Replace High 4:4:4 profile lossless with High 4:4:4 Predictive.
[x262.git] / encoder / encoder.c
1 /*****************************************************************************
2  * x264: h264 encoder
3  *****************************************************************************
4  * Copyright (C) 2003-2008 x264 project
5  *
6  * Authors: Laurent Aimar <fenrir@via.ecp.fr>
7  *          Loren Merritt <lorenm@u.washington.edu>
8  *
9  * This program is free software; you can redistribute it and/or modify
10  * it under the terms of the GNU General Public License as published by
11  * the Free Software Foundation; either version 2 of the License, or
12  * (at your option) any later version.
13  *
14  * This program is distributed in the hope that it will be useful,
15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
17  * GNU General Public License for more details.
18  *
19  * You should have received a copy of the GNU General Public License
20  * along with this program; if not, write to the Free Software
21  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02111, USA.
22  *****************************************************************************/
23
24 #include <math.h>
25
26 #include "common/common.h"
27 #include "common/cpu.h"
28
29 #include "set.h"
30 #include "analyse.h"
31 #include "ratecontrol.h"
32 #include "macroblock.h"
33
34 #if VISUALIZE
35 #include "common/visualize.h"
36 #endif
37
38 //#define DEBUG_MB_TYPE
39
40 #define NALU_OVERHEAD 5 // startcode + NAL type costs 5 bytes per frame
41
42 #define bs_write_ue bs_write_ue_big
43
44 static void x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
45                                     x264_nal_t **pp_nal, int *pi_nal,
46                                     x264_picture_t *pic_out );
47
48 /****************************************************************************
49  *
50  ******************************* x264 libs **********************************
51  *
52  ****************************************************************************/
53 static float x264_psnr( int64_t i_sqe, int64_t i_size )
54 {
55     double f_mse = (double)i_sqe / ((double)65025.0 * (double)i_size);
56     if( f_mse <= 0.0000000001 ) /* Max 100dB */
57         return 100;
58
59     return (float)(-10.0 * log( f_mse ) / log( 10.0 ));
60 }
61
62 static void x264_frame_dump( x264_t *h )
63 {
64     FILE *f = fopen( h->param.psz_dump_yuv, "r+b" );
65     int i, y;
66     if( !f )
67         return;
68     /* Write the frame in display order */
69     fseek( f, h->fdec->i_frame * h->param.i_height * h->param.i_width * 3/2, SEEK_SET );
70     for( i = 0; i < h->fdec->i_plane; i++ )
71         for( y = 0; y < h->param.i_height >> !!i; y++ )
72             fwrite( &h->fdec->plane[i][y*h->fdec->i_stride[i]], 1, h->param.i_width >> !!i, f );
73     fclose( f );
74 }
75
76
77 /* Fill "default" values */
78 static void x264_slice_header_init( x264_t *h, x264_slice_header_t *sh,
79                                     x264_sps_t *sps, x264_pps_t *pps,
80                                     int i_idr_pic_id, int i_frame, int i_qp )
81 {
82     x264_param_t *param = &h->param;
83     int i;
84
85     /* First we fill all field */
86     sh->sps = sps;
87     sh->pps = pps;
88
89     sh->i_first_mb  = 0;
90     sh->i_last_mb   = h->sps->i_mb_width * h->sps->i_mb_height;
91     sh->i_pps_id    = pps->i_id;
92
93     sh->i_frame_num = i_frame;
94
95     sh->b_mbaff = h->param.b_interlaced;
96     sh->b_field_pic = 0;    /* no field support for now */
97     sh->b_bottom_field = 0; /* not yet used */
98
99     sh->i_idr_pic_id = i_idr_pic_id;
100
101     /* poc stuff, fixed later */
102     sh->i_poc_lsb = 0;
103     sh->i_delta_poc_bottom = 0;
104     sh->i_delta_poc[0] = 0;
105     sh->i_delta_poc[1] = 0;
106
107     sh->i_redundant_pic_cnt = 0;
108
109     if( !h->mb.b_direct_auto_read )
110     {
111         if( h->mb.b_direct_auto_write )
112             sh->b_direct_spatial_mv_pred = ( h->stat.i_direct_score[1] > h->stat.i_direct_score[0] );
113         else
114             sh->b_direct_spatial_mv_pred = ( param->analyse.i_direct_mv_pred == X264_DIRECT_PRED_SPATIAL );
115     }
116     /* else b_direct_spatial_mv_pred was read from the 2pass statsfile */
117
118     sh->b_num_ref_idx_override = 0;
119     sh->i_num_ref_idx_l0_active = 1;
120     sh->i_num_ref_idx_l1_active = 1;
121
122     sh->b_ref_pic_list_reordering_l0 = h->b_ref_reorder[0];
123     sh->b_ref_pic_list_reordering_l1 = h->b_ref_reorder[1];
124
125     /* If the ref list isn't in the default order, construct reordering header */
126     /* List1 reordering isn't needed yet */
127     if( sh->b_ref_pic_list_reordering_l0 )
128     {
129         int pred_frame_num = i_frame;
130         for( i = 0; i < h->i_ref0; i++ )
131         {
132             int diff = h->fref0[i]->i_frame_num - pred_frame_num;
133             if( diff == 0 )
134                 x264_log( h, X264_LOG_ERROR, "diff frame num == 0\n" );
135             sh->ref_pic_list_order[0][i].idc = ( diff > 0 );
136             sh->ref_pic_list_order[0][i].arg = abs( diff ) - 1;
137             pred_frame_num = h->fref0[i]->i_frame_num;
138         }
139     }
140
141     sh->i_cabac_init_idc = param->i_cabac_init_idc;
142
143     sh->i_qp = i_qp;
144     sh->i_qp_delta = i_qp - pps->i_pic_init_qp;
145     sh->b_sp_for_swidth = 0;
146     sh->i_qs_delta = 0;
147
148     /* If effective qp <= 15, deblocking would have no effect anyway */
149     if( param->b_deblocking_filter
150         && ( h->mb.b_variable_qp
151         || 15 < i_qp + 2 * X264_MIN(param->i_deblocking_filter_alphac0, param->i_deblocking_filter_beta) ) )
152     {
153         sh->i_disable_deblocking_filter_idc = 0;
154     }
155     else
156     {
157         sh->i_disable_deblocking_filter_idc = 1;
158     }
159     sh->i_alpha_c0_offset = param->i_deblocking_filter_alphac0 << 1;
160     sh->i_beta_offset = param->i_deblocking_filter_beta << 1;
161 }
162
163 static void x264_slice_header_write( bs_t *s, x264_slice_header_t *sh, int i_nal_ref_idc )
164 {
165     int i;
166
167     if( sh->b_mbaff )
168     {
169         assert( sh->i_first_mb % (2*sh->sps->i_mb_width) == 0 );
170         bs_write_ue( s, sh->i_first_mb >> 1 );
171     }
172     else
173         bs_write_ue( s, sh->i_first_mb );
174
175     bs_write_ue( s, sh->i_type + 5 );   /* same type things */
176     bs_write_ue( s, sh->i_pps_id );
177     bs_write( s, sh->sps->i_log2_max_frame_num, sh->i_frame_num );
178
179     if( !sh->sps->b_frame_mbs_only )
180     {
181         bs_write1( s, sh->b_field_pic );
182         if ( sh->b_field_pic )
183             bs_write1( s, sh->b_bottom_field );
184     }
185
186     if( sh->i_idr_pic_id >= 0 ) /* NAL IDR */
187     {
188         bs_write_ue( s, sh->i_idr_pic_id );
189     }
190
191     if( sh->sps->i_poc_type == 0 )
192     {
193         bs_write( s, sh->sps->i_log2_max_poc_lsb, sh->i_poc_lsb );
194         if( sh->pps->b_pic_order && !sh->b_field_pic )
195         {
196             bs_write_se( s, sh->i_delta_poc_bottom );
197         }
198     }
199     else if( sh->sps->i_poc_type == 1 && !sh->sps->b_delta_pic_order_always_zero )
200     {
201         bs_write_se( s, sh->i_delta_poc[0] );
202         if( sh->pps->b_pic_order && !sh->b_field_pic )
203         {
204             bs_write_se( s, sh->i_delta_poc[1] );
205         }
206     }
207
208     if( sh->pps->b_redundant_pic_cnt )
209     {
210         bs_write_ue( s, sh->i_redundant_pic_cnt );
211     }
212
213     if( sh->i_type == SLICE_TYPE_B )
214     {
215         bs_write1( s, sh->b_direct_spatial_mv_pred );
216     }
217     if( sh->i_type == SLICE_TYPE_P || sh->i_type == SLICE_TYPE_SP || sh->i_type == SLICE_TYPE_B )
218     {
219         bs_write1( s, sh->b_num_ref_idx_override );
220         if( sh->b_num_ref_idx_override )
221         {
222             bs_write_ue( s, sh->i_num_ref_idx_l0_active - 1 );
223             if( sh->i_type == SLICE_TYPE_B )
224             {
225                 bs_write_ue( s, sh->i_num_ref_idx_l1_active - 1 );
226             }
227         }
228     }
229
230     /* ref pic list reordering */
231     if( sh->i_type != SLICE_TYPE_I )
232     {
233         bs_write1( s, sh->b_ref_pic_list_reordering_l0 );
234         if( sh->b_ref_pic_list_reordering_l0 )
235         {
236             for( i = 0; i < sh->i_num_ref_idx_l0_active; i++ )
237             {
238                 bs_write_ue( s, sh->ref_pic_list_order[0][i].idc );
239                 bs_write_ue( s, sh->ref_pic_list_order[0][i].arg );
240
241             }
242             bs_write_ue( s, 3 );
243         }
244     }
245     if( sh->i_type == SLICE_TYPE_B )
246     {
247         bs_write1( s, sh->b_ref_pic_list_reordering_l1 );
248         if( sh->b_ref_pic_list_reordering_l1 )
249         {
250             for( i = 0; i < sh->i_num_ref_idx_l1_active; i++ )
251             {
252                 bs_write_ue( s, sh->ref_pic_list_order[1][i].idc );
253                 bs_write_ue( s, sh->ref_pic_list_order[1][i].arg );
254             }
255             bs_write_ue( s, 3 );
256         }
257     }
258
259     if( ( sh->pps->b_weighted_pred && ( sh->i_type == SLICE_TYPE_P || sh->i_type == SLICE_TYPE_SP ) ) ||
260         ( sh->pps->b_weighted_bipred == 1 && sh->i_type == SLICE_TYPE_B ) )
261     {
262         /* FIXME */
263     }
264
265     if( i_nal_ref_idc != 0 )
266     {
267         if( sh->i_idr_pic_id >= 0 )
268         {
269             bs_write1( s, 0 );  /* no output of prior pics flag */
270             bs_write1( s, 0 );  /* long term reference flag */
271         }
272         else
273         {
274             bs_write1( s, 0 );  /* adaptive_ref_pic_marking_mode_flag */
275         }
276     }
277
278     if( sh->pps->b_cabac && sh->i_type != SLICE_TYPE_I )
279     {
280         bs_write_ue( s, sh->i_cabac_init_idc );
281     }
282     bs_write_se( s, sh->i_qp_delta );      /* slice qp delta */
283
284     if( sh->pps->b_deblocking_filter_control )
285     {
286         bs_write_ue( s, sh->i_disable_deblocking_filter_idc );
287         if( sh->i_disable_deblocking_filter_idc != 1 )
288         {
289             bs_write_se( s, sh->i_alpha_c0_offset >> 1 );
290             bs_write_se( s, sh->i_beta_offset >> 1 );
291         }
292     }
293 }
294
295 /* If we are within a reasonable distance of the end of the memory allocated for the bitstream, */
296 /* reallocate, adding an arbitrary amount of space (100 kilobytes). */
297 static void x264_bitstream_check_buffer( x264_t *h )
298 {
299     if( ( h->param.b_cabac && (h->cabac.p_end - h->cabac.p < 2500) )
300      || ( h->out.bs.p_end - h->out.bs.p < 2500 ) )
301     {
302         uint8_t *bs_bak = h->out.p_bitstream;
303         intptr_t delta;
304         int i;
305
306         h->out.i_bitstream += 100000;
307         h->out.p_bitstream = x264_realloc( h->out.p_bitstream, h->out.i_bitstream );
308         delta = h->out.p_bitstream - bs_bak;
309
310         h->out.bs.p_start += delta;
311         h->out.bs.p += delta;
312         h->out.bs.p_end = h->out.p_bitstream + h->out.i_bitstream;
313
314         h->cabac.p_start += delta;
315         h->cabac.p += delta;
316         h->cabac.p_end = h->out.p_bitstream + h->out.i_bitstream;
317
318         for( i = 0; i <= h->out.i_nal; i++ )
319             h->out.nal[i].p_payload += delta;
320     }
321 }
322
323 /****************************************************************************
324  *
325  ****************************************************************************
326  ****************************** External API*********************************
327  ****************************************************************************
328  *
329  ****************************************************************************/
330
331 static int x264_validate_parameters( x264_t *h )
332 {
333 #ifdef HAVE_MMX
334     if( !(x264_cpu_detect() & X264_CPU_MMXEXT) )
335     {
336         x264_log( h, X264_LOG_ERROR, "your cpu does not support MMXEXT, but x264 was compiled with asm support\n");
337         x264_log( h, X264_LOG_ERROR, "to run x264, recompile without asm support (configure --disable-asm)\n");
338         return -1;
339     }
340 #endif
341     if( h->param.i_width <= 0 || h->param.i_height <= 0 )
342     {
343         x264_log( h, X264_LOG_ERROR, "invalid width x height (%dx%d)\n",
344                   h->param.i_width, h->param.i_height );
345         return -1;
346     }
347
348     if( h->param.i_width % 2 || h->param.i_height % 2 )
349     {
350         x264_log( h, X264_LOG_ERROR, "width or height not divisible by 2 (%dx%d)\n",
351                   h->param.i_width, h->param.i_height );
352         return -1;
353     }
354     if( h->param.i_csp != X264_CSP_I420 )
355     {
356         x264_log( h, X264_LOG_ERROR, "invalid CSP (only I420 supported)\n" );
357         return -1;
358     }
359
360     if( h->param.i_threads == 0 )
361         h->param.i_threads = x264_cpu_num_processors() * 3/2;
362     h->param.i_threads = x264_clip3( h->param.i_threads, 1, X264_THREAD_MAX );
363     if( h->param.i_threads > 1 )
364     {
365 #ifndef HAVE_PTHREAD
366         x264_log( h, X264_LOG_WARNING, "not compiled with pthread support!\n");
367         h->param.i_threads = 1;
368 #else
369         if( h->param.i_scenecut_threshold >= 0 )
370             h->param.b_pre_scenecut = 1;
371 #endif
372     }
373
374     if( h->param.b_interlaced )
375     {
376         if( h->param.analyse.i_me_method >= X264_ME_ESA )
377         {
378             x264_log( h, X264_LOG_WARNING, "interlace + me=esa is not implemented\n" );
379             h->param.analyse.i_me_method = X264_ME_UMH;
380         }
381         if( h->param.analyse.i_direct_mv_pred > X264_DIRECT_PRED_SPATIAL )
382         {
383             x264_log( h, X264_LOG_WARNING, "interlace + direct=temporal is not implemented\n" );
384             h->param.analyse.i_direct_mv_pred = X264_DIRECT_PRED_SPATIAL;
385         }
386     }
387
388     if( h->param.rc.i_rc_method < 0 || h->param.rc.i_rc_method > 2 )
389     {
390         x264_log( h, X264_LOG_ERROR, "no ratecontrol method specified\n" );
391         return -1;
392     }
393     h->param.rc.f_rf_constant = x264_clip3f( h->param.rc.f_rf_constant, 0, 51 );
394     h->param.rc.i_qp_constant = x264_clip3( h->param.rc.i_qp_constant, 0, 51 );
395     if( h->param.rc.i_rc_method == X264_RC_CRF )
396         h->param.rc.i_qp_constant = h->param.rc.f_rf_constant;
397     if( (h->param.rc.i_rc_method == X264_RC_CQP || h->param.rc.i_rc_method == X264_RC_CRF)
398         && h->param.rc.i_qp_constant == 0 )
399     {
400         h->mb.b_lossless = 1;
401         h->param.i_cqm_preset = X264_CQM_FLAT;
402         h->param.psz_cqm_file = NULL;
403         h->param.rc.i_rc_method = X264_RC_CQP;
404         h->param.rc.f_ip_factor = 1;
405         h->param.rc.f_pb_factor = 1;
406         h->param.analyse.b_psnr = 0;
407         h->param.analyse.b_ssim = 0;
408         h->param.analyse.i_chroma_qp_offset = 0;
409         h->param.analyse.i_trellis = 0;
410         h->param.analyse.b_fast_pskip = 0;
411         h->param.analyse.i_noise_reduction = 0;
412         h->param.analyse.f_psy_rd = 0;
413         /* 8x8dct is not useful at all in CAVLC lossless */
414         if( !h->param.b_cabac )
415             h->param.analyse.b_transform_8x8 = 0;
416     }
417     if( h->param.rc.i_rc_method == X264_RC_CQP )
418     {
419         float qp_p = h->param.rc.i_qp_constant;
420         float qp_i = qp_p - 6*log(h->param.rc.f_ip_factor)/log(2);
421         float qp_b = qp_p + 6*log(h->param.rc.f_pb_factor)/log(2);
422         h->param.rc.i_qp_min = x264_clip3( (int)(X264_MIN3( qp_p, qp_i, qp_b )), 0, 51 );
423         h->param.rc.i_qp_max = x264_clip3( (int)(X264_MAX3( qp_p, qp_i, qp_b ) + .999), 0, 51 );
424         h->param.rc.i_aq_mode = 0;
425     }
426     h->param.rc.i_qp_max = x264_clip3( h->param.rc.i_qp_max, 0, 51 );
427     h->param.rc.i_qp_min = x264_clip3( h->param.rc.i_qp_min, 0, h->param.rc.i_qp_max );
428
429     if( ( h->param.i_width % 16 || h->param.i_height % 16 )
430         && h->param.i_height != 1080 && !h->mb.b_lossless )
431     {
432         // There's nothing special about 1080 in that the warning still applies to it,
433         // but chances are the user can't help it if his content is already 1080p,
434         // so there's no point in warning in that case.
435         x264_log( h, X264_LOG_WARNING,
436                   "width or height not divisible by 16 (%dx%d), compression will suffer.\n",
437                   h->param.i_width, h->param.i_height );
438     }
439
440     h->param.i_frame_reference = x264_clip3( h->param.i_frame_reference, 1, 16 );
441     if( h->param.i_keyint_max <= 0 )
442         h->param.i_keyint_max = 1;
443     h->param.i_keyint_min = x264_clip3( h->param.i_keyint_min, 1, h->param.i_keyint_max/2+1 );
444
445     h->param.i_bframe = x264_clip3( h->param.i_bframe, 0, X264_BFRAME_MAX );
446     h->param.i_bframe_bias = x264_clip3( h->param.i_bframe_bias, -90, 100 );
447     h->param.b_bframe_pyramid = h->param.b_bframe_pyramid && h->param.i_bframe > 1;
448     if( !h->param.i_bframe )
449         h->param.i_bframe_adaptive = X264_B_ADAPT_NONE;
450     h->param.analyse.b_weighted_bipred = h->param.analyse.b_weighted_bipred && h->param.i_bframe > 0;
451     h->mb.b_direct_auto_write = h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO
452                                 && h->param.i_bframe
453                                 && ( h->param.rc.b_stat_write || !h->param.rc.b_stat_read );
454     if( h->param.i_scenecut_threshold < 0 )
455         h->param.b_pre_scenecut = 0;
456
457     h->param.i_deblocking_filter_alphac0 = x264_clip3( h->param.i_deblocking_filter_alphac0, -6, 6 );
458     h->param.i_deblocking_filter_beta    = x264_clip3( h->param.i_deblocking_filter_beta, -6, 6 );
459     h->param.analyse.i_luma_deadzone[0] = x264_clip3( h->param.analyse.i_luma_deadzone[0], 0, 32 );
460     h->param.analyse.i_luma_deadzone[1] = x264_clip3( h->param.analyse.i_luma_deadzone[1], 0, 32 );
461
462     h->param.i_cabac_init_idc = x264_clip3( h->param.i_cabac_init_idc, 0, 2 );
463
464     if( h->param.i_cqm_preset < X264_CQM_FLAT || h->param.i_cqm_preset > X264_CQM_CUSTOM )
465         h->param.i_cqm_preset = X264_CQM_FLAT;
466
467     if( h->param.analyse.i_me_method < X264_ME_DIA ||
468         h->param.analyse.i_me_method > X264_ME_TESA )
469         h->param.analyse.i_me_method = X264_ME_HEX;
470     if( h->param.analyse.i_me_range < 4 )
471         h->param.analyse.i_me_range = 4;
472     if( h->param.analyse.i_me_range > 16 && h->param.analyse.i_me_method <= X264_ME_HEX )
473         h->param.analyse.i_me_range = 16;
474     if( h->param.analyse.i_me_method == X264_ME_TESA &&
475         (h->mb.b_lossless || h->param.analyse.i_subpel_refine <= 1) )
476         h->param.analyse.i_me_method = X264_ME_ESA;
477     h->param.analyse.i_subpel_refine = x264_clip3( h->param.analyse.i_subpel_refine, 1, 7 );
478     h->param.analyse.b_bframe_rdo = h->param.analyse.b_bframe_rdo && h->param.analyse.i_subpel_refine >= 6;
479     h->param.analyse.b_mixed_references = h->param.analyse.b_mixed_references && h->param.i_frame_reference > 1;
480     h->param.analyse.inter &= X264_ANALYSE_PSUB16x16|X264_ANALYSE_PSUB8x8|X264_ANALYSE_BSUB16x16|
481                               X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
482     h->param.analyse.intra &= X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
483     if( !(h->param.analyse.inter & X264_ANALYSE_PSUB16x16) )
484         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
485     if( !h->param.analyse.b_transform_8x8 )
486     {
487         h->param.analyse.inter &= ~X264_ANALYSE_I8x8;
488         h->param.analyse.intra &= ~X264_ANALYSE_I8x8;
489     }
490     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -12, 12);
491     if( !h->param.b_cabac )
492         h->param.analyse.i_trellis = 0;
493     h->param.analyse.i_trellis = x264_clip3( h->param.analyse.i_trellis, 0, 2 );
494     if( !h->param.analyse.i_trellis )
495         h->param.analyse.f_psy_trellis = 0;
496     h->param.analyse.f_psy_rd = x264_clip3f( h->param.analyse.f_psy_rd, 0, 10 );
497     h->param.analyse.f_psy_trellis = x264_clip3f( h->param.analyse.f_psy_trellis, 0, 10 );
498     if( h->param.analyse.i_subpel_refine < 6 )
499         h->param.analyse.f_psy_rd = 0;
500     h->mb.i_psy_rd = FIX8( h->param.analyse.f_psy_rd );
501     /* Psy RDO increases overall quantizers to improve the quality of luma--this indirectly hurts chroma quality */
502     /* so we lower the chroma QP offset to compensate */
503     /* This can be triggered repeatedly on multiple calls to parameter_validate, but since encoding
504      * uses the pps chroma qp offset not the param chroma qp offset, this is not a problem. */
505     if( h->mb.i_psy_rd )
506         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_rd < 0.25 ? 1 : 2;
507     h->mb.i_psy_trellis = FIX8( h->param.analyse.f_psy_trellis / 4 );
508     /* Psy trellis has a similar effect. */
509     if( h->mb.i_psy_trellis )
510         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_trellis < 0.25 ? 1 : 2;
511     else
512         h->mb.i_psy_trellis = 0;
513     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -12, 12);
514     h->param.rc.i_aq_mode = x264_clip3( h->param.rc.i_aq_mode, 0, 1 );
515     if( h->param.rc.f_aq_strength <= 0 )
516         h->param.rc.i_aq_mode = 0;
517     h->param.analyse.i_noise_reduction = x264_clip3( h->param.analyse.i_noise_reduction, 0, 1<<16 );
518
519     {
520         const x264_level_t *l = x264_levels;
521         if( h->param.i_level_idc < 0 )
522         {
523             if( h->param.rc.i_rc_method == X264_RC_ABR && h->param.rc.i_vbv_buffer_size <= 0 )
524                 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate * 2;
525             h->sps = h->sps_array;
526             x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
527             do h->param.i_level_idc = l->level_idc;
528                 while( l[1].level_idc && x264_validate_levels( h, 0 ) && l++ );
529             if( h->param.rc.i_vbv_buffer_size <= 0 )
530                 h->param.rc.i_vbv_max_bitrate = 0;
531             x264_log( h, X264_LOG_DEBUG, "level_idc: %d\n", h->param.i_level_idc );
532         }
533         else
534         {
535             while( l->level_idc && l->level_idc != h->param.i_level_idc )
536                 l++;
537             if( l->level_idc == 0 )
538             {
539                 x264_log( h, X264_LOG_ERROR, "invalid level_idc: %d\n", h->param.i_level_idc );
540                 return -1;
541             }
542         }
543         if( h->param.analyse.i_mv_range <= 0 )
544             h->param.analyse.i_mv_range = l->mv_range >> h->param.b_interlaced;
545         else
546             h->param.analyse.i_mv_range = x264_clip3(h->param.analyse.i_mv_range, 32, 512 >> h->param.b_interlaced);
547         if( h->param.analyse.i_direct_8x8_inference < 0 )
548             h->param.analyse.i_direct_8x8_inference = l->direct8x8;
549     }
550
551     if( h->param.i_threads > 1 )
552     {
553         int r = h->param.analyse.i_mv_range_thread;
554         int r2;
555         if( r <= 0 )
556         {
557             // half of the available space is reserved and divided evenly among the threads,
558             // the rest is allocated to whichever thread is far enough ahead to use it.
559             // reserving more space increases quality for some videos, but costs more time
560             // in thread synchronization.
561             int max_range = (h->param.i_height + X264_THREAD_HEIGHT) / h->param.i_threads - X264_THREAD_HEIGHT;
562             r = max_range / 2;
563         }
564         r = X264_MAX( r, h->param.analyse.i_me_range );
565         r = X264_MIN( r, h->param.analyse.i_mv_range );
566         // round up to use the whole mb row
567         r2 = (r & ~15) + ((-X264_THREAD_HEIGHT) & 15);
568         if( r2 < r )
569             r2 += 16;
570         x264_log( h, X264_LOG_DEBUG, "using mv_range_thread = %d\n", r2 );
571         h->param.analyse.i_mv_range_thread = r2;
572     }
573
574     if( h->param.rc.f_qblur < 0 )
575         h->param.rc.f_qblur = 0;
576     if( h->param.rc.f_complexity_blur < 0 )
577         h->param.rc.f_complexity_blur = 0;
578
579     h->param.i_sps_id &= 31;
580
581     if( h->param.i_log_level < X264_LOG_INFO )
582     {
583         h->param.analyse.b_psnr = 0;
584         h->param.analyse.b_ssim = 0;
585     }
586
587     /* ensure the booleans are 0 or 1 so they can be used in math */
588 #define BOOLIFY(x) h->param.x = !!h->param.x
589     BOOLIFY( b_cabac );
590     BOOLIFY( b_deblocking_filter );
591     BOOLIFY( b_interlaced );
592     BOOLIFY( analyse.b_transform_8x8 );
593     BOOLIFY( analyse.i_direct_8x8_inference );
594     BOOLIFY( analyse.b_bidir_me );
595     BOOLIFY( analyse.b_chroma_me );
596     BOOLIFY( analyse.b_fast_pskip );
597     BOOLIFY( rc.b_stat_write );
598     BOOLIFY( rc.b_stat_read );
599 #undef BOOLIFY
600
601     return 0;
602 }
603
604 static void mbcmp_init( x264_t *h )
605 {
606     int satd = !h->mb.b_lossless && h->param.analyse.i_subpel_refine > 1;
607     memcpy( h->pixf.mbcmp, satd ? h->pixf.satd : h->pixf.sad_aligned, sizeof(h->pixf.mbcmp) );
608     memcpy( h->pixf.mbcmp_unaligned, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.mbcmp_unaligned) );
609     h->pixf.intra_mbcmp_x3_16x16 = satd ? h->pixf.intra_satd_x3_16x16 : h->pixf.intra_sad_x3_16x16;
610     satd &= h->param.analyse.i_me_method == X264_ME_TESA;
611     memcpy( h->pixf.fpelcmp, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.fpelcmp) );
612     memcpy( h->pixf.fpelcmp_x3, satd ? h->pixf.satd_x3 : h->pixf.sad_x3, sizeof(h->pixf.fpelcmp_x3) );
613     memcpy( h->pixf.fpelcmp_x4, satd ? h->pixf.satd_x4 : h->pixf.sad_x4, sizeof(h->pixf.fpelcmp_x4) );
614 }
615
616 /****************************************************************************
617  * x264_encoder_open:
618  ****************************************************************************/
619 x264_t *x264_encoder_open   ( x264_param_t *param )
620 {
621     x264_t *h = x264_malloc( sizeof( x264_t ) );
622     char buf[1000], *p;
623     int i;
624
625     memset( h, 0, sizeof( x264_t ) );
626
627     /* Create a copy of param */
628     memcpy( &h->param, param, sizeof( x264_param_t ) );
629
630     if( x264_validate_parameters( h ) < 0 )
631     {
632         x264_free( h );
633         return NULL;
634     }
635
636     if( h->param.psz_cqm_file )
637         if( x264_cqm_parse_file( h, h->param.psz_cqm_file ) < 0 )
638         {
639             x264_free( h );
640             return NULL;
641         }
642
643     if( h->param.rc.psz_stat_out )
644         h->param.rc.psz_stat_out = strdup( h->param.rc.psz_stat_out );
645     if( h->param.rc.psz_stat_in )
646         h->param.rc.psz_stat_in = strdup( h->param.rc.psz_stat_in );
647
648     /* VUI */
649     if( h->param.vui.i_sar_width > 0 && h->param.vui.i_sar_height > 0 )
650     {
651         int i_w = param->vui.i_sar_width;
652         int i_h = param->vui.i_sar_height;
653
654         x264_reduce_fraction( &i_w, &i_h );
655
656         while( i_w > 65535 || i_h > 65535 )
657         {
658             i_w /= 2;
659             i_h /= 2;
660         }
661
662         h->param.vui.i_sar_width = 0;
663         h->param.vui.i_sar_height = 0;
664         if( i_w == 0 || i_h == 0 )
665         {
666             x264_log( h, X264_LOG_WARNING, "cannot create valid sample aspect ratio\n" );
667         }
668         else
669         {
670             x264_log( h, X264_LOG_INFO, "using SAR=%d/%d\n", i_w, i_h );
671             h->param.vui.i_sar_width = i_w;
672             h->param.vui.i_sar_height = i_h;
673         }
674     }
675
676     x264_reduce_fraction( &h->param.i_fps_num, &h->param.i_fps_den );
677
678     /* Init x264_t */
679     h->i_frame = 0;
680     h->i_frame_num = 0;
681     h->i_idr_pic_id = 0;
682
683     h->sps = &h->sps_array[0];
684     x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
685
686     h->pps = &h->pps_array[0];
687     x264_pps_init( h->pps, h->param.i_sps_id, &h->param, h->sps);
688
689     x264_validate_levels( h, 1 );
690
691     if( x264_cqm_init( h ) < 0 )
692     {
693         x264_free( h );
694         return NULL;
695     }
696
697     h->mb.i_mb_count = h->sps->i_mb_width * h->sps->i_mb_height;
698
699     /* Init frames. */
700     if( h->param.i_bframe_adaptive == X264_B_ADAPT_TRELLIS )
701         h->frames.i_delay = X264_MAX(h->param.i_bframe,3)*4 + h->param.i_threads - 1;
702     else
703         h->frames.i_delay = h->param.i_bframe + h->param.i_threads - 1;
704     h->frames.i_max_ref0 = h->param.i_frame_reference;
705     h->frames.i_max_ref1 = h->sps->vui.i_num_reorder_frames;
706     h->frames.i_max_dpb  = h->sps->vui.i_max_dec_frame_buffering;
707     h->frames.b_have_lowres = !h->param.rc.b_stat_read
708         && ( h->param.rc.i_rc_method == X264_RC_ABR
709           || h->param.rc.i_rc_method == X264_RC_CRF
710           || h->param.i_bframe_adaptive
711           || h->param.b_pre_scenecut );
712     h->frames.b_have_lowres |= (h->param.rc.b_stat_read && h->param.rc.i_vbv_buffer_size > 0);
713
714     h->frames.i_last_idr = - h->param.i_keyint_max;
715     h->frames.i_input    = 0;
716     h->frames.last_nonb  = NULL;
717
718     h->i_ref0 = 0;
719     h->i_ref1 = 0;
720
721     h->chroma_qp_table = i_chroma_qp_table + 12 + h->pps->i_chroma_qp_index_offset;
722
723     x264_rdo_init( );
724
725     /* init CPU functions */
726     x264_predict_16x16_init( h->param.cpu, h->predict_16x16 );
727     x264_predict_8x8c_init( h->param.cpu, h->predict_8x8c );
728     x264_predict_8x8_init( h->param.cpu, h->predict_8x8 );
729     x264_predict_4x4_init( h->param.cpu, h->predict_4x4 );
730
731     x264_pixel_init( h->param.cpu, &h->pixf );
732     x264_dct_init( h->param.cpu, &h->dctf );
733     x264_zigzag_init( h->param.cpu, &h->zigzagf, h->param.b_interlaced );
734     x264_mc_init( h->param.cpu, &h->mc );
735     x264_quant_init( h, h->param.cpu, &h->quantf );
736     x264_deblock_init( h->param.cpu, &h->loopf );
737     x264_dct_init_weights();
738
739     mbcmp_init( h );
740
741     p = buf + sprintf( buf, "using cpu capabilities:" );
742     for( i=0; x264_cpu_names[i].flags; i++ )
743     {
744         if( !strcmp(x264_cpu_names[i].name, "SSE2")
745             && param->cpu & (X264_CPU_SSE2_IS_FAST|X264_CPU_SSE2_IS_SLOW) )
746             continue;
747         if( !strcmp(x264_cpu_names[i].name, "SSE3")
748             && (param->cpu & X264_CPU_SSSE3 || !(param->cpu & X264_CPU_CACHELINE_64)) )
749             continue;
750         if( (param->cpu & x264_cpu_names[i].flags) == x264_cpu_names[i].flags
751             && (!i || x264_cpu_names[i].flags != x264_cpu_names[i-1].flags) )
752             p += sprintf( p, " %s", x264_cpu_names[i].name );
753     }
754     if( !param->cpu )
755         p += sprintf( p, " none!" );
756     x264_log( h, X264_LOG_INFO, "%s\n", buf );
757
758     h->out.i_nal = 0;
759     h->out.i_bitstream = X264_MAX( 1000000, h->param.i_width * h->param.i_height * 4
760         * ( h->param.rc.i_rc_method == X264_RC_ABR ? pow( 0.95, h->param.rc.i_qp_min )
761           : pow( 0.95, h->param.rc.i_qp_constant ) * X264_MAX( 1, h->param.rc.f_ip_factor )));
762
763     h->thread[0] = h;
764     h->i_thread_num = 0;
765     for( i = 1; i < h->param.i_threads; i++ )
766         h->thread[i] = x264_malloc( sizeof(x264_t) );
767
768     for( i = 0; i < h->param.i_threads; i++ )
769     {
770         if( i > 0 )
771             *h->thread[i] = *h;
772         h->thread[i]->fdec = x264_frame_pop_unused( h );
773         h->thread[i]->out.p_bitstream = x264_malloc( h->out.i_bitstream );
774         if( x264_macroblock_cache_init( h->thread[i] ) < 0 )
775             return NULL;
776     }
777
778     if( x264_ratecontrol_new( h ) < 0 )
779         return NULL;
780
781     if( h->param.psz_dump_yuv )
782     {
783         /* create or truncate the reconstructed video file */
784         FILE *f = fopen( h->param.psz_dump_yuv, "w" );
785         if( f )
786             fclose( f );
787         else
788         {
789             x264_log( h, X264_LOG_ERROR, "can't write to fdec.yuv\n" );
790             x264_free( h );
791             return NULL;
792         }
793     }
794
795     return h;
796 }
797
798 /****************************************************************************
799  * x264_encoder_reconfig:
800  ****************************************************************************/
801 int x264_encoder_reconfig( x264_t *h, x264_param_t *param )
802 {
803 #define COPY(var) h->param.var = param->var
804     COPY( i_frame_reference ); // but never uses more refs than initially specified
805     COPY( i_bframe_bias );
806     if( h->param.i_scenecut_threshold >= 0 && param->i_scenecut_threshold >= 0 )
807         COPY( i_scenecut_threshold ); // can't turn it on or off, only vary the threshold
808     COPY( b_deblocking_filter );
809     COPY( i_deblocking_filter_alphac0 );
810     COPY( i_deblocking_filter_beta );
811     COPY( analyse.intra );
812     COPY( analyse.inter );
813     COPY( analyse.i_direct_mv_pred );
814     COPY( analyse.i_me_method );
815     COPY( analyse.i_me_range );
816     COPY( analyse.i_noise_reduction );
817     COPY( analyse.i_subpel_refine );
818     COPY( analyse.i_trellis );
819     COPY( analyse.b_bidir_me );
820     COPY( analyse.b_bframe_rdo );
821     COPY( analyse.b_chroma_me );
822     COPY( analyse.b_dct_decimate );
823     COPY( analyse.b_fast_pskip );
824     COPY( analyse.b_mixed_references );
825     // can only twiddle these if they were enabled to begin with:
826     if( h->pps->b_transform_8x8_mode )
827         COPY( analyse.b_transform_8x8 );
828     if( h->frames.i_max_ref1 > 1 )
829         COPY( b_bframe_pyramid );
830 #undef COPY
831
832     mbcmp_init( h );
833
834     return x264_validate_parameters( h );
835 }
836
837 /* internal usage */
838 static void x264_nal_start( x264_t *h, int i_type, int i_ref_idc )
839 {
840     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
841
842     nal->i_ref_idc = i_ref_idc;
843     nal->i_type    = i_type;
844
845     nal->i_payload= 0;
846     nal->p_payload= &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8];
847 }
848 static void x264_nal_end( x264_t *h )
849 {
850     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
851     nal->i_payload = &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8] - nal->p_payload;
852     h->out.i_nal++;
853 }
854
855 /****************************************************************************
856  * x264_encoder_headers:
857  ****************************************************************************/
858 int x264_encoder_headers( x264_t *h, x264_nal_t **pp_nal, int *pi_nal )
859 {
860     /* init bitstream context */
861     h->out.i_nal = 0;
862     bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
863
864     /* Put SPS and PPS */
865     if( h->i_frame == 0 )
866     {
867         /* identify ourself */
868         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
869         x264_sei_version_write( h, &h->out.bs );
870         x264_nal_end( h );
871
872         /* generate sequence parameters */
873         x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
874         x264_sps_write( &h->out.bs, h->sps );
875         x264_nal_end( h );
876
877         /* generate picture parameters */
878         x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
879         x264_pps_write( &h->out.bs, h->pps );
880         x264_nal_end( h );
881     }
882     /* now set output*/
883     *pi_nal = h->out.i_nal;
884     *pp_nal = &h->out.nal[0];
885     h->out.i_nal = 0;
886
887     return 0;
888 }
889
890 static inline void x264_reference_build_list( x264_t *h, int i_poc )
891 {
892     int i;
893     int b_ok;
894
895     /* build ref list 0/1 */
896     h->i_ref0 = 0;
897     h->i_ref1 = 0;
898     for( i = 0; h->frames.reference[i]; i++ )
899     {
900         if( h->frames.reference[i]->i_poc < i_poc )
901         {
902             h->fref0[h->i_ref0++] = h->frames.reference[i];
903         }
904         else if( h->frames.reference[i]->i_poc > i_poc )
905         {
906             h->fref1[h->i_ref1++] = h->frames.reference[i];
907         }
908     }
909
910     /* Order ref0 from higher to lower poc */
911     do
912     {
913         b_ok = 1;
914         for( i = 0; i < h->i_ref0 - 1; i++ )
915         {
916             if( h->fref0[i]->i_poc < h->fref0[i+1]->i_poc )
917             {
918                 XCHG( x264_frame_t*, h->fref0[i], h->fref0[i+1] );
919                 b_ok = 0;
920                 break;
921             }
922         }
923     } while( !b_ok );
924     /* Order ref1 from lower to higher poc (bubble sort) for B-frame */
925     do
926     {
927         b_ok = 1;
928         for( i = 0; i < h->i_ref1 - 1; i++ )
929         {
930             if( h->fref1[i]->i_poc > h->fref1[i+1]->i_poc )
931             {
932                 XCHG( x264_frame_t*, h->fref1[i], h->fref1[i+1] );
933                 b_ok = 0;
934                 break;
935             }
936         }
937     } while( !b_ok );
938
939     /* In the standard, a P-frame's ref list is sorted by frame_num.
940      * We use POC, but check whether explicit reordering is needed */
941     h->b_ref_reorder[0] =
942     h->b_ref_reorder[1] = 0;
943     if( h->sh.i_type == SLICE_TYPE_P )
944     {
945         for( i = 0; i < h->i_ref0 - 1; i++ )
946             if( h->fref0[i]->i_frame_num < h->fref0[i+1]->i_frame_num )
947             {
948                 h->b_ref_reorder[0] = 1;
949                 break;
950             }
951     }
952
953     h->i_ref1 = X264_MIN( h->i_ref1, h->frames.i_max_ref1 );
954     h->i_ref0 = X264_MIN( h->i_ref0, h->frames.i_max_ref0 );
955     h->i_ref0 = X264_MIN( h->i_ref0, h->param.i_frame_reference ); // if reconfig() has lowered the limit
956     assert( h->i_ref0 + h->i_ref1 <= 16 );
957     h->mb.pic.i_fref[0] = h->i_ref0;
958     h->mb.pic.i_fref[1] = h->i_ref1;
959 }
960
961 static void x264_fdec_filter_row( x264_t *h, int mb_y )
962 {
963     /* mb_y is the mb to be encoded next, not the mb to be filtered here */
964     int b_hpel = h->fdec->b_kept_as_ref;
965     int b_deblock = !h->sh.i_disable_deblocking_filter_idc;
966     int b_end = mb_y == h->sps->i_mb_height;
967     int min_y = mb_y - (1 << h->sh.b_mbaff);
968     int max_y = b_end ? h->sps->i_mb_height : mb_y;
969     b_deblock &= b_hpel || h->param.psz_dump_yuv;
970     if( mb_y & h->sh.b_mbaff )
971         return;
972     if( min_y < 0 )
973         return;
974
975     if( !b_end )
976     {
977         int i, j;
978         for( j=0; j<=h->sh.b_mbaff; j++ )
979             for( i=0; i<3; i++ )
980             {
981                 memcpy( h->mb.intra_border_backup[j][i],
982                         h->fdec->plane[i] + ((mb_y*16 >> !!i) + j - 1 - h->sh.b_mbaff) * h->fdec->i_stride[i],
983                         h->sps->i_mb_width*16 >> !!i );
984             }
985     }
986
987     if( b_deblock )
988     {
989         int y;
990         for( y = min_y; y < max_y; y += (1 << h->sh.b_mbaff) )
991             x264_frame_deblock_row( h, y );
992     }
993
994     if( b_hpel )
995     {
996         x264_frame_expand_border( h, h->fdec, min_y, b_end );
997         x264_frame_filter( h, h->fdec, min_y, b_end );
998         x264_frame_expand_border_filtered( h, h->fdec, min_y, b_end );
999     }
1000
1001     if( h->param.i_threads > 1 && h->fdec->b_kept_as_ref )
1002     {
1003         x264_frame_cond_broadcast( h->fdec, mb_y*16 + (b_end ? 10000 : -(X264_THREAD_HEIGHT << h->sh.b_mbaff)) );
1004     }
1005
1006     min_y = X264_MAX( min_y*16-8, 0 );
1007     max_y = b_end ? h->param.i_height : mb_y*16-8;
1008
1009     if( h->param.analyse.b_psnr )
1010     {
1011         int i;
1012         for( i=0; i<3; i++ )
1013             h->stat.frame.i_ssd[i] +=
1014                 x264_pixel_ssd_wxh( &h->pixf,
1015                     h->fdec->plane[i] + (min_y>>!!i) * h->fdec->i_stride[i], h->fdec->i_stride[i],
1016                     h->fenc->plane[i] + (min_y>>!!i) * h->fenc->i_stride[i], h->fenc->i_stride[i],
1017                     h->param.i_width >> !!i, (max_y-min_y) >> !!i );
1018     }
1019
1020     if( h->param.analyse.b_ssim )
1021     {
1022         x264_emms();
1023         /* offset by 2 pixels to avoid alignment of ssim blocks with dct blocks,
1024          * and overlap by 4 */
1025         min_y += min_y == 0 ? 2 : -6;
1026         h->stat.frame.f_ssim +=
1027             x264_pixel_ssim_wxh( &h->pixf,
1028                 h->fdec->plane[0] + 2+min_y*h->fdec->i_stride[0], h->fdec->i_stride[0],
1029                 h->fenc->plane[0] + 2+min_y*h->fenc->i_stride[0], h->fenc->i_stride[0],
1030                 h->param.i_width-2, max_y-min_y );
1031     }
1032 }
1033
1034 static inline void x264_reference_update( x264_t *h )
1035 {
1036     int i;
1037
1038     if( h->fdec->i_frame >= 0 )
1039         h->i_frame++;
1040
1041     if( !h->fdec->b_kept_as_ref )
1042     {
1043         if( h->param.i_threads > 1 )
1044         {
1045             x264_frame_push_unused( h, h->fdec );
1046             h->fdec = x264_frame_pop_unused( h );
1047         }
1048         return;
1049     }
1050
1051     /* move lowres copy of the image to the ref frame */
1052     for( i = 0; i < 4; i++)
1053     {
1054         XCHG( uint8_t*, h->fdec->lowres[i], h->fenc->lowres[i] );
1055         XCHG( uint8_t*, h->fdec->buffer_lowres[i], h->fenc->buffer_lowres[i] );
1056     }
1057
1058     /* adaptive B decision needs a pointer, since it can't use the ref lists */
1059     if( h->sh.i_type != SLICE_TYPE_B )
1060         h->frames.last_nonb = h->fdec;
1061
1062     /* move frame in the buffer */
1063     x264_frame_push( h->frames.reference, h->fdec );
1064     if( h->frames.reference[h->frames.i_max_dpb] )
1065         x264_frame_push_unused( h, x264_frame_shift( h->frames.reference ) );
1066     h->fdec = x264_frame_pop_unused( h );
1067 }
1068
1069 static inline void x264_reference_reset( x264_t *h )
1070 {
1071     while( h->frames.reference[0] )
1072         x264_frame_push_unused( h, x264_frame_pop( h->frames.reference ) );
1073     h->fdec->i_poc =
1074     h->fenc->i_poc = 0;
1075 }
1076
1077 static inline void x264_slice_init( x264_t *h, int i_nal_type, int i_global_qp )
1078 {
1079     /* ------------------------ Create slice header  ----------------------- */
1080     if( i_nal_type == NAL_SLICE_IDR )
1081     {
1082         x264_slice_header_init( h, &h->sh, h->sps, h->pps, h->i_idr_pic_id, h->i_frame_num, i_global_qp );
1083
1084         /* increment id */
1085         h->i_idr_pic_id = ( h->i_idr_pic_id + 1 ) % 65536;
1086     }
1087     else
1088     {
1089         x264_slice_header_init( h, &h->sh, h->sps, h->pps, -1, h->i_frame_num, i_global_qp );
1090
1091         /* always set the real higher num of ref frame used */
1092         h->sh.b_num_ref_idx_override = 1;
1093         h->sh.i_num_ref_idx_l0_active = h->i_ref0 <= 0 ? 1 : h->i_ref0;
1094         h->sh.i_num_ref_idx_l1_active = h->i_ref1 <= 0 ? 1 : h->i_ref1;
1095     }
1096
1097     h->fdec->i_frame_num = h->sh.i_frame_num;
1098
1099     if( h->sps->i_poc_type == 0 )
1100     {
1101         h->sh.i_poc_lsb = h->fdec->i_poc & ( (1 << h->sps->i_log2_max_poc_lsb) - 1 );
1102         h->sh.i_delta_poc_bottom = 0;   /* XXX won't work for field */
1103     }
1104     else if( h->sps->i_poc_type == 1 )
1105     {
1106         /* FIXME TODO FIXME */
1107     }
1108     else
1109     {
1110         /* Nothing to do ? */
1111     }
1112
1113     x264_macroblock_slice_init( h );
1114 }
1115
1116 static void x264_slice_write( x264_t *h )
1117 {
1118     int i_skip;
1119     int mb_xy, i_mb_x, i_mb_y;
1120     int i, i_list, i_ref;
1121
1122     /* init stats */
1123     memset( &h->stat.frame, 0, sizeof(h->stat.frame) );
1124
1125     /* Slice */
1126     x264_nal_start( h, h->i_nal_type, h->i_nal_ref_idc );
1127
1128     /* Slice header */
1129     x264_slice_header_write( &h->out.bs, &h->sh, h->i_nal_ref_idc );
1130     if( h->param.b_cabac )
1131     {
1132         /* alignment needed */
1133         bs_align_1( &h->out.bs );
1134
1135         /* init cabac */
1136         x264_cabac_context_init( &h->cabac, h->sh.i_type, h->sh.i_qp, h->sh.i_cabac_init_idc );
1137         x264_cabac_encode_init ( &h->cabac, h->out.bs.p, h->out.bs.p_end );
1138     }
1139     h->mb.i_last_qp = h->sh.i_qp;
1140     h->mb.i_last_dqp = 0;
1141
1142     i_mb_y = h->sh.i_first_mb / h->sps->i_mb_width;
1143     i_mb_x = h->sh.i_first_mb % h->sps->i_mb_width;
1144     i_skip = 0;
1145
1146     while( (mb_xy = i_mb_x + i_mb_y * h->sps->i_mb_width) < h->sh.i_last_mb )
1147     {
1148         int mb_spos = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
1149
1150         if( i_mb_x == 0 )
1151             x264_fdec_filter_row( h, i_mb_y );
1152
1153         /* load cache */
1154         x264_macroblock_cache_load( h, i_mb_x, i_mb_y );
1155
1156         /* analyse parameters
1157          * Slice I: choose I_4x4 or I_16x16 mode
1158          * Slice P: choose between using P mode or intra (4x4 or 16x16)
1159          * */
1160         x264_macroblock_analyse( h );
1161
1162         /* encode this macroblock -> be careful it can change the mb type to P_SKIP if needed */
1163         x264_macroblock_encode( h );
1164
1165         x264_bitstream_check_buffer( h );
1166
1167         if( h->param.b_cabac )
1168         {
1169             if( mb_xy > h->sh.i_first_mb && !(h->sh.b_mbaff && (i_mb_y&1)) )
1170                 x264_cabac_encode_terminal( &h->cabac );
1171
1172             if( IS_SKIP( h->mb.i_type ) )
1173                 x264_cabac_mb_skip( h, 1 );
1174             else
1175             {
1176                 if( h->sh.i_type != SLICE_TYPE_I )
1177                     x264_cabac_mb_skip( h, 0 );
1178                 x264_macroblock_write_cabac( h, &h->cabac );
1179             }
1180         }
1181         else
1182         {
1183             if( IS_SKIP( h->mb.i_type ) )
1184                 i_skip++;
1185             else
1186             {
1187                 if( h->sh.i_type != SLICE_TYPE_I )
1188                 {
1189                     bs_write_ue( &h->out.bs, i_skip );  /* skip run */
1190                     i_skip = 0;
1191                 }
1192                 x264_macroblock_write_cavlc( h, &h->out.bs );
1193             }
1194         }
1195
1196 #if VISUALIZE
1197         if( h->param.b_visualize )
1198             x264_visualize_mb( h );
1199 #endif
1200
1201         /* save cache */
1202         x264_macroblock_cache_save( h );
1203
1204         /* accumulate mb stats */
1205         h->stat.frame.i_mb_count[h->mb.i_type]++;
1206         if( !IS_SKIP(h->mb.i_type) && !IS_INTRA(h->mb.i_type) && !IS_DIRECT(h->mb.i_type) )
1207         {
1208             if( h->mb.i_partition != D_8x8 )
1209                 h->stat.frame.i_mb_partition[h->mb.i_partition] += 4;
1210             else
1211                 for( i = 0; i < 4; i++ )
1212                     h->stat.frame.i_mb_partition[h->mb.i_sub_partition[i]] ++;
1213             if( h->param.i_frame_reference > 1 )
1214                 for( i_list = 0; i_list <= (h->sh.i_type == SLICE_TYPE_B); i_list++ )
1215                     for( i = 0; i < 4; i++ )
1216                     {
1217                         i_ref = h->mb.cache.ref[i_list][ x264_scan8[4*i] ];
1218                         if( i_ref >= 0 )
1219                             h->stat.frame.i_mb_count_ref[i_list][i_ref] ++;
1220                     }
1221         }
1222         if( h->mb.i_cbp_luma && !IS_INTRA(h->mb.i_type) )
1223         {
1224             h->stat.frame.i_mb_count_8x8dct[0] ++;
1225             h->stat.frame.i_mb_count_8x8dct[1] += h->mb.b_transform_8x8;
1226         }
1227
1228         x264_ratecontrol_mb( h, bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac) - mb_spos );
1229
1230         if( h->sh.b_mbaff )
1231         {
1232             i_mb_x += i_mb_y & 1;
1233             i_mb_y ^= i_mb_x < h->sps->i_mb_width;
1234         }
1235         else
1236             i_mb_x++;
1237         if(i_mb_x == h->sps->i_mb_width)
1238         {
1239             i_mb_y++;
1240             i_mb_x = 0;
1241         }
1242     }
1243
1244     if( h->param.b_cabac )
1245     {
1246         x264_cabac_encode_flush( h, &h->cabac );
1247         h->out.bs.p = h->cabac.p;
1248     }
1249     else
1250     {
1251         if( i_skip > 0 )
1252             bs_write_ue( &h->out.bs, i_skip );  /* last skip run */
1253         /* rbsp_slice_trailing_bits */
1254         bs_rbsp_trailing( &h->out.bs );
1255     }
1256
1257     x264_nal_end( h );
1258
1259     x264_fdec_filter_row( h, h->sps->i_mb_height );
1260
1261     /* Compute misc bits */
1262     h->stat.frame.i_misc_bits = bs_pos( &h->out.bs )
1263                               + NALU_OVERHEAD * 8
1264                               - h->stat.frame.i_tex_bits
1265                               - h->stat.frame.i_mv_bits;
1266 }
1267
1268 static void x264_thread_sync_context( x264_t *dst, x264_t *src )
1269 {
1270     x264_frame_t **f;
1271     if( dst == src )
1272         return;
1273
1274     // reference counting
1275     for( f = src->frames.reference; *f; f++ )
1276         (*f)->i_reference_count++;
1277     for( f = dst->frames.reference; *f; f++ )
1278         x264_frame_push_unused( src, *f );
1279     src->fdec->i_reference_count++;
1280     x264_frame_push_unused( src, dst->fdec );
1281
1282     // copy everything except the per-thread pointers and the constants.
1283     memcpy( &dst->i_frame, &src->i_frame, offsetof(x264_t, mb.type) - offsetof(x264_t, i_frame) );
1284     dst->stat = src->stat;
1285 }
1286
1287 static void x264_thread_sync_stat( x264_t *dst, x264_t *src )
1288 {
1289     if( dst == src )
1290         return;
1291     memcpy( &dst->stat.i_slice_count, &src->stat.i_slice_count, sizeof(dst->stat) - sizeof(dst->stat.frame) );
1292 }
1293
1294 static int x264_slices_write( x264_t *h )
1295 {
1296     int i_frame_size;
1297
1298 #if VISUALIZE
1299     if( h->param.b_visualize )
1300         x264_visualize_init( h );
1301 #endif
1302
1303     x264_stack_align( x264_slice_write, h );
1304     i_frame_size = h->out.nal[h->out.i_nal-1].i_payload;
1305
1306 #if VISUALIZE
1307     if( h->param.b_visualize )
1308     {
1309         x264_visualize_show( h );
1310         x264_visualize_close( h );
1311     }
1312 #endif
1313
1314     h->out.i_frame_size = i_frame_size;
1315     return 0;
1316 }
1317
1318 /****************************************************************************
1319  * x264_encoder_encode:
1320  *  XXX: i_poc   : is the poc of the current given picture
1321  *       i_frame : is the number of the frame being coded
1322  *  ex:  type frame poc
1323  *       I      0   2*0
1324  *       P      1   2*3
1325  *       B      2   2*1
1326  *       B      3   2*2
1327  *       P      4   2*6
1328  *       B      5   2*4
1329  *       B      6   2*5
1330  ****************************************************************************/
1331 int     x264_encoder_encode( x264_t *h,
1332                              x264_nal_t **pp_nal, int *pi_nal,
1333                              x264_picture_t *pic_in,
1334                              x264_picture_t *pic_out )
1335 {
1336     x264_t *thread_current, *thread_prev, *thread_oldest;
1337     int     i_nal_type;
1338     int     i_nal_ref_idc;
1339
1340     int   i_global_qp;
1341
1342     if( h->param.i_threads > 1)
1343     {
1344         int i = ++h->i_thread_phase;
1345         int t = h->param.i_threads;
1346         thread_current = h->thread[ i%t ];
1347         thread_prev    = h->thread[ (i-1)%t ];
1348         thread_oldest  = h->thread[ (i+1)%t ];
1349         x264_thread_sync_context( thread_current, thread_prev );
1350         x264_thread_sync_ratecontrol( thread_current, thread_prev, thread_oldest );
1351         h = thread_current;
1352 //      fprintf(stderr, "current: %p  prev: %p  oldest: %p \n", thread_current, thread_prev, thread_oldest);
1353     }
1354     else
1355     {
1356         thread_current =
1357         thread_prev    =
1358         thread_oldest  = h;
1359     }
1360
1361     // ok to call this before encoding any frames, since the initial values of fdec have b_kept_as_ref=0
1362     x264_reference_update( h );
1363     h->fdec->i_lines_completed = -1;
1364
1365     /* no data out */
1366     *pi_nal = 0;
1367     *pp_nal = NULL;
1368
1369     /* ------------------- Setup new frame from picture -------------------- */
1370     if( pic_in != NULL )
1371     {
1372         /* 1: Copy the picture to a frame and move it to a buffer */
1373         x264_frame_t *fenc = x264_frame_pop_unused( h );
1374
1375         if( x264_frame_copy_picture( h, fenc, pic_in ) < 0 )
1376             return -1;
1377
1378         if( h->param.i_width != 16 * h->sps->i_mb_width ||
1379             h->param.i_height != 16 * h->sps->i_mb_height )
1380             x264_frame_expand_border_mod16( h, fenc );
1381
1382         fenc->i_frame = h->frames.i_input++;
1383
1384         x264_frame_push( h->frames.next, fenc );
1385
1386         if( h->frames.b_have_lowres )
1387             x264_frame_init_lowres( h, fenc );
1388
1389         if( h->param.rc.i_aq_mode )
1390             x264_adaptive_quant_frame( h, fenc );
1391
1392         if( h->frames.i_input <= h->frames.i_delay + 1 - h->param.i_threads )
1393         {
1394             /* Nothing yet to encode */
1395             /* waiting for filling bframe buffer */
1396             pic_out->i_type = X264_TYPE_AUTO;
1397             return 0;
1398         }
1399     }
1400
1401     if( h->frames.current[0] == NULL )
1402     {
1403         int bframes = 0;
1404         /* 2: Select frame types */
1405         if( h->frames.next[0] == NULL )
1406         {
1407             x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
1408             return 0;
1409         }
1410
1411         x264_slicetype_decide( h );
1412
1413         /* 3: move some B-frames and 1 non-B to encode queue */
1414         while( IS_X264_TYPE_B( h->frames.next[bframes]->i_type ) )
1415             bframes++;
1416         x264_frame_push( h->frames.current, x264_frame_shift( &h->frames.next[bframes] ) );
1417         /* FIXME: when max B-frames > 3, BREF may no longer be centered after GOP closing */
1418         if( h->param.b_bframe_pyramid && bframes > 1 )
1419         {
1420             x264_frame_t *mid = x264_frame_shift( &h->frames.next[bframes/2] );
1421             mid->i_type = X264_TYPE_BREF;
1422             x264_frame_push( h->frames.current, mid );
1423             bframes--;
1424         }
1425         while( bframes-- )
1426             x264_frame_push( h->frames.current, x264_frame_shift( h->frames.next ) );
1427     }
1428
1429     /* ------------------- Get frame to be encoded ------------------------- */
1430     /* 4: get picture to encode */
1431     h->fenc = x264_frame_shift( h->frames.current );
1432     if( h->fenc == NULL )
1433     {
1434         /* Nothing yet to encode (ex: waiting for I/P with B frames) */
1435         /* waiting for filling bframe buffer */
1436         pic_out->i_type = X264_TYPE_AUTO;
1437         return 0;
1438     }
1439
1440 do_encode:
1441
1442     if( h->fenc->i_type == X264_TYPE_IDR )
1443     {
1444         h->frames.i_last_idr = h->fenc->i_frame;
1445     }
1446
1447     /* ------------------- Setup frame context ----------------------------- */
1448     /* 5: Init data dependent of frame type */
1449     if( h->fenc->i_type == X264_TYPE_IDR )
1450     {
1451         /* reset ref pictures */
1452         x264_reference_reset( h );
1453
1454         i_nal_type    = NAL_SLICE_IDR;
1455         i_nal_ref_idc = NAL_PRIORITY_HIGHEST;
1456         h->sh.i_type = SLICE_TYPE_I;
1457     }
1458     else if( h->fenc->i_type == X264_TYPE_I )
1459     {
1460         i_nal_type    = NAL_SLICE;
1461         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
1462         h->sh.i_type = SLICE_TYPE_I;
1463     }
1464     else if( h->fenc->i_type == X264_TYPE_P )
1465     {
1466         i_nal_type    = NAL_SLICE;
1467         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
1468         h->sh.i_type = SLICE_TYPE_P;
1469     }
1470     else if( h->fenc->i_type == X264_TYPE_BREF )
1471     {
1472         i_nal_type    = NAL_SLICE;
1473         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* maybe add MMCO to forget it? -> low */
1474         h->sh.i_type = SLICE_TYPE_B;
1475     }
1476     else    /* B frame */
1477     {
1478         i_nal_type    = NAL_SLICE;
1479         i_nal_ref_idc = NAL_PRIORITY_DISPOSABLE;
1480         h->sh.i_type = SLICE_TYPE_B;
1481     }
1482
1483     h->fdec->i_poc =
1484     h->fenc->i_poc = 2 * (h->fenc->i_frame - h->frames.i_last_idr);
1485     h->fdec->i_type = h->fenc->i_type;
1486     h->fdec->i_frame = h->fenc->i_frame;
1487     h->fenc->b_kept_as_ref =
1488     h->fdec->b_kept_as_ref = i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE && h->param.i_keyint_max > 1;
1489
1490
1491
1492     /* ------------------- Init                ----------------------------- */
1493     /* build ref list 0/1 */
1494     x264_reference_build_list( h, h->fdec->i_poc );
1495
1496     /* Init the rate control */
1497     x264_ratecontrol_start( h, h->fenc->i_qpplus1 );
1498     i_global_qp = x264_ratecontrol_qp( h );
1499
1500     pic_out->i_qpplus1 =
1501     h->fdec->i_qpplus1 = i_global_qp + 1;
1502
1503     if( h->sh.i_type == SLICE_TYPE_B )
1504         x264_macroblock_bipred_init( h );
1505
1506     /* ------------------------ Create slice header  ----------------------- */
1507     x264_slice_init( h, i_nal_type, i_global_qp );
1508
1509     if( i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE )
1510         h->i_frame_num++;
1511
1512     /* ---------------------- Write the bitstream -------------------------- */
1513     /* Init bitstream context */
1514     h->out.i_nal = 0;
1515     bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
1516
1517     if(h->param.b_aud){
1518         int pic_type;
1519
1520         if(h->sh.i_type == SLICE_TYPE_I)
1521             pic_type = 0;
1522         else if(h->sh.i_type == SLICE_TYPE_P)
1523             pic_type = 1;
1524         else if(h->sh.i_type == SLICE_TYPE_B)
1525             pic_type = 2;
1526         else
1527             pic_type = 7;
1528
1529         x264_nal_start(h, NAL_AUD, NAL_PRIORITY_DISPOSABLE);
1530         bs_write(&h->out.bs, 3, pic_type);
1531         bs_rbsp_trailing(&h->out.bs);
1532         x264_nal_end(h);
1533     }
1534
1535     h->i_nal_type = i_nal_type;
1536     h->i_nal_ref_idc = i_nal_ref_idc;
1537
1538     /* Write SPS and PPS */
1539     if( i_nal_type == NAL_SLICE_IDR && h->param.b_repeat_headers )
1540     {
1541         if( h->fenc->i_frame == 0 )
1542         {
1543             /* identify ourself */
1544             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
1545             x264_sei_version_write( h, &h->out.bs );
1546             x264_nal_end( h );
1547         }
1548
1549         /* generate sequence parameters */
1550         x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
1551         x264_sps_write( &h->out.bs, h->sps );
1552         x264_nal_end( h );
1553
1554         /* generate picture parameters */
1555         x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
1556         x264_pps_write( &h->out.bs, h->pps );
1557         x264_nal_end( h );
1558     }
1559
1560     /* Write frame */
1561     if( h->param.i_threads > 1 )
1562     {
1563         x264_pthread_create( &h->thread_handle, NULL, (void*)x264_slices_write, h );
1564         h->b_thread_active = 1;
1565     }
1566     else
1567         x264_slices_write( h );
1568
1569     /* restore CPU state (before using float again) */
1570     x264_emms();
1571
1572     if( h->sh.i_type == SLICE_TYPE_P && !h->param.rc.b_stat_read
1573         && h->param.i_scenecut_threshold >= 0
1574         && !h->param.b_pre_scenecut )
1575     {
1576         const int *mbs = h->stat.frame.i_mb_count;
1577         int i_mb_i = mbs[I_16x16] + mbs[I_8x8] + mbs[I_4x4];
1578         int i_mb_p = mbs[P_L0] + mbs[P_8x8];
1579         int i_mb_s = mbs[P_SKIP];
1580         int i_mb   = h->sps->i_mb_width * h->sps->i_mb_height;
1581         int64_t i_inter_cost = h->stat.frame.i_inter_cost;
1582         int64_t i_intra_cost = h->stat.frame.i_intra_cost;
1583
1584         float f_bias;
1585         int i_gop_size = h->fenc->i_frame - h->frames.i_last_idr;
1586         float f_thresh_max = h->param.i_scenecut_threshold / 100.0;
1587         /* magic numbers pulled out of thin air */
1588         float f_thresh_min = f_thresh_max * h->param.i_keyint_min
1589                              / ( h->param.i_keyint_max * 4 );
1590         if( h->param.i_keyint_min == h->param.i_keyint_max )
1591              f_thresh_min= f_thresh_max;
1592
1593         /* macroblock_analyse() doesn't further analyse skipped mbs,
1594          * so we have to guess their cost */
1595         if( h->stat.frame.i_mbs_analysed > 0 )
1596             i_intra_cost = i_intra_cost * i_mb / h->stat.frame.i_mbs_analysed;
1597
1598         if( i_gop_size < h->param.i_keyint_min / 4 )
1599             f_bias = f_thresh_min / 4;
1600         else if( i_gop_size <= h->param.i_keyint_min )
1601             f_bias = f_thresh_min * i_gop_size / h->param.i_keyint_min;
1602         else
1603         {
1604             f_bias = f_thresh_min
1605                      + ( f_thresh_max - f_thresh_min )
1606                        * ( i_gop_size - h->param.i_keyint_min )
1607                        / ( h->param.i_keyint_max - h->param.i_keyint_min );
1608         }
1609         f_bias = X264_MIN( f_bias, 1.0 );
1610
1611         /* Bad P will be reencoded as I */
1612         if( h->stat.frame.i_mbs_analysed > 0 &&
1613             i_inter_cost >= (1.0 - f_bias) * i_intra_cost )
1614         {
1615             int b;
1616
1617             x264_log( h, X264_LOG_DEBUG, "scene cut at %d Icost:%.0f Pcost:%.0f ratio:%.4f bias:%.4f gop:%d (imb:%d pmb:%d smb:%d)\n",
1618                       h->fenc->i_frame,
1619                       (double)i_intra_cost, (double)i_inter_cost,
1620                       1. - (double)i_inter_cost / i_intra_cost,
1621                       f_bias, i_gop_size,
1622                       i_mb_i, i_mb_p, i_mb_s );
1623
1624             /* Restore frame num */
1625             h->i_frame_num--;
1626
1627             for( b = 0; h->frames.current[b] && IS_X264_TYPE_B( h->frames.current[b]->i_type ); b++ );
1628             if( b > 0 )
1629             {
1630                 /* If using B-frames, force GOP to be closed.
1631                  * Even if this frame is going to be I and not IDR, forcing a
1632                  * P-frame before the scenecut will probably help compression.
1633                  *
1634                  * We don't yet know exactly which frame is the scene cut, so
1635                  * we can't assign an I-frame. Instead, change the previous
1636                  * B-frame to P, and rearrange coding order. */
1637
1638                 if( h->param.i_bframe_adaptive || b > 1 )
1639                     h->fenc->i_type = X264_TYPE_AUTO;
1640                 x264_frame_sort_pts( h->frames.current );
1641                 x264_frame_unshift( h->frames.next, h->fenc );
1642                 h->fenc = h->frames.current[b-1];
1643                 h->frames.current[b-1] = NULL;
1644                 h->fenc->i_type = X264_TYPE_P;
1645                 x264_frame_sort_dts( h->frames.current );
1646             }
1647             /* Do IDR if needed */
1648             else if( i_gop_size >= h->param.i_keyint_min )
1649             {
1650                 /* Reset */
1651                 h->i_frame_num = 0;
1652
1653                 /* Reinit field of fenc */
1654                 h->fenc->i_type = X264_TYPE_IDR;
1655                 h->fenc->i_poc = 0;
1656
1657                 /* Put enqueued frames back in the pool */
1658                 while( h->frames.current[0] )
1659                     x264_frame_push( h->frames.next, x264_frame_shift( h->frames.current ) );
1660                 x264_frame_sort_pts( h->frames.next );
1661             }
1662             else
1663             {
1664                 h->fenc->i_type = X264_TYPE_I;
1665             }
1666             goto do_encode;
1667         }
1668     }
1669
1670     x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
1671     return 0;
1672 }
1673
1674 static void x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
1675                                     x264_nal_t **pp_nal, int *pi_nal,
1676                                     x264_picture_t *pic_out )
1677 {
1678     int i, i_list;
1679     char psz_message[80];
1680
1681     if( h->b_thread_active )
1682     {
1683         x264_pthread_join( h->thread_handle, NULL );
1684         h->b_thread_active = 0;
1685     }
1686     if( !h->out.i_nal )
1687     {
1688         pic_out->i_type = X264_TYPE_AUTO;
1689         return;
1690     }
1691
1692     x264_frame_push_unused( thread_current, h->fenc );
1693
1694     /* End bitstream, set output  */
1695     *pi_nal = h->out.i_nal;
1696     *pp_nal = h->out.nal;
1697     h->out.i_nal = 0;
1698
1699     /* Set output picture properties */
1700     if( h->sh.i_type == SLICE_TYPE_I )
1701         pic_out->i_type = h->i_nal_type == NAL_SLICE_IDR ? X264_TYPE_IDR : X264_TYPE_I;
1702     else if( h->sh.i_type == SLICE_TYPE_P )
1703         pic_out->i_type = X264_TYPE_P;
1704     else
1705         pic_out->i_type = X264_TYPE_B;
1706     pic_out->i_pts = h->fenc->i_pts;
1707
1708     pic_out->img.i_plane = h->fdec->i_plane;
1709     for(i = 0; i < 4; i++){
1710         pic_out->img.i_stride[i] = h->fdec->i_stride[i];
1711         pic_out->img.plane[i] = h->fdec->plane[i];
1712     }
1713
1714     /* ---------------------- Update encoder state ------------------------- */
1715
1716     /* update rc */
1717     x264_emms();
1718     x264_ratecontrol_end( h, h->out.i_frame_size * 8 );
1719
1720     /* restore CPU state (before using float again) */
1721     x264_emms();
1722
1723     x264_noise_reduction_update( thread_current );
1724
1725     /* ---------------------- Compute/Print statistics --------------------- */
1726     x264_thread_sync_stat( h, h->thread[0] );
1727
1728     /* Slice stat */
1729     h->stat.i_slice_count[h->sh.i_type]++;
1730     h->stat.i_slice_size[h->sh.i_type] += h->out.i_frame_size + NALU_OVERHEAD;
1731     h->stat.f_slice_qp[h->sh.i_type] += h->fdec->f_qp_avg_aq;
1732
1733     for( i = 0; i < X264_MBTYPE_MAX; i++ )
1734         h->stat.i_mb_count[h->sh.i_type][i] += h->stat.frame.i_mb_count[i];
1735     for( i = 0; i < X264_PARTTYPE_MAX; i++ )
1736         h->stat.i_mb_partition[h->sh.i_type][i] += h->stat.frame.i_mb_partition[i];
1737     for( i = 0; i < 2; i++ )
1738         h->stat.i_mb_count_8x8dct[i] += h->stat.frame.i_mb_count_8x8dct[i];
1739     if( h->sh.i_type != SLICE_TYPE_I )
1740         for( i_list = 0; i_list < 2; i_list++ )
1741             for( i = 0; i < 32; i++ )
1742                 h->stat.i_mb_count_ref[h->sh.i_type][i_list][i] += h->stat.frame.i_mb_count_ref[i_list][i];
1743     if( h->sh.i_type == SLICE_TYPE_P )
1744         h->stat.i_consecutive_bframes[h->fdec->i_frame - h->fref0[0]->i_frame - 1]++;
1745     if( h->sh.i_type == SLICE_TYPE_B )
1746     {
1747         h->stat.i_direct_frames[ h->sh.b_direct_spatial_mv_pred ] ++;
1748         if( h->mb.b_direct_auto_write )
1749         {
1750             //FIXME somewhat arbitrary time constants
1751             if( h->stat.i_direct_score[0] + h->stat.i_direct_score[1] > h->mb.i_mb_count )
1752             {
1753                 for( i = 0; i < 2; i++ )
1754                     h->stat.i_direct_score[i] = h->stat.i_direct_score[i] * 9/10;
1755             }
1756             for( i = 0; i < 2; i++ )
1757                 h->stat.i_direct_score[i] += h->stat.frame.i_direct_score[i];
1758         }
1759     }
1760
1761     psz_message[0] = '\0';
1762     if( h->param.analyse.b_psnr )
1763     {
1764         int64_t ssd[3] = {
1765             h->stat.frame.i_ssd[0],
1766             h->stat.frame.i_ssd[1],
1767             h->stat.frame.i_ssd[2],
1768         };
1769
1770         h->stat.i_ssd_global[h->sh.i_type] += ssd[0] + ssd[1] + ssd[2];
1771         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 );
1772         h->stat.f_psnr_mean_y[h->sh.i_type] += x264_psnr( ssd[0], h->param.i_width * h->param.i_height );
1773         h->stat.f_psnr_mean_u[h->sh.i_type] += x264_psnr( ssd[1], h->param.i_width * h->param.i_height / 4 );
1774         h->stat.f_psnr_mean_v[h->sh.i_type] += x264_psnr( ssd[2], h->param.i_width * h->param.i_height / 4 );
1775
1776         snprintf( psz_message, 80, " PSNR Y:%5.2f U:%5.2f V:%5.2f",
1777                   x264_psnr( ssd[0], h->param.i_width * h->param.i_height ),
1778                   x264_psnr( ssd[1], h->param.i_width * h->param.i_height / 4),
1779                   x264_psnr( ssd[2], h->param.i_width * h->param.i_height / 4) );
1780     }
1781
1782     if( h->param.analyse.b_ssim )
1783     {
1784         double ssim_y = h->stat.frame.f_ssim
1785                       / (((h->param.i_width-6)>>2) * ((h->param.i_height-6)>>2));
1786         h->stat.f_ssim_mean_y[h->sh.i_type] += ssim_y;
1787         snprintf( psz_message + strlen(psz_message), 80 - strlen(psz_message),
1788                   " SSIM Y:%.5f", ssim_y );
1789     }
1790     psz_message[79] = '\0';
1791
1792     x264_log( h, X264_LOG_DEBUG,
1793                   "frame=%4d QP=%.2f NAL=%d Slice:%c Poc:%-3d I:%-4d P:%-4d SKIP:%-4d size=%d bytes%s\n",
1794               h->i_frame,
1795               h->fdec->f_qp_avg_aq,
1796               h->i_nal_ref_idc,
1797               h->sh.i_type == SLICE_TYPE_I ? 'I' : (h->sh.i_type == SLICE_TYPE_P ? 'P' : 'B' ),
1798               h->fdec->i_poc,
1799               h->stat.frame.i_mb_count_i,
1800               h->stat.frame.i_mb_count_p,
1801               h->stat.frame.i_mb_count_skip,
1802               h->out.i_frame_size,
1803               psz_message );
1804
1805     // keep stats all in one place
1806     x264_thread_sync_stat( h->thread[0], h );
1807     // for the use of the next frame
1808     x264_thread_sync_stat( thread_current, h );
1809
1810 #ifdef DEBUG_MB_TYPE
1811 {
1812     static const char mb_chars[] = { 'i', 'i', 'I', 'C', 'P', '8', 'S',
1813         'D', '<', 'X', 'B', 'X', '>', 'B', 'B', 'B', 'B', '8', 'S' };
1814     int mb_xy;
1815     for( mb_xy = 0; mb_xy < h->sps->i_mb_width * h->sps->i_mb_height; mb_xy++ )
1816     {
1817         if( h->mb.type[mb_xy] < X264_MBTYPE_MAX && h->mb.type[mb_xy] >= 0 )
1818             fprintf( stderr, "%c ", mb_chars[ h->mb.type[mb_xy] ] );
1819         else
1820             fprintf( stderr, "? " );
1821
1822         if( (mb_xy+1) % h->sps->i_mb_width == 0 )
1823             fprintf( stderr, "\n" );
1824     }
1825 }
1826 #endif
1827
1828     if( h->param.psz_dump_yuv )
1829         x264_frame_dump( h );
1830 }
1831
1832 static void x264_print_intra( int64_t *i_mb_count, double i_count, int b_print_pcm, char *intra )
1833 {
1834     intra += sprintf( intra, "I16..4%s: %4.1f%% %4.1f%% %4.1f%%",
1835         b_print_pcm ? "..PCM" : "",
1836         i_mb_count[I_16x16]/ i_count,
1837         i_mb_count[I_8x8]  / i_count,
1838         i_mb_count[I_4x4]  / i_count );
1839     if( b_print_pcm )
1840         sprintf( intra, " %4.1f%%", i_mb_count[I_PCM]  / i_count );
1841 }
1842
1843 /****************************************************************************
1844  * x264_encoder_close:
1845  ****************************************************************************/
1846 void    x264_encoder_close  ( x264_t *h )
1847 {
1848     int64_t i_yuv_size = 3 * h->param.i_width * h->param.i_height / 2;
1849     int64_t i_mb_count_size[2][7] = {{0}};
1850     char buf[200];
1851     int i, j, i_list, i_type;
1852     int b_print_pcm = h->stat.i_mb_count[SLICE_TYPE_I][I_PCM]
1853                    || h->stat.i_mb_count[SLICE_TYPE_P][I_PCM]
1854                    || h->stat.i_mb_count[SLICE_TYPE_B][I_PCM];
1855
1856     for( i=0; i<h->param.i_threads; i++ )
1857     {
1858         // don't strictly have to wait for the other threads, but it's simpler than canceling them
1859         if( h->thread[i]->b_thread_active )
1860         {
1861             x264_pthread_join( h->thread[i]->thread_handle, NULL );
1862             assert( h->thread[i]->fenc->i_reference_count == 1 );
1863             x264_frame_delete( h->thread[i]->fenc );
1864         }
1865     }
1866
1867     /* Slices used and PSNR */
1868     for( i=0; i<5; i++ )
1869     {
1870         static const int slice_order[] = { SLICE_TYPE_I, SLICE_TYPE_SI, SLICE_TYPE_P, SLICE_TYPE_SP, SLICE_TYPE_B };
1871         static const char *slice_name[] = { "P", "B", "I", "SP", "SI" };
1872         int i_slice = slice_order[i];
1873
1874         if( h->stat.i_slice_count[i_slice] > 0 )
1875         {
1876             const int i_count = h->stat.i_slice_count[i_slice];
1877             if( h->param.analyse.b_psnr )
1878             {
1879                 x264_log( h, X264_LOG_INFO,
1880                           "slice %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",
1881                           slice_name[i_slice],
1882                           i_count,
1883                           h->stat.f_slice_qp[i_slice] / i_count,
1884                           (double)h->stat.i_slice_size[i_slice] / i_count,
1885                           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,
1886                           h->stat.f_psnr_average[i_slice] / i_count,
1887                           x264_psnr( h->stat.i_ssd_global[i_slice], i_count * i_yuv_size ) );
1888             }
1889             else
1890             {
1891                 x264_log( h, X264_LOG_INFO,
1892                           "slice %s:%-5d Avg QP:%5.2f  size:%6.0f\n",
1893                           slice_name[i_slice],
1894                           i_count,
1895                           h->stat.f_slice_qp[i_slice] / i_count,
1896                           (double)h->stat.i_slice_size[i_slice] / i_count );
1897             }
1898         }
1899     }
1900     if( h->param.i_bframe && h->stat.i_slice_count[SLICE_TYPE_P] )
1901     {
1902         char *p = buf;
1903         int den = 0;
1904         // weight by number of frames (including the P-frame) that are in a sequence of N B-frames
1905         for( i=0; i<=h->param.i_bframe; i++ )
1906             den += (i+1) * h->stat.i_consecutive_bframes[i];
1907         for( i=0; i<=h->param.i_bframe; i++ )
1908             p += sprintf( p, " %4.1f%%", 100. * (i+1) * h->stat.i_consecutive_bframes[i] / den );
1909         x264_log( h, X264_LOG_INFO, "consecutive B-frames:%s\n", buf );
1910     }
1911
1912     for( i_type = 0; i_type < 2; i_type++ )
1913         for( i = 0; i < X264_PARTTYPE_MAX; i++ )
1914         {
1915             if( i == D_DIRECT_8x8 ) continue; /* direct is counted as its own type */
1916             i_mb_count_size[i_type][x264_mb_partition_pixel_table[i]] += h->stat.i_mb_partition[i_type][i];
1917         }
1918
1919     /* MB types used */
1920     if( h->stat.i_slice_count[SLICE_TYPE_I] > 0 )
1921     {
1922         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_I];
1923         double i_count = h->stat.i_slice_count[SLICE_TYPE_I] * h->mb.i_mb_count / 100.0;
1924         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
1925         x264_log( h, X264_LOG_INFO, "mb I  %s\n", buf );
1926     }
1927     if( h->stat.i_slice_count[SLICE_TYPE_P] > 0 )
1928     {
1929         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_P];
1930         double i_count = h->stat.i_slice_count[SLICE_TYPE_P] * h->mb.i_mb_count / 100.0;
1931         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_P];
1932         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
1933         x264_log( h, X264_LOG_INFO,
1934                   "mb P  %s  P16..4: %4.1f%% %4.1f%% %4.1f%% %4.1f%% %4.1f%%    skip:%4.1f%%\n",
1935                   buf,
1936                   i_mb_size[PIXEL_16x16] / (i_count*4),
1937                   (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
1938                   i_mb_size[PIXEL_8x8] / (i_count*4),
1939                   (i_mb_size[PIXEL_8x4] + i_mb_size[PIXEL_4x8]) / (i_count*4),
1940                   i_mb_size[PIXEL_4x4] / (i_count*4),
1941                   i_mb_count[P_SKIP] / i_count );
1942     }
1943     if( h->stat.i_slice_count[SLICE_TYPE_B] > 0 )
1944     {
1945         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_B];
1946         double i_count = h->stat.i_slice_count[SLICE_TYPE_B] * h->mb.i_mb_count / 100.0;
1947         double i_mb_list_count;
1948         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_B];
1949         int64_t list_count[3] = {0}; /* 0 == L0, 1 == L1, 2 == BI */
1950         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
1951         for( i = 0; i < X264_PARTTYPE_MAX; i++ )
1952             for( j = 0; j < 2; j++ )
1953             {
1954                 int l0 = x264_mb_type_list0_table[i][j];
1955                 int l1 = x264_mb_type_list1_table[i][j];
1956                 if( l0 || l1 )
1957                     list_count[l1+l0*l1] += h->stat.i_mb_count[SLICE_TYPE_B][i] * 2;
1958             }
1959         list_count[0] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L0_8x8];
1960         list_count[1] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L1_8x8];
1961         list_count[2] += h->stat.i_mb_partition[SLICE_TYPE_B][D_BI_8x8];
1962         i_mb_count[B_DIRECT] += (h->stat.i_mb_partition[SLICE_TYPE_B][D_DIRECT_8x8]+2)/4;
1963         i_mb_list_count = (list_count[0] + list_count[1] + list_count[2]) / 100.0;
1964         x264_log( h, X264_LOG_INFO,
1965                   "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",
1966                   buf,
1967                   i_mb_size[PIXEL_16x16] / (i_count*4),
1968                   (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
1969                   i_mb_size[PIXEL_8x8] / (i_count*4),
1970                   i_mb_count[B_DIRECT] / i_count,
1971                   i_mb_count[B_SKIP]   / i_count,
1972                   list_count[0] / i_mb_list_count,
1973                   list_count[1] / i_mb_list_count,
1974                   list_count[2] / i_mb_list_count );
1975     }
1976
1977     x264_ratecontrol_summary( h );
1978
1979     if( h->stat.i_slice_count[SLICE_TYPE_I] + h->stat.i_slice_count[SLICE_TYPE_P] + h->stat.i_slice_count[SLICE_TYPE_B] > 0 )
1980     {
1981         const int i_count = h->stat.i_slice_count[SLICE_TYPE_I] +
1982                             h->stat.i_slice_count[SLICE_TYPE_P] +
1983                             h->stat.i_slice_count[SLICE_TYPE_B];
1984         float fps = (float) h->param.i_fps_num / h->param.i_fps_den;
1985 #define SUM3(p) (p[SLICE_TYPE_I] + p[SLICE_TYPE_P] + p[SLICE_TYPE_B])
1986 #define SUM3b(p,o) (p[SLICE_TYPE_I][o] + p[SLICE_TYPE_P][o] + p[SLICE_TYPE_B][o])
1987         float f_bitrate = fps * SUM3(h->stat.i_slice_size) / i_count / 125;
1988
1989         if( h->pps->b_transform_8x8_mode )
1990         {
1991             int64_t i_i8x8 = SUM3b( h->stat.i_mb_count, I_8x8 );
1992             int64_t i_intra = i_i8x8 + SUM3b( h->stat.i_mb_count, I_4x4 )
1993                                      + SUM3b( h->stat.i_mb_count, I_16x16 );
1994             x264_log( h, X264_LOG_INFO, "8x8 transform  intra:%.1f%%  inter:%.1f%%\n",
1995                       100. * i_i8x8 / i_intra,
1996                       100. * h->stat.i_mb_count_8x8dct[1] / h->stat.i_mb_count_8x8dct[0] );
1997         }
1998
1999         if( h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO
2000             && h->stat.i_slice_count[SLICE_TYPE_B] )
2001         {
2002             x264_log( h, X264_LOG_INFO, "direct mvs  spatial:%.1f%%  temporal:%.1f%%\n",
2003                       h->stat.i_direct_frames[1] * 100. / h->stat.i_slice_count[SLICE_TYPE_B],
2004                       h->stat.i_direct_frames[0] * 100. / h->stat.i_slice_count[SLICE_TYPE_B] );
2005         }
2006
2007         for( i_list = 0; i_list < 2; i_list++ )
2008         {
2009             int i_slice;
2010             for( i_slice = 0; i_slice < 2; i_slice++ )
2011             {
2012                 char *p = buf;
2013                 int64_t i_den = 0;
2014                 int i_max = 0;
2015                 for( i = 0; i < 32; i++ )
2016                     if( h->stat.i_mb_count_ref[i_slice][i_list][i] )
2017                     {
2018                         i_den += h->stat.i_mb_count_ref[i_slice][i_list][i];
2019                         i_max = i;
2020                     }
2021                 if( i_max == 0 )
2022                     continue;
2023                 for( i = 0; i <= i_max; i++ )
2024                     p += sprintf( p, " %4.1f%%", 100. * h->stat.i_mb_count_ref[i_slice][i_list][i] / i_den );
2025                 x264_log( h, X264_LOG_INFO, "ref %c L%d %s\n", "PB"[i_slice], i_list, buf );
2026             }
2027         }
2028
2029         if( h->param.analyse.b_ssim )
2030         {
2031             x264_log( h, X264_LOG_INFO,
2032                       "SSIM Mean Y:%.7f\n",
2033                       SUM3( h->stat.f_ssim_mean_y ) / i_count );
2034         }
2035         if( h->param.analyse.b_psnr )
2036         {
2037             x264_log( h, X264_LOG_INFO,
2038                       "PSNR Mean Y:%6.3f U:%6.3f V:%6.3f Avg:%6.3f Global:%6.3f kb/s:%.2f\n",
2039                       SUM3( h->stat.f_psnr_mean_y ) / i_count,
2040                       SUM3( h->stat.f_psnr_mean_u ) / i_count,
2041                       SUM3( h->stat.f_psnr_mean_v ) / i_count,
2042                       SUM3( h->stat.f_psnr_average ) / i_count,
2043                       x264_psnr( SUM3( h->stat.i_ssd_global ), i_count * i_yuv_size ),
2044                       f_bitrate );
2045         }
2046         else
2047             x264_log( h, X264_LOG_INFO, "kb/s:%.1f\n", f_bitrate );
2048     }
2049
2050     /* rc */
2051     x264_ratecontrol_delete( h );
2052
2053     /* param */
2054     if( h->param.rc.psz_stat_out )
2055         free( h->param.rc.psz_stat_out );
2056     if( h->param.rc.psz_stat_in )
2057         free( h->param.rc.psz_stat_in );
2058
2059     x264_cqm_delete( h );
2060
2061     if( h->param.i_threads > 1)
2062         h = h->thread[ h->i_thread_phase % h->param.i_threads ];
2063
2064     /* frames */
2065     for( i = 0; h->frames.current[i]; i++ )
2066     {
2067         assert( h->frames.current[i]->i_reference_count == 1 );
2068         x264_frame_delete( h->frames.current[i] );
2069     }
2070     for( i = 0; h->frames.next[i]; i++ )
2071     {
2072         assert( h->frames.next[i]->i_reference_count == 1 );
2073         x264_frame_delete( h->frames.next[i] );
2074     }
2075     for( i = 0; h->frames.unused[i]; i++ )
2076     {
2077         assert( h->frames.unused[i]->i_reference_count == 0 );
2078         x264_frame_delete( h->frames.unused[i] );
2079     }
2080
2081     h = h->thread[0];
2082
2083     for( i = h->param.i_threads - 1; i >= 0; i-- )
2084     {
2085         x264_frame_t **frame;
2086
2087         for( frame = h->thread[i]->frames.reference; *frame; frame++ )
2088         {
2089             assert( (*frame)->i_reference_count > 0 );
2090             (*frame)->i_reference_count--;
2091             if( (*frame)->i_reference_count == 0 )
2092                 x264_frame_delete( *frame );
2093         }
2094         frame = &h->thread[i]->fdec;
2095         assert( (*frame)->i_reference_count > 0 );
2096         (*frame)->i_reference_count--;
2097         if( (*frame)->i_reference_count == 0 )
2098             x264_frame_delete( *frame );
2099
2100         x264_macroblock_cache_end( h->thread[i] );
2101         x264_free( h->thread[i]->out.p_bitstream );
2102         x264_free( h->thread[i] );
2103     }
2104 }