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