4:2:2 encoding support
[x262.git] / encoder / encoder.c
1 /*****************************************************************************
2  * encoder.c: top-level encoder functions
3  *****************************************************************************
4  * Copyright (C) 2003-2011 x264 project
5  *
6  * Authors: Laurent Aimar <fenrir@via.ecp.fr>
7  *          Loren Merritt <lorenm@u.washington.edu>
8  *          Jason Garrett-Glaser <darkshikari@gmail.com>
9  *
10  * This program is free software; you can redistribute it and/or modify
11  * it under the terms of the GNU General Public License as published by
12  * the Free Software Foundation; either version 2 of the License, or
13  * (at your option) any later version.
14  *
15  * This program is distributed in the hope that it will be useful,
16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
18  * GNU General Public License for more details.
19  *
20  * You should have received a copy of the GNU General Public License
21  * along with this program; if not, write to the Free Software
22  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02111, USA.
23  *
24  * This program is also available under a commercial proprietary license.
25  * For more information, contact us at licensing@x264.com.
26  *****************************************************************************/
27
28 #include "common/common.h"
29
30 #include "set.h"
31 #include "analyse.h"
32 #include "ratecontrol.h"
33 #include "macroblock.h"
34 #include "me.h"
35
36 #if HAVE_VISUALIZE
37 #include "common/visualize.h"
38 #endif
39
40 //#define DEBUG_MB_TYPE
41
42 #define bs_write_ue bs_write_ue_big
43
44 static int x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
45                                    x264_nal_t **pp_nal, int *pi_nal,
46                                    x264_picture_t *pic_out );
47
48 /****************************************************************************
49  *
50  ******************************* x264 libs **********************************
51  *
52  ****************************************************************************/
53 static double x264_psnr( double sqe, double size )
54 {
55     double mse = sqe / (PIXEL_MAX*PIXEL_MAX * size);
56     if( mse <= 0.0000000001 ) /* Max 100dB */
57         return 100;
58
59     return -10.0 * log10( mse );
60 }
61
62 static double x264_ssim( double ssim )
63 {
64     return -10.0 * log10( 1 - ssim );
65 }
66
67 static void x264_frame_dump( x264_t *h )
68 {
69     FILE *f = fopen( h->param.psz_dump_yuv, "r+b" );
70     if( !f )
71         return;
72
73     /* Write the frame in display order */
74     int frame_size = FRAME_SIZE( h->param.i_height * h->param.i_width * sizeof(pixel) );
75     fseek( f, (uint64_t)h->fdec->i_frame * frame_size, SEEK_SET );
76     for( int p = 0; p < (CHROMA444 ? 3 : 1); p++ )
77         for( int y = 0; y < h->param.i_height; y++ )
78             fwrite( &h->fdec->plane[p][y*h->fdec->i_stride[p]], sizeof(pixel), h->param.i_width, f );
79     if( !CHROMA444 )
80     {
81         int cw = h->param.i_width>>1;
82         int ch = h->param.i_height>>h->mb.chroma_v_shift;
83         pixel *planeu = x264_malloc( (cw*ch*2+32)*sizeof(pixel) );
84         pixel *planev = planeu + cw*ch + 16;
85         h->mc.plane_copy_deinterleave( planeu, cw, planev, cw, h->fdec->plane[1], h->fdec->i_stride[1], cw, ch );
86         fwrite( planeu, 1, cw*ch*sizeof(pixel), f );
87         fwrite( planev, 1, cw*ch*sizeof(pixel), f );
88         x264_free( planeu );
89     }
90     fclose( f );
91 }
92
93 /* Fill "default" values */
94 static void x264_slice_header_init( x264_t *h, x264_slice_header_t *sh,
95                                     x264_sps_t *sps, x264_pps_t *pps,
96                                     int i_idr_pic_id, int i_frame, int i_qp )
97 {
98     x264_param_t *param = &h->param;
99
100     /* First we fill all fields */
101     sh->sps = sps;
102     sh->pps = pps;
103
104     sh->i_first_mb  = 0;
105     sh->i_last_mb   = h->mb.i_mb_count - 1;
106     sh->i_pps_id    = pps->i_id;
107
108     sh->i_frame_num = i_frame;
109
110     sh->b_mbaff = PARAM_INTERLACED;
111     sh->b_field_pic = 0;    /* no field support for now */
112     sh->b_bottom_field = 0; /* not yet used */
113
114     sh->i_idr_pic_id = i_idr_pic_id;
115
116     /* poc stuff, fixed later */
117     sh->i_poc = 0;
118     sh->i_delta_poc_bottom = 0;
119     sh->i_delta_poc[0] = 0;
120     sh->i_delta_poc[1] = 0;
121
122     sh->i_redundant_pic_cnt = 0;
123
124     h->mb.b_direct_auto_write = h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO
125                                 && h->param.i_bframe
126                                 && ( h->param.rc.b_stat_write || !h->param.rc.b_stat_read );
127
128     if( !h->mb.b_direct_auto_read && sh->i_type == SLICE_TYPE_B )
129     {
130         if( h->fref[1][0]->i_poc_l0ref0 == h->fref[0][0]->i_poc )
131         {
132             if( h->mb.b_direct_auto_write )
133                 sh->b_direct_spatial_mv_pred = ( h->stat.i_direct_score[1] > h->stat.i_direct_score[0] );
134             else
135                 sh->b_direct_spatial_mv_pred = ( param->analyse.i_direct_mv_pred == X264_DIRECT_PRED_SPATIAL );
136         }
137         else
138         {
139             h->mb.b_direct_auto_write = 0;
140             sh->b_direct_spatial_mv_pred = 1;
141         }
142     }
143     /* else b_direct_spatial_mv_pred was read from the 2pass statsfile */
144
145     sh->b_num_ref_idx_override = 0;
146     sh->i_num_ref_idx_l0_active = 1;
147     sh->i_num_ref_idx_l1_active = 1;
148
149     sh->b_ref_pic_list_reordering[0] = h->b_ref_reorder[0];
150     sh->b_ref_pic_list_reordering[1] = h->b_ref_reorder[1];
151
152     /* If the ref list isn't in the default order, construct reordering header */
153     for( int list = 0; list < 2; list++ )
154     {
155         if( sh->b_ref_pic_list_reordering[list] )
156         {
157             int pred_frame_num = i_frame;
158             for( int i = 0; i < h->i_ref[list]; i++ )
159             {
160                 int diff = h->fref[list][i]->i_frame_num - pred_frame_num;
161                 sh->ref_pic_list_order[list][i].idc = ( diff > 0 );
162                 sh->ref_pic_list_order[list][i].arg = (abs(diff) - 1) & ((1 << sps->i_log2_max_frame_num) - 1);
163                 pred_frame_num = h->fref[list][i]->i_frame_num;
164             }
165         }
166     }
167
168     sh->i_cabac_init_idc = param->i_cabac_init_idc;
169
170     sh->i_qp = SPEC_QP(i_qp);
171     sh->i_qp_delta = sh->i_qp - pps->i_pic_init_qp;
172     sh->b_sp_for_swidth = 0;
173     sh->i_qs_delta = 0;
174
175     int deblock_thresh = i_qp + 2 * X264_MIN(param->i_deblocking_filter_alphac0, param->i_deblocking_filter_beta);
176     /* If effective qp <= 15, deblocking would have no effect anyway */
177     if( param->b_deblocking_filter && (h->mb.b_variable_qp || 15 < deblock_thresh ) )
178         sh->i_disable_deblocking_filter_idc = param->b_sliced_threads ? 2 : 0;
179     else
180         sh->i_disable_deblocking_filter_idc = 1;
181     sh->i_alpha_c0_offset = param->i_deblocking_filter_alphac0 << 1;
182     sh->i_beta_offset = param->i_deblocking_filter_beta << 1;
183 }
184
185 static void x264_slice_header_write( bs_t *s, x264_slice_header_t *sh, int i_nal_ref_idc )
186 {
187     if( sh->b_mbaff )
188     {
189         int first_x = sh->i_first_mb % sh->sps->i_mb_width;
190         int first_y = sh->i_first_mb / sh->sps->i_mb_width;
191         assert( (first_y&1) == 0 );
192         bs_write_ue( s, (2*first_x + sh->sps->i_mb_width*(first_y&~1) + (first_y&1)) >> 1 );
193     }
194     else
195         bs_write_ue( s, sh->i_first_mb );
196
197     bs_write_ue( s, sh->i_type + 5 );   /* same type things */
198     bs_write_ue( s, sh->i_pps_id );
199     bs_write( s, sh->sps->i_log2_max_frame_num, sh->i_frame_num & ((1<<sh->sps->i_log2_max_frame_num)-1) );
200
201     if( !sh->sps->b_frame_mbs_only )
202     {
203         bs_write1( s, sh->b_field_pic );
204         if( sh->b_field_pic )
205             bs_write1( s, sh->b_bottom_field );
206     }
207
208     if( sh->i_idr_pic_id >= 0 ) /* NAL IDR */
209         bs_write_ue( s, sh->i_idr_pic_id );
210
211     if( sh->sps->i_poc_type == 0 )
212     {
213         bs_write( s, sh->sps->i_log2_max_poc_lsb, sh->i_poc & ((1<<sh->sps->i_log2_max_poc_lsb)-1) );
214         if( sh->pps->b_pic_order && !sh->b_field_pic )
215             bs_write_se( s, sh->i_delta_poc_bottom );
216     }
217
218     if( sh->pps->b_redundant_pic_cnt )
219         bs_write_ue( s, sh->i_redundant_pic_cnt );
220
221     if( sh->i_type == SLICE_TYPE_B )
222         bs_write1( s, sh->b_direct_spatial_mv_pred );
223
224     if( sh->i_type == SLICE_TYPE_P || sh->i_type == SLICE_TYPE_B )
225     {
226         bs_write1( s, sh->b_num_ref_idx_override );
227         if( sh->b_num_ref_idx_override )
228         {
229             bs_write_ue( s, sh->i_num_ref_idx_l0_active - 1 );
230             if( sh->i_type == SLICE_TYPE_B )
231                 bs_write_ue( s, sh->i_num_ref_idx_l1_active - 1 );
232         }
233     }
234
235     /* ref pic list reordering */
236     if( sh->i_type != SLICE_TYPE_I )
237     {
238         bs_write1( s, sh->b_ref_pic_list_reordering[0] );
239         if( sh->b_ref_pic_list_reordering[0] )
240         {
241             for( int i = 0; i < sh->i_num_ref_idx_l0_active; i++ )
242             {
243                 bs_write_ue( s, sh->ref_pic_list_order[0][i].idc );
244                 bs_write_ue( s, sh->ref_pic_list_order[0][i].arg );
245             }
246             bs_write_ue( s, 3 );
247         }
248     }
249     if( sh->i_type == SLICE_TYPE_B )
250     {
251         bs_write1( s, sh->b_ref_pic_list_reordering[1] );
252         if( sh->b_ref_pic_list_reordering[1] )
253         {
254             for( int i = 0; i < sh->i_num_ref_idx_l1_active; i++ )
255             {
256                 bs_write_ue( s, sh->ref_pic_list_order[1][i].idc );
257                 bs_write_ue( s, sh->ref_pic_list_order[1][i].arg );
258             }
259             bs_write_ue( s, 3 );
260         }
261     }
262
263     if( sh->pps->b_weighted_pred && sh->i_type == SLICE_TYPE_P )
264     {
265         /* pred_weight_table() */
266         bs_write_ue( s, sh->weight[0][0].i_denom );
267         bs_write_ue( s, sh->weight[0][1].i_denom );
268         for( int i = 0; i < sh->i_num_ref_idx_l0_active; i++ )
269         {
270             int luma_weight_l0_flag = !!sh->weight[i][0].weightfn;
271             int chroma_weight_l0_flag = !!sh->weight[i][1].weightfn || !!sh->weight[i][2].weightfn;
272             bs_write1( s, luma_weight_l0_flag );
273             if( luma_weight_l0_flag )
274             {
275                 bs_write_se( s, sh->weight[i][0].i_scale );
276                 bs_write_se( s, sh->weight[i][0].i_offset );
277             }
278             bs_write1( s, chroma_weight_l0_flag );
279             if( chroma_weight_l0_flag )
280             {
281                 for( int j = 1; j < 3; j++ )
282                 {
283                     bs_write_se( s, sh->weight[i][j].i_scale );
284                     bs_write_se( s, sh->weight[i][j].i_offset );
285                 }
286             }
287         }
288     }
289     else if( sh->pps->b_weighted_bipred == 1 && sh->i_type == SLICE_TYPE_B )
290     {
291       /* TODO */
292     }
293
294     if( i_nal_ref_idc != 0 )
295     {
296         if( sh->i_idr_pic_id >= 0 )
297         {
298             bs_write1( s, 0 );  /* no output of prior pics flag */
299             bs_write1( s, 0 );  /* long term reference flag */
300         }
301         else
302         {
303             bs_write1( s, sh->i_mmco_command_count > 0 ); /* adaptive_ref_pic_marking_mode_flag */
304             if( sh->i_mmco_command_count > 0 )
305             {
306                 for( int i = 0; i < sh->i_mmco_command_count; i++ )
307                 {
308                     bs_write_ue( s, 1 ); /* mark short term ref as unused */
309                     bs_write_ue( s, sh->mmco[i].i_difference_of_pic_nums - 1 );
310                 }
311                 bs_write_ue( s, 0 ); /* end command list */
312             }
313         }
314     }
315
316     if( sh->pps->b_cabac && sh->i_type != SLICE_TYPE_I )
317         bs_write_ue( s, sh->i_cabac_init_idc );
318
319     bs_write_se( s, sh->i_qp_delta );      /* slice qp delta */
320
321     if( sh->pps->b_deblocking_filter_control )
322     {
323         bs_write_ue( s, sh->i_disable_deblocking_filter_idc );
324         if( sh->i_disable_deblocking_filter_idc != 1 )
325         {
326             bs_write_se( s, sh->i_alpha_c0_offset >> 1 );
327             bs_write_se( s, sh->i_beta_offset >> 1 );
328         }
329     }
330 }
331
332 /* If we are within a reasonable distance of the end of the memory allocated for the bitstream, */
333 /* reallocate, adding an arbitrary amount of space (100 kilobytes). */
334 static int x264_bitstream_check_buffer( x264_t *h )
335 {
336     uint8_t *bs_bak = h->out.p_bitstream;
337     int max_mb_size = 2500 << SLICE_MBAFF;
338     if( (h->param.b_cabac && (h->cabac.p_end - h->cabac.p < max_mb_size)) ||
339         (h->out.bs.p_end - h->out.bs.p < max_mb_size) )
340     {
341         h->out.i_bitstream += 100000;
342         CHECKED_MALLOC( h->out.p_bitstream, h->out.i_bitstream );
343         h->mc.memcpy_aligned( h->out.p_bitstream, bs_bak, (h->out.i_bitstream - 100000) & ~15 );
344         intptr_t delta = h->out.p_bitstream - bs_bak;
345
346         h->out.bs.p_start += delta;
347         h->out.bs.p += delta;
348         h->out.bs.p_end = h->out.p_bitstream + h->out.i_bitstream;
349
350         h->cabac.p_start += delta;
351         h->cabac.p += delta;
352         h->cabac.p_end = h->out.p_bitstream + h->out.i_bitstream;
353
354         for( int i = 0; i <= h->out.i_nal; i++ )
355             h->out.nal[i].p_payload += delta;
356         x264_free( bs_bak );
357     }
358     return 0;
359 fail:
360     x264_free( bs_bak );
361     return -1;
362 }
363
364 #if HAVE_THREAD
365 static void x264_encoder_thread_init( x264_t *h )
366 {
367     if( h->param.i_sync_lookahead )
368         x264_lower_thread_priority( 10 );
369
370 #if HAVE_MMX
371     /* Misalign mask has to be set separately for each thread. */
372     if( h->param.cpu&X264_CPU_SSE_MISALIGN )
373         x264_cpu_mask_misalign_sse();
374 #endif
375 }
376 #endif
377
378 /****************************************************************************
379  *
380  ****************************************************************************
381  ****************************** External API*********************************
382  ****************************************************************************
383  *
384  ****************************************************************************/
385
386 static int x264_validate_parameters( x264_t *h, int b_open )
387 {
388 #if HAVE_MMX
389 #ifdef __SSE__
390     if( b_open && !(x264_cpu_detect() & X264_CPU_SSE) )
391     {
392         x264_log( h, X264_LOG_ERROR, "your cpu does not support SSE1, but x264 was compiled with asm support\n");
393 #else
394     if( b_open && !(x264_cpu_detect() & X264_CPU_MMX2) )
395     {
396         x264_log( h, X264_LOG_ERROR, "your cpu does not support MMXEXT, but x264 was compiled with asm support\n");
397 #endif
398         x264_log( h, X264_LOG_ERROR, "to run x264, recompile without asm support (configure --disable-asm)\n");
399         return -1;
400     }
401 #endif
402
403 #if HAVE_INTERLACED
404     h->param.b_interlaced = !!PARAM_INTERLACED;
405 #else
406     if( h->param.b_interlaced )
407     {
408         x264_log( h, X264_LOG_ERROR, "not compiled with interlaced support\n" );
409         return -1;
410     }
411 #endif
412
413     if( h->param.i_width <= 0 || h->param.i_height <= 0 )
414     {
415         x264_log( h, X264_LOG_ERROR, "invalid width x height (%dx%d)\n",
416                   h->param.i_width, h->param.i_height );
417         return -1;
418     }
419
420     int i_csp = h->param.i_csp & X264_CSP_MASK;
421     if( i_csp <= X264_CSP_NONE || i_csp >= X264_CSP_MAX )
422     {
423         x264_log( h, X264_LOG_ERROR, "invalid CSP (only I420/YV12/NV12/I422/YV16/NV16/I444/YV24/BGR/BGRA/RGB supported)\n" );
424         return -1;
425     }
426
427     if( i_csp < X264_CSP_I444 && h->param.i_width % 2 )
428     {
429         x264_log( h, X264_LOG_ERROR, "width not divisible by 2 (%dx%d)\n",
430                   h->param.i_width, h->param.i_height );
431         return -1;
432     }
433
434     if( i_csp < X264_CSP_I422 && PARAM_INTERLACED && h->param.i_height % 4 )
435     {
436         x264_log( h, X264_LOG_ERROR, "height not divisible by 4 (%dx%d)\n",
437                   h->param.i_width, h->param.i_height );
438         return -1;
439     }
440
441     if( (i_csp < X264_CSP_I422 || PARAM_INTERLACED) && h->param.i_height % 2 )
442     {
443         x264_log( h, X264_LOG_ERROR, "height not divisible by 2 (%dx%d)\n",
444                   h->param.i_width, h->param.i_height );
445         return -1;
446     }
447
448     if( (h->param.crop_rect.i_left + h->param.crop_rect.i_right ) >= h->param.i_width ||
449         (h->param.crop_rect.i_top  + h->param.crop_rect.i_bottom) >= h->param.i_height )
450     {
451         x264_log( h, X264_LOG_ERROR, "invalid crop-rect %u,%u,%u,%u\n", h->param.crop_rect.i_left,
452                   h->param.crop_rect.i_top, h->param.crop_rect.i_right,  h->param.crop_rect.i_bottom );
453         return -1;
454     }
455
456     if( h->param.i_threads == X264_THREADS_AUTO )
457         h->param.i_threads = x264_cpu_num_processors() * (h->param.b_sliced_threads?2:3)/2;
458     h->param.i_threads = x264_clip3( h->param.i_threads, 1, X264_THREAD_MAX );
459     if( h->param.i_threads > 1 )
460     {
461 #if !HAVE_THREAD
462         x264_log( h, X264_LOG_WARNING, "not compiled with thread support!\n");
463         h->param.i_threads = 1;
464 #endif
465         /* Avoid absurdly small thread slices as they can reduce performance
466          * and VBV compliance.  Capped at an arbitrary 4 rows per thread. */
467         if( h->param.b_sliced_threads )
468         {
469             int max_threads = (h->param.i_height+15)/16 / 4;
470             h->param.i_threads = X264_MIN( h->param.i_threads, max_threads );
471         }
472     }
473     else
474         h->param.b_sliced_threads = 0;
475     h->i_thread_frames = h->param.b_sliced_threads ? 1 : h->param.i_threads;
476     if( h->i_thread_frames > 1 )
477         h->param.nalu_process = NULL;
478
479     h->param.i_keyint_max = x264_clip3( h->param.i_keyint_max, 1, X264_KEYINT_MAX_INFINITE );
480     if( h->param.i_keyint_max == 1 )
481     {
482         h->param.b_intra_refresh = 0;
483         h->param.analyse.i_weighted_pred = 0;
484     }
485
486     h->param.i_frame_packing = x264_clip3( h->param.i_frame_packing, -1, 5 );
487
488     /* Detect default ffmpeg settings and terminate with an error. */
489     if( b_open )
490     {
491         int score = 0;
492         score += h->param.analyse.i_me_range == 0;
493         score += h->param.rc.i_qp_step == 3;
494         score += h->param.i_keyint_max == 12;
495         score += h->param.rc.i_qp_min == 2;
496         score += h->param.rc.i_qp_max == 31;
497         score += h->param.rc.f_qcompress == 0.5;
498         score += fabs(h->param.rc.f_ip_factor - 1.25) < 0.01;
499         score += fabs(h->param.rc.f_pb_factor - 1.25) < 0.01;
500         score += h->param.analyse.inter == 0 && h->param.analyse.i_subpel_refine == 8;
501         if( score >= 5 )
502         {
503             x264_log( h, X264_LOG_ERROR, "broken ffmpeg default settings detected\n" );
504             x264_log( h, X264_LOG_ERROR, "use an encoding preset (e.g. -vpre medium)\n" );
505             x264_log( h, X264_LOG_ERROR, "preset usage: -vpre <speed> -vpre <profile>\n" );
506             x264_log( h, X264_LOG_ERROR, "speed presets are listed in x264 --help\n" );
507             x264_log( h, X264_LOG_ERROR, "profile is optional; x264 defaults to high\n" );
508             return -1;
509         }
510     }
511
512     if( h->param.rc.i_rc_method < 0 || h->param.rc.i_rc_method > 2 )
513     {
514         x264_log( h, X264_LOG_ERROR, "no ratecontrol method specified\n" );
515         return -1;
516     }
517     h->param.rc.f_rf_constant = x264_clip3f( h->param.rc.f_rf_constant, -QP_BD_OFFSET, 51 );
518     h->param.rc.f_rf_constant_max = x264_clip3f( h->param.rc.f_rf_constant_max, -QP_BD_OFFSET, 51 );
519     h->param.rc.i_qp_constant = x264_clip3( h->param.rc.i_qp_constant, 0, QP_MAX );
520     h->param.analyse.i_subpel_refine = x264_clip3( h->param.analyse.i_subpel_refine, 0, 11 );
521     h->param.rc.f_ip_factor = X264_MAX( h->param.rc.f_ip_factor, 0.01f );
522     h->param.rc.f_pb_factor = X264_MAX( h->param.rc.f_pb_factor, 0.01f );
523     if( h->param.rc.i_rc_method == X264_RC_CRF )
524     {
525         h->param.rc.i_qp_constant = h->param.rc.f_rf_constant + QP_BD_OFFSET;
526         h->param.rc.i_bitrate = 0;
527     }
528     if( (h->param.rc.i_rc_method == X264_RC_CQP || h->param.rc.i_rc_method == X264_RC_CRF)
529         && h->param.rc.i_qp_constant == 0 )
530     {
531         h->mb.b_lossless = 1;
532         h->param.i_cqm_preset = X264_CQM_FLAT;
533         h->param.psz_cqm_file = NULL;
534         h->param.rc.i_rc_method = X264_RC_CQP;
535         h->param.rc.f_ip_factor = 1;
536         h->param.rc.f_pb_factor = 1;
537         h->param.analyse.b_psnr = 0;
538         h->param.analyse.b_ssim = 0;
539         h->param.analyse.i_chroma_qp_offset = 0;
540         h->param.analyse.i_trellis = 0;
541         h->param.analyse.b_fast_pskip = 0;
542         h->param.analyse.i_noise_reduction = 0;
543         h->param.analyse.b_psy = 0;
544         h->param.i_bframe = 0;
545         /* 8x8dct is not useful without RD in CAVLC lossless */
546         if( !h->param.b_cabac && h->param.analyse.i_subpel_refine < 6 )
547             h->param.analyse.b_transform_8x8 = 0;
548     }
549     if( h->param.rc.i_rc_method == X264_RC_CQP )
550     {
551         float qp_p = h->param.rc.i_qp_constant;
552         float qp_i = qp_p - 6*log2f( h->param.rc.f_ip_factor );
553         float qp_b = qp_p + 6*log2f( h->param.rc.f_pb_factor );
554         h->param.rc.i_qp_min = x264_clip3( (int)(X264_MIN3( qp_p, qp_i, qp_b )), 0, QP_MAX );
555         h->param.rc.i_qp_max = x264_clip3( (int)(X264_MAX3( qp_p, qp_i, qp_b ) + .999), 0, QP_MAX );
556         h->param.rc.i_aq_mode = 0;
557         h->param.rc.b_mb_tree = 0;
558         h->param.rc.i_bitrate = 0;
559     }
560     h->param.rc.i_qp_max = x264_clip3( h->param.rc.i_qp_max, 0, QP_MAX );
561     h->param.rc.i_qp_min = x264_clip3( h->param.rc.i_qp_min, 0, h->param.rc.i_qp_max );
562     h->param.rc.i_qp_step = x264_clip3( h->param.rc.i_qp_step, 0, QP_MAX );
563     h->param.rc.i_bitrate = x264_clip3( h->param.rc.i_bitrate, 0, 2000000 );
564     h->param.rc.i_vbv_buffer_size = x264_clip3( h->param.rc.i_vbv_buffer_size, 0, 2000000 );
565     h->param.rc.i_vbv_max_bitrate = x264_clip3( h->param.rc.i_vbv_max_bitrate, 0, 2000000 );
566     h->param.rc.f_vbv_buffer_init = x264_clip3f( h->param.rc.f_vbv_buffer_init, 0, 2000000 );
567     if( h->param.rc.i_vbv_buffer_size )
568     {
569         if( h->param.rc.i_rc_method == X264_RC_CQP )
570         {
571             x264_log( h, X264_LOG_WARNING, "VBV is incompatible with constant QP, ignored.\n" );
572             h->param.rc.i_vbv_max_bitrate = 0;
573             h->param.rc.i_vbv_buffer_size = 0;
574         }
575         else if( h->param.rc.i_vbv_max_bitrate == 0 )
576         {
577             if( h->param.rc.i_rc_method == X264_RC_ABR )
578             {
579                 x264_log( h, X264_LOG_WARNING, "VBV maxrate unspecified, assuming CBR\n" );
580                 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate;
581             }
582             else
583             {
584                 x264_log( h, X264_LOG_WARNING, "VBV bufsize set but maxrate unspecified, ignored\n" );
585                 h->param.rc.i_vbv_buffer_size = 0;
586             }
587         }
588         else if( h->param.rc.i_vbv_max_bitrate < h->param.rc.i_bitrate &&
589                  h->param.rc.i_rc_method == X264_RC_ABR )
590         {
591             x264_log( h, X264_LOG_WARNING, "max bitrate less than average bitrate, assuming CBR\n" );
592             h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate;
593         }
594     }
595     else if( h->param.rc.i_vbv_max_bitrate )
596     {
597         x264_log( h, X264_LOG_WARNING, "VBV maxrate specified, but no bufsize, ignored\n" );
598         h->param.rc.i_vbv_max_bitrate = 0;
599     }
600
601     h->param.i_slice_max_size = X264_MAX( h->param.i_slice_max_size, 0 );
602     h->param.i_slice_max_mbs = X264_MAX( h->param.i_slice_max_mbs, 0 );
603
604     int max_slices = (h->param.i_height+((16<<PARAM_INTERLACED)-1))/(16<<PARAM_INTERLACED);
605     if( h->param.b_sliced_threads )
606         h->param.i_slice_count = x264_clip3( h->param.i_threads, 0, max_slices );
607     else
608     {
609         h->param.i_slice_count = x264_clip3( h->param.i_slice_count, 0, max_slices );
610         if( h->param.i_slice_max_mbs || h->param.i_slice_max_size )
611             h->param.i_slice_count = 0;
612     }
613
614     if( h->param.b_bluray_compat )
615     {
616         h->param.i_bframe_pyramid = X264_MIN( X264_B_PYRAMID_STRICT, h->param.i_bframe_pyramid );
617         h->param.i_bframe = X264_MIN( h->param.i_bframe, 3 );
618         h->param.b_aud = 1;
619         h->param.i_nal_hrd = X264_MAX( h->param.i_nal_hrd, X264_NAL_HRD_VBR );
620         h->param.i_slice_max_size = 0;
621         h->param.i_slice_max_mbs = 0;
622         h->param.b_intra_refresh = 0;
623         h->param.i_frame_reference = X264_MIN( h->param.i_frame_reference, 6 );
624         h->param.i_dpb_size = X264_MIN( h->param.i_dpb_size, 6 );
625         /* Due to the proliferation of broken players that don't handle dupes properly. */
626         h->param.analyse.i_weighted_pred = X264_MIN( h->param.analyse.i_weighted_pred, X264_WEIGHTP_SIMPLE );
627         if( h->param.b_fake_interlaced )
628             h->param.b_pic_struct = 1;
629     }
630
631     h->param.i_frame_reference = x264_clip3( h->param.i_frame_reference, 1, X264_REF_MAX );
632     h->param.i_dpb_size = x264_clip3( h->param.i_dpb_size, 1, X264_REF_MAX );
633     if( h->param.i_scenecut_threshold < 0 )
634         h->param.i_scenecut_threshold = 0;
635     h->param.analyse.i_direct_mv_pred = x264_clip3( h->param.analyse.i_direct_mv_pred, X264_DIRECT_PRED_NONE, X264_DIRECT_PRED_AUTO );
636     if( !h->param.analyse.i_subpel_refine && h->param.analyse.i_direct_mv_pred > X264_DIRECT_PRED_SPATIAL )
637     {
638         x264_log( h, X264_LOG_WARNING, "subme=0 + direct=temporal is not supported\n" );
639         h->param.analyse.i_direct_mv_pred = X264_DIRECT_PRED_SPATIAL;
640     }
641     h->param.i_bframe = x264_clip3( h->param.i_bframe, 0, X264_MIN( X264_BFRAME_MAX, h->param.i_keyint_max-1 ) );
642     h->param.i_bframe_bias = x264_clip3( h->param.i_bframe_bias, -90, 100 );
643     if( h->param.i_bframe <= 1 )
644         h->param.i_bframe_pyramid = X264_B_PYRAMID_NONE;
645     h->param.i_bframe_pyramid = x264_clip3( h->param.i_bframe_pyramid, X264_B_PYRAMID_NONE, X264_B_PYRAMID_NORMAL );
646     h->param.i_bframe_adaptive = x264_clip3( h->param.i_bframe_adaptive, X264_B_ADAPT_NONE, X264_B_ADAPT_TRELLIS );
647     if( !h->param.i_bframe )
648     {
649         h->param.i_bframe_adaptive = X264_B_ADAPT_NONE;
650         h->param.analyse.i_direct_mv_pred = 0;
651         h->param.analyse.b_weighted_bipred = 0;
652         h->param.b_open_gop = 0;
653     }
654     if( h->param.b_intra_refresh && h->param.i_bframe_pyramid == X264_B_PYRAMID_NORMAL )
655     {
656         x264_log( h, X264_LOG_WARNING, "b-pyramid normal + intra-refresh is not supported\n" );
657         h->param.i_bframe_pyramid = X264_B_PYRAMID_STRICT;
658     }
659     if( h->param.b_intra_refresh && (h->param.i_frame_reference > 1 || h->param.i_dpb_size > 1) )
660     {
661         x264_log( h, X264_LOG_WARNING, "ref > 1 + intra-refresh is not supported\n" );
662         h->param.i_frame_reference = 1;
663         h->param.i_dpb_size = 1;
664     }
665     if( h->param.b_intra_refresh && h->param.b_open_gop )
666     {
667         x264_log( h, X264_LOG_WARNING, "intra-refresh is not compatible with open-gop\n" );
668         h->param.b_open_gop = 0;
669     }
670     float fps = h->param.i_fps_num > 0 && h->param.i_fps_den > 0 ? (float) h->param.i_fps_num / h->param.i_fps_den : 25.0;
671     if( h->param.i_keyint_min == X264_KEYINT_MIN_AUTO )
672         h->param.i_keyint_min = X264_MIN( h->param.i_keyint_max / 10, fps );
673     h->param.i_keyint_min = x264_clip3( h->param.i_keyint_min, 1, h->param.i_keyint_max/2+1 );
674     h->param.rc.i_lookahead = x264_clip3( h->param.rc.i_lookahead, 0, X264_LOOKAHEAD_MAX );
675     {
676         int maxrate = X264_MAX( h->param.rc.i_vbv_max_bitrate, h->param.rc.i_bitrate );
677         float bufsize = maxrate ? (float)h->param.rc.i_vbv_buffer_size / maxrate : 0;
678         h->param.rc.i_lookahead = X264_MIN( h->param.rc.i_lookahead, X264_MAX( h->param.i_keyint_max, bufsize*fps ) );
679     }
680
681     if( !h->param.i_timebase_num || !h->param.i_timebase_den || !(h->param.b_vfr_input || h->param.b_pulldown) )
682     {
683         h->param.i_timebase_num = h->param.i_fps_den;
684         h->param.i_timebase_den = h->param.i_fps_num;
685     }
686
687     h->param.rc.f_qcompress = x264_clip3f( h->param.rc.f_qcompress, 0.0, 1.0 );
688     if( h->param.i_keyint_max == 1 || h->param.rc.f_qcompress == 1 )
689         h->param.rc.b_mb_tree = 0;
690     if( (!h->param.b_intra_refresh && h->param.i_keyint_max != X264_KEYINT_MAX_INFINITE) &&
691         !h->param.rc.i_lookahead && h->param.rc.b_mb_tree )
692     {
693         x264_log( h, X264_LOG_WARNING, "lookaheadless mb-tree requires intra refresh or infinite keyint\n" );
694         h->param.rc.b_mb_tree = 0;
695     }
696     if( h->param.rc.b_stat_read )
697         h->param.rc.i_lookahead = 0;
698 #if HAVE_THREAD
699     if( h->param.i_sync_lookahead < 0 )
700         h->param.i_sync_lookahead = h->param.i_bframe + 1;
701     h->param.i_sync_lookahead = X264_MIN( h->param.i_sync_lookahead, X264_LOOKAHEAD_MAX );
702     if( h->param.rc.b_stat_read || h->i_thread_frames == 1 )
703         h->param.i_sync_lookahead = 0;
704 #else
705     h->param.i_sync_lookahead = 0;
706 #endif
707
708     h->param.i_deblocking_filter_alphac0 = x264_clip3( h->param.i_deblocking_filter_alphac0, -6, 6 );
709     h->param.i_deblocking_filter_beta    = x264_clip3( h->param.i_deblocking_filter_beta, -6, 6 );
710     h->param.analyse.i_luma_deadzone[0] = x264_clip3( h->param.analyse.i_luma_deadzone[0], 0, 32 );
711     h->param.analyse.i_luma_deadzone[1] = x264_clip3( h->param.analyse.i_luma_deadzone[1], 0, 32 );
712
713     h->param.i_cabac_init_idc = x264_clip3( h->param.i_cabac_init_idc, 0, 2 );
714
715     if( h->param.i_cqm_preset < X264_CQM_FLAT || h->param.i_cqm_preset > X264_CQM_CUSTOM )
716         h->param.i_cqm_preset = X264_CQM_FLAT;
717
718     if( h->param.analyse.i_me_method < X264_ME_DIA ||
719         h->param.analyse.i_me_method > X264_ME_TESA )
720         h->param.analyse.i_me_method = X264_ME_HEX;
721     h->param.analyse.i_me_range = x264_clip3( h->param.analyse.i_me_range, 4, 1024 );
722     if( h->param.analyse.i_me_range > 16 && h->param.analyse.i_me_method <= X264_ME_HEX )
723         h->param.analyse.i_me_range = 16;
724     if( h->param.analyse.i_me_method == X264_ME_TESA &&
725         (h->mb.b_lossless || h->param.analyse.i_subpel_refine <= 1) )
726         h->param.analyse.i_me_method = X264_ME_ESA;
727     h->param.analyse.b_mixed_references = h->param.analyse.b_mixed_references && h->param.i_frame_reference > 1;
728     h->param.analyse.inter &= X264_ANALYSE_PSUB16x16|X264_ANALYSE_PSUB8x8|X264_ANALYSE_BSUB16x16|
729                               X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
730     h->param.analyse.intra &= X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
731     if( !(h->param.analyse.inter & X264_ANALYSE_PSUB16x16) )
732         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
733     if( !h->param.analyse.b_transform_8x8 )
734     {
735         h->param.analyse.inter &= ~X264_ANALYSE_I8x8;
736         h->param.analyse.intra &= ~X264_ANALYSE_I8x8;
737     }
738     h->param.analyse.i_trellis = x264_clip3( h->param.analyse.i_trellis, 0, 2 );
739     h->param.rc.i_aq_mode = x264_clip3( h->param.rc.i_aq_mode, 0, 2 );
740     h->param.rc.f_aq_strength = x264_clip3f( h->param.rc.f_aq_strength, 0, 3 );
741     if( h->param.rc.f_aq_strength == 0 )
742         h->param.rc.i_aq_mode = 0;
743
744     if( h->param.i_log_level < X264_LOG_INFO )
745     {
746         h->param.analyse.b_psnr = 0;
747         h->param.analyse.b_ssim = 0;
748     }
749     /* Warn users trying to measure PSNR/SSIM with psy opts on. */
750     if( b_open && (h->param.analyse.b_psnr || h->param.analyse.b_ssim) )
751     {
752         char *s = NULL;
753
754         if( h->param.analyse.b_psy )
755         {
756             s = h->param.analyse.b_psnr ? "psnr" : "ssim";
757             x264_log( h, X264_LOG_WARNING, "--%s used with psy on: results will be invalid!\n", s );
758         }
759         else if( !h->param.rc.i_aq_mode && h->param.analyse.b_ssim )
760         {
761             x264_log( h, X264_LOG_WARNING, "--ssim used with AQ off: results will be invalid!\n" );
762             s = "ssim";
763         }
764         else if(  h->param.rc.i_aq_mode && h->param.analyse.b_psnr )
765         {
766             x264_log( h, X264_LOG_WARNING, "--psnr used with AQ on: results will be invalid!\n" );
767             s = "psnr";
768         }
769         if( s )
770             x264_log( h, X264_LOG_WARNING, "--tune %s should be used if attempting to benchmark %s!\n", s, s );
771     }
772
773     if( !h->param.analyse.b_psy )
774     {
775         h->param.analyse.f_psy_rd = 0;
776         h->param.analyse.f_psy_trellis = 0;
777     }
778     h->param.analyse.f_psy_rd = x264_clip3f( h->param.analyse.f_psy_rd, 0, 10 );
779     h->param.analyse.f_psy_trellis = x264_clip3f( h->param.analyse.f_psy_trellis, 0, 10 );
780     h->mb.i_psy_rd = h->param.analyse.i_subpel_refine >= 6 ? FIX8( h->param.analyse.f_psy_rd ) : 0;
781     h->mb.i_psy_trellis = h->param.analyse.i_trellis ? FIX8( h->param.analyse.f_psy_trellis / 4 ) : 0;
782     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -32, 32);
783     /* In 4:4:4 mode, chroma gets twice as much resolution, so we can halve its quality. */
784     if( b_open && i_csp >= X264_CSP_I444 && i_csp < X264_CSP_BGR && h->param.analyse.b_psy )
785         h->param.analyse.i_chroma_qp_offset += 6;
786     /* Psy RDO increases overall quantizers to improve the quality of luma--this indirectly hurts chroma quality */
787     /* so we lower the chroma QP offset to compensate */
788     if( b_open && h->mb.i_psy_rd )
789         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_rd < 0.25 ? 1 : 2;
790     /* Psy trellis has a similar effect. */
791     if( b_open && h->mb.i_psy_trellis )
792         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_trellis < 0.25 ? 1 : 2;
793     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -12, 12);
794     /* MB-tree requires AQ to be on, even if the strength is zero. */
795     if( !h->param.rc.i_aq_mode && h->param.rc.b_mb_tree )
796     {
797         h->param.rc.i_aq_mode = 1;
798         h->param.rc.f_aq_strength = 0;
799     }
800     h->param.analyse.i_noise_reduction = x264_clip3( h->param.analyse.i_noise_reduction, 0, 1<<16 );
801     if( h->param.analyse.i_subpel_refine >= 10 && (h->param.analyse.i_trellis != 2 || !h->param.rc.i_aq_mode) )
802         h->param.analyse.i_subpel_refine = 9;
803
804     {
805         const x264_level_t *l = x264_levels;
806         if( h->param.i_level_idc < 0 )
807         {
808             int maxrate_bak = h->param.rc.i_vbv_max_bitrate;
809             if( h->param.rc.i_rc_method == X264_RC_ABR && h->param.rc.i_vbv_buffer_size <= 0 )
810                 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate * 2;
811             x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
812             do h->param.i_level_idc = l->level_idc;
813                 while( l[1].level_idc && x264_validate_levels( h, 0 ) && l++ );
814             h->param.rc.i_vbv_max_bitrate = maxrate_bak;
815         }
816         else
817         {
818             while( l->level_idc && l->level_idc != h->param.i_level_idc )
819                 l++;
820             if( l->level_idc == 0 )
821             {
822                 x264_log( h, X264_LOG_ERROR, "invalid level_idc: %d\n", h->param.i_level_idc );
823                 return -1;
824             }
825         }
826         if( h->param.analyse.i_mv_range <= 0 )
827             h->param.analyse.i_mv_range = l->mv_range >> PARAM_INTERLACED;
828         else
829             h->param.analyse.i_mv_range = x264_clip3(h->param.analyse.i_mv_range, 32, 512 >> PARAM_INTERLACED);
830     }
831
832     h->param.analyse.i_weighted_pred = x264_clip3( h->param.analyse.i_weighted_pred, X264_WEIGHTP_NONE, X264_WEIGHTP_SMART );
833
834     if( PARAM_INTERLACED )
835     {
836         if( h->param.analyse.i_me_method >= X264_ME_ESA )
837         {
838             x264_log( h, X264_LOG_WARNING, "interlace + me=esa is not implemented\n" );
839             h->param.analyse.i_me_method = X264_ME_UMH;
840         }
841         if( h->param.analyse.i_weighted_pred > 0 )
842         {
843             x264_log( h, X264_LOG_WARNING, "interlace + weightp is not implemented\n" );
844             h->param.analyse.i_weighted_pred = X264_WEIGHTP_NONE;
845         }
846     }
847
848     if( !h->param.analyse.i_weighted_pred && h->param.rc.b_mb_tree && h->param.analyse.b_psy )
849         h->param.analyse.i_weighted_pred = X264_WEIGHTP_FAKE;
850
851     if( h->i_thread_frames > 1 )
852     {
853         int r = h->param.analyse.i_mv_range_thread;
854         int r2;
855         if( r <= 0 )
856         {
857             // half of the available space is reserved and divided evenly among the threads,
858             // the rest is allocated to whichever thread is far enough ahead to use it.
859             // reserving more space increases quality for some videos, but costs more time
860             // in thread synchronization.
861             int max_range = (h->param.i_height + X264_THREAD_HEIGHT) / h->i_thread_frames - X264_THREAD_HEIGHT;
862             r = max_range / 2;
863         }
864         r = X264_MAX( r, h->param.analyse.i_me_range );
865         r = X264_MIN( r, h->param.analyse.i_mv_range );
866         // round up to use the whole mb row
867         r2 = (r & ~15) + ((-X264_THREAD_HEIGHT) & 15);
868         if( r2 < r )
869             r2 += 16;
870         x264_log( h, X264_LOG_DEBUG, "using mv_range_thread = %d\n", r2 );
871         h->param.analyse.i_mv_range_thread = r2;
872     }
873
874     if( h->param.rc.f_rate_tolerance < 0 )
875         h->param.rc.f_rate_tolerance = 0;
876     if( h->param.rc.f_qblur < 0 )
877         h->param.rc.f_qblur = 0;
878     if( h->param.rc.f_complexity_blur < 0 )
879         h->param.rc.f_complexity_blur = 0;
880
881     h->param.i_sps_id &= 31;
882
883     if( PARAM_INTERLACED )
884         h->param.b_pic_struct = 1;
885
886     h->param.i_nal_hrd = x264_clip3( h->param.i_nal_hrd, X264_NAL_HRD_NONE, X264_NAL_HRD_CBR );
887
888     if( h->param.i_nal_hrd && !h->param.rc.i_vbv_buffer_size )
889     {
890         x264_log( h, X264_LOG_WARNING, "NAL HRD parameters require VBV parameters\n" );
891         h->param.i_nal_hrd = X264_NAL_HRD_NONE;
892     }
893
894     if( h->param.i_nal_hrd == X264_NAL_HRD_CBR &&
895        (h->param.rc.i_bitrate != h->param.rc.i_vbv_max_bitrate || !h->param.rc.i_vbv_max_bitrate) )
896     {
897         x264_log( h, X264_LOG_WARNING, "CBR HRD requires constant bitrate\n" );
898         h->param.i_nal_hrd = X264_NAL_HRD_VBR;
899     }
900
901     /* ensure the booleans are 0 or 1 so they can be used in math */
902 #define BOOLIFY(x) h->param.x = !!h->param.x
903     BOOLIFY( b_cabac );
904     BOOLIFY( b_constrained_intra );
905     BOOLIFY( b_deblocking_filter );
906     BOOLIFY( b_deterministic );
907     BOOLIFY( b_sliced_threads );
908     BOOLIFY( b_interlaced );
909     BOOLIFY( b_intra_refresh );
910     BOOLIFY( b_visualize );
911     BOOLIFY( b_aud );
912     BOOLIFY( b_repeat_headers );
913     BOOLIFY( b_annexb );
914     BOOLIFY( b_vfr_input );
915     BOOLIFY( b_pulldown );
916     BOOLIFY( b_tff );
917     BOOLIFY( b_pic_struct );
918     BOOLIFY( b_fake_interlaced );
919     BOOLIFY( b_open_gop );
920     BOOLIFY( b_bluray_compat );
921     BOOLIFY( analyse.b_transform_8x8 );
922     BOOLIFY( analyse.b_weighted_bipred );
923     BOOLIFY( analyse.b_chroma_me );
924     BOOLIFY( analyse.b_mixed_references );
925     BOOLIFY( analyse.b_fast_pskip );
926     BOOLIFY( analyse.b_dct_decimate );
927     BOOLIFY( analyse.b_psy );
928     BOOLIFY( analyse.b_psnr );
929     BOOLIFY( analyse.b_ssim );
930     BOOLIFY( rc.b_stat_write );
931     BOOLIFY( rc.b_stat_read );
932     BOOLIFY( rc.b_mb_tree );
933 #undef BOOLIFY
934
935     return 0;
936 }
937
938 static void mbcmp_init( x264_t *h )
939 {
940     int satd = !h->mb.b_lossless && h->param.analyse.i_subpel_refine > 1;
941     memcpy( h->pixf.mbcmp, satd ? h->pixf.satd : h->pixf.sad_aligned, sizeof(h->pixf.mbcmp) );
942     memcpy( h->pixf.mbcmp_unaligned, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.mbcmp_unaligned) );
943     h->pixf.intra_mbcmp_x3_16x16 = satd ? h->pixf.intra_satd_x3_16x16 : h->pixf.intra_sad_x3_16x16;
944     h->pixf.intra_mbcmp_x3_8x16c = satd ? h->pixf.intra_satd_x3_8x16c : h->pixf.intra_sad_x3_8x16c;
945     h->pixf.intra_mbcmp_x3_8x8c  = satd ? h->pixf.intra_satd_x3_8x8c  : h->pixf.intra_sad_x3_8x8c;
946     h->pixf.intra_mbcmp_x3_8x8 = satd ? h->pixf.intra_sa8d_x3_8x8 : h->pixf.intra_sad_x3_8x8;
947     h->pixf.intra_mbcmp_x3_4x4 = satd ? h->pixf.intra_satd_x3_4x4 : h->pixf.intra_sad_x3_4x4;
948     h->pixf.intra_mbcmp_x9_4x4 = h->param.b_cpu_independent || h->mb.b_lossless ? NULL
949                                : satd ? h->pixf.intra_satd_x9_4x4 : h->pixf.intra_sad_x9_4x4;
950     satd &= h->param.analyse.i_me_method == X264_ME_TESA;
951     memcpy( h->pixf.fpelcmp, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.fpelcmp) );
952     memcpy( h->pixf.fpelcmp_x3, satd ? h->pixf.satd_x3 : h->pixf.sad_x3, sizeof(h->pixf.fpelcmp_x3) );
953     memcpy( h->pixf.fpelcmp_x4, satd ? h->pixf.satd_x4 : h->pixf.sad_x4, sizeof(h->pixf.fpelcmp_x4) );
954 }
955
956 static void chroma_dsp_init( x264_t *h )
957 {
958     memcpy( h->luma2chroma_pixel, x264_luma2chroma_pixel[CHROMA_FORMAT], sizeof(h->luma2chroma_pixel) );
959
960     switch( CHROMA_FORMAT )
961     {
962         case CHROMA_420:
963             memcpy( h->predict_chroma, h->predict_8x8c, sizeof(h->predict_chroma) );
964             h->loopf.deblock_chroma[0] = h->loopf.deblock_h_chroma_420;
965             h->loopf.deblock_chroma_intra[0] = h->loopf.deblock_h_chroma_420_intra;
966             h->loopf.deblock_chroma_mbaff = h->loopf.deblock_chroma_420_mbaff;
967             h->loopf.deblock_chroma_intra_mbaff = h->loopf.deblock_chroma_420_intra_mbaff;
968             h->pixf.intra_mbcmp_x3_chroma = h->pixf.intra_mbcmp_x3_8x8c;
969             h->quantf.coeff_last[DCT_CHROMA_DC] = h->quantf.coeff_last4;
970             h->quantf.coeff_level_run[DCT_CHROMA_DC] = h->quantf.coeff_level_run4;
971             break;
972         case CHROMA_422:
973             memcpy( h->predict_chroma, h->predict_8x16c, sizeof(h->predict_chroma) );
974             h->loopf.deblock_chroma[0] = h->loopf.deblock_h_chroma_422;
975             h->loopf.deblock_chroma_intra[0] = h->loopf.deblock_h_chroma_422_intra;
976             h->loopf.deblock_chroma_mbaff = h->loopf.deblock_chroma_422_mbaff;
977             h->loopf.deblock_chroma_intra_mbaff = h->loopf.deblock_chroma_422_intra_mbaff;
978             h->pixf.intra_mbcmp_x3_chroma = h->pixf.intra_mbcmp_x3_8x16c;
979             h->quantf.coeff_last[DCT_CHROMA_DC] = h->quantf.coeff_last8;
980             h->quantf.coeff_level_run[DCT_CHROMA_DC] = h->quantf.coeff_level_run8;
981             break;
982         case CHROMA_444:
983             h->loopf.deblock_chroma_mbaff = h->loopf.deblock_luma_mbaff;
984             h->loopf.deblock_chroma_intra_mbaff = h->loopf.deblock_luma_intra_mbaff;
985             break;
986     }
987 }
988
989 static void x264_set_aspect_ratio( x264_t *h, x264_param_t *param, int initial )
990 {
991     /* VUI */
992     if( param->vui.i_sar_width > 0 && param->vui.i_sar_height > 0 )
993     {
994         uint32_t i_w = param->vui.i_sar_width;
995         uint32_t i_h = param->vui.i_sar_height;
996         uint32_t old_w = h->param.vui.i_sar_width;
997         uint32_t old_h = h->param.vui.i_sar_height;
998
999         x264_reduce_fraction( &i_w, &i_h );
1000
1001         while( i_w > 65535 || i_h > 65535 )
1002         {
1003             i_w /= 2;
1004             i_h /= 2;
1005         }
1006
1007         x264_reduce_fraction( &i_w, &i_h );
1008
1009         if( i_w != old_w || i_h != old_h || initial )
1010         {
1011             h->param.vui.i_sar_width = 0;
1012             h->param.vui.i_sar_height = 0;
1013             if( i_w == 0 || i_h == 0 )
1014                 x264_log( h, X264_LOG_WARNING, "cannot create valid sample aspect ratio\n" );
1015             else
1016             {
1017                 x264_log( h, initial?X264_LOG_INFO:X264_LOG_DEBUG, "using SAR=%d/%d\n", i_w, i_h );
1018                 h->param.vui.i_sar_width = i_w;
1019                 h->param.vui.i_sar_height = i_h;
1020             }
1021             x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
1022         }
1023     }
1024 }
1025
1026 /****************************************************************************
1027  * x264_encoder_open:
1028  ****************************************************************************/
1029 x264_t *x264_encoder_open( x264_param_t *param )
1030 {
1031     x264_t *h;
1032     char buf[1000], *p;
1033     int qp, i_slicetype_length;
1034
1035     CHECKED_MALLOCZERO( h, sizeof(x264_t) );
1036
1037     /* Create a copy of param */
1038     memcpy( &h->param, param, sizeof(x264_param_t) );
1039
1040     if( param->param_free )
1041         param->param_free( param );
1042
1043     if( x264_threading_init() )
1044     {
1045         x264_log( h, X264_LOG_ERROR, "unable to initialize threading\n" );
1046         goto fail;
1047     }
1048
1049     if( x264_validate_parameters( h, 1 ) < 0 )
1050         goto fail;
1051
1052     if( h->param.psz_cqm_file )
1053         if( x264_cqm_parse_file( h, h->param.psz_cqm_file ) < 0 )
1054             goto fail;
1055
1056     if( h->param.rc.psz_stat_out )
1057         h->param.rc.psz_stat_out = strdup( h->param.rc.psz_stat_out );
1058     if( h->param.rc.psz_stat_in )
1059         h->param.rc.psz_stat_in = strdup( h->param.rc.psz_stat_in );
1060
1061     x264_reduce_fraction( &h->param.i_fps_num, &h->param.i_fps_den );
1062     x264_reduce_fraction( &h->param.i_timebase_num, &h->param.i_timebase_den );
1063
1064     /* Init x264_t */
1065     h->i_frame = -1;
1066     h->i_frame_num = 0;
1067     h->i_idr_pic_id = 0;
1068
1069     if( (uint64_t)h->param.i_timebase_den * 2 > UINT32_MAX )
1070     {
1071         x264_log( h, X264_LOG_ERROR, "Effective timebase denominator %u exceeds H.264 maximum\n", h->param.i_timebase_den );
1072         goto fail;
1073     }
1074
1075     x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
1076     x264_pps_init( h->pps, h->param.i_sps_id, &h->param, h->sps );
1077
1078     x264_set_aspect_ratio( h, &h->param, 1 );
1079
1080     x264_validate_levels( h, 1 );
1081
1082     h->chroma_qp_table = i_chroma_qp_table + 12 + h->pps->i_chroma_qp_index_offset;
1083
1084     if( x264_cqm_init( h ) < 0 )
1085         goto fail;
1086
1087     h->mb.i_mb_width = h->sps->i_mb_width;
1088     h->mb.i_mb_height = h->sps->i_mb_height;
1089     h->mb.i_mb_count = h->mb.i_mb_width * h->mb.i_mb_height;
1090
1091     h->mb.chroma_h_shift = CHROMA_FORMAT == CHROMA_420 || CHROMA_FORMAT == CHROMA_422;
1092     h->mb.chroma_v_shift = CHROMA_FORMAT == CHROMA_420;
1093
1094     /* Adaptive MBAFF and subme 0 are not supported as we require halving motion
1095      * vectors during prediction, resulting in hpel mvs.
1096      * The chosen solution is to make MBAFF non-adaptive in this case. */
1097     h->mb.b_adaptive_mbaff = PARAM_INTERLACED && h->param.analyse.i_subpel_refine;
1098
1099     /* Init frames. */
1100     if( h->param.i_bframe_adaptive == X264_B_ADAPT_TRELLIS && !h->param.rc.b_stat_read )
1101         h->frames.i_delay = X264_MAX(h->param.i_bframe,3)*4;
1102     else
1103         h->frames.i_delay = h->param.i_bframe;
1104     if( h->param.rc.b_mb_tree || h->param.rc.i_vbv_buffer_size )
1105         h->frames.i_delay = X264_MAX( h->frames.i_delay, h->param.rc.i_lookahead );
1106     i_slicetype_length = h->frames.i_delay;
1107     h->frames.i_delay += h->i_thread_frames - 1;
1108     h->frames.i_delay += h->param.i_sync_lookahead;
1109     h->frames.i_delay += h->param.b_vfr_input;
1110     h->frames.i_bframe_delay = h->param.i_bframe ? (h->param.i_bframe_pyramid ? 2 : 1) : 0;
1111
1112     h->frames.i_max_ref0 = h->param.i_frame_reference;
1113     h->frames.i_max_ref1 = X264_MIN( h->sps->vui.i_num_reorder_frames, h->param.i_frame_reference );
1114     h->frames.i_max_dpb  = h->sps->vui.i_max_dec_frame_buffering;
1115     h->frames.b_have_lowres = !h->param.rc.b_stat_read
1116         && ( h->param.rc.i_rc_method == X264_RC_ABR
1117           || h->param.rc.i_rc_method == X264_RC_CRF
1118           || h->param.i_bframe_adaptive
1119           || h->param.i_scenecut_threshold
1120           || h->param.rc.b_mb_tree
1121           || h->param.analyse.i_weighted_pred );
1122     h->frames.b_have_lowres |= h->param.rc.b_stat_read && h->param.rc.i_vbv_buffer_size > 0;
1123     h->frames.b_have_sub8x8_esa = !!(h->param.analyse.inter & X264_ANALYSE_PSUB8x8);
1124
1125     h->frames.i_last_idr =
1126     h->frames.i_last_keyframe = - h->param.i_keyint_max;
1127     h->frames.i_input    = 0;
1128     h->frames.i_largest_pts = h->frames.i_second_largest_pts = -1;
1129     h->frames.i_poc_last_open_gop = -1;
1130
1131     CHECKED_MALLOCZERO( h->frames.unused[0], (h->frames.i_delay + 3) * sizeof(x264_frame_t *) );
1132     /* Allocate room for max refs plus a few extra just in case. */
1133     CHECKED_MALLOCZERO( h->frames.unused[1], (h->i_thread_frames + X264_REF_MAX + 4) * sizeof(x264_frame_t *) );
1134     CHECKED_MALLOCZERO( h->frames.current, (h->param.i_sync_lookahead + h->param.i_bframe
1135                         + h->i_thread_frames + 3) * sizeof(x264_frame_t *) );
1136     if( h->param.analyse.i_weighted_pred > 0 )
1137         CHECKED_MALLOCZERO( h->frames.blank_unused, h->i_thread_frames * 4 * sizeof(x264_frame_t *) );
1138     h->i_ref[0] = h->i_ref[1] = 0;
1139     h->i_cpb_delay = h->i_coded_fields = h->i_disp_fields = 0;
1140     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);
1141     h->i_disp_fields_last_frame = -1;
1142     x264_rdo_init();
1143
1144     /* init CPU functions */
1145     x264_predict_16x16_init( h->param.cpu, h->predict_16x16 );
1146     x264_predict_8x8c_init( h->param.cpu, h->predict_8x8c );
1147     x264_predict_8x16c_init( h->param.cpu, h->predict_8x16c );
1148     x264_predict_8x8_init( h->param.cpu, h->predict_8x8, &h->predict_8x8_filter );
1149     x264_predict_4x4_init( h->param.cpu, h->predict_4x4 );
1150     if( h->param.b_cabac )
1151         x264_cabac_init( h );
1152     else
1153         x264_cavlc_init();
1154     x264_pixel_init( h->param.cpu, &h->pixf );
1155     x264_dct_init( h->param.cpu, &h->dctf );
1156     x264_zigzag_init( h->param.cpu, &h->zigzagf_progressive, &h->zigzagf_interlaced );
1157     memcpy( &h->zigzagf, PARAM_INTERLACED ? &h->zigzagf_interlaced : &h->zigzagf_progressive, sizeof(h->zigzagf) );
1158     x264_mc_init( h->param.cpu, &h->mc );
1159     x264_quant_init( h, h->param.cpu, &h->quantf );
1160     x264_deblock_init( h->param.cpu, &h->loopf, PARAM_INTERLACED );
1161     x264_bitstream_init( h->param.cpu, &h->bsf );
1162     x264_dct_init_weights();
1163
1164     mbcmp_init( h );
1165     chroma_dsp_init( h );
1166
1167     p = buf + sprintf( buf, "using cpu capabilities:" );
1168     for( int i = 0; x264_cpu_names[i].flags; i++ )
1169     {
1170         if( !strcmp(x264_cpu_names[i].name, "SSE2")
1171             && h->param.cpu & (X264_CPU_SSE2_IS_FAST|X264_CPU_SSE2_IS_SLOW) )
1172             continue;
1173         if( !strcmp(x264_cpu_names[i].name, "SSE3")
1174             && (h->param.cpu & X264_CPU_SSSE3 || !(h->param.cpu & X264_CPU_CACHELINE_64)) )
1175             continue;
1176         if( !strcmp(x264_cpu_names[i].name, "SSE4.1")
1177             && (h->param.cpu & X264_CPU_SSE42) )
1178             continue;
1179         if( (h->param.cpu & x264_cpu_names[i].flags) == x264_cpu_names[i].flags
1180             && (!i || x264_cpu_names[i].flags != x264_cpu_names[i-1].flags) )
1181             p += sprintf( p, " %s", x264_cpu_names[i].name );
1182     }
1183     if( !h->param.cpu )
1184         p += sprintf( p, " none!" );
1185     x264_log( h, X264_LOG_INFO, "%s\n", buf );
1186
1187     float *logs = x264_analyse_prepare_costs( h );
1188     if( !logs )
1189         goto fail;
1190     for( qp = X264_MIN( h->param.rc.i_qp_min, QP_MAX_SPEC ); qp <= h->param.rc.i_qp_max; qp++ )
1191         if( x264_analyse_init_costs( h, logs, qp ) )
1192             goto fail;
1193     if( x264_analyse_init_costs( h, logs, X264_LOOKAHEAD_QP ) )
1194         goto fail;
1195     x264_free( logs );
1196
1197     static const uint16_t cost_mv_correct[7] = { 24, 47, 95, 189, 379, 757, 1515 };
1198     /* Checks for known miscompilation issues. */
1199     if( h->cost_mv[X264_LOOKAHEAD_QP][2013] != cost_mv_correct[BIT_DEPTH-8] )
1200     {
1201         x264_log( h, X264_LOG_ERROR, "MV cost test failed: x264 has been miscompiled!\n" );
1202         goto fail;
1203     }
1204
1205     /* Must be volatile or else GCC will optimize it out. */
1206     volatile int temp = 392;
1207     if( x264_clz( temp ) != 23 )
1208     {
1209         x264_log( h, X264_LOG_ERROR, "CLZ test failed: x264 has been miscompiled!\n" );
1210 #if ARCH_X86 || ARCH_X86_64
1211         x264_log( h, X264_LOG_ERROR, "Are you attempting to run an SSE4a-targeted build on a CPU that\n" );
1212         x264_log( h, X264_LOG_ERROR, "doesn't support it?\n" );
1213 #endif
1214         goto fail;
1215     }
1216
1217     h->out.i_nal = 0;
1218     h->out.i_bitstream = X264_MAX( 1000000, h->param.i_width * h->param.i_height * 4
1219         * ( h->param.rc.i_rc_method == X264_RC_ABR ? pow( 0.95, h->param.rc.i_qp_min )
1220           : pow( 0.95, h->param.rc.i_qp_constant ) * X264_MAX( 1, h->param.rc.f_ip_factor )));
1221
1222     h->nal_buffer_size = h->out.i_bitstream * 3/2 + 4;
1223     CHECKED_MALLOC( h->nal_buffer, h->nal_buffer_size );
1224
1225     if( h->param.i_threads > 1 &&
1226         x264_threadpool_init( &h->threadpool, h->param.i_threads, (void*)x264_encoder_thread_init, h ) )
1227         goto fail;
1228
1229     h->thread[0] = h;
1230     for( int i = 1; i < h->param.i_threads + !!h->param.i_sync_lookahead; i++ )
1231         CHECKED_MALLOC( h->thread[i], sizeof(x264_t) );
1232
1233     for( int i = 0; i < h->param.i_threads; i++ )
1234     {
1235         int init_nal_count = h->param.i_slice_count + 3;
1236         int allocate_threadlocal_data = !h->param.b_sliced_threads || !i;
1237         if( i > 0 )
1238             *h->thread[i] = *h;
1239
1240         if( allocate_threadlocal_data )
1241         {
1242             h->thread[i]->fdec = x264_frame_pop_unused( h, 1 );
1243             if( !h->thread[i]->fdec )
1244                 goto fail;
1245         }
1246         else
1247             h->thread[i]->fdec = h->thread[0]->fdec;
1248
1249         CHECKED_MALLOC( h->thread[i]->out.p_bitstream, h->out.i_bitstream );
1250         /* Start each thread with room for init_nal_count NAL units; it'll realloc later if needed. */
1251         CHECKED_MALLOC( h->thread[i]->out.nal, init_nal_count*sizeof(x264_nal_t) );
1252         h->thread[i]->out.i_nals_allocated = init_nal_count;
1253
1254         if( allocate_threadlocal_data && x264_macroblock_cache_allocate( h->thread[i] ) < 0 )
1255             goto fail;
1256     }
1257
1258     if( x264_lookahead_init( h, i_slicetype_length ) )
1259         goto fail;
1260
1261     for( int i = 0; i < h->param.i_threads; i++ )
1262         if( x264_macroblock_thread_allocate( h->thread[i], 0 ) < 0 )
1263             goto fail;
1264
1265     if( x264_ratecontrol_new( h ) < 0 )
1266         goto fail;
1267
1268     if( h->param.i_nal_hrd )
1269     {
1270         x264_log( h, X264_LOG_DEBUG, "HRD bitrate: %i bits/sec\n", h->sps->vui.hrd.i_bit_rate_unscaled );
1271         x264_log( h, X264_LOG_DEBUG, "CPB size: %i bits\n", h->sps->vui.hrd.i_cpb_size_unscaled );
1272     }
1273
1274     if( h->param.psz_dump_yuv )
1275     {
1276         /* create or truncate the reconstructed video file */
1277         FILE *f = fopen( h->param.psz_dump_yuv, "w" );
1278         if( !f )
1279         {
1280             x264_log( h, X264_LOG_ERROR, "dump_yuv: can't write to %s\n", h->param.psz_dump_yuv );
1281             goto fail;
1282         }
1283         else if( !x264_is_regular_file( f ) )
1284         {
1285             x264_log( h, X264_LOG_ERROR, "dump_yuv: incompatible with non-regular file %s\n", h->param.psz_dump_yuv );
1286             goto fail;
1287         }
1288         fclose( f );
1289     }
1290
1291     const char *profile = h->sps->i_profile_idc == PROFILE_BASELINE ? "Constrained Baseline" :
1292                           h->sps->i_profile_idc == PROFILE_MAIN ? "Main" :
1293                           h->sps->i_profile_idc == PROFILE_HIGH ? "High" :
1294                           h->sps->i_profile_idc == PROFILE_HIGH10 ? (h->sps->b_constraint_set3 == 1 ? "High 10 Intra" : "High 10") :
1295                           h->sps->i_profile_idc == PROFILE_HIGH422 ? (h->sps->b_constraint_set3 == 1 ? "High 4:2:2 Intra" : "High 4:2:2") :
1296                           h->sps->b_constraint_set3 == 1 ? "High 4:4:4 Intra" : "High 4:4:4 Predictive";
1297     char level[4];
1298     snprintf( level, sizeof(level), "%d.%d", h->sps->i_level_idc/10, h->sps->i_level_idc%10 );
1299     if( h->sps->i_level_idc == 9 || ( h->sps->i_level_idc == 11 && h->sps->b_constraint_set3 &&
1300         (h->sps->i_profile_idc >= PROFILE_BASELINE && h->sps->i_profile_idc <= PROFILE_EXTENDED) ) )
1301         strcpy( level, "1b" );
1302
1303     if( h->sps->i_profile_idc < PROFILE_HIGH10 )
1304     {
1305         x264_log( h, X264_LOG_INFO, "profile %s, level %s\n",
1306             profile, level );
1307     }
1308     else
1309     {
1310         static const char * const subsampling[4] = { "4:0:0", "4:2:0", "4:2:2", "4:4:4" };
1311         x264_log( h, X264_LOG_INFO, "profile %s, level %s, %s %d-bit\n",
1312             profile, level, subsampling[CHROMA_FORMAT], BIT_DEPTH );
1313     }
1314
1315     return h;
1316 fail:
1317     x264_free( h );
1318     return NULL;
1319 }
1320
1321 /****************************************************************************
1322  * x264_encoder_reconfig:
1323  ****************************************************************************/
1324 int x264_encoder_reconfig( x264_t *h, x264_param_t *param )
1325 {
1326     int rc_reconfig = 0;
1327     h = h->thread[h->thread[0]->i_thread_phase];
1328     x264_set_aspect_ratio( h, param, 0 );
1329 #define COPY(var) h->param.var = param->var
1330     COPY( i_frame_reference ); // but never uses more refs than initially specified
1331     COPY( i_bframe_bias );
1332     if( h->param.i_scenecut_threshold )
1333         COPY( i_scenecut_threshold ); // can't turn it on or off, only vary the threshold
1334     COPY( b_deblocking_filter );
1335     COPY( i_deblocking_filter_alphac0 );
1336     COPY( i_deblocking_filter_beta );
1337     COPY( i_frame_packing );
1338     COPY( analyse.inter );
1339     COPY( analyse.intra );
1340     COPY( analyse.i_direct_mv_pred );
1341     /* Scratch buffer prevents me_range from being increased for esa/tesa */
1342     if( h->param.analyse.i_me_method < X264_ME_ESA || param->analyse.i_me_range < h->param.analyse.i_me_range )
1343         COPY( analyse.i_me_range );
1344     COPY( analyse.i_noise_reduction );
1345     /* We can't switch out of subme=0 during encoding. */
1346     if( h->param.analyse.i_subpel_refine )
1347         COPY( analyse.i_subpel_refine );
1348     COPY( analyse.i_trellis );
1349     COPY( analyse.b_chroma_me );
1350     COPY( analyse.b_dct_decimate );
1351     COPY( analyse.b_fast_pskip );
1352     COPY( analyse.b_mixed_references );
1353     COPY( analyse.f_psy_rd );
1354     COPY( analyse.f_psy_trellis );
1355     COPY( crop_rect );
1356     // can only twiddle these if they were enabled to begin with:
1357     if( h->param.analyse.i_me_method >= X264_ME_ESA || param->analyse.i_me_method < X264_ME_ESA )
1358         COPY( analyse.i_me_method );
1359     if( h->param.analyse.i_me_method >= X264_ME_ESA && !h->frames.b_have_sub8x8_esa )
1360         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
1361     if( h->pps->b_transform_8x8_mode )
1362         COPY( analyse.b_transform_8x8 );
1363     if( h->frames.i_max_ref1 > 1 )
1364         COPY( i_bframe_pyramid );
1365     COPY( i_slice_max_size );
1366     COPY( i_slice_max_mbs );
1367     COPY( i_slice_count );
1368     COPY( b_tff );
1369
1370     /* VBV can't be turned on if it wasn't on to begin with */
1371     if( h->param.rc.i_vbv_max_bitrate > 0 && h->param.rc.i_vbv_buffer_size > 0 &&
1372           param->rc.i_vbv_max_bitrate > 0 &&   param->rc.i_vbv_buffer_size > 0 )
1373     {
1374         rc_reconfig |= h->param.rc.i_vbv_max_bitrate != param->rc.i_vbv_max_bitrate;
1375         rc_reconfig |= h->param.rc.i_vbv_buffer_size != param->rc.i_vbv_buffer_size;
1376         rc_reconfig |= h->param.rc.i_bitrate != param->rc.i_bitrate;
1377         COPY( rc.i_vbv_max_bitrate );
1378         COPY( rc.i_vbv_buffer_size );
1379         COPY( rc.i_bitrate );
1380     }
1381     rc_reconfig |= h->param.rc.f_rf_constant != param->rc.f_rf_constant;
1382     rc_reconfig |= h->param.rc.f_rf_constant_max != param->rc.f_rf_constant_max;
1383     COPY( rc.f_rf_constant );
1384     COPY( rc.f_rf_constant_max );
1385 #undef COPY
1386
1387     mbcmp_init( h );
1388
1389     int ret = x264_validate_parameters( h, 0 );
1390
1391     /* Supported reconfiguration options (1-pass only):
1392      * vbv-maxrate
1393      * vbv-bufsize
1394      * crf
1395      * bitrate (CBR only) */
1396     if( !ret && rc_reconfig )
1397         x264_ratecontrol_init_reconfigurable( h, 0 );
1398
1399     return ret;
1400 }
1401
1402 /****************************************************************************
1403  * x264_encoder_parameters:
1404  ****************************************************************************/
1405 void x264_encoder_parameters( x264_t *h, x264_param_t *param )
1406 {
1407     memcpy( param, &h->thread[h->i_thread_phase]->param, sizeof(x264_param_t) );
1408 }
1409
1410 /* internal usage */
1411 static void x264_nal_start( x264_t *h, int i_type, int i_ref_idc )
1412 {
1413     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1414
1415     nal->i_ref_idc        = i_ref_idc;
1416     nal->i_type           = i_type;
1417     nal->b_long_startcode = 1;
1418
1419     nal->i_payload= 0;
1420     nal->p_payload= &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8];
1421 }
1422
1423 /* if number of allocated nals is not enough, re-allocate a larger one. */
1424 static int x264_nal_check_buffer( x264_t *h )
1425 {
1426     if( h->out.i_nal >= h->out.i_nals_allocated )
1427     {
1428         x264_nal_t *new_out = x264_malloc( sizeof(x264_nal_t) * (h->out.i_nals_allocated*2) );
1429         if( !new_out )
1430             return -1;
1431         memcpy( new_out, h->out.nal, sizeof(x264_nal_t) * (h->out.i_nals_allocated) );
1432         x264_free( h->out.nal );
1433         h->out.nal = new_out;
1434         h->out.i_nals_allocated *= 2;
1435     }
1436     return 0;
1437 }
1438
1439 static int x264_nal_end( x264_t *h )
1440 {
1441     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1442     uint8_t *end = &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8];
1443     nal->i_payload = end - nal->p_payload;
1444     /* nal_escape_mmx reads past the end of the input.
1445      * While undefined padding wouldn't actually affect the output, it makes valgrind unhappy. */
1446     memset( end, 0xff, 32 );
1447     if( h->param.nalu_process )
1448         h->param.nalu_process( h, nal );
1449     h->out.i_nal++;
1450
1451     return x264_nal_check_buffer( h );
1452 }
1453
1454 static int x264_encoder_encapsulate_nals( x264_t *h, int start )
1455 {
1456     int nal_size = 0, previous_nal_size = 0;
1457
1458     if( h->param.nalu_process )
1459     {
1460         for( int i = start; i < h->out.i_nal; i++ )
1461             nal_size += h->out.nal[i].i_payload;
1462         return nal_size;
1463     }
1464
1465     for( int i = 0; i < start; i++ )
1466         previous_nal_size += h->out.nal[i].i_payload;
1467
1468     for( int i = start; i < h->out.i_nal; i++ )
1469         nal_size += h->out.nal[i].i_payload;
1470
1471     /* Worst-case NAL unit escaping: reallocate the buffer if it's too small. */
1472     int necessary_size = nal_size * 3/2 + h->out.i_nal * 4;
1473     if( h->nal_buffer_size < necessary_size )
1474     {
1475         h->nal_buffer_size = necessary_size * 2;
1476         uint8_t *buf = x264_malloc( h->nal_buffer_size );
1477         if( !buf )
1478             return -1;
1479         if( previous_nal_size )
1480             memcpy( buf, h->nal_buffer, previous_nal_size );
1481         x264_free( h->nal_buffer );
1482         h->nal_buffer = buf;
1483     }
1484
1485     uint8_t *nal_buffer = h->nal_buffer + previous_nal_size;
1486
1487     for( int i = start; i < h->out.i_nal; i++ )
1488     {
1489         h->out.nal[i].b_long_startcode = !i || h->out.nal[i].i_type == NAL_SPS || h->out.nal[i].i_type == NAL_PPS;
1490         x264_nal_encode( h, nal_buffer, &h->out.nal[i] );
1491         nal_buffer += h->out.nal[i].i_payload;
1492     }
1493
1494     x264_emms();
1495
1496     return nal_buffer - (h->nal_buffer + previous_nal_size);
1497 }
1498
1499 /****************************************************************************
1500  * x264_encoder_headers:
1501  ****************************************************************************/
1502 int x264_encoder_headers( x264_t *h, x264_nal_t **pp_nal, int *pi_nal )
1503 {
1504     int frame_size = 0;
1505     /* init bitstream context */
1506     h->out.i_nal = 0;
1507     bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
1508
1509     /* Write SEI, SPS and PPS. */
1510
1511     /* generate sequence parameters */
1512     x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
1513     x264_sps_write( &h->out.bs, h->sps );
1514     if( x264_nal_end( h ) )
1515         return -1;
1516
1517     /* generate picture parameters */
1518     x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
1519     x264_pps_write( &h->out.bs, h->sps, h->pps );
1520     if( x264_nal_end( h ) )
1521         return -1;
1522
1523     /* identify ourselves */
1524     x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
1525     if( x264_sei_version_write( h, &h->out.bs ) )
1526         return -1;
1527     if( x264_nal_end( h ) )
1528         return -1;
1529
1530     frame_size = x264_encoder_encapsulate_nals( h, 0 );
1531     if( frame_size < 0 )
1532         return -1;
1533
1534     /* now set output*/
1535     *pi_nal = h->out.i_nal;
1536     *pp_nal = &h->out.nal[0];
1537     h->out.i_nal = 0;
1538
1539     return frame_size;
1540 }
1541
1542 /* Check to see whether we have chosen a reference list ordering different
1543  * from the standard's default. */
1544 static inline void x264_reference_check_reorder( x264_t *h )
1545 {
1546     /* The reorder check doesn't check for missing frames, so just
1547      * force a reorder if one of the reference list is corrupt. */
1548     for( int i = 0; h->frames.reference[i]; i++ )
1549         if( h->frames.reference[i]->b_corrupt )
1550         {
1551             h->b_ref_reorder[0] = 1;
1552             return;
1553         }
1554     for( int list = 0; list <= (h->sh.i_type == SLICE_TYPE_B); list++ )
1555         for( int i = 0; i < h->i_ref[list] - 1; i++ )
1556         {
1557             int framenum_diff = h->fref[list][i+1]->i_frame_num - h->fref[list][i]->i_frame_num;
1558             int poc_diff = h->fref[list][i+1]->i_poc - h->fref[list][i]->i_poc;
1559             /* P and B-frames use different default orders. */
1560             if( h->sh.i_type == SLICE_TYPE_P ? framenum_diff > 0 : list == 1 ? poc_diff < 0 : poc_diff > 0 )
1561             {
1562                 h->b_ref_reorder[list] = 1;
1563                 return;
1564             }
1565         }
1566 }
1567
1568 /* return -1 on failure, else return the index of the new reference frame */
1569 int x264_weighted_reference_duplicate( x264_t *h, int i_ref, const x264_weight_t *w )
1570 {
1571     int i = h->i_ref[0];
1572     int j = 1;
1573     x264_frame_t *newframe;
1574     if( i <= 1 ) /* empty list, definitely can't duplicate frame */
1575         return -1;
1576
1577     //Duplication is only used in X264_WEIGHTP_SMART
1578     if( h->param.analyse.i_weighted_pred != X264_WEIGHTP_SMART )
1579         return -1;
1580
1581     /* Duplication is a hack to compensate for crappy rounding in motion compensation.
1582      * With high bit depth, it's not worth doing, so turn it off except in the case of
1583      * unweighted dupes. */
1584     if( BIT_DEPTH > 8 && w != x264_weight_none )
1585         return -1;
1586
1587     newframe = x264_frame_pop_blank_unused( h );
1588     if( !newframe )
1589         return -1;
1590
1591     //FIXME: probably don't need to copy everything
1592     *newframe = *h->fref[0][i_ref];
1593     newframe->i_reference_count = 1;
1594     newframe->orig = h->fref[0][i_ref];
1595     newframe->b_duplicate = 1;
1596     memcpy( h->fenc->weight[j], w, sizeof(h->fenc->weight[i]) );
1597
1598     /* shift the frames to make space for the dupe. */
1599     h->b_ref_reorder[0] = 1;
1600     if( h->i_ref[0] < X264_REF_MAX )
1601         ++h->i_ref[0];
1602     h->fref[0][X264_REF_MAX-1] = NULL;
1603     x264_frame_unshift( &h->fref[0][j], newframe );
1604
1605     return j;
1606 }
1607
1608 static void x264_weighted_pred_init( x264_t *h )
1609 {
1610     /* for now no analysis and set all weights to nothing */
1611     for( int i_ref = 0; i_ref < h->i_ref[0]; i_ref++ )
1612         h->fenc->weighted[i_ref] = h->fref[0][i_ref]->filtered[0][0];
1613
1614     // FIXME: This only supports weighting of one reference frame
1615     // and duplicates of that frame.
1616     h->fenc->i_lines_weighted = 0;
1617
1618     for( int i_ref = 0; i_ref < (h->i_ref[0] << SLICE_MBAFF); i_ref++ )
1619         for( int i = 0; i < 3; i++ )
1620             h->sh.weight[i_ref][i].weightfn = NULL;
1621
1622
1623     if( h->sh.i_type != SLICE_TYPE_P || h->param.analyse.i_weighted_pred <= 0 )
1624         return;
1625
1626     int i_padv = PADV << PARAM_INTERLACED;
1627     int denom = -1;
1628     int weightplane[2] = { 0, 0 };
1629     int buffer_next = 0;
1630     for( int i = 0; i < 3; i++ )
1631     {
1632         for( int j = 0; j < h->i_ref[0]; j++ )
1633         {
1634             if( h->fenc->weight[j][i].weightfn )
1635             {
1636                 h->sh.weight[j][i] = h->fenc->weight[j][i];
1637                 // if weight is useless, don't write it to stream
1638                 if( h->sh.weight[j][i].i_scale == 1<<h->sh.weight[j][i].i_denom && h->sh.weight[j][i].i_offset == 0 )
1639                     h->sh.weight[j][i].weightfn = NULL;
1640                 else
1641                 {
1642                     if( !weightplane[!!i] )
1643                     {
1644                         weightplane[!!i] = 1;
1645                         h->sh.weight[0][!!i].i_denom = denom = h->sh.weight[j][i].i_denom;
1646                         assert( x264_clip3( denom, 0, 7 ) == denom );
1647                     }
1648
1649                     assert( h->sh.weight[j][i].i_denom == denom );
1650                     if( !i )
1651                     {
1652                         h->fenc->weighted[j] = h->mb.p_weight_buf[buffer_next++] + h->fenc->i_stride[0] * i_padv + PADH;
1653                         //scale full resolution frame
1654                         if( h->param.i_threads == 1 )
1655                         {
1656                             pixel *src = h->fref[0][j]->filtered[0][0] - h->fref[0][j]->i_stride[0]*i_padv - PADH;
1657                             pixel *dst = h->fenc->weighted[j] - h->fenc->i_stride[0]*i_padv - PADH;
1658                             int stride = h->fenc->i_stride[0];
1659                             int width = h->fenc->i_width[0] + PADH*2;
1660                             int height = h->fenc->i_lines[0] + i_padv*2;
1661                             x264_weight_scale_plane( h, dst, stride, src, stride, width, height, &h->sh.weight[j][0] );
1662                             h->fenc->i_lines_weighted = height;
1663                         }
1664                     }
1665                 }
1666             }
1667         }
1668     }
1669
1670     if( weightplane[1] )
1671         for( int i = 0; i < h->i_ref[0]; i++ )
1672         {
1673             if( h->sh.weight[i][1].weightfn && !h->sh.weight[i][2].weightfn )
1674             {
1675                 h->sh.weight[i][2].i_scale = 1 << h->sh.weight[0][1].i_denom;
1676                 h->sh.weight[i][2].i_offset = 0;
1677             }
1678             else if( h->sh.weight[i][2].weightfn && !h->sh.weight[i][1].weightfn )
1679             {
1680                 h->sh.weight[i][1].i_scale = 1 << h->sh.weight[0][1].i_denom;
1681                 h->sh.weight[i][1].i_offset = 0;
1682             }
1683         }
1684
1685     if( !weightplane[0] )
1686         h->sh.weight[0][0].i_denom = 0;
1687     if( !weightplane[1] )
1688         h->sh.weight[0][1].i_denom = 0;
1689     h->sh.weight[0][2].i_denom = h->sh.weight[0][1].i_denom;
1690 }
1691
1692 static inline int x264_reference_distance( x264_t *h, x264_frame_t *frame )
1693 {
1694     if( h->param.i_frame_packing == 5 )
1695         return abs((h->fenc->i_frame&~1) - (frame->i_frame&~1)) +
1696                   ((h->fenc->i_frame&1) != (frame->i_frame&1));
1697     else
1698         return abs(h->fenc->i_frame - frame->i_frame);
1699 }
1700
1701 static inline void x264_reference_build_list( x264_t *h, int i_poc )
1702 {
1703     int b_ok;
1704
1705     /* build ref list 0/1 */
1706     h->mb.pic.i_fref[0] = h->i_ref[0] = 0;
1707     h->mb.pic.i_fref[1] = h->i_ref[1] = 0;
1708     if( h->sh.i_type == SLICE_TYPE_I )
1709         return;
1710
1711     for( int i = 0; h->frames.reference[i]; i++ )
1712     {
1713         if( h->frames.reference[i]->b_corrupt )
1714             continue;
1715         if( h->frames.reference[i]->i_poc < i_poc )
1716             h->fref[0][h->i_ref[0]++] = h->frames.reference[i];
1717         else if( h->frames.reference[i]->i_poc > i_poc )
1718             h->fref[1][h->i_ref[1]++] = h->frames.reference[i];
1719     }
1720
1721     /* Order reference lists by distance from the current frame. */
1722     for( int list = 0; list < 2; list++ )
1723     {
1724         h->fref_nearest[list] = h->fref[list][0];
1725         do
1726         {
1727             b_ok = 1;
1728             for( int i = 0; i < h->i_ref[list] - 1; i++ )
1729             {
1730                 if( list ? h->fref[list][i+1]->i_poc < h->fref_nearest[list]->i_poc
1731                          : h->fref[list][i+1]->i_poc > h->fref_nearest[list]->i_poc )
1732                     h->fref_nearest[list] = h->fref[list][i+1];
1733                 if( x264_reference_distance( h, h->fref[list][i] ) > x264_reference_distance( h, h->fref[list][i+1] ) )
1734                 {
1735                     XCHG( x264_frame_t*, h->fref[list][i], h->fref[list][i+1] );
1736                     b_ok = 0;
1737                     break;
1738                 }
1739             }
1740         } while( !b_ok );
1741     }
1742
1743     if( h->sh.i_mmco_remove_from_end )
1744         for( int i = h->i_ref[0]-1; i >= h->i_ref[0] - h->sh.i_mmco_remove_from_end; i-- )
1745         {
1746             int diff = h->i_frame_num - h->fref[0][i]->i_frame_num;
1747             h->sh.mmco[h->sh.i_mmco_command_count].i_poc = h->fref[0][i]->i_poc;
1748             h->sh.mmco[h->sh.i_mmco_command_count++].i_difference_of_pic_nums = diff;
1749         }
1750
1751     x264_reference_check_reorder( h );
1752
1753     h->i_ref[1] = X264_MIN( h->i_ref[1], h->frames.i_max_ref1 );
1754     h->i_ref[0] = X264_MIN( h->i_ref[0], h->frames.i_max_ref0 );
1755     h->i_ref[0] = X264_MIN( h->i_ref[0], h->param.i_frame_reference ); // if reconfig() has lowered the limit
1756
1757     /* For Blu-ray compliance, don't reference frames outside of the minigop. */
1758     if( IS_X264_TYPE_B( h->fenc->i_type ) && h->param.b_bluray_compat )
1759         h->i_ref[0] = X264_MIN( h->i_ref[0], IS_X264_TYPE_B( h->fref[0][0]->i_type ) + 1 );
1760
1761     /* add duplicates */
1762     if( h->fenc->i_type == X264_TYPE_P )
1763     {
1764         int idx = -1;
1765         if( h->param.analyse.i_weighted_pred >= X264_WEIGHTP_SIMPLE )
1766         {
1767             x264_weight_t w[3];
1768             w[1].weightfn = w[2].weightfn = NULL;
1769             if( h->param.rc.b_stat_read )
1770                 x264_ratecontrol_set_weights( h, h->fenc );
1771
1772             if( !h->fenc->weight[0][0].weightfn )
1773             {
1774                 h->fenc->weight[0][0].i_denom = 0;
1775                 SET_WEIGHT( w[0], 1, 1, 0, -1 );
1776                 idx = x264_weighted_reference_duplicate( h, 0, w );
1777             }
1778             else
1779             {
1780                 if( h->fenc->weight[0][0].i_scale == 1<<h->fenc->weight[0][0].i_denom )
1781                 {
1782                     SET_WEIGHT( h->fenc->weight[0][0], 1, 1, 0, h->fenc->weight[0][0].i_offset );
1783                 }
1784                 x264_weighted_reference_duplicate( h, 0, x264_weight_none );
1785                 if( h->fenc->weight[0][0].i_offset > -128 )
1786                 {
1787                     w[0] = h->fenc->weight[0][0];
1788                     w[0].i_offset--;
1789                     h->mc.weight_cache( h, &w[0] );
1790                     idx = x264_weighted_reference_duplicate( h, 0, w );
1791                 }
1792             }
1793         }
1794         h->mb.ref_blind_dupe = idx;
1795     }
1796
1797     assert( h->i_ref[0] + h->i_ref[1] <= X264_REF_MAX );
1798     h->mb.pic.i_fref[0] = h->i_ref[0];
1799     h->mb.pic.i_fref[1] = h->i_ref[1];
1800 }
1801
1802 static void x264_fdec_filter_row( x264_t *h, int mb_y, int b_inloop )
1803 {
1804     /* mb_y is the mb to be encoded next, not the mb to be filtered here */
1805     int b_hpel = h->fdec->b_kept_as_ref;
1806     int b_deblock = h->sh.i_disable_deblocking_filter_idc != 1;
1807     int b_end = mb_y == h->i_threadslice_end;
1808     int b_measure_quality = 1;
1809     int min_y = mb_y - (1 << SLICE_MBAFF);
1810     int b_start = min_y == h->i_threadslice_start;
1811     /* Even in interlaced mode, deblocking never modifies more than 4 pixels
1812      * above each MB, as bS=4 doesn't happen for the top of interlaced mbpairs. */
1813     int minpix_y = min_y*16 - 4 * !b_start;
1814     int maxpix_y = mb_y*16 - 4 * !b_end;
1815     b_deblock &= b_hpel || h->param.psz_dump_yuv;
1816     if( h->param.b_sliced_threads && b_start && min_y && !b_inloop )
1817     {
1818         b_deblock = 0;         /* We already deblocked on the inloop pass. */
1819         b_measure_quality = 0; /* We already measured quality on the inloop pass. */
1820     }
1821     if( mb_y & SLICE_MBAFF )
1822         return;
1823     if( min_y < h->i_threadslice_start )
1824         return;
1825
1826     if( b_deblock )
1827         for( int y = min_y; y < mb_y; y += (1 << SLICE_MBAFF) )
1828             x264_frame_deblock_row( h, y );
1829
1830     /* FIXME: Prediction requires different borders for interlaced/progressive mc,
1831      * but the actual image data is equivalent. For now, maintain this
1832      * consistency by copying deblocked pixels between planes. */
1833     if( PARAM_INTERLACED )
1834         for( int p = 0; p < h->fdec->i_plane; p++ )
1835             for( int i = minpix_y>>(h->mb.chroma_v_shift && p); i < maxpix_y>>(h->mb.chroma_v_shift && p); i++ )
1836                 memcpy( h->fdec->plane_fld[p] + i*h->fdec->i_stride[p],
1837                         h->fdec->plane[p]     + i*h->fdec->i_stride[p],
1838                         h->mb.i_mb_width*16*sizeof(pixel) );
1839
1840     if( b_hpel )
1841     {
1842         int end = mb_y == h->mb.i_mb_height;
1843         x264_frame_expand_border( h, h->fdec, min_y, end );
1844         if( h->param.analyse.i_subpel_refine )
1845         {
1846             x264_frame_filter( h, h->fdec, min_y, end );
1847             x264_frame_expand_border_filtered( h, h->fdec, min_y, end );
1848         }
1849     }
1850
1851     if( SLICE_MBAFF )
1852         for( int i = 0; i < 3; i++ )
1853         {
1854             XCHG( pixel *, h->intra_border_backup[0][i], h->intra_border_backup[3][i] );
1855             XCHG( pixel *, h->intra_border_backup[1][i], h->intra_border_backup[4][i] );
1856         }
1857
1858     if( h->i_thread_frames > 1 && h->fdec->b_kept_as_ref )
1859         x264_frame_cond_broadcast( h->fdec, mb_y*16 + (b_end ? 10000 : -(X264_THREAD_HEIGHT << SLICE_MBAFF)) );
1860
1861     if( b_measure_quality )
1862     {
1863         maxpix_y = X264_MIN( maxpix_y, h->param.i_height );
1864         if( h->param.analyse.b_psnr )
1865         {
1866             for( int p = 0; p < (CHROMA444 ? 3 : 1); p++ )
1867                 h->stat.frame.i_ssd[p] += x264_pixel_ssd_wxh( &h->pixf,
1868                     h->fdec->plane[p] + minpix_y * h->fdec->i_stride[p], h->fdec->i_stride[p],
1869                     h->fenc->plane[p] + minpix_y * h->fenc->i_stride[p], h->fenc->i_stride[p],
1870                     h->param.i_width, maxpix_y-minpix_y );
1871             if( !CHROMA444 )
1872             {
1873                 uint64_t ssd_u, ssd_v;
1874                 int v_shift = h->mb.chroma_v_shift;
1875                 x264_pixel_ssd_nv12( &h->pixf,
1876                     h->fdec->plane[1] + (minpix_y>>v_shift) * h->fdec->i_stride[1], h->fdec->i_stride[1],
1877                     h->fenc->plane[1] + (minpix_y>>v_shift) * h->fenc->i_stride[1], h->fenc->i_stride[1],
1878                     h->param.i_width>>1, (maxpix_y-minpix_y)>>v_shift, &ssd_u, &ssd_v );
1879                 h->stat.frame.i_ssd[1] += ssd_u;
1880                 h->stat.frame.i_ssd[2] += ssd_v;
1881             }
1882         }
1883
1884         if( h->param.analyse.b_ssim )
1885         {
1886             int ssim_cnt;
1887             x264_emms();
1888             /* offset by 2 pixels to avoid alignment of ssim blocks with dct blocks,
1889              * and overlap by 4 */
1890             minpix_y += b_start ? 2 : -6;
1891             h->stat.frame.f_ssim +=
1892                 x264_pixel_ssim_wxh( &h->pixf,
1893                     h->fdec->plane[0] + 2+minpix_y*h->fdec->i_stride[0], h->fdec->i_stride[0],
1894                     h->fenc->plane[0] + 2+minpix_y*h->fenc->i_stride[0], h->fenc->i_stride[0],
1895                     h->param.i_width-2, maxpix_y-minpix_y, h->scratch_buffer, &ssim_cnt );
1896             h->stat.frame.i_ssim_cnt += ssim_cnt;
1897         }
1898     }
1899 }
1900
1901 static inline int x264_reference_update( x264_t *h )
1902 {
1903     if( !h->fdec->b_kept_as_ref )
1904     {
1905         if( h->i_thread_frames > 1 )
1906         {
1907             x264_frame_push_unused( h, h->fdec );
1908             h->fdec = x264_frame_pop_unused( h, 1 );
1909             if( !h->fdec )
1910                 return -1;
1911         }
1912         return 0;
1913     }
1914
1915     /* apply mmco from previous frame. */
1916     for( int i = 0; i < h->sh.i_mmco_command_count; i++ )
1917         for( int j = 0; h->frames.reference[j]; j++ )
1918             if( h->frames.reference[j]->i_poc == h->sh.mmco[i].i_poc )
1919                 x264_frame_push_unused( h, x264_frame_shift( &h->frames.reference[j] ) );
1920
1921     /* move frame in the buffer */
1922     x264_frame_push( h->frames.reference, h->fdec );
1923     if( h->frames.reference[h->sps->i_num_ref_frames] )
1924         x264_frame_push_unused( h, x264_frame_shift( h->frames.reference ) );
1925     h->fdec = x264_frame_pop_unused( h, 1 );
1926     if( !h->fdec )
1927         return -1;
1928     return 0;
1929 }
1930
1931 static inline void x264_reference_reset( x264_t *h )
1932 {
1933     while( h->frames.reference[0] )
1934         x264_frame_push_unused( h, x264_frame_pop( h->frames.reference ) );
1935     h->fdec->i_poc =
1936     h->fenc->i_poc = 0;
1937 }
1938
1939 static inline void x264_reference_hierarchy_reset( x264_t *h )
1940 {
1941     int ref;
1942     int b_hasdelayframe = 0;
1943
1944     /* look for delay frames -- chain must only contain frames that are disposable */
1945     for( int i = 0; h->frames.current[i] && IS_DISPOSABLE( h->frames.current[i]->i_type ); i++ )
1946         b_hasdelayframe |= h->frames.current[i]->i_coded
1947                         != h->frames.current[i]->i_frame + h->sps->vui.i_num_reorder_frames;
1948
1949     /* This function must handle b-pyramid and clear frames for open-gop */
1950     if( h->param.i_bframe_pyramid != X264_B_PYRAMID_STRICT && !b_hasdelayframe && h->frames.i_poc_last_open_gop == -1 )
1951         return;
1952
1953     /* Remove last BREF. There will never be old BREFs in the
1954      * dpb during a BREF decode when pyramid == STRICT */
1955     for( ref = 0; h->frames.reference[ref]; ref++ )
1956     {
1957         if( ( h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT
1958             && h->frames.reference[ref]->i_type == X264_TYPE_BREF )
1959             || ( h->frames.reference[ref]->i_poc < h->frames.i_poc_last_open_gop
1960             && h->sh.i_type != SLICE_TYPE_B ) )
1961         {
1962             int diff = h->i_frame_num - h->frames.reference[ref]->i_frame_num;
1963             h->sh.mmco[h->sh.i_mmco_command_count].i_difference_of_pic_nums = diff;
1964             h->sh.mmco[h->sh.i_mmco_command_count++].i_poc = h->frames.reference[ref]->i_poc;
1965             x264_frame_push_unused( h, x264_frame_shift( &h->frames.reference[ref] ) );
1966             h->b_ref_reorder[0] = 1;
1967             ref--;
1968         }
1969     }
1970
1971     /* Prepare room in the dpb for the delayed display time of the later b-frame's */
1972     if( h->param.i_bframe_pyramid )
1973         h->sh.i_mmco_remove_from_end = X264_MAX( ref + 2 - h->frames.i_max_dpb, 0 );
1974 }
1975
1976 static inline void x264_slice_init( x264_t *h, int i_nal_type, int i_global_qp )
1977 {
1978     /* ------------------------ Create slice header  ----------------------- */
1979     if( i_nal_type == NAL_SLICE_IDR )
1980     {
1981         x264_slice_header_init( h, &h->sh, h->sps, h->pps, h->i_idr_pic_id, h->i_frame_num, i_global_qp );
1982
1983         /* alternate id */
1984         h->i_idr_pic_id ^= 1;
1985     }
1986     else
1987     {
1988         x264_slice_header_init( h, &h->sh, h->sps, h->pps, -1, h->i_frame_num, i_global_qp );
1989
1990         h->sh.i_num_ref_idx_l0_active = h->i_ref[0] <= 0 ? 1 : h->i_ref[0];
1991         h->sh.i_num_ref_idx_l1_active = h->i_ref[1] <= 0 ? 1 : h->i_ref[1];
1992         if( h->sh.i_num_ref_idx_l0_active != h->pps->i_num_ref_idx_l0_default_active ||
1993             (h->sh.i_type == SLICE_TYPE_B && h->sh.i_num_ref_idx_l1_active != h->pps->i_num_ref_idx_l1_default_active) )
1994         {
1995             h->sh.b_num_ref_idx_override = 1;
1996         }
1997     }
1998
1999     if( h->fenc->i_type == X264_TYPE_BREF && h->param.b_bluray_compat && h->sh.i_mmco_command_count )
2000     {
2001         h->b_sh_backup = 1;
2002         h->sh_backup = h->sh;
2003     }
2004
2005     h->fdec->i_frame_num = h->sh.i_frame_num;
2006
2007     if( h->sps->i_poc_type == 0 )
2008     {
2009         h->sh.i_poc = h->fdec->i_poc;
2010         if( PARAM_INTERLACED )
2011         {
2012             h->sh.i_delta_poc_bottom = h->param.b_tff ? 1 : -1;
2013             h->sh.i_poc += h->sh.i_delta_poc_bottom == -1;
2014         }
2015         else
2016             h->sh.i_delta_poc_bottom = 0;
2017         h->fdec->i_delta_poc[0] = h->sh.i_delta_poc_bottom == -1;
2018         h->fdec->i_delta_poc[1] = h->sh.i_delta_poc_bottom ==  1;
2019     }
2020     else
2021     {
2022         /* Nothing to do ? */
2023     }
2024
2025     x264_macroblock_slice_init( h );
2026 }
2027
2028 static int x264_slice_write( x264_t *h )
2029 {
2030     int i_skip;
2031     int mb_xy, i_mb_x, i_mb_y;
2032     int i_skip_bak = 0; /* Shut up GCC. */
2033     bs_t UNINIT(bs_bak);
2034     x264_cabac_t cabac_bak;
2035     uint8_t cabac_prevbyte_bak = 0; /* Shut up GCC. */
2036     int mv_bits_bak = 0;
2037     int tex_bits_bak = 0;
2038     /* NALUs other than the first use a 3-byte startcode.
2039      * Add one extra byte for the rbsp, and one more for the final CABAC putbyte.
2040      * Then add an extra 5 bytes just in case, to account for random NAL escapes and
2041      * other inaccuracies. */
2042     int overhead_guess = (NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal)) + 1 + h->param.b_cabac + 5;
2043     int slice_max_size = h->param.i_slice_max_size > 0 ? (h->param.i_slice_max_size-overhead_guess)*8 : 0;
2044     int back_up_bitstream = slice_max_size || (!h->param.b_cabac && h->sps->i_profile_idc < PROFILE_HIGH);
2045     int starting_bits = bs_pos(&h->out.bs);
2046     int b_deblock = h->sh.i_disable_deblocking_filter_idc != 1;
2047     int b_hpel = h->fdec->b_kept_as_ref;
2048     uint8_t *last_emu_check;
2049     b_deblock &= b_hpel || h->param.psz_dump_yuv;
2050     bs_realign( &h->out.bs );
2051
2052     /* Slice */
2053     x264_nal_start( h, h->i_nal_type, h->i_nal_ref_idc );
2054     h->out.nal[h->out.i_nal].i_first_mb = h->sh.i_first_mb;
2055
2056     /* Slice header */
2057     x264_macroblock_thread_init( h );
2058
2059     /* If this isn't the first slice in the threadslice, set the slice QP
2060      * equal to the last QP in the previous slice for more accurate
2061      * CABAC initialization. */
2062     if( h->sh.i_first_mb != h->i_threadslice_start * h->mb.i_mb_width )
2063     {
2064         h->sh.i_qp = h->mb.i_last_qp;
2065         h->sh.i_qp_delta = h->sh.i_qp - h->pps->i_pic_init_qp;
2066     }
2067
2068     x264_slice_header_write( &h->out.bs, &h->sh, h->i_nal_ref_idc );
2069     if( h->param.b_cabac )
2070     {
2071         /* alignment needed */
2072         bs_align_1( &h->out.bs );
2073
2074         /* init cabac */
2075         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 );
2076         x264_cabac_encode_init ( &h->cabac, h->out.bs.p, h->out.bs.p_end );
2077         last_emu_check = h->cabac.p;
2078     }
2079     else
2080         last_emu_check = h->out.bs.p;
2081     h->mb.i_last_qp = h->sh.i_qp;
2082     h->mb.i_last_dqp = 0;
2083     h->mb.field_decoding_flag = 0;
2084
2085     i_mb_y = h->sh.i_first_mb / h->mb.i_mb_width;
2086     i_mb_x = h->sh.i_first_mb % h->mb.i_mb_width;
2087     i_skip = 0;
2088
2089     while( 1 )
2090     {
2091         mb_xy = i_mb_x + i_mb_y * h->mb.i_mb_width;
2092         int mb_spos = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
2093
2094         if( !(i_mb_y & SLICE_MBAFF) )
2095         {
2096             if( x264_bitstream_check_buffer( h ) )
2097                 return -1;
2098
2099             if( back_up_bitstream )
2100             {
2101                 mv_bits_bak = h->stat.frame.i_mv_bits;
2102                 tex_bits_bak = h->stat.frame.i_tex_bits;
2103                 /* We don't need the contexts because flushing the CABAC encoder has no context
2104                  * dependency and macroblocks are only re-encoded in the case where a slice is
2105                  * ended (and thus the content of all contexts are thrown away). */
2106                 if( h->param.b_cabac )
2107                 {
2108                     memcpy( &cabac_bak, &h->cabac, offsetof(x264_cabac_t, f8_bits_encoded) );
2109                     /* x264's CABAC writer modifies the previous byte during carry, so it has to be
2110                      * backed up. */
2111                     cabac_prevbyte_bak = h->cabac.p[-1];
2112                 }
2113                 else
2114                 {
2115                     bs_bak = h->out.bs;
2116                     i_skip_bak = i_skip;
2117                 }
2118             }
2119         }
2120
2121         if( i_mb_x == 0 && !h->mb.b_reencode_mb )
2122             x264_fdec_filter_row( h, i_mb_y, 1 );
2123
2124         if( PARAM_INTERLACED )
2125         {
2126             if( h->mb.b_adaptive_mbaff )
2127             {
2128                 if( !(i_mb_y&1) )
2129                 {
2130                     /* FIXME: VSAD is fast but fairly poor at choosing the best interlace type. */
2131                     h->mb.b_interlaced = x264_field_vsad( h, i_mb_x, i_mb_y );
2132                     memcpy( &h->zigzagf, MB_INTERLACED ? &h->zigzagf_interlaced : &h->zigzagf_progressive, sizeof(h->zigzagf) );
2133                     if( !MB_INTERLACED && (i_mb_y+2) == h->mb.i_mb_height )
2134                         x264_expand_border_mbpair( h, i_mb_x, i_mb_y );
2135                 }
2136             }
2137             h->mb.field[mb_xy] = MB_INTERLACED;
2138         }
2139
2140         /* load cache */
2141         if( SLICE_MBAFF )
2142             x264_macroblock_cache_load_interlaced( h, i_mb_x, i_mb_y );
2143         else
2144             x264_macroblock_cache_load_progressive( h, i_mb_x, i_mb_y );
2145
2146         x264_macroblock_analyse( h );
2147
2148         /* encode this macroblock -> be careful it can change the mb type to P_SKIP if needed */
2149 reencode:
2150         x264_macroblock_encode( h );
2151
2152         if( h->param.b_cabac )
2153         {
2154             if( mb_xy > h->sh.i_first_mb && !(SLICE_MBAFF && (i_mb_y&1)) )
2155                 x264_cabac_encode_terminal( &h->cabac );
2156
2157             if( IS_SKIP( h->mb.i_type ) )
2158                 x264_cabac_mb_skip( h, 1 );
2159             else
2160             {
2161                 if( h->sh.i_type != SLICE_TYPE_I )
2162                     x264_cabac_mb_skip( h, 0 );
2163                 x264_macroblock_write_cabac( h, &h->cabac );
2164             }
2165         }
2166         else
2167         {
2168             if( IS_SKIP( h->mb.i_type ) )
2169                 i_skip++;
2170             else
2171             {
2172                 if( h->sh.i_type != SLICE_TYPE_I )
2173                 {
2174                     bs_write_ue( &h->out.bs, i_skip );  /* skip run */
2175                     i_skip = 0;
2176                 }
2177                 x264_macroblock_write_cavlc( h );
2178                 /* If there was a CAVLC level code overflow, try again at a higher QP. */
2179                 if( h->mb.b_overflow )
2180                 {
2181                     h->mb.i_chroma_qp = h->chroma_qp_table[++h->mb.i_qp];
2182                     h->mb.i_skip_intra = 0;
2183                     h->mb.b_skip_mc = 0;
2184                     h->mb.b_overflow = 0;
2185                     h->out.bs = bs_bak;
2186                     i_skip = i_skip_bak;
2187                     h->stat.frame.i_mv_bits = mv_bits_bak;
2188                     h->stat.frame.i_tex_bits = tex_bits_bak;
2189                     goto reencode;
2190                 }
2191             }
2192         }
2193
2194         int total_bits = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
2195         int mb_size = total_bits - mb_spos;
2196
2197         if( slice_max_size )
2198         {
2199             /* Count the skip run, just in case. */
2200             if( !h->param.b_cabac )
2201                 total_bits += bs_size_ue_big( i_skip );
2202             /* Check for escape bytes. */
2203             uint8_t *end = h->param.b_cabac ? h->cabac.p : h->out.bs.p;
2204             for( ; last_emu_check < end - 2; last_emu_check++ )
2205                 if( last_emu_check[0] == 0 && last_emu_check[1] == 0 && last_emu_check[2] <= 3 )
2206                 {
2207                     slice_max_size -= 8;
2208                     last_emu_check++;
2209                 }
2210             /* We'll just re-encode this last macroblock if we go over the max slice size. */
2211             if( total_bits - starting_bits > slice_max_size && !h->mb.b_reencode_mb )
2212             {
2213                 if( mb_xy != h->sh.i_first_mb )
2214                 {
2215                     h->stat.frame.i_mv_bits = mv_bits_bak;
2216                     h->stat.frame.i_tex_bits = tex_bits_bak;
2217                     if( h->param.b_cabac )
2218                     {
2219                         memcpy( &h->cabac, &cabac_bak, offsetof(x264_cabac_t, f8_bits_encoded) );
2220                         h->cabac.p[-1] = cabac_prevbyte_bak;
2221                     }
2222                     else
2223                     {
2224                         h->out.bs = bs_bak;
2225                         i_skip = i_skip_bak;
2226                     }
2227                     h->mb.b_reencode_mb = 1;
2228                     if( SLICE_MBAFF )
2229                     {
2230                         // set to bottom of previous mbpair
2231                         if( i_mb_x )
2232                             h->sh.i_last_mb = mb_xy-1+h->mb.i_mb_stride*(!(i_mb_y&1));
2233                         else
2234                             h->sh.i_last_mb = (i_mb_y-2+!(i_mb_y&1))*h->mb.i_mb_stride + h->mb.i_mb_width - 1;
2235                     }
2236                     else
2237                         h->sh.i_last_mb = mb_xy-1;
2238                     break;
2239                 }
2240                 else
2241                 {
2242                     h->sh.i_last_mb = mb_xy;
2243                     h->mb.b_reencode_mb = 0;
2244                 }
2245             }
2246             else
2247                 h->mb.b_reencode_mb = 0;
2248         }
2249
2250 #if HAVE_VISUALIZE
2251         if( h->param.b_visualize )
2252             x264_visualize_mb( h );
2253 #endif
2254
2255         /* save cache */
2256         x264_macroblock_cache_save( h );
2257
2258         /* accumulate mb stats */
2259         h->stat.frame.i_mb_count[h->mb.i_type]++;
2260
2261         int b_intra = IS_INTRA( h->mb.i_type );
2262         int b_skip = IS_SKIP( h->mb.i_type );
2263         if( h->param.i_log_level >= X264_LOG_INFO || h->param.rc.b_stat_write )
2264         {
2265             if( !b_intra && !b_skip && !IS_DIRECT( h->mb.i_type ) )
2266             {
2267                 if( h->mb.i_partition != D_8x8 )
2268                         h->stat.frame.i_mb_partition[h->mb.i_partition] += 4;
2269                     else
2270                         for( int i = 0; i < 4; i++ )
2271                             h->stat.frame.i_mb_partition[h->mb.i_sub_partition[i]] ++;
2272                 if( h->param.i_frame_reference > 1 )
2273                     for( int i_list = 0; i_list <= (h->sh.i_type == SLICE_TYPE_B); i_list++ )
2274                         for( int i = 0; i < 4; i++ )
2275                         {
2276                             int i_ref = h->mb.cache.ref[i_list][ x264_scan8[4*i] ];
2277                             if( i_ref >= 0 )
2278                                 h->stat.frame.i_mb_count_ref[i_list][i_ref] ++;
2279                         }
2280             }
2281         }
2282
2283         if( h->param.i_log_level >= X264_LOG_INFO )
2284         {
2285             if( h->mb.i_cbp_luma | h->mb.i_cbp_chroma )
2286             {
2287                 if( CHROMA444 )
2288                 {
2289                     for( int i = 0; i < 4; i++ )
2290                         if( h->mb.i_cbp_luma & (1 << i) )
2291                             for( int p = 0; p < 3; p++ )
2292                             {
2293                                 int s8 = i*4+p*16;
2294                                 int nnz8x8 = M16( &h->mb.cache.non_zero_count[x264_scan8[s8]+0] )
2295                                            | M16( &h->mb.cache.non_zero_count[x264_scan8[s8]+8] );
2296                                 h->stat.frame.i_mb_cbp[!b_intra + p*2] += !!nnz8x8;
2297                             }
2298                 }
2299                 else
2300                 {
2301                     int cbpsum = (h->mb.i_cbp_luma&1) + ((h->mb.i_cbp_luma>>1)&1)
2302                                + ((h->mb.i_cbp_luma>>2)&1) + (h->mb.i_cbp_luma>>3);
2303                     h->stat.frame.i_mb_cbp[!b_intra + 0] += cbpsum;
2304                     h->stat.frame.i_mb_cbp[!b_intra + 2] += !!h->mb.i_cbp_chroma;
2305                     h->stat.frame.i_mb_cbp[!b_intra + 4] += h->mb.i_cbp_chroma >> 1;
2306                 }
2307             }
2308             if( h->mb.i_cbp_luma && !b_intra )
2309             {
2310                 h->stat.frame.i_mb_count_8x8dct[0] ++;
2311                 h->stat.frame.i_mb_count_8x8dct[1] += h->mb.b_transform_8x8;
2312             }
2313             if( b_intra && h->mb.i_type != I_PCM )
2314             {
2315                 if( h->mb.i_type == I_16x16 )
2316                     h->stat.frame.i_mb_pred_mode[0][h->mb.i_intra16x16_pred_mode]++;
2317                 else if( h->mb.i_type == I_8x8 )
2318                     for( int i = 0; i < 16; i += 4 )
2319                         h->stat.frame.i_mb_pred_mode[1][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
2320                 else //if( h->mb.i_type == I_4x4 )
2321                     for( int i = 0; i < 16; i++ )
2322                         h->stat.frame.i_mb_pred_mode[2][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
2323                 h->stat.frame.i_mb_pred_mode[3][x264_mb_chroma_pred_mode_fix[h->mb.i_chroma_pred_mode]]++;
2324             }
2325             h->stat.frame.i_mb_field[b_intra?0:b_skip?2:1] += MB_INTERLACED;
2326         }
2327
2328         /* calculate deblock strength values (actual deblocking is done per-row along with hpel) */
2329         if( b_deblock )
2330             x264_macroblock_deblock_strength( h );
2331
2332         x264_ratecontrol_mb( h, mb_size );
2333
2334         if( mb_xy == h->sh.i_last_mb )
2335             break;
2336
2337         if( SLICE_MBAFF )
2338         {
2339             i_mb_x += i_mb_y & 1;
2340             i_mb_y ^= i_mb_x < h->mb.i_mb_width;
2341         }
2342         else
2343             i_mb_x++;
2344         if( i_mb_x == h->mb.i_mb_width )
2345         {
2346             i_mb_y++;
2347             i_mb_x = 0;
2348         }
2349     }
2350     h->out.nal[h->out.i_nal].i_last_mb = h->sh.i_last_mb;
2351
2352     if( h->param.b_cabac )
2353     {
2354         x264_cabac_encode_flush( h, &h->cabac );
2355         h->out.bs.p = h->cabac.p;
2356     }
2357     else
2358     {
2359         if( i_skip > 0 )
2360             bs_write_ue( &h->out.bs, i_skip );  /* last skip run */
2361         /* rbsp_slice_trailing_bits */
2362         bs_rbsp_trailing( &h->out.bs );
2363         bs_flush( &h->out.bs );
2364     }
2365     if( x264_nal_end( h ) )
2366         return -1;
2367
2368     if( h->sh.i_last_mb == (h->i_threadslice_end * h->mb.i_mb_width - 1) )
2369     {
2370         h->stat.frame.i_misc_bits = bs_pos( &h->out.bs )
2371                                   + (h->out.i_nal*NALU_OVERHEAD * 8)
2372                                   - h->stat.frame.i_tex_bits
2373                                   - h->stat.frame.i_mv_bits;
2374         x264_fdec_filter_row( h, h->i_threadslice_end, 1 );
2375     }
2376
2377     return 0;
2378 }
2379
2380 static void x264_thread_sync_context( x264_t *dst, x264_t *src )
2381 {
2382     if( dst == src )
2383         return;
2384
2385     // reference counting
2386     for( x264_frame_t **f = src->frames.reference; *f; f++ )
2387         (*f)->i_reference_count++;
2388     for( x264_frame_t **f = dst->frames.reference; *f; f++ )
2389         x264_frame_push_unused( src, *f );
2390     src->fdec->i_reference_count++;
2391     x264_frame_push_unused( src, dst->fdec );
2392
2393     // copy everything except the per-thread pointers and the constants.
2394     memcpy( &dst->i_frame, &src->i_frame, offsetof(x264_t, mb.type) - offsetof(x264_t, i_frame) );
2395     dst->param = src->param;
2396     dst->stat = src->stat;
2397     dst->pixf = src->pixf;
2398 }
2399
2400 static void x264_thread_sync_stat( x264_t *dst, x264_t *src )
2401 {
2402     if( dst == src )
2403         return;
2404     memcpy( &dst->stat.i_frame_count, &src->stat.i_frame_count, sizeof(dst->stat) - sizeof(dst->stat.frame) );
2405 }
2406
2407 static void *x264_slices_write( x264_t *h )
2408 {
2409     int i_slice_num = 0;
2410     int last_thread_mb = h->sh.i_last_mb;
2411
2412 #if HAVE_VISUALIZE
2413     if( h->param.b_visualize )
2414         if( x264_visualize_init( h ) )
2415             return (void *)-1;
2416 #endif
2417
2418     /* init stats */
2419     memset( &h->stat.frame, 0, sizeof(h->stat.frame) );
2420     h->mb.b_reencode_mb = 0;
2421     while( h->sh.i_first_mb + SLICE_MBAFF*h->mb.i_mb_stride <= last_thread_mb )
2422     {
2423         h->sh.i_last_mb = last_thread_mb;
2424         if( h->param.i_slice_max_mbs )
2425         {
2426             if( SLICE_MBAFF )
2427             {
2428                 // convert first to mbaff form, add slice-max-mbs, then convert back to normal form
2429                 int last_mbaff = 2*(h->sh.i_first_mb % h->mb.i_mb_width)
2430                     + h->mb.i_mb_width*(h->sh.i_first_mb / h->mb.i_mb_width)
2431                     + h->param.i_slice_max_mbs - 1;
2432                 int last_x = (last_mbaff % (2*h->mb.i_mb_width))/2;
2433                 int last_y = (last_mbaff / (2*h->mb.i_mb_width))*2 + 1;
2434                 h->sh.i_last_mb = last_x + h->mb.i_mb_stride*last_y;
2435             }
2436             else
2437                 h->sh.i_last_mb = h->sh.i_first_mb + h->param.i_slice_max_mbs - 1;
2438         }
2439         else if( h->param.i_slice_count && !h->param.b_sliced_threads )
2440         {
2441             int height = h->mb.i_mb_height >> PARAM_INTERLACED;
2442             int width = h->mb.i_mb_width << PARAM_INTERLACED;
2443             i_slice_num++;
2444             h->sh.i_last_mb = (height * i_slice_num + h->param.i_slice_count/2) / h->param.i_slice_count * width - 1;
2445         }
2446         h->sh.i_last_mb = X264_MIN( h->sh.i_last_mb, last_thread_mb );
2447         if( x264_stack_align( x264_slice_write, h ) )
2448             return (void *)-1;
2449         h->sh.i_first_mb = h->sh.i_last_mb + 1;
2450         // if i_first_mb is not the last mb in a row then go to the next mb in MBAFF order
2451         if( SLICE_MBAFF && h->sh.i_first_mb % h->mb.i_mb_width )
2452             h->sh.i_first_mb -= h->mb.i_mb_stride;
2453     }
2454
2455 #if HAVE_VISUALIZE
2456     if( h->param.b_visualize )
2457     {
2458         x264_visualize_show( h );
2459         x264_visualize_close( h );
2460     }
2461 #endif
2462
2463     return (void *)0;
2464 }
2465
2466 static int x264_threaded_slices_write( x264_t *h )
2467 {
2468     /* set first/last mb and sync contexts */
2469     for( int i = 0; i < h->param.i_threads; i++ )
2470     {
2471         x264_t *t = h->thread[i];
2472         if( i )
2473         {
2474             t->param = h->param;
2475             memcpy( &t->i_frame, &h->i_frame, offsetof(x264_t, rc) - offsetof(x264_t, i_frame) );
2476         }
2477         int height = h->mb.i_mb_height >> PARAM_INTERLACED;
2478         t->i_threadslice_start = ((height *  i    + h->param.i_slice_count/2) / h->param.i_threads) << PARAM_INTERLACED;
2479         t->i_threadslice_end   = ((height * (i+1) + h->param.i_slice_count/2) / h->param.i_threads) << PARAM_INTERLACED;
2480         t->sh.i_first_mb = t->i_threadslice_start * h->mb.i_mb_width;
2481         t->sh.i_last_mb  =   t->i_threadslice_end * h->mb.i_mb_width - 1;
2482     }
2483
2484     x264_stack_align( x264_analyse_weight_frame, h, h->mb.i_mb_height*16 + 16 );
2485
2486     x264_threads_distribute_ratecontrol( h );
2487
2488     /* dispatch */
2489     for( int i = 0; i < h->param.i_threads; i++ )
2490     {
2491         x264_threadpool_run( h->threadpool, (void*)x264_slices_write, h->thread[i] );
2492         h->thread[i]->b_thread_active = 1;
2493     }
2494     for( int i = 0; i < h->param.i_threads; i++ )
2495     {
2496         h->thread[i]->b_thread_active = 0;
2497         if( (intptr_t)x264_threadpool_wait( h->threadpool, h->thread[i] ) )
2498             return -1;
2499     }
2500
2501     /* Go back and fix up the hpel on the borders between slices. */
2502     for( int i = 1; i < h->param.i_threads; i++ )
2503     {
2504         x264_fdec_filter_row( h->thread[i], h->thread[i]->i_threadslice_start + 1, 0 );
2505         if( SLICE_MBAFF )
2506             x264_fdec_filter_row( h->thread[i], h->thread[i]->i_threadslice_start + 2, 0 );
2507     }
2508
2509     x264_threads_merge_ratecontrol( h );
2510
2511     for( int i = 1; i < h->param.i_threads; i++ )
2512     {
2513         x264_t *t = h->thread[i];
2514         for( int j = 0; j < t->out.i_nal; j++ )
2515         {
2516             h->out.nal[h->out.i_nal] = t->out.nal[j];
2517             h->out.i_nal++;
2518             x264_nal_check_buffer( h );
2519         }
2520         /* All entries in stat.frame are ints except for ssd/ssim. */
2521         for( int j = 0; j < (offsetof(x264_t,stat.frame.i_ssd) - offsetof(x264_t,stat.frame.i_mv_bits)) / sizeof(int); j++ )
2522             ((int*)&h->stat.frame)[j] += ((int*)&t->stat.frame)[j];
2523         for( int j = 0; j < 3; j++ )
2524             h->stat.frame.i_ssd[j] += t->stat.frame.i_ssd[j];
2525         h->stat.frame.f_ssim += t->stat.frame.f_ssim;
2526         h->stat.frame.i_ssim_cnt += t->stat.frame.i_ssim_cnt;
2527     }
2528
2529     return 0;
2530 }
2531
2532 void x264_encoder_intra_refresh( x264_t *h )
2533 {
2534     h = h->thread[h->i_thread_phase];
2535     h->b_queued_intra_refresh = 1;
2536 }
2537
2538 int x264_encoder_invalidate_reference( x264_t *h, int64_t pts )
2539 {
2540     if( h->param.i_bframe )
2541     {
2542         x264_log( h, X264_LOG_ERROR, "x264_encoder_invalidate_reference is not supported with B-frames enabled\n" );
2543         return -1;
2544     }
2545     if( h->param.b_intra_refresh )
2546     {
2547         x264_log( h, X264_LOG_ERROR, "x264_encoder_invalidate_reference is not supported with intra refresh enabled\n" );
2548         return -1;
2549     }
2550     h = h->thread[h->i_thread_phase];
2551     if( pts >= h->i_last_idr_pts )
2552     {
2553         for( int i = 0; h->frames.reference[i]; i++ )
2554             if( pts <= h->frames.reference[i]->i_pts )
2555                 h->frames.reference[i]->b_corrupt = 1;
2556         if( pts <= h->fdec->i_pts )
2557             h->fdec->b_corrupt = 1;
2558     }
2559     return 0;
2560 }
2561
2562 /****************************************************************************
2563  * x264_encoder_encode:
2564  *  XXX: i_poc   : is the poc of the current given picture
2565  *       i_frame : is the number of the frame being coded
2566  *  ex:  type frame poc
2567  *       I      0   2*0
2568  *       P      1   2*3
2569  *       B      2   2*1
2570  *       B      3   2*2
2571  *       P      4   2*6
2572  *       B      5   2*4
2573  *       B      6   2*5
2574  ****************************************************************************/
2575 int     x264_encoder_encode( x264_t *h,
2576                              x264_nal_t **pp_nal, int *pi_nal,
2577                              x264_picture_t *pic_in,
2578                              x264_picture_t *pic_out )
2579 {
2580     x264_t *thread_current, *thread_prev, *thread_oldest;
2581     int i_nal_type, i_nal_ref_idc, i_global_qp;
2582     int overhead = NALU_OVERHEAD;
2583
2584     if( h->i_thread_frames > 1 )
2585     {
2586         thread_prev    = h->thread[ h->i_thread_phase ];
2587         h->i_thread_phase = (h->i_thread_phase + 1) % h->i_thread_frames;
2588         thread_current = h->thread[ h->i_thread_phase ];
2589         thread_oldest  = h->thread[ (h->i_thread_phase + 1) % h->i_thread_frames ];
2590         x264_thread_sync_context( thread_current, thread_prev );
2591         x264_thread_sync_ratecontrol( thread_current, thread_prev, thread_oldest );
2592         h = thread_current;
2593     }
2594     else
2595     {
2596         thread_current =
2597         thread_oldest  = h;
2598     }
2599 #if HAVE_MMX
2600     if( h->param.cpu&X264_CPU_SSE_MISALIGN )
2601         x264_cpu_mask_misalign_sse();
2602 #endif
2603
2604     // ok to call this before encoding any frames, since the initial values of fdec have b_kept_as_ref=0
2605     if( x264_reference_update( h ) )
2606         return -1;
2607     h->fdec->i_lines_completed = -1;
2608
2609     /* no data out */
2610     *pi_nal = 0;
2611     *pp_nal = NULL;
2612
2613     /* ------------------- Setup new frame from picture -------------------- */
2614     if( pic_in != NULL )
2615     {
2616         /* 1: Copy the picture to a frame and move it to a buffer */
2617         x264_frame_t *fenc = x264_frame_pop_unused( h, 0 );
2618         if( !fenc )
2619             return -1;
2620
2621         if( x264_frame_copy_picture( h, fenc, pic_in ) < 0 )
2622             return -1;
2623
2624         if( h->param.i_width != 16 * h->mb.i_mb_width ||
2625             h->param.i_height != 16 * h->mb.i_mb_height )
2626             x264_frame_expand_border_mod16( h, fenc );
2627
2628         fenc->i_frame = h->frames.i_input++;
2629
2630         if( fenc->i_frame == 0 )
2631             h->frames.i_first_pts = fenc->i_pts;
2632         if( h->frames.i_bframe_delay && fenc->i_frame == h->frames.i_bframe_delay )
2633             h->frames.i_bframe_delay_time = fenc->i_pts - h->frames.i_first_pts;
2634
2635         if( h->param.b_vfr_input && fenc->i_pts <= h->frames.i_largest_pts )
2636             x264_log( h, X264_LOG_WARNING, "non-strictly-monotonic PTS\n" );
2637
2638         h->frames.i_second_largest_pts = h->frames.i_largest_pts;
2639         h->frames.i_largest_pts = fenc->i_pts;
2640
2641         if( (fenc->i_pic_struct < PIC_STRUCT_AUTO) || (fenc->i_pic_struct > PIC_STRUCT_TRIPLE) )
2642             fenc->i_pic_struct = PIC_STRUCT_AUTO;
2643
2644         if( fenc->i_pic_struct == PIC_STRUCT_AUTO )
2645         {
2646 #if HAVE_INTERLACED
2647             int b_interlaced = fenc->param ? fenc->param->b_interlaced : h->param.b_interlaced;
2648 #else
2649             int b_interlaced = 0;
2650 #endif
2651             if( b_interlaced )
2652             {
2653                 int b_tff = fenc->param ? fenc->param->b_tff : h->param.b_tff;
2654                 fenc->i_pic_struct = b_tff ? PIC_STRUCT_TOP_BOTTOM : PIC_STRUCT_BOTTOM_TOP;
2655             }
2656             else
2657                 fenc->i_pic_struct = PIC_STRUCT_PROGRESSIVE;
2658         }
2659
2660         if( h->param.rc.b_mb_tree && h->param.rc.b_stat_read )
2661         {
2662             if( x264_macroblock_tree_read( h, fenc, pic_in->prop.quant_offsets ) )
2663                 return -1;
2664         }
2665         else
2666             x264_stack_align( x264_adaptive_quant_frame, h, fenc, pic_in->prop.quant_offsets );
2667
2668         if( pic_in->prop.quant_offsets_free )
2669             pic_in->prop.quant_offsets_free( pic_in->prop.quant_offsets );
2670
2671         if( h->frames.b_have_lowres )
2672             x264_frame_init_lowres( h, fenc );
2673
2674         /* 2: Place the frame into the queue for its slice type decision */
2675         x264_lookahead_put_frame( h, fenc );
2676
2677         if( h->frames.i_input <= h->frames.i_delay + 1 - h->i_thread_frames )
2678         {
2679             /* Nothing yet to encode, waiting for filling of buffers */
2680             pic_out->i_type = X264_TYPE_AUTO;
2681             return 0;
2682         }
2683     }
2684     else
2685     {
2686         /* signal kills for lookahead thread */
2687         x264_pthread_mutex_lock( &h->lookahead->ifbuf.mutex );
2688         h->lookahead->b_exit_thread = 1;
2689         x264_pthread_cond_broadcast( &h->lookahead->ifbuf.cv_fill );
2690         x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
2691     }
2692
2693     h->i_frame++;
2694     /* 3: The picture is analyzed in the lookahead */
2695     if( !h->frames.current[0] )
2696         x264_lookahead_get_frames( h );
2697
2698     if( !h->frames.current[0] && x264_lookahead_is_empty( h ) )
2699         return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
2700
2701     /* ------------------- Get frame to be encoded ------------------------- */
2702     /* 4: get picture to encode */
2703     h->fenc = x264_frame_shift( h->frames.current );
2704     if( h->i_frame == h->i_thread_frames - 1 )
2705         h->i_reordered_pts_delay = h->fenc->i_reordered_pts;
2706     if( h->fenc->param )
2707     {
2708         x264_encoder_reconfig( h, h->fenc->param );
2709         if( h->fenc->param->param_free )
2710             h->fenc->param->param_free( h->fenc->param );
2711     }
2712
2713     if( !IS_X264_TYPE_I( h->fenc->i_type ) )
2714     {
2715         int valid_refs_left = 0;
2716         for( int i = 0; h->frames.reference[i]; i++ )
2717             if( !h->frames.reference[i]->b_corrupt )
2718                 valid_refs_left++;
2719         /* No valid reference frames left: force an IDR. */
2720         if( !valid_refs_left )
2721         {
2722             h->fenc->b_keyframe = 1;
2723             h->fenc->i_type = X264_TYPE_IDR;
2724         }
2725     }
2726
2727     if( h->fenc->b_keyframe )
2728     {
2729         h->frames.i_last_keyframe = h->fenc->i_frame;
2730         if( h->fenc->i_type == X264_TYPE_IDR )
2731         {
2732             h->i_frame_num = 0;
2733             h->frames.i_last_idr = h->fenc->i_frame;
2734         }
2735     }
2736     h->sh.i_mmco_command_count =
2737     h->sh.i_mmco_remove_from_end = 0;
2738     h->b_ref_reorder[0] =
2739     h->b_ref_reorder[1] = 0;
2740     h->fdec->i_poc =
2741     h->fenc->i_poc = 2 * ( h->fenc->i_frame - X264_MAX( h->frames.i_last_idr, 0 ) );
2742
2743     /* ------------------- Setup frame context ----------------------------- */
2744     /* 5: Init data dependent of frame type */
2745     if( h->fenc->i_type == X264_TYPE_IDR )
2746     {
2747         /* reset ref pictures */
2748         i_nal_type    = NAL_SLICE_IDR;
2749         i_nal_ref_idc = NAL_PRIORITY_HIGHEST;
2750         h->sh.i_type = SLICE_TYPE_I;
2751         x264_reference_reset( h );
2752         h->frames.i_poc_last_open_gop = -1;
2753     }
2754     else if( h->fenc->i_type == X264_TYPE_I )
2755     {
2756         i_nal_type    = NAL_SLICE;
2757         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
2758         h->sh.i_type = SLICE_TYPE_I;
2759         x264_reference_hierarchy_reset( h );
2760         if( h->param.b_open_gop )
2761             h->frames.i_poc_last_open_gop = h->fenc->b_keyframe ? h->fenc->i_poc : -1;
2762     }
2763     else if( h->fenc->i_type == X264_TYPE_P )
2764     {
2765         i_nal_type    = NAL_SLICE;
2766         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
2767         h->sh.i_type = SLICE_TYPE_P;
2768         x264_reference_hierarchy_reset( h );
2769         h->frames.i_poc_last_open_gop = -1;
2770     }
2771     else if( h->fenc->i_type == X264_TYPE_BREF )
2772     {
2773         i_nal_type    = NAL_SLICE;
2774         i_nal_ref_idc = h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT ? NAL_PRIORITY_LOW : NAL_PRIORITY_HIGH;
2775         h->sh.i_type = SLICE_TYPE_B;
2776         x264_reference_hierarchy_reset( h );
2777     }
2778     else    /* B frame */
2779     {
2780         i_nal_type    = NAL_SLICE;
2781         i_nal_ref_idc = NAL_PRIORITY_DISPOSABLE;
2782         h->sh.i_type = SLICE_TYPE_B;
2783     }
2784
2785     h->fdec->i_type = h->fenc->i_type;
2786     h->fdec->i_frame = h->fenc->i_frame;
2787     h->fenc->b_kept_as_ref =
2788     h->fdec->b_kept_as_ref = i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE && h->param.i_keyint_max > 1;
2789
2790     h->fdec->i_pts = h->fenc->i_pts;
2791     if( h->frames.i_bframe_delay )
2792     {
2793         int64_t *prev_reordered_pts = thread_current->frames.i_prev_reordered_pts;
2794         h->fdec->i_dts = h->i_frame > h->frames.i_bframe_delay
2795                        ? prev_reordered_pts[ (h->i_frame - h->frames.i_bframe_delay) % h->frames.i_bframe_delay ]
2796                        : h->fenc->i_reordered_pts - h->frames.i_bframe_delay_time;
2797         prev_reordered_pts[ h->i_frame % h->frames.i_bframe_delay ] = h->fenc->i_reordered_pts;
2798     }
2799     else
2800         h->fdec->i_dts = h->fenc->i_reordered_pts;
2801     if( h->fenc->i_type == X264_TYPE_IDR )
2802         h->i_last_idr_pts = h->fdec->i_pts;
2803
2804     /* ------------------- Init                ----------------------------- */
2805     /* build ref list 0/1 */
2806     x264_reference_build_list( h, h->fdec->i_poc );
2807
2808     /* ---------------------- Write the bitstream -------------------------- */
2809     /* Init bitstream context */
2810     if( h->param.b_sliced_threads )
2811     {
2812         for( int i = 0; i < h->param.i_threads; i++ )
2813         {
2814             bs_init( &h->thread[i]->out.bs, h->thread[i]->out.p_bitstream, h->thread[i]->out.i_bitstream );
2815             h->thread[i]->out.i_nal = 0;
2816         }
2817     }
2818     else
2819     {
2820         bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
2821         h->out.i_nal = 0;
2822     }
2823
2824     if( h->param.b_aud )
2825     {
2826         int pic_type;
2827
2828         if( h->sh.i_type == SLICE_TYPE_I )
2829             pic_type = 0;
2830         else if( h->sh.i_type == SLICE_TYPE_P )
2831             pic_type = 1;
2832         else if( h->sh.i_type == SLICE_TYPE_B )
2833             pic_type = 2;
2834         else
2835             pic_type = 7;
2836
2837         x264_nal_start( h, NAL_AUD, NAL_PRIORITY_DISPOSABLE );
2838         bs_write( &h->out.bs, 3, pic_type );
2839         bs_rbsp_trailing( &h->out.bs );
2840         if( x264_nal_end( h ) )
2841             return -1;
2842         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2843     }
2844
2845     h->i_nal_type = i_nal_type;
2846     h->i_nal_ref_idc = i_nal_ref_idc;
2847
2848     if( h->param.b_intra_refresh )
2849     {
2850         if( IS_X264_TYPE_I( h->fenc->i_type ) )
2851         {
2852             h->fdec->i_frames_since_pir = 0;
2853             h->b_queued_intra_refresh = 0;
2854             /* PIR is currently only supported with ref == 1, so any intra frame effectively refreshes
2855              * the whole frame and counts as an intra refresh. */
2856             h->fdec->f_pir_position = h->mb.i_mb_width;
2857         }
2858         else if( h->fenc->i_type == X264_TYPE_P )
2859         {
2860             int pocdiff = (h->fdec->i_poc - h->fref[0][0]->i_poc)/2;
2861             float increment = X264_MAX( ((float)h->mb.i_mb_width-1) / h->param.i_keyint_max, 1 );
2862             h->fdec->f_pir_position = h->fref[0][0]->f_pir_position;
2863             h->fdec->i_frames_since_pir = h->fref[0][0]->i_frames_since_pir + pocdiff;
2864             if( h->fdec->i_frames_since_pir >= h->param.i_keyint_max ||
2865                 (h->b_queued_intra_refresh && h->fdec->f_pir_position + 0.5 >= h->mb.i_mb_width) )
2866             {
2867                 h->fdec->f_pir_position = 0;
2868                 h->fdec->i_frames_since_pir = 0;
2869                 h->b_queued_intra_refresh = 0;
2870                 h->fenc->b_keyframe = 1;
2871             }
2872             h->fdec->i_pir_start_col = h->fdec->f_pir_position+0.5;
2873             h->fdec->f_pir_position += increment * pocdiff;
2874             h->fdec->i_pir_end_col = h->fdec->f_pir_position+0.5;
2875             /* If our intra refresh has reached the right side of the frame, we're done. */
2876             if( h->fdec->i_pir_end_col >= h->mb.i_mb_width - 1 )
2877                 h->fdec->f_pir_position = h->mb.i_mb_width;
2878         }
2879     }
2880
2881     if( h->fenc->b_keyframe )
2882     {
2883         /* Write SPS and PPS */
2884         if( h->param.b_repeat_headers )
2885         {
2886             /* generate sequence parameters */
2887             x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
2888             x264_sps_write( &h->out.bs, h->sps );
2889             if( x264_nal_end( h ) )
2890                 return -1;
2891             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
2892
2893             /* generate picture parameters */
2894             x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
2895             x264_pps_write( &h->out.bs, h->sps, h->pps );
2896             if( x264_nal_end( h ) )
2897                 return -1;
2898             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
2899         }
2900
2901         /* when frame threading is used, buffering period sei is written in x264_encoder_frame_end */
2902         if( h->i_thread_frames == 1 && h->sps->vui.b_nal_hrd_parameters_present )
2903         {
2904             x264_hrd_fullness( h );
2905             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2906             x264_sei_buffering_period_write( h, &h->out.bs );
2907             if( x264_nal_end( h ) )
2908                return -1;
2909             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
2910         }
2911     }
2912
2913     /* write extra sei */
2914     for( int i = 0; i < h->fenc->extra_sei.num_payloads; i++ )
2915     {
2916         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2917         x264_sei_write( &h->out.bs, h->fenc->extra_sei.payloads[i].payload, h->fenc->extra_sei.payloads[i].payload_size,
2918                         h->fenc->extra_sei.payloads[i].payload_type );
2919         if( x264_nal_end( h ) )
2920             return -1;
2921         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2922         if( h->fenc->extra_sei.sei_free && h->fenc->extra_sei.payloads[i].payload )
2923             h->fenc->extra_sei.sei_free( h->fenc->extra_sei.payloads[i].payload );
2924     }
2925
2926     if( h->fenc->extra_sei.sei_free && h->fenc->extra_sei.payloads )
2927         h->fenc->extra_sei.sei_free( h->fenc->extra_sei.payloads );
2928
2929     if( h->fenc->b_keyframe )
2930     {
2931         if( h->param.b_repeat_headers && h->fenc->i_frame == 0 )
2932         {
2933             /* identify ourself */
2934             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2935             if( x264_sei_version_write( h, &h->out.bs ) )
2936                 return -1;
2937             if( x264_nal_end( h ) )
2938                 return -1;
2939             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2940         }
2941
2942         if( h->fenc->i_type != X264_TYPE_IDR )
2943         {
2944             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;
2945             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2946             x264_sei_recovery_point_write( h, &h->out.bs, time_to_recovery );
2947             if( x264_nal_end( h ) )
2948                 return -1;
2949             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2950         }
2951
2952         if ( h->param.i_frame_packing >= 0 )
2953         {
2954             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2955             x264_sei_frame_packing_write( h, &h->out.bs );
2956             if( x264_nal_end( h ) )
2957                 return -1;
2958             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2959         }
2960     }
2961
2962     /* generate sei pic timing */
2963     if( h->sps->vui.b_pic_struct_present || h->sps->vui.b_nal_hrd_parameters_present )
2964     {
2965         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2966         x264_sei_pic_timing_write( h, &h->out.bs );
2967         if( x264_nal_end( h ) )
2968             return -1;
2969         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2970     }
2971
2972     /* As required by Blu-ray. */
2973     if( !IS_X264_TYPE_B( h->fenc->i_type ) && h->b_sh_backup )
2974     {
2975         h->b_sh_backup = 0;
2976         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2977         x264_sei_dec_ref_pic_marking_write( h, &h->out.bs );
2978         if( x264_nal_end( h ) )
2979             return -1;
2980         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2981     }
2982
2983     if( h->fenc->b_keyframe && h->param.b_intra_refresh )
2984         h->i_cpb_delay_pir_offset = h->fenc->i_cpb_delay;
2985
2986     /* Init the rate control */
2987     /* FIXME: Include slice header bit cost. */
2988     x264_ratecontrol_start( h, h->fenc->i_qpplus1, overhead*8 );
2989     i_global_qp = x264_ratecontrol_qp( h );
2990
2991     pic_out->i_qpplus1 =
2992     h->fdec->i_qpplus1 = i_global_qp + 1;
2993
2994     if( h->param.rc.b_stat_read && h->sh.i_type != SLICE_TYPE_I )
2995     {
2996         x264_reference_build_list_optimal( h );
2997         x264_reference_check_reorder( h );
2998     }
2999
3000     if( h->i_ref[0] )
3001         h->fdec->i_poc_l0ref0 = h->fref[0][0]->i_poc;
3002
3003     /* ------------------------ Create slice header  ----------------------- */
3004     x264_slice_init( h, i_nal_type, i_global_qp );
3005
3006     /*------------------------- Weights -------------------------------------*/
3007     if( h->sh.i_type == SLICE_TYPE_B )
3008         x264_macroblock_bipred_init( h );
3009
3010     x264_weighted_pred_init( h );
3011
3012     if( i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE )
3013         h->i_frame_num++;
3014
3015     /* Write frame */
3016     h->i_threadslice_start = 0;
3017     h->i_threadslice_end = h->mb.i_mb_height;
3018     if( h->i_thread_frames > 1 )
3019     {
3020         x264_threadpool_run( h->threadpool, (void*)x264_slices_write, h );
3021         h->b_thread_active = 1;
3022     }
3023     else if( h->param.b_sliced_threads )
3024     {
3025         if( x264_threaded_slices_write( h ) )
3026             return -1;
3027     }
3028     else
3029         if( (intptr_t)x264_slices_write( h ) )
3030             return -1;
3031
3032     return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
3033 }
3034
3035 static int x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
3036                                    x264_nal_t **pp_nal, int *pi_nal,
3037                                    x264_picture_t *pic_out )
3038 {
3039     char psz_message[80];
3040
3041     if( h->b_thread_active )
3042     {
3043         h->b_thread_active = 0;
3044         if( (intptr_t)x264_threadpool_wait( h->threadpool, h ) )
3045             return -1;
3046     }
3047     if( !h->out.i_nal )
3048     {
3049         pic_out->i_type = X264_TYPE_AUTO;
3050         return 0;
3051     }
3052
3053     x264_emms();
3054     /* generate buffering period sei and insert it into place */
3055     if( h->i_thread_frames > 1 && h->fenc->b_keyframe && h->sps->vui.b_nal_hrd_parameters_present )
3056     {
3057         x264_hrd_fullness( h );
3058         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3059         x264_sei_buffering_period_write( h, &h->out.bs );
3060         if( x264_nal_end( h ) )
3061            return -1;
3062         /* buffering period sei must follow AUD, SPS and PPS and precede all other SEIs */
3063         int idx = 0;
3064         while( h->out.nal[idx].i_type == NAL_AUD ||
3065                h->out.nal[idx].i_type == NAL_SPS ||
3066                h->out.nal[idx].i_type == NAL_PPS )
3067             idx++;
3068         x264_nal_t nal_tmp = h->out.nal[h->out.i_nal-1];
3069         memmove( &h->out.nal[idx+1], &h->out.nal[idx], (h->out.i_nal-idx-1)*sizeof(x264_nal_t) );
3070         h->out.nal[idx] = nal_tmp;
3071     }
3072
3073     int frame_size = x264_encoder_encapsulate_nals( h, 0 );
3074     if( frame_size < 0 )
3075         return -1;
3076
3077     /* Set output picture properties */
3078     pic_out->i_type = h->fenc->i_type;
3079
3080     pic_out->b_keyframe = h->fenc->b_keyframe;
3081     pic_out->i_pic_struct = h->fenc->i_pic_struct;
3082
3083     pic_out->i_pts = h->fdec->i_pts;
3084     pic_out->i_dts = h->fdec->i_dts;
3085
3086     if( pic_out->i_pts < pic_out->i_dts )
3087         x264_log( h, X264_LOG_WARNING, "invalid DTS: PTS is less than DTS\n" );
3088
3089     pic_out->img.i_csp = h->fdec->i_csp;
3090 #if HIGH_BIT_DEPTH
3091     pic_out->img.i_csp |= X264_CSP_HIGH_DEPTH;
3092 #endif
3093     pic_out->img.i_plane = h->fdec->i_plane;
3094     for( int i = 0; i < pic_out->img.i_plane; i++ )
3095     {
3096         pic_out->img.i_stride[i] = h->fdec->i_stride[i] * sizeof(pixel);
3097         pic_out->img.plane[i] = (uint8_t*)h->fdec->plane[i];
3098     }
3099
3100     x264_frame_push_unused( thread_current, h->fenc );
3101
3102     /* ---------------------- Update encoder state ------------------------- */
3103
3104     /* update rc */
3105     int filler = 0;
3106     if( x264_ratecontrol_end( h, frame_size * 8, &filler ) < 0 )
3107         return -1;
3108
3109     pic_out->hrd_timing = h->fenc->hrd_timing;
3110
3111     while( filler > 0 )
3112     {
3113         int f, overhead;
3114         overhead = (FILLER_OVERHEAD - h->param.b_annexb);
3115         if( h->param.i_slice_max_size && filler > h->param.i_slice_max_size )
3116         {
3117             int next_size = filler - h->param.i_slice_max_size;
3118             int overflow = X264_MAX( overhead - next_size, 0 );
3119             f = h->param.i_slice_max_size - overhead - overflow;
3120         }
3121         else
3122             f = X264_MAX( 0, filler - overhead );
3123
3124         x264_nal_start( h, NAL_FILLER, NAL_PRIORITY_DISPOSABLE );
3125         x264_filler_write( h, &h->out.bs, f );
3126         if( x264_nal_end( h ) )
3127             return -1;
3128         int total_size = x264_encoder_encapsulate_nals( h, h->out.i_nal-1 );
3129         if( total_size < 0 )
3130             return -1;
3131         frame_size += total_size;
3132         filler -= total_size;
3133     }
3134
3135     /* End bitstream, set output  */
3136     *pi_nal = h->out.i_nal;
3137     *pp_nal = h->out.nal;
3138
3139     h->out.i_nal = 0;
3140
3141     x264_noise_reduction_update( h );
3142
3143     /* ---------------------- Compute/Print statistics --------------------- */
3144     x264_thread_sync_stat( h, h->thread[0] );
3145
3146     /* Slice stat */
3147     h->stat.i_frame_count[h->sh.i_type]++;
3148     h->stat.i_frame_size[h->sh.i_type] += frame_size;
3149     h->stat.f_frame_qp[h->sh.i_type] += h->fdec->f_qp_avg_aq;
3150
3151     for( int i = 0; i < X264_MBTYPE_MAX; i++ )
3152         h->stat.i_mb_count[h->sh.i_type][i] += h->stat.frame.i_mb_count[i];
3153     for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
3154         h->stat.i_mb_partition[h->sh.i_type][i] += h->stat.frame.i_mb_partition[i];
3155     for( int i = 0; i < 2; i++ )
3156         h->stat.i_mb_count_8x8dct[i] += h->stat.frame.i_mb_count_8x8dct[i];
3157     for( int i = 0; i < 6; i++ )
3158         h->stat.i_mb_cbp[i] += h->stat.frame.i_mb_cbp[i];
3159     for( int i = 0; i < 4; i++ )
3160         for( int j = 0; j < 13; j++ )
3161             h->stat.i_mb_pred_mode[i][j] += h->stat.frame.i_mb_pred_mode[i][j];
3162     if( h->sh.i_type != SLICE_TYPE_I )
3163         for( int i_list = 0; i_list < 2; i_list++ )
3164             for( int i = 0; i < X264_REF_MAX*2; i++ )
3165                 h->stat.i_mb_count_ref[h->sh.i_type][i_list][i] += h->stat.frame.i_mb_count_ref[i_list][i];
3166     for( int i = 0; i < 3; i++ )
3167         h->stat.i_mb_field[i] += h->stat.frame.i_mb_field[i];
3168     if( h->sh.i_type == SLICE_TYPE_P && h->param.analyse.i_weighted_pred >= X264_WEIGHTP_SIMPLE )
3169     {
3170         h->stat.i_wpred[0] += !!h->sh.weight[0][0].weightfn;
3171         h->stat.i_wpred[1] += !!h->sh.weight[0][1].weightfn || !!h->sh.weight[0][2].weightfn;
3172     }
3173     if( h->sh.i_type == SLICE_TYPE_B )
3174     {
3175         h->stat.i_direct_frames[ h->sh.b_direct_spatial_mv_pred ] ++;
3176         if( h->mb.b_direct_auto_write )
3177         {
3178             //FIXME somewhat arbitrary time constants
3179             if( h->stat.i_direct_score[0] + h->stat.i_direct_score[1] > h->mb.i_mb_count )
3180                 for( int i = 0; i < 2; i++ )
3181                     h->stat.i_direct_score[i] = h->stat.i_direct_score[i] * 9/10;
3182             for( int i = 0; i < 2; i++ )
3183                 h->stat.i_direct_score[i] += h->stat.frame.i_direct_score[i];
3184         }
3185     }
3186     else
3187         h->stat.i_consecutive_bframes[h->fenc->i_bframes]++;
3188
3189     psz_message[0] = '\0';
3190     double dur = h->fenc->f_duration;
3191     h->stat.f_frame_duration[h->sh.i_type] += dur;
3192     if( h->param.analyse.b_psnr )
3193     {
3194         int64_t ssd[3] =
3195         {
3196             h->stat.frame.i_ssd[0],
3197             h->stat.frame.i_ssd[1],
3198             h->stat.frame.i_ssd[2],
3199         };
3200         int luma_size = h->param.i_width * h->param.i_height;
3201         int chroma_size = CHROMA_SIZE( luma_size );
3202         double psnr_y = x264_psnr( ssd[0], luma_size );
3203         double psnr_u = x264_psnr( ssd[1], chroma_size );
3204         double psnr_v = x264_psnr( ssd[2], chroma_size );
3205
3206         h->stat.f_ssd_global[h->sh.i_type]   += dur * (ssd[0] + ssd[1] + ssd[2]);
3207         h->stat.f_psnr_average[h->sh.i_type] += dur * x264_psnr( ssd[0] + ssd[1] + ssd[2], luma_size + chroma_size*2 );
3208         h->stat.f_psnr_mean_y[h->sh.i_type]  += dur * psnr_y;
3209         h->stat.f_psnr_mean_u[h->sh.i_type]  += dur * psnr_u;
3210         h->stat.f_psnr_mean_v[h->sh.i_type]  += dur * psnr_v;
3211
3212         snprintf( psz_message, 80, " PSNR Y:%5.2f U:%5.2f V:%5.2f", psnr_y, psnr_u, psnr_v );
3213     }
3214
3215     if( h->param.analyse.b_ssim )
3216     {
3217         double ssim_y = h->stat.frame.f_ssim
3218                       / h->stat.frame.i_ssim_cnt;
3219         h->stat.f_ssim_mean_y[h->sh.i_type] += ssim_y * dur;
3220         snprintf( psz_message + strlen(psz_message), 80 - strlen(psz_message),
3221                   " SSIM Y:%.5f", ssim_y );
3222     }
3223     psz_message[79] = '\0';
3224
3225     x264_log( h, X264_LOG_DEBUG,
3226                   "frame=%4d QP=%.2f NAL=%d Slice:%c Poc:%-3d I:%-4d P:%-4d SKIP:%-4d size=%d bytes%s\n",
3227               h->i_frame,
3228               h->fdec->f_qp_avg_aq,
3229               h->i_nal_ref_idc,
3230               h->sh.i_type == SLICE_TYPE_I ? 'I' : (h->sh.i_type == SLICE_TYPE_P ? 'P' : 'B' ),
3231               h->fdec->i_poc,
3232               h->stat.frame.i_mb_count_i,
3233               h->stat.frame.i_mb_count_p,
3234               h->stat.frame.i_mb_count_skip,
3235               frame_size,
3236               psz_message );
3237
3238     // keep stats all in one place
3239     x264_thread_sync_stat( h->thread[0], h );
3240     // for the use of the next frame
3241     x264_thread_sync_stat( thread_current, h );
3242
3243 #ifdef DEBUG_MB_TYPE
3244 {
3245     static const char mb_chars[] = { 'i', 'i', 'I', 'C', 'P', '8', 'S',
3246         'D', '<', 'X', 'B', 'X', '>', 'B', 'B', 'B', 'B', '8', 'S' };
3247     for( int mb_xy = 0; mb_xy < h->mb.i_mb_width * h->mb.i_mb_height; mb_xy++ )
3248     {
3249         if( h->mb.type[mb_xy] < X264_MBTYPE_MAX && h->mb.type[mb_xy] >= 0 )
3250             fprintf( stderr, "%c ", mb_chars[ h->mb.type[mb_xy] ] );
3251         else
3252             fprintf( stderr, "? " );
3253
3254         if( (mb_xy+1) % h->mb.i_mb_width == 0 )
3255             fprintf( stderr, "\n" );
3256     }
3257 }
3258 #endif
3259
3260     /* Remove duplicates, must be done near the end as breaks h->fref0 array
3261      * by freeing some of its pointers. */
3262     for( int i = 0; i < h->i_ref[0]; i++ )
3263         if( h->fref[0][i] && h->fref[0][i]->b_duplicate )
3264         {
3265             x264_frame_push_blank_unused( h, h->fref[0][i] );
3266             h->fref[0][i] = 0;
3267         }
3268
3269     if( h->param.psz_dump_yuv )
3270         x264_frame_dump( h );
3271     x264_emms();
3272
3273     return frame_size;
3274 }
3275
3276 static void x264_print_intra( int64_t *i_mb_count, double i_count, int b_print_pcm, char *intra )
3277 {
3278     intra += sprintf( intra, "I16..4%s: %4.1f%% %4.1f%% %4.1f%%",
3279         b_print_pcm ? "..PCM" : "",
3280         i_mb_count[I_16x16]/ i_count,
3281         i_mb_count[I_8x8]  / i_count,
3282         i_mb_count[I_4x4]  / i_count );
3283     if( b_print_pcm )
3284         sprintf( intra, " %4.1f%%", i_mb_count[I_PCM]  / i_count );
3285 }
3286
3287 /****************************************************************************
3288  * x264_encoder_close:
3289  ****************************************************************************/
3290 void    x264_encoder_close  ( x264_t *h )
3291 {
3292     int64_t i_yuv_size = FRAME_SIZE( h->param.i_width * h->param.i_height );
3293     int64_t i_mb_count_size[2][7] = {{0}};
3294     char buf[200];
3295     int b_print_pcm = h->stat.i_mb_count[SLICE_TYPE_I][I_PCM]
3296                    || h->stat.i_mb_count[SLICE_TYPE_P][I_PCM]
3297                    || h->stat.i_mb_count[SLICE_TYPE_B][I_PCM];
3298
3299     x264_lookahead_delete( h );
3300
3301     if( h->param.i_threads > 1 )
3302         x264_threadpool_delete( h->threadpool );
3303     if( h->i_thread_frames > 1 )
3304     {
3305         for( int i = 0; i < h->i_thread_frames; i++ )
3306             if( h->thread[i]->b_thread_active )
3307             {
3308                 assert( h->thread[i]->fenc->i_reference_count == 1 );
3309                 x264_frame_delete( h->thread[i]->fenc );
3310             }
3311
3312         x264_t *thread_prev = h->thread[h->i_thread_phase];
3313         x264_thread_sync_ratecontrol( h, thread_prev, h );
3314         x264_thread_sync_ratecontrol( thread_prev, thread_prev, h );
3315         h->i_frame = thread_prev->i_frame + 1 - h->i_thread_frames;
3316     }
3317     h->i_frame++;
3318
3319     /* Slices used and PSNR */
3320     for( int i = 0; i < 3; i++ )
3321     {
3322         static const uint8_t slice_order[] = { SLICE_TYPE_I, SLICE_TYPE_P, SLICE_TYPE_B };
3323         int i_slice = slice_order[i];
3324
3325         if( h->stat.i_frame_count[i_slice] > 0 )
3326         {
3327             int i_count = h->stat.i_frame_count[i_slice];
3328             double dur =  h->stat.f_frame_duration[i_slice];
3329             if( h->param.analyse.b_psnr )
3330             {
3331                 x264_log( h, X264_LOG_INFO,
3332                           "frame %c:%-5d Avg QP:%5.2f  size:%6.0f  PSNR Mean Y:%5.2f U:%5.2f V:%5.2f Avg:%5.2f Global:%5.2f\n",
3333                           slice_type_to_char[i_slice],
3334                           i_count,
3335                           h->stat.f_frame_qp[i_slice] / i_count,
3336                           (double)h->stat.i_frame_size[i_slice] / i_count,
3337                           h->stat.f_psnr_mean_y[i_slice] / dur, h->stat.f_psnr_mean_u[i_slice] / dur, h->stat.f_psnr_mean_v[i_slice] / dur,
3338                           h->stat.f_psnr_average[i_slice] / dur,
3339                           x264_psnr( h->stat.f_ssd_global[i_slice], dur * i_yuv_size ) );
3340             }
3341             else
3342             {
3343                 x264_log( h, X264_LOG_INFO,
3344                           "frame %c:%-5d Avg QP:%5.2f  size:%6.0f\n",
3345                           slice_type_to_char[i_slice],
3346                           i_count,
3347                           h->stat.f_frame_qp[i_slice] / i_count,
3348                           (double)h->stat.i_frame_size[i_slice] / i_count );
3349             }
3350         }
3351     }
3352     if( h->param.i_bframe && h->stat.i_frame_count[SLICE_TYPE_B] )
3353     {
3354         char *p = buf;
3355         int den = 0;
3356         // weight by number of frames (including the I/P-frames) that are in a sequence of N B-frames
3357         for( int i = 0; i <= h->param.i_bframe; i++ )
3358             den += (i+1) * h->stat.i_consecutive_bframes[i];
3359         for( int i = 0; i <= h->param.i_bframe; i++ )
3360             p += sprintf( p, " %4.1f%%", 100. * (i+1) * h->stat.i_consecutive_bframes[i] / den );
3361         x264_log( h, X264_LOG_INFO, "consecutive B-frames:%s\n", buf );
3362     }
3363
3364     for( int i_type = 0; i_type < 2; i_type++ )
3365         for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
3366         {
3367             if( i == D_DIRECT_8x8 ) continue; /* direct is counted as its own type */
3368             i_mb_count_size[i_type][x264_mb_partition_pixel_table[i]] += h->stat.i_mb_partition[i_type][i];
3369         }
3370
3371     /* MB types used */
3372     if( h->stat.i_frame_count[SLICE_TYPE_I] > 0 )
3373     {
3374         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_I];
3375         double i_count = h->stat.i_frame_count[SLICE_TYPE_I] * h->mb.i_mb_count / 100.0;
3376         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
3377         x264_log( h, X264_LOG_INFO, "mb I  %s\n", buf );
3378     }
3379     if( h->stat.i_frame_count[SLICE_TYPE_P] > 0 )
3380     {
3381         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_P];
3382         double i_count = h->stat.i_frame_count[SLICE_TYPE_P] * h->mb.i_mb_count / 100.0;
3383         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_P];
3384         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
3385         x264_log( h, X264_LOG_INFO,
3386                   "mb P  %s  P16..4: %4.1f%% %4.1f%% %4.1f%% %4.1f%% %4.1f%%    skip:%4.1f%%\n",
3387                   buf,
3388                   i_mb_size[PIXEL_16x16] / (i_count*4),
3389                   (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
3390                   i_mb_size[PIXEL_8x8] / (i_count*4),
3391                   (i_mb_size[PIXEL_8x4] + i_mb_size[PIXEL_4x8]) / (i_count*4),
3392                   i_mb_size[PIXEL_4x4] / (i_count*4),
3393                   i_mb_count[P_SKIP] / i_count );
3394     }
3395     if( h->stat.i_frame_count[SLICE_TYPE_B] > 0 )
3396     {
3397         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_B];
3398         double i_count = h->stat.i_frame_count[SLICE_TYPE_B] * h->mb.i_mb_count / 100.0;
3399         double i_mb_list_count;
3400         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_B];
3401         int64_t list_count[3] = {0}; /* 0 == L0, 1 == L1, 2 == BI */
3402         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
3403         for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
3404             for( int j = 0; j < 2; j++ )
3405             {
3406                 int l0 = x264_mb_type_list_table[i][0][j];
3407                 int l1 = x264_mb_type_list_table[i][1][j];
3408                 if( l0 || l1 )
3409                     list_count[l1+l0*l1] += h->stat.i_mb_count[SLICE_TYPE_B][i] * 2;
3410             }
3411         list_count[0] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L0_8x8];
3412         list_count[1] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L1_8x8];
3413         list_count[2] += h->stat.i_mb_partition[SLICE_TYPE_B][D_BI_8x8];
3414         i_mb_count[B_DIRECT] += (h->stat.i_mb_partition[SLICE_TYPE_B][D_DIRECT_8x8]+2)/4;
3415         i_mb_list_count = (list_count[0] + list_count[1] + list_count[2]) / 100.0;
3416         sprintf( buf + strlen(buf), "  B16..8: %4.1f%% %4.1f%% %4.1f%%  direct:%4.1f%%  skip:%4.1f%%",
3417                  i_mb_size[PIXEL_16x16] / (i_count*4),
3418                  (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
3419                  i_mb_size[PIXEL_8x8] / (i_count*4),
3420                  i_mb_count[B_DIRECT] / i_count,
3421                  i_mb_count[B_SKIP]   / i_count );
3422         if( i_mb_list_count != 0 )
3423             sprintf( buf + strlen(buf), "  L0:%4.1f%% L1:%4.1f%% BI:%4.1f%%",
3424                      list_count[0] / i_mb_list_count,
3425                      list_count[1] / i_mb_list_count,
3426                      list_count[2] / i_mb_list_count );
3427         x264_log( h, X264_LOG_INFO, "mb B  %s\n", buf );
3428     }
3429
3430     x264_ratecontrol_summary( h );
3431
3432     if( h->stat.i_frame_count[SLICE_TYPE_I] + h->stat.i_frame_count[SLICE_TYPE_P] + h->stat.i_frame_count[SLICE_TYPE_B] > 0 )
3433     {
3434 #define SUM3(p) (p[SLICE_TYPE_I] + p[SLICE_TYPE_P] + p[SLICE_TYPE_B])
3435 #define SUM3b(p,o) (p[SLICE_TYPE_I][o] + p[SLICE_TYPE_P][o] + p[SLICE_TYPE_B][o])
3436         int64_t i_i8x8 = SUM3b( h->stat.i_mb_count, I_8x8 );
3437         int64_t i_intra = i_i8x8 + SUM3b( h->stat.i_mb_count, I_4x4 )
3438                                  + SUM3b( h->stat.i_mb_count, I_16x16 );
3439         int64_t i_all_intra = i_intra + SUM3b( h->stat.i_mb_count, I_PCM);
3440         int64_t i_skip = SUM3b( h->stat.i_mb_count, P_SKIP )
3441                        + SUM3b( h->stat.i_mb_count, B_SKIP );
3442         const int i_count = h->stat.i_frame_count[SLICE_TYPE_I] +
3443                             h->stat.i_frame_count[SLICE_TYPE_P] +
3444                             h->stat.i_frame_count[SLICE_TYPE_B];
3445         int64_t i_mb_count = (int64_t)i_count * h->mb.i_mb_count;
3446         int64_t i_inter = i_mb_count - i_skip - i_intra;
3447         const double duration = h->stat.f_frame_duration[SLICE_TYPE_I] +
3448                                 h->stat.f_frame_duration[SLICE_TYPE_P] +
3449                                 h->stat.f_frame_duration[SLICE_TYPE_B];
3450         float f_bitrate = SUM3(h->stat.i_frame_size) / duration / 125;
3451
3452         if( PARAM_INTERLACED )
3453         {
3454             char *fieldstats = buf;
3455             fieldstats[0] = 0;
3456             if( i_inter )
3457                 fieldstats += sprintf( fieldstats, " inter:%.1f%%", h->stat.i_mb_field[1] * 100.0 / i_inter );
3458             if( i_skip )
3459                 fieldstats += sprintf( fieldstats, " skip:%.1f%%", h->stat.i_mb_field[2] * 100.0 / i_skip );
3460             x264_log( h, X264_LOG_INFO, "field mbs: intra: %.1f%%%s\n",
3461                       h->stat.i_mb_field[0] * 100.0 / i_intra, buf );
3462         }
3463
3464         if( h->pps->b_transform_8x8_mode )
3465         {
3466             buf[0] = 0;
3467             if( h->stat.i_mb_count_8x8dct[0] )
3468                 sprintf( buf, " inter:%.1f%%", 100. * h->stat.i_mb_count_8x8dct[1] / h->stat.i_mb_count_8x8dct[0] );
3469             x264_log( h, X264_LOG_INFO, "8x8 transform intra:%.1f%%%s\n", 100. * i_i8x8 / i_intra, buf );
3470         }
3471
3472         if( (h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO ||
3473             (h->stat.i_direct_frames[0] && h->stat.i_direct_frames[1]))
3474             && h->stat.i_frame_count[SLICE_TYPE_B] )
3475         {
3476             x264_log( h, X264_LOG_INFO, "direct mvs  spatial:%.1f%% temporal:%.1f%%\n",
3477                       h->stat.i_direct_frames[1] * 100. / h->stat.i_frame_count[SLICE_TYPE_B],
3478                       h->stat.i_direct_frames[0] * 100. / h->stat.i_frame_count[SLICE_TYPE_B] );
3479         }
3480
3481         buf[0] = 0;
3482         int csize = CHROMA444 ? 4 : 1;
3483         if( i_mb_count != i_all_intra )
3484             sprintf( buf, " inter: %.1f%% %.1f%% %.1f%%",
3485                      h->stat.i_mb_cbp[1] * 100.0 / ((i_mb_count - i_all_intra)*4),
3486                      h->stat.i_mb_cbp[3] * 100.0 / ((i_mb_count - i_all_intra)*csize),
3487                      h->stat.i_mb_cbp[5] * 100.0 / ((i_mb_count - i_all_intra)*csize) );
3488         x264_log( h, X264_LOG_INFO, "coded y,%s,%s intra: %.1f%% %.1f%% %.1f%%%s\n",
3489                   CHROMA444?"u":"uvDC", CHROMA444?"v":"uvAC",
3490                   h->stat.i_mb_cbp[0] * 100.0 / (i_all_intra*4),
3491                   h->stat.i_mb_cbp[2] * 100.0 / (i_all_intra*csize),
3492                   h->stat.i_mb_cbp[4] * 100.0 / (i_all_intra*csize), buf );
3493
3494         int64_t fixed_pred_modes[4][9] = {{0}};
3495         int64_t sum_pred_modes[4] = {0};
3496         for( int i = 0; i <= I_PRED_16x16_DC_128; i++ )
3497         {
3498             fixed_pred_modes[0][x264_mb_pred_mode16x16_fix[i]] += h->stat.i_mb_pred_mode[0][i];
3499             sum_pred_modes[0] += h->stat.i_mb_pred_mode[0][i];
3500         }
3501         if( sum_pred_modes[0] )
3502             x264_log( h, X264_LOG_INFO, "i16 v,h,dc,p: %2.0f%% %2.0f%% %2.0f%% %2.0f%%\n",
3503                       fixed_pred_modes[0][0] * 100.0 / sum_pred_modes[0],
3504                       fixed_pred_modes[0][1] * 100.0 / sum_pred_modes[0],
3505                       fixed_pred_modes[0][2] * 100.0 / sum_pred_modes[0],
3506                       fixed_pred_modes[0][3] * 100.0 / sum_pred_modes[0] );
3507         for( int i = 1; i <= 2; i++ )
3508         {
3509             for( int j = 0; j <= I_PRED_8x8_DC_128; j++ )
3510             {
3511                 fixed_pred_modes[i][x264_mb_pred_mode4x4_fix(j)] += h->stat.i_mb_pred_mode[i][j];
3512                 sum_pred_modes[i] += h->stat.i_mb_pred_mode[i][j];
3513             }
3514             if( sum_pred_modes[i] )
3515                 x264_log( h, X264_LOG_INFO, "i%d v,h,dc,ddl,ddr,vr,hd,vl,hu: %2.0f%% %2.0f%% %2.0f%% %2.0f%% %2.0f%% %2.0f%% %2.0f%% %2.0f%% %2.0f%%\n", (3-i)*4,
3516                           fixed_pred_modes[i][0] * 100.0 / sum_pred_modes[i],
3517                           fixed_pred_modes[i][1] * 100.0 / sum_pred_modes[i],
3518                           fixed_pred_modes[i][2] * 100.0 / sum_pred_modes[i],
3519                           fixed_pred_modes[i][3] * 100.0 / sum_pred_modes[i],
3520                           fixed_pred_modes[i][4] * 100.0 / sum_pred_modes[i],
3521                           fixed_pred_modes[i][5] * 100.0 / sum_pred_modes[i],
3522                           fixed_pred_modes[i][6] * 100.0 / sum_pred_modes[i],
3523                           fixed_pred_modes[i][7] * 100.0 / sum_pred_modes[i],
3524                           fixed_pred_modes[i][8] * 100.0 / sum_pred_modes[i] );
3525         }
3526         for( int i = 0; i <= I_PRED_CHROMA_DC_128; i++ )
3527         {
3528             fixed_pred_modes[3][x264_mb_chroma_pred_mode_fix[i]] += h->stat.i_mb_pred_mode[3][i];
3529             sum_pred_modes[3] += h->stat.i_mb_pred_mode[3][i];
3530         }
3531         if( sum_pred_modes[3] && !CHROMA444 )
3532             x264_log( h, X264_LOG_INFO, "i8c dc,h,v,p: %2.0f%% %2.0f%% %2.0f%% %2.0f%%\n",
3533                       fixed_pred_modes[3][0] * 100.0 / sum_pred_modes[3],
3534                       fixed_pred_modes[3][1] * 100.0 / sum_pred_modes[3],
3535                       fixed_pred_modes[3][2] * 100.0 / sum_pred_modes[3],
3536                       fixed_pred_modes[3][3] * 100.0 / sum_pred_modes[3] );
3537
3538         if( h->param.analyse.i_weighted_pred >= X264_WEIGHTP_SIMPLE && h->stat.i_frame_count[SLICE_TYPE_P] > 0 )
3539             x264_log( h, X264_LOG_INFO, "Weighted P-Frames: Y:%.1f%% UV:%.1f%%\n",
3540                       h->stat.i_wpred[0] * 100.0 / h->stat.i_frame_count[SLICE_TYPE_P],
3541                       h->stat.i_wpred[1] * 100.0 / h->stat.i_frame_count[SLICE_TYPE_P] );
3542
3543         for( int i_list = 0; i_list < 2; i_list++ )
3544             for( int i_slice = 0; i_slice < 2; i_slice++ )
3545             {
3546                 char *p = buf;
3547                 int64_t i_den = 0;
3548                 int i_max = 0;
3549                 for( int i = 0; i < X264_REF_MAX*2; i++ )
3550                     if( h->stat.i_mb_count_ref[i_slice][i_list][i] )
3551                     {
3552                         i_den += h->stat.i_mb_count_ref[i_slice][i_list][i];
3553                         i_max = i;
3554                     }
3555                 if( i_max == 0 )
3556                     continue;
3557                 for( int i = 0; i <= i_max; i++ )
3558                     p += sprintf( p, " %4.1f%%", 100. * h->stat.i_mb_count_ref[i_slice][i_list][i] / i_den );
3559                 x264_log( h, X264_LOG_INFO, "ref %c L%d:%s\n", "PB"[i_slice], i_list, buf );
3560             }
3561
3562         if( h->param.analyse.b_ssim )
3563         {
3564             float ssim = SUM3( h->stat.f_ssim_mean_y ) / duration;
3565             x264_log( h, X264_LOG_INFO, "SSIM Mean Y:%.7f (%6.3fdb)\n", ssim, x264_ssim( ssim ) );
3566         }
3567         if( h->param.analyse.b_psnr )
3568         {
3569             x264_log( h, X264_LOG_INFO,
3570                       "PSNR Mean Y:%6.3f U:%6.3f V:%6.3f Avg:%6.3f Global:%6.3f kb/s:%.2f\n",
3571                       SUM3( h->stat.f_psnr_mean_y ) / duration,
3572                       SUM3( h->stat.f_psnr_mean_u ) / duration,
3573                       SUM3( h->stat.f_psnr_mean_v ) / duration,
3574                       SUM3( h->stat.f_psnr_average ) / duration,
3575                       x264_psnr( SUM3( h->stat.f_ssd_global ), duration * i_yuv_size ),
3576                       f_bitrate );
3577         }
3578         else
3579             x264_log( h, X264_LOG_INFO, "kb/s:%.2f\n", f_bitrate );
3580     }
3581
3582     /* rc */
3583     x264_ratecontrol_delete( h );
3584
3585     /* param */
3586     if( h->param.rc.psz_stat_out )
3587         free( h->param.rc.psz_stat_out );
3588     if( h->param.rc.psz_stat_in )
3589         free( h->param.rc.psz_stat_in );
3590
3591     x264_cqm_delete( h );
3592     x264_free( h->nal_buffer );
3593     x264_analyse_free_costs( h );
3594
3595     if( h->i_thread_frames > 1)
3596         h = h->thread[h->i_thread_phase];
3597
3598     /* frames */
3599     x264_frame_delete_list( h->frames.unused[0] );
3600     x264_frame_delete_list( h->frames.unused[1] );
3601     x264_frame_delete_list( h->frames.current );
3602     x264_frame_delete_list( h->frames.blank_unused );
3603
3604     h = h->thread[0];
3605
3606     for( int i = 0; i < h->i_thread_frames; i++ )
3607         if( h->thread[i]->b_thread_active )
3608             for( int j = 0; j < h->thread[i]->i_ref[0]; j++ )
3609                 if( h->thread[i]->fref[0][j] && h->thread[i]->fref[0][j]->b_duplicate )
3610                     x264_frame_delete( h->thread[i]->fref[0][j] );
3611
3612     for( int i = h->param.i_threads - 1; i >= 0; i-- )
3613     {
3614         x264_frame_t **frame;
3615
3616         if( !h->param.b_sliced_threads || i == 0 )
3617         {
3618             for( frame = h->thread[i]->frames.reference; *frame; frame++ )
3619             {
3620                 assert( (*frame)->i_reference_count > 0 );
3621                 (*frame)->i_reference_count--;
3622                 if( (*frame)->i_reference_count == 0 )
3623                     x264_frame_delete( *frame );
3624             }
3625             frame = &h->thread[i]->fdec;
3626             if( *frame )
3627             {
3628                 assert( (*frame)->i_reference_count > 0 );
3629                 (*frame)->i_reference_count--;
3630                 if( (*frame)->i_reference_count == 0 )
3631                     x264_frame_delete( *frame );
3632             }
3633             x264_macroblock_cache_free( h->thread[i] );
3634         }
3635         x264_macroblock_thread_free( h->thread[i], 0 );
3636         x264_free( h->thread[i]->out.p_bitstream );
3637         x264_free( h->thread[i]->out.nal);
3638         x264_free( h->thread[i] );
3639     }
3640 }
3641
3642 int x264_encoder_delayed_frames( x264_t *h )
3643 {
3644     int delayed_frames = 0;
3645     if( h->i_thread_frames > 1 )
3646     {
3647         for( int i = 0; i < h->i_thread_frames; i++ )
3648             delayed_frames += h->thread[i]->b_thread_active;
3649         h = h->thread[h->i_thread_phase];
3650     }
3651     for( int i = 0; h->frames.current[i]; i++ )
3652         delayed_frames++;
3653     x264_pthread_mutex_lock( &h->lookahead->ofbuf.mutex );
3654     x264_pthread_mutex_lock( &h->lookahead->ifbuf.mutex );
3655     x264_pthread_mutex_lock( &h->lookahead->next.mutex );
3656     delayed_frames += h->lookahead->ifbuf.i_size + h->lookahead->next.i_size + h->lookahead->ofbuf.i_size;
3657     x264_pthread_mutex_unlock( &h->lookahead->next.mutex );
3658     x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
3659     x264_pthread_mutex_unlock( &h->lookahead->ofbuf.mutex );
3660     return delayed_frames;
3661 }
3662
3663 int x264_encoder_maximum_delayed_frames( x264_t *h )
3664 {
3665     return h->frames.i_delay;
3666 }