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_field_pic = 0; /* Not field support for now */
121 sh->b_bottom_field = 1; /* not yet used */
123 sh->i_idr_pic_id = i_idr_pic_id;
125 /* poc stuff, fixed later */
127 sh->i_delta_poc_bottom = 0;
128 sh->i_delta_poc[0] = 0;
129 sh->i_delta_poc[1] = 0;
131 sh->i_redundant_pic_cnt = 0;
133 if( !h->mb.b_direct_auto_read )
135 if( h->mb.b_direct_auto_write )
136 sh->b_direct_spatial_mv_pred = ( h->stat.i_direct_score[1] > h->stat.i_direct_score[0] );
138 sh->b_direct_spatial_mv_pred = ( param->analyse.i_direct_mv_pred == X264_DIRECT_PRED_SPATIAL );
140 /* else b_direct_spatial_mv_pred was read from the 2pass statsfile */
142 sh->b_num_ref_idx_override = 0;
143 sh->i_num_ref_idx_l0_active = 1;
144 sh->i_num_ref_idx_l1_active = 1;
146 sh->b_ref_pic_list_reordering_l0 = h->b_ref_reorder[0];
147 sh->b_ref_pic_list_reordering_l1 = h->b_ref_reorder[1];
149 /* If the ref list isn't in the default order, construct reordering header */
150 /* List1 reordering isn't needed yet */
151 if( sh->b_ref_pic_list_reordering_l0 )
153 int pred_frame_num = i_frame;
154 for( i = 0; i < h->i_ref0; i++ )
156 int diff = h->fref0[i]->i_frame_num - pred_frame_num;
158 x264_log( h, X264_LOG_ERROR, "diff frame num == 0\n" );
159 sh->ref_pic_list_order[0][i].idc = ( diff > 0 );
160 sh->ref_pic_list_order[0][i].arg = abs( diff ) - 1;
161 pred_frame_num = h->fref0[i]->i_frame_num;
165 sh->i_cabac_init_idc = param->i_cabac_init_idc;
168 sh->i_qp_delta = i_qp - pps->i_pic_init_qp;
169 sh->b_sp_for_swidth = 0;
172 /* If effective qp <= 15, deblocking would have no effect anyway */
173 if( param->b_deblocking_filter
174 && ( h->mb.b_variable_qp
175 || 15 < i_qp + 2 * X264_MAX(param->i_deblocking_filter_alphac0, param->i_deblocking_filter_beta) ) )
177 sh->i_disable_deblocking_filter_idc = 0;
181 sh->i_disable_deblocking_filter_idc = 1;
183 sh->i_alpha_c0_offset = param->i_deblocking_filter_alphac0 << 1;
184 sh->i_beta_offset = param->i_deblocking_filter_beta << 1;
187 static void x264_slice_header_write( bs_t *s, x264_slice_header_t *sh, int i_nal_ref_idc )
191 bs_write_ue( s, sh->i_first_mb );
192 bs_write_ue( s, sh->i_type + 5 ); /* same type things */
193 bs_write_ue( s, sh->i_pps_id );
194 bs_write( s, sh->sps->i_log2_max_frame_num, sh->i_frame_num );
196 if( sh->i_idr_pic_id >= 0 ) /* NAL IDR */
198 bs_write_ue( s, sh->i_idr_pic_id );
201 if( sh->sps->i_poc_type == 0 )
203 bs_write( s, sh->sps->i_log2_max_poc_lsb, sh->i_poc_lsb );
204 if( sh->pps->b_pic_order && !sh->b_field_pic )
206 bs_write_se( s, sh->i_delta_poc_bottom );
209 else if( sh->sps->i_poc_type == 1 && !sh->sps->b_delta_pic_order_always_zero )
211 bs_write_se( s, sh->i_delta_poc[0] );
212 if( sh->pps->b_pic_order && !sh->b_field_pic )
214 bs_write_se( s, sh->i_delta_poc[1] );
218 if( sh->pps->b_redundant_pic_cnt )
220 bs_write_ue( s, sh->i_redundant_pic_cnt );
223 if( sh->i_type == SLICE_TYPE_B )
225 bs_write1( s, sh->b_direct_spatial_mv_pred );
227 if( sh->i_type == SLICE_TYPE_P || sh->i_type == SLICE_TYPE_SP || sh->i_type == SLICE_TYPE_B )
229 bs_write1( s, sh->b_num_ref_idx_override );
230 if( sh->b_num_ref_idx_override )
232 bs_write_ue( s, sh->i_num_ref_idx_l0_active - 1 );
233 if( sh->i_type == SLICE_TYPE_B )
235 bs_write_ue( s, sh->i_num_ref_idx_l1_active - 1 );
240 /* ref pic list reordering */
241 if( sh->i_type != SLICE_TYPE_I )
243 bs_write1( s, sh->b_ref_pic_list_reordering_l0 );
244 if( sh->b_ref_pic_list_reordering_l0 )
246 for( i = 0; i < sh->i_num_ref_idx_l0_active; i++ )
248 bs_write_ue( s, sh->ref_pic_list_order[0][i].idc );
249 bs_write_ue( s, sh->ref_pic_list_order[0][i].arg );
255 if( sh->i_type == SLICE_TYPE_B )
257 bs_write1( s, sh->b_ref_pic_list_reordering_l1 );
258 if( sh->b_ref_pic_list_reordering_l1 )
260 for( i = 0; i < sh->i_num_ref_idx_l1_active; i++ )
262 bs_write_ue( s, sh->ref_pic_list_order[1][i].idc );
263 bs_write_ue( s, sh->ref_pic_list_order[1][i].arg );
269 if( ( sh->pps->b_weighted_pred && ( sh->i_type == SLICE_TYPE_P || sh->i_type == SLICE_TYPE_SP ) ) ||
270 ( sh->pps->b_weighted_bipred == 1 && sh->i_type == SLICE_TYPE_B ) )
275 if( i_nal_ref_idc != 0 )
277 if( sh->i_idr_pic_id >= 0 )
279 bs_write1( s, 0 ); /* no output of prior pics flag */
280 bs_write1( s, 0 ); /* long term reference flag */
284 bs_write1( s, 0 ); /* adaptive_ref_pic_marking_mode_flag */
288 if( sh->pps->b_cabac && sh->i_type != SLICE_TYPE_I )
290 bs_write_ue( s, sh->i_cabac_init_idc );
292 bs_write_se( s, sh->i_qp_delta ); /* slice qp delta */
294 if( sh->pps->b_deblocking_filter_control )
296 bs_write_ue( s, sh->i_disable_deblocking_filter_idc );
297 if( sh->i_disable_deblocking_filter_idc != 1 )
299 bs_write_se( s, sh->i_alpha_c0_offset >> 1 );
300 bs_write_se( s, sh->i_beta_offset >> 1 );
305 /****************************************************************************
307 ****************************************************************************
308 ****************************** External API*********************************
309 ****************************************************************************
311 ****************************************************************************/
313 static int x264_validate_parameters( x264_t *h )
315 if( h->param.i_width <= 0 || h->param.i_height <= 0 )
317 x264_log( h, X264_LOG_ERROR, "invalid width x height (%dx%d)\n",
318 h->param.i_width, h->param.i_height );
322 if( h->param.i_width % 2 || h->param.i_height % 2 )
324 x264_log( h, X264_LOG_ERROR, "width or height not divisible by 2 (%dx%d)\n",
325 h->param.i_width, h->param.i_height );
328 if( h->param.i_csp != X264_CSP_I420 )
330 x264_log( h, X264_LOG_ERROR, "invalid CSP (only I420 supported)\n" );
334 h->param.i_threads = x264_clip3( h->param.i_threads, 1, X264_SLICE_MAX );
335 h->param.i_threads = X264_MIN( h->param.i_threads, (h->param.i_height + 15) / 16 );
337 if( h->param.i_threads > 1 )
339 x264_log( h, X264_LOG_WARNING, "not compiled with pthread support!\n");
340 x264_log( h, X264_LOG_WARNING, "multislicing anyway, but you won't see any speed gain.\n" );
344 if( h->param.rc.b_cbr )
345 h->param.rc.i_rf_constant = 0;
346 if( h->param.rc.i_rf_constant > 0 )
347 h->param.rc.i_qp_constant = h->param.rc.i_rf_constant;
348 h->param.rc.i_rf_constant = x264_clip3( h->param.rc.i_rf_constant, 0, 51 );
349 h->param.rc.i_qp_constant = x264_clip3( h->param.rc.i_qp_constant, 0, 51 );
350 if( !h->param.rc.b_cbr && h->param.rc.i_qp_constant == 0 )
352 h->mb.b_lossless = 1;
353 h->param.analyse.b_transform_8x8 = 0;
354 h->param.i_cqm_preset = X264_CQM_FLAT;
355 h->param.psz_cqm_file = NULL;
356 h->param.rc.f_ip_factor = 1;
357 h->param.rc.f_pb_factor = 1;
358 h->param.analyse.b_psnr = 0;
359 h->param.analyse.i_chroma_qp_offset = 0;
360 h->param.analyse.i_trellis = 0;
361 h->param.analyse.b_fast_pskip = 0;
362 h->param.analyse.i_noise_reduction = 0;
363 h->param.analyse.i_subpel_refine = x264_clip3( h->param.analyse.i_subpel_refine, 1, 6 );
366 if( ( h->param.i_width % 16 || h->param.i_height % 16 ) && !h->mb.b_lossless )
368 x264_log( h, X264_LOG_WARNING,
369 "width or height not divisible by 16 (%dx%d), compression will suffer.\n",
370 h->param.i_width, h->param.i_height );
373 h->param.i_frame_reference = x264_clip3( h->param.i_frame_reference, 1, 16 );
374 if( h->param.i_keyint_max <= 0 )
375 h->param.i_keyint_max = 1;
376 h->param.i_keyint_min = x264_clip3( h->param.i_keyint_min, 1, h->param.i_keyint_max/2+1 );
378 h->param.i_bframe = x264_clip3( h->param.i_bframe, 0, X264_BFRAME_MAX );
379 h->param.i_bframe_bias = x264_clip3( h->param.i_bframe_bias, -90, 100 );
380 h->param.b_bframe_pyramid = h->param.b_bframe_pyramid && h->param.i_bframe > 1;
381 h->param.b_bframe_adaptive = h->param.b_bframe_adaptive && h->param.i_bframe > 0;
382 h->param.analyse.b_weighted_bipred = h->param.analyse.b_weighted_bipred && h->param.i_bframe > 0;
383 h->mb.b_direct_auto_write = h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO
385 && ( h->param.rc.b_stat_write || !h->param.rc.b_stat_read );
387 h->param.i_deblocking_filter_alphac0 = x264_clip3( h->param.i_deblocking_filter_alphac0, -6, 6 );
388 h->param.i_deblocking_filter_beta = x264_clip3( h->param.i_deblocking_filter_beta, -6, 6 );
390 h->param.i_cabac_init_idc = x264_clip3( h->param.i_cabac_init_idc, 0, 2 );
392 if( h->param.i_cqm_preset < X264_CQM_FLAT || h->param.i_cqm_preset > X264_CQM_CUSTOM )
393 h->param.i_cqm_preset = X264_CQM_FLAT;
395 if( h->param.analyse.i_me_method < X264_ME_DIA ||
396 h->param.analyse.i_me_method > X264_ME_ESA )
397 h->param.analyse.i_me_method = X264_ME_HEX;
398 if( h->param.analyse.i_me_range < 4 )
399 h->param.analyse.i_me_range = 4;
400 if( h->param.analyse.i_me_range > 16 && h->param.analyse.i_me_method <= X264_ME_HEX )
401 h->param.analyse.i_me_range = 16;
402 h->param.analyse.i_subpel_refine = x264_clip3( h->param.analyse.i_subpel_refine, 1, 7 );
403 h->param.analyse.b_bframe_rdo = h->param.analyse.b_bframe_rdo && h->param.analyse.i_subpel_refine >= 6;
404 h->param.analyse.b_mixed_references = h->param.analyse.b_mixed_references && h->param.i_frame_reference > 1;
405 h->param.analyse.inter &= X264_ANALYSE_PSUB16x16|X264_ANALYSE_PSUB8x8|X264_ANALYSE_BSUB16x16|
406 X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
407 h->param.analyse.intra &= X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
408 if( !(h->param.analyse.inter & X264_ANALYSE_PSUB16x16) )
409 h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
410 if( !h->param.analyse.b_transform_8x8 )
412 h->param.analyse.inter &= ~X264_ANALYSE_I8x8;
413 h->param.analyse.intra &= ~X264_ANALYSE_I8x8;
415 h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -12, 12);
416 if( !h->param.b_cabac )
417 h->param.analyse.i_trellis = 0;
418 h->param.analyse.i_trellis = x264_clip3( h->param.analyse.i_trellis, 0, 2 );
419 h->param.analyse.i_noise_reduction = x264_clip3( h->param.analyse.i_noise_reduction, 0, 1<<16 );
422 const x264_level_t *l = x264_levels;
423 while( l->level_idc != 0 && l->level_idc != h->param.i_level_idc )
425 if( l->level_idc == 0 )
427 x264_log( h, X264_LOG_ERROR, "invalid level_idc: %d\n", h->param.i_level_idc );
430 if( h->param.analyse.i_mv_range <= 0 )
431 h->param.analyse.i_mv_range = l->mv_range;
433 h->param.analyse.i_mv_range = x264_clip3(h->param.analyse.i_mv_range, 32, 2048);
436 if( h->param.rc.f_qblur < 0 )
437 h->param.rc.f_qblur = 0;
438 if( h->param.rc.f_complexity_blur < 0 )
439 h->param.rc.f_complexity_blur = 0;
441 h->param.i_sps_id &= 31;
443 /* ensure the booleans are 0 or 1 so they can be used in math */
444 #define BOOLIFY(x) h->param.x = !!h->param.x
446 BOOLIFY( b_deblocking_filter );
447 BOOLIFY( analyse.b_transform_8x8 );
448 BOOLIFY( analyse.b_bidir_me );
449 BOOLIFY( analyse.b_chroma_me );
450 BOOLIFY( analyse.b_fast_pskip );
452 BOOLIFY( rc.b_stat_write );
453 BOOLIFY( rc.b_stat_read );
459 /****************************************************************************
461 ****************************************************************************/
462 x264_t *x264_encoder_open ( x264_param_t *param )
464 x264_t *h = x264_malloc( sizeof( x264_t ) );
467 memset( h, 0, sizeof( x264_t ) );
469 /* Create a copy of param */
470 memcpy( &h->param, param, sizeof( x264_param_t ) );
472 if( x264_validate_parameters( h ) < 0 )
478 if( h->param.psz_cqm_file )
479 if( x264_cqm_parse_file( h, h->param.psz_cqm_file ) < 0 )
485 if( h->param.rc.psz_stat_out )
486 h->param.rc.psz_stat_out = strdup( h->param.rc.psz_stat_out );
487 if( h->param.rc.psz_stat_in )
488 h->param.rc.psz_stat_in = strdup( h->param.rc.psz_stat_in );
489 if( h->param.rc.psz_rc_eq )
490 h->param.rc.psz_rc_eq = strdup( h->param.rc.psz_rc_eq );
493 if( h->param.vui.i_sar_width > 0 && h->param.vui.i_sar_height > 0 )
495 int i_w = param->vui.i_sar_width;
496 int i_h = param->vui.i_sar_height;
498 x264_reduce_fraction( &i_w, &i_h );
500 while( i_w > 65535 || i_h > 65535 )
506 h->param.vui.i_sar_width = 0;
507 h->param.vui.i_sar_height = 0;
508 if( i_w == 0 || i_h == 0 )
510 x264_log( h, X264_LOG_ERROR, "cannot create valid sample aspect ratio\n" );
514 x264_log( h, X264_LOG_INFO, "using SAR=%d/%d\n", i_w, i_h );
515 h->param.vui.i_sar_width = i_w;
516 h->param.vui.i_sar_height = i_h;
520 x264_reduce_fraction( &h->param.i_fps_num, &h->param.i_fps_den );
524 h->out.i_bitstream = X264_MAX( 1000000, h->param.i_width * h->param.i_height * 1.7
525 * ( h->param.rc.b_cbr ? pow( 0.5, h->param.rc.i_qp_min )
526 : pow( 0.5, h->param.rc.i_qp_constant ) * X264_MAX( 1, h->param.rc.f_ip_factor )));
527 h->out.p_bitstream = x264_malloc( h->out.i_bitstream );
533 h->sps = &h->sps_array[0];
534 x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
536 h->pps = &h->pps_array[0];
537 x264_pps_init( h->pps, h->param.i_sps_id, &h->param, h->sps);
539 x264_validate_levels( h );
543 h->mb.i_mb_count = h->sps->i_mb_width * h->sps->i_mb_height;
546 h->frames.i_delay = h->param.i_bframe;
547 h->frames.i_max_ref0 = h->param.i_frame_reference;
548 h->frames.i_max_ref1 = h->sps->vui.i_num_reorder_frames;
549 h->frames.i_max_dpb = h->sps->vui.i_max_dec_frame_buffering + 1;
550 h->frames.b_have_lowres = !h->param.rc.b_stat_read
551 && ( h->param.rc.b_cbr || h->param.rc.i_rf_constant || h->param.b_bframe_adaptive );
553 for( i = 0; i < X264_BFRAME_MAX + 3; i++ )
555 h->frames.current[i] = NULL;
556 h->frames.next[i] = NULL;
557 h->frames.unused[i] = NULL;
559 for( i = 0; i < 1 + h->frames.i_delay; i++ )
561 h->frames.unused[i] = x264_frame_new( h );
563 for( i = 0; i < h->frames.i_max_dpb; i++ )
565 h->frames.reference[i] = x264_frame_new( h );
567 h->frames.reference[h->frames.i_max_dpb] = NULL;
568 h->frames.i_last_idr = - h->param.i_keyint_max;
569 h->frames.i_input = 0;
570 h->frames.last_nonb = NULL;
575 h->fdec = h->frames.reference[0];
577 x264_macroblock_cache_init( h );
580 /* init CPU functions */
581 x264_predict_16x16_init( h->param.cpu, h->predict_16x16 );
582 x264_predict_8x8c_init( h->param.cpu, h->predict_8x8c );
583 x264_predict_8x8_init( h->param.cpu, h->predict_8x8 );
584 x264_predict_4x4_init( h->param.cpu, h->predict_4x4 );
586 x264_pixel_init( h->param.cpu, &h->pixf );
587 x264_dct_init( h->param.cpu, &h->dctf );
588 x264_mc_init( h->param.cpu, &h->mc );
589 x264_csp_init( h->param.cpu, h->param.i_csp, &h->csp );
590 x264_quant_init( h, h->param.cpu, &h->quantf );
591 x264_deblock_init( h->param.cpu, &h->loopf );
593 memcpy( h->pixf.mbcmp,
594 ( h->mb.b_lossless || h->param.analyse.i_subpel_refine <= 1 ) ? h->pixf.sad : h->pixf.satd,
595 sizeof(h->pixf.mbcmp) );
598 if( x264_ratecontrol_new( h ) < 0 )
601 x264_log( h, X264_LOG_INFO, "using cpu capabilities %s%s%s%s%s%s\n",
602 param->cpu&X264_CPU_MMX ? "MMX " : "",
603 param->cpu&X264_CPU_MMXEXT ? "MMXEXT " : "",
604 param->cpu&X264_CPU_SSE ? "SSE " : "",
605 param->cpu&X264_CPU_SSE2 ? "SSE2 " : "",
606 param->cpu&X264_CPU_3DNOW ? "3DNow! " : "",
607 param->cpu&X264_CPU_ALTIVEC ? "Altivec " : "" );
611 for( i = 1; i < param->i_threads; i++ )
612 h->thread[i] = x264_malloc( sizeof(x264_t) );
614 #ifdef DEBUG_DUMP_FRAME
616 /* create or truncate the reconstructed video file */
617 FILE *f = fopen( "fdec.yuv", "w" );
622 x264_log( h, X264_LOG_ERROR, "can't write to fdec.yuv\n" );
632 /****************************************************************************
633 * x264_encoder_reconfig:
634 ****************************************************************************/
635 int x264_encoder_reconfig( x264_t *h, x264_param_t *param )
637 h->param.i_bframe_bias = param->i_bframe_bias;
638 h->param.i_deblocking_filter_alphac0 = param->i_deblocking_filter_alphac0;
639 h->param.i_deblocking_filter_beta = param->i_deblocking_filter_beta;
640 h->param.analyse.i_me_method = param->analyse.i_me_method;
641 h->param.analyse.i_me_range = param->analyse.i_me_range;
642 h->param.analyse.i_subpel_refine = param->analyse.i_subpel_refine;
643 h->param.analyse.i_trellis = param->analyse.i_trellis;
644 h->param.analyse.intra = param->analyse.intra;
645 h->param.analyse.inter = param->analyse.inter;
647 memcpy( h->pixf.mbcmp,
648 ( h->mb.b_lossless || h->param.analyse.i_subpel_refine <= 1 ) ? h->pixf.sad : h->pixf.satd,
649 sizeof(h->pixf.mbcmp) );
651 return x264_validate_parameters( h );
655 static void x264_nal_start( x264_t *h, int i_type, int i_ref_idc )
657 x264_nal_t *nal = &h->out.nal[h->out.i_nal];
659 nal->i_ref_idc = i_ref_idc;
660 nal->i_type = i_type;
662 bs_align_0( &h->out.bs ); /* not needed */
665 nal->p_payload= &h->out.p_bitstream[bs_pos( &h->out.bs) / 8];
667 static void x264_nal_end( x264_t *h )
669 x264_nal_t *nal = &h->out.nal[h->out.i_nal];
671 bs_align_0( &h->out.bs ); /* not needed */
673 nal->i_payload = &h->out.p_bitstream[bs_pos( &h->out.bs)/8] - nal->p_payload;
678 /****************************************************************************
679 * x264_encoder_headers:
680 ****************************************************************************/
681 int x264_encoder_headers( x264_t *h, x264_nal_t **pp_nal, int *pi_nal )
683 /* init bitstream context */
685 bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
687 /* Put SPS and PPS */
688 if( h->i_frame == 0 )
690 /* identify ourself */
691 x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
692 x264_sei_version_write( h, &h->out.bs );
695 /* generate sequence parameters */
696 x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
697 x264_sps_write( &h->out.bs, h->sps );
700 /* generate picture parameters */
701 x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
702 x264_pps_write( &h->out.bs, h->pps );
706 *pi_nal = h->out.i_nal;
707 *pp_nal = &h->out.nal[0];
713 static void x264_frame_put( x264_frame_t *list[X264_BFRAME_MAX], x264_frame_t *frame )
716 while( list[i] ) i++;
720 static void x264_frame_push( x264_frame_t *list[X264_BFRAME_MAX], x264_frame_t *frame )
723 while( list[i] ) i++;
729 static x264_frame_t *x264_frame_get( x264_frame_t *list[X264_BFRAME_MAX+1] )
731 x264_frame_t *frame = list[0];
733 for( i = 0; list[i]; i++ )
738 static void x264_frame_sort( x264_frame_t *list[X264_BFRAME_MAX+1], int b_dts )
743 for( i = 0; list[i+1]; i++ )
745 int dtype = list[i]->i_type - list[i+1]->i_type;
746 int dtime = list[i]->i_frame - list[i+1]->i_frame;
747 int swap = b_dts ? dtype > 0 || ( dtype == 0 && dtime > 0 )
751 XCHG( x264_frame_t*, list[i], list[i+1] );
757 #define x264_frame_sort_dts(list) x264_frame_sort(list, 1)
758 #define x264_frame_sort_pts(list) x264_frame_sort(list, 0)
760 static inline void x264_reference_build_list( x264_t *h, int i_poc, int i_slice_type )
765 /* build ref list 0/1 */
768 for( i = 1; i < h->frames.i_max_dpb; i++ )
770 if( h->frames.reference[i]->i_poc >= 0 )
772 if( h->frames.reference[i]->i_poc < i_poc )
774 h->fref0[h->i_ref0++] = h->frames.reference[i];
776 else if( h->frames.reference[i]->i_poc > i_poc )
778 h->fref1[h->i_ref1++] = h->frames.reference[i];
783 /* Order ref0 from higher to lower poc */
787 for( i = 0; i < h->i_ref0 - 1; i++ )
789 if( h->fref0[i]->i_poc < h->fref0[i+1]->i_poc )
791 XCHG( x264_frame_t*, h->fref0[i], h->fref0[i+1] );
797 /* Order ref1 from lower to higher poc (bubble sort) for B-frame */
801 for( i = 0; i < h->i_ref1 - 1; i++ )
803 if( h->fref1[i]->i_poc > h->fref1[i+1]->i_poc )
805 XCHG( x264_frame_t*, h->fref1[i], h->fref1[i+1] );
812 /* In the standard, a P-frame's ref list is sorted by frame_num.
813 * We use POC, but check whether explicit reordering is needed */
814 h->b_ref_reorder[0] =
815 h->b_ref_reorder[1] = 0;
816 if( i_slice_type == SLICE_TYPE_P )
818 for( i = 0; i < h->i_ref0 - 1; i++ )
819 if( h->fref0[i]->i_frame_num < h->fref0[i+1]->i_frame_num )
821 h->b_ref_reorder[0] = 1;
826 h->i_ref1 = X264_MIN( h->i_ref1, h->frames.i_max_ref1 );
827 h->i_ref0 = X264_MIN( h->i_ref0, h->frames.i_max_ref0 );
828 h->i_ref0 = X264_MIN( h->i_ref0, 16 - h->i_ref1 );
831 static inline void x264_fdec_deblock( x264_t *h )
833 /* apply deblocking filter to the current decoded picture */
834 if( !h->sh.i_disable_deblocking_filter_idc )
836 TIMER_START( i_mtime_filter );
837 x264_frame_deblocking_filter( h, h->sh.i_type );
838 TIMER_STOP( i_mtime_filter );
842 static inline void x264_reference_update( x264_t *h )
846 x264_fdec_deblock( h );
849 x264_frame_expand_border( h->fdec );
851 /* create filtered images */
852 x264_frame_filter( h->param.cpu, h->fdec );
854 /* expand border of filtered images */
855 x264_frame_expand_border_filtered( h->fdec );
857 /* move lowres copy of the image to the ref frame */
858 for( i = 0; i < 4; i++)
859 XCHG( uint8_t*, h->fdec->lowres[i], h->fenc->lowres[i] );
861 /* adaptive B decision needs a pointer, since it can't use the ref lists */
862 if( h->sh.i_type != SLICE_TYPE_B )
863 h->frames.last_nonb = h->fdec;
865 /* move frame in the buffer */
866 h->fdec = h->frames.reference[h->frames.i_max_dpb-1];
867 for( i = h->frames.i_max_dpb-1; i > 0; i-- )
869 h->frames.reference[i] = h->frames.reference[i-1];
871 h->frames.reference[0] = h->fdec;
874 static inline void x264_reference_reset( x264_t *h )
878 /* reset ref pictures */
879 for( i = 1; i < h->frames.i_max_dpb; i++ )
881 h->frames.reference[i]->i_poc = -1;
883 h->frames.reference[0]->i_poc = 0;
886 static inline void x264_slice_init( x264_t *h, int i_nal_type, int i_slice_type, int i_global_qp )
888 /* ------------------------ Create slice header ----------------------- */
889 if( i_nal_type == NAL_SLICE_IDR )
891 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 );
894 h->i_idr_pic_id = ( h->i_idr_pic_id + 1 ) % 65536;
898 x264_slice_header_init( h, &h->sh, h->sps, h->pps, i_slice_type, -1, h->i_frame_num, i_global_qp );
900 /* always set the real higher num of ref frame used */
901 h->sh.b_num_ref_idx_override = 1;
902 h->sh.i_num_ref_idx_l0_active = h->i_ref0 <= 0 ? 1 : h->i_ref0;
903 h->sh.i_num_ref_idx_l1_active = h->i_ref1 <= 0 ? 1 : h->i_ref1;
906 h->fdec->i_frame_num = h->sh.i_frame_num;
908 if( h->sps->i_poc_type == 0 )
910 h->sh.i_poc_lsb = h->fdec->i_poc & ( (1 << h->sps->i_log2_max_poc_lsb) - 1 );
911 h->sh.i_delta_poc_bottom = 0; /* XXX won't work for field */
913 else if( h->sps->i_poc_type == 1 )
915 /* FIXME TODO FIXME */
919 /* Nothing to do ? */
922 x264_macroblock_slice_init( h );
925 static int x264_slice_write( x264_t *h )
932 memset( &h->stat.frame, 0, sizeof(h->stat.frame) );
935 x264_nal_start( h, h->i_nal_type, h->i_nal_ref_idc );
938 x264_slice_header_write( &h->out.bs, &h->sh, h->i_nal_ref_idc );
939 if( h->param.b_cabac )
941 /* alignment needed */
942 bs_align_1( &h->out.bs );
945 x264_cabac_context_init( &h->cabac, h->sh.i_type, h->sh.i_qp, h->sh.i_cabac_init_idc );
946 x264_cabac_encode_init ( &h->cabac, &h->out.bs );
948 h->mb.i_last_qp = h->sh.i_qp;
949 h->mb.i_last_dqp = 0;
951 for( mb_xy = h->sh.i_first_mb, i_skip = 0; mb_xy < h->sh.i_last_mb; mb_xy++ )
953 const int i_mb_y = mb_xy / h->sps->i_mb_width;
954 const int i_mb_x = mb_xy % h->sps->i_mb_width;
956 int mb_spos = bs_pos(&h->out.bs);
959 x264_macroblock_cache_load( h, i_mb_x, i_mb_y );
961 /* analyse parameters
962 * Slice I: choose I_4x4 or I_16x16 mode
963 * Slice P: choose between using P mode or intra (4x4 or 16x16)
965 TIMER_START( i_mtime_analyse );
966 x264_macroblock_analyse( h );
967 TIMER_STOP( i_mtime_analyse );
969 /* encode this macrobock -> be carefull it can change the mb type to P_SKIP if needed */
970 TIMER_START( i_mtime_encode );
971 x264_macroblock_encode( h );
972 TIMER_STOP( i_mtime_encode );
974 TIMER_START( i_mtime_write );
975 if( h->param.b_cabac )
977 if( mb_xy > h->sh.i_first_mb )
978 x264_cabac_encode_terminal( &h->cabac, 0 );
980 if( IS_SKIP( h->mb.i_type ) )
981 x264_cabac_mb_skip( h, 1 );
984 if( h->sh.i_type != SLICE_TYPE_I )
985 x264_cabac_mb_skip( h, 0 );
986 x264_macroblock_write_cabac( h, &h->cabac );
991 if( IS_SKIP( h->mb.i_type ) )
995 if( h->sh.i_type != SLICE_TYPE_I )
997 bs_write_ue( &h->out.bs, i_skip ); /* skip run */
1000 x264_macroblock_write_cavlc( h, &h->out.bs );
1003 TIMER_STOP( i_mtime_write );
1006 if( h->param.b_visualize )
1007 x264_visualize_mb( h );
1011 x264_macroblock_cache_save( h );
1013 /* accumulate mb stats */
1014 h->stat.frame.i_mb_count[h->mb.i_type]++;
1015 if( !IS_SKIP(h->mb.i_type) && !IS_INTRA(h->mb.i_type) && !IS_DIRECT(h->mb.i_type) )
1017 if( h->mb.i_partition != D_8x8 )
1018 h->stat.frame.i_mb_count_size[ x264_mb_partition_pixel_table[ h->mb.i_partition ] ] += 4;
1020 for( i = 0; i < 4; i++ )
1021 h->stat.frame.i_mb_count_size[ x264_mb_partition_pixel_table[ h->mb.i_sub_partition[i] ] ] ++;
1022 if( h->param.i_frame_reference > 1 )
1024 for( i = 0; i < 4; i++ )
1026 int i_ref = h->mb.cache.ref[0][ x264_scan8[4*i] ];
1028 h->stat.frame.i_mb_count_ref[i_ref] ++;
1032 if( h->mb.i_cbp_luma && !IS_INTRA(h->mb.i_type) )
1034 h->stat.frame.i_mb_count_8x8dct[0] ++;
1035 h->stat.frame.i_mb_count_8x8dct[1] += h->mb.b_transform_8x8;
1038 if( h->mb.b_variable_qp )
1039 x264_ratecontrol_mb(h, bs_pos(&h->out.bs) - mb_spos);
1042 if( h->param.b_cabac )
1045 x264_cabac_encode_terminal( &h->cabac, 1 );
1047 else if( i_skip > 0 )
1049 bs_write_ue( &h->out.bs, i_skip ); /* last skip run */
1052 if( h->param.b_cabac )
1054 x264_cabac_encode_flush( &h->cabac );
1059 /* rbsp_slice_trailing_bits */
1060 bs_rbsp_trailing( &h->out.bs );
1065 /* Compute misc bits */
1066 h->stat.frame.i_misc_bits = bs_pos( &h->out.bs )
1068 - h->stat.frame.i_itex_bits
1069 - h->stat.frame.i_ptex_bits
1070 - h->stat.frame.i_hdr_bits;
1075 static inline int x264_slices_write( x264_t *h )
1080 if( h->param.b_visualize )
1081 x264_visualize_init( h );
1084 if( h->param.i_threads == 1 )
1086 x264_ratecontrol_threads_start( h );
1087 x264_slice_write( h );
1088 i_frame_size = h->out.nal[h->out.i_nal-1].i_payload;
1092 int i_nal = h->out.i_nal;
1093 int i_bs_size = h->out.i_bitstream / h->param.i_threads;
1095 /* duplicate contexts */
1096 for( i = 0; i < h->param.i_threads; i++ )
1098 x264_t *t = h->thread[i];
1101 memcpy( t, h, sizeof(x264_t) );
1102 t->out.p_bitstream += i*i_bs_size;
1103 bs_init( &t->out.bs, t->out.p_bitstream, i_bs_size );
1104 t->i_thread_num = i;
1106 t->sh.i_first_mb = (i * h->sps->i_mb_height / h->param.i_threads) * h->sps->i_mb_width;
1107 t->sh.i_last_mb = ((i+1) * h->sps->i_mb_height / h->param.i_threads) * h->sps->i_mb_width;
1108 t->out.i_nal = i_nal + i;
1110 x264_ratecontrol_threads_start( h );
1115 pthread_t handles[X264_SLICE_MAX];
1116 for( i = 0; i < h->param.i_threads; i++ )
1117 pthread_create( &handles[i], NULL, (void*)x264_slice_write, (void*)h->thread[i] );
1118 for( i = 0; i < h->param.i_threads; i++ )
1119 pthread_join( handles[i], NULL );
1122 for( i = 0; i < h->param.i_threads; i++ )
1123 x264_slice_write( h->thread[i] );
1126 /* merge contexts */
1127 i_frame_size = h->out.nal[i_nal].i_payload;
1128 for( i = 1; i < h->param.i_threads; i++ )
1131 x264_t *t = h->thread[i];
1132 h->out.nal[i_nal+i] = t->out.nal[i_nal+i];
1133 i_frame_size += t->out.nal[i_nal+i].i_payload;
1134 // all entries in stat.frame are ints
1135 for( j = 0; j < sizeof(h->stat.frame) / sizeof(int); j++ )
1136 ((int*)&h->stat.frame)[j] += ((int*)&t->stat.frame)[j];
1138 h->out.i_nal = i_nal + h->param.i_threads;
1142 if( h->param.b_visualize )
1144 x264_visualize_show( h );
1145 x264_visualize_close( h );
1149 return i_frame_size;
1152 /****************************************************************************
1153 * x264_encoder_encode:
1154 * XXX: i_poc : is the poc of the current given picture
1155 * i_frame : is the number of the frame being coded
1156 * ex: type frame poc
1164 ****************************************************************************/
1165 int x264_encoder_encode( x264_t *h,
1166 x264_nal_t **pp_nal, int *pi_nal,
1167 x264_picture_t *pic_in,
1168 x264_picture_t *pic_out )
1170 x264_frame_t *frame_psnr = h->fdec; /* just to keep the current decoded frame for psnr calculation */
1180 char psz_message[80];
1187 /* ------------------- Setup new frame from picture -------------------- */
1188 TIMER_START( i_mtime_encode_frame );
1189 if( pic_in != NULL )
1191 /* 1: Copy the picture to a frame and move it to a buffer */
1192 x264_frame_t *fenc = x264_frame_get( h->frames.unused );
1194 x264_frame_copy_picture( h, fenc, pic_in );
1196 if( h->param.i_width % 16 || h->param.i_height % 16 )
1197 x264_frame_expand_border_mod16( h, fenc );
1199 fenc->i_frame = h->frames.i_input++;
1201 x264_frame_put( h->frames.next, fenc );
1203 if( h->frames.b_have_lowres )
1204 x264_frame_init_lowres( h->param.cpu, fenc );
1206 if( h->frames.i_input <= h->frames.i_delay )
1208 /* Nothing yet to encode */
1209 /* waiting for filling bframe buffer */
1210 pic_out->i_type = X264_TYPE_AUTO;
1215 if( h->frames.current[0] == NULL )
1218 /* 2: Select frame types */
1219 if( h->frames.next[0] == NULL )
1222 x264_slicetype_decide( h );
1224 /* 3: move some B-frames and 1 non-B to encode queue */
1225 while( IS_X264_TYPE_B( h->frames.next[bframes]->i_type ) )
1227 x264_frame_put( h->frames.current, x264_frame_get( &h->frames.next[bframes] ) );
1228 /* FIXME: when max B-frames > 3, BREF may no longer be centered after GOP closing */
1229 if( h->param.b_bframe_pyramid && bframes > 1 )
1231 x264_frame_t *mid = x264_frame_get( &h->frames.next[bframes/2] );
1232 mid->i_type = X264_TYPE_BREF;
1233 x264_frame_put( h->frames.current, mid );
1237 x264_frame_put( h->frames.current, x264_frame_get( h->frames.next ) );
1239 TIMER_STOP( i_mtime_encode_frame );
1241 /* ------------------- Get frame to be encoded ------------------------- */
1242 /* 4: get picture to encode */
1243 h->fenc = x264_frame_get( h->frames.current );
1244 if( h->fenc == NULL )
1246 /* Nothing yet to encode (ex: waiting for I/P with B frames) */
1247 /* waiting for filling bframe buffer */
1248 pic_out->i_type = X264_TYPE_AUTO;
1254 if( h->fenc->i_type == X264_TYPE_IDR )
1256 h->frames.i_last_idr = h->fenc->i_frame;
1259 /* ------------------- Setup frame context ----------------------------- */
1260 /* 5: Init data dependant of frame type */
1261 TIMER_START( i_mtime_encode_frame );
1262 if( h->fenc->i_type == X264_TYPE_IDR )
1264 /* reset ref pictures */
1265 x264_reference_reset( h );
1267 i_nal_type = NAL_SLICE_IDR;
1268 i_nal_ref_idc = NAL_PRIORITY_HIGHEST;
1269 i_slice_type = SLICE_TYPE_I;
1271 else if( h->fenc->i_type == X264_TYPE_I )
1273 i_nal_type = NAL_SLICE;
1274 i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
1275 i_slice_type = SLICE_TYPE_I;
1277 else if( h->fenc->i_type == X264_TYPE_P )
1279 i_nal_type = NAL_SLICE;
1280 i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
1281 i_slice_type = SLICE_TYPE_P;
1283 else if( h->fenc->i_type == X264_TYPE_BREF )
1285 i_nal_type = NAL_SLICE;
1286 i_nal_ref_idc = NAL_PRIORITY_HIGH; /* maybe add MMCO to forget it? -> low */
1287 i_slice_type = SLICE_TYPE_B;
1291 i_nal_type = NAL_SLICE;
1292 i_nal_ref_idc = NAL_PRIORITY_DISPOSABLE;
1293 i_slice_type = SLICE_TYPE_B;
1297 h->fenc->i_poc = 2 * (h->fenc->i_frame - h->frames.i_last_idr);
1298 h->fdec->i_type = h->fenc->i_type;
1299 h->fdec->i_frame = h->fenc->i_frame;
1300 h->fenc->b_kept_as_ref =
1301 h->fdec->b_kept_as_ref = i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE;
1305 /* ------------------- Init ----------------------------- */
1306 /* build ref list 0/1 */
1307 x264_reference_build_list( h, h->fdec->i_poc, i_slice_type );
1309 /* Init the rate control */
1310 x264_ratecontrol_start( h, i_slice_type, h->fenc->i_qpplus1 );
1311 i_global_qp = x264_ratecontrol_qp( h );
1313 pic_out->i_qpplus1 =
1314 h->fdec->i_qpplus1 = i_global_qp + 1;
1316 if( i_slice_type == SLICE_TYPE_B )
1317 x264_macroblock_bipred_init( h );
1319 /* ------------------------ Create slice header ----------------------- */
1320 x264_slice_init( h, i_nal_type, i_slice_type, i_global_qp );
1322 if( h->fenc->b_kept_as_ref )
1325 /* ---------------------- Write the bitstream -------------------------- */
1326 /* Init bitstream context */
1328 bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
1333 if(i_slice_type == SLICE_TYPE_I)
1335 else if(i_slice_type == SLICE_TYPE_P)
1337 else if(i_slice_type == SLICE_TYPE_B)
1342 x264_nal_start(h, NAL_AUD, NAL_PRIORITY_DISPOSABLE);
1343 bs_write(&h->out.bs, 3, pic_type);
1344 bs_rbsp_trailing(&h->out.bs);
1348 h->i_nal_type = i_nal_type;
1349 h->i_nal_ref_idc = i_nal_ref_idc;
1351 /* Write SPS and PPS */
1352 if( i_nal_type == NAL_SLICE_IDR && h->param.b_repeat_headers )
1354 if( h->fenc->i_frame == 0 )
1356 /* identify ourself */
1357 x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
1358 x264_sei_version_write( h, &h->out.bs );
1362 /* generate sequence parameters */
1363 x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
1364 x264_sps_write( &h->out.bs, h->sps );
1367 /* generate picture parameters */
1368 x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
1369 x264_pps_write( &h->out.bs, h->pps );
1374 i_frame_size = x264_slices_write( h );
1376 /* restore CPU state (before using float again) */
1377 x264_cpu_restore( h->param.cpu );
1379 if( i_slice_type == SLICE_TYPE_P && !h->param.rc.b_stat_read
1380 && h->param.i_scenecut_threshold >= 0 )
1382 const int *mbs = h->stat.frame.i_mb_count;
1383 int i_mb_i = mbs[I_16x16] + mbs[I_8x8] + mbs[I_4x4];
1384 int i_mb_p = mbs[P_L0] + mbs[P_8x8];
1385 int i_mb_s = mbs[P_SKIP];
1386 int i_mb = h->sps->i_mb_width * h->sps->i_mb_height;
1387 int64_t i_inter_cost = h->stat.frame.i_inter_cost;
1388 int64_t i_intra_cost = h->stat.frame.i_intra_cost;
1391 int i_gop_size = h->fenc->i_frame - h->frames.i_last_idr;
1392 float f_thresh_max = h->param.i_scenecut_threshold / 100.0;
1393 /* magic numbers pulled out of thin air */
1394 float f_thresh_min = f_thresh_max * h->param.i_keyint_min
1395 / ( h->param.i_keyint_max * 4 );
1396 if( h->param.i_keyint_min == h->param.i_keyint_max )
1397 f_thresh_min= f_thresh_max;
1399 /* macroblock_analyse() doesn't further analyse skipped mbs,
1400 * so we have to guess their cost */
1402 i_intra_cost = i_intra_cost * i_mb / (i_mb - i_mb_s);
1404 if( i_gop_size < h->param.i_keyint_min / 4 )
1405 f_bias = f_thresh_min / 4;
1406 else if( i_gop_size <= h->param.i_keyint_min )
1407 f_bias = f_thresh_min * i_gop_size / h->param.i_keyint_min;
1410 f_bias = f_thresh_min
1411 + ( f_thresh_max - f_thresh_min )
1412 * ( i_gop_size - h->param.i_keyint_min )
1413 / ( h->param.i_keyint_max - h->param.i_keyint_min );
1415 f_bias = X264_MIN( f_bias, 1.0 );
1417 /* Bad P will be reencoded as I */
1418 if( i_mb_s < i_mb &&
1419 i_inter_cost >= (1.0 - f_bias) * i_intra_cost )
1423 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",
1425 (double)i_intra_cost, (double)i_inter_cost,
1426 (double)i_inter_cost / i_intra_cost,
1428 i_mb_i, i_mb_p, i_mb_s );
1430 /* Restore frame num */
1433 for( b = 0; h->frames.current[b] && IS_X264_TYPE_B( h->frames.current[b]->i_type ); b++ );
1436 /* If using B-frames, force GOP to be closed.
1437 * Even if this frame is going to be I and not IDR, forcing a
1438 * P-frame before the scenecut will probably help compression.
1440 * We don't yet know exactly which frame is the scene cut, so
1441 * we can't assign an I-frame. Instead, change the previous
1442 * B-frame to P, and rearrange coding order. */
1444 if( h->param.b_bframe_adaptive || b > 1 )
1445 h->fenc->i_type = X264_TYPE_AUTO;
1446 x264_frame_sort_pts( h->frames.current );
1447 x264_frame_push( h->frames.next, h->fenc );
1448 h->fenc = h->frames.current[b-1];
1449 h->frames.current[b-1] = NULL;
1450 h->fenc->i_type = X264_TYPE_P;
1451 x264_frame_sort_dts( h->frames.current );
1453 /* Do IDR if needed */
1454 else if( i_gop_size >= h->param.i_keyint_min )
1461 /* Reinit field of fenc */
1462 h->fenc->i_type = X264_TYPE_IDR;
1465 /* Put enqueued frames back in the pool */
1466 while( (tmp = x264_frame_get( h->frames.current ) ) != NULL )
1467 x264_frame_put( h->frames.next, tmp );
1468 x264_frame_sort_pts( h->frames.next );
1472 h->fenc->i_type = X264_TYPE_I;
1478 /* End bitstream, set output */
1479 *pi_nal = h->out.i_nal;
1480 *pp_nal = h->out.nal;
1482 /* Set output picture properties */
1483 if( i_slice_type == SLICE_TYPE_I )
1484 pic_out->i_type = i_nal_type == NAL_SLICE_IDR ? X264_TYPE_IDR : X264_TYPE_I;
1485 else if( i_slice_type == SLICE_TYPE_P )
1486 pic_out->i_type = X264_TYPE_P;
1488 pic_out->i_type = X264_TYPE_B;
1489 pic_out->i_pts = h->fenc->i_pts;
1491 pic_out->img.i_plane = h->fdec->i_plane;
1492 for(i = 0; i < 4; i++){
1493 pic_out->img.i_stride[i] = h->fdec->i_stride[i];
1494 pic_out->img.plane[i] = h->fdec->plane[i];
1497 /* ---------------------- Update encoder state ------------------------- */
1500 x264_cpu_restore( h->param.cpu );
1501 x264_ratecontrol_end( h, i_frame_size * 8 );
1503 /* handle references */
1504 if( i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE )
1505 x264_reference_update( h );
1506 #ifdef DEBUG_DUMP_FRAME
1508 x264_fdec_deblock( h );
1510 x264_frame_put( h->frames.unused, h->fenc );
1512 /* increase frame count */
1515 /* restore CPU state (before using float again) */
1516 x264_cpu_restore( h->param.cpu );
1518 x264_noise_reduction_update( h );
1520 TIMER_STOP( i_mtime_encode_frame );
1522 /* ---------------------- Compute/Print statistics --------------------- */
1524 h->stat.i_slice_count[i_slice_type]++;
1525 h->stat.i_slice_size[i_slice_type] += i_frame_size + NALU_OVERHEAD;
1526 h->stat.i_slice_qp[i_slice_type] += i_global_qp;
1528 for( i = 0; i < 19; i++ )
1529 h->stat.i_mb_count[h->sh.i_type][i] += h->stat.frame.i_mb_count[i];
1530 for( i = 0; i < 2; i++ )
1531 h->stat.i_mb_count_8x8dct[i] += h->stat.frame.i_mb_count_8x8dct[i];
1532 if( h->sh.i_type != SLICE_TYPE_I )
1534 for( i = 0; i < 7; i++ )
1535 h->stat.i_mb_count_size[h->sh.i_type][i] += h->stat.frame.i_mb_count_size[i];
1536 for( i = 0; i < 16; i++ )
1537 h->stat.i_mb_count_ref[h->sh.i_type][i] += h->stat.frame.i_mb_count_ref[i];
1539 if( i_slice_type == SLICE_TYPE_B )
1541 h->stat.i_direct_frames[ h->sh.b_direct_spatial_mv_pred ] ++;
1542 if( h->mb.b_direct_auto_write )
1544 //FIXME somewhat arbitrary time constants
1545 if( h->stat.i_direct_score[0] + h->stat.i_direct_score[1] > h->mb.i_mb_count )
1547 for( i = 0; i < 2; i++ )
1548 h->stat.i_direct_score[i] = h->stat.i_direct_score[i] * 9/10;
1550 for( i = 0; i < 2; i++ )
1551 h->stat.i_direct_score[i] += h->stat.frame.i_direct_score[i];
1555 if( h->param.analyse.b_psnr )
1557 int64_t i_sqe_y, i_sqe_u, i_sqe_v;
1560 i_sqe_y = x264_pixel_ssd_wxh( &h->pixf, frame_psnr->plane[0], frame_psnr->i_stride[0], h->fenc->plane[0], h->fenc->i_stride[0], h->param.i_width, h->param.i_height );
1561 i_sqe_u = x264_pixel_ssd_wxh( &h->pixf, frame_psnr->plane[1], frame_psnr->i_stride[1], h->fenc->plane[1], h->fenc->i_stride[1], h->param.i_width/2, h->param.i_height/2);
1562 i_sqe_v = x264_pixel_ssd_wxh( &h->pixf, frame_psnr->plane[2], frame_psnr->i_stride[2], h->fenc->plane[2], h->fenc->i_stride[2], h->param.i_width/2, h->param.i_height/2);
1563 x264_cpu_restore( h->param.cpu );
1565 h->stat.i_sqe_global[i_slice_type] += i_sqe_y + i_sqe_u + i_sqe_v;
1566 h->stat.f_psnr_average[i_slice_type] += x264_psnr( i_sqe_y + i_sqe_u + i_sqe_v, 3 * h->param.i_width * h->param.i_height / 2 );
1567 h->stat.f_psnr_mean_y[i_slice_type] += x264_psnr( i_sqe_y, h->param.i_width * h->param.i_height );
1568 h->stat.f_psnr_mean_u[i_slice_type] += x264_psnr( i_sqe_u, h->param.i_width * h->param.i_height / 4 );
1569 h->stat.f_psnr_mean_v[i_slice_type] += x264_psnr( i_sqe_v, h->param.i_width * h->param.i_height / 4 );
1571 snprintf( psz_message, 80, " PSNR Y:%2.2f U:%2.2f V:%2.2f",
1572 x264_psnr( i_sqe_y, h->param.i_width * h->param.i_height ),
1573 x264_psnr( i_sqe_u, h->param.i_width * h->param.i_height / 4),
1574 x264_psnr( i_sqe_v, h->param.i_width * h->param.i_height / 4) );
1575 psz_message[79] = '\0';
1579 psz_message[0] = '\0';
1582 x264_log( h, X264_LOG_DEBUG,
1583 "frame=%4d QP=%i NAL=%d Slice:%c Poc:%-3d I:%-4d P:%-4d SKIP:%-4d size=%d bytes%s\n",
1587 i_slice_type == SLICE_TYPE_I ? 'I' : (i_slice_type == SLICE_TYPE_P ? 'P' : 'B' ),
1589 h->stat.frame.i_mb_count_i,
1590 h->stat.frame.i_mb_count_p,
1591 h->stat.frame.i_mb_count_skip,
1596 #ifdef DEBUG_MB_TYPE
1598 static const char mb_chars[] = { 'i', 'i', 'I', 'C', 'P', '8', 'S',
1599 'D', '<', 'X', 'B', 'X', '>', 'B', 'B', 'B', 'B', '8', 'S' };
1601 for( mb_xy = 0; mb_xy < h->sps->i_mb_width * h->sps->i_mb_height; mb_xy++ )
1603 if( h->mb.type[mb_xy] < 19 && h->mb.type[mb_xy] >= 0 )
1604 fprintf( stderr, "%c ", mb_chars[ h->mb.type[mb_xy] ] );
1606 fprintf( stderr, "? " );
1608 if( (mb_xy+1) % h->sps->i_mb_width == 0 )
1609 fprintf( stderr, "\n" );
1614 #ifdef DEBUG_DUMP_FRAME
1615 /* Dump reconstructed frame */
1616 x264_frame_dump( h, frame_psnr, "fdec.yuv" );
1621 /****************************************************************************
1622 * x264_encoder_close:
1623 ****************************************************************************/
1624 void x264_encoder_close ( x264_t *h )
1626 #ifdef DEBUG_BENCHMARK
1627 int64_t i_mtime_total = i_mtime_analyse + i_mtime_encode + i_mtime_write + i_mtime_filter + 1;
1629 int64_t i_yuv_size = 3 * h->param.i_width * h->param.i_height / 2;
1632 #ifdef DEBUG_BENCHMARK
1633 x264_log( h, X264_LOG_INFO,
1634 "analyse=%d(%lldms) encode=%d(%lldms) write=%d(%lldms) filter=%d(%lldms)\n",
1635 (int)(100*i_mtime_analyse/i_mtime_total), i_mtime_analyse/1000,
1636 (int)(100*i_mtime_encode/i_mtime_total), i_mtime_encode/1000,
1637 (int)(100*i_mtime_write/i_mtime_total), i_mtime_write/1000,
1638 (int)(100*i_mtime_filter/i_mtime_total), i_mtime_filter/1000 );
1641 /* Slices used and PSNR */
1642 for( i=0; i<5; i++ )
1644 static const int slice_order[] = { SLICE_TYPE_I, SLICE_TYPE_SI, SLICE_TYPE_P, SLICE_TYPE_SP, SLICE_TYPE_B };
1645 static const char *slice_name[] = { "P", "B", "I", "SP", "SI" };
1646 int i_slice = slice_order[i];
1648 if( h->stat.i_slice_count[i_slice] > 0 )
1650 const int i_count = h->stat.i_slice_count[i_slice];
1651 if( h->param.analyse.b_psnr )
1653 x264_log( h, X264_LOG_INFO,
1654 "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",
1655 slice_name[i_slice],
1657 (double)h->stat.i_slice_qp[i_slice] / i_count,
1658 (double)h->stat.i_slice_size[i_slice] / i_count,
1659 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,
1660 h->stat.f_psnr_average[i_slice] / i_count,
1661 x264_psnr( h->stat.i_sqe_global[i_slice], i_count * i_yuv_size ) );
1665 x264_log( h, X264_LOG_INFO,
1666 "slice %s:%-5d Avg QP:%5.2f size:%6.0f\n",
1667 slice_name[i_slice],
1669 (double)h->stat.i_slice_qp[i_slice] / i_count,
1670 (double)h->stat.i_slice_size[i_slice] / i_count );
1676 if( h->stat.i_slice_count[SLICE_TYPE_I] > 0 )
1678 const int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_I];
1679 const double i_count = h->stat.i_slice_count[SLICE_TYPE_I] * h->mb.i_mb_count / 100.0;
1680 x264_log( h, X264_LOG_INFO,
1681 "mb I I16..4: %4.1f%% %4.1f%% %4.1f%%\n",
1682 i_mb_count[I_16x16]/ i_count,
1683 i_mb_count[I_8x8] / i_count,
1684 i_mb_count[I_4x4] / i_count );
1686 if( h->stat.i_slice_count[SLICE_TYPE_P] > 0 )
1688 const int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_P];
1689 const int64_t *i_mb_size = h->stat.i_mb_count_size[SLICE_TYPE_P];
1690 const double i_count = h->stat.i_slice_count[SLICE_TYPE_P] * h->mb.i_mb_count / 100.0;
1691 x264_log( h, X264_LOG_INFO,
1692 "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",
1693 i_mb_count[I_16x16]/ i_count,
1694 i_mb_count[I_8x8] / i_count,
1695 i_mb_count[I_4x4] / i_count,
1696 i_mb_size[PIXEL_16x16] / (i_count*4),
1697 (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
1698 i_mb_size[PIXEL_8x8] / (i_count*4),
1699 (i_mb_size[PIXEL_8x4] + i_mb_size[PIXEL_4x8]) / (i_count*4),
1700 i_mb_size[PIXEL_4x4] / (i_count*4),
1701 i_mb_count[P_SKIP] / i_count );
1703 if( h->stat.i_slice_count[SLICE_TYPE_B] > 0 )
1705 const int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_B];
1706 const int64_t *i_mb_size = h->stat.i_mb_count_size[SLICE_TYPE_B];
1707 const double i_count = h->stat.i_slice_count[SLICE_TYPE_B] * h->mb.i_mb_count / 100.0;
1708 x264_log( h, X264_LOG_INFO,
1709 "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",
1710 i_mb_count[I_16x16] / i_count,
1711 i_mb_count[I_8x8] / i_count,
1712 i_mb_count[I_4x4] / i_count,
1713 i_mb_size[PIXEL_16x16] / (i_count*4),
1714 (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
1715 i_mb_size[PIXEL_8x8] / (i_count*4),
1716 i_mb_count[B_DIRECT] / i_count,
1717 i_mb_count[B_SKIP] / i_count );
1720 x264_ratecontrol_summary( h );
1722 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 )
1724 const int i_count = h->stat.i_slice_count[SLICE_TYPE_I] +
1725 h->stat.i_slice_count[SLICE_TYPE_P] +
1726 h->stat.i_slice_count[SLICE_TYPE_B];
1727 float fps = (float) h->param.i_fps_num / h->param.i_fps_den;
1728 #define SUM3(p) (p[SLICE_TYPE_I] + p[SLICE_TYPE_P] + p[SLICE_TYPE_B])
1729 #define SUM3b(p,o) (p[SLICE_TYPE_I][o] + p[SLICE_TYPE_P][o] + p[SLICE_TYPE_B][o])
1730 float f_bitrate = fps * SUM3(h->stat.i_slice_size) / i_count / 125;
1732 if( h->param.analyse.b_transform_8x8 )
1734 int64_t i_i8x8 = SUM3b( h->stat.i_mb_count, I_8x8 );
1735 int64_t i_intra = i_i8x8 + SUM3b( h->stat.i_mb_count, I_4x4 )
1736 + SUM3b( h->stat.i_mb_count, I_16x16 );
1737 x264_log( h, X264_LOG_INFO, "8x8 transform intra:%.1f%% inter:%.1f%%\n",
1738 100. * i_i8x8 / i_intra,
1739 100. * h->stat.i_mb_count_8x8dct[1] / h->stat.i_mb_count_8x8dct[0] );
1742 if( h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO
1743 && h->stat.i_slice_count[SLICE_TYPE_B] )
1745 x264_log( h, X264_LOG_INFO, "direct mvs spatial:%.1f%% temporal:%.1f%%\n",
1746 h->stat.i_direct_frames[1] * 100. / h->stat.i_slice_count[SLICE_TYPE_B],
1747 h->stat.i_direct_frames[0] * 100. / h->stat.i_slice_count[SLICE_TYPE_B] );
1750 if( h->param.i_frame_reference > 1 )
1753 for( i_slice = 0; i_slice < 2; i_slice++ )
1759 for( i = 0; i < h->param.i_frame_reference; i++ )
1760 if( h->stat.i_mb_count_ref[i_slice][i] )
1762 i_den += h->stat.i_mb_count_ref[i_slice][i];
1767 for( i = 0; i <= i_max; i++ )
1768 p += sprintf( p, " %4.1f%%", 100. * h->stat.i_mb_count_ref[i_slice][i] / i_den );
1769 x264_log( h, X264_LOG_INFO, "ref %c %s\n", i_slice==SLICE_TYPE_P ? 'P' : 'B', buf );
1773 if( h->param.analyse.b_psnr )
1774 x264_log( h, X264_LOG_INFO,
1775 "PSNR Mean Y:%6.3f U:%6.3f V:%6.3f Avg:%6.3f Global:%6.3f kb/s:%.2f\n",
1776 SUM3( h->stat.f_psnr_mean_y ) / i_count,
1777 SUM3( h->stat.f_psnr_mean_u ) / i_count,
1778 SUM3( h->stat.f_psnr_mean_v ) / i_count,
1779 SUM3( h->stat.f_psnr_average ) / i_count,
1780 x264_psnr( SUM3( h->stat.i_sqe_global ), i_count * i_yuv_size ),
1783 x264_log( h, X264_LOG_INFO, "kb/s:%.1f\n", f_bitrate );
1787 for( i = 0; i < X264_BFRAME_MAX + 3; i++ )
1789 if( h->frames.current[i] ) x264_frame_delete( h->frames.current[i] );
1790 if( h->frames.next[i] ) x264_frame_delete( h->frames.next[i] );
1791 if( h->frames.unused[i] ) x264_frame_delete( h->frames.unused[i] );
1794 for( i = 0; i < h->frames.i_max_dpb; i++ )
1796 x264_frame_delete( h->frames.reference[i] );
1800 x264_ratecontrol_delete( h );
1803 if( h->param.rc.psz_stat_out )
1804 free( h->param.rc.psz_stat_out );
1805 if( h->param.rc.psz_stat_in )
1806 free( h->param.rc.psz_stat_in );
1807 if( h->param.rc.psz_rc_eq )
1808 free( h->param.rc.psz_rc_eq );
1810 x264_macroblock_cache_end( h );
1811 x264_free( h->out.p_bitstream );
1812 for( i = 1; i < h->param.i_threads; i++ )
1813 x264_free( h->thread[i] );