Support for 9 and 10-bit encoding
[x262.git] / encoder / encoder.c
1 /*****************************************************************************
2  * x264: h264 encoder
3  *****************************************************************************
4  * Copyright (C) 2003-2008 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
25 #include <math.h>
26
27 #include "common/common.h"
28
29 #include "set.h"
30 #include "analyse.h"
31 #include "ratecontrol.h"
32 #include "macroblock.h"
33 #include "me.h"
34
35 #if HAVE_VISUALIZE
36 #include "common/visualize.h"
37 #endif
38
39 //#define DEBUG_MB_TYPE
40
41 #define bs_write_ue bs_write_ue_big
42
43 static int x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
44                                    x264_nal_t **pp_nal, int *pi_nal,
45                                    x264_picture_t *pic_out );
46
47 /****************************************************************************
48  *
49  ******************************* x264 libs **********************************
50  *
51  ****************************************************************************/
52 static float x264_psnr( int64_t i_sqe, int64_t i_size )
53 {
54     double f_mse = (double)i_sqe / (PIXEL_MAX*PIXEL_MAX * (double)i_size);
55     if( f_mse <= 0.0000000001 ) /* Max 100dB */
56         return 100;
57
58     return -10.0 * log10( f_mse );
59 }
60
61 static float x264_ssim( float ssim )
62 {
63     return -10.0 * log10( 1 - ssim );
64 }
65
66 static void x264_frame_dump( x264_t *h )
67 {
68     FILE *f = fopen( h->param.psz_dump_yuv, "r+b" );
69     if( !f )
70         return;
71     int bytes_per_pixel = (BIT_DEPTH+7)/8;
72     /* Write the frame in display order */
73     fseek( f, (uint64_t)h->fdec->i_frame * h->param.i_height * h->param.i_width * 3/2 * bytes_per_pixel, SEEK_SET );
74     for( int i = 0; i < h->fdec->i_plane; i++ )
75         for( int y = 0; y < h->param.i_height >> !!i; y++ )
76             for( int j = 0; j < h->param.i_width >> !!i; j++ )
77                 fwrite( &h->fdec->plane[i][y*h->fdec->i_stride[i]]+j, bytes_per_pixel, 1, f );
78     fclose( f );
79 }
80
81
82 /* Fill "default" values */
83 static void x264_slice_header_init( x264_t *h, x264_slice_header_t *sh,
84                                     x264_sps_t *sps, x264_pps_t *pps,
85                                     int i_idr_pic_id, int i_frame, int i_qp )
86 {
87     x264_param_t *param = &h->param;
88
89     /* First we fill all fields */
90     sh->sps = sps;
91     sh->pps = pps;
92
93     sh->i_first_mb  = 0;
94     sh->i_last_mb   = h->mb.i_mb_count - 1;
95     sh->i_pps_id    = pps->i_id;
96
97     sh->i_frame_num = i_frame;
98
99     sh->b_mbaff = h->param.b_interlaced;
100     sh->b_field_pic = 0;    /* no field support for now */
101     sh->b_bottom_field = 0; /* not yet used */
102
103     sh->i_idr_pic_id = i_idr_pic_id;
104
105     /* poc stuff, fixed later */
106     sh->i_poc = 0;
107     sh->i_delta_poc_bottom = 0;
108     sh->i_delta_poc[0] = 0;
109     sh->i_delta_poc[1] = 0;
110
111     sh->i_redundant_pic_cnt = 0;
112
113     h->mb.b_direct_auto_write = h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO
114                                 && h->param.i_bframe
115                                 && ( h->param.rc.b_stat_write || !h->param.rc.b_stat_read );
116
117     if( !h->mb.b_direct_auto_read && sh->i_type == SLICE_TYPE_B )
118     {
119         if( h->fref1[0]->i_poc_l0ref0 == h->fref0[0]->i_poc )
120         {
121             if( h->mb.b_direct_auto_write )
122                 sh->b_direct_spatial_mv_pred = ( h->stat.i_direct_score[1] > h->stat.i_direct_score[0] );
123             else
124                 sh->b_direct_spatial_mv_pred = ( param->analyse.i_direct_mv_pred == X264_DIRECT_PRED_SPATIAL );
125         }
126         else
127         {
128             h->mb.b_direct_auto_write = 0;
129             sh->b_direct_spatial_mv_pred = 1;
130         }
131     }
132     /* else b_direct_spatial_mv_pred was read from the 2pass statsfile */
133
134     sh->b_num_ref_idx_override = 0;
135     sh->i_num_ref_idx_l0_active = 1;
136     sh->i_num_ref_idx_l1_active = 1;
137
138     sh->b_ref_pic_list_reordering_l0 = h->b_ref_reorder[0];
139     sh->b_ref_pic_list_reordering_l1 = h->b_ref_reorder[1];
140
141     /* If the ref list isn't in the default order, construct reordering header */
142     /* List1 reordering isn't needed yet */
143     if( sh->b_ref_pic_list_reordering_l0 )
144     {
145         int pred_frame_num = i_frame;
146         for( int i = 0; i < h->i_ref0; i++ )
147         {
148             int diff = h->fref0[i]->i_frame_num - pred_frame_num;
149             sh->ref_pic_list_order[0][i].idc = ( diff > 0 );
150             sh->ref_pic_list_order[0][i].arg = (abs(diff) - 1) & ((1 << sps->i_log2_max_frame_num) - 1);
151             pred_frame_num = h->fref0[i]->i_frame_num;
152         }
153     }
154
155     sh->i_cabac_init_idc = param->i_cabac_init_idc;
156
157     sh->i_qp = i_qp;
158     sh->i_qp_delta = i_qp - pps->i_pic_init_qp;
159     sh->b_sp_for_swidth = 0;
160     sh->i_qs_delta = 0;
161
162     int deblock_thresh = i_qp + 2 * X264_MIN(param->i_deblocking_filter_alphac0, param->i_deblocking_filter_beta);
163     /* If effective qp <= 15, deblocking would have no effect anyway */
164     if( param->b_deblocking_filter && (h->mb.b_variable_qp || 15 < deblock_thresh ) )
165         sh->i_disable_deblocking_filter_idc = param->b_sliced_threads ? 2 : 0;
166     else
167         sh->i_disable_deblocking_filter_idc = 1;
168     sh->i_alpha_c0_offset = param->i_deblocking_filter_alphac0 << 1;
169     sh->i_beta_offset = param->i_deblocking_filter_beta << 1;
170 }
171
172 static void x264_slice_header_write( bs_t *s, x264_slice_header_t *sh, int i_nal_ref_idc )
173 {
174     if( sh->b_mbaff )
175     {
176         assert( sh->i_first_mb % (2*sh->sps->i_mb_width) == 0 );
177         bs_write_ue( s, sh->i_first_mb >> 1 );
178     }
179     else
180         bs_write_ue( s, sh->i_first_mb );
181
182     bs_write_ue( s, sh->i_type + 5 );   /* same type things */
183     bs_write_ue( s, sh->i_pps_id );
184     bs_write( s, sh->sps->i_log2_max_frame_num, sh->i_frame_num & ((1<<sh->sps->i_log2_max_frame_num)-1) );
185
186     if( !sh->sps->b_frame_mbs_only )
187     {
188         bs_write1( s, sh->b_field_pic );
189         if( sh->b_field_pic )
190             bs_write1( s, sh->b_bottom_field );
191     }
192
193     if( sh->i_idr_pic_id >= 0 ) /* NAL IDR */
194         bs_write_ue( s, sh->i_idr_pic_id );
195
196     if( sh->sps->i_poc_type == 0 )
197     {
198         bs_write( s, sh->sps->i_log2_max_poc_lsb, sh->i_poc & ((1<<sh->sps->i_log2_max_poc_lsb)-1) );
199         if( sh->pps->b_pic_order && !sh->b_field_pic )
200             bs_write_se( s, sh->i_delta_poc_bottom );
201     }
202     else if( sh->sps->i_poc_type == 1 && !sh->sps->b_delta_pic_order_always_zero )
203     {
204         bs_write_se( s, sh->i_delta_poc[0] );
205         if( sh->pps->b_pic_order && !sh->b_field_pic )
206             bs_write_se( s, sh->i_delta_poc[1] );
207     }
208
209     if( sh->pps->b_redundant_pic_cnt )
210         bs_write_ue( s, sh->i_redundant_pic_cnt );
211
212     if( sh->i_type == SLICE_TYPE_B )
213         bs_write1( s, sh->b_direct_spatial_mv_pred );
214
215     if( sh->i_type == SLICE_TYPE_P || sh->i_type == SLICE_TYPE_SP || sh->i_type == SLICE_TYPE_B )
216     {
217         bs_write1( s, sh->b_num_ref_idx_override );
218         if( sh->b_num_ref_idx_override )
219         {
220             bs_write_ue( s, sh->i_num_ref_idx_l0_active - 1 );
221             if( sh->i_type == SLICE_TYPE_B )
222                 bs_write_ue( s, sh->i_num_ref_idx_l1_active - 1 );
223         }
224     }
225
226     /* ref pic list reordering */
227     if( sh->i_type != SLICE_TYPE_I )
228     {
229         bs_write1( s, sh->b_ref_pic_list_reordering_l0 );
230         if( sh->b_ref_pic_list_reordering_l0 )
231         {
232             for( int i = 0; i < sh->i_num_ref_idx_l0_active; i++ )
233             {
234                 bs_write_ue( s, sh->ref_pic_list_order[0][i].idc );
235                 bs_write_ue( s, sh->ref_pic_list_order[0][i].arg );
236             }
237             bs_write_ue( s, 3 );
238         }
239     }
240     if( sh->i_type == SLICE_TYPE_B )
241     {
242         bs_write1( s, sh->b_ref_pic_list_reordering_l1 );
243         if( sh->b_ref_pic_list_reordering_l1 )
244         {
245             for( int i = 0; i < sh->i_num_ref_idx_l1_active; i++ )
246             {
247                 bs_write_ue( s, sh->ref_pic_list_order[1][i].idc );
248                 bs_write_ue( s, sh->ref_pic_list_order[1][i].arg );
249             }
250             bs_write_ue( s, 3 );
251         }
252     }
253
254     if( sh->pps->b_weighted_pred && ( sh->i_type == SLICE_TYPE_P || sh->i_type == SLICE_TYPE_SP ) )
255     {
256         /* pred_weight_table() */
257         bs_write_ue( s, sh->weight[0][0].i_denom );
258         bs_write_ue( s, sh->weight[0][1].i_denom );
259         for( int i = 0; i < sh->i_num_ref_idx_l0_active; i++ )
260         {
261             int luma_weight_l0_flag = !!sh->weight[i][0].weightfn;
262             int chroma_weight_l0_flag = !!sh->weight[i][1].weightfn || !!sh->weight[i][2].weightfn;
263             bs_write1( s, luma_weight_l0_flag );
264             if( luma_weight_l0_flag )
265             {
266                 bs_write_se( s, sh->weight[i][0].i_scale );
267                 bs_write_se( s, sh->weight[i][0].i_offset );
268             }
269             bs_write1( s, chroma_weight_l0_flag );
270             if( chroma_weight_l0_flag )
271             {
272                 for( int j = 1; j < 3; j++ )
273                 {
274                     bs_write_se( s, sh->weight[i][j].i_scale );
275                     bs_write_se( s, sh->weight[i][j].i_offset );
276                 }
277             }
278         }
279     }
280     else if( sh->pps->b_weighted_bipred == 1 && sh->i_type == SLICE_TYPE_B )
281     {
282       /* TODO */
283     }
284
285     if( i_nal_ref_idc != 0 )
286     {
287         if( sh->i_idr_pic_id >= 0 )
288         {
289             bs_write1( s, 0 );  /* no output of prior pics flag */
290             bs_write1( s, 0 );  /* long term reference flag */
291         }
292         else
293         {
294             bs_write1( s, sh->i_mmco_command_count > 0 ); /* adaptive_ref_pic_marking_mode_flag */
295             if( sh->i_mmco_command_count > 0 )
296             {
297                 for( int i = 0; i < sh->i_mmco_command_count; i++ )
298                 {
299                     bs_write_ue( s, 1 ); /* mark short term ref as unused */
300                     bs_write_ue( s, sh->mmco[i].i_difference_of_pic_nums - 1 );
301                 }
302                 bs_write_ue( s, 0 ); /* end command list */
303             }
304         }
305     }
306
307     if( sh->pps->b_cabac && sh->i_type != SLICE_TYPE_I )
308         bs_write_ue( s, sh->i_cabac_init_idc );
309
310     bs_write_se( s, sh->i_qp_delta );      /* slice qp delta */
311
312     if( sh->pps->b_deblocking_filter_control )
313     {
314         bs_write_ue( s, sh->i_disable_deblocking_filter_idc );
315         if( sh->i_disable_deblocking_filter_idc != 1 )
316         {
317             bs_write_se( s, sh->i_alpha_c0_offset >> 1 );
318             bs_write_se( s, sh->i_beta_offset >> 1 );
319         }
320     }
321 }
322
323 /* If we are within a reasonable distance of the end of the memory allocated for the bitstream, */
324 /* reallocate, adding an arbitrary amount of space (100 kilobytes). */
325 static int x264_bitstream_check_buffer( x264_t *h )
326 {
327     uint8_t *bs_bak = h->out.p_bitstream;
328     if( (h->param.b_cabac && (h->cabac.p_end - h->cabac.p < 2500)) ||
329         (h->out.bs.p_end - h->out.bs.p < 2500) )
330     {
331         h->out.i_bitstream += 100000;
332         CHECKED_MALLOC( h->out.p_bitstream, h->out.i_bitstream );
333         h->mc.memcpy_aligned( h->out.p_bitstream, bs_bak, (h->out.i_bitstream - 100000) & ~15 );
334         intptr_t delta = h->out.p_bitstream - bs_bak;
335
336         h->out.bs.p_start += delta;
337         h->out.bs.p += delta;
338         h->out.bs.p_end = h->out.p_bitstream + h->out.i_bitstream;
339
340         h->cabac.p_start += delta;
341         h->cabac.p += delta;
342         h->cabac.p_end = h->out.p_bitstream + h->out.i_bitstream;
343
344         for( int i = 0; i <= h->out.i_nal; i++ )
345             h->out.nal[i].p_payload += delta;
346         x264_free( bs_bak );
347     }
348     return 0;
349 fail:
350     x264_free( bs_bak );
351     return -1;
352 }
353
354 #if HAVE_PTHREAD
355 static void x264_encoder_thread_init( x264_t *h )
356 {
357     if( h->param.i_sync_lookahead )
358         x264_lower_thread_priority( 10 );
359
360 #if HAVE_MMX
361     /* Misalign mask has to be set separately for each thread. */
362     if( h->param.cpu&X264_CPU_SSE_MISALIGN )
363         x264_cpu_mask_misalign_sse();
364 #endif
365 }
366 #endif
367
368 /****************************************************************************
369  *
370  ****************************************************************************
371  ****************************** External API*********************************
372  ****************************************************************************
373  *
374  ****************************************************************************/
375
376 static int x264_validate_parameters( x264_t *h )
377 {
378 #if HAVE_MMX
379 #ifdef __SSE__
380     if( !(x264_cpu_detect() & X264_CPU_SSE) )
381     {
382         x264_log( h, X264_LOG_ERROR, "your cpu does not support SSE1, but x264 was compiled with asm support\n");
383 #else
384     if( !(x264_cpu_detect() & X264_CPU_MMXEXT) )
385     {
386         x264_log( h, X264_LOG_ERROR, "your cpu does not support MMXEXT, but x264 was compiled with asm support\n");
387 #endif
388         x264_log( h, X264_LOG_ERROR, "to run x264, recompile without asm support (configure --disable-asm)\n");
389         return -1;
390     }
391 #endif
392     if( h->param.i_width <= 0 || h->param.i_height <= 0 )
393     {
394         x264_log( h, X264_LOG_ERROR, "invalid width x height (%dx%d)\n",
395                   h->param.i_width, h->param.i_height );
396         return -1;
397     }
398
399     if( h->param.i_width % 2 || h->param.i_height % 2 )
400     {
401         x264_log( h, X264_LOG_ERROR, "width or height not divisible by 2 (%dx%d)\n",
402                   h->param.i_width, h->param.i_height );
403         return -1;
404     }
405     int i_csp = h->param.i_csp & X264_CSP_MASK;
406     if( i_csp != X264_CSP_I420 && i_csp != X264_CSP_YV12 )
407     {
408         x264_log( h, X264_LOG_ERROR, "invalid CSP (only I420/YV12 supported)\n" );
409         return -1;
410     }
411
412     if( h->param.i_threads == X264_THREADS_AUTO )
413         h->param.i_threads = x264_cpu_num_processors() * (h->param.b_sliced_threads?2:3)/2;
414     h->param.i_threads = x264_clip3( h->param.i_threads, 1, X264_THREAD_MAX );
415     if( h->param.i_threads > 1 )
416     {
417 #if !HAVE_PTHREAD
418         x264_log( h, X264_LOG_WARNING, "not compiled with pthread support!\n");
419         h->param.i_threads = 1;
420 #endif
421         /* Avoid absurdly small thread slices as they can reduce performance
422          * and VBV compliance.  Capped at an arbitrary 4 rows per thread. */
423         if( h->param.b_sliced_threads )
424         {
425             int max_threads = (h->param.i_height+15)/16 / 4;
426             h->param.i_threads = X264_MIN( h->param.i_threads, max_threads );
427         }
428     }
429     else
430         h->param.b_sliced_threads = 0;
431     h->i_thread_frames = h->param.b_sliced_threads ? 1 : h->param.i_threads;
432     if( h->i_thread_frames > 1 )
433         h->param.nalu_process = NULL;
434
435     if( h->param.b_interlaced )
436     {
437         if( h->param.analyse.i_me_method >= X264_ME_ESA )
438         {
439             x264_log( h, X264_LOG_WARNING, "interlace + me=esa is not implemented\n" );
440             h->param.analyse.i_me_method = X264_ME_UMH;
441         }
442         if( h->param.analyse.i_weighted_pred > 0 )
443         {
444             x264_log( h, X264_LOG_WARNING, "interlace + weightp is not implemented\n" );
445             h->param.analyse.i_weighted_pred = X264_WEIGHTP_NONE;
446         }
447     }
448
449     /* Detect default ffmpeg settings and terminate with an error. */
450     {
451         int score = 0;
452         score += h->param.analyse.i_me_range == 0;
453         score += h->param.rc.i_qp_step == 3;
454         score += h->param.i_keyint_max == 12;
455         score += h->param.rc.i_qp_min == 2;
456         score += h->param.rc.i_qp_max == 31;
457         score += h->param.rc.f_qcompress == 0.5;
458         score += fabs(h->param.rc.f_ip_factor - 1.25) < 0.01;
459         score += fabs(h->param.rc.f_pb_factor - 1.25) < 0.01;
460         score += h->param.analyse.inter == 0 && h->param.analyse.i_subpel_refine == 8;
461         if( score >= 5 )
462         {
463             x264_log( h, X264_LOG_ERROR, "broken ffmpeg default settings detected\n" );
464             x264_log( h, X264_LOG_ERROR, "use an encoding preset (vpre)\n" );
465             return -1;
466         }
467     }
468
469     if( h->param.rc.i_rc_method < 0 || h->param.rc.i_rc_method > 2 )
470     {
471         x264_log( h, X264_LOG_ERROR, "no ratecontrol method specified\n" );
472         return -1;
473     }
474     h->param.rc.f_rf_constant = x264_clip3f( h->param.rc.f_rf_constant, 0, QP_MAX );
475     h->param.rc.i_qp_constant = x264_clip3( h->param.rc.i_qp_constant, 0, QP_MAX );
476     if( h->param.rc.i_rc_method == X264_RC_CRF )
477     {
478         h->param.rc.i_qp_constant = h->param.rc.f_rf_constant;
479         h->param.rc.i_bitrate = 0;
480     }
481     if( (h->param.rc.i_rc_method == X264_RC_CQP || h->param.rc.i_rc_method == X264_RC_CRF)
482         && h->param.rc.i_qp_constant == 0 )
483     {
484         h->mb.b_lossless = 1;
485         h->param.i_cqm_preset = X264_CQM_FLAT;
486         h->param.psz_cqm_file = NULL;
487         h->param.rc.i_rc_method = X264_RC_CQP;
488         h->param.rc.f_ip_factor = 1;
489         h->param.rc.f_pb_factor = 1;
490         h->param.analyse.b_psnr = 0;
491         h->param.analyse.b_ssim = 0;
492         h->param.analyse.i_chroma_qp_offset = 0;
493         h->param.analyse.i_trellis = 0;
494         h->param.analyse.b_fast_pskip = 0;
495         h->param.analyse.i_noise_reduction = 0;
496         h->param.analyse.b_psy = 0;
497         h->param.i_bframe = 0;
498         /* 8x8dct is not useful at all in CAVLC lossless */
499         if( !h->param.b_cabac )
500             h->param.analyse.b_transform_8x8 = 0;
501     }
502     if( h->param.rc.i_rc_method == X264_RC_CQP )
503     {
504         float qp_p = h->param.rc.i_qp_constant;
505         float qp_i = qp_p - 6*log2f( h->param.rc.f_ip_factor );
506         float qp_b = qp_p + 6*log2f( h->param.rc.f_pb_factor );
507         h->param.rc.i_qp_min = x264_clip3( (int)(X264_MIN3( qp_p, qp_i, qp_b )), 0, QP_MAX );
508         h->param.rc.i_qp_max = x264_clip3( (int)(X264_MAX3( qp_p, qp_i, qp_b ) + .999), 0, QP_MAX );
509         h->param.rc.i_aq_mode = 0;
510         h->param.rc.b_mb_tree = 0;
511     }
512     h->param.rc.i_qp_max = x264_clip3( h->param.rc.i_qp_max, 0, QP_MAX );
513     h->param.rc.i_qp_min = x264_clip3( h->param.rc.i_qp_min, 0, h->param.rc.i_qp_max );
514     if( h->param.rc.i_vbv_buffer_size )
515     {
516         if( h->param.rc.i_rc_method == X264_RC_CQP )
517         {
518             x264_log( h, X264_LOG_WARNING, "VBV is incompatible with constant QP, ignored.\n" );
519             h->param.rc.i_vbv_max_bitrate = 0;
520             h->param.rc.i_vbv_buffer_size = 0;
521         }
522         else if( h->param.rc.i_vbv_max_bitrate == 0 )
523         {
524             if( h->param.rc.i_rc_method == X264_RC_ABR )
525             {
526                 x264_log( h, X264_LOG_WARNING, "VBV maxrate unspecified, assuming CBR\n" );
527                 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate;
528             }
529             else
530             {
531                 x264_log( h, X264_LOG_WARNING, "VBV bufsize set but maxrate unspecified, ignored\n" );
532                 h->param.rc.i_vbv_buffer_size = 0;
533             }
534         }
535         else if( h->param.rc.i_vbv_max_bitrate < h->param.rc.i_bitrate &&
536                  h->param.rc.i_rc_method == X264_RC_ABR )
537         {
538             x264_log( h, X264_LOG_WARNING, "max bitrate less than average bitrate, assuming CBR\n" );
539             h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate;
540         }
541     }
542     else if( h->param.rc.i_vbv_max_bitrate )
543     {
544         x264_log( h, X264_LOG_WARNING, "VBV maxrate specified, but no bufsize, ignored\n" );
545         h->param.rc.i_vbv_max_bitrate = 0;
546     }
547
548     if( h->param.b_interlaced && h->param.i_slice_max_size )
549     {
550         x264_log( h, X264_LOG_WARNING, "interlaced + slice-max-size is not implemented\n" );
551         h->param.i_slice_max_size = 0;
552     }
553     if( h->param.b_interlaced && h->param.i_slice_max_mbs )
554     {
555         x264_log( h, X264_LOG_WARNING, "interlaced + slice-max-mbs is not implemented\n" );
556         h->param.i_slice_max_mbs = 0;
557     }
558     int max_slices = (h->param.i_height+((16<<h->param.b_interlaced)-1))/(16<<h->param.b_interlaced);
559     if( h->param.b_sliced_threads )
560         h->param.i_slice_count = x264_clip3( h->param.i_threads, 0, max_slices );
561     else
562     {
563         h->param.i_slice_count = x264_clip3( h->param.i_slice_count, 0, max_slices );
564         h->param.i_slice_max_size = X264_MAX( h->param.i_slice_max_size, 0 );
565         h->param.i_slice_max_mbs = X264_MAX( h->param.i_slice_max_mbs, 0 );
566         if( h->param.i_slice_max_mbs || h->param.i_slice_max_size )
567             h->param.i_slice_count = 0;
568     }
569
570     h->param.i_frame_reference = x264_clip3( h->param.i_frame_reference, 1, 16 );
571     h->param.i_dpb_size = x264_clip3( h->param.i_dpb_size, 1, 16 );
572     h->param.i_keyint_max = x264_clip3( h->param.i_keyint_max, 1, X264_KEYINT_MAX_INFINITE );
573     if( h->param.i_scenecut_threshold < 0 )
574         h->param.i_scenecut_threshold = 0;
575     if( !h->param.analyse.i_subpel_refine && h->param.analyse.i_direct_mv_pred > X264_DIRECT_PRED_SPATIAL )
576     {
577         x264_log( h, X264_LOG_WARNING, "subme=0 + direct=temporal is not supported\n" );
578         h->param.analyse.i_direct_mv_pred = X264_DIRECT_PRED_SPATIAL;
579     }
580     h->param.i_bframe = x264_clip3( h->param.i_bframe, 0, X264_MIN( X264_BFRAME_MAX, h->param.i_keyint_max-1 ) );
581     h->param.i_open_gop = x264_clip3( h->param.i_open_gop, X264_OPEN_GOP_NONE, X264_OPEN_GOP_BLURAY );
582     if( h->param.i_keyint_max == 1 )
583         h->param.b_intra_refresh = 0;
584     h->param.i_bframe_bias = x264_clip3( h->param.i_bframe_bias, -90, 100 );
585     if( h->param.i_bframe <= 1 )
586         h->param.i_bframe_pyramid = X264_B_PYRAMID_NONE;
587     h->param.i_bframe_pyramid = x264_clip3( h->param.i_bframe_pyramid, X264_B_PYRAMID_NONE, X264_B_PYRAMID_NORMAL );
588     if( !h->param.i_bframe )
589     {
590         h->param.i_bframe_adaptive = X264_B_ADAPT_NONE;
591         h->param.analyse.i_direct_mv_pred = 0;
592         h->param.analyse.b_weighted_bipred = 0;
593         h->param.i_open_gop = X264_OPEN_GOP_NONE;
594     }
595     if( h->param.b_intra_refresh && h->param.i_bframe_pyramid == X264_B_PYRAMID_NORMAL )
596     {
597         x264_log( h, X264_LOG_WARNING, "b-pyramid normal + intra-refresh is not supported\n" );
598         h->param.i_bframe_pyramid = X264_B_PYRAMID_STRICT;
599     }
600     if( h->param.b_intra_refresh && (h->param.i_frame_reference > 1 || h->param.i_dpb_size > 1) )
601     {
602         x264_log( h, X264_LOG_WARNING, "ref > 1 + intra-refresh is not supported\n" );
603         h->param.i_frame_reference = 1;
604         h->param.i_dpb_size = 1;
605     }
606     if( h->param.b_intra_refresh && h->param.i_open_gop )
607     {
608         x264_log( h, X264_LOG_WARNING, "intra-refresh is not compatible with open-gop\n" );
609         h->param.i_open_gop = X264_OPEN_GOP_NONE;
610     }
611     if( h->param.i_keyint_min == X264_KEYINT_MIN_AUTO )
612         h->param.i_keyint_min = h->param.i_keyint_max / 10;
613     h->param.i_keyint_min = x264_clip3( h->param.i_keyint_min, 1, h->param.i_keyint_max/2+1 );
614     h->param.rc.i_lookahead = x264_clip3( h->param.rc.i_lookahead, 0, X264_LOOKAHEAD_MAX );
615     {
616         int maxrate = X264_MAX( h->param.rc.i_vbv_max_bitrate, h->param.rc.i_bitrate );
617         float bufsize = maxrate ? (float)h->param.rc.i_vbv_buffer_size / maxrate : 0;
618         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;
619         h->param.rc.i_lookahead = X264_MIN( h->param.rc.i_lookahead, X264_MAX( h->param.i_keyint_max, bufsize*fps ) );
620     }
621
622     if( !h->param.i_timebase_num || !h->param.i_timebase_den )
623     {
624         h->param.i_timebase_num = h->param.i_fps_den;
625         h->param.i_timebase_den = h->param.i_fps_num;
626     }
627
628     h->param.rc.f_qcompress = x264_clip3f( h->param.rc.f_qcompress, 0.0, 1.0 );
629     if( h->param.i_keyint_max == 1 || h->param.rc.f_qcompress == 1 )
630         h->param.rc.b_mb_tree = 0;
631     if( (!h->param.b_intra_refresh && h->param.i_keyint_max != X264_KEYINT_MAX_INFINITE) &&
632         !h->param.rc.i_lookahead && h->param.rc.b_mb_tree )
633     {
634         x264_log( h, X264_LOG_WARNING, "lookaheadless mb-tree requires intra refresh or infinite keyint\n" );
635         h->param.rc.b_mb_tree = 0;
636     }
637     if( h->param.rc.b_stat_read )
638         h->param.rc.i_lookahead = 0;
639 #if HAVE_PTHREAD
640     if( h->param.i_sync_lookahead < 0 )
641         h->param.i_sync_lookahead = h->param.i_bframe + 1;
642     h->param.i_sync_lookahead = X264_MIN( h->param.i_sync_lookahead, X264_LOOKAHEAD_MAX );
643     if( h->param.rc.b_stat_read || h->i_thread_frames == 1 )
644         h->param.i_sync_lookahead = 0;
645 #else
646     h->param.i_sync_lookahead = 0;
647 #endif
648
649     h->param.i_deblocking_filter_alphac0 = x264_clip3( h->param.i_deblocking_filter_alphac0, -6, 6 );
650     h->param.i_deblocking_filter_beta    = x264_clip3( h->param.i_deblocking_filter_beta, -6, 6 );
651     h->param.analyse.i_luma_deadzone[0] = x264_clip3( h->param.analyse.i_luma_deadzone[0], 0, 32 );
652     h->param.analyse.i_luma_deadzone[1] = x264_clip3( h->param.analyse.i_luma_deadzone[1], 0, 32 );
653
654     h->param.i_cabac_init_idc = x264_clip3( h->param.i_cabac_init_idc, 0, 2 );
655
656     if( h->param.i_cqm_preset < X264_CQM_FLAT || h->param.i_cqm_preset > X264_CQM_CUSTOM )
657         h->param.i_cqm_preset = X264_CQM_FLAT;
658
659     if( h->param.analyse.i_me_method < X264_ME_DIA ||
660         h->param.analyse.i_me_method > X264_ME_TESA )
661         h->param.analyse.i_me_method = X264_ME_HEX;
662     if( h->param.analyse.i_me_range < 4 )
663         h->param.analyse.i_me_range = 4;
664     if( h->param.analyse.i_me_range > 16 && h->param.analyse.i_me_method <= X264_ME_HEX )
665         h->param.analyse.i_me_range = 16;
666     if( h->param.analyse.i_me_method == X264_ME_TESA &&
667         (h->mb.b_lossless || h->param.analyse.i_subpel_refine <= 1) )
668         h->param.analyse.i_me_method = X264_ME_ESA;
669     h->param.analyse.i_subpel_refine = x264_clip3( h->param.analyse.i_subpel_refine, 0, 10 );
670     h->param.analyse.b_mixed_references = h->param.analyse.b_mixed_references && h->param.i_frame_reference > 1;
671     h->param.analyse.inter &= X264_ANALYSE_PSUB16x16|X264_ANALYSE_PSUB8x8|X264_ANALYSE_BSUB16x16|
672                               X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
673     h->param.analyse.intra &= X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
674     if( !(h->param.analyse.inter & X264_ANALYSE_PSUB16x16) )
675         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
676     if( !h->param.analyse.b_transform_8x8 )
677     {
678         h->param.analyse.inter &= ~X264_ANALYSE_I8x8;
679         h->param.analyse.intra &= ~X264_ANALYSE_I8x8;
680     }
681     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -12, 12);
682     if( !h->param.b_cabac )
683         h->param.analyse.i_trellis = 0;
684     h->param.analyse.i_trellis = x264_clip3( h->param.analyse.i_trellis, 0, 2 );
685     if( !h->param.analyse.b_psy )
686     {
687         h->param.analyse.f_psy_rd = 0;
688         h->param.analyse.f_psy_trellis = 0;
689     }
690     if( !h->param.analyse.i_trellis )
691         h->param.analyse.f_psy_trellis = 0;
692     h->param.analyse.f_psy_rd = x264_clip3f( h->param.analyse.f_psy_rd, 0, 10 );
693     h->param.analyse.f_psy_trellis = x264_clip3f( h->param.analyse.f_psy_trellis, 0, 10 );
694     if( h->param.analyse.i_subpel_refine < 6 )
695         h->param.analyse.f_psy_rd = 0;
696     h->mb.i_psy_rd = FIX8( h->param.analyse.f_psy_rd );
697     /* Psy RDO increases overall quantizers to improve the quality of luma--this indirectly hurts chroma quality */
698     /* so we lower the chroma QP offset to compensate */
699     /* This can be triggered repeatedly on multiple calls to parameter_validate, but since encoding
700      * uses the pps chroma qp offset not the param chroma qp offset, this is not a problem. */
701     if( h->mb.i_psy_rd )
702         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_rd < 0.25 ? 1 : 2;
703     h->mb.i_psy_trellis = FIX8( h->param.analyse.f_psy_trellis / 4 );
704     /* Psy trellis has a similar effect. */
705     if( h->mb.i_psy_trellis )
706         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_trellis < 0.25 ? 1 : 2;
707     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -12, 12);
708     h->param.rc.i_aq_mode = x264_clip3( h->param.rc.i_aq_mode, 0, 2 );
709     h->param.rc.f_aq_strength = x264_clip3f( h->param.rc.f_aq_strength, 0, 3 );
710     if( h->param.rc.f_aq_strength == 0 )
711         h->param.rc.i_aq_mode = 0;
712     /* MB-tree requires AQ to be on, even if the strength is zero. */
713     if( !h->param.rc.i_aq_mode && h->param.rc.b_mb_tree )
714     {
715         h->param.rc.i_aq_mode = 1;
716         h->param.rc.f_aq_strength = 0;
717     }
718     h->param.analyse.i_noise_reduction = x264_clip3( h->param.analyse.i_noise_reduction, 0, 1<<16 );
719     if( h->param.analyse.i_subpel_refine == 10 && (h->param.analyse.i_trellis != 2 || !h->param.rc.i_aq_mode) )
720         h->param.analyse.i_subpel_refine = 9;
721
722     {
723         const x264_level_t *l = x264_levels;
724         if( h->param.i_level_idc < 0 )
725         {
726             int maxrate_bak = h->param.rc.i_vbv_max_bitrate;
727             if( h->param.rc.i_rc_method == X264_RC_ABR && h->param.rc.i_vbv_buffer_size <= 0 )
728                 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate * 2;
729             h->sps = h->sps_array;
730             x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
731             do h->param.i_level_idc = l->level_idc;
732                 while( l[1].level_idc && x264_validate_levels( h, 0 ) && l++ );
733             h->param.rc.i_vbv_max_bitrate = maxrate_bak;
734         }
735         else
736         {
737             while( l->level_idc && l->level_idc != h->param.i_level_idc )
738                 l++;
739             if( l->level_idc == 0 )
740             {
741                 x264_log( h, X264_LOG_ERROR, "invalid level_idc: %d\n", h->param.i_level_idc );
742                 return -1;
743             }
744         }
745         if( h->param.analyse.i_mv_range <= 0 )
746             h->param.analyse.i_mv_range = l->mv_range >> h->param.b_interlaced;
747         else
748             h->param.analyse.i_mv_range = x264_clip3(h->param.analyse.i_mv_range, 32, 512 >> h->param.b_interlaced);
749     }
750
751     h->param.analyse.i_weighted_pred = x264_clip3( h->param.analyse.i_weighted_pred, 0, X264_WEIGHTP_SMART );
752     if( !h->param.analyse.i_weighted_pred && h->param.rc.b_mb_tree && h->param.analyse.b_psy && !h->param.b_interlaced )
753         h->param.analyse.i_weighted_pred = X264_WEIGHTP_FAKE;
754
755     if( h->i_thread_frames > 1 )
756     {
757         int r = h->param.analyse.i_mv_range_thread;
758         int r2;
759         if( r <= 0 )
760         {
761             // half of the available space is reserved and divided evenly among the threads,
762             // the rest is allocated to whichever thread is far enough ahead to use it.
763             // reserving more space increases quality for some videos, but costs more time
764             // in thread synchronization.
765             int max_range = (h->param.i_height + X264_THREAD_HEIGHT) / h->i_thread_frames - X264_THREAD_HEIGHT;
766             r = max_range / 2;
767         }
768         r = X264_MAX( r, h->param.analyse.i_me_range );
769         r = X264_MIN( r, h->param.analyse.i_mv_range );
770         // round up to use the whole mb row
771         r2 = (r & ~15) + ((-X264_THREAD_HEIGHT) & 15);
772         if( r2 < r )
773             r2 += 16;
774         x264_log( h, X264_LOG_DEBUG, "using mv_range_thread = %d\n", r2 );
775         h->param.analyse.i_mv_range_thread = r2;
776     }
777
778     if( h->param.rc.f_qblur < 0 )
779         h->param.rc.f_qblur = 0;
780     if( h->param.rc.f_complexity_blur < 0 )
781         h->param.rc.f_complexity_blur = 0;
782
783     h->param.i_sps_id &= 31;
784
785     if( h->param.i_log_level < X264_LOG_INFO )
786     {
787         h->param.analyse.b_psnr = 0;
788         h->param.analyse.b_ssim = 0;
789     }
790
791     if( h->param.b_interlaced )
792         h->param.b_pic_struct = 1;
793
794     if( h->param.i_nal_hrd && !h->param.rc.i_vbv_buffer_size )
795     {
796         x264_log( h, X264_LOG_WARNING, "NAL HRD parameters require VBV parameters\n" );
797         h->param.i_nal_hrd = X264_NAL_HRD_NONE;
798     }
799
800     if( h->param.i_nal_hrd == X264_NAL_HRD_CBR &&
801        (h->param.rc.i_bitrate != h->param.rc.i_vbv_max_bitrate || !h->param.rc.i_vbv_max_bitrate) )
802     {
803         x264_log( h, X264_LOG_WARNING, "CBR HRD requires constant bitrate\n" );
804         h->param.i_nal_hrd = X264_NAL_HRD_VBR;
805     }
806
807     /* ensure the booleans are 0 or 1 so they can be used in math */
808 #define BOOLIFY(x) h->param.x = !!h->param.x
809     BOOLIFY( b_cabac );
810     BOOLIFY( b_constrained_intra );
811     BOOLIFY( b_deblocking_filter );
812     BOOLIFY( b_deterministic );
813     BOOLIFY( b_sliced_threads );
814     BOOLIFY( b_interlaced );
815     BOOLIFY( b_intra_refresh );
816     BOOLIFY( b_visualize );
817     BOOLIFY( b_aud );
818     BOOLIFY( b_repeat_headers );
819     BOOLIFY( b_annexb );
820     BOOLIFY( b_vfr_input );
821     BOOLIFY( b_pic_struct );
822     BOOLIFY( b_fake_interlaced );
823     BOOLIFY( analyse.b_transform_8x8 );
824     BOOLIFY( analyse.b_weighted_bipred );
825     BOOLIFY( analyse.b_chroma_me );
826     BOOLIFY( analyse.b_mixed_references );
827     BOOLIFY( analyse.b_fast_pskip );
828     BOOLIFY( analyse.b_dct_decimate );
829     BOOLIFY( analyse.b_psy );
830     BOOLIFY( analyse.b_psnr );
831     BOOLIFY( analyse.b_ssim );
832     BOOLIFY( rc.b_stat_write );
833     BOOLIFY( rc.b_stat_read );
834     BOOLIFY( rc.b_mb_tree );
835 #undef BOOLIFY
836
837     return 0;
838 }
839
840 static void mbcmp_init( x264_t *h )
841 {
842     int satd = !h->mb.b_lossless && h->param.analyse.i_subpel_refine > 1;
843     memcpy( h->pixf.mbcmp, satd ? h->pixf.satd : h->pixf.sad_aligned, sizeof(h->pixf.mbcmp) );
844     memcpy( h->pixf.mbcmp_unaligned, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.mbcmp_unaligned) );
845     h->pixf.intra_mbcmp_x3_16x16 = satd ? h->pixf.intra_satd_x3_16x16 : h->pixf.intra_sad_x3_16x16;
846     h->pixf.intra_mbcmp_x3_8x8c = satd ? h->pixf.intra_satd_x3_8x8c : h->pixf.intra_sad_x3_8x8c;
847     h->pixf.intra_mbcmp_x3_8x8 = satd ? h->pixf.intra_sa8d_x3_8x8 : h->pixf.intra_sad_x3_8x8;
848     h->pixf.intra_mbcmp_x3_4x4 = satd ? h->pixf.intra_satd_x3_4x4 : h->pixf.intra_sad_x3_4x4;
849     satd &= h->param.analyse.i_me_method == X264_ME_TESA;
850     memcpy( h->pixf.fpelcmp, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.fpelcmp) );
851     memcpy( h->pixf.fpelcmp_x3, satd ? h->pixf.satd_x3 : h->pixf.sad_x3, sizeof(h->pixf.fpelcmp_x3) );
852     memcpy( h->pixf.fpelcmp_x4, satd ? h->pixf.satd_x4 : h->pixf.sad_x4, sizeof(h->pixf.fpelcmp_x4) );
853 }
854
855 static void x264_set_aspect_ratio( x264_t *h, x264_param_t *param, int initial )
856 {
857     /* VUI */
858     if( param->vui.i_sar_width > 0 && param->vui.i_sar_height > 0 )
859     {
860         uint32_t i_w = param->vui.i_sar_width;
861         uint32_t i_h = param->vui.i_sar_height;
862         uint32_t old_w = h->param.vui.i_sar_width;
863         uint32_t old_h = h->param.vui.i_sar_height;
864
865         x264_reduce_fraction( &i_w, &i_h );
866
867         while( i_w > 65535 || i_h > 65535 )
868         {
869             i_w /= 2;
870             i_h /= 2;
871         }
872
873         x264_reduce_fraction( &i_w, &i_h );
874
875         if( i_w != old_w || i_h != old_h || initial )
876         {
877             h->param.vui.i_sar_width = 0;
878             h->param.vui.i_sar_height = 0;
879             if( i_w == 0 || i_h == 0 )
880                 x264_log( h, X264_LOG_WARNING, "cannot create valid sample aspect ratio\n" );
881             else
882             {
883                 x264_log( h, initial?X264_LOG_INFO:X264_LOG_DEBUG, "using SAR=%d/%d\n", i_w, i_h );
884                 h->param.vui.i_sar_width = i_w;
885                 h->param.vui.i_sar_height = i_h;
886             }
887         }
888     }
889 }
890
891 /****************************************************************************
892  * x264_encoder_open:
893  ****************************************************************************/
894 x264_t *x264_encoder_open( x264_param_t *param )
895 {
896     x264_t *h;
897     char buf[1000], *p;
898     int qp, i_slicetype_length;
899
900     CHECKED_MALLOCZERO( h, sizeof(x264_t) );
901
902     /* Create a copy of param */
903     memcpy( &h->param, param, sizeof(x264_param_t) );
904
905     if( param->param_free )
906         param->param_free( param );
907
908     if( x264_validate_parameters( h ) < 0 )
909         goto fail;
910
911     if( h->param.psz_cqm_file )
912         if( x264_cqm_parse_file( h, h->param.psz_cqm_file ) < 0 )
913             goto fail;
914
915     if( h->param.rc.psz_stat_out )
916         h->param.rc.psz_stat_out = strdup( h->param.rc.psz_stat_out );
917     if( h->param.rc.psz_stat_in )
918         h->param.rc.psz_stat_in = strdup( h->param.rc.psz_stat_in );
919
920     x264_set_aspect_ratio( h, &h->param, 1 );
921
922     x264_reduce_fraction( &h->param.i_fps_num, &h->param.i_fps_den );
923     x264_reduce_fraction( &h->param.i_timebase_num, &h->param.i_timebase_den );
924
925     /* Init x264_t */
926     h->i_frame = -1;
927     h->i_frame_num = 0;
928     h->i_idr_pic_id = 0;
929     uint64_t new_timebase_den = h->param.i_timebase_den;
930     if( h->param.b_dts_compress )
931     {
932         /* h->i_dts_compress_multiplier == h->frames.i_bframe_delay + 1 */
933         h->i_dts_compress_multiplier = h->param.i_bframe ? (h->param.i_bframe_pyramid ? 3 : 2) : 1;
934         if( h->i_dts_compress_multiplier != 1 )
935         {
936             new_timebase_den = h->param.i_timebase_den * h->i_dts_compress_multiplier;
937             x264_log( h, X264_LOG_DEBUG, "DTS compresion changed timebase: %u/%u -> %u/%"PRIu64"\n",
938                       h->param.i_timebase_num, h->param.i_timebase_den,
939                       h->param.i_timebase_num, new_timebase_den );
940         }
941     }
942     else
943         h->i_dts_compress_multiplier = 1;
944
945     if( new_timebase_den * 2 > UINT32_MAX )
946     {
947         x264_log( h, X264_LOG_ERROR, "Effective timebase denominator %"PRIu64" exceeds H.264 maximum\n", new_timebase_den );
948         goto fail;
949     }
950     h->param.i_timebase_den = new_timebase_den;
951
952     h->sps = &h->sps_array[0];
953     x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
954
955     h->pps = &h->pps_array[0];
956     x264_pps_init( h->pps, h->param.i_sps_id, &h->param, h->sps );
957
958     x264_validate_levels( h, 1 );
959
960     h->chroma_qp_table = i_chroma_qp_table + 12 + h->pps->i_chroma_qp_index_offset;
961
962     if( x264_cqm_init( h ) < 0 )
963         goto fail;
964
965     h->mb.i_mb_width = h->sps->i_mb_width;
966     h->mb.i_mb_height = h->sps->i_mb_height;
967     h->mb.i_mb_count = h->mb.i_mb_width * h->mb.i_mb_height;
968
969     /* Init frames. */
970     if( h->param.i_bframe_adaptive == X264_B_ADAPT_TRELLIS && !h->param.rc.b_stat_read )
971         h->frames.i_delay = X264_MAX(h->param.i_bframe,3)*4;
972     else
973         h->frames.i_delay = h->param.i_bframe;
974     if( h->param.rc.b_mb_tree || h->param.rc.i_vbv_buffer_size )
975         h->frames.i_delay = X264_MAX( h->frames.i_delay, h->param.rc.i_lookahead );
976     i_slicetype_length = h->frames.i_delay;
977     h->frames.i_delay += h->i_thread_frames - 1;
978     h->frames.i_delay += h->param.i_sync_lookahead;
979     h->frames.i_delay += h->param.b_vfr_input && (h->param.rc.i_rc_method == X264_RC_ABR || h->param.rc.b_stat_write
980                                                  || h->param.rc.i_vbv_buffer_size);
981     h->frames.i_bframe_delay = h->param.i_bframe ? (h->param.i_bframe_pyramid ? 2 : 1) : 0;
982
983     h->frames.i_max_ref0 = h->param.i_frame_reference;
984     h->frames.i_max_ref1 = X264_MIN( h->sps->vui.i_num_reorder_frames, h->param.i_frame_reference );
985     h->frames.i_max_dpb  = h->sps->vui.i_max_dec_frame_buffering;
986     h->frames.b_have_lowres = !h->param.rc.b_stat_read
987         && ( h->param.rc.i_rc_method == X264_RC_ABR
988           || h->param.rc.i_rc_method == X264_RC_CRF
989           || h->param.i_bframe_adaptive
990           || h->param.i_scenecut_threshold
991           || h->param.rc.b_mb_tree
992           || h->param.analyse.i_weighted_pred == X264_WEIGHTP_SMART );
993     h->frames.b_have_lowres |= h->param.rc.b_stat_read && h->param.rc.i_vbv_buffer_size > 0;
994     h->frames.b_have_sub8x8_esa = !!(h->param.analyse.inter & X264_ANALYSE_PSUB8x8);
995
996     h->frames.i_last_idr =
997     h->frames.i_last_keyframe = - h->param.i_keyint_max;
998     h->frames.i_input    = 0;
999     h->frames.i_largest_pts = h->frames.i_second_largest_pts = -1;
1000     h->frames.i_poc_last_open_gop = -1;
1001
1002     CHECKED_MALLOCZERO( h->frames.unused[0], (h->frames.i_delay + 3) * sizeof(x264_frame_t *) );
1003     /* Allocate room for max refs plus a few extra just in case. */
1004     CHECKED_MALLOCZERO( h->frames.unused[1], (h->i_thread_frames + 20) * sizeof(x264_frame_t *) );
1005     CHECKED_MALLOCZERO( h->frames.current, (h->param.i_sync_lookahead + h->param.i_bframe
1006                         + h->i_thread_frames + 3) * sizeof(x264_frame_t *) );
1007     if( h->param.analyse.i_weighted_pred > 0 )
1008         CHECKED_MALLOCZERO( h->frames.blank_unused, h->i_thread_frames * 4 * sizeof(x264_frame_t *) );
1009     h->i_ref0 = 0;
1010     h->i_ref1 = 0;
1011     h->i_cpb_delay = h->i_coded_fields = h->i_disp_fields = h->i_prev_duration = 0;
1012     h->i_disp_fields_last_frame = -1;
1013     x264_rdo_init();
1014
1015     /* init CPU functions */
1016     x264_predict_16x16_init( h->param.cpu, h->predict_16x16 );
1017     x264_predict_8x8c_init( h->param.cpu, h->predict_8x8c );
1018     x264_predict_8x8_init( h->param.cpu, h->predict_8x8, &h->predict_8x8_filter );
1019     x264_predict_4x4_init( h->param.cpu, h->predict_4x4 );
1020     if( !h->param.b_cabac )
1021         x264_init_vlc_tables();
1022     x264_pixel_init( h->param.cpu, &h->pixf );
1023     x264_dct_init( h->param.cpu, &h->dctf );
1024     x264_zigzag_init( h->param.cpu, &h->zigzagf, h->param.b_interlaced );
1025     x264_mc_init( h->param.cpu, &h->mc );
1026     x264_quant_init( h, h->param.cpu, &h->quantf );
1027     x264_deblock_init( h->param.cpu, &h->loopf );
1028     x264_bitstream_init( h->param.cpu, &h->bsf );
1029     x264_dct_init_weights();
1030
1031     mbcmp_init( h );
1032
1033     p = buf + sprintf( buf, "using cpu capabilities:" );
1034     for( int i = 0; x264_cpu_names[i].flags; i++ )
1035     {
1036         if( !strcmp(x264_cpu_names[i].name, "SSE2")
1037             && h->param.cpu & (X264_CPU_SSE2_IS_FAST|X264_CPU_SSE2_IS_SLOW) )
1038             continue;
1039         if( !strcmp(x264_cpu_names[i].name, "SSE3")
1040             && (h->param.cpu & X264_CPU_SSSE3 || !(h->param.cpu & X264_CPU_CACHELINE_64)) )
1041             continue;
1042         if( !strcmp(x264_cpu_names[i].name, "SSE4.1")
1043             && (h->param.cpu & X264_CPU_SSE42) )
1044             continue;
1045         if( (h->param.cpu & x264_cpu_names[i].flags) == x264_cpu_names[i].flags
1046             && (!i || x264_cpu_names[i].flags != x264_cpu_names[i-1].flags) )
1047             p += sprintf( p, " %s", x264_cpu_names[i].name );
1048     }
1049     if( !h->param.cpu )
1050         p += sprintf( p, " none!" );
1051     x264_log( h, X264_LOG_INFO, "%s\n", buf );
1052
1053     for( qp = h->param.rc.i_qp_min; qp <= h->param.rc.i_qp_max; qp++ )
1054         if( x264_analyse_init_costs( h, qp ) )
1055             goto fail;
1056     if( x264_analyse_init_costs( h, X264_LOOKAHEAD_QP ) )
1057         goto fail;
1058
1059     static const uint16_t cost_mv_correct[7] = { 24, 47, 95, 189, 379, 757, 1515 };
1060     /* Checks for known miscompilation issues. */
1061     if( h->cost_mv[x264_lambda_tab[X264_LOOKAHEAD_QP]][2013] != cost_mv_correct[BIT_DEPTH-8] )
1062     {
1063         x264_log( h, X264_LOG_ERROR, "MV cost test failed: x264 has been miscompiled!\n" );
1064         goto fail;
1065     }
1066
1067     /* Must be volatile or else GCC will optimize it out. */
1068     volatile int temp = 392;
1069     if( x264_clz( temp ) != 23 )
1070     {
1071         x264_log( h, X264_LOG_ERROR, "CLZ test failed: x264 has been miscompiled!\n" );
1072 #if ARCH_X86 || ARCH_X86_64
1073         x264_log( h, X264_LOG_ERROR, "Are you attempting to run an SSE4a-targeted build on a CPU that\n" );
1074         x264_log( h, X264_LOG_ERROR, "doesn't support it?\n" );
1075 #endif
1076         goto fail;
1077     }
1078
1079     h->out.i_nal = 0;
1080     h->out.i_bitstream = X264_MAX( 1000000, h->param.i_width * h->param.i_height * 4
1081         * ( h->param.rc.i_rc_method == X264_RC_ABR ? pow( 0.95, h->param.rc.i_qp_min )
1082           : pow( 0.95, h->param.rc.i_qp_constant ) * X264_MAX( 1, h->param.rc.f_ip_factor )));
1083
1084     CHECKED_MALLOC( h->nal_buffer, h->out.i_bitstream * 3/2 + 4 );
1085     h->nal_buffer_size = h->out.i_bitstream * 3/2 + 4;
1086
1087     if( h->param.i_threads > 1 &&
1088         x264_threadpool_init( &h->threadpool, h->param.i_threads, (void*)x264_encoder_thread_init, h ) )
1089         goto fail;
1090
1091     h->thread[0] = h;
1092     for( int i = 1; i < h->param.i_threads + !!h->param.i_sync_lookahead; i++ )
1093         CHECKED_MALLOC( h->thread[i], sizeof(x264_t) );
1094
1095     if( x264_lookahead_init( h, i_slicetype_length ) )
1096         goto fail;
1097
1098     for( int i = 0; i < h->param.i_threads; i++ )
1099     {
1100         int init_nal_count = h->param.i_slice_count + 3;
1101         int allocate_threadlocal_data = !h->param.b_sliced_threads || !i;
1102         if( i > 0 )
1103             *h->thread[i] = *h;
1104
1105         if( allocate_threadlocal_data )
1106         {
1107             h->thread[i]->fdec = x264_frame_pop_unused( h, 1 );
1108             if( !h->thread[i]->fdec )
1109                 goto fail;
1110         }
1111         else
1112             h->thread[i]->fdec = h->thread[0]->fdec;
1113
1114         CHECKED_MALLOC( h->thread[i]->out.p_bitstream, h->out.i_bitstream );
1115         /* Start each thread with room for init_nal_count NAL units; it'll realloc later if needed. */
1116         CHECKED_MALLOC( h->thread[i]->out.nal, init_nal_count*sizeof(x264_nal_t) );
1117         h->thread[i]->out.i_nals_allocated = init_nal_count;
1118
1119         if( allocate_threadlocal_data && x264_macroblock_cache_allocate( h->thread[i] ) < 0 )
1120             goto fail;
1121     }
1122
1123     for( int i = 0; i < h->param.i_threads; i++ )
1124         if( x264_macroblock_thread_allocate( h->thread[i], 0 ) < 0 )
1125             goto fail;
1126
1127     if( x264_ratecontrol_new( h ) < 0 )
1128         goto fail;
1129
1130     if( h->param.i_nal_hrd )
1131     {
1132         x264_log( h, X264_LOG_DEBUG, "HRD bitrate: %i bits/sec\n", h->sps->vui.hrd.i_bit_rate_unscaled );
1133         x264_log( h, X264_LOG_DEBUG, "CPB size: %i bits\n", h->sps->vui.hrd.i_cpb_size_unscaled );
1134     }
1135
1136     if( h->param.psz_dump_yuv )
1137     {
1138         /* create or truncate the reconstructed video file */
1139         FILE *f = fopen( h->param.psz_dump_yuv, "w" );
1140         if( !f )
1141         {
1142             x264_log( h, X264_LOG_ERROR, "dump_yuv: can't write to %s\n", h->param.psz_dump_yuv );
1143             goto fail;
1144         }
1145         else if( !x264_is_regular_file( f ) )
1146         {
1147             x264_log( h, X264_LOG_ERROR, "dump_yuv: incompatible with non-regular file %s\n", h->param.psz_dump_yuv );
1148             goto fail;
1149         }
1150         fclose( f );
1151     }
1152
1153     const char *profile = h->sps->i_profile_idc == PROFILE_BASELINE ? "Baseline" :
1154                           h->sps->i_profile_idc == PROFILE_MAIN ? "Main" :
1155                           h->sps->i_profile_idc == PROFILE_HIGH ? "High" :
1156                           h->sps->i_profile_idc == PROFILE_HIGH10 ? "High 10" :
1157                           "High 4:4:4 Predictive";
1158
1159     if( h->sps->i_profile_idc < PROFILE_HIGH10 )
1160     {
1161         x264_log( h, X264_LOG_INFO, "profile %s, level %d.%d\n",
1162             profile, h->sps->i_level_idc/10, h->sps->i_level_idc%10 );
1163     }
1164     else
1165     {
1166         x264_log( h, X264_LOG_INFO, "profile %s, level %d.%d, bit depth %d\n",
1167             profile, h->sps->i_level_idc/10, h->sps->i_level_idc%10, BIT_DEPTH );
1168     }
1169
1170     return h;
1171 fail:
1172     x264_free( h );
1173     return NULL;
1174 }
1175
1176 /****************************************************************************
1177  * x264_encoder_reconfig:
1178  ****************************************************************************/
1179 int x264_encoder_reconfig( x264_t *h, x264_param_t *param )
1180 {
1181     int rc_reconfig = 0;
1182     h = h->thread[h->thread[0]->i_thread_phase];
1183     x264_set_aspect_ratio( h, param, 0 );
1184 #define COPY(var) h->param.var = param->var
1185     COPY( i_frame_reference ); // but never uses more refs than initially specified
1186     COPY( i_bframe_bias );
1187     if( h->param.i_scenecut_threshold )
1188         COPY( i_scenecut_threshold ); // can't turn it on or off, only vary the threshold
1189     COPY( b_deblocking_filter );
1190     COPY( i_deblocking_filter_alphac0 );
1191     COPY( i_deblocking_filter_beta );
1192     COPY( analyse.inter );
1193     COPY( analyse.intra );
1194     COPY( analyse.i_direct_mv_pred );
1195     /* Scratch buffer prevents me_range from being increased for esa/tesa */
1196     if( h->param.analyse.i_me_method < X264_ME_ESA || param->analyse.i_me_range < h->param.analyse.i_me_range )
1197         COPY( analyse.i_me_range );
1198     COPY( analyse.i_noise_reduction );
1199     /* We can't switch out of subme=0 during encoding. */
1200     if( h->param.analyse.i_subpel_refine )
1201         COPY( analyse.i_subpel_refine );
1202     COPY( analyse.i_trellis );
1203     COPY( analyse.b_chroma_me );
1204     COPY( analyse.b_dct_decimate );
1205     COPY( analyse.b_fast_pskip );
1206     COPY( analyse.b_mixed_references );
1207     COPY( analyse.f_psy_rd );
1208     COPY( analyse.f_psy_trellis );
1209     // can only twiddle these if they were enabled to begin with:
1210     if( h->param.analyse.i_me_method >= X264_ME_ESA || param->analyse.i_me_method < X264_ME_ESA )
1211         COPY( analyse.i_me_method );
1212     if( h->param.analyse.i_me_method >= X264_ME_ESA && !h->frames.b_have_sub8x8_esa )
1213         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
1214     if( h->pps->b_transform_8x8_mode )
1215         COPY( analyse.b_transform_8x8 );
1216     if( h->frames.i_max_ref1 > 1 )
1217         COPY( i_bframe_pyramid );
1218     COPY( i_slice_max_size );
1219     COPY( i_slice_max_mbs );
1220     COPY( i_slice_count );
1221
1222     /* VBV can't be turned on if it wasn't on to begin with */
1223     if( h->param.rc.i_vbv_max_bitrate > 0 && h->param.rc.i_vbv_buffer_size > 0 &&
1224           param->rc.i_vbv_max_bitrate > 0 &&   param->rc.i_vbv_buffer_size > 0 )
1225     {
1226         COPY( rc.i_vbv_max_bitrate );
1227         COPY( rc.i_vbv_buffer_size );
1228         COPY( rc.i_bitrate );
1229         rc_reconfig = 1;
1230     }
1231     if( h->param.rc.f_rf_constant != param->rc.f_rf_constant )
1232     {
1233         COPY( rc.f_rf_constant );
1234         rc_reconfig = 1;
1235     }
1236     if( h->param.rc.f_rf_constant_max != param->rc.f_rf_constant_max )
1237     {
1238         COPY( rc.f_rf_constant_max );
1239         rc_reconfig = 1;
1240     }
1241
1242 #undef COPY
1243
1244     mbcmp_init( h );
1245
1246     int ret = x264_validate_parameters( h );
1247
1248     /* Supported reconfiguration options (1-pass only):
1249      * vbv-maxrate
1250      * vbv-bufsize
1251      * crf
1252      * bitrate (CBR only) */
1253     if( !ret && rc_reconfig )
1254         x264_ratecontrol_init_reconfigurable( h, 0 );
1255
1256     return ret;
1257 }
1258
1259 /****************************************************************************
1260  * x264_encoder_parameters:
1261  ****************************************************************************/
1262 void x264_encoder_parameters( x264_t *h, x264_param_t *param )
1263 {
1264     memcpy( param, &h->thread[h->i_thread_phase]->param, sizeof(x264_param_t) );
1265 }
1266
1267 /* internal usage */
1268 static void x264_nal_start( x264_t *h, int i_type, int i_ref_idc )
1269 {
1270     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1271
1272     nal->i_ref_idc        = i_ref_idc;
1273     nal->i_type           = i_type;
1274     nal->b_long_startcode = 1;
1275
1276     nal->i_payload= 0;
1277     nal->p_payload= &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8];
1278 }
1279
1280 /* if number of allocated nals is not enough, re-allocate a larger one. */
1281 static int x264_nal_check_buffer( x264_t *h )
1282 {
1283     if( h->out.i_nal >= h->out.i_nals_allocated )
1284     {
1285         x264_nal_t *new_out = x264_malloc( sizeof(x264_nal_t) * (h->out.i_nals_allocated*2) );
1286         if( !new_out )
1287             return -1;
1288         memcpy( new_out, h->out.nal, sizeof(x264_nal_t) * (h->out.i_nals_allocated) );
1289         x264_free( h->out.nal );
1290         h->out.nal = new_out;
1291         h->out.i_nals_allocated *= 2;
1292     }
1293     return 0;
1294 }
1295
1296 static int x264_nal_end( x264_t *h )
1297 {
1298     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1299     nal->i_payload = &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8] - nal->p_payload;
1300     if( h->param.nalu_process )
1301         h->param.nalu_process( h, nal );
1302     h->out.i_nal++;
1303
1304     return x264_nal_check_buffer( h );
1305 }
1306
1307 static int x264_encoder_encapsulate_nals( x264_t *h, int start )
1308 {
1309     int nal_size = 0, previous_nal_size = 0;
1310
1311     if( h->param.nalu_process )
1312     {
1313         for( int i = start; i < h->out.i_nal; i++ )
1314             nal_size += h->out.nal[i].i_payload;
1315         return nal_size;
1316     }
1317
1318     for( int i = 0; i < start; i++ )
1319         previous_nal_size += h->out.nal[i].i_payload;
1320
1321     for( int i = start; i < h->out.i_nal; i++ )
1322         nal_size += h->out.nal[i].i_payload;
1323
1324     /* Worst-case NAL unit escaping: reallocate the buffer if it's too small. */
1325     if( h->nal_buffer_size < nal_size * 3/2 + h->out.i_nal * 4 )
1326     {
1327         uint8_t *buf = x264_malloc( nal_size * 2 + h->out.i_nal * 4 );
1328         if( !buf )
1329             return -1;
1330         if( previous_nal_size )
1331             memcpy( buf, h->nal_buffer, previous_nal_size );
1332         x264_free( h->nal_buffer );
1333         h->nal_buffer = buf;
1334     }
1335
1336     uint8_t *nal_buffer = h->nal_buffer + previous_nal_size;
1337
1338     for( int i = start; i < h->out.i_nal; i++ )
1339     {
1340         h->out.nal[i].b_long_startcode = !i || h->out.nal[i].i_type == NAL_SPS || h->out.nal[i].i_type == NAL_PPS;
1341         x264_nal_encode( h, nal_buffer, &h->out.nal[i] );
1342         nal_buffer += h->out.nal[i].i_payload;
1343     }
1344
1345     x264_emms();
1346
1347     return nal_buffer - (h->nal_buffer + previous_nal_size);
1348 }
1349
1350 /****************************************************************************
1351  * x264_encoder_headers:
1352  ****************************************************************************/
1353 int x264_encoder_headers( x264_t *h, x264_nal_t **pp_nal, int *pi_nal )
1354 {
1355     int frame_size = 0;
1356     /* init bitstream context */
1357     h->out.i_nal = 0;
1358     bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
1359
1360     /* Write SEI, SPS and PPS. */
1361
1362     /* generate sequence parameters */
1363     x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
1364     x264_sps_write( &h->out.bs, h->sps );
1365     if( x264_nal_end( h ) )
1366         return -1;
1367
1368     /* generate picture parameters */
1369     x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
1370     x264_pps_write( &h->out.bs, h->pps );
1371     if( x264_nal_end( h ) )
1372         return -1;
1373
1374     /* identify ourselves */
1375     x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
1376     if( x264_sei_version_write( h, &h->out.bs ) )
1377         return -1;
1378     if( x264_nal_end( h ) )
1379         return -1;
1380
1381     frame_size = x264_encoder_encapsulate_nals( h, 0 );
1382
1383     /* now set output*/
1384     *pi_nal = h->out.i_nal;
1385     *pp_nal = &h->out.nal[0];
1386     h->out.i_nal = 0;
1387
1388     return frame_size;
1389 }
1390
1391 /* Check to see whether we have chosen a reference list ordering different
1392  * from the standard's default. */
1393 static inline void x264_reference_check_reorder( x264_t *h )
1394 {
1395     for( int i = 0; i < h->i_ref0 - 1; i++ )
1396         /* P and B-frames use different default orders. */
1397         if( h->sh.i_type == SLICE_TYPE_P ? h->fref0[i]->i_frame_num < h->fref0[i+1]->i_frame_num
1398                                          : h->fref0[i]->i_poc < h->fref0[i+1]->i_poc )
1399         {
1400             h->b_ref_reorder[0] = 1;
1401             break;
1402         }
1403 }
1404
1405 /* return -1 on failure, else return the index of the new reference frame */
1406 int x264_weighted_reference_duplicate( x264_t *h, int i_ref, const x264_weight_t *w )
1407 {
1408     int i = h->i_ref0;
1409     int j = 1;
1410     x264_frame_t *newframe;
1411     if( i <= 1 ) /* empty list, definitely can't duplicate frame */
1412         return -1;
1413
1414     newframe = x264_frame_pop_blank_unused( h );
1415
1416     //FIXME: probably don't need to copy everything
1417     *newframe = *h->fref0[i_ref];
1418     newframe->i_reference_count = 1;
1419     newframe->orig = h->fref0[i_ref];
1420     newframe->b_duplicate = 1;
1421     memcpy( h->fenc->weight[j], w, sizeof(h->fenc->weight[i]) );
1422
1423     /* shift the frames to make space for the dupe. */
1424     h->b_ref_reorder[0] = 1;
1425     if( h->i_ref0 < 16 )
1426         ++h->i_ref0;
1427     h->fref0[15] = NULL;
1428     x264_frame_unshift( &h->fref0[j], newframe );
1429
1430     return j;
1431 }
1432
1433 static void x264_weighted_pred_init( x264_t *h )
1434 {
1435     /* for now no analysis and set all weights to nothing */
1436     for( int i_ref = 0; i_ref < h->i_ref0; i_ref++ )
1437         h->fenc->weighted[i_ref] = h->fref0[i_ref]->filtered[0];
1438
1439     // FIXME: This only supports weighting of one reference frame
1440     // and duplicates of that frame.
1441     h->fenc->i_lines_weighted = 0;
1442
1443     for( int i_ref = 0; i_ref < (h->i_ref0 << h->sh.b_mbaff); i_ref++ )
1444         for( int i = 0; i < 3; i++ )
1445             h->sh.weight[i_ref][i].weightfn = NULL;
1446
1447
1448     if( h->sh.i_type != SLICE_TYPE_P || h->param.analyse.i_weighted_pred <= 0 )
1449         return;
1450
1451     int i_padv = PADV << h->param.b_interlaced;
1452     int denom = -1;
1453     int weightluma = 0;
1454     int buffer_next = 0;
1455     //FIXME: when chroma support is added, move this into loop
1456     h->sh.weight[0][1].weightfn = h->sh.weight[0][2].weightfn = NULL;
1457     h->sh.weight[0][1].i_denom = h->sh.weight[0][2].i_denom = 0;
1458     for( int j = 0; j < h->i_ref0; j++ )
1459     {
1460         if( h->fenc->weight[j][0].weightfn )
1461         {
1462             h->sh.weight[j][0] = h->fenc->weight[j][0];
1463             // if weight is useless, don't write it to stream
1464             if( h->sh.weight[j][0].i_scale == 1<<h->sh.weight[j][0].i_denom && h->sh.weight[j][0].i_offset == 0 )
1465                 h->sh.weight[j][0].weightfn = NULL;
1466             else
1467             {
1468                 if( !weightluma )
1469                 {
1470                     weightluma = 1;
1471                     h->sh.weight[0][0].i_denom = denom = h->sh.weight[j][0].i_denom;
1472                     assert( x264_clip3( denom, 0, 7 ) == denom );
1473                 }
1474                 assert( h->sh.weight[j][0].i_denom == denom );
1475                 assert( x264_clip3( h->sh.weight[j][0].i_scale, 0, 127 ) == h->sh.weight[j][0].i_scale );
1476                 assert( x264_clip3( h->sh.weight[j][0].i_offset, -128, 127 ) == h->sh.weight[j][0].i_offset );
1477                 h->fenc->weighted[j] = h->mb.p_weight_buf[buffer_next++] +
1478                     h->fenc->i_stride[0] * i_padv + PADH;
1479             }
1480         }
1481
1482         //scale full resolution frame
1483         if( h->sh.weight[j][0].weightfn && h->param.i_threads == 1 )
1484         {
1485             pixel *src = h->fref0[j]->filtered[0] - h->fref0[j]->i_stride[0]*i_padv - PADH;
1486             pixel *dst = h->fenc->weighted[j] - h->fenc->i_stride[0]*i_padv - PADH;
1487             int stride = h->fenc->i_stride[0];
1488             int width = h->fenc->i_width[0] + PADH*2;
1489             int height = h->fenc->i_lines[0] + i_padv*2;
1490             x264_weight_scale_plane( h, dst, stride, src, stride, width, height, &h->sh.weight[j][0] );
1491             h->fenc->i_lines_weighted = height;
1492         }
1493     }
1494     if( !weightluma )
1495         h->sh.weight[0][0].i_denom = 0;
1496 }
1497
1498 static inline void x264_reference_build_list( x264_t *h, int i_poc )
1499 {
1500     int b_ok;
1501
1502     /* build ref list 0/1 */
1503     h->mb.pic.i_fref[0] = h->i_ref0 = 0;
1504     h->mb.pic.i_fref[1] = h->i_ref1 = 0;
1505     if( h->sh.i_type == SLICE_TYPE_I )
1506         return;
1507
1508     for( int i = 0; h->frames.reference[i]; i++ )
1509     {
1510         if( h->frames.reference[i]->b_corrupt )
1511             continue;
1512         if( h->frames.reference[i]->i_poc < i_poc )
1513             h->fref0[h->i_ref0++] = h->frames.reference[i];
1514         else if( h->frames.reference[i]->i_poc > i_poc )
1515             h->fref1[h->i_ref1++] = h->frames.reference[i];
1516     }
1517
1518     /* Order ref0 from higher to lower poc */
1519     do
1520     {
1521         b_ok = 1;
1522         for( int i = 0; i < h->i_ref0 - 1; i++ )
1523         {
1524             if( h->fref0[i]->i_poc < h->fref0[i+1]->i_poc )
1525             {
1526                 XCHG( x264_frame_t*, h->fref0[i], h->fref0[i+1] );
1527                 b_ok = 0;
1528                 break;
1529             }
1530         }
1531     } while( !b_ok );
1532
1533     if( h->sh.i_mmco_remove_from_end )
1534         for( int i = h->i_ref0-1; i >= h->i_ref0 - h->sh.i_mmco_remove_from_end; i-- )
1535         {
1536             int diff = h->i_frame_num - h->fref0[i]->i_frame_num;
1537             h->sh.mmco[h->sh.i_mmco_command_count].i_poc = h->fref0[i]->i_poc;
1538             h->sh.mmco[h->sh.i_mmco_command_count++].i_difference_of_pic_nums = diff;
1539         }
1540
1541     /* Order ref1 from lower to higher poc (bubble sort) for B-frame */
1542     do
1543     {
1544         b_ok = 1;
1545         for( int i = 0; i < h->i_ref1 - 1; i++ )
1546         {
1547             if( h->fref1[i]->i_poc > h->fref1[i+1]->i_poc )
1548             {
1549                 XCHG( x264_frame_t*, h->fref1[i], h->fref1[i+1] );
1550                 b_ok = 0;
1551                 break;
1552             }
1553         }
1554     } while( !b_ok );
1555
1556     x264_reference_check_reorder( h );
1557
1558     h->i_ref1 = X264_MIN( h->i_ref1, h->frames.i_max_ref1 );
1559     h->i_ref0 = X264_MIN( h->i_ref0, h->frames.i_max_ref0 );
1560     h->i_ref0 = X264_MIN( h->i_ref0, h->param.i_frame_reference ); // if reconfig() has lowered the limit
1561
1562     /* add duplicates */
1563     if( h->fenc->i_type == X264_TYPE_P )
1564     {
1565         int idx = -1;
1566         if( h->param.analyse.i_weighted_pred == X264_WEIGHTP_SMART )
1567         {
1568             x264_weight_t w[3];
1569             w[1].weightfn = w[2].weightfn = NULL;
1570             if( h->param.rc.b_stat_read )
1571                 x264_ratecontrol_set_weights( h, h->fenc );
1572
1573             if( !h->fenc->weight[0][0].weightfn )
1574             {
1575                 h->fenc->weight[0][0].i_denom = 0;
1576                 SET_WEIGHT( w[0], 1, 1, 0, -1 );
1577                 idx = x264_weighted_reference_duplicate( h, 0, w );
1578             }
1579             else
1580             {
1581                 if( h->fenc->weight[0][0].i_scale == 1<<h->fenc->weight[0][0].i_denom )
1582                 {
1583                     SET_WEIGHT( h->fenc->weight[0][0], 1, 1, 0, h->fenc->weight[0][0].i_offset );
1584                 }
1585                 x264_weighted_reference_duplicate( h, 0, weight_none );
1586                 if( h->fenc->weight[0][0].i_offset > -128 )
1587                 {
1588                     w[0] = h->fenc->weight[0][0];
1589                     w[0].i_offset--;
1590                     h->mc.weight_cache( h, &w[0] );
1591                     idx = x264_weighted_reference_duplicate( h, 0, w );
1592                 }
1593             }
1594         }
1595         else if( h->param.analyse.i_weighted_pred == X264_WEIGHTP_BLIND )
1596         {
1597             //weighted offset=-1
1598             x264_weight_t w[3];
1599             SET_WEIGHT( w[0], 1, 1, 0, -1 );
1600             h->fenc->weight[0][0].i_denom = 0;
1601             w[1].weightfn = w[2].weightfn = NULL;
1602             idx = x264_weighted_reference_duplicate( h, 0, w );
1603         }
1604         h->mb.ref_blind_dupe = idx;
1605     }
1606
1607     assert( h->i_ref0 + h->i_ref1 <= 16 );
1608     h->mb.pic.i_fref[0] = h->i_ref0;
1609     h->mb.pic.i_fref[1] = h->i_ref1;
1610 }
1611
1612 static void x264_fdec_filter_row( x264_t *h, int mb_y, int b_inloop )
1613 {
1614     /* mb_y is the mb to be encoded next, not the mb to be filtered here */
1615     int b_hpel = h->fdec->b_kept_as_ref;
1616     int b_deblock = h->sh.i_disable_deblocking_filter_idc != 1;
1617     int b_end = mb_y == h->i_threadslice_end;
1618     int b_measure_quality = 1;
1619     int min_y = mb_y - (1 << h->sh.b_mbaff);
1620     int b_start = min_y == h->i_threadslice_start;
1621     int max_y = b_end ? h->i_threadslice_end : mb_y;
1622     b_deblock &= b_hpel || h->param.psz_dump_yuv;
1623     if( h->param.b_sliced_threads && b_start && min_y && !b_inloop )
1624     {
1625         b_deblock = 0;         /* We already deblocked on the inloop pass. */
1626         b_measure_quality = 0; /* We already measured quality on the inloop pass. */
1627     }
1628     if( mb_y & h->sh.b_mbaff )
1629         return;
1630     if( min_y < h->i_threadslice_start )
1631         return;
1632
1633     if( !b_end && b_inloop )
1634         for( int j = 0; j <= h->sh.b_mbaff; j++ )
1635             for( int i = 0; i < 3; i++ )
1636             {
1637                 memcpy( h->intra_border_backup[j][i],
1638                         h->fdec->plane[i] + ((mb_y*16 >> !!i) + j - 1 - h->sh.b_mbaff) * h->fdec->i_stride[i],
1639                         (h->mb.i_mb_width*16 >> !!i) * sizeof(pixel) );
1640             }
1641
1642     if( b_deblock )
1643         for( int y = min_y; y < max_y; y += (1 << h->sh.b_mbaff) )
1644             x264_frame_deblock_row( h, y );
1645
1646     if( b_hpel )
1647     {
1648         int end = mb_y == h->mb.i_mb_height;
1649         x264_frame_expand_border( h, h->fdec, min_y, end );
1650         if( h->param.analyse.i_subpel_refine )
1651         {
1652             x264_frame_filter( h, h->fdec, min_y, end );
1653             x264_frame_expand_border_filtered( h, h->fdec, min_y, end );
1654         }
1655     }
1656
1657     if( h->i_thread_frames > 1 && h->fdec->b_kept_as_ref )
1658         x264_frame_cond_broadcast( h->fdec, mb_y*16 + (b_end ? 10000 : -(X264_THREAD_HEIGHT << h->sh.b_mbaff)) );
1659
1660     min_y = min_y*16 - 8 * !b_start;
1661     max_y = b_end ? X264_MIN( h->i_threadslice_end*16 , h->param.i_height ) : mb_y*16 - 8;
1662
1663     if( b_measure_quality )
1664     {
1665         if( h->param.analyse.b_psnr )
1666             for( int i = 0; i < 3; i++ )
1667                 h->stat.frame.i_ssd[i] +=
1668                     x264_pixel_ssd_wxh( &h->pixf,
1669                         h->fdec->plane[i] + (min_y>>!!i) * h->fdec->i_stride[i], h->fdec->i_stride[i],
1670                         h->fenc->plane[i] + (min_y>>!!i) * h->fenc->i_stride[i], h->fenc->i_stride[i],
1671                         h->param.i_width >> !!i, (max_y-min_y) >> !!i );
1672
1673         if( h->param.analyse.b_ssim )
1674         {
1675             x264_emms();
1676             /* offset by 2 pixels to avoid alignment of ssim blocks with dct blocks,
1677              * and overlap by 4 */
1678             min_y += b_start ? 2 : -6;
1679             h->stat.frame.f_ssim +=
1680                 x264_pixel_ssim_wxh( &h->pixf,
1681                     h->fdec->plane[0] + 2+min_y*h->fdec->i_stride[0], h->fdec->i_stride[0],
1682                     h->fenc->plane[0] + 2+min_y*h->fenc->i_stride[0], h->fenc->i_stride[0],
1683                     h->param.i_width-2, max_y-min_y, h->scratch_buffer );
1684         }
1685     }
1686 }
1687
1688 static inline int x264_reference_update( x264_t *h )
1689 {
1690     if( !h->fdec->b_kept_as_ref )
1691     {
1692         if( h->i_thread_frames > 1 )
1693         {
1694             x264_frame_push_unused( h, h->fdec );
1695             h->fdec = x264_frame_pop_unused( h, 1 );
1696             if( !h->fdec )
1697                 return -1;
1698         }
1699         return 0;
1700     }
1701
1702     /* apply mmco from previous frame. */
1703     for( int i = 0; i < h->sh.i_mmco_command_count; i++ )
1704         for( int j = 0; h->frames.reference[j]; j++ )
1705             if( h->frames.reference[j]->i_poc == h->sh.mmco[i].i_poc )
1706                 x264_frame_push_unused( h, x264_frame_shift( &h->frames.reference[j] ) );
1707
1708     /* move frame in the buffer */
1709     x264_frame_push( h->frames.reference, h->fdec );
1710     if( h->frames.reference[h->sps->i_num_ref_frames] )
1711         x264_frame_push_unused( h, x264_frame_shift( h->frames.reference ) );
1712     h->fdec = x264_frame_pop_unused( h, 1 );
1713     if( !h->fdec )
1714         return -1;
1715     return 0;
1716 }
1717
1718 static inline void x264_reference_reset( x264_t *h )
1719 {
1720     while( h->frames.reference[0] )
1721         x264_frame_push_unused( h, x264_frame_pop( h->frames.reference ) );
1722     h->fdec->i_poc =
1723     h->fenc->i_poc = 0;
1724 }
1725
1726 static inline void x264_reference_hierarchy_reset( x264_t *h )
1727 {
1728     int ref;
1729     int b_hasdelayframe = 0;
1730
1731     /* look for delay frames -- chain must only contain frames that are disposable */
1732     for( int i = 0; h->frames.current[i] && IS_DISPOSABLE( h->frames.current[i]->i_type ); i++ )
1733         b_hasdelayframe |= h->frames.current[i]->i_coded
1734                         != h->frames.current[i]->i_frame + h->sps->vui.i_num_reorder_frames;
1735
1736     /* This function must handle b-pyramid and clear frames for open-gop */
1737     if( h->param.i_bframe_pyramid != X264_B_PYRAMID_STRICT && !b_hasdelayframe && h->frames.i_poc_last_open_gop == -1 )
1738         return;
1739
1740     /* Remove last BREF. There will never be old BREFs in the
1741      * dpb during a BREF decode when pyramid == STRICT */
1742     for( ref = 0; h->frames.reference[ref]; ref++ )
1743     {
1744         if( ( h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT
1745             && h->frames.reference[ref]->i_type == X264_TYPE_BREF )
1746             || ( h->frames.reference[ref]->i_poc < h->frames.i_poc_last_open_gop
1747             && h->sh.i_type != SLICE_TYPE_B ) )
1748         {
1749             int diff = h->i_frame_num - h->frames.reference[ref]->i_frame_num;
1750             h->sh.mmco[h->sh.i_mmco_command_count].i_difference_of_pic_nums = diff;
1751             h->sh.mmco[h->sh.i_mmco_command_count++].i_poc = h->frames.reference[ref]->i_poc;
1752             x264_frame_push_unused( h, x264_frame_shift( &h->frames.reference[ref] ) );
1753             h->b_ref_reorder[0] = 1;
1754             ref--;
1755         }
1756     }
1757
1758     /* Prepare room in the dpb for the delayed display time of the later b-frame's */
1759     if( h->param.i_bframe_pyramid )
1760         h->sh.i_mmco_remove_from_end = X264_MAX( ref + 2 - h->frames.i_max_dpb, 0 );
1761 }
1762
1763 static inline void x264_slice_init( x264_t *h, int i_nal_type, int i_global_qp )
1764 {
1765     /* ------------------------ Create slice header  ----------------------- */
1766     if( i_nal_type == NAL_SLICE_IDR )
1767     {
1768         x264_slice_header_init( h, &h->sh, h->sps, h->pps, h->i_idr_pic_id, h->i_frame_num, i_global_qp );
1769
1770         /* increment id */
1771         h->i_idr_pic_id = ( h->i_idr_pic_id + 1 ) % 65536;
1772     }
1773     else
1774     {
1775         x264_slice_header_init( h, &h->sh, h->sps, h->pps, -1, h->i_frame_num, i_global_qp );
1776
1777         h->sh.i_num_ref_idx_l0_active = h->i_ref0 <= 0 ? 1 : h->i_ref0;
1778         h->sh.i_num_ref_idx_l1_active = h->i_ref1 <= 0 ? 1 : h->i_ref1;
1779         if( h->sh.i_num_ref_idx_l0_active != h->pps->i_num_ref_idx_l0_default_active ||
1780             (h->sh.i_type == SLICE_TYPE_B && h->sh.i_num_ref_idx_l1_active != h->pps->i_num_ref_idx_l1_default_active) )
1781         {
1782             h->sh.b_num_ref_idx_override = 1;
1783         }
1784     }
1785
1786     h->fdec->i_frame_num = h->sh.i_frame_num;
1787
1788     if( h->sps->i_poc_type == 0 )
1789     {
1790         h->sh.i_poc = h->fdec->i_poc;
1791         if( h->param.b_interlaced )
1792         {
1793             h->sh.i_delta_poc_bottom = h->param.b_tff ? 1 : -1;
1794             if( h->sh.i_delta_poc_bottom == -1 )
1795                 h->sh.i_poc = h->fdec->i_poc + 1;
1796         }
1797         else
1798             h->sh.i_delta_poc_bottom = 0;
1799     }
1800     else if( h->sps->i_poc_type == 1 )
1801     {
1802         /* FIXME TODO FIXME */
1803     }
1804     else
1805     {
1806         /* Nothing to do ? */
1807     }
1808
1809     x264_macroblock_slice_init( h );
1810 }
1811
1812 static int x264_slice_write( x264_t *h )
1813 {
1814     int i_skip;
1815     int mb_xy, i_mb_x, i_mb_y;
1816     int i_skip_bak = 0; /* Shut up GCC. */
1817     bs_t bs_bak;
1818     x264_cabac_t cabac_bak;
1819     uint8_t cabac_prevbyte_bak = 0; /* Shut up GCC. */
1820     /* Assume no more than 3 bytes of NALU escaping.
1821      * NALUs other than the first use a 3-byte startcode. */
1822     int overhead_guess = (NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal)) + 3;
1823     int slice_max_size = h->param.i_slice_max_size > 0 ? (h->param.i_slice_max_size-overhead_guess)*8 : INT_MAX;
1824     int starting_bits = bs_pos(&h->out.bs);
1825     int b_deblock = h->sh.i_disable_deblocking_filter_idc != 1;
1826     int b_hpel = h->fdec->b_kept_as_ref;
1827     b_deblock &= b_hpel || h->param.psz_dump_yuv;
1828     bs_realign( &h->out.bs );
1829
1830     /* Slice */
1831     x264_nal_start( h, h->i_nal_type, h->i_nal_ref_idc );
1832     h->out.nal[h->out.i_nal].i_first_mb = h->sh.i_first_mb;
1833
1834     /* Slice header */
1835     x264_macroblock_thread_init( h );
1836
1837     /* If this isn't the first slice in the threadslice, set the slice QP
1838      * equal to the last QP in the previous slice for more accurate
1839      * CABAC initialization. */
1840     if( h->sh.i_first_mb != h->i_threadslice_start * h->mb.i_mb_width )
1841     {
1842         h->sh.i_qp = h->mb.i_last_qp;
1843         h->sh.i_qp_delta = h->sh.i_qp - h->pps->i_pic_init_qp;
1844     }
1845
1846     x264_slice_header_write( &h->out.bs, &h->sh, h->i_nal_ref_idc );
1847     if( h->param.b_cabac )
1848     {
1849         /* alignment needed */
1850         bs_align_1( &h->out.bs );
1851
1852         /* init cabac */
1853         x264_cabac_context_init( &h->cabac, h->sh.i_type, x264_clip3( h->sh.i_qp-QP_BD_OFFSET, 0, 51 ), h->sh.i_cabac_init_idc );
1854         x264_cabac_encode_init ( &h->cabac, h->out.bs.p, h->out.bs.p_end );
1855     }
1856     h->mb.i_last_qp = h->sh.i_qp;
1857     h->mb.i_last_dqp = 0;
1858
1859     i_mb_y = h->sh.i_first_mb / h->mb.i_mb_width;
1860     i_mb_x = h->sh.i_first_mb % h->mb.i_mb_width;
1861     i_skip = 0;
1862
1863     while( (mb_xy = i_mb_x + i_mb_y * h->mb.i_mb_width) <= h->sh.i_last_mb )
1864     {
1865         int mb_spos = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
1866         if( h->param.i_slice_max_size > 0 )
1867         {
1868             /* We don't need the contexts because flushing the CABAC encoder has no context
1869              * dependency and macroblocks are only re-encoded in the case where a slice is
1870              * ended (and thus the content of all contexts are thrown away). */
1871             if( h->param.b_cabac )
1872             {
1873                 memcpy( &cabac_bak, &h->cabac, offsetof(x264_cabac_t, f8_bits_encoded) );
1874                 /* x264's CABAC writer modifies the previous byte during carry, so it has to be
1875                  * backed up. */
1876                 cabac_prevbyte_bak = h->cabac.p[-1];
1877             }
1878             else
1879             {
1880                 bs_bak = h->out.bs;
1881                 i_skip_bak = i_skip;
1882             }
1883         }
1884
1885         if( i_mb_x == 0 && !h->mb.b_reencode_mb )
1886             x264_fdec_filter_row( h, i_mb_y, 1 );
1887
1888         /* load cache */
1889         x264_macroblock_cache_load( h, i_mb_x, i_mb_y );
1890
1891         x264_macroblock_analyse( h );
1892
1893         /* encode this macroblock -> be careful it can change the mb type to P_SKIP if needed */
1894         x264_macroblock_encode( h );
1895
1896         if( x264_bitstream_check_buffer( h ) )
1897             return -1;
1898
1899         if( h->param.b_cabac )
1900         {
1901             if( mb_xy > h->sh.i_first_mb && !(h->sh.b_mbaff && (i_mb_y&1)) )
1902                 x264_cabac_encode_terminal( &h->cabac );
1903
1904             if( IS_SKIP( h->mb.i_type ) )
1905                 x264_cabac_mb_skip( h, 1 );
1906             else
1907             {
1908                 if( h->sh.i_type != SLICE_TYPE_I )
1909                     x264_cabac_mb_skip( h, 0 );
1910                 x264_macroblock_write_cabac( h, &h->cabac );
1911             }
1912         }
1913         else
1914         {
1915             if( IS_SKIP( h->mb.i_type ) )
1916                 i_skip++;
1917             else
1918             {
1919                 if( h->sh.i_type != SLICE_TYPE_I )
1920                 {
1921                     bs_write_ue( &h->out.bs, i_skip );  /* skip run */
1922                     i_skip = 0;
1923                 }
1924                 x264_macroblock_write_cavlc( h );
1925             }
1926         }
1927
1928         int total_bits = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
1929         int mb_size = total_bits - mb_spos;
1930
1931         /* We'll just re-encode this last macroblock if we go over the max slice size. */
1932         if( total_bits - starting_bits > slice_max_size && !h->mb.b_reencode_mb )
1933         {
1934             if( mb_xy != h->sh.i_first_mb )
1935             {
1936                 if( h->param.b_cabac )
1937                 {
1938                     memcpy( &h->cabac, &cabac_bak, offsetof(x264_cabac_t, f8_bits_encoded) );
1939                     h->cabac.p[-1] = cabac_prevbyte_bak;
1940                 }
1941                 else
1942                 {
1943                     h->out.bs = bs_bak;
1944                     i_skip = i_skip_bak;
1945                 }
1946                 h->mb.b_reencode_mb = 1;
1947                 h->sh.i_last_mb = mb_xy-1;
1948                 break;
1949             }
1950             else
1951             {
1952                 h->sh.i_last_mb = mb_xy;
1953                 h->mb.b_reencode_mb = 0;
1954             }
1955         }
1956         else
1957             h->mb.b_reencode_mb = 0;
1958
1959 #if HAVE_VISUALIZE
1960         if( h->param.b_visualize )
1961             x264_visualize_mb( h );
1962 #endif
1963
1964         /* save cache */
1965         x264_macroblock_cache_save( h );
1966
1967         /* accumulate mb stats */
1968         h->stat.frame.i_mb_count[h->mb.i_type]++;
1969
1970         int b_intra = IS_INTRA( h->mb.i_type );
1971         if( h->param.i_log_level >= X264_LOG_INFO || h->param.rc.b_stat_write )
1972         {
1973             if( !b_intra && !IS_SKIP( h->mb.i_type ) && !IS_DIRECT( h->mb.i_type ) )
1974             {
1975                 if( h->mb.i_partition != D_8x8 )
1976                         h->stat.frame.i_mb_partition[h->mb.i_partition] += 4;
1977                     else
1978                         for( int i = 0; i < 4; i++ )
1979                             h->stat.frame.i_mb_partition[h->mb.i_sub_partition[i]] ++;
1980                 if( h->param.i_frame_reference > 1 )
1981                     for( int i_list = 0; i_list <= (h->sh.i_type == SLICE_TYPE_B); i_list++ )
1982                         for( int i = 0; i < 4; i++ )
1983                         {
1984                             int i_ref = h->mb.cache.ref[i_list][ x264_scan8[4*i] ];
1985                             if( i_ref >= 0 )
1986                                 h->stat.frame.i_mb_count_ref[i_list][i_ref] ++;
1987                         }
1988             }
1989         }
1990
1991         if( h->param.i_log_level >= X264_LOG_INFO )
1992         {
1993             if( h->mb.i_cbp_luma | h->mb.i_cbp_chroma )
1994             {
1995                 int cbpsum = (h->mb.i_cbp_luma&1) + ((h->mb.i_cbp_luma>>1)&1)
1996                            + ((h->mb.i_cbp_luma>>2)&1) + (h->mb.i_cbp_luma>>3);
1997                 h->stat.frame.i_mb_cbp[!b_intra + 0] += cbpsum;
1998                 h->stat.frame.i_mb_cbp[!b_intra + 2] += !!h->mb.i_cbp_chroma;
1999                 h->stat.frame.i_mb_cbp[!b_intra + 4] += h->mb.i_cbp_chroma >> 1;
2000             }
2001             if( h->mb.i_cbp_luma && !b_intra )
2002             {
2003                 h->stat.frame.i_mb_count_8x8dct[0] ++;
2004                 h->stat.frame.i_mb_count_8x8dct[1] += h->mb.b_transform_8x8;
2005             }
2006             if( b_intra && h->mb.i_type != I_PCM )
2007             {
2008                 if( h->mb.i_type == I_16x16 )
2009                     h->stat.frame.i_mb_pred_mode[0][h->mb.i_intra16x16_pred_mode]++;
2010                 else if( h->mb.i_type == I_8x8 )
2011                     for( int i = 0; i < 16; i += 4 )
2012                         h->stat.frame.i_mb_pred_mode[1][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
2013                 else //if( h->mb.i_type == I_4x4 )
2014                     for( int i = 0; i < 16; i++ )
2015                         h->stat.frame.i_mb_pred_mode[2][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
2016                 h->stat.frame.i_mb_pred_mode[3][x264_mb_pred_mode8x8c_fix[h->mb.i_chroma_pred_mode]]++;
2017             }
2018         }
2019
2020         /* calculate deblock strength values (actual deblocking is done per-row along with hpel) */
2021         if( b_deblock )
2022         {
2023             int mvy_limit = 4 >> h->sh.b_mbaff;
2024             uint8_t (*bs)[4][4] = h->deblock_strength[h->mb.i_mb_y&h->sh.b_mbaff][h->mb.i_mb_x];
2025             x264_macroblock_cache_load_deblock( h );
2026             if( IS_INTRA( h->mb.type[h->mb.i_mb_xy] ) )
2027                 memset( bs, 3, 2*4*4*sizeof(uint8_t) );
2028             else
2029                 h->loopf.deblock_strength( h->mb.cache.non_zero_count, h->mb.cache.ref, h->mb.cache.mv,
2030                                            bs, mvy_limit, h->sh.i_type == SLICE_TYPE_B );
2031         }
2032
2033         x264_ratecontrol_mb( h, mb_size );
2034
2035         if( h->sh.b_mbaff )
2036         {
2037             i_mb_x += i_mb_y & 1;
2038             i_mb_y ^= i_mb_x < h->mb.i_mb_width;
2039         }
2040         else
2041             i_mb_x++;
2042         if( i_mb_x == h->mb.i_mb_width )
2043         {
2044             i_mb_y++;
2045             i_mb_x = 0;
2046         }
2047     }
2048     h->out.nal[h->out.i_nal].i_last_mb = h->sh.i_last_mb;
2049
2050     if( h->param.b_cabac )
2051     {
2052         x264_cabac_encode_flush( h, &h->cabac );
2053         h->out.bs.p = h->cabac.p;
2054     }
2055     else
2056     {
2057         if( i_skip > 0 )
2058             bs_write_ue( &h->out.bs, i_skip );  /* last skip run */
2059         /* rbsp_slice_trailing_bits */
2060         bs_rbsp_trailing( &h->out.bs );
2061         bs_flush( &h->out.bs );
2062     }
2063     if( x264_nal_end( h ) )
2064         return -1;
2065
2066     if( h->sh.i_last_mb == (h->i_threadslice_end * h->mb.i_mb_width - 1) )
2067     {
2068         h->stat.frame.i_misc_bits = bs_pos( &h->out.bs )
2069                                   + (h->out.i_nal*NALU_OVERHEAD * 8)
2070                                   - h->stat.frame.i_tex_bits
2071                                   - h->stat.frame.i_mv_bits;
2072         x264_fdec_filter_row( h, h->i_threadslice_end, 1 );
2073     }
2074
2075     return 0;
2076 }
2077
2078 static void x264_thread_sync_context( x264_t *dst, x264_t *src )
2079 {
2080     if( dst == src )
2081         return;
2082
2083     // reference counting
2084     for( x264_frame_t **f = src->frames.reference; *f; f++ )
2085         (*f)->i_reference_count++;
2086     for( x264_frame_t **f = dst->frames.reference; *f; f++ )
2087         x264_frame_push_unused( src, *f );
2088     src->fdec->i_reference_count++;
2089     x264_frame_push_unused( src, dst->fdec );
2090
2091     // copy everything except the per-thread pointers and the constants.
2092     memcpy( &dst->i_frame, &src->i_frame, offsetof(x264_t, mb.type) - offsetof(x264_t, i_frame) );
2093     dst->param = src->param;
2094     dst->stat = src->stat;
2095 }
2096
2097 static void x264_thread_sync_stat( x264_t *dst, x264_t *src )
2098 {
2099     if( dst == src )
2100         return;
2101     memcpy( &dst->stat.i_frame_count, &src->stat.i_frame_count, sizeof(dst->stat) - sizeof(dst->stat.frame) );
2102 }
2103
2104 static void *x264_slices_write( x264_t *h )
2105 {
2106     int i_slice_num = 0;
2107     int last_thread_mb = h->sh.i_last_mb;
2108
2109 #if HAVE_VISUALIZE
2110     if( h->param.b_visualize )
2111         if( x264_visualize_init( h ) )
2112             return (void *)-1;
2113 #endif
2114
2115     /* init stats */
2116     memset( &h->stat.frame, 0, sizeof(h->stat.frame) );
2117     h->mb.b_reencode_mb = 0;
2118     while( h->sh.i_first_mb <= last_thread_mb )
2119     {
2120         h->sh.i_last_mb = last_thread_mb;
2121         if( h->param.i_slice_max_mbs )
2122             h->sh.i_last_mb = h->sh.i_first_mb + h->param.i_slice_max_mbs - 1;
2123         else if( h->param.i_slice_count && !h->param.b_sliced_threads )
2124         {
2125             int height = h->mb.i_mb_height >> h->param.b_interlaced;
2126             int width = h->mb.i_mb_width << h->param.b_interlaced;
2127             i_slice_num++;
2128             h->sh.i_last_mb = (height * i_slice_num + h->param.i_slice_count/2) / h->param.i_slice_count * width - 1;
2129         }
2130         h->sh.i_last_mb = X264_MIN( h->sh.i_last_mb, last_thread_mb );
2131         if( x264_stack_align( x264_slice_write, h ) )
2132             return (void *)-1;
2133         h->sh.i_first_mb = h->sh.i_last_mb + 1;
2134     }
2135
2136 #if HAVE_VISUALIZE
2137     if( h->param.b_visualize )
2138     {
2139         x264_visualize_show( h );
2140         x264_visualize_close( h );
2141     }
2142 #endif
2143
2144     return (void *)0;
2145 }
2146
2147 static int x264_threaded_slices_write( x264_t *h )
2148 {
2149     /* set first/last mb and sync contexts */
2150     for( int i = 0; i < h->param.i_threads; i++ )
2151     {
2152         x264_t *t = h->thread[i];
2153         if( i )
2154         {
2155             t->param = h->param;
2156             memcpy( &t->i_frame, &h->i_frame, offsetof(x264_t, rc) - offsetof(x264_t, i_frame) );
2157         }
2158         int height = h->mb.i_mb_height >> h->param.b_interlaced;
2159         t->i_threadslice_start = ((height *  i    + h->param.i_slice_count/2) / h->param.i_threads) << h->param.b_interlaced;
2160         t->i_threadslice_end   = ((height * (i+1) + h->param.i_slice_count/2) / h->param.i_threads) << h->param.b_interlaced;
2161         t->sh.i_first_mb = t->i_threadslice_start * h->mb.i_mb_width;
2162         t->sh.i_last_mb  =   t->i_threadslice_end * h->mb.i_mb_width - 1;
2163     }
2164
2165     x264_stack_align( x264_analyse_weight_frame, h, h->mb.i_mb_height*16 + 16 );
2166
2167     x264_threads_distribute_ratecontrol( h );
2168
2169     /* dispatch */
2170     for( int i = 0; i < h->param.i_threads; i++ )
2171     {
2172         x264_threadpool_run( h->threadpool, (void*)x264_slices_write, h->thread[i] );
2173         h->thread[i]->b_thread_active = 1;
2174     }
2175     for( int i = 0; i < h->param.i_threads; i++ )
2176     {
2177         h->thread[i]->b_thread_active = 0;
2178         if( (intptr_t)x264_threadpool_wait( h->threadpool, h->thread[i] ) )
2179             return -1;
2180     }
2181
2182     /* Go back and fix up the hpel on the borders between slices. */
2183     for( int i = 1; i < h->param.i_threads; i++ )
2184     {
2185         x264_fdec_filter_row( h->thread[i], h->thread[i]->i_threadslice_start + 1, 0 );
2186         if( h->sh.b_mbaff )
2187             x264_fdec_filter_row( h->thread[i], h->thread[i]->i_threadslice_start + 2, 0 );
2188     }
2189
2190     x264_threads_merge_ratecontrol( h );
2191
2192     for( int i = 1; i < h->param.i_threads; i++ )
2193     {
2194         x264_t *t = h->thread[i];
2195         for( int j = 0; j < t->out.i_nal; j++ )
2196         {
2197             h->out.nal[h->out.i_nal] = t->out.nal[j];
2198             h->out.i_nal++;
2199             x264_nal_check_buffer( h );
2200         }
2201         /* All entries in stat.frame are ints except for ssd/ssim. */
2202         for( int j = 0; j < (offsetof(x264_t,stat.frame.i_ssd) - offsetof(x264_t,stat.frame.i_mv_bits)) / sizeof(int); j++ )
2203             ((int*)&h->stat.frame)[j] += ((int*)&t->stat.frame)[j];
2204         for( int j = 0; j < 3; j++ )
2205             h->stat.frame.i_ssd[j] += t->stat.frame.i_ssd[j];
2206         h->stat.frame.f_ssim += t->stat.frame.f_ssim;
2207     }
2208
2209     return 0;
2210 }
2211
2212 void x264_encoder_intra_refresh( x264_t *h )
2213 {
2214     h = h->thread[h->i_thread_phase];
2215     h->b_queued_intra_refresh = 1;
2216 }
2217
2218 int x264_encoder_invalidate_reference( x264_t *h, int64_t pts )
2219 {
2220     if( h->param.i_bframe )
2221     {
2222         x264_log( h, X264_LOG_ERROR, "x264_encoder_invalidate_reference is not supported with B-frames enabled\n" );
2223         return -1;
2224     }
2225     if( h->param.b_intra_refresh )
2226     {
2227         x264_log( h, X264_LOG_ERROR, "x264_encoder_invalidate_reference is not supported with intra refresh enabled\n" );
2228         return -1;
2229     }
2230     h = h->thread[h->i_thread_phase];
2231     h->i_reference_invalidate_pts = pts;
2232     return 0;
2233 }
2234
2235 /****************************************************************************
2236  * x264_encoder_encode:
2237  *  XXX: i_poc   : is the poc of the current given picture
2238  *       i_frame : is the number of the frame being coded
2239  *  ex:  type frame poc
2240  *       I      0   2*0
2241  *       P      1   2*3
2242  *       B      2   2*1
2243  *       B      3   2*2
2244  *       P      4   2*6
2245  *       B      5   2*4
2246  *       B      6   2*5
2247  ****************************************************************************/
2248 int     x264_encoder_encode( x264_t *h,
2249                              x264_nal_t **pp_nal, int *pi_nal,
2250                              x264_picture_t *pic_in,
2251                              x264_picture_t *pic_out )
2252 {
2253     x264_t *thread_current, *thread_prev, *thread_oldest;
2254     int i_nal_type, i_nal_ref_idc, i_global_qp;
2255     int overhead = NALU_OVERHEAD;
2256
2257     if( h->i_thread_frames > 1 )
2258     {
2259         thread_prev    = h->thread[ h->i_thread_phase ];
2260         h->i_thread_phase = (h->i_thread_phase + 1) % h->i_thread_frames;
2261         thread_current = h->thread[ h->i_thread_phase ];
2262         thread_oldest  = h->thread[ (h->i_thread_phase + 1) % h->i_thread_frames ];
2263         x264_thread_sync_context( thread_current, thread_prev );
2264         x264_thread_sync_ratecontrol( thread_current, thread_prev, thread_oldest );
2265         h = thread_current;
2266     }
2267     else
2268     {
2269         thread_current =
2270         thread_oldest  = h;
2271     }
2272 #if HAVE_MMX
2273     if( h->i_thread_frames == 1 && h->param.cpu&X264_CPU_SSE_MISALIGN )
2274         x264_cpu_mask_misalign_sse();
2275 #endif
2276
2277     // ok to call this before encoding any frames, since the initial values of fdec have b_kept_as_ref=0
2278     if( x264_reference_update( h ) )
2279         return -1;
2280     h->fdec->i_lines_completed = -1;
2281
2282     /* no data out */
2283     *pi_nal = 0;
2284     *pp_nal = NULL;
2285
2286     /* ------------------- Setup new frame from picture -------------------- */
2287     if( pic_in != NULL )
2288     {
2289         /* 1: Copy the picture to a frame and move it to a buffer */
2290         x264_frame_t *fenc = x264_frame_pop_unused( h, 0 );
2291         if( !fenc )
2292             return -1;
2293
2294         if( x264_frame_copy_picture( h, fenc, pic_in ) < 0 )
2295             return -1;
2296
2297         if( h->param.i_width != 16 * h->mb.i_mb_width ||
2298             h->param.i_height != 16 * h->mb.i_mb_height )
2299             x264_frame_expand_border_mod16( h, fenc );
2300
2301         fenc->i_frame = h->frames.i_input++;
2302
2303         if( h->frames.i_bframe_delay && fenc->i_frame == h->frames.i_bframe_delay )
2304             h->frames.i_bframe_delay_time = fenc->i_pts;
2305
2306         if( h->param.b_vfr_input && fenc->i_pts <= h->frames.i_largest_pts )
2307             x264_log( h, X264_LOG_WARNING, "non-strictly-monotonic PTS\n" );
2308
2309         h->frames.i_second_largest_pts = h->frames.i_largest_pts;
2310         h->frames.i_largest_pts = fenc->i_pts;
2311
2312         if( (fenc->i_pic_struct < PIC_STRUCT_AUTO) || (fenc->i_pic_struct > PIC_STRUCT_TRIPLE) )
2313             fenc->i_pic_struct = PIC_STRUCT_AUTO;
2314
2315         if( fenc->i_pic_struct == PIC_STRUCT_AUTO )
2316         {
2317             int b_interlaced = fenc->param ? fenc->param->b_interlaced : h->param.b_interlaced;
2318             if( b_interlaced )
2319             {
2320                 int b_tff = fenc->param ? fenc->param->b_tff : h->param.b_tff;
2321                 fenc->i_pic_struct = b_tff ? PIC_STRUCT_TOP_BOTTOM : PIC_STRUCT_BOTTOM_TOP;
2322             }
2323             else
2324                 fenc->i_pic_struct = PIC_STRUCT_PROGRESSIVE;
2325         }
2326
2327         if( h->param.rc.b_mb_tree && h->param.rc.b_stat_read )
2328         {
2329             if( x264_macroblock_tree_read( h, fenc, pic_in->prop.quant_offsets ) )
2330                 return -1;
2331         }
2332         else
2333             x264_adaptive_quant_frame( h, fenc, pic_in->prop.quant_offsets );
2334
2335         if( pic_in->prop.quant_offsets_free )
2336             pic_in->prop.quant_offsets_free( pic_in->prop.quant_offsets );
2337
2338         if( h->frames.b_have_lowres )
2339             x264_frame_init_lowres( h, fenc );
2340
2341         /* 2: Place the frame into the queue for its slice type decision */
2342         x264_lookahead_put_frame( h, fenc );
2343
2344         if( h->frames.i_input <= h->frames.i_delay + 1 - h->i_thread_frames )
2345         {
2346             /* Nothing yet to encode, waiting for filling of buffers */
2347             pic_out->i_type = X264_TYPE_AUTO;
2348             return 0;
2349         }
2350     }
2351     else
2352     {
2353         /* signal kills for lookahead thread */
2354         x264_pthread_mutex_lock( &h->lookahead->ifbuf.mutex );
2355         h->lookahead->b_exit_thread = 1;
2356         x264_pthread_cond_broadcast( &h->lookahead->ifbuf.cv_fill );
2357         x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
2358     }
2359
2360     h->i_frame++;
2361     /* 3: The picture is analyzed in the lookahead */
2362     if( !h->frames.current[0] )
2363         x264_lookahead_get_frames( h );
2364
2365     if( !h->frames.current[0] && x264_lookahead_is_empty( h ) )
2366         return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
2367
2368     /* ------------------- Get frame to be encoded ------------------------- */
2369     /* 4: get picture to encode */
2370     h->fenc = x264_frame_shift( h->frames.current );
2371     if( h->i_frame == h->i_thread_frames - 1 )
2372         h->i_reordered_pts_delay = h->fenc->i_reordered_pts;
2373     if( h->fenc->param )
2374     {
2375         x264_encoder_reconfig( h, h->fenc->param );
2376         if( h->fenc->param->param_free )
2377             h->fenc->param->param_free( h->fenc->param );
2378     }
2379
2380     if( h->i_reference_invalidate_pts )
2381     {
2382         if( h->i_reference_invalidate_pts >= h->i_last_idr_pts )
2383             for( int i = 0; h->frames.reference[i]; i++ )
2384                 if( h->i_reference_invalidate_pts <= h->frames.reference[i]->i_pts )
2385                     h->frames.reference[i]->b_corrupt = 1;
2386         h->i_reference_invalidate_pts = 0;
2387     }
2388
2389     if( !IS_X264_TYPE_I( h->fenc->i_type ) )
2390     {
2391         int valid_refs_left = 0;
2392         for( int i = 0; h->frames.reference[i]; i++ )
2393             if( !h->frames.reference[i]->b_corrupt )
2394                 valid_refs_left++;
2395         /* No valid reference frames left: force an IDR. */
2396         if( !valid_refs_left )
2397         {
2398             h->fenc->b_keyframe = 1;
2399             h->fenc->i_type = X264_TYPE_IDR;
2400         }
2401     }
2402
2403     if( h->fenc->b_keyframe )
2404     {
2405         h->frames.i_last_keyframe = h->fenc->i_frame;
2406         if( h->fenc->i_type == X264_TYPE_IDR )
2407         {
2408             h->i_frame_num = 0;
2409             h->frames.i_last_idr = h->fenc->i_frame;
2410         }
2411     }
2412     h->sh.i_mmco_command_count =
2413     h->sh.i_mmco_remove_from_end = 0;
2414     h->b_ref_reorder[0] =
2415     h->b_ref_reorder[1] = 0;
2416     h->fdec->i_poc =
2417     h->fenc->i_poc = 2 * ( h->fenc->i_frame - X264_MAX( h->frames.i_last_idr, 0 ) );
2418
2419     /* ------------------- Setup frame context ----------------------------- */
2420     /* 5: Init data dependent of frame type */
2421     if( h->fenc->i_type == X264_TYPE_IDR )
2422     {
2423         /* reset ref pictures */
2424         i_nal_type    = NAL_SLICE_IDR;
2425         i_nal_ref_idc = NAL_PRIORITY_HIGHEST;
2426         h->sh.i_type = SLICE_TYPE_I;
2427         x264_reference_reset( h );
2428         h->frames.i_poc_last_open_gop = -1;
2429     }
2430     else if( h->fenc->i_type == X264_TYPE_I )
2431     {
2432         i_nal_type    = NAL_SLICE;
2433         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
2434         h->sh.i_type = SLICE_TYPE_I;
2435         x264_reference_hierarchy_reset( h );
2436         if( h->param.i_open_gop )
2437             h->frames.i_poc_last_open_gop = h->fenc->b_keyframe ? h->fenc->i_poc : -1;
2438     }
2439     else if( h->fenc->i_type == X264_TYPE_P )
2440     {
2441         i_nal_type    = NAL_SLICE;
2442         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
2443         h->sh.i_type = SLICE_TYPE_P;
2444         x264_reference_hierarchy_reset( h );
2445         h->frames.i_poc_last_open_gop = -1;
2446     }
2447     else if( h->fenc->i_type == X264_TYPE_BREF )
2448     {
2449         i_nal_type    = NAL_SLICE;
2450         i_nal_ref_idc = h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT ? NAL_PRIORITY_LOW : NAL_PRIORITY_HIGH;
2451         h->sh.i_type = SLICE_TYPE_B;
2452         x264_reference_hierarchy_reset( h );
2453     }
2454     else    /* B frame */
2455     {
2456         i_nal_type    = NAL_SLICE;
2457         i_nal_ref_idc = NAL_PRIORITY_DISPOSABLE;
2458         h->sh.i_type = SLICE_TYPE_B;
2459     }
2460
2461     h->fdec->i_type = h->fenc->i_type;
2462     h->fdec->i_frame = h->fenc->i_frame;
2463     h->fenc->b_kept_as_ref =
2464     h->fdec->b_kept_as_ref = i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE && h->param.i_keyint_max > 1;
2465
2466     h->fdec->i_pts = h->fenc->i_pts *= h->i_dts_compress_multiplier;
2467     if( h->frames.i_bframe_delay )
2468     {
2469         int64_t *prev_reordered_pts = thread_current->frames.i_prev_reordered_pts;
2470         if( h->i_frame <= h->frames.i_bframe_delay )
2471         {
2472             if( h->i_dts_compress_multiplier == 1 )
2473                 h->fdec->i_dts = h->fenc->i_reordered_pts - h->frames.i_bframe_delay_time;
2474             else
2475             {
2476                 /* DTS compression */
2477                 if( h->i_frame == 1 )
2478                     thread_current->frames.i_init_delta = h->fenc->i_reordered_pts * h->i_dts_compress_multiplier;
2479                 h->fdec->i_dts = h->i_frame * thread_current->frames.i_init_delta / h->i_dts_compress_multiplier;
2480             }
2481         }
2482         else
2483             h->fdec->i_dts = prev_reordered_pts[ (h->i_frame - h->frames.i_bframe_delay) % h->frames.i_bframe_delay ];
2484         prev_reordered_pts[ h->i_frame % h->frames.i_bframe_delay ] = h->fenc->i_reordered_pts * h->i_dts_compress_multiplier;
2485     }
2486     else
2487         h->fdec->i_dts = h->fenc->i_reordered_pts;
2488     if( h->fenc->i_type == X264_TYPE_IDR )
2489         h->i_last_idr_pts = h->fdec->i_pts;
2490
2491     /* ------------------- Init                ----------------------------- */
2492     /* build ref list 0/1 */
2493     x264_reference_build_list( h, h->fdec->i_poc );
2494
2495     /* ---------------------- Write the bitstream -------------------------- */
2496     /* Init bitstream context */
2497     if( h->param.b_sliced_threads )
2498     {
2499         for( int i = 0; i < h->param.i_threads; i++ )
2500         {
2501             bs_init( &h->thread[i]->out.bs, h->thread[i]->out.p_bitstream, h->thread[i]->out.i_bitstream );
2502             h->thread[i]->out.i_nal = 0;
2503         }
2504     }
2505     else
2506     {
2507         bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
2508         h->out.i_nal = 0;
2509     }
2510
2511     if( h->param.b_aud )
2512     {
2513         int pic_type;
2514
2515         if( h->sh.i_type == SLICE_TYPE_I )
2516             pic_type = 0;
2517         else if( h->sh.i_type == SLICE_TYPE_P )
2518             pic_type = 1;
2519         else if( h->sh.i_type == SLICE_TYPE_B )
2520             pic_type = 2;
2521         else
2522             pic_type = 7;
2523
2524         x264_nal_start( h, NAL_AUD, NAL_PRIORITY_DISPOSABLE );
2525         bs_write( &h->out.bs, 3, pic_type );
2526         bs_rbsp_trailing( &h->out.bs );
2527         if( x264_nal_end( h ) )
2528             return -1;
2529         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2530     }
2531
2532     h->i_nal_type = i_nal_type;
2533     h->i_nal_ref_idc = i_nal_ref_idc;
2534
2535     if( h->param.b_intra_refresh )
2536     {
2537         if( IS_X264_TYPE_I( h->fenc->i_type ) )
2538         {
2539             h->fdec->i_frames_since_pir = 0;
2540             h->b_queued_intra_refresh = 0;
2541             /* PIR is currently only supported with ref == 1, so any intra frame effectively refreshes
2542              * the whole frame and counts as an intra refresh. */
2543             h->fdec->f_pir_position = h->mb.i_mb_width;
2544         }
2545         else if( h->fenc->i_type == X264_TYPE_P )
2546         {
2547             int pocdiff = (h->fdec->i_poc - h->fref0[0]->i_poc)/2;
2548             float increment = X264_MAX( ((float)h->mb.i_mb_width-1) / h->param.i_keyint_max, 1 );
2549             h->fdec->f_pir_position = h->fref0[0]->f_pir_position;
2550             h->fdec->i_frames_since_pir = h->fref0[0]->i_frames_since_pir + pocdiff;
2551             if( h->fdec->i_frames_since_pir >= h->param.i_keyint_max ||
2552                 (h->b_queued_intra_refresh && h->fdec->f_pir_position + 0.5 >= h->mb.i_mb_width) )
2553             {
2554                 h->fdec->f_pir_position = 0;
2555                 h->fdec->i_frames_since_pir = 0;
2556                 h->b_queued_intra_refresh = 0;
2557                 h->fenc->b_keyframe = 1;
2558             }
2559             h->fdec->i_pir_start_col = h->fdec->f_pir_position+0.5;
2560             h->fdec->f_pir_position += increment * pocdiff;
2561             h->fdec->i_pir_end_col = h->fdec->f_pir_position+0.5;
2562         }
2563     }
2564
2565     if( h->fenc->b_keyframe )
2566     {
2567         /* Write SPS and PPS */
2568         if( h->param.b_repeat_headers )
2569         {
2570             /* generate sequence parameters */
2571             x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
2572             x264_sps_write( &h->out.bs, h->sps );
2573             if( x264_nal_end( h ) )
2574                 return -1;
2575             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
2576
2577             /* generate picture parameters */
2578             x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
2579             x264_pps_write( &h->out.bs, h->pps );
2580             if( x264_nal_end( h ) )
2581                 return -1;
2582             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
2583         }
2584
2585         /* buffering period sei is written in x264_encoder_frame_end */
2586
2587         if( h->param.b_repeat_headers && h->fenc->i_frame == 0 )
2588         {
2589             /* identify ourself */
2590             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2591             if( x264_sei_version_write( h, &h->out.bs ) )
2592                 return -1;
2593             if( x264_nal_end( h ) )
2594                 return -1;
2595             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2596         }
2597
2598         if( h->fenc->i_type != X264_TYPE_IDR )
2599         {
2600             int time_to_recovery = h->param.i_open_gop ? 0 : X264_MIN( h->mb.i_mb_width - 1, h->param.i_keyint_max ) + h->param.i_bframe;
2601             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2602             x264_sei_recovery_point_write( h, &h->out.bs, time_to_recovery );
2603             x264_nal_end( h );
2604             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2605         }
2606     }
2607
2608     /* generate sei pic timing */
2609     if( h->sps->vui.b_pic_struct_present || h->sps->vui.b_nal_hrd_parameters_present )
2610     {
2611         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2612         x264_sei_pic_timing_write( h, &h->out.bs );
2613         if( x264_nal_end( h ) )
2614             return -1;
2615         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2616     }
2617
2618     /* Init the rate control */
2619     /* FIXME: Include slice header bit cost. */
2620     x264_ratecontrol_start( h, h->fenc->i_qpplus1, overhead*8 );
2621     i_global_qp = x264_ratecontrol_qp( h );
2622
2623     pic_out->i_qpplus1 =
2624     h->fdec->i_qpplus1 = i_global_qp + 1;
2625
2626     if( h->param.rc.b_stat_read && h->sh.i_type != SLICE_TYPE_I )
2627     {
2628         x264_reference_build_list_optimal( h );
2629         x264_reference_check_reorder( h );
2630     }
2631
2632     if( h->i_ref0 )
2633         h->fdec->i_poc_l0ref0 = h->fref0[0]->i_poc;
2634
2635     if( h->sh.i_type == SLICE_TYPE_B )
2636         x264_macroblock_bipred_init( h );
2637
2638     /*------------------------- Weights -------------------------------------*/
2639     x264_weighted_pred_init( h );
2640
2641     /* ------------------------ Create slice header  ----------------------- */
2642     x264_slice_init( h, i_nal_type, i_global_qp );
2643
2644     if( i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE )
2645         h->i_frame_num++;
2646
2647     /* Write frame */
2648     h->i_threadslice_start = 0;
2649     h->i_threadslice_end = h->mb.i_mb_height;
2650     if( h->i_thread_frames > 1 )
2651     {
2652         x264_threadpool_run( h->threadpool, (void*)x264_slices_write, h );
2653         h->b_thread_active = 1;
2654     }
2655     else if( h->param.b_sliced_threads )
2656     {
2657         if( x264_threaded_slices_write( h ) )
2658             return -1;
2659     }
2660     else
2661         if( (intptr_t)x264_slices_write( h ) )
2662             return -1;
2663
2664     return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
2665 }
2666
2667 static int x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
2668                                    x264_nal_t **pp_nal, int *pi_nal,
2669                                    x264_picture_t *pic_out )
2670 {
2671     char psz_message[80];
2672
2673     if( h->b_thread_active )
2674     {
2675         h->b_thread_active = 0;
2676         if( (intptr_t)x264_threadpool_wait( h->threadpool, h ) )
2677             return -1;
2678     }
2679     if( !h->out.i_nal )
2680     {
2681         pic_out->i_type = X264_TYPE_AUTO;
2682         return 0;
2683     }
2684
2685     x264_emms();
2686     /* generate sei buffering period and insert it into place */
2687     if( h->fenc->b_keyframe && h->sps->vui.b_nal_hrd_parameters_present )
2688     {
2689         x264_hrd_fullness( h );
2690         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2691         x264_sei_buffering_period_write( h, &h->out.bs );
2692         if( x264_nal_end( h ) )
2693            return -1;
2694         /* buffering period sei must follow AUD, SPS and PPS and precede all other SEIs */
2695         int idx = 0;
2696         while( h->out.nal[idx].i_type == NAL_AUD ||
2697                h->out.nal[idx].i_type == NAL_SPS ||
2698                h->out.nal[idx].i_type == NAL_PPS )
2699             idx++;
2700         x264_nal_t nal_tmp = h->out.nal[h->out.i_nal-1];
2701         memmove( &h->out.nal[idx+1], &h->out.nal[idx], (h->out.i_nal-idx-1)*sizeof(x264_nal_t) );
2702         h->out.nal[idx] = nal_tmp;
2703     }
2704
2705     int frame_size = x264_encoder_encapsulate_nals( h, 0 );
2706
2707     /* Set output picture properties */
2708     pic_out->i_type = h->fenc->i_type;
2709
2710     pic_out->b_keyframe = h->fenc->b_keyframe;
2711
2712     pic_out->i_pts = h->fdec->i_pts;
2713     pic_out->i_dts = h->fdec->i_dts;
2714
2715     if( pic_out->i_pts < pic_out->i_dts )
2716         x264_log( h, X264_LOG_WARNING, "invalid DTS: PTS is less than DTS\n" );
2717
2718     pic_out->img.i_plane = h->fdec->i_plane;
2719     for( int i = 0; i < 3; i++ )
2720     {
2721         pic_out->img.i_stride[i] = h->fdec->i_stride[i];
2722         // FIXME This breaks the API when pixel != uint8_t.
2723         pic_out->img.plane[i] = h->fdec->plane[i];
2724     }
2725
2726     x264_frame_push_unused( thread_current, h->fenc );
2727
2728     /* ---------------------- Update encoder state ------------------------- */
2729
2730     /* update rc */
2731     int filler = 0;
2732     if( x264_ratecontrol_end( h, frame_size * 8, &filler ) < 0 )
2733         return -1;
2734
2735     pic_out->hrd_timing = h->fenc->hrd_timing;
2736
2737     while( filler > 0 )
2738     {
2739         int f, overhead;
2740         overhead = (FILLER_OVERHEAD - h->param.b_annexb);
2741         if( h->param.i_slice_max_size && filler > h->param.i_slice_max_size )
2742         {
2743             int next_size = filler - h->param.i_slice_max_size;
2744             int overflow = X264_MAX( overhead - next_size, 0 );
2745             f = h->param.i_slice_max_size - overhead - overflow;
2746         }
2747         else
2748             f = X264_MAX( 0, filler - overhead );
2749
2750         x264_nal_start( h, NAL_FILLER, NAL_PRIORITY_DISPOSABLE );
2751         x264_filler_write( h, &h->out.bs, f );
2752         if( x264_nal_end( h ) )
2753             return -1;
2754         int total_size = x264_encoder_encapsulate_nals( h, h->out.i_nal-1 );
2755         frame_size += total_size;
2756         filler -= total_size;
2757     }
2758
2759     /* End bitstream, set output  */
2760     *pi_nal = h->out.i_nal;
2761     *pp_nal = h->out.nal;
2762
2763     h->out.i_nal = 0;
2764
2765     x264_noise_reduction_update( h );
2766
2767     /* ---------------------- Compute/Print statistics --------------------- */
2768     x264_thread_sync_stat( h, h->thread[0] );
2769
2770     /* Slice stat */
2771     h->stat.i_frame_count[h->sh.i_type]++;
2772     h->stat.i_frame_size[h->sh.i_type] += frame_size;
2773     h->stat.f_frame_qp[h->sh.i_type] += h->fdec->f_qp_avg_aq;
2774
2775     for( int i = 0; i < X264_MBTYPE_MAX; i++ )
2776         h->stat.i_mb_count[h->sh.i_type][i] += h->stat.frame.i_mb_count[i];
2777     for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
2778         h->stat.i_mb_partition[h->sh.i_type][i] += h->stat.frame.i_mb_partition[i];
2779     for( int i = 0; i < 2; i++ )
2780         h->stat.i_mb_count_8x8dct[i] += h->stat.frame.i_mb_count_8x8dct[i];
2781     for( int i = 0; i < 6; i++ )
2782         h->stat.i_mb_cbp[i] += h->stat.frame.i_mb_cbp[i];
2783     for( int i = 0; i < 4; i++ )
2784         for( int j = 0; j < 13; j++ )
2785             h->stat.i_mb_pred_mode[i][j] += h->stat.frame.i_mb_pred_mode[i][j];
2786     if( h->sh.i_type != SLICE_TYPE_I )
2787         for( int i_list = 0; i_list < 2; i_list++ )
2788             for( int i = 0; i < 32; i++ )
2789                 h->stat.i_mb_count_ref[h->sh.i_type][i_list][i] += h->stat.frame.i_mb_count_ref[i_list][i];
2790     if( h->sh.i_type == SLICE_TYPE_P )
2791     {
2792         h->stat.i_consecutive_bframes[h->fdec->i_frame - h->fref0[0]->i_frame - 1]++;
2793         if( h->param.analyse.i_weighted_pred == X264_WEIGHTP_SMART )
2794             for( int i = 0; i < 3; i++ )
2795                 for( int j = 0; j < h->i_ref0; j++ )
2796                     if( h->sh.weight[0][i].i_denom != 0 )
2797                     {
2798                         h->stat.i_wpred[i]++;
2799                         break;
2800                     }
2801     }
2802     if( h->sh.i_type == SLICE_TYPE_B )
2803     {
2804         h->stat.i_direct_frames[ h->sh.b_direct_spatial_mv_pred ] ++;
2805         if( h->mb.b_direct_auto_write )
2806         {
2807             //FIXME somewhat arbitrary time constants
2808             if( h->stat.i_direct_score[0] + h->stat.i_direct_score[1] > h->mb.i_mb_count )
2809                 for( int i = 0; i < 2; i++ )
2810                     h->stat.i_direct_score[i] = h->stat.i_direct_score[i] * 9/10;
2811             for( int i = 0; i < 2; i++ )
2812                 h->stat.i_direct_score[i] += h->stat.frame.i_direct_score[i];
2813         }
2814     }
2815
2816     psz_message[0] = '\0';
2817     if( h->param.analyse.b_psnr )
2818     {
2819         int64_t ssd[3] = {
2820             h->stat.frame.i_ssd[0],
2821             h->stat.frame.i_ssd[1],
2822             h->stat.frame.i_ssd[2],
2823         };
2824
2825         h->stat.i_ssd_global[h->sh.i_type] += ssd[0] + ssd[1] + ssd[2];
2826         h->stat.f_psnr_average[h->sh.i_type] += x264_psnr( ssd[0] + ssd[1] + ssd[2], 3 * h->param.i_width * h->param.i_height / 2 );
2827         h->stat.f_psnr_mean_y[h->sh.i_type] += x264_psnr( ssd[0], h->param.i_width * h->param.i_height );
2828         h->stat.f_psnr_mean_u[h->sh.i_type] += x264_psnr( ssd[1], h->param.i_width * h->param.i_height / 4 );
2829         h->stat.f_psnr_mean_v[h->sh.i_type] += x264_psnr( ssd[2], h->param.i_width * h->param.i_height / 4 );
2830
2831         snprintf( psz_message, 80, " PSNR Y:%5.2f U:%5.2f V:%5.2f",
2832                   x264_psnr( ssd[0], h->param.i_width * h->param.i_height ),
2833                   x264_psnr( ssd[1], h->param.i_width * h->param.i_height / 4),
2834                   x264_psnr( ssd[2], h->param.i_width * h->param.i_height / 4) );
2835     }
2836
2837     if( h->param.analyse.b_ssim )
2838     {
2839         double ssim_y = h->stat.frame.f_ssim
2840                       / (((h->param.i_width-6)>>2) * ((h->param.i_height-6)>>2));
2841         h->stat.f_ssim_mean_y[h->sh.i_type] += ssim_y;
2842         snprintf( psz_message + strlen(psz_message), 80 - strlen(psz_message),
2843                   " SSIM Y:%.5f", ssim_y );
2844     }
2845     psz_message[79] = '\0';
2846
2847     x264_log( h, X264_LOG_DEBUG,
2848                   "frame=%4d QP=%.2f NAL=%d Slice:%c Poc:%-3d I:%-4d P:%-4d SKIP:%-4d size=%d bytes%s\n",
2849               h->i_frame,
2850               h->fdec->f_qp_avg_aq,
2851               h->i_nal_ref_idc,
2852               h->sh.i_type == SLICE_TYPE_I ? 'I' : (h->sh.i_type == SLICE_TYPE_P ? 'P' : 'B' ),
2853               h->fdec->i_poc,
2854               h->stat.frame.i_mb_count_i,
2855               h->stat.frame.i_mb_count_p,
2856               h->stat.frame.i_mb_count_skip,
2857               frame_size,
2858               psz_message );
2859
2860     // keep stats all in one place
2861     x264_thread_sync_stat( h->thread[0], h );
2862     // for the use of the next frame
2863     x264_thread_sync_stat( thread_current, h );
2864
2865 #ifdef DEBUG_MB_TYPE
2866 {
2867     static const char mb_chars[] = { 'i', 'i', 'I', 'C', 'P', '8', 'S',
2868         'D', '<', 'X', 'B', 'X', '>', 'B', 'B', 'B', 'B', '8', 'S' };
2869     for( int mb_xy = 0; mb_xy < h->mb.i_mb_width * h->mb.i_mb_height; mb_xy++ )
2870     {
2871         if( h->mb.type[mb_xy] < X264_MBTYPE_MAX && h->mb.type[mb_xy] >= 0 )
2872             fprintf( stderr, "%c ", mb_chars[ h->mb.type[mb_xy] ] );
2873         else
2874             fprintf( stderr, "? " );
2875
2876         if( (mb_xy+1) % h->mb.i_mb_width == 0 )
2877             fprintf( stderr, "\n" );
2878     }
2879 }
2880 #endif
2881
2882     /* Remove duplicates, must be done near the end as breaks h->fref0 array
2883      * by freeing some of its pointers. */
2884      for( int i = 0; i < h->i_ref0; i++ )
2885          if( h->fref0[i] && h->fref0[i]->b_duplicate )
2886          {
2887              x264_frame_push_blank_unused( h, h->fref0[i] );
2888              h->fref0[i] = 0;
2889          }
2890
2891     if( h->param.psz_dump_yuv )
2892         x264_frame_dump( h );
2893
2894     return frame_size;
2895 }
2896
2897 static void x264_print_intra( int64_t *i_mb_count, double i_count, int b_print_pcm, char *intra )
2898 {
2899     intra += sprintf( intra, "I16..4%s: %4.1f%% %4.1f%% %4.1f%%",
2900         b_print_pcm ? "..PCM" : "",
2901         i_mb_count[I_16x16]/ i_count,
2902         i_mb_count[I_8x8]  / i_count,
2903         i_mb_count[I_4x4]  / i_count );
2904     if( b_print_pcm )
2905         sprintf( intra, " %4.1f%%", i_mb_count[I_PCM]  / i_count );
2906 }
2907
2908 /****************************************************************************
2909  * x264_encoder_close:
2910  ****************************************************************************/
2911 void    x264_encoder_close  ( x264_t *h )
2912 {
2913     int64_t i_yuv_size = 3 * h->param.i_width * h->param.i_height / 2;
2914     int64_t i_mb_count_size[2][7] = {{0}};
2915     char buf[200];
2916     int b_print_pcm = h->stat.i_mb_count[SLICE_TYPE_I][I_PCM]
2917                    || h->stat.i_mb_count[SLICE_TYPE_P][I_PCM]
2918                    || h->stat.i_mb_count[SLICE_TYPE_B][I_PCM];
2919
2920     x264_lookahead_delete( h );
2921
2922     if( h->param.i_threads > 1 )
2923         x264_threadpool_delete( h->threadpool );
2924     if( h->i_thread_frames > 1 )
2925     {
2926         for( int i = 0; i < h->i_thread_frames; i++ )
2927             if( h->thread[i]->b_thread_active )
2928             {
2929                 assert( h->thread[i]->fenc->i_reference_count == 1 );
2930                 x264_frame_delete( h->thread[i]->fenc );
2931             }
2932
2933         x264_t *thread_prev = h->thread[h->i_thread_phase];
2934         x264_thread_sync_ratecontrol( h, thread_prev, h );
2935         x264_thread_sync_ratecontrol( thread_prev, thread_prev, h );
2936         h->i_frame = thread_prev->i_frame + 1 - h->i_thread_frames;
2937     }
2938     h->i_frame++;
2939
2940     /* Slices used and PSNR */
2941     for( int i = 0; i < 5; i++ )
2942     {
2943         static const uint8_t slice_order[] = { SLICE_TYPE_I, SLICE_TYPE_SI, SLICE_TYPE_P, SLICE_TYPE_SP, SLICE_TYPE_B };
2944         static const char * const slice_name[] = { "P", "B", "I", "SP", "SI" };
2945         int i_slice = slice_order[i];
2946
2947         if( h->stat.i_frame_count[i_slice] > 0 )
2948         {
2949             const int i_count = h->stat.i_frame_count[i_slice];
2950             if( h->param.analyse.b_psnr )
2951             {
2952                 x264_log( h, X264_LOG_INFO,
2953                           "frame %s:%-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",
2954                           slice_name[i_slice],
2955                           i_count,
2956                           h->stat.f_frame_qp[i_slice] / i_count,
2957                           (double)h->stat.i_frame_size[i_slice] / i_count,
2958                           h->stat.f_psnr_mean_y[i_slice] / i_count, h->stat.f_psnr_mean_u[i_slice] / i_count, h->stat.f_psnr_mean_v[i_slice] / i_count,
2959                           h->stat.f_psnr_average[i_slice] / i_count,
2960                           x264_psnr( h->stat.i_ssd_global[i_slice], i_count * i_yuv_size ) );
2961             }
2962             else
2963             {
2964                 x264_log( h, X264_LOG_INFO,
2965                           "frame %s:%-5d Avg QP:%5.2f  size:%6.0f\n",
2966                           slice_name[i_slice],
2967                           i_count,
2968                           h->stat.f_frame_qp[i_slice] / i_count,
2969                           (double)h->stat.i_frame_size[i_slice] / i_count );
2970             }
2971         }
2972     }
2973     if( h->param.i_bframe && h->stat.i_frame_count[SLICE_TYPE_P] )
2974     {
2975         char *p = buf;
2976         int den = 0;
2977         // weight by number of frames (including the P-frame) that are in a sequence of N B-frames
2978         for( int i = 0; i <= h->param.i_bframe; i++ )
2979             den += (i+1) * h->stat.i_consecutive_bframes[i];
2980         for( int i = 0; i <= h->param.i_bframe; i++ )
2981             p += sprintf( p, " %4.1f%%", 100. * (i+1) * h->stat.i_consecutive_bframes[i] / den );
2982         x264_log( h, X264_LOG_INFO, "consecutive B-frames:%s\n", buf );
2983     }
2984
2985     for( int i_type = 0; i_type < 2; i_type++ )
2986         for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
2987         {
2988             if( i == D_DIRECT_8x8 ) continue; /* direct is counted as its own type */
2989             i_mb_count_size[i_type][x264_mb_partition_pixel_table[i]] += h->stat.i_mb_partition[i_type][i];
2990         }
2991
2992     /* MB types used */
2993     if( h->stat.i_frame_count[SLICE_TYPE_I] > 0 )
2994     {
2995         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_I];
2996         double i_count = h->stat.i_frame_count[SLICE_TYPE_I] * h->mb.i_mb_count / 100.0;
2997         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
2998         x264_log( h, X264_LOG_INFO, "mb I  %s\n", buf );
2999     }
3000     if( h->stat.i_frame_count[SLICE_TYPE_P] > 0 )
3001     {
3002         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_P];
3003         double i_count = h->stat.i_frame_count[SLICE_TYPE_P] * h->mb.i_mb_count / 100.0;
3004         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_P];
3005         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
3006         x264_log( h, X264_LOG_INFO,
3007                   "mb P  %s  P16..4: %4.1f%% %4.1f%% %4.1f%% %4.1f%% %4.1f%%    skip:%4.1f%%\n",
3008                   buf,
3009                   i_mb_size[PIXEL_16x16] / (i_count*4),
3010                   (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
3011                   i_mb_size[PIXEL_8x8] / (i_count*4),
3012                   (i_mb_size[PIXEL_8x4] + i_mb_size[PIXEL_4x8]) / (i_count*4),
3013                   i_mb_size[PIXEL_4x4] / (i_count*4),
3014                   i_mb_count[P_SKIP] / i_count );
3015     }
3016     if( h->stat.i_frame_count[SLICE_TYPE_B] > 0 )
3017     {
3018         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_B];
3019         double i_count = h->stat.i_frame_count[SLICE_TYPE_B] * h->mb.i_mb_count / 100.0;
3020         double i_mb_list_count;
3021         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_B];
3022         int64_t list_count[3] = {0}; /* 0 == L0, 1 == L1, 2 == BI */
3023         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
3024         for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
3025             for( int j = 0; j < 2; j++ )
3026             {
3027                 int l0 = x264_mb_type_list_table[i][0][j];
3028                 int l1 = x264_mb_type_list_table[i][1][j];
3029                 if( l0 || l1 )
3030                     list_count[l1+l0*l1] += h->stat.i_mb_count[SLICE_TYPE_B][i] * 2;
3031             }
3032         list_count[0] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L0_8x8];
3033         list_count[1] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L1_8x8];
3034         list_count[2] += h->stat.i_mb_partition[SLICE_TYPE_B][D_BI_8x8];
3035         i_mb_count[B_DIRECT] += (h->stat.i_mb_partition[SLICE_TYPE_B][D_DIRECT_8x8]+2)/4;
3036         i_mb_list_count = (list_count[0] + list_count[1] + list_count[2]) / 100.0;
3037         x264_log( h, X264_LOG_INFO,
3038                   "mb B  %s  B16..8: %4.1f%% %4.1f%% %4.1f%%  direct:%4.1f%%  skip:%4.1f%%  L0:%4.1f%% L1:%4.1f%% BI:%4.1f%%\n",
3039                   buf,
3040                   i_mb_size[PIXEL_16x16] / (i_count*4),
3041                   (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
3042                   i_mb_size[PIXEL_8x8] / (i_count*4),
3043                   i_mb_count[B_DIRECT] / i_count,
3044                   i_mb_count[B_SKIP]   / i_count,
3045                   list_count[0] / i_mb_list_count,
3046                   list_count[1] / i_mb_list_count,
3047                   list_count[2] / i_mb_list_count );
3048     }
3049
3050     x264_ratecontrol_summary( h );
3051
3052     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 )
3053     {
3054 #define SUM3(p) (p[SLICE_TYPE_I] + p[SLICE_TYPE_P] + p[SLICE_TYPE_B])
3055 #define SUM3b(p,o) (p[SLICE_TYPE_I][o] + p[SLICE_TYPE_P][o] + p[SLICE_TYPE_B][o])
3056         int64_t i_i8x8 = SUM3b( h->stat.i_mb_count, I_8x8 );
3057         int64_t i_intra = i_i8x8 + SUM3b( h->stat.i_mb_count, I_4x4 )
3058                                  + SUM3b( h->stat.i_mb_count, I_16x16 );
3059         int64_t i_all_intra = i_intra + SUM3b( h->stat.i_mb_count, I_PCM);
3060         const int i_count = h->stat.i_frame_count[SLICE_TYPE_I] +
3061                             h->stat.i_frame_count[SLICE_TYPE_P] +
3062                             h->stat.i_frame_count[SLICE_TYPE_B];
3063         int64_t i_mb_count = i_count * h->mb.i_mb_count;
3064         float fps = (float) h->param.i_fps_num / h->param.i_fps_den;
3065         float f_bitrate;
3066         /* duration algorithm fails with one frame */
3067         if( !h->param.b_vfr_input || i_count == 1 )
3068             f_bitrate = fps * SUM3(h->stat.i_frame_size) / i_count / 125;
3069         else
3070         {
3071             float duration = (float)(2 * h->frames.i_largest_pts - h->frames.i_second_largest_pts) * h->param.i_timebase_num / h->param.i_timebase_den;
3072             f_bitrate = SUM3(h->stat.i_frame_size) / duration / 125;
3073         }
3074
3075         if( h->pps->b_transform_8x8_mode )
3076         {
3077             buf[0] = 0;
3078             if( h->stat.i_mb_count_8x8dct[0] )
3079                 sprintf( buf, " inter:%.1f%%", 100. * h->stat.i_mb_count_8x8dct[1] / h->stat.i_mb_count_8x8dct[0] );
3080             x264_log( h, X264_LOG_INFO, "8x8 transform intra:%.1f%%%s\n", 100. * i_i8x8 / i_intra, buf );
3081         }
3082
3083         if( (h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO ||
3084             (h->stat.i_direct_frames[0] && h->stat.i_direct_frames[1]))
3085             && h->stat.i_frame_count[SLICE_TYPE_B] )
3086         {
3087             x264_log( h, X264_LOG_INFO, "direct mvs  spatial:%.1f%% temporal:%.1f%%\n",
3088                       h->stat.i_direct_frames[1] * 100. / h->stat.i_frame_count[SLICE_TYPE_B],
3089                       h->stat.i_direct_frames[0] * 100. / h->stat.i_frame_count[SLICE_TYPE_B] );
3090         }
3091
3092         buf[0] = 0;
3093         if( i_mb_count != i_all_intra )
3094             sprintf( buf, " inter: %.1f%% %.1f%% %.1f%%",
3095                      h->stat.i_mb_cbp[1] * 100.0 / ((i_mb_count - i_all_intra)*4),
3096                      h->stat.i_mb_cbp[3] * 100.0 / ((i_mb_count - i_all_intra)  ),
3097                      h->stat.i_mb_cbp[5] * 100.0 / ((i_mb_count - i_all_intra)) );
3098         x264_log( h, X264_LOG_INFO, "coded y,uvDC,uvAC intra: %.1f%% %.1f%% %.1f%%%s\n",
3099                   h->stat.i_mb_cbp[0] * 100.0 / (i_all_intra*4),
3100                   h->stat.i_mb_cbp[2] * 100.0 / (i_all_intra  ),
3101                   h->stat.i_mb_cbp[4] * 100.0 / (i_all_intra  ), buf );
3102
3103         int64_t fixed_pred_modes[4][9] = {{0}};
3104         int64_t sum_pred_modes[4] = {0};
3105         for( int i = 0; i <= I_PRED_16x16_DC_128; i++ )
3106         {
3107             fixed_pred_modes[0][x264_mb_pred_mode16x16_fix[i]] += h->stat.i_mb_pred_mode[0][i];
3108             sum_pred_modes[0] += h->stat.i_mb_pred_mode[0][i];
3109         }
3110         if( sum_pred_modes[0] )
3111             x264_log( h, X264_LOG_INFO, "i16 v,h,dc,p: %2.0f%% %2.0f%% %2.0f%% %2.0f%%\n",
3112                       fixed_pred_modes[0][0] * 100.0 / sum_pred_modes[0],
3113                       fixed_pred_modes[0][1] * 100.0 / sum_pred_modes[0],
3114                       fixed_pred_modes[0][2] * 100.0 / sum_pred_modes[0],
3115                       fixed_pred_modes[0][3] * 100.0 / sum_pred_modes[0] );
3116         for( int i = 1; i <= 2; i++ )
3117         {
3118             for( int j = 0; j <= I_PRED_8x8_DC_128; j++ )
3119             {
3120                 fixed_pred_modes[i][x264_mb_pred_mode4x4_fix(j)] += h->stat.i_mb_pred_mode[i][j];
3121                 sum_pred_modes[i] += h->stat.i_mb_pred_mode[i][j];
3122             }
3123             if( sum_pred_modes[i] )
3124                 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,
3125                           fixed_pred_modes[i][0] * 100.0 / sum_pred_modes[i],
3126                           fixed_pred_modes[i][1] * 100.0 / sum_pred_modes[i],
3127                           fixed_pred_modes[i][2] * 100.0 / sum_pred_modes[i],
3128                           fixed_pred_modes[i][3] * 100.0 / sum_pred_modes[i],
3129                           fixed_pred_modes[i][4] * 100.0 / sum_pred_modes[i],
3130                           fixed_pred_modes[i][5] * 100.0 / sum_pred_modes[i],
3131                           fixed_pred_modes[i][6] * 100.0 / sum_pred_modes[i],
3132                           fixed_pred_modes[i][7] * 100.0 / sum_pred_modes[i],
3133                           fixed_pred_modes[i][8] * 100.0 / sum_pred_modes[i] );
3134         }
3135         for( int i = 0; i <= I_PRED_CHROMA_DC_128; i++ )
3136         {
3137             fixed_pred_modes[3][x264_mb_pred_mode8x8c_fix[i]] += h->stat.i_mb_pred_mode[3][i];
3138             sum_pred_modes[3] += h->stat.i_mb_pred_mode[3][i];
3139         }
3140         if( sum_pred_modes[3] )
3141             x264_log( h, X264_LOG_INFO, "i8c dc,h,v,p: %2.0f%% %2.0f%% %2.0f%% %2.0f%%\n",
3142                       fixed_pred_modes[3][0] * 100.0 / sum_pred_modes[3],
3143                       fixed_pred_modes[3][1] * 100.0 / sum_pred_modes[3],
3144                       fixed_pred_modes[3][2] * 100.0 / sum_pred_modes[3],
3145                       fixed_pred_modes[3][3] * 100.0 / sum_pred_modes[3] );
3146
3147         if( h->param.analyse.i_weighted_pred == X264_WEIGHTP_SMART && h->stat.i_frame_count[SLICE_TYPE_P] > 0 )
3148             x264_log( h, X264_LOG_INFO, "Weighted P-Frames: Y:%.1f%%\n",
3149                       h->stat.i_wpred[0] * 100.0 / h->stat.i_frame_count[SLICE_TYPE_P] );
3150
3151         for( int i_list = 0; i_list < 2; i_list++ )
3152             for( int i_slice = 0; i_slice < 2; i_slice++ )
3153             {
3154                 char *p = buf;
3155                 int64_t i_den = 0;
3156                 int i_max = 0;
3157                 for( int i = 0; i < 32; i++ )
3158                     if( h->stat.i_mb_count_ref[i_slice][i_list][i] )
3159                     {
3160                         i_den += h->stat.i_mb_count_ref[i_slice][i_list][i];
3161                         i_max = i;
3162                     }
3163                 if( i_max == 0 )
3164                     continue;
3165                 for( int i = 0; i <= i_max; i++ )
3166                     p += sprintf( p, " %4.1f%%", 100. * h->stat.i_mb_count_ref[i_slice][i_list][i] / i_den );
3167                 x264_log( h, X264_LOG_INFO, "ref %c L%d:%s\n", "PB"[i_slice], i_list, buf );
3168             }
3169
3170         if( h->param.analyse.b_ssim )
3171         {
3172             float ssim = SUM3( h->stat.f_ssim_mean_y ) / i_count;
3173             x264_log( h, X264_LOG_INFO, "SSIM Mean Y:%.7f (%6.3fdb)\n", ssim, x264_ssim( ssim ) );
3174         }
3175         if( h->param.analyse.b_psnr )
3176         {
3177             x264_log( h, X264_LOG_INFO,
3178                       "PSNR Mean Y:%6.3f U:%6.3f V:%6.3f Avg:%6.3f Global:%6.3f kb/s:%.2f\n",
3179                       SUM3( h->stat.f_psnr_mean_y ) / i_count,
3180                       SUM3( h->stat.f_psnr_mean_u ) / i_count,
3181                       SUM3( h->stat.f_psnr_mean_v ) / i_count,
3182                       SUM3( h->stat.f_psnr_average ) / i_count,
3183                       x264_psnr( SUM3( h->stat.i_ssd_global ), i_count * i_yuv_size ),
3184                       f_bitrate );
3185         }
3186         else
3187             x264_log( h, X264_LOG_INFO, "kb/s:%.2f\n", f_bitrate );
3188     }
3189
3190     /* rc */
3191     x264_ratecontrol_delete( h );
3192
3193     /* param */
3194     if( h->param.rc.psz_stat_out )
3195         free( h->param.rc.psz_stat_out );
3196     if( h->param.rc.psz_stat_in )
3197         free( h->param.rc.psz_stat_in );
3198
3199     x264_cqm_delete( h );
3200     x264_free( h->nal_buffer );
3201     x264_analyse_free_costs( h );
3202
3203     if( h->i_thread_frames > 1)
3204         h = h->thread[h->i_thread_phase];
3205
3206     /* frames */
3207     x264_frame_delete_list( h->frames.unused[0] );
3208     x264_frame_delete_list( h->frames.unused[1] );
3209     x264_frame_delete_list( h->frames.current );
3210     x264_frame_delete_list( h->frames.blank_unused );
3211
3212     h = h->thread[0];
3213
3214     for( int i = h->param.i_threads - 1; i >= 0; i-- )
3215     {
3216         x264_frame_t **frame;
3217
3218         if( !h->param.b_sliced_threads || i == 0 )
3219         {
3220             for( frame = h->thread[i]->frames.reference; *frame; frame++ )
3221             {
3222                 assert( (*frame)->i_reference_count > 0 );
3223                 (*frame)->i_reference_count--;
3224                 if( (*frame)->i_reference_count == 0 )
3225                     x264_frame_delete( *frame );
3226             }
3227             frame = &h->thread[i]->fdec;
3228             assert( (*frame)->i_reference_count > 0 );
3229             (*frame)->i_reference_count--;
3230             if( (*frame)->i_reference_count == 0 )
3231                 x264_frame_delete( *frame );
3232             x264_macroblock_cache_free( h->thread[i] );
3233         }
3234         x264_macroblock_thread_free( h->thread[i], 0 );
3235         x264_free( h->thread[i]->out.p_bitstream );
3236         x264_free( h->thread[i]->out.nal);
3237         x264_free( h->thread[i] );
3238     }
3239 }
3240
3241 /****************************************************************************
3242  * x264_encoder_delayed_frames:
3243  ****************************************************************************/
3244 int x264_encoder_delayed_frames( x264_t *h )
3245 {
3246     int delayed_frames = 0;
3247     if( h->i_thread_frames > 1 )
3248     {
3249         for( int i = 0; i < h->i_thread_frames; i++ )
3250             delayed_frames += h->thread[i]->b_thread_active;
3251         h = h->thread[h->i_thread_phase];
3252     }
3253     for( int i = 0; h->frames.current[i]; i++ )
3254         delayed_frames++;
3255     x264_pthread_mutex_lock( &h->lookahead->ofbuf.mutex );
3256     x264_pthread_mutex_lock( &h->lookahead->ifbuf.mutex );
3257     x264_pthread_mutex_lock( &h->lookahead->next.mutex );
3258     delayed_frames += h->lookahead->ifbuf.i_size + h->lookahead->next.i_size + h->lookahead->ofbuf.i_size;
3259     x264_pthread_mutex_unlock( &h->lookahead->next.mutex );
3260     x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
3261     x264_pthread_mutex_unlock( &h->lookahead->ofbuf.mutex );
3262     return delayed_frames;
3263 }