1 /*****************************************************************************
3 *****************************************************************************
4 * Copyright (C) 2003-2008 x264 project
6 * Authors: Laurent Aimar <fenrir@via.ecp.fr>
7 * Loren Merritt <lorenm@u.washington.edu>
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., 51 Franklin Street, Fifth Floor, Boston, MA 02111, USA.
22 *****************************************************************************/
26 #include "common/common.h"
27 #include "common/cpu.h"
31 #include "ratecontrol.h"
32 #include "macroblock.h"
35 #include "common/visualize.h"
38 //#define DEBUG_MB_TYPE
40 #define NALU_OVERHEAD 5 // startcode + NAL type costs 5 bytes per frame
42 #define bs_write_ue bs_write_ue_big
44 static void x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
45 x264_nal_t **pp_nal, int *pi_nal,
46 x264_picture_t *pic_out );
48 /****************************************************************************
50 ******************************* x264 libs **********************************
52 ****************************************************************************/
53 static float x264_psnr( int64_t i_sqe, int64_t i_size )
55 double f_mse = (double)i_sqe / ((double)65025.0 * (double)i_size);
56 if( f_mse <= 0.0000000001 ) /* Max 100dB */
59 return (float)(-10.0 * log( f_mse ) / log( 10.0 ));
62 static void x264_frame_dump( x264_t *h )
64 FILE *f = fopen( h->param.psz_dump_yuv, "r+b" );
68 /* Write the frame in display order */
69 fseek( f, h->fdec->i_frame * h->param.i_height * h->param.i_width * 3/2, SEEK_SET );
70 for( i = 0; i < h->fdec->i_plane; i++ )
71 for( y = 0; y < h->param.i_height >> !!i; y++ )
72 fwrite( &h->fdec->plane[i][y*h->fdec->i_stride[i]], 1, h->param.i_width >> !!i, f );
77 /* Fill "default" values */
78 static void x264_slice_header_init( x264_t *h, x264_slice_header_t *sh,
79 x264_sps_t *sps, x264_pps_t *pps,
80 int i_idr_pic_id, int i_frame, int i_qp )
82 x264_param_t *param = &h->param;
85 /* First we fill all field */
90 sh->i_last_mb = h->sps->i_mb_width * h->sps->i_mb_height;
91 sh->i_pps_id = pps->i_id;
93 sh->i_frame_num = i_frame;
95 sh->b_mbaff = h->param.b_interlaced;
96 sh->b_field_pic = 0; /* no field support for now */
97 sh->b_bottom_field = 0; /* not yet used */
99 sh->i_idr_pic_id = i_idr_pic_id;
101 /* poc stuff, fixed later */
103 sh->i_delta_poc_bottom = 0;
104 sh->i_delta_poc[0] = 0;
105 sh->i_delta_poc[1] = 0;
107 sh->i_redundant_pic_cnt = 0;
109 if( !h->mb.b_direct_auto_read )
111 if( h->mb.b_direct_auto_write )
112 sh->b_direct_spatial_mv_pred = ( h->stat.i_direct_score[1] > h->stat.i_direct_score[0] );
114 sh->b_direct_spatial_mv_pred = ( param->analyse.i_direct_mv_pred == X264_DIRECT_PRED_SPATIAL );
116 /* else b_direct_spatial_mv_pred was read from the 2pass statsfile */
118 sh->b_num_ref_idx_override = 0;
119 sh->i_num_ref_idx_l0_active = 1;
120 sh->i_num_ref_idx_l1_active = 1;
122 sh->b_ref_pic_list_reordering_l0 = h->b_ref_reorder[0];
123 sh->b_ref_pic_list_reordering_l1 = h->b_ref_reorder[1];
125 /* If the ref list isn't in the default order, construct reordering header */
126 /* List1 reordering isn't needed yet */
127 if( sh->b_ref_pic_list_reordering_l0 )
129 int pred_frame_num = i_frame;
130 for( i = 0; i < h->i_ref0; i++ )
132 int diff = h->fref0[i]->i_frame_num - pred_frame_num;
134 x264_log( h, X264_LOG_ERROR, "diff frame num == 0\n" );
135 sh->ref_pic_list_order[0][i].idc = ( diff > 0 );
136 sh->ref_pic_list_order[0][i].arg = abs( diff ) - 1;
137 pred_frame_num = h->fref0[i]->i_frame_num;
141 sh->i_cabac_init_idc = param->i_cabac_init_idc;
144 sh->i_qp_delta = i_qp - pps->i_pic_init_qp;
145 sh->b_sp_for_swidth = 0;
148 /* If effective qp <= 15, deblocking would have no effect anyway */
149 if( param->b_deblocking_filter
150 && ( h->mb.b_variable_qp
151 || 15 < i_qp + 2 * X264_MIN(param->i_deblocking_filter_alphac0, param->i_deblocking_filter_beta) ) )
153 sh->i_disable_deblocking_filter_idc = 0;
157 sh->i_disable_deblocking_filter_idc = 1;
159 sh->i_alpha_c0_offset = param->i_deblocking_filter_alphac0 << 1;
160 sh->i_beta_offset = param->i_deblocking_filter_beta << 1;
163 static void x264_slice_header_write( bs_t *s, x264_slice_header_t *sh, int i_nal_ref_idc )
169 assert( sh->i_first_mb % (2*sh->sps->i_mb_width) == 0 );
170 bs_write_ue( s, sh->i_first_mb >> 1 );
173 bs_write_ue( s, sh->i_first_mb );
175 bs_write_ue( s, sh->i_type + 5 ); /* same type things */
176 bs_write_ue( s, sh->i_pps_id );
177 bs_write( s, sh->sps->i_log2_max_frame_num, sh->i_frame_num );
179 if( !sh->sps->b_frame_mbs_only )
181 bs_write1( s, sh->b_field_pic );
182 if ( sh->b_field_pic )
183 bs_write1( s, sh->b_bottom_field );
186 if( sh->i_idr_pic_id >= 0 ) /* NAL IDR */
188 bs_write_ue( s, sh->i_idr_pic_id );
191 if( sh->sps->i_poc_type == 0 )
193 bs_write( s, sh->sps->i_log2_max_poc_lsb, sh->i_poc_lsb );
194 if( sh->pps->b_pic_order && !sh->b_field_pic )
196 bs_write_se( s, sh->i_delta_poc_bottom );
199 else if( sh->sps->i_poc_type == 1 && !sh->sps->b_delta_pic_order_always_zero )
201 bs_write_se( s, sh->i_delta_poc[0] );
202 if( sh->pps->b_pic_order && !sh->b_field_pic )
204 bs_write_se( s, sh->i_delta_poc[1] );
208 if( sh->pps->b_redundant_pic_cnt )
210 bs_write_ue( s, sh->i_redundant_pic_cnt );
213 if( sh->i_type == SLICE_TYPE_B )
215 bs_write1( s, sh->b_direct_spatial_mv_pred );
217 if( sh->i_type == SLICE_TYPE_P || sh->i_type == SLICE_TYPE_SP || sh->i_type == SLICE_TYPE_B )
219 bs_write1( s, sh->b_num_ref_idx_override );
220 if( sh->b_num_ref_idx_override )
222 bs_write_ue( s, sh->i_num_ref_idx_l0_active - 1 );
223 if( sh->i_type == SLICE_TYPE_B )
225 bs_write_ue( s, sh->i_num_ref_idx_l1_active - 1 );
230 /* ref pic list reordering */
231 if( sh->i_type != SLICE_TYPE_I )
233 bs_write1( s, sh->b_ref_pic_list_reordering_l0 );
234 if( sh->b_ref_pic_list_reordering_l0 )
236 for( i = 0; i < sh->i_num_ref_idx_l0_active; i++ )
238 bs_write_ue( s, sh->ref_pic_list_order[0][i].idc );
239 bs_write_ue( s, sh->ref_pic_list_order[0][i].arg );
245 if( sh->i_type == SLICE_TYPE_B )
247 bs_write1( s, sh->b_ref_pic_list_reordering_l1 );
248 if( sh->b_ref_pic_list_reordering_l1 )
250 for( i = 0; i < sh->i_num_ref_idx_l1_active; i++ )
252 bs_write_ue( s, sh->ref_pic_list_order[1][i].idc );
253 bs_write_ue( s, sh->ref_pic_list_order[1][i].arg );
259 if( ( sh->pps->b_weighted_pred && ( sh->i_type == SLICE_TYPE_P || sh->i_type == SLICE_TYPE_SP ) ) ||
260 ( sh->pps->b_weighted_bipred == 1 && sh->i_type == SLICE_TYPE_B ) )
265 if( i_nal_ref_idc != 0 )
267 if( sh->i_idr_pic_id >= 0 )
269 bs_write1( s, 0 ); /* no output of prior pics flag */
270 bs_write1( s, 0 ); /* long term reference flag */
274 bs_write1( s, 0 ); /* adaptive_ref_pic_marking_mode_flag */
278 if( sh->pps->b_cabac && sh->i_type != SLICE_TYPE_I )
280 bs_write_ue( s, sh->i_cabac_init_idc );
282 bs_write_se( s, sh->i_qp_delta ); /* slice qp delta */
284 if( sh->pps->b_deblocking_filter_control )
286 bs_write_ue( s, sh->i_disable_deblocking_filter_idc );
287 if( sh->i_disable_deblocking_filter_idc != 1 )
289 bs_write_se( s, sh->i_alpha_c0_offset >> 1 );
290 bs_write_se( s, sh->i_beta_offset >> 1 );
295 /* If we are within a reasonable distance of the end of the memory allocated for the bitstream, */
296 /* reallocate, adding an arbitrary amount of space (100 kilobytes). */
297 static void x264_bitstream_check_buffer( x264_t *h )
299 if( ( h->param.b_cabac && (h->cabac.p_end - h->cabac.p < 2500) )
300 || ( h->out.bs.p_end - h->out.bs.p < 2500 ) )
302 uint8_t *bs_bak = h->out.p_bitstream;
306 h->out.i_bitstream += 100000;
307 h->out.p_bitstream = x264_realloc( h->out.p_bitstream, h->out.i_bitstream );
308 delta = h->out.p_bitstream - bs_bak;
310 h->out.bs.p_start += delta;
311 h->out.bs.p += delta;
312 h->out.bs.p_end = h->out.p_bitstream + h->out.i_bitstream;
314 h->cabac.p_start += delta;
316 h->cabac.p_end = h->out.p_bitstream + h->out.i_bitstream;
318 for( i = 0; i <= h->out.i_nal; i++ )
319 h->out.nal[i].p_payload += delta;
323 /****************************************************************************
325 ****************************************************************************
326 ****************************** External API*********************************
327 ****************************************************************************
329 ****************************************************************************/
331 static int x264_validate_parameters( x264_t *h )
334 if( !(x264_cpu_detect() & X264_CPU_MMXEXT) )
336 x264_log( h, X264_LOG_ERROR, "your cpu does not support MMXEXT, but x264 was compiled with asm support\n");
337 x264_log( h, X264_LOG_ERROR, "to run x264, recompile without asm support (configure --disable-asm)\n");
341 if( h->param.i_width <= 0 || h->param.i_height <= 0 )
343 x264_log( h, X264_LOG_ERROR, "invalid width x height (%dx%d)\n",
344 h->param.i_width, h->param.i_height );
348 if( h->param.i_width % 2 || h->param.i_height % 2 )
350 x264_log( h, X264_LOG_ERROR, "width or height not divisible by 2 (%dx%d)\n",
351 h->param.i_width, h->param.i_height );
354 if( h->param.i_csp != X264_CSP_I420 )
356 x264_log( h, X264_LOG_ERROR, "invalid CSP (only I420 supported)\n" );
360 if( h->param.i_threads == 0 )
361 h->param.i_threads = x264_cpu_num_processors() * 3/2;
362 h->param.i_threads = x264_clip3( h->param.i_threads, 1, X264_THREAD_MAX );
363 if( h->param.i_threads > 1 )
366 x264_log( h, X264_LOG_WARNING, "not compiled with pthread support!\n");
367 h->param.i_threads = 1;
369 if( h->param.i_scenecut_threshold >= 0 )
370 h->param.b_pre_scenecut = 1;
374 if( h->param.b_interlaced )
376 if( h->param.analyse.i_me_method >= X264_ME_ESA )
378 x264_log( h, X264_LOG_WARNING, "interlace + me=esa is not implemented\n" );
379 h->param.analyse.i_me_method = X264_ME_UMH;
381 if( h->param.analyse.i_direct_mv_pred > X264_DIRECT_PRED_SPATIAL )
383 x264_log( h, X264_LOG_WARNING, "interlace + direct=temporal is not implemented\n" );
384 h->param.analyse.i_direct_mv_pred = X264_DIRECT_PRED_SPATIAL;
388 if( h->param.rc.i_rc_method < 0 || h->param.rc.i_rc_method > 2 )
390 x264_log( h, X264_LOG_ERROR, "no ratecontrol method specified\n" );
393 h->param.rc.f_rf_constant = x264_clip3f( h->param.rc.f_rf_constant, 0, 51 );
394 h->param.rc.i_qp_constant = x264_clip3( h->param.rc.i_qp_constant, 0, 51 );
395 if( h->param.rc.i_rc_method == X264_RC_CRF )
396 h->param.rc.i_qp_constant = h->param.rc.f_rf_constant;
397 if( (h->param.rc.i_rc_method == X264_RC_CQP || h->param.rc.i_rc_method == X264_RC_CRF)
398 && h->param.rc.i_qp_constant == 0 )
400 h->mb.b_lossless = 1;
401 h->param.i_cqm_preset = X264_CQM_FLAT;
402 h->param.psz_cqm_file = NULL;
403 h->param.rc.i_rc_method = X264_RC_CQP;
404 h->param.rc.f_ip_factor = 1;
405 h->param.rc.f_pb_factor = 1;
406 h->param.analyse.b_psnr = 0;
407 h->param.analyse.b_ssim = 0;
408 h->param.analyse.i_chroma_qp_offset = 0;
409 h->param.analyse.i_trellis = 0;
410 h->param.analyse.b_fast_pskip = 0;
411 h->param.analyse.i_noise_reduction = 0;
412 h->param.analyse.f_psy_rd = 0;
413 /* 8x8dct is not useful at all in CAVLC lossless */
414 if( !h->param.b_cabac )
415 h->param.analyse.b_transform_8x8 = 0;
417 if( h->param.rc.i_rc_method == X264_RC_CQP )
419 float qp_p = h->param.rc.i_qp_constant;
420 float qp_i = qp_p - 6*log(h->param.rc.f_ip_factor)/log(2);
421 float qp_b = qp_p + 6*log(h->param.rc.f_pb_factor)/log(2);
422 h->param.rc.i_qp_min = x264_clip3( (int)(X264_MIN3( qp_p, qp_i, qp_b )), 0, 51 );
423 h->param.rc.i_qp_max = x264_clip3( (int)(X264_MAX3( qp_p, qp_i, qp_b ) + .999), 0, 51 );
424 h->param.rc.i_aq_mode = 0;
426 h->param.rc.i_qp_max = x264_clip3( h->param.rc.i_qp_max, 0, 51 );
427 h->param.rc.i_qp_min = x264_clip3( h->param.rc.i_qp_min, 0, h->param.rc.i_qp_max );
429 if( ( h->param.i_width % 16 || h->param.i_height % 16 )
430 && h->param.i_height != 1080 && !h->mb.b_lossless )
432 // There's nothing special about 1080 in that the warning still applies to it,
433 // but chances are the user can't help it if his content is already 1080p,
434 // so there's no point in warning in that case.
435 x264_log( h, X264_LOG_WARNING,
436 "width or height not divisible by 16 (%dx%d), compression will suffer.\n",
437 h->param.i_width, h->param.i_height );
440 h->param.i_frame_reference = x264_clip3( h->param.i_frame_reference, 1, 16 );
441 if( h->param.i_keyint_max <= 0 )
442 h->param.i_keyint_max = 1;
443 h->param.i_keyint_min = x264_clip3( h->param.i_keyint_min, 1, h->param.i_keyint_max/2+1 );
444 if( !h->param.analyse.i_subpel_refine && h->param.analyse.i_direct_mv_pred > X264_DIRECT_PRED_SPATIAL )
446 x264_log( h, X264_LOG_WARNING, "subme=0 + direct=temporal is not supported\n" );
447 h->param.analyse.i_direct_mv_pred = X264_DIRECT_PRED_SPATIAL;
449 h->param.i_bframe = x264_clip3( h->param.i_bframe, 0, X264_BFRAME_MAX );
450 h->param.i_bframe_bias = x264_clip3( h->param.i_bframe_bias, -90, 100 );
451 h->param.b_bframe_pyramid = h->param.b_bframe_pyramid && h->param.i_bframe > 1;
452 if( !h->param.i_bframe )
453 h->param.i_bframe_adaptive = X264_B_ADAPT_NONE;
454 h->param.analyse.b_weighted_bipred = h->param.analyse.b_weighted_bipred && h->param.i_bframe > 0;
455 h->mb.b_direct_auto_write = h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO
457 && ( h->param.rc.b_stat_write || !h->param.rc.b_stat_read );
458 if( h->param.i_scenecut_threshold < 0 )
459 h->param.b_pre_scenecut = 0;
461 h->param.i_deblocking_filter_alphac0 = x264_clip3( h->param.i_deblocking_filter_alphac0, -6, 6 );
462 h->param.i_deblocking_filter_beta = x264_clip3( h->param.i_deblocking_filter_beta, -6, 6 );
463 h->param.analyse.i_luma_deadzone[0] = x264_clip3( h->param.analyse.i_luma_deadzone[0], 0, 32 );
464 h->param.analyse.i_luma_deadzone[1] = x264_clip3( h->param.analyse.i_luma_deadzone[1], 0, 32 );
466 h->param.i_cabac_init_idc = x264_clip3( h->param.i_cabac_init_idc, 0, 2 );
468 if( h->param.i_cqm_preset < X264_CQM_FLAT || h->param.i_cqm_preset > X264_CQM_CUSTOM )
469 h->param.i_cqm_preset = X264_CQM_FLAT;
471 if( h->param.analyse.i_me_method < X264_ME_DIA ||
472 h->param.analyse.i_me_method > X264_ME_TESA )
473 h->param.analyse.i_me_method = X264_ME_HEX;
474 if( h->param.analyse.i_me_range < 4 )
475 h->param.analyse.i_me_range = 4;
476 if( h->param.analyse.i_me_range > 16 && h->param.analyse.i_me_method <= X264_ME_HEX )
477 h->param.analyse.i_me_range = 16;
478 if( h->param.analyse.i_me_method == X264_ME_TESA &&
479 (h->mb.b_lossless || h->param.analyse.i_subpel_refine <= 1) )
480 h->param.analyse.i_me_method = X264_ME_ESA;
481 h->param.analyse.i_subpel_refine = x264_clip3( h->param.analyse.i_subpel_refine, 0, 9 );
482 h->param.analyse.b_mixed_references = h->param.analyse.b_mixed_references && h->param.i_frame_reference > 1;
483 h->param.analyse.inter &= X264_ANALYSE_PSUB16x16|X264_ANALYSE_PSUB8x8|X264_ANALYSE_BSUB16x16|
484 X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
485 h->param.analyse.intra &= X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
486 if( !(h->param.analyse.inter & X264_ANALYSE_PSUB16x16) )
487 h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
488 if( !h->param.analyse.b_transform_8x8 )
490 h->param.analyse.inter &= ~X264_ANALYSE_I8x8;
491 h->param.analyse.intra &= ~X264_ANALYSE_I8x8;
493 h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -12, 12);
494 if( !h->param.b_cabac )
495 h->param.analyse.i_trellis = 0;
496 h->param.analyse.i_trellis = x264_clip3( h->param.analyse.i_trellis, 0, 2 );
497 if( !h->param.analyse.i_trellis )
498 h->param.analyse.f_psy_trellis = 0;
499 h->param.analyse.f_psy_rd = x264_clip3f( h->param.analyse.f_psy_rd, 0, 10 );
500 h->param.analyse.f_psy_trellis = x264_clip3f( h->param.analyse.f_psy_trellis, 0, 10 );
501 if( h->param.analyse.i_subpel_refine < 6 )
502 h->param.analyse.f_psy_rd = 0;
503 h->mb.i_psy_rd = FIX8( h->param.analyse.f_psy_rd );
504 /* Psy RDO increases overall quantizers to improve the quality of luma--this indirectly hurts chroma quality */
505 /* so we lower the chroma QP offset to compensate */
506 /* This can be triggered repeatedly on multiple calls to parameter_validate, but since encoding
507 * uses the pps chroma qp offset not the param chroma qp offset, this is not a problem. */
509 h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_rd < 0.25 ? 1 : 2;
510 h->mb.i_psy_trellis = FIX8( h->param.analyse.f_psy_trellis / 4 );
511 /* Psy trellis has a similar effect. */
512 if( h->mb.i_psy_trellis )
513 h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_trellis < 0.25 ? 1 : 2;
515 h->mb.i_psy_trellis = 0;
516 h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -12, 12);
517 h->param.rc.i_aq_mode = x264_clip3( h->param.rc.i_aq_mode, 0, 1 );
518 h->param.rc.f_aq_strength = x264_clip3f( h->param.rc.f_aq_strength, 0, 3 );
519 if( h->param.rc.f_aq_strength == 0 )
520 h->param.rc.i_aq_mode = 0;
521 h->param.analyse.i_noise_reduction = x264_clip3( h->param.analyse.i_noise_reduction, 0, 1<<16 );
524 const x264_level_t *l = x264_levels;
525 if( h->param.i_level_idc < 0 )
527 if( h->param.rc.i_rc_method == X264_RC_ABR && h->param.rc.i_vbv_buffer_size <= 0 )
528 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate * 2;
529 h->sps = h->sps_array;
530 x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
531 do h->param.i_level_idc = l->level_idc;
532 while( l[1].level_idc && x264_validate_levels( h, 0 ) && l++ );
533 if( h->param.rc.i_vbv_buffer_size <= 0 )
534 h->param.rc.i_vbv_max_bitrate = 0;
538 while( l->level_idc && l->level_idc != h->param.i_level_idc )
540 if( l->level_idc == 0 )
542 x264_log( h, X264_LOG_ERROR, "invalid level_idc: %d\n", h->param.i_level_idc );
546 if( h->param.analyse.i_mv_range <= 0 )
547 h->param.analyse.i_mv_range = l->mv_range >> h->param.b_interlaced;
549 h->param.analyse.i_mv_range = x264_clip3(h->param.analyse.i_mv_range, 32, 512 >> h->param.b_interlaced);
550 if( h->param.analyse.i_direct_8x8_inference < 0 )
551 h->param.analyse.i_direct_8x8_inference = l->direct8x8;
554 if( h->param.i_threads > 1 )
556 int r = h->param.analyse.i_mv_range_thread;
560 // half of the available space is reserved and divided evenly among the threads,
561 // the rest is allocated to whichever thread is far enough ahead to use it.
562 // reserving more space increases quality for some videos, but costs more time
563 // in thread synchronization.
564 int max_range = (h->param.i_height + X264_THREAD_HEIGHT) / h->param.i_threads - X264_THREAD_HEIGHT;
567 r = X264_MAX( r, h->param.analyse.i_me_range );
568 r = X264_MIN( r, h->param.analyse.i_mv_range );
569 // round up to use the whole mb row
570 r2 = (r & ~15) + ((-X264_THREAD_HEIGHT) & 15);
573 x264_log( h, X264_LOG_DEBUG, "using mv_range_thread = %d\n", r2 );
574 h->param.analyse.i_mv_range_thread = r2;
577 if( h->param.rc.f_qblur < 0 )
578 h->param.rc.f_qblur = 0;
579 if( h->param.rc.f_complexity_blur < 0 )
580 h->param.rc.f_complexity_blur = 0;
582 h->param.i_sps_id &= 31;
584 if( h->param.i_log_level < X264_LOG_INFO )
586 h->param.analyse.b_psnr = 0;
587 h->param.analyse.b_ssim = 0;
590 /* ensure the booleans are 0 or 1 so they can be used in math */
591 #define BOOLIFY(x) h->param.x = !!h->param.x
593 BOOLIFY( b_deblocking_filter );
594 BOOLIFY( b_interlaced );
595 BOOLIFY( analyse.b_transform_8x8 );
596 BOOLIFY( analyse.i_direct_8x8_inference );
597 BOOLIFY( analyse.b_chroma_me );
598 BOOLIFY( analyse.b_fast_pskip );
599 BOOLIFY( rc.b_stat_write );
600 BOOLIFY( rc.b_stat_read );
606 static void mbcmp_init( x264_t *h )
608 int satd = !h->mb.b_lossless && h->param.analyse.i_subpel_refine > 1;
609 memcpy( h->pixf.mbcmp, satd ? h->pixf.satd : h->pixf.sad_aligned, sizeof(h->pixf.mbcmp) );
610 memcpy( h->pixf.mbcmp_unaligned, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.mbcmp_unaligned) );
611 h->pixf.intra_mbcmp_x3_16x16 = satd ? h->pixf.intra_satd_x3_16x16 : h->pixf.intra_sad_x3_16x16;
612 satd &= h->param.analyse.i_me_method == X264_ME_TESA;
613 memcpy( h->pixf.fpelcmp, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.fpelcmp) );
614 memcpy( h->pixf.fpelcmp_x3, satd ? h->pixf.satd_x3 : h->pixf.sad_x3, sizeof(h->pixf.fpelcmp_x3) );
615 memcpy( h->pixf.fpelcmp_x4, satd ? h->pixf.satd_x4 : h->pixf.sad_x4, sizeof(h->pixf.fpelcmp_x4) );
618 /****************************************************************************
620 ****************************************************************************/
621 x264_t *x264_encoder_open ( x264_param_t *param )
623 x264_t *h = x264_malloc( sizeof( x264_t ) );
627 memset( h, 0, sizeof( x264_t ) );
629 /* Create a copy of param */
630 memcpy( &h->param, param, sizeof( x264_param_t ) );
632 if( x264_validate_parameters( h ) < 0 )
638 if( h->param.psz_cqm_file )
639 if( x264_cqm_parse_file( h, h->param.psz_cqm_file ) < 0 )
645 if( h->param.rc.psz_stat_out )
646 h->param.rc.psz_stat_out = strdup( h->param.rc.psz_stat_out );
647 if( h->param.rc.psz_stat_in )
648 h->param.rc.psz_stat_in = strdup( h->param.rc.psz_stat_in );
651 if( h->param.vui.i_sar_width > 0 && h->param.vui.i_sar_height > 0 )
653 int i_w = param->vui.i_sar_width;
654 int i_h = param->vui.i_sar_height;
656 x264_reduce_fraction( &i_w, &i_h );
658 while( i_w > 65535 || i_h > 65535 )
664 h->param.vui.i_sar_width = 0;
665 h->param.vui.i_sar_height = 0;
666 if( i_w == 0 || i_h == 0 )
668 x264_log( h, X264_LOG_WARNING, "cannot create valid sample aspect ratio\n" );
672 x264_log( h, X264_LOG_INFO, "using SAR=%d/%d\n", i_w, i_h );
673 h->param.vui.i_sar_width = i_w;
674 h->param.vui.i_sar_height = i_h;
678 x264_reduce_fraction( &h->param.i_fps_num, &h->param.i_fps_den );
685 h->sps = &h->sps_array[0];
686 x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
688 h->pps = &h->pps_array[0];
689 x264_pps_init( h->pps, h->param.i_sps_id, &h->param, h->sps);
691 x264_validate_levels( h, 1 );
693 if( x264_cqm_init( h ) < 0 )
699 h->mb.i_mb_count = h->sps->i_mb_width * h->sps->i_mb_height;
702 if( h->param.i_bframe_adaptive == X264_B_ADAPT_TRELLIS )
703 h->frames.i_delay = X264_MAX(h->param.i_bframe,3)*4 + h->param.i_threads - 1;
705 h->frames.i_delay = h->param.i_bframe + h->param.i_threads - 1;
706 h->frames.i_max_ref0 = h->param.i_frame_reference;
707 h->frames.i_max_ref1 = h->sps->vui.i_num_reorder_frames;
708 h->frames.i_max_dpb = h->sps->vui.i_max_dec_frame_buffering;
709 h->frames.b_have_lowres = !h->param.rc.b_stat_read
710 && ( h->param.rc.i_rc_method == X264_RC_ABR
711 || h->param.rc.i_rc_method == X264_RC_CRF
712 || h->param.i_bframe_adaptive
713 || h->param.b_pre_scenecut );
714 h->frames.b_have_lowres |= (h->param.rc.b_stat_read && h->param.rc.i_vbv_buffer_size > 0);
716 h->frames.i_last_idr = - h->param.i_keyint_max;
717 h->frames.i_input = 0;
718 h->frames.last_nonb = NULL;
723 h->chroma_qp_table = i_chroma_qp_table + 12 + h->pps->i_chroma_qp_index_offset;
727 /* init CPU functions */
728 x264_predict_16x16_init( h->param.cpu, h->predict_16x16 );
729 x264_predict_8x8c_init( h->param.cpu, h->predict_8x8c );
730 x264_predict_8x8_init( h->param.cpu, h->predict_8x8 );
731 x264_predict_4x4_init( h->param.cpu, h->predict_4x4 );
733 x264_pixel_init( h->param.cpu, &h->pixf );
734 x264_dct_init( h->param.cpu, &h->dctf );
735 x264_zigzag_init( h->param.cpu, &h->zigzagf, h->param.b_interlaced );
736 x264_mc_init( h->param.cpu, &h->mc );
737 x264_quant_init( h, h->param.cpu, &h->quantf );
738 x264_deblock_init( h->param.cpu, &h->loopf );
739 x264_dct_init_weights();
743 p = buf + sprintf( buf, "using cpu capabilities:" );
744 for( i=0; x264_cpu_names[i].flags; i++ )
746 if( !strcmp(x264_cpu_names[i].name, "SSE2")
747 && param->cpu & (X264_CPU_SSE2_IS_FAST|X264_CPU_SSE2_IS_SLOW) )
749 if( !strcmp(x264_cpu_names[i].name, "SSE3")
750 && (param->cpu & X264_CPU_SSSE3 || !(param->cpu & X264_CPU_CACHELINE_64)) )
752 if( !strcmp(x264_cpu_names[i].name, "SSE4.1")
753 && (param->cpu & X264_CPU_SSE42) )
755 if( (param->cpu & x264_cpu_names[i].flags) == x264_cpu_names[i].flags
756 && (!i || x264_cpu_names[i].flags != x264_cpu_names[i-1].flags) )
757 p += sprintf( p, " %s", x264_cpu_names[i].name );
760 p += sprintf( p, " none!" );
761 x264_log( h, X264_LOG_INFO, "%s\n", buf );
764 h->out.i_bitstream = X264_MAX( 1000000, h->param.i_width * h->param.i_height * 4
765 * ( h->param.rc.i_rc_method == X264_RC_ABR ? pow( 0.95, h->param.rc.i_qp_min )
766 : pow( 0.95, h->param.rc.i_qp_constant ) * X264_MAX( 1, h->param.rc.f_ip_factor )));
770 for( i = 1; i < h->param.i_threads; i++ )
771 h->thread[i] = x264_malloc( sizeof(x264_t) );
773 for( i = 0; i < h->param.i_threads; i++ )
777 h->thread[i]->fdec = x264_frame_pop_unused( h );
778 h->thread[i]->out.p_bitstream = x264_malloc( h->out.i_bitstream );
779 if( x264_macroblock_cache_init( h->thread[i] ) < 0 )
783 if( x264_ratecontrol_new( h ) < 0 )
786 if( h->param.psz_dump_yuv )
788 /* create or truncate the reconstructed video file */
789 FILE *f = fopen( h->param.psz_dump_yuv, "w" );
794 x264_log( h, X264_LOG_ERROR, "can't write to fdec.yuv\n" );
800 x264_log( h, X264_LOG_INFO, "profile %s, level %d.%d\n",
801 h->sps->i_profile_idc == PROFILE_BASELINE ? "Baseline" :
802 h->sps->i_profile_idc == PROFILE_MAIN ? "Main" :
803 h->sps->i_profile_idc == PROFILE_HIGH ? "High" :
804 "High 4:4:4 Predictive", h->sps->i_level_idc/10, h->sps->i_level_idc%10 );
809 /****************************************************************************
810 * x264_encoder_reconfig:
811 ****************************************************************************/
812 int x264_encoder_reconfig( x264_t *h, x264_param_t *param )
814 #define COPY(var) h->param.var = param->var
815 COPY( i_frame_reference ); // but never uses more refs than initially specified
816 COPY( i_bframe_bias );
817 if( h->param.i_scenecut_threshold >= 0 && param->i_scenecut_threshold >= 0 )
818 COPY( i_scenecut_threshold ); // can't turn it on or off, only vary the threshold
819 COPY( b_deblocking_filter );
820 COPY( i_deblocking_filter_alphac0 );
821 COPY( i_deblocking_filter_beta );
822 COPY( analyse.intra );
823 COPY( analyse.inter );
824 COPY( analyse.i_direct_mv_pred );
825 COPY( analyse.i_me_range );
826 COPY( analyse.i_noise_reduction );
827 /* We can't switch out of subme=0 during encoding. */
828 if( h->param.analyse.i_subpel_refine )
829 COPY( analyse.i_subpel_refine );
830 COPY( analyse.i_trellis );
831 COPY( analyse.b_chroma_me );
832 COPY( analyse.b_dct_decimate );
833 COPY( analyse.b_fast_pskip );
834 COPY( analyse.b_mixed_references );
835 COPY( analyse.f_psy_rd );
836 COPY( analyse.f_psy_trellis );
837 // can only twiddle these if they were enabled to begin with:
838 if( h->param.analyse.i_me_method >= X264_ME_ESA || param->analyse.i_me_method < X264_ME_ESA )
839 COPY( analyse.i_me_method );
840 if( h->pps->b_transform_8x8_mode )
841 COPY( analyse.b_transform_8x8 );
842 if( h->frames.i_max_ref1 > 1 )
843 COPY( b_bframe_pyramid );
848 return x264_validate_parameters( h );
852 static void x264_nal_start( x264_t *h, int i_type, int i_ref_idc )
854 x264_nal_t *nal = &h->out.nal[h->out.i_nal];
856 nal->i_ref_idc = i_ref_idc;
857 nal->i_type = i_type;
860 nal->p_payload= &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8];
862 static void x264_nal_end( x264_t *h )
864 x264_nal_t *nal = &h->out.nal[h->out.i_nal];
865 nal->i_payload = &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8] - nal->p_payload;
869 /****************************************************************************
870 * x264_encoder_headers:
871 ****************************************************************************/
872 int x264_encoder_headers( x264_t *h, x264_nal_t **pp_nal, int *pi_nal )
874 /* init bitstream context */
876 bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
878 /* Put SPS and PPS */
879 if( h->i_frame == 0 )
881 /* identify ourself */
882 x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
883 x264_sei_version_write( h, &h->out.bs );
886 /* generate sequence parameters */
887 x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
888 x264_sps_write( &h->out.bs, h->sps );
891 /* generate picture parameters */
892 x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
893 x264_pps_write( &h->out.bs, h->pps );
897 *pi_nal = h->out.i_nal;
898 *pp_nal = &h->out.nal[0];
904 static inline void x264_reference_build_list( x264_t *h, int i_poc )
909 /* build ref list 0/1 */
912 for( i = 0; h->frames.reference[i]; i++ )
914 if( h->frames.reference[i]->i_poc < i_poc )
916 h->fref0[h->i_ref0++] = h->frames.reference[i];
918 else if( h->frames.reference[i]->i_poc > i_poc )
920 h->fref1[h->i_ref1++] = h->frames.reference[i];
924 /* Order ref0 from higher to lower poc */
928 for( i = 0; i < h->i_ref0 - 1; i++ )
930 if( h->fref0[i]->i_poc < h->fref0[i+1]->i_poc )
932 XCHG( x264_frame_t*, h->fref0[i], h->fref0[i+1] );
938 /* Order ref1 from lower to higher poc (bubble sort) for B-frame */
942 for( i = 0; i < h->i_ref1 - 1; i++ )
944 if( h->fref1[i]->i_poc > h->fref1[i+1]->i_poc )
946 XCHG( x264_frame_t*, h->fref1[i], h->fref1[i+1] );
953 /* In the standard, a P-frame's ref list is sorted by frame_num.
954 * We use POC, but check whether explicit reordering is needed */
955 h->b_ref_reorder[0] =
956 h->b_ref_reorder[1] = 0;
957 if( h->sh.i_type == SLICE_TYPE_P )
959 for( i = 0; i < h->i_ref0 - 1; i++ )
960 if( h->fref0[i]->i_frame_num < h->fref0[i+1]->i_frame_num )
962 h->b_ref_reorder[0] = 1;
967 h->i_ref1 = X264_MIN( h->i_ref1, h->frames.i_max_ref1 );
968 h->i_ref0 = X264_MIN( h->i_ref0, h->frames.i_max_ref0 );
969 h->i_ref0 = X264_MIN( h->i_ref0, h->param.i_frame_reference ); // if reconfig() has lowered the limit
970 assert( h->i_ref0 + h->i_ref1 <= 16 );
971 h->mb.pic.i_fref[0] = h->i_ref0;
972 h->mb.pic.i_fref[1] = h->i_ref1;
975 static void x264_fdec_filter_row( x264_t *h, int mb_y )
977 /* mb_y is the mb to be encoded next, not the mb to be filtered here */
978 int b_hpel = h->fdec->b_kept_as_ref;
979 int b_deblock = !h->sh.i_disable_deblocking_filter_idc;
980 int b_end = mb_y == h->sps->i_mb_height;
981 int min_y = mb_y - (1 << h->sh.b_mbaff);
982 int max_y = b_end ? h->sps->i_mb_height : mb_y;
983 b_deblock &= b_hpel || h->param.psz_dump_yuv;
984 if( mb_y & h->sh.b_mbaff )
992 for( j=0; j<=h->sh.b_mbaff; j++ )
995 memcpy( h->mb.intra_border_backup[j][i],
996 h->fdec->plane[i] + ((mb_y*16 >> !!i) + j - 1 - h->sh.b_mbaff) * h->fdec->i_stride[i],
997 h->sps->i_mb_width*16 >> !!i );
1004 for( y = min_y; y < max_y; y += (1 << h->sh.b_mbaff) )
1005 x264_frame_deblock_row( h, y );
1010 x264_frame_expand_border( h, h->fdec, min_y, b_end );
1011 if( h->param.analyse.i_subpel_refine )
1013 x264_frame_filter( h, h->fdec, min_y, b_end );
1014 x264_frame_expand_border_filtered( h, h->fdec, min_y, b_end );
1018 if( h->param.i_threads > 1 && h->fdec->b_kept_as_ref )
1020 x264_frame_cond_broadcast( h->fdec, mb_y*16 + (b_end ? 10000 : -(X264_THREAD_HEIGHT << h->sh.b_mbaff)) );
1023 min_y = X264_MAX( min_y*16-8, 0 );
1024 max_y = b_end ? h->param.i_height : mb_y*16-8;
1026 if( h->param.analyse.b_psnr )
1029 for( i=0; i<3; i++ )
1030 h->stat.frame.i_ssd[i] +=
1031 x264_pixel_ssd_wxh( &h->pixf,
1032 h->fdec->plane[i] + (min_y>>!!i) * h->fdec->i_stride[i], h->fdec->i_stride[i],
1033 h->fenc->plane[i] + (min_y>>!!i) * h->fenc->i_stride[i], h->fenc->i_stride[i],
1034 h->param.i_width >> !!i, (max_y-min_y) >> !!i );
1037 if( h->param.analyse.b_ssim )
1040 /* offset by 2 pixels to avoid alignment of ssim blocks with dct blocks,
1041 * and overlap by 4 */
1042 min_y += min_y == 0 ? 2 : -6;
1043 h->stat.frame.f_ssim +=
1044 x264_pixel_ssim_wxh( &h->pixf,
1045 h->fdec->plane[0] + 2+min_y*h->fdec->i_stride[0], h->fdec->i_stride[0],
1046 h->fenc->plane[0] + 2+min_y*h->fenc->i_stride[0], h->fenc->i_stride[0],
1047 h->param.i_width-2, max_y-min_y );
1051 static inline void x264_reference_update( x264_t *h )
1055 if( h->fdec->i_frame >= 0 )
1058 if( !h->fdec->b_kept_as_ref )
1060 if( h->param.i_threads > 1 )
1062 x264_frame_push_unused( h, h->fdec );
1063 h->fdec = x264_frame_pop_unused( h );
1068 /* move lowres copy of the image to the ref frame */
1069 for( i = 0; i < 4; i++)
1071 XCHG( uint8_t*, h->fdec->lowres[i], h->fenc->lowres[i] );
1072 XCHG( uint8_t*, h->fdec->buffer_lowres[i], h->fenc->buffer_lowres[i] );
1075 /* adaptive B decision needs a pointer, since it can't use the ref lists */
1076 if( h->sh.i_type != SLICE_TYPE_B )
1077 h->frames.last_nonb = h->fdec;
1079 /* move frame in the buffer */
1080 x264_frame_push( h->frames.reference, h->fdec );
1081 if( h->frames.reference[h->frames.i_max_dpb] )
1082 x264_frame_push_unused( h, x264_frame_shift( h->frames.reference ) );
1083 h->fdec = x264_frame_pop_unused( h );
1086 static inline void x264_reference_reset( x264_t *h )
1088 while( h->frames.reference[0] )
1089 x264_frame_push_unused( h, x264_frame_pop( h->frames.reference ) );
1094 static inline void x264_slice_init( x264_t *h, int i_nal_type, int i_global_qp )
1096 /* ------------------------ Create slice header ----------------------- */
1097 if( i_nal_type == NAL_SLICE_IDR )
1099 x264_slice_header_init( h, &h->sh, h->sps, h->pps, h->i_idr_pic_id, h->i_frame_num, i_global_qp );
1102 h->i_idr_pic_id = ( h->i_idr_pic_id + 1 ) % 65536;
1106 x264_slice_header_init( h, &h->sh, h->sps, h->pps, -1, h->i_frame_num, i_global_qp );
1108 /* always set the real higher num of ref frame used */
1109 h->sh.b_num_ref_idx_override = 1;
1110 h->sh.i_num_ref_idx_l0_active = h->i_ref0 <= 0 ? 1 : h->i_ref0;
1111 h->sh.i_num_ref_idx_l1_active = h->i_ref1 <= 0 ? 1 : h->i_ref1;
1114 h->fdec->i_frame_num = h->sh.i_frame_num;
1116 if( h->sps->i_poc_type == 0 )
1118 h->sh.i_poc_lsb = h->fdec->i_poc & ( (1 << h->sps->i_log2_max_poc_lsb) - 1 );
1119 h->sh.i_delta_poc_bottom = 0; /* XXX won't work for field */
1121 else if( h->sps->i_poc_type == 1 )
1123 /* FIXME TODO FIXME */
1127 /* Nothing to do ? */
1130 x264_macroblock_slice_init( h );
1133 static void x264_slice_write( x264_t *h )
1136 int mb_xy, i_mb_x, i_mb_y;
1137 int i, i_list, i_ref;
1140 memset( &h->stat.frame, 0, sizeof(h->stat.frame) );
1143 x264_nal_start( h, h->i_nal_type, h->i_nal_ref_idc );
1146 x264_slice_header_write( &h->out.bs, &h->sh, h->i_nal_ref_idc );
1147 if( h->param.b_cabac )
1149 /* alignment needed */
1150 bs_align_1( &h->out.bs );
1153 x264_cabac_context_init( &h->cabac, h->sh.i_type, h->sh.i_qp, h->sh.i_cabac_init_idc );
1154 x264_cabac_encode_init ( &h->cabac, h->out.bs.p, h->out.bs.p_end );
1156 h->mb.i_last_qp = h->sh.i_qp;
1157 h->mb.i_last_dqp = 0;
1159 i_mb_y = h->sh.i_first_mb / h->sps->i_mb_width;
1160 i_mb_x = h->sh.i_first_mb % h->sps->i_mb_width;
1163 while( (mb_xy = i_mb_x + i_mb_y * h->sps->i_mb_width) < h->sh.i_last_mb )
1165 int mb_spos = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
1168 x264_fdec_filter_row( h, i_mb_y );
1171 x264_macroblock_cache_load( h, i_mb_x, i_mb_y );
1173 /* analyse parameters
1174 * Slice I: choose I_4x4 or I_16x16 mode
1175 * Slice P: choose between using P mode or intra (4x4 or 16x16)
1177 x264_macroblock_analyse( h );
1179 /* encode this macroblock -> be careful it can change the mb type to P_SKIP if needed */
1180 x264_macroblock_encode( h );
1182 x264_bitstream_check_buffer( h );
1184 if( h->param.b_cabac )
1186 if( mb_xy > h->sh.i_first_mb && !(h->sh.b_mbaff && (i_mb_y&1)) )
1187 x264_cabac_encode_terminal( &h->cabac );
1189 if( IS_SKIP( h->mb.i_type ) )
1190 x264_cabac_mb_skip( h, 1 );
1193 if( h->sh.i_type != SLICE_TYPE_I )
1194 x264_cabac_mb_skip( h, 0 );
1195 x264_macroblock_write_cabac( h, &h->cabac );
1200 if( IS_SKIP( h->mb.i_type ) )
1204 if( h->sh.i_type != SLICE_TYPE_I )
1206 bs_write_ue( &h->out.bs, i_skip ); /* skip run */
1209 x264_macroblock_write_cavlc( h, &h->out.bs );
1214 if( h->param.b_visualize )
1215 x264_visualize_mb( h );
1219 x264_macroblock_cache_save( h );
1221 /* accumulate mb stats */
1222 h->stat.frame.i_mb_count[h->mb.i_type]++;
1223 if( !IS_SKIP(h->mb.i_type) && !IS_INTRA(h->mb.i_type) && !IS_DIRECT(h->mb.i_type) )
1225 if( h->mb.i_partition != D_8x8 )
1226 h->stat.frame.i_mb_partition[h->mb.i_partition] += 4;
1228 for( i = 0; i < 4; i++ )
1229 h->stat.frame.i_mb_partition[h->mb.i_sub_partition[i]] ++;
1230 if( h->param.i_frame_reference > 1 )
1231 for( i_list = 0; i_list <= (h->sh.i_type == SLICE_TYPE_B); i_list++ )
1232 for( i = 0; i < 4; i++ )
1234 i_ref = h->mb.cache.ref[i_list][ x264_scan8[4*i] ];
1236 h->stat.frame.i_mb_count_ref[i_list][i_ref] ++;
1239 if( h->mb.i_cbp_luma && !IS_INTRA(h->mb.i_type) )
1241 h->stat.frame.i_mb_count_8x8dct[0] ++;
1242 h->stat.frame.i_mb_count_8x8dct[1] += h->mb.b_transform_8x8;
1245 x264_ratecontrol_mb( h, bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac) - mb_spos );
1249 i_mb_x += i_mb_y & 1;
1250 i_mb_y ^= i_mb_x < h->sps->i_mb_width;
1254 if(i_mb_x == h->sps->i_mb_width)
1261 if( h->param.b_cabac )
1263 x264_cabac_encode_flush( h, &h->cabac );
1264 h->out.bs.p = h->cabac.p;
1269 bs_write_ue( &h->out.bs, i_skip ); /* last skip run */
1270 /* rbsp_slice_trailing_bits */
1271 bs_rbsp_trailing( &h->out.bs );
1276 x264_fdec_filter_row( h, h->sps->i_mb_height );
1278 /* Compute misc bits */
1279 h->stat.frame.i_misc_bits = bs_pos( &h->out.bs )
1281 - h->stat.frame.i_tex_bits
1282 - h->stat.frame.i_mv_bits;
1285 static void x264_thread_sync_context( x264_t *dst, x264_t *src )
1291 // reference counting
1292 for( f = src->frames.reference; *f; f++ )
1293 (*f)->i_reference_count++;
1294 for( f = dst->frames.reference; *f; f++ )
1295 x264_frame_push_unused( src, *f );
1296 src->fdec->i_reference_count++;
1297 x264_frame_push_unused( src, dst->fdec );
1299 // copy everything except the per-thread pointers and the constants.
1300 memcpy( &dst->i_frame, &src->i_frame, offsetof(x264_t, mb.type) - offsetof(x264_t, i_frame) );
1301 dst->stat = src->stat;
1304 static void x264_thread_sync_stat( x264_t *dst, x264_t *src )
1308 memcpy( &dst->stat.i_slice_count, &src->stat.i_slice_count, sizeof(dst->stat) - sizeof(dst->stat.frame) );
1311 static int x264_slices_write( x264_t *h )
1316 /* Misalign mask has to be set separately for each thread. */
1317 if( h->param.cpu&X264_CPU_SSE_MISALIGN )
1318 x264_cpu_mask_misalign_sse();
1322 if( h->param.b_visualize )
1323 x264_visualize_init( h );
1326 x264_stack_align( x264_slice_write, h );
1327 i_frame_size = h->out.nal[h->out.i_nal-1].i_payload;
1330 if( h->param.b_visualize )
1332 x264_visualize_show( h );
1333 x264_visualize_close( h );
1337 h->out.i_frame_size = i_frame_size;
1341 /****************************************************************************
1342 * x264_encoder_encode:
1343 * XXX: i_poc : is the poc of the current given picture
1344 * i_frame : is the number of the frame being coded
1345 * ex: type frame poc
1353 ****************************************************************************/
1354 int x264_encoder_encode( x264_t *h,
1355 x264_nal_t **pp_nal, int *pi_nal,
1356 x264_picture_t *pic_in,
1357 x264_picture_t *pic_out )
1359 x264_t *thread_current, *thread_prev, *thread_oldest;
1365 if( h->param.i_threads > 1)
1367 int i = ++h->i_thread_phase;
1368 int t = h->param.i_threads;
1369 thread_current = h->thread[ i%t ];
1370 thread_prev = h->thread[ (i-1)%t ];
1371 thread_oldest = h->thread[ (i+1)%t ];
1372 x264_thread_sync_context( thread_current, thread_prev );
1373 x264_thread_sync_ratecontrol( thread_current, thread_prev, thread_oldest );
1375 // fprintf(stderr, "current: %p prev: %p oldest: %p \n", thread_current, thread_prev, thread_oldest);
1384 // ok to call this before encoding any frames, since the initial values of fdec have b_kept_as_ref=0
1385 x264_reference_update( h );
1386 h->fdec->i_lines_completed = -1;
1392 /* ------------------- Setup new frame from picture -------------------- */
1393 if( pic_in != NULL )
1395 /* 1: Copy the picture to a frame and move it to a buffer */
1396 x264_frame_t *fenc = x264_frame_pop_unused( h );
1398 if( x264_frame_copy_picture( h, fenc, pic_in ) < 0 )
1401 if( h->param.i_width != 16 * h->sps->i_mb_width ||
1402 h->param.i_height != 16 * h->sps->i_mb_height )
1403 x264_frame_expand_border_mod16( h, fenc );
1405 fenc->i_frame = h->frames.i_input++;
1407 x264_frame_push( h->frames.next, fenc );
1409 if( h->frames.b_have_lowres )
1410 x264_frame_init_lowres( h, fenc );
1412 if( h->param.rc.i_aq_mode )
1413 x264_adaptive_quant_frame( h, fenc );
1415 if( h->frames.i_input <= h->frames.i_delay + 1 - h->param.i_threads )
1417 /* Nothing yet to encode */
1418 /* waiting for filling bframe buffer */
1419 pic_out->i_type = X264_TYPE_AUTO;
1424 if( h->frames.current[0] == NULL )
1427 /* 2: Select frame types */
1428 if( h->frames.next[0] == NULL )
1430 x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
1434 x264_slicetype_decide( h );
1436 /* 3: move some B-frames and 1 non-B to encode queue */
1437 while( IS_X264_TYPE_B( h->frames.next[bframes]->i_type ) )
1439 x264_frame_push( h->frames.current, x264_frame_shift( &h->frames.next[bframes] ) );
1440 /* FIXME: when max B-frames > 3, BREF may no longer be centered after GOP closing */
1441 if( h->param.b_bframe_pyramid && bframes > 1 )
1443 x264_frame_t *mid = x264_frame_shift( &h->frames.next[bframes/2] );
1444 mid->i_type = X264_TYPE_BREF;
1445 x264_frame_push( h->frames.current, mid );
1449 x264_frame_push( h->frames.current, x264_frame_shift( h->frames.next ) );
1452 /* ------------------- Get frame to be encoded ------------------------- */
1453 /* 4: get picture to encode */
1454 h->fenc = x264_frame_shift( h->frames.current );
1455 if( h->fenc == NULL )
1457 /* Nothing yet to encode (ex: waiting for I/P with B frames) */
1458 /* waiting for filling bframe buffer */
1459 pic_out->i_type = X264_TYPE_AUTO;
1465 if( h->fenc->i_type == X264_TYPE_IDR )
1467 h->frames.i_last_idr = h->fenc->i_frame;
1470 /* ------------------- Setup frame context ----------------------------- */
1471 /* 5: Init data dependent of frame type */
1472 if( h->fenc->i_type == X264_TYPE_IDR )
1474 /* reset ref pictures */
1475 x264_reference_reset( h );
1477 i_nal_type = NAL_SLICE_IDR;
1478 i_nal_ref_idc = NAL_PRIORITY_HIGHEST;
1479 h->sh.i_type = SLICE_TYPE_I;
1481 else if( h->fenc->i_type == X264_TYPE_I )
1483 i_nal_type = NAL_SLICE;
1484 i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
1485 h->sh.i_type = SLICE_TYPE_I;
1487 else if( h->fenc->i_type == X264_TYPE_P )
1489 i_nal_type = NAL_SLICE;
1490 i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
1491 h->sh.i_type = SLICE_TYPE_P;
1493 else if( h->fenc->i_type == X264_TYPE_BREF )
1495 i_nal_type = NAL_SLICE;
1496 i_nal_ref_idc = NAL_PRIORITY_HIGH; /* maybe add MMCO to forget it? -> low */
1497 h->sh.i_type = SLICE_TYPE_B;
1501 i_nal_type = NAL_SLICE;
1502 i_nal_ref_idc = NAL_PRIORITY_DISPOSABLE;
1503 h->sh.i_type = SLICE_TYPE_B;
1507 h->fenc->i_poc = 2 * (h->fenc->i_frame - h->frames.i_last_idr);
1508 h->fdec->i_type = h->fenc->i_type;
1509 h->fdec->i_frame = h->fenc->i_frame;
1510 h->fenc->b_kept_as_ref =
1511 h->fdec->b_kept_as_ref = i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE && h->param.i_keyint_max > 1;
1515 /* ------------------- Init ----------------------------- */
1516 /* build ref list 0/1 */
1517 x264_reference_build_list( h, h->fdec->i_poc );
1519 /* Init the rate control */
1520 x264_ratecontrol_start( h, h->fenc->i_qpplus1 );
1521 i_global_qp = x264_ratecontrol_qp( h );
1523 pic_out->i_qpplus1 =
1524 h->fdec->i_qpplus1 = i_global_qp + 1;
1526 if( h->sh.i_type == SLICE_TYPE_B )
1527 x264_macroblock_bipred_init( h );
1529 /* ------------------------ Create slice header ----------------------- */
1530 x264_slice_init( h, i_nal_type, i_global_qp );
1532 if( i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE )
1535 /* ---------------------- Write the bitstream -------------------------- */
1536 /* Init bitstream context */
1538 bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
1543 if(h->sh.i_type == SLICE_TYPE_I)
1545 else if(h->sh.i_type == SLICE_TYPE_P)
1547 else if(h->sh.i_type == SLICE_TYPE_B)
1552 x264_nal_start(h, NAL_AUD, NAL_PRIORITY_DISPOSABLE);
1553 bs_write(&h->out.bs, 3, pic_type);
1554 bs_rbsp_trailing(&h->out.bs);
1558 h->i_nal_type = i_nal_type;
1559 h->i_nal_ref_idc = i_nal_ref_idc;
1561 /* Write SPS and PPS */
1562 if( i_nal_type == NAL_SLICE_IDR && h->param.b_repeat_headers )
1564 if( h->fenc->i_frame == 0 )
1566 /* identify ourself */
1567 x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
1568 x264_sei_version_write( h, &h->out.bs );
1572 /* generate sequence parameters */
1573 x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
1574 x264_sps_write( &h->out.bs, h->sps );
1577 /* generate picture parameters */
1578 x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
1579 x264_pps_write( &h->out.bs, h->pps );
1584 if( h->param.i_threads > 1 )
1586 x264_pthread_create( &h->thread_handle, NULL, (void*)x264_slices_write, h );
1587 h->b_thread_active = 1;
1590 x264_slices_write( h );
1592 /* restore CPU state (before using float again) */
1595 if( h->sh.i_type == SLICE_TYPE_P && !h->param.rc.b_stat_read
1596 && h->param.i_scenecut_threshold >= 0
1597 && !h->param.b_pre_scenecut )
1599 const int *mbs = h->stat.frame.i_mb_count;
1600 int i_mb_i = mbs[I_16x16] + mbs[I_8x8] + mbs[I_4x4];
1601 int i_mb_p = mbs[P_L0] + mbs[P_8x8];
1602 int i_mb_s = mbs[P_SKIP];
1603 int i_mb = h->sps->i_mb_width * h->sps->i_mb_height;
1604 int64_t i_inter_cost = h->stat.frame.i_inter_cost;
1605 int64_t i_intra_cost = h->stat.frame.i_intra_cost;
1608 int i_gop_size = h->fenc->i_frame - h->frames.i_last_idr;
1609 float f_thresh_max = h->param.i_scenecut_threshold / 100.0;
1610 /* magic numbers pulled out of thin air */
1611 float f_thresh_min = f_thresh_max * h->param.i_keyint_min
1612 / ( h->param.i_keyint_max * 4 );
1613 if( h->param.i_keyint_min == h->param.i_keyint_max )
1614 f_thresh_min= f_thresh_max;
1616 /* macroblock_analyse() doesn't further analyse skipped mbs,
1617 * so we have to guess their cost */
1618 if( h->stat.frame.i_mbs_analysed > 0 )
1619 i_intra_cost = i_intra_cost * i_mb / h->stat.frame.i_mbs_analysed;
1621 if( i_gop_size < h->param.i_keyint_min / 4 )
1622 f_bias = f_thresh_min / 4;
1623 else if( i_gop_size <= h->param.i_keyint_min )
1624 f_bias = f_thresh_min * i_gop_size / h->param.i_keyint_min;
1627 f_bias = f_thresh_min
1628 + ( f_thresh_max - f_thresh_min )
1629 * ( i_gop_size - h->param.i_keyint_min )
1630 / ( h->param.i_keyint_max - h->param.i_keyint_min );
1632 f_bias = X264_MIN( f_bias, 1.0 );
1634 /* Bad P will be reencoded as I */
1635 if( h->stat.frame.i_mbs_analysed > 0 &&
1636 i_inter_cost >= (1.0 - f_bias) * i_intra_cost )
1640 x264_log( h, X264_LOG_DEBUG, "scene cut at %d Icost:%.0f Pcost:%.0f ratio:%.4f bias:%.4f gop:%d (imb:%d pmb:%d smb:%d)\n",
1642 (double)i_intra_cost, (double)i_inter_cost,
1643 1. - (double)i_inter_cost / i_intra_cost,
1645 i_mb_i, i_mb_p, i_mb_s );
1647 /* Restore frame num */
1650 for( b = 0; h->frames.current[b] && IS_X264_TYPE_B( h->frames.current[b]->i_type ); b++ );
1653 /* If using B-frames, force GOP to be closed.
1654 * Even if this frame is going to be I and not IDR, forcing a
1655 * P-frame before the scenecut will probably help compression.
1657 * We don't yet know exactly which frame is the scene cut, so
1658 * we can't assign an I-frame. Instead, change the previous
1659 * B-frame to P, and rearrange coding order. */
1661 if( h->param.i_bframe_adaptive || b > 1 )
1662 h->fenc->i_type = X264_TYPE_AUTO;
1663 x264_frame_sort_pts( h->frames.current );
1664 x264_frame_unshift( h->frames.next, h->fenc );
1665 h->fenc = h->frames.current[b-1];
1666 h->frames.current[b-1] = NULL;
1667 h->fenc->i_type = X264_TYPE_P;
1668 x264_frame_sort_dts( h->frames.current );
1670 /* Do IDR if needed */
1671 else if( i_gop_size >= h->param.i_keyint_min )
1676 /* Reinit field of fenc */
1677 h->fenc->i_type = X264_TYPE_IDR;
1680 /* Put enqueued frames back in the pool */
1681 while( h->frames.current[0] )
1682 x264_frame_push( h->frames.next, x264_frame_shift( h->frames.current ) );
1683 x264_frame_sort_pts( h->frames.next );
1687 h->fenc->i_type = X264_TYPE_I;
1693 x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
1697 static void x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
1698 x264_nal_t **pp_nal, int *pi_nal,
1699 x264_picture_t *pic_out )
1702 char psz_message[80];
1704 if( h->b_thread_active )
1706 x264_pthread_join( h->thread_handle, NULL );
1707 h->b_thread_active = 0;
1711 pic_out->i_type = X264_TYPE_AUTO;
1715 x264_frame_push_unused( thread_current, h->fenc );
1717 /* End bitstream, set output */
1718 *pi_nal = h->out.i_nal;
1719 *pp_nal = h->out.nal;
1722 /* Set output picture properties */
1723 if( h->sh.i_type == SLICE_TYPE_I )
1724 pic_out->i_type = h->i_nal_type == NAL_SLICE_IDR ? X264_TYPE_IDR : X264_TYPE_I;
1725 else if( h->sh.i_type == SLICE_TYPE_P )
1726 pic_out->i_type = X264_TYPE_P;
1728 pic_out->i_type = X264_TYPE_B;
1729 pic_out->i_pts = h->fenc->i_pts;
1731 pic_out->img.i_plane = h->fdec->i_plane;
1732 for(i = 0; i < 4; i++){
1733 pic_out->img.i_stride[i] = h->fdec->i_stride[i];
1734 pic_out->img.plane[i] = h->fdec->plane[i];
1737 /* ---------------------- Update encoder state ------------------------- */
1741 x264_ratecontrol_end( h, h->out.i_frame_size * 8 );
1743 /* restore CPU state (before using float again) */
1746 x264_noise_reduction_update( thread_current );
1748 /* ---------------------- Compute/Print statistics --------------------- */
1749 x264_thread_sync_stat( h, h->thread[0] );
1752 h->stat.i_slice_count[h->sh.i_type]++;
1753 h->stat.i_slice_size[h->sh.i_type] += h->out.i_frame_size + NALU_OVERHEAD;
1754 h->stat.f_slice_qp[h->sh.i_type] += h->fdec->f_qp_avg_aq;
1756 for( i = 0; i < X264_MBTYPE_MAX; i++ )
1757 h->stat.i_mb_count[h->sh.i_type][i] += h->stat.frame.i_mb_count[i];
1758 for( i = 0; i < X264_PARTTYPE_MAX; i++ )
1759 h->stat.i_mb_partition[h->sh.i_type][i] += h->stat.frame.i_mb_partition[i];
1760 for( i = 0; i < 2; i++ )
1761 h->stat.i_mb_count_8x8dct[i] += h->stat.frame.i_mb_count_8x8dct[i];
1762 if( h->sh.i_type != SLICE_TYPE_I )
1763 for( i_list = 0; i_list < 2; i_list++ )
1764 for( i = 0; i < 32; i++ )
1765 h->stat.i_mb_count_ref[h->sh.i_type][i_list][i] += h->stat.frame.i_mb_count_ref[i_list][i];
1766 if( h->sh.i_type == SLICE_TYPE_P )
1767 h->stat.i_consecutive_bframes[h->fdec->i_frame - h->fref0[0]->i_frame - 1]++;
1768 if( h->sh.i_type == SLICE_TYPE_B )
1770 h->stat.i_direct_frames[ h->sh.b_direct_spatial_mv_pred ] ++;
1771 if( h->mb.b_direct_auto_write )
1773 //FIXME somewhat arbitrary time constants
1774 if( h->stat.i_direct_score[0] + h->stat.i_direct_score[1] > h->mb.i_mb_count )
1776 for( i = 0; i < 2; i++ )
1777 h->stat.i_direct_score[i] = h->stat.i_direct_score[i] * 9/10;
1779 for( i = 0; i < 2; i++ )
1780 h->stat.i_direct_score[i] += h->stat.frame.i_direct_score[i];
1784 psz_message[0] = '\0';
1785 if( h->param.analyse.b_psnr )
1788 h->stat.frame.i_ssd[0],
1789 h->stat.frame.i_ssd[1],
1790 h->stat.frame.i_ssd[2],
1793 h->stat.i_ssd_global[h->sh.i_type] += ssd[0] + ssd[1] + ssd[2];
1794 h->stat.f_psnr_average[h->sh.i_type] += x264_psnr( ssd[0] + ssd[1] + ssd[2], 3 * h->param.i_width * h->param.i_height / 2 );
1795 h->stat.f_psnr_mean_y[h->sh.i_type] += x264_psnr( ssd[0], h->param.i_width * h->param.i_height );
1796 h->stat.f_psnr_mean_u[h->sh.i_type] += x264_psnr( ssd[1], h->param.i_width * h->param.i_height / 4 );
1797 h->stat.f_psnr_mean_v[h->sh.i_type] += x264_psnr( ssd[2], h->param.i_width * h->param.i_height / 4 );
1799 snprintf( psz_message, 80, " PSNR Y:%5.2f U:%5.2f V:%5.2f",
1800 x264_psnr( ssd[0], h->param.i_width * h->param.i_height ),
1801 x264_psnr( ssd[1], h->param.i_width * h->param.i_height / 4),
1802 x264_psnr( ssd[2], h->param.i_width * h->param.i_height / 4) );
1805 if( h->param.analyse.b_ssim )
1807 double ssim_y = h->stat.frame.f_ssim
1808 / (((h->param.i_width-6)>>2) * ((h->param.i_height-6)>>2));
1809 h->stat.f_ssim_mean_y[h->sh.i_type] += ssim_y;
1810 snprintf( psz_message + strlen(psz_message), 80 - strlen(psz_message),
1811 " SSIM Y:%.5f", ssim_y );
1813 psz_message[79] = '\0';
1815 x264_log( h, X264_LOG_DEBUG,
1816 "frame=%4d QP=%.2f NAL=%d Slice:%c Poc:%-3d I:%-4d P:%-4d SKIP:%-4d size=%d bytes%s\n",
1818 h->fdec->f_qp_avg_aq,
1820 h->sh.i_type == SLICE_TYPE_I ? 'I' : (h->sh.i_type == SLICE_TYPE_P ? 'P' : 'B' ),
1822 h->stat.frame.i_mb_count_i,
1823 h->stat.frame.i_mb_count_p,
1824 h->stat.frame.i_mb_count_skip,
1825 h->out.i_frame_size,
1828 // keep stats all in one place
1829 x264_thread_sync_stat( h->thread[0], h );
1830 // for the use of the next frame
1831 x264_thread_sync_stat( thread_current, h );
1833 #ifdef DEBUG_MB_TYPE
1835 static const char mb_chars[] = { 'i', 'i', 'I', 'C', 'P', '8', 'S',
1836 'D', '<', 'X', 'B', 'X', '>', 'B', 'B', 'B', 'B', '8', 'S' };
1838 for( mb_xy = 0; mb_xy < h->sps->i_mb_width * h->sps->i_mb_height; mb_xy++ )
1840 if( h->mb.type[mb_xy] < X264_MBTYPE_MAX && h->mb.type[mb_xy] >= 0 )
1841 fprintf( stderr, "%c ", mb_chars[ h->mb.type[mb_xy] ] );
1843 fprintf( stderr, "? " );
1845 if( (mb_xy+1) % h->sps->i_mb_width == 0 )
1846 fprintf( stderr, "\n" );
1851 if( h->param.psz_dump_yuv )
1852 x264_frame_dump( h );
1855 static void x264_print_intra( int64_t *i_mb_count, double i_count, int b_print_pcm, char *intra )
1857 intra += sprintf( intra, "I16..4%s: %4.1f%% %4.1f%% %4.1f%%",
1858 b_print_pcm ? "..PCM" : "",
1859 i_mb_count[I_16x16]/ i_count,
1860 i_mb_count[I_8x8] / i_count,
1861 i_mb_count[I_4x4] / i_count );
1863 sprintf( intra, " %4.1f%%", i_mb_count[I_PCM] / i_count );
1866 /****************************************************************************
1867 * x264_encoder_close:
1868 ****************************************************************************/
1869 void x264_encoder_close ( x264_t *h )
1871 int64_t i_yuv_size = 3 * h->param.i_width * h->param.i_height / 2;
1872 int64_t i_mb_count_size[2][7] = {{0}};
1874 int i, j, i_list, i_type;
1875 int b_print_pcm = h->stat.i_mb_count[SLICE_TYPE_I][I_PCM]
1876 || h->stat.i_mb_count[SLICE_TYPE_P][I_PCM]
1877 || h->stat.i_mb_count[SLICE_TYPE_B][I_PCM];
1879 for( i=0; i<h->param.i_threads; i++ )
1881 // don't strictly have to wait for the other threads, but it's simpler than canceling them
1882 if( h->thread[i]->b_thread_active )
1884 x264_pthread_join( h->thread[i]->thread_handle, NULL );
1885 assert( h->thread[i]->fenc->i_reference_count == 1 );
1886 x264_frame_delete( h->thread[i]->fenc );
1890 /* Slices used and PSNR */
1891 for( i=0; i<5; i++ )
1893 static const int slice_order[] = { SLICE_TYPE_I, SLICE_TYPE_SI, SLICE_TYPE_P, SLICE_TYPE_SP, SLICE_TYPE_B };
1894 static const char *slice_name[] = { "P", "B", "I", "SP", "SI" };
1895 int i_slice = slice_order[i];
1897 if( h->stat.i_slice_count[i_slice] > 0 )
1899 const int i_count = h->stat.i_slice_count[i_slice];
1900 if( h->param.analyse.b_psnr )
1902 x264_log( h, X264_LOG_INFO,
1903 "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",
1904 slice_name[i_slice],
1906 h->stat.f_slice_qp[i_slice] / i_count,
1907 (double)h->stat.i_slice_size[i_slice] / i_count,
1908 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,
1909 h->stat.f_psnr_average[i_slice] / i_count,
1910 x264_psnr( h->stat.i_ssd_global[i_slice], i_count * i_yuv_size ) );
1914 x264_log( h, X264_LOG_INFO,
1915 "slice %s:%-5d Avg QP:%5.2f size:%6.0f\n",
1916 slice_name[i_slice],
1918 h->stat.f_slice_qp[i_slice] / i_count,
1919 (double)h->stat.i_slice_size[i_slice] / i_count );
1923 if( h->param.i_bframe && h->stat.i_slice_count[SLICE_TYPE_P] )
1927 // weight by number of frames (including the P-frame) that are in a sequence of N B-frames
1928 for( i=0; i<=h->param.i_bframe; i++ )
1929 den += (i+1) * h->stat.i_consecutive_bframes[i];
1930 for( i=0; i<=h->param.i_bframe; i++ )
1931 p += sprintf( p, " %4.1f%%", 100. * (i+1) * h->stat.i_consecutive_bframes[i] / den );
1932 x264_log( h, X264_LOG_INFO, "consecutive B-frames:%s\n", buf );
1935 for( i_type = 0; i_type < 2; i_type++ )
1936 for( i = 0; i < X264_PARTTYPE_MAX; i++ )
1938 if( i == D_DIRECT_8x8 ) continue; /* direct is counted as its own type */
1939 i_mb_count_size[i_type][x264_mb_partition_pixel_table[i]] += h->stat.i_mb_partition[i_type][i];
1943 if( h->stat.i_slice_count[SLICE_TYPE_I] > 0 )
1945 int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_I];
1946 double i_count = h->stat.i_slice_count[SLICE_TYPE_I] * h->mb.i_mb_count / 100.0;
1947 x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
1948 x264_log( h, X264_LOG_INFO, "mb I %s\n", buf );
1950 if( h->stat.i_slice_count[SLICE_TYPE_P] > 0 )
1952 int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_P];
1953 double i_count = h->stat.i_slice_count[SLICE_TYPE_P] * h->mb.i_mb_count / 100.0;
1954 int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_P];
1955 x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
1956 x264_log( h, X264_LOG_INFO,
1957 "mb P %s P16..4: %4.1f%% %4.1f%% %4.1f%% %4.1f%% %4.1f%% skip:%4.1f%%\n",
1959 i_mb_size[PIXEL_16x16] / (i_count*4),
1960 (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
1961 i_mb_size[PIXEL_8x8] / (i_count*4),
1962 (i_mb_size[PIXEL_8x4] + i_mb_size[PIXEL_4x8]) / (i_count*4),
1963 i_mb_size[PIXEL_4x4] / (i_count*4),
1964 i_mb_count[P_SKIP] / i_count );
1966 if( h->stat.i_slice_count[SLICE_TYPE_B] > 0 )
1968 int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_B];
1969 double i_count = h->stat.i_slice_count[SLICE_TYPE_B] * h->mb.i_mb_count / 100.0;
1970 double i_mb_list_count;
1971 int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_B];
1972 int64_t list_count[3] = {0}; /* 0 == L0, 1 == L1, 2 == BI */
1973 x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
1974 for( i = 0; i < X264_PARTTYPE_MAX; i++ )
1975 for( j = 0; j < 2; j++ )
1977 int l0 = x264_mb_type_list0_table[i][j];
1978 int l1 = x264_mb_type_list1_table[i][j];
1980 list_count[l1+l0*l1] += h->stat.i_mb_count[SLICE_TYPE_B][i] * 2;
1982 list_count[0] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L0_8x8];
1983 list_count[1] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L1_8x8];
1984 list_count[2] += h->stat.i_mb_partition[SLICE_TYPE_B][D_BI_8x8];
1985 i_mb_count[B_DIRECT] += (h->stat.i_mb_partition[SLICE_TYPE_B][D_DIRECT_8x8]+2)/4;
1986 i_mb_list_count = (list_count[0] + list_count[1] + list_count[2]) / 100.0;
1987 x264_log( h, X264_LOG_INFO,
1988 "mb B %s B16..8: %4.1f%% %4.1f%% %4.1f%% direct:%4.1f%% skip:%4.1f%% L0:%4.1f%% L1:%4.1f%% BI:%4.1f%%\n",
1990 i_mb_size[PIXEL_16x16] / (i_count*4),
1991 (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
1992 i_mb_size[PIXEL_8x8] / (i_count*4),
1993 i_mb_count[B_DIRECT] / i_count,
1994 i_mb_count[B_SKIP] / i_count,
1995 list_count[0] / i_mb_list_count,
1996 list_count[1] / i_mb_list_count,
1997 list_count[2] / i_mb_list_count );
2000 x264_ratecontrol_summary( h );
2002 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 )
2004 const int i_count = h->stat.i_slice_count[SLICE_TYPE_I] +
2005 h->stat.i_slice_count[SLICE_TYPE_P] +
2006 h->stat.i_slice_count[SLICE_TYPE_B];
2007 float fps = (float) h->param.i_fps_num / h->param.i_fps_den;
2008 #define SUM3(p) (p[SLICE_TYPE_I] + p[SLICE_TYPE_P] + p[SLICE_TYPE_B])
2009 #define SUM3b(p,o) (p[SLICE_TYPE_I][o] + p[SLICE_TYPE_P][o] + p[SLICE_TYPE_B][o])
2010 float f_bitrate = fps * SUM3(h->stat.i_slice_size) / i_count / 125;
2012 if( h->pps->b_transform_8x8_mode )
2014 int64_t i_i8x8 = SUM3b( h->stat.i_mb_count, I_8x8 );
2015 int64_t i_intra = i_i8x8 + SUM3b( h->stat.i_mb_count, I_4x4 )
2016 + SUM3b( h->stat.i_mb_count, I_16x16 );
2017 x264_log( h, X264_LOG_INFO, "8x8 transform intra:%.1f%% inter:%.1f%%\n",
2018 100. * i_i8x8 / i_intra,
2019 100. * h->stat.i_mb_count_8x8dct[1] / h->stat.i_mb_count_8x8dct[0] );
2022 if( h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO
2023 && h->stat.i_slice_count[SLICE_TYPE_B] )
2025 x264_log( h, X264_LOG_INFO, "direct mvs spatial:%.1f%% temporal:%.1f%%\n",
2026 h->stat.i_direct_frames[1] * 100. / h->stat.i_slice_count[SLICE_TYPE_B],
2027 h->stat.i_direct_frames[0] * 100. / h->stat.i_slice_count[SLICE_TYPE_B] );
2030 for( i_list = 0; i_list < 2; i_list++ )
2033 for( i_slice = 0; i_slice < 2; i_slice++ )
2038 for( i = 0; i < 32; i++ )
2039 if( h->stat.i_mb_count_ref[i_slice][i_list][i] )
2041 i_den += h->stat.i_mb_count_ref[i_slice][i_list][i];
2046 for( i = 0; i <= i_max; i++ )
2047 p += sprintf( p, " %4.1f%%", 100. * h->stat.i_mb_count_ref[i_slice][i_list][i] / i_den );
2048 x264_log( h, X264_LOG_INFO, "ref %c L%d %s\n", "PB"[i_slice], i_list, buf );
2052 if( h->param.analyse.b_ssim )
2054 x264_log( h, X264_LOG_INFO,
2055 "SSIM Mean Y:%.7f\n",
2056 SUM3( h->stat.f_ssim_mean_y ) / i_count );
2058 if( h->param.analyse.b_psnr )
2060 x264_log( h, X264_LOG_INFO,
2061 "PSNR Mean Y:%6.3f U:%6.3f V:%6.3f Avg:%6.3f Global:%6.3f kb/s:%.2f\n",
2062 SUM3( h->stat.f_psnr_mean_y ) / i_count,
2063 SUM3( h->stat.f_psnr_mean_u ) / i_count,
2064 SUM3( h->stat.f_psnr_mean_v ) / i_count,
2065 SUM3( h->stat.f_psnr_average ) / i_count,
2066 x264_psnr( SUM3( h->stat.i_ssd_global ), i_count * i_yuv_size ),
2070 x264_log( h, X264_LOG_INFO, "kb/s:%.1f\n", f_bitrate );
2074 x264_ratecontrol_delete( h );
2077 if( h->param.rc.psz_stat_out )
2078 free( h->param.rc.psz_stat_out );
2079 if( h->param.rc.psz_stat_in )
2080 free( h->param.rc.psz_stat_in );
2082 x264_cqm_delete( h );
2084 if( h->param.i_threads > 1)
2085 h = h->thread[ h->i_thread_phase % h->param.i_threads ];
2088 for( i = 0; h->frames.current[i]; i++ )
2090 assert( h->frames.current[i]->i_reference_count == 1 );
2091 x264_frame_delete( h->frames.current[i] );
2093 for( i = 0; h->frames.next[i]; i++ )
2095 assert( h->frames.next[i]->i_reference_count == 1 );
2096 x264_frame_delete( h->frames.next[i] );
2098 for( i = 0; h->frames.unused[i]; i++ )
2100 assert( h->frames.unused[i]->i_reference_count == 0 );
2101 x264_frame_delete( h->frames.unused[i] );
2106 for( i = h->param.i_threads - 1; i >= 0; i-- )
2108 x264_frame_t **frame;
2110 for( frame = h->thread[i]->frames.reference; *frame; frame++ )
2112 assert( (*frame)->i_reference_count > 0 );
2113 (*frame)->i_reference_count--;
2114 if( (*frame)->i_reference_count == 0 )
2115 x264_frame_delete( *frame );
2117 frame = &h->thread[i]->fdec;
2118 assert( (*frame)->i_reference_count > 0 );
2119 (*frame)->i_reference_count--;
2120 if( (*frame)->i_reference_count == 0 )
2121 x264_frame_delete( *frame );
2123 x264_macroblock_cache_end( h->thread[i] );
2124 x264_free( h->thread[i]->out.p_bitstream );
2125 x264_free( h->thread[i] );