1 /*****************************************************************************
3 *****************************************************************************
4 * Copyright (C) 2003 Laurent Aimar
5 * $Id: encoder.c,v 1.1 2004/06/03 19:27:08 fenrir Exp $
7 * Authors: Laurent Aimar <fenrir@via.ecp.fr>
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.
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.
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., 59 Temple Place - Suite 330, Boston, MA 02111, USA.
22 *****************************************************************************/
28 #include "common/common.h"
29 #include "common/cpu.h"
33 #include "ratecontrol.h"
34 #include "macroblock.h"
37 #include "common/visualize.h"
40 //#define DEBUG_MB_TYPE
41 //#define DEBUG_DUMP_FRAME
42 //#define DEBUG_BENCHMARK
44 #ifdef DEBUG_BENCHMARK
45 static int64_t i_mtime_encode_frame = 0;
46 static int64_t i_mtime_analyse = 0;
47 static int64_t i_mtime_encode = 0;
48 static int64_t i_mtime_write = 0;
49 static int64_t i_mtime_filter = 0;
50 #define TIMER_START( d ) \
52 int64_t d##start = x264_mdate();
54 #define TIMER_STOP( d ) \
55 d += x264_mdate() - d##start;\
58 #define TIMER_START( d )
59 #define TIMER_STOP( d )
62 #define NALU_OVERHEAD 5 // startcode + NAL type costs 5 bytes per frame
64 /****************************************************************************
66 ******************************* x264 libs **********************************
68 ****************************************************************************/
69 static float x264_psnr( int64_t i_sqe, int64_t i_size )
71 double f_mse = (double)i_sqe / ((double)65025.0 * (double)i_size);
72 if( f_mse <= 0.0000000001 ) /* Max 100dB */
75 return (float)(-10.0 * log( f_mse ) / log( 10.0 ));
78 #ifdef DEBUG_DUMP_FRAME
79 static void x264_frame_dump( x264_t *h, x264_frame_t *fr, char *name )
81 FILE *f = fopen( name, "r+b" );
86 /* Write the frame in display order */
87 fseek( f, fr->i_frame * h->param.i_height * h->param.i_width * 3 / 2, SEEK_SET );
89 for( i = 0; i < fr->i_plane; i++ )
91 for( y = 0; y < h->param.i_height / ( i == 0 ? 1 : 2 ); y++ )
93 fwrite( &fr->plane[i][y*fr->i_stride[i]], 1, h->param.i_width / ( i == 0 ? 1 : 2 ), f );
101 /* Fill "default" values */
102 static void x264_slice_header_init( x264_t *h, x264_slice_header_t *sh,
103 x264_sps_t *sps, x264_pps_t *pps,
104 int i_type, int i_idr_pic_id, int i_frame, int i_qp )
106 x264_param_t *param = &h->param;
109 /* First we fill all field */
115 sh->i_last_mb = h->sps->i_mb_width * h->sps->i_mb_height;
116 sh->i_pps_id = pps->i_id;
118 sh->i_frame_num = i_frame;
120 sh->b_mbaff = h->param.b_interlaced;
121 sh->b_field_pic = 0; /* no field support for now */
122 sh->b_bottom_field = 0; /* not yet used */
124 sh->i_idr_pic_id = i_idr_pic_id;
126 /* poc stuff, fixed later */
128 sh->i_delta_poc_bottom = 0;
129 sh->i_delta_poc[0] = 0;
130 sh->i_delta_poc[1] = 0;
132 sh->i_redundant_pic_cnt = 0;
134 if( !h->mb.b_direct_auto_read )
136 if( h->mb.b_direct_auto_write )
137 sh->b_direct_spatial_mv_pred = ( h->stat.i_direct_score[1] > h->stat.i_direct_score[0] );
139 sh->b_direct_spatial_mv_pred = ( param->analyse.i_direct_mv_pred == X264_DIRECT_PRED_SPATIAL );
141 /* else b_direct_spatial_mv_pred was read from the 2pass statsfile */
143 sh->b_num_ref_idx_override = 0;
144 sh->i_num_ref_idx_l0_active = 1;
145 sh->i_num_ref_idx_l1_active = 1;
147 sh->b_ref_pic_list_reordering_l0 = h->b_ref_reorder[0];
148 sh->b_ref_pic_list_reordering_l1 = h->b_ref_reorder[1];
150 /* If the ref list isn't in the default order, construct reordering header */
151 /* List1 reordering isn't needed yet */
152 if( sh->b_ref_pic_list_reordering_l0 )
154 int pred_frame_num = i_frame;
155 for( i = 0; i < h->i_ref0; i++ )
157 int diff = h->fref0[i]->i_frame_num - pred_frame_num;
159 x264_log( h, X264_LOG_ERROR, "diff frame num == 0\n" );
160 sh->ref_pic_list_order[0][i].idc = ( diff > 0 );
161 sh->ref_pic_list_order[0][i].arg = abs( diff ) - 1;
162 pred_frame_num = h->fref0[i]->i_frame_num;
166 sh->i_cabac_init_idc = param->i_cabac_init_idc;
169 sh->i_qp_delta = i_qp - pps->i_pic_init_qp;
170 sh->b_sp_for_swidth = 0;
173 /* If effective qp <= 15, deblocking would have no effect anyway */
174 if( param->b_deblocking_filter
175 && ( h->mb.b_variable_qp
176 || 15 < i_qp + 2 * X264_MAX(param->i_deblocking_filter_alphac0, param->i_deblocking_filter_beta) ) )
178 sh->i_disable_deblocking_filter_idc = 0;
182 sh->i_disable_deblocking_filter_idc = 1;
184 sh->i_alpha_c0_offset = param->i_deblocking_filter_alphac0 << 1;
185 sh->i_beta_offset = param->i_deblocking_filter_beta << 1;
188 static void x264_slice_header_write( bs_t *s, x264_slice_header_t *sh, int i_nal_ref_idc )
194 assert( sh->i_first_mb % (2*sh->sps->i_mb_width) == 0 );
195 bs_write_ue( s, sh->i_first_mb >> 1 );
198 bs_write_ue( s, sh->i_first_mb );
200 bs_write_ue( s, sh->i_type + 5 ); /* same type things */
201 bs_write_ue( s, sh->i_pps_id );
202 bs_write( s, sh->sps->i_log2_max_frame_num, sh->i_frame_num );
204 if( !sh->sps->b_frame_mbs_only )
206 bs_write1( s, sh->b_field_pic );
207 if ( sh->b_field_pic )
208 bs_write1( s, sh->b_bottom_field );
211 if( sh->i_idr_pic_id >= 0 ) /* NAL IDR */
213 bs_write_ue( s, sh->i_idr_pic_id );
216 if( sh->sps->i_poc_type == 0 )
218 bs_write( s, sh->sps->i_log2_max_poc_lsb, sh->i_poc_lsb );
219 if( sh->pps->b_pic_order && !sh->b_field_pic )
221 bs_write_se( s, sh->i_delta_poc_bottom );
224 else if( sh->sps->i_poc_type == 1 && !sh->sps->b_delta_pic_order_always_zero )
226 bs_write_se( s, sh->i_delta_poc[0] );
227 if( sh->pps->b_pic_order && !sh->b_field_pic )
229 bs_write_se( s, sh->i_delta_poc[1] );
233 if( sh->pps->b_redundant_pic_cnt )
235 bs_write_ue( s, sh->i_redundant_pic_cnt );
238 if( sh->i_type == SLICE_TYPE_B )
240 bs_write1( s, sh->b_direct_spatial_mv_pred );
242 if( sh->i_type == SLICE_TYPE_P || sh->i_type == SLICE_TYPE_SP || sh->i_type == SLICE_TYPE_B )
244 bs_write1( s, sh->b_num_ref_idx_override );
245 if( sh->b_num_ref_idx_override )
247 bs_write_ue( s, sh->i_num_ref_idx_l0_active - 1 );
248 if( sh->i_type == SLICE_TYPE_B )
250 bs_write_ue( s, sh->i_num_ref_idx_l1_active - 1 );
255 /* ref pic list reordering */
256 if( sh->i_type != SLICE_TYPE_I )
258 bs_write1( s, sh->b_ref_pic_list_reordering_l0 );
259 if( sh->b_ref_pic_list_reordering_l0 )
261 for( i = 0; i < sh->i_num_ref_idx_l0_active; i++ )
263 bs_write_ue( s, sh->ref_pic_list_order[0][i].idc );
264 bs_write_ue( s, sh->ref_pic_list_order[0][i].arg );
270 if( sh->i_type == SLICE_TYPE_B )
272 bs_write1( s, sh->b_ref_pic_list_reordering_l1 );
273 if( sh->b_ref_pic_list_reordering_l1 )
275 for( i = 0; i < sh->i_num_ref_idx_l1_active; i++ )
277 bs_write_ue( s, sh->ref_pic_list_order[1][i].idc );
278 bs_write_ue( s, sh->ref_pic_list_order[1][i].arg );
284 if( ( sh->pps->b_weighted_pred && ( sh->i_type == SLICE_TYPE_P || sh->i_type == SLICE_TYPE_SP ) ) ||
285 ( sh->pps->b_weighted_bipred == 1 && sh->i_type == SLICE_TYPE_B ) )
290 if( i_nal_ref_idc != 0 )
292 if( sh->i_idr_pic_id >= 0 )
294 bs_write1( s, 0 ); /* no output of prior pics flag */
295 bs_write1( s, 0 ); /* long term reference flag */
299 bs_write1( s, 0 ); /* adaptive_ref_pic_marking_mode_flag */
303 if( sh->pps->b_cabac && sh->i_type != SLICE_TYPE_I )
305 bs_write_ue( s, sh->i_cabac_init_idc );
307 bs_write_se( s, sh->i_qp_delta ); /* slice qp delta */
309 if( sh->pps->b_deblocking_filter_control )
311 bs_write_ue( s, sh->i_disable_deblocking_filter_idc );
312 if( sh->i_disable_deblocking_filter_idc != 1 )
314 bs_write_se( s, sh->i_alpha_c0_offset >> 1 );
315 bs_write_se( s, sh->i_beta_offset >> 1 );
320 /****************************************************************************
322 ****************************************************************************
323 ****************************** External API*********************************
324 ****************************************************************************
326 ****************************************************************************/
328 static int x264_validate_parameters( x264_t *h )
330 if( h->param.i_width <= 0 || h->param.i_height <= 0 )
332 x264_log( h, X264_LOG_ERROR, "invalid width x height (%dx%d)\n",
333 h->param.i_width, h->param.i_height );
337 if( h->param.i_width % 2 || h->param.i_height % 2 )
339 x264_log( h, X264_LOG_ERROR, "width or height not divisible by 2 (%dx%d)\n",
340 h->param.i_width, h->param.i_height );
343 if( h->param.i_csp != X264_CSP_I420 )
345 x264_log( h, X264_LOG_ERROR, "invalid CSP (only I420 supported)\n" );
349 if( h->param.i_threads == 0 )
350 h->param.i_threads = x264_cpu_num_processors();
351 h->param.i_threads = x264_clip3( h->param.i_threads, 1, X264_SLICE_MAX );
352 h->param.i_threads = X264_MIN( h->param.i_threads, (h->param.i_height + 15) >> (4 + h->param.b_interlaced) );
354 if( h->param.i_threads > 1 )
356 x264_log( h, X264_LOG_WARNING, "not compiled with pthread support!\n");
357 x264_log( h, X264_LOG_WARNING, "multislicing anyway, but you won't see any speed gain.\n" );
361 if( h->param.b_interlaced )
363 if( h->param.analyse.i_me_method == X264_ME_ESA )
365 x264_log( h, X264_LOG_WARNING, "interlace + me=esa is not implemented\n" );
366 h->param.analyse.i_me_method = X264_ME_UMH;
368 if( h->param.analyse.i_direct_mv_pred > X264_DIRECT_PRED_SPATIAL )
370 x264_log( h, X264_LOG_WARNING, "interlace + direct=temporal is not implemented\n" );
371 h->param.analyse.i_direct_mv_pred = X264_DIRECT_PRED_SPATIAL;
375 if( h->param.rc.i_rc_method < 0 || h->param.rc.i_rc_method > 2 )
377 x264_log( h, X264_LOG_ERROR, "invalid RC method\n" );
380 h->param.rc.i_rf_constant = x264_clip3( h->param.rc.i_rf_constant, 0, 51 );
381 h->param.rc.i_qp_constant = x264_clip3( h->param.rc.i_qp_constant, 0, 51 );
382 if( h->param.rc.i_rc_method == X264_RC_CRF )
383 h->param.rc.i_qp_constant = h->param.rc.i_rf_constant;
384 if( (h->param.rc.i_rc_method == X264_RC_CQP || h->param.rc.i_rc_method == X264_RC_CRF)
385 && h->param.rc.i_qp_constant == 0 )
387 h->mb.b_lossless = 1;
388 h->param.i_cqm_preset = X264_CQM_FLAT;
389 h->param.psz_cqm_file = NULL;
390 h->param.rc.i_rc_method = X264_RC_CQP;
391 h->param.rc.f_ip_factor = 1;
392 h->param.rc.f_pb_factor = 1;
393 h->param.analyse.b_transform_8x8 = 0;
394 h->param.analyse.b_psnr = 0;
395 h->param.analyse.b_ssim = 0;
396 h->param.analyse.i_chroma_qp_offset = 0;
397 h->param.analyse.i_trellis = 0;
398 h->param.analyse.b_fast_pskip = 0;
399 h->param.analyse.i_noise_reduction = 0;
400 h->param.analyse.i_subpel_refine = x264_clip3( h->param.analyse.i_subpel_refine, 1, 6 );
403 if( ( h->param.i_width % 16 || h->param.i_height % 16 ) && !h->mb.b_lossless )
405 x264_log( h, X264_LOG_WARNING,
406 "width or height not divisible by 16 (%dx%d), compression will suffer.\n",
407 h->param.i_width, h->param.i_height );
410 h->param.i_frame_reference = x264_clip3( h->param.i_frame_reference, 1, 16 );
411 if( h->param.i_keyint_max <= 0 )
412 h->param.i_keyint_max = 1;
413 h->param.i_keyint_min = x264_clip3( h->param.i_keyint_min, 1, h->param.i_keyint_max/2+1 );
415 h->param.i_bframe = x264_clip3( h->param.i_bframe, 0, X264_BFRAME_MAX );
416 h->param.i_bframe_bias = x264_clip3( h->param.i_bframe_bias, -90, 100 );
417 h->param.b_bframe_pyramid = h->param.b_bframe_pyramid && h->param.i_bframe > 1;
418 h->param.b_bframe_adaptive = h->param.b_bframe_adaptive && h->param.i_bframe > 0;
419 h->param.analyse.b_weighted_bipred = h->param.analyse.b_weighted_bipred && h->param.i_bframe > 0;
420 h->mb.b_direct_auto_write = h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO
422 && ( h->param.rc.b_stat_write || !h->param.rc.b_stat_read );
424 h->param.i_deblocking_filter_alphac0 = x264_clip3( h->param.i_deblocking_filter_alphac0, -6, 6 );
425 h->param.i_deblocking_filter_beta = x264_clip3( h->param.i_deblocking_filter_beta, -6, 6 );
427 h->param.i_cabac_init_idc = x264_clip3( h->param.i_cabac_init_idc, 0, 2 );
429 if( h->param.i_cqm_preset < X264_CQM_FLAT || h->param.i_cqm_preset > X264_CQM_CUSTOM )
430 h->param.i_cqm_preset = X264_CQM_FLAT;
432 if( h->param.analyse.i_me_method < X264_ME_DIA ||
433 h->param.analyse.i_me_method > X264_ME_ESA )
434 h->param.analyse.i_me_method = X264_ME_HEX;
435 if( h->param.analyse.i_me_range < 4 )
436 h->param.analyse.i_me_range = 4;
437 if( h->param.analyse.i_me_range > 16 && h->param.analyse.i_me_method <= X264_ME_HEX )
438 h->param.analyse.i_me_range = 16;
439 h->param.analyse.i_subpel_refine = x264_clip3( h->param.analyse.i_subpel_refine, 1, 7 );
440 h->param.analyse.b_bframe_rdo = h->param.analyse.b_bframe_rdo && h->param.analyse.i_subpel_refine >= 6;
441 h->param.analyse.b_mixed_references = h->param.analyse.b_mixed_references && h->param.i_frame_reference > 1;
442 h->param.analyse.inter &= X264_ANALYSE_PSUB16x16|X264_ANALYSE_PSUB8x8|X264_ANALYSE_BSUB16x16|
443 X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
444 h->param.analyse.intra &= X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
445 if( !(h->param.analyse.inter & X264_ANALYSE_PSUB16x16) )
446 h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
447 if( !h->param.analyse.b_transform_8x8 )
449 h->param.analyse.inter &= ~X264_ANALYSE_I8x8;
450 h->param.analyse.intra &= ~X264_ANALYSE_I8x8;
452 h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -12, 12);
453 if( !h->param.b_cabac )
454 h->param.analyse.i_trellis = 0;
455 h->param.analyse.i_trellis = x264_clip3( h->param.analyse.i_trellis, 0, 2 );
456 h->param.analyse.i_noise_reduction = x264_clip3( h->param.analyse.i_noise_reduction, 0, 1<<16 );
459 const x264_level_t *l = x264_levels;
460 while( l->level_idc != 0 && l->level_idc != h->param.i_level_idc )
462 if( l->level_idc == 0 )
464 x264_log( h, X264_LOG_ERROR, "invalid level_idc: %d\n", h->param.i_level_idc );
467 if( h->param.analyse.i_mv_range <= 0 )
468 h->param.analyse.i_mv_range = l->mv_range;
470 h->param.analyse.i_mv_range = x264_clip3(h->param.analyse.i_mv_range, 32, 2048);
473 if( h->param.rc.f_qblur < 0 )
474 h->param.rc.f_qblur = 0;
475 if( h->param.rc.f_complexity_blur < 0 )
476 h->param.rc.f_complexity_blur = 0;
478 h->param.i_sps_id &= 31;
480 if( h->param.i_log_level < X264_LOG_INFO )
482 h->param.analyse.b_psnr = 0;
483 h->param.analyse.b_ssim = 0;
486 /* ensure the booleans are 0 or 1 so they can be used in math */
487 #define BOOLIFY(x) h->param.x = !!h->param.x
489 BOOLIFY( b_deblocking_filter );
490 BOOLIFY( b_interlaced );
491 BOOLIFY( analyse.b_transform_8x8 );
492 BOOLIFY( analyse.b_bidir_me );
493 BOOLIFY( analyse.b_chroma_me );
494 BOOLIFY( analyse.b_fast_pskip );
495 BOOLIFY( rc.b_stat_write );
496 BOOLIFY( rc.b_stat_read );
502 static void mbcmp_init( x264_t *h )
504 memcpy( h->pixf.mbcmp,
505 ( h->mb.b_lossless || h->param.analyse.i_subpel_refine <= 1 ) ? h->pixf.sad : h->pixf.satd,
506 sizeof(h->pixf.mbcmp) );
509 /****************************************************************************
511 ****************************************************************************/
512 x264_t *x264_encoder_open ( x264_param_t *param )
514 x264_t *h = x264_malloc( sizeof( x264_t ) );
517 memset( h, 0, sizeof( x264_t ) );
519 /* Create a copy of param */
520 memcpy( &h->param, param, sizeof( x264_param_t ) );
522 if( x264_validate_parameters( h ) < 0 )
528 if( h->param.psz_cqm_file )
529 if( x264_cqm_parse_file( h, h->param.psz_cqm_file ) < 0 )
535 if( h->param.rc.psz_stat_out )
536 h->param.rc.psz_stat_out = strdup( h->param.rc.psz_stat_out );
537 if( h->param.rc.psz_stat_in )
538 h->param.rc.psz_stat_in = strdup( h->param.rc.psz_stat_in );
539 if( h->param.rc.psz_rc_eq )
540 h->param.rc.psz_rc_eq = strdup( h->param.rc.psz_rc_eq );
543 if( h->param.vui.i_sar_width > 0 && h->param.vui.i_sar_height > 0 )
545 int i_w = param->vui.i_sar_width;
546 int i_h = param->vui.i_sar_height;
548 x264_reduce_fraction( &i_w, &i_h );
550 while( i_w > 65535 || i_h > 65535 )
556 h->param.vui.i_sar_width = 0;
557 h->param.vui.i_sar_height = 0;
558 if( i_w == 0 || i_h == 0 )
560 x264_log( h, X264_LOG_WARNING, "cannot create valid sample aspect ratio\n" );
564 x264_log( h, X264_LOG_INFO, "using SAR=%d/%d\n", i_w, i_h );
565 h->param.vui.i_sar_width = i_w;
566 h->param.vui.i_sar_height = i_h;
570 x264_reduce_fraction( &h->param.i_fps_num, &h->param.i_fps_den );
574 h->out.i_bitstream = X264_MAX( 1000000, h->param.i_width * h->param.i_height * 1.7
575 * ( h->param.rc.i_rc_method == X264_RC_ABR ? pow( 0.5, h->param.rc.i_qp_min )
576 : pow( 0.5, h->param.rc.i_qp_constant ) * X264_MAX( 1, h->param.rc.f_ip_factor )));
577 h->out.p_bitstream = x264_malloc( h->out.i_bitstream );
583 h->sps = &h->sps_array[0];
584 x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
586 h->pps = &h->pps_array[0];
587 x264_pps_init( h->pps, h->param.i_sps_id, &h->param, h->sps);
589 x264_validate_levels( h );
593 h->mb.i_mb_count = h->sps->i_mb_width * h->sps->i_mb_height;
596 h->frames.i_delay = h->param.i_bframe;
597 h->frames.i_max_ref0 = h->param.i_frame_reference;
598 h->frames.i_max_ref1 = h->sps->vui.i_num_reorder_frames;
599 h->frames.i_max_dpb = h->sps->vui.i_max_dec_frame_buffering + 1;
600 h->frames.b_have_lowres = !h->param.rc.b_stat_read
601 && ( h->param.rc.i_rc_method == X264_RC_ABR
602 || h->param.rc.i_rc_method == X264_RC_CRF
603 || h->param.b_bframe_adaptive );
605 for( i = 0; i < X264_BFRAME_MAX + 3; i++ )
607 h->frames.current[i] = NULL;
608 h->frames.next[i] = NULL;
609 h->frames.unused[i] = NULL;
611 for( i = 0; i < 1 + h->frames.i_delay; i++ )
613 h->frames.unused[i] = x264_frame_new( h );
614 if( !h->frames.unused[i] )
617 for( i = 0; i < h->frames.i_max_dpb; i++ )
619 h->frames.reference[i] = x264_frame_new( h );
620 if( !h->frames.reference[i] )
623 h->frames.reference[h->frames.i_max_dpb] = NULL;
624 h->frames.i_last_idr = - h->param.i_keyint_max;
625 h->frames.i_input = 0;
626 h->frames.last_nonb = NULL;
631 h->fdec = h->frames.reference[0];
633 if( x264_macroblock_cache_init( h ) < 0 )
637 /* init CPU functions */
638 x264_predict_16x16_init( h->param.cpu, h->predict_16x16 );
639 x264_predict_8x8c_init( h->param.cpu, h->predict_8x8c );
640 x264_predict_8x8_init( h->param.cpu, h->predict_8x8 );
641 x264_predict_4x4_init( h->param.cpu, h->predict_4x4 );
643 x264_pixel_init( h->param.cpu, &h->pixf );
644 x264_dct_init( h->param.cpu, &h->dctf );
645 x264_mc_init( h->param.cpu, &h->mc );
646 x264_csp_init( h->param.cpu, h->param.i_csp, &h->csp );
647 x264_quant_init( h, h->param.cpu, &h->quantf );
648 x264_deblock_init( h->param.cpu, &h->loopf );
649 x264_dct_init_weights();
654 if( x264_ratecontrol_new( h ) < 0 )
657 x264_log( h, X264_LOG_INFO, "using cpu capabilities %s%s%s%s%s%s\n",
658 param->cpu&X264_CPU_MMX ? "MMX " : "",
659 param->cpu&X264_CPU_MMXEXT ? "MMXEXT " : "",
660 param->cpu&X264_CPU_SSE ? "SSE " : "",
661 param->cpu&X264_CPU_SSE2 ? "SSE2 " : "",
662 param->cpu&X264_CPU_3DNOW ? "3DNow! " : "",
663 param->cpu&X264_CPU_ALTIVEC ? "Altivec " : "" );
667 for( i = 1; i < h->param.i_threads; i++ )
668 h->thread[i] = x264_malloc( sizeof(x264_t) );
670 #ifdef DEBUG_DUMP_FRAME
672 /* create or truncate the reconstructed video file */
673 FILE *f = fopen( "fdec.yuv", "w" );
678 x264_log( h, X264_LOG_ERROR, "can't write to fdec.yuv\n" );
688 /****************************************************************************
689 * x264_encoder_reconfig:
690 ****************************************************************************/
691 int x264_encoder_reconfig( x264_t *h, x264_param_t *param )
693 #define COPY(var) h->param.var = param->var
694 COPY( i_frame_reference ); // but never uses more refs than initially specified
695 COPY( i_bframe_bias );
696 COPY( i_scenecut_threshold );
697 COPY( b_deblocking_filter );
698 COPY( i_deblocking_filter_alphac0 );
699 COPY( i_deblocking_filter_beta );
700 COPY( analyse.intra );
701 COPY( analyse.inter );
702 COPY( analyse.i_direct_mv_pred );
703 COPY( analyse.i_me_method );
704 COPY( analyse.i_me_range );
705 COPY( analyse.i_noise_reduction );
706 COPY( analyse.i_subpel_refine );
707 COPY( analyse.i_trellis );
708 COPY( analyse.b_bidir_me );
709 COPY( analyse.b_bframe_rdo );
710 COPY( analyse.b_chroma_me );
711 COPY( analyse.b_dct_decimate );
712 COPY( analyse.b_fast_pskip );
713 COPY( analyse.b_mixed_references );
716 if( h->pps->b_transform_8x8_mode )
717 h->param.analyse.b_transform_8x8 = param->analyse.b_transform_8x8;
721 return x264_validate_parameters( h );
725 static void x264_nal_start( x264_t *h, int i_type, int i_ref_idc )
727 x264_nal_t *nal = &h->out.nal[h->out.i_nal];
729 nal->i_ref_idc = i_ref_idc;
730 nal->i_type = i_type;
732 bs_align_0( &h->out.bs ); /* not needed */
735 nal->p_payload= &h->out.p_bitstream[bs_pos( &h->out.bs) / 8];
737 static void x264_nal_end( x264_t *h )
739 x264_nal_t *nal = &h->out.nal[h->out.i_nal];
741 bs_align_0( &h->out.bs ); /* not needed */
743 nal->i_payload = &h->out.p_bitstream[bs_pos( &h->out.bs)/8] - nal->p_payload;
748 /****************************************************************************
749 * x264_encoder_headers:
750 ****************************************************************************/
751 int x264_encoder_headers( x264_t *h, x264_nal_t **pp_nal, int *pi_nal )
753 /* init bitstream context */
755 bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
757 /* Put SPS and PPS */
758 if( h->i_frame == 0 )
760 /* identify ourself */
761 x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
762 x264_sei_version_write( h, &h->out.bs );
765 /* generate sequence parameters */
766 x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
767 x264_sps_write( &h->out.bs, h->sps );
770 /* generate picture parameters */
771 x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
772 x264_pps_write( &h->out.bs, h->pps );
776 *pi_nal = h->out.i_nal;
777 *pp_nal = &h->out.nal[0];
783 static void x264_frame_put( x264_frame_t *list[X264_BFRAME_MAX], x264_frame_t *frame )
786 while( list[i] ) i++;
790 static void x264_frame_push( x264_frame_t *list[X264_BFRAME_MAX], x264_frame_t *frame )
793 while( list[i] ) i++;
799 static x264_frame_t *x264_frame_get( x264_frame_t *list[X264_BFRAME_MAX+1] )
801 x264_frame_t *frame = list[0];
803 for( i = 0; list[i]; i++ )
808 static void x264_frame_sort( x264_frame_t *list[X264_BFRAME_MAX+1], int b_dts )
813 for( i = 0; list[i+1]; i++ )
815 int dtype = list[i]->i_type - list[i+1]->i_type;
816 int dtime = list[i]->i_frame - list[i+1]->i_frame;
817 int swap = b_dts ? dtype > 0 || ( dtype == 0 && dtime > 0 )
821 XCHG( x264_frame_t*, list[i], list[i+1] );
827 #define x264_frame_sort_dts(list) x264_frame_sort(list, 1)
828 #define x264_frame_sort_pts(list) x264_frame_sort(list, 0)
830 static inline void x264_reference_build_list( x264_t *h, int i_poc, int i_slice_type )
835 /* build ref list 0/1 */
838 for( i = 1; i < h->frames.i_max_dpb; i++ )
840 if( h->frames.reference[i]->i_poc >= 0 )
842 if( h->frames.reference[i]->i_poc < i_poc )
844 h->fref0[h->i_ref0++] = h->frames.reference[i];
846 else if( h->frames.reference[i]->i_poc > i_poc )
848 h->fref1[h->i_ref1++] = h->frames.reference[i];
853 /* Order ref0 from higher to lower poc */
857 for( i = 0; i < h->i_ref0 - 1; i++ )
859 if( h->fref0[i]->i_poc < h->fref0[i+1]->i_poc )
861 XCHG( x264_frame_t*, h->fref0[i], h->fref0[i+1] );
867 /* Order ref1 from lower to higher poc (bubble sort) for B-frame */
871 for( i = 0; i < h->i_ref1 - 1; i++ )
873 if( h->fref1[i]->i_poc > h->fref1[i+1]->i_poc )
875 XCHG( x264_frame_t*, h->fref1[i], h->fref1[i+1] );
882 /* In the standard, a P-frame's ref list is sorted by frame_num.
883 * We use POC, but check whether explicit reordering is needed */
884 h->b_ref_reorder[0] =
885 h->b_ref_reorder[1] = 0;
886 if( i_slice_type == SLICE_TYPE_P )
888 for( i = 0; i < h->i_ref0 - 1; i++ )
889 if( h->fref0[i]->i_frame_num < h->fref0[i+1]->i_frame_num )
891 h->b_ref_reorder[0] = 1;
896 h->i_ref1 = X264_MIN( h->i_ref1, h->frames.i_max_ref1 );
897 h->i_ref0 = X264_MIN( h->i_ref0, h->frames.i_max_ref0 );
898 h->i_ref0 = X264_MIN( h->i_ref0, h->param.i_frame_reference ); // if reconfig() has lowered the limit
899 h->i_ref0 = X264_MIN( h->i_ref0, 16 - h->i_ref1 );
900 h->mb.pic.i_fref[0] = h->i_ref0;
901 h->mb.pic.i_fref[1] = h->i_ref1;
904 static inline void x264_fdec_deblock( x264_t *h )
906 /* apply deblocking filter to the current decoded picture */
907 if( !h->sh.i_disable_deblocking_filter_idc )
909 TIMER_START( i_mtime_filter );
910 x264_frame_deblocking_filter( h, h->sh.i_type );
911 TIMER_STOP( i_mtime_filter );
915 static inline void x264_reference_update( x264_t *h )
919 x264_fdec_deblock( h );
922 x264_frame_expand_border( h, h->fdec );
924 /* create filtered images */
925 x264_frame_filter( h->param.cpu, h->fdec, h->sh.b_mbaff );
927 /* expand border of filtered images */
928 x264_frame_expand_border_filtered( h, h->fdec );
930 /* move lowres copy of the image to the ref frame */
931 for( i = 0; i < 4; i++)
933 XCHG( uint8_t*, h->fdec->lowres[i], h->fenc->lowres[i] );
934 XCHG( uint8_t*, h->fdec->buffer_lowres[i], h->fenc->buffer_lowres[i] );
937 /* adaptive B decision needs a pointer, since it can't use the ref lists */
938 if( h->sh.i_type != SLICE_TYPE_B )
939 h->frames.last_nonb = h->fdec;
941 /* move frame in the buffer */
942 h->fdec = h->frames.reference[h->frames.i_max_dpb-1];
943 for( i = h->frames.i_max_dpb-1; i > 0; i-- )
945 h->frames.reference[i] = h->frames.reference[i-1];
947 h->frames.reference[0] = h->fdec;
950 static inline void x264_reference_reset( x264_t *h )
954 /* reset ref pictures */
955 for( i = 1; i < h->frames.i_max_dpb; i++ )
957 h->frames.reference[i]->i_poc = -1;
959 h->frames.reference[0]->i_poc = 0;
962 static inline void x264_slice_init( x264_t *h, int i_nal_type, int i_slice_type, int i_global_qp )
964 /* ------------------------ Create slice header ----------------------- */
965 if( i_nal_type == NAL_SLICE_IDR )
967 x264_slice_header_init( h, &h->sh, h->sps, h->pps, i_slice_type, h->i_idr_pic_id, h->i_frame_num, i_global_qp );
970 h->i_idr_pic_id = ( h->i_idr_pic_id + 1 ) % 65536;
974 x264_slice_header_init( h, &h->sh, h->sps, h->pps, i_slice_type, -1, h->i_frame_num, i_global_qp );
976 /* always set the real higher num of ref frame used */
977 h->sh.b_num_ref_idx_override = 1;
978 h->sh.i_num_ref_idx_l0_active = h->i_ref0 <= 0 ? 1 : h->i_ref0;
979 h->sh.i_num_ref_idx_l1_active = h->i_ref1 <= 0 ? 1 : h->i_ref1;
982 h->fdec->i_frame_num = h->sh.i_frame_num;
984 if( h->sps->i_poc_type == 0 )
986 h->sh.i_poc_lsb = h->fdec->i_poc & ( (1 << h->sps->i_log2_max_poc_lsb) - 1 );
987 h->sh.i_delta_poc_bottom = 0; /* XXX won't work for field */
989 else if( h->sps->i_poc_type == 1 )
991 /* FIXME TODO FIXME */
995 /* Nothing to do ? */
998 x264_macroblock_slice_init( h );
1001 static int x264_slice_write( x264_t *h )
1008 memset( &h->stat.frame, 0, sizeof(h->stat.frame) );
1011 x264_nal_start( h, h->i_nal_type, h->i_nal_ref_idc );
1014 x264_slice_header_write( &h->out.bs, &h->sh, h->i_nal_ref_idc );
1015 if( h->param.b_cabac )
1017 /* alignment needed */
1018 bs_align_1( &h->out.bs );
1021 x264_cabac_context_init( &h->cabac, h->sh.i_type, h->sh.i_qp, h->sh.i_cabac_init_idc );
1022 x264_cabac_encode_init ( &h->cabac, &h->out.bs );
1024 h->mb.i_last_qp = h->sh.i_qp;
1025 h->mb.i_last_dqp = 0;
1027 for( mb_xy = h->sh.i_first_mb, i_skip = 0; mb_xy < h->sh.i_last_mb; )
1029 const int i_mb_y = mb_xy / h->sps->i_mb_width;
1030 const int i_mb_x = mb_xy % h->sps->i_mb_width;
1032 int mb_spos = bs_pos(&h->out.bs);
1035 x264_macroblock_cache_load( h, i_mb_x, i_mb_y );
1037 /* analyse parameters
1038 * Slice I: choose I_4x4 or I_16x16 mode
1039 * Slice P: choose between using P mode or intra (4x4 or 16x16)
1041 TIMER_START( i_mtime_analyse );
1042 x264_macroblock_analyse( h );
1043 TIMER_STOP( i_mtime_analyse );
1045 /* encode this macroblock -> be careful it can change the mb type to P_SKIP if needed */
1046 TIMER_START( i_mtime_encode );
1047 x264_macroblock_encode( h );
1048 TIMER_STOP( i_mtime_encode );
1050 TIMER_START( i_mtime_write );
1051 if( h->param.b_cabac )
1053 if( mb_xy > h->sh.i_first_mb && !(h->sh.b_mbaff && (i_mb_y&1)) )
1054 x264_cabac_encode_terminal( &h->cabac, 0 );
1056 if( IS_SKIP( h->mb.i_type ) )
1057 x264_cabac_mb_skip( h, 1 );
1060 if( h->sh.i_type != SLICE_TYPE_I )
1061 x264_cabac_mb_skip( h, 0 );
1062 x264_macroblock_write_cabac( h, &h->cabac );
1067 if( IS_SKIP( h->mb.i_type ) )
1071 if( h->sh.i_type != SLICE_TYPE_I )
1073 bs_write_ue( &h->out.bs, i_skip ); /* skip run */
1076 x264_macroblock_write_cavlc( h, &h->out.bs );
1079 TIMER_STOP( i_mtime_write );
1082 if( h->param.b_visualize )
1083 x264_visualize_mb( h );
1087 x264_macroblock_cache_save( h );
1089 /* accumulate mb stats */
1090 h->stat.frame.i_mb_count[h->mb.i_type]++;
1091 if( !IS_SKIP(h->mb.i_type) && !IS_INTRA(h->mb.i_type) && !IS_DIRECT(h->mb.i_type) )
1093 if( h->mb.i_partition != D_8x8 )
1094 h->stat.frame.i_mb_count_size[ x264_mb_partition_pixel_table[ h->mb.i_partition ] ] += 4;
1096 for( i = 0; i < 4; i++ )
1097 h->stat.frame.i_mb_count_size[ x264_mb_partition_pixel_table[ h->mb.i_sub_partition[i] ] ] ++;
1098 if( h->param.i_frame_reference > 1 )
1100 for( i = 0; i < 4; i++ )
1102 int i_ref = h->mb.cache.ref[0][ x264_scan8[4*i] ];
1104 h->stat.frame.i_mb_count_ref[i_ref] ++;
1108 if( h->mb.i_cbp_luma && !IS_INTRA(h->mb.i_type) )
1110 h->stat.frame.i_mb_count_8x8dct[0] ++;
1111 h->stat.frame.i_mb_count_8x8dct[1] += h->mb.b_transform_8x8;
1114 if( h->mb.b_variable_qp )
1115 x264_ratecontrol_mb(h, bs_pos(&h->out.bs) - mb_spos);
1119 if( (i_mb_y&1) && i_mb_x == h->sps->i_mb_width - 1 )
1122 mb_xy += 1 - h->sps->i_mb_width;
1124 mb_xy += h->sps->i_mb_width;
1130 if( h->param.b_cabac )
1133 x264_cabac_encode_terminal( &h->cabac, 1 );
1135 else if( i_skip > 0 )
1137 bs_write_ue( &h->out.bs, i_skip ); /* last skip run */
1140 if( h->param.b_cabac )
1142 x264_cabac_encode_flush( &h->cabac );
1147 /* rbsp_slice_trailing_bits */
1148 bs_rbsp_trailing( &h->out.bs );
1153 /* Compute misc bits */
1154 h->stat.frame.i_misc_bits = bs_pos( &h->out.bs )
1156 - h->stat.frame.i_itex_bits
1157 - h->stat.frame.i_ptex_bits
1158 - h->stat.frame.i_hdr_bits;
1163 static inline int x264_slices_write( x264_t *h )
1168 if( h->param.b_visualize )
1169 x264_visualize_init( h );
1172 if( h->param.i_threads == 1 )
1174 x264_ratecontrol_threads_start( h );
1175 x264_slice_write( h );
1176 i_frame_size = h->out.nal[h->out.i_nal-1].i_payload;
1180 int i_nal = h->out.i_nal;
1181 int i_bs_size = h->out.i_bitstream / h->param.i_threads;
1183 /* duplicate contexts */
1184 for( i = 0; i < h->param.i_threads; i++ )
1186 x264_t *t = h->thread[i];
1187 int mb_height = h->sps->i_mb_height >> h->sh.b_mbaff;
1188 int mb_width = h->sps->i_mb_width << h->sh.b_mbaff;
1191 memcpy( t, h, sizeof(x264_t) );
1192 t->out.p_bitstream += i*i_bs_size;
1193 bs_init( &t->out.bs, t->out.p_bitstream, i_bs_size );
1194 t->i_thread_num = i;
1196 t->sh.i_first_mb = (i * mb_height / h->param.i_threads) * mb_width;
1197 t->sh.i_last_mb = ((i+1) * mb_height / h->param.i_threads) * mb_width;
1198 t->out.i_nal = i_nal + i;
1200 x264_ratecontrol_threads_start( h );
1205 pthread_t handles[X264_SLICE_MAX];
1206 for( i = 0; i < h->param.i_threads; i++ )
1207 pthread_create( &handles[i], NULL, (void*)x264_slice_write, (void*)h->thread[i] );
1208 for( i = 0; i < h->param.i_threads; i++ )
1209 pthread_join( handles[i], NULL );
1212 for( i = 0; i < h->param.i_threads; i++ )
1213 x264_slice_write( h->thread[i] );
1216 /* merge contexts */
1217 i_frame_size = h->out.nal[i_nal].i_payload;
1218 for( i = 1; i < h->param.i_threads; i++ )
1221 x264_t *t = h->thread[i];
1222 h->out.nal[i_nal+i] = t->out.nal[i_nal+i];
1223 i_frame_size += t->out.nal[i_nal+i].i_payload;
1224 // all entries in stat.frame are ints
1225 for( j = 0; j < sizeof(h->stat.frame) / sizeof(int); j++ )
1226 ((int*)&h->stat.frame)[j] += ((int*)&t->stat.frame)[j];
1228 h->out.i_nal = i_nal + h->param.i_threads;
1232 if( h->param.b_visualize )
1234 x264_visualize_show( h );
1235 x264_visualize_close( h );
1239 return i_frame_size;
1242 /****************************************************************************
1243 * x264_encoder_encode:
1244 * XXX: i_poc : is the poc of the current given picture
1245 * i_frame : is the number of the frame being coded
1246 * ex: type frame poc
1254 ****************************************************************************/
1255 int x264_encoder_encode( x264_t *h,
1256 x264_nal_t **pp_nal, int *pi_nal,
1257 x264_picture_t *pic_in,
1258 x264_picture_t *pic_out )
1260 x264_frame_t *frame_psnr = h->fdec; /* just to keep the current decoded frame for psnr calculation */
1270 char psz_message[80];
1277 /* ------------------- Setup new frame from picture -------------------- */
1278 TIMER_START( i_mtime_encode_frame );
1279 if( pic_in != NULL )
1281 /* 1: Copy the picture to a frame and move it to a buffer */
1282 x264_frame_t *fenc = x264_frame_get( h->frames.unused );
1284 x264_frame_copy_picture( h, fenc, pic_in );
1286 if( h->param.i_width != 16 * h->sps->i_mb_width ||
1287 h->param.i_height != 16 * h->sps->i_mb_height )
1288 x264_frame_expand_border_mod16( h, fenc );
1290 fenc->i_frame = h->frames.i_input++;
1292 x264_frame_put( h->frames.next, fenc );
1294 if( h->frames.b_have_lowres )
1295 x264_frame_init_lowres( h->param.cpu, fenc );
1297 if( h->frames.i_input <= h->frames.i_delay )
1299 /* Nothing yet to encode */
1300 /* waiting for filling bframe buffer */
1301 pic_out->i_type = X264_TYPE_AUTO;
1306 if( h->frames.current[0] == NULL )
1309 /* 2: Select frame types */
1310 if( h->frames.next[0] == NULL )
1313 x264_slicetype_decide( h );
1315 /* 3: move some B-frames and 1 non-B to encode queue */
1316 while( IS_X264_TYPE_B( h->frames.next[bframes]->i_type ) )
1318 x264_frame_put( h->frames.current, x264_frame_get( &h->frames.next[bframes] ) );
1319 /* FIXME: when max B-frames > 3, BREF may no longer be centered after GOP closing */
1320 if( h->param.b_bframe_pyramid && bframes > 1 )
1322 x264_frame_t *mid = x264_frame_get( &h->frames.next[bframes/2] );
1323 mid->i_type = X264_TYPE_BREF;
1324 x264_frame_put( h->frames.current, mid );
1328 x264_frame_put( h->frames.current, x264_frame_get( h->frames.next ) );
1330 TIMER_STOP( i_mtime_encode_frame );
1332 /* ------------------- Get frame to be encoded ------------------------- */
1333 /* 4: get picture to encode */
1334 h->fenc = x264_frame_get( h->frames.current );
1335 if( h->fenc == NULL )
1337 /* Nothing yet to encode (ex: waiting for I/P with B frames) */
1338 /* waiting for filling bframe buffer */
1339 pic_out->i_type = X264_TYPE_AUTO;
1345 if( h->fenc->i_type == X264_TYPE_IDR )
1347 h->frames.i_last_idr = h->fenc->i_frame;
1350 /* ------------------- Setup frame context ----------------------------- */
1351 /* 5: Init data dependent of frame type */
1352 TIMER_START( i_mtime_encode_frame );
1353 if( h->fenc->i_type == X264_TYPE_IDR )
1355 /* reset ref pictures */
1356 x264_reference_reset( h );
1358 i_nal_type = NAL_SLICE_IDR;
1359 i_nal_ref_idc = NAL_PRIORITY_HIGHEST;
1360 i_slice_type = SLICE_TYPE_I;
1362 else if( h->fenc->i_type == X264_TYPE_I )
1364 i_nal_type = NAL_SLICE;
1365 i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
1366 i_slice_type = SLICE_TYPE_I;
1368 else if( h->fenc->i_type == X264_TYPE_P )
1370 i_nal_type = NAL_SLICE;
1371 i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
1372 i_slice_type = SLICE_TYPE_P;
1374 else if( h->fenc->i_type == X264_TYPE_BREF )
1376 i_nal_type = NAL_SLICE;
1377 i_nal_ref_idc = NAL_PRIORITY_HIGH; /* maybe add MMCO to forget it? -> low */
1378 i_slice_type = SLICE_TYPE_B;
1382 i_nal_type = NAL_SLICE;
1383 i_nal_ref_idc = NAL_PRIORITY_DISPOSABLE;
1384 i_slice_type = SLICE_TYPE_B;
1388 h->fenc->i_poc = 2 * (h->fenc->i_frame - h->frames.i_last_idr);
1389 h->fdec->i_type = h->fenc->i_type;
1390 h->fdec->i_frame = h->fenc->i_frame;
1391 h->fenc->b_kept_as_ref =
1392 h->fdec->b_kept_as_ref = i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE;
1396 /* ------------------- Init ----------------------------- */
1397 /* build ref list 0/1 */
1398 x264_reference_build_list( h, h->fdec->i_poc, i_slice_type );
1400 /* Init the rate control */
1401 x264_ratecontrol_start( h, i_slice_type, h->fenc->i_qpplus1 );
1402 i_global_qp = x264_ratecontrol_qp( h );
1404 pic_out->i_qpplus1 =
1405 h->fdec->i_qpplus1 = i_global_qp + 1;
1407 if( i_slice_type == SLICE_TYPE_B )
1408 x264_macroblock_bipred_init( h );
1410 /* ------------------------ Create slice header ----------------------- */
1411 x264_slice_init( h, i_nal_type, i_slice_type, i_global_qp );
1413 if( h->fenc->b_kept_as_ref )
1416 /* ---------------------- Write the bitstream -------------------------- */
1417 /* Init bitstream context */
1419 bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
1424 if(i_slice_type == SLICE_TYPE_I)
1426 else if(i_slice_type == SLICE_TYPE_P)
1428 else if(i_slice_type == SLICE_TYPE_B)
1433 x264_nal_start(h, NAL_AUD, NAL_PRIORITY_DISPOSABLE);
1434 bs_write(&h->out.bs, 3, pic_type);
1435 bs_rbsp_trailing(&h->out.bs);
1439 h->i_nal_type = i_nal_type;
1440 h->i_nal_ref_idc = i_nal_ref_idc;
1442 /* Write SPS and PPS */
1443 if( i_nal_type == NAL_SLICE_IDR && h->param.b_repeat_headers )
1445 if( h->fenc->i_frame == 0 )
1447 /* identify ourself */
1448 x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
1449 x264_sei_version_write( h, &h->out.bs );
1453 /* generate sequence parameters */
1454 x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
1455 x264_sps_write( &h->out.bs, h->sps );
1458 /* generate picture parameters */
1459 x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
1460 x264_pps_write( &h->out.bs, h->pps );
1465 i_frame_size = x264_slices_write( h );
1467 /* restore CPU state (before using float again) */
1468 x264_cpu_restore( h->param.cpu );
1470 if( i_slice_type == SLICE_TYPE_P && !h->param.rc.b_stat_read
1471 && h->param.i_scenecut_threshold >= 0 )
1473 const int *mbs = h->stat.frame.i_mb_count;
1474 int i_mb_i = mbs[I_16x16] + mbs[I_8x8] + mbs[I_4x4];
1475 int i_mb_p = mbs[P_L0] + mbs[P_8x8];
1476 int i_mb_s = mbs[P_SKIP];
1477 int i_mb = h->sps->i_mb_width * h->sps->i_mb_height;
1478 int64_t i_inter_cost = h->stat.frame.i_inter_cost;
1479 int64_t i_intra_cost = h->stat.frame.i_intra_cost;
1482 int i_gop_size = h->fenc->i_frame - h->frames.i_last_idr;
1483 float f_thresh_max = h->param.i_scenecut_threshold / 100.0;
1484 /* magic numbers pulled out of thin air */
1485 float f_thresh_min = f_thresh_max * h->param.i_keyint_min
1486 / ( h->param.i_keyint_max * 4 );
1487 if( h->param.i_keyint_min == h->param.i_keyint_max )
1488 f_thresh_min= f_thresh_max;
1490 /* macroblock_analyse() doesn't further analyse skipped mbs,
1491 * so we have to guess their cost */
1493 i_intra_cost = i_intra_cost * i_mb / (i_mb - i_mb_s);
1495 if( i_gop_size < h->param.i_keyint_min / 4 )
1496 f_bias = f_thresh_min / 4;
1497 else if( i_gop_size <= h->param.i_keyint_min )
1498 f_bias = f_thresh_min * i_gop_size / h->param.i_keyint_min;
1501 f_bias = f_thresh_min
1502 + ( f_thresh_max - f_thresh_min )
1503 * ( i_gop_size - h->param.i_keyint_min )
1504 / ( h->param.i_keyint_max - h->param.i_keyint_min );
1506 f_bias = X264_MIN( f_bias, 1.0 );
1508 /* Bad P will be reencoded as I */
1509 if( i_mb_s < i_mb &&
1510 i_inter_cost >= (1.0 - f_bias) * i_intra_cost )
1514 x264_log( h, X264_LOG_DEBUG, "scene cut at %d Icost:%.0f Pcost:%.0f ratio:%.3f bias=%.3f lastIDR:%d (I:%d P:%d S:%d)\n",
1516 (double)i_intra_cost, (double)i_inter_cost,
1517 (double)i_inter_cost / i_intra_cost,
1519 i_mb_i, i_mb_p, i_mb_s );
1521 /* Restore frame num */
1524 for( b = 0; h->frames.current[b] && IS_X264_TYPE_B( h->frames.current[b]->i_type ); b++ );
1527 /* If using B-frames, force GOP to be closed.
1528 * Even if this frame is going to be I and not IDR, forcing a
1529 * P-frame before the scenecut will probably help compression.
1531 * We don't yet know exactly which frame is the scene cut, so
1532 * we can't assign an I-frame. Instead, change the previous
1533 * B-frame to P, and rearrange coding order. */
1535 if( h->param.b_bframe_adaptive || b > 1 )
1536 h->fenc->i_type = X264_TYPE_AUTO;
1537 x264_frame_sort_pts( h->frames.current );
1538 x264_frame_push( h->frames.next, h->fenc );
1539 h->fenc = h->frames.current[b-1];
1540 h->frames.current[b-1] = NULL;
1541 h->fenc->i_type = X264_TYPE_P;
1542 x264_frame_sort_dts( h->frames.current );
1544 /* Do IDR if needed */
1545 else if( i_gop_size >= h->param.i_keyint_min )
1552 /* Reinit field of fenc */
1553 h->fenc->i_type = X264_TYPE_IDR;
1556 /* Put enqueued frames back in the pool */
1557 while( (tmp = x264_frame_get( h->frames.current ) ) != NULL )
1558 x264_frame_put( h->frames.next, tmp );
1559 x264_frame_sort_pts( h->frames.next );
1563 h->fenc->i_type = X264_TYPE_I;
1569 /* End bitstream, set output */
1570 *pi_nal = h->out.i_nal;
1571 *pp_nal = h->out.nal;
1573 /* Set output picture properties */
1574 if( i_slice_type == SLICE_TYPE_I )
1575 pic_out->i_type = i_nal_type == NAL_SLICE_IDR ? X264_TYPE_IDR : X264_TYPE_I;
1576 else if( i_slice_type == SLICE_TYPE_P )
1577 pic_out->i_type = X264_TYPE_P;
1579 pic_out->i_type = X264_TYPE_B;
1580 pic_out->i_pts = h->fenc->i_pts;
1582 pic_out->img.i_plane = h->fdec->i_plane;
1583 for(i = 0; i < 4; i++){
1584 pic_out->img.i_stride[i] = h->fdec->i_stride[i];
1585 pic_out->img.plane[i] = h->fdec->plane[i];
1588 /* ---------------------- Update encoder state ------------------------- */
1591 x264_cpu_restore( h->param.cpu );
1592 x264_ratecontrol_end( h, i_frame_size * 8 );
1594 /* handle references */
1595 if( i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE )
1596 x264_reference_update( h );
1597 #ifdef DEBUG_DUMP_FRAME
1599 x264_fdec_deblock( h );
1601 x264_frame_put( h->frames.unused, h->fenc );
1603 /* increase frame count */
1606 /* restore CPU state (before using float again) */
1607 x264_cpu_restore( h->param.cpu );
1609 x264_noise_reduction_update( h );
1611 TIMER_STOP( i_mtime_encode_frame );
1613 /* ---------------------- Compute/Print statistics --------------------- */
1615 h->stat.i_slice_count[i_slice_type]++;
1616 h->stat.i_slice_size[i_slice_type] += i_frame_size + NALU_OVERHEAD;
1617 h->stat.i_slice_qp[i_slice_type] += i_global_qp;
1619 for( i = 0; i < X264_MBTYPE_MAX; i++ )
1620 h->stat.i_mb_count[h->sh.i_type][i] += h->stat.frame.i_mb_count[i];
1621 for( i = 0; i < 2; i++ )
1622 h->stat.i_mb_count_8x8dct[i] += h->stat.frame.i_mb_count_8x8dct[i];
1623 if( h->sh.i_type != SLICE_TYPE_I )
1625 for( i = 0; i < 7; i++ )
1626 h->stat.i_mb_count_size[h->sh.i_type][i] += h->stat.frame.i_mb_count_size[i];
1627 for( i = 0; i < 32; i++ )
1628 h->stat.i_mb_count_ref[h->sh.i_type][i] += h->stat.frame.i_mb_count_ref[i];
1630 if( i_slice_type == SLICE_TYPE_B )
1632 h->stat.i_direct_frames[ h->sh.b_direct_spatial_mv_pred ] ++;
1633 if( h->mb.b_direct_auto_write )
1635 //FIXME somewhat arbitrary time constants
1636 if( h->stat.i_direct_score[0] + h->stat.i_direct_score[1] > h->mb.i_mb_count )
1638 for( i = 0; i < 2; i++ )
1639 h->stat.i_direct_score[i] = h->stat.i_direct_score[i] * 9/10;
1641 for( i = 0; i < 2; i++ )
1642 h->stat.i_direct_score[i] += h->stat.frame.i_direct_score[i];
1646 psz_message[0] = '\0';
1647 if( h->param.analyse.b_psnr )
1651 for( i=0; i<3; i++ )
1653 sqe[i] = x264_pixel_ssd_wxh( &h->pixf,
1654 frame_psnr->plane[i], frame_psnr->i_stride[i],
1655 h->fenc->plane[i], h->fenc->i_stride[i],
1656 h->param.i_width >> !!i, h->param.i_height >> !!i );
1658 x264_cpu_restore( h->param.cpu );
1660 h->stat.i_sqe_global[i_slice_type] += sqe[0] + sqe[1] + sqe[2];
1661 h->stat.f_psnr_average[i_slice_type] += x264_psnr( sqe[0] + sqe[1] + sqe[2], 3 * h->param.i_width * h->param.i_height / 2 );
1662 h->stat.f_psnr_mean_y[i_slice_type] += x264_psnr( sqe[0], h->param.i_width * h->param.i_height );
1663 h->stat.f_psnr_mean_u[i_slice_type] += x264_psnr( sqe[1], h->param.i_width * h->param.i_height / 4 );
1664 h->stat.f_psnr_mean_v[i_slice_type] += x264_psnr( sqe[2], h->param.i_width * h->param.i_height / 4 );
1666 snprintf( psz_message, 80, " PSNR Y:%5.2f U:%5.2f V:%5.2f",
1667 x264_psnr( sqe[0], h->param.i_width * h->param.i_height ),
1668 x264_psnr( sqe[1], h->param.i_width * h->param.i_height / 4),
1669 x264_psnr( sqe[2], h->param.i_width * h->param.i_height / 4) );
1672 if( h->param.analyse.b_ssim )
1674 // offset by 2 pixels to avoid alignment of ssim blocks with dct blocks
1675 float ssim_y = x264_pixel_ssim_wxh( &h->pixf,
1676 frame_psnr->plane[0] + 2+2*frame_psnr->i_stride[0], frame_psnr->i_stride[0],
1677 h->fenc->plane[0] + 2+2*h->fenc->i_stride[0], h->fenc->i_stride[0],
1678 h->param.i_width-2, h->param.i_height-2 );
1679 h->stat.f_ssim_mean_y[i_slice_type] += ssim_y;
1680 snprintf( psz_message + strlen(psz_message), 80 - strlen(psz_message),
1681 " SSIM Y:%.5f", ssim_y );
1683 psz_message[79] = '\0';
1685 x264_log( h, X264_LOG_DEBUG,
1686 "frame=%4d QP=%i NAL=%d Slice:%c Poc:%-3d I:%-4d P:%-4d SKIP:%-4d size=%d bytes%s\n",
1690 i_slice_type == SLICE_TYPE_I ? 'I' : (i_slice_type == SLICE_TYPE_P ? 'P' : 'B' ),
1692 h->stat.frame.i_mb_count_i,
1693 h->stat.frame.i_mb_count_p,
1694 h->stat.frame.i_mb_count_skip,
1699 #ifdef DEBUG_MB_TYPE
1701 static const char mb_chars[] = { 'i', 'i', 'I', 'C', 'P', '8', 'S',
1702 'D', '<', 'X', 'B', 'X', '>', 'B', 'B', 'B', 'B', '8', 'S' };
1704 for( mb_xy = 0; mb_xy < h->sps->i_mb_width * h->sps->i_mb_height; mb_xy++ )
1706 if( h->mb.type[mb_xy] < X264_MBTYPE_MAX && h->mb.type[mb_xy] >= 0 )
1707 fprintf( stderr, "%c ", mb_chars[ h->mb.type[mb_xy] ] );
1709 fprintf( stderr, "? " );
1711 if( (mb_xy+1) % h->sps->i_mb_width == 0 )
1712 fprintf( stderr, "\n" );
1717 #ifdef DEBUG_DUMP_FRAME
1718 /* Dump reconstructed frame */
1719 x264_frame_dump( h, frame_psnr, "fdec.yuv" );
1724 /****************************************************************************
1725 * x264_encoder_close:
1726 ****************************************************************************/
1727 void x264_encoder_close ( x264_t *h )
1729 #ifdef DEBUG_BENCHMARK
1730 int64_t i_mtime_total = i_mtime_analyse + i_mtime_encode + i_mtime_write + i_mtime_filter + 1;
1732 int64_t i_yuv_size = 3 * h->param.i_width * h->param.i_height / 2;
1735 #ifdef DEBUG_BENCHMARK
1736 x264_log( h, X264_LOG_INFO,
1737 "analyse=%d(%lldms) encode=%d(%lldms) write=%d(%lldms) filter=%d(%lldms)\n",
1738 (int)(100*i_mtime_analyse/i_mtime_total), i_mtime_analyse/1000,
1739 (int)(100*i_mtime_encode/i_mtime_total), i_mtime_encode/1000,
1740 (int)(100*i_mtime_write/i_mtime_total), i_mtime_write/1000,
1741 (int)(100*i_mtime_filter/i_mtime_total), i_mtime_filter/1000 );
1744 /* Slices used and PSNR */
1745 for( i=0; i<5; i++ )
1747 static const int slice_order[] = { SLICE_TYPE_I, SLICE_TYPE_SI, SLICE_TYPE_P, SLICE_TYPE_SP, SLICE_TYPE_B };
1748 static const char *slice_name[] = { "P", "B", "I", "SP", "SI" };
1749 int i_slice = slice_order[i];
1751 if( h->stat.i_slice_count[i_slice] > 0 )
1753 const int i_count = h->stat.i_slice_count[i_slice];
1754 if( h->param.analyse.b_psnr )
1756 x264_log( h, X264_LOG_INFO,
1757 "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",
1758 slice_name[i_slice],
1760 (double)h->stat.i_slice_qp[i_slice] / i_count,
1761 (double)h->stat.i_slice_size[i_slice] / i_count,
1762 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,
1763 h->stat.f_psnr_average[i_slice] / i_count,
1764 x264_psnr( h->stat.i_sqe_global[i_slice], i_count * i_yuv_size ) );
1768 x264_log( h, X264_LOG_INFO,
1769 "slice %s:%-5d Avg QP:%5.2f size:%6.0f\n",
1770 slice_name[i_slice],
1772 (double)h->stat.i_slice_qp[i_slice] / i_count,
1773 (double)h->stat.i_slice_size[i_slice] / i_count );
1779 if( h->stat.i_slice_count[SLICE_TYPE_I] > 0 )
1781 const int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_I];
1782 const double i_count = h->stat.i_slice_count[SLICE_TYPE_I] * h->mb.i_mb_count / 100.0;
1783 x264_log( h, X264_LOG_INFO,
1784 "mb I I16..4: %4.1f%% %4.1f%% %4.1f%%\n",
1785 i_mb_count[I_16x16]/ i_count,
1786 i_mb_count[I_8x8] / i_count,
1787 i_mb_count[I_4x4] / i_count );
1789 if( h->stat.i_slice_count[SLICE_TYPE_P] > 0 )
1791 const int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_P];
1792 const int64_t *i_mb_size = h->stat.i_mb_count_size[SLICE_TYPE_P];
1793 const double i_count = h->stat.i_slice_count[SLICE_TYPE_P] * h->mb.i_mb_count / 100.0;
1794 x264_log( h, X264_LOG_INFO,
1795 "mb P I16..4: %4.1f%% %4.1f%% %4.1f%% P16..4: %4.1f%% %4.1f%% %4.1f%% %4.1f%% %4.1f%% skip:%4.1f%%\n",
1796 i_mb_count[I_16x16]/ i_count,
1797 i_mb_count[I_8x8] / i_count,
1798 i_mb_count[I_4x4] / i_count,
1799 i_mb_size[PIXEL_16x16] / (i_count*4),
1800 (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
1801 i_mb_size[PIXEL_8x8] / (i_count*4),
1802 (i_mb_size[PIXEL_8x4] + i_mb_size[PIXEL_4x8]) / (i_count*4),
1803 i_mb_size[PIXEL_4x4] / (i_count*4),
1804 i_mb_count[P_SKIP] / i_count );
1806 if( h->stat.i_slice_count[SLICE_TYPE_B] > 0 )
1808 const int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_B];
1809 const int64_t *i_mb_size = h->stat.i_mb_count_size[SLICE_TYPE_B];
1810 const double i_count = h->stat.i_slice_count[SLICE_TYPE_B] * h->mb.i_mb_count / 100.0;
1811 x264_log( h, X264_LOG_INFO,
1812 "mb B I16..4: %4.1f%% %4.1f%% %4.1f%% B16..8: %4.1f%% %4.1f%% %4.1f%% direct:%4.1f%% skip:%4.1f%%\n",
1813 i_mb_count[I_16x16] / i_count,
1814 i_mb_count[I_8x8] / i_count,
1815 i_mb_count[I_4x4] / i_count,
1816 i_mb_size[PIXEL_16x16] / (i_count*4),
1817 (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
1818 i_mb_size[PIXEL_8x8] / (i_count*4),
1819 i_mb_count[B_DIRECT] / i_count,
1820 i_mb_count[B_SKIP] / i_count );
1823 x264_ratecontrol_summary( h );
1825 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 )
1827 const int i_count = h->stat.i_slice_count[SLICE_TYPE_I] +
1828 h->stat.i_slice_count[SLICE_TYPE_P] +
1829 h->stat.i_slice_count[SLICE_TYPE_B];
1830 float fps = (float) h->param.i_fps_num / h->param.i_fps_den;
1831 #define SUM3(p) (p[SLICE_TYPE_I] + p[SLICE_TYPE_P] + p[SLICE_TYPE_B])
1832 #define SUM3b(p,o) (p[SLICE_TYPE_I][o] + p[SLICE_TYPE_P][o] + p[SLICE_TYPE_B][o])
1833 float f_bitrate = fps * SUM3(h->stat.i_slice_size) / i_count / 125;
1835 if( h->pps->b_transform_8x8_mode )
1837 int64_t i_i8x8 = SUM3b( h->stat.i_mb_count, I_8x8 );
1838 int64_t i_intra = i_i8x8 + SUM3b( h->stat.i_mb_count, I_4x4 )
1839 + SUM3b( h->stat.i_mb_count, I_16x16 );
1840 x264_log( h, X264_LOG_INFO, "8x8 transform intra:%.1f%% inter:%.1f%%\n",
1841 100. * i_i8x8 / i_intra,
1842 100. * h->stat.i_mb_count_8x8dct[1] / h->stat.i_mb_count_8x8dct[0] );
1845 if( h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO
1846 && h->stat.i_slice_count[SLICE_TYPE_B] )
1848 x264_log( h, X264_LOG_INFO, "direct mvs spatial:%.1f%% temporal:%.1f%%\n",
1849 h->stat.i_direct_frames[1] * 100. / h->stat.i_slice_count[SLICE_TYPE_B],
1850 h->stat.i_direct_frames[0] * 100. / h->stat.i_slice_count[SLICE_TYPE_B] );
1853 if( h->frames.i_max_ref0 > 1 )
1856 for( i_slice = 0; i_slice < 2; i_slice++ )
1862 for( i = 0; i < h->frames.i_max_ref0 << h->param.b_interlaced; i++ )
1863 if( h->stat.i_mb_count_ref[i_slice][i] )
1865 i_den += h->stat.i_mb_count_ref[i_slice][i];
1870 for( i = 0; i <= i_max; i++ )
1871 p += sprintf( p, " %4.1f%%", 100. * h->stat.i_mb_count_ref[i_slice][i] / i_den );
1872 x264_log( h, X264_LOG_INFO, "ref %c %s\n", i_slice==SLICE_TYPE_P ? 'P' : 'B', buf );
1876 if( h->param.analyse.b_ssim )
1878 x264_log( h, X264_LOG_INFO,
1879 "SSIM Mean Y:%.7f\n",
1880 SUM3( h->stat.f_ssim_mean_y ) / i_count );
1882 if( h->param.analyse.b_psnr )
1884 x264_log( h, X264_LOG_INFO,
1885 "PSNR Mean Y:%6.3f U:%6.3f V:%6.3f Avg:%6.3f Global:%6.3f kb/s:%.2f\n",
1886 SUM3( h->stat.f_psnr_mean_y ) / i_count,
1887 SUM3( h->stat.f_psnr_mean_u ) / i_count,
1888 SUM3( h->stat.f_psnr_mean_v ) / i_count,
1889 SUM3( h->stat.f_psnr_average ) / i_count,
1890 x264_psnr( SUM3( h->stat.i_sqe_global ), i_count * i_yuv_size ),
1894 x264_log( h, X264_LOG_INFO, "kb/s:%.1f\n", f_bitrate );
1898 for( i = 0; i < X264_BFRAME_MAX + 3; i++ )
1900 if( h->frames.current[i] ) x264_frame_delete( h->frames.current[i] );
1901 if( h->frames.next[i] ) x264_frame_delete( h->frames.next[i] );
1902 if( h->frames.unused[i] ) x264_frame_delete( h->frames.unused[i] );
1905 for( i = 0; i < h->frames.i_max_dpb; i++ )
1907 x264_frame_delete( h->frames.reference[i] );
1911 x264_ratecontrol_delete( h );
1914 if( h->param.rc.psz_stat_out )
1915 free( h->param.rc.psz_stat_out );
1916 if( h->param.rc.psz_stat_in )
1917 free( h->param.rc.psz_stat_in );
1918 if( h->param.rc.psz_rc_eq )
1919 free( h->param.rc.psz_rc_eq );
1921 x264_cqm_delete( h );
1922 x264_macroblock_cache_end( h );
1923 x264_free( h->out.p_bitstream );
1924 for( i = 1; i < h->param.i_threads; i++ )
1925 x264_free( h->thread[i] );