]> git.sesse.net Git - x264/blob - encoder/encoder.c
bb6d8535a2681500a3a01c18d7e506a832ab9649
[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  *
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_MAX(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_transform_8x8 = 0;
407         h->param.analyse.b_psnr = 0;
408         h->param.analyse.b_ssim = 0;
409         h->param.analyse.i_chroma_qp_offset = 0;
410         h->param.analyse.i_trellis = 0;
411         h->param.analyse.b_fast_pskip = 0;
412         h->param.analyse.i_noise_reduction = 0;
413         h->param.analyse.i_subpel_refine = x264_clip3( h->param.analyse.i_subpel_refine, 1, 6 );
414     }
415     if( h->param.rc.i_rc_method == X264_RC_CQP )
416     {
417         float qp_p = h->param.rc.i_qp_constant;
418         float qp_i = qp_p - 6*log(h->param.rc.f_ip_factor)/log(2);
419         float qp_b = qp_p + 6*log(h->param.rc.f_pb_factor)/log(2);
420         h->param.rc.i_qp_min = x264_clip3( (int)(X264_MIN3( qp_p, qp_i, qp_b )), 0, 51 );
421         h->param.rc.i_qp_max = x264_clip3( (int)(X264_MAX3( qp_p, qp_i, qp_b ) + .999), 0, 51 );
422         h->param.rc.i_aq_mode = 0;
423     }
424
425     if( ( h->param.i_width % 16 || h->param.i_height % 16 )
426         && h->param.i_height != 1080 && !h->mb.b_lossless )
427     {
428         // There's nothing special about 1080 in that the warning still applies to it,
429         // but chances are the user can't help it if his content is already 1080p,
430         // so there's no point in warning in that case.
431         x264_log( h, X264_LOG_WARNING, 
432                   "width or height not divisible by 16 (%dx%d), compression will suffer.\n",
433                   h->param.i_width, h->param.i_height );
434     }
435
436     h->param.i_frame_reference = x264_clip3( h->param.i_frame_reference, 1, 16 );
437     if( h->param.i_keyint_max <= 0 )
438         h->param.i_keyint_max = 1;
439     h->param.i_keyint_min = x264_clip3( h->param.i_keyint_min, 1, h->param.i_keyint_max/2+1 );
440
441     h->param.i_bframe = x264_clip3( h->param.i_bframe, 0, X264_BFRAME_MAX );
442     h->param.i_bframe_bias = x264_clip3( h->param.i_bframe_bias, -90, 100 );
443     h->param.b_bframe_pyramid = h->param.b_bframe_pyramid && h->param.i_bframe > 1;
444     h->param.b_bframe_adaptive = h->param.b_bframe_adaptive && h->param.i_bframe > 0;
445     h->param.analyse.b_weighted_bipred = h->param.analyse.b_weighted_bipred && h->param.i_bframe > 0;
446     h->mb.b_direct_auto_write = h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO
447                                 && h->param.i_bframe
448                                 && ( h->param.rc.b_stat_write || !h->param.rc.b_stat_read );
449     if( h->param.i_scenecut_threshold < 0 )
450         h->param.b_pre_scenecut = 0;
451
452     h->param.i_deblocking_filter_alphac0 = x264_clip3( h->param.i_deblocking_filter_alphac0, -6, 6 );
453     h->param.i_deblocking_filter_beta    = x264_clip3( h->param.i_deblocking_filter_beta, -6, 6 );
454     h->param.analyse.i_luma_deadzone[0] = x264_clip3( h->param.analyse.i_luma_deadzone[0], 0, 32 );
455     h->param.analyse.i_luma_deadzone[1] = x264_clip3( h->param.analyse.i_luma_deadzone[1], 0, 32 );
456
457     h->param.i_cabac_init_idc = x264_clip3( h->param.i_cabac_init_idc, 0, 2 );
458
459     if( h->param.i_cqm_preset < X264_CQM_FLAT || h->param.i_cqm_preset > X264_CQM_CUSTOM )
460         h->param.i_cqm_preset = X264_CQM_FLAT;
461
462     if( h->param.analyse.i_me_method < X264_ME_DIA ||
463         h->param.analyse.i_me_method > X264_ME_TESA )
464         h->param.analyse.i_me_method = X264_ME_HEX;
465     if( h->param.analyse.i_me_range < 4 )
466         h->param.analyse.i_me_range = 4;
467     if( h->param.analyse.i_me_range > 16 && h->param.analyse.i_me_method <= X264_ME_HEX )
468         h->param.analyse.i_me_range = 16;
469     if( h->param.analyse.i_me_method == X264_ME_TESA &&
470         (h->mb.b_lossless || h->param.analyse.i_subpel_refine <= 1) )
471         h->param.analyse.i_me_method = X264_ME_ESA;
472     h->param.analyse.i_subpel_refine = x264_clip3( h->param.analyse.i_subpel_refine, 1, 7 );
473     h->param.analyse.b_bframe_rdo = h->param.analyse.b_bframe_rdo && h->param.analyse.i_subpel_refine >= 6;
474     h->param.analyse.b_mixed_references = h->param.analyse.b_mixed_references && h->param.i_frame_reference > 1;
475     h->param.analyse.inter &= X264_ANALYSE_PSUB16x16|X264_ANALYSE_PSUB8x8|X264_ANALYSE_BSUB16x16|
476                               X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
477     h->param.analyse.intra &= X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
478     if( !(h->param.analyse.inter & X264_ANALYSE_PSUB16x16) )
479         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
480     if( !h->param.analyse.b_transform_8x8 )
481     {
482         h->param.analyse.inter &= ~X264_ANALYSE_I8x8;
483         h->param.analyse.intra &= ~X264_ANALYSE_I8x8;
484     }
485     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -12, 12);
486     if( !h->param.b_cabac )
487         h->param.analyse.i_trellis = 0;
488     h->param.analyse.i_trellis = x264_clip3( h->param.analyse.i_trellis, 0, 2 );
489     h->param.rc.i_aq_mode = x264_clip3( h->param.rc.i_aq_mode, 0, 2 );
490     if( h->param.rc.f_aq_strength <= 0 )
491         h->param.rc.i_aq_mode = 0;
492     /* VAQ effectively replaces qcomp, so qcomp is raised towards 1 to compensate. */
493     if( h->param.rc.i_aq_mode == X264_AQ_GLOBAL )
494         h->param.rc.f_qcompress = x264_clip3f(h->param.rc.f_qcompress + h->param.rc.f_aq_strength / 0.7, 0, 1);
495     h->param.analyse.i_noise_reduction = x264_clip3( h->param.analyse.i_noise_reduction, 0, 1<<16 );
496
497     {
498         const x264_level_t *l = x264_levels;
499         if( h->param.i_level_idc < 0 )
500         {
501             if( h->param.rc.i_rc_method == X264_RC_ABR && h->param.rc.i_vbv_buffer_size <= 0 )
502                 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate * 2;
503             h->sps = h->sps_array;
504             x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
505             do h->param.i_level_idc = l->level_idc;
506                 while( l[1].level_idc && x264_validate_levels( h, 0 ) && l++ );
507             if( h->param.rc.i_vbv_buffer_size <= 0 )
508                 h->param.rc.i_vbv_max_bitrate = 0;
509             x264_log( h, X264_LOG_DEBUG, "level_idc: %d\n", h->param.i_level_idc );
510         }
511         else
512         {
513             while( l->level_idc && l->level_idc != h->param.i_level_idc )
514                 l++;
515             if( l->level_idc == 0 )
516             {
517                 x264_log( h, X264_LOG_ERROR, "invalid level_idc: %d\n", h->param.i_level_idc );
518                 return -1;
519             }
520         }
521         if( h->param.analyse.i_mv_range <= 0 )
522             h->param.analyse.i_mv_range = l->mv_range >> h->param.b_interlaced;
523         else
524             h->param.analyse.i_mv_range = x264_clip3(h->param.analyse.i_mv_range, 32, 512 >> h->param.b_interlaced);
525         if( h->param.analyse.i_direct_8x8_inference < 0 )
526             h->param.analyse.i_direct_8x8_inference = l->direct8x8;
527     }
528
529     if( h->param.i_threads > 1 )
530     {
531         int r = h->param.analyse.i_mv_range_thread;
532         int r2;
533         if( r <= 0 )
534         {
535             // half of the available space is reserved and divided evenly among the threads,
536             // the rest is allocated to whichever thread is far enough ahead to use it.
537             // reserving more space increases quality for some videos, but costs more time
538             // in thread synchronization.
539             int max_range = (h->param.i_height + X264_THREAD_HEIGHT) / h->param.i_threads - X264_THREAD_HEIGHT;
540             r = max_range / 2;
541         }
542         r = X264_MAX( r, h->param.analyse.i_me_range );
543         r = X264_MIN( r, h->param.analyse.i_mv_range );
544         // round up to use the whole mb row
545         r2 = (r & ~15) + ((-X264_THREAD_HEIGHT) & 15);
546         if( r2 < r )
547             r2 += 16;
548         x264_log( h, X264_LOG_DEBUG, "using mv_range_thread = %d\n", r2 );
549         h->param.analyse.i_mv_range_thread = r2;
550     }
551
552     if( h->param.rc.f_qblur < 0 )
553         h->param.rc.f_qblur = 0;
554     if( h->param.rc.f_complexity_blur < 0 )
555         h->param.rc.f_complexity_blur = 0;
556
557     h->param.i_sps_id &= 31;
558
559     if( h->param.i_log_level < X264_LOG_INFO )
560     {
561         h->param.analyse.b_psnr = 0;
562         h->param.analyse.b_ssim = 0;
563     }
564
565     /* ensure the booleans are 0 or 1 so they can be used in math */
566 #define BOOLIFY(x) h->param.x = !!h->param.x
567     BOOLIFY( b_cabac );
568     BOOLIFY( b_deblocking_filter );
569     BOOLIFY( b_interlaced );
570     BOOLIFY( analyse.b_transform_8x8 );
571     BOOLIFY( analyse.i_direct_8x8_inference );
572     BOOLIFY( analyse.b_bidir_me );
573     BOOLIFY( analyse.b_chroma_me );
574     BOOLIFY( analyse.b_fast_pskip );
575     BOOLIFY( rc.b_stat_write );
576     BOOLIFY( rc.b_stat_read );
577 #undef BOOLIFY
578
579     return 0;
580 }
581
582 static void mbcmp_init( x264_t *h )
583 {
584     int satd = !h->mb.b_lossless && h->param.analyse.i_subpel_refine > 1;
585     memcpy( h->pixf.mbcmp, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.mbcmp) );
586     satd &= h->param.analyse.i_me_method == X264_ME_TESA;
587     memcpy( h->pixf.fpelcmp, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.fpelcmp) );
588     memcpy( h->pixf.fpelcmp_x3, satd ? h->pixf.satd_x3 : h->pixf.sad_x3, sizeof(h->pixf.fpelcmp_x3) );
589     memcpy( h->pixf.fpelcmp_x4, satd ? h->pixf.satd_x4 : h->pixf.sad_x4, sizeof(h->pixf.fpelcmp_x4) );
590 }
591
592 /****************************************************************************
593  * x264_encoder_open:
594  ****************************************************************************/
595 x264_t *x264_encoder_open   ( x264_param_t *param )
596 {
597     x264_t *h = x264_malloc( sizeof( x264_t ) );
598     char buf[1000], *p;
599     int i;
600
601     memset( h, 0, sizeof( x264_t ) );
602
603     /* Create a copy of param */
604     memcpy( &h->param, param, sizeof( x264_param_t ) );
605
606     if( x264_validate_parameters( h ) < 0 )
607     {
608         x264_free( h );
609         return NULL;
610     }
611
612     if( h->param.psz_cqm_file )
613         if( x264_cqm_parse_file( h, h->param.psz_cqm_file ) < 0 )
614         {
615             x264_free( h );
616             return NULL;
617         }
618
619     if( h->param.rc.psz_stat_out )
620         h->param.rc.psz_stat_out = strdup( h->param.rc.psz_stat_out );
621     if( h->param.rc.psz_stat_in )
622         h->param.rc.psz_stat_in = strdup( h->param.rc.psz_stat_in );
623     if( h->param.rc.psz_rc_eq )
624         h->param.rc.psz_rc_eq = strdup( h->param.rc.psz_rc_eq );
625
626     /* VUI */
627     if( h->param.vui.i_sar_width > 0 && h->param.vui.i_sar_height > 0 )
628     {
629         int i_w = param->vui.i_sar_width;
630         int i_h = param->vui.i_sar_height;
631
632         x264_reduce_fraction( &i_w, &i_h );
633
634         while( i_w > 65535 || i_h > 65535 )
635         {
636             i_w /= 2;
637             i_h /= 2;
638         }
639
640         h->param.vui.i_sar_width = 0;
641         h->param.vui.i_sar_height = 0;
642         if( i_w == 0 || i_h == 0 )
643         {
644             x264_log( h, X264_LOG_WARNING, "cannot create valid sample aspect ratio\n" );
645         }
646         else
647         {
648             x264_log( h, X264_LOG_INFO, "using SAR=%d/%d\n", i_w, i_h );
649             h->param.vui.i_sar_width = i_w;
650             h->param.vui.i_sar_height = i_h;
651         }
652     }
653
654     x264_reduce_fraction( &h->param.i_fps_num, &h->param.i_fps_den );
655
656     /* Init x264_t */
657     h->i_frame = 0;
658     h->i_frame_num = 0;
659     h->i_idr_pic_id = 0;
660
661     h->sps = &h->sps_array[0];
662     x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
663
664     h->pps = &h->pps_array[0];
665     x264_pps_init( h->pps, h->param.i_sps_id, &h->param, h->sps);
666
667     x264_validate_levels( h, 1 );
668
669     if( x264_cqm_init( h ) < 0 )
670     {
671         x264_free( h );
672         return NULL;
673     }
674     
675     h->mb.i_mb_count = h->sps->i_mb_width * h->sps->i_mb_height;
676
677     /* Init frames. */
678     h->frames.i_delay = h->param.i_bframe + h->param.i_threads - 1;
679     h->frames.i_max_ref0 = h->param.i_frame_reference;
680     h->frames.i_max_ref1 = h->sps->vui.i_num_reorder_frames;
681     h->frames.i_max_dpb  = h->sps->vui.i_max_dec_frame_buffering;
682     h->frames.b_have_lowres = !h->param.rc.b_stat_read
683         && ( h->param.rc.i_rc_method == X264_RC_ABR
684           || h->param.rc.i_rc_method == X264_RC_CRF
685           || h->param.b_bframe_adaptive
686           || h->param.b_pre_scenecut );
687     h->frames.b_have_lowres |= (h->param.rc.b_stat_read && h->param.rc.i_vbv_buffer_size > 0);
688
689     h->frames.i_last_idr = - h->param.i_keyint_max;
690     h->frames.i_input    = 0;
691     h->frames.last_nonb  = NULL;
692
693     h->i_ref0 = 0;
694     h->i_ref1 = 0;
695
696     x264_rdo_init( );
697
698     /* init CPU functions */
699     x264_predict_16x16_init( h->param.cpu, h->predict_16x16 );
700     x264_predict_8x8c_init( h->param.cpu, h->predict_8x8c );
701     x264_predict_8x8_init( h->param.cpu, h->predict_8x8 );
702     x264_predict_4x4_init( h->param.cpu, h->predict_4x4 );
703
704     x264_pixel_init( h->param.cpu, &h->pixf );
705     x264_dct_init( h->param.cpu, &h->dctf );
706     x264_zigzag_init( h->param.cpu, &h->zigzagf, h->param.b_interlaced );
707     x264_mc_init( h->param.cpu, &h->mc );
708     x264_quant_init( h, h->param.cpu, &h->quantf );
709     x264_deblock_init( h->param.cpu, &h->loopf );
710     x264_dct_init_weights();
711
712     mbcmp_init( h );
713
714     p = buf + sprintf( buf, "using cpu capabilities:" );
715     for( i=0; x264_cpu_names[i].flags; i++ )
716     {
717         if( !strcmp(x264_cpu_names[i].name, "SSE2")
718             && param->cpu & (X264_CPU_SSE2_IS_FAST|X264_CPU_SSE2_IS_SLOW) )
719             continue;
720         if( !strcmp(x264_cpu_names[i].name, "SSE3")
721             && (param->cpu & X264_CPU_SSSE3 || !(param->cpu & X264_CPU_CACHELINE_64)) )
722             continue;
723         if( (param->cpu & x264_cpu_names[i].flags) == x264_cpu_names[i].flags
724             && (!i || x264_cpu_names[i].flags != x264_cpu_names[i-1].flags) )
725             p += sprintf( p, " %s", x264_cpu_names[i].name );
726     }
727     if( !param->cpu )
728         p += sprintf( p, " none!" );
729     x264_log( h, X264_LOG_INFO, "%s\n", buf );
730
731     h->out.i_nal = 0;
732     h->out.i_bitstream = X264_MAX( 1000000, h->param.i_width * h->param.i_height * 4
733         * ( h->param.rc.i_rc_method == X264_RC_ABR ? pow( 0.95, h->param.rc.i_qp_min )
734           : pow( 0.95, h->param.rc.i_qp_constant ) * X264_MAX( 1, h->param.rc.f_ip_factor )));
735
736     h->thread[0] = h;
737     h->i_thread_num = 0;
738     for( i = 1; i < h->param.i_threads; i++ )
739         h->thread[i] = x264_malloc( sizeof(x264_t) );
740
741     for( i = 0; i < h->param.i_threads; i++ )
742     {
743         if( i > 0 )
744             *h->thread[i] = *h;
745         h->thread[i]->fdec = x264_frame_pop_unused( h );
746         h->thread[i]->out.p_bitstream = x264_malloc( h->out.i_bitstream );
747         if( x264_macroblock_cache_init( h->thread[i] ) < 0 )
748             return NULL;
749     }
750
751     if( x264_ratecontrol_new( h ) < 0 )
752         return NULL;
753
754     if( h->param.psz_dump_yuv )
755     {
756         /* create or truncate the reconstructed video file */
757         FILE *f = fopen( h->param.psz_dump_yuv, "w" );
758         if( f )
759             fclose( f );
760         else
761         {
762             x264_log( h, X264_LOG_ERROR, "can't write to fdec.yuv\n" );
763             x264_free( h );
764             return NULL;
765         }
766     }
767
768     return h;
769 }
770
771 /****************************************************************************
772  * x264_encoder_reconfig:
773  ****************************************************************************/
774 int x264_encoder_reconfig( x264_t *h, x264_param_t *param )
775 {
776 #define COPY(var) h->param.var = param->var
777     COPY( i_frame_reference ); // but never uses more refs than initially specified
778     COPY( i_bframe_bias );
779     if( h->param.i_scenecut_threshold >= 0 && param->i_scenecut_threshold >= 0 )
780         COPY( i_scenecut_threshold ); // can't turn it on or off, only vary the threshold
781     COPY( b_deblocking_filter );
782     COPY( i_deblocking_filter_alphac0 );
783     COPY( i_deblocking_filter_beta );
784     COPY( analyse.intra );
785     COPY( analyse.inter );
786     COPY( analyse.i_direct_mv_pred );
787     COPY( analyse.i_me_method );
788     COPY( analyse.i_me_range );
789     COPY( analyse.i_noise_reduction );
790     COPY( analyse.i_subpel_refine );
791     COPY( analyse.i_trellis );
792     COPY( analyse.b_bidir_me );
793     COPY( analyse.b_bframe_rdo );
794     COPY( analyse.b_chroma_me );
795     COPY( analyse.b_dct_decimate );
796     COPY( analyse.b_fast_pskip );
797     COPY( analyse.b_mixed_references );
798     // can only twiddle these if they were enabled to begin with:
799     if( h->pps->b_transform_8x8_mode )
800         COPY( analyse.b_transform_8x8 );
801     if( h->frames.i_max_ref1 > 1 )
802         COPY( b_bframe_pyramid );
803 #undef COPY
804
805     mbcmp_init( h );
806
807     return x264_validate_parameters( h );
808 }
809
810 /* internal usage */
811 static void x264_nal_start( x264_t *h, int i_type, int i_ref_idc )
812 {
813     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
814
815     nal->i_ref_idc = i_ref_idc;
816     nal->i_type    = i_type;
817
818     nal->i_payload= 0;
819     nal->p_payload= &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8];
820 }
821 static void x264_nal_end( x264_t *h )
822 {
823     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
824     nal->i_payload = &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8] - nal->p_payload;
825     h->out.i_nal++;
826 }
827
828 /****************************************************************************
829  * x264_encoder_headers:
830  ****************************************************************************/
831 int x264_encoder_headers( x264_t *h, x264_nal_t **pp_nal, int *pi_nal )
832 {
833     /* init bitstream context */
834     h->out.i_nal = 0;
835     bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
836
837     /* Put SPS and PPS */
838     if( h->i_frame == 0 )
839     {
840         /* identify ourself */
841         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
842         x264_sei_version_write( h, &h->out.bs );
843         x264_nal_end( h );
844
845         /* generate sequence parameters */
846         x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
847         x264_sps_write( &h->out.bs, h->sps );
848         x264_nal_end( h );
849
850         /* generate picture parameters */
851         x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
852         x264_pps_write( &h->out.bs, h->pps );
853         x264_nal_end( h );
854     }
855     /* now set output*/
856     *pi_nal = h->out.i_nal;
857     *pp_nal = &h->out.nal[0];
858     h->out.i_nal = 0;
859
860     return 0;
861 }
862
863 static inline void x264_reference_build_list( x264_t *h, int i_poc )
864 {
865     int i;
866     int b_ok;
867
868     /* build ref list 0/1 */
869     h->i_ref0 = 0;
870     h->i_ref1 = 0;
871     for( i = 0; h->frames.reference[i]; i++ )
872     {
873         if( h->frames.reference[i]->i_poc < i_poc )
874         {
875             h->fref0[h->i_ref0++] = h->frames.reference[i];
876         }
877         else if( h->frames.reference[i]->i_poc > i_poc )
878         {
879             h->fref1[h->i_ref1++] = h->frames.reference[i];
880         }
881     }
882
883     /* Order ref0 from higher to lower poc */
884     do
885     {
886         b_ok = 1;
887         for( i = 0; i < h->i_ref0 - 1; i++ )
888         {
889             if( h->fref0[i]->i_poc < h->fref0[i+1]->i_poc )
890             {
891                 XCHG( x264_frame_t*, h->fref0[i], h->fref0[i+1] );
892                 b_ok = 0;
893                 break;
894             }
895         }
896     } while( !b_ok );
897     /* Order ref1 from lower to higher poc (bubble sort) for B-frame */
898     do
899     {
900         b_ok = 1;
901         for( i = 0; i < h->i_ref1 - 1; i++ )
902         {
903             if( h->fref1[i]->i_poc > h->fref1[i+1]->i_poc )
904             {
905                 XCHG( x264_frame_t*, h->fref1[i], h->fref1[i+1] );
906                 b_ok = 0;
907                 break;
908             }
909         }
910     } while( !b_ok );
911
912     /* In the standard, a P-frame's ref list is sorted by frame_num.
913      * We use POC, but check whether explicit reordering is needed */
914     h->b_ref_reorder[0] =
915     h->b_ref_reorder[1] = 0;
916     if( h->sh.i_type == SLICE_TYPE_P )
917     {
918         for( i = 0; i < h->i_ref0 - 1; i++ )
919             if( h->fref0[i]->i_frame_num < h->fref0[i+1]->i_frame_num )
920             {
921                 h->b_ref_reorder[0] = 1;
922                 break;
923             }
924     }
925
926     h->i_ref1 = X264_MIN( h->i_ref1, h->frames.i_max_ref1 );
927     h->i_ref0 = X264_MIN( h->i_ref0, h->frames.i_max_ref0 );
928     h->i_ref0 = X264_MIN( h->i_ref0, h->param.i_frame_reference ); // if reconfig() has lowered the limit
929     assert( h->i_ref0 + h->i_ref1 <= 16 );
930     h->mb.pic.i_fref[0] = h->i_ref0;
931     h->mb.pic.i_fref[1] = h->i_ref1;
932 }
933
934 static void x264_fdec_filter_row( x264_t *h, int mb_y )
935 {
936     /* mb_y is the mb to be encoded next, not the mb to be filtered here */
937     int b_hpel = h->fdec->b_kept_as_ref;
938     int b_deblock = !h->sh.i_disable_deblocking_filter_idc;
939     int b_end = mb_y == h->sps->i_mb_height;
940     int min_y = mb_y - (1 << h->sh.b_mbaff);
941     int max_y = b_end ? h->sps->i_mb_height : mb_y;
942     b_deblock &= b_hpel || h->param.psz_dump_yuv;
943     if( mb_y & h->sh.b_mbaff )
944         return;
945     if( min_y < 0 )
946         return;
947
948     if( !b_end )
949     {
950         int i, j;
951         for( j=0; j<=h->sh.b_mbaff; j++ )
952             for( i=0; i<3; i++ )
953             {
954                 memcpy( h->mb.intra_border_backup[j][i],
955                         h->fdec->plane[i] + ((mb_y*16 >> !!i) + j - 1 - h->sh.b_mbaff) * h->fdec->i_stride[i],
956                         h->sps->i_mb_width*16 >> !!i );
957             }
958     }
959
960     if( b_deblock )
961     {
962         int y;
963         for( y = min_y; y < max_y; y += (1 << h->sh.b_mbaff) )
964             x264_frame_deblock_row( h, y );
965     }
966
967     if( b_hpel )
968     {
969         x264_frame_expand_border( h, h->fdec, min_y, b_end );
970         x264_frame_filter( h, h->fdec, min_y, b_end );
971         x264_frame_expand_border_filtered( h, h->fdec, min_y, b_end );
972     }
973
974     if( h->param.i_threads > 1 && h->fdec->b_kept_as_ref )
975     {
976         x264_frame_cond_broadcast( h->fdec, mb_y*16 + (b_end ? 10000 : -(X264_THREAD_HEIGHT << h->sh.b_mbaff)) );
977     }
978
979     min_y = X264_MAX( min_y*16-8, 0 );
980     max_y = b_end ? h->param.i_height : mb_y*16-8;
981
982     if( h->param.analyse.b_psnr )
983     {
984         int i;
985         for( i=0; i<3; i++ )
986             h->stat.frame.i_ssd[i] +=
987                 x264_pixel_ssd_wxh( &h->pixf,
988                     h->fdec->plane[i] + (min_y>>!!i) * h->fdec->i_stride[i], h->fdec->i_stride[i],
989                     h->fenc->plane[i] + (min_y>>!!i) * h->fenc->i_stride[i], h->fenc->i_stride[i],
990                     h->param.i_width >> !!i, (max_y-min_y) >> !!i );
991     }
992
993     if( h->param.analyse.b_ssim )
994     {
995         x264_emms();
996         /* offset by 2 pixels to avoid alignment of ssim blocks with dct blocks,
997          * and overlap by 4 */
998         min_y += min_y == 0 ? 2 : -6;
999         h->stat.frame.f_ssim +=
1000             x264_pixel_ssim_wxh( &h->pixf,
1001                 h->fdec->plane[0] + 2+min_y*h->fdec->i_stride[0], h->fdec->i_stride[0],
1002                 h->fenc->plane[0] + 2+min_y*h->fenc->i_stride[0], h->fenc->i_stride[0],
1003                 h->param.i_width-2, max_y-min_y );
1004     }
1005 }
1006
1007 static inline void x264_reference_update( x264_t *h )
1008 {
1009     int i;
1010
1011     if( h->fdec->i_frame >= 0 )
1012         h->i_frame++;
1013
1014     if( !h->fdec->b_kept_as_ref )
1015     {
1016         if( h->param.i_threads > 1 )
1017         {
1018             x264_frame_push_unused( h, h->fdec );
1019             h->fdec = x264_frame_pop_unused( h );
1020         }
1021         return;
1022     }
1023
1024     /* move lowres copy of the image to the ref frame */
1025     for( i = 0; i < 4; i++)
1026     {
1027         XCHG( uint8_t*, h->fdec->lowres[i], h->fenc->lowres[i] );
1028         XCHG( uint8_t*, h->fdec->buffer_lowres[i], h->fenc->buffer_lowres[i] );
1029     }
1030
1031     /* adaptive B decision needs a pointer, since it can't use the ref lists */
1032     if( h->sh.i_type != SLICE_TYPE_B )
1033         h->frames.last_nonb = h->fdec;
1034
1035     /* move frame in the buffer */
1036     x264_frame_push( h->frames.reference, h->fdec );
1037     if( h->frames.reference[h->frames.i_max_dpb] )
1038         x264_frame_push_unused( h, x264_frame_shift( h->frames.reference ) );
1039     h->fdec = x264_frame_pop_unused( h );
1040 }
1041
1042 static inline void x264_reference_reset( x264_t *h )
1043 {
1044     while( h->frames.reference[0] )
1045         x264_frame_push_unused( h, x264_frame_pop( h->frames.reference ) );
1046     h->fdec->i_poc =
1047     h->fenc->i_poc = 0;
1048 }
1049
1050 static inline void x264_slice_init( x264_t *h, int i_nal_type, int i_global_qp )
1051 {
1052     /* ------------------------ Create slice header  ----------------------- */
1053     if( i_nal_type == NAL_SLICE_IDR )
1054     {
1055         x264_slice_header_init( h, &h->sh, h->sps, h->pps, h->i_idr_pic_id, h->i_frame_num, i_global_qp );
1056
1057         /* increment id */
1058         h->i_idr_pic_id = ( h->i_idr_pic_id + 1 ) % 65536;
1059     }
1060     else
1061     {
1062         x264_slice_header_init( h, &h->sh, h->sps, h->pps, -1, h->i_frame_num, i_global_qp );
1063
1064         /* always set the real higher num of ref frame used */
1065         h->sh.b_num_ref_idx_override = 1;
1066         h->sh.i_num_ref_idx_l0_active = h->i_ref0 <= 0 ? 1 : h->i_ref0;
1067         h->sh.i_num_ref_idx_l1_active = h->i_ref1 <= 0 ? 1 : h->i_ref1;
1068     }
1069
1070     h->fdec->i_frame_num = h->sh.i_frame_num;
1071
1072     if( h->sps->i_poc_type == 0 )
1073     {
1074         h->sh.i_poc_lsb = h->fdec->i_poc & ( (1 << h->sps->i_log2_max_poc_lsb) - 1 );
1075         h->sh.i_delta_poc_bottom = 0;   /* XXX won't work for field */
1076     }
1077     else if( h->sps->i_poc_type == 1 )
1078     {
1079         /* FIXME TODO FIXME */
1080     }
1081     else
1082     {
1083         /* Nothing to do ? */
1084     }
1085
1086     x264_macroblock_slice_init( h );
1087 }
1088
1089 static void x264_slice_write( x264_t *h )
1090 {
1091     int i_skip;
1092     int mb_xy, i_mb_x, i_mb_y;
1093     int i, i_list, i_ref;
1094
1095     /* init stats */
1096     memset( &h->stat.frame, 0, sizeof(h->stat.frame) );
1097
1098     /* Slice */
1099     x264_nal_start( h, h->i_nal_type, h->i_nal_ref_idc );
1100
1101     /* Slice header */
1102     x264_slice_header_write( &h->out.bs, &h->sh, h->i_nal_ref_idc );
1103     if( h->param.b_cabac )
1104     {
1105         /* alignment needed */
1106         bs_align_1( &h->out.bs );
1107
1108         /* init cabac */
1109         x264_cabac_context_init( &h->cabac, h->sh.i_type, h->sh.i_qp, h->sh.i_cabac_init_idc );
1110         x264_cabac_encode_init ( &h->cabac, h->out.bs.p, h->out.bs.p_end );
1111     }
1112     h->mb.i_last_qp = h->sh.i_qp;
1113     h->mb.i_last_dqp = 0;
1114
1115     i_mb_y = h->sh.i_first_mb / h->sps->i_mb_width;
1116     i_mb_x = h->sh.i_first_mb % h->sps->i_mb_width;
1117     i_skip = 0;
1118
1119     while( (mb_xy = i_mb_x + i_mb_y * h->sps->i_mb_width) < h->sh.i_last_mb )
1120     {
1121         int mb_spos = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
1122
1123         if( i_mb_x == 0 )
1124             x264_fdec_filter_row( h, i_mb_y );
1125
1126         /* load cache */
1127         x264_macroblock_cache_load( h, i_mb_x, i_mb_y );
1128
1129         /* analyse parameters
1130          * Slice I: choose I_4x4 or I_16x16 mode
1131          * Slice P: choose between using P mode or intra (4x4 or 16x16)
1132          * */
1133         x264_macroblock_analyse( h );
1134
1135         /* encode this macroblock -> be careful it can change the mb type to P_SKIP if needed */
1136         x264_macroblock_encode( h );
1137
1138         x264_bitstream_check_buffer( h );
1139
1140         if( h->param.b_cabac )
1141         {
1142             if( mb_xy > h->sh.i_first_mb && !(h->sh.b_mbaff && (i_mb_y&1)) )
1143                 x264_cabac_encode_terminal( &h->cabac );
1144
1145             if( IS_SKIP( h->mb.i_type ) )
1146                 x264_cabac_mb_skip( h, 1 );
1147             else
1148             {
1149                 if( h->sh.i_type != SLICE_TYPE_I )
1150                     x264_cabac_mb_skip( h, 0 );
1151                 x264_macroblock_write_cabac( h, &h->cabac );
1152             }
1153         }
1154         else
1155         {
1156             if( IS_SKIP( h->mb.i_type ) )
1157                 i_skip++;
1158             else
1159             {
1160                 if( h->sh.i_type != SLICE_TYPE_I )
1161                 {
1162                     bs_write_ue( &h->out.bs, i_skip );  /* skip run */
1163                     i_skip = 0;
1164                 }
1165                 x264_macroblock_write_cavlc( h, &h->out.bs );
1166             }
1167         }
1168
1169 #if VISUALIZE
1170         if( h->param.b_visualize )
1171             x264_visualize_mb( h );
1172 #endif
1173
1174         /* save cache */
1175         x264_macroblock_cache_save( h );
1176
1177         /* accumulate mb stats */
1178         h->stat.frame.i_mb_count[h->mb.i_type]++;
1179         if( !IS_SKIP(h->mb.i_type) && !IS_INTRA(h->mb.i_type) && !IS_DIRECT(h->mb.i_type) )
1180         {
1181             if( h->mb.i_partition != D_8x8 )
1182                 h->stat.frame.i_mb_partition[h->mb.i_partition] += 4;
1183             else
1184                 for( i = 0; i < 4; i++ )
1185                     h->stat.frame.i_mb_partition[h->mb.i_sub_partition[i]] ++;
1186             if( h->param.i_frame_reference > 1 )
1187                 for( i_list = 0; i_list <= (h->sh.i_type == SLICE_TYPE_B); i_list++ )
1188                     for( i = 0; i < 4; i++ )
1189                     {
1190                         i_ref = h->mb.cache.ref[i_list][ x264_scan8[4*i] ];
1191                         if( i_ref >= 0 )
1192                             h->stat.frame.i_mb_count_ref[i_list][i_ref] ++;
1193                     }
1194         }
1195         if( h->mb.i_cbp_luma && !IS_INTRA(h->mb.i_type) )
1196         {
1197             h->stat.frame.i_mb_count_8x8dct[0] ++;
1198             h->stat.frame.i_mb_count_8x8dct[1] += h->mb.b_transform_8x8;
1199         }
1200
1201         x264_ratecontrol_mb( h, bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac) - mb_spos );
1202
1203         if( h->sh.b_mbaff )
1204         {
1205             i_mb_x += i_mb_y & 1;
1206             i_mb_y ^= i_mb_x < h->sps->i_mb_width;
1207         }
1208         else
1209             i_mb_x++;
1210         if(i_mb_x == h->sps->i_mb_width)
1211         {
1212             i_mb_y++;
1213             i_mb_x = 0;
1214         }
1215     }
1216
1217     if( h->param.b_cabac )
1218     {
1219         x264_cabac_encode_flush( h, &h->cabac );
1220         h->out.bs.p = h->cabac.p;
1221     }
1222     else
1223     {
1224         if( i_skip > 0 )
1225             bs_write_ue( &h->out.bs, i_skip );  /* last skip run */
1226         /* rbsp_slice_trailing_bits */
1227         bs_rbsp_trailing( &h->out.bs );
1228     }
1229
1230     x264_nal_end( h );
1231
1232     x264_fdec_filter_row( h, h->sps->i_mb_height );
1233
1234     /* Compute misc bits */
1235     h->stat.frame.i_misc_bits = bs_pos( &h->out.bs )
1236                               + NALU_OVERHEAD * 8
1237                               - h->stat.frame.i_itex_bits
1238                               - h->stat.frame.i_ptex_bits
1239                               - h->stat.frame.i_hdr_bits;
1240 }
1241
1242 static void x264_thread_sync_context( x264_t *dst, x264_t *src )
1243 {
1244     x264_frame_t **f;
1245     if( dst == src )
1246         return;
1247
1248     // reference counting
1249     for( f = src->frames.reference; *f; f++ )
1250         (*f)->i_reference_count++;
1251     for( f = dst->frames.reference; *f; f++ )
1252         x264_frame_push_unused( src, *f );
1253     src->fdec->i_reference_count++;
1254     x264_frame_push_unused( src, dst->fdec );
1255
1256     // copy everything except the per-thread pointers and the constants.
1257     memcpy( &dst->i_frame, &src->i_frame, offsetof(x264_t, mb.type) - offsetof(x264_t, i_frame) );
1258     memcpy( &dst->mb.i_type, &src->mb.i_type, offsetof(x264_t, rc) - offsetof(x264_t, mb.i_type) );
1259     dst->stat = src->stat;
1260 }
1261
1262 static void x264_thread_sync_stat( x264_t *dst, x264_t *src )
1263 {
1264     if( dst == src )
1265         return;
1266     memcpy( &dst->stat.i_slice_count, &src->stat.i_slice_count, sizeof(dst->stat) - sizeof(dst->stat.frame) );
1267 }
1268
1269 static int x264_slices_write( x264_t *h )
1270 {
1271     int i_frame_size;
1272
1273 #if VISUALIZE
1274     if( h->param.b_visualize )
1275         x264_visualize_init( h );
1276 #endif
1277
1278     x264_stack_align( x264_slice_write, h );
1279     i_frame_size = h->out.nal[h->out.i_nal-1].i_payload;
1280
1281 #if VISUALIZE
1282     if( h->param.b_visualize )
1283     {
1284         x264_visualize_show( h );
1285         x264_visualize_close( h );
1286     }
1287 #endif
1288
1289     h->out.i_frame_size = i_frame_size;
1290     return 0;
1291 }
1292
1293 /****************************************************************************
1294  * x264_encoder_encode:
1295  *  XXX: i_poc   : is the poc of the current given picture
1296  *       i_frame : is the number of the frame being coded
1297  *  ex:  type frame poc
1298  *       I      0   2*0
1299  *       P      1   2*3
1300  *       B      2   2*1
1301  *       B      3   2*2
1302  *       P      4   2*6
1303  *       B      5   2*4
1304  *       B      6   2*5
1305  ****************************************************************************/
1306 int     x264_encoder_encode( x264_t *h,
1307                              x264_nal_t **pp_nal, int *pi_nal,
1308                              x264_picture_t *pic_in,
1309                              x264_picture_t *pic_out )
1310 {
1311     x264_t *thread_current, *thread_prev, *thread_oldest;
1312     int     i_nal_type;
1313     int     i_nal_ref_idc;
1314
1315     int   i_global_qp;
1316
1317     if( h->param.i_threads > 1)
1318     {
1319         int i = ++h->i_thread_phase;
1320         int t = h->param.i_threads;
1321         thread_current = h->thread[ i%t ];
1322         thread_prev    = h->thread[ (i-1)%t ];
1323         thread_oldest  = h->thread[ (i+1)%t ];
1324         x264_thread_sync_context( thread_current, thread_prev );
1325         x264_thread_sync_ratecontrol( thread_current, thread_prev, thread_oldest );
1326         h = thread_current;
1327 //      fprintf(stderr, "current: %p  prev: %p  oldest: %p \n", thread_current, thread_prev, thread_oldest);
1328     }
1329     else
1330     {
1331         thread_current =
1332         thread_prev    =
1333         thread_oldest  = h;
1334     }
1335
1336     // ok to call this before encoding any frames, since the initial values of fdec have b_kept_as_ref=0
1337     x264_reference_update( h );
1338     h->fdec->i_lines_completed = -1;
1339
1340     /* no data out */
1341     *pi_nal = 0;
1342     *pp_nal = NULL;
1343
1344     /* ------------------- Setup new frame from picture -------------------- */
1345     if( pic_in != NULL )
1346     {
1347         /* 1: Copy the picture to a frame and move it to a buffer */
1348         x264_frame_t *fenc = x264_frame_pop_unused( h );
1349
1350         if( x264_frame_copy_picture( h, fenc, pic_in ) < 0 )
1351             return -1;
1352
1353         if( h->param.i_width != 16 * h->sps->i_mb_width ||
1354             h->param.i_height != 16 * h->sps->i_mb_height )
1355             x264_frame_expand_border_mod16( h, fenc );
1356
1357         fenc->i_frame = h->frames.i_input++;
1358
1359         x264_frame_push( h->frames.next, fenc );
1360
1361         if( h->frames.b_have_lowres )
1362             x264_frame_init_lowres( h, fenc );
1363
1364         if( h->frames.i_input <= h->frames.i_delay + 1 - h->param.i_threads )
1365         {
1366             /* Nothing yet to encode */
1367             /* waiting for filling bframe buffer */
1368             pic_out->i_type = X264_TYPE_AUTO;
1369             return 0;
1370         }
1371     }
1372
1373     if( h->frames.current[0] == NULL )
1374     {
1375         int bframes = 0;
1376         /* 2: Select frame types */
1377         if( h->frames.next[0] == NULL )
1378         {
1379             x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
1380             return 0;
1381         }
1382
1383         x264_slicetype_decide( h );
1384
1385         /* 3: move some B-frames and 1 non-B to encode queue */
1386         while( IS_X264_TYPE_B( h->frames.next[bframes]->i_type ) )
1387             bframes++;
1388         x264_frame_push( h->frames.current, x264_frame_shift( &h->frames.next[bframes] ) );
1389         /* FIXME: when max B-frames > 3, BREF may no longer be centered after GOP closing */
1390         if( h->param.b_bframe_pyramid && bframes > 1 )
1391         {
1392             x264_frame_t *mid = x264_frame_shift( &h->frames.next[bframes/2] );
1393             mid->i_type = X264_TYPE_BREF;
1394             x264_frame_push( h->frames.current, mid );
1395             bframes--;
1396         }
1397         while( bframes-- )
1398             x264_frame_push( h->frames.current, x264_frame_shift( h->frames.next ) );
1399     }
1400
1401     /* ------------------- Get frame to be encoded ------------------------- */
1402     /* 4: get picture to encode */
1403     h->fenc = x264_frame_shift( h->frames.current );
1404     if( h->fenc == NULL )
1405     {
1406         /* Nothing yet to encode (ex: waiting for I/P with B frames) */
1407         /* waiting for filling bframe buffer */
1408         pic_out->i_type = X264_TYPE_AUTO;
1409         return 0;
1410     }
1411
1412 do_encode:
1413
1414     if( h->fenc->i_type == X264_TYPE_IDR )
1415     {
1416         h->frames.i_last_idr = h->fenc->i_frame;
1417     }
1418
1419     /* ------------------- Setup frame context ----------------------------- */
1420     /* 5: Init data dependent of frame type */
1421     if( h->fenc->i_type == X264_TYPE_IDR )
1422     {
1423         /* reset ref pictures */
1424         x264_reference_reset( h );
1425
1426         i_nal_type    = NAL_SLICE_IDR;
1427         i_nal_ref_idc = NAL_PRIORITY_HIGHEST;
1428         h->sh.i_type = SLICE_TYPE_I;
1429     }
1430     else if( h->fenc->i_type == X264_TYPE_I )
1431     {
1432         i_nal_type    = NAL_SLICE;
1433         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
1434         h->sh.i_type = SLICE_TYPE_I;
1435     }
1436     else if( h->fenc->i_type == X264_TYPE_P )
1437     {
1438         i_nal_type    = NAL_SLICE;
1439         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
1440         h->sh.i_type = SLICE_TYPE_P;
1441     }
1442     else if( h->fenc->i_type == X264_TYPE_BREF )
1443     {
1444         i_nal_type    = NAL_SLICE;
1445         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* maybe add MMCO to forget it? -> low */
1446         h->sh.i_type = SLICE_TYPE_B;
1447     }
1448     else    /* B frame */
1449     {
1450         i_nal_type    = NAL_SLICE;
1451         i_nal_ref_idc = NAL_PRIORITY_DISPOSABLE;
1452         h->sh.i_type = SLICE_TYPE_B;
1453     }
1454
1455     h->fdec->i_poc =
1456     h->fenc->i_poc = 2 * (h->fenc->i_frame - h->frames.i_last_idr);
1457     h->fdec->i_type = h->fenc->i_type;
1458     h->fdec->i_frame = h->fenc->i_frame;
1459     h->fenc->b_kept_as_ref =
1460     h->fdec->b_kept_as_ref = i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE && h->param.i_keyint_max > 1;
1461
1462
1463
1464     /* ------------------- Init                ----------------------------- */
1465     /* build ref list 0/1 */
1466     x264_reference_build_list( h, h->fdec->i_poc );
1467
1468     /* Init the rate control */
1469     x264_ratecontrol_start( h, h->fenc->i_qpplus1 );
1470     i_global_qp = x264_ratecontrol_qp( h );
1471
1472     pic_out->i_qpplus1 =
1473     h->fdec->i_qpplus1 = i_global_qp + 1;
1474
1475     if( h->sh.i_type == SLICE_TYPE_B )
1476         x264_macroblock_bipred_init( h );
1477
1478     /* ------------------------ Create slice header  ----------------------- */
1479     x264_slice_init( h, i_nal_type, i_global_qp );
1480
1481     if( i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE )
1482         h->i_frame_num++;
1483
1484     /* ---------------------- Write the bitstream -------------------------- */
1485     /* Init bitstream context */
1486     h->out.i_nal = 0;
1487     bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
1488
1489     if(h->param.b_aud){
1490         int pic_type;
1491
1492         if(h->sh.i_type == SLICE_TYPE_I)
1493             pic_type = 0;
1494         else if(h->sh.i_type == SLICE_TYPE_P)
1495             pic_type = 1;
1496         else if(h->sh.i_type == SLICE_TYPE_B)
1497             pic_type = 2;
1498         else
1499             pic_type = 7;
1500
1501         x264_nal_start(h, NAL_AUD, NAL_PRIORITY_DISPOSABLE);
1502         bs_write(&h->out.bs, 3, pic_type);
1503         bs_rbsp_trailing(&h->out.bs);
1504         x264_nal_end(h);
1505     }
1506
1507     h->i_nal_type = i_nal_type;
1508     h->i_nal_ref_idc = i_nal_ref_idc;
1509
1510     /* Write SPS and PPS */
1511     if( i_nal_type == NAL_SLICE_IDR && h->param.b_repeat_headers )
1512     {
1513         if( h->fenc->i_frame == 0 )
1514         {
1515             /* identify ourself */
1516             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
1517             x264_sei_version_write( h, &h->out.bs );
1518             x264_nal_end( h );
1519         }
1520
1521         /* generate sequence parameters */
1522         x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
1523         x264_sps_write( &h->out.bs, h->sps );
1524         x264_nal_end( h );
1525
1526         /* generate picture parameters */
1527         x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
1528         x264_pps_write( &h->out.bs, h->pps );
1529         x264_nal_end( h );
1530     }
1531
1532     /* Write frame */
1533     if( h->param.i_threads > 1 )
1534     {
1535         x264_pthread_create( &h->thread_handle, NULL, (void*)x264_slices_write, h );
1536         h->b_thread_active = 1;
1537     }
1538     else
1539         x264_slices_write( h );
1540
1541     /* restore CPU state (before using float again) */
1542     x264_emms();
1543
1544     if( h->sh.i_type == SLICE_TYPE_P && !h->param.rc.b_stat_read 
1545         && h->param.i_scenecut_threshold >= 0
1546         && !h->param.b_pre_scenecut )
1547     {
1548         const int *mbs = h->stat.frame.i_mb_count;
1549         int i_mb_i = mbs[I_16x16] + mbs[I_8x8] + mbs[I_4x4];
1550         int i_mb_p = mbs[P_L0] + mbs[P_8x8];
1551         int i_mb_s = mbs[P_SKIP];
1552         int i_mb   = h->sps->i_mb_width * h->sps->i_mb_height;
1553         int64_t i_inter_cost = h->stat.frame.i_inter_cost;
1554         int64_t i_intra_cost = h->stat.frame.i_intra_cost;
1555
1556         float f_bias;
1557         int i_gop_size = h->fenc->i_frame - h->frames.i_last_idr;
1558         float f_thresh_max = h->param.i_scenecut_threshold / 100.0;
1559         /* magic numbers pulled out of thin air */
1560         float f_thresh_min = f_thresh_max * h->param.i_keyint_min
1561                              / ( h->param.i_keyint_max * 4 );
1562         if( h->param.i_keyint_min == h->param.i_keyint_max )
1563              f_thresh_min= f_thresh_max;
1564
1565         /* macroblock_analyse() doesn't further analyse skipped mbs,
1566          * so we have to guess their cost */
1567         if( h->stat.frame.i_mbs_analysed > 0 )
1568             i_intra_cost = i_intra_cost * i_mb / h->stat.frame.i_mbs_analysed;
1569
1570         if( i_gop_size < h->param.i_keyint_min / 4 )
1571             f_bias = f_thresh_min / 4;
1572         else if( i_gop_size <= h->param.i_keyint_min )
1573             f_bias = f_thresh_min * i_gop_size / h->param.i_keyint_min;
1574         else
1575         {
1576             f_bias = f_thresh_min
1577                      + ( f_thresh_max - f_thresh_min )
1578                        * ( i_gop_size - h->param.i_keyint_min )
1579                        / ( h->param.i_keyint_max - h->param.i_keyint_min );
1580         }
1581         f_bias = X264_MIN( f_bias, 1.0 );
1582
1583         /* Bad P will be reencoded as I */
1584         if( h->stat.frame.i_mbs_analysed > 0 &&
1585             i_inter_cost >= (1.0 - f_bias) * i_intra_cost )
1586         {
1587             int b;
1588
1589             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",
1590                       h->fenc->i_frame,
1591                       (double)i_intra_cost, (double)i_inter_cost,
1592                       1. - (double)i_inter_cost / i_intra_cost,
1593                       f_bias, i_gop_size,
1594                       i_mb_i, i_mb_p, i_mb_s );
1595
1596             /* Restore frame num */
1597             h->i_frame_num--;
1598
1599             for( b = 0; h->frames.current[b] && IS_X264_TYPE_B( h->frames.current[b]->i_type ); b++ );
1600             if( b > 0 )
1601             {
1602                 /* If using B-frames, force GOP to be closed.
1603                  * Even if this frame is going to be I and not IDR, forcing a
1604                  * P-frame before the scenecut will probably help compression.
1605                  * 
1606                  * We don't yet know exactly which frame is the scene cut, so
1607                  * we can't assign an I-frame. Instead, change the previous
1608                  * B-frame to P, and rearrange coding order. */
1609
1610                 if( h->param.b_bframe_adaptive || b > 1 )
1611                     h->fenc->i_type = X264_TYPE_AUTO;
1612                 x264_frame_sort_pts( h->frames.current );
1613                 x264_frame_unshift( h->frames.next, h->fenc );
1614                 h->fenc = h->frames.current[b-1];
1615                 h->frames.current[b-1] = NULL;
1616                 h->fenc->i_type = X264_TYPE_P;
1617                 x264_frame_sort_dts( h->frames.current );
1618             }
1619             /* Do IDR if needed */
1620             else if( i_gop_size >= h->param.i_keyint_min )
1621             {
1622                 /* Reset */
1623                 h->i_frame_num = 0;
1624
1625                 /* Reinit field of fenc */
1626                 h->fenc->i_type = X264_TYPE_IDR;
1627                 h->fenc->i_poc = 0;
1628
1629                 /* Put enqueued frames back in the pool */
1630                 while( h->frames.current[0] )
1631                     x264_frame_push( h->frames.next, x264_frame_shift( h->frames.current ) );
1632                 x264_frame_sort_pts( h->frames.next );
1633             }
1634             else
1635             {
1636                 h->fenc->i_type = X264_TYPE_I;
1637             }
1638             goto do_encode;
1639         }
1640     }
1641
1642     x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
1643     return 0;
1644 }
1645
1646 static void x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
1647                                     x264_nal_t **pp_nal, int *pi_nal,
1648                                     x264_picture_t *pic_out )
1649 {
1650     int i, i_list;
1651     char psz_message[80];
1652
1653     if( h->b_thread_active )
1654     {
1655         x264_pthread_join( h->thread_handle, NULL );
1656         h->b_thread_active = 0;
1657     }
1658     if( !h->out.i_nal )
1659     {
1660         pic_out->i_type = X264_TYPE_AUTO;
1661         return;
1662     }
1663
1664     x264_frame_push_unused( thread_current, h->fenc );
1665
1666     /* End bitstream, set output  */
1667     *pi_nal = h->out.i_nal;
1668     *pp_nal = h->out.nal;
1669     h->out.i_nal = 0;
1670
1671     /* Set output picture properties */
1672     if( h->sh.i_type == SLICE_TYPE_I )
1673         pic_out->i_type = h->i_nal_type == NAL_SLICE_IDR ? X264_TYPE_IDR : X264_TYPE_I;
1674     else if( h->sh.i_type == SLICE_TYPE_P )
1675         pic_out->i_type = X264_TYPE_P;
1676     else
1677         pic_out->i_type = X264_TYPE_B;
1678     pic_out->i_pts = h->fenc->i_pts;
1679
1680     pic_out->img.i_plane = h->fdec->i_plane;
1681     for(i = 0; i < 4; i++){
1682         pic_out->img.i_stride[i] = h->fdec->i_stride[i];
1683         pic_out->img.plane[i] = h->fdec->plane[i];
1684     }
1685
1686     /* ---------------------- Update encoder state ------------------------- */
1687
1688     /* update rc */
1689     x264_emms();
1690     x264_ratecontrol_end( h, h->out.i_frame_size * 8 );
1691
1692     /* restore CPU state (before using float again) */
1693     x264_emms();
1694
1695     x264_noise_reduction_update( thread_current );
1696
1697     /* ---------------------- Compute/Print statistics --------------------- */
1698     x264_thread_sync_stat( h, h->thread[0] );
1699
1700     /* Slice stat */
1701     h->stat.i_slice_count[h->sh.i_type]++;
1702     h->stat.i_slice_size[h->sh.i_type] += h->out.i_frame_size + NALU_OVERHEAD;
1703     h->stat.f_slice_qp[h->sh.i_type] += h->fdec->f_qp_avg_aq;
1704
1705     for( i = 0; i < X264_MBTYPE_MAX; i++ )
1706         h->stat.i_mb_count[h->sh.i_type][i] += h->stat.frame.i_mb_count[i];
1707     for( i = 0; i < X264_PARTTYPE_MAX; i++ )
1708         h->stat.i_mb_partition[h->sh.i_type][i] += h->stat.frame.i_mb_partition[i];
1709     for( i = 0; i < 2; i++ )
1710         h->stat.i_mb_count_8x8dct[i] += h->stat.frame.i_mb_count_8x8dct[i];
1711     if( h->sh.i_type != SLICE_TYPE_I )
1712         for( i_list = 0; i_list < 2; i_list++ )
1713             for( i = 0; i < 32; i++ )
1714                 h->stat.i_mb_count_ref[h->sh.i_type][i_list][i] += h->stat.frame.i_mb_count_ref[i_list][i];
1715     if( h->sh.i_type == SLICE_TYPE_B )
1716     {
1717         h->stat.i_direct_frames[ h->sh.b_direct_spatial_mv_pred ] ++;
1718         if( h->mb.b_direct_auto_write )
1719         {
1720             //FIXME somewhat arbitrary time constants
1721             if( h->stat.i_direct_score[0] + h->stat.i_direct_score[1] > h->mb.i_mb_count )
1722             {
1723                 for( i = 0; i < 2; i++ )
1724                     h->stat.i_direct_score[i] = h->stat.i_direct_score[i] * 9/10;
1725             }
1726             for( i = 0; i < 2; i++ )
1727                 h->stat.i_direct_score[i] += h->stat.frame.i_direct_score[i];
1728         }
1729     }
1730
1731     psz_message[0] = '\0';
1732     if( h->param.analyse.b_psnr )
1733     {
1734         int64_t sqe[3] = {
1735             h->stat.frame.i_ssd[0],
1736             h->stat.frame.i_ssd[1],
1737             h->stat.frame.i_ssd[2],
1738         };
1739
1740         h->stat.i_sqe_global[h->sh.i_type] += sqe[0] + sqe[1] + sqe[2];
1741         h->stat.f_psnr_average[h->sh.i_type] += x264_psnr( sqe[0] + sqe[1] + sqe[2], 3 * h->param.i_width * h->param.i_height / 2 );
1742         h->stat.f_psnr_mean_y[h->sh.i_type] += x264_psnr( sqe[0], h->param.i_width * h->param.i_height );
1743         h->stat.f_psnr_mean_u[h->sh.i_type] += x264_psnr( sqe[1], h->param.i_width * h->param.i_height / 4 );
1744         h->stat.f_psnr_mean_v[h->sh.i_type] += x264_psnr( sqe[2], h->param.i_width * h->param.i_height / 4 );
1745
1746         snprintf( psz_message, 80, " PSNR Y:%5.2f U:%5.2f V:%5.2f",
1747                   x264_psnr( sqe[0], h->param.i_width * h->param.i_height ),
1748                   x264_psnr( sqe[1], h->param.i_width * h->param.i_height / 4),
1749                   x264_psnr( sqe[2], h->param.i_width * h->param.i_height / 4) );
1750     }
1751
1752     if( h->param.analyse.b_ssim )
1753     {
1754         double ssim_y = h->stat.frame.f_ssim
1755                       / (((h->param.i_width-6)>>2) * ((h->param.i_height-6)>>2));
1756         h->stat.f_ssim_mean_y[h->sh.i_type] += ssim_y;
1757         snprintf( psz_message + strlen(psz_message), 80 - strlen(psz_message),
1758                   " SSIM Y:%.5f", ssim_y );
1759     }
1760     psz_message[79] = '\0';
1761     
1762     x264_log( h, X264_LOG_DEBUG,
1763                   "frame=%4d QP=%.2f NAL=%d Slice:%c Poc:%-3d I:%-4d P:%-4d SKIP:%-4d size=%d bytes%s\n",
1764               h->i_frame,
1765               h->fdec->f_qp_avg_aq,
1766               h->i_nal_ref_idc,
1767               h->sh.i_type == SLICE_TYPE_I ? 'I' : (h->sh.i_type == SLICE_TYPE_P ? 'P' : 'B' ),
1768               h->fdec->i_poc,
1769               h->stat.frame.i_mb_count_i,
1770               h->stat.frame.i_mb_count_p,
1771               h->stat.frame.i_mb_count_skip,
1772               h->out.i_frame_size,
1773               psz_message );
1774
1775     // keep stats all in one place
1776     x264_thread_sync_stat( h->thread[0], h );
1777     // for the use of the next frame
1778     x264_thread_sync_stat( thread_current, h );
1779
1780 #ifdef DEBUG_MB_TYPE
1781 {
1782     static const char mb_chars[] = { 'i', 'i', 'I', 'C', 'P', '8', 'S',
1783         'D', '<', 'X', 'B', 'X', '>', 'B', 'B', 'B', 'B', '8', 'S' };
1784     int mb_xy;
1785     for( mb_xy = 0; mb_xy < h->sps->i_mb_width * h->sps->i_mb_height; mb_xy++ )
1786     {
1787         if( h->mb.type[mb_xy] < X264_MBTYPE_MAX && h->mb.type[mb_xy] >= 0 )
1788             fprintf( stderr, "%c ", mb_chars[ h->mb.type[mb_xy] ] );
1789         else
1790             fprintf( stderr, "? " );
1791
1792         if( (mb_xy+1) % h->sps->i_mb_width == 0 )
1793             fprintf( stderr, "\n" );
1794     }
1795 }
1796 #endif
1797
1798     if( h->param.psz_dump_yuv )
1799         x264_frame_dump( h );
1800 }
1801
1802 static void x264_print_intra( int64_t *i_mb_count, double i_count, int b_print_pcm, char *intra )
1803 {
1804     intra += sprintf( intra, "I16..4%s: %4.1f%% %4.1f%% %4.1f%%",
1805         b_print_pcm ? "..PCM" : "",
1806         i_mb_count[I_16x16]/ i_count,
1807         i_mb_count[I_8x8]  / i_count,
1808         i_mb_count[I_4x4]  / i_count );
1809     if( b_print_pcm )
1810         sprintf( intra, " %4.1f%%", i_mb_count[I_PCM]  / i_count );
1811 }
1812
1813 /****************************************************************************
1814  * x264_encoder_close:
1815  ****************************************************************************/
1816 void    x264_encoder_close  ( x264_t *h )
1817 {
1818     int64_t i_yuv_size = 3 * h->param.i_width * h->param.i_height / 2;
1819     int64_t i_mb_count_size[2][7] = {{0}};
1820     char intra[40];
1821     int i, j, i_list, i_type;
1822     int b_print_pcm = h->stat.i_mb_count[SLICE_TYPE_I][I_PCM]
1823                    || h->stat.i_mb_count[SLICE_TYPE_P][I_PCM]
1824                    || h->stat.i_mb_count[SLICE_TYPE_B][I_PCM];
1825
1826     for( i=0; i<h->param.i_threads; i++ )
1827     {
1828         // don't strictly have to wait for the other threads, but it's simpler than canceling them
1829         if( h->thread[i]->b_thread_active )
1830         {
1831             x264_pthread_join( h->thread[i]->thread_handle, NULL );
1832             assert( h->thread[i]->fenc->i_reference_count == 1 );
1833             x264_frame_delete( h->thread[i]->fenc );
1834         }
1835     }
1836
1837     /* Slices used and PSNR */
1838     for( i=0; i<5; i++ )
1839     {
1840         static const int slice_order[] = { SLICE_TYPE_I, SLICE_TYPE_SI, SLICE_TYPE_P, SLICE_TYPE_SP, SLICE_TYPE_B };
1841         static const char *slice_name[] = { "P", "B", "I", "SP", "SI" };
1842         int i_slice = slice_order[i];
1843
1844         if( h->stat.i_slice_count[i_slice] > 0 )
1845         {
1846             const int i_count = h->stat.i_slice_count[i_slice];
1847             if( h->param.analyse.b_psnr )
1848             {
1849                 x264_log( h, X264_LOG_INFO,
1850                           "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",
1851                           slice_name[i_slice],
1852                           i_count,
1853                           h->stat.f_slice_qp[i_slice] / i_count,
1854                           (double)h->stat.i_slice_size[i_slice] / i_count,
1855                           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,
1856                           h->stat.f_psnr_average[i_slice] / i_count,
1857                           x264_psnr( h->stat.i_sqe_global[i_slice], i_count * i_yuv_size ) );
1858             }
1859             else
1860             {
1861                 x264_log( h, X264_LOG_INFO,
1862                           "slice %s:%-5d Avg QP:%5.2f  size:%6.0f\n",
1863                           slice_name[i_slice],
1864                           i_count,
1865                           h->stat.f_slice_qp[i_slice] / i_count,
1866                           (double)h->stat.i_slice_size[i_slice] / i_count );
1867             }
1868         }
1869     }
1870
1871     for( i_type = 0; i_type < 2; i_type++ )
1872         for( i = 0; i < X264_PARTTYPE_MAX; i++ )
1873         {
1874             if( i == D_DIRECT_8x8 ) continue; /* direct is counted as its own type */
1875             i_mb_count_size[i_type][x264_mb_partition_pixel_table[i]] += h->stat.i_mb_partition[i_type][i];
1876         }
1877
1878     /* MB types used */
1879     if( h->stat.i_slice_count[SLICE_TYPE_I] > 0 )
1880     {
1881         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_I];
1882         double i_count = h->stat.i_slice_count[SLICE_TYPE_I] * h->mb.i_mb_count / 100.0;
1883         x264_print_intra( i_mb_count, i_count, b_print_pcm, intra );
1884         x264_log( h, X264_LOG_INFO, "mb I  %s\n", intra );
1885     }
1886     if( h->stat.i_slice_count[SLICE_TYPE_P] > 0 )
1887     {
1888         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_P];
1889         double i_count = h->stat.i_slice_count[SLICE_TYPE_P] * h->mb.i_mb_count / 100.0;
1890         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_P];
1891         x264_print_intra( i_mb_count, i_count, b_print_pcm, intra );
1892         x264_log( h, X264_LOG_INFO,
1893                   "mb P  %s  P16..4: %4.1f%% %4.1f%% %4.1f%% %4.1f%% %4.1f%%    skip:%4.1f%%\n",
1894                   intra,
1895                   i_mb_size[PIXEL_16x16] / (i_count*4),
1896                   (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
1897                   i_mb_size[PIXEL_8x8] / (i_count*4),
1898                   (i_mb_size[PIXEL_8x4] + i_mb_size[PIXEL_4x8]) / (i_count*4),
1899                   i_mb_size[PIXEL_4x4] / (i_count*4),
1900                   i_mb_count[P_SKIP] / i_count );
1901     }
1902     if( h->stat.i_slice_count[SLICE_TYPE_B] > 0 )
1903     {
1904         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_B];
1905         double i_count = h->stat.i_slice_count[SLICE_TYPE_B] * h->mb.i_mb_count / 100.0;
1906         double i_mb_list_count;
1907         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_B];
1908         int64_t list_count[3] = {0}; /* 0 == L0, 1 == L1, 2 == BI */
1909         x264_print_intra( i_mb_count, i_count, b_print_pcm, intra );
1910         for( i = 0; i < X264_PARTTYPE_MAX; i++ )
1911             for( j = 0; j < 2; j++ )
1912             {
1913                 int l0 = x264_mb_type_list0_table[i][j];
1914                 int l1 = x264_mb_type_list1_table[i][j];
1915                 if( l0 || l1 )
1916                     list_count[l1+l0*l1] += h->stat.i_mb_count[SLICE_TYPE_B][i] * 2;
1917             }
1918         list_count[0] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L0_8x8];
1919         list_count[1] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L1_8x8];
1920         list_count[2] += h->stat.i_mb_partition[SLICE_TYPE_B][D_BI_8x8];
1921         i_mb_count[B_DIRECT] += (h->stat.i_mb_partition[SLICE_TYPE_B][D_DIRECT_8x8]+2)/4;
1922         i_mb_list_count = (list_count[0] + list_count[1] + list_count[2]) / 100.0;
1923         x264_log( h, X264_LOG_INFO,
1924                   "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",
1925                   intra,
1926                   i_mb_size[PIXEL_16x16] / (i_count*4),
1927                   (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
1928                   i_mb_size[PIXEL_8x8] / (i_count*4),
1929                   i_mb_count[B_DIRECT] / i_count,
1930                   i_mb_count[B_SKIP]   / i_count,
1931                   list_count[0] / i_mb_list_count,
1932                   list_count[1] / i_mb_list_count,
1933                   list_count[2] / i_mb_list_count );
1934     }
1935
1936     x264_ratecontrol_summary( h );
1937
1938     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 )
1939     {
1940         const int i_count = h->stat.i_slice_count[SLICE_TYPE_I] +
1941                             h->stat.i_slice_count[SLICE_TYPE_P] +
1942                             h->stat.i_slice_count[SLICE_TYPE_B];
1943         float fps = (float) h->param.i_fps_num / h->param.i_fps_den;
1944 #define SUM3(p) (p[SLICE_TYPE_I] + p[SLICE_TYPE_P] + p[SLICE_TYPE_B])
1945 #define SUM3b(p,o) (p[SLICE_TYPE_I][o] + p[SLICE_TYPE_P][o] + p[SLICE_TYPE_B][o])
1946         float f_bitrate = fps * SUM3(h->stat.i_slice_size) / i_count / 125;
1947
1948         if( h->pps->b_transform_8x8_mode )
1949         {
1950             int64_t i_i8x8 = SUM3b( h->stat.i_mb_count, I_8x8 );
1951             int64_t i_intra = i_i8x8 + SUM3b( h->stat.i_mb_count, I_4x4 )
1952                                      + SUM3b( h->stat.i_mb_count, I_16x16 );
1953             x264_log( h, X264_LOG_INFO, "8x8 transform  intra:%.1f%%  inter:%.1f%%\n",
1954                       100. * i_i8x8 / i_intra,
1955                       100. * h->stat.i_mb_count_8x8dct[1] / h->stat.i_mb_count_8x8dct[0] );
1956         }
1957
1958         if( h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO
1959             && h->stat.i_slice_count[SLICE_TYPE_B] )
1960         {
1961             x264_log( h, X264_LOG_INFO, "direct mvs  spatial:%.1f%%  temporal:%.1f%%\n",
1962                       h->stat.i_direct_frames[1] * 100. / h->stat.i_slice_count[SLICE_TYPE_B],
1963                       h->stat.i_direct_frames[0] * 100. / h->stat.i_slice_count[SLICE_TYPE_B] );
1964         }
1965
1966         for( i_list = 0; i_list < 2; i_list++ )
1967         {
1968             int i_slice;
1969             for( i_slice = 0; i_slice < 2; i_slice++ )
1970             {
1971                 char buf[200];
1972                 char *p = buf;
1973                 int64_t i_den = 0;
1974                 int i_max = 0;
1975                 for( i = 0; i < 32; i++ )
1976                     if( h->stat.i_mb_count_ref[i_slice][i_list][i] )
1977                     {
1978                         i_den += h->stat.i_mb_count_ref[i_slice][i_list][i];
1979                         i_max = i;
1980                     }
1981                 if( i_max == 0 )
1982                     continue;
1983                 for( i = 0; i <= i_max; i++ )
1984                     p += sprintf( p, " %4.1f%%", 100. * h->stat.i_mb_count_ref[i_slice][i_list][i] / i_den );
1985                 x264_log( h, X264_LOG_INFO, "ref %c L%d %s\n", "PB"[i_slice], i_list, buf );
1986             }
1987         }
1988
1989         if( h->param.analyse.b_ssim )
1990         {
1991             x264_log( h, X264_LOG_INFO,
1992                       "SSIM Mean Y:%.7f\n",
1993                       SUM3( h->stat.f_ssim_mean_y ) / i_count );
1994         }
1995         if( h->param.analyse.b_psnr )
1996         {
1997             x264_log( h, X264_LOG_INFO,
1998                       "PSNR Mean Y:%6.3f U:%6.3f V:%6.3f Avg:%6.3f Global:%6.3f kb/s:%.2f\n",
1999                       SUM3( h->stat.f_psnr_mean_y ) / i_count,
2000                       SUM3( h->stat.f_psnr_mean_u ) / i_count,
2001                       SUM3( h->stat.f_psnr_mean_v ) / i_count,
2002                       SUM3( h->stat.f_psnr_average ) / i_count,
2003                       x264_psnr( SUM3( h->stat.i_sqe_global ), i_count * i_yuv_size ),
2004                       f_bitrate );
2005         }
2006         else
2007             x264_log( h, X264_LOG_INFO, "kb/s:%.1f\n", f_bitrate );
2008     }
2009
2010     /* rc */
2011     x264_ratecontrol_delete( h );
2012
2013     /* param */
2014     if( h->param.rc.psz_stat_out )
2015         free( h->param.rc.psz_stat_out );
2016     if( h->param.rc.psz_stat_in )
2017         free( h->param.rc.psz_stat_in );
2018     if( h->param.rc.psz_rc_eq )
2019         free( h->param.rc.psz_rc_eq );
2020
2021     x264_cqm_delete( h );
2022
2023     if( h->param.i_threads > 1)
2024         h = h->thread[ h->i_thread_phase % h->param.i_threads ];
2025
2026     /* frames */
2027     for( i = 0; h->frames.current[i]; i++ )
2028     {
2029         assert( h->frames.current[i]->i_reference_count == 1 );
2030         x264_frame_delete( h->frames.current[i] );
2031     }
2032     for( i = 0; h->frames.next[i]; i++ )
2033     {
2034         assert( h->frames.next[i]->i_reference_count == 1 );
2035         x264_frame_delete( h->frames.next[i] );
2036     }
2037     for( i = 0; h->frames.unused[i]; i++ )
2038     {
2039         assert( h->frames.unused[i]->i_reference_count == 0 );
2040         x264_frame_delete( h->frames.unused[i] );
2041     }
2042
2043     h = h->thread[0];
2044
2045     for( i = h->param.i_threads - 1; i >= 0; i-- )
2046     {
2047         x264_frame_t **frame;
2048
2049         for( frame = h->thread[i]->frames.reference; *frame; frame++ )
2050         {
2051             assert( (*frame)->i_reference_count > 0 );
2052             (*frame)->i_reference_count--;
2053             if( (*frame)->i_reference_count == 0 )
2054                 x264_frame_delete( *frame );
2055         }
2056         frame = &h->thread[i]->fdec;
2057         assert( (*frame)->i_reference_count > 0 );
2058         (*frame)->i_reference_count--;
2059         if( (*frame)->i_reference_count == 0 )
2060             x264_frame_delete( *frame );
2061
2062         x264_macroblock_cache_end( h->thread[i] );
2063         x264_free( h->thread[i]->out.p_bitstream );
2064         x264_free( h->thread[i] );
2065     }
2066 }