1 /*****************************************************************************
2 * encoder.c: top-level encoder functions
3 *****************************************************************************
4 * Copyright (C) 2003-2011 x264 project
6 * Authors: Laurent Aimar <fenrir@via.ecp.fr>
7 * Loren Merritt <lorenm@u.washington.edu>
8 * Fiona Glaser <fiona@x264.com>
10 * This program is free software; you can redistribute it and/or modify
11 * it under the terms of the GNU General Public License as published by
12 * the Free Software Foundation; either version 2 of the License, or
13 * (at your option) any later version.
15 * This program is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
18 * GNU General Public License for more details.
20 * You should have received a copy of the GNU General Public License
21 * along with this program; if not, write to the Free Software
22 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02111, USA.
24 * This program is also available under a commercial proprietary license.
25 * For more information, contact us at licensing@x264.com.
26 *****************************************************************************/
28 #include "common/common.h"
32 #include "ratecontrol.h"
33 #include "macroblock.h"
37 #include "common/visualize.h"
40 //#define DEBUG_MB_TYPE
42 #define bs_write_ue bs_write_ue_big
44 static int 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 double x264_psnr( double sqe, double size )
55 double mse = sqe / (PIXEL_MAX*PIXEL_MAX * size);
56 if( mse <= 0.0000000001 ) /* Max 100dB */
59 return -10.0 * log10( mse );
62 static double x264_ssim( double ssim )
64 return -10.0 * log10( 1 - ssim );
67 static void x264_frame_dump( x264_t *h )
69 FILE *f = fopen( h->param.psz_dump_yuv, "r+b" );
72 /* Write the frame in display order */
73 fseek( f, (uint64_t)h->fdec->i_frame * h->param.i_height * h->param.i_width * 3/2 * sizeof(pixel), SEEK_SET );
74 for( int y = 0; y < h->param.i_height; y++ )
75 fwrite( &h->fdec->plane[0][y*h->fdec->i_stride[0]], sizeof(pixel), h->param.i_width, f );
76 int cw = h->param.i_width>>1;
77 int ch = h->param.i_height>>1;
78 pixel *planeu = x264_malloc( (cw*ch*2+32)*sizeof(pixel) );
79 pixel *planev = planeu + cw*ch + 16;
80 h->mc.plane_copy_deinterleave( planeu, cw, planev, cw, h->fdec->plane[1], h->fdec->i_stride[1], cw, ch );
81 fwrite( planeu, 1, cw*ch*sizeof(pixel), f );
82 fwrite( planev, 1, cw*ch*sizeof(pixel), f );
88 /* Fill "default" values */
89 static void x264_slice_header_init( x264_t *h, x264_slice_header_t *sh,
90 x264_sps_t *sps, x264_pps_t *pps,
91 int i_idr_pic_id, int i_frame, int i_qp )
93 x264_param_t *param = &h->param;
95 /* First we fill all fields */
100 sh->i_last_mb = h->mb.i_mb_count - 1;
101 sh->i_pps_id = pps->i_id;
103 sh->i_frame_num = i_frame;
105 sh->b_mbaff = h->param.b_interlaced;
106 sh->b_field_pic = 0; /* no field support for now */
107 sh->b_bottom_field = 0; /* not yet used */
109 sh->i_idr_pic_id = i_idr_pic_id;
111 /* poc stuff, fixed later */
113 sh->i_delta_poc_bottom = 0;
114 sh->i_delta_poc[0] = 0;
115 sh->i_delta_poc[1] = 0;
117 sh->i_redundant_pic_cnt = 0;
119 h->mb.b_direct_auto_write = h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO
121 && ( h->param.rc.b_stat_write || !h->param.rc.b_stat_read );
123 if( !h->mb.b_direct_auto_read && sh->i_type == SLICE_TYPE_B )
125 if( h->fref[1][0]->i_poc_l0ref0 == h->fref[0][0]->i_poc )
127 if( h->mb.b_direct_auto_write )
128 sh->b_direct_spatial_mv_pred = ( h->stat.i_direct_score[1] > h->stat.i_direct_score[0] );
130 sh->b_direct_spatial_mv_pred = ( param->analyse.i_direct_mv_pred == X264_DIRECT_PRED_SPATIAL );
134 h->mb.b_direct_auto_write = 0;
135 sh->b_direct_spatial_mv_pred = 1;
138 /* else b_direct_spatial_mv_pred was read from the 2pass statsfile */
140 sh->b_num_ref_idx_override = 0;
141 sh->i_num_ref_idx_l0_active = 1;
142 sh->i_num_ref_idx_l1_active = 1;
144 sh->b_ref_pic_list_reordering[0] = h->b_ref_reorder[0];
145 sh->b_ref_pic_list_reordering[1] = h->b_ref_reorder[1];
147 /* If the ref list isn't in the default order, construct reordering header */
148 for( int list = 0; list < 2; list++ )
150 if( sh->b_ref_pic_list_reordering[list] )
152 int pred_frame_num = i_frame;
153 for( int i = 0; i < h->i_ref[list]; i++ )
155 int diff = h->fref[list][i]->i_frame_num - pred_frame_num;
156 sh->ref_pic_list_order[list][i].idc = ( diff > 0 );
157 sh->ref_pic_list_order[list][i].arg = (abs(diff) - 1) & ((1 << sps->i_log2_max_frame_num) - 1);
158 pred_frame_num = h->fref[list][i]->i_frame_num;
163 sh->i_cabac_init_idc = param->i_cabac_init_idc;
165 sh->i_qp = SPEC_QP(i_qp);
166 sh->i_qp_delta = sh->i_qp - pps->i_pic_init_qp;
167 sh->b_sp_for_swidth = 0;
170 int deblock_thresh = i_qp + 2 * X264_MIN(param->i_deblocking_filter_alphac0, param->i_deblocking_filter_beta);
171 /* If effective qp <= 15, deblocking would have no effect anyway */
172 if( param->b_deblocking_filter && (h->mb.b_variable_qp || 15 < deblock_thresh ) )
173 sh->i_disable_deblocking_filter_idc = param->b_sliced_threads ? 2 : 0;
175 sh->i_disable_deblocking_filter_idc = 1;
176 sh->i_alpha_c0_offset = param->i_deblocking_filter_alphac0 << 1;
177 sh->i_beta_offset = param->i_deblocking_filter_beta << 1;
180 static void x264_slice_header_write( bs_t *s, x264_slice_header_t *sh, int i_nal_ref_idc )
184 int first_x = sh->i_first_mb % sh->sps->i_mb_width;
185 int first_y = sh->i_first_mb / sh->sps->i_mb_width;
186 assert( (first_y&1) == 0 );
187 bs_write_ue( s, (2*first_x + sh->sps->i_mb_width*(first_y&~1) + (first_y&1)) >> 1 );
190 bs_write_ue( s, sh->i_first_mb );
192 bs_write_ue( s, sh->i_type + 5 ); /* same type things */
193 bs_write_ue( s, sh->i_pps_id );
194 bs_write( s, sh->sps->i_log2_max_frame_num, sh->i_frame_num & ((1<<sh->sps->i_log2_max_frame_num)-1) );
196 if( !sh->sps->b_frame_mbs_only )
198 bs_write1( s, sh->b_field_pic );
199 if( sh->b_field_pic )
200 bs_write1( s, sh->b_bottom_field );
203 if( sh->i_idr_pic_id >= 0 ) /* NAL IDR */
204 bs_write_ue( s, sh->i_idr_pic_id );
206 if( sh->sps->i_poc_type == 0 )
208 bs_write( s, sh->sps->i_log2_max_poc_lsb, sh->i_poc & ((1<<sh->sps->i_log2_max_poc_lsb)-1) );
209 if( sh->pps->b_pic_order && !sh->b_field_pic )
210 bs_write_se( s, sh->i_delta_poc_bottom );
212 else if( sh->sps->i_poc_type == 1 && !sh->sps->b_delta_pic_order_always_zero )
214 bs_write_se( s, sh->i_delta_poc[0] );
215 if( sh->pps->b_pic_order && !sh->b_field_pic )
216 bs_write_se( s, sh->i_delta_poc[1] );
219 if( sh->pps->b_redundant_pic_cnt )
220 bs_write_ue( s, sh->i_redundant_pic_cnt );
222 if( sh->i_type == SLICE_TYPE_B )
223 bs_write1( s, sh->b_direct_spatial_mv_pred );
225 if( sh->i_type == SLICE_TYPE_P || sh->i_type == SLICE_TYPE_B )
227 bs_write1( s, sh->b_num_ref_idx_override );
228 if( sh->b_num_ref_idx_override )
230 bs_write_ue( s, sh->i_num_ref_idx_l0_active - 1 );
231 if( sh->i_type == SLICE_TYPE_B )
232 bs_write_ue( s, sh->i_num_ref_idx_l1_active - 1 );
236 /* ref pic list reordering */
237 if( sh->i_type != SLICE_TYPE_I )
239 bs_write1( s, sh->b_ref_pic_list_reordering[0] );
240 if( sh->b_ref_pic_list_reordering[0] )
242 for( int i = 0; i < sh->i_num_ref_idx_l0_active; i++ )
244 bs_write_ue( s, sh->ref_pic_list_order[0][i].idc );
245 bs_write_ue( s, sh->ref_pic_list_order[0][i].arg );
250 if( sh->i_type == SLICE_TYPE_B )
252 bs_write1( s, sh->b_ref_pic_list_reordering[1] );
253 if( sh->b_ref_pic_list_reordering[1] )
255 for( int i = 0; i < sh->i_num_ref_idx_l1_active; i++ )
257 bs_write_ue( s, sh->ref_pic_list_order[1][i].idc );
258 bs_write_ue( s, sh->ref_pic_list_order[1][i].arg );
264 if( sh->pps->b_weighted_pred && sh->i_type == SLICE_TYPE_P )
266 /* pred_weight_table() */
267 bs_write_ue( s, sh->weight[0][0].i_denom );
268 bs_write_ue( s, sh->weight[0][1].i_denom );
269 for( int i = 0; i < sh->i_num_ref_idx_l0_active; i++ )
271 int luma_weight_l0_flag = !!sh->weight[i][0].weightfn;
272 int chroma_weight_l0_flag = !!sh->weight[i][1].weightfn || !!sh->weight[i][2].weightfn;
273 bs_write1( s, luma_weight_l0_flag );
274 if( luma_weight_l0_flag )
276 bs_write_se( s, sh->weight[i][0].i_scale );
277 bs_write_se( s, sh->weight[i][0].i_offset );
279 bs_write1( s, chroma_weight_l0_flag );
280 if( chroma_weight_l0_flag )
282 for( int j = 1; j < 3; j++ )
284 bs_write_se( s, sh->weight[i][j].i_scale );
285 bs_write_se( s, sh->weight[i][j].i_offset );
290 else if( sh->pps->b_weighted_bipred == 1 && sh->i_type == SLICE_TYPE_B )
295 if( i_nal_ref_idc != 0 )
297 if( sh->i_idr_pic_id >= 0 )
299 bs_write1( s, 0 ); /* no output of prior pics flag */
300 bs_write1( s, 0 ); /* long term reference flag */
304 bs_write1( s, sh->i_mmco_command_count > 0 ); /* adaptive_ref_pic_marking_mode_flag */
305 if( sh->i_mmco_command_count > 0 )
307 for( int i = 0; i < sh->i_mmco_command_count; i++ )
309 bs_write_ue( s, 1 ); /* mark short term ref as unused */
310 bs_write_ue( s, sh->mmco[i].i_difference_of_pic_nums - 1 );
312 bs_write_ue( s, 0 ); /* end command list */
317 if( sh->pps->b_cabac && sh->i_type != SLICE_TYPE_I )
318 bs_write_ue( s, sh->i_cabac_init_idc );
320 bs_write_se( s, sh->i_qp_delta ); /* slice qp delta */
322 if( sh->pps->b_deblocking_filter_control )
324 bs_write_ue( s, sh->i_disable_deblocking_filter_idc );
325 if( sh->i_disable_deblocking_filter_idc != 1 )
327 bs_write_se( s, sh->i_alpha_c0_offset >> 1 );
328 bs_write_se( s, sh->i_beta_offset >> 1 );
333 /* If we are within a reasonable distance of the end of the memory allocated for the bitstream, */
334 /* reallocate, adding an arbitrary amount of space (100 kilobytes). */
335 static int x264_bitstream_check_buffer( x264_t *h )
337 uint8_t *bs_bak = h->out.p_bitstream;
338 if( (h->param.b_cabac && (h->cabac.p_end - h->cabac.p < 2500)) ||
339 (h->out.bs.p_end - h->out.bs.p < 2500) )
341 h->out.i_bitstream += 100000;
342 CHECKED_MALLOC( h->out.p_bitstream, h->out.i_bitstream );
343 h->mc.memcpy_aligned( h->out.p_bitstream, bs_bak, (h->out.i_bitstream - 100000) & ~15 );
344 intptr_t delta = h->out.p_bitstream - bs_bak;
346 h->out.bs.p_start += delta;
347 h->out.bs.p += delta;
348 h->out.bs.p_end = h->out.p_bitstream + h->out.i_bitstream;
350 h->cabac.p_start += delta;
352 h->cabac.p_end = h->out.p_bitstream + h->out.i_bitstream;
354 for( int i = 0; i <= h->out.i_nal; i++ )
355 h->out.nal[i].p_payload += delta;
365 static void x264_encoder_thread_init( x264_t *h )
367 if( h->param.i_sync_lookahead )
368 x264_lower_thread_priority( 10 );
371 /* Misalign mask has to be set separately for each thread. */
372 if( h->param.cpu&X264_CPU_SSE_MISALIGN )
373 x264_cpu_mask_misalign_sse();
378 /****************************************************************************
380 ****************************************************************************
381 ****************************** External API*********************************
382 ****************************************************************************
384 ****************************************************************************/
386 static int x264_validate_parameters( x264_t *h, int b_open )
390 if( b_open && !(x264_cpu_detect() & X264_CPU_SSE) )
392 x264_log( h, X264_LOG_ERROR, "your cpu does not support SSE1, but x264 was compiled with asm support\n");
394 if( b_open && !(x264_cpu_detect() & X264_CPU_MMXEXT) )
396 x264_log( h, X264_LOG_ERROR, "your cpu does not support MMXEXT, but x264 was compiled with asm support\n");
398 x264_log( h, X264_LOG_ERROR, "to run x264, recompile without asm support (configure --disable-asm)\n");
402 if( h->param.i_width <= 0 || h->param.i_height <= 0 )
404 x264_log( h, X264_LOG_ERROR, "invalid width x height (%dx%d)\n",
405 h->param.i_width, h->param.i_height );
409 if( h->param.i_width % 2 || h->param.i_height % 2 )
411 x264_log( h, X264_LOG_ERROR, "width or height not divisible by 2 (%dx%d)\n",
412 h->param.i_width, h->param.i_height );
415 int i_csp = h->param.i_csp & X264_CSP_MASK;
416 if( i_csp <= X264_CSP_NONE || i_csp >= X264_CSP_MAX )
418 x264_log( h, X264_LOG_ERROR, "invalid CSP (only I420/YV12/NV12 supported)\n" );
422 if( (h->param.crop_rect.i_left + h->param.crop_rect.i_right ) >= h->param.i_width ||
423 (h->param.crop_rect.i_top + h->param.crop_rect.i_bottom) >= h->param.i_height )
425 x264_log( h, X264_LOG_ERROR, "invalid crop-rect %u,%u,%u,%u\n", h->param.crop_rect.i_left,
426 h->param.crop_rect.i_top, h->param.crop_rect.i_right, h->param.crop_rect.i_bottom );
430 if( h->param.i_threads == X264_THREADS_AUTO )
431 h->param.i_threads = x264_cpu_num_processors() * (h->param.b_sliced_threads?2:3)/2;
432 h->param.i_threads = x264_clip3( h->param.i_threads, 1, X264_THREAD_MAX );
433 if( h->param.i_threads > 1 )
436 x264_log( h, X264_LOG_WARNING, "not compiled with thread support!\n");
437 h->param.i_threads = 1;
439 /* Avoid absurdly small thread slices as they can reduce performance
440 * and VBV compliance. Capped at an arbitrary 4 rows per thread. */
441 if( h->param.b_sliced_threads )
443 int max_threads = (h->param.i_height+15)/16 / 4;
444 h->param.i_threads = X264_MIN( h->param.i_threads, max_threads );
448 h->param.b_sliced_threads = 0;
449 h->i_thread_frames = h->param.b_sliced_threads ? 1 : h->param.i_threads;
450 if( h->i_thread_frames > 1 )
451 h->param.nalu_process = NULL;
453 h->param.i_keyint_max = x264_clip3( h->param.i_keyint_max, 1, X264_KEYINT_MAX_INFINITE );
454 if( h->param.i_keyint_max == 1 )
456 h->param.b_intra_refresh = 0;
457 h->param.analyse.i_weighted_pred = 0;
460 h->param.i_frame_packing = x264_clip3( h->param.i_frame_packing, -1, 5 );
462 /* Detect default ffmpeg settings and terminate with an error. */
466 score += h->param.analyse.i_me_range == 0;
467 score += h->param.rc.i_qp_step == 3;
468 score += h->param.i_keyint_max == 12;
469 score += h->param.rc.i_qp_min == 2;
470 score += h->param.rc.i_qp_max == 31;
471 score += h->param.rc.f_qcompress == 0.5;
472 score += fabs(h->param.rc.f_ip_factor - 1.25) < 0.01;
473 score += fabs(h->param.rc.f_pb_factor - 1.25) < 0.01;
474 score += h->param.analyse.inter == 0 && h->param.analyse.i_subpel_refine == 8;
477 x264_log( h, X264_LOG_ERROR, "broken ffmpeg default settings detected\n" );
478 x264_log( h, X264_LOG_ERROR, "use an encoding preset (e.g. -vpre medium)\n" );
479 x264_log( h, X264_LOG_ERROR, "preset usage: -vpre <speed> -vpre <profile>\n" );
480 x264_log( h, X264_LOG_ERROR, "speed presets are listed in x264 --help\n" );
481 x264_log( h, X264_LOG_ERROR, "profile is optional; x264 defaults to high\n" );
486 if( h->param.rc.i_rc_method < 0 || h->param.rc.i_rc_method > 2 )
488 x264_log( h, X264_LOG_ERROR, "no ratecontrol method specified\n" );
491 h->param.rc.f_rf_constant = x264_clip3f( h->param.rc.f_rf_constant, -QP_BD_OFFSET, 51 );
492 h->param.rc.f_rf_constant_max = x264_clip3f( h->param.rc.f_rf_constant_max, -QP_BD_OFFSET, 51 );
493 h->param.rc.i_qp_constant = x264_clip3( h->param.rc.i_qp_constant, 0, QP_MAX );
494 h->param.analyse.i_subpel_refine = x264_clip3( h->param.analyse.i_subpel_refine, 0, 10 );
495 h->param.rc.f_ip_factor = X264_MAX( h->param.rc.f_ip_factor, 0.01f );
496 h->param.rc.f_pb_factor = X264_MAX( h->param.rc.f_pb_factor, 0.01f );
497 if( h->param.rc.i_rc_method == X264_RC_CRF )
499 h->param.rc.i_qp_constant = h->param.rc.f_rf_constant + QP_BD_OFFSET;
500 h->param.rc.i_bitrate = 0;
502 if( (h->param.rc.i_rc_method == X264_RC_CQP || h->param.rc.i_rc_method == X264_RC_CRF)
503 && h->param.rc.i_qp_constant == 0 )
505 h->mb.b_lossless = 1;
506 h->param.i_cqm_preset = X264_CQM_FLAT;
507 h->param.psz_cqm_file = NULL;
508 h->param.rc.i_rc_method = X264_RC_CQP;
509 h->param.rc.f_ip_factor = 1;
510 h->param.rc.f_pb_factor = 1;
511 h->param.analyse.b_psnr = 0;
512 h->param.analyse.b_ssim = 0;
513 h->param.analyse.i_chroma_qp_offset = 0;
514 h->param.analyse.i_trellis = 0;
515 h->param.analyse.b_fast_pskip = 0;
516 h->param.analyse.i_noise_reduction = 0;
517 h->param.analyse.b_psy = 0;
518 h->param.i_bframe = 0;
519 /* 8x8dct is not useful without RD in CAVLC lossless */
520 if( !h->param.b_cabac && h->param.analyse.i_subpel_refine < 6 )
521 h->param.analyse.b_transform_8x8 = 0;
523 if( h->param.rc.i_rc_method == X264_RC_CQP )
525 float qp_p = h->param.rc.i_qp_constant;
526 float qp_i = qp_p - 6*log2f( h->param.rc.f_ip_factor );
527 float qp_b = qp_p + 6*log2f( h->param.rc.f_pb_factor );
528 h->param.rc.i_qp_min = x264_clip3( (int)(X264_MIN3( qp_p, qp_i, qp_b )), 0, QP_MAX );
529 h->param.rc.i_qp_max = x264_clip3( (int)(X264_MAX3( qp_p, qp_i, qp_b ) + .999), 0, QP_MAX );
530 h->param.rc.i_aq_mode = 0;
531 h->param.rc.b_mb_tree = 0;
532 h->param.rc.i_bitrate = 0;
534 h->param.rc.i_qp_max = x264_clip3( h->param.rc.i_qp_max, 0, QP_MAX );
535 h->param.rc.i_qp_min = x264_clip3( h->param.rc.i_qp_min, 0, h->param.rc.i_qp_max );
536 h->param.rc.i_qp_step = x264_clip3( h->param.rc.i_qp_step, 0, QP_MAX );
537 h->param.rc.i_bitrate = x264_clip3( h->param.rc.i_bitrate, 0, 2000000 );
538 h->param.rc.i_vbv_buffer_size = x264_clip3( h->param.rc.i_vbv_buffer_size, 0, 2000000 );
539 h->param.rc.i_vbv_max_bitrate = x264_clip3( h->param.rc.i_vbv_max_bitrate, 0, 2000000 );
540 h->param.rc.f_vbv_buffer_init = x264_clip3f( h->param.rc.f_vbv_buffer_init, 0, 2000000 );
541 if( h->param.rc.i_vbv_buffer_size )
543 if( h->param.rc.i_rc_method == X264_RC_CQP )
545 x264_log( h, X264_LOG_WARNING, "VBV is incompatible with constant QP, ignored.\n" );
546 h->param.rc.i_vbv_max_bitrate = 0;
547 h->param.rc.i_vbv_buffer_size = 0;
549 else if( h->param.rc.i_vbv_max_bitrate == 0 )
551 if( h->param.rc.i_rc_method == X264_RC_ABR )
553 x264_log( h, X264_LOG_WARNING, "VBV maxrate unspecified, assuming CBR\n" );
554 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate;
558 x264_log( h, X264_LOG_WARNING, "VBV bufsize set but maxrate unspecified, ignored\n" );
559 h->param.rc.i_vbv_buffer_size = 0;
562 else if( h->param.rc.i_vbv_max_bitrate < h->param.rc.i_bitrate &&
563 h->param.rc.i_rc_method == X264_RC_ABR )
565 x264_log( h, X264_LOG_WARNING, "max bitrate less than average bitrate, assuming CBR\n" );
566 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate;
569 else if( h->param.rc.i_vbv_max_bitrate )
571 x264_log( h, X264_LOG_WARNING, "VBV maxrate specified, but no bufsize, ignored\n" );
572 h->param.rc.i_vbv_max_bitrate = 0;
575 h->param.i_slice_max_size = X264_MAX( h->param.i_slice_max_size, 0 );
576 h->param.i_slice_max_mbs = X264_MAX( h->param.i_slice_max_mbs, 0 );
578 h->param.b_interlaced = !!h->param.b_interlaced;
579 int max_slices = (h->param.i_height+((16<<h->param.b_interlaced)-1))/(16<<h->param.b_interlaced);
580 if( h->param.b_sliced_threads )
581 h->param.i_slice_count = x264_clip3( h->param.i_threads, 0, max_slices );
584 h->param.i_slice_count = x264_clip3( h->param.i_slice_count, 0, max_slices );
585 if( h->param.i_slice_max_mbs || h->param.i_slice_max_size )
586 h->param.i_slice_count = 0;
589 if( h->param.b_bluray_compat )
591 h->param.i_bframe_pyramid = X264_MIN( X264_B_PYRAMID_STRICT, h->param.i_bframe_pyramid );
592 h->param.i_bframe = X264_MIN( h->param.i_bframe, 3 );
594 h->param.i_nal_hrd = X264_MAX( h->param.i_nal_hrd, X264_NAL_HRD_VBR );
595 h->param.i_slice_max_size = 0;
596 h->param.i_slice_max_mbs = 0;
597 h->param.b_intra_refresh = 0;
598 h->param.i_frame_reference = X264_MIN( h->param.i_frame_reference, 6 );
599 h->param.i_dpb_size = X264_MIN( h->param.i_dpb_size, 6 );
600 /* Due to the proliferation of broken players that don't handle dupes properly. */
601 h->param.analyse.i_weighted_pred = X264_MIN( h->param.analyse.i_weighted_pred, X264_WEIGHTP_SIMPLE );
602 if( h->param.b_fake_interlaced )
603 h->param.b_pic_struct = 1;
606 h->param.i_frame_reference = x264_clip3( h->param.i_frame_reference, 1, X264_REF_MAX );
607 h->param.i_dpb_size = x264_clip3( h->param.i_dpb_size, 1, X264_REF_MAX );
608 if( h->param.i_scenecut_threshold < 0 )
609 h->param.i_scenecut_threshold = 0;
610 h->param.analyse.i_direct_mv_pred = x264_clip3( h->param.analyse.i_direct_mv_pred, X264_DIRECT_PRED_NONE, X264_DIRECT_PRED_AUTO );
611 if( !h->param.analyse.i_subpel_refine && h->param.analyse.i_direct_mv_pred > X264_DIRECT_PRED_SPATIAL )
613 x264_log( h, X264_LOG_WARNING, "subme=0 + direct=temporal is not supported\n" );
614 h->param.analyse.i_direct_mv_pred = X264_DIRECT_PRED_SPATIAL;
616 h->param.i_bframe = x264_clip3( h->param.i_bframe, 0, X264_MIN( X264_BFRAME_MAX, h->param.i_keyint_max-1 ) );
617 h->param.i_bframe_bias = x264_clip3( h->param.i_bframe_bias, -90, 100 );
618 if( h->param.i_bframe <= 1 )
619 h->param.i_bframe_pyramid = X264_B_PYRAMID_NONE;
620 h->param.i_bframe_pyramid = x264_clip3( h->param.i_bframe_pyramid, X264_B_PYRAMID_NONE, X264_B_PYRAMID_NORMAL );
621 h->param.i_bframe_adaptive = x264_clip3( h->param.i_bframe_adaptive, X264_B_ADAPT_NONE, X264_B_ADAPT_TRELLIS );
622 if( !h->param.i_bframe )
624 h->param.i_bframe_adaptive = X264_B_ADAPT_NONE;
625 h->param.analyse.i_direct_mv_pred = 0;
626 h->param.analyse.b_weighted_bipred = 0;
627 h->param.b_open_gop = 0;
629 if( h->param.b_intra_refresh && h->param.i_bframe_pyramid == X264_B_PYRAMID_NORMAL )
631 x264_log( h, X264_LOG_WARNING, "b-pyramid normal + intra-refresh is not supported\n" );
632 h->param.i_bframe_pyramid = X264_B_PYRAMID_STRICT;
634 if( h->param.b_intra_refresh && (h->param.i_frame_reference > 1 || h->param.i_dpb_size > 1) )
636 x264_log( h, X264_LOG_WARNING, "ref > 1 + intra-refresh is not supported\n" );
637 h->param.i_frame_reference = 1;
638 h->param.i_dpb_size = 1;
640 if( h->param.b_intra_refresh && h->param.b_open_gop )
642 x264_log( h, X264_LOG_WARNING, "intra-refresh is not compatible with open-gop\n" );
643 h->param.b_open_gop = 0;
645 float fps = h->param.i_fps_num > 0 && h->param.i_fps_den > 0 ? (float) h->param.i_fps_num / h->param.i_fps_den : 25.0;
646 if( h->param.i_keyint_min == X264_KEYINT_MIN_AUTO )
647 h->param.i_keyint_min = X264_MIN( h->param.i_keyint_max / 10, fps );
648 h->param.i_keyint_min = x264_clip3( h->param.i_keyint_min, 1, h->param.i_keyint_max/2+1 );
649 h->param.rc.i_lookahead = x264_clip3( h->param.rc.i_lookahead, 0, X264_LOOKAHEAD_MAX );
651 int maxrate = X264_MAX( h->param.rc.i_vbv_max_bitrate, h->param.rc.i_bitrate );
652 float bufsize = maxrate ? (float)h->param.rc.i_vbv_buffer_size / maxrate : 0;
653 h->param.rc.i_lookahead = X264_MIN( h->param.rc.i_lookahead, X264_MAX( h->param.i_keyint_max, bufsize*fps ) );
656 if( !h->param.i_timebase_num || !h->param.i_timebase_den || !(h->param.b_vfr_input || h->param.b_pulldown) )
658 h->param.i_timebase_num = h->param.i_fps_den;
659 h->param.i_timebase_den = h->param.i_fps_num;
662 h->param.rc.f_qcompress = x264_clip3f( h->param.rc.f_qcompress, 0.0, 1.0 );
663 if( h->param.i_keyint_max == 1 || h->param.rc.f_qcompress == 1 )
664 h->param.rc.b_mb_tree = 0;
665 if( (!h->param.b_intra_refresh && h->param.i_keyint_max != X264_KEYINT_MAX_INFINITE) &&
666 !h->param.rc.i_lookahead && h->param.rc.b_mb_tree )
668 x264_log( h, X264_LOG_WARNING, "lookaheadless mb-tree requires intra refresh or infinite keyint\n" );
669 h->param.rc.b_mb_tree = 0;
671 if( h->param.rc.b_stat_read )
672 h->param.rc.i_lookahead = 0;
674 if( h->param.i_sync_lookahead < 0 )
675 h->param.i_sync_lookahead = h->param.i_bframe + 1;
676 h->param.i_sync_lookahead = X264_MIN( h->param.i_sync_lookahead, X264_LOOKAHEAD_MAX );
677 if( h->param.rc.b_stat_read || h->i_thread_frames == 1 )
678 h->param.i_sync_lookahead = 0;
680 h->param.i_sync_lookahead = 0;
683 h->param.i_deblocking_filter_alphac0 = x264_clip3( h->param.i_deblocking_filter_alphac0, -6, 6 );
684 h->param.i_deblocking_filter_beta = x264_clip3( h->param.i_deblocking_filter_beta, -6, 6 );
685 h->param.analyse.i_luma_deadzone[0] = x264_clip3( h->param.analyse.i_luma_deadzone[0], 0, 32 );
686 h->param.analyse.i_luma_deadzone[1] = x264_clip3( h->param.analyse.i_luma_deadzone[1], 0, 32 );
688 h->param.i_cabac_init_idc = x264_clip3( h->param.i_cabac_init_idc, 0, 2 );
690 if( h->param.i_cqm_preset < X264_CQM_FLAT || h->param.i_cqm_preset > X264_CQM_CUSTOM )
691 h->param.i_cqm_preset = X264_CQM_FLAT;
693 if( h->param.analyse.i_me_method < X264_ME_DIA ||
694 h->param.analyse.i_me_method > X264_ME_TESA )
695 h->param.analyse.i_me_method = X264_ME_HEX;
696 h->param.analyse.i_me_range = x264_clip3( h->param.analyse.i_me_range, 4, 1024 );
697 if( h->param.analyse.i_me_range > 16 && h->param.analyse.i_me_method <= X264_ME_HEX )
698 h->param.analyse.i_me_range = 16;
699 if( h->param.analyse.i_me_method == X264_ME_TESA &&
700 (h->mb.b_lossless || h->param.analyse.i_subpel_refine <= 1) )
701 h->param.analyse.i_me_method = X264_ME_ESA;
702 h->param.analyse.b_mixed_references = h->param.analyse.b_mixed_references && h->param.i_frame_reference > 1;
703 h->param.analyse.inter &= X264_ANALYSE_PSUB16x16|X264_ANALYSE_PSUB8x8|X264_ANALYSE_BSUB16x16|
704 X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
705 h->param.analyse.intra &= X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
706 if( !(h->param.analyse.inter & X264_ANALYSE_PSUB16x16) )
707 h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
708 if( !h->param.analyse.b_transform_8x8 )
710 h->param.analyse.inter &= ~X264_ANALYSE_I8x8;
711 h->param.analyse.intra &= ~X264_ANALYSE_I8x8;
713 h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -12, 12);
714 h->param.analyse.i_trellis = x264_clip3( h->param.analyse.i_trellis, 0, 2 );
715 h->param.rc.i_aq_mode = x264_clip3( h->param.rc.i_aq_mode, 0, 2 );
716 h->param.rc.f_aq_strength = x264_clip3f( h->param.rc.f_aq_strength, 0, 3 );
717 if( h->param.rc.f_aq_strength == 0 )
718 h->param.rc.i_aq_mode = 0;
720 if( h->param.i_log_level < X264_LOG_INFO )
722 h->param.analyse.b_psnr = 0;
723 h->param.analyse.b_ssim = 0;
725 /* Warn users trying to measure PSNR/SSIM with psy opts on. */
726 if( b_open && (h->param.analyse.b_psnr || h->param.analyse.b_ssim) )
730 if( h->param.analyse.b_psy )
732 s = h->param.analyse.b_psnr ? "psnr" : "ssim";
733 x264_log( h, X264_LOG_WARNING, "--%s used with psy on: results will be invalid!\n", s );
735 else if( !h->param.rc.i_aq_mode && h->param.analyse.b_ssim )
737 x264_log( h, X264_LOG_WARNING, "--ssim used with AQ off: results will be invalid!\n" );
740 else if( h->param.rc.i_aq_mode && h->param.analyse.b_psnr )
742 x264_log( h, X264_LOG_WARNING, "--psnr used with AQ on: results will be invalid!\n" );
746 x264_log( h, X264_LOG_WARNING, "--tune %s should be used if attempting to benchmark %s!\n", s, s );
749 if( !h->param.analyse.b_psy )
751 h->param.analyse.f_psy_rd = 0;
752 h->param.analyse.f_psy_trellis = 0;
754 h->param.analyse.f_psy_rd = x264_clip3f( h->param.analyse.f_psy_rd, 0, 10 );
755 h->param.analyse.f_psy_trellis = x264_clip3f( h->param.analyse.f_psy_trellis, 0, 10 );
756 h->mb.i_psy_rd = h->param.analyse.i_subpel_refine >= 6 ? FIX8( h->param.analyse.f_psy_rd ) : 0;
757 h->mb.i_psy_trellis = h->param.analyse.i_trellis ? FIX8( h->param.analyse.f_psy_trellis / 4 ) : 0;
758 /* Psy RDO increases overall quantizers to improve the quality of luma--this indirectly hurts chroma quality */
759 /* so we lower the chroma QP offset to compensate */
760 if( b_open && h->mb.i_psy_rd )
761 h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_rd < 0.25 ? 1 : 2;
762 /* Psy trellis has a similar effect. */
763 if( b_open && h->mb.i_psy_trellis )
764 h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_trellis < 0.25 ? 1 : 2;
765 h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -12, 12);
766 /* MB-tree requires AQ to be on, even if the strength is zero. */
767 if( !h->param.rc.i_aq_mode && h->param.rc.b_mb_tree )
769 h->param.rc.i_aq_mode = 1;
770 h->param.rc.f_aq_strength = 0;
772 h->param.analyse.i_noise_reduction = x264_clip3( h->param.analyse.i_noise_reduction, 0, 1<<16 );
773 if( h->param.analyse.i_subpel_refine == 10 && (h->param.analyse.i_trellis != 2 || !h->param.rc.i_aq_mode) )
774 h->param.analyse.i_subpel_refine = 9;
777 const x264_level_t *l = x264_levels;
778 if( h->param.i_level_idc < 0 )
780 int maxrate_bak = h->param.rc.i_vbv_max_bitrate;
781 if( h->param.rc.i_rc_method == X264_RC_ABR && h->param.rc.i_vbv_buffer_size <= 0 )
782 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate * 2;
783 h->sps = h->sps_array;
784 x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
785 do h->param.i_level_idc = l->level_idc;
786 while( l[1].level_idc && x264_validate_levels( h, 0 ) && l++ );
787 h->param.rc.i_vbv_max_bitrate = maxrate_bak;
791 while( l->level_idc && l->level_idc != h->param.i_level_idc )
793 if( l->level_idc == 0 )
795 x264_log( h, X264_LOG_ERROR, "invalid level_idc: %d\n", h->param.i_level_idc );
799 if( h->param.analyse.i_mv_range <= 0 )
800 h->param.analyse.i_mv_range = l->mv_range >> h->param.b_interlaced;
802 h->param.analyse.i_mv_range = x264_clip3(h->param.analyse.i_mv_range, 32, 512 >> h->param.b_interlaced);
805 h->param.analyse.i_weighted_pred = x264_clip3( h->param.analyse.i_weighted_pred, X264_WEIGHTP_NONE, X264_WEIGHTP_SMART );
807 if( h->param.b_interlaced )
809 if( h->param.analyse.i_me_method >= X264_ME_ESA )
811 x264_log( h, X264_LOG_WARNING, "interlace + me=esa is not implemented\n" );
812 h->param.analyse.i_me_method = X264_ME_UMH;
814 if( h->param.analyse.i_weighted_pred > 0 )
816 x264_log( h, X264_LOG_WARNING, "interlace + weightp is not implemented\n" );
817 h->param.analyse.i_weighted_pred = X264_WEIGHTP_NONE;
821 if( !h->param.analyse.i_weighted_pred && h->param.rc.b_mb_tree && h->param.analyse.b_psy )
822 h->param.analyse.i_weighted_pred = X264_WEIGHTP_FAKE;
824 if( h->i_thread_frames > 1 )
826 int r = h->param.analyse.i_mv_range_thread;
830 // half of the available space is reserved and divided evenly among the threads,
831 // the rest is allocated to whichever thread is far enough ahead to use it.
832 // reserving more space increases quality for some videos, but costs more time
833 // in thread synchronization.
834 int max_range = (h->param.i_height + X264_THREAD_HEIGHT) / h->i_thread_frames - X264_THREAD_HEIGHT;
837 r = X264_MAX( r, h->param.analyse.i_me_range );
838 r = X264_MIN( r, h->param.analyse.i_mv_range );
839 // round up to use the whole mb row
840 r2 = (r & ~15) + ((-X264_THREAD_HEIGHT) & 15);
843 x264_log( h, X264_LOG_DEBUG, "using mv_range_thread = %d\n", r2 );
844 h->param.analyse.i_mv_range_thread = r2;
847 if( h->param.rc.f_rate_tolerance < 0 )
848 h->param.rc.f_rate_tolerance = 0;
849 if( h->param.rc.f_qblur < 0 )
850 h->param.rc.f_qblur = 0;
851 if( h->param.rc.f_complexity_blur < 0 )
852 h->param.rc.f_complexity_blur = 0;
854 h->param.i_sps_id &= 31;
856 if( h->param.b_interlaced )
857 h->param.b_pic_struct = 1;
859 h->param.i_nal_hrd = x264_clip3( h->param.i_nal_hrd, X264_NAL_HRD_NONE, X264_NAL_HRD_CBR );
861 if( h->param.i_nal_hrd && !h->param.rc.i_vbv_buffer_size )
863 x264_log( h, X264_LOG_WARNING, "NAL HRD parameters require VBV parameters\n" );
864 h->param.i_nal_hrd = X264_NAL_HRD_NONE;
867 if( h->param.i_nal_hrd == X264_NAL_HRD_CBR &&
868 (h->param.rc.i_bitrate != h->param.rc.i_vbv_max_bitrate || !h->param.rc.i_vbv_max_bitrate) )
870 x264_log( h, X264_LOG_WARNING, "CBR HRD requires constant bitrate\n" );
871 h->param.i_nal_hrd = X264_NAL_HRD_VBR;
874 /* ensure the booleans are 0 or 1 so they can be used in math */
875 #define BOOLIFY(x) h->param.x = !!h->param.x
877 BOOLIFY( b_constrained_intra );
878 BOOLIFY( b_deblocking_filter );
879 BOOLIFY( b_deterministic );
880 BOOLIFY( b_sliced_threads );
881 BOOLIFY( b_interlaced );
882 BOOLIFY( b_intra_refresh );
883 BOOLIFY( b_visualize );
885 BOOLIFY( b_repeat_headers );
887 BOOLIFY( b_vfr_input );
888 BOOLIFY( b_pulldown );
890 BOOLIFY( b_pic_struct );
891 BOOLIFY( b_fake_interlaced );
892 BOOLIFY( b_open_gop );
893 BOOLIFY( b_bluray_compat );
894 BOOLIFY( analyse.b_transform_8x8 );
895 BOOLIFY( analyse.b_weighted_bipred );
896 BOOLIFY( analyse.b_chroma_me );
897 BOOLIFY( analyse.b_mixed_references );
898 BOOLIFY( analyse.b_fast_pskip );
899 BOOLIFY( analyse.b_dct_decimate );
900 BOOLIFY( analyse.b_psy );
901 BOOLIFY( analyse.b_psnr );
902 BOOLIFY( analyse.b_ssim );
903 BOOLIFY( rc.b_stat_write );
904 BOOLIFY( rc.b_stat_read );
905 BOOLIFY( rc.b_mb_tree );
911 static void mbcmp_init( x264_t *h )
913 int satd = !h->mb.b_lossless && h->param.analyse.i_subpel_refine > 1;
914 memcpy( h->pixf.mbcmp, satd ? h->pixf.satd : h->pixf.sad_aligned, sizeof(h->pixf.mbcmp) );
915 memcpy( h->pixf.mbcmp_unaligned, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.mbcmp_unaligned) );
916 h->pixf.intra_mbcmp_x3_16x16 = satd ? h->pixf.intra_satd_x3_16x16 : h->pixf.intra_sad_x3_16x16;
917 h->pixf.intra_mbcmp_x3_8x8c = satd ? h->pixf.intra_satd_x3_8x8c : h->pixf.intra_sad_x3_8x8c;
918 h->pixf.intra_mbcmp_x3_8x8 = satd ? h->pixf.intra_sa8d_x3_8x8 : h->pixf.intra_sad_x3_8x8;
919 h->pixf.intra_mbcmp_x3_4x4 = satd ? h->pixf.intra_satd_x3_4x4 : h->pixf.intra_sad_x3_4x4;
920 satd &= h->param.analyse.i_me_method == X264_ME_TESA;
921 memcpy( h->pixf.fpelcmp, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.fpelcmp) );
922 memcpy( h->pixf.fpelcmp_x3, satd ? h->pixf.satd_x3 : h->pixf.sad_x3, sizeof(h->pixf.fpelcmp_x3) );
923 memcpy( h->pixf.fpelcmp_x4, satd ? h->pixf.satd_x4 : h->pixf.sad_x4, sizeof(h->pixf.fpelcmp_x4) );
926 static void x264_set_aspect_ratio( x264_t *h, x264_param_t *param, int initial )
929 if( param->vui.i_sar_width > 0 && param->vui.i_sar_height > 0 )
931 uint32_t i_w = param->vui.i_sar_width;
932 uint32_t i_h = param->vui.i_sar_height;
933 uint32_t old_w = h->param.vui.i_sar_width;
934 uint32_t old_h = h->param.vui.i_sar_height;
936 x264_reduce_fraction( &i_w, &i_h );
938 while( i_w > 65535 || i_h > 65535 )
944 x264_reduce_fraction( &i_w, &i_h );
946 if( i_w != old_w || i_h != old_h || initial )
948 h->param.vui.i_sar_width = 0;
949 h->param.vui.i_sar_height = 0;
950 if( i_w == 0 || i_h == 0 )
951 x264_log( h, X264_LOG_WARNING, "cannot create valid sample aspect ratio\n" );
954 x264_log( h, initial?X264_LOG_INFO:X264_LOG_DEBUG, "using SAR=%d/%d\n", i_w, i_h );
955 h->param.vui.i_sar_width = i_w;
956 h->param.vui.i_sar_height = i_h;
958 x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
963 /****************************************************************************
965 ****************************************************************************/
966 x264_t *x264_encoder_open( x264_param_t *param )
970 int qp, i_slicetype_length;
972 CHECKED_MALLOCZERO( h, sizeof(x264_t) );
974 /* Create a copy of param */
975 memcpy( &h->param, param, sizeof(x264_param_t) );
977 if( param->param_free )
978 param->param_free( param );
980 if( x264_threading_init() )
982 x264_log( h, X264_LOG_ERROR, "unable to initialize threading\n" );
986 if( x264_validate_parameters( h, 1 ) < 0 )
989 if( h->param.psz_cqm_file )
990 if( x264_cqm_parse_file( h, h->param.psz_cqm_file ) < 0 )
993 if( h->param.rc.psz_stat_out )
994 h->param.rc.psz_stat_out = strdup( h->param.rc.psz_stat_out );
995 if( h->param.rc.psz_stat_in )
996 h->param.rc.psz_stat_in = strdup( h->param.rc.psz_stat_in );
998 x264_reduce_fraction( &h->param.i_fps_num, &h->param.i_fps_den );
999 x264_reduce_fraction( &h->param.i_timebase_num, &h->param.i_timebase_den );
1004 h->i_idr_pic_id = 0;
1006 if( (uint64_t)h->param.i_timebase_den * 2 > UINT32_MAX )
1008 x264_log( h, X264_LOG_ERROR, "Effective timebase denominator %u exceeds H.264 maximum\n", h->param.i_timebase_den );
1012 h->sps = &h->sps_array[0];
1013 x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
1015 h->pps = &h->pps_array[0];
1016 x264_pps_init( h->pps, h->param.i_sps_id, &h->param, h->sps );
1018 x264_set_aspect_ratio( h, &h->param, 1 );
1020 x264_validate_levels( h, 1 );
1022 h->chroma_qp_table = i_chroma_qp_table + 12 + h->pps->i_chroma_qp_index_offset;
1024 if( x264_cqm_init( h ) < 0 )
1027 h->mb.i_mb_width = h->sps->i_mb_width;
1028 h->mb.i_mb_height = h->sps->i_mb_height;
1029 h->mb.i_mb_count = h->mb.i_mb_width * h->mb.i_mb_height;
1030 /* Adaptive MBAFF and subme 0 are not supported as we require halving motion
1031 * vectors during prediction, resulting in hpel mvs.
1032 * The chosen solution is to make MBAFF non-adaptive in this case. */
1033 h->mb.b_adaptive_mbaff = h->param.b_interlaced && h->param.analyse.i_subpel_refine;
1036 if( h->param.i_bframe_adaptive == X264_B_ADAPT_TRELLIS && !h->param.rc.b_stat_read )
1037 h->frames.i_delay = X264_MAX(h->param.i_bframe,3)*4;
1039 h->frames.i_delay = h->param.i_bframe;
1040 if( h->param.rc.b_mb_tree || h->param.rc.i_vbv_buffer_size )
1041 h->frames.i_delay = X264_MAX( h->frames.i_delay, h->param.rc.i_lookahead );
1042 i_slicetype_length = h->frames.i_delay;
1043 h->frames.i_delay += h->i_thread_frames - 1;
1044 h->frames.i_delay += h->param.i_sync_lookahead;
1045 h->frames.i_delay += h->param.b_vfr_input;
1046 h->frames.i_bframe_delay = h->param.i_bframe ? (h->param.i_bframe_pyramid ? 2 : 1) : 0;
1048 h->frames.i_max_ref0 = h->param.i_frame_reference;
1049 h->frames.i_max_ref1 = X264_MIN( h->sps->vui.i_num_reorder_frames, h->param.i_frame_reference );
1050 h->frames.i_max_dpb = h->sps->vui.i_max_dec_frame_buffering;
1051 h->frames.b_have_lowres = !h->param.rc.b_stat_read
1052 && ( h->param.rc.i_rc_method == X264_RC_ABR
1053 || h->param.rc.i_rc_method == X264_RC_CRF
1054 || h->param.i_bframe_adaptive
1055 || h->param.i_scenecut_threshold
1056 || h->param.rc.b_mb_tree
1057 || h->param.analyse.i_weighted_pred );
1058 h->frames.b_have_lowres |= h->param.rc.b_stat_read && h->param.rc.i_vbv_buffer_size > 0;
1059 h->frames.b_have_sub8x8_esa = !!(h->param.analyse.inter & X264_ANALYSE_PSUB8x8);
1061 h->frames.i_last_idr =
1062 h->frames.i_last_keyframe = - h->param.i_keyint_max;
1063 h->frames.i_input = 0;
1064 h->frames.i_largest_pts = h->frames.i_second_largest_pts = -1;
1065 h->frames.i_poc_last_open_gop = -1;
1067 CHECKED_MALLOCZERO( h->frames.unused[0], (h->frames.i_delay + 3) * sizeof(x264_frame_t *) );
1068 /* Allocate room for max refs plus a few extra just in case. */
1069 CHECKED_MALLOCZERO( h->frames.unused[1], (h->i_thread_frames + X264_REF_MAX + 4) * sizeof(x264_frame_t *) );
1070 CHECKED_MALLOCZERO( h->frames.current, (h->param.i_sync_lookahead + h->param.i_bframe
1071 + h->i_thread_frames + 3) * sizeof(x264_frame_t *) );
1072 if( h->param.analyse.i_weighted_pred > 0 )
1073 CHECKED_MALLOCZERO( h->frames.blank_unused, h->i_thread_frames * 4 * sizeof(x264_frame_t *) );
1074 h->i_ref[0] = h->i_ref[1] = 0;
1075 h->i_cpb_delay = h->i_coded_fields = h->i_disp_fields = 0;
1076 h->i_prev_duration = ((uint64_t)h->param.i_fps_den * h->sps->vui.i_time_scale) / ((uint64_t)h->param.i_fps_num * h->sps->vui.i_num_units_in_tick);
1077 h->i_disp_fields_last_frame = -1;
1080 /* init CPU functions */
1081 x264_predict_16x16_init( h->param.cpu, h->predict_16x16 );
1082 x264_predict_8x8c_init( h->param.cpu, h->predict_8x8c );
1083 x264_predict_8x8_init( h->param.cpu, h->predict_8x8, &h->predict_8x8_filter );
1084 x264_predict_4x4_init( h->param.cpu, h->predict_4x4 );
1085 if( h->param.b_cabac )
1089 x264_pixel_init( h->param.cpu, &h->pixf );
1090 x264_dct_init( h->param.cpu, &h->dctf );
1091 x264_zigzag_init( h->param.cpu, &h->zigzagf, h->param.b_interlaced );
1092 x264_mc_init( h->param.cpu, &h->mc );
1093 x264_quant_init( h, h->param.cpu, &h->quantf );
1094 x264_deblock_init( h->param.cpu, &h->loopf, h->param.b_interlaced );
1095 x264_bitstream_init( h->param.cpu, &h->bsf );
1096 x264_dct_init_weights();
1100 p = buf + sprintf( buf, "using cpu capabilities:" );
1101 for( int i = 0; x264_cpu_names[i].flags; i++ )
1103 if( !strcmp(x264_cpu_names[i].name, "SSE2")
1104 && h->param.cpu & (X264_CPU_SSE2_IS_FAST|X264_CPU_SSE2_IS_SLOW) )
1106 if( !strcmp(x264_cpu_names[i].name, "SSE3")
1107 && (h->param.cpu & X264_CPU_SSSE3 || !(h->param.cpu & X264_CPU_CACHELINE_64)) )
1109 if( !strcmp(x264_cpu_names[i].name, "SSE4.1")
1110 && (h->param.cpu & X264_CPU_SSE42) )
1112 if( (h->param.cpu & x264_cpu_names[i].flags) == x264_cpu_names[i].flags
1113 && (!i || x264_cpu_names[i].flags != x264_cpu_names[i-1].flags) )
1114 p += sprintf( p, " %s", x264_cpu_names[i].name );
1117 p += sprintf( p, " none!" );
1118 x264_log( h, X264_LOG_INFO, "%s\n", buf );
1120 float *logs = x264_analyse_prepare_costs( h );
1123 for( qp = X264_MIN( h->param.rc.i_qp_min, QP_MAX_SPEC ); qp <= h->param.rc.i_qp_max; qp++ )
1124 if( x264_analyse_init_costs( h, logs, qp ) )
1126 if( x264_analyse_init_costs( h, logs, X264_LOOKAHEAD_QP ) )
1130 static const uint16_t cost_mv_correct[7] = { 24, 47, 95, 189, 379, 757, 1515 };
1131 /* Checks for known miscompilation issues. */
1132 if( h->cost_mv[X264_LOOKAHEAD_QP][2013] != cost_mv_correct[BIT_DEPTH-8] )
1134 x264_log( h, X264_LOG_ERROR, "MV cost test failed: x264 has been miscompiled!\n" );
1138 /* Must be volatile or else GCC will optimize it out. */
1139 volatile int temp = 392;
1140 if( x264_clz( temp ) != 23 )
1142 x264_log( h, X264_LOG_ERROR, "CLZ test failed: x264 has been miscompiled!\n" );
1143 #if ARCH_X86 || ARCH_X86_64
1144 x264_log( h, X264_LOG_ERROR, "Are you attempting to run an SSE4a-targeted build on a CPU that\n" );
1145 x264_log( h, X264_LOG_ERROR, "doesn't support it?\n" );
1151 h->out.i_bitstream = X264_MAX( 1000000, h->param.i_width * h->param.i_height * 4
1152 * ( h->param.rc.i_rc_method == X264_RC_ABR ? pow( 0.95, h->param.rc.i_qp_min )
1153 : pow( 0.95, h->param.rc.i_qp_constant ) * X264_MAX( 1, h->param.rc.f_ip_factor )));
1155 h->nal_buffer_size = h->out.i_bitstream * 3/2 + 4;
1156 CHECKED_MALLOC( h->nal_buffer, h->nal_buffer_size );
1158 if( h->param.i_threads > 1 &&
1159 x264_threadpool_init( &h->threadpool, h->param.i_threads, (void*)x264_encoder_thread_init, h ) )
1163 for( int i = 1; i < h->param.i_threads + !!h->param.i_sync_lookahead; i++ )
1164 CHECKED_MALLOC( h->thread[i], sizeof(x264_t) );
1166 for( int i = 0; i < h->param.i_threads; i++ )
1168 int init_nal_count = h->param.i_slice_count + 3;
1169 int allocate_threadlocal_data = !h->param.b_sliced_threads || !i;
1173 if( allocate_threadlocal_data )
1175 h->thread[i]->fdec = x264_frame_pop_unused( h, 1 );
1176 if( !h->thread[i]->fdec )
1180 h->thread[i]->fdec = h->thread[0]->fdec;
1182 h->thread[i]->sps = &h->thread[i]->sps_array[0];
1183 h->thread[i]->pps = &h->thread[i]->pps_array[0];
1185 CHECKED_MALLOC( h->thread[i]->out.p_bitstream, h->out.i_bitstream );
1186 /* Start each thread with room for init_nal_count NAL units; it'll realloc later if needed. */
1187 CHECKED_MALLOC( h->thread[i]->out.nal, init_nal_count*sizeof(x264_nal_t) );
1188 h->thread[i]->out.i_nals_allocated = init_nal_count;
1190 if( allocate_threadlocal_data && x264_macroblock_cache_allocate( h->thread[i] ) < 0 )
1194 if( x264_lookahead_init( h, i_slicetype_length ) )
1197 for( int i = 0; i < h->param.i_threads; i++ )
1198 if( x264_macroblock_thread_allocate( h->thread[i], 0 ) < 0 )
1201 if( x264_ratecontrol_new( h ) < 0 )
1204 if( h->param.i_nal_hrd )
1206 x264_log( h, X264_LOG_DEBUG, "HRD bitrate: %i bits/sec\n", h->sps->vui.hrd.i_bit_rate_unscaled );
1207 x264_log( h, X264_LOG_DEBUG, "CPB size: %i bits\n", h->sps->vui.hrd.i_cpb_size_unscaled );
1210 if( h->param.psz_dump_yuv )
1212 /* create or truncate the reconstructed video file */
1213 FILE *f = fopen( h->param.psz_dump_yuv, "w" );
1216 x264_log( h, X264_LOG_ERROR, "dump_yuv: can't write to %s\n", h->param.psz_dump_yuv );
1219 else if( !x264_is_regular_file( f ) )
1221 x264_log( h, X264_LOG_ERROR, "dump_yuv: incompatible with non-regular file %s\n", h->param.psz_dump_yuv );
1227 const char *profile = h->sps->i_profile_idc == PROFILE_BASELINE ? "Constrained Baseline" :
1228 h->sps->i_profile_idc == PROFILE_MAIN ? "Main" :
1229 h->sps->i_profile_idc == PROFILE_HIGH ? "High" :
1230 h->sps->i_profile_idc == PROFILE_HIGH10 ? (h->sps->b_constraint_set3 == 1 ? "High 10 Intra" : "High 10") :
1231 "High 4:4:4 Predictive";
1233 snprintf( level, sizeof(level), "%d.%d", h->sps->i_level_idc/10, h->sps->i_level_idc%10 );
1234 if( h->sps->i_level_idc == 9 || ( h->sps->i_level_idc == 11 && h->sps->b_constraint_set3 ) )
1235 strcpy( level, "1b" );
1237 if( h->sps->i_profile_idc < PROFILE_HIGH10 )
1239 x264_log( h, X264_LOG_INFO, "profile %s, level %s\n",
1244 x264_log( h, X264_LOG_INFO, "profile %s, level %s, bit depth %d\n",
1245 profile, level, BIT_DEPTH );
1254 /****************************************************************************
1255 * x264_encoder_reconfig:
1256 ****************************************************************************/
1257 int x264_encoder_reconfig( x264_t *h, x264_param_t *param )
1259 int rc_reconfig = 0;
1260 h = h->thread[h->thread[0]->i_thread_phase];
1261 x264_set_aspect_ratio( h, param, 0 );
1262 #define COPY(var) h->param.var = param->var
1263 COPY( i_frame_reference ); // but never uses more refs than initially specified
1264 COPY( i_bframe_bias );
1265 if( h->param.i_scenecut_threshold )
1266 COPY( i_scenecut_threshold ); // can't turn it on or off, only vary the threshold
1267 COPY( b_deblocking_filter );
1268 COPY( i_deblocking_filter_alphac0 );
1269 COPY( i_deblocking_filter_beta );
1270 COPY( i_frame_packing );
1271 COPY( analyse.inter );
1272 COPY( analyse.intra );
1273 COPY( analyse.i_direct_mv_pred );
1274 /* Scratch buffer prevents me_range from being increased for esa/tesa */
1275 if( h->param.analyse.i_me_method < X264_ME_ESA || param->analyse.i_me_range < h->param.analyse.i_me_range )
1276 COPY( analyse.i_me_range );
1277 COPY( analyse.i_noise_reduction );
1278 /* We can't switch out of subme=0 during encoding. */
1279 if( h->param.analyse.i_subpel_refine )
1280 COPY( analyse.i_subpel_refine );
1281 COPY( analyse.i_trellis );
1282 COPY( analyse.b_chroma_me );
1283 COPY( analyse.b_dct_decimate );
1284 COPY( analyse.b_fast_pskip );
1285 COPY( analyse.b_mixed_references );
1286 COPY( analyse.f_psy_rd );
1287 COPY( analyse.f_psy_trellis );
1289 // can only twiddle these if they were enabled to begin with:
1290 if( h->param.analyse.i_me_method >= X264_ME_ESA || param->analyse.i_me_method < X264_ME_ESA )
1291 COPY( analyse.i_me_method );
1292 if( h->param.analyse.i_me_method >= X264_ME_ESA && !h->frames.b_have_sub8x8_esa )
1293 h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
1294 if( h->pps->b_transform_8x8_mode )
1295 COPY( analyse.b_transform_8x8 );
1296 if( h->frames.i_max_ref1 > 1 )
1297 COPY( i_bframe_pyramid );
1298 COPY( i_slice_max_size );
1299 COPY( i_slice_max_mbs );
1300 COPY( i_slice_count );
1303 /* VBV can't be turned on if it wasn't on to begin with */
1304 if( h->param.rc.i_vbv_max_bitrate > 0 && h->param.rc.i_vbv_buffer_size > 0 &&
1305 param->rc.i_vbv_max_bitrate > 0 && param->rc.i_vbv_buffer_size > 0 )
1307 rc_reconfig |= h->param.rc.i_vbv_max_bitrate != param->rc.i_vbv_max_bitrate;
1308 rc_reconfig |= h->param.rc.i_vbv_buffer_size != param->rc.i_vbv_buffer_size;
1309 rc_reconfig |= h->param.rc.i_bitrate != param->rc.i_bitrate;
1310 COPY( rc.i_vbv_max_bitrate );
1311 COPY( rc.i_vbv_buffer_size );
1312 COPY( rc.i_bitrate );
1314 rc_reconfig |= h->param.rc.f_rf_constant != param->rc.f_rf_constant;
1315 rc_reconfig |= h->param.rc.f_rf_constant_max != param->rc.f_rf_constant_max;
1316 COPY( rc.f_rf_constant );
1317 COPY( rc.f_rf_constant_max );
1322 int ret = x264_validate_parameters( h, 0 );
1324 /* Supported reconfiguration options (1-pass only):
1328 * bitrate (CBR only) */
1329 if( !ret && rc_reconfig )
1330 x264_ratecontrol_init_reconfigurable( h, 0 );
1335 /****************************************************************************
1336 * x264_encoder_parameters:
1337 ****************************************************************************/
1338 void x264_encoder_parameters( x264_t *h, x264_param_t *param )
1340 memcpy( param, &h->thread[h->i_thread_phase]->param, sizeof(x264_param_t) );
1343 /* internal usage */
1344 static void x264_nal_start( x264_t *h, int i_type, int i_ref_idc )
1346 x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1348 nal->i_ref_idc = i_ref_idc;
1349 nal->i_type = i_type;
1350 nal->b_long_startcode = 1;
1353 nal->p_payload= &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8];
1356 /* if number of allocated nals is not enough, re-allocate a larger one. */
1357 static int x264_nal_check_buffer( x264_t *h )
1359 if( h->out.i_nal >= h->out.i_nals_allocated )
1361 x264_nal_t *new_out = x264_malloc( sizeof(x264_nal_t) * (h->out.i_nals_allocated*2) );
1364 memcpy( new_out, h->out.nal, sizeof(x264_nal_t) * (h->out.i_nals_allocated) );
1365 x264_free( h->out.nal );
1366 h->out.nal = new_out;
1367 h->out.i_nals_allocated *= 2;
1372 static int x264_nal_end( x264_t *h )
1374 x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1375 nal->i_payload = &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8] - nal->p_payload;
1376 if( h->param.nalu_process )
1377 h->param.nalu_process( h, nal );
1380 return x264_nal_check_buffer( h );
1383 static int x264_encoder_encapsulate_nals( x264_t *h, int start )
1385 int nal_size = 0, previous_nal_size = 0;
1387 if( h->param.nalu_process )
1389 for( int i = start; i < h->out.i_nal; i++ )
1390 nal_size += h->out.nal[i].i_payload;
1394 for( int i = 0; i < start; i++ )
1395 previous_nal_size += h->out.nal[i].i_payload;
1397 for( int i = start; i < h->out.i_nal; i++ )
1398 nal_size += h->out.nal[i].i_payload;
1400 /* Worst-case NAL unit escaping: reallocate the buffer if it's too small. */
1401 int necessary_size = nal_size * 3/2 + h->out.i_nal * 4;
1402 if( h->nal_buffer_size < necessary_size )
1404 h->nal_buffer_size = necessary_size * 2;
1405 uint8_t *buf = x264_malloc( h->nal_buffer_size );
1408 if( previous_nal_size )
1409 memcpy( buf, h->nal_buffer, previous_nal_size );
1410 x264_free( h->nal_buffer );
1411 h->nal_buffer = buf;
1414 uint8_t *nal_buffer = h->nal_buffer + previous_nal_size;
1416 for( int i = start; i < h->out.i_nal; i++ )
1418 h->out.nal[i].b_long_startcode = !i || h->out.nal[i].i_type == NAL_SPS || h->out.nal[i].i_type == NAL_PPS;
1419 x264_nal_encode( h, nal_buffer, &h->out.nal[i] );
1420 nal_buffer += h->out.nal[i].i_payload;
1425 return nal_buffer - (h->nal_buffer + previous_nal_size);
1428 /****************************************************************************
1429 * x264_encoder_headers:
1430 ****************************************************************************/
1431 int x264_encoder_headers( x264_t *h, x264_nal_t **pp_nal, int *pi_nal )
1434 /* init bitstream context */
1436 bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
1438 /* Write SEI, SPS and PPS. */
1440 /* generate sequence parameters */
1441 x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
1442 x264_sps_write( &h->out.bs, h->sps );
1443 if( x264_nal_end( h ) )
1446 /* generate picture parameters */
1447 x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
1448 x264_pps_write( &h->out.bs, h->pps );
1449 if( x264_nal_end( h ) )
1452 /* identify ourselves */
1453 x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
1454 if( x264_sei_version_write( h, &h->out.bs ) )
1456 if( x264_nal_end( h ) )
1459 frame_size = x264_encoder_encapsulate_nals( h, 0 );
1460 if( frame_size < 0 )
1464 *pi_nal = h->out.i_nal;
1465 *pp_nal = &h->out.nal[0];
1471 /* Check to see whether we have chosen a reference list ordering different
1472 * from the standard's default. */
1473 static inline void x264_reference_check_reorder( x264_t *h )
1475 /* The reorder check doesn't check for missing frames, so just
1476 * force a reorder if one of the reference list is corrupt. */
1477 for( int i = 0; h->frames.reference[i]; i++ )
1478 if( h->frames.reference[i]->b_corrupt )
1480 h->b_ref_reorder[0] = 1;
1483 for( int list = 0; list <= (h->sh.i_type == SLICE_TYPE_B); list++ )
1484 for( int i = 0; i < h->i_ref[list] - 1; i++ )
1486 int framenum_diff = h->fref[list][i+1]->i_frame_num - h->fref[list][i]->i_frame_num;
1487 int poc_diff = h->fref[list][i+1]->i_poc - h->fref[list][i]->i_poc;
1488 /* P and B-frames use different default orders. */
1489 if( h->sh.i_type == SLICE_TYPE_P ? framenum_diff > 0 : list == 1 ? poc_diff < 0 : poc_diff > 0 )
1491 h->b_ref_reorder[list] = 1;
1497 /* return -1 on failure, else return the index of the new reference frame */
1498 int x264_weighted_reference_duplicate( x264_t *h, int i_ref, const x264_weight_t *w )
1500 int i = h->i_ref[0];
1502 x264_frame_t *newframe;
1503 if( i <= 1 ) /* empty list, definitely can't duplicate frame */
1506 //Duplication is only used in X264_WEIGHTP_SMART
1507 if( h->param.analyse.i_weighted_pred != X264_WEIGHTP_SMART )
1510 /* Duplication is a hack to compensate for crappy rounding in motion compensation.
1511 * With high bit depth, it's not worth doing, so turn it off except in the case of
1512 * unweighted dupes. */
1513 if( BIT_DEPTH > 8 && w != weight_none )
1516 newframe = x264_frame_pop_blank_unused( h );
1520 //FIXME: probably don't need to copy everything
1521 *newframe = *h->fref[0][i_ref];
1522 newframe->i_reference_count = 1;
1523 newframe->orig = h->fref[0][i_ref];
1524 newframe->b_duplicate = 1;
1525 memcpy( h->fenc->weight[j], w, sizeof(h->fenc->weight[i]) );
1527 /* shift the frames to make space for the dupe. */
1528 h->b_ref_reorder[0] = 1;
1529 if( h->i_ref[0] < X264_REF_MAX )
1531 h->fref[0][X264_REF_MAX-1] = NULL;
1532 x264_frame_unshift( &h->fref[0][j], newframe );
1537 static void x264_weighted_pred_init( x264_t *h )
1539 /* for now no analysis and set all weights to nothing */
1540 for( int i_ref = 0; i_ref < h->i_ref[0]; i_ref++ )
1541 h->fenc->weighted[i_ref] = h->fref[0][i_ref]->filtered[0];
1543 // FIXME: This only supports weighting of one reference frame
1544 // and duplicates of that frame.
1545 h->fenc->i_lines_weighted = 0;
1547 for( int i_ref = 0; i_ref < (h->i_ref[0] << h->sh.b_mbaff); i_ref++ )
1548 for( int i = 0; i < 3; i++ )
1549 h->sh.weight[i_ref][i].weightfn = NULL;
1552 if( h->sh.i_type != SLICE_TYPE_P || h->param.analyse.i_weighted_pred <= 0 )
1555 int i_padv = PADV << h->param.b_interlaced;
1557 int weightplane[2] = { 0, 0 };
1558 int buffer_next = 0;
1559 for( int i = 0; i < 3; i++ )
1561 for( int j = 0; j < h->i_ref[0]; j++ )
1563 if( h->fenc->weight[j][i].weightfn )
1565 h->sh.weight[j][i] = h->fenc->weight[j][i];
1566 // if weight is useless, don't write it to stream
1567 if( h->sh.weight[j][i].i_scale == 1<<h->sh.weight[j][i].i_denom && h->sh.weight[j][i].i_offset == 0 )
1568 h->sh.weight[j][i].weightfn = NULL;
1571 if( !weightplane[!!i] )
1573 weightplane[!!i] = 1;
1574 h->sh.weight[0][!!i].i_denom = denom = h->sh.weight[j][i].i_denom;
1575 assert( x264_clip3( denom, 0, 7 ) == denom );
1578 assert( h->sh.weight[j][i].i_denom == denom );
1581 h->fenc->weighted[j] = h->mb.p_weight_buf[buffer_next++] + h->fenc->i_stride[0] * i_padv + PADH;
1582 //scale full resolution frame
1583 if( h->param.i_threads == 1 )
1585 pixel *src = h->fref[0][j]->filtered[0] - h->fref[0][j]->i_stride[0]*i_padv - PADH;
1586 pixel *dst = h->fenc->weighted[j] - h->fenc->i_stride[0]*i_padv - PADH;
1587 int stride = h->fenc->i_stride[0];
1588 int width = h->fenc->i_width[0] + PADH*2;
1589 int height = h->fenc->i_lines[0] + i_padv*2;
1590 x264_weight_scale_plane( h, dst, stride, src, stride, width, height, &h->sh.weight[j][0] );
1591 h->fenc->i_lines_weighted = height;
1599 if( weightplane[1] )
1600 for( int i = 0; i < h->i_ref[0]; i++ )
1602 if( h->sh.weight[i][1].weightfn && !h->sh.weight[i][2].weightfn )
1604 h->sh.weight[i][2].i_scale = 1 << h->sh.weight[0][1].i_denom;
1605 h->sh.weight[i][2].i_offset = 0;
1607 else if( h->sh.weight[i][2].weightfn && !h->sh.weight[i][1].weightfn )
1609 h->sh.weight[i][1].i_scale = 1 << h->sh.weight[0][1].i_denom;
1610 h->sh.weight[i][1].i_offset = 0;
1614 if( !weightplane[0] )
1615 h->sh.weight[0][0].i_denom = 0;
1616 if( !weightplane[1] )
1617 h->sh.weight[0][1].i_denom = 0;
1618 h->sh.weight[0][2].i_denom = h->sh.weight[0][1].i_denom;
1621 static inline int x264_reference_distance( x264_t *h, x264_frame_t *frame )
1623 if( h->param.i_frame_packing == 5 )
1624 return abs((h->fenc->i_frame&~1) - (frame->i_frame&~1)) +
1625 ((h->fenc->i_frame&1) != (frame->i_frame&1));
1627 return abs(h->fenc->i_frame - frame->i_frame);
1630 static inline void x264_reference_build_list( x264_t *h, int i_poc )
1634 /* build ref list 0/1 */
1635 h->mb.pic.i_fref[0] = h->i_ref[0] = 0;
1636 h->mb.pic.i_fref[1] = h->i_ref[1] = 0;
1637 if( h->sh.i_type == SLICE_TYPE_I )
1640 for( int i = 0; h->frames.reference[i]; i++ )
1642 if( h->frames.reference[i]->b_corrupt )
1644 if( h->frames.reference[i]->i_poc < i_poc )
1645 h->fref[0][h->i_ref[0]++] = h->frames.reference[i];
1646 else if( h->frames.reference[i]->i_poc > i_poc )
1647 h->fref[1][h->i_ref[1]++] = h->frames.reference[i];
1650 /* Order reference lists by distance from the current frame. */
1651 for( int list = 0; list < 2; list++ )
1653 h->fref_nearest[list] = h->fref[list][0];
1657 for( int i = 0; i < h->i_ref[list] - 1; i++ )
1659 if( list ? h->fref[list][i+1]->i_poc < h->fref_nearest[list]->i_poc
1660 : h->fref[list][i+1]->i_poc > h->fref_nearest[list]->i_poc )
1661 h->fref_nearest[list] = h->fref[list][i+1];
1662 if( x264_reference_distance( h, h->fref[list][i] ) > x264_reference_distance( h, h->fref[list][i+1] ) )
1664 XCHG( x264_frame_t*, h->fref[list][i], h->fref[list][i+1] );
1672 if( h->sh.i_mmco_remove_from_end )
1673 for( int i = h->i_ref[0]-1; i >= h->i_ref[0] - h->sh.i_mmco_remove_from_end; i-- )
1675 int diff = h->i_frame_num - h->fref[0][i]->i_frame_num;
1676 h->sh.mmco[h->sh.i_mmco_command_count].i_poc = h->fref[0][i]->i_poc;
1677 h->sh.mmco[h->sh.i_mmco_command_count++].i_difference_of_pic_nums = diff;
1680 x264_reference_check_reorder( h );
1682 h->i_ref[1] = X264_MIN( h->i_ref[1], h->frames.i_max_ref1 );
1683 h->i_ref[0] = X264_MIN( h->i_ref[0], h->frames.i_max_ref0 );
1684 h->i_ref[0] = X264_MIN( h->i_ref[0], h->param.i_frame_reference ); // if reconfig() has lowered the limit
1686 /* For Blu-ray compliance, don't reference frames outside of the minigop. */
1687 if( IS_X264_TYPE_B( h->fenc->i_type ) && h->param.b_bluray_compat )
1688 h->i_ref[0] = X264_MIN( h->i_ref[0], IS_X264_TYPE_B( h->fref[0][0]->i_type ) + 1 );
1690 /* add duplicates */
1691 if( h->fenc->i_type == X264_TYPE_P )
1694 if( h->param.analyse.i_weighted_pred >= X264_WEIGHTP_SIMPLE )
1697 w[1].weightfn = w[2].weightfn = NULL;
1698 if( h->param.rc.b_stat_read )
1699 x264_ratecontrol_set_weights( h, h->fenc );
1701 if( !h->fenc->weight[0][0].weightfn )
1703 h->fenc->weight[0][0].i_denom = 0;
1704 SET_WEIGHT( w[0], 1, 1, 0, -1 );
1705 idx = x264_weighted_reference_duplicate( h, 0, w );
1709 if( h->fenc->weight[0][0].i_scale == 1<<h->fenc->weight[0][0].i_denom )
1711 SET_WEIGHT( h->fenc->weight[0][0], 1, 1, 0, h->fenc->weight[0][0].i_offset );
1713 x264_weighted_reference_duplicate( h, 0, weight_none );
1714 if( h->fenc->weight[0][0].i_offset > -128 )
1716 w[0] = h->fenc->weight[0][0];
1718 h->mc.weight_cache( h, &w[0] );
1719 idx = x264_weighted_reference_duplicate( h, 0, w );
1723 h->mb.ref_blind_dupe = idx;
1726 assert( h->i_ref[0] + h->i_ref[1] <= X264_REF_MAX );
1727 h->mb.pic.i_fref[0] = h->i_ref[0];
1728 h->mb.pic.i_fref[1] = h->i_ref[1];
1731 static void x264_fdec_filter_row( x264_t *h, int mb_y, int b_inloop )
1733 /* mb_y is the mb to be encoded next, not the mb to be filtered here */
1734 int b_hpel = h->fdec->b_kept_as_ref;
1735 int b_deblock = h->sh.i_disable_deblocking_filter_idc != 1;
1736 int b_end = mb_y == h->i_threadslice_end;
1737 int b_measure_quality = 1;
1738 int min_y = mb_y - (1 << h->sh.b_mbaff);
1739 int b_start = min_y == h->i_threadslice_start;
1740 /* Even in interlaced mode, deblocking never modifies more than 4 pixels
1741 * above each MB, as bS=4 doesn't happen for the top of interlaced mbpairs. */
1742 int minpix_y = min_y*16 - 4 * !b_start;
1743 int maxpix_y = mb_y*16 - 4 * !b_end;
1744 b_deblock &= b_hpel || h->param.psz_dump_yuv;
1745 if( h->param.b_sliced_threads && b_start && min_y && !b_inloop )
1747 b_deblock = 0; /* We already deblocked on the inloop pass. */
1748 b_measure_quality = 0; /* We already measured quality on the inloop pass. */
1750 if( mb_y & h->sh.b_mbaff )
1752 if( min_y < h->i_threadslice_start )
1756 for( int y = min_y; y < mb_y; y += (1 << h->sh.b_mbaff) )
1757 x264_frame_deblock_row( h, y );
1759 /* FIXME: Prediction requires different borders for interlaced/progressive mc,
1760 * but the actual image data is equivalent. For now, maintain this
1761 * consistency by copying deblocked pixels between planes. */
1762 if( h->param.b_interlaced )
1763 for( int p = 0; p < 2; p++ )
1764 for( int i = minpix_y>>p; i < maxpix_y>>p; i++ )
1765 memcpy( h->fdec->plane_fld[p] + i*h->fdec->i_stride[p],
1766 h->fdec->plane[p] + i*h->fdec->i_stride[p],
1767 h->mb.i_mb_width*16*sizeof(pixel) );
1771 int end = mb_y == h->mb.i_mb_height;
1772 x264_frame_expand_border( h, h->fdec, min_y, end );
1773 if( h->param.analyse.i_subpel_refine )
1775 x264_frame_filter( h, h->fdec, min_y, end );
1776 x264_frame_expand_border_filtered( h, h->fdec, min_y, end );
1781 for( int i = 0; i < 2; i++ )
1783 XCHG( pixel *, h->intra_border_backup[0][i], h->intra_border_backup[3][i] );
1784 XCHG( pixel *, h->intra_border_backup[1][i], h->intra_border_backup[4][i] );
1787 if( h->i_thread_frames > 1 && h->fdec->b_kept_as_ref )
1788 x264_frame_cond_broadcast( h->fdec, mb_y*16 + (b_end ? 10000 : -(X264_THREAD_HEIGHT << h->sh.b_mbaff)) );
1790 if( b_measure_quality )
1792 maxpix_y = X264_MIN( maxpix_y, h->param.i_height );
1793 if( h->param.analyse.b_psnr )
1795 uint64_t ssd_y = x264_pixel_ssd_wxh( &h->pixf,
1796 h->fdec->plane[0] + minpix_y * h->fdec->i_stride[0], h->fdec->i_stride[0],
1797 h->fenc->plane[0] + minpix_y * h->fenc->i_stride[0], h->fenc->i_stride[0],
1798 h->param.i_width, maxpix_y-minpix_y );
1799 uint64_t ssd_u, ssd_v;
1800 x264_pixel_ssd_nv12( &h->pixf,
1801 h->fdec->plane[1] + (minpix_y>>1) * h->fdec->i_stride[1], h->fdec->i_stride[1],
1802 h->fenc->plane[1] + (minpix_y>>1) * h->fenc->i_stride[1], h->fenc->i_stride[1],
1803 h->param.i_width>>1, (maxpix_y-minpix_y)>>1, &ssd_u, &ssd_v );
1804 h->stat.frame.i_ssd[0] += ssd_y;
1805 h->stat.frame.i_ssd[1] += ssd_u;
1806 h->stat.frame.i_ssd[2] += ssd_v;
1809 if( h->param.analyse.b_ssim )
1812 /* offset by 2 pixels to avoid alignment of ssim blocks with dct blocks,
1813 * and overlap by 4 */
1814 minpix_y += b_start ? 2 : -6;
1815 h->stat.frame.f_ssim +=
1816 x264_pixel_ssim_wxh( &h->pixf,
1817 h->fdec->plane[0] + 2+minpix_y*h->fdec->i_stride[0], h->fdec->i_stride[0],
1818 h->fenc->plane[0] + 2+minpix_y*h->fenc->i_stride[0], h->fenc->i_stride[0],
1819 h->param.i_width-2, maxpix_y-minpix_y, h->scratch_buffer );
1824 static inline int x264_reference_update( x264_t *h )
1826 if( !h->fdec->b_kept_as_ref )
1828 if( h->i_thread_frames > 1 )
1830 x264_frame_push_unused( h, h->fdec );
1831 h->fdec = x264_frame_pop_unused( h, 1 );
1838 /* apply mmco from previous frame. */
1839 for( int i = 0; i < h->sh.i_mmco_command_count; i++ )
1840 for( int j = 0; h->frames.reference[j]; j++ )
1841 if( h->frames.reference[j]->i_poc == h->sh.mmco[i].i_poc )
1842 x264_frame_push_unused( h, x264_frame_shift( &h->frames.reference[j] ) );
1844 /* move frame in the buffer */
1845 x264_frame_push( h->frames.reference, h->fdec );
1846 if( h->frames.reference[h->sps->i_num_ref_frames] )
1847 x264_frame_push_unused( h, x264_frame_shift( h->frames.reference ) );
1848 h->fdec = x264_frame_pop_unused( h, 1 );
1854 static inline void x264_reference_reset( x264_t *h )
1856 while( h->frames.reference[0] )
1857 x264_frame_push_unused( h, x264_frame_pop( h->frames.reference ) );
1862 static inline void x264_reference_hierarchy_reset( x264_t *h )
1865 int b_hasdelayframe = 0;
1867 /* look for delay frames -- chain must only contain frames that are disposable */
1868 for( int i = 0; h->frames.current[i] && IS_DISPOSABLE( h->frames.current[i]->i_type ); i++ )
1869 b_hasdelayframe |= h->frames.current[i]->i_coded
1870 != h->frames.current[i]->i_frame + h->sps->vui.i_num_reorder_frames;
1872 /* This function must handle b-pyramid and clear frames for open-gop */
1873 if( h->param.i_bframe_pyramid != X264_B_PYRAMID_STRICT && !b_hasdelayframe && h->frames.i_poc_last_open_gop == -1 )
1876 /* Remove last BREF. There will never be old BREFs in the
1877 * dpb during a BREF decode when pyramid == STRICT */
1878 for( ref = 0; h->frames.reference[ref]; ref++ )
1880 if( ( h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT
1881 && h->frames.reference[ref]->i_type == X264_TYPE_BREF )
1882 || ( h->frames.reference[ref]->i_poc < h->frames.i_poc_last_open_gop
1883 && h->sh.i_type != SLICE_TYPE_B ) )
1885 int diff = h->i_frame_num - h->frames.reference[ref]->i_frame_num;
1886 h->sh.mmco[h->sh.i_mmco_command_count].i_difference_of_pic_nums = diff;
1887 h->sh.mmco[h->sh.i_mmco_command_count++].i_poc = h->frames.reference[ref]->i_poc;
1888 x264_frame_push_unused( h, x264_frame_shift( &h->frames.reference[ref] ) );
1889 h->b_ref_reorder[0] = 1;
1894 /* Prepare room in the dpb for the delayed display time of the later b-frame's */
1895 if( h->param.i_bframe_pyramid )
1896 h->sh.i_mmco_remove_from_end = X264_MAX( ref + 2 - h->frames.i_max_dpb, 0 );
1899 static inline void x264_slice_init( x264_t *h, int i_nal_type, int i_global_qp )
1901 /* ------------------------ Create slice header ----------------------- */
1902 if( i_nal_type == NAL_SLICE_IDR )
1904 x264_slice_header_init( h, &h->sh, h->sps, h->pps, h->i_idr_pic_id, h->i_frame_num, i_global_qp );
1907 h->i_idr_pic_id ^= 1;
1911 x264_slice_header_init( h, &h->sh, h->sps, h->pps, -1, h->i_frame_num, i_global_qp );
1913 h->sh.i_num_ref_idx_l0_active = h->i_ref[0] <= 0 ? 1 : h->i_ref[0];
1914 h->sh.i_num_ref_idx_l1_active = h->i_ref[1] <= 0 ? 1 : h->i_ref[1];
1915 if( h->sh.i_num_ref_idx_l0_active != h->pps->i_num_ref_idx_l0_default_active ||
1916 (h->sh.i_type == SLICE_TYPE_B && h->sh.i_num_ref_idx_l1_active != h->pps->i_num_ref_idx_l1_default_active) )
1918 h->sh.b_num_ref_idx_override = 1;
1922 if( h->fenc->i_type == X264_TYPE_BREF && h->param.b_bluray_compat && h->sh.i_mmco_command_count )
1925 h->sh_backup = h->sh;
1928 h->fdec->i_frame_num = h->sh.i_frame_num;
1930 if( h->sps->i_poc_type == 0 )
1932 h->sh.i_poc = h->fdec->i_poc;
1933 if( h->param.b_interlaced )
1935 h->sh.i_delta_poc_bottom = h->param.b_tff ? 1 : -1;
1936 h->sh.i_poc += h->sh.i_delta_poc_bottom == -1;
1939 h->sh.i_delta_poc_bottom = 0;
1940 h->fdec->i_delta_poc[0] = h->sh.i_delta_poc_bottom == -1;
1941 h->fdec->i_delta_poc[1] = h->sh.i_delta_poc_bottom == 1;
1943 else if( h->sps->i_poc_type == 1 )
1945 /* FIXME TODO FIXME */
1949 /* Nothing to do ? */
1952 x264_macroblock_slice_init( h );
1955 static int x264_slice_write( x264_t *h )
1958 int mb_xy, i_mb_x, i_mb_y;
1959 int i_skip_bak = 0; /* Shut up GCC. */
1960 bs_t UNINIT(bs_bak);
1961 x264_cabac_t cabac_bak;
1962 uint8_t cabac_prevbyte_bak = 0; /* Shut up GCC. */
1963 int mv_bits_bak = 0;
1964 int tex_bits_bak = 0;
1965 /* NALUs other than the first use a 3-byte startcode.
1966 * Add one extra byte for the rbsp, and one more for the final CABAC putbyte.
1967 * Then add an extra 5 bytes just in case, to account for random NAL escapes and
1968 * other inaccuracies. */
1969 int overhead_guess = (NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal)) + 1 + h->param.b_cabac + 5;
1970 int slice_max_size = h->param.i_slice_max_size > 0 ? (h->param.i_slice_max_size-overhead_guess)*8 : 0;
1971 int back_up_bitstream = slice_max_size || (!h->param.b_cabac && h->sps->i_profile_idc < PROFILE_HIGH);
1972 int starting_bits = bs_pos(&h->out.bs);
1973 int b_deblock = h->sh.i_disable_deblocking_filter_idc != 1;
1974 int b_hpel = h->fdec->b_kept_as_ref;
1975 uint8_t *last_emu_check;
1976 b_deblock &= b_hpel || h->param.psz_dump_yuv;
1977 bs_realign( &h->out.bs );
1980 x264_nal_start( h, h->i_nal_type, h->i_nal_ref_idc );
1981 h->out.nal[h->out.i_nal].i_first_mb = h->sh.i_first_mb;
1984 x264_macroblock_thread_init( h );
1986 /* If this isn't the first slice in the threadslice, set the slice QP
1987 * equal to the last QP in the previous slice for more accurate
1988 * CABAC initialization. */
1989 if( h->sh.i_first_mb != h->i_threadslice_start * h->mb.i_mb_width )
1991 h->sh.i_qp = h->mb.i_last_qp;
1992 h->sh.i_qp_delta = h->sh.i_qp - h->pps->i_pic_init_qp;
1995 x264_slice_header_write( &h->out.bs, &h->sh, h->i_nal_ref_idc );
1996 if( h->param.b_cabac )
1998 /* alignment needed */
1999 bs_align_1( &h->out.bs );
2002 x264_cabac_context_init( &h->cabac, h->sh.i_type, x264_clip3( h->sh.i_qp-QP_BD_OFFSET, 0, 51 ), h->sh.i_cabac_init_idc );
2003 x264_cabac_encode_init ( &h->cabac, h->out.bs.p, h->out.bs.p_end );
2004 last_emu_check = h->cabac.p;
2007 last_emu_check = h->out.bs.p;
2008 h->mb.i_last_qp = h->sh.i_qp;
2009 h->mb.i_last_dqp = 0;
2010 h->mb.field_decoding_flag = 0;
2012 i_mb_y = h->sh.i_first_mb / h->mb.i_mb_width;
2013 i_mb_x = h->sh.i_first_mb % h->mb.i_mb_width;
2019 mb_xy = i_mb_x + i_mb_y * h->mb.i_mb_width;
2020 int mb_spos = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
2022 if( x264_bitstream_check_buffer( h ) )
2025 if( back_up_bitstream && (!h->sh.b_mbaff || (i_mb_y&1) == 0) )
2027 mv_bits_bak = h->stat.frame.i_mv_bits;
2028 tex_bits_bak = h->stat.frame.i_tex_bits;
2029 /* We don't need the contexts because flushing the CABAC encoder has no context
2030 * dependency and macroblocks are only re-encoded in the case where a slice is
2031 * ended (and thus the content of all contexts are thrown away). */
2032 if( h->param.b_cabac )
2034 memcpy( &cabac_bak, &h->cabac, offsetof(x264_cabac_t, f8_bits_encoded) );
2035 /* x264's CABAC writer modifies the previous byte during carry, so it has to be
2037 cabac_prevbyte_bak = h->cabac.p[-1];
2042 i_skip_bak = i_skip;
2046 if( i_mb_x == 0 && !h->mb.b_reencode_mb )
2047 x264_fdec_filter_row( h, i_mb_y, 1 );
2049 if( h->param.b_interlaced )
2051 if( h->mb.b_adaptive_mbaff )
2055 /* FIXME: VSAD is fast but fairly poor at choosing the best interlace type. */
2056 int stride = h->fenc->i_stride[0];
2057 pixel *fenc = h->fenc->plane[0] + 16 * (i_mb_x + i_mb_y * stride);
2058 h->mb.b_interlaced = x264_field_vsad( h, fenc, stride );
2060 x264_zigzag_init( h->param.cpu, &h->zigzagf, h->mb.b_interlaced );
2062 h->mb.field[mb_xy] = h->mb.b_interlaced;
2066 x264_macroblock_cache_load( h, i_mb_x, i_mb_y );
2068 x264_macroblock_analyse( h );
2070 /* encode this macroblock -> be careful it can change the mb type to P_SKIP if needed */
2072 x264_macroblock_encode( h );
2074 if( h->param.b_cabac )
2076 if( mb_xy > h->sh.i_first_mb && !(h->sh.b_mbaff && (i_mb_y&1)) )
2077 x264_cabac_encode_terminal( &h->cabac );
2079 if( IS_SKIP( h->mb.i_type ) )
2080 x264_cabac_mb_skip( h, 1 );
2083 if( h->sh.i_type != SLICE_TYPE_I )
2084 x264_cabac_mb_skip( h, 0 );
2085 x264_macroblock_write_cabac( h, &h->cabac );
2090 if( IS_SKIP( h->mb.i_type ) )
2094 if( h->sh.i_type != SLICE_TYPE_I )
2096 bs_write_ue( &h->out.bs, i_skip ); /* skip run */
2099 x264_macroblock_write_cavlc( h );
2100 /* If there was a CAVLC level code overflow, try again at a higher QP. */
2101 if( h->mb.b_overflow )
2103 h->mb.i_chroma_qp = h->chroma_qp_table[++h->mb.i_qp];
2104 h->mb.i_skip_intra = 0;
2105 h->mb.b_skip_mc = 0;
2106 h->mb.b_overflow = 0;
2108 i_skip = i_skip_bak;
2109 h->stat.frame.i_mv_bits = mv_bits_bak;
2110 h->stat.frame.i_tex_bits = tex_bits_bak;
2116 int total_bits = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
2117 mb_size[i_mb_y&1] = total_bits - mb_spos;
2119 if( slice_max_size )
2121 /* Count the skip run, just in case. */
2122 if( !h->param.b_cabac )
2123 total_bits += bs_size_ue_big( i_skip );
2124 /* Check for escape bytes. */
2125 uint8_t *end = h->param.b_cabac ? h->cabac.p : h->out.bs.p;
2126 for( ; last_emu_check < end - 2; last_emu_check++ )
2127 if( last_emu_check[0] == 0 && last_emu_check[1] == 0 && last_emu_check[2] <= 3 )
2129 slice_max_size -= 8;
2132 /* We'll just re-encode this last macroblock if we go over the max slice size. */
2133 if( total_bits - starting_bits > slice_max_size && !h->mb.b_reencode_mb )
2135 if( mb_xy != h->sh.i_first_mb )
2137 h->stat.frame.i_mv_bits = mv_bits_bak;
2138 h->stat.frame.i_tex_bits = tex_bits_bak;
2139 if( h->param.b_cabac )
2141 memcpy( &h->cabac, &cabac_bak, offsetof(x264_cabac_t, f8_bits_encoded) );
2142 h->cabac.p[-1] = cabac_prevbyte_bak;
2147 i_skip = i_skip_bak;
2149 h->mb.b_reencode_mb = 1;
2152 // set to bottom of previous mbpair
2154 h->sh.i_last_mb = mb_xy-1+h->mb.i_mb_stride*(!(i_mb_y&1));
2156 h->sh.i_last_mb = (i_mb_y-2+!(i_mb_y&1))*h->mb.i_mb_stride + h->mb.i_mb_width - 1;
2159 h->sh.i_last_mb = mb_xy-1;
2164 h->sh.i_last_mb = mb_xy;
2165 h->mb.b_reencode_mb = 0;
2169 h->mb.b_reencode_mb = 0;
2173 if( h->param.b_visualize )
2174 x264_visualize_mb( h );
2178 x264_macroblock_cache_save( h );
2180 /* accumulate mb stats */
2181 h->stat.frame.i_mb_count[h->mb.i_type]++;
2183 int b_intra = IS_INTRA( h->mb.i_type );
2184 if( h->param.i_log_level >= X264_LOG_INFO || h->param.rc.b_stat_write )
2186 if( !b_intra && !IS_SKIP( h->mb.i_type ) && !IS_DIRECT( h->mb.i_type ) )
2188 if( h->mb.i_partition != D_8x8 )
2189 h->stat.frame.i_mb_partition[h->mb.i_partition] += 4;
2191 for( int i = 0; i < 4; i++ )
2192 h->stat.frame.i_mb_partition[h->mb.i_sub_partition[i]] ++;
2193 if( h->param.i_frame_reference > 1 )
2194 for( int i_list = 0; i_list <= (h->sh.i_type == SLICE_TYPE_B); i_list++ )
2195 for( int i = 0; i < 4; i++ )
2197 int i_ref = h->mb.cache.ref[i_list][ x264_scan8[4*i] ];
2199 h->stat.frame.i_mb_count_ref[i_list][i_ref] ++;
2204 if( h->param.i_log_level >= X264_LOG_INFO )
2206 if( h->mb.i_cbp_luma | h->mb.i_cbp_chroma )
2208 int cbpsum = (h->mb.i_cbp_luma&1) + ((h->mb.i_cbp_luma>>1)&1)
2209 + ((h->mb.i_cbp_luma>>2)&1) + (h->mb.i_cbp_luma>>3);
2210 h->stat.frame.i_mb_cbp[!b_intra + 0] += cbpsum;
2211 h->stat.frame.i_mb_cbp[!b_intra + 2] += !!h->mb.i_cbp_chroma;
2212 h->stat.frame.i_mb_cbp[!b_intra + 4] += h->mb.i_cbp_chroma >> 1;
2214 if( h->mb.i_cbp_luma && !b_intra )
2216 h->stat.frame.i_mb_count_8x8dct[0] ++;
2217 h->stat.frame.i_mb_count_8x8dct[1] += h->mb.b_transform_8x8;
2219 if( b_intra && h->mb.i_type != I_PCM )
2221 if( h->mb.i_type == I_16x16 )
2222 h->stat.frame.i_mb_pred_mode[0][h->mb.i_intra16x16_pred_mode]++;
2223 else if( h->mb.i_type == I_8x8 )
2224 for( int i = 0; i < 16; i += 4 )
2225 h->stat.frame.i_mb_pred_mode[1][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
2226 else //if( h->mb.i_type == I_4x4 )
2227 for( int i = 0; i < 16; i++ )
2228 h->stat.frame.i_mb_pred_mode[2][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
2229 h->stat.frame.i_mb_pred_mode[3][x264_mb_pred_mode8x8c_fix[h->mb.i_chroma_pred_mode]]++;
2233 /* calculate deblock strength values (actual deblocking is done per-row along with hpel) */
2235 x264_macroblock_deblock_strength( h );
2239 /* update ratecontrol per-mbpair in MBAFF */
2241 x264_ratecontrol_mb( h, mb_size[0]+mb_size[1] );
2244 x264_ratecontrol_mb( h, mb_size[i_mb_y&1] );
2246 if( mb_xy == h->sh.i_last_mb )
2251 i_mb_x += i_mb_y & 1;
2252 i_mb_y ^= i_mb_x < h->mb.i_mb_width;
2256 if( i_mb_x == h->mb.i_mb_width )
2262 h->out.nal[h->out.i_nal].i_last_mb = h->sh.i_last_mb;
2264 if( h->param.b_cabac )
2266 x264_cabac_encode_flush( h, &h->cabac );
2267 h->out.bs.p = h->cabac.p;
2272 bs_write_ue( &h->out.bs, i_skip ); /* last skip run */
2273 /* rbsp_slice_trailing_bits */
2274 bs_rbsp_trailing( &h->out.bs );
2275 bs_flush( &h->out.bs );
2277 if( x264_nal_end( h ) )
2280 if( h->sh.i_last_mb == (h->i_threadslice_end * h->mb.i_mb_width - 1) )
2282 h->stat.frame.i_misc_bits = bs_pos( &h->out.bs )
2283 + (h->out.i_nal*NALU_OVERHEAD * 8)
2284 - h->stat.frame.i_tex_bits
2285 - h->stat.frame.i_mv_bits;
2286 x264_fdec_filter_row( h, h->i_threadslice_end, 1 );
2292 static void x264_thread_sync_context( x264_t *dst, x264_t *src )
2297 // reference counting
2298 for( x264_frame_t **f = src->frames.reference; *f; f++ )
2299 (*f)->i_reference_count++;
2300 for( x264_frame_t **f = dst->frames.reference; *f; f++ )
2301 x264_frame_push_unused( src, *f );
2302 src->fdec->i_reference_count++;
2303 x264_frame_push_unused( src, dst->fdec );
2305 // copy everything except the per-thread pointers and the constants.
2306 memcpy( &dst->i_frame, &src->i_frame, offsetof(x264_t, mb.type) - offsetof(x264_t, i_frame) );
2307 dst->param = src->param;
2308 dst->stat = src->stat;
2309 dst->pixf = src->pixf;
2312 static void x264_thread_sync_stat( x264_t *dst, x264_t *src )
2316 memcpy( &dst->stat.i_frame_count, &src->stat.i_frame_count, sizeof(dst->stat) - sizeof(dst->stat.frame) );
2319 static void *x264_slices_write( x264_t *h )
2321 int i_slice_num = 0;
2322 int last_thread_mb = h->sh.i_last_mb;
2325 if( h->param.b_visualize )
2326 if( x264_visualize_init( h ) )
2331 memset( &h->stat.frame, 0, sizeof(h->stat.frame) );
2332 h->mb.b_reencode_mb = 0;
2333 while( h->sh.i_first_mb + h->sh.b_mbaff*h->mb.i_mb_stride <= last_thread_mb )
2335 h->sh.i_last_mb = last_thread_mb;
2336 if( h->param.i_slice_max_mbs )
2340 // convert first to mbaff form, add slice-max-mbs, then convert back to normal form
2341 int last_mbaff = 2*(h->sh.i_first_mb % h->mb.i_mb_width)
2342 + h->mb.i_mb_width*(h->sh.i_first_mb / h->mb.i_mb_width)
2343 + h->param.i_slice_max_mbs - 1;
2344 int last_x = (last_mbaff % (2*h->mb.i_mb_width))/2;
2345 int last_y = (last_mbaff / (2*h->mb.i_mb_width))*2 + 1;
2346 h->sh.i_last_mb = last_x + h->mb.i_mb_stride*last_y;
2349 h->sh.i_last_mb = h->sh.i_first_mb + h->param.i_slice_max_mbs - 1;
2351 else if( h->param.i_slice_count && !h->param.b_sliced_threads )
2353 int height = h->mb.i_mb_height >> h->param.b_interlaced;
2354 int width = h->mb.i_mb_width << h->param.b_interlaced;
2356 h->sh.i_last_mb = (height * i_slice_num + h->param.i_slice_count/2) / h->param.i_slice_count * width - 1;
2358 h->sh.i_last_mb = X264_MIN( h->sh.i_last_mb, last_thread_mb );
2359 if( x264_stack_align( x264_slice_write, h ) )
2361 h->sh.i_first_mb = h->sh.i_last_mb + 1;
2362 // if i_first_mb is not the last mb in a row then go to the next mb in MBAFF order
2363 if( h->sh.b_mbaff && h->sh.i_first_mb % h->mb.i_mb_width )
2364 h->sh.i_first_mb -= h->mb.i_mb_stride;
2368 if( h->param.b_visualize )
2370 x264_visualize_show( h );
2371 x264_visualize_close( h );
2378 static int x264_threaded_slices_write( x264_t *h )
2380 /* set first/last mb and sync contexts */
2381 for( int i = 0; i < h->param.i_threads; i++ )
2383 x264_t *t = h->thread[i];
2386 t->param = h->param;
2387 memcpy( &t->i_frame, &h->i_frame, offsetof(x264_t, rc) - offsetof(x264_t, i_frame) );
2389 int height = h->mb.i_mb_height >> h->param.b_interlaced;
2390 t->i_threadslice_start = ((height * i + h->param.i_slice_count/2) / h->param.i_threads) << h->param.b_interlaced;
2391 t->i_threadslice_end = ((height * (i+1) + h->param.i_slice_count/2) / h->param.i_threads) << h->param.b_interlaced;
2392 t->sh.i_first_mb = t->i_threadslice_start * h->mb.i_mb_width;
2393 t->sh.i_last_mb = t->i_threadslice_end * h->mb.i_mb_width - 1;
2396 x264_stack_align( x264_analyse_weight_frame, h, h->mb.i_mb_height*16 + 16 );
2398 x264_threads_distribute_ratecontrol( h );
2401 for( int i = 0; i < h->param.i_threads; i++ )
2403 x264_threadpool_run( h->threadpool, (void*)x264_slices_write, h->thread[i] );
2404 h->thread[i]->b_thread_active = 1;
2406 for( int i = 0; i < h->param.i_threads; i++ )
2408 h->thread[i]->b_thread_active = 0;
2409 if( (intptr_t)x264_threadpool_wait( h->threadpool, h->thread[i] ) )
2413 /* Go back and fix up the hpel on the borders between slices. */
2414 for( int i = 1; i < h->param.i_threads; i++ )
2416 x264_fdec_filter_row( h->thread[i], h->thread[i]->i_threadslice_start + 1, 0 );
2418 x264_fdec_filter_row( h->thread[i], h->thread[i]->i_threadslice_start + 2, 0 );
2421 x264_threads_merge_ratecontrol( h );
2423 for( int i = 1; i < h->param.i_threads; i++ )
2425 x264_t *t = h->thread[i];
2426 for( int j = 0; j < t->out.i_nal; j++ )
2428 h->out.nal[h->out.i_nal] = t->out.nal[j];
2430 x264_nal_check_buffer( h );
2432 /* All entries in stat.frame are ints except for ssd/ssim. */
2433 for( int j = 0; j < (offsetof(x264_t,stat.frame.i_ssd) - offsetof(x264_t,stat.frame.i_mv_bits)) / sizeof(int); j++ )
2434 ((int*)&h->stat.frame)[j] += ((int*)&t->stat.frame)[j];
2435 for( int j = 0; j < 3; j++ )
2436 h->stat.frame.i_ssd[j] += t->stat.frame.i_ssd[j];
2437 h->stat.frame.f_ssim += t->stat.frame.f_ssim;
2443 void x264_encoder_intra_refresh( x264_t *h )
2445 h = h->thread[h->i_thread_phase];
2446 h->b_queued_intra_refresh = 1;
2449 int x264_encoder_invalidate_reference( x264_t *h, int64_t pts )
2451 if( h->param.i_bframe )
2453 x264_log( h, X264_LOG_ERROR, "x264_encoder_invalidate_reference is not supported with B-frames enabled\n" );
2456 if( h->param.b_intra_refresh )
2458 x264_log( h, X264_LOG_ERROR, "x264_encoder_invalidate_reference is not supported with intra refresh enabled\n" );
2461 h = h->thread[h->i_thread_phase];
2462 if( pts >= h->i_last_idr_pts )
2464 for( int i = 0; h->frames.reference[i]; i++ )
2465 if( pts <= h->frames.reference[i]->i_pts )
2466 h->frames.reference[i]->b_corrupt = 1;
2467 if( pts <= h->fdec->i_pts )
2468 h->fdec->b_corrupt = 1;
2473 /****************************************************************************
2474 * x264_encoder_encode:
2475 * XXX: i_poc : is the poc of the current given picture
2476 * i_frame : is the number of the frame being coded
2477 * ex: type frame poc
2485 ****************************************************************************/
2486 int x264_encoder_encode( x264_t *h,
2487 x264_nal_t **pp_nal, int *pi_nal,
2488 x264_picture_t *pic_in,
2489 x264_picture_t *pic_out )
2491 x264_t *thread_current, *thread_prev, *thread_oldest;
2492 int i_nal_type, i_nal_ref_idc, i_global_qp;
2493 int overhead = NALU_OVERHEAD;
2495 if( h->i_thread_frames > 1 )
2497 thread_prev = h->thread[ h->i_thread_phase ];
2498 h->i_thread_phase = (h->i_thread_phase + 1) % h->i_thread_frames;
2499 thread_current = h->thread[ h->i_thread_phase ];
2500 thread_oldest = h->thread[ (h->i_thread_phase + 1) % h->i_thread_frames ];
2501 x264_thread_sync_context( thread_current, thread_prev );
2502 x264_thread_sync_ratecontrol( thread_current, thread_prev, thread_oldest );
2511 if( h->param.cpu&X264_CPU_SSE_MISALIGN )
2512 x264_cpu_mask_misalign_sse();
2515 // ok to call this before encoding any frames, since the initial values of fdec have b_kept_as_ref=0
2516 if( x264_reference_update( h ) )
2518 h->fdec->i_lines_completed = -1;
2524 /* ------------------- Setup new frame from picture -------------------- */
2525 if( pic_in != NULL )
2527 /* 1: Copy the picture to a frame and move it to a buffer */
2528 x264_frame_t *fenc = x264_frame_pop_unused( h, 0 );
2532 if( x264_frame_copy_picture( h, fenc, pic_in ) < 0 )
2535 if( h->param.i_width != 16 * h->mb.i_mb_width ||
2536 h->param.i_height != 16 * h->mb.i_mb_height )
2537 x264_frame_expand_border_mod16( h, fenc );
2539 fenc->i_frame = h->frames.i_input++;
2541 if( fenc->i_frame == 0 )
2542 h->frames.i_first_pts = fenc->i_pts;
2543 if( h->frames.i_bframe_delay && fenc->i_frame == h->frames.i_bframe_delay )
2544 h->frames.i_bframe_delay_time = fenc->i_pts - h->frames.i_first_pts;
2546 if( h->param.b_vfr_input && fenc->i_pts <= h->frames.i_largest_pts )
2547 x264_log( h, X264_LOG_WARNING, "non-strictly-monotonic PTS\n" );
2549 h->frames.i_second_largest_pts = h->frames.i_largest_pts;
2550 h->frames.i_largest_pts = fenc->i_pts;
2552 if( (fenc->i_pic_struct < PIC_STRUCT_AUTO) || (fenc->i_pic_struct > PIC_STRUCT_TRIPLE) )
2553 fenc->i_pic_struct = PIC_STRUCT_AUTO;
2555 if( fenc->i_pic_struct == PIC_STRUCT_AUTO )
2557 int b_interlaced = fenc->param ? fenc->param->b_interlaced : h->param.b_interlaced;
2560 int b_tff = fenc->param ? fenc->param->b_tff : h->param.b_tff;
2561 fenc->i_pic_struct = b_tff ? PIC_STRUCT_TOP_BOTTOM : PIC_STRUCT_BOTTOM_TOP;
2564 fenc->i_pic_struct = PIC_STRUCT_PROGRESSIVE;
2567 if( h->param.rc.b_mb_tree && h->param.rc.b_stat_read )
2569 if( x264_macroblock_tree_read( h, fenc, pic_in->prop.quant_offsets ) )
2573 x264_stack_align( x264_adaptive_quant_frame, h, fenc, pic_in->prop.quant_offsets );
2575 if( pic_in->prop.quant_offsets_free )
2576 pic_in->prop.quant_offsets_free( pic_in->prop.quant_offsets );
2578 if( h->frames.b_have_lowres )
2579 x264_frame_init_lowres( h, fenc );
2581 /* 2: Place the frame into the queue for its slice type decision */
2582 x264_lookahead_put_frame( h, fenc );
2584 if( h->frames.i_input <= h->frames.i_delay + 1 - h->i_thread_frames )
2586 /* Nothing yet to encode, waiting for filling of buffers */
2587 pic_out->i_type = X264_TYPE_AUTO;
2593 /* signal kills for lookahead thread */
2594 x264_pthread_mutex_lock( &h->lookahead->ifbuf.mutex );
2595 h->lookahead->b_exit_thread = 1;
2596 x264_pthread_cond_broadcast( &h->lookahead->ifbuf.cv_fill );
2597 x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
2601 /* 3: The picture is analyzed in the lookahead */
2602 if( !h->frames.current[0] )
2603 x264_lookahead_get_frames( h );
2605 if( !h->frames.current[0] && x264_lookahead_is_empty( h ) )
2606 return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
2608 /* ------------------- Get frame to be encoded ------------------------- */
2609 /* 4: get picture to encode */
2610 h->fenc = x264_frame_shift( h->frames.current );
2611 if( h->i_frame == h->i_thread_frames - 1 )
2612 h->i_reordered_pts_delay = h->fenc->i_reordered_pts;
2613 if( h->fenc->param )
2615 x264_encoder_reconfig( h, h->fenc->param );
2616 if( h->fenc->param->param_free )
2617 h->fenc->param->param_free( h->fenc->param );
2620 if( !IS_X264_TYPE_I( h->fenc->i_type ) )
2622 int valid_refs_left = 0;
2623 for( int i = 0; h->frames.reference[i]; i++ )
2624 if( !h->frames.reference[i]->b_corrupt )
2626 /* No valid reference frames left: force an IDR. */
2627 if( !valid_refs_left )
2629 h->fenc->b_keyframe = 1;
2630 h->fenc->i_type = X264_TYPE_IDR;
2634 if( h->fenc->b_keyframe )
2636 h->frames.i_last_keyframe = h->fenc->i_frame;
2637 if( h->fenc->i_type == X264_TYPE_IDR )
2640 h->frames.i_last_idr = h->fenc->i_frame;
2643 h->sh.i_mmco_command_count =
2644 h->sh.i_mmco_remove_from_end = 0;
2645 h->b_ref_reorder[0] =
2646 h->b_ref_reorder[1] = 0;
2648 h->fenc->i_poc = 2 * ( h->fenc->i_frame - X264_MAX( h->frames.i_last_idr, 0 ) );
2650 /* ------------------- Setup frame context ----------------------------- */
2651 /* 5: Init data dependent of frame type */
2652 if( h->fenc->i_type == X264_TYPE_IDR )
2654 /* reset ref pictures */
2655 i_nal_type = NAL_SLICE_IDR;
2656 i_nal_ref_idc = NAL_PRIORITY_HIGHEST;
2657 h->sh.i_type = SLICE_TYPE_I;
2658 x264_reference_reset( h );
2659 h->frames.i_poc_last_open_gop = -1;
2661 else if( h->fenc->i_type == X264_TYPE_I )
2663 i_nal_type = NAL_SLICE;
2664 i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
2665 h->sh.i_type = SLICE_TYPE_I;
2666 x264_reference_hierarchy_reset( h );
2667 if( h->param.b_open_gop )
2668 h->frames.i_poc_last_open_gop = h->fenc->b_keyframe ? h->fenc->i_poc : -1;
2670 else if( h->fenc->i_type == X264_TYPE_P )
2672 i_nal_type = NAL_SLICE;
2673 i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
2674 h->sh.i_type = SLICE_TYPE_P;
2675 x264_reference_hierarchy_reset( h );
2676 h->frames.i_poc_last_open_gop = -1;
2678 else if( h->fenc->i_type == X264_TYPE_BREF )
2680 i_nal_type = NAL_SLICE;
2681 i_nal_ref_idc = h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT ? NAL_PRIORITY_LOW : NAL_PRIORITY_HIGH;
2682 h->sh.i_type = SLICE_TYPE_B;
2683 x264_reference_hierarchy_reset( h );
2687 i_nal_type = NAL_SLICE;
2688 i_nal_ref_idc = NAL_PRIORITY_DISPOSABLE;
2689 h->sh.i_type = SLICE_TYPE_B;
2692 h->fdec->i_type = h->fenc->i_type;
2693 h->fdec->i_frame = h->fenc->i_frame;
2694 h->fenc->b_kept_as_ref =
2695 h->fdec->b_kept_as_ref = i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE && h->param.i_keyint_max > 1;
2697 h->fdec->i_pts = h->fenc->i_pts;
2698 if( h->frames.i_bframe_delay )
2700 int64_t *prev_reordered_pts = thread_current->frames.i_prev_reordered_pts;
2701 h->fdec->i_dts = h->i_frame > h->frames.i_bframe_delay
2702 ? prev_reordered_pts[ (h->i_frame - h->frames.i_bframe_delay) % h->frames.i_bframe_delay ]
2703 : h->fenc->i_reordered_pts - h->frames.i_bframe_delay_time;
2704 prev_reordered_pts[ h->i_frame % h->frames.i_bframe_delay ] = h->fenc->i_reordered_pts;
2707 h->fdec->i_dts = h->fenc->i_reordered_pts;
2708 if( h->fenc->i_type == X264_TYPE_IDR )
2709 h->i_last_idr_pts = h->fdec->i_pts;
2711 /* ------------------- Init ----------------------------- */
2712 /* build ref list 0/1 */
2713 x264_reference_build_list( h, h->fdec->i_poc );
2715 /* ---------------------- Write the bitstream -------------------------- */
2716 /* Init bitstream context */
2717 if( h->param.b_sliced_threads )
2719 for( int i = 0; i < h->param.i_threads; i++ )
2721 bs_init( &h->thread[i]->out.bs, h->thread[i]->out.p_bitstream, h->thread[i]->out.i_bitstream );
2722 h->thread[i]->out.i_nal = 0;
2727 bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
2731 if( h->param.b_aud )
2735 if( h->sh.i_type == SLICE_TYPE_I )
2737 else if( h->sh.i_type == SLICE_TYPE_P )
2739 else if( h->sh.i_type == SLICE_TYPE_B )
2744 x264_nal_start( h, NAL_AUD, NAL_PRIORITY_DISPOSABLE );
2745 bs_write( &h->out.bs, 3, pic_type );
2746 bs_rbsp_trailing( &h->out.bs );
2747 if( x264_nal_end( h ) )
2749 overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2752 h->i_nal_type = i_nal_type;
2753 h->i_nal_ref_idc = i_nal_ref_idc;
2755 if( h->param.b_intra_refresh )
2757 if( IS_X264_TYPE_I( h->fenc->i_type ) )
2759 h->fdec->i_frames_since_pir = 0;
2760 h->b_queued_intra_refresh = 0;
2761 /* PIR is currently only supported with ref == 1, so any intra frame effectively refreshes
2762 * the whole frame and counts as an intra refresh. */
2763 h->fdec->f_pir_position = h->mb.i_mb_width;
2765 else if( h->fenc->i_type == X264_TYPE_P )
2767 int pocdiff = (h->fdec->i_poc - h->fref[0][0]->i_poc)/2;
2768 float increment = X264_MAX( ((float)h->mb.i_mb_width-1) / h->param.i_keyint_max, 1 );
2769 h->fdec->f_pir_position = h->fref[0][0]->f_pir_position;
2770 h->fdec->i_frames_since_pir = h->fref[0][0]->i_frames_since_pir + pocdiff;
2771 if( h->fdec->i_frames_since_pir >= h->param.i_keyint_max ||
2772 (h->b_queued_intra_refresh && h->fdec->f_pir_position + 0.5 >= h->mb.i_mb_width) )
2774 h->fdec->f_pir_position = 0;
2775 h->fdec->i_frames_since_pir = 0;
2776 h->b_queued_intra_refresh = 0;
2777 h->fenc->b_keyframe = 1;
2779 h->fdec->i_pir_start_col = h->fdec->f_pir_position+0.5;
2780 h->fdec->f_pir_position += increment * pocdiff;
2781 h->fdec->i_pir_end_col = h->fdec->f_pir_position+0.5;
2782 /* If our intra refresh has reached the right side of the frame, we're done. */
2783 if( h->fdec->i_pir_end_col >= h->mb.i_mb_width - 1 )
2784 h->fdec->f_pir_position = h->mb.i_mb_width;
2788 if( h->fenc->b_keyframe )
2790 /* Write SPS and PPS */
2791 if( h->param.b_repeat_headers )
2793 /* generate sequence parameters */
2794 x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
2795 x264_sps_write( &h->out.bs, h->sps );
2796 if( x264_nal_end( h ) )
2798 overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
2800 /* generate picture parameters */
2801 x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
2802 x264_pps_write( &h->out.bs, h->pps );
2803 if( x264_nal_end( h ) )
2805 overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
2808 /* buffering period sei is written in x264_encoder_frame_end */
2811 /* write extra sei */
2812 for( int i = 0; i < h->fenc->extra_sei.num_payloads; i++ )
2814 x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2815 x264_sei_write( &h->out.bs, h->fenc->extra_sei.payloads[i].payload, h->fenc->extra_sei.payloads[i].payload_size,
2816 h->fenc->extra_sei.payloads[i].payload_type );
2817 if( x264_nal_end( h ) )
2819 overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2820 if( h->fenc->extra_sei.sei_free && h->fenc->extra_sei.payloads[i].payload )
2821 h->fenc->extra_sei.sei_free( h->fenc->extra_sei.payloads[i].payload );
2824 if( h->fenc->extra_sei.sei_free && h->fenc->extra_sei.payloads )
2825 h->fenc->extra_sei.sei_free( h->fenc->extra_sei.payloads );
2827 if( h->fenc->b_keyframe )
2829 if( h->param.b_repeat_headers && h->fenc->i_frame == 0 )
2831 /* identify ourself */
2832 x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2833 if( x264_sei_version_write( h, &h->out.bs ) )
2835 if( x264_nal_end( h ) )
2837 overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2840 if( h->fenc->i_type != X264_TYPE_IDR )
2842 int time_to_recovery = h->param.b_open_gop ? 0 : X264_MIN( h->mb.i_mb_width - 1, h->param.i_keyint_max ) + h->param.i_bframe - 1;
2843 x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2844 x264_sei_recovery_point_write( h, &h->out.bs, time_to_recovery );
2845 if( x264_nal_end( h ) )
2847 overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2850 if ( h->param.i_frame_packing >= 0 )
2852 x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2853 x264_sei_frame_packing_write( h, &h->out.bs );
2854 if( x264_nal_end( h ) )
2856 overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2860 /* generate sei pic timing */
2861 if( h->sps->vui.b_pic_struct_present || h->sps->vui.b_nal_hrd_parameters_present )
2863 x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2864 x264_sei_pic_timing_write( h, &h->out.bs );
2865 if( x264_nal_end( h ) )
2867 overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2870 /* As required by Blu-ray. */
2871 if( !IS_X264_TYPE_B( h->fenc->i_type ) && h->b_sh_backup )
2874 x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2875 x264_sei_dec_ref_pic_marking_write( h, &h->out.bs );
2876 if( x264_nal_end( h ) )
2878 overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2881 if( h->fenc->b_keyframe && h->param.b_intra_refresh )
2882 h->i_cpb_delay_pir_offset = h->fenc->i_cpb_delay;
2884 /* Init the rate control */
2885 /* FIXME: Include slice header bit cost. */
2886 x264_ratecontrol_start( h, h->fenc->i_qpplus1, overhead*8 );
2887 i_global_qp = x264_ratecontrol_qp( h );
2889 pic_out->i_qpplus1 =
2890 h->fdec->i_qpplus1 = i_global_qp + 1;
2892 if( h->param.rc.b_stat_read && h->sh.i_type != SLICE_TYPE_I )
2894 x264_reference_build_list_optimal( h );
2895 x264_reference_check_reorder( h );
2899 h->fdec->i_poc_l0ref0 = h->fref[0][0]->i_poc;
2901 /* ------------------------ Create slice header ----------------------- */
2902 x264_slice_init( h, i_nal_type, i_global_qp );
2904 /*------------------------- Weights -------------------------------------*/
2905 if( h->sh.i_type == SLICE_TYPE_B )
2906 x264_macroblock_bipred_init( h );
2908 x264_weighted_pred_init( h );
2910 if( i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE )
2914 h->i_threadslice_start = 0;
2915 h->i_threadslice_end = h->mb.i_mb_height;
2916 if( h->i_thread_frames > 1 )
2918 x264_threadpool_run( h->threadpool, (void*)x264_slices_write, h );
2919 h->b_thread_active = 1;
2921 else if( h->param.b_sliced_threads )
2923 if( x264_threaded_slices_write( h ) )
2927 if( (intptr_t)x264_slices_write( h ) )
2930 return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
2933 static int x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
2934 x264_nal_t **pp_nal, int *pi_nal,
2935 x264_picture_t *pic_out )
2937 char psz_message[80];
2939 if( h->b_thread_active )
2941 h->b_thread_active = 0;
2942 if( (intptr_t)x264_threadpool_wait( h->threadpool, h ) )
2947 pic_out->i_type = X264_TYPE_AUTO;
2952 /* generate sei buffering period and insert it into place */
2953 if( h->fenc->b_keyframe && h->sps->vui.b_nal_hrd_parameters_present )
2955 x264_hrd_fullness( h );
2956 x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2957 x264_sei_buffering_period_write( h, &h->out.bs );
2958 if( x264_nal_end( h ) )
2960 /* buffering period sei must follow AUD, SPS and PPS and precede all other SEIs */
2962 while( h->out.nal[idx].i_type == NAL_AUD ||
2963 h->out.nal[idx].i_type == NAL_SPS ||
2964 h->out.nal[idx].i_type == NAL_PPS )
2966 x264_nal_t nal_tmp = h->out.nal[h->out.i_nal-1];
2967 memmove( &h->out.nal[idx+1], &h->out.nal[idx], (h->out.i_nal-idx-1)*sizeof(x264_nal_t) );
2968 h->out.nal[idx] = nal_tmp;
2971 int frame_size = x264_encoder_encapsulate_nals( h, 0 );
2972 if( frame_size < 0 )
2975 /* Set output picture properties */
2976 pic_out->i_type = h->fenc->i_type;
2978 pic_out->b_keyframe = h->fenc->b_keyframe;
2979 pic_out->i_pic_struct = h->fenc->i_pic_struct;
2981 pic_out->i_pts = h->fdec->i_pts;
2982 pic_out->i_dts = h->fdec->i_dts;
2984 if( pic_out->i_pts < pic_out->i_dts )
2985 x264_log( h, X264_LOG_WARNING, "invalid DTS: PTS is less than DTS\n" );
2987 pic_out->img.i_csp = X264_CSP_NV12;
2989 pic_out->img.i_csp |= X264_CSP_HIGH_DEPTH;
2991 pic_out->img.i_plane = h->fdec->i_plane;
2992 for( int i = 0; i < 2; i++ )
2994 pic_out->img.i_stride[i] = h->fdec->i_stride[i] * sizeof(pixel);
2995 pic_out->img.plane[i] = (uint8_t*)h->fdec->plane[i];
2998 x264_frame_push_unused( thread_current, h->fenc );
3000 /* ---------------------- Update encoder state ------------------------- */
3004 if( x264_ratecontrol_end( h, frame_size * 8, &filler ) < 0 )
3007 pic_out->hrd_timing = h->fenc->hrd_timing;
3012 overhead = (FILLER_OVERHEAD - h->param.b_annexb);
3013 if( h->param.i_slice_max_size && filler > h->param.i_slice_max_size )
3015 int next_size = filler - h->param.i_slice_max_size;
3016 int overflow = X264_MAX( overhead - next_size, 0 );
3017 f = h->param.i_slice_max_size - overhead - overflow;
3020 f = X264_MAX( 0, filler - overhead );
3022 x264_nal_start( h, NAL_FILLER, NAL_PRIORITY_DISPOSABLE );
3023 x264_filler_write( h, &h->out.bs, f );
3024 if( x264_nal_end( h ) )
3026 int total_size = x264_encoder_encapsulate_nals( h, h->out.i_nal-1 );
3027 if( total_size < 0 )
3029 frame_size += total_size;
3030 filler -= total_size;
3033 /* End bitstream, set output */
3034 *pi_nal = h->out.i_nal;
3035 *pp_nal = h->out.nal;
3039 x264_noise_reduction_update( h );
3041 /* ---------------------- Compute/Print statistics --------------------- */
3042 x264_thread_sync_stat( h, h->thread[0] );
3045 h->stat.i_frame_count[h->sh.i_type]++;
3046 h->stat.i_frame_size[h->sh.i_type] += frame_size;
3047 h->stat.f_frame_qp[h->sh.i_type] += h->fdec->f_qp_avg_aq;
3049 for( int i = 0; i < X264_MBTYPE_MAX; i++ )
3050 h->stat.i_mb_count[h->sh.i_type][i] += h->stat.frame.i_mb_count[i];
3051 for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
3052 h->stat.i_mb_partition[h->sh.i_type][i] += h->stat.frame.i_mb_partition[i];
3053 for( int i = 0; i < 2; i++ )
3054 h->stat.i_mb_count_8x8dct[i] += h->stat.frame.i_mb_count_8x8dct[i];
3055 for( int i = 0; i < 6; i++ )
3056 h->stat.i_mb_cbp[i] += h->stat.frame.i_mb_cbp[i];
3057 for( int i = 0; i < 4; i++ )
3058 for( int j = 0; j < 13; j++ )
3059 h->stat.i_mb_pred_mode[i][j] += h->stat.frame.i_mb_pred_mode[i][j];
3060 if( h->sh.i_type != SLICE_TYPE_I )
3061 for( int i_list = 0; i_list < 2; i_list++ )
3062 for( int i = 0; i < X264_REF_MAX*2; i++ )
3063 h->stat.i_mb_count_ref[h->sh.i_type][i_list][i] += h->stat.frame.i_mb_count_ref[i_list][i];
3064 if( h->sh.i_type == SLICE_TYPE_P && h->param.analyse.i_weighted_pred >= X264_WEIGHTP_SIMPLE )
3066 h->stat.i_wpred[0] += !!h->sh.weight[0][0].weightfn;
3067 h->stat.i_wpred[1] += !!h->sh.weight[0][1].weightfn || !!h->sh.weight[0][2].weightfn;
3069 if( h->sh.i_type == SLICE_TYPE_B )
3071 h->stat.i_direct_frames[ h->sh.b_direct_spatial_mv_pred ] ++;
3072 if( h->mb.b_direct_auto_write )
3074 //FIXME somewhat arbitrary time constants
3075 if( h->stat.i_direct_score[0] + h->stat.i_direct_score[1] > h->mb.i_mb_count )
3076 for( int i = 0; i < 2; i++ )
3077 h->stat.i_direct_score[i] = h->stat.i_direct_score[i] * 9/10;
3078 for( int i = 0; i < 2; i++ )
3079 h->stat.i_direct_score[i] += h->stat.frame.i_direct_score[i];
3083 h->stat.i_consecutive_bframes[h->fenc->i_bframes]++;
3085 psz_message[0] = '\0';
3086 double dur = h->fenc->f_duration;
3087 h->stat.f_frame_duration[h->sh.i_type] += dur;
3088 if( h->param.analyse.b_psnr )
3092 h->stat.frame.i_ssd[0],
3093 h->stat.frame.i_ssd[1],
3094 h->stat.frame.i_ssd[2],
3097 h->stat.f_ssd_global[h->sh.i_type] += dur * (ssd[0] + ssd[1] + ssd[2]);
3098 h->stat.f_psnr_average[h->sh.i_type] += dur * x264_psnr( ssd[0] + ssd[1] + ssd[2], 3 * h->param.i_width * h->param.i_height / 2 );
3099 h->stat.f_psnr_mean_y[h->sh.i_type] += dur * x264_psnr( ssd[0], h->param.i_width * h->param.i_height );
3100 h->stat.f_psnr_mean_u[h->sh.i_type] += dur * x264_psnr( ssd[1], h->param.i_width * h->param.i_height / 4 );
3101 h->stat.f_psnr_mean_v[h->sh.i_type] += dur * x264_psnr( ssd[2], h->param.i_width * h->param.i_height / 4 );
3103 snprintf( psz_message, 80, " PSNR Y:%5.2f U:%5.2f V:%5.2f",
3104 x264_psnr( ssd[0], h->param.i_width * h->param.i_height ),
3105 x264_psnr( ssd[1], h->param.i_width * h->param.i_height / 4),
3106 x264_psnr( ssd[2], h->param.i_width * h->param.i_height / 4) );
3109 if( h->param.analyse.b_ssim )
3111 double ssim_y = h->stat.frame.f_ssim
3112 / (((h->param.i_width-6)>>2) * ((h->param.i_height-6)>>2));
3113 h->stat.f_ssim_mean_y[h->sh.i_type] += ssim_y * dur;
3114 snprintf( psz_message + strlen(psz_message), 80 - strlen(psz_message),
3115 " SSIM Y:%.5f", ssim_y );
3117 psz_message[79] = '\0';
3119 x264_log( h, X264_LOG_DEBUG,
3120 "frame=%4d QP=%.2f NAL=%d Slice:%c Poc:%-3d I:%-4d P:%-4d SKIP:%-4d size=%d bytes%s\n",
3122 h->fdec->f_qp_avg_aq,
3124 h->sh.i_type == SLICE_TYPE_I ? 'I' : (h->sh.i_type == SLICE_TYPE_P ? 'P' : 'B' ),
3126 h->stat.frame.i_mb_count_i,
3127 h->stat.frame.i_mb_count_p,
3128 h->stat.frame.i_mb_count_skip,
3132 // keep stats all in one place
3133 x264_thread_sync_stat( h->thread[0], h );
3134 // for the use of the next frame
3135 x264_thread_sync_stat( thread_current, h );
3137 #ifdef DEBUG_MB_TYPE
3139 static const char mb_chars[] = { 'i', 'i', 'I', 'C', 'P', '8', 'S',
3140 'D', '<', 'X', 'B', 'X', '>', 'B', 'B', 'B', 'B', '8', 'S' };
3141 for( int mb_xy = 0; mb_xy < h->mb.i_mb_width * h->mb.i_mb_height; mb_xy++ )
3143 if( h->mb.type[mb_xy] < X264_MBTYPE_MAX && h->mb.type[mb_xy] >= 0 )
3144 fprintf( stderr, "%c ", mb_chars[ h->mb.type[mb_xy] ] );
3146 fprintf( stderr, "? " );
3148 if( (mb_xy+1) % h->mb.i_mb_width == 0 )
3149 fprintf( stderr, "\n" );
3154 /* Remove duplicates, must be done near the end as breaks h->fref0 array
3155 * by freeing some of its pointers. */
3156 for( int i = 0; i < h->i_ref[0]; i++ )
3157 if( h->fref[0][i] && h->fref[0][i]->b_duplicate )
3159 x264_frame_push_blank_unused( h, h->fref[0][i] );
3163 if( h->param.psz_dump_yuv )
3164 x264_frame_dump( h );
3170 static void x264_print_intra( int64_t *i_mb_count, double i_count, int b_print_pcm, char *intra )
3172 intra += sprintf( intra, "I16..4%s: %4.1f%% %4.1f%% %4.1f%%",
3173 b_print_pcm ? "..PCM" : "",
3174 i_mb_count[I_16x16]/ i_count,
3175 i_mb_count[I_8x8] / i_count,
3176 i_mb_count[I_4x4] / i_count );
3178 sprintf( intra, " %4.1f%%", i_mb_count[I_PCM] / i_count );
3181 /****************************************************************************
3182 * x264_encoder_close:
3183 ****************************************************************************/
3184 void x264_encoder_close ( x264_t *h )
3186 int64_t i_yuv_size = 3 * h->param.i_width * h->param.i_height / 2;
3187 int64_t i_mb_count_size[2][7] = {{0}};
3189 int b_print_pcm = h->stat.i_mb_count[SLICE_TYPE_I][I_PCM]
3190 || h->stat.i_mb_count[SLICE_TYPE_P][I_PCM]
3191 || h->stat.i_mb_count[SLICE_TYPE_B][I_PCM];
3193 x264_lookahead_delete( h );
3195 if( h->param.i_threads > 1 )
3196 x264_threadpool_delete( h->threadpool );
3197 if( h->i_thread_frames > 1 )
3199 for( int i = 0; i < h->i_thread_frames; i++ )
3200 if( h->thread[i]->b_thread_active )
3202 assert( h->thread[i]->fenc->i_reference_count == 1 );
3203 x264_frame_delete( h->thread[i]->fenc );
3206 x264_t *thread_prev = h->thread[h->i_thread_phase];
3207 x264_thread_sync_ratecontrol( h, thread_prev, h );
3208 x264_thread_sync_ratecontrol( thread_prev, thread_prev, h );
3209 h->i_frame = thread_prev->i_frame + 1 - h->i_thread_frames;
3213 /* Slices used and PSNR */
3214 for( int i = 0; i < 3; i++ )
3216 static const uint8_t slice_order[] = { SLICE_TYPE_I, SLICE_TYPE_P, SLICE_TYPE_B };
3217 int i_slice = slice_order[i];
3219 if( h->stat.i_frame_count[i_slice] > 0 )
3221 int i_count = h->stat.i_frame_count[i_slice];
3222 double dur = h->stat.f_frame_duration[i_slice];
3223 if( h->param.analyse.b_psnr )
3225 x264_log( h, X264_LOG_INFO,
3226 "frame %c:%-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",
3227 slice_type_to_char[i_slice],
3229 h->stat.f_frame_qp[i_slice] / i_count,
3230 (double)h->stat.i_frame_size[i_slice] / i_count,
3231 h->stat.f_psnr_mean_y[i_slice] / dur, h->stat.f_psnr_mean_u[i_slice] / dur, h->stat.f_psnr_mean_v[i_slice] / dur,
3232 h->stat.f_psnr_average[i_slice] / dur,
3233 x264_psnr( h->stat.f_ssd_global[i_slice], dur * i_yuv_size ) );
3237 x264_log( h, X264_LOG_INFO,
3238 "frame %c:%-5d Avg QP:%5.2f size:%6.0f\n",
3239 slice_type_to_char[i_slice],
3241 h->stat.f_frame_qp[i_slice] / i_count,
3242 (double)h->stat.i_frame_size[i_slice] / i_count );
3246 if( h->param.i_bframe && h->stat.i_frame_count[SLICE_TYPE_B] )
3250 // weight by number of frames (including the I/P-frames) that are in a sequence of N B-frames
3251 for( int i = 0; i <= h->param.i_bframe; i++ )
3252 den += (i+1) * h->stat.i_consecutive_bframes[i];
3253 for( int i = 0; i <= h->param.i_bframe; i++ )
3254 p += sprintf( p, " %4.1f%%", 100. * (i+1) * h->stat.i_consecutive_bframes[i] / den );
3255 x264_log( h, X264_LOG_INFO, "consecutive B-frames:%s\n", buf );
3258 for( int i_type = 0; i_type < 2; i_type++ )
3259 for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
3261 if( i == D_DIRECT_8x8 ) continue; /* direct is counted as its own type */
3262 i_mb_count_size[i_type][x264_mb_partition_pixel_table[i]] += h->stat.i_mb_partition[i_type][i];
3266 if( h->stat.i_frame_count[SLICE_TYPE_I] > 0 )
3268 int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_I];
3269 double i_count = h->stat.i_frame_count[SLICE_TYPE_I] * h->mb.i_mb_count / 100.0;
3270 x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
3271 x264_log( h, X264_LOG_INFO, "mb I %s\n", buf );
3273 if( h->stat.i_frame_count[SLICE_TYPE_P] > 0 )
3275 int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_P];
3276 double i_count = h->stat.i_frame_count[SLICE_TYPE_P] * h->mb.i_mb_count / 100.0;
3277 int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_P];
3278 x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
3279 x264_log( h, X264_LOG_INFO,
3280 "mb P %s P16..4: %4.1f%% %4.1f%% %4.1f%% %4.1f%% %4.1f%% skip:%4.1f%%\n",
3282 i_mb_size[PIXEL_16x16] / (i_count*4),
3283 (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
3284 i_mb_size[PIXEL_8x8] / (i_count*4),
3285 (i_mb_size[PIXEL_8x4] + i_mb_size[PIXEL_4x8]) / (i_count*4),
3286 i_mb_size[PIXEL_4x4] / (i_count*4),
3287 i_mb_count[P_SKIP] / i_count );
3289 if( h->stat.i_frame_count[SLICE_TYPE_B] > 0 )
3291 int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_B];
3292 double i_count = h->stat.i_frame_count[SLICE_TYPE_B] * h->mb.i_mb_count / 100.0;
3293 double i_mb_list_count;
3294 int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_B];
3295 int64_t list_count[3] = {0}; /* 0 == L0, 1 == L1, 2 == BI */
3296 x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
3297 for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
3298 for( int j = 0; j < 2; j++ )
3300 int l0 = x264_mb_type_list_table[i][0][j];
3301 int l1 = x264_mb_type_list_table[i][1][j];
3303 list_count[l1+l0*l1] += h->stat.i_mb_count[SLICE_TYPE_B][i] * 2;
3305 list_count[0] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L0_8x8];
3306 list_count[1] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L1_8x8];
3307 list_count[2] += h->stat.i_mb_partition[SLICE_TYPE_B][D_BI_8x8];
3308 i_mb_count[B_DIRECT] += (h->stat.i_mb_partition[SLICE_TYPE_B][D_DIRECT_8x8]+2)/4;
3309 i_mb_list_count = (list_count[0] + list_count[1] + list_count[2]) / 100.0;
3310 x264_log( h, X264_LOG_INFO,
3311 "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",
3313 i_mb_size[PIXEL_16x16] / (i_count*4),
3314 (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
3315 i_mb_size[PIXEL_8x8] / (i_count*4),
3316 i_mb_count[B_DIRECT] / i_count,
3317 i_mb_count[B_SKIP] / i_count,
3318 list_count[0] / i_mb_list_count,
3319 list_count[1] / i_mb_list_count,
3320 list_count[2] / i_mb_list_count );
3323 x264_ratecontrol_summary( h );
3325 if( h->stat.i_frame_count[SLICE_TYPE_I] + h->stat.i_frame_count[SLICE_TYPE_P] + h->stat.i_frame_count[SLICE_TYPE_B] > 0 )
3327 #define SUM3(p) (p[SLICE_TYPE_I] + p[SLICE_TYPE_P] + p[SLICE_TYPE_B])
3328 #define SUM3b(p,o) (p[SLICE_TYPE_I][o] + p[SLICE_TYPE_P][o] + p[SLICE_TYPE_B][o])
3329 int64_t i_i8x8 = SUM3b( h->stat.i_mb_count, I_8x8 );
3330 int64_t i_intra = i_i8x8 + SUM3b( h->stat.i_mb_count, I_4x4 )
3331 + SUM3b( h->stat.i_mb_count, I_16x16 );
3332 int64_t i_all_intra = i_intra + SUM3b( h->stat.i_mb_count, I_PCM);
3333 const int i_count = h->stat.i_frame_count[SLICE_TYPE_I] +
3334 h->stat.i_frame_count[SLICE_TYPE_P] +
3335 h->stat.i_frame_count[SLICE_TYPE_B];
3336 const double duration = h->stat.f_frame_duration[SLICE_TYPE_I] +
3337 h->stat.f_frame_duration[SLICE_TYPE_P] +
3338 h->stat.f_frame_duration[SLICE_TYPE_B];
3339 int64_t i_mb_count = (int64_t)i_count * h->mb.i_mb_count;
3340 float f_bitrate = SUM3(h->stat.i_frame_size) / duration / 125;
3342 if( h->pps->b_transform_8x8_mode )
3345 if( h->stat.i_mb_count_8x8dct[0] )
3346 sprintf( buf, " inter:%.1f%%", 100. * h->stat.i_mb_count_8x8dct[1] / h->stat.i_mb_count_8x8dct[0] );
3347 x264_log( h, X264_LOG_INFO, "8x8 transform intra:%.1f%%%s\n", 100. * i_i8x8 / i_intra, buf );
3350 if( (h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO ||
3351 (h->stat.i_direct_frames[0] && h->stat.i_direct_frames[1]))
3352 && h->stat.i_frame_count[SLICE_TYPE_B] )
3354 x264_log( h, X264_LOG_INFO, "direct mvs spatial:%.1f%% temporal:%.1f%%\n",
3355 h->stat.i_direct_frames[1] * 100. / h->stat.i_frame_count[SLICE_TYPE_B],
3356 h->stat.i_direct_frames[0] * 100. / h->stat.i_frame_count[SLICE_TYPE_B] );
3360 if( i_mb_count != i_all_intra )
3361 sprintf( buf, " inter: %.1f%% %.1f%% %.1f%%",
3362 h->stat.i_mb_cbp[1] * 100.0 / ((i_mb_count - i_all_intra)*4),
3363 h->stat.i_mb_cbp[3] * 100.0 / ((i_mb_count - i_all_intra) ),
3364 h->stat.i_mb_cbp[5] * 100.0 / ((i_mb_count - i_all_intra)) );
3365 x264_log( h, X264_LOG_INFO, "coded y,uvDC,uvAC intra: %.1f%% %.1f%% %.1f%%%s\n",
3366 h->stat.i_mb_cbp[0] * 100.0 / (i_all_intra*4),
3367 h->stat.i_mb_cbp[2] * 100.0 / (i_all_intra ),
3368 h->stat.i_mb_cbp[4] * 100.0 / (i_all_intra ), buf );
3370 int64_t fixed_pred_modes[4][9] = {{0}};
3371 int64_t sum_pred_modes[4] = {0};
3372 for( int i = 0; i <= I_PRED_16x16_DC_128; i++ )
3374 fixed_pred_modes[0][x264_mb_pred_mode16x16_fix[i]] += h->stat.i_mb_pred_mode[0][i];
3375 sum_pred_modes[0] += h->stat.i_mb_pred_mode[0][i];
3377 if( sum_pred_modes[0] )
3378 x264_log( h, X264_LOG_INFO, "i16 v,h,dc,p: %2.0f%% %2.0f%% %2.0f%% %2.0f%%\n",
3379 fixed_pred_modes[0][0] * 100.0 / sum_pred_modes[0],
3380 fixed_pred_modes[0][1] * 100.0 / sum_pred_modes[0],
3381 fixed_pred_modes[0][2] * 100.0 / sum_pred_modes[0],
3382 fixed_pred_modes[0][3] * 100.0 / sum_pred_modes[0] );
3383 for( int i = 1; i <= 2; i++ )
3385 for( int j = 0; j <= I_PRED_8x8_DC_128; j++ )
3387 fixed_pred_modes[i][x264_mb_pred_mode4x4_fix(j)] += h->stat.i_mb_pred_mode[i][j];
3388 sum_pred_modes[i] += h->stat.i_mb_pred_mode[i][j];
3390 if( sum_pred_modes[i] )
3391 x264_log( h, X264_LOG_INFO, "i%d v,h,dc,ddl,ddr,vr,hd,vl,hu: %2.0f%% %2.0f%% %2.0f%% %2.0f%% %2.0f%% %2.0f%% %2.0f%% %2.0f%% %2.0f%%\n", (3-i)*4,
3392 fixed_pred_modes[i][0] * 100.0 / sum_pred_modes[i],
3393 fixed_pred_modes[i][1] * 100.0 / sum_pred_modes[i],
3394 fixed_pred_modes[i][2] * 100.0 / sum_pred_modes[i],
3395 fixed_pred_modes[i][3] * 100.0 / sum_pred_modes[i],
3396 fixed_pred_modes[i][4] * 100.0 / sum_pred_modes[i],
3397 fixed_pred_modes[i][5] * 100.0 / sum_pred_modes[i],
3398 fixed_pred_modes[i][6] * 100.0 / sum_pred_modes[i],
3399 fixed_pred_modes[i][7] * 100.0 / sum_pred_modes[i],
3400 fixed_pred_modes[i][8] * 100.0 / sum_pred_modes[i] );
3402 for( int i = 0; i <= I_PRED_CHROMA_DC_128; i++ )
3404 fixed_pred_modes[3][x264_mb_pred_mode8x8c_fix[i]] += h->stat.i_mb_pred_mode[3][i];
3405 sum_pred_modes[3] += h->stat.i_mb_pred_mode[3][i];
3407 if( sum_pred_modes[3] )
3408 x264_log( h, X264_LOG_INFO, "i8c dc,h,v,p: %2.0f%% %2.0f%% %2.0f%% %2.0f%%\n",
3409 fixed_pred_modes[3][0] * 100.0 / sum_pred_modes[3],
3410 fixed_pred_modes[3][1] * 100.0 / sum_pred_modes[3],
3411 fixed_pred_modes[3][2] * 100.0 / sum_pred_modes[3],
3412 fixed_pred_modes[3][3] * 100.0 / sum_pred_modes[3] );
3414 if( h->param.analyse.i_weighted_pred >= X264_WEIGHTP_SIMPLE && h->stat.i_frame_count[SLICE_TYPE_P] > 0 )
3415 x264_log( h, X264_LOG_INFO, "Weighted P-Frames: Y:%.1f%% UV:%.1f%%\n",
3416 h->stat.i_wpred[0] * 100.0 / h->stat.i_frame_count[SLICE_TYPE_P],
3417 h->stat.i_wpred[1] * 100.0 / h->stat.i_frame_count[SLICE_TYPE_P] );
3419 for( int i_list = 0; i_list < 2; i_list++ )
3420 for( int i_slice = 0; i_slice < 2; i_slice++ )
3425 for( int i = 0; i < X264_REF_MAX*2; i++ )
3426 if( h->stat.i_mb_count_ref[i_slice][i_list][i] )
3428 i_den += h->stat.i_mb_count_ref[i_slice][i_list][i];
3433 for( int i = 0; i <= i_max; i++ )
3434 p += sprintf( p, " %4.1f%%", 100. * h->stat.i_mb_count_ref[i_slice][i_list][i] / i_den );
3435 x264_log( h, X264_LOG_INFO, "ref %c L%d:%s\n", "PB"[i_slice], i_list, buf );
3438 if( h->param.analyse.b_ssim )
3440 float ssim = SUM3( h->stat.f_ssim_mean_y ) / duration;
3441 x264_log( h, X264_LOG_INFO, "SSIM Mean Y:%.7f (%6.3fdb)\n", ssim, x264_ssim( ssim ) );
3443 if( h->param.analyse.b_psnr )
3445 x264_log( h, X264_LOG_INFO,
3446 "PSNR Mean Y:%6.3f U:%6.3f V:%6.3f Avg:%6.3f Global:%6.3f kb/s:%.2f\n",
3447 SUM3( h->stat.f_psnr_mean_y ) / duration,
3448 SUM3( h->stat.f_psnr_mean_u ) / duration,
3449 SUM3( h->stat.f_psnr_mean_v ) / duration,
3450 SUM3( h->stat.f_psnr_average ) / duration,
3451 x264_psnr( SUM3( h->stat.f_ssd_global ), duration * i_yuv_size ),
3455 x264_log( h, X264_LOG_INFO, "kb/s:%.2f\n", f_bitrate );
3459 x264_ratecontrol_delete( h );
3462 if( h->param.rc.psz_stat_out )
3463 free( h->param.rc.psz_stat_out );
3464 if( h->param.rc.psz_stat_in )
3465 free( h->param.rc.psz_stat_in );
3467 x264_cqm_delete( h );
3468 x264_free( h->nal_buffer );
3469 x264_analyse_free_costs( h );
3471 if( h->i_thread_frames > 1)
3472 h = h->thread[h->i_thread_phase];
3475 x264_frame_delete_list( h->frames.unused[0] );
3476 x264_frame_delete_list( h->frames.unused[1] );
3477 x264_frame_delete_list( h->frames.current );
3478 x264_frame_delete_list( h->frames.blank_unused );
3482 for( int i = 0; i < h->i_thread_frames; i++ )
3483 if( h->thread[i]->b_thread_active )
3484 for( int j = 0; j < h->thread[i]->i_ref[0]; j++ )
3485 if( h->thread[i]->fref[0][j] && h->thread[i]->fref[0][j]->b_duplicate )
3486 x264_frame_delete( h->thread[i]->fref[0][j] );
3488 for( int i = h->param.i_threads - 1; i >= 0; i-- )
3490 x264_frame_t **frame;
3492 if( !h->param.b_sliced_threads || i == 0 )
3494 for( frame = h->thread[i]->frames.reference; *frame; frame++ )
3496 assert( (*frame)->i_reference_count > 0 );
3497 (*frame)->i_reference_count--;
3498 if( (*frame)->i_reference_count == 0 )
3499 x264_frame_delete( *frame );
3501 frame = &h->thread[i]->fdec;
3504 assert( (*frame)->i_reference_count > 0 );
3505 (*frame)->i_reference_count--;
3506 if( (*frame)->i_reference_count == 0 )
3507 x264_frame_delete( *frame );
3509 x264_macroblock_cache_free( h->thread[i] );
3511 x264_macroblock_thread_free( h->thread[i], 0 );
3512 x264_free( h->thread[i]->out.p_bitstream );
3513 x264_free( h->thread[i]->out.nal);
3514 x264_free( h->thread[i] );
3518 int x264_encoder_delayed_frames( x264_t *h )
3520 int delayed_frames = 0;
3521 if( h->i_thread_frames > 1 )
3523 for( int i = 0; i < h->i_thread_frames; i++ )
3524 delayed_frames += h->thread[i]->b_thread_active;
3525 h = h->thread[h->i_thread_phase];
3527 for( int i = 0; h->frames.current[i]; i++ )
3529 x264_pthread_mutex_lock( &h->lookahead->ofbuf.mutex );
3530 x264_pthread_mutex_lock( &h->lookahead->ifbuf.mutex );
3531 x264_pthread_mutex_lock( &h->lookahead->next.mutex );
3532 delayed_frames += h->lookahead->ifbuf.i_size + h->lookahead->next.i_size + h->lookahead->ofbuf.i_size;
3533 x264_pthread_mutex_unlock( &h->lookahead->next.mutex );
3534 x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
3535 x264_pthread_mutex_unlock( &h->lookahead->ofbuf.mutex );
3536 return delayed_frames;
3539 int x264_encoder_maximum_delayed_frames( x264_t *h )
3541 return h->frames.i_delay;