1 /*****************************************************************************
3 *****************************************************************************
4 * Copyright (C) 2003 Laurent Aimar
5 * $Id: encoder.c,v 1.1 2004/06/03 19:27:08 fenrir Exp $
7 * Authors: Laurent Aimar <fenrir@via.ecp.fr>
9 * This program is free software; you can redistribute it and/or modify
10 * it under the terms of the GNU General Public License as published by
11 * the Free Software Foundation; either version 2 of the License, or
12 * (at your option) any later version.
14 * This program is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
17 * GNU General Public License for more details.
19 * You should have received a copy of the GNU General Public License
20 * along with this program; if not, write to the Free Software
21 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111, USA.
22 *****************************************************************************/
28 #include "common/common.h"
29 #include "common/cpu.h"
33 #include "ratecontrol.h"
34 #include "macroblock.h"
37 #include "common/visualize.h"
40 //#define DEBUG_MB_TYPE
41 //#define DEBUG_DUMP_FRAME
42 //#define DEBUG_BENCHMARK
44 #ifdef DEBUG_BENCHMARK
45 static int64_t i_mtime_encode_frame = 0;
46 static int64_t i_mtime_analyse = 0;
47 static int64_t i_mtime_encode = 0;
48 static int64_t i_mtime_write = 0;
49 static int64_t i_mtime_filter = 0;
50 #define TIMER_START( d ) \
52 int64_t d##start = x264_mdate();
54 #define TIMER_STOP( d ) \
55 d += x264_mdate() - d##start;\
58 #define TIMER_START( d )
59 #define TIMER_STOP( d )
62 #define NALU_OVERHEAD 5 // startcode + NAL type costs 5 bytes per frame
64 /****************************************************************************
66 ******************************* x264 libs **********************************
68 ****************************************************************************/
69 static float x264_psnr( int64_t i_sqe, int64_t i_size )
71 double f_mse = (double)i_sqe / ((double)65025.0 * (double)i_size);
72 if( f_mse <= 0.0000000001 ) /* Max 100dB */
75 return (float)(-10.0 * log( f_mse ) / log( 10.0 ));
78 #ifdef DEBUG_DUMP_FRAME
79 static void x264_frame_dump( x264_t *h, x264_frame_t *fr, char *name )
81 FILE *f = fopen( name, "r+b" );
86 /* Write the frame in display order */
87 fseek( f, fr->i_frame * h->param.i_height * h->param.i_width * 3 / 2, SEEK_SET );
89 for( i = 0; i < fr->i_plane; i++ )
91 for( y = 0; y < h->param.i_height / ( i == 0 ? 1 : 2 ); y++ )
93 fwrite( &fr->plane[i][y*fr->i_stride[i]], 1, h->param.i_width / ( i == 0 ? 1 : 2 ), f );
101 /* Fill "default" values */
102 static void x264_slice_header_init( x264_t *h, x264_slice_header_t *sh,
103 x264_sps_t *sps, x264_pps_t *pps,
104 int i_type, int i_idr_pic_id, int i_frame, int i_qp )
106 x264_param_t *param = &h->param;
109 /* First we fill all field */
115 sh->i_last_mb = h->sps->i_mb_width * h->sps->i_mb_height;
116 sh->i_pps_id = pps->i_id;
118 sh->i_frame_num = i_frame;
120 sh->b_mbaff = h->param.b_interlaced;
121 sh->b_field_pic = 0; /* no field support for now */
122 sh->b_bottom_field = 0; /* not yet used */
124 sh->i_idr_pic_id = i_idr_pic_id;
126 /* poc stuff, fixed later */
128 sh->i_delta_poc_bottom = 0;
129 sh->i_delta_poc[0] = 0;
130 sh->i_delta_poc[1] = 0;
132 sh->i_redundant_pic_cnt = 0;
134 if( !h->mb.b_direct_auto_read )
136 if( h->mb.b_direct_auto_write )
137 sh->b_direct_spatial_mv_pred = ( h->stat.i_direct_score[1] > h->stat.i_direct_score[0] );
139 sh->b_direct_spatial_mv_pred = ( param->analyse.i_direct_mv_pred == X264_DIRECT_PRED_SPATIAL );
141 /* else b_direct_spatial_mv_pred was read from the 2pass statsfile */
143 sh->b_num_ref_idx_override = 0;
144 sh->i_num_ref_idx_l0_active = 1;
145 sh->i_num_ref_idx_l1_active = 1;
147 sh->b_ref_pic_list_reordering_l0 = h->b_ref_reorder[0];
148 sh->b_ref_pic_list_reordering_l1 = h->b_ref_reorder[1];
150 /* If the ref list isn't in the default order, construct reordering header */
151 /* List1 reordering isn't needed yet */
152 if( sh->b_ref_pic_list_reordering_l0 )
154 int pred_frame_num = i_frame;
155 for( i = 0; i < h->i_ref0; i++ )
157 int diff = h->fref0[i]->i_frame_num - pred_frame_num;
159 x264_log( h, X264_LOG_ERROR, "diff frame num == 0\n" );
160 sh->ref_pic_list_order[0][i].idc = ( diff > 0 );
161 sh->ref_pic_list_order[0][i].arg = abs( diff ) - 1;
162 pred_frame_num = h->fref0[i]->i_frame_num;
166 sh->i_cabac_init_idc = param->i_cabac_init_idc;
169 sh->i_qp_delta = i_qp - pps->i_pic_init_qp;
170 sh->b_sp_for_swidth = 0;
173 /* If effective qp <= 15, deblocking would have no effect anyway */
174 if( param->b_deblocking_filter
175 && ( h->mb.b_variable_qp
176 || 15 < i_qp + 2 * X264_MAX(param->i_deblocking_filter_alphac0, param->i_deblocking_filter_beta) ) )
178 sh->i_disable_deblocking_filter_idc = 0;
182 sh->i_disable_deblocking_filter_idc = 1;
184 sh->i_alpha_c0_offset = param->i_deblocking_filter_alphac0 << 1;
185 sh->i_beta_offset = param->i_deblocking_filter_beta << 1;
188 static void x264_slice_header_write( bs_t *s, x264_slice_header_t *sh, int i_nal_ref_idc )
194 assert( sh->i_first_mb % (2*sh->sps->i_mb_width) == 0 );
195 bs_write_ue( s, sh->i_first_mb >> 1 );
198 bs_write_ue( s, sh->i_first_mb );
200 bs_write_ue( s, sh->i_type + 5 ); /* same type things */
201 bs_write_ue( s, sh->i_pps_id );
202 bs_write( s, sh->sps->i_log2_max_frame_num, sh->i_frame_num );
204 if( !sh->sps->b_frame_mbs_only )
206 bs_write1( s, sh->b_field_pic );
207 if ( sh->b_field_pic )
208 bs_write1( s, sh->b_bottom_field );
211 if( sh->i_idr_pic_id >= 0 ) /* NAL IDR */
213 bs_write_ue( s, sh->i_idr_pic_id );
216 if( sh->sps->i_poc_type == 0 )
218 bs_write( s, sh->sps->i_log2_max_poc_lsb, sh->i_poc_lsb );
219 if( sh->pps->b_pic_order && !sh->b_field_pic )
221 bs_write_se( s, sh->i_delta_poc_bottom );
224 else if( sh->sps->i_poc_type == 1 && !sh->sps->b_delta_pic_order_always_zero )
226 bs_write_se( s, sh->i_delta_poc[0] );
227 if( sh->pps->b_pic_order && !sh->b_field_pic )
229 bs_write_se( s, sh->i_delta_poc[1] );
233 if( sh->pps->b_redundant_pic_cnt )
235 bs_write_ue( s, sh->i_redundant_pic_cnt );
238 if( sh->i_type == SLICE_TYPE_B )
240 bs_write1( s, sh->b_direct_spatial_mv_pred );
242 if( sh->i_type == SLICE_TYPE_P || sh->i_type == SLICE_TYPE_SP || sh->i_type == SLICE_TYPE_B )
244 bs_write1( s, sh->b_num_ref_idx_override );
245 if( sh->b_num_ref_idx_override )
247 bs_write_ue( s, sh->i_num_ref_idx_l0_active - 1 );
248 if( sh->i_type == SLICE_TYPE_B )
250 bs_write_ue( s, sh->i_num_ref_idx_l1_active - 1 );
255 /* ref pic list reordering */
256 if( sh->i_type != SLICE_TYPE_I )
258 bs_write1( s, sh->b_ref_pic_list_reordering_l0 );
259 if( sh->b_ref_pic_list_reordering_l0 )
261 for( i = 0; i < sh->i_num_ref_idx_l0_active; i++ )
263 bs_write_ue( s, sh->ref_pic_list_order[0][i].idc );
264 bs_write_ue( s, sh->ref_pic_list_order[0][i].arg );
270 if( sh->i_type == SLICE_TYPE_B )
272 bs_write1( s, sh->b_ref_pic_list_reordering_l1 );
273 if( sh->b_ref_pic_list_reordering_l1 )
275 for( i = 0; i < sh->i_num_ref_idx_l1_active; i++ )
277 bs_write_ue( s, sh->ref_pic_list_order[1][i].idc );
278 bs_write_ue( s, sh->ref_pic_list_order[1][i].arg );
284 if( ( sh->pps->b_weighted_pred && ( sh->i_type == SLICE_TYPE_P || sh->i_type == SLICE_TYPE_SP ) ) ||
285 ( sh->pps->b_weighted_bipred == 1 && sh->i_type == SLICE_TYPE_B ) )
290 if( i_nal_ref_idc != 0 )
292 if( sh->i_idr_pic_id >= 0 )
294 bs_write1( s, 0 ); /* no output of prior pics flag */
295 bs_write1( s, 0 ); /* long term reference flag */
299 bs_write1( s, 0 ); /* adaptive_ref_pic_marking_mode_flag */
303 if( sh->pps->b_cabac && sh->i_type != SLICE_TYPE_I )
305 bs_write_ue( s, sh->i_cabac_init_idc );
307 bs_write_se( s, sh->i_qp_delta ); /* slice qp delta */
309 if( sh->pps->b_deblocking_filter_control )
311 bs_write_ue( s, sh->i_disable_deblocking_filter_idc );
312 if( sh->i_disable_deblocking_filter_idc != 1 )
314 bs_write_se( s, sh->i_alpha_c0_offset >> 1 );
315 bs_write_se( s, sh->i_beta_offset >> 1 );
320 /****************************************************************************
322 ****************************************************************************
323 ****************************** External API*********************************
324 ****************************************************************************
326 ****************************************************************************/
328 static int x264_validate_parameters( x264_t *h )
330 if( h->param.i_width <= 0 || h->param.i_height <= 0 )
332 x264_log( h, X264_LOG_ERROR, "invalid width x height (%dx%d)\n",
333 h->param.i_width, h->param.i_height );
337 if( h->param.i_width % 2 || h->param.i_height % 2 )
339 x264_log( h, X264_LOG_ERROR, "width or height not divisible by 2 (%dx%d)\n",
340 h->param.i_width, h->param.i_height );
343 if( h->param.i_csp != X264_CSP_I420 )
345 x264_log( h, X264_LOG_ERROR, "invalid CSP (only I420 supported)\n" );
349 if( h->param.i_threads == 0 )
350 h->param.i_threads = x264_cpu_num_processors();
351 h->param.i_threads = x264_clip3( h->param.i_threads, 1, X264_SLICE_MAX );
352 h->param.i_threads = X264_MIN( h->param.i_threads, (h->param.i_height + 15) >> (4 + h->param.b_interlaced) );
354 if( h->param.i_threads > 1 )
356 x264_log( h, X264_LOG_WARNING, "not compiled with pthread support!\n");
357 x264_log( h, X264_LOG_WARNING, "multislicing anyway, but you won't see any speed gain.\n" );
361 if( h->param.b_interlaced )
363 if( h->param.analyse.i_me_method == X264_ME_ESA )
365 x264_log( h, X264_LOG_WARNING, "interlace + me=esa is not implemented\n" );
366 h->param.analyse.i_me_method = X264_ME_UMH;
368 if( h->param.analyse.i_direct_mv_pred > X264_DIRECT_PRED_SPATIAL )
370 x264_log( h, X264_LOG_WARNING, "interlace + direct=temporal is not implemented\n" );
371 h->param.analyse.i_direct_mv_pred = X264_DIRECT_PRED_SPATIAL;
375 if( h->param.rc.i_rc_method < 0 || h->param.rc.i_rc_method > 2 )
377 x264_log( h, X264_LOG_ERROR, "invalid RC method\n" );
380 h->param.rc.f_rf_constant = x264_clip3f( h->param.rc.f_rf_constant, 0, 51 );
381 h->param.rc.i_qp_constant = x264_clip3( h->param.rc.i_qp_constant, 0, 51 );
382 if( h->param.rc.i_rc_method == X264_RC_CRF )
383 h->param.rc.i_qp_constant = h->param.rc.f_rf_constant;
384 if( (h->param.rc.i_rc_method == X264_RC_CQP || h->param.rc.i_rc_method == X264_RC_CRF)
385 && h->param.rc.i_qp_constant == 0 )
387 h->mb.b_lossless = 1;
388 h->param.i_cqm_preset = X264_CQM_FLAT;
389 h->param.psz_cqm_file = NULL;
390 h->param.rc.i_rc_method = X264_RC_CQP;
391 h->param.rc.f_ip_factor = 1;
392 h->param.rc.f_pb_factor = 1;
393 h->param.analyse.b_transform_8x8 = 0;
394 h->param.analyse.b_psnr = 0;
395 h->param.analyse.b_ssim = 0;
396 h->param.analyse.i_chroma_qp_offset = 0;
397 h->param.analyse.i_trellis = 0;
398 h->param.analyse.b_fast_pskip = 0;
399 h->param.analyse.i_noise_reduction = 0;
400 h->param.analyse.i_subpel_refine = x264_clip3( h->param.analyse.i_subpel_refine, 1, 6 );
403 if( ( h->param.i_width % 16 || h->param.i_height % 16 ) && !h->mb.b_lossless )
405 x264_log( h, X264_LOG_WARNING,
406 "width or height not divisible by 16 (%dx%d), compression will suffer.\n",
407 h->param.i_width, h->param.i_height );
410 h->param.i_frame_reference = x264_clip3( h->param.i_frame_reference, 1, 16 );
411 if( h->param.i_keyint_max <= 0 )
412 h->param.i_keyint_max = 1;
413 h->param.i_keyint_min = x264_clip3( h->param.i_keyint_min, 1, h->param.i_keyint_max/2+1 );
415 h->param.i_bframe = x264_clip3( h->param.i_bframe, 0, X264_BFRAME_MAX );
416 h->param.i_bframe_bias = x264_clip3( h->param.i_bframe_bias, -90, 100 );
417 h->param.b_bframe_pyramid = h->param.b_bframe_pyramid && h->param.i_bframe > 1;
418 h->param.b_bframe_adaptive = h->param.b_bframe_adaptive && h->param.i_bframe > 0;
419 h->param.analyse.b_weighted_bipred = h->param.analyse.b_weighted_bipred && h->param.i_bframe > 0;
420 h->mb.b_direct_auto_write = h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO
422 && ( h->param.rc.b_stat_write || !h->param.rc.b_stat_read );
424 h->param.i_deblocking_filter_alphac0 = x264_clip3( h->param.i_deblocking_filter_alphac0, -6, 6 );
425 h->param.i_deblocking_filter_beta = x264_clip3( h->param.i_deblocking_filter_beta, -6, 6 );
426 h->param.analyse.i_luma_deadzone[0] = x264_clip3( h->param.analyse.i_luma_deadzone[0], 0, 32 );
427 h->param.analyse.i_luma_deadzone[1] = x264_clip3( h->param.analyse.i_luma_deadzone[1], 0, 32 );
428 h->mb.i_luma_deadzone[0] = 32 - h->param.analyse.i_luma_deadzone[0];
429 h->mb.i_luma_deadzone[1] = 32 - h->param.analyse.i_luma_deadzone[1];
431 h->param.i_cabac_init_idc = x264_clip3( h->param.i_cabac_init_idc, 0, 2 );
433 if( h->param.i_cqm_preset < X264_CQM_FLAT || h->param.i_cqm_preset > X264_CQM_CUSTOM )
434 h->param.i_cqm_preset = X264_CQM_FLAT;
436 if( h->param.analyse.i_me_method < X264_ME_DIA ||
437 h->param.analyse.i_me_method > X264_ME_ESA )
438 h->param.analyse.i_me_method = X264_ME_HEX;
439 if( h->param.analyse.i_me_range < 4 )
440 h->param.analyse.i_me_range = 4;
441 if( h->param.analyse.i_me_range > 16 && h->param.analyse.i_me_method <= X264_ME_HEX )
442 h->param.analyse.i_me_range = 16;
443 h->param.analyse.i_subpel_refine = x264_clip3( h->param.analyse.i_subpel_refine, 1, 7 );
444 h->param.analyse.b_bframe_rdo = h->param.analyse.b_bframe_rdo && h->param.analyse.i_subpel_refine >= 6;
445 h->param.analyse.b_mixed_references = h->param.analyse.b_mixed_references && h->param.i_frame_reference > 1;
446 h->param.analyse.inter &= X264_ANALYSE_PSUB16x16|X264_ANALYSE_PSUB8x8|X264_ANALYSE_BSUB16x16|
447 X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
448 h->param.analyse.intra &= X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
449 if( !(h->param.analyse.inter & X264_ANALYSE_PSUB16x16) )
450 h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
451 if( !h->param.analyse.b_transform_8x8 )
453 h->param.analyse.inter &= ~X264_ANALYSE_I8x8;
454 h->param.analyse.intra &= ~X264_ANALYSE_I8x8;
456 h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -12, 12);
457 if( !h->param.b_cabac )
458 h->param.analyse.i_trellis = 0;
459 h->param.analyse.i_trellis = x264_clip3( h->param.analyse.i_trellis, 0, 2 );
460 h->param.analyse.i_noise_reduction = x264_clip3( h->param.analyse.i_noise_reduction, 0, 1<<16 );
463 const x264_level_t *l = x264_levels;
464 while( l->level_idc != 0 && l->level_idc != h->param.i_level_idc )
466 if( l->level_idc == 0 )
468 x264_log( h, X264_LOG_ERROR, "invalid level_idc: %d\n", h->param.i_level_idc );
471 if( h->param.analyse.i_mv_range <= 0 )
472 h->param.analyse.i_mv_range = l->mv_range;
474 h->param.analyse.i_mv_range = x264_clip3(h->param.analyse.i_mv_range, 32, 2048);
475 if( h->param.analyse.i_direct_8x8_inference < 0 )
476 h->param.analyse.i_direct_8x8_inference = l->direct8x8;
479 if( h->param.rc.f_qblur < 0 )
480 h->param.rc.f_qblur = 0;
481 if( h->param.rc.f_complexity_blur < 0 )
482 h->param.rc.f_complexity_blur = 0;
484 h->param.i_sps_id &= 31;
486 if( h->param.i_log_level < X264_LOG_INFO )
488 h->param.analyse.b_psnr = 0;
489 h->param.analyse.b_ssim = 0;
492 /* ensure the booleans are 0 or 1 so they can be used in math */
493 #define BOOLIFY(x) h->param.x = !!h->param.x
495 BOOLIFY( b_deblocking_filter );
496 BOOLIFY( b_interlaced );
497 BOOLIFY( analyse.b_transform_8x8 );
498 BOOLIFY( analyse.i_direct_8x8_inference );
499 BOOLIFY( analyse.b_bidir_me );
500 BOOLIFY( analyse.b_chroma_me );
501 BOOLIFY( analyse.b_fast_pskip );
502 BOOLIFY( rc.b_stat_write );
503 BOOLIFY( rc.b_stat_read );
509 static void mbcmp_init( x264_t *h )
511 memcpy( h->pixf.mbcmp,
512 ( h->mb.b_lossless || h->param.analyse.i_subpel_refine <= 1 ) ? h->pixf.sad : h->pixf.satd,
513 sizeof(h->pixf.mbcmp) );
516 /****************************************************************************
518 ****************************************************************************/
519 x264_t *x264_encoder_open ( x264_param_t *param )
521 x264_t *h = x264_malloc( sizeof( x264_t ) );
524 memset( h, 0, sizeof( x264_t ) );
526 /* Create a copy of param */
527 memcpy( &h->param, param, sizeof( x264_param_t ) );
529 if( x264_validate_parameters( h ) < 0 )
535 if( h->param.psz_cqm_file )
536 if( x264_cqm_parse_file( h, h->param.psz_cqm_file ) < 0 )
542 if( h->param.rc.psz_stat_out )
543 h->param.rc.psz_stat_out = strdup( h->param.rc.psz_stat_out );
544 if( h->param.rc.psz_stat_in )
545 h->param.rc.psz_stat_in = strdup( h->param.rc.psz_stat_in );
546 if( h->param.rc.psz_rc_eq )
547 h->param.rc.psz_rc_eq = strdup( h->param.rc.psz_rc_eq );
550 if( h->param.vui.i_sar_width > 0 && h->param.vui.i_sar_height > 0 )
552 int i_w = param->vui.i_sar_width;
553 int i_h = param->vui.i_sar_height;
555 x264_reduce_fraction( &i_w, &i_h );
557 while( i_w > 65535 || i_h > 65535 )
563 h->param.vui.i_sar_width = 0;
564 h->param.vui.i_sar_height = 0;
565 if( i_w == 0 || i_h == 0 )
567 x264_log( h, X264_LOG_WARNING, "cannot create valid sample aspect ratio\n" );
571 x264_log( h, X264_LOG_INFO, "using SAR=%d/%d\n", i_w, i_h );
572 h->param.vui.i_sar_width = i_w;
573 h->param.vui.i_sar_height = i_h;
577 x264_reduce_fraction( &h->param.i_fps_num, &h->param.i_fps_den );
581 h->out.i_bitstream = X264_MAX( 1000000, h->param.i_width * h->param.i_height * 1.7
582 * ( h->param.rc.i_rc_method == X264_RC_ABR ? pow( 0.5, h->param.rc.i_qp_min )
583 : pow( 0.5, h->param.rc.i_qp_constant ) * X264_MAX( 1, h->param.rc.f_ip_factor )));
584 h->out.p_bitstream = x264_malloc( h->out.i_bitstream );
590 h->sps = &h->sps_array[0];
591 x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
593 h->pps = &h->pps_array[0];
594 x264_pps_init( h->pps, h->param.i_sps_id, &h->param, h->sps);
596 x264_validate_levels( h );
600 h->mb.i_mb_count = h->sps->i_mb_width * h->sps->i_mb_height;
603 h->frames.i_delay = h->param.i_bframe;
604 h->frames.i_max_ref0 = h->param.i_frame_reference;
605 h->frames.i_max_ref1 = h->sps->vui.i_num_reorder_frames;
606 h->frames.i_max_dpb = h->sps->vui.i_max_dec_frame_buffering + 1;
607 h->frames.b_have_lowres = !h->param.rc.b_stat_read
608 && ( h->param.rc.i_rc_method == X264_RC_ABR
609 || h->param.rc.i_rc_method == X264_RC_CRF
610 || h->param.b_bframe_adaptive );
612 for( i = 0; i < X264_BFRAME_MAX + 3; i++ )
614 h->frames.current[i] = NULL;
615 h->frames.next[i] = NULL;
616 h->frames.unused[i] = NULL;
618 for( i = 0; i < 1 + h->frames.i_delay; i++ )
620 h->frames.unused[i] = x264_frame_new( h );
621 if( !h->frames.unused[i] )
624 for( i = 0; i < h->frames.i_max_dpb; i++ )
626 h->frames.reference[i] = x264_frame_new( h );
627 if( !h->frames.reference[i] )
630 h->frames.reference[h->frames.i_max_dpb] = NULL;
631 h->frames.i_last_idr = - h->param.i_keyint_max;
632 h->frames.i_input = 0;
633 h->frames.last_nonb = NULL;
638 h->fdec = h->frames.reference[0];
640 if( x264_macroblock_cache_init( h ) < 0 )
644 /* init CPU functions */
645 x264_predict_16x16_init( h->param.cpu, h->predict_16x16 );
646 x264_predict_8x8c_init( h->param.cpu, h->predict_8x8c );
647 x264_predict_8x8_init( h->param.cpu, h->predict_8x8 );
648 x264_predict_4x4_init( h->param.cpu, h->predict_4x4 );
650 x264_pixel_init( h->param.cpu, &h->pixf );
651 x264_dct_init( h->param.cpu, &h->dctf );
652 x264_zigzag_init( h->param.cpu, &h->zigzagf, h->param.b_interlaced );
653 x264_mc_init( h->param.cpu, &h->mc );
654 x264_csp_init( h->param.cpu, h->param.i_csp, &h->csp );
655 x264_quant_init( h, h->param.cpu, &h->quantf );
656 x264_deblock_init( h->param.cpu, &h->loopf );
657 x264_dct_init_weights();
662 if( x264_ratecontrol_new( h ) < 0 )
665 x264_log( h, X264_LOG_INFO, "using cpu capabilities %s%s%s%s%s%s\n",
666 param->cpu&X264_CPU_MMX ? "MMX " : "",
667 param->cpu&X264_CPU_MMXEXT ? "MMXEXT " : "",
668 param->cpu&X264_CPU_SSE ? "SSE " : "",
669 param->cpu&X264_CPU_SSE2 ? "SSE2 " : "",
670 param->cpu&X264_CPU_3DNOW ? "3DNow! " : "",
671 param->cpu&X264_CPU_ALTIVEC ? "Altivec " : "" );
675 for( i = 1; i < h->param.i_threads; i++ )
676 h->thread[i] = x264_malloc( sizeof(x264_t) );
678 #ifdef DEBUG_DUMP_FRAME
680 /* create or truncate the reconstructed video file */
681 FILE *f = fopen( "fdec.yuv", "w" );
686 x264_log( h, X264_LOG_ERROR, "can't write to fdec.yuv\n" );
696 /****************************************************************************
697 * x264_encoder_reconfig:
698 ****************************************************************************/
699 int x264_encoder_reconfig( x264_t *h, x264_param_t *param )
701 #define COPY(var) h->param.var = param->var
702 COPY( i_frame_reference ); // but never uses more refs than initially specified
703 COPY( i_bframe_bias );
704 COPY( i_scenecut_threshold );
705 COPY( b_deblocking_filter );
706 COPY( i_deblocking_filter_alphac0 );
707 COPY( i_deblocking_filter_beta );
708 COPY( analyse.intra );
709 COPY( analyse.inter );
710 COPY( analyse.i_direct_mv_pred );
711 COPY( analyse.i_me_method );
712 COPY( analyse.i_me_range );
713 COPY( analyse.i_noise_reduction );
714 COPY( analyse.i_subpel_refine );
715 COPY( analyse.i_trellis );
716 COPY( analyse.b_bidir_me );
717 COPY( analyse.b_bframe_rdo );
718 COPY( analyse.b_chroma_me );
719 COPY( analyse.b_dct_decimate );
720 COPY( analyse.b_fast_pskip );
721 COPY( analyse.b_mixed_references );
724 if( h->pps->b_transform_8x8_mode )
725 h->param.analyse.b_transform_8x8 = param->analyse.b_transform_8x8;
729 return x264_validate_parameters( h );
733 static void x264_nal_start( x264_t *h, int i_type, int i_ref_idc )
735 x264_nal_t *nal = &h->out.nal[h->out.i_nal];
737 nal->i_ref_idc = i_ref_idc;
738 nal->i_type = i_type;
740 bs_align_0( &h->out.bs ); /* not needed */
743 nal->p_payload= &h->out.p_bitstream[bs_pos( &h->out.bs) / 8];
745 static void x264_nal_end( x264_t *h )
747 x264_nal_t *nal = &h->out.nal[h->out.i_nal];
749 bs_align_0( &h->out.bs ); /* not needed */
751 nal->i_payload = &h->out.p_bitstream[bs_pos( &h->out.bs)/8] - nal->p_payload;
756 /****************************************************************************
757 * x264_encoder_headers:
758 ****************************************************************************/
759 int x264_encoder_headers( x264_t *h, x264_nal_t **pp_nal, int *pi_nal )
761 /* init bitstream context */
763 bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
765 /* Put SPS and PPS */
766 if( h->i_frame == 0 )
768 /* identify ourself */
769 x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
770 x264_sei_version_write( h, &h->out.bs );
773 /* generate sequence parameters */
774 x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
775 x264_sps_write( &h->out.bs, h->sps );
778 /* generate picture parameters */
779 x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
780 x264_pps_write( &h->out.bs, h->pps );
784 *pi_nal = h->out.i_nal;
785 *pp_nal = &h->out.nal[0];
791 static void x264_frame_put( x264_frame_t *list[X264_BFRAME_MAX], x264_frame_t *frame )
794 while( list[i] ) i++;
798 static void x264_frame_push( x264_frame_t *list[X264_BFRAME_MAX], x264_frame_t *frame )
801 while( list[i] ) i++;
807 static x264_frame_t *x264_frame_get( x264_frame_t *list[X264_BFRAME_MAX+1] )
809 x264_frame_t *frame = list[0];
811 for( i = 0; list[i]; i++ )
816 static void x264_frame_sort( x264_frame_t *list[X264_BFRAME_MAX+1], int b_dts )
821 for( i = 0; list[i+1]; i++ )
823 int dtype = list[i]->i_type - list[i+1]->i_type;
824 int dtime = list[i]->i_frame - list[i+1]->i_frame;
825 int swap = b_dts ? dtype > 0 || ( dtype == 0 && dtime > 0 )
829 XCHG( x264_frame_t*, list[i], list[i+1] );
835 #define x264_frame_sort_dts(list) x264_frame_sort(list, 1)
836 #define x264_frame_sort_pts(list) x264_frame_sort(list, 0)
838 static inline void x264_reference_build_list( x264_t *h, int i_poc, int i_slice_type )
843 /* build ref list 0/1 */
846 for( i = 1; i < h->frames.i_max_dpb; i++ )
848 if( h->frames.reference[i]->i_poc >= 0 )
850 if( h->frames.reference[i]->i_poc < i_poc )
852 h->fref0[h->i_ref0++] = h->frames.reference[i];
854 else if( h->frames.reference[i]->i_poc > i_poc )
856 h->fref1[h->i_ref1++] = h->frames.reference[i];
861 /* Order ref0 from higher to lower poc */
865 for( i = 0; i < h->i_ref0 - 1; i++ )
867 if( h->fref0[i]->i_poc < h->fref0[i+1]->i_poc )
869 XCHG( x264_frame_t*, h->fref0[i], h->fref0[i+1] );
875 /* Order ref1 from lower to higher poc (bubble sort) for B-frame */
879 for( i = 0; i < h->i_ref1 - 1; i++ )
881 if( h->fref1[i]->i_poc > h->fref1[i+1]->i_poc )
883 XCHG( x264_frame_t*, h->fref1[i], h->fref1[i+1] );
890 /* In the standard, a P-frame's ref list is sorted by frame_num.
891 * We use POC, but check whether explicit reordering is needed */
892 h->b_ref_reorder[0] =
893 h->b_ref_reorder[1] = 0;
894 if( i_slice_type == SLICE_TYPE_P )
896 for( i = 0; i < h->i_ref0 - 1; i++ )
897 if( h->fref0[i]->i_frame_num < h->fref0[i+1]->i_frame_num )
899 h->b_ref_reorder[0] = 1;
904 h->i_ref1 = X264_MIN( h->i_ref1, h->frames.i_max_ref1 );
905 h->i_ref0 = X264_MIN( h->i_ref0, h->frames.i_max_ref0 );
906 h->i_ref0 = X264_MIN( h->i_ref0, h->param.i_frame_reference ); // if reconfig() has lowered the limit
907 h->i_ref0 = X264_MIN( h->i_ref0, 16 - h->i_ref1 );
908 h->mb.pic.i_fref[0] = h->i_ref0;
909 h->mb.pic.i_fref[1] = h->i_ref1;
912 static inline void x264_fdec_deblock( x264_t *h )
914 /* apply deblocking filter to the current decoded picture */
915 if( !h->sh.i_disable_deblocking_filter_idc )
917 TIMER_START( i_mtime_filter );
918 x264_frame_deblocking_filter( h, h->sh.i_type );
919 TIMER_STOP( i_mtime_filter );
923 static inline void x264_reference_update( x264_t *h )
927 x264_fdec_deblock( h );
930 x264_frame_expand_border( h, h->fdec );
932 /* create filtered images */
933 x264_frame_filter( h->param.cpu, h->fdec, h->sh.b_mbaff );
935 /* expand border of filtered images */
936 x264_frame_expand_border_filtered( h, h->fdec );
938 /* move lowres copy of the image to the ref frame */
939 for( i = 0; i < 4; i++)
941 XCHG( uint8_t*, h->fdec->lowres[i], h->fenc->lowres[i] );
942 XCHG( uint8_t*, h->fdec->buffer_lowres[i], h->fenc->buffer_lowres[i] );
945 /* adaptive B decision needs a pointer, since it can't use the ref lists */
946 if( h->sh.i_type != SLICE_TYPE_B )
947 h->frames.last_nonb = h->fdec;
949 /* move frame in the buffer */
950 h->fdec = h->frames.reference[h->frames.i_max_dpb-1];
951 for( i = h->frames.i_max_dpb-1; i > 0; i-- )
953 h->frames.reference[i] = h->frames.reference[i-1];
955 h->frames.reference[0] = h->fdec;
958 static inline void x264_reference_reset( x264_t *h )
962 /* reset ref pictures */
963 for( i = 1; i < h->frames.i_max_dpb; i++ )
965 h->frames.reference[i]->i_poc = -1;
967 h->frames.reference[0]->i_poc = 0;
970 static inline void x264_slice_init( x264_t *h, int i_nal_type, int i_slice_type, int i_global_qp )
972 /* ------------------------ Create slice header ----------------------- */
973 if( i_nal_type == NAL_SLICE_IDR )
975 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 );
978 h->i_idr_pic_id = ( h->i_idr_pic_id + 1 ) % 65536;
982 x264_slice_header_init( h, &h->sh, h->sps, h->pps, i_slice_type, -1, h->i_frame_num, i_global_qp );
984 /* always set the real higher num of ref frame used */
985 h->sh.b_num_ref_idx_override = 1;
986 h->sh.i_num_ref_idx_l0_active = h->i_ref0 <= 0 ? 1 : h->i_ref0;
987 h->sh.i_num_ref_idx_l1_active = h->i_ref1 <= 0 ? 1 : h->i_ref1;
990 h->fdec->i_frame_num = h->sh.i_frame_num;
992 if( h->sps->i_poc_type == 0 )
994 h->sh.i_poc_lsb = h->fdec->i_poc & ( (1 << h->sps->i_log2_max_poc_lsb) - 1 );
995 h->sh.i_delta_poc_bottom = 0; /* XXX won't work for field */
997 else if( h->sps->i_poc_type == 1 )
999 /* FIXME TODO FIXME */
1003 /* Nothing to do ? */
1006 x264_macroblock_slice_init( h );
1009 static int x264_slice_write( x264_t *h )
1016 memset( &h->stat.frame, 0, sizeof(h->stat.frame) );
1019 x264_nal_start( h, h->i_nal_type, h->i_nal_ref_idc );
1022 x264_slice_header_write( &h->out.bs, &h->sh, h->i_nal_ref_idc );
1023 if( h->param.b_cabac )
1025 /* alignment needed */
1026 bs_align_1( &h->out.bs );
1029 x264_cabac_context_init( &h->cabac, h->sh.i_type, h->sh.i_qp, h->sh.i_cabac_init_idc );
1030 x264_cabac_encode_init ( &h->cabac, &h->out.bs );
1032 h->mb.i_last_qp = h->sh.i_qp;
1033 h->mb.i_last_dqp = 0;
1035 for( mb_xy = h->sh.i_first_mb, i_skip = 0; mb_xy < h->sh.i_last_mb; )
1037 const int i_mb_y = mb_xy / h->sps->i_mb_width;
1038 const int i_mb_x = mb_xy % h->sps->i_mb_width;
1040 int mb_spos = bs_pos(&h->out.bs);
1043 x264_macroblock_cache_load( h, i_mb_x, i_mb_y );
1045 /* analyse parameters
1046 * Slice I: choose I_4x4 or I_16x16 mode
1047 * Slice P: choose between using P mode or intra (4x4 or 16x16)
1049 TIMER_START( i_mtime_analyse );
1050 x264_macroblock_analyse( h );
1051 TIMER_STOP( i_mtime_analyse );
1053 /* encode this macroblock -> be careful it can change the mb type to P_SKIP if needed */
1054 TIMER_START( i_mtime_encode );
1055 x264_macroblock_encode( h );
1056 TIMER_STOP( i_mtime_encode );
1058 TIMER_START( i_mtime_write );
1059 if( h->param.b_cabac )
1061 if( mb_xy > h->sh.i_first_mb && !(h->sh.b_mbaff && (i_mb_y&1)) )
1062 x264_cabac_encode_terminal( &h->cabac, 0 );
1064 if( IS_SKIP( h->mb.i_type ) )
1065 x264_cabac_mb_skip( h, 1 );
1068 if( h->sh.i_type != SLICE_TYPE_I )
1069 x264_cabac_mb_skip( h, 0 );
1070 x264_macroblock_write_cabac( h, &h->cabac );
1075 if( IS_SKIP( h->mb.i_type ) )
1079 if( h->sh.i_type != SLICE_TYPE_I )
1081 bs_write_ue( &h->out.bs, i_skip ); /* skip run */
1084 x264_macroblock_write_cavlc( h, &h->out.bs );
1087 TIMER_STOP( i_mtime_write );
1090 if( h->param.b_visualize )
1091 x264_visualize_mb( h );
1095 x264_macroblock_cache_save( h );
1097 /* accumulate mb stats */
1098 h->stat.frame.i_mb_count[h->mb.i_type]++;
1099 if( !IS_SKIP(h->mb.i_type) && !IS_INTRA(h->mb.i_type) && !IS_DIRECT(h->mb.i_type) )
1101 if( h->mb.i_partition != D_8x8 )
1102 h->stat.frame.i_mb_count_size[ x264_mb_partition_pixel_table[ h->mb.i_partition ] ] += 4;
1104 for( i = 0; i < 4; i++ )
1105 h->stat.frame.i_mb_count_size[ x264_mb_partition_pixel_table[ h->mb.i_sub_partition[i] ] ] ++;
1106 if( h->param.i_frame_reference > 1 )
1108 for( i = 0; i < 4; i++ )
1110 int i_ref = h->mb.cache.ref[0][ x264_scan8[4*i] ];
1112 h->stat.frame.i_mb_count_ref[i_ref] ++;
1116 if( h->mb.i_cbp_luma && !IS_INTRA(h->mb.i_type) )
1118 h->stat.frame.i_mb_count_8x8dct[0] ++;
1119 h->stat.frame.i_mb_count_8x8dct[1] += h->mb.b_transform_8x8;
1122 if( h->mb.b_variable_qp )
1123 x264_ratecontrol_mb(h, bs_pos(&h->out.bs) - mb_spos);
1127 if( (i_mb_y&1) && i_mb_x == h->sps->i_mb_width - 1 )
1130 mb_xy += 1 - h->sps->i_mb_width;
1132 mb_xy += h->sps->i_mb_width;
1138 if( h->param.b_cabac )
1141 x264_cabac_encode_terminal( &h->cabac, 1 );
1143 else if( i_skip > 0 )
1145 bs_write_ue( &h->out.bs, i_skip ); /* last skip run */
1148 if( h->param.b_cabac )
1150 x264_cabac_encode_flush( &h->cabac );
1155 /* rbsp_slice_trailing_bits */
1156 bs_rbsp_trailing( &h->out.bs );
1161 /* Compute misc bits */
1162 h->stat.frame.i_misc_bits = bs_pos( &h->out.bs )
1164 - h->stat.frame.i_itex_bits
1165 - h->stat.frame.i_ptex_bits
1166 - h->stat.frame.i_hdr_bits;
1171 static inline int x264_slices_write( x264_t *h )
1176 if( h->param.b_visualize )
1177 x264_visualize_init( h );
1180 if( h->param.i_threads == 1 )
1182 x264_ratecontrol_threads_start( h );
1183 x264_slice_write( h );
1184 i_frame_size = h->out.nal[h->out.i_nal-1].i_payload;
1188 int i_nal = h->out.i_nal;
1189 int i_bs_size = h->out.i_bitstream / h->param.i_threads;
1191 /* duplicate contexts */
1192 for( i = 0; i < h->param.i_threads; i++ )
1194 x264_t *t = h->thread[i];
1195 int mb_height = h->sps->i_mb_height >> h->sh.b_mbaff;
1196 int mb_width = h->sps->i_mb_width << h->sh.b_mbaff;
1199 memcpy( t, h, sizeof(x264_t) );
1200 t->out.p_bitstream += i*i_bs_size;
1201 bs_init( &t->out.bs, t->out.p_bitstream, i_bs_size );
1202 t->i_thread_num = i;
1204 t->sh.i_first_mb = (i * mb_height / h->param.i_threads) * mb_width;
1205 t->sh.i_last_mb = ((i+1) * mb_height / h->param.i_threads) * mb_width;
1206 t->out.i_nal = i_nal + i;
1208 x264_ratecontrol_threads_start( h );
1213 pthread_t handles[X264_SLICE_MAX];
1214 for( i = 0; i < h->param.i_threads; i++ )
1215 pthread_create( &handles[i], NULL, (void*)x264_slice_write, (void*)h->thread[i] );
1216 for( i = 0; i < h->param.i_threads; i++ )
1217 pthread_join( handles[i], NULL );
1220 for( i = 0; i < h->param.i_threads; i++ )
1221 x264_slice_write( h->thread[i] );
1224 /* merge contexts */
1225 i_frame_size = h->out.nal[i_nal].i_payload;
1226 for( i = 1; i < h->param.i_threads; i++ )
1229 x264_t *t = h->thread[i];
1230 h->out.nal[i_nal+i] = t->out.nal[i_nal+i];
1231 i_frame_size += t->out.nal[i_nal+i].i_payload;
1232 // all entries in stat.frame are ints
1233 for( j = 0; j < sizeof(h->stat.frame) / sizeof(int); j++ )
1234 ((int*)&h->stat.frame)[j] += ((int*)&t->stat.frame)[j];
1236 h->out.i_nal = i_nal + h->param.i_threads;
1240 if( h->param.b_visualize )
1242 x264_visualize_show( h );
1243 x264_visualize_close( h );
1247 return i_frame_size;
1250 /****************************************************************************
1251 * x264_encoder_encode:
1252 * XXX: i_poc : is the poc of the current given picture
1253 * i_frame : is the number of the frame being coded
1254 * ex: type frame poc
1262 ****************************************************************************/
1263 int x264_encoder_encode( x264_t *h,
1264 x264_nal_t **pp_nal, int *pi_nal,
1265 x264_picture_t *pic_in,
1266 x264_picture_t *pic_out )
1268 x264_frame_t *frame_psnr = h->fdec; /* just to keep the current decoded frame for psnr calculation */
1278 char psz_message[80];
1285 /* ------------------- Setup new frame from picture -------------------- */
1286 TIMER_START( i_mtime_encode_frame );
1287 if( pic_in != NULL )
1289 /* 1: Copy the picture to a frame and move it to a buffer */
1290 x264_frame_t *fenc = x264_frame_get( h->frames.unused );
1292 x264_frame_copy_picture( h, fenc, pic_in );
1294 if( h->param.i_width != 16 * h->sps->i_mb_width ||
1295 h->param.i_height != 16 * h->sps->i_mb_height )
1296 x264_frame_expand_border_mod16( h, fenc );
1298 fenc->i_frame = h->frames.i_input++;
1300 x264_frame_put( h->frames.next, fenc );
1302 if( h->frames.b_have_lowres )
1303 x264_frame_init_lowres( h->param.cpu, fenc );
1305 if( h->frames.i_input <= h->frames.i_delay )
1307 /* Nothing yet to encode */
1308 /* waiting for filling bframe buffer */
1309 pic_out->i_type = X264_TYPE_AUTO;
1314 if( h->frames.current[0] == NULL )
1317 /* 2: Select frame types */
1318 if( h->frames.next[0] == NULL )
1321 x264_slicetype_decide( h );
1323 /* 3: move some B-frames and 1 non-B to encode queue */
1324 while( IS_X264_TYPE_B( h->frames.next[bframes]->i_type ) )
1326 x264_frame_put( h->frames.current, x264_frame_get( &h->frames.next[bframes] ) );
1327 /* FIXME: when max B-frames > 3, BREF may no longer be centered after GOP closing */
1328 if( h->param.b_bframe_pyramid && bframes > 1 )
1330 x264_frame_t *mid = x264_frame_get( &h->frames.next[bframes/2] );
1331 mid->i_type = X264_TYPE_BREF;
1332 x264_frame_put( h->frames.current, mid );
1336 x264_frame_put( h->frames.current, x264_frame_get( h->frames.next ) );
1338 TIMER_STOP( i_mtime_encode_frame );
1340 /* ------------------- Get frame to be encoded ------------------------- */
1341 /* 4: get picture to encode */
1342 h->fenc = x264_frame_get( h->frames.current );
1343 if( h->fenc == NULL )
1345 /* Nothing yet to encode (ex: waiting for I/P with B frames) */
1346 /* waiting for filling bframe buffer */
1347 pic_out->i_type = X264_TYPE_AUTO;
1353 if( h->fenc->i_type == X264_TYPE_IDR )
1355 h->frames.i_last_idr = h->fenc->i_frame;
1358 /* ------------------- Setup frame context ----------------------------- */
1359 /* 5: Init data dependent of frame type */
1360 TIMER_START( i_mtime_encode_frame );
1361 if( h->fenc->i_type == X264_TYPE_IDR )
1363 /* reset ref pictures */
1364 x264_reference_reset( h );
1366 i_nal_type = NAL_SLICE_IDR;
1367 i_nal_ref_idc = NAL_PRIORITY_HIGHEST;
1368 i_slice_type = SLICE_TYPE_I;
1370 else if( h->fenc->i_type == X264_TYPE_I )
1372 i_nal_type = NAL_SLICE;
1373 i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
1374 i_slice_type = SLICE_TYPE_I;
1376 else if( h->fenc->i_type == X264_TYPE_P )
1378 i_nal_type = NAL_SLICE;
1379 i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
1380 i_slice_type = SLICE_TYPE_P;
1382 else if( h->fenc->i_type == X264_TYPE_BREF )
1384 i_nal_type = NAL_SLICE;
1385 i_nal_ref_idc = NAL_PRIORITY_HIGH; /* maybe add MMCO to forget it? -> low */
1386 i_slice_type = SLICE_TYPE_B;
1390 i_nal_type = NAL_SLICE;
1391 i_nal_ref_idc = NAL_PRIORITY_DISPOSABLE;
1392 i_slice_type = SLICE_TYPE_B;
1396 h->fenc->i_poc = 2 * (h->fenc->i_frame - h->frames.i_last_idr);
1397 h->fdec->i_type = h->fenc->i_type;
1398 h->fdec->i_frame = h->fenc->i_frame;
1399 h->fenc->b_kept_as_ref =
1400 h->fdec->b_kept_as_ref = i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE;
1404 /* ------------------- Init ----------------------------- */
1405 /* build ref list 0/1 */
1406 x264_reference_build_list( h, h->fdec->i_poc, i_slice_type );
1408 /* Init the rate control */
1409 x264_ratecontrol_start( h, i_slice_type, h->fenc->i_qpplus1 );
1410 i_global_qp = x264_ratecontrol_qp( h );
1412 pic_out->i_qpplus1 =
1413 h->fdec->i_qpplus1 = i_global_qp + 1;
1415 if( i_slice_type == SLICE_TYPE_B )
1416 x264_macroblock_bipred_init( h );
1418 /* ------------------------ Create slice header ----------------------- */
1419 x264_slice_init( h, i_nal_type, i_slice_type, i_global_qp );
1421 if( h->fenc->b_kept_as_ref )
1424 /* ---------------------- Write the bitstream -------------------------- */
1425 /* Init bitstream context */
1427 bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
1432 if(i_slice_type == SLICE_TYPE_I)
1434 else if(i_slice_type == SLICE_TYPE_P)
1436 else if(i_slice_type == SLICE_TYPE_B)
1441 x264_nal_start(h, NAL_AUD, NAL_PRIORITY_DISPOSABLE);
1442 bs_write(&h->out.bs, 3, pic_type);
1443 bs_rbsp_trailing(&h->out.bs);
1447 h->i_nal_type = i_nal_type;
1448 h->i_nal_ref_idc = i_nal_ref_idc;
1450 /* Write SPS and PPS */
1451 if( i_nal_type == NAL_SLICE_IDR && h->param.b_repeat_headers )
1453 if( h->fenc->i_frame == 0 )
1455 /* identify ourself */
1456 x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
1457 x264_sei_version_write( h, &h->out.bs );
1461 /* generate sequence parameters */
1462 x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
1463 x264_sps_write( &h->out.bs, h->sps );
1466 /* generate picture parameters */
1467 x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
1468 x264_pps_write( &h->out.bs, h->pps );
1473 i_frame_size = x264_slices_write( h );
1475 /* restore CPU state (before using float again) */
1476 x264_cpu_restore( h->param.cpu );
1478 if( i_slice_type == SLICE_TYPE_P && !h->param.rc.b_stat_read
1479 && h->param.i_scenecut_threshold >= 0 )
1481 const int *mbs = h->stat.frame.i_mb_count;
1482 int i_mb_i = mbs[I_16x16] + mbs[I_8x8] + mbs[I_4x4];
1483 int i_mb_p = mbs[P_L0] + mbs[P_8x8];
1484 int i_mb_s = mbs[P_SKIP];
1485 int i_mb = h->sps->i_mb_width * h->sps->i_mb_height;
1486 int64_t i_inter_cost = h->stat.frame.i_inter_cost;
1487 int64_t i_intra_cost = h->stat.frame.i_intra_cost;
1490 int i_gop_size = h->fenc->i_frame - h->frames.i_last_idr;
1491 float f_thresh_max = h->param.i_scenecut_threshold / 100.0;
1492 /* magic numbers pulled out of thin air */
1493 float f_thresh_min = f_thresh_max * h->param.i_keyint_min
1494 / ( h->param.i_keyint_max * 4 );
1495 if( h->param.i_keyint_min == h->param.i_keyint_max )
1496 f_thresh_min= f_thresh_max;
1498 /* macroblock_analyse() doesn't further analyse skipped mbs,
1499 * so we have to guess their cost */
1501 i_intra_cost = i_intra_cost * i_mb / (i_mb - i_mb_s);
1503 if( i_gop_size < h->param.i_keyint_min / 4 )
1504 f_bias = f_thresh_min / 4;
1505 else if( i_gop_size <= h->param.i_keyint_min )
1506 f_bias = f_thresh_min * i_gop_size / h->param.i_keyint_min;
1509 f_bias = f_thresh_min
1510 + ( f_thresh_max - f_thresh_min )
1511 * ( i_gop_size - h->param.i_keyint_min )
1512 / ( h->param.i_keyint_max - h->param.i_keyint_min );
1514 f_bias = X264_MIN( f_bias, 1.0 );
1516 /* Bad P will be reencoded as I */
1517 if( i_mb_s < i_mb &&
1518 i_inter_cost >= (1.0 - f_bias) * i_intra_cost )
1522 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",
1524 (double)i_intra_cost, (double)i_inter_cost,
1525 (double)i_inter_cost / i_intra_cost,
1527 i_mb_i, i_mb_p, i_mb_s );
1529 /* Restore frame num */
1532 for( b = 0; h->frames.current[b] && IS_X264_TYPE_B( h->frames.current[b]->i_type ); b++ );
1535 /* If using B-frames, force GOP to be closed.
1536 * Even if this frame is going to be I and not IDR, forcing a
1537 * P-frame before the scenecut will probably help compression.
1539 * We don't yet know exactly which frame is the scene cut, so
1540 * we can't assign an I-frame. Instead, change the previous
1541 * B-frame to P, and rearrange coding order. */
1543 if( h->param.b_bframe_adaptive || b > 1 )
1544 h->fenc->i_type = X264_TYPE_AUTO;
1545 x264_frame_sort_pts( h->frames.current );
1546 x264_frame_push( h->frames.next, h->fenc );
1547 h->fenc = h->frames.current[b-1];
1548 h->frames.current[b-1] = NULL;
1549 h->fenc->i_type = X264_TYPE_P;
1550 x264_frame_sort_dts( h->frames.current );
1552 /* Do IDR if needed */
1553 else if( i_gop_size >= h->param.i_keyint_min )
1560 /* Reinit field of fenc */
1561 h->fenc->i_type = X264_TYPE_IDR;
1564 /* Put enqueued frames back in the pool */
1565 while( (tmp = x264_frame_get( h->frames.current ) ) != NULL )
1566 x264_frame_put( h->frames.next, tmp );
1567 x264_frame_sort_pts( h->frames.next );
1571 h->fenc->i_type = X264_TYPE_I;
1577 /* End bitstream, set output */
1578 *pi_nal = h->out.i_nal;
1579 *pp_nal = h->out.nal;
1581 /* Set output picture properties */
1582 if( i_slice_type == SLICE_TYPE_I )
1583 pic_out->i_type = i_nal_type == NAL_SLICE_IDR ? X264_TYPE_IDR : X264_TYPE_I;
1584 else if( i_slice_type == SLICE_TYPE_P )
1585 pic_out->i_type = X264_TYPE_P;
1587 pic_out->i_type = X264_TYPE_B;
1588 pic_out->i_pts = h->fenc->i_pts;
1590 pic_out->img.i_plane = h->fdec->i_plane;
1591 for(i = 0; i < 4; i++){
1592 pic_out->img.i_stride[i] = h->fdec->i_stride[i];
1593 pic_out->img.plane[i] = h->fdec->plane[i];
1596 /* ---------------------- Update encoder state ------------------------- */
1599 x264_cpu_restore( h->param.cpu );
1600 x264_ratecontrol_end( h, i_frame_size * 8 );
1602 /* handle references */
1603 if( i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE && h->param.i_keyint_max > 1 )
1604 x264_reference_update( h );
1605 #ifdef DEBUG_DUMP_FRAME
1607 x264_fdec_deblock( h );
1609 x264_frame_put( h->frames.unused, h->fenc );
1611 /* increase frame count */
1614 /* restore CPU state (before using float again) */
1615 x264_cpu_restore( h->param.cpu );
1617 x264_noise_reduction_update( h );
1619 TIMER_STOP( i_mtime_encode_frame );
1621 /* ---------------------- Compute/Print statistics --------------------- */
1623 h->stat.i_slice_count[i_slice_type]++;
1624 h->stat.i_slice_size[i_slice_type] += i_frame_size + NALU_OVERHEAD;
1625 h->stat.i_slice_qp[i_slice_type] += i_global_qp;
1627 for( i = 0; i < X264_MBTYPE_MAX; i++ )
1628 h->stat.i_mb_count[h->sh.i_type][i] += h->stat.frame.i_mb_count[i];
1629 for( i = 0; i < 2; i++ )
1630 h->stat.i_mb_count_8x8dct[i] += h->stat.frame.i_mb_count_8x8dct[i];
1631 if( h->sh.i_type != SLICE_TYPE_I )
1633 for( i = 0; i < 7; i++ )
1634 h->stat.i_mb_count_size[h->sh.i_type][i] += h->stat.frame.i_mb_count_size[i];
1635 for( i = 0; i < 32; i++ )
1636 h->stat.i_mb_count_ref[h->sh.i_type][i] += h->stat.frame.i_mb_count_ref[i];
1638 if( i_slice_type == SLICE_TYPE_B )
1640 h->stat.i_direct_frames[ h->sh.b_direct_spatial_mv_pred ] ++;
1641 if( h->mb.b_direct_auto_write )
1643 //FIXME somewhat arbitrary time constants
1644 if( h->stat.i_direct_score[0] + h->stat.i_direct_score[1] > h->mb.i_mb_count )
1646 for( i = 0; i < 2; i++ )
1647 h->stat.i_direct_score[i] = h->stat.i_direct_score[i] * 9/10;
1649 for( i = 0; i < 2; i++ )
1650 h->stat.i_direct_score[i] += h->stat.frame.i_direct_score[i];
1654 psz_message[0] = '\0';
1655 if( h->param.analyse.b_psnr )
1659 for( i=0; i<3; i++ )
1661 sqe[i] = x264_pixel_ssd_wxh( &h->pixf,
1662 frame_psnr->plane[i], frame_psnr->i_stride[i],
1663 h->fenc->plane[i], h->fenc->i_stride[i],
1664 h->param.i_width >> !!i, h->param.i_height >> !!i );
1666 x264_cpu_restore( h->param.cpu );
1668 h->stat.i_sqe_global[i_slice_type] += sqe[0] + sqe[1] + sqe[2];
1669 h->stat.f_psnr_average[i_slice_type] += x264_psnr( sqe[0] + sqe[1] + sqe[2], 3 * h->param.i_width * h->param.i_height / 2 );
1670 h->stat.f_psnr_mean_y[i_slice_type] += x264_psnr( sqe[0], h->param.i_width * h->param.i_height );
1671 h->stat.f_psnr_mean_u[i_slice_type] += x264_psnr( sqe[1], h->param.i_width * h->param.i_height / 4 );
1672 h->stat.f_psnr_mean_v[i_slice_type] += x264_psnr( sqe[2], h->param.i_width * h->param.i_height / 4 );
1674 snprintf( psz_message, 80, " PSNR Y:%5.2f U:%5.2f V:%5.2f",
1675 x264_psnr( sqe[0], h->param.i_width * h->param.i_height ),
1676 x264_psnr( sqe[1], h->param.i_width * h->param.i_height / 4),
1677 x264_psnr( sqe[2], h->param.i_width * h->param.i_height / 4) );
1680 if( h->param.analyse.b_ssim )
1682 // offset by 2 pixels to avoid alignment of ssim blocks with dct blocks
1683 float ssim_y = x264_pixel_ssim_wxh( &h->pixf,
1684 frame_psnr->plane[0] + 2+2*frame_psnr->i_stride[0], frame_psnr->i_stride[0],
1685 h->fenc->plane[0] + 2+2*h->fenc->i_stride[0], h->fenc->i_stride[0],
1686 h->param.i_width-2, h->param.i_height-2 );
1687 h->stat.f_ssim_mean_y[i_slice_type] += ssim_y;
1688 snprintf( psz_message + strlen(psz_message), 80 - strlen(psz_message),
1689 " SSIM Y:%.5f", ssim_y );
1691 psz_message[79] = '\0';
1693 x264_log( h, X264_LOG_DEBUG,
1694 "frame=%4d QP=%i NAL=%d Slice:%c Poc:%-3d I:%-4d P:%-4d SKIP:%-4d size=%d bytes%s\n",
1698 i_slice_type == SLICE_TYPE_I ? 'I' : (i_slice_type == SLICE_TYPE_P ? 'P' : 'B' ),
1700 h->stat.frame.i_mb_count_i,
1701 h->stat.frame.i_mb_count_p,
1702 h->stat.frame.i_mb_count_skip,
1707 #ifdef DEBUG_MB_TYPE
1709 static const char mb_chars[] = { 'i', 'i', 'I', 'C', 'P', '8', 'S',
1710 'D', '<', 'X', 'B', 'X', '>', 'B', 'B', 'B', 'B', '8', 'S' };
1712 for( mb_xy = 0; mb_xy < h->sps->i_mb_width * h->sps->i_mb_height; mb_xy++ )
1714 if( h->mb.type[mb_xy] < X264_MBTYPE_MAX && h->mb.type[mb_xy] >= 0 )
1715 fprintf( stderr, "%c ", mb_chars[ h->mb.type[mb_xy] ] );
1717 fprintf( stderr, "? " );
1719 if( (mb_xy+1) % h->sps->i_mb_width == 0 )
1720 fprintf( stderr, "\n" );
1725 #ifdef DEBUG_DUMP_FRAME
1726 /* Dump reconstructed frame */
1727 x264_frame_dump( h, frame_psnr, "fdec.yuv" );
1732 /****************************************************************************
1733 * x264_encoder_close:
1734 ****************************************************************************/
1735 void x264_encoder_close ( x264_t *h )
1737 #ifdef DEBUG_BENCHMARK
1738 int64_t i_mtime_total = i_mtime_analyse + i_mtime_encode + i_mtime_write + i_mtime_filter + 1;
1740 int64_t i_yuv_size = 3 * h->param.i_width * h->param.i_height / 2;
1743 #ifdef DEBUG_BENCHMARK
1744 x264_log( h, X264_LOG_INFO,
1745 "analyse=%d(%lldms) encode=%d(%lldms) write=%d(%lldms) filter=%d(%lldms)\n",
1746 (int)(100*i_mtime_analyse/i_mtime_total), i_mtime_analyse/1000,
1747 (int)(100*i_mtime_encode/i_mtime_total), i_mtime_encode/1000,
1748 (int)(100*i_mtime_write/i_mtime_total), i_mtime_write/1000,
1749 (int)(100*i_mtime_filter/i_mtime_total), i_mtime_filter/1000 );
1752 /* Slices used and PSNR */
1753 for( i=0; i<5; i++ )
1755 static const int slice_order[] = { SLICE_TYPE_I, SLICE_TYPE_SI, SLICE_TYPE_P, SLICE_TYPE_SP, SLICE_TYPE_B };
1756 static const char *slice_name[] = { "P", "B", "I", "SP", "SI" };
1757 int i_slice = slice_order[i];
1759 if( h->stat.i_slice_count[i_slice] > 0 )
1761 const int i_count = h->stat.i_slice_count[i_slice];
1762 if( h->param.analyse.b_psnr )
1764 x264_log( h, X264_LOG_INFO,
1765 "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",
1766 slice_name[i_slice],
1768 (double)h->stat.i_slice_qp[i_slice] / i_count,
1769 (double)h->stat.i_slice_size[i_slice] / i_count,
1770 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,
1771 h->stat.f_psnr_average[i_slice] / i_count,
1772 x264_psnr( h->stat.i_sqe_global[i_slice], i_count * i_yuv_size ) );
1776 x264_log( h, X264_LOG_INFO,
1777 "slice %s:%-5d Avg QP:%5.2f size:%6.0f\n",
1778 slice_name[i_slice],
1780 (double)h->stat.i_slice_qp[i_slice] / i_count,
1781 (double)h->stat.i_slice_size[i_slice] / i_count );
1787 if( h->stat.i_slice_count[SLICE_TYPE_I] > 0 )
1789 const int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_I];
1790 const double i_count = h->stat.i_slice_count[SLICE_TYPE_I] * h->mb.i_mb_count / 100.0;
1791 x264_log( h, X264_LOG_INFO,
1792 "mb I I16..4: %4.1f%% %4.1f%% %4.1f%%\n",
1793 i_mb_count[I_16x16]/ i_count,
1794 i_mb_count[I_8x8] / i_count,
1795 i_mb_count[I_4x4] / i_count );
1797 if( h->stat.i_slice_count[SLICE_TYPE_P] > 0 )
1799 const int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_P];
1800 const int64_t *i_mb_size = h->stat.i_mb_count_size[SLICE_TYPE_P];
1801 const double i_count = h->stat.i_slice_count[SLICE_TYPE_P] * h->mb.i_mb_count / 100.0;
1802 x264_log( h, X264_LOG_INFO,
1803 "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",
1804 i_mb_count[I_16x16]/ i_count,
1805 i_mb_count[I_8x8] / i_count,
1806 i_mb_count[I_4x4] / i_count,
1807 i_mb_size[PIXEL_16x16] / (i_count*4),
1808 (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
1809 i_mb_size[PIXEL_8x8] / (i_count*4),
1810 (i_mb_size[PIXEL_8x4] + i_mb_size[PIXEL_4x8]) / (i_count*4),
1811 i_mb_size[PIXEL_4x4] / (i_count*4),
1812 i_mb_count[P_SKIP] / i_count );
1814 if( h->stat.i_slice_count[SLICE_TYPE_B] > 0 )
1816 const int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_B];
1817 const int64_t *i_mb_size = h->stat.i_mb_count_size[SLICE_TYPE_B];
1818 const double i_count = h->stat.i_slice_count[SLICE_TYPE_B] * h->mb.i_mb_count / 100.0;
1819 x264_log( h, X264_LOG_INFO,
1820 "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",
1821 i_mb_count[I_16x16] / i_count,
1822 i_mb_count[I_8x8] / i_count,
1823 i_mb_count[I_4x4] / i_count,
1824 i_mb_size[PIXEL_16x16] / (i_count*4),
1825 (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
1826 i_mb_size[PIXEL_8x8] / (i_count*4),
1827 i_mb_count[B_DIRECT] / i_count,
1828 i_mb_count[B_SKIP] / i_count );
1831 x264_ratecontrol_summary( h );
1833 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 )
1835 const int i_count = h->stat.i_slice_count[SLICE_TYPE_I] +
1836 h->stat.i_slice_count[SLICE_TYPE_P] +
1837 h->stat.i_slice_count[SLICE_TYPE_B];
1838 float fps = (float) h->param.i_fps_num / h->param.i_fps_den;
1839 #define SUM3(p) (p[SLICE_TYPE_I] + p[SLICE_TYPE_P] + p[SLICE_TYPE_B])
1840 #define SUM3b(p,o) (p[SLICE_TYPE_I][o] + p[SLICE_TYPE_P][o] + p[SLICE_TYPE_B][o])
1841 float f_bitrate = fps * SUM3(h->stat.i_slice_size) / i_count / 125;
1843 if( h->pps->b_transform_8x8_mode )
1845 int64_t i_i8x8 = SUM3b( h->stat.i_mb_count, I_8x8 );
1846 int64_t i_intra = i_i8x8 + SUM3b( h->stat.i_mb_count, I_4x4 )
1847 + SUM3b( h->stat.i_mb_count, I_16x16 );
1848 x264_log( h, X264_LOG_INFO, "8x8 transform intra:%.1f%% inter:%.1f%%\n",
1849 100. * i_i8x8 / i_intra,
1850 100. * h->stat.i_mb_count_8x8dct[1] / h->stat.i_mb_count_8x8dct[0] );
1853 if( h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO
1854 && h->stat.i_slice_count[SLICE_TYPE_B] )
1856 x264_log( h, X264_LOG_INFO, "direct mvs spatial:%.1f%% temporal:%.1f%%\n",
1857 h->stat.i_direct_frames[1] * 100. / h->stat.i_slice_count[SLICE_TYPE_B],
1858 h->stat.i_direct_frames[0] * 100. / h->stat.i_slice_count[SLICE_TYPE_B] );
1861 if( h->frames.i_max_ref0 > 1 )
1864 for( i_slice = 0; i_slice < 2; i_slice++ )
1870 for( i = 0; i < h->frames.i_max_ref0 << h->param.b_interlaced; i++ )
1871 if( h->stat.i_mb_count_ref[i_slice][i] )
1873 i_den += h->stat.i_mb_count_ref[i_slice][i];
1878 for( i = 0; i <= i_max; i++ )
1879 p += sprintf( p, " %4.1f%%", 100. * h->stat.i_mb_count_ref[i_slice][i] / i_den );
1880 x264_log( h, X264_LOG_INFO, "ref %c %s\n", i_slice==SLICE_TYPE_P ? 'P' : 'B', buf );
1884 if( h->param.analyse.b_ssim )
1886 x264_log( h, X264_LOG_INFO,
1887 "SSIM Mean Y:%.7f\n",
1888 SUM3( h->stat.f_ssim_mean_y ) / i_count );
1890 if( h->param.analyse.b_psnr )
1892 x264_log( h, X264_LOG_INFO,
1893 "PSNR Mean Y:%6.3f U:%6.3f V:%6.3f Avg:%6.3f Global:%6.3f kb/s:%.2f\n",
1894 SUM3( h->stat.f_psnr_mean_y ) / i_count,
1895 SUM3( h->stat.f_psnr_mean_u ) / i_count,
1896 SUM3( h->stat.f_psnr_mean_v ) / i_count,
1897 SUM3( h->stat.f_psnr_average ) / i_count,
1898 x264_psnr( SUM3( h->stat.i_sqe_global ), i_count * i_yuv_size ),
1902 x264_log( h, X264_LOG_INFO, "kb/s:%.1f\n", f_bitrate );
1906 for( i = 0; i < X264_BFRAME_MAX + 3; i++ )
1908 if( h->frames.current[i] ) x264_frame_delete( h->frames.current[i] );
1909 if( h->frames.next[i] ) x264_frame_delete( h->frames.next[i] );
1910 if( h->frames.unused[i] ) x264_frame_delete( h->frames.unused[i] );
1913 for( i = 0; i < h->frames.i_max_dpb; i++ )
1915 x264_frame_delete( h->frames.reference[i] );
1919 x264_ratecontrol_delete( h );
1922 if( h->param.rc.psz_stat_out )
1923 free( h->param.rc.psz_stat_out );
1924 if( h->param.rc.psz_stat_in )
1925 free( h->param.rc.psz_stat_in );
1926 if( h->param.rc.psz_rc_eq )
1927 free( h->param.rc.psz_rc_eq );
1929 x264_cqm_delete( h );
1930 x264_macroblock_cache_end( h );
1931 x264_free( h->out.p_bitstream );
1932 for( i = 1; i < h->param.i_threads; i++ )
1933 x264_free( h->thread[i] );