bitstream, h->out.i_bitstream );
1594         /* Start each thread with room for init_nal_count NAL units; it'll realloc later if needed. */
1595         CHECKED_MALLOC( h->thread[i]->out.nal, init_nal_count*sizeof(x264_nal_t) );
1596         h->thread[i]->out.i_nals_allocated = init_nal_count;
1597
1598         if( allocate_threadlocal_data && x264_macroblock_cache_allocate( h->thread[i] ) < 0 )
1599             goto fail;
1600     }
1601
1602 #if HAVE_OPENCL
1603     if( h->param.b_opencl && x264_opencl_lookahead_init( h ) < 0 )
1604         h->param.b_opencl = 0;
1605 #endif
1606
1607     if( x264_lookahead_init( h, i_slicetype_length ) )
1608         goto fail;
1609
1610     for( int i = 0; i < h->param.i_threads; i++ )
1611         if( x264_macroblock_thread_allocate( h->thread[i], 0 ) < 0 )
1612             goto fail;
1613
1614     if( x264_ratecontrol_new( h ) < 0 )
1615         goto fail;
1616
1617     if( h->param.i_nal_hrd )
1618     {
1619         x264_log( h, X264_LOG_DEBUG, "HRD bitrate: %i bits/sec\n", h->sps->vui.hrd.i_bit_rate_unscaled );
1620         x264_log( h, X264_LOG_DEBUG, "CPB size: %i bits\n", h->sps->vui.hrd.i_cpb_size_unscaled );
1621     }
1622
1623     if( h->param.psz_dump_yuv )
1624     {
1625         /* create or truncate the reconstructed video file */
1626         FILE *f = fopen( h->param.psz_dump_yuv, "w" );
1627         if( !f )
1628         {
1629             x264_log( h, X264_LOG_ERROR, "dump_yuv: can't write to %s\n", h->param.psz_dump_yuv );
1630             goto fail;
1631         }
1632         else if( !x264_is_regular_file( f ) )
1633         {
1634             x264_log( h, X264_LOG_ERROR, "dump_yuv: incompatible with non-regular file %s\n", h->param.psz_dump_yuv );
1635             goto fail;
1636         }
1637         fclose( f );
1638     }
1639
1640     const char *profile = h->sps->i_profile_idc == PROFILE_BASELINE ? "Constrained Baseline" :
1641                           h->sps->i_profile_idc == PROFILE_MAIN ? "Main" :
1642                           h->sps->i_profile_idc == PROFILE_HIGH ? "High" :
1643                           h->sps->i_profile_idc == PROFILE_HIGH10 ? (h->sps->b_constraint_set3 == 1 ? "High 10 Intra" : "High 10") :
1644                           h->sps->i_profile_idc == PROFILE_HIGH422 ? (h->sps->b_constraint_set3 == 1 ? "High 4:2:2 Intra" : "High 4:2:2") :
1645                           h->sps->b_constraint_set3 == 1 ? "High 4:4:4 Intra" : "High 4:4:4 Predictive";
1646     char level[4];
1647     snprintf( level, sizeof(level), "%d.%d", h->sps->i_level_idc/10, h->sps->i_level_idc%10 );
1648     if( h->sps->i_level_idc == 9 || ( h->sps->i_level_idc == 11 && h->sps->b_constraint_set3 &&
1649         (h->sps->i_profile_idc == PROFILE_BASELINE || h->sps->i_profile_idc == PROFILE_MAIN) ) )
1650         strcpy( level, "1b" );
1651
1652     if( h->sps->i_profile_idc < PROFILE_HIGH10 )
1653     {
1654         x264_log( h, X264_LOG_INFO, "profile %s, level %s\n",
1655             profile, level );
1656     }
1657     else
1658     {
1659         static const char * const subsampling[4] = { "4:0:0", "4:2:0", "4:2:2", "4:4:4" };
1660         x264_log( h, X264_LOG_INFO, "profile %s, level %s, %s %d-bit\n",
1661             profile, level, subsampling[CHROMA_FORMAT], BIT_DEPTH );
1662     }
1663
1664     return h;
1665 fail:
1666     x264_free( h );
1667     return NULL;
1668 }
1669
1670 /****************************************************************************
1671  * x264_encoder_reconfig:
1672  ****************************************************************************/
1673 int x264_encoder_reconfig( x264_t *h, x264_param_t *param )
1674 {
1675     /* If the previous frame isn't done encoding, reconfiguring is probably dangerous. */
1676     if( h->param.b_sliced_threads )
1677         if( x264_threadpool_wait_all( h ) < 0 )
1678             return -1;
1679
1680     int rc_reconfig = 0;
1681     h = h->thread[h->thread[0]->i_thread_phase];
1682     x264_set_aspect_ratio( h, param, 0 );
1683 #define COPY(var) h->param.var = param->var
1684     COPY( i_frame_reference ); // but never uses more refs than initially specified
1685     COPY( i_bframe_bias );
1686     if( h->param.i_scenecut_threshold )
1687         COPY( i_scenecut_threshold ); // can't turn it on or off, only vary the threshold
1688     COPY( b_deblocking_filter );
1689     COPY( i_deblocking_filter_alphac0 );
1690     COPY( i_deblocking_filter_beta );
1691     COPY( i_frame_packing );
1692     COPY( analyse.inter );
1693     COPY( analyse.intra );
1694     COPY( analyse.i_direct_mv_pred );
1695     /* Scratch buffer prevents me_range from being increased for esa/tesa */
1696     if( h->param.analyse.i_me_method < X264_ME_ESA || param->analyse.i_me_range < h->param.analyse.i_me_range )
1697         COPY( analyse.i_me_range );
1698     COPY( analyse.i_noise_reduction );
1699     /* We can't switch out of subme=0 during encoding. */
1700     if( h->param.analyse.i_subpel_refine )
1701         COPY( analyse.i_subpel_refine );
1702     COPY( analyse.i_trellis );
1703     COPY( analyse.b_chroma_me );
1704     COPY( analyse.b_dct_decimate );
1705     COPY( analyse.b_fast_pskip );
1706     COPY( analyse.b_mixed_references );
1707     COPY( analyse.f_psy_rd );
1708     COPY( analyse.f_psy_trellis );
1709     COPY( crop_rect );
1710     // can only twiddle these if they were enabled to begin with:
1711     if( h->param.analyse.i_me_method >= X264_ME_ESA || param->analyse.i_me_method < X264_ME_ESA )
1712         COPY( analyse.i_me_method );
1713     if( h->param.analyse.i_me_method >= X264_ME_ESA && !h->frames.b_have_sub8x8_esa )
1714         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
1715     if( h->pps->b_transform_8x8_mode )
1716         COPY( analyse.b_transform_8x8 );
1717     if( h->frames.i_max_ref1 > 1 )
1718         COPY( i_bframe_pyramid );
1719     COPY( i_slice_max_size );
1720     COPY( i_slice_max_mbs );
1721     COPY( i_slice_min_mbs );
1722     COPY( i_slice_count );
1723     COPY( i_slice_count_max );
1724     COPY( b_tff );
1725
1726     /* VBV can't be turned on if it wasn't on to begin with */
1727     if( h->param.rc.i_vbv_max_bitrate > 0 && h->param.rc.i_vbv_buffer_size > 0 &&
1728           param->rc.i_vbv_max_bitrate > 0 &&   param->rc.i_vbv_buffer_size > 0 )
1729     {
1730         rc_reconfig |= h->param.rc.i_vbv_max_bitrate != param->rc.i_vbv_max_bitrate;
1731         rc_reconfig |= h->param.rc.i_vbv_buffer_size != param->rc.i_vbv_buffer_size;
1732         rc_reconfig |= h->param.rc.i_bitrate != param->rc.i_bitrate;
1733         COPY( rc.i_vbv_max_bitrate );
1734         COPY( rc.i_vbv_buffer_size );
1735         COPY( rc.i_bitrate );
1736     }
1737     rc_reconfig |= h->param.rc.f_rf_constant != param->rc.f_rf_constant;
1738     rc_reconfig |= h->param.rc.f_rf_constant_max != param->rc.f_rf_constant_max;
1739     COPY( rc.f_rf_constant );
1740     COPY( rc.f_rf_constant_max );
1741 #undef COPY
1742
1743     mbcmp_init( h );
1744
1745     int ret = x264_validate_parameters( h, 0 );
1746
1747     /* Supported reconfiguration options (1-pass only):
1748      * vbv-maxrate
1749      * vbv-bufsize
1750      * crf
1751      * bitrate (CBR only) */
1752     if( !ret && rc_reconfig )
1753         x264_ratecontrol_init_reconfigurable( h, 0 );
1754
1755     return ret;
1756 }
1757
1758 /****************************************************************************
1759  * x264_encoder_parameters:
1760  ****************************************************************************/
1761 void x264_encoder_parameters( x264_t *h, x264_param_t *param )
1762 {
1763     memcpy( param, &h->thread[h->i_thread_phase]->param, sizeof(x264_param_t) );
1764 }
1765
1766 /* internal usage */
1767 static void x264_nal_start( x264_t *h, int i_type, int i_ref_idc )
1768 {
1769     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1770
1771     nal->i_ref_idc        = i_ref_idc;
1772     nal->i_type           = i_type;
1773     nal->b_long_startcode = 1;
1774
1775     nal->i_payload= 0;
1776     nal->p_payload= &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8];
1777     nal->i_padding= 0;
1778 }
1779
1780 /* if number of allocated nals is not enough, re-allocate a larger one. */
1781 static int x264_nal_check_buffer( x264_t *h )
1782 {
1783     if( h->out.i_nal >= h->out.i_nals_allocated )
1784     {
1785         x264_nal_t *new_out = x264_malloc( sizeof(x264_nal_t) * (h->out.i_nals_allocated*2) );
1786         if( !new_out )
1787             return -1;
1788         memcpy( new_out, h->out.nal, sizeof(x264_nal_t) * (h->out.i_nals_allocated) );
1789         x264_free( h->out.nal );
1790         h->out.nal = new_out;
1791         h->out.i_nals_allocated *= 2;
1792     }
1793     return 0;
1794 }
1795
1796 static int x264_nal_end( x264_t *h )
1797 {
1798     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1799     uint8_t *end = &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8];
1800     nal->i_payload = end - nal->p_payload;
1801     /* Assembly implementation of nal_escape reads past the end of the input.
1802      * While undefined padding wouldn't actually affect the output, it makes valgrind unhappy. */
1803     memset( end, 0xff, 64 );
1804     if( h->param.nalu_process )
1805         h->param.nalu_process( h, nal, h->fenc->opaque );
1806     h->out.i_nal++;
1807
1808     return x264_nal_check_buffer( h );
1809 }
1810
1811 static int x264_check_encapsulated_buffer( x264_t *h, x264_t *h0, int start,
1812                                            int previous_nal_size, int necessary_size )
1813 {
1814     if( h0->nal_buffer_size < necessary_size )
1815     {
1816         necessary_size *= 2;
1817         uint8_t *buf = x264_malloc( necessary_size );
1818         if( !buf )
1819             return -1;
1820         if( previous_nal_size )
1821             memcpy( buf, h0->nal_buffer, previous_nal_size );
1822
1823         intptr_t delta = buf - h0->nal_buffer;
1824         for( int i = 0; i < start; i++ )
1825             h->out.nal[i].p_payload += delta;
1826
1827         x264_free( h0->nal_buffer );
1828         h0->nal_buffer = buf;
1829         h0->nal_buffer_size = necessary_size;
1830     }
1831
1832     return 0;
1833 }
1834
1835 static int x264_encoder_encapsulate_nals( x264_t *h, int start )
1836 {
1837     x264_t *h0 = h->thread[0];
1838     int nal_size = 0, previous_nal_size = 0;
1839
1840     if( h->param.nalu_process )
1841     {
1842         for( int i = start; i < h->out.i_nal; i++ )
1843             nal_size += h->out.nal[i].i_payload;
1844         return nal_size;
1845     }
1846
1847     for( int i = 0; i < start; i++ )
1848         previous_nal_size += h->out.nal[i].i_payload;
1849
1850     for( int i = start; i < h->out.i_nal; i++ )
1851         nal_size += h->out.nal[i].i_payload;
1852
1853     /* Worst-case NAL unit escaping: reallocate the buffer if it's too small. */
1854     int necessary_size = previous_nal_size + nal_size * 3/2 + h->out.i_nal * 4 + 4 + 64;
1855     for( int i = start; i < h->out.i_nal; i++ )
1856         necessary_size += h->out.nal[i].i_padding;
1857     if( x264_check_encapsulated_buffer( h, h0, start, previous_nal_size, necessary_size ) )
1858         return -1;
1859
1860     uint8_t *nal_buffer = h0->nal_buffer + previous_nal_size;
1861
1862     for( int i = start; i < h->out.i_nal; i++ )
1863     {
1864         int old_payload_len = h->out.nal[i].i_payload;
1865         h->out.nal[i].b_long_startcode = !i || h->out.nal[i].i_type == NAL_SPS || h->out.nal[i].i_type == NAL_PPS ||
1866                                          h->param.b_avcintra_compat;
1867         x264_nal_encode( h, nal_buffer, &h->out.nal[i] );
1868         nal_buffer += h->out.nal[i].i_payload;
1869         if( h->param.b_avcintra_compat )
1870         {
1871             h->out.nal[i].i_padding -= h->out.nal[i].i_payload - (old_payload_len + NALU_OVERHEAD);
1872             if( h->out.nal[i].i_padding > 0 )
1873             {
1874                 memset( nal_buffer, 0, h->out.nal[i].i_padding );
1875                 nal_buffer += h->out.nal[i].i_padding;
1876                 h->out.nal[i].i_payload += h->out.nal[i].i_padding;
1877             }
1878             h->out.nal[i].i_padding = X264_MAX( h->out.nal[i].i_padding, 0 );
1879         }
1880     }
1881
1882     x264_emms();
1883
1884     return nal_buffer - (h0->nal_buffer + previous_nal_size);
1885 }
1886
1887 /****************************************************************************
1888  * x264_encoder_headers:
1889  ****************************************************************************/
1890 int x264_encoder_headers( x264_t *h, x264_nal_t **pp_nal, int *pi_nal )
1891 {
1892     int frame_size = 0;
1893     /* init bitstream context */
1894     h->out.i_nal = 0;
1895     bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
1896
1897     /* Write SEI, SPS and PPS. */
1898
1899     /* generate sequence parameters */
1900     x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
1901     x264_sps_write( &h->out.bs, h->sps );
1902     if( x264_nal_end( h ) )
1903         return -1;
1904
1905     /* generate picture parameters */
1906     x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
1907     x264_pps_write( &h->out.bs, h->sps, h->pps );
1908     if( x264_nal_end( h ) )
1909         return -1;
1910
1911     /* identify ourselves */
1912     x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
1913     if( x264_sei_version_write( h, &h->out.bs ) )
1914         return -1;
1915     if( x264_nal_end( h ) )
1916         return -1;
1917
1918     frame_size = x264_encoder_encapsulate_nals( h, 0 );
1919     if( frame_size < 0 )
1920         return -1;
1921
1922     /* now set output*/
1923     *pi_nal = h->out.i_nal;
1924     *pp_nal = &h->out.nal[0];
1925     h->out.i_nal = 0;
1926
1927     return frame_size;
1928 }
1929
1930 /* Check to see whether we have chosen a reference list ordering different
1931  * from the standard's default. */
1932 static inline void x264_reference_check_reorder( x264_t *h )
1933 {
1934     /* The reorder check doesn't check for missing frames, so just
1935      * force a reorder if one of the reference list is corrupt. */
1936     for( int i = 0; h->frames.reference[i]; i++ )
1937         if( h->frames.reference[i]->b_corrupt )
1938         {
1939             h->b_ref_reorder[0] = 1;
1940             return;
1941         }
1942     for( int list = 0; list <= (h->sh.i_type == SLICE_TYPE_B); list++ )
1943         for( int i = 0; i < h->i_ref[list] - 1; i++ )
1944         {
1945             int framenum_diff = h->fref[list][i+1]->i_frame_num - h->fref[list][i]->i_frame_num;
1946             int poc_diff = h->fref[list][i+1]->i_poc - h->fref[list][i]->i_poc;
1947             /* P and B-frames use different default orders. */
1948             if( h->sh.i_type == SLICE_TYPE_P ? framenum_diff > 0 : list == 1 ? poc_diff < 0 : poc_diff > 0 )
1949             {
1950                 h->b_ref_reorder[list] = 1;
1951                 return;
1952             }
1953         }
1954 }
1955
1956 /* return -1 on failure, else return the index of the new reference frame */
1957 int x264_weighted_reference_duplicate( x264_t *h, int i_ref, const x264_weight_t *w )
1958 {
1959     int i = h->i_ref[0];
1960     int j = 1;
1961     x264_frame_t *newframe;
1962     if( i <= 1 ) /* empty list, definitely can't duplicate frame */
1963         return -1;
1964
1965     //Duplication is only used in X264_WEIGHTP_SMART
1966     if( h->param.analyse.i_weighted_pred != X264_WEIGHTP_SMART )
1967         return -1;
1968
1969     /* Duplication is a hack to compensate for crappy rounding in motion compensation.
1970      * With high bit depth, it's not worth doing, so turn it off except in the case of
1971      * unweighted dupes. */
1972     if( BIT_DEPTH > 8 && w != x264_weight_none )
1973         return -1;
1974
1975     newframe = x264_frame_pop_blank_unused( h );
1976     if( !newframe )
1977         return -1;
1978
1979     //FIXME: probably don't need to copy everything
1980     *newframe = *h->fref[0][i_ref];
1981     newframe->i_reference_count = 1;
1982     newframe->orig = h->fref[0][i_ref];
1983     newframe->b_duplicate = 1;
1984     memcpy( h->fenc->weight[j], w, sizeof(h->fenc->weight[i]) );
1985
1986     /* shift the frames to make space for the dupe. */
1987     h->b_ref_reorder[0] = 1;
1988     if( h->i_ref[0] < X264_REF_MAX )
1989         ++h->i_ref[0];
1990     h->fref[0][X264_REF_MAX-1] = NULL;
1991     x264_frame_unshift( &h->fref[0][j], newframe );
1992
1993     return j;
1994 }
1995
1996 static void x264_weighted_pred_init( x264_t *h )
1997 {
1998     /* for now no analysis and set all weights to nothing */
1999     for( int i_ref = 0; i_ref < h->i_ref[0]; i_ref++ )
2000         h->fenc->weighted[i_ref] = h->fref[0][i_ref]->filtered[0][0];
2001
2002     // FIXME: This only supports weighting of one reference frame
2003     // and duplicates of that frame.
2004     h->fenc->i_lines_weighted = 0;
2005
2006     for( int i_ref = 0; i_ref < (h->i_ref[0] << SLICE_MBAFF); i_ref++ )
2007         for( int i = 0; i < 3; i++ )
2008             h->sh.weight[i_ref][i].weightfn = NULL;
2009
2010
2011     if( h->sh.i_type != SLICE_TYPE_P || h->param.analyse.i_weighted_pred <= 0 )
2012         return;
2013
2014     int i_padv = PADV << PARAM_INTERLACED;
2015     int denom = -1;
2016     int weightplane[2] = { 0, 0 };
2017     int buffer_next = 0;
2018     for( int i = 0; i < 3; i++ )
2019     {
2020         for( int j = 0; j < h->i_ref[0]; j++ )
2021         {
2022             if( h->fenc->weight[j][i].weightfn )
2023             {
2024                 h->sh.weight[j][i] = h->fenc->weight[j][i];
2025                 // if weight is useless, don't write it to stream
2026                 if( h->sh.weight[j][i].i_scale == 1<<h->sh.weight[j][i].i_denom && h->sh.weight[j][i].i_offset == 0 )
2027                     h->sh.weight[j][i].weightfn = NULL;
2028                 else
2029                 {
2030                     if( !weightplane[!!i] )
2031                     {
2032                         weightplane[!!i] = 1;
2033                         h->sh.weight[0][!!i].i_denom = denom = h->sh.weight[j][i].i_denom;
2034                         assert( x264_clip3( denom, 0, 7 ) == denom );
2035                     }
2036
2037                     assert( h->sh.weight[j][i].i_denom == denom );
2038                     if( !i )
2039                     {
2040                         h->fenc->weighted[j] = h->mb.p_weight_buf[buffer_next++] + h->fenc->i_stride[0] * i_padv + PADH;
2041                         //scale full resolution frame
2042                         if( h->param.i_threads == 1 )
2043                         {
2044                             pixel *src = h->fref[0][j]->filtered[0][0] - h->fref[0][j]->i_stride[0]*i_padv - PADH;
2045                             pixel *dst = h->fenc->weighted[j] - h->fenc->i_stride[0]*i_padv - PADH;
2046                             int stride = h->fenc->i_stride[0];
2047                             int width = h->fenc->i_width[0] + PADH*2;
2048                             int height = h->fenc->i_lines[0] + i_padv*2;
2049                             x264_weight_scale_plane( h, dst, stride, src, stride, width, height, &h->sh.weight[j][0] );
2050                             h->fenc->i_lines_weighted = height;
2051                         }
2052                     }
2053                 }
2054             }
2055         }
2056     }
2057
2058     if( weightplane[1] )
2059         for( int i = 0; i < h->i_ref[0]; i++ )
2060         {
2061             if( h->sh.weight[i][1].weightfn && !h->sh.weight[i][2].weightfn )
2062             {
2063                 h->sh.weight[i][2].i_scale = 1 << h->sh.weight[0][1].i_denom;
2064                 h->sh.weight[i][2].i_offset = 0;
2065             }
2066             else if( h->sh.weight[i][2].weightfn && !h->sh.weight[i][1].weightfn )
2067             {
2068                 h->sh.weight[i][1].i_scale = 1 << h->sh.weight[0][1].i_denom;
2069                 h->sh.weight[i][1].i_offset = 0;
2070             }
2071         }
2072
2073     if( !weightplane[0] )
2074         h->sh.weight[0][0].i_denom = 0;
2075     if( !weightplane[1] )
2076         h->sh.weight[0][1].i_denom = 0;
2077     h->sh.weight[0][2].i_denom = h->sh.weight[0][1].i_denom;
2078 }
2079
2080 static inline int x264_reference_distance( x264_t *h, x264_frame_t *frame )
2081 {
2082     if( h->param.i_frame_packing == 5 )
2083         return abs((h->fenc->i_frame&~1) - (frame->i_frame&~1)) +
2084                   ((h->fenc->i_frame&1) != (frame->i_frame&1));
2085     else
2086         return abs(h->fenc->i_frame - frame->i_frame);
2087 }
2088
2089 static inline void x264_reference_build_list( x264_t *h, int i_poc )
2090 {
2091     int b_ok;
2092
2093     /* build ref list 0/1 */
2094     h->mb.pic.i_fref[0] = h->i_ref[0] = 0;
2095     h->mb.pic.i_fref[1] = h->i_ref[1] = 0;
2096     if( h->sh.i_type == SLICE_TYPE_I )
2097         return;
2098
2099     for( int i = 0; h->frames.reference[i]; i++ )
2100     {
2101         if( h->frames.reference[i]->b_corrupt )
2102             continue;
2103         if( h->frames.reference[i]->i_poc < i_poc )
2104             h->fref[0][h->i_ref[0]++] = h->frames.reference[i];
2105         else if( h->frames.reference[i]->i_poc > i_poc )
2106             h->fref[1][h->i_ref[1]++] = h->frames.reference[i];
2107     }
2108
2109     /* Order reference lists by distance from the current frame. */
2110     for( int list = 0; list < 2; list++ )
2111     {
2112         h->fref_nearest[list] = h->fref[list][0];
2113         do
2114         {
2115             b_ok = 1;
2116             for( int i = 0; i < h->i_ref[list] - 1; i++ )
2117             {
2118                 if( list ? h->fref[list][i+1]->i_poc < h->fref_nearest[list]->i_poc
2119                          : h->fref[list][i+1]->i_poc > h->fref_nearest[list]->i_poc )
2120                     h->fref_nearest[list] = h->fref[list][i+1];
2121                 if( x264_reference_distance( h, h->fref[list][i] ) > x264_reference_distance( h, h->fref[list][i+1] ) )
2122                 {
2123                     XCHG( x264_frame_t*, h->fref[list][i], h->fref[list][i+1] );
2124                     b_ok = 0;
2125                     break;
2126                 }
2127             }
2128         } while( !b_ok );
2129     }
2130
2131     if( h->sh.i_mmco_remove_from_end )
2132         for( int i = h->i_ref[0]-1; i >= h->i_ref[0] - h->sh.i_mmco_remove_from_end; i-- )
2133         {
2134             int diff = h->i_frame_num - h->fref[0][i]->i_frame_num;
2135             h->sh.mmco[h->sh.i_mmco_command_count].i_poc = h->fref[0][i]->i_poc;
2136             h->sh.mmco[h->sh.i_mmco_command_count++].i_difference_of_pic_nums = diff;
2137         }
2138
2139     x264_reference_check_reorder( h );
2140
2141     h->i_ref[1] = X264_MIN( h->i_ref[1], h->frames.i_max_ref1 );
2142     h->i_ref[0] = X264_MIN( h->i_ref[0], h->frames.i_max_ref0 );
2143     h->i_ref[0] = X264_MIN( h->i_ref[0], h->param.i_frame_reference ); // if reconfig() has lowered the limit
2144
2145     /* For Blu-ray compliance, don't reference frames outside of the minigop. */
2146     if( IS_X264_TYPE_B( h->fenc->i_type ) && h->param.b_bluray_compat )
2147         h->i_ref[0] = X264_MIN( h->i_ref[0], IS_X264_TYPE_B( h->fref[0][0]->i_type ) + 1 );
2148
2149     /* add duplicates */
2150     if( h->fenc->i_type == X264_TYPE_P )
2151     {
2152         int idx = -1;
2153         if( h->param.analyse.i_weighted_pred >= X264_WEIGHTP_SIMPLE )
2154         {
2155             x264_weight_t w[3];
2156             w[1].weightfn = w[2].weightfn = NULL;
2157             if( h->param.rc.b_stat_read )
2158                 x264_ratecontrol_set_weights( h, h->fenc );
2159
2160             if( !h->fenc->weight[0][0].weightfn )
2161             {
2162                 h->fenc->weight[0][0].i_denom = 0;
2163                 SET_WEIGHT( w[0], 1, 1, 0, -1 );
2164                 idx = x264_weighted_reference_duplicate( h, 0, w );
2165             }
2166             else
2167             {
2168                 if( h->fenc->weight[0][0].i_scale == 1<<h->fenc->weight[0][0].i_denom )
2169                 {
2170                     SET_WEIGHT( h->fenc->weight[0][0], 1, 1, 0, h->fenc->weight[0][0].i_offset );
2171                 }
2172                 x264_weighted_reference_duplicate( h, 0, x264_weight_none );
2173                 if( h->fenc->weight[0][0].i_offset > -128 )
2174                 {
2175                     w[0] = h->fenc->weight[0][0];
2176                     w[0].i_offset--;
2177                     h->mc.weight_cache( h, &w[0] );
2178                     idx = x264_weighted_reference_duplicate( h, 0, w );
2179                 }
2180             }
2181         }
2182         h->mb.ref_blind_dupe = idx;
2183     }
2184
2185     assert( h->i_ref[0] + h->i_ref[1] <= X264_REF_MAX );
2186     h->mb.pic.i_fref[0] = h->i_ref[0];
2187     h->mb.pic.i_fref[1] = h->i_ref[1];
2188 }
2189
2190 static void x264_fdec_filter_row( x264_t *h, int mb_y, int pass )
2191 {
2192     /* mb_y is the mb to be encoded next, not the mb to be filtered here */
2193     int b_hpel = h->fdec->b_kept_as_ref;
2194     int b_deblock = h->sh.i_disable_deblocking_filter_idc != 1;
2195     int b_end = mb_y == h->i_threadslice_end;
2196     int b_measure_quality = 1;
2197     int min_y = mb_y - (1 << SLICE_MBAFF);
2198     int b_start = min_y == h->i_threadslice_start;
2199     /* Even in interlaced mode, deblocking never modifies more than 4 pixels
2200      * above each MB, as bS=4 doesn't happen for the top of interlaced mbpairs. */
2201     int minpix_y = min_y*16 - 4 * !b_start;
2202     int maxpix_y = mb_y*16 - 4 * !b_end;
2203     b_deblock &= b_hpel || h->param.b_full_recon || h->param.psz_dump_yuv;
2204     if( h->param.b_sliced_threads )
2205     {
2206         switch( pass )
2207         {
2208             /* During encode: only do deblock if asked for */
2209             default:
2210             case 0:
2211                 b_deblock &= h->param.b_full_recon;
2212                 b_hpel = 0;
2213                 break;
2214             /* During post-encode pass: do deblock if not done yet, do hpel for all
2215              * rows except those between slices. */
2216             case 1:
2217                 b_deblock &= !h->param.b_full_recon;
2218                 b_hpel &= !(b_start && min_y > 0);
2219                 b_measure_quality = 0;
2220                 break;
2221             /* Final pass: do the rows between slices in sequence. */
2222             case 2:
2223                 b_deblock = 0;
2224                 b_measure_quality = 0;
2225                 break;
2226         }
2227     }
2228     if( mb_y & SLICE_MBAFF )
2229         return;
2230     if( min_y < h->i_threadslice_start )
2231         return;
2232
2233     if( b_deblock )
2234         for( int y = min_y; y < mb_y; y += (1 << SLICE_MBAFF) )
2235             x264_frame_deblock_row( h, y );
2236
2237     /* FIXME: Prediction requires different borders for interlaced/progressive mc,
2238      * but the actual image data is equivalent. For now, maintain this
2239      * consistency by copying deblocked pixels between planes. */
2240     if( PARAM_INTERLACED && (!h->param.b_sliced_threads || pass == 1) )
2241         for( int p = 0; p < h->fdec->i_plane; p++ )
2242             for( int i = minpix_y>>(CHROMA_V_SHIFT && p); i < maxpix_y>>(CHROMA_V_SHIFT && p); i++ )
2243                 memcpy( h->fdec->plane_fld[p] + i*h->fdec->i_stride[p],
2244                         h->fdec->plane[p]     + i*h->fdec->i_stride[p],
2245                         h->mb.i_mb_width*16*sizeof(pixel) );
2246
2247     if( h->fdec->b_kept_as_ref && (!h->param.b_sliced_threads || pass == 1) )
2248         x264_frame_expand_border( h, h->fdec, min_y );
2249     if( b_hpel )
2250     {
2251         int end = mb_y == h->mb.i_mb_height;
2252         /* Can't do hpel until the previous slice is done encoding. */
2253         if( h->param.analyse.i_subpel_refine )
2254         {
2255             x264_frame_filter( h, h->fdec, min_y, end );
2256             x264_frame_expand_border_filtered( h, h->fdec, min_y, end );
2257         }
2258     }
2259
2260     if( SLICE_MBAFF && pass == 0 )
2261         for( int i = 0; i < 3; i++ )
2262         {
2263             XCHG( pixel *, h->intra_border_backup[0][i], h->intra_border_backup[3][i] );
2264             XCHG( pixel *, h->intra_border_backup[1][i], h->intra_border_backup[4][i] );
2265         }
2266
2267     if( h->i_thread_frames > 1 && h->fdec->b_kept_as_ref )
2268         x264_frame_cond_broadcast( h->fdec, mb_y*16 + (b_end ? 10000 : -(X264_THREAD_HEIGHT << SLICE_MBAFF)) );
2269
2270     if( b_measure_quality )
2271     {
2272         maxpix_y = X264_MIN( maxpix_y, h->param.i_height );
2273         if( h->param.analyse.b_psnr )
2274         {
2275             for( int p = 0; p < (CHROMA444 ? 3 : 1); p++ )
2276                 h->stat.frame.i_ssd[p] += x264_pixel_ssd_wxh( &h->pixf,
2277                     h->fdec->plane[p] + minpix_y * h->fdec->i_stride[p], h->fdec->i_stride[p],
2278                     h->fenc->plane[p] + minpix_y * h->fenc->i_stride[p], h->fenc->i_stride[p],
2279                     h->param.i_width, maxpix_y-minpix_y );
2280             if( !CHROMA444 )
2281             {
2282                 uint64_t ssd_u, ssd_v;
2283                 int v_shift = CHROMA_V_SHIFT;
2284                 x264_pixel_ssd_nv12( &h->pixf,
2285                     h->fdec->plane[1] + (minpix_y>>v_shift) * h->fdec->i_stride[1], h->fdec->i_stride[1],
2286                     h->fenc->plane[1] + (minpix_y>>v_shift) * h->fenc->i_stride[1], h->fenc->i_stride[1],
2287                     h->param.i_width>>1, (maxpix_y-minpix_y)>>v_shift, &ssd_u, &ssd_v );
2288                 h->stat.frame.i_ssd[1] += ssd_u;
2289                 h->stat.frame.i_ssd[2] += ssd_v;
2290             }
2291         }
2292
2293         if( h->param.analyse.b_ssim )
2294         {
2295             int ssim_cnt;
2296             x264_emms();
2297             /* offset by 2 pixels to avoid alignment of ssim blocks with dct blocks,
2298              * and overlap by 4 */
2299             minpix_y += b_start ? 2 : -6;
2300             h->stat.frame.f_ssim +=
2301                 x264_pixel_ssim_wxh( &h->pixf,
2302                     h->fdec->plane[0] + 2+minpix_y*h->fdec->i_stride[0], h->fdec->i_stride[0],
2303                     h->fenc->plane[0] + 2+minpix_y*h->fenc->i_stride[0], h->fenc->i_stride[0],
2304                     h->param.i_width-2, maxpix_y-minpix_y, h->scratch_buffer, &ssim_cnt );
2305             h->stat.frame.i_ssim_cnt += ssim_cnt;
2306         }
2307     }
2308 }
2309
2310 static inline int x264_reference_update( x264_t *h )
2311 {
2312     if( !h->fdec->b_kept_as_ref )
2313     {
2314         if( h->i_thread_frames > 1 )
2315         {
2316             x264_frame_push_unused( h, h->fdec );
2317             h->fdec = x264_frame_pop_unused( h, 1 );
2318             if( !h->fdec )
2319                 return -1;
2320         }
2321         return 0;
2322     }
2323
2324     /* apply mmco from previous frame. */
2325     for( int i = 0; i < h->sh.i_mmco_command_count; i++ )
2326         for( int j = 0; h->frames.reference[j]; j++ )
2327             if( h->frames.reference[j]->i_poc == h->sh.mmco[i].i_poc )
2328                 x264_frame_push_unused( h, x264_frame_shift( &h->frames.reference[j] ) );
2329
2330     /* move frame in the buffer */
2331     x264_frame_push( h->frames.reference, h->fdec );
2332     if( h->frames.reference[h->sps->i_num_ref_frames] )
2333         x264_frame_push_unused( h, x264_frame_shift( h->frames.reference ) );
2334     h->fdec = x264_frame_pop_unused( h, 1 );
2335     if( !h->fdec )
2336         return -1;
2337     return 0;
2338 }
2339
2340 static inline void x264_reference_reset( x264_t *h )
2341 {
2342     while( h->frames.reference[0] )
2343         x264_frame_push_unused( h, x264_frame_pop( h->frames.reference ) );
2344     h->fdec->i_poc =
2345     h->fenc->i_poc = 0;
2346 }
2347
2348 static inline void x264_reference_hierarchy_reset( x264_t *h )
2349 {
2350     int ref;
2351     int b_hasdelayframe = 0;
2352
2353     /* look for delay frames -- chain must only contain frames that are disposable */
2354     for( int i = 0; h->frames.current[i] && IS_DISPOSABLE( h->frames.current[i]->i_type ); i++ )
2355         b_hasdelayframe |= h->frames.current[i]->i_coded
2356                         != h->frames.current[i]->i_frame + h->sps->vui.i_num_reorder_frames;
2357
2358     /* This function must handle b-pyramid and clear frames for open-gop */
2359     if( h->param.i_bframe_pyramid != X264_B_PYRAMID_STRICT && !b_hasdelayframe && h->frames.i_poc_last_open_gop == -1 )
2360         return;
2361
2362     /* Remove last BREF. There will never be old BREFs in the
2363      * dpb during a BREF decode when pyramid == STRICT */
2364     for( ref = 0; h->frames.reference[ref]; ref++ )
2365     {
2366         if( ( h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT
2367             && h->frames.reference[ref]->i_type == X264_TYPE_BREF )
2368             || ( h->frames.reference[ref]->i_poc < h->frames.i_poc_last_open_gop
2369             && h->sh.i_type != SLICE_TYPE_B ) )
2370         {
2371             int diff = h->i_frame_num - h->frames.reference[ref]->i_frame_num;
2372             h->sh.mmco[h->sh.i_mmco_command_count].i_difference_of_pic_nums = diff;
2373             h->sh.mmco[h->sh.i_mmco_command_count++].i_poc = h->frames.reference[ref]->i_poc;
2374             x264_frame_push_unused( h, x264_frame_shift( &h->frames.reference[ref] ) );
2375             h->b_ref_reorder[0] = 1;
2376             ref--;
2377         }
2378     }
2379
2380     /* Prepare room in the dpb for the delayed display time of the later b-frame's */
2381     if( h->param.i_bframe_pyramid )
2382         h->sh.i_mmco_remove_from_end = X264_MAX( ref + 2 - h->frames.i_max_dpb, 0 );
2383 }
2384
2385 static inline void x264_slice_init( x264_t *h, int i_nal_type, int i_global_qp )
2386 {
2387     /* ------------------------ Create slice header  ----------------------- */
2388     if( i_nal_type == NAL_SLICE_IDR )
2389     {
2390         x264_slice_header_init( h, &h->sh, h->sps, h->pps, h->i_idr_pic_id, h->i_frame_num, i_global_qp );
2391
2392         /* alternate id */
2393         h->i_idr_pic_id ^= 1;
2394     }
2395     else
2396     {
2397         x264_slice_header_init( h, &h->sh, h->sps, h->pps, -1, h->i_frame_num, i_global_qp );
2398
2399         h->sh.i_num_ref_idx_l0_active = h->i_ref[0] <= 0 ? 1 : h->i_ref[0];
2400         h->sh.i_num_ref_idx_l1_active = h->i_ref[1] <= 0 ? 1 : h->i_ref[1];
2401         if( h->sh.i_num_ref_idx_l0_active != h->pps->i_num_ref_idx_l0_default_active ||
2402             (h->sh.i_type == SLICE_TYPE_B && h->sh.i_num_ref_idx_l1_active != h->pps->i_num_ref_idx_l1_default_active) )
2403         {
2404             h->sh.b_num_ref_idx_override = 1;
2405         }
2406     }
2407
2408     if( h->fenc->i_type == X264_TYPE_BREF && h->param.b_bluray_compat && h->sh.i_mmco_command_count )
2409     {
2410         h->b_sh_backup = 1;
2411         h->sh_backup = h->sh;
2412     }
2413
2414     h->fdec->i_frame_num = h->sh.i_frame_num;
2415
2416     if( h->sps->i_poc_type == 0 )
2417     {
2418         h->sh.i_poc = h->fdec->i_poc;
2419         if( PARAM_INTERLACED )
2420         {
2421             h->sh.i_delta_poc_bottom = h->param.b_tff ? 1 : -1;
2422             h->sh.i_poc += h->sh.i_delta_poc_bottom == -1;
2423         }
2424         else
2425             h->sh.i_delta_poc_bottom = 0;
2426         h->fdec->i_delta_poc[0] = h->sh.i_delta_poc_bottom == -1;
2427         h->fdec->i_delta_poc[1] = h->sh.i_delta_poc_bottom ==  1;
2428     }
2429     else
2430     {
2431         /* Nothing to do ? */
2432     }
2433
2434     x264_macroblock_slice_init( h );
2435 }
2436
2437 typedef struct
2438 {
2439     int skip;
2440     uint8_t cabac_prevbyte;
2441     bs_t bs;
2442     x264_cabac_t cabac;
2443     x264_frame_stat_t stat;
2444     int last_qp;
2445     int last_dqp;
2446     int field_decoding_flag;
2447 } x264_bs_bak_t;
2448
2449 static ALWAYS_INLINE void x264_bitstream_backup( x264_t *h, x264_bs_bak_t *bak, int i_skip, int full )
2450 {
2451     if( full )
2452     {
2453         bak->stat = h->stat.frame;
2454         bak->last_qp = h->mb.i_last_qp;
2455         bak->last_dqp = h->mb.i_last_dqp;
2456         bak->field_decoding_flag = h->mb.field_decoding_flag;
2457     }
2458     else
2459     {
2460         bak->stat.i_mv_bits = h->stat.frame.i_mv_bits;
2461         bak->stat.i_tex_bits = h->stat.frame.i_tex_bits;
2462     }
2463     /* In the per-MB backup, we don't need the contexts because flushing the CABAC
2464      * encoder has no context dependency and in this case, a slice is ended (and
2465      * thus the content of all contexts are thrown away). */
2466     if( h->param.b_cabac )
2467     {
2468         if( full )
2469             memcpy( &bak->cabac, &h->cabac, sizeof(x264_cabac_t) );
2470         else
2471             memcpy( &bak->cabac, &h->cabac, offsetof(x264_cabac_t, f8_bits_encoded) );
2472         /* x264's CABAC writer modifies the previous byte during carry, so it has to be
2473          * backed up. */
2474         bak->cabac_prevbyte = h->cabac.p[-1];
2475     }
2476     else
2477     {
2478         bak->bs = h->out.bs;
2479         bak->skip = i_skip;
2480     }
2481 }
2482
2483 static ALWAYS_INLINE void x264_bitstream_restore( x264_t *h, x264_bs_bak_t *bak, int *skip, int full )
2484 {
2485     if( full )
2486     {
2487         h->stat.frame = bak->stat;
2488         h->mb.i_last_qp = bak->last_qp;
2489         h->mb.i_last_dqp = bak->last_dqp;
2490         h->mb.field_decoding_flag = bak->field_decoding_flag;
2491     }
2492     else
2493     {
2494         h->stat.frame.i_mv_bits = bak->stat.i_mv_bits;
2495         h->stat.frame.i_tex_bits = bak->stat.i_tex_bits;
2496     }
2497     if( h->param.b_cabac )
2498     {
2499         if( full )
2500             memcpy( &h->cabac, &bak->cabac, sizeof(x264_cabac_t) );
2501         else
2502             memcpy( &h->cabac, &bak->cabac, offsetof(x264_cabac_t, f8_bits_encoded) );
2503         h->cabac.p[-1] = bak->cabac_prevbyte;
2504     }
2505     else
2506     {
2507         h->out.bs = bak->bs;
2508         *skip = bak->skip;
2509     }
2510 }
2511
2512 static int x264_slice_write( x264_t *h )
2513 {
2514     int i_skip;
2515     int mb_xy, i_mb_x, i_mb_y;
2516     /* NALUs other than the first use a 3-byte startcode.
2517      * Add one extra byte for the rbsp, and one more for the final CABAC putbyte.
2518      * Then add an extra 5 bytes just in case, to account for random NAL escapes and
2519      * other inaccuracies. */
2520     int overhead_guess = (NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal)) + 1 + h->param.b_cabac + 5;
2521     int slice_max_size = h->param.i_slice_max_size > 0 ? (h->param.i_slice_max_size-overhead_guess)*8 : 0;
2522     int back_up_bitstream = slice_max_size || (!h->param.b_cabac && h->sps->i_profile_idc < PROFILE_HIGH);
2523     int starting_bits = bs_pos(&h->out.bs);
2524     int b_deblock = h->sh.i_disable_deblocking_filter_idc != 1;
2525     int b_hpel = h->fdec->b_kept_as_ref;
2526     int orig_last_mb = h->sh.i_last_mb;
2527     int thread_last_mb = h->i_threadslice_end * h->mb.i_mb_width - 1;
2528     uint8_t *last_emu_check;
2529 #define BS_BAK_SLICE_MAX_SIZE 0
2530 #define BS_BAK_SLICE_MIN_MBS  1
2531 #define BS_BAK_ROW_VBV        2
2532     x264_bs_bak_t bs_bak[3];
2533     b_deblock &= b_hpel || h->param.b_full_recon || h->param.psz_dump_yuv;
2534     bs_realign( &h->out.bs );
2535
2536     /* Slice */
2537     x264_nal_start( h, h->i_nal_type, h->i_nal_ref_idc );
2538     h->out.nal[h->out.i_nal].i_first_mb = h->sh.i_first_mb;
2539
2540     /* Slice header */
2541     x264_macroblock_thread_init( h );
2542
2543     /* Set the QP equal to the first QP in the slice for more accurate CABAC initialization. */
2544     h->mb.i_mb_xy = h->sh.i_first_mb;
2545     h->sh.i_qp = x264_ratecontrol_mb_qp( h );
2546     h->sh.i_qp = SPEC_QP( h->sh.i_qp );
2547     h->sh.i_qp_delta = h->sh.i_qp - h->pps->i_pic_init_qp;
2548
2549     x264_slice_header_write( &h->out.bs, &h->sh, h->i_nal_ref_idc );
2550     if( h->param.b_cabac )
2551     {
2552         /* alignment needed */
2553         bs_align_1( &h->out.bs );
2554
2555         /* init cabac */
2556         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 );
2557         x264_cabac_encode_init ( &h->cabac, h->out.bs.p, h->out.bs.p_end );
2558         last_emu_check = h->cabac.p;
2559     }
2560     else
2561         last_emu_check = h->out.bs.p;
2562     h->mb.i_last_qp = h->sh.i_qp;
2563     h->mb.i_last_dqp = 0;
2564     h->mb.field_decoding_flag = 0;
2565
2566     i_mb_y = h->sh.i_first_mb / h->mb.i_mb_width;
2567     i_mb_x = h->sh.i_first_mb % h->mb.i_mb_width;
2568     i_skip = 0;
2569
2570     while( 1 )
2571     {
2572         mb_xy = i_mb_x + i_mb_y * h->mb.i_mb_width;
2573         int mb_spos = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
2574
2575         if( i_mb_x == 0 )
2576         {
2577             if( x264_bitstream_check_buffer( h ) )
2578                 return -1;
2579             if( !(i_mb_y & SLICE_MBAFF) && h->param.rc.i_vbv_buffer_size )
2580                 x264_bitstream_backup( h, &bs_bak[BS_BAK_ROW_VBV], i_skip, 1 );
2581             if( !h->mb.b_reencode_mb )
2582                 x264_fdec_filter_row( h, i_mb_y, 0 );
2583         }
2584
2585         if( !(i_mb_y & SLICE_MBAFF) && back_up_bitstream )
2586         {
2587             x264_bitstream_backup( h, &bs_bak[BS_BAK_SLICE_MAX_SIZE], i_skip, 0 );
2588             if( slice_max_size && (thread_last_mb+1-mb_xy) == h->param.i_slice_min_mbs )
2589                 x264_bitstream_backup( h, &bs_bak[BS_BAK_SLICE_MIN_MBS], i_skip, 0 );
2590         }
2591
2592         if( PARAM_INTERLACED )
2593         {
2594             if( h->mb.b_adaptive_mbaff )
2595             {
2596                 if( !(i_mb_y&1) )
2597                 {
2598                     /* FIXME: VSAD is fast but fairly poor at choosing the best interlace type. */
2599                     h->mb.b_interlaced = x264_field_vsad( h, i_mb_x, i_mb_y );
2600                     memcpy( &h->zigzagf, MB_INTERLACED ? &h->zigzagf_interlaced : &h->zigzagf_progressive, sizeof(h->zigzagf) );
2601                     if( !MB_INTERLACED && (i_mb_y+2) == h->mb.i_mb_height )
2602                         x264_expand_border_mbpair( h, i_mb_x, i_mb_y );
2603                 }
2604             }
2605             h->mb.field[mb_xy] = MB_INTERLACED;
2606         }
2607
2608         /* load cache */
2609         if( SLICE_MBAFF )
2610             x264_macroblock_cache_load_interlaced( h, i_mb_x, i_mb_y );
2611         else
2612             x264_macroblock_cache_load_progressive( h, i_mb_x, i_mb_y );
2613
2614         x264_macroblock_analyse( h );
2615
2616         /* encode this macroblock -> be careful it can change the mb type to P_SKIP if needed */
2617 reencode:
2618         x264_macroblock_encode( h );
2619
2620         if( h->param.b_cabac )
2621         {
2622             if( mb_xy > h->sh.i_first_mb && !(SLICE_MBAFF && (i_mb_y&1)) )
2623                 x264_cabac_encode_terminal( &h->cabac );
2624
2625             if( IS_SKIP( h->mb.i_type ) )
2626                 x264_cabac_mb_skip( h, 1 );
2627             else
2628             {
2629                 if( h->sh.i_type != SLICE_TYPE_I )
2630                     x264_cabac_mb_skip( h, 0 );
2631                 x264_macroblock_write_cabac( h, &h->cabac );
2632             }
2633         }
2634         else
2635         {
2636             if( IS_SKIP( h->mb.i_type ) )
2637                 i_skip++;
2638             else
2639             {
2640                 if( h->sh.i_type != SLICE_TYPE_I )
2641                 {
2642                     bs_write_ue( &h->out.bs, i_skip );  /* skip run */
2643                     i_skip = 0;
2644                 }
2645                 x264_macroblock_write_cavlc( h );
2646                 /* If there was a CAVLC level code overflow, try again at a higher QP. */
2647                 if( h->mb.b_overflow )
2648                 {
2649                     h->mb.i_chroma_qp = h->chroma_qp_table[++h->mb.i_qp];
2650                     h->mb.i_skip_intra = 0;
2651                     h->mb.b_skip_mc = 0;
2652                     h->mb.b_overflow = 0;
2653                     x264_bitstream_restore( h, &bs_bak[BS_BAK_SLICE_MAX_SIZE], &i_skip, 0 );
2654                     goto reencode;
2655                 }
2656             }
2657         }
2658
2659         int total_bits = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
2660         int mb_size = total_bits - mb_spos;
2661
2662         if( slice_max_size && (!SLICE_MBAFF || (i_mb_y&1)) )
2663         {
2664             /* Count the skip run, just in case. */
2665             if( !h->param.b_cabac )
2666                 total_bits += bs_size_ue_big( i_skip );
2667             /* Check for escape bytes. */
2668             uint8_t *end = h->param.b_cabac ? h->cabac.p : h->out.bs.p;
2669             for( ; last_emu_check < end - 2; last_emu_check++ )
2670                 if( last_emu_check[0] == 0 && last_emu_check[1] == 0 && last_emu_check[2] <= 3 )
2671                 {
2672                     slice_max_size -= 8;
2673                     last_emu_check++;
2674                 }
2675             /* We'll just re-encode this last macroblock if we go over the max slice size. */
2676             if( total_bits - starting_bits > slice_max_size && !h->mb.b_reencode_mb )
2677             {
2678                 if( !x264_frame_new_slice( h, h->fdec ) )
2679                 {
2680                     /* Handle the most obnoxious slice-min-mbs edge case: we need to end the slice
2681                      * because it's gone over the maximum size, but doing so would violate slice-min-mbs.
2682                      * If possible, roll back to the last checkpoint and try again.
2683                      * We could try raising QP, but that would break in the case where a slice spans multiple
2684                      * rows, which the re-encoding infrastructure can't currently handle. */
2685                     if( mb_xy <= thread_last_mb && (thread_last_mb+1-mb_xy) < h->param.i_slice_min_mbs )
2686                     {
2687                         if( thread_last_mb-h->param.i_slice_min_mbs < h->sh.i_first_mb+h->param.i_slice_min_mbs )
2688                         {
2689                             x264_log( h, X264_LOG_WARNING, "slice-max-size violated (frame %d, cause: slice-min-mbs)\n", h->i_frame );
2690                             slice_max_size = 0;
2691                             goto cont;
2692                         }
2693                         x264_bitstream_restore( h, &bs_bak[BS_BAK_SLICE_MIN_MBS], &i_skip, 0 );
2694                         h->mb.b_reencode_mb = 1;
2695                         h->sh.i_last_mb = thread_last_mb-h->param.i_slice_min_mbs;
2696                         break;
2697                     }
2698                     if( mb_xy-SLICE_MBAFF*h->mb.i_mb_stride != h->sh.i_first_mb )
2699                     {
2700                         x264_bitstream_restore( h, &bs_bak[BS_BAK_SLICE_MAX_SIZE], &i_skip, 0 );
2701                         h->mb.b_reencode_mb = 1;
2702                         if( SLICE_MBAFF )
2703                         {
2704                             // set to bottom of previous mbpair
2705                             if( i_mb_x )
2706                                 h->sh.i_last_mb = mb_xy-1+h->mb.i_mb_stride*(!(i_mb_y&1));
2707                             else
2708                                 h->sh.i_last_mb = (i_mb_y-2+!(i_mb_y&1))*h->mb.i_mb_stride + h->mb.i_mb_width - 1;
2709                         }
2710                         else
2711                             h->sh.i_last_mb = mb_xy-1;
2712                         break;
2713                     }
2714                     else
2715                         h->sh.i_last_mb = mb_xy;
2716                 }
2717                 else
2718                     slice_max_size = 0;
2719             }
2720         }
2721 cont:
2722         h->mb.b_reencode_mb = 0;
2723
2724 #if HAVE_VISUALIZE
2725         if( h->param.b_visualize )
2726             x264_visualize_mb( h );
2727 #endif
2728
2729         /* save cache */
2730         x264_macroblock_cache_save( h );
2731
2732         if( x264_ratecontrol_mb( h, mb_size ) < 0 )
2733         {
2734             x264_bitstream_restore( h, &bs_bak[BS_BAK_ROW_VBV], &i_skip, 1 );
2735             h->mb.b_reencode_mb = 1;
2736             i_mb_x = 0;
2737             i_mb_y = i_mb_y - SLICE_MBAFF;
2738             h->mb.i_mb_prev_xy = i_mb_y * h->mb.i_mb_stride - 1;
2739             h->sh.i_last_mb = orig_last_mb;
2740             continue;
2741         }
2742
2743         /* accumulate mb stats */
2744         h->stat.frame.i_mb_count[h->mb.i_type]++;
2745
2746         int b_intra = IS_INTRA( h->mb.i_type );
2747         int b_skip = IS_SKIP( h->mb.i_type );
2748         if( h->param.i_log_level >= X264_LOG_INFO || h->param.rc.b_stat_write )
2749         {
2750             if( !b_intra && !b_skip && !IS_DIRECT( h->mb.i_type ) )
2751             {
2752                 if( h->mb.i_partition != D_8x8 )
2753                         h->stat.frame.i_mb_partition[h->mb.i_partition] += 4;
2754                     else
2755                         for( int i = 0; i < 4; i++ )
2756                             h->stat.frame.i_mb_partition[h->mb.i_sub_partition[i]] ++;
2757                 if( h->param.i_frame_reference > 1 )
2758                     for( int i_list = 0; i_list <= (h->sh.i_type == SLICE_TYPE_B); i_list++ )
2759                         for( int i = 0; i < 4; i++ )
2760                         {
2761                             int i_ref = h->mb.cache.ref[i_list][ x264_scan8[4*i] ];
2762                             if( i_ref >= 0 )
2763                                 h->stat.frame.i_mb_count_ref[i_list][i_ref] ++;
2764                         }
2765             }
2766         }
2767
2768         if( h->param.i_log_level >= X264_LOG_INFO )
2769         {
2770             if( h->mb.i_cbp_luma | h->mb.i_cbp_chroma )
2771             {
2772                 if( CHROMA444 )
2773                 {
2774                     for( int i = 0; i < 4; i++ )
2775                         if( h->mb.i_cbp_luma & (1 << i) )
2776                             for( int p = 0; p < 3; p++ )
2777                             {
2778                                 int s8 = i*4+p*16;
2779                                 int nnz8x8 = M16( &h->mb.cache.non_zero_count[x264_scan8[s8]+0] )
2780                                            | M16( &h->mb.cache.non_zero_count[x264_scan8[s8]+8] );
2781                                 h->stat.frame.i_mb_cbp[!b_intra + p*2] += !!nnz8x8;
2782                             }
2783                 }
2784                 else
2785                 {
2786                     int cbpsum = (h->mb.i_cbp_luma&1) + ((h->mb.i_cbp_luma>>1)&1)
2787                                + ((h->mb.i_cbp_luma>>2)&1) + (h->mb.i_cbp_luma>>3);
2788                     h->stat.frame.i_mb_cbp[!b_intra + 0] += cbpsum;
2789                     h->stat.frame.i_mb_cbp[!b_intra + 2] += !!h->mb.i_cbp_chroma;
2790                     h->stat.frame.i_mb_cbp[!b_intra + 4] += h->mb.i_cbp_chroma >> 1;
2791                 }
2792             }
2793             if( h->mb.i_cbp_luma && !b_intra )
2794             {
2795                 h->stat.frame.i_mb_count_8x8dct[0] ++;
2796                 h->stat.frame.i_mb_count_8x8dct[1] += h->mb.b_transform_8x8;
2797             }
2798             if( b_intra && h->mb.i_type != I_PCM )
2799             {
2800                 if( h->mb.i_type == I_16x16 )
2801                     h->stat.frame.i_mb_pred_mode[0][h->mb.i_intra16x16_pred_mode]++;
2802                 else if( h->mb.i_type == I_8x8 )
2803                     for( int i = 0; i < 16; i += 4 )
2804                         h->stat.frame.i_mb_pred_mode[1][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
2805                 else //if( h->mb.i_type == I_4x4 )
2806                     for( int i = 0; i < 16; i++ )
2807                         h->stat.frame.i_mb_pred_mode[2][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
2808                 h->stat.frame.i_mb_pred_mode[3][x264_mb_chroma_pred_mode_fix[h->mb.i_chroma_pred_mode]]++;
2809             }
2810             h->stat.frame.i_mb_field[b_intra?0:b_skip?2:1] += MB_INTERLACED;
2811         }
2812
2813         /* calculate deblock strength values (actual deblocking is done per-row along with hpel) */
2814         if( b_deblock )
2815             x264_macroblock_deblock_strength( h );
2816
2817         if( mb_xy == h->sh.i_last_mb )
2818             break;
2819
2820         if( SLICE_MBAFF )
2821         {
2822             i_mb_x += i_mb_y & 1;
2823             i_mb_y ^= i_mb_x < h->mb.i_mb_width;
2824         }
2825         else
2826             i_mb_x++;
2827         if( i_mb_x == h->mb.i_mb_width )
2828         {
2829             i_mb_y++;
2830             i_mb_x = 0;
2831         }
2832     }
2833     if( h->sh.i_last_mb < h->sh.i_first_mb )
2834         return 0;
2835
2836     h->out.nal[h->out.i_nal].i_last_mb = h->sh.i_last_mb;
2837
2838     if( h->param.b_cabac )
2839     {
2840         x264_cabac_encode_flush( h, &h->cabac );
2841         h->out.bs.p = h->cabac.p;
2842     }
2843     else
2844     {
2845         if( i_skip > 0 )
2846             bs_write_ue( &h->out.bs, i_skip );  /* last skip run */
2847         /* rbsp_slice_trailing_bits */
2848         bs_rbsp_trailing( &h->out.bs );
2849         bs_flush( &h->out.bs );
2850     }
2851     if( x264_nal_end( h ) )
2852         return -1;
2853
2854     if( h->sh.i_last_mb == (h->i_threadslice_end * h->mb.i_mb_width - 1) )
2855     {
2856         h->stat.frame.i_misc_bits = bs_pos( &h->out.bs )
2857                                   + (h->out.i_nal*NALU_OVERHEAD * 8)
2858                                   - h->stat.frame.i_tex_bits
2859                                   - h->stat.frame.i_mv_bits;
2860         x264_fdec_filter_row( h, h->i_threadslice_end, 0 );
2861
2862         if( h->param.b_sliced_threads )
2863         {
2864             /* Tell the main thread we're done. */
2865             x264_threadslice_cond_broadcast( h, 1 );
2866             /* Do hpel now */
2867             for( int mb_y = h->i_threadslice_start; mb_y <= h->i_threadslice_end; mb_y++ )
2868                 x264_fdec_filter_row( h, mb_y, 1 );
2869             x264_threadslice_cond_broadcast( h, 2 );
2870             /* Do the first row of hpel, now that the previous slice is done */
2871             if( h->i_thread_idx > 0 )
2872             {
2873                 x264_threadslice_cond_wait( h->thread[h->i_thread_idx-1], 2 );
2874                 x264_fdec_filter_row( h, h->i_threadslice_start + (1 << SLICE_MBAFF), 2 );
2875             }
2876         }
2877
2878         /* Free mb info after the last thread's done using it */
2879         if( h->fdec->mb_info_free && (!h->param.b_sliced_threads || h->i_thread_idx == (h->param.i_threads-1)) )
2880         {
2881             h->fdec->mb_info_free( h->fdec->mb_info );
2882             h->fdec->mb_info = NULL;
2883             h->fdec->mb_info_free = NULL;
2884         }
2885     }
2886
2887     return 0;
2888 }
2889
2890 static void x264_thread_sync_context( x264_t *dst, x264_t *src )
2891 {
2892     if( dst == src )
2893         return;
2894
2895     // reference counting
2896     for( x264_frame_t **f = src->frames.reference; *f; f++ )
2897         (*f)->i_reference_count++;
2898     for( x264_frame_t **f = dst->frames.reference; *f; f++ )
2899         x264_frame_push_unused( src, *f );
2900     src->fdec->i_reference_count++;
2901     x264_frame_push_unused( src, dst->fdec );
2902
2903     // copy everything except the per-thread pointers and the constants.
2904     memcpy( &dst->i_frame, &src->i_frame, offsetof(x264_t, mb.base) - offsetof(x264_t, i_frame) );
2905     dst->param = src->param;
2906     dst->stat = src->stat;
2907     dst->pixf = src->pixf;
2908 }
2909
2910 static void x264_thread_sync_stat( x264_t *dst, x264_t *src )
2911 {
2912     if( dst == src )
2913         return;
2914     memcpy( &dst->stat.i_frame_count, &src->stat.i_frame_count, sizeof(dst->stat) - sizeof(dst->stat.frame) );
2915 }
2916
2917 static void *x264_slices_write( x264_t *h )
2918 {
2919     int i_slice_num = 0;
2920     int last_thread_mb = h->sh.i_last_mb;
2921
2922 #if HAVE_VISUALIZE
2923     if( h->param.b_visualize )
2924         if( x264_visualize_init( h ) )
2925             goto fail;
2926 #endif
2927
2928     /* init stats */
2929     memset( &h->stat.frame, 0, sizeof(h->stat.frame) );
2930     h->mb.b_reencode_mb = 0;
2931     while( h->sh.i_first_mb + SLICE_MBAFF*h->mb.i_mb_stride <= last_thread_mb )
2932     {
2933         h->sh.i_last_mb = last_thread_mb;
2934         if( !i_slice_num || !x264_frame_new_slice( h, h->fdec ) )
2935         {
2936             if( h->param.i_slice_max_mbs )
2937             {
2938                 if( SLICE_MBAFF )
2939                 {
2940                     // convert first to mbaff form, add slice-max-mbs, then convert back to normal form
2941                     int last_mbaff = 2*(h->sh.i_first_mb % h->mb.i_mb_width)
2942                         + h->mb.i_mb_width*(h->sh.i_first_mb / h->mb.i_mb_width)
2943                         + h->param.i_slice_max_mbs - 1;
2944                     int last_x = (last_mbaff % (2*h->mb.i_mb_width))/2;
2945                     int last_y = (last_mbaff / (2*h->mb.i_mb_width))*2 + 1;
2946                     h->sh.i_last_mb = last_x + h->mb.i_mb_stride*last_y;
2947                 }
2948                 else
2949                 {
2950                     h->sh.i_last_mb = h->sh.i_first_mb + h->param.i_slice_max_mbs - 1;
2951                     if( h->sh.i_last_mb < last_thread_mb && last_thread_mb - h->sh.i_last_mb < h->param.i_slice_min_mbs )
2952                         h->sh.i_last_mb = last_thread_mb - h->param.i_slice_min_mbs;
2953                 }
2954                 i_slice_num++;
2955             }
2956             else if( h->param.i_slice_count && !h->param.b_sliced_threads )
2957             {
2958                 int height = h->mb.i_mb_height >> PARAM_INTERLACED;
2959                 int width = h->mb.i_mb_width << PARAM_INTERLACED;
2960                 i_slice_num++;
2961                 h->sh.i_last_mb = (height * i_slice_num + h->param.i_slice_count/2) / h->param.i_slice_count * width - 1;
2962             }
2963         }
2964         h->sh.i_last_mb = X264_MIN( h->sh.i_last_mb, last_thread_mb );
2965         if( x264_stack_align( x264_slice_write, h ) )
2966             goto fail;
2967         h->sh.i_first_mb = h->sh.i_last_mb + 1;
2968         // if i_first_mb is not the last mb in a row then go to the next mb in MBAFF order
2969         if( SLICE_MBAFF && h->sh.i_first_mb % h->mb.i_mb_width )
2970             h->sh.i_first_mb -= h->mb.i_mb_stride;
2971     }
2972
2973 #if HAVE_VISUALIZE
2974     if( h->param.b_visualize )
2975     {
2976         x264_visualize_show( h );
2977         x264_visualize_close( h );
2978     }
2979 #endif
2980
2981     return (void *)0;
2982
2983 fail:
2984     /* Tell other threads we're done, so they wouldn't wait for it */
2985     if( h->param.b_sliced_threads )
2986         x264_threadslice_cond_broadcast( h, 2 );
2987     return (void *)-1;
2988 }
2989
2990 static int x264_threaded_slices_write( x264_t *h )
2991 {
2992     /* set first/last mb and sync contexts */
2993     for( int i = 0; i < h->param.i_threads; i++ )
2994     {
2995         x264_t *t = h->thread[i];
2996         if( i )
2997         {
2998             t->param = h->param;
2999             memcpy( &t->i_frame, &h->i_frame, offsetof(x264_t, rc) - offsetof(x264_t, i_frame) );
3000         }
3001         int height = h->mb.i_mb_height >> PARAM_INTERLACED;
3002         t->i_threadslice_start = ((height *  i    + h->param.i_slice_count/2) / h->param.i_threads) << PARAM_INTERLACED;
3003         t->i_threadslice_end   = ((height * (i+1) + h->param.i_slice_count/2) / h->param.i_threads) << PARAM_INTERLACED;
3004         t->sh.i_first_mb = t->i_threadslice_start * h->mb.i_mb_width;
3005         t->sh.i_last_mb  =   t->i_threadslice_end * h->mb.i_mb_width - 1;
3006     }
3007
3008     x264_stack_align( x264_analyse_weight_frame, h, h->mb.i_mb_height*16 + 16 );
3009
3010     x264_threads_distribute_ratecontrol( h );
3011
3012     /* setup */
3013     for( int i = 0; i < h->param.i_threads; i++ )
3014     {
3015         h->thread[i]->i_thread_idx = i;
3016         h->thread[i]->b_thread_active = 1;
3017         x264_threadslice_cond_broadcast( h->thread[i], 0 );
3018     }
3019     /* dispatch */
3020     for( int i = 0; i < h->param.i_threads; i++ )
3021         x264_threadpool_run( h->threadpool, (void*)x264_slices_write, h->thread[i] );
3022     /* wait */
3023     for( int i = 0; i < h->param.i_threads; i++ )
3024         x264_threadslice_cond_wait( h->thread[i], 1 );
3025
3026     x264_threads_merge_ratecontrol( h );
3027
3028     for( int i = 1; i < h->param.i_threads; i++ )
3029     {
3030         x264_t *t = h->thread[i];
3031         for( int j = 0; j < t->out.i_nal; j++ )
3032         {
3033             h->out.nal[h->out.i_nal] = t->out.nal[j];
3034             h->out.i_nal++;
3035             x264_nal_check_buffer( h );
3036         }
3037         /* All entries in stat.frame are ints except for ssd/ssim. */
3038         for( int j = 0; j < (offsetof(x264_t,stat.frame.i_ssd) - offsetof(x264_t,stat.frame.i_mv_bits)) / sizeof(int); j++ )
3039             ((int*)&h->stat.frame)[j] += ((int*)&t->stat.frame)[j];
3040         for( int j = 0; j < 3; j++ )
3041             h->stat.frame.i_ssd[j] += t->stat.frame.i_ssd[j];
3042         h->stat.frame.f_ssim += t->stat.frame.f_ssim;
3043         h->stat.frame.i_ssim_cnt += t->stat.frame.i_ssim_cnt;
3044     }
3045
3046     return 0;
3047 }
3048
3049 void x264_encoder_intra_refresh( x264_t *h )
3050 {
3051     h = h->thread[h->i_thread_phase];
3052     h->b_queued_intra_refresh = 1;
3053 }
3054
3055 int x264_encoder_invalidate_reference( x264_t *h, int64_t pts )
3056 {
3057     if( h->param.i_bframe )
3058     {
3059         x264_log( h, X264_LOG_ERROR, "x264_encoder_invalidate_reference is not supported with B-frames enabled\n" );
3060         return -1;
3061     }
3062     if( h->param.b_intra_refresh )
3063     {
3064         x264_log( h, X264_LOG_ERROR, "x264_encoder_invalidate_reference is not supported with intra refresh enabled\n" );
3065         return -1;
3066     }
3067     h = h->thread[h->i_thread_phase];
3068     if( pts >= h->i_last_idr_pts )
3069     {
3070         for( int i = 0; h->frames.reference[i]; i++ )
3071             if( pts <= h->frames.reference[i]->i_pts )
3072                 h->frames.reference[i]->b_corrupt = 1;
3073         if( pts <= h->fdec->i_pts )
3074             h->fdec->b_corrupt = 1;
3075     }
3076     return 0;
3077 }
3078
3079 /****************************************************************************
3080  * x264_encoder_encode:
3081  *  XXX: i_poc   : is the poc of the current given picture
3082  *       i_frame : is the number of the frame being coded
3083  *  ex:  type frame poc
3084  *       I      0   2*0
3085  *       P      1   2*3
3086  *       B      2   2*1
3087  *       B      3   2*2
3088  *       P      4   2*6
3089  *       B      5   2*4
3090  *       B      6   2*5
3091  ****************************************************************************/
3092 int     x264_encoder_encode( x264_t *h,
3093                              x264_nal_t **pp_nal, int *pi_nal,
3094                              x264_picture_t *pic_in,
3095                              x264_picture_t *pic_out )
3096 {
3097     x264_t *thread_current, *thread_prev, *thread_oldest;
3098     int i_nal_type, i_nal_ref_idc, i_global_qp;
3099     int overhead = NALU_OVERHEAD;
3100
3101 #if HAVE_OPENCL
3102     if( h->opencl.b_fatal_error )
3103         return -1;
3104 #endif
3105
3106     if( h->i_thread_frames > 1 )
3107     {
3108         thread_prev    = h->thread[ h->i_thread_phase ];
3109         h->i_thread_phase = (h->i_thread_phase + 1) % h->i_thread_frames;
3110         thread_current = h->thread[ h->i_thread_phase ];
3111         thread_oldest  = h->thread[ (h->i_thread_phase + 1) % h->i_thread_frames ];
3112         x264_thread_sync_context( thread_current, thread_prev );
3113         x264_thread_sync_ratecontrol( thread_current, thread_prev, thread_oldest );
3114         h = thread_current;
3115     }
3116     else
3117     {
3118         thread_current =
3119         thread_oldest  = h;
3120     }
3121     h->i_cpb_delay_pir_offset = h->i_cpb_delay_pir_offset_next;
3122
3123     /* no data out */
3124     *pi_nal = 0;
3125     *pp_nal = NULL;
3126
3127     /* ------------------- Setup new frame from picture -------------------- */
3128     if( pic_in != NULL )
3129     {
3130         /* 1: Copy the picture to a frame and move it to a buffer */
3131         x264_frame_t *fenc = x264_frame_pop_unused( h, 0 );
3132         if( !fenc )
3133             return -1;
3134
3135         if( x264_frame_copy_picture( h, fenc, pic_in ) < 0 )
3136             return -1;
3137
3138         if( h->param.i_width != 16 * h->mb.i_mb_width ||
3139             h->param.i_height != 16 * h->mb.i_mb_height )
3140             x264_frame_expand_border_mod16( h, fenc );
3141
3142         fenc->i_frame = h->frames.i_input++;
3143
3144         if( fenc->i_frame == 0 )
3145             h->frames.i_first_pts = fenc->i_pts;
3146         if( h->frames.i_bframe_delay && fenc->i_frame == h->frames.i_bframe_delay )
3147             h->frames.i_bframe_delay_time = fenc->i_pts - h->frames.i_first_pts;
3148
3149         if( h->param.b_vfr_input && fenc->i_pts <= h->frames.i_largest_pts )
3150             x264_log( h, X264_LOG_WARNING, "non-strictly-monotonic PTS\n" );
3151
3152         h->frames.i_second_largest_pts = h->frames.i_largest_pts;
3153         h->frames.i_largest_pts = fenc->i_pts;
3154
3155         if( (fenc->i_pic_struct < PIC_STRUCT_AUTO) || (fenc->i_pic_struct > PIC_STRUCT_TRIPLE) )
3156             fenc->i_pic_struct = PIC_STRUCT_AUTO;
3157
3158         if( fenc->i_pic_struct == PIC_STRUCT_AUTO )
3159         {
3160 #if HAVE_INTERLACED
3161             int b_interlaced = fenc->param ? fenc->param->b_interlaced : h->param.b_interlaced;
3162 #else
3163             int b_interlaced = 0;
3164 #endif
3165             if( b_interlaced )
3166             {
3167                 int b_tff = fenc->param ? fenc->param->b_tff : h->param.b_tff;
3168                 fenc->i_pic_struct = b_tff ? PIC_STRUCT_TOP_BOTTOM : PIC_STRUCT_BOTTOM_TOP;
3169             }
3170             else
3171                 fenc->i_pic_struct = PIC_STRUCT_PROGRESSIVE;
3172         }
3173
3174         if( h->param.rc.b_mb_tree && h->param.rc.b_stat_read )
3175         {
3176             if( x264_macroblock_tree_read( h, fenc, pic_in->prop.quant_offsets ) )
3177                 return -1;
3178         }
3179         else
3180             x264_stack_align( x264_adaptive_quant_frame, h, fenc, pic_in->prop.quant_offsets );
3181
3182         if( pic_in->prop.quant_offsets_free )
3183             pic_in->prop.quant_offsets_free( pic_in->prop.quant_offsets );
3184
3185         if( h->frames.b_have_lowres )
3186             x264_frame_init_lowres( h, fenc );
3187
3188         /* 2: Place the frame into the queue for its slice type decision */
3189         x264_lookahead_put_frame( h, fenc );
3190
3191         if( h->frames.i_input <= h->frames.i_delay + 1 - h->i_thread_frames )
3192         {
3193             /* Nothing yet to encode, waiting for filling of buffers */
3194             pic_out->i_type = X264_TYPE_AUTO;
3195             return 0;
3196         }
3197     }
3198     else
3199     {
3200         /* signal kills for lookahead thread */
3201         x264_pthread_mutex_lock( &h->lookahead->ifbuf.mutex );
3202         h->lookahead->b_exit_thread = 1;
3203         x264_pthread_cond_broadcast( &h->lookahead->ifbuf.cv_fill );
3204         x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
3205     }
3206
3207     h->i_frame++;
3208     /* 3: The picture is analyzed in the lookahead */
3209     if( !h->frames.current[0] )
3210         x264_lookahead_get_frames( h );
3211
3212     if( !h->frames.current[0] && x264_lookahead_is_empty( h ) )
3213         return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
3214
3215     /* ------------------- Get frame to be encoded ------------------------- */
3216     /* 4: get picture to encode */
3217     h->fenc = x264_frame_shift( h->frames.current );
3218
3219     /* If applicable, wait for previous frame reconstruction to finish */
3220     if( h->param.b_sliced_threads )
3221         if( x264_threadpool_wait_all( h ) < 0 )
3222             return -1;
3223
3224     if( h->i_frame == h->i_thread_frames - 1 )
3225         h->i_reordered_pts_delay = h->fenc->i_reordered_pts;
3226     if( h->fenc->param )
3227     {
3228         x264_encoder_reconfig( h, h->fenc->param );
3229         if( h->fenc->param->param_free )
3230         {
3231             h->fenc->param->param_free( h->fenc->param );
3232             h->fenc->param = NULL;
3233         }
3234     }
3235
3236     // ok to call this before encoding any frames, since the initial values of fdec have b_kept_as_ref=0
3237     if( x264_reference_update( h ) )
3238         return -1;
3239     h->fdec->i_lines_completed = -1;
3240
3241     if( !IS_X264_TYPE_I( h->fenc->i_type ) )
3242     {
3243         int valid_refs_left = 0;
3244         for( int i = 0; h->frames.reference[i]; i++ )
3245             if( !h->frames.reference[i]->b_corrupt )
3246                 valid_refs_left++;
3247         /* No valid reference frames left: force an IDR. */
3248         if( !valid_refs_left )
3249         {
3250             h->fenc->b_keyframe = 1;
3251             h->fenc->i_type = X264_TYPE_IDR;
3252         }
3253     }
3254
3255     if( h->fenc->b_keyframe )
3256     {
3257         h->frames.i_last_keyframe = h->fenc->i_frame;
3258         if( h->fenc->i_type == X264_TYPE_IDR )
3259         {
3260             h->i_frame_num = 0;
3261             h->frames.i_last_idr = h->fenc->i_frame;
3262         }
3263     }
3264     h->sh.i_mmco_command_count =
3265     h->sh.i_mmco_remove_from_end = 0;
3266     h->b_ref_reorder[0] =
3267     h->b_ref_reorder[1] = 0;
3268     h->fdec->i_poc =
3269     h->fenc->i_poc = 2 * ( h->fenc->i_frame - X264_MAX( h->frames.i_last_idr, 0 ) );
3270
3271     /* ------------------- Setup frame context ----------------------------- */
3272     /* 5: Init data dependent of frame type */
3273     if( h->fenc->i_type == X264_TYPE_IDR )
3274     {
3275         /* reset ref pictures */
3276         i_nal_type    = NAL_SLICE_IDR;
3277         i_nal_ref_idc = NAL_PRIORITY_HIGHEST;
3278         h->sh.i_type = SLICE_TYPE_I;
3279         x264_reference_reset( h );
3280         h->frames.i_poc_last_open_gop = -1;
3281     }
3282     else if( h->fenc->i_type == X264_TYPE_I )
3283     {
3284         i_nal_type    = NAL_SLICE;
3285         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
3286         h->sh.i_type = SLICE_TYPE_I;
3287         x264_reference_hierarchy_reset( h );
3288         if( h->param.b_open_gop )
3289             h->frames.i_poc_last_open_gop = h->fenc->b_keyframe ? h->fenc->i_poc : -1;
3290     }
3291     else if( h->fenc->i_type == X264_TYPE_P )
3292     {
3293         i_nal_type    = NAL_SLICE;
3294         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
3295         h->sh.i_type = SLICE_TYPE_P;
3296         x264_reference_hierarchy_reset( h );
3297         h->frames.i_poc_last_open_gop = -1;
3298     }
3299     else if( h->fenc->i_type == X264_TYPE_BREF )
3300     {
3301         i_nal_type    = NAL_SLICE;
3302         i_nal_ref_idc = h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT ? NAL_PRIORITY_LOW : NAL_PRIORITY_HIGH;
3303         h->sh.i_type = SLICE_TYPE_B;
3304         x264_reference_hierarchy_reset( h );
3305     }
3306     else    /* B frame */
3307     {
3308         i_nal_type    = NAL_SLICE;
3309         i_nal_ref_idc = NAL_PRIORITY_DISPOSABLE;
3310         h->sh.i_type = SLICE_TYPE_B;
3311     }
3312
3313     h->fdec->i_type = h->fenc->i_type;
3314     h->fdec->i_frame = h->fenc->i_frame;
3315     h->fenc->b_kept_as_ref =
3316     h->fdec->b_kept_as_ref = i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE && h->param.i_keyint_max > 1;
3317
3318     h->fdec->mb_info = h->fenc->mb_info;
3319     h->fdec->mb_info_free = h->fenc->mb_info_free;
3320     h->fenc->mb_info = NULL;
3321     h->fenc->mb_info_free = NULL;
3322
3323     h->fdec->i_pts = h->fenc->i_pts;
3324     if( h->frames.i_bframe_delay )
3325     {
3326         int64_t *prev_reordered_pts = thread_current->frames.i_prev_reordered_pts;
3327         h->fdec->i_dts = h->i_frame > h->frames.i_bframe_delay
3328                        ? prev_reordered_pts[ (h->i_frame - h->frames.i_bframe_delay) % h->frames.i_bframe_delay ]
3329                        : h->fenc->i_reordered_pts - h->frames.i_bframe_delay_time;
3330         prev_reordered_pts[ h->i_frame % h->frames.i_bframe_delay ] = h->fenc->i_reordered_pts;
3331     }
3332     else
3333         h->fdec->i_dts = h->fenc->i_reordered_pts;
3334     if( h->fenc->i_type == X264_TYPE_IDR )
3335         h->i_last_idr_pts = h->fdec->i_pts;
3336
3337     /* ------------------- Init                ----------------------------- */
3338     /* build ref list 0/1 */
3339     x264_reference_build_list( h, h->fdec->i_poc );
3340
3341     /* ---------------------- Write the bitstream -------------------------- */
3342     /* Init bitstream context */
3343     if( h->param.b_sliced_threads )
3344     {
3345         for( int i = 0; i < h->param.i_threads; i++ )
3346         {
3347             bs_init( &h->thread[i]->out.bs, h->thread[i]->out.p_bitstream, h->thread[i]->out.i_bitstream );
3348             h->thread[i]->out.i_nal = 0;
3349         }
3350     }
3351     else
3352     {
3353         bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
3354         h->out.i_nal = 0;
3355     }
3356
3357     if( h->param.b_aud )
3358     {
3359         int pic_type;
3360
3361         if( h->sh.i_type == SLICE_TYPE_I )
3362             pic_type = 0;
3363         else if( h->sh.i_type == SLICE_TYPE_P )
3364             pic_type = 1;
3365         else if( h->sh.i_type == SLICE_TYPE_B )
3366             pic_type = 2;
3367         else
3368             pic_type = 7;
3369
3370         x264_nal_start( h, NAL_AUD, NAL_PRIORITY_DISPOSABLE );
3371         bs_write( &h->out.bs, 3, pic_type );
3372         bs_rbsp_trailing( &h->out.bs );
3373         if( x264_nal_end( h ) )
3374             return -1;
3375         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
3376     }
3377
3378     h->i_nal_type = i_nal_type;
3379     h->i_nal_ref_idc = i_nal_ref_idc;
3380
3381     if( h->param.b_intra_refresh )
3382     {
3383         if( IS_X264_TYPE_I( h->fenc->i_type ) )
3384         {
3385             h->fdec->i_frames_since_pir = 0;
3386             h->b_queued_intra_refresh = 0;
3387             /* PIR is currently only supported with ref == 1, so any intra frame effectively refreshes
3388              * the whole frame and counts as an intra refresh. */
3389             h->fdec->f_pir_position = h->mb.i_mb_width;
3390         }
3391         else if( h->fenc->i_type == X264_TYPE_P )
3392         {
3393             int pocdiff = (h->fdec->i_poc - h->fref[0][0]->i_poc)/2;
3394             float increment = X264_MAX( ((float)h->mb.i_mb_width-1) / h->param.i_keyint_max, 1 );
3395             h->fdec->f_pir_position = h->fref[0][0]->f_pir_position;
3396             h->fdec->i_frames_since_pir = h->fref[0][0]->i_frames_since_pir + pocdiff;
3397             if( h->fdec->i_frames_since_pir >= h->param.i_keyint_max ||
3398                 (h->b_queued_intra_refresh && h->fdec->f_pir_position + 0.5 >= h->mb.i_mb_width) )
3399             {
3400                 h->fdec->f_pir_position = 0;
3401                 h->fdec->i_frames_since_pir = 0;
3402                 h->b_queued_intra_refresh = 0;
3403                 h->fenc->b_keyframe = 1;
3404             }
3405             h->fdec->i_pir_start_col = h->fdec->f_pir_position+0.5;
3406             h->fdec->f_pir_position += increment * pocdiff;
3407             h->fdec->i_pir_end_col = h->fdec->f_pir_position+0.5;
3408             /* If our intra refresh has reached the right side of the frame, we're done. */
3409             if( h->fdec->i_pir_end_col >= h->mb.i_mb_width - 1 )
3410             {
3411                 h->fdec->f_pir_position = h->mb.i_mb_width;
3412                 h->fdec->i_pir_end_col = h->mb.i_mb_width - 1;
3413             }
3414         }
3415     }
3416
3417     if( h->fenc->b_keyframe )
3418     {
3419         /* Write SPS and PPS */
3420         if( h->param.b_repeat_headers )
3421         {
3422             /* generate sequence parameters */
3423             x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
3424             x264_sps_write( &h->out.bs, h->sps );
3425             if( x264_nal_end( h ) )
3426                 return -1;
3427             /* Pad AUD/SPS to 256 bytes like Panasonic */
3428             if( h->param.b_avcintra_compat )
3429                 h->out.nal[h->out.i_nal-1].i_padding = 256 - bs_pos( &h->out.bs ) / 8 - 2*NALU_OVERHEAD;
3430             overhead += h->out.nal[h->out.i_nal-1].i_payload + h->out.nal[h->out.i_nal-1].i_padding + NALU_OVERHEAD;
3431
3432             /* generate picture parameters */
3433             x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
3434             x264_pps_write( &h->out.bs, h->sps, h->pps );
3435             if( x264_nal_end( h ) )
3436                 return -1;
3437             if( h->param.b_avcintra_compat )
3438                 h->out.nal[h->out.i_nal-1].i_padding = 256 - h->out.nal[h->out.i_nal-1].i_payload - NALU_OVERHEAD;
3439             overhead += h->out.nal[h->out.i_nal-1].i_payload + h->out.nal[h->out.i_nal-1].i_padding + NALU_OVERHEAD;
3440         }
3441
3442         /* when frame threading is used, buffering period sei is written in x264_encoder_frame_end */
3443         if( h->i_thread_frames == 1 && h->sps->vui.b_nal_hrd_parameters_present )
3444         {
3445             x264_hrd_fullness( h );
3446             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3447             x264_sei_buffering_period_write( h, &h->out.bs );
3448             if( x264_nal_end( h ) )
3449                return -1;
3450             overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3451         }
3452     }
3453
3454     /* write extra sei */
3455     for( int i = 0; i < h->fenc->extra_sei.num_payloads; i++ )
3456     {
3457         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3458         x264_sei_write( &h->out.bs, h->fenc->extra_sei.payloads[i].payload, h->fenc->extra_sei.payloads[i].payload_size,
3459                         h->fenc->extra_sei.payloads[i].payload_type );
3460         if( x264_nal_end( h ) )
3461             return -1;
3462         overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3463         if( h->fenc->extra_sei.sei_free )
3464         {
3465             h->fenc->extra_sei.sei_free( h->fenc->extra_sei.payloads[i].payload );
3466             h->fenc->extra_sei.payloads[i].payload = NULL;
3467         }
3468     }
3469
3470     if( h->fenc->extra_sei.sei_free )
3471     {
3472         h->fenc->extra_sei.sei_free( h->fenc->extra_sei.payloads );
3473         h->fenc->extra_sei.payloads = NULL;
3474         h->fenc->extra_sei.sei_free = NULL;
3475     }
3476
3477     if( h->fenc->b_keyframe )
3478     {
3479         /* Avid's decoder strictly wants two SEIs for AVC-Intra so we can't insert the x264 SEI */
3480         if( h->param.b_repeat_headers && h->fenc->i_frame == 0 && !h->param.b_avcintra_compat )
3481         {
3482             /* identify ourself */
3483             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3484             if( x264_sei_version_write( h, &h->out.bs ) )
3485                 return -1;
3486             if( x264_nal_end( h ) )
3487                 return -1;
3488             overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3489         }
3490
3491         if( h->fenc->i_type != X264_TYPE_IDR )
3492         {
3493             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;
3494             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3495             x264_sei_recovery_point_write( h, &h->out.bs, time_to_recovery );
3496             if( x264_nal_end( h ) )
3497                 return -1;
3498             overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3499         }
3500
3501         if( h->param.i_frame_packing >= 0 )
3502         {
3503             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3504             x264_sei_frame_packing_write( h, &h->out.bs );
3505             if( x264_nal_end( h ) )
3506                 return -1;
3507             overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3508         }
3509     }
3510
3511     /* generate sei pic timing */
3512     if( h->sps->vui.b_pic_struct_present || h->sps->vui.b_nal_hrd_parameters_present )
3513     {
3514         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3515         x264_sei_pic_timing_write( h, &h->out.bs );
3516         if( x264_nal_end( h ) )
3517             return -1;
3518         overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3519     }
3520
3521     /* As required by Blu-ray. */
3522     if( !IS_X264_TYPE_B( h->fenc->i_type ) && h->b_sh_backup )
3523     {
3524         h->b_sh_backup = 0;
3525         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3526         x264_sei_dec_ref_pic_marking_write( h, &h->out.bs );
3527         if( x264_nal_end( h ) )
3528             return -1;
3529         overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3530     }
3531
3532     if( h->fenc->b_keyframe && h->param.b_intra_refresh )
3533         h->i_cpb_delay_pir_offset_next = h->fenc->i_cpb_delay;
3534
3535     /* Filler space: 10 or 18 SEIs' worth of space, depending on resolution */
3536     if( h->param.b_avcintra_compat )
3537     {
3538         /* Write an empty filler NAL to mimic the AUD in the P2 format*/
3539         if( h->param.b_avcintra_compat )
3540         {
3541             x264_nal_start( h, NAL_FILLER, NAL_PRIORITY_DISPOSABLE );
3542             x264_filler_write( h, &h->out.bs, 0 );
3543             if( x264_nal_end( h ) )
3544                 return -1;
3545             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
3546         }
3547
3548         /* All lengths are magic lengths that decoders expect to see */
3549         /* "UMID" SEI */
3550         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3551         if( x264_sei_avcintra_write( h, &h->out.bs, 497, "UMID" ) < 0 )
3552             return -1;
3553         if( x264_nal_end( h ) )
3554             return -1;
3555         overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3556
3557
3558         int unpadded_len;
3559         int total_len;
3560         if( h->param.i_height == 1080 )
3561         {
3562             unpadded_len = 5780;
3563             total_len = 17*512;
3564         }
3565         else
3566         {
3567             unpadded_len = 2900;
3568             total_len = 9*512;
3569         }
3570         /* "VANC" SEI */
3571         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3572         if( x264_sei_avcintra_write( h, &h->out.bs, unpadded_len, "VANC" ) < 0 )
3573             return -1;
3574         if( x264_nal_end( h ) )
3575             return -1;
3576
3577         h->out.nal[h->out.i_nal-1].i_padding = total_len - h->out.nal[h->out.i_nal-1].i_payload - SEI_OVERHEAD;
3578         overhead += h->out.nal[h->out.i_nal-1].i_payload + h->out.nal[h->out.i_nal-1].i_padding + SEI_OVERHEAD;
3579     }
3580
3581     /* Init the rate control */
3582     /* FIXME: Include slice header bit cost. */
3583     x264_ratecontrol_start( h, h->fenc->i_qpplus1, overhead*8 );
3584     i_global_qp = x264_ratecontrol_qp( h );
3585
3586     pic_out->i_qpplus1 =
3587     h->fdec->i_qpplus1 = i_global_qp + 1;
3588
3589     if( h->param.rc.b_stat_read && h->sh.i_type != SLICE_TYPE_I )
3590     {
3591         x264_reference_build_list_optimal( h );
3592         x264_reference_check_reorder( h );
3593     }
3594
3595     if( h->i_ref[0] )
3596         h->fdec->i_poc_l0ref0 = h->fref[0][0]->i_poc;
3597
3598     /* ------------------------ Create slice header  ----------------------- */
3599     x264_slice_init( h, i_nal_type, i_global_qp );
3600
3601     /*------------------------- Weights -------------------------------------*/
3602     if( h->sh.i_type == SLICE_TYPE_B )
3603         x264_macroblock_bipred_init( h );
3604
3605     x264_weighted_pred_init( h );
3606
3607     if( i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE )
3608         h->i_frame_num++;
3609
3610     /* Write frame */
3611     h->i_threadslice_start = 0;
3612     h->i_threadslice_end = h->mb.i_mb_height;
3613     if( h->i_thread_frames > 1 )
3614     {
3615         x264_threadpool_run( h->threadpool, (void*)x264_slices_write, h );
3616         h->b_thread_active = 1;
3617     }
3618     else if( h->param.b_sliced_threads )
3619     {
3620         if( x264_threaded_slices_write( h ) )
3621             return -1;
3622     }
3623     else
3624         if( (intptr_t)x264_slices_write( h ) )
3625             return -1;
3626
3627     return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
3628 }
3629
3630 static int x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
3631                                    x264_nal_t **pp_nal, int *pi_nal,
3632                                    x264_picture_t *pic_out )
3633 {
3634     char psz_message[80];
3635
3636     if( !h->param.b_sliced_threads && h->b_thread_active )
3637     {
3638         h->b_thread_active = 0;
3639         if( (intptr_t)x264_threadpool_wait( h->threadpool, h ) )
3640             return -1;
3641     }
3642     if( !h->out.i_nal )
3643     {
3644         pic_out->i_type = X264_TYPE_AUTO;
3645         return 0;
3646     }
3647
3648     x264_emms();
3649
3650     /* generate buffering period sei and insert it into place */
3651     if( h->i_thread_frames > 1 && h->fenc->b_keyframe && h->sps->vui.b_nal_hrd_parameters_present )
3652     {
3653         x264_hrd_fullness( h );
3654         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3655         x264_sei_buffering_period_write( h, &h->out.bs );
3656         if( x264_nal_end( h ) )
3657            return -1;
3658         /* buffering period sei must follow AUD, SPS and PPS and precede all other SEIs */
3659         int idx = 0;
3660         while( h->out.nal[idx].i_type == NAL_AUD ||
3661                h->out.nal[idx].i_type == NAL_SPS ||
3662                h->out.nal[idx].i_type == NAL_PPS )
3663             idx++;
3664         x264_nal_t nal_tmp = h->out.nal[h->out.i_nal-1];
3665         memmove( &h->out.nal[idx+1], &h->out.nal[idx], (h->out.i_nal-idx-1)*sizeof(x264_nal_t) );
3666         h->out.nal[idx] = nal_tmp;
3667     }
3668
3669     int frame_size = x264_encoder_encapsulate_nals( h, 0 );
3670     if( frame_size < 0 )
3671         return -1;
3672
3673     /* Set output picture properties */
3674     pic_out->i_type = h->fenc->i_type;
3675
3676     pic_out->b_keyframe = h->fenc->b_keyframe;
3677     pic_out->i_pic_struct = h->fenc->i_pic_struct;
3678
3679     pic_out->i_pts = h->fdec->i_pts;
3680     pic_out->i_dts = h->fdec->i_dts;
3681
3682     if( pic_out->i_pts < pic_out->i_dts )
3683         x264_log( h, X264_LOG_WARNING, "invalid DTS: PTS is less than DTS\n" );
3684
3685     pic_out->opaque = h->fenc->opaque;
3686
3687     pic_out->img.i_csp = h->fdec->i_csp;
3688 #if HIGH_BIT_DEPTH
3689     pic_out->img.i_csp |= X264_CSP_HIGH_DEPTH;
3690 #endif
3691     pic_out->img.i_plane = h->fdec->i_plane;
3692     for( int i = 0; i < pic_out->img.i_plane; i++ )
3693     {
3694         pic_out->img.i_stride[i] = h->fdec->i_stride[i] * sizeof(pixel);
3695         pic_out->img.plane[i] = (uint8_t*)h->fdec->plane[i];
3696     }
3697
3698     x264_frame_push_unused( thread_current, h->fenc );
3699
3700     /* ---------------------- Update encoder state ------------------------- */
3701
3702     /* update rc */
3703     int filler = 0;
3704     if( x264_ratecontrol_end( h, frame_size * 8, &filler ) < 0 )
3705         return -1;
3706
3707     pic_out->hrd_timing = h->fenc->hrd_timing;
3708     pic_out->prop.f_crf_avg = h->fdec->f_crf_avg;
3709
3710     /* Filler in AVC-Intra mode is written as zero bytes to the last slice
3711      * We don't know the size of the last slice until encapsulation so we add filler to the encapsulated NAL */
3712     if( h->param.b_avcintra_compat )
3713     {
3714         x264_t *h0 = h->thread[0];
3715         int ret = x264_check_encapsulated_buffer( h, h0, h->out.i_nal, frame_size, frame_size + filler );
3716         if( ret < 0 )
3717             return -1;
3718         memset( h->out.nal[0].p_payload + frame_size, 0, filler );
3719         h->out.nal[h->out.i_nal-1].i_payload += filler;
3720         h->out.nal[h->out.i_nal-1].i_padding = filler;
3721         frame_size += filler;
3722     }
3723     else
3724     {
3725         while( filler > 0 )
3726         {
3727             int f, overhead;
3728             overhead = (FILLER_OVERHEAD - h->param.b_annexb);
3729             if( h->param.i_slice_max_size && filler > h->param.i_slice_max_size )
3730             {
3731                 int next_size = filler - h->param.i_slice_max_size;
3732                 int overflow = X264_MAX( overhead - next_size, 0 );
3733                 f = h->param.i_slice_max_size - overhead - overflow;
3734             }
3735             else
3736                 f = X264_MAX( 0, filler - overhead );
3737
3738             if( x264_bitstream_check_buffer_filler( h, f ) )
3739                 return -1;
3740             x264_nal_start( h, NAL_FILLER, NAL_PRIORITY_DISPOSABLE );
3741             x264_filler_write( h, &h->out.bs, f );
3742             if( x264_nal_end( h ) )
3743                 return -1;
3744             int total_size = x264_encoder_encapsulate_nals( h, h->out.i_nal-1 );
3745             if( total_size < 0 )
3746                 return -1;
3747             frame_size += total_size;
3748             filler -= total_size;
3749         }
3750     }
3751
3752     /* End bitstream, set output  */
3753     *pi_nal = h->out.i_nal;
3754     *pp_nal = h->out.nal;
3755
3756     h->out.i_nal = 0;
3757
3758     x264_noise_reduction_update( h );
3759
3760     /* ---------------------- Compute/Print statistics --------------------- */
3761     x264_thread_sync_stat( h, h->thread[0] );
3762
3763     /* Slice stat */
3764     h->stat.i_frame_count[h->sh.i_type]++;
3765     h->stat.i_frame_size[h->sh.i_type] += frame_size;
3766     h->stat.f_frame_qp[h->sh.i_type] += h->fdec->f_qp_avg_aq;
3767
3768     for( int i = 0; i < X264_MBTYPE_MAX; i++ )
3769         h->stat.i_mb_count[h->sh.i_type][i] += h->stat.frame.i_mb_count[i];
3770     for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
3771         h->stat.i_mb_partition[h->sh.i_type][i] += h->stat.frame.i_mb_partition[i];
3772     for( int i = 0; i < 2; i++ )
3773         h->stat.i_mb_count_8x8dct[i] += h->stat.frame.i_mb_count_8x8dct[i];
3774     for( int i = 0; i < 6; i++ )
3775         h->stat.i_mb_cbp[i] += h->stat.frame.i_mb_cbp[i];
3776     for( int i = 0; i < 4; i++ )
3777         for( int j = 0; j < 13; j++ )
3778             h->stat.i_mb_pred_mode[i][j] += h->stat.frame.i_mb_pred_mode[i][j];
3779     if( h->sh.i_type != SLICE_TYPE_I )
3780         for( int i_list = 0; i_list < 2; i_list++ )
3781             for( int i = 0; i < X264_REF_MAX*2; i++ )
3782                 h->stat.i_mb_count_ref[h->sh.i_type][i_list][i] += h->stat.frame.i_mb_count_ref[i_list][i];
3783     for( int i = 0; i < 3; i++ )
3784         h->stat.i_mb_field[i] += h->stat.frame.i_mb_field[i];
3785     if( h->sh.i_type == SLICE_TYPE_P && h->param.analyse.i_weighted_pred >= X264_WEIGHTP_SIMPLE )
3786     {
3787         h->stat.i_wpred[0] += !!h->sh.weight[0][0].weightfn;
3788         h->stat.i_wpred[1] += !!h->sh.weight[0][1].weightfn || !!h->sh.weight[0][2].weightfn;
3789     }
3790     if( h->sh.i_type == SLICE_TYPE_B )
3791     {
3792         h->stat.i_direct_frames[ h->sh.b_direct_spatial_mv_pred ] ++;
3793         if( h->mb.b_direct_auto_write )
3794         {
3795             //FIXME somewhat arbitrary time constants
3796             if( h->stat.i_direct_score[0] + h->stat.i_direct_score[1] > h->mb.i_mb_count )
3797                 for( int i = 0; i < 2; i++ )
3798                     h->stat.i_direct_score[i] = h->stat.i_direct_score[i] * 9/10;
3799             for( int i = 0; i < 2; i++ )
3800                 h->stat.i_direct_score[i] += h->stat.frame.i_direct_score[i];
3801         }
3802     }
3803     else
3804         h->stat.i_consecutive_bframes[h->fenc->i_bframes]++;
3805
3806     psz_message[0] = '\0';
3807     double dur = h->fenc->f_duration;
3808     h->stat.f_frame_duration[h->sh.i_type] += dur;
3809     if( h->param.analyse.b_psnr )
3810     {
3811         int64_t ssd[3] =
3812         {
3813             h->stat.frame.i_ssd[0],
3814             h->stat.frame.i_ssd[1],
3815             h->stat.frame.i_ssd[2],
3816         };
3817         int luma_size = h->param.i_width * h->param.i_height;
3818         int chroma_size = CHROMA_SIZE( luma_size );
3819         pic_out->prop.f_psnr[0] = x264_psnr( ssd[0], luma_size );
3820         pic_out->prop.f_psnr[1] = x264_psnr( ssd[1], chroma_size );
3821         pic_out->prop.f_psnr[2] = x264_psnr( ssd[2], chroma_size );
3822         pic_out->prop.f_psnr_avg = x264_psnr( ssd[0] + ssd[1] + ssd[2], luma_size + chroma_size*2 );
3823
3824         h->stat.f_ssd_global[h->sh.i_type]   += dur * (ssd[0] + ssd[1] + ssd[2]);
3825         h->stat.f_psnr_average[h->sh.i_type] += dur * pic_out->prop.f_psnr_avg;
3826         h->stat.f_psnr_mean_y[h->sh.i_type]  += dur * pic_out->prop.f_psnr[0];
3827         h->stat.f_psnr_mean_u[h->sh.i_type]  += dur * pic_out->prop.f_psnr[1];
3828         h->stat.f_psnr_mean_v[h->sh.i_type]  += dur * pic_out->prop.f_psnr[2];
3829
3830         snprintf( psz_message, 80, " PSNR Y:%5.2f U:%5.2f V:%5.2f", pic_out->prop.f_psnr[0],
3831                                                                     pic_out->prop.f_psnr[1],
3832                                                                     pic_out->prop.f_psnr[2] );
3833     }
3834
3835     if( h->param.analyse.b_ssim )
3836     {
3837         pic_out->prop.f_ssim = h->stat.frame.f_ssim / h->stat.frame.i_ssim_cnt;
3838         h->stat.f_ssim_mean_y[h->sh.i_type] += pic_out->prop.f_ssim * dur;
3839         snprintf( psz_message + strlen(psz_message), 80 - strlen(psz_message),
3840                   " SSIM Y:%.5f", pic_out->prop.f_ssim );
3841     }
3842     psz_message[79] = '\0';
3843
3844     x264_log( h, X264_LOG_DEBUG,
3845                   "frame=%4d QP=%.2f NAL=%d Slice:%c Poc:%-3d I:%-4d P:%-4d SKIP:%-4d size=%d bytes%s\n",
3846               h->i_frame,
3847               h->fdec->f_qp_avg_aq,
3848               h->i_nal_ref_idc,
3849               h->sh.i_type == SLICE_TYPE_I ? 'I' : (h->sh.i_type == SLICE_TYPE_P ? 'P' : 'B' ),
3850               h->fdec->i_poc,
3851               h->stat.frame.i_mb_count_i,
3852               h->stat.frame.i_mb_count_p,
3853               h->stat.frame.i_mb_count_skip,
3854               frame_size,
3855               psz_message );
3856
3857     // keep stats all in one place
3858     x264_thread_sync_stat( h->thread[0], h );
3859     // for the use of the next frame
3860     x264_thread_sync_stat( thread_current, h );
3861
3862 #ifdef DEBUG_MB_TYPE
3863 {
3864     static const char mb_chars[] = { 'i', 'i', 'I', 'C', 'P', '8', 'S',
3865         'D', '<', 'X', 'B', 'X', '>', 'B', 'B', 'B', 'B', '8', 'S' };
3866     for( int mb_xy = 0; mb_xy < h->mb.i_mb_width * h->mb.i_mb_height; mb_xy++ )
3867     {
3868         if( h->mb.type[mb_xy] < X264_MBTYPE_MAX && h->mb.type[mb_xy] >= 0 )
3869             fprintf( stderr, "%c ", mb_chars[ h->mb.type[mb_xy] ] );
3870         else
3871             fprintf( stderr, "? " );
3872
3873         if( (mb_xy+1) % h->mb.i_mb_width == 0 )
3874             fprintf( stderr, "\n" );
3875     }
3876 }
3877 #endif
3878
3879     /* Remove duplicates, must be done near the end as breaks h->fref0 array
3880      * by freeing some of its pointers. */
3881     for( int i = 0; i < h->i_ref[0]; i++ )
3882         if( h->fref[0][i] && h->fref[0][i]->b_duplicate )
3883         {
3884             x264_frame_push_blank_unused( h, h->fref[0][i] );
3885             h->fref[0][i] = 0;
3886         }
3887
3888     if( h->param.psz_dump_yuv )
3889         x264_frame_dump( h );
3890     x264_emms();
3891
3892     return frame_size;
3893 }
3894
3895 static void x264_print_intra( int64_t *i_mb_count, double i_count, int b_print_pcm, char *intra )
3896 {
3897     intra += sprintf( intra, "I16..4%s: %4.1f%% %4.1f%% %4.1f%%",
3898         b_print_pcm ? "..PCM" : "",
3899         i_mb_count[I_16x16]/ i_count,
3900         i_mb_count[I_8x8]  / i_count,
3901         i_mb_count[I_4x4]  / i_count );
3902     if( b_print_pcm )
3903         sprintf( intra, " %4.1f%%", i_mb_count[I_PCM]  / i_count );
3904 }
3905
3906 /****************************************************************************
3907  * x264_encoder_close:
3908  ****************************************************************************/
3909 void    x264_encoder_close  ( x264_t *h )
3910 {
3911     int64_t i_yuv_size = FRAME_SIZE( h->param.i_width * h->param.i_height );
3912     int64_t i_mb_count_size[2][7] = {{0}};
3913     char buf[200];
3914     int b_print_pcm = h->stat.i_mb_count[SLICE_TYPE_I][I_PCM]
3915                    || h->stat.i_mb_count[SLICE_TYPE_P][I_PCM]
3916                    || h->stat.i_mb_count[SLICE_TYPE_B][I_PCM];
3917
3918     x264_lookahead_delete( h );
3919
3920 #if HAVE_OPENCL
3921     x264_opencl_lookahead_delete( h );
3922     x264_opencl_function_t *ocl = h->opencl.ocl;
3923 #endif
3924
3925     if( h->param.b_sliced_threads )
3926         x264_threadpool_wait_all( h );
3927     if( h->param.i_threads > 1 )
3928         x264_threadpool_delete( h->threadpool );
3929     if( h->param.i_lookahead_threads > 1 )
3930         x264_threadpool_delete( h->lookaheadpool );
3931     if( h->i_thread_frames > 1 )
3932     {
3933         for( int i = 0; i < h->i_thread_frames; i++ )
3934             if( h->thread[i]->b_thread_active )
3935             {
3936                 assert( h->thread[i]->fenc->i_reference_count == 1 );
3937                 x264_frame_delete( h->thread[i]->fenc );
3938             }
3939
3940         x264_t *thread_prev = h->thread[h->i_thread_phase];
3941         x264_thread_sync_ratecontrol( h, thread_prev, h );
3942         x264_thread_sync_ratecontrol( thread_prev, thread_prev, h );
3943         h->i_frame = thread_prev->i_frame + 1 - h->i_thread_frames;
3944     }
3945     h->i_frame++;
3946
3947     /* Slices used and PSNR */
3948     for( int i = 0; i < 3; i++ )
3949     {
3950         static const uint8_t slice_order[] = { SLICE_TYPE_I, SLICE_TYPE_P, SLICE_TYPE_B };
3951         int i_slice = slice_order[i];
3952
3953         if( h->stat.i_frame_count[i_slice] > 0 )
3954         {
3955             int i_count = h->stat.i_frame_count[i_slice];
3956             double dur =  h->stat.f_frame_duration[i_slice];
3957             if( h->param.analyse.b_psnr )
3958             {
3959                 x264_log( h, X264_LOG_INFO,
3960                           "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",
3961                           slice_type_to_char[i_slice],
3962                           i_count,
3963                           h->stat.f_frame_qp[i_slice] / i_count,
3964                           (double)h->stat.i_frame_size[i_slice] / i_count,
3965                           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,
3966                           h->stat.f_psnr_average[i_slice] / dur,
3967                           x264_psnr( h->stat.f_ssd_global[i_slice], dur * i_yuv_size ) );
3968             }
3969             else
3970             {
3971                 x264_log( h, X264_LOG_INFO,
3972                           "frame %c:%-5d Avg QP:%5.2f  size:%6.0f\n",
3973                           slice_type_to_char[i_slice],
3974                           i_count,
3975                           h->stat.f_frame_qp[i_slice] / i_count,
3976                           (double)h->stat.i_frame_size[i_slice] / i_count );
3977             }
3978         }
3979     }
3980     if( h->param.i_bframe && h->stat.i_frame_count[SLICE_TYPE_B] )
3981     {
3982         char *p = buf;
3983         int den = 0;
3984         // weight by number of frames (including the I/P-frames) that are in a sequence of N B-frames
3985         for( int i = 0; i <= h->param.i_bframe; i++ )
3986             den += (i+1) * h->stat.i_consecutive_bframes[i];
3987         for( int i = 0; i <= h->param.i_bframe; i++ )
3988             p += sprintf( p, " %4.1f%%", 100. * (i+1) * h->stat.i_consecutive_bframes[i] / den );
3989         x264_log( h, X264_LOG_INFO, "consecutive B-frames:%s\n", buf );
3990     }
3991
3992     for( int i_type = 0; i_type < 2; i_type++ )
3993         for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
3994         {
3995             if( i == D_DIRECT_8x8 ) continue; /* direct is counted as its own type */
3996             i_mb_count_size[i_type][x264_mb_partition_pixel_table[i]] += h->stat.i_mb_partition[i_type][i];
3997         }
3998
3999     /* MB types used */
4000     if( h->stat.i_frame_count[SLICE_TYPE_I] > 0 )
4001     {
4002         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_I];
4003         double i_count = h->stat.i_frame_count[SLICE_TYPE_I] * h->mb.i_mb_count / 100.0;
4004         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
4005         x264_log( h, X264_LOG_INFO, "mb I  %s\n", buf );
4006     }
4007     if( h->stat.i_frame_count[SLICE_TYPE_P] > 0 )
4008     {
4009         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_P];
4010         double i_count = h->stat.i_frame_count[SLICE_TYPE_P] * h->mb.i_mb_count / 100.0;
4011         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_P];
4012         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
4013         x264_log( h, X264_LOG_INFO,
4014                   "mb P  %s  P16..4: %4.1f%% %4.1f%% %4.1f%% %4.1f%% %4.1f%%    skip:%4.1f%%\n",
4015                   buf,
4016                   i_mb_size[PIXEL_16x16] / (i_count*4),
4017                   (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
4018                   i_mb_size[PIXEL_8x8] / (i_count*4),
4019                   (i_mb_size[PIXEL_8x4] + i_mb_size[PIXEL_4x8]) / (i_count*4),
4020                   i_mb_size[PIXEL_4x4] / (i_count*4),
4021                   i_mb_count[P_SKIP] / i_count );
4022     }
4023     if( h->stat.i_frame_count[SLICE_TYPE_B] > 0 )
4024     {
4025         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_B];
4026         double i_count = h->stat.i_frame_count[SLICE_TYPE_B] * h->mb.i_mb_count / 100.0;
4027         double i_mb_list_count;
4028         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_B];
4029         int64_t list_count[3] = {0}; /* 0 == L0, 1 == L1, 2 == BI */
4030         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
4031         for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
4032             for( int j = 0; j < 2; j++ )
4033             {
4034                 int l0 = x264_mb_type_list_table[i][0][j];
4035                 int l1 = x264_mb_type_list_table[i][1][j];
4036                 if( l0 || l1 )
4037                     list_count[l1+l0*l1] += h->stat.i_mb_count[SLICE_TYPE_B][i] * 2;
4038             }
4039         list_count[0] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L0_8x8];
4040         list_count[1] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L1_8x8];
4041         list_count[2] += h->stat.i_mb_partition[SLICE_TYPE_B][D_BI_8x8];
4042         i_mb_count[B_DIRECT] += (h->stat.i_mb_partition[SLICE_TYPE_B][D_DIRECT_8x8]+2)/4;
4043         i_mb_list_count = (list_count[0] + list_count[1] + list_count[2]) / 100.0;
4044         sprintf( buf + strlen(buf), "  B16..8: %4.1f%% %4.1f%% %4.1f%%  direct:%4.1f%%  skip:%4.1f%%",
4045                  i_mb_size[PIXEL_16x16] / (i_count*4),
4046                  (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
4047                  i_mb_size[PIXEL_8x8] / (i_count*4),
4048                  i_mb_count[B_DIRECT] / i_count,
4049                  i_mb_count[B_SKIP]   / i_count );
4050         if( i_mb_list_count != 0 )
4051             sprintf( buf + strlen(buf), "  L0:%4.1f%% L1:%4.1f%% BI:%4.1f%%",
4052                      list_count[0] / i_mb_list_count,
4053                      list_count[1] / i_mb_list_count,
4054                      list_count[2] / i_mb_list_count );
4055         x264_log( h, X264_LOG_INFO, "mb B  %s\n", buf );
4056     }
4057
4058     x264_ratecontrol_summary( h );
4059
4060     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 )
4061     {
4062 #define SUM3(p) (p[SLICE_TYPE_I] + p[SLICE_TYPE_P] + p[SLICE_TYPE_B])
4063 #define SUM3b(p,o) (p[SLICE_TYPE_I][o] + p[SLICE_TYPE_P][o] + p[SLICE_TYPE_B][o])
4064         int64_t i_i8x8 = SUM3b( h->stat.i_mb_count, I_8x8 );
4065         int64_t i_intra = i_i8x8 + SUM3b( h->stat.i_mb_count, I_4x4 )
4066                                  + SUM3b( h->stat.i_mb_count, I_16x16 );
4067         int64_t i_all_intra = i_intra + SUM3b( h->stat.i_mb_count, I_PCM);
4068         int64_t i_skip = SUM3b( h->stat.i_mb_count, P_SKIP )
4069                        + SUM3b( h->stat.i_mb_count, B_SKIP );
4070         const int i_count = h->stat.i_frame_count[SLICE_TYPE_I] +
4071                             h->stat.i_frame_count[SLICE_TYPE_P] +
4072                             h->stat.i_frame_count[SLICE_TYPE_B];
4073         int64_t i_mb_count = (int64_t)i_count * h->mb.i_mb_count;
4074         int64_t i_inter = i_mb_count - i_skip - i_intra;
4075         const double duration = h->stat.f_frame_duration[SLICE_TYPE_I] +
4076                                 h->stat.f_frame_duration[SLICE_TYPE_P] +
4077                                 h->stat.f_frame_duration[SLICE_TYPE_B];
4078         float f_bitrate = SUM3(h->stat.i_frame_size) / duration / 125;
4079
4080         if( PARAM_INTERLACED )
4081         {
4082             char *fieldstats = buf;
4083             fieldstats[0] = 0;
4084             if( i_inter )
4085                 fieldstats += sprintf( fieldstats, " inter:%.1f%%", h->stat.i_mb_field[1] * 100.0 / i_inter );
4086             if( i_skip )
4087                 fieldstats += sprintf( fieldstats, " skip:%.1f%%", h->stat.i_mb_field[2] * 100.0 / i_skip );
4088             x264_log( h, X264_LOG_INFO, "field mbs: intra: %.1f%%%s\n",
4089                       h->stat.i_mb_field[0] * 100.0 / i_intra, buf );
4090         }
4091
4092         if( h->pps->b_transform_8x8_mode )
4093         {
4094             buf[0] = 0;
4095             if( h->stat.i_mb_count_8x8dct[0] )
4096                 sprintf( buf, " inter:%.1f%%", 100. * h->stat.i_mb_count_8x8dct[1] / h->stat.i_mb_count_8x8dct[0] );
4097             x264_log( h, X264_LOG_INFO, "8x8 transform intra:%.1f%%%s\n", 100. * i_i8x8 / i_intra, buf );
4098         }
4099
4100         if( (h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO ||
4101             (h->stat.i_direct_frames[0] && h->stat.i_direct_frames[1]))
4102             && h->stat.i_frame_count[SLICE_TYPE_B] )
4103         {
4104             x264_log( h, X264_LOG_INFO, "direct mvs  spatial:%.1f%% temporal:%.1f%%\n",
4105                       h->stat.i_direct_frames[1] * 100. / h->stat.i_frame_count[SLICE_TYPE_B],
4106                       h->stat.i_direct_frames[0] * 100. / h->stat.i_frame_count[SLICE_TYPE_B] );
4107         }
4108
4109         buf[0] = 0;
4110         int csize = CHROMA444 ? 4 : 1;
4111         if( i_mb_count != i_all_intra )
4112             sprintf( buf, " inter: %.1f%% %.1f%% %.1f%%",
4113                      h->stat.i_mb_cbp[1] * 100.0 / ((i_mb_count - i_all_intra)*4),
4114                      h->stat.i_mb_cbp[3] * 100.0 / ((i_mb_count - i_all_intra)*csize),
4115                      h->stat.i_mb_cbp[5] * 100.0 / ((i_mb_count - i_all_intra)*csize) );
4116         x264_log( h, X264_LOG_INFO, "coded y,%s,%s intra: %.1f%% %.1f%% %.1f%%%s\n",
4117                   CHROMA444?"u":"uvDC", CHROMA444?"v":"uvAC",
4118                   h->stat.i_mb_cbp[0] * 100.0 / (i_all_intra*4),
4119                   h->stat.i_mb_cbp[2] * 100.0 / (i_all_intra*csize),
4120                   h->stat.i_mb_cbp[4] * 100.0 / (i_all_intra*csize), buf );
4121
4122         int64_t fixed_pred_modes[4][9] = {{0}};
4123         int64_t sum_pred_modes[4] = {0};
4124         for( int i = 0; i <= I_PRED_16x16_DC_128; i++ )
4125         {
4126             fixed_pred_modes[0][x264_mb_pred_mode16x16_fix[i]] += h->stat.i_mb_pred_mode[0][i];
4127             sum_pred_modes[0] += h->stat.i_mb_pred_mode[0][i];
4128         }
4129         if( sum_pred_modes[0] )
4130             x264_log( h, X264_LOG_INFO, "i16 v,h,dc,p: %2.0f%% %2.0f%% %2.0f%% %2.0f%%\n",
4131                       fixed_pred_modes[0][0] * 100.0 / sum_pred_modes[0],
4132                       fixed_pred_modes[0][1] * 100.0 / sum_pred_modes[0],
4133                       fixed_pred_modes[0][2] * 100.0 / sum_pred_modes[0],
4134                       fixed_pred_modes[0][3] * 100.0 / sum_pred_modes[0] );
4135         for( int i = 1; i <= 2; i++ )
4136         {
4137             for( int j = 0; j <= I_PRED_8x8_DC_128; j++ )