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, "a" );
84 fseek( f, 0, SEEK_END );
86 for( i = 0; i < fr->i_plane; i++ )
88 for( y = 0; y < h->param.i_height / ( i == 0 ? 1 : 2 ); y++ )
90 fwrite( &fr->plane[i][y*fr->i_stride[i]], 1, h->param.i_width / ( i == 0 ? 1 : 2 ), f );
98 /* Fill "default" values */
99 static void x264_slice_header_init( x264_t *h, x264_slice_header_t *sh,
100 x264_sps_t *sps, x264_pps_t *pps,
101 int i_type, int i_idr_pic_id, int i_frame, int i_qp )
103 x264_param_t *param = &h->param;
106 /* First we fill all field */
112 sh->i_last_mb = h->sps->i_mb_width * h->sps->i_mb_height;
113 sh->i_pps_id = pps->i_id;
115 sh->i_frame_num = i_frame;
117 sh->b_field_pic = 0; /* Not field support for now */
118 sh->b_bottom_field = 1; /* not yet used */
120 sh->i_idr_pic_id = i_idr_pic_id;
122 /* poc stuff, fixed later */
124 sh->i_delta_poc_bottom = 0;
125 sh->i_delta_poc[0] = 0;
126 sh->i_delta_poc[1] = 0;
128 sh->i_redundant_pic_cnt = 0;
130 if( !h->mb.b_direct_auto_read )
132 if( h->mb.b_direct_auto_write )
133 sh->b_direct_spatial_mv_pred = ( h->stat.i_direct_score[1] > h->stat.i_direct_score[0] );
135 sh->b_direct_spatial_mv_pred = ( param->analyse.i_direct_mv_pred == X264_DIRECT_PRED_SPATIAL );
137 /* else b_direct_spatial_mv_pred was read from the 2pass statsfile */
139 sh->b_num_ref_idx_override = 0;
140 sh->i_num_ref_idx_l0_active = 1;
141 sh->i_num_ref_idx_l1_active = 1;
143 sh->b_ref_pic_list_reordering_l0 = h->b_ref_reorder[0];
144 sh->b_ref_pic_list_reordering_l1 = h->b_ref_reorder[1];
146 /* If the ref list isn't in the default order, construct reordering header */
147 /* List1 reordering isn't needed yet */
148 if( sh->b_ref_pic_list_reordering_l0 )
150 int pred_frame_num = i_frame;
151 for( i = 0; i < h->i_ref0; i++ )
153 int diff = h->fref0[i]->i_frame_num - pred_frame_num;
155 x264_log( h, X264_LOG_ERROR, "diff frame num == 0\n" );
156 sh->ref_pic_list_order[0][i].idc = ( diff > 0 );
157 sh->ref_pic_list_order[0][i].arg = abs( diff ) - 1;
158 pred_frame_num = h->fref0[i]->i_frame_num;
162 sh->i_cabac_init_idc = param->i_cabac_init_idc;
165 sh->i_qp_delta = i_qp - pps->i_pic_init_qp;
166 sh->b_sp_for_swidth = 0;
169 /* If effective qp <= 15, deblocking would have no effect anyway */
170 if( param->b_deblocking_filter
171 && ( h->mb.b_variable_qp
172 || 15 < i_qp + 2 * X264_MAX(param->i_deblocking_filter_alphac0, param->i_deblocking_filter_beta) ) )
174 sh->i_disable_deblocking_filter_idc = 0;
178 sh->i_disable_deblocking_filter_idc = 1;
180 sh->i_alpha_c0_offset = param->i_deblocking_filter_alphac0 << 1;
181 sh->i_beta_offset = param->i_deblocking_filter_beta << 1;
184 static void x264_slice_header_write( bs_t *s, x264_slice_header_t *sh, int i_nal_ref_idc )
188 bs_write_ue( s, sh->i_first_mb );
189 bs_write_ue( s, sh->i_type + 5 ); /* same type things */
190 bs_write_ue( s, sh->i_pps_id );
191 bs_write( s, sh->sps->i_log2_max_frame_num, sh->i_frame_num );
193 if( sh->i_idr_pic_id >= 0 ) /* NAL IDR */
195 bs_write_ue( s, sh->i_idr_pic_id );
198 if( sh->sps->i_poc_type == 0 )
200 bs_write( s, sh->sps->i_log2_max_poc_lsb, sh->i_poc_lsb );
201 if( sh->pps->b_pic_order && !sh->b_field_pic )
203 bs_write_se( s, sh->i_delta_poc_bottom );
206 else if( sh->sps->i_poc_type == 1 && !sh->sps->b_delta_pic_order_always_zero )
208 bs_write_se( s, sh->i_delta_poc[0] );
209 if( sh->pps->b_pic_order && !sh->b_field_pic )
211 bs_write_se( s, sh->i_delta_poc[1] );
215 if( sh->pps->b_redundant_pic_cnt )
217 bs_write_ue( s, sh->i_redundant_pic_cnt );
220 if( sh->i_type == SLICE_TYPE_B )
222 bs_write1( s, sh->b_direct_spatial_mv_pred );
224 if( sh->i_type == SLICE_TYPE_P || sh->i_type == SLICE_TYPE_SP || sh->i_type == SLICE_TYPE_B )
226 bs_write1( s, sh->b_num_ref_idx_override );
227 if( sh->b_num_ref_idx_override )
229 bs_write_ue( s, sh->i_num_ref_idx_l0_active - 1 );
230 if( sh->i_type == SLICE_TYPE_B )
232 bs_write_ue( s, sh->i_num_ref_idx_l1_active - 1 );
237 /* ref pic list reordering */
238 if( sh->i_type != SLICE_TYPE_I )
240 bs_write1( s, sh->b_ref_pic_list_reordering_l0 );
241 if( sh->b_ref_pic_list_reordering_l0 )
243 for( i = 0; i < sh->i_num_ref_idx_l0_active; i++ )
245 bs_write_ue( s, sh->ref_pic_list_order[0][i].idc );
246 bs_write_ue( s, sh->ref_pic_list_order[0][i].arg );
252 if( sh->i_type == SLICE_TYPE_B )
254 bs_write1( s, sh->b_ref_pic_list_reordering_l1 );
255 if( sh->b_ref_pic_list_reordering_l1 )
257 for( i = 0; i < sh->i_num_ref_idx_l1_active; i++ )
259 bs_write_ue( s, sh->ref_pic_list_order[1][i].idc );
260 bs_write_ue( s, sh->ref_pic_list_order[1][i].arg );
266 if( ( sh->pps->b_weighted_pred && ( sh->i_type == SLICE_TYPE_P || sh->i_type == SLICE_TYPE_SP ) ) ||
267 ( sh->pps->b_weighted_bipred == 1 && sh->i_type == SLICE_TYPE_B ) )
272 if( i_nal_ref_idc != 0 )
274 if( sh->i_idr_pic_id >= 0 )
276 bs_write1( s, 0 ); /* no output of prior pics flag */
277 bs_write1( s, 0 ); /* long term reference flag */
281 bs_write1( s, 0 ); /* adaptive_ref_pic_marking_mode_flag */
285 if( sh->pps->b_cabac && sh->i_type != SLICE_TYPE_I )
287 bs_write_ue( s, sh->i_cabac_init_idc );
289 bs_write_se( s, sh->i_qp_delta ); /* slice qp delta */
291 if( sh->pps->b_deblocking_filter_control )
293 bs_write_ue( s, sh->i_disable_deblocking_filter_idc );
294 if( sh->i_disable_deblocking_filter_idc != 1 )
296 bs_write_se( s, sh->i_alpha_c0_offset >> 1 );
297 bs_write_se( s, sh->i_beta_offset >> 1 );
302 /****************************************************************************
304 ****************************************************************************
305 ****************************** External API*********************************
306 ****************************************************************************
308 ****************************************************************************/
310 static int x264_validate_parameters( x264_t *h )
312 if( h->param.i_width <= 0 || h->param.i_height <= 0 )
314 x264_log( h, X264_LOG_ERROR, "invalid width x height (%dx%d)\n",
315 h->param.i_width, h->param.i_height );
319 if( h->param.i_width % 2 || h->param.i_height % 2 )
321 x264_log( h, X264_LOG_ERROR, "width or height not divisible by 2 (%dx%d)\n",
322 h->param.i_width, h->param.i_height );
325 if( h->param.i_csp != X264_CSP_I420 )
327 x264_log( h, X264_LOG_ERROR, "invalid CSP (only I420 supported)\n" );
331 h->param.i_threads = x264_clip3( h->param.i_threads, 1, X264_SLICE_MAX );
332 h->param.i_threads = X264_MIN( h->param.i_threads, (h->param.i_height + 15) / 16 );
334 if( h->param.i_threads > 1 )
336 x264_log( h, X264_LOG_WARNING, "not compiled with pthread support!\n");
337 x264_log( h, X264_LOG_WARNING, "multislicing anyway, but you won't see any speed gain.\n" );
341 if( h->param.rc.b_cbr )
342 h->param.rc.i_rf_constant = 0;
343 if( h->param.rc.i_rf_constant > 0 )
344 h->param.rc.i_qp_constant = h->param.rc.i_rf_constant;
345 h->param.rc.i_rf_constant = x264_clip3( h->param.rc.i_rf_constant, 0, 51 );
346 h->param.rc.i_qp_constant = x264_clip3( h->param.rc.i_qp_constant, 0, 51 );
347 if( !h->param.rc.b_cbr && h->param.rc.i_qp_constant == 0 )
349 h->mb.b_lossless = 1;
350 h->param.analyse.b_transform_8x8 = 0;
351 h->param.i_cqm_preset = X264_CQM_FLAT;
352 h->param.psz_cqm_file = NULL;
353 h->param.rc.f_ip_factor = 1;
354 h->param.rc.f_pb_factor = 1;
355 h->param.analyse.b_psnr = 0;
356 h->param.analyse.i_chroma_qp_offset = 0;
357 h->param.analyse.i_trellis = 0;
358 h->param.analyse.b_fast_pskip = 0;
359 h->param.analyse.i_noise_reduction = 0;
360 h->param.analyse.i_subpel_refine = x264_clip3( h->param.analyse.i_subpel_refine, 1, 6 );
363 if( ( h->param.i_width % 16 || h->param.i_height % 16 ) && !h->mb.b_lossless )
365 x264_log( h, X264_LOG_WARNING,
366 "width or height not divisible by 16 (%dx%d), compression will suffer.\n",
367 h->param.i_width, h->param.i_height );
370 h->param.i_frame_reference = x264_clip3( h->param.i_frame_reference, 1, 16 );
371 if( h->param.i_keyint_max <= 0 )
372 h->param.i_keyint_max = 1;
373 h->param.i_keyint_min = x264_clip3( h->param.i_keyint_min, 1, h->param.i_keyint_max/2+1 );
375 h->param.i_bframe = x264_clip3( h->param.i_bframe, 0, X264_BFRAME_MAX );
376 h->param.i_bframe_bias = x264_clip3( h->param.i_bframe_bias, -90, 100 );
377 h->param.b_bframe_pyramid = h->param.b_bframe_pyramid && h->param.i_bframe > 1;
378 h->param.b_bframe_adaptive = h->param.b_bframe_adaptive && h->param.i_bframe > 0;
379 h->param.analyse.b_weighted_bipred = h->param.analyse.b_weighted_bipred && h->param.i_bframe > 0;
380 h->mb.b_direct_auto_write = h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO
382 && ( h->param.rc.b_stat_write || !h->param.rc.b_stat_read );
384 h->param.i_deblocking_filter_alphac0 = x264_clip3( h->param.i_deblocking_filter_alphac0, -6, 6 );
385 h->param.i_deblocking_filter_beta = x264_clip3( h->param.i_deblocking_filter_beta, -6, 6 );
387 h->param.i_cabac_init_idc = x264_clip3( h->param.i_cabac_init_idc, 0, 2 );
389 if( h->param.i_cqm_preset < X264_CQM_FLAT || h->param.i_cqm_preset > X264_CQM_CUSTOM )
390 h->param.i_cqm_preset = X264_CQM_FLAT;
392 if( h->param.analyse.i_me_method < X264_ME_DIA ||
393 h->param.analyse.i_me_method > X264_ME_ESA )
394 h->param.analyse.i_me_method = X264_ME_HEX;
395 if( h->param.analyse.i_me_range < 4 )
396 h->param.analyse.i_me_range = 4;
397 if( h->param.analyse.i_me_range > 16 && h->param.analyse.i_me_method <= X264_ME_HEX )
398 h->param.analyse.i_me_range = 16;
399 h->param.analyse.i_subpel_refine = x264_clip3( h->param.analyse.i_subpel_refine, 1, 7 );
400 h->param.analyse.b_bframe_rdo = h->param.analyse.b_bframe_rdo && h->param.analyse.i_subpel_refine >= 6;
401 h->param.analyse.b_mixed_references = h->param.analyse.b_mixed_references && h->param.i_frame_reference > 1;
402 h->param.analyse.inter &= X264_ANALYSE_PSUB16x16|X264_ANALYSE_PSUB8x8|X264_ANALYSE_BSUB16x16|
403 X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
404 h->param.analyse.intra &= X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
405 if( !(h->param.analyse.inter & X264_ANALYSE_PSUB16x16) )
406 h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
407 if( !h->param.analyse.b_transform_8x8 )
409 h->param.analyse.inter &= ~X264_ANALYSE_I8x8;
410 h->param.analyse.intra &= ~X264_ANALYSE_I8x8;
412 h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -12, 12);
413 if( !h->param.b_cabac )
414 h->param.analyse.i_trellis = 0;
415 h->param.analyse.i_trellis = x264_clip3( h->param.analyse.i_trellis, 0, 2 );
416 h->param.analyse.i_noise_reduction = x264_clip3( h->param.analyse.i_noise_reduction, 0, 1<<16 );
419 const x264_level_t *l = x264_levels;
420 while( l->level_idc != 0 && l->level_idc != h->param.i_level_idc )
422 if( l->level_idc == 0 )
424 x264_log( h, X264_LOG_ERROR, "invalid level_idc: %d\n", h->param.i_level_idc );
427 if( h->param.analyse.i_mv_range <= 0 )
428 h->param.analyse.i_mv_range = l->mv_range;
430 h->param.analyse.i_mv_range = x264_clip3(h->param.analyse.i_mv_range, 32, 2048);
433 if( h->param.rc.f_qblur < 0 )
434 h->param.rc.f_qblur = 0;
435 if( h->param.rc.f_complexity_blur < 0 )
436 h->param.rc.f_complexity_blur = 0;
438 h->param.i_sps_id &= 31;
440 /* ensure the booleans are 0 or 1 so they can be used in math */
441 #define BOOLIFY(x) h->param.x = !!h->param.x
443 BOOLIFY( b_deblocking_filter );
444 BOOLIFY( analyse.b_transform_8x8 );
445 BOOLIFY( analyse.b_bidir_me );
446 BOOLIFY( analyse.b_chroma_me );
447 BOOLIFY( analyse.b_fast_pskip );
449 BOOLIFY( rc.b_stat_write );
450 BOOLIFY( rc.b_stat_read );
456 /****************************************************************************
458 ****************************************************************************/
459 x264_t *x264_encoder_open ( x264_param_t *param )
461 x264_t *h = x264_malloc( sizeof( x264_t ) );
464 memset( h, 0, sizeof( x264_t ) );
466 /* Create a copy of param */
467 memcpy( &h->param, param, sizeof( x264_param_t ) );
469 if( x264_validate_parameters( h ) < 0 )
475 if( h->param.psz_cqm_file )
476 if( x264_cqm_parse_file( h, h->param.psz_cqm_file ) < 0 )
482 if( h->param.rc.psz_stat_out )
483 h->param.rc.psz_stat_out = strdup( h->param.rc.psz_stat_out );
484 if( h->param.rc.psz_stat_in )
485 h->param.rc.psz_stat_in = strdup( h->param.rc.psz_stat_in );
486 if( h->param.rc.psz_rc_eq )
487 h->param.rc.psz_rc_eq = strdup( h->param.rc.psz_rc_eq );
490 if( h->param.vui.i_sar_width > 0 && h->param.vui.i_sar_height > 0 )
492 int i_w = param->vui.i_sar_width;
493 int i_h = param->vui.i_sar_height;
495 x264_reduce_fraction( &i_w, &i_h );
497 while( i_w > 65535 || i_h > 65535 )
503 h->param.vui.i_sar_width = 0;
504 h->param.vui.i_sar_height = 0;
505 if( i_w == 0 || i_h == 0 )
507 x264_log( h, X264_LOG_ERROR, "cannot create valid sample aspect ratio\n" );
511 x264_log( h, X264_LOG_INFO, "using SAR=%d/%d\n", i_w, i_h );
512 h->param.vui.i_sar_width = i_w;
513 h->param.vui.i_sar_height = i_h;
517 x264_reduce_fraction( &h->param.i_fps_num, &h->param.i_fps_den );
521 h->out.i_bitstream = X264_MAX( 1000000, h->param.i_width * h->param.i_height * 1.7
522 * ( h->param.rc.b_cbr ? pow( 0.5, h->param.rc.i_qp_min )
523 : pow( 0.5, h->param.rc.i_qp_constant ) * X264_MAX( 1, h->param.rc.f_ip_factor )));
524 h->out.p_bitstream = x264_malloc( h->out.i_bitstream );
530 h->sps = &h->sps_array[0];
531 x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
533 h->pps = &h->pps_array[0];
534 x264_pps_init( h->pps, h->param.i_sps_id, &h->param, h->sps);
536 x264_validate_levels( h );
540 h->mb.i_mb_count = h->sps->i_mb_width * h->sps->i_mb_height;
543 h->frames.i_delay = h->param.i_bframe;
544 h->frames.i_max_ref0 = h->param.i_frame_reference;
545 h->frames.i_max_ref1 = h->sps->vui.i_num_reorder_frames;
546 h->frames.i_max_dpb = h->sps->vui.i_max_dec_frame_buffering + 1;
547 h->frames.b_have_lowres = !h->param.rc.b_stat_read
548 && ( h->param.rc.b_cbr || h->param.rc.i_rf_constant || h->param.b_bframe_adaptive );
550 for( i = 0; i < X264_BFRAME_MAX + 3; i++ )
552 h->frames.current[i] = NULL;
553 h->frames.next[i] = NULL;
554 h->frames.unused[i] = NULL;
556 for( i = 0; i < 1 + h->frames.i_delay; i++ )
558 h->frames.unused[i] = x264_frame_new( h );
560 for( i = 0; i < h->frames.i_max_dpb; i++ )
562 h->frames.reference[i] = x264_frame_new( h );
564 h->frames.reference[h->frames.i_max_dpb] = NULL;
565 h->frames.i_last_idr = - h->param.i_keyint_max;
566 h->frames.i_input = 0;
567 h->frames.last_nonb = NULL;
572 h->fdec = h->frames.reference[0];
574 x264_macroblock_cache_init( h );
577 /* init CPU functions */
578 x264_predict_16x16_init( h->param.cpu, h->predict_16x16 );
579 x264_predict_8x8c_init( h->param.cpu, h->predict_8x8c );
580 x264_predict_8x8_init( h->param.cpu, h->predict_8x8 );
581 x264_predict_4x4_init( h->param.cpu, h->predict_4x4 );
583 x264_pixel_init( h->param.cpu, &h->pixf );
584 x264_dct_init( h->param.cpu, &h->dctf );
585 x264_mc_init( h->param.cpu, &h->mc );
586 x264_csp_init( h->param.cpu, h->param.i_csp, &h->csp );
587 x264_quant_init( h, h->param.cpu, &h->quantf );
588 x264_deblock_init( h->param.cpu, &h->loopf );
590 memcpy( h->pixf.mbcmp,
591 ( h->mb.b_lossless || h->param.analyse.i_subpel_refine <= 1 ) ? h->pixf.sad : h->pixf.satd,
592 sizeof(h->pixf.mbcmp) );
595 if( x264_ratecontrol_new( h ) < 0 )
598 x264_log( h, X264_LOG_INFO, "using cpu capabilities %s%s%s%s%s%s\n",
599 param->cpu&X264_CPU_MMX ? "MMX " : "",
600 param->cpu&X264_CPU_MMXEXT ? "MMXEXT " : "",
601 param->cpu&X264_CPU_SSE ? "SSE " : "",
602 param->cpu&X264_CPU_SSE2 ? "SSE2 " : "",
603 param->cpu&X264_CPU_3DNOW ? "3DNow! " : "",
604 param->cpu&X264_CPU_ALTIVEC ? "Altivec " : "" );
608 for( i = 1; i < param->i_threads; i++ )
609 h->thread[i] = x264_malloc( sizeof(x264_t) );
614 /****************************************************************************
615 * x264_encoder_reconfig:
616 ****************************************************************************/
617 int x264_encoder_reconfig( x264_t *h, x264_param_t *param )
619 h->param.i_bframe_bias = param->i_bframe_bias;
620 h->param.i_deblocking_filter_alphac0 = param->i_deblocking_filter_alphac0;
621 h->param.i_deblocking_filter_beta = param->i_deblocking_filter_beta;
622 h->param.analyse.i_me_method = param->analyse.i_me_method;
623 h->param.analyse.i_me_range = param->analyse.i_me_range;
624 h->param.analyse.i_subpel_refine = param->analyse.i_subpel_refine;
625 h->param.analyse.i_trellis = param->analyse.i_trellis;
626 h->param.analyse.intra = param->analyse.intra;
627 h->param.analyse.inter = param->analyse.inter;
629 memcpy( h->pixf.mbcmp,
630 ( h->mb.b_lossless || h->param.analyse.i_subpel_refine <= 1 ) ? h->pixf.sad : h->pixf.satd,
631 sizeof(h->pixf.mbcmp) );
633 return x264_validate_parameters( h );
637 static void x264_nal_start( x264_t *h, int i_type, int i_ref_idc )
639 x264_nal_t *nal = &h->out.nal[h->out.i_nal];
641 nal->i_ref_idc = i_ref_idc;
642 nal->i_type = i_type;
644 bs_align_0( &h->out.bs ); /* not needed */
647 nal->p_payload= &h->out.p_bitstream[bs_pos( &h->out.bs) / 8];
649 static void x264_nal_end( x264_t *h )
651 x264_nal_t *nal = &h->out.nal[h->out.i_nal];
653 bs_align_0( &h->out.bs ); /* not needed */
655 nal->i_payload = &h->out.p_bitstream[bs_pos( &h->out.bs)/8] - nal->p_payload;
660 /****************************************************************************
661 * x264_encoder_headers:
662 ****************************************************************************/
663 int x264_encoder_headers( x264_t *h, x264_nal_t **pp_nal, int *pi_nal )
665 /* init bitstream context */
667 bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
669 /* Put SPS and PPS */
670 if( h->i_frame == 0 )
672 /* identify ourself */
673 x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
674 x264_sei_version_write( h, &h->out.bs );
677 /* generate sequence parameters */
678 x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
679 x264_sps_write( &h->out.bs, h->sps );
682 /* generate picture parameters */
683 x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
684 x264_pps_write( &h->out.bs, h->pps );
688 *pi_nal = h->out.i_nal;
689 *pp_nal = &h->out.nal[0];
695 static void x264_frame_put( x264_frame_t *list[X264_BFRAME_MAX], x264_frame_t *frame )
698 while( list[i] ) i++;
702 static void x264_frame_push( x264_frame_t *list[X264_BFRAME_MAX], x264_frame_t *frame )
705 while( list[i] ) i++;
711 static x264_frame_t *x264_frame_get( x264_frame_t *list[X264_BFRAME_MAX+1] )
713 x264_frame_t *frame = list[0];
715 for( i = 0; list[i]; i++ )
720 static void x264_frame_sort( x264_frame_t *list[X264_BFRAME_MAX+1], int b_dts )
725 for( i = 0; list[i+1]; i++ )
727 int dtype = list[i]->i_type - list[i+1]->i_type;
728 int dtime = list[i]->i_frame - list[i+1]->i_frame;
729 int swap = b_dts ? dtype > 0 || ( dtype == 0 && dtime > 0 )
733 XCHG( x264_frame_t*, list[i], list[i+1] );
739 #define x264_frame_sort_dts(list) x264_frame_sort(list, 1)
740 #define x264_frame_sort_pts(list) x264_frame_sort(list, 0)
742 static inline void x264_reference_build_list( x264_t *h, int i_poc, int i_slice_type )
747 /* build ref list 0/1 */
750 for( i = 1; i < h->frames.i_max_dpb; i++ )
752 if( h->frames.reference[i]->i_poc >= 0 )
754 if( h->frames.reference[i]->i_poc < i_poc )
756 h->fref0[h->i_ref0++] = h->frames.reference[i];
758 else if( h->frames.reference[i]->i_poc > i_poc )
760 h->fref1[h->i_ref1++] = h->frames.reference[i];
765 /* Order ref0 from higher to lower poc */
769 for( i = 0; i < h->i_ref0 - 1; i++ )
771 if( h->fref0[i]->i_poc < h->fref0[i+1]->i_poc )
773 XCHG( x264_frame_t*, h->fref0[i], h->fref0[i+1] );
779 /* Order ref1 from lower to higher poc (bubble sort) for B-frame */
783 for( i = 0; i < h->i_ref1 - 1; i++ )
785 if( h->fref1[i]->i_poc > h->fref1[i+1]->i_poc )
787 XCHG( x264_frame_t*, h->fref1[i], h->fref1[i+1] );
794 /* In the standard, a P-frame's ref list is sorted by frame_num.
795 * We use POC, but check whether explicit reordering is needed */
796 h->b_ref_reorder[0] =
797 h->b_ref_reorder[1] = 0;
798 if( i_slice_type == SLICE_TYPE_P )
800 for( i = 0; i < h->i_ref0 - 1; i++ )
801 if( h->fref0[i]->i_frame_num < h->fref0[i+1]->i_frame_num )
803 h->b_ref_reorder[0] = 1;
808 h->i_ref1 = X264_MIN( h->i_ref1, h->frames.i_max_ref1 );
809 h->i_ref0 = X264_MIN( h->i_ref0, h->frames.i_max_ref0 );
810 h->i_ref0 = X264_MIN( h->i_ref0, 16 - h->i_ref1 );
813 static inline void x264_reference_update( x264_t *h )
817 /* apply deblocking filter to the current decoded picture */
818 if( !h->sh.i_disable_deblocking_filter_idc )
820 TIMER_START( i_mtime_filter );
821 x264_frame_deblocking_filter( h, h->sh.i_type );
822 TIMER_STOP( i_mtime_filter );
825 x264_frame_expand_border( h->fdec );
827 /* create filtered images */
828 x264_frame_filter( h->param.cpu, h->fdec );
830 /* expand border of filtered images */
831 x264_frame_expand_border_filtered( h->fdec );
833 /* move lowres copy of the image to the ref frame */
834 for( i = 0; i < 4; i++)
835 XCHG( uint8_t*, h->fdec->lowres[i], h->fenc->lowres[i] );
837 /* adaptive B decision needs a pointer, since it can't use the ref lists */
838 if( h->sh.i_type != SLICE_TYPE_B )
839 h->frames.last_nonb = h->fdec;
841 /* move frame in the buffer */
842 h->fdec = h->frames.reference[h->frames.i_max_dpb-1];
843 for( i = h->frames.i_max_dpb-1; i > 0; i-- )
845 h->frames.reference[i] = h->frames.reference[i-1];
847 h->frames.reference[0] = h->fdec;
850 static inline void x264_reference_reset( x264_t *h )
854 /* reset ref pictures */
855 for( i = 1; i < h->frames.i_max_dpb; i++ )
857 h->frames.reference[i]->i_poc = -1;
859 h->frames.reference[0]->i_poc = 0;
862 static inline void x264_slice_init( x264_t *h, int i_nal_type, int i_slice_type, int i_global_qp )
864 /* ------------------------ Create slice header ----------------------- */
865 if( i_nal_type == NAL_SLICE_IDR )
867 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 );
870 h->i_idr_pic_id = ( h->i_idr_pic_id + 1 ) % 65536;
874 x264_slice_header_init( h, &h->sh, h->sps, h->pps, i_slice_type, -1, h->i_frame_num, i_global_qp );
876 /* always set the real higher num of ref frame used */
877 h->sh.b_num_ref_idx_override = 1;
878 h->sh.i_num_ref_idx_l0_active = h->i_ref0 <= 0 ? 1 : h->i_ref0;
879 h->sh.i_num_ref_idx_l1_active = h->i_ref1 <= 0 ? 1 : h->i_ref1;
882 h->fdec->i_frame_num = h->sh.i_frame_num;
884 if( h->sps->i_poc_type == 0 )
886 h->sh.i_poc_lsb = h->fdec->i_poc & ( (1 << h->sps->i_log2_max_poc_lsb) - 1 );
887 h->sh.i_delta_poc_bottom = 0; /* XXX won't work for field */
889 else if( h->sps->i_poc_type == 1 )
891 /* FIXME TODO FIXME */
895 /* Nothing to do ? */
898 x264_macroblock_slice_init( h );
901 static int x264_slice_write( x264_t *h )
908 memset( &h->stat.frame, 0, sizeof(h->stat.frame) );
911 x264_nal_start( h, h->i_nal_type, h->i_nal_ref_idc );
914 x264_slice_header_write( &h->out.bs, &h->sh, h->i_nal_ref_idc );
915 if( h->param.b_cabac )
917 /* alignment needed */
918 bs_align_1( &h->out.bs );
921 x264_cabac_context_init( &h->cabac, h->sh.i_type, h->sh.i_qp, h->sh.i_cabac_init_idc );
922 x264_cabac_encode_init ( &h->cabac, &h->out.bs );
924 h->mb.i_last_qp = h->sh.i_qp;
925 h->mb.i_last_dqp = 0;
927 for( mb_xy = h->sh.i_first_mb, i_skip = 0; mb_xy < h->sh.i_last_mb; mb_xy++ )
929 const int i_mb_y = mb_xy / h->sps->i_mb_width;
930 const int i_mb_x = mb_xy % h->sps->i_mb_width;
932 int mb_spos = bs_pos(&h->out.bs);
935 x264_macroblock_cache_load( h, i_mb_x, i_mb_y );
937 /* analyse parameters
938 * Slice I: choose I_4x4 or I_16x16 mode
939 * Slice P: choose between using P mode or intra (4x4 or 16x16)
941 TIMER_START( i_mtime_analyse );
942 x264_macroblock_analyse( h );
943 TIMER_STOP( i_mtime_analyse );
945 /* encode this macrobock -> be carefull it can change the mb type to P_SKIP if needed */
946 TIMER_START( i_mtime_encode );
947 x264_macroblock_encode( h );
948 TIMER_STOP( i_mtime_encode );
950 TIMER_START( i_mtime_write );
951 if( h->param.b_cabac )
953 if( mb_xy > h->sh.i_first_mb )
954 x264_cabac_encode_terminal( &h->cabac, 0 );
956 if( IS_SKIP( h->mb.i_type ) )
957 x264_cabac_mb_skip( h, 1 );
960 if( h->sh.i_type != SLICE_TYPE_I )
961 x264_cabac_mb_skip( h, 0 );
962 x264_macroblock_write_cabac( h, &h->cabac );
967 if( IS_SKIP( h->mb.i_type ) )
971 if( h->sh.i_type != SLICE_TYPE_I )
973 bs_write_ue( &h->out.bs, i_skip ); /* skip run */
976 x264_macroblock_write_cavlc( h, &h->out.bs );
979 TIMER_STOP( i_mtime_write );
982 if( h->param.b_visualize )
983 x264_visualize_mb( h );
987 x264_macroblock_cache_save( h );
989 /* accumulate mb stats */
990 h->stat.frame.i_mb_count[h->mb.i_type]++;
991 if( !IS_SKIP(h->mb.i_type) && !IS_INTRA(h->mb.i_type) && !IS_DIRECT(h->mb.i_type) )
993 if( h->mb.i_partition != D_8x8 )
994 h->stat.frame.i_mb_count_size[ x264_mb_partition_pixel_table[ h->mb.i_partition ] ] += 4;
996 for( i = 0; i < 4; i++ )
997 h->stat.frame.i_mb_count_size[ x264_mb_partition_pixel_table[ h->mb.i_sub_partition[i] ] ] ++;
998 if( h->param.i_frame_reference > 1 )
1000 for( i = 0; i < 4; i++ )
1002 int i_ref = h->mb.cache.ref[0][ x264_scan8[4*i] ];
1004 h->stat.frame.i_mb_count_ref[i_ref] ++;
1008 if( h->mb.i_cbp_luma && !IS_INTRA(h->mb.i_type) )
1010 h->stat.frame.i_mb_count_8x8dct[0] ++;
1011 h->stat.frame.i_mb_count_8x8dct[1] += h->mb.b_transform_8x8;
1014 if( h->mb.b_variable_qp )
1015 x264_ratecontrol_mb(h, bs_pos(&h->out.bs) - mb_spos);
1018 if( h->param.b_cabac )
1021 x264_cabac_encode_terminal( &h->cabac, 1 );
1023 else if( i_skip > 0 )
1025 bs_write_ue( &h->out.bs, i_skip ); /* last skip run */
1028 if( h->param.b_cabac )
1030 x264_cabac_encode_flush( &h->cabac );
1035 /* rbsp_slice_trailing_bits */
1036 bs_rbsp_trailing( &h->out.bs );
1041 /* Compute misc bits */
1042 h->stat.frame.i_misc_bits = bs_pos( &h->out.bs )
1044 - h->stat.frame.i_itex_bits
1045 - h->stat.frame.i_ptex_bits
1046 - h->stat.frame.i_hdr_bits;
1051 static inline int x264_slices_write( x264_t *h )
1056 if( h->param.b_visualize )
1057 x264_visualize_init( h );
1060 if( h->param.i_threads == 1 )
1062 x264_ratecontrol_threads_start( h );
1063 x264_slice_write( h );
1064 i_frame_size = h->out.nal[h->out.i_nal-1].i_payload;
1068 int i_nal = h->out.i_nal;
1069 int i_bs_size = h->out.i_bitstream / h->param.i_threads;
1071 /* duplicate contexts */
1072 for( i = 0; i < h->param.i_threads; i++ )
1074 x264_t *t = h->thread[i];
1077 memcpy( t, h, sizeof(x264_t) );
1078 t->out.p_bitstream += i*i_bs_size;
1079 bs_init( &t->out.bs, t->out.p_bitstream, i_bs_size );
1080 t->i_thread_num = i;
1082 t->sh.i_first_mb = (i * h->sps->i_mb_height / h->param.i_threads) * h->sps->i_mb_width;
1083 t->sh.i_last_mb = ((i+1) * h->sps->i_mb_height / h->param.i_threads) * h->sps->i_mb_width;
1084 t->out.i_nal = i_nal + i;
1086 x264_ratecontrol_threads_start( h );
1091 pthread_t handles[X264_SLICE_MAX];
1092 for( i = 0; i < h->param.i_threads; i++ )
1093 pthread_create( &handles[i], NULL, (void*)x264_slice_write, (void*)h->thread[i] );
1094 for( i = 0; i < h->param.i_threads; i++ )
1095 pthread_join( handles[i], NULL );
1098 for( i = 0; i < h->param.i_threads; i++ )
1099 x264_slice_write( h->thread[i] );
1102 /* merge contexts */
1103 i_frame_size = h->out.nal[i_nal].i_payload;
1104 for( i = 1; i < h->param.i_threads; i++ )
1107 x264_t *t = h->thread[i];
1108 h->out.nal[i_nal+i] = t->out.nal[i_nal+i];
1109 i_frame_size += t->out.nal[i_nal+i].i_payload;
1110 // all entries in stat.frame are ints
1111 for( j = 0; j < sizeof(h->stat.frame) / sizeof(int); j++ )
1112 ((int*)&h->stat.frame)[j] += ((int*)&t->stat.frame)[j];
1114 h->out.i_nal = i_nal + h->param.i_threads;
1118 if( h->param.b_visualize )
1120 x264_visualize_show( h );
1121 x264_visualize_close( h );
1125 return i_frame_size;
1128 /****************************************************************************
1129 * x264_encoder_encode:
1130 * XXX: i_poc : is the poc of the current given picture
1131 * i_frame : is the number of the frame being coded
1132 * ex: type frame poc
1140 ****************************************************************************/
1141 int x264_encoder_encode( x264_t *h,
1142 x264_nal_t **pp_nal, int *pi_nal,
1143 x264_picture_t *pic_in,
1144 x264_picture_t *pic_out )
1146 x264_frame_t *frame_psnr = h->fdec; /* just to keep the current decoded frame for psnr calculation */
1156 char psz_message[80];
1163 /* ------------------- Setup new frame from picture -------------------- */
1164 TIMER_START( i_mtime_encode_frame );
1165 if( pic_in != NULL )
1167 /* 1: Copy the picture to a frame and move it to a buffer */
1168 x264_frame_t *fenc = x264_frame_get( h->frames.unused );
1170 x264_frame_copy_picture( h, fenc, pic_in );
1172 if( h->param.i_width % 16 || h->param.i_height % 16 )
1173 x264_frame_expand_border_mod16( h, fenc );
1175 fenc->i_frame = h->frames.i_input++;
1177 x264_frame_put( h->frames.next, fenc );
1179 if( h->frames.b_have_lowres )
1180 x264_frame_init_lowres( h->param.cpu, fenc );
1182 if( h->frames.i_input <= h->frames.i_delay )
1184 /* Nothing yet to encode */
1185 /* waiting for filling bframe buffer */
1186 pic_out->i_type = X264_TYPE_AUTO;
1191 if( h->frames.current[0] == NULL )
1194 /* 2: Select frame types */
1195 if( h->frames.next[0] == NULL )
1198 x264_slicetype_decide( h );
1200 /* 3: move some B-frames and 1 non-B to encode queue */
1201 while( IS_X264_TYPE_B( h->frames.next[bframes]->i_type ) )
1203 x264_frame_put( h->frames.current, x264_frame_get( &h->frames.next[bframes] ) );
1204 /* FIXME: when max B-frames > 3, BREF may no longer be centered after GOP closing */
1205 if( h->param.b_bframe_pyramid && bframes > 1 )
1207 x264_frame_t *mid = x264_frame_get( &h->frames.next[bframes/2] );
1208 mid->i_type = X264_TYPE_BREF;
1209 x264_frame_put( h->frames.current, mid );
1213 x264_frame_put( h->frames.current, x264_frame_get( h->frames.next ) );
1215 TIMER_STOP( i_mtime_encode_frame );
1217 /* ------------------- Get frame to be encoded ------------------------- */
1218 /* 4: get picture to encode */
1219 h->fenc = x264_frame_get( h->frames.current );
1220 if( h->fenc == NULL )
1222 /* Nothing yet to encode (ex: waiting for I/P with B frames) */
1223 /* waiting for filling bframe buffer */
1224 pic_out->i_type = X264_TYPE_AUTO;
1230 if( h->fenc->i_type == X264_TYPE_IDR )
1232 h->frames.i_last_idr = h->fenc->i_frame;
1235 /* ------------------- Setup frame context ----------------------------- */
1236 /* 5: Init data dependant of frame type */
1237 TIMER_START( i_mtime_encode_frame );
1238 if( h->fenc->i_type == X264_TYPE_IDR )
1240 /* reset ref pictures */
1241 x264_reference_reset( h );
1243 i_nal_type = NAL_SLICE_IDR;
1244 i_nal_ref_idc = NAL_PRIORITY_HIGHEST;
1245 i_slice_type = SLICE_TYPE_I;
1247 else if( h->fenc->i_type == X264_TYPE_I )
1249 i_nal_type = NAL_SLICE;
1250 i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
1251 i_slice_type = SLICE_TYPE_I;
1253 else if( h->fenc->i_type == X264_TYPE_P )
1255 i_nal_type = NAL_SLICE;
1256 i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
1257 i_slice_type = SLICE_TYPE_P;
1259 else if( h->fenc->i_type == X264_TYPE_BREF )
1261 i_nal_type = NAL_SLICE;
1262 i_nal_ref_idc = NAL_PRIORITY_HIGH; /* maybe add MMCO to forget it? -> low */
1263 i_slice_type = SLICE_TYPE_B;
1267 i_nal_type = NAL_SLICE;
1268 i_nal_ref_idc = NAL_PRIORITY_DISPOSABLE;
1269 i_slice_type = SLICE_TYPE_B;
1273 h->fenc->i_poc = 2 * (h->fenc->i_frame - h->frames.i_last_idr);
1274 h->fdec->i_type = h->fenc->i_type;
1275 h->fdec->i_frame = h->fenc->i_frame;
1276 h->fenc->b_kept_as_ref =
1277 h->fdec->b_kept_as_ref = i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE;
1281 /* ------------------- Init ----------------------------- */
1282 /* build ref list 0/1 */
1283 x264_reference_build_list( h, h->fdec->i_poc, i_slice_type );
1285 /* Init the rate control */
1286 x264_ratecontrol_start( h, i_slice_type, h->fenc->i_qpplus1 );
1287 i_global_qp = x264_ratecontrol_qp( h );
1289 pic_out->i_qpplus1 =
1290 h->fdec->i_qpplus1 = i_global_qp + 1;
1292 if( i_slice_type == SLICE_TYPE_B )
1293 x264_macroblock_bipred_init( h );
1295 /* ------------------------ Create slice header ----------------------- */
1296 x264_slice_init( h, i_nal_type, i_slice_type, i_global_qp );
1298 if( h->fenc->b_kept_as_ref )
1301 /* ---------------------- Write the bitstream -------------------------- */
1302 /* Init bitstream context */
1304 bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
1309 if(i_slice_type == SLICE_TYPE_I)
1311 else if(i_slice_type == SLICE_TYPE_P)
1313 else if(i_slice_type == SLICE_TYPE_B)
1318 x264_nal_start(h, NAL_AUD, NAL_PRIORITY_DISPOSABLE);
1319 bs_write(&h->out.bs, 3, pic_type);
1320 bs_rbsp_trailing(&h->out.bs);
1324 h->i_nal_type = i_nal_type;
1325 h->i_nal_ref_idc = i_nal_ref_idc;
1327 /* Write SPS and PPS */
1328 if( i_nal_type == NAL_SLICE_IDR && h->param.b_repeat_headers )
1330 if( h->fenc->i_frame == 0 )
1332 /* identify ourself */
1333 x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
1334 x264_sei_version_write( h, &h->out.bs );
1338 /* generate sequence parameters */
1339 x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
1340 x264_sps_write( &h->out.bs, h->sps );
1343 /* generate picture parameters */
1344 x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
1345 x264_pps_write( &h->out.bs, h->pps );
1350 i_frame_size = x264_slices_write( h );
1352 /* restore CPU state (before using float again) */
1353 x264_cpu_restore( h->param.cpu );
1355 if( i_slice_type == SLICE_TYPE_P && !h->param.rc.b_stat_read
1356 && h->param.i_scenecut_threshold >= 0 )
1358 const int *mbs = h->stat.frame.i_mb_count;
1359 int i_mb_i = mbs[I_16x16] + mbs[I_8x8] + mbs[I_4x4];
1360 int i_mb_p = mbs[P_L0] + mbs[P_8x8];
1361 int i_mb_s = mbs[P_SKIP];
1362 int i_mb = h->sps->i_mb_width * h->sps->i_mb_height;
1363 int64_t i_inter_cost = h->stat.frame.i_inter_cost;
1364 int64_t i_intra_cost = h->stat.frame.i_intra_cost;
1367 int i_gop_size = h->fenc->i_frame - h->frames.i_last_idr;
1368 float f_thresh_max = h->param.i_scenecut_threshold / 100.0;
1369 /* magic numbers pulled out of thin air */
1370 float f_thresh_min = f_thresh_max * h->param.i_keyint_min
1371 / ( h->param.i_keyint_max * 4 );
1372 if( h->param.i_keyint_min == h->param.i_keyint_max )
1373 f_thresh_min= f_thresh_max;
1375 /* macroblock_analyse() doesn't further analyse skipped mbs,
1376 * so we have to guess their cost */
1378 i_intra_cost = i_intra_cost * i_mb / (i_mb - i_mb_s);
1380 if( i_gop_size < h->param.i_keyint_min / 4 )
1381 f_bias = f_thresh_min / 4;
1382 else if( i_gop_size <= h->param.i_keyint_min )
1383 f_bias = f_thresh_min * i_gop_size / h->param.i_keyint_min;
1386 f_bias = f_thresh_min
1387 + ( f_thresh_max - f_thresh_min )
1388 * ( i_gop_size - h->param.i_keyint_min )
1389 / ( h->param.i_keyint_max - h->param.i_keyint_min );
1391 f_bias = X264_MIN( f_bias, 1.0 );
1393 /* Bad P will be reencoded as I */
1394 if( i_mb_s < i_mb &&
1395 i_inter_cost >= (1.0 - f_bias) * i_intra_cost )
1399 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",
1401 (double)i_intra_cost, (double)i_inter_cost,
1402 (double)i_inter_cost / i_intra_cost,
1404 i_mb_i, i_mb_p, i_mb_s );
1406 /* Restore frame num */
1409 for( b = 0; h->frames.current[b] && IS_X264_TYPE_B( h->frames.current[b]->i_type ); b++ );
1412 /* If using B-frames, force GOP to be closed.
1413 * Even if this frame is going to be I and not IDR, forcing a
1414 * P-frame before the scenecut will probably help compression.
1416 * We don't yet know exactly which frame is the scene cut, so
1417 * we can't assign an I-frame. Instead, change the previous
1418 * B-frame to P, and rearrange coding order. */
1420 if( h->param.b_bframe_adaptive || b > 1 )
1421 h->fenc->i_type = X264_TYPE_AUTO;
1422 x264_frame_sort_pts( h->frames.current );
1423 x264_frame_push( h->frames.next, h->fenc );
1424 h->fenc = h->frames.current[b-1];
1425 h->frames.current[b-1] = NULL;
1426 h->fenc->i_type = X264_TYPE_P;
1427 x264_frame_sort_dts( h->frames.current );
1429 /* Do IDR if needed */
1430 else if( i_gop_size >= h->param.i_keyint_min )
1437 /* Reinit field of fenc */
1438 h->fenc->i_type = X264_TYPE_IDR;
1441 /* Put enqueued frames back in the pool */
1442 while( (tmp = x264_frame_get( h->frames.current ) ) != NULL )
1443 x264_frame_put( h->frames.next, tmp );
1444 x264_frame_sort_pts( h->frames.next );
1448 h->fenc->i_type = X264_TYPE_I;
1454 /* End bitstream, set output */
1455 *pi_nal = h->out.i_nal;
1456 *pp_nal = h->out.nal;
1458 /* Set output picture properties */
1459 if( i_slice_type == SLICE_TYPE_I )
1460 pic_out->i_type = i_nal_type == NAL_SLICE_IDR ? X264_TYPE_IDR : X264_TYPE_I;
1461 else if( i_slice_type == SLICE_TYPE_P )
1462 pic_out->i_type = X264_TYPE_P;
1464 pic_out->i_type = X264_TYPE_B;
1465 pic_out->i_pts = h->fenc->i_pts;
1467 pic_out->img.i_plane = h->fdec->i_plane;
1468 for(i = 0; i < 4; i++){
1469 pic_out->img.i_stride[i] = h->fdec->i_stride[i];
1470 pic_out->img.plane[i] = h->fdec->plane[i];
1473 /* ---------------------- Update encoder state ------------------------- */
1476 x264_cpu_restore( h->param.cpu );
1477 x264_ratecontrol_end( h, i_frame_size * 8 );
1479 /* handle references */
1480 if( i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE )
1481 x264_reference_update( h );
1482 x264_frame_put( h->frames.unused, h->fenc );
1484 /* increase frame count */
1487 /* restore CPU state (before using float again) */
1488 x264_cpu_restore( h->param.cpu );
1490 x264_noise_reduction_update( h );
1492 TIMER_STOP( i_mtime_encode_frame );
1494 /* ---------------------- Compute/Print statistics --------------------- */
1496 h->stat.i_slice_count[i_slice_type]++;
1497 h->stat.i_slice_size[i_slice_type] += i_frame_size + NALU_OVERHEAD;
1498 h->stat.i_slice_qp[i_slice_type] += i_global_qp;
1500 for( i = 0; i < 19; i++ )
1501 h->stat.i_mb_count[h->sh.i_type][i] += h->stat.frame.i_mb_count[i];
1502 for( i = 0; i < 2; i++ )
1503 h->stat.i_mb_count_8x8dct[i] += h->stat.frame.i_mb_count_8x8dct[i];
1504 if( h->sh.i_type != SLICE_TYPE_I )
1506 for( i = 0; i < 7; i++ )
1507 h->stat.i_mb_count_size[h->sh.i_type][i] += h->stat.frame.i_mb_count_size[i];
1508 for( i = 0; i < 16; i++ )
1509 h->stat.i_mb_count_ref[h->sh.i_type][i] += h->stat.frame.i_mb_count_ref[i];
1511 if( i_slice_type == SLICE_TYPE_B )
1513 h->stat.i_direct_frames[ h->sh.b_direct_spatial_mv_pred ] ++;
1514 if( h->mb.b_direct_auto_write )
1516 //FIXME somewhat arbitrary time constants
1517 if( h->stat.i_direct_score[0] + h->stat.i_direct_score[1] > h->mb.i_mb_count )
1519 for( i = 0; i < 2; i++ )
1520 h->stat.i_direct_score[i] = h->stat.i_direct_score[i] * 9/10;
1522 for( i = 0; i < 2; i++ )
1523 h->stat.i_direct_score[i] += h->stat.frame.i_direct_score[i];
1527 if( h->param.analyse.b_psnr )
1529 int64_t i_sqe_y, i_sqe_u, i_sqe_v;
1532 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 );
1533 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);
1534 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);
1535 x264_cpu_restore( h->param.cpu );
1537 h->stat.i_sqe_global[i_slice_type] += i_sqe_y + i_sqe_u + i_sqe_v;
1538 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 );
1539 h->stat.f_psnr_mean_y[i_slice_type] += x264_psnr( i_sqe_y, h->param.i_width * h->param.i_height );
1540 h->stat.f_psnr_mean_u[i_slice_type] += x264_psnr( i_sqe_u, h->param.i_width * h->param.i_height / 4 );
1541 h->stat.f_psnr_mean_v[i_slice_type] += x264_psnr( i_sqe_v, h->param.i_width * h->param.i_height / 4 );
1543 snprintf( psz_message, 80, " PSNR Y:%2.2f U:%2.2f V:%2.2f",
1544 x264_psnr( i_sqe_y, h->param.i_width * h->param.i_height ),
1545 x264_psnr( i_sqe_u, h->param.i_width * h->param.i_height / 4),
1546 x264_psnr( i_sqe_v, h->param.i_width * h->param.i_height / 4) );
1547 psz_message[79] = '\0';
1551 psz_message[0] = '\0';
1554 x264_log( h, X264_LOG_DEBUG,
1555 "frame=%4d QP=%i NAL=%d Slice:%c Poc:%-3d I:%-4d P:%-4d SKIP:%-4d size=%d bytes%s\n",
1559 i_slice_type == SLICE_TYPE_I ? 'I' : (i_slice_type == SLICE_TYPE_P ? 'P' : 'B' ),
1561 h->stat.frame.i_mb_count_i,
1562 h->stat.frame.i_mb_count_p,
1563 h->stat.frame.i_mb_count_skip,
1568 #ifdef DEBUG_MB_TYPE
1570 static const char mb_chars[] = { 'i', 'i', 'I', 'C', 'P', '8', 'S',
1571 'D', '<', 'X', 'B', 'X', '>', 'B', 'B', 'B', 'B', '8', 'S' };
1573 for( mb_xy = 0; mb_xy < h->sps->i_mb_width * h->sps->i_mb_height; mb_xy++ )
1575 if( h->mb.type[mb_xy] < 19 && h->mb.type[mb_xy] >= 0 )
1576 fprintf( stderr, "%c ", mb_chars[ h->mb.type[mb_xy] ] );
1578 fprintf( stderr, "? " );
1580 if( (mb_xy+1) % h->sps->i_mb_width == 0 )
1581 fprintf( stderr, "\n" );
1586 #ifdef DEBUG_DUMP_FRAME
1587 /* Dump reconstructed frame */
1588 x264_frame_dump( h, frame_psnr, "fdec.yuv" );
1593 /****************************************************************************
1594 * x264_encoder_close:
1595 ****************************************************************************/
1596 void x264_encoder_close ( x264_t *h )
1598 #ifdef DEBUG_BENCHMARK
1599 int64_t i_mtime_total = i_mtime_analyse + i_mtime_encode + i_mtime_write + i_mtime_filter + 1;
1601 int64_t i_yuv_size = 3 * h->param.i_width * h->param.i_height / 2;
1604 #ifdef DEBUG_BENCHMARK
1605 x264_log( h, X264_LOG_INFO,
1606 "analyse=%d(%lldms) encode=%d(%lldms) write=%d(%lldms) filter=%d(%lldms)\n",
1607 (int)(100*i_mtime_analyse/i_mtime_total), i_mtime_analyse/1000,
1608 (int)(100*i_mtime_encode/i_mtime_total), i_mtime_encode/1000,
1609 (int)(100*i_mtime_write/i_mtime_total), i_mtime_write/1000,
1610 (int)(100*i_mtime_filter/i_mtime_total), i_mtime_filter/1000 );
1613 /* Slices used and PSNR */
1614 for( i=0; i<5; i++ )
1616 static const int slice_order[] = { SLICE_TYPE_I, SLICE_TYPE_SI, SLICE_TYPE_P, SLICE_TYPE_SP, SLICE_TYPE_B };
1617 static const char *slice_name[] = { "P", "B", "I", "SP", "SI" };
1618 int i_slice = slice_order[i];
1620 if( h->stat.i_slice_count[i_slice] > 0 )
1622 const int i_count = h->stat.i_slice_count[i_slice];
1623 if( h->param.analyse.b_psnr )
1625 x264_log( h, X264_LOG_INFO,
1626 "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",
1627 slice_name[i_slice],
1629 (double)h->stat.i_slice_qp[i_slice] / i_count,
1630 (double)h->stat.i_slice_size[i_slice] / i_count,
1631 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,
1632 h->stat.f_psnr_average[i_slice] / i_count,
1633 x264_psnr( h->stat.i_sqe_global[i_slice], i_count * i_yuv_size ) );
1637 x264_log( h, X264_LOG_INFO,
1638 "slice %s:%-5d Avg QP:%5.2f size:%6.0f\n",
1639 slice_name[i_slice],
1641 (double)h->stat.i_slice_qp[i_slice] / i_count,
1642 (double)h->stat.i_slice_size[i_slice] / i_count );
1648 if( h->stat.i_slice_count[SLICE_TYPE_I] > 0 )
1650 const int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_I];
1651 const double i_count = h->stat.i_slice_count[SLICE_TYPE_I] * h->mb.i_mb_count / 100.0;
1652 x264_log( h, X264_LOG_INFO,
1653 "mb I I16..4: %4.1f%% %4.1f%% %4.1f%%\n",
1654 i_mb_count[I_16x16]/ i_count,
1655 i_mb_count[I_8x8] / i_count,
1656 i_mb_count[I_4x4] / i_count );
1658 if( h->stat.i_slice_count[SLICE_TYPE_P] > 0 )
1660 const int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_P];
1661 const int64_t *i_mb_size = h->stat.i_mb_count_size[SLICE_TYPE_P];
1662 const double i_count = h->stat.i_slice_count[SLICE_TYPE_P] * h->mb.i_mb_count / 100.0;
1663 x264_log( h, X264_LOG_INFO,
1664 "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",
1665 i_mb_count[I_16x16]/ i_count,
1666 i_mb_count[I_8x8] / i_count,
1667 i_mb_count[I_4x4] / i_count,
1668 i_mb_size[PIXEL_16x16] / (i_count*4),
1669 (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
1670 i_mb_size[PIXEL_8x8] / (i_count*4),
1671 (i_mb_size[PIXEL_8x4] + i_mb_size[PIXEL_4x8]) / (i_count*4),
1672 i_mb_size[PIXEL_4x4] / (i_count*4),
1673 i_mb_count[P_SKIP] / i_count );
1675 if( h->stat.i_slice_count[SLICE_TYPE_B] > 0 )
1677 const int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_B];
1678 const int64_t *i_mb_size = h->stat.i_mb_count_size[SLICE_TYPE_B];
1679 const double i_count = h->stat.i_slice_count[SLICE_TYPE_B] * h->mb.i_mb_count / 100.0;
1680 x264_log( h, X264_LOG_INFO,
1681 "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",
1682 i_mb_count[I_16x16] / i_count,
1683 i_mb_count[I_8x8] / i_count,
1684 i_mb_count[I_4x4] / i_count,
1685 i_mb_size[PIXEL_16x16] / (i_count*4),
1686 (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
1687 i_mb_size[PIXEL_8x8] / (i_count*4),
1688 i_mb_count[B_DIRECT] / i_count,
1689 i_mb_count[B_SKIP] / i_count );
1692 x264_ratecontrol_summary( h );
1694 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 )
1696 const int i_count = h->stat.i_slice_count[SLICE_TYPE_I] +
1697 h->stat.i_slice_count[SLICE_TYPE_P] +
1698 h->stat.i_slice_count[SLICE_TYPE_B];
1699 float fps = (float) h->param.i_fps_num / h->param.i_fps_den;
1700 #define SUM3(p) (p[SLICE_TYPE_I] + p[SLICE_TYPE_P] + p[SLICE_TYPE_B])
1701 #define SUM3b(p,o) (p[SLICE_TYPE_I][o] + p[SLICE_TYPE_P][o] + p[SLICE_TYPE_B][o])
1702 float f_bitrate = fps * SUM3(h->stat.i_slice_size) / i_count / 125;
1704 if( h->param.analyse.b_transform_8x8 )
1706 int64_t i_i8x8 = SUM3b( h->stat.i_mb_count, I_8x8 );
1707 int64_t i_intra = i_i8x8 + SUM3b( h->stat.i_mb_count, I_4x4 )
1708 + SUM3b( h->stat.i_mb_count, I_16x16 );
1709 x264_log( h, X264_LOG_INFO, "8x8 transform intra:%.1f%% inter:%.1f%%\n",
1710 100. * i_i8x8 / i_intra,
1711 100. * h->stat.i_mb_count_8x8dct[1] / h->stat.i_mb_count_8x8dct[0] );
1714 if( h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO
1715 && h->stat.i_slice_count[SLICE_TYPE_B] )
1717 x264_log( h, X264_LOG_INFO, "direct mvs spatial:%.1f%% temporal:%.1f%%\n",
1718 h->stat.i_direct_frames[1] * 100. / h->stat.i_slice_count[SLICE_TYPE_B],
1719 h->stat.i_direct_frames[0] * 100. / h->stat.i_slice_count[SLICE_TYPE_B] );
1722 if( h->param.i_frame_reference > 1 )
1725 for( i_slice = 0; i_slice < 2; i_slice++ )
1731 for( i = 0; i < h->param.i_frame_reference; i++ )
1732 if( h->stat.i_mb_count_ref[i_slice][i] )
1734 i_den += h->stat.i_mb_count_ref[i_slice][i];
1739 for( i = 0; i <= i_max; i++ )
1740 p += sprintf( p, " %4.1f%%", 100. * h->stat.i_mb_count_ref[i_slice][i] / i_den );
1741 x264_log( h, X264_LOG_INFO, "ref %c %s\n", i_slice==SLICE_TYPE_P ? 'P' : 'B', buf );
1745 if( h->param.analyse.b_psnr )
1746 x264_log( h, X264_LOG_INFO,
1747 "PSNR Mean Y:%6.3f U:%6.3f V:%6.3f Avg:%6.3f Global:%6.3f kb/s:%.2f\n",
1748 SUM3( h->stat.f_psnr_mean_y ) / i_count,
1749 SUM3( h->stat.f_psnr_mean_u ) / i_count,
1750 SUM3( h->stat.f_psnr_mean_v ) / i_count,
1751 SUM3( h->stat.f_psnr_average ) / i_count,
1752 x264_psnr( SUM3( h->stat.i_sqe_global ), i_count * i_yuv_size ),
1755 x264_log( h, X264_LOG_INFO, "kb/s:%.1f\n", f_bitrate );
1759 for( i = 0; i < X264_BFRAME_MAX + 3; i++ )
1761 if( h->frames.current[i] ) x264_frame_delete( h->frames.current[i] );
1762 if( h->frames.next[i] ) x264_frame_delete( h->frames.next[i] );
1763 if( h->frames.unused[i] ) x264_frame_delete( h->frames.unused[i] );
1766 for( i = 0; i < h->frames.i_max_dpb; i++ )
1768 x264_frame_delete( h->frames.reference[i] );
1772 x264_ratecontrol_delete( h );
1775 if( h->param.rc.psz_stat_out )
1776 free( h->param.rc.psz_stat_out );
1777 if( h->param.rc.psz_stat_in )
1778 free( h->param.rc.psz_stat_in );
1779 if( h->param.rc.psz_rc_eq )
1780 free( h->param.rc.psz_rc_eq );
1782 x264_macroblock_cache_end( h );
1783 x264_free( h->out.p_bitstream );
1784 for( i = 1; i < h->param.i_threads; i++ )
1785 x264_free( h->thread[i] );