]> git.sesse.net Git - x264/blob - encoder/encoder.c
Faster probe_skip, 2x2 DC transform handling
[x264] / 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  *          Fiona Glaser <fiona@x264.com>
9  *
10  * This program is free software; you can redistribute it and/or modify
11  * it under the terms of the GNU General Public License as published by
12  * the Free Software Foundation; either version 2 of the License, or
13  * (at your option) any later version.
14  *
15  * This program is distributed in the hope that it will be useful,
16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
18  * GNU General Public License for more details.
19  *
20  * You should have received a copy of the GNU General Public License
21  * along with this program; if not, write to the Free Software
22  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02111, USA.
23  *****************************************************************************/
24
25 #include <math.h>
26
27 #include "common/common.h"
28 #include "common/cpu.h"
29
30 #include "set.h"
31 #include "analyse.h"
32 #include "ratecontrol.h"
33 #include "macroblock.h"
34 #include "me.h"
35
36 #ifdef HAVE_VISUALIZE
37 #include "common/visualize.h"
38 #endif
39
40 //#define DEBUG_MB_TYPE
41
42 #define NALU_OVERHEAD 5 // startcode + NAL type costs 5 bytes per frame
43
44 #define bs_write_ue bs_write_ue_big
45
46 static int x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
47                                    x264_nal_t **pp_nal, int *pi_nal,
48                                    x264_picture_t *pic_out );
49
50 /****************************************************************************
51  *
52  ******************************* x264 libs **********************************
53  *
54  ****************************************************************************/
55 static float x264_psnr( int64_t i_sqe, int64_t i_size )
56 {
57     double f_mse = (double)i_sqe / ((double)65025.0 * (double)i_size);
58     if( f_mse <= 0.0000000001 ) /* Max 100dB */
59         return 100;
60
61     return (float)(-10.0 * log( f_mse ) / log( 10.0 ));
62 }
63
64 static void x264_frame_dump( x264_t *h )
65 {
66     FILE *f = fopen( h->param.psz_dump_yuv, "r+b" );
67     int i, y;
68     if( !f )
69         return;
70     /* Write the frame in display order */
71     fseek( f, (uint64_t)h->fdec->i_frame * h->param.i_height * h->param.i_width * 3/2, SEEK_SET );
72     for( i = 0; i < h->fdec->i_plane; i++ )
73         for( y = 0; y < h->param.i_height >> !!i; y++ )
74             fwrite( &h->fdec->plane[i][y*h->fdec->i_stride[i]], 1, h->param.i_width >> !!i, f );
75     fclose( f );
76 }
77
78
79 /* Fill "default" values */
80 static void x264_slice_header_init( x264_t *h, x264_slice_header_t *sh,
81                                     x264_sps_t *sps, x264_pps_t *pps,
82                                     int i_idr_pic_id, int i_frame, int i_qp )
83 {
84     x264_param_t *param = &h->param;
85     int i;
86
87     /* First we fill all fields */
88     sh->sps = sps;
89     sh->pps = pps;
90
91     sh->i_first_mb  = 0;
92     sh->i_last_mb   = h->mb.i_mb_count - 1;
93     sh->i_pps_id    = pps->i_id;
94
95     sh->i_frame_num = i_frame;
96
97     sh->b_mbaff = h->param.b_interlaced;
98     sh->b_field_pic = 0;    /* no field support for now */
99     sh->b_bottom_field = 0; /* not yet used */
100
101     sh->i_idr_pic_id = i_idr_pic_id;
102
103     /* poc stuff, fixed later */
104     sh->i_poc_lsb = 0;
105     sh->i_delta_poc_bottom = 0;
106     sh->i_delta_poc[0] = 0;
107     sh->i_delta_poc[1] = 0;
108
109     sh->i_redundant_pic_cnt = 0;
110
111     h->mb.b_direct_auto_write = h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO
112                                 && h->param.i_bframe
113                                 && ( h->param.rc.b_stat_write || !h->param.rc.b_stat_read );
114
115     if( !h->mb.b_direct_auto_read && sh->i_type == SLICE_TYPE_B )
116     {
117         if( h->fref1[0]->i_poc_l0ref0 == h->fref0[0]->i_poc )
118         {
119             if( h->mb.b_direct_auto_write )
120                 sh->b_direct_spatial_mv_pred = ( h->stat.i_direct_score[1] > h->stat.i_direct_score[0] );
121             else
122                 sh->b_direct_spatial_mv_pred = ( param->analyse.i_direct_mv_pred == X264_DIRECT_PRED_SPATIAL );
123         }
124         else
125         {
126             h->mb.b_direct_auto_write = 0;
127             sh->b_direct_spatial_mv_pred = 1;
128         }
129     }
130     /* else b_direct_spatial_mv_pred was read from the 2pass statsfile */
131
132     sh->b_num_ref_idx_override = 0;
133     sh->i_num_ref_idx_l0_active = 1;
134     sh->i_num_ref_idx_l1_active = 1;
135
136     sh->b_ref_pic_list_reordering_l0 = h->b_ref_reorder[0];
137     sh->b_ref_pic_list_reordering_l1 = h->b_ref_reorder[1];
138
139     /* If the ref list isn't in the default order, construct reordering header */
140     /* List1 reordering isn't needed yet */
141     if( sh->b_ref_pic_list_reordering_l0 )
142     {
143         int pred_frame_num = i_frame;
144         for( i = 0; i < h->i_ref0; i++ )
145         {
146             int diff = h->fref0[i]->i_frame_num - pred_frame_num;
147             if( diff == 0 )
148                 x264_log( h, X264_LOG_ERROR, "diff frame num == 0\n" );
149             sh->ref_pic_list_order[0][i].idc = ( diff > 0 );
150             sh->ref_pic_list_order[0][i].arg = abs( diff ) - 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     /* If effective qp <= 15, deblocking would have no effect anyway */
163     if( param->b_deblocking_filter
164         && ( h->mb.b_variable_qp
165         || 15 < i_qp + 2 * X264_MIN(param->i_deblocking_filter_alphac0, param->i_deblocking_filter_beta) ) )
166     {
167         sh->i_disable_deblocking_filter_idc = 0;
168     }
169     else
170     {
171         sh->i_disable_deblocking_filter_idc = 1;
172     }
173     sh->i_alpha_c0_offset = param->i_deblocking_filter_alphac0 << 1;
174     sh->i_beta_offset = param->i_deblocking_filter_beta << 1;
175 }
176
177 static void x264_slice_header_write( bs_t *s, x264_slice_header_t *sh, int i_nal_ref_idc )
178 {
179     int i;
180
181     if( sh->b_mbaff )
182     {
183         assert( sh->i_first_mb % (2*sh->sps->i_mb_width) == 0 );
184         bs_write_ue( s, sh->i_first_mb >> 1 );
185     }
186     else
187         bs_write_ue( s, sh->i_first_mb );
188
189     bs_write_ue( s, sh->i_type + 5 );   /* same type things */
190     bs_write_ue( s, sh->i_pps_id );
191     bs_write( s, sh->sps->i_log2_max_frame_num, sh->i_frame_num & ((1<<sh->sps->i_log2_max_frame_num)-1) );
192
193     if( !sh->sps->b_frame_mbs_only )
194     {
195         bs_write1( s, sh->b_field_pic );
196         if( sh->b_field_pic )
197             bs_write1( s, sh->b_bottom_field );
198     }
199
200     if( sh->i_idr_pic_id >= 0 ) /* NAL IDR */
201     {
202         bs_write_ue( s, sh->i_idr_pic_id );
203     }
204
205     if( sh->sps->i_poc_type == 0 )
206     {
207         bs_write( s, sh->sps->i_log2_max_poc_lsb, sh->i_poc_lsb & ((1<<sh->sps->i_log2_max_poc_lsb)-1) );
208         if( sh->pps->b_pic_order && !sh->b_field_pic )
209         {
210             bs_write_se( s, sh->i_delta_poc_bottom );
211         }
212     }
213     else if( sh->sps->i_poc_type == 1 && !sh->sps->b_delta_pic_order_always_zero )
214     {
215         bs_write_se( s, sh->i_delta_poc[0] );
216         if( sh->pps->b_pic_order && !sh->b_field_pic )
217         {
218             bs_write_se( s, sh->i_delta_poc[1] );
219         }
220     }
221
222     if( sh->pps->b_redundant_pic_cnt )
223     {
224         bs_write_ue( s, sh->i_redundant_pic_cnt );
225     }
226
227     if( sh->i_type == SLICE_TYPE_B )
228     {
229         bs_write1( s, sh->b_direct_spatial_mv_pred );
230     }
231     if( sh->i_type == SLICE_TYPE_P || sh->i_type == SLICE_TYPE_SP || sh->i_type == SLICE_TYPE_B )
232     {
233         bs_write1( s, sh->b_num_ref_idx_override );
234         if( sh->b_num_ref_idx_override )
235         {
236             bs_write_ue( s, sh->i_num_ref_idx_l0_active - 1 );
237             if( sh->i_type == SLICE_TYPE_B )
238             {
239                 bs_write_ue( s, sh->i_num_ref_idx_l1_active - 1 );
240             }
241         }
242     }
243
244     /* ref pic list reordering */
245     if( sh->i_type != SLICE_TYPE_I )
246     {
247         bs_write1( s, sh->b_ref_pic_list_reordering_l0 );
248         if( sh->b_ref_pic_list_reordering_l0 )
249         {
250             for( i = 0; i < sh->i_num_ref_idx_l0_active; i++ )
251             {
252                 bs_write_ue( s, sh->ref_pic_list_order[0][i].idc );
253                 bs_write_ue( s, sh->ref_pic_list_order[0][i].arg );
254
255             }
256             bs_write_ue( s, 3 );
257         }
258     }
259     if( sh->i_type == SLICE_TYPE_B )
260     {
261         bs_write1( s, sh->b_ref_pic_list_reordering_l1 );
262         if( sh->b_ref_pic_list_reordering_l1 )
263         {
264             for( i = 0; i < sh->i_num_ref_idx_l1_active; i++ )
265             {
266                 bs_write_ue( s, sh->ref_pic_list_order[1][i].idc );
267                 bs_write_ue( s, sh->ref_pic_list_order[1][i].arg );
268             }
269             bs_write_ue( s, 3 );
270         }
271     }
272
273     if( sh->pps->b_weighted_pred && ( sh->i_type == SLICE_TYPE_P || sh->i_type == SLICE_TYPE_SP ) )
274     {
275         /* pred_weight_table() */
276         bs_write_ue( s, sh->weight[0][0].i_denom );
277         bs_write_ue( s, sh->weight[0][1].i_denom );
278         for( i = 0; i < sh->i_num_ref_idx_l0_active; i++ )
279         {
280             int luma_weight_l0_flag = !!sh->weight[i][0].weightfn;
281             int chroma_weight_l0_flag = !!sh->weight[i][1].weightfn || !!sh->weight[i][2].weightfn;
282             bs_write1( s, luma_weight_l0_flag );
283             if( luma_weight_l0_flag )
284             {
285                 bs_write_se( s, sh->weight[i][0].i_scale );
286                 bs_write_se( s, sh->weight[i][0].i_offset );
287             }
288             bs_write1( s, chroma_weight_l0_flag );
289             if( chroma_weight_l0_flag )
290             {
291                 int j;
292                 for( j = 1; j < 3; j++ )
293                 {
294                     bs_write_se( s, sh->weight[i][j].i_scale );
295                     bs_write_se( s, sh->weight[i][j].i_offset );
296                 }
297             }
298         }
299     }
300     else if( sh->pps->b_weighted_bipred == 1 && sh->i_type == SLICE_TYPE_B )
301     {
302       /* TODO */
303     }
304
305     if( i_nal_ref_idc != 0 )
306     {
307         if( sh->i_idr_pic_id >= 0 )
308         {
309             bs_write1( s, 0 );  /* no output of prior pics flag */
310             bs_write1( s, 0 );  /* long term reference flag */
311         }
312         else
313         {
314             bs_write1( s, sh->i_mmco_command_count > 0 ); /* adaptive_ref_pic_marking_mode_flag */
315             if( sh->i_mmco_command_count > 0 )
316             {
317                 int i;
318                 for( i = 0; i < sh->i_mmco_command_count; i++ )
319                 {
320                     bs_write_ue( s, 1 ); /* mark short term ref as unused */
321                     bs_write_ue( s, sh->mmco[i].i_difference_of_pic_nums - 1 );
322                 }
323                 bs_write_ue( s, 0 ); /* end command list */
324             }
325         }
326     }
327
328     if( sh->pps->b_cabac && sh->i_type != SLICE_TYPE_I )
329     {
330         bs_write_ue( s, sh->i_cabac_init_idc );
331     }
332     bs_write_se( s, sh->i_qp_delta );      /* slice qp delta */
333
334     if( sh->pps->b_deblocking_filter_control )
335     {
336         bs_write_ue( s, sh->i_disable_deblocking_filter_idc );
337         if( sh->i_disable_deblocking_filter_idc != 1 )
338         {
339             bs_write_se( s, sh->i_alpha_c0_offset >> 1 );
340             bs_write_se( s, sh->i_beta_offset >> 1 );
341         }
342     }
343 }
344
345 /* If we are within a reasonable distance of the end of the memory allocated for the bitstream, */
346 /* reallocate, adding an arbitrary amount of space (100 kilobytes). */
347 static int x264_bitstream_check_buffer( x264_t *h )
348 {
349     uint8_t *bs_bak = h->out.p_bitstream;
350     if( ( h->param.b_cabac && (h->cabac.p_end - h->cabac.p < 2500) )
351      || ( h->out.bs.p_end - h->out.bs.p < 2500 ) )
352     {
353         intptr_t delta;
354         int i;
355
356         h->out.i_bitstream += 100000;
357         CHECKED_MALLOC( h->out.p_bitstream, h->out.i_bitstream );
358         h->mc.memcpy_aligned( h->out.p_bitstream, bs_bak, (h->out.i_bitstream - 100000) & ~15 );
359         delta = h->out.p_bitstream - bs_bak;
360
361         h->out.bs.p_start += delta;
362         h->out.bs.p += delta;
363         h->out.bs.p_end = h->out.p_bitstream + h->out.i_bitstream;
364
365         h->cabac.p_start += delta;
366         h->cabac.p += delta;
367         h->cabac.p_end = h->out.p_bitstream + h->out.i_bitstream;
368
369         for( i = 0; i <= h->out.i_nal; i++ )
370             h->out.nal[i].p_payload += delta;
371         x264_free( bs_bak );
372     }
373     return 0;
374 fail:
375     x264_free( bs_bak );
376     return -1;
377 }
378
379 /****************************************************************************
380  *
381  ****************************************************************************
382  ****************************** External API*********************************
383  ****************************************************************************
384  *
385  ****************************************************************************/
386
387 static int x264_validate_parameters( x264_t *h )
388 {
389 #ifdef HAVE_MMX
390     if( !(x264_cpu_detect() & X264_CPU_SSE) )
391     {
392         x264_log( h, X264_LOG_ERROR, "your cpu does not support SSE1, but x264 was compiled with asm support\n");
393         x264_log( h, X264_LOG_ERROR, "to run x264, recompile without asm support (configure --disable-asm)\n");
394         return -1;
395     }
396 #endif
397     if( h->param.i_width <= 0 || h->param.i_height <= 0 )
398     {
399         x264_log( h, X264_LOG_ERROR, "invalid width x height (%dx%d)\n",
400                   h->param.i_width, h->param.i_height );
401         return -1;
402     }
403
404     if( h->param.i_width % 2 || h->param.i_height % 2 )
405     {
406         x264_log( h, X264_LOG_ERROR, "width or height not divisible by 2 (%dx%d)\n",
407                   h->param.i_width, h->param.i_height );
408         return -1;
409     }
410     int i_csp = h->param.i_csp & X264_CSP_MASK;
411     if( i_csp != X264_CSP_I420 && i_csp != X264_CSP_YV12 )
412     {
413         x264_log( h, X264_LOG_ERROR, "invalid CSP (only I420/YV12 supported)\n" );
414         return -1;
415     }
416
417     if( h->param.i_threads == X264_THREADS_AUTO )
418         h->param.i_threads = x264_cpu_num_processors() * (h->param.b_sliced_threads?2:3)/2;
419     h->param.i_threads = x264_clip3( h->param.i_threads, 1, X264_THREAD_MAX );
420     if( h->param.i_threads > 1 )
421     {
422 #ifndef HAVE_PTHREAD
423         x264_log( h, X264_LOG_WARNING, "not compiled with pthread support!\n");
424         h->param.i_threads = 1;
425 #endif
426         /* Avoid absurdly small thread slices as they can reduce performance
427          * and VBV compliance.  Capped at an arbitrary 4 rows per thread. */
428         if( h->param.b_sliced_threads )
429         {
430             int max_threads = (h->param.i_height+15)/16 / 4;
431             h->param.i_threads = X264_MIN( h->param.i_threads, max_threads );
432         }
433     }
434     else
435         h->param.b_sliced_threads = 0;
436     h->i_thread_frames = h->param.b_sliced_threads ? 1 : h->param.i_threads;
437
438     if( h->param.b_interlaced )
439     {
440         if( h->param.analyse.i_me_method >= X264_ME_ESA )
441         {
442             x264_log( h, X264_LOG_WARNING, "interlace + me=esa is not implemented\n" );
443             h->param.analyse.i_me_method = X264_ME_UMH;
444         }
445         if( h->param.analyse.i_weighted_pred > 0 )
446         {
447             x264_log( h, X264_LOG_WARNING, "interlace + weightp is not implemented\n" );
448             h->param.analyse.i_weighted_pred = X264_WEIGHTP_NONE;
449         }
450     }
451
452     /* Detect default ffmpeg settings and terminate with an error. */
453     {
454         int score = 0;
455         score += h->param.analyse.i_me_range == 0;
456         score += h->param.rc.i_qp_step == 3;
457         score += h->param.i_keyint_max == 12;
458         score += h->param.rc.i_qp_min == 2;
459         score += h->param.rc.i_qp_max == 31;
460         score += h->param.rc.f_qcompress == 0.5;
461         score += fabs(h->param.rc.f_ip_factor - 1.25) < 0.01;
462         score += fabs(h->param.rc.f_pb_factor - 1.25) < 0.01;
463         score += h->param.analyse.inter == 0 && h->param.analyse.i_subpel_refine == 8;
464         if( score >= 5 )
465         {
466             x264_log( h, X264_LOG_ERROR, "broken ffmpeg default settings detected\n" );
467             x264_log( h, X264_LOG_ERROR, "use an encoding preset (vpre)\n" );
468             return -1;
469         }
470     }
471
472     if( h->param.rc.i_rc_method < 0 || h->param.rc.i_rc_method > 2 )
473     {
474         x264_log( h, X264_LOG_ERROR, "no ratecontrol method specified\n" );
475         return -1;
476     }
477     h->param.rc.f_rf_constant = x264_clip3f( h->param.rc.f_rf_constant, 0, 51 );
478     h->param.rc.i_qp_constant = x264_clip3( h->param.rc.i_qp_constant, 0, 51 );
479     if( h->param.rc.i_rc_method == X264_RC_CRF )
480     {
481         h->param.rc.i_qp_constant = h->param.rc.f_rf_constant;
482         h->param.rc.i_bitrate = 0;
483     }
484     if( (h->param.rc.i_rc_method == X264_RC_CQP || h->param.rc.i_rc_method == X264_RC_CRF)
485         && h->param.rc.i_qp_constant == 0 )
486     {
487         h->mb.b_lossless = 1;
488         h->param.i_cqm_preset = X264_CQM_FLAT;
489         h->param.psz_cqm_file = NULL;
490         h->param.rc.i_rc_method = X264_RC_CQP;
491         h->param.rc.f_ip_factor = 1;
492         h->param.rc.f_pb_factor = 1;
493         h->param.analyse.b_psnr = 0;
494         h->param.analyse.b_ssim = 0;
495         h->param.analyse.i_chroma_qp_offset = 0;
496         h->param.analyse.i_trellis = 0;
497         h->param.analyse.b_fast_pskip = 0;
498         h->param.analyse.i_noise_reduction = 0;
499         h->param.analyse.f_psy_rd = 0;
500         h->param.i_bframe = 0;
501         /* 8x8dct is not useful at all in CAVLC lossless */
502         if( !h->param.b_cabac )
503             h->param.analyse.b_transform_8x8 = 0;
504     }
505     if( h->param.rc.i_rc_method == X264_RC_CQP )
506     {
507         float qp_p = h->param.rc.i_qp_constant;
508         float qp_i = qp_p - 6*log(h->param.rc.f_ip_factor)/log(2);
509         float qp_b = qp_p + 6*log(h->param.rc.f_pb_factor)/log(2);
510         h->param.rc.i_qp_min = x264_clip3( (int)(X264_MIN3( qp_p, qp_i, qp_b )), 0, 51 );
511         h->param.rc.i_qp_max = x264_clip3( (int)(X264_MAX3( qp_p, qp_i, qp_b ) + .999), 0, 51 );
512         h->param.rc.i_aq_mode = 0;
513         h->param.rc.b_mb_tree = 0;
514     }
515     h->param.rc.i_qp_max = x264_clip3( h->param.rc.i_qp_max, 0, 51 );
516     h->param.rc.i_qp_min = x264_clip3( h->param.rc.i_qp_min, 0, h->param.rc.i_qp_max );
517     if( h->param.rc.i_vbv_buffer_size )
518     {
519         if( h->param.rc.i_rc_method == X264_RC_CQP )
520         {
521             x264_log( h, X264_LOG_WARNING, "VBV is incompatible with constant QP, ignored.\n" );
522             h->param.rc.i_vbv_max_bitrate = 0;
523             h->param.rc.i_vbv_buffer_size = 0;
524         }
525         else if( h->param.rc.i_vbv_max_bitrate == 0 )
526         {
527             if( h->param.rc.i_rc_method == X264_RC_ABR )
528             {
529                 x264_log( h, X264_LOG_WARNING, "VBV maxrate unspecified, assuming CBR\n" );
530                 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate;
531             }
532             else
533             {
534                 x264_log( h, X264_LOG_WARNING, "VBV bufsize set but maxrate unspecified, ignored\n" );
535                 h->param.rc.i_vbv_buffer_size = 0;
536             }
537         }
538         else if( h->param.rc.i_vbv_max_bitrate < h->param.rc.i_bitrate &&
539                  h->param.rc.i_rc_method == X264_RC_ABR )
540         {
541             x264_log( h, X264_LOG_WARNING, "max bitrate less than average bitrate, assuming CBR\n" );
542             h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate;
543         }
544     }
545     else if( h->param.rc.i_vbv_max_bitrate )
546     {
547         x264_log( h, X264_LOG_WARNING, "VBV maxrate specified, but no bufsize, ignored\n" );
548         h->param.rc.i_vbv_max_bitrate = 0;
549     }
550
551     int max_slices = (h->param.i_height+((16<<h->param.b_interlaced)-1))/(16<<h->param.b_interlaced);
552     if( h->param.b_sliced_threads )
553         h->param.i_slice_count = x264_clip3( h->param.i_threads, 0, max_slices );
554     else
555     {
556         h->param.i_slice_count = x264_clip3( h->param.i_slice_count, 0, max_slices );
557         h->param.i_slice_max_size = X264_MAX( h->param.i_slice_max_size, 0 );
558         h->param.i_slice_max_mbs = X264_MAX( h->param.i_slice_max_mbs, 0 );
559         if( h->param.b_interlaced && h->param.i_slice_max_size )
560         {
561             x264_log( h, X264_LOG_WARNING, "interlaced + slice-max-size is not implemented\n" );
562             h->param.i_slice_max_size = 0;
563         }
564         if( h->param.b_interlaced && h->param.i_slice_max_mbs )
565         {
566             x264_log( h, X264_LOG_WARNING, "interlaced + slice-max-mbs is not implemented\n" );
567             h->param.i_slice_max_mbs = 0;
568         }
569         if( h->param.i_slice_max_mbs || h->param.i_slice_max_size )
570             h->param.i_slice_count = 0;
571     }
572
573     h->param.i_frame_reference = x264_clip3( h->param.i_frame_reference, 1, 16 );
574     if( h->param.i_keyint_max <= 0 )
575         h->param.i_keyint_max = 1;
576     if( h->param.i_scenecut_threshold < 0 )
577         h->param.i_scenecut_threshold = 0;
578     if( !h->param.analyse.i_subpel_refine && h->param.analyse.i_direct_mv_pred > X264_DIRECT_PRED_SPATIAL )
579     {
580         x264_log( h, X264_LOG_WARNING, "subme=0 + direct=temporal is not supported\n" );
581         h->param.analyse.i_direct_mv_pred = X264_DIRECT_PRED_SPATIAL;
582     }
583     h->param.i_bframe = x264_clip3( h->param.i_bframe, 0, X264_BFRAME_MAX );
584     if( h->param.i_keyint_max == 1 )
585     {
586         h->param.i_bframe = 0;
587         h->param.b_intra_refresh = 0;
588     }
589     h->param.i_bframe_bias = x264_clip3( h->param.i_bframe_bias, -90, 100 );
590     if( h->param.i_bframe <= 1 )
591         h->param.i_bframe_pyramid = X264_B_PYRAMID_NONE;
592     h->param.i_bframe_pyramid = x264_clip3( h->param.i_bframe_pyramid, X264_B_PYRAMID_NONE, X264_B_PYRAMID_NORMAL );
593     if( !h->param.i_bframe )
594     {
595         h->param.i_bframe_adaptive = X264_B_ADAPT_NONE;
596         h->param.analyse.i_direct_mv_pred = 0;
597         h->param.analyse.b_weighted_bipred = 0;
598     }
599     if( h->param.b_intra_refresh && h->param.i_bframe_pyramid == X264_B_PYRAMID_NORMAL )
600     {
601         x264_log( h, X264_LOG_WARNING, "b-pyramid normal + intra-refresh is not supported\n" );
602         h->param.i_bframe_pyramid = X264_B_PYRAMID_STRICT;
603     }
604     if( h->param.b_intra_refresh && h->param.i_frame_reference > 1 )
605     {
606         x264_log( h, X264_LOG_WARNING, "ref > 1 + intra-refresh is not supported\n" );
607         h->param.i_frame_reference = 1;
608     }
609     h->param.i_keyint_min = x264_clip3( h->param.i_keyint_min, 1, h->param.i_keyint_max/2+1 );
610     h->param.rc.i_lookahead = x264_clip3( h->param.rc.i_lookahead, 0, X264_LOOKAHEAD_MAX );
611     {
612         int maxrate = X264_MAX( h->param.rc.i_vbv_max_bitrate, h->param.rc.i_bitrate );
613         float bufsize = maxrate ? (float)h->param.rc.i_vbv_buffer_size / maxrate : 0;
614         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;
615         h->param.rc.i_lookahead = X264_MIN( h->param.rc.i_lookahead, X264_MAX( h->param.i_keyint_max, bufsize*fps ) );
616     }
617
618     if( !h->param.i_timebase_num || !h->param.i_timebase_den )
619     {
620         h->param.i_timebase_num = h->param.i_fps_den;
621         h->param.i_timebase_den = h->param.i_fps_num;
622     }
623
624     h->param.rc.f_qcompress = x264_clip3f( h->param.rc.f_qcompress, 0.0, 1.0 );
625     if( !h->param.rc.i_lookahead || h->param.i_keyint_max == 1 || h->param.rc.f_qcompress == 1 )
626         h->param.rc.b_mb_tree = 0;
627     if( h->param.rc.b_stat_read )
628         h->param.rc.i_lookahead = 0;
629 #ifdef HAVE_PTHREAD
630     if( h->param.i_sync_lookahead )
631         h->param.i_sync_lookahead = x264_clip3( h->param.i_sync_lookahead, h->i_thread_frames + h->param.i_bframe, X264_LOOKAHEAD_MAX );
632     if( h->param.rc.b_stat_read || h->i_thread_frames == 1 )
633         h->param.i_sync_lookahead = 0;
634 #else
635     h->param.i_sync_lookahead = 0;
636 #endif
637
638     h->param.i_deblocking_filter_alphac0 = x264_clip3( h->param.i_deblocking_filter_alphac0, -6, 6 );
639     h->param.i_deblocking_filter_beta    = x264_clip3( h->param.i_deblocking_filter_beta, -6, 6 );
640     h->param.analyse.i_luma_deadzone[0] = x264_clip3( h->param.analyse.i_luma_deadzone[0], 0, 32 );
641     h->param.analyse.i_luma_deadzone[1] = x264_clip3( h->param.analyse.i_luma_deadzone[1], 0, 32 );
642
643     h->param.i_cabac_init_idc = x264_clip3( h->param.i_cabac_init_idc, 0, 2 );
644
645     if( h->param.i_cqm_preset < X264_CQM_FLAT || h->param.i_cqm_preset > X264_CQM_CUSTOM )
646         h->param.i_cqm_preset = X264_CQM_FLAT;
647
648     if( h->param.analyse.i_me_method < X264_ME_DIA ||
649         h->param.analyse.i_me_method > X264_ME_TESA )
650         h->param.analyse.i_me_method = X264_ME_HEX;
651     if( h->param.analyse.i_me_range < 4 )
652         h->param.analyse.i_me_range = 4;
653     if( h->param.analyse.i_me_range > 16 && h->param.analyse.i_me_method <= X264_ME_HEX )
654         h->param.analyse.i_me_range = 16;
655     if( h->param.analyse.i_me_method == X264_ME_TESA &&
656         (h->mb.b_lossless || h->param.analyse.i_subpel_refine <= 1) )
657         h->param.analyse.i_me_method = X264_ME_ESA;
658     h->param.analyse.i_subpel_refine = x264_clip3( h->param.analyse.i_subpel_refine, 0, 10 );
659     h->param.analyse.b_mixed_references = h->param.analyse.b_mixed_references && h->param.i_frame_reference > 1;
660     h->param.analyse.inter &= X264_ANALYSE_PSUB16x16|X264_ANALYSE_PSUB8x8|X264_ANALYSE_BSUB16x16|
661                               X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
662     h->param.analyse.intra &= X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
663     if( !(h->param.analyse.inter & X264_ANALYSE_PSUB16x16) )
664         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
665     if( !h->param.analyse.b_transform_8x8 )
666     {
667         h->param.analyse.inter &= ~X264_ANALYSE_I8x8;
668         h->param.analyse.intra &= ~X264_ANALYSE_I8x8;
669     }
670     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -12, 12);
671     if( !h->param.b_cabac )
672         h->param.analyse.i_trellis = 0;
673     h->param.analyse.i_trellis = x264_clip3( h->param.analyse.i_trellis, 0, 2 );
674     if( !h->param.analyse.b_psy )
675     {
676         h->param.analyse.f_psy_rd = 0;
677         h->param.analyse.f_psy_trellis = 0;
678     }
679     if( !h->param.analyse.i_trellis )
680         h->param.analyse.f_psy_trellis = 0;
681     h->param.analyse.f_psy_rd = x264_clip3f( h->param.analyse.f_psy_rd, 0, 10 );
682     h->param.analyse.f_psy_trellis = x264_clip3f( h->param.analyse.f_psy_trellis, 0, 10 );
683     if( h->param.analyse.i_subpel_refine < 6 )
684         h->param.analyse.f_psy_rd = 0;
685     h->mb.i_psy_rd = FIX8( h->param.analyse.f_psy_rd );
686     /* Psy RDO increases overall quantizers to improve the quality of luma--this indirectly hurts chroma quality */
687     /* so we lower the chroma QP offset to compensate */
688     /* This can be triggered repeatedly on multiple calls to parameter_validate, but since encoding
689      * uses the pps chroma qp offset not the param chroma qp offset, this is not a problem. */
690     if( h->mb.i_psy_rd )
691         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_rd < 0.25 ? 1 : 2;
692     h->mb.i_psy_trellis = FIX8( h->param.analyse.f_psy_trellis / 4 );
693     /* Psy trellis has a similar effect. */
694     if( h->mb.i_psy_trellis )
695         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_trellis < 0.25 ? 1 : 2;
696     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -12, 12);
697     h->param.rc.i_aq_mode = x264_clip3( h->param.rc.i_aq_mode, 0, 2 );
698     h->param.rc.f_aq_strength = x264_clip3f( h->param.rc.f_aq_strength, 0, 3 );
699     if( h->param.rc.f_aq_strength == 0 )
700         h->param.rc.i_aq_mode = 0;
701     /* MB-tree requires AQ to be on, even if the strength is zero. */
702     if( !h->param.rc.i_aq_mode && h->param.rc.b_mb_tree )
703     {
704         h->param.rc.i_aq_mode = 1;
705         h->param.rc.f_aq_strength = 0;
706     }
707     h->param.analyse.i_noise_reduction = x264_clip3( h->param.analyse.i_noise_reduction, 0, 1<<16 );
708     if( h->param.analyse.i_subpel_refine == 10 && (h->param.analyse.i_trellis != 2 || !h->param.rc.i_aq_mode) )
709         h->param.analyse.i_subpel_refine = 9;
710
711     {
712         const x264_level_t *l = x264_levels;
713         if( h->param.i_level_idc < 0 )
714         {
715             int maxrate_bak = h->param.rc.i_vbv_max_bitrate;
716             if( h->param.rc.i_rc_method == X264_RC_ABR && h->param.rc.i_vbv_buffer_size <= 0 )
717                 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate * 2;
718             h->sps = h->sps_array;
719             x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
720             do h->param.i_level_idc = l->level_idc;
721                 while( l[1].level_idc && x264_validate_levels( h, 0 ) && l++ );
722             h->param.rc.i_vbv_max_bitrate = maxrate_bak;
723         }
724         else
725         {
726             while( l->level_idc && l->level_idc != h->param.i_level_idc )
727                 l++;
728             if( l->level_idc == 0 )
729             {
730                 x264_log( h, X264_LOG_ERROR, "invalid level_idc: %d\n", h->param.i_level_idc );
731                 return -1;
732             }
733         }
734         if( h->param.analyse.i_mv_range <= 0 )
735             h->param.analyse.i_mv_range = l->mv_range >> h->param.b_interlaced;
736         else
737             h->param.analyse.i_mv_range = x264_clip3(h->param.analyse.i_mv_range, 32, 512 >> h->param.b_interlaced);
738     }
739
740     h->param.analyse.i_weighted_pred = x264_clip3( h->param.analyse.i_weighted_pred, 0, X264_WEIGHTP_SMART );
741     if( !h->param.analyse.i_weighted_pred && h->param.rc.b_mb_tree && h->param.analyse.b_psy && !h->param.b_interlaced )
742         h->param.analyse.i_weighted_pred = X264_WEIGHTP_FAKE;
743
744     if( h->i_thread_frames > 1 )
745     {
746         int r = h->param.analyse.i_mv_range_thread;
747         int r2;
748         if( r <= 0 )
749         {
750             // half of the available space is reserved and divided evenly among the threads,
751             // the rest is allocated to whichever thread is far enough ahead to use it.
752             // reserving more space increases quality for some videos, but costs more time
753             // in thread synchronization.
754             int max_range = (h->param.i_height + X264_THREAD_HEIGHT) / h->i_thread_frames - X264_THREAD_HEIGHT;
755             r = max_range / 2;
756         }
757         r = X264_MAX( r, h->param.analyse.i_me_range );
758         r = X264_MIN( r, h->param.analyse.i_mv_range );
759         // round up to use the whole mb row
760         r2 = (r & ~15) + ((-X264_THREAD_HEIGHT) & 15);
761         if( r2 < r )
762             r2 += 16;
763         x264_log( h, X264_LOG_DEBUG, "using mv_range_thread = %d\n", r2 );
764         h->param.analyse.i_mv_range_thread = r2;
765     }
766
767     if( h->param.rc.f_qblur < 0 )
768         h->param.rc.f_qblur = 0;
769     if( h->param.rc.f_complexity_blur < 0 )
770         h->param.rc.f_complexity_blur = 0;
771
772     h->param.i_sps_id &= 31;
773
774     if( h->param.i_log_level < X264_LOG_INFO )
775     {
776         h->param.analyse.b_psnr = 0;
777         h->param.analyse.b_ssim = 0;
778     }
779
780     /* ensure the booleans are 0 or 1 so they can be used in math */
781 #define BOOLIFY(x) h->param.x = !!h->param.x
782     BOOLIFY( b_cabac );
783     BOOLIFY( b_constrained_intra );
784     BOOLIFY( b_deblocking_filter );
785     BOOLIFY( b_deterministic );
786     BOOLIFY( b_sliced_threads );
787     BOOLIFY( b_interlaced );
788     BOOLIFY( b_intra_refresh );
789     BOOLIFY( b_visualize );
790     BOOLIFY( b_aud );
791     BOOLIFY( b_repeat_headers );
792     BOOLIFY( b_annexb );
793     BOOLIFY( analyse.b_transform_8x8 );
794     BOOLIFY( analyse.b_weighted_bipred );
795     BOOLIFY( analyse.b_chroma_me );
796     BOOLIFY( analyse.b_mixed_references );
797     BOOLIFY( analyse.b_fast_pskip );
798     BOOLIFY( analyse.b_dct_decimate );
799     BOOLIFY( analyse.b_psy );
800     BOOLIFY( analyse.b_psnr );
801     BOOLIFY( analyse.b_ssim );
802     BOOLIFY( rc.b_stat_write );
803     BOOLIFY( rc.b_stat_read );
804     BOOLIFY( rc.b_mb_tree );
805 #undef BOOLIFY
806
807     return 0;
808 }
809
810 static void mbcmp_init( x264_t *h )
811 {
812     int satd = !h->mb.b_lossless && h->param.analyse.i_subpel_refine > 1;
813     memcpy( h->pixf.mbcmp, satd ? h->pixf.satd : h->pixf.sad_aligned, sizeof(h->pixf.mbcmp) );
814     memcpy( h->pixf.mbcmp_unaligned, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.mbcmp_unaligned) );
815     h->pixf.intra_mbcmp_x3_16x16 = satd ? h->pixf.intra_satd_x3_16x16 : h->pixf.intra_sad_x3_16x16;
816     h->pixf.intra_mbcmp_x3_8x8c = satd ? h->pixf.intra_satd_x3_8x8c : h->pixf.intra_sad_x3_8x8c;
817     h->pixf.intra_mbcmp_x3_4x4 = satd ? h->pixf.intra_satd_x3_4x4 : h->pixf.intra_sad_x3_4x4;
818     satd &= h->param.analyse.i_me_method == X264_ME_TESA;
819     memcpy( h->pixf.fpelcmp, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.fpelcmp) );
820     memcpy( h->pixf.fpelcmp_x3, satd ? h->pixf.satd_x3 : h->pixf.sad_x3, sizeof(h->pixf.fpelcmp_x3) );
821     memcpy( h->pixf.fpelcmp_x4, satd ? h->pixf.satd_x4 : h->pixf.sad_x4, sizeof(h->pixf.fpelcmp_x4) );
822 }
823
824 static void x264_set_aspect_ratio( x264_t *h, x264_param_t *param, int initial )
825 {
826     /* VUI */
827     if( param->vui.i_sar_width > 0 && param->vui.i_sar_height > 0 )
828     {
829         int i_w = param->vui.i_sar_width;
830         int i_h = param->vui.i_sar_height;
831         int old_w = h->param.vui.i_sar_width;
832         int old_h = h->param.vui.i_sar_height;
833
834         x264_reduce_fraction( &i_w, &i_h );
835
836         while( i_w > 65535 || i_h > 65535 )
837         {
838             i_w /= 2;
839             i_h /= 2;
840         }
841
842         x264_reduce_fraction( &i_w, &i_h );
843
844         if( i_w != old_w || i_h != old_h || initial )
845         {
846             h->param.vui.i_sar_width = 0;
847             h->param.vui.i_sar_height = 0;
848             if( i_w == 0 || i_h == 0 )
849                 x264_log( h, X264_LOG_WARNING, "cannot create valid sample aspect ratio\n" );
850             else
851             {
852                 x264_log( h, initial?X264_LOG_INFO:X264_LOG_DEBUG, "using SAR=%d/%d\n", i_w, i_h );
853                 h->param.vui.i_sar_width = i_w;
854                 h->param.vui.i_sar_height = i_h;
855             }
856         }
857     }
858 }
859
860 /****************************************************************************
861  * x264_encoder_open:
862  ****************************************************************************/
863 x264_t *x264_encoder_open( x264_param_t *param )
864 {
865     x264_t *h;
866     char buf[1000], *p;
867     int i, qp, i_slicetype_length;
868
869     CHECKED_MALLOCZERO( h, sizeof(x264_t) );
870
871     /* Create a copy of param */
872     memcpy( &h->param, param, sizeof(x264_param_t) );
873
874     if( param->param_free )
875         param->param_free( param );
876
877     if( x264_validate_parameters( h ) < 0 )
878         goto fail;
879
880     if( h->param.psz_cqm_file )
881         if( x264_cqm_parse_file( h, h->param.psz_cqm_file ) < 0 )
882             goto fail;
883
884     if( h->param.rc.psz_stat_out )
885         h->param.rc.psz_stat_out = strdup( h->param.rc.psz_stat_out );
886     if( h->param.rc.psz_stat_in )
887         h->param.rc.psz_stat_in = strdup( h->param.rc.psz_stat_in );
888
889     x264_set_aspect_ratio( h, &h->param, 1 );
890
891     x264_reduce_fraction( &h->param.i_fps_num, &h->param.i_fps_den );
892     x264_reduce_fraction( &h->param.i_timebase_num, &h->param.i_timebase_den );
893
894     /* Init x264_t */
895     h->i_frame = -1;
896     h->i_frame_num = 0;
897     h->i_idr_pic_id = 0;
898     if( h->param.b_dts_compress )
899     {
900         /* h->i_dts_compress_multiplier == h->frames.i_bframe_delay + 1 */
901         h->i_dts_compress_multiplier = h->param.i_bframe ? (h->param.i_bframe_pyramid ? 3 : 2) : 1;
902         if( h->i_dts_compress_multiplier != 1 )
903             x264_log( h, X264_LOG_DEBUG, "DTS compresion changed timebase: %d/%d -> %d/%d\n",
904                       h->param.i_timebase_num, h->param.i_timebase_den,
905                       h->param.i_timebase_num, h->param.i_timebase_den * h->i_dts_compress_multiplier );
906         h->param.i_timebase_den *= h->i_dts_compress_multiplier;
907     }
908     else
909         h->i_dts_compress_multiplier = 1;
910
911     h->sps = &h->sps_array[0];
912     x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
913
914     h->pps = &h->pps_array[0];
915     x264_pps_init( h->pps, h->param.i_sps_id, &h->param, h->sps );
916
917     x264_validate_levels( h, 1 );
918
919     h->chroma_qp_table = i_chroma_qp_table + 12 + h->pps->i_chroma_qp_index_offset;
920
921     if( x264_cqm_init( h ) < 0 )
922         goto fail;
923
924     h->mb.i_mb_count = h->sps->i_mb_width * h->sps->i_mb_height;
925
926     /* Init frames. */
927     if( h->param.i_bframe_adaptive == X264_B_ADAPT_TRELLIS )
928         h->frames.i_delay = X264_MAX(h->param.i_bframe,3)*4;
929     else
930         h->frames.i_delay = h->param.i_bframe;
931     if( h->param.rc.b_mb_tree || h->param.rc.i_vbv_buffer_size )
932         h->frames.i_delay = X264_MAX( h->frames.i_delay, h->param.rc.i_lookahead );
933     i_slicetype_length = h->frames.i_delay;
934     h->frames.i_delay += h->i_thread_frames - 1;
935     h->frames.i_delay = X264_MIN( h->frames.i_delay, X264_LOOKAHEAD_MAX );
936     h->frames.i_delay += h->param.i_sync_lookahead;
937     h->frames.i_bframe_delay = h->param.i_bframe ? (h->param.i_bframe_pyramid ? 2 : 1) : 0;
938
939     h->frames.i_max_ref0 = h->param.i_frame_reference;
940     h->frames.i_max_ref1 = h->sps->vui.i_num_reorder_frames;
941     h->frames.i_max_dpb  = h->sps->vui.i_max_dec_frame_buffering;
942     h->frames.b_have_lowres = !h->param.rc.b_stat_read
943         && ( h->param.rc.i_rc_method == X264_RC_ABR
944           || h->param.rc.i_rc_method == X264_RC_CRF
945           || h->param.i_bframe_adaptive
946           || h->param.i_scenecut_threshold
947           || h->param.rc.b_mb_tree
948           || h->param.analyse.i_weighted_pred == X264_WEIGHTP_SMART );
949     h->frames.b_have_lowres |= h->param.rc.b_stat_read && h->param.rc.i_vbv_buffer_size > 0;
950     h->frames.b_have_sub8x8_esa = !!(h->param.analyse.inter & X264_ANALYSE_PSUB8x8);
951
952     h->frames.i_last_keyframe = - h->param.i_keyint_max;
953     h->frames.i_input    = 0;
954
955     CHECKED_MALLOCZERO( h->frames.unused[0], (h->frames.i_delay + 3) * sizeof(x264_frame_t *) );
956     /* Allocate room for max refs plus a few extra just in case. */
957     CHECKED_MALLOCZERO( h->frames.unused[1], (h->i_thread_frames + 20) * sizeof(x264_frame_t *) );
958     CHECKED_MALLOCZERO( h->frames.current, (h->param.i_sync_lookahead + h->param.i_bframe
959                         + h->i_thread_frames + 3) * sizeof(x264_frame_t *) );
960     if( h->param.analyse.i_weighted_pred > 0 )
961         CHECKED_MALLOCZERO( h->frames.blank_unused, h->i_thread_frames * 4 * sizeof(x264_frame_t *) );
962     h->i_ref0 = 0;
963     h->i_ref1 = 0;
964
965     x264_rdo_init();
966
967     /* init CPU functions */
968     x264_predict_16x16_init( h->param.cpu, h->predict_16x16 );
969     x264_predict_8x8c_init( h->param.cpu, h->predict_8x8c );
970     x264_predict_8x8_init( h->param.cpu, h->predict_8x8, &h->predict_8x8_filter );
971     x264_predict_4x4_init( h->param.cpu, h->predict_4x4 );
972     if( !h->param.b_cabac )
973         x264_init_vlc_tables();
974     x264_pixel_init( h->param.cpu, &h->pixf );
975     x264_dct_init( h->param.cpu, &h->dctf );
976     x264_zigzag_init( h->param.cpu, &h->zigzagf, h->param.b_interlaced );
977     x264_mc_init( h->param.cpu, &h->mc );
978     x264_quant_init( h, h->param.cpu, &h->quantf );
979     x264_deblock_init( h->param.cpu, &h->loopf );
980     x264_dct_init_weights();
981
982     mbcmp_init( h );
983
984     p = buf + sprintf( buf, "using cpu capabilities:" );
985     for( i=0; x264_cpu_names[i].flags; i++ )
986     {
987         if( !strcmp(x264_cpu_names[i].name, "SSE2")
988             && h->param.cpu & (X264_CPU_SSE2_IS_FAST|X264_CPU_SSE2_IS_SLOW) )
989             continue;
990         if( !strcmp(x264_cpu_names[i].name, "SSE3")
991             && (h->param.cpu & X264_CPU_SSSE3 || !(h->param.cpu & X264_CPU_CACHELINE_64)) )
992             continue;
993         if( !strcmp(x264_cpu_names[i].name, "SSE4.1")
994             && (h->param.cpu & X264_CPU_SSE42) )
995             continue;
996         if( (h->param.cpu & x264_cpu_names[i].flags) == x264_cpu_names[i].flags
997             && (!i || x264_cpu_names[i].flags != x264_cpu_names[i-1].flags) )
998             p += sprintf( p, " %s", x264_cpu_names[i].name );
999     }
1000     if( !h->param.cpu )
1001         p += sprintf( p, " none!" );
1002     x264_log( h, X264_LOG_INFO, "%s\n", buf );
1003
1004     for( qp = h->param.rc.i_qp_min; qp <= h->param.rc.i_qp_max; qp++ )
1005         if( x264_analyse_init_costs( h, qp ) )
1006             goto fail;
1007     if( x264_analyse_init_costs( h, X264_LOOKAHEAD_QP ) )
1008         goto fail;
1009     if( h->cost_mv[1][2013] != 24 )
1010     {
1011         x264_log( h, X264_LOG_ERROR, "MV cost test failed: x264 has been miscompiled!\n" );
1012         goto fail;
1013     }
1014
1015     h->out.i_nal = 0;
1016     h->out.i_bitstream = X264_MAX( 1000000, h->param.i_width * h->param.i_height * 4
1017         * ( h->param.rc.i_rc_method == X264_RC_ABR ? pow( 0.95, h->param.rc.i_qp_min )
1018           : pow( 0.95, h->param.rc.i_qp_constant ) * X264_MAX( 1, h->param.rc.f_ip_factor )));
1019
1020     CHECKED_MALLOC( h->nal_buffer, h->out.i_bitstream * 3/2 + 4 );
1021     h->nal_buffer_size = h->out.i_bitstream * 3/2 + 4;
1022
1023     h->thread[0] = h;
1024     for( i = 1; i < h->param.i_threads + !!h->param.i_sync_lookahead; i++ )
1025         CHECKED_MALLOC( h->thread[i], sizeof(x264_t) );
1026
1027     if( x264_lookahead_init( h, i_slicetype_length ) )
1028         goto fail;
1029
1030     for( i = 0; i < h->param.i_threads; i++ )
1031     {
1032         int init_nal_count = h->param.i_slice_count + 3;
1033         int allocate_threadlocal_data = !h->param.b_sliced_threads || !i;
1034         if( i > 0 )
1035             *h->thread[i] = *h;
1036
1037         if( allocate_threadlocal_data )
1038         {
1039             h->thread[i]->fdec = x264_frame_pop_unused( h, 1 );
1040             if( !h->thread[i]->fdec )
1041                 goto fail;
1042         }
1043         else
1044             h->thread[i]->fdec = h->thread[0]->fdec;
1045
1046         CHECKED_MALLOC( h->thread[i]->out.p_bitstream, h->out.i_bitstream );
1047         /* Start each thread with room for init_nal_count NAL units; it'll realloc later if needed. */
1048         CHECKED_MALLOC( h->thread[i]->out.nal, init_nal_count*sizeof(x264_nal_t) );
1049         h->thread[i]->out.i_nals_allocated = init_nal_count;
1050
1051         if( allocate_threadlocal_data && x264_macroblock_cache_init( h->thread[i] ) < 0 )
1052             goto fail;
1053     }
1054
1055     /* Allocate scratch buffer */
1056     for( i = 0; i < h->param.i_threads + !!h->param.i_sync_lookahead; i++ )
1057     {
1058         int buf_hpel = (h->fdec->i_width[0]+48) * sizeof(int16_t);
1059         int buf_ssim = h->param.analyse.b_ssim * 8 * (h->param.i_width/4+3) * sizeof(int);
1060         int me_range = X264_MIN(h->param.analyse.i_me_range, h->param.analyse.i_mv_range);
1061         int buf_tesa = (h->param.analyse.i_me_method >= X264_ME_ESA) *
1062             ((me_range*2+18) * sizeof(int16_t) + (me_range+4) * (me_range+1) * 4 * sizeof(mvsad_t));
1063         int buf_mbtree = h->param.rc.b_mb_tree * ((h->sps->i_mb_width+3)&~3) * sizeof(int);
1064         int buf_nnz = !h->param.b_cabac * h->pps->b_transform_8x8_mode * (h->sps->i_mb_width * 4 * 16 * sizeof(uint8_t));
1065         int scratch_size = X264_MAX4( buf_hpel, buf_ssim, buf_tesa, X264_MAX( buf_mbtree, buf_nnz ) );
1066         CHECKED_MALLOC( h->thread[i]->scratch_buffer, scratch_size );
1067     }
1068
1069     if( x264_ratecontrol_new( h ) < 0 )
1070         goto fail;
1071
1072     if( h->param.psz_dump_yuv )
1073     {
1074         /* create or truncate the reconstructed video file */
1075         FILE *f = fopen( h->param.psz_dump_yuv, "w" );
1076         if( !f )
1077         {
1078             x264_log( h, X264_LOG_ERROR, "dump_yuv: can't write to %s\n", h->param.psz_dump_yuv );
1079             goto fail;
1080         }
1081         else if( !x264_is_regular_file( f ) )
1082         {
1083             x264_log( h, X264_LOG_ERROR, "dump_yuv: incompatible with non-regular file %s\n", h->param.psz_dump_yuv );
1084             goto fail;
1085         }
1086         fclose( f );
1087     }
1088
1089     x264_log( h, X264_LOG_INFO, "profile %s, level %d.%d\n",
1090         h->sps->i_profile_idc == PROFILE_BASELINE ? "Baseline" :
1091         h->sps->i_profile_idc == PROFILE_MAIN ? "Main" :
1092         h->sps->i_profile_idc == PROFILE_HIGH ? "High" :
1093         "High 4:4:4 Predictive", h->sps->i_level_idc/10, h->sps->i_level_idc%10 );
1094
1095     return h;
1096 fail:
1097     x264_free( h );
1098     return NULL;
1099 }
1100
1101 /****************************************************************************
1102  * x264_encoder_reconfig:
1103  ****************************************************************************/
1104 int x264_encoder_reconfig( x264_t *h, x264_param_t *param )
1105 {
1106     h = h->thread[h->thread[0]->i_thread_phase];
1107     x264_set_aspect_ratio( h, param, 0 );
1108 #define COPY(var) h->param.var = param->var
1109     COPY( i_frame_reference ); // but never uses more refs than initially specified
1110     COPY( i_bframe_bias );
1111     if( h->param.i_scenecut_threshold )
1112         COPY( i_scenecut_threshold ); // can't turn it on or off, only vary the threshold
1113     COPY( b_deblocking_filter );
1114     COPY( i_deblocking_filter_alphac0 );
1115     COPY( i_deblocking_filter_beta );
1116     COPY( analyse.inter );
1117     COPY( analyse.intra );
1118     COPY( analyse.i_direct_mv_pred );
1119     /* Scratch buffer prevents me_range from being increased for esa/tesa */
1120     if( h->param.analyse.i_me_method < X264_ME_ESA || param->analyse.i_me_range < h->param.analyse.i_me_range )
1121         COPY( analyse.i_me_range );
1122     COPY( analyse.i_noise_reduction );
1123     /* We can't switch out of subme=0 during encoding. */
1124     if( h->param.analyse.i_subpel_refine )
1125         COPY( analyse.i_subpel_refine );
1126     COPY( analyse.i_trellis );
1127     COPY( analyse.b_chroma_me );
1128     COPY( analyse.b_dct_decimate );
1129     COPY( analyse.b_fast_pskip );
1130     COPY( analyse.b_mixed_references );
1131     COPY( analyse.f_psy_rd );
1132     COPY( analyse.f_psy_trellis );
1133     // can only twiddle these if they were enabled to begin with:
1134     if( h->param.analyse.i_me_method >= X264_ME_ESA || param->analyse.i_me_method < X264_ME_ESA )
1135         COPY( analyse.i_me_method );
1136     if( h->param.analyse.i_me_method >= X264_ME_ESA && !h->frames.b_have_sub8x8_esa )
1137         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
1138     if( h->pps->b_transform_8x8_mode )
1139         COPY( analyse.b_transform_8x8 );
1140     if( h->frames.i_max_ref1 > 1 )
1141         COPY( i_bframe_pyramid );
1142     COPY( i_slice_max_size );
1143     COPY( i_slice_max_mbs );
1144     COPY( i_slice_count );
1145     /* VBV can't be turned on if it wasn't on to begin with */
1146     if( h->param.rc.i_vbv_max_bitrate > 0 && h->param.rc.i_vbv_buffer_size > 0 &&
1147           param->rc.i_vbv_max_bitrate > 0 &&   param->rc.i_vbv_buffer_size > 0 )
1148     {
1149         COPY( rc.i_vbv_max_bitrate );
1150         COPY( rc.i_vbv_buffer_size );
1151         COPY( rc.i_bitrate );
1152     }
1153     COPY( rc.f_rf_constant );
1154 #undef COPY
1155
1156     mbcmp_init( h );
1157
1158     int ret = x264_validate_parameters( h );
1159
1160     /* Supported reconfiguration options (1-pass only):
1161      * vbv-maxrate
1162      * vbv-bufsize
1163      * crf
1164      * bitrate (CBR only) */
1165     if( !ret )
1166         x264_ratecontrol_init_reconfigurable( h, 0 );
1167
1168     return ret;
1169 }
1170
1171 /****************************************************************************
1172  * x264_encoder_parameters:
1173  ****************************************************************************/
1174 void x264_encoder_parameters( x264_t *h, x264_param_t *param )
1175 {
1176     memcpy( param, &h->thread[h->i_thread_phase]->param, sizeof(x264_param_t) );
1177 }
1178
1179 /* internal usage */
1180 static void x264_nal_start( x264_t *h, int i_type, int i_ref_idc )
1181 {
1182     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1183
1184     nal->i_ref_idc = i_ref_idc;
1185     nal->i_type    = i_type;
1186
1187     nal->i_payload= 0;
1188     nal->p_payload= &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8];
1189 }
1190 /* if number of allocated nals is not enough, re-allocate a larger one. */
1191 static int x264_nal_check_buffer( x264_t *h )
1192 {
1193     if( h->out.i_nal >= h->out.i_nals_allocated )
1194     {
1195         x264_nal_t *new_out = x264_malloc( sizeof(x264_nal_t) * (h->out.i_nals_allocated*2) );
1196         if( !new_out )
1197             return -1;
1198         memcpy( new_out, h->out.nal, sizeof(x264_nal_t) * (h->out.i_nals_allocated) );
1199         x264_free( h->out.nal );
1200         h->out.nal = new_out;
1201         h->out.i_nals_allocated *= 2;
1202     }
1203     return 0;
1204 }
1205 static int x264_nal_end( x264_t *h )
1206 {
1207     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1208     nal->i_payload = &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8] - nal->p_payload;
1209     h->out.i_nal++;
1210
1211     return x264_nal_check_buffer( h );
1212 }
1213
1214 static int x264_encoder_encapsulate_nals( x264_t *h )
1215 {
1216     int nal_size = 0, i;
1217     for( i = 0; i < h->out.i_nal; i++ )
1218         nal_size += h->out.nal[i].i_payload;
1219
1220     /* Worst-case NAL unit escaping: reallocate the buffer if it's too small. */
1221     if( h->nal_buffer_size < nal_size * 3/2 + h->out.i_nal * 4 )
1222     {
1223         uint8_t *buf = x264_malloc( nal_size * 2 + h->out.i_nal * 4 );
1224         if( !buf )
1225             return -1;
1226         x264_free( h->nal_buffer );
1227         h->nal_buffer = buf;
1228     }
1229
1230     uint8_t *nal_buffer = h->nal_buffer;
1231     int long_startcode = 1;
1232
1233     for( i = 0; i < h->out.i_nal; i++ )
1234     {
1235         int size = x264_nal_encode( nal_buffer, &h->out.nal[i], h->param.b_annexb, long_startcode );
1236         /* Don't use long startcodes for any slice beyond the first. */
1237         if( h->out.nal[i].i_type >= NAL_SLICE && h->out.nal[i].i_type <= NAL_SLICE_IDR )
1238             long_startcode = 0;
1239         h->out.nal[i].i_payload = size;
1240         h->out.nal[i].p_payload = nal_buffer;
1241         nal_buffer += size;
1242     }
1243
1244     return nal_buffer - h->nal_buffer;
1245 }
1246
1247 /****************************************************************************
1248  * x264_encoder_headers:
1249  ****************************************************************************/
1250 int x264_encoder_headers( x264_t *h, x264_nal_t **pp_nal, int *pi_nal )
1251 {
1252     int frame_size = 0;
1253     /* init bitstream context */
1254     h->out.i_nal = 0;
1255     bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
1256
1257     /* Write SEI, SPS and PPS. */
1258     x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
1259     if( x264_sei_version_write( h, &h->out.bs ) )
1260         return -1;
1261     if( x264_nal_end( h ) )
1262         return -1;
1263
1264     /* generate sequence parameters */
1265     x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
1266     x264_sps_write( &h->out.bs, h->sps );
1267     if( x264_nal_end( h ) )
1268         return -1;
1269
1270     /* generate picture parameters */
1271     x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
1272     x264_pps_write( &h->out.bs, h->pps );
1273     if( x264_nal_end( h ) )
1274         return -1;
1275
1276     frame_size = x264_encoder_encapsulate_nals( h );
1277
1278     /* now set output*/
1279     *pi_nal = h->out.i_nal;
1280     *pp_nal = &h->out.nal[0];
1281     h->out.i_nal = 0;
1282
1283     return frame_size;
1284 }
1285
1286 /* Check to see whether we have chosen a reference list ordering different
1287  * from the standard's default. */
1288 static inline void x264_reference_check_reorder( x264_t *h )
1289 {
1290     int i;
1291     for( i = 0; i < h->i_ref0 - 1; i++ )
1292         /* P and B-frames use different default orders. */
1293         if( h->sh.i_type == SLICE_TYPE_P ? h->fref0[i]->i_frame_num < h->fref0[i+1]->i_frame_num
1294                                          : h->fref0[i]->i_poc < h->fref0[i+1]->i_poc )
1295         {
1296             h->b_ref_reorder[0] = 1;
1297             break;
1298         }
1299 }
1300
1301 /* return -1 on failure, else return the index of the new reference frame */
1302 int x264_weighted_reference_duplicate( x264_t *h, int i_ref, const x264_weight_t *w )
1303 {
1304     int i = h->i_ref0;
1305     int j;
1306     x264_frame_t *newframe;
1307     if( i <= 1 ) /* empty list, definitely can't duplicate frame */
1308         return -1;
1309
1310     /* Find a place to insert the duplicate in the reference list. */
1311     for( j = 0; j < i; j++ )
1312         if( h->fref0[i_ref]->i_frame != h->fref0[j]->i_frame )
1313         {
1314             /* found a place, after j, make sure there is not already a duplicate there */
1315             if( j == i-1 || ( h->fref0[j+1] && h->fref0[i_ref]->i_frame != h->fref0[j+1]->i_frame ) )
1316                 break;
1317         }
1318
1319     if( j == i ) /* No room in the reference list for the duplicate. */
1320         return -1;
1321     j++;
1322
1323     newframe = x264_frame_pop_blank_unused( h );
1324
1325     //FIXME: probably don't need to copy everything
1326     *newframe = *h->fref0[i_ref];
1327     newframe->i_reference_count = 1;
1328     newframe->orig = h->fref0[i_ref];
1329     newframe->b_duplicate = 1;
1330     memcpy( h->fenc->weight[j], w, sizeof(h->fenc->weight[i]) );
1331
1332     /* shift the frames to make space for the dupe. */
1333     h->b_ref_reorder[0] = 1;
1334     if( h->i_ref0 < 16 )
1335         ++h->i_ref0;
1336     h->fref0[15] = NULL;
1337     x264_frame_unshift( &h->fref0[j], newframe );
1338
1339     return j;
1340 }
1341
1342 static void x264_weighted_pred_init( x264_t *h )
1343 {
1344     int i_ref;
1345     int i;
1346
1347     /* for now no analysis and set all weights to nothing */
1348     for( i_ref = 0; i_ref < h->i_ref0; i_ref++ )
1349         h->fenc->weighted[i_ref] = h->fref0[i_ref]->filtered[0];
1350
1351     // FIXME: This only supports weighting of one reference frame
1352     // and duplicates of that frame.
1353     h->fenc->i_lines_weighted = 0;
1354
1355     for( i_ref = 0; i_ref < (h->i_ref0 << h->sh.b_mbaff); i_ref++ )
1356         for( i = 0; i < 3; i++ )
1357             h->sh.weight[i_ref][i].weightfn = NULL;
1358
1359
1360     if( h->sh.i_type != SLICE_TYPE_P || h->param.analyse.i_weighted_pred <= 0 )
1361         return;
1362
1363     int i_padv = PADV << h->param.b_interlaced;
1364     int denom = -1;
1365     int weightluma = 0;
1366     int buffer_next = 0;
1367     int j;
1368     //FIXME: when chroma support is added, move this into loop
1369     h->sh.weight[0][1].weightfn = h->sh.weight[0][2].weightfn = NULL;
1370     h->sh.weight[0][1].i_denom = h->sh.weight[0][2].i_denom = 0;
1371     for( j = 0; j < h->i_ref0; j++ )
1372     {
1373         if( h->fenc->weight[j][0].weightfn )
1374         {
1375             h->sh.weight[j][0] = h->fenc->weight[j][0];
1376             // if weight is useless, don't write it to stream
1377             if( h->sh.weight[j][0].i_scale == 1<<h->sh.weight[j][0].i_denom && h->sh.weight[j][0].i_offset == 0 )
1378                 h->sh.weight[j][0].weightfn = NULL;
1379             else
1380             {
1381                 if( !weightluma )
1382                 {
1383                     weightluma = 1;
1384                     h->sh.weight[0][0].i_denom = denom = h->sh.weight[j][0].i_denom;
1385                     assert( x264_clip3( denom, 0, 7 ) == denom );
1386                 }
1387                 assert( h->sh.weight[j][0].i_denom == denom );
1388                 assert( x264_clip3( h->sh.weight[j][0].i_scale, 0, 127 ) == h->sh.weight[j][0].i_scale );
1389                 assert( x264_clip3( h->sh.weight[j][0].i_offset, -128, 127 ) == h->sh.weight[j][0].i_offset );
1390                 h->fenc->weighted[j] = h->mb.p_weight_buf[buffer_next++] +
1391                     h->fenc->i_stride[0] * i_padv + PADH;
1392             }
1393         }
1394
1395         //scale full resolution frame
1396         if( h->sh.weight[j][0].weightfn && h->param.i_threads == 1 )
1397         {
1398             uint8_t *src = h->fref0[j]->filtered[0] - h->fref0[j]->i_stride[0]*i_padv - PADH;
1399             uint8_t *dst = h->fenc->weighted[j] - h->fenc->i_stride[0]*i_padv - PADH;
1400             int stride = h->fenc->i_stride[0];
1401             int width = h->fenc->i_width[0] + PADH*2;
1402             int height = h->fenc->i_lines[0] + i_padv*2;
1403             x264_weight_scale_plane( h, dst, stride, src, stride, width, height, &h->sh.weight[j][0] );
1404             h->fenc->i_lines_weighted = height;
1405         }
1406     }
1407     if( !weightluma )
1408         h->sh.weight[0][0].i_denom = 0;
1409 }
1410
1411 static inline void x264_reference_build_list( x264_t *h, int i_poc )
1412 {
1413     int i, b_ok;
1414
1415     /* build ref list 0/1 */
1416     h->mb.pic.i_fref[0] = h->i_ref0 = 0;
1417     h->mb.pic.i_fref[1] = h->i_ref1 = 0;
1418     if( h->sh.i_type == SLICE_TYPE_I )
1419         return;
1420
1421     for( i = 0; h->frames.reference[i]; i++ )
1422     {
1423         if( h->frames.reference[i]->i_poc < i_poc )
1424         {
1425             h->fref0[h->i_ref0++] = h->frames.reference[i];
1426         }
1427         else if( h->frames.reference[i]->i_poc > i_poc )
1428         {
1429             h->fref1[h->i_ref1++] = h->frames.reference[i];
1430         }
1431     }
1432
1433     /* Order ref0 from higher to lower poc */
1434     do
1435     {
1436         b_ok = 1;
1437         for( i = 0; i < h->i_ref0 - 1; i++ )
1438         {
1439             if( h->fref0[i]->i_poc < h->fref0[i+1]->i_poc )
1440             {
1441                 XCHG( x264_frame_t*, h->fref0[i], h->fref0[i+1] );
1442                 b_ok = 0;
1443                 break;
1444             }
1445         }
1446     } while( !b_ok );
1447
1448     if( h->sh.i_mmco_remove_from_end )
1449         for( i = h->i_ref0-1; i >= h->i_ref0 - h->sh.i_mmco_remove_from_end; i-- )
1450         {
1451             int diff = h->i_frame_num - h->fref0[i]->i_frame_num;
1452             h->sh.mmco[h->sh.i_mmco_command_count].i_poc = h->fref0[i]->i_poc;
1453             h->sh.mmco[h->sh.i_mmco_command_count++].i_difference_of_pic_nums = diff;
1454         }
1455
1456     /* Order ref1 from lower to higher poc (bubble sort) for B-frame */
1457     do
1458     {
1459         b_ok = 1;
1460         for( i = 0; i < h->i_ref1 - 1; i++ )
1461         {
1462             if( h->fref1[i]->i_poc > h->fref1[i+1]->i_poc )
1463             {
1464                 XCHG( x264_frame_t*, h->fref1[i], h->fref1[i+1] );
1465                 b_ok = 0;
1466                 break;
1467             }
1468         }
1469     } while( !b_ok );
1470
1471     x264_reference_check_reorder( h );
1472
1473     h->i_ref1 = X264_MIN( h->i_ref1, h->frames.i_max_ref1 );
1474     h->i_ref0 = X264_MIN( h->i_ref0, h->frames.i_max_ref0 );
1475     h->i_ref0 = X264_MIN( h->i_ref0, h->param.i_frame_reference ); // if reconfig() has lowered the limit
1476
1477     /* add duplicates */
1478     if( h->fenc->i_type == X264_TYPE_P )
1479     {
1480         int idx = -1;
1481         if( h->param.analyse.i_weighted_pred == X264_WEIGHTP_SMART )
1482         {
1483             x264_weight_t w[3];
1484             w[1].weightfn = w[2].weightfn = NULL;
1485             if( h->param.rc.b_stat_read )
1486                 x264_ratecontrol_set_weights( h, h->fenc );
1487
1488             if( !h->fenc->weight[0][0].weightfn )
1489             {
1490                 h->fenc->weight[0][0].i_denom = 0;
1491                 SET_WEIGHT( w[0], 1, 1, 0, -1 );
1492                 idx = x264_weighted_reference_duplicate( h, 0, w );
1493             }
1494             else
1495             {
1496                 if( h->fenc->weight[0][0].i_scale == 1<<h->fenc->weight[0][0].i_denom )
1497                 {
1498                     SET_WEIGHT( h->fenc->weight[0][0], 1, 1, 0, h->fenc->weight[0][0].i_offset );
1499                 }
1500                 x264_weighted_reference_duplicate( h, 0, weight_none );
1501                 if( h->fenc->weight[0][0].i_offset > -128 )
1502                 {
1503                     w[0] = h->fenc->weight[0][0];
1504                     w[0].i_offset--;
1505                     h->mc.weight_cache( h, &w[0] );
1506                     idx = x264_weighted_reference_duplicate( h, 0, w );
1507                 }
1508             }
1509         }
1510         else if( h->param.analyse.i_weighted_pred == X264_WEIGHTP_BLIND )
1511         {
1512             //weighted offset=-1
1513             x264_weight_t w[3];
1514             SET_WEIGHT( w[0], 1, 1, 0, -1 );
1515             h->fenc->weight[0][0].i_denom = 0;
1516             w[1].weightfn = w[2].weightfn = NULL;
1517             idx = x264_weighted_reference_duplicate( h, 0, w );
1518         }
1519         h->mb.ref_blind_dupe = idx;
1520     }
1521
1522     assert( h->i_ref0 + h->i_ref1 <= 16 );
1523     h->mb.pic.i_fref[0] = h->i_ref0;
1524     h->mb.pic.i_fref[1] = h->i_ref1;
1525 }
1526
1527 static void x264_fdec_filter_row( x264_t *h, int mb_y )
1528 {
1529     /* mb_y is the mb to be encoded next, not the mb to be filtered here */
1530     int b_hpel = h->fdec->b_kept_as_ref;
1531     int b_deblock = !h->sh.i_disable_deblocking_filter_idc;
1532     int b_end = mb_y == h->sps->i_mb_height;
1533     int min_y = mb_y - (1 << h->sh.b_mbaff);
1534     int max_y = b_end ? h->sps->i_mb_height : mb_y;
1535     b_deblock &= b_hpel || h->param.psz_dump_yuv;
1536     if( mb_y & h->sh.b_mbaff )
1537         return;
1538     if( min_y < 0 )
1539         return;
1540
1541     if( !b_end && !h->param.b_sliced_threads )
1542     {
1543         int i, j;
1544         for( j=0; j<=h->sh.b_mbaff; j++ )
1545             for( i=0; i<3; i++ )
1546             {
1547                 memcpy( h->mb.intra_border_backup[j][i],
1548                         h->fdec->plane[i] + ((mb_y*16 >> !!i) + j - 1 - h->sh.b_mbaff) * h->fdec->i_stride[i],
1549                         h->sps->i_mb_width*16 >> !!i );
1550             }
1551     }
1552
1553     if( b_deblock )
1554     {
1555         int y;
1556         for( y = min_y; y < max_y; y += (1 << h->sh.b_mbaff) )
1557             x264_frame_deblock_row( h, y );
1558     }
1559
1560     if( b_hpel )
1561     {
1562         x264_frame_expand_border( h, h->fdec, min_y, b_end );
1563         if( h->param.analyse.i_subpel_refine )
1564         {
1565             x264_frame_filter( h, h->fdec, min_y, b_end );
1566             x264_frame_expand_border_filtered( h, h->fdec, min_y, b_end );
1567         }
1568     }
1569
1570     if( h->i_thread_frames > 1 && h->fdec->b_kept_as_ref )
1571         x264_frame_cond_broadcast( h->fdec, mb_y*16 + (b_end ? 10000 : -(X264_THREAD_HEIGHT << h->sh.b_mbaff)) );
1572
1573     min_y = X264_MAX( min_y*16-8, 0 );
1574     max_y = b_end ? h->param.i_height : mb_y*16-8;
1575
1576     if( h->param.analyse.b_psnr )
1577     {
1578         int i;
1579         for( i=0; i<3; i++ )
1580             h->stat.frame.i_ssd[i] +=
1581                 x264_pixel_ssd_wxh( &h->pixf,
1582                     h->fdec->plane[i] + (min_y>>!!i) * h->fdec->i_stride[i], h->fdec->i_stride[i],
1583                     h->fenc->plane[i] + (min_y>>!!i) * h->fenc->i_stride[i], h->fenc->i_stride[i],
1584                     h->param.i_width >> !!i, (max_y-min_y) >> !!i );
1585     }
1586
1587     if( h->param.analyse.b_ssim )
1588     {
1589         x264_emms();
1590         /* offset by 2 pixels to avoid alignment of ssim blocks with dct blocks,
1591          * and overlap by 4 */
1592         min_y += min_y == 0 ? 2 : -6;
1593         h->stat.frame.f_ssim +=
1594             x264_pixel_ssim_wxh( &h->pixf,
1595                 h->fdec->plane[0] + 2+min_y*h->fdec->i_stride[0], h->fdec->i_stride[0],
1596                 h->fenc->plane[0] + 2+min_y*h->fenc->i_stride[0], h->fenc->i_stride[0],
1597                 h->param.i_width-2, max_y-min_y, h->scratch_buffer );
1598     }
1599 }
1600
1601 static inline int x264_reference_update( x264_t *h )
1602 {
1603     int i, j;
1604     if( !h->fdec->b_kept_as_ref )
1605     {
1606         if( h->i_thread_frames > 1 )
1607         {
1608             x264_frame_push_unused( h, h->fdec );
1609             h->fdec = x264_frame_pop_unused( h, 1 );
1610             if( !h->fdec )
1611                 return -1;
1612         }
1613         return 0;
1614     }
1615
1616     /* apply mmco from previous frame. */
1617     for( i = 0; i < h->sh.i_mmco_command_count; i++ )
1618         for( j = 0; h->frames.reference[j]; j++ )
1619             if( h->frames.reference[j]->i_poc == h->sh.mmco[i].i_poc )
1620                 x264_frame_push_unused( h, x264_frame_shift( &h->frames.reference[j] ) );
1621
1622     /* move frame in the buffer */
1623     x264_frame_push( h->frames.reference, h->fdec );
1624     if( h->frames.reference[h->sps->i_num_ref_frames] )
1625         x264_frame_push_unused( h, x264_frame_shift( h->frames.reference ) );
1626     h->fdec = x264_frame_pop_unused( h, 1 );
1627     if( !h->fdec )
1628         return -1;
1629     return 0;
1630 }
1631
1632 static inline void x264_reference_reset( x264_t *h )
1633 {
1634     while( h->frames.reference[0] )
1635         x264_frame_push_unused( h, x264_frame_pop( h->frames.reference ) );
1636     h->fdec->i_poc =
1637     h->fenc->i_poc = 0;
1638 }
1639
1640 static inline void x264_reference_hierarchy_reset( x264_t *h )
1641 {
1642     int i, ref;
1643     int b_hasdelayframe = 0;
1644     if( !h->param.i_bframe_pyramid )
1645         return;
1646
1647     /* look for delay frames -- chain must only contain frames that are disposable */
1648     for( i = 0; h->frames.current[i] && IS_DISPOSABLE( h->frames.current[i]->i_type ); i++ )
1649         b_hasdelayframe |= h->frames.current[i]->i_coded
1650                         != h->frames.current[i]->i_frame + h->sps->vui.i_num_reorder_frames;
1651
1652     if( h->param.i_bframe_pyramid != X264_B_PYRAMID_STRICT && !b_hasdelayframe )
1653         return;
1654
1655     /* Remove last BREF. There will never be old BREFs in the
1656      * dpb during a BREF decode when pyramid == STRICT */
1657     for( ref = 0; h->frames.reference[ref]; ref++ )
1658     {
1659         if( h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT
1660             && h->frames.reference[ref]->i_type == X264_TYPE_BREF )
1661         {
1662             int diff = h->i_frame_num - h->frames.reference[ref]->i_frame_num;
1663             h->sh.mmco[h->sh.i_mmco_command_count].i_difference_of_pic_nums = diff;
1664             h->sh.mmco[h->sh.i_mmco_command_count++].i_poc = h->frames.reference[ref]->i_poc;
1665             x264_frame_push_unused( h, x264_frame_pop( h->frames.reference ) );
1666             h->b_ref_reorder[0] = 1;
1667             break;
1668         }
1669     }
1670
1671     /* Prepare to room in the dpb for the delayed display time of the later b-frame's */
1672     h->sh.i_mmco_remove_from_end = X264_MAX( ref + 2 - h->frames.i_max_dpb, 0 );
1673 }
1674
1675 static inline void x264_slice_init( x264_t *h, int i_nal_type, int i_global_qp )
1676 {
1677     /* ------------------------ Create slice header  ----------------------- */
1678     if( i_nal_type == NAL_SLICE_IDR )
1679     {
1680         x264_slice_header_init( h, &h->sh, h->sps, h->pps, h->i_idr_pic_id, h->i_frame_num, i_global_qp );
1681
1682         /* increment id */
1683         h->i_idr_pic_id = ( h->i_idr_pic_id + 1 ) % 65536;
1684     }
1685     else
1686     {
1687         x264_slice_header_init( h, &h->sh, h->sps, h->pps, -1, h->i_frame_num, i_global_qp );
1688
1689         /* always set the real higher num of ref frame used */
1690         h->sh.b_num_ref_idx_override = 1;
1691         h->sh.i_num_ref_idx_l0_active = h->i_ref0 <= 0 ? 1 : h->i_ref0;
1692         h->sh.i_num_ref_idx_l1_active = h->i_ref1 <= 0 ? 1 : h->i_ref1;
1693     }
1694
1695     h->fdec->i_frame_num = h->sh.i_frame_num;
1696
1697     if( h->sps->i_poc_type == 0 )
1698     {
1699         h->sh.i_poc_lsb = h->fdec->i_poc & ( (1 << h->sps->i_log2_max_poc_lsb) - 1 );
1700         h->sh.i_delta_poc_bottom = 0;
1701     }
1702     else if( h->sps->i_poc_type == 1 )
1703     {
1704         /* FIXME TODO FIXME */
1705     }
1706     else
1707     {
1708         /* Nothing to do ? */
1709     }
1710
1711     x264_macroblock_slice_init( h );
1712 }
1713
1714 static int x264_slice_write( x264_t *h )
1715 {
1716     int i_skip;
1717     int mb_xy, i_mb_x, i_mb_y;
1718     int i, i_list, i_ref, i_skip_bak = 0; /* Shut up GCC. */
1719     bs_t bs_bak;
1720     x264_cabac_t cabac_bak;
1721     uint8_t cabac_prevbyte_bak = 0; /* Shut up GCC. */
1722     /* Assume no more than 3 bytes of NALU escaping.
1723      * Slices other than the first use a 3-byte startcode. */
1724     int overhead_guess = (NALU_OVERHEAD - (h->param.b_annexb && h->sh.i_first_mb)) + 3;
1725     int slice_max_size = h->param.i_slice_max_size > 0 ? (h->param.i_slice_max_size-overhead_guess)*8 : INT_MAX;
1726     int starting_bits = bs_pos(&h->out.bs);
1727     bs_realign( &h->out.bs );
1728
1729     /* Slice */
1730     x264_nal_start( h, h->i_nal_type, h->i_nal_ref_idc );
1731
1732     /* Slice header */
1733     x264_macroblock_thread_init( h );
1734     x264_slice_header_write( &h->out.bs, &h->sh, h->i_nal_ref_idc );
1735     if( h->param.b_cabac )
1736     {
1737         /* alignment needed */
1738         bs_align_1( &h->out.bs );
1739
1740         /* init cabac */
1741         x264_cabac_context_init( &h->cabac, h->sh.i_type, h->sh.i_qp, h->sh.i_cabac_init_idc );
1742         x264_cabac_encode_init ( &h->cabac, h->out.bs.p, h->out.bs.p_end );
1743     }
1744     h->mb.i_last_qp = h->sh.i_qp;
1745     h->mb.i_last_dqp = 0;
1746
1747     i_mb_y = h->sh.i_first_mb / h->sps->i_mb_width;
1748     i_mb_x = h->sh.i_first_mb % h->sps->i_mb_width;
1749     i_skip = 0;
1750
1751     while( (mb_xy = i_mb_x + i_mb_y * h->sps->i_mb_width) <= h->sh.i_last_mb )
1752     {
1753         int mb_spos = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
1754         if( h->param.i_slice_max_size > 0 )
1755         {
1756             /* We don't need the contexts because flushing the CABAC encoder has no context
1757              * dependency and macroblocks are only re-encoded in the case where a slice is
1758              * ended (and thus the content of all contexts are thrown away). */
1759             if( h->param.b_cabac )
1760             {
1761                 memcpy( &cabac_bak, &h->cabac, offsetof(x264_cabac_t, f8_bits_encoded) );
1762                 /* x264's CABAC writer modifies the previous byte during carry, so it has to be
1763                  * backed up. */
1764                 cabac_prevbyte_bak = h->cabac.p[-1];
1765             }
1766             else
1767             {
1768                 bs_bak = h->out.bs;
1769                 i_skip_bak = i_skip;
1770             }
1771         }
1772
1773         if( i_mb_x == 0 && !h->mb.b_reencode_mb && !h->param.b_sliced_threads )
1774             x264_fdec_filter_row( h, i_mb_y );
1775
1776         /* load cache */
1777         x264_macroblock_cache_load( h, i_mb_x, i_mb_y );
1778
1779         x264_macroblock_analyse( h );
1780
1781         /* encode this macroblock -> be careful it can change the mb type to P_SKIP if needed */
1782         x264_macroblock_encode( h );
1783
1784         if( x264_bitstream_check_buffer( h ) )
1785             return -1;
1786
1787         if( h->param.b_cabac )
1788         {
1789             if( mb_xy > h->sh.i_first_mb && !(h->sh.b_mbaff && (i_mb_y&1)) )
1790                 x264_cabac_encode_terminal( &h->cabac );
1791
1792             if( IS_SKIP( h->mb.i_type ) )
1793                 x264_cabac_mb_skip( h, 1 );
1794             else
1795             {
1796                 if( h->sh.i_type != SLICE_TYPE_I )
1797                     x264_cabac_mb_skip( h, 0 );
1798                 x264_macroblock_write_cabac( h, &h->cabac );
1799             }
1800         }
1801         else
1802         {
1803             if( IS_SKIP( h->mb.i_type ) )
1804                 i_skip++;
1805             else
1806             {
1807                 if( h->sh.i_type != SLICE_TYPE_I )
1808                 {
1809                     bs_write_ue( &h->out.bs, i_skip );  /* skip run */
1810                     i_skip = 0;
1811                 }
1812                 x264_macroblock_write_cavlc( h );
1813             }
1814         }
1815
1816         int total_bits = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
1817         int mb_size = total_bits - mb_spos;
1818
1819         /* We'll just re-encode this last macroblock if we go over the max slice size. */
1820         if( total_bits - starting_bits > slice_max_size && !h->mb.b_reencode_mb )
1821         {
1822             if( mb_xy != h->sh.i_first_mb )
1823             {
1824                 if( h->param.b_cabac )
1825                 {
1826                     memcpy( &h->cabac, &cabac_bak, offsetof(x264_cabac_t, f8_bits_encoded) );
1827                     h->cabac.p[-1] = cabac_prevbyte_bak;
1828                 }
1829                 else
1830                 {
1831                     h->out.bs = bs_bak;
1832                     i_skip = i_skip_bak;
1833                 }
1834                 h->mb.b_reencode_mb = 1;
1835                 h->sh.i_last_mb = mb_xy-1;
1836                 break;
1837             }
1838             else
1839             {
1840                 h->sh.i_last_mb = mb_xy;
1841                 h->mb.b_reencode_mb = 0;
1842             }
1843         }
1844         else
1845             h->mb.b_reencode_mb = 0;
1846
1847 #ifdef HAVE_VISUALIZE
1848         if( h->param.b_visualize )
1849             x264_visualize_mb( h );
1850 #endif
1851
1852         /* save cache */
1853         x264_macroblock_cache_save( h );
1854
1855         /* accumulate mb stats */
1856         h->stat.frame.i_mb_count[h->mb.i_type]++;
1857
1858         if( !IS_INTRA(h->mb.i_type) && !IS_SKIP(h->mb.i_type) && !IS_DIRECT(h->mb.i_type) )
1859         {
1860             if( h->mb.i_partition != D_8x8 )
1861                     h->stat.frame.i_mb_partition[h->mb.i_partition] += 4;
1862                 else
1863                     for( i = 0; i < 4; i++ )
1864                         h->stat.frame.i_mb_partition[h->mb.i_sub_partition[i]] ++;
1865             if( h->param.i_frame_reference > 1 )
1866                 for( i_list = 0; i_list <= (h->sh.i_type == SLICE_TYPE_B); i_list++ )
1867                     for( i = 0; i < 4; i++ )
1868                     {
1869                         i_ref = h->mb.cache.ref[i_list][ x264_scan8[4*i] ];
1870                         if( i_ref >= 0 )
1871                             h->stat.frame.i_mb_count_ref[i_list][i_ref] ++;
1872                     }
1873         }
1874
1875         if( h->param.i_log_level >= X264_LOG_INFO )
1876         {
1877             if( h->mb.i_cbp_luma || h->mb.i_cbp_chroma )
1878             {
1879                 int cbpsum = (h->mb.i_cbp_luma&1) + ((h->mb.i_cbp_luma>>1)&1)
1880                            + ((h->mb.i_cbp_luma>>2)&1) + (h->mb.i_cbp_luma>>3);
1881                 int b_intra = IS_INTRA(h->mb.i_type);
1882                 h->stat.frame.i_mb_cbp[!b_intra + 0] += cbpsum;
1883                 h->stat.frame.i_mb_cbp[!b_intra + 2] += h->mb.i_cbp_chroma >= 1;
1884                 h->stat.frame.i_mb_cbp[!b_intra + 4] += h->mb.i_cbp_chroma == 2;
1885             }
1886             if( h->mb.i_cbp_luma && !IS_INTRA(h->mb.i_type) )
1887             {
1888                 h->stat.frame.i_mb_count_8x8dct[0] ++;
1889                 h->stat.frame.i_mb_count_8x8dct[1] += h->mb.b_transform_8x8;
1890             }
1891             if( IS_INTRA(h->mb.i_type) && h->mb.i_type != I_PCM )
1892             {
1893                 if( h->mb.i_type == I_16x16 )
1894                     h->stat.frame.i_mb_pred_mode[0][h->mb.i_intra16x16_pred_mode]++;
1895                 else if( h->mb.i_type == I_8x8 )
1896                     for( i = 0; i < 16; i += 4 )
1897                         h->stat.frame.i_mb_pred_mode[1][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
1898                 else //if( h->mb.i_type == I_4x4 )
1899                     for( i = 0; i < 16; i++ )
1900                         h->stat.frame.i_mb_pred_mode[2][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
1901             }
1902         }
1903
1904         x264_ratecontrol_mb( h, mb_size );
1905
1906         if( h->sh.b_mbaff )
1907         {
1908             i_mb_x += i_mb_y & 1;
1909             i_mb_y ^= i_mb_x < h->sps->i_mb_width;
1910         }
1911         else
1912             i_mb_x++;
1913         if( i_mb_x == h->sps->i_mb_width )
1914         {
1915             i_mb_y++;
1916             i_mb_x = 0;
1917         }
1918     }
1919
1920     if( h->param.b_cabac )
1921     {
1922         x264_cabac_encode_flush( h, &h->cabac );
1923         h->out.bs.p = h->cabac.p;
1924     }
1925     else
1926     {
1927         if( i_skip > 0 )
1928             bs_write_ue( &h->out.bs, i_skip );  /* last skip run */
1929         /* rbsp_slice_trailing_bits */
1930         bs_rbsp_trailing( &h->out.bs );
1931         bs_flush( &h->out.bs );
1932     }
1933     if( x264_nal_end( h ) )
1934         return -1;
1935
1936     if( h->sh.i_last_mb == h->mb.i_mb_count-1 )
1937     {
1938         h->stat.frame.i_misc_bits = bs_pos( &h->out.bs )
1939                                   + (h->out.i_nal*NALU_OVERHEAD * 8)
1940                                   - h->stat.frame.i_tex_bits
1941                                   - h->stat.frame.i_mv_bits;
1942         if( !h->param.b_sliced_threads )
1943             x264_fdec_filter_row( h, h->sps->i_mb_height );
1944     }
1945
1946     return 0;
1947 }
1948
1949 static void x264_thread_sync_context( x264_t *dst, x264_t *src )
1950 {
1951     if( dst == src )
1952         return;
1953
1954     // reference counting
1955     x264_frame_t **f;
1956     for( f = src->frames.reference; *f; f++ )
1957         (*f)->i_reference_count++;
1958     for( f = dst->frames.reference; *f; f++ )
1959         x264_frame_push_unused( src, *f );
1960     src->fdec->i_reference_count++;
1961     x264_frame_push_unused( src, dst->fdec );
1962
1963     // copy everything except the per-thread pointers and the constants.
1964     memcpy( &dst->i_frame, &src->i_frame, offsetof(x264_t, mb.type) - offsetof(x264_t, i_frame) );
1965     dst->param = src->param;
1966     dst->stat = src->stat;
1967 }
1968
1969 static void x264_thread_sync_stat( x264_t *dst, x264_t *src )
1970 {
1971     if( dst == src )
1972         return;
1973     memcpy( &dst->stat.i_frame_count, &src->stat.i_frame_count, sizeof(dst->stat) - sizeof(dst->stat.frame) );
1974 }
1975
1976 static void *x264_slices_write( x264_t *h )
1977 {
1978     int i_slice_num = 0;
1979     int last_thread_mb = h->sh.i_last_mb;
1980     if( h->param.i_sync_lookahead )
1981         x264_lower_thread_priority( 10 );
1982
1983 #ifdef HAVE_MMX
1984     /* Misalign mask has to be set separately for each thread. */
1985     if( h->param.cpu&X264_CPU_SSE_MISALIGN )
1986         x264_cpu_mask_misalign_sse();
1987 #endif
1988
1989 #ifdef HAVE_VISUALIZE
1990     if( h->param.b_visualize )
1991         if( x264_visualize_init( h ) )
1992             return (void *)-1;
1993 #endif
1994
1995     /* init stats */
1996     memset( &h->stat.frame, 0, sizeof(h->stat.frame) );
1997     h->mb.b_reencode_mb = 0;
1998     while( h->sh.i_first_mb <= last_thread_mb )
1999     {
2000         h->sh.i_last_mb = last_thread_mb;
2001         if( h->param.i_slice_max_mbs )
2002             h->sh.i_last_mb = h->sh.i_first_mb + h->param.i_slice_max_mbs - 1;
2003         else if( h->param.i_slice_count && !h->param.b_sliced_threads )
2004         {
2005             int height = h->sps->i_mb_height >> h->param.b_interlaced;
2006             int width = h->sps->i_mb_width << h->param.b_interlaced;
2007             i_slice_num++;
2008             h->sh.i_last_mb = (height * i_slice_num + h->param.i_slice_count/2) / h->param.i_slice_count * width - 1;
2009         }
2010         h->sh.i_last_mb = X264_MIN( h->sh.i_last_mb, last_thread_mb );
2011         if( x264_stack_align( x264_slice_write, h ) )
2012             return (void *)-1;
2013         h->sh.i_first_mb = h->sh.i_last_mb + 1;
2014     }
2015
2016 #ifdef HAVE_VISUALIZE
2017     if( h->param.b_visualize )
2018     {
2019         x264_visualize_show( h );
2020         x264_visualize_close( h );
2021     }
2022 #endif
2023
2024     return (void *)0;
2025 }
2026
2027 static int x264_threaded_slices_write( x264_t *h )
2028 {
2029     int i, j;
2030     void *ret = NULL;
2031     /* set first/last mb and sync contexts */
2032     for( i = 0; i < h->param.i_threads; i++ )
2033     {
2034         x264_t *t = h->thread[i];
2035         if( i )
2036         {
2037             t->param = h->param;
2038             memcpy( &t->i_frame, &h->i_frame, offsetof(x264_t, rc) - offsetof(x264_t, i_frame) );
2039         }
2040         int height = h->sps->i_mb_height >> h->param.b_interlaced;
2041         t->i_threadslice_start = ((height *  i    + h->param.i_slice_count/2) / h->param.i_threads) << h->param.b_interlaced;
2042         t->i_threadslice_end   = ((height * (i+1) + h->param.i_slice_count/2) / h->param.i_threads) << h->param.b_interlaced;
2043         t->sh.i_first_mb = t->i_threadslice_start * h->sps->i_mb_width;
2044         t->sh.i_last_mb  =   t->i_threadslice_end * h->sps->i_mb_width - 1;
2045     }
2046
2047     x264_analyse_weight_frame( h, h->sps->i_mb_height*16 + 16 );
2048
2049     x264_threads_distribute_ratecontrol( h );
2050
2051     /* dispatch */
2052     for( i = 0; i < h->param.i_threads; i++ )
2053     {
2054         if( x264_pthread_create( &h->thread[i]->thread_handle, NULL, (void*)x264_slices_write, (void*)h->thread[i] ) )
2055             return -1;
2056         h->thread[i]->b_thread_active = 1;
2057     }
2058     for( i = 0; i < h->param.i_threads; i++ )
2059     {
2060         x264_pthread_join( h->thread[i]->thread_handle, &ret );
2061         h->thread[i]->b_thread_active = 0;
2062         if( (intptr_t)ret )
2063             return (intptr_t)ret;
2064     }
2065
2066     /* deblocking and hpel filtering */
2067     for( i = 0; i <= h->sps->i_mb_height; i++ )
2068         x264_fdec_filter_row( h, i );
2069
2070     x264_threads_merge_ratecontrol( h );
2071
2072     for( i = 1; i < h->param.i_threads; i++ )
2073     {
2074         x264_t *t = h->thread[i];
2075         for( j = 0; j < t->out.i_nal; j++ )
2076         {
2077             h->out.nal[h->out.i_nal] = t->out.nal[j];
2078             h->out.i_nal++;
2079             x264_nal_check_buffer( h );
2080         }
2081         /* All entries in stat.frame are ints except for ssd/ssim,
2082          * which are only calculated in the main thread. */
2083         for( j = 0; j < (offsetof(x264_t,stat.frame.i_ssd) - offsetof(x264_t,stat.frame.i_mv_bits)) / sizeof(int); j++ )
2084             ((int*)&h->stat.frame)[j] += ((int*)&t->stat.frame)[j];
2085     }
2086
2087     return 0;
2088 }
2089
2090 /****************************************************************************
2091  * x264_encoder_encode:
2092  *  XXX: i_poc   : is the poc of the current given picture
2093  *       i_frame : is the number of the frame being coded
2094  *  ex:  type frame poc
2095  *       I      0   2*0
2096  *       P      1   2*3
2097  *       B      2   2*1
2098  *       B      3   2*2
2099  *       P      4   2*6
2100  *       B      5   2*4
2101  *       B      6   2*5
2102  ****************************************************************************/
2103 int     x264_encoder_encode( x264_t *h,
2104                              x264_nal_t **pp_nal, int *pi_nal,
2105                              x264_picture_t *pic_in,
2106                              x264_picture_t *pic_out )
2107 {
2108     x264_t *thread_current, *thread_prev, *thread_oldest;
2109     int i_nal_type, i_nal_ref_idc, i_global_qp, i;
2110
2111     if( h->i_thread_frames > 1 )
2112     {
2113         thread_prev    = h->thread[ h->i_thread_phase ];
2114         h->i_thread_phase = (h->i_thread_phase + 1) % h->i_thread_frames;
2115         thread_current = h->thread[ h->i_thread_phase ];
2116         thread_oldest  = h->thread[ (h->i_thread_phase + 1) % h->i_thread_frames ];
2117         x264_thread_sync_context( thread_current, thread_prev );
2118         x264_thread_sync_ratecontrol( thread_current, thread_prev, thread_oldest );
2119         h = thread_current;
2120 //      fprintf(stderr, "current: %p  prev: %p  oldest: %p \n", thread_current, thread_prev, thread_oldest);
2121     }
2122     else
2123     {
2124         thread_current =
2125         thread_oldest  = h;
2126     }
2127
2128     // ok to call this before encoding any frames, since the initial values of fdec have b_kept_as_ref=0
2129     if( x264_reference_update( h ) )
2130         return -1;
2131     h->fdec->i_lines_completed = -1;
2132
2133     /* no data out */
2134     *pi_nal = 0;
2135     *pp_nal = NULL;
2136
2137     /* ------------------- Setup new frame from picture -------------------- */
2138     if( pic_in != NULL )
2139     {
2140         /* 1: Copy the picture to a frame and move it to a buffer */
2141         x264_frame_t *fenc = x264_frame_pop_unused( h, 0 );
2142         if( !fenc )
2143             return -1;
2144
2145         if( x264_frame_copy_picture( h, fenc, pic_in ) < 0 )
2146             return -1;
2147
2148         if( h->param.i_width != 16 * h->sps->i_mb_width ||
2149             h->param.i_height != 16 * h->sps->i_mb_height )
2150             x264_frame_expand_border_mod16( h, fenc );
2151
2152         fenc->i_frame = h->frames.i_input++;
2153
2154         if( h->frames.i_bframe_delay && fenc->i_frame == h->frames.i_bframe_delay )
2155             h->frames.i_bframe_delay_time = fenc->i_pts;
2156
2157         if( h->frames.b_have_lowres )
2158         {
2159             if( h->param.analyse.i_weighted_pred == X264_WEIGHTP_FAKE || h->param.analyse.i_weighted_pred == X264_WEIGHTP_SMART )
2160                 x264_weight_plane_analyse( h, fenc );
2161             x264_frame_init_lowres( h, fenc );
2162         }
2163
2164         if( h->param.rc.b_mb_tree && h->param.rc.b_stat_read )
2165         {
2166             if( x264_macroblock_tree_read( h, fenc ) )
2167                 return -1;
2168         }
2169         else if( h->param.rc.i_aq_mode )
2170             x264_adaptive_quant_frame( h, fenc );
2171
2172         /* 2: Place the frame into the queue for its slice type decision */
2173         x264_lookahead_put_frame( h, fenc );
2174
2175         if( h->frames.i_input <= h->frames.i_delay + 1 - h->i_thread_frames )
2176         {
2177             /* Nothing yet to encode, waiting for filling of buffers */
2178             pic_out->i_type = X264_TYPE_AUTO;
2179             return 0;
2180         }
2181     }
2182     else
2183     {
2184         /* signal kills for lookahead thread */
2185         x264_pthread_mutex_lock( &h->lookahead->ifbuf.mutex );
2186         h->lookahead->b_exit_thread = 1;
2187         x264_pthread_cond_broadcast( &h->lookahead->ifbuf.cv_fill );
2188         x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
2189     }
2190
2191     h->i_frame++;
2192     /* 3: The picture is analyzed in the lookahead */
2193     if( !h->frames.current[0] )
2194         x264_lookahead_get_frames( h );
2195
2196     if( !h->frames.current[0] && x264_lookahead_is_empty( h ) )
2197         return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
2198
2199     /* ------------------- Get frame to be encoded ------------------------- */
2200     /* 4: get picture to encode */
2201     h->fenc = x264_frame_shift( h->frames.current );
2202     if( h->fenc->param )
2203     {
2204         x264_encoder_reconfig( h, h->fenc->param );
2205         if( h->fenc->param->param_free )
2206             h->fenc->param->param_free( h->fenc->param );
2207     }
2208
2209     if( h->fenc->b_keyframe )
2210     {
2211         h->frames.i_last_keyframe = h->fenc->i_frame;
2212         if( h->fenc->i_type == X264_TYPE_IDR )
2213             h->i_frame_num = 0;
2214     }
2215     h->sh.i_mmco_command_count =
2216     h->sh.i_mmco_remove_from_end = 0;
2217     h->b_ref_reorder[0] =
2218     h->b_ref_reorder[1] = 0;
2219
2220     /* ------------------- Setup frame context ----------------------------- */
2221     /* 5: Init data dependent of frame type */
2222     if( h->fenc->i_type == X264_TYPE_IDR )
2223     {
2224         /* reset ref pictures */
2225         i_nal_type    = NAL_SLICE_IDR;
2226         i_nal_ref_idc = NAL_PRIORITY_HIGHEST;
2227         h->sh.i_type = SLICE_TYPE_I;
2228         x264_reference_reset( h );
2229     }
2230     else if( h->fenc->i_type == X264_TYPE_I )
2231     {
2232         i_nal_type    = NAL_SLICE;
2233         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
2234         h->sh.i_type = SLICE_TYPE_I;
2235         x264_reference_hierarchy_reset( h );
2236     }
2237     else if( h->fenc->i_type == X264_TYPE_P )
2238     {
2239         i_nal_type    = NAL_SLICE;
2240         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
2241         h->sh.i_type = SLICE_TYPE_P;
2242         x264_reference_hierarchy_reset( h );
2243     }
2244     else if( h->fenc->i_type == X264_TYPE_BREF )
2245     {
2246         i_nal_type    = NAL_SLICE;
2247         i_nal_ref_idc = h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT ? NAL_PRIORITY_LOW : NAL_PRIORITY_HIGH;
2248         h->sh.i_type = SLICE_TYPE_B;
2249         x264_reference_hierarchy_reset( h );
2250     }
2251     else    /* B frame */
2252     {
2253         i_nal_type    = NAL_SLICE;
2254         i_nal_ref_idc = NAL_PRIORITY_DISPOSABLE;
2255         h->sh.i_type = SLICE_TYPE_B;
2256     }
2257
2258     h->fdec->i_poc =
2259     h->fenc->i_poc = 2 * (h->fenc->i_frame - h->frames.i_last_keyframe);
2260     h->fdec->i_type = h->fenc->i_type;
2261     h->fdec->i_frame = h->fenc->i_frame;
2262     h->fenc->b_kept_as_ref =
2263     h->fdec->b_kept_as_ref = i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE && h->param.i_keyint_max > 1;
2264
2265
2266
2267     /* ------------------- Init                ----------------------------- */
2268     /* build ref list 0/1 */
2269     x264_reference_build_list( h, h->fdec->i_poc );
2270
2271     /* ---------------------- Write the bitstream -------------------------- */
2272     /* Init bitstream context */
2273     if( h->param.b_sliced_threads )
2274     {
2275         for( i = 0; i < h->param.i_threads; i++ )
2276         {
2277             bs_init( &h->thread[i]->out.bs, h->thread[i]->out.p_bitstream, h->thread[i]->out.i_bitstream );
2278             h->thread[i]->out.i_nal = 0;
2279         }
2280     }
2281     else
2282     {
2283         bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
2284         h->out.i_nal = 0;
2285     }
2286
2287     if( h->param.b_aud )
2288     {
2289         int pic_type;
2290
2291         if( h->sh.i_type == SLICE_TYPE_I )
2292             pic_type = 0;
2293         else if( h->sh.i_type == SLICE_TYPE_P )
2294             pic_type = 1;
2295         else if( h->sh.i_type == SLICE_TYPE_B )
2296             pic_type = 2;
2297         else
2298             pic_type = 7;
2299
2300         x264_nal_start( h, NAL_AUD, NAL_PRIORITY_DISPOSABLE );
2301         bs_write( &h->out.bs, 3, pic_type );
2302         bs_rbsp_trailing( &h->out.bs );
2303         if( x264_nal_end( h ) )
2304             return -1;
2305     }
2306
2307     h->i_nal_type = i_nal_type;
2308     h->i_nal_ref_idc = i_nal_ref_idc;
2309
2310     int overhead = NALU_OVERHEAD;
2311
2312     if( h->param.b_intra_refresh && h->fenc->i_type == X264_TYPE_P )
2313     {
2314         int pocdiff = (h->fdec->i_poc - h->fref0[0]->i_poc)/2;
2315         float increment = X264_MAX( ((float)h->sps->i_mb_width-1) / h->param.i_keyint_max, 1 );
2316         int max_position = (int)(increment * h->param.i_keyint_max);
2317         if( IS_X264_TYPE_I( h->fref0[0]->i_type ) )
2318             h->fdec->f_pir_position = 0;
2319         else
2320         {
2321             h->fdec->f_pir_position = h->fref0[0]->f_pir_position;
2322             if( h->fdec->f_pir_position+0.5 >= max_position )
2323             {
2324                 h->fdec->f_pir_position = 0;
2325                 h->fenc->b_keyframe = 1;
2326             }
2327         }
2328         h->fdec->i_pir_start_col = h->fdec->f_pir_position+0.5;
2329         h->fdec->f_pir_position += increment * pocdiff;
2330         h->fdec->i_pir_end_col = h->fdec->f_pir_position+0.5;
2331     }
2332
2333     /* Write SPS and PPS */
2334     if( h->fenc->b_keyframe )
2335     {
2336         if( h->param.b_repeat_headers )
2337         {
2338             if( h->fenc->i_frame == 0 )
2339             {
2340                 /* identify ourself */
2341                 x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2342                 if( x264_sei_version_write( h, &h->out.bs ) )
2343                     return -1;
2344                 if( x264_nal_end( h ) )
2345                     return -1;
2346                 overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
2347             }
2348
2349             /* generate sequence parameters */
2350             x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
2351             x264_sps_write( &h->out.bs, h->sps );
2352             if( x264_nal_end( h ) )
2353                 return -1;
2354             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
2355
2356             /* generate picture parameters */
2357             x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
2358             x264_pps_write( &h->out.bs, h->pps );
2359             if( x264_nal_end( h ) )
2360                 return -1;
2361             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
2362         }
2363
2364         if( h->fenc->i_type != X264_TYPE_IDR )
2365         {
2366             int time_to_recovery = X264_MIN( h->sps->i_mb_width - 1, h->param.i_keyint_max ) + h->param.i_bframe;
2367             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2368             x264_sei_recovery_point_write( h, &h->out.bs, time_to_recovery );
2369             x264_nal_end( h );
2370             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
2371         }
2372     }
2373
2374     /* Init the rate control */
2375     /* FIXME: Include slice header bit cost. */
2376     x264_ratecontrol_start( h, h->fenc->i_qpplus1, overhead*8 );
2377     i_global_qp = x264_ratecontrol_qp( h );
2378
2379     pic_out->i_qpplus1 =
2380     h->fdec->i_qpplus1 = i_global_qp + 1;
2381
2382     if( h->param.rc.b_stat_read && h->sh.i_type != SLICE_TYPE_I )
2383     {
2384         x264_reference_build_list_optimal( h );
2385         x264_reference_check_reorder( h );
2386     }
2387
2388     if( h->i_ref0 )
2389         h->fdec->i_poc_l0ref0 = h->fref0[0]->i_poc;
2390
2391     if( h->sh.i_type == SLICE_TYPE_B )
2392         x264_macroblock_bipred_init( h );
2393
2394     /*------------------------- Weights -------------------------------------*/
2395     x264_weighted_pred_init( h );
2396
2397     /* ------------------------ Create slice header  ----------------------- */
2398     x264_slice_init( h, i_nal_type, i_global_qp );
2399
2400     if( i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE )
2401         h->i_frame_num++;
2402
2403     /* Write frame */
2404     h->i_threadslice_start = 0;
2405     h->i_threadslice_end = h->sps->i_mb_height;
2406     if( h->i_thread_frames > 1 )
2407     {
2408         if( x264_pthread_create( &h->thread_handle, NULL, (void*)x264_slices_write, h ) )
2409             return -1;
2410         h->b_thread_active = 1;
2411     }
2412     else if( h->param.b_sliced_threads )
2413     {
2414         if( x264_threaded_slices_write( h ) )
2415             return -1;
2416     }
2417     else
2418         if( (intptr_t)x264_slices_write( h ) )
2419             return -1;
2420
2421     return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
2422 }
2423
2424 static int x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
2425                                    x264_nal_t **pp_nal, int *pi_nal,
2426                                    x264_picture_t *pic_out )
2427 {
2428     int i, j, i_list, frame_size;
2429     char psz_message[80];
2430
2431     if( h->b_thread_active )
2432     {
2433         void *ret = NULL;
2434         x264_pthread_join( h->thread_handle, &ret );
2435         h->b_thread_active = 0;
2436         if( (intptr_t)ret )
2437             return (intptr_t)ret;
2438     }
2439     if( !h->out.i_nal )
2440     {
2441         pic_out->i_type = X264_TYPE_AUTO;
2442         return 0;
2443     }
2444
2445     x264_frame_push_unused( thread_current, h->fenc );
2446
2447     /* End bitstream, set output  */
2448     *pi_nal = h->out.i_nal;
2449     *pp_nal = h->out.nal;
2450
2451     frame_size = x264_encoder_encapsulate_nals( h );
2452
2453     h->out.i_nal = 0;
2454
2455     /* Set output picture properties */
2456     if( h->sh.i_type == SLICE_TYPE_I )
2457         pic_out->i_type = h->i_nal_type == NAL_SLICE_IDR ? X264_TYPE_IDR : X264_TYPE_I;
2458     else if( h->sh.i_type == SLICE_TYPE_P )
2459         pic_out->i_type = X264_TYPE_P;
2460     else
2461         pic_out->i_type = X264_TYPE_B;
2462
2463     pic_out->b_keyframe = h->fenc->b_keyframe;
2464
2465     pic_out->i_pts = h->fenc->i_pts *= h->i_dts_compress_multiplier;
2466     if( h->frames.i_bframe_delay )
2467     {
2468         int64_t *i_prev_dts = thread_current->frames.i_prev_dts;
2469         if( h->i_frame <= h->frames.i_bframe_delay )
2470         {
2471             if( h->i_dts_compress_multiplier == 1 )
2472                 pic_out->i_dts = h->fenc->i_reordered_pts - h->frames.i_bframe_delay_time;
2473             else
2474             {
2475                 /* DTS compression */
2476                 if( h->i_frame == 1 )
2477                     thread_current->frames.i_init_delta = h->fenc->i_reordered_pts * h->i_dts_compress_multiplier;
2478                 pic_out->i_dts = h->i_frame * thread_current->frames.i_init_delta / h->i_dts_compress_multiplier;
2479             }
2480         }
2481         else
2482             pic_out->i_dts = i_prev_dts[ (h->i_frame - h->frames.i_bframe_delay) % h->frames.i_bframe_delay ];
2483         i_prev_dts[ h->i_frame % h->frames.i_bframe_delay ] = h->fenc->i_reordered_pts * h->i_dts_compress_multiplier;
2484     }
2485     else
2486         pic_out->i_dts = h->fenc->i_reordered_pts;
2487     assert( pic_out->i_pts >= pic_out->i_dts );
2488
2489     pic_out->img.i_plane = h->fdec->i_plane;
2490     for(i = 0; i < 3; i++)
2491     {
2492         pic_out->img.i_stride[i] = h->fdec->i_stride[i];
2493         pic_out->img.plane[i] = h->fdec->plane[i];
2494     }
2495
2496     /* ---------------------- Update encoder state ------------------------- */
2497
2498     /* update rc */
2499     x264_emms();
2500     if( x264_ratecontrol_end( h, frame_size * 8 ) < 0 )
2501         return -1;
2502
2503     x264_noise_reduction_update( thread_current );
2504
2505     /* ---------------------- Compute/Print statistics --------------------- */
2506     x264_thread_sync_stat( h, h->thread[0] );
2507
2508     /* Slice stat */
2509     h->stat.i_frame_count[h->sh.i_type]++;
2510     h->stat.i_frame_size[h->sh.i_type] += frame_size;
2511     h->stat.f_frame_qp[h->sh.i_type] += h->fdec->f_qp_avg_aq;
2512
2513     for( i = 0; i < X264_MBTYPE_MAX; i++ )
2514         h->stat.i_mb_count[h->sh.i_type][i] += h->stat.frame.i_mb_count[i];
2515     for( i = 0; i < X264_PARTTYPE_MAX; i++ )
2516         h->stat.i_mb_partition[h->sh.i_type][i] += h->stat.frame.i_mb_partition[i];
2517     for( i = 0; i < 2; i++ )
2518         h->stat.i_mb_count_8x8dct[i] += h->stat.frame.i_mb_count_8x8dct[i];
2519     for( i = 0; i < 6; i++ )
2520         h->stat.i_mb_cbp[i] += h->stat.frame.i_mb_cbp[i];
2521     for( i = 0; i < 3; i++ )
2522         for( j = 0; j < 13; j++ )
2523             h->stat.i_mb_pred_mode[i][j] += h->stat.frame.i_mb_pred_mode[i][j];
2524     if( h->sh.i_type != SLICE_TYPE_I )
2525         for( i_list = 0; i_list < 2; i_list++ )
2526             for( i = 0; i < 32; i++ )
2527                 h->stat.i_mb_count_ref[h->sh.i_type][i_list][i] += h->stat.frame.i_mb_count_ref[i_list][i];
2528     if( h->sh.i_type == SLICE_TYPE_P )
2529     {
2530         h->stat.i_consecutive_bframes[h->fdec->i_frame - h->fref0[0]->i_frame - 1]++;
2531         if( h->param.analyse.i_weighted_pred == X264_WEIGHTP_SMART )
2532         {
2533             for( i = 0; i < 3; i++ )
2534                 for( j = 0; j < h->i_ref0; j++ )
2535                     if( h->sh.weight[0][i].i_denom != 0 )
2536                     {
2537                         h->stat.i_wpred[i]++;
2538                         break;
2539                     }
2540         }
2541     }
2542     if( h->sh.i_type == SLICE_TYPE_B )
2543     {
2544         h->stat.i_direct_frames[ h->sh.b_direct_spatial_mv_pred ] ++;
2545         if( h->mb.b_direct_auto_write )
2546         {
2547             //FIXME somewhat arbitrary time constants
2548             if( h->stat.i_direct_score[0] + h->stat.i_direct_score[1] > h->mb.i_mb_count )
2549             {
2550                 for( i = 0; i < 2; i++ )
2551                     h->stat.i_direct_score[i] = h->stat.i_direct_score[i] * 9/10;
2552             }
2553             for( i = 0; i < 2; i++ )
2554                 h->stat.i_direct_score[i] += h->stat.frame.i_direct_score[i];
2555         }
2556     }
2557
2558     psz_message[0] = '\0';
2559     if( h->param.analyse.b_psnr )
2560     {
2561         int64_t ssd[3] = {
2562             h->stat.frame.i_ssd[0],
2563             h->stat.frame.i_ssd[1],
2564             h->stat.frame.i_ssd[2],
2565         };
2566
2567         h->stat.i_ssd_global[h->sh.i_type] += ssd[0] + ssd[1] + ssd[2];
2568         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 );
2569         h->stat.f_psnr_mean_y[h->sh.i_type] += x264_psnr( ssd[0], h->param.i_width * h->param.i_height );
2570         h->stat.f_psnr_mean_u[h->sh.i_type] += x264_psnr( ssd[1], h->param.i_width * h->param.i_height / 4 );
2571         h->stat.f_psnr_mean_v[h->sh.i_type] += x264_psnr( ssd[2], h->param.i_width * h->param.i_height / 4 );
2572
2573         snprintf( psz_message, 80, " PSNR Y:%5.2f U:%5.2f V:%5.2f",
2574                   x264_psnr( ssd[0], h->param.i_width * h->param.i_height ),
2575                   x264_psnr( ssd[1], h->param.i_width * h->param.i_height / 4),
2576                   x264_psnr( ssd[2], h->param.i_width * h->param.i_height / 4) );
2577     }
2578
2579     if( h->param.analyse.b_ssim )
2580     {
2581         double ssim_y = h->stat.frame.f_ssim
2582                       / (((h->param.i_width-6)>>2) * ((h->param.i_height-6)>>2));
2583         h->stat.f_ssim_mean_y[h->sh.i_type] += ssim_y;
2584         snprintf( psz_message + strlen(psz_message), 80 - strlen(psz_message),
2585                   " SSIM Y:%.5f", ssim_y );
2586     }
2587     psz_message[79] = '\0';
2588
2589     x264_log( h, X264_LOG_DEBUG,
2590                   "frame=%4d QP=%.2f NAL=%d Slice:%c Poc:%-3d I:%-4d P:%-4d SKIP:%-4d size=%d bytes%s\n",
2591               h->i_frame,
2592               h->fdec->f_qp_avg_aq,
2593               h->i_nal_ref_idc,
2594               h->sh.i_type == SLICE_TYPE_I ? 'I' : (h->sh.i_type == SLICE_TYPE_P ? 'P' : 'B' ),
2595               h->fdec->i_poc,
2596               h->stat.frame.i_mb_count_i,
2597               h->stat.frame.i_mb_count_p,
2598               h->stat.frame.i_mb_count_skip,
2599               frame_size,
2600               psz_message );
2601
2602     // keep stats all in one place
2603     x264_thread_sync_stat( h->thread[0], h );
2604     // for the use of the next frame
2605     x264_thread_sync_stat( thread_current, h );
2606
2607 #ifdef DEBUG_MB_TYPE
2608 {
2609     static const char mb_chars[] = { 'i', 'i', 'I', 'C', 'P', '8', 'S',
2610         'D', '<', 'X', 'B', 'X', '>', 'B', 'B', 'B', 'B', '8', 'S' };
2611     int mb_xy;
2612     for( mb_xy = 0; mb_xy < h->sps->i_mb_width * h->sps->i_mb_height; mb_xy++ )
2613     {
2614         if( h->mb.type[mb_xy] < X264_MBTYPE_MAX && h->mb.type[mb_xy] >= 0 )
2615             fprintf( stderr, "%c ", mb_chars[ h->mb.type[mb_xy] ] );
2616         else
2617             fprintf( stderr, "? " );
2618
2619         if( (mb_xy+1) % h->sps->i_mb_width == 0 )
2620             fprintf( stderr, "\n" );
2621     }
2622 }
2623 #endif
2624
2625     /* Remove duplicates, must be done near the end as breaks h->fref0 array
2626      * by freeing some of its pointers. */
2627      for( i = 0; i < h->i_ref0; i++ )
2628          if( h->fref0[i] && h->fref0[i]->b_duplicate )
2629          {
2630              x264_frame_push_blank_unused( h, h->fref0[i] );
2631              h->fref0[i] = 0;
2632          }
2633
2634     if( h->param.psz_dump_yuv )
2635         x264_frame_dump( h );
2636
2637     return frame_size;
2638 }
2639
2640 static void x264_print_intra( int64_t *i_mb_count, double i_count, int b_print_pcm, char *intra )
2641 {
2642     intra += sprintf( intra, "I16..4%s: %4.1f%% %4.1f%% %4.1f%%",
2643         b_print_pcm ? "..PCM" : "",
2644         i_mb_count[I_16x16]/ i_count,
2645         i_mb_count[I_8x8]  / i_count,
2646         i_mb_count[I_4x4]  / i_count );
2647     if( b_print_pcm )
2648         sprintf( intra, " %4.1f%%", i_mb_count[I_PCM]  / i_count );
2649 }
2650
2651 /****************************************************************************
2652  * x264_encoder_close:
2653  ****************************************************************************/
2654 void    x264_encoder_close  ( x264_t *h )
2655 {
2656     int64_t i_yuv_size = 3 * h->param.i_width * h->param.i_height / 2;
2657     int64_t i_mb_count_size[2][7] = {{0}};
2658     char buf[200];
2659     int i, j, i_list, i_type;
2660     int b_print_pcm = h->stat.i_mb_count[SLICE_TYPE_I][I_PCM]
2661                    || h->stat.i_mb_count[SLICE_TYPE_P][I_PCM]
2662                    || h->stat.i_mb_count[SLICE_TYPE_B][I_PCM];
2663
2664     x264_lookahead_delete( h );
2665
2666     if( h->param.i_threads > 1 )
2667     {
2668         // don't strictly have to wait for the other threads, but it's simpler than canceling them
2669         for( i = 0; i < h->param.i_threads; i++ )
2670             if( h->thread[i]->b_thread_active )
2671                 x264_pthread_join( h->thread[i]->thread_handle, NULL );
2672         if( h->i_thread_frames > 1 )
2673         {
2674             for( i = 0; i < h->i_thread_frames; i++ )
2675             {
2676                 if( h->thread[i]->b_thread_active )
2677                 {
2678                     assert( h->thread[i]->fenc->i_reference_count == 1 );
2679                     x264_frame_delete( h->thread[i]->fenc );
2680                 }
2681             }
2682
2683             x264_t *thread_prev = h->thread[h->i_thread_phase];
2684             x264_thread_sync_ratecontrol( h, thread_prev, h );
2685             x264_thread_sync_ratecontrol( thread_prev, thread_prev, h );
2686             h->i_frame = thread_prev->i_frame + 1 - h->i_thread_frames;
2687         }
2688     }
2689     h->i_frame++;
2690
2691     /* Slices used and PSNR */
2692     for( i=0; i<5; i++ )
2693     {
2694         static const int slice_order[] = { SLICE_TYPE_I, SLICE_TYPE_SI, SLICE_TYPE_P, SLICE_TYPE_SP, SLICE_TYPE_B };
2695         static const char *slice_name[] = { "P", "B", "I", "SP", "SI" };
2696         int i_slice = slice_order[i];
2697
2698         if( h->stat.i_frame_count[i_slice] > 0 )
2699         {
2700             const int i_count = h->stat.i_frame_count[i_slice];
2701             if( h->param.analyse.b_psnr )
2702             {
2703                 x264_log( h, X264_LOG_INFO,
2704                           "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",
2705                           slice_name[i_slice],
2706                           i_count,
2707                           h->stat.f_frame_qp[i_slice] / i_count,
2708                           (double)h->stat.i_frame_size[i_slice] / i_count,
2709                           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,
2710                           h->stat.f_psnr_average[i_slice] / i_count,
2711                           x264_psnr( h->stat.i_ssd_global[i_slice], i_count * i_yuv_size ) );
2712             }
2713             else
2714             {
2715                 x264_log( h, X264_LOG_INFO,
2716                           "frame %s:%-5d Avg QP:%5.2f  size:%6.0f\n",
2717                           slice_name[i_slice],
2718                           i_count,
2719                           h->stat.f_frame_qp[i_slice] / i_count,
2720                           (double)h->stat.i_frame_size[i_slice] / i_count );
2721             }
2722         }
2723     }
2724     if( h->param.i_bframe && h->stat.i_frame_count[SLICE_TYPE_P] )
2725     {
2726         char *p = buf;
2727         int den = 0;
2728         // weight by number of frames (including the P-frame) that are in a sequence of N B-frames
2729         for( i=0; i<=h->param.i_bframe; i++ )
2730             den += (i+1) * h->stat.i_consecutive_bframes[i];
2731         for( i=0; i<=h->param.i_bframe; i++ )
2732             p += sprintf( p, " %4.1f%%", 100. * (i+1) * h->stat.i_consecutive_bframes[i] / den );
2733         x264_log( h, X264_LOG_INFO, "consecutive B-frames:%s\n", buf );
2734     }
2735
2736     for( i_type = 0; i_type < 2; i_type++ )
2737         for( i = 0; i < X264_PARTTYPE_MAX; i++ )
2738         {
2739             if( i == D_DIRECT_8x8 ) continue; /* direct is counted as its own type */
2740             i_mb_count_size[i_type][x264_mb_partition_pixel_table[i]] += h->stat.i_mb_partition[i_type][i];
2741         }
2742
2743     /* MB types used */
2744     if( h->stat.i_frame_count[SLICE_TYPE_I] > 0 )
2745     {
2746         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_I];
2747         double i_count = h->stat.i_frame_count[SLICE_TYPE_I] * h->mb.i_mb_count / 100.0;
2748         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
2749         x264_log( h, X264_LOG_INFO, "mb I  %s\n", buf );
2750     }
2751     if( h->stat.i_frame_count[SLICE_TYPE_P] > 0 )
2752     {
2753         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_P];
2754         double i_count = h->stat.i_frame_count[SLICE_TYPE_P] * h->mb.i_mb_count / 100.0;
2755         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_P];
2756         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
2757         x264_log( h, X264_LOG_INFO,
2758                   "mb P  %s  P16..4: %4.1f%% %4.1f%% %4.1f%% %4.1f%% %4.1f%%    skip:%4.1f%%\n",
2759                   buf,
2760                   i_mb_size[PIXEL_16x16] / (i_count*4),
2761                   (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
2762                   i_mb_size[PIXEL_8x8] / (i_count*4),
2763                   (i_mb_size[PIXEL_8x4] + i_mb_size[PIXEL_4x8]) / (i_count*4),
2764                   i_mb_size[PIXEL_4x4] / (i_count*4),
2765                   i_mb_count[P_SKIP] / i_count );
2766     }
2767     if( h->stat.i_frame_count[SLICE_TYPE_B] > 0 )
2768     {
2769         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_B];
2770         double i_count = h->stat.i_frame_count[SLICE_TYPE_B] * h->mb.i_mb_count / 100.0;
2771         double i_mb_list_count;
2772         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_B];
2773         int64_t list_count[3] = {0}; /* 0 == L0, 1 == L1, 2 == BI */
2774         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
2775         for( i = 0; i < X264_PARTTYPE_MAX; i++ )
2776             for( j = 0; j < 2; j++ )
2777             {
2778                 int l0 = x264_mb_type_list_table[i][0][j];
2779                 int l1 = x264_mb_type_list_table[i][1][j];
2780                 if( l0 || l1 )
2781                     list_count[l1+l0*l1] += h->stat.i_mb_count[SLICE_TYPE_B][i] * 2;
2782             }
2783         list_count[0] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L0_8x8];
2784         list_count[1] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L1_8x8];
2785         list_count[2] += h->stat.i_mb_partition[SLICE_TYPE_B][D_BI_8x8];
2786         i_mb_count[B_DIRECT] += (h->stat.i_mb_partition[SLICE_TYPE_B][D_DIRECT_8x8]+2)/4;
2787         i_mb_list_count = (list_count[0] + list_count[1] + list_count[2]) / 100.0;
2788         x264_log( h, X264_LOG_INFO,
2789                   "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",
2790                   buf,
2791                   i_mb_size[PIXEL_16x16] / (i_count*4),
2792                   (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
2793                   i_mb_size[PIXEL_8x8] / (i_count*4),
2794                   i_mb_count[B_DIRECT] / i_count,
2795                   i_mb_count[B_SKIP]   / i_count,
2796                   list_count[0] / i_mb_list_count,
2797                   list_count[1] / i_mb_list_count,
2798                   list_count[2] / i_mb_list_count );
2799     }
2800
2801     x264_ratecontrol_summary( h );
2802
2803     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 )
2804     {
2805 #define SUM3(p) (p[SLICE_TYPE_I] + p[SLICE_TYPE_P] + p[SLICE_TYPE_B])
2806 #define SUM3b(p,o) (p[SLICE_TYPE_I][o] + p[SLICE_TYPE_P][o] + p[SLICE_TYPE_B][o])
2807         int64_t i_i8x8 = SUM3b( h->stat.i_mb_count, I_8x8 );
2808         int64_t i_intra = i_i8x8 + SUM3b( h->stat.i_mb_count, I_4x4 )
2809                                  + SUM3b( h->stat.i_mb_count, I_16x16 );
2810         int64_t i_all_intra = i_intra + SUM3b( h->stat.i_mb_count, I_PCM);
2811         const int i_count = h->stat.i_frame_count[SLICE_TYPE_I] +
2812                             h->stat.i_frame_count[SLICE_TYPE_P] +
2813                             h->stat.i_frame_count[SLICE_TYPE_B];
2814         int64_t i_mb_count = i_count * h->mb.i_mb_count;
2815         float fps = (float) h->param.i_fps_num / h->param.i_fps_den;
2816         float f_bitrate = fps * SUM3(h->stat.i_frame_size) / i_count / 125;
2817
2818         if( h->pps->b_transform_8x8_mode )
2819         {
2820             buf[0] = 0;
2821             if( h->stat.i_mb_count_8x8dct[0] )
2822                 sprintf( buf, " inter:%.1f%%", 100. * h->stat.i_mb_count_8x8dct[1] / h->stat.i_mb_count_8x8dct[0] );
2823             x264_log( h, X264_LOG_INFO, "8x8 transform intra:%.1f%%%s\n", 100. * i_i8x8 / i_intra, buf );
2824         }
2825
2826         if( (h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO ||
2827             (h->stat.i_direct_frames[0] && h->stat.i_direct_frames[1]))
2828             && h->stat.i_frame_count[SLICE_TYPE_B] )
2829         {
2830             x264_log( h, X264_LOG_INFO, "direct mvs  spatial:%.1f%% temporal:%.1f%%\n",
2831                       h->stat.i_direct_frames[1] * 100. / h->stat.i_frame_count[SLICE_TYPE_B],
2832                       h->stat.i_direct_frames[0] * 100. / h->stat.i_frame_count[SLICE_TYPE_B] );
2833         }
2834
2835         buf[0] = 0;
2836         if( i_mb_count != i_all_intra )
2837             sprintf( buf, " inter: %.1f%% %.1f%% %.1f%%",
2838                      h->stat.i_mb_cbp[1] * 100.0 / ((i_mb_count - i_all_intra)*4),
2839                      h->stat.i_mb_cbp[3] * 100.0 / ((i_mb_count - i_all_intra)  ),
2840                      h->stat.i_mb_cbp[5] * 100.0 / ((i_mb_count - i_all_intra)) );
2841         x264_log( h, X264_LOG_INFO, "coded y,uvDC,uvAC intra: %.1f%% %.1f%% %.1f%%%s\n",
2842                   h->stat.i_mb_cbp[0] * 100.0 / (i_all_intra*4),
2843                   h->stat.i_mb_cbp[2] * 100.0 / (i_all_intra  ),
2844                   h->stat.i_mb_cbp[4] * 100.0 / (i_all_intra  ), buf );
2845
2846         int64_t fixed_pred_modes[3][9] = {{0}};
2847         int64_t sum_pred_modes[3] = {0};
2848         for( i = 0; i <= I_PRED_16x16_DC_128; i++ )
2849         {
2850             fixed_pred_modes[0][x264_mb_pred_mode16x16_fix[i]] += h->stat.i_mb_pred_mode[0][i];
2851             sum_pred_modes[0] += h->stat.i_mb_pred_mode[0][i];
2852         }
2853         if( sum_pred_modes[0] )
2854             x264_log( h, X264_LOG_INFO, "i16 v,h,dc,p: %2.0f%% %2.0f%% %2.0f%% %2.0f%%\n",
2855                       fixed_pred_modes[0][0] * 100.0 / sum_pred_modes[0],
2856                       fixed_pred_modes[0][1] * 100.0 / sum_pred_modes[0],
2857                       fixed_pred_modes[0][2] * 100.0 / sum_pred_modes[0],
2858                       fixed_pred_modes[0][3] * 100.0 / sum_pred_modes[0] );
2859         for( i = 1; i <= 2; i++ )
2860         {
2861             for( j = 0; j <= I_PRED_8x8_DC_128; j++ )
2862             {
2863                 fixed_pred_modes[i][x264_mb_pred_mode4x4_fix(j)] += h->stat.i_mb_pred_mode[i][j];
2864                 sum_pred_modes[i] += h->stat.i_mb_pred_mode[i][j];
2865             }
2866             if( sum_pred_modes[i] )
2867                 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,
2868                           fixed_pred_modes[i][0] * 100.0 / sum_pred_modes[i],
2869                           fixed_pred_modes[i][1] * 100.0 / sum_pred_modes[i],
2870                           fixed_pred_modes[i][2] * 100.0 / sum_pred_modes[i],
2871                           fixed_pred_modes[i][3] * 100.0 / sum_pred_modes[i],
2872                           fixed_pred_modes[i][4] * 100.0 / sum_pred_modes[i],
2873                           fixed_pred_modes[i][5] * 100.0 / sum_pred_modes[i],
2874                           fixed_pred_modes[i][6] * 100.0 / sum_pred_modes[i],
2875                           fixed_pred_modes[i][7] * 100.0 / sum_pred_modes[i],
2876                           fixed_pred_modes[i][8] * 100.0 / sum_pred_modes[i] );
2877         }
2878
2879         if( h->param.analyse.i_weighted_pred == X264_WEIGHTP_SMART && h->stat.i_frame_count[SLICE_TYPE_P] > 0 )
2880             x264_log( h, X264_LOG_INFO, "Weighted P-Frames: Y:%.1f%%\n",
2881                       h->stat.i_wpred[0] * 100.0 / h->stat.i_frame_count[SLICE_TYPE_P] );
2882
2883         for( i_list = 0; i_list < 2; i_list++ )
2884         {
2885             int i_slice;
2886             for( i_slice = 0; i_slice < 2; i_slice++ )
2887             {
2888                 char *p = buf;
2889                 int64_t i_den = 0;
2890                 int i_max = 0;
2891                 for( i = 0; i < 32; i++ )
2892                     if( h->stat.i_mb_count_ref[i_slice][i_list][i] )
2893                     {
2894                         i_den += h->stat.i_mb_count_ref[i_slice][i_list][i];
2895                         i_max = i;
2896                     }
2897                 if( i_max == 0 )
2898                     continue;
2899                 for( i = 0; i <= i_max; i++ )
2900                     p += sprintf( p, " %4.1f%%", 100. * h->stat.i_mb_count_ref[i_slice][i_list][i] / i_den );
2901                 x264_log( h, X264_LOG_INFO, "ref %c L%d:%s\n", "PB"[i_slice], i_list, buf );
2902             }
2903         }
2904
2905         if( h->param.analyse.b_ssim )
2906         {
2907             x264_log( h, X264_LOG_INFO,
2908                       "SSIM Mean Y:%.7f\n",
2909                       SUM3( h->stat.f_ssim_mean_y ) / i_count );
2910         }
2911         if( h->param.analyse.b_psnr )
2912         {
2913             x264_log( h, X264_LOG_INFO,
2914                       "PSNR Mean Y:%6.3f U:%6.3f V:%6.3f Avg:%6.3f Global:%6.3f kb/s:%.2f\n",
2915                       SUM3( h->stat.f_psnr_mean_y ) / i_count,
2916                       SUM3( h->stat.f_psnr_mean_u ) / i_count,
2917                       SUM3( h->stat.f_psnr_mean_v ) / i_count,
2918                       SUM3( h->stat.f_psnr_average ) / i_count,
2919                       x264_psnr( SUM3( h->stat.i_ssd_global ), i_count * i_yuv_size ),
2920                       f_bitrate );
2921         }
2922         else
2923             x264_log( h, X264_LOG_INFO, "kb/s:%.2f\n", f_bitrate );
2924     }
2925
2926     /* rc */
2927     x264_ratecontrol_delete( h );
2928
2929     /* param */
2930     if( h->param.rc.psz_stat_out )
2931         free( h->param.rc.psz_stat_out );
2932     if( h->param.rc.psz_stat_in )
2933         free( h->param.rc.psz_stat_in );
2934
2935     x264_cqm_delete( h );
2936     x264_free( h->nal_buffer );
2937     x264_analyse_free_costs( h );
2938
2939     if( h->i_thread_frames > 1)
2940         h = h->thread[h->i_thread_phase];
2941
2942     /* frames */
2943     x264_frame_delete_list( h->frames.unused[0] );
2944     x264_frame_delete_list( h->frames.unused[1] );
2945     x264_frame_delete_list( h->frames.current );
2946     x264_frame_delete_list( h->frames.blank_unused );
2947
2948     h = h->thread[0];
2949
2950     for( i = h->param.i_threads - 1; i >= 0; i-- )
2951     {
2952         x264_frame_t **frame;
2953
2954         if( !h->param.b_sliced_threads || i == 0 )
2955         {
2956             for( frame = h->thread[i]->frames.reference; *frame; frame++ )
2957             {
2958                 assert( (*frame)->i_reference_count > 0 );
2959                 (*frame)->i_reference_count--;
2960                 if( (*frame)->i_reference_count == 0 )
2961                     x264_frame_delete( *frame );
2962             }
2963             frame = &h->thread[i]->fdec;
2964             assert( (*frame)->i_reference_count > 0 );
2965             (*frame)->i_reference_count--;
2966             if( (*frame)->i_reference_count == 0 )
2967                 x264_frame_delete( *frame );
2968             x264_macroblock_cache_end( h->thread[i] );
2969         }
2970         x264_free( h->thread[i]->scratch_buffer );
2971         x264_free( h->thread[i]->out.p_bitstream );
2972         x264_free( h->thread[i]->out.nal);
2973         x264_free( h->thread[i] );
2974     }
2975 }
2976
2977 /****************************************************************************
2978  * x264_encoder_delayed_frames:
2979  ****************************************************************************/
2980 int x264_encoder_delayed_frames( x264_t *h )
2981 {
2982     int delayed_frames = 0;
2983     int i;
2984     if( h->i_thread_frames > 1 )
2985     {
2986         for( i=0; i<h->i_thread_frames; i++ )
2987             delayed_frames += h->thread[i]->b_thread_active;
2988         h = h->thread[h->i_thread_phase];
2989     }
2990     for( i=0; h->frames.current[i]; i++ )
2991         delayed_frames++;
2992     x264_pthread_mutex_lock( &h->lookahead->ofbuf.mutex );
2993     x264_pthread_mutex_lock( &h->lookahead->ifbuf.mutex );
2994     x264_pthread_mutex_lock( &h->lookahead->next.mutex );
2995     delayed_frames += h->lookahead->ifbuf.i_size + h->lookahead->next.i_size + h->lookahead->ofbuf.i_size;
2996     x264_pthread_mutex_unlock( &h->lookahead->next.mutex );
2997     x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
2998     x264_pthread_mutex_unlock( &h->lookahead->ofbuf.mutex );
2999     return delayed_frames;
3000 }