1 /*****************************************************************************
3 *****************************************************************************
4 * Copyright (C) 2003-2008 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.
23 *****************************************************************************/
27 #include "common/common.h"
28 #include "common/cpu.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 float x264_psnr( int64_t i_sqe, int64_t i_size )
55 double f_mse = (double)i_sqe / ((double)65025.0 * (double)i_size);
56 if( f_mse <= 0.0000000001 ) /* Max 100dB */
59 return -10.0 * log10( f_mse );
62 static void x264_frame_dump( x264_t *h )
64 FILE *f = fopen( h->param.psz_dump_yuv, "r+b" );
67 /* Write the frame in display order */
68 fseek( f, (uint64_t)h->fdec->i_frame * h->param.i_height * h->param.i_width * 3/2, SEEK_SET );
69 for( int i = 0; i < h->fdec->i_plane; i++ )
70 for( int y = 0; y < h->param.i_height >> !!i; y++ )
71 fwrite( &h->fdec->plane[i][y*h->fdec->i_stride[i]], 1, h->param.i_width >> !!i, f );
76 /* Fill "default" values */
77 static void x264_slice_header_init( x264_t *h, x264_slice_header_t *sh,
78 x264_sps_t *sps, x264_pps_t *pps,
79 int i_idr_pic_id, int i_frame, int i_qp )
81 x264_param_t *param = &h->param;
83 /* First we fill all fields */
88 sh->i_last_mb = h->mb.i_mb_count - 1;
89 sh->i_pps_id = pps->i_id;
91 sh->i_frame_num = i_frame;
93 sh->b_mbaff = h->param.b_interlaced;
94 sh->b_field_pic = 0; /* no field support for now */
95 sh->b_bottom_field = 0; /* not yet used */
97 sh->i_idr_pic_id = i_idr_pic_id;
99 /* poc stuff, fixed later */
101 sh->i_delta_poc_bottom = 0;
102 sh->i_delta_poc[0] = 0;
103 sh->i_delta_poc[1] = 0;
105 sh->i_redundant_pic_cnt = 0;
107 h->mb.b_direct_auto_write = h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO
109 && ( h->param.rc.b_stat_write || !h->param.rc.b_stat_read );
111 if( !h->mb.b_direct_auto_read && sh->i_type == SLICE_TYPE_B )
113 if( h->fref1[0]->i_poc_l0ref0 == h->fref0[0]->i_poc )
115 if( h->mb.b_direct_auto_write )
116 sh->b_direct_spatial_mv_pred = ( h->stat.i_direct_score[1] > h->stat.i_direct_score[0] );
118 sh->b_direct_spatial_mv_pred = ( param->analyse.i_direct_mv_pred == X264_DIRECT_PRED_SPATIAL );
122 h->mb.b_direct_auto_write = 0;
123 sh->b_direct_spatial_mv_pred = 1;
126 /* else b_direct_spatial_mv_pred was read from the 2pass statsfile */
128 sh->b_num_ref_idx_override = 0;
129 sh->i_num_ref_idx_l0_active = 1;
130 sh->i_num_ref_idx_l1_active = 1;
132 sh->b_ref_pic_list_reordering_l0 = h->b_ref_reorder[0];
133 sh->b_ref_pic_list_reordering_l1 = h->b_ref_reorder[1];
135 /* If the ref list isn't in the default order, construct reordering header */
136 /* List1 reordering isn't needed yet */
137 if( sh->b_ref_pic_list_reordering_l0 )
139 int pred_frame_num = i_frame;
140 for( int i = 0; i < h->i_ref0; i++ )
142 int diff = h->fref0[i]->i_frame_num - pred_frame_num;
144 x264_log( h, X264_LOG_ERROR, "diff frame num == 0\n" );
145 sh->ref_pic_list_order[0][i].idc = ( diff > 0 );
146 sh->ref_pic_list_order[0][i].arg = abs( diff ) - 1;
147 pred_frame_num = h->fref0[i]->i_frame_num;
151 sh->i_cabac_init_idc = param->i_cabac_init_idc;
154 sh->i_qp_delta = i_qp - pps->i_pic_init_qp;
155 sh->b_sp_for_swidth = 0;
158 int deblock_thresh = i_qp + 2 * X264_MIN(param->i_deblocking_filter_alphac0, param->i_deblocking_filter_beta);
159 /* If effective qp <= 15, deblocking would have no effect anyway */
160 if( param->b_deblocking_filter && (h->mb.b_variable_qp || 15 < deblock_thresh ) )
161 sh->i_disable_deblocking_filter_idc = param->b_sliced_threads ? 2 : 0;
163 sh->i_disable_deblocking_filter_idc = 1;
164 sh->i_alpha_c0_offset = param->i_deblocking_filter_alphac0 << 1;
165 sh->i_beta_offset = param->i_deblocking_filter_beta << 1;
168 static void x264_slice_header_write( bs_t *s, x264_slice_header_t *sh, int i_nal_ref_idc )
172 assert( sh->i_first_mb % (2*sh->sps->i_mb_width) == 0 );
173 bs_write_ue( s, sh->i_first_mb >> 1 );
176 bs_write_ue( s, sh->i_first_mb );
178 bs_write_ue( s, sh->i_type + 5 ); /* same type things */
179 bs_write_ue( s, sh->i_pps_id );
180 bs_write( s, sh->sps->i_log2_max_frame_num, sh->i_frame_num & ((1<<sh->sps->i_log2_max_frame_num)-1) );
182 if( !sh->sps->b_frame_mbs_only )
184 bs_write1( s, sh->b_field_pic );
185 if( sh->b_field_pic )
186 bs_write1( s, sh->b_bottom_field );
189 if( sh->i_idr_pic_id >= 0 ) /* NAL IDR */
190 bs_write_ue( s, sh->i_idr_pic_id );
192 if( sh->sps->i_poc_type == 0 )
194 bs_write( s, sh->sps->i_log2_max_poc_lsb, sh->i_poc & ((1<<sh->sps->i_log2_max_poc_lsb)-1) );
195 if( sh->pps->b_pic_order && !sh->b_field_pic )
196 bs_write_se( s, sh->i_delta_poc_bottom );
198 else if( sh->sps->i_poc_type == 1 && !sh->sps->b_delta_pic_order_always_zero )
200 bs_write_se( s, sh->i_delta_poc[0] );
201 if( sh->pps->b_pic_order && !sh->b_field_pic )
202 bs_write_se( s, sh->i_delta_poc[1] );
205 if( sh->pps->b_redundant_pic_cnt )
206 bs_write_ue( s, sh->i_redundant_pic_cnt );
208 if( sh->i_type == SLICE_TYPE_B )
209 bs_write1( s, sh->b_direct_spatial_mv_pred );
211 if( sh->i_type == SLICE_TYPE_P || sh->i_type == SLICE_TYPE_SP || sh->i_type == SLICE_TYPE_B )
213 bs_write1( s, sh->b_num_ref_idx_override );
214 if( sh->b_num_ref_idx_override )
216 bs_write_ue( s, sh->i_num_ref_idx_l0_active - 1 );
217 if( sh->i_type == SLICE_TYPE_B )
218 bs_write_ue( s, sh->i_num_ref_idx_l1_active - 1 );
222 /* ref pic list reordering */
223 if( sh->i_type != SLICE_TYPE_I )
225 bs_write1( s, sh->b_ref_pic_list_reordering_l0 );
226 if( sh->b_ref_pic_list_reordering_l0 )
228 for( int i = 0; i < sh->i_num_ref_idx_l0_active; i++ )
230 bs_write_ue( s, sh->ref_pic_list_order[0][i].idc );
231 bs_write_ue( s, sh->ref_pic_list_order[0][i].arg );
236 if( sh->i_type == SLICE_TYPE_B )
238 bs_write1( s, sh->b_ref_pic_list_reordering_l1 );
239 if( sh->b_ref_pic_list_reordering_l1 )
241 for( int i = 0; i < sh->i_num_ref_idx_l1_active; i++ )
243 bs_write_ue( s, sh->ref_pic_list_order[1][i].idc );
244 bs_write_ue( s, sh->ref_pic_list_order[1][i].arg );
250 if( sh->pps->b_weighted_pred && ( sh->i_type == SLICE_TYPE_P || sh->i_type == SLICE_TYPE_SP ) )
252 /* pred_weight_table() */
253 bs_write_ue( s, sh->weight[0][0].i_denom );
254 bs_write_ue( s, sh->weight[0][1].i_denom );
255 for( int i = 0; i < sh->i_num_ref_idx_l0_active; i++ )
257 int luma_weight_l0_flag = !!sh->weight[i][0].weightfn;
258 int chroma_weight_l0_flag = !!sh->weight[i][1].weightfn || !!sh->weight[i][2].weightfn;
259 bs_write1( s, luma_weight_l0_flag );
260 if( luma_weight_l0_flag )
262 bs_write_se( s, sh->weight[i][0].i_scale );
263 bs_write_se( s, sh->weight[i][0].i_offset );
265 bs_write1( s, chroma_weight_l0_flag );
266 if( chroma_weight_l0_flag )
268 for( int j = 1; j < 3; j++ )
270 bs_write_se( s, sh->weight[i][j].i_scale );
271 bs_write_se( s, sh->weight[i][j].i_offset );
276 else if( sh->pps->b_weighted_bipred == 1 && sh->i_type == SLICE_TYPE_B )
281 if( i_nal_ref_idc != 0 )
283 if( sh->i_idr_pic_id >= 0 )
285 bs_write1( s, 0 ); /* no output of prior pics flag */
286 bs_write1( s, 0 ); /* long term reference flag */
290 bs_write1( s, sh->i_mmco_command_count > 0 ); /* adaptive_ref_pic_marking_mode_flag */
291 if( sh->i_mmco_command_count > 0 )
293 for( int i = 0; i < sh->i_mmco_command_count; i++ )
295 bs_write_ue( s, 1 ); /* mark short term ref as unused */
296 bs_write_ue( s, sh->mmco[i].i_difference_of_pic_nums - 1 );
298 bs_write_ue( s, 0 ); /* end command list */
303 if( sh->pps->b_cabac && sh->i_type != SLICE_TYPE_I )
304 bs_write_ue( s, sh->i_cabac_init_idc );
306 bs_write_se( s, sh->i_qp_delta ); /* slice qp delta */
308 if( sh->pps->b_deblocking_filter_control )
310 bs_write_ue( s, sh->i_disable_deblocking_filter_idc );
311 if( sh->i_disable_deblocking_filter_idc != 1 )
313 bs_write_se( s, sh->i_alpha_c0_offset >> 1 );
314 bs_write_se( s, sh->i_beta_offset >> 1 );
319 /* If we are within a reasonable distance of the end of the memory allocated for the bitstream, */
320 /* reallocate, adding an arbitrary amount of space (100 kilobytes). */
321 static int x264_bitstream_check_buffer( x264_t *h )
323 uint8_t *bs_bak = h->out.p_bitstream;
324 if( (h->param.b_cabac && (h->cabac.p_end - h->cabac.p < 2500)) ||
325 (h->out.bs.p_end - h->out.bs.p < 2500) )
327 h->out.i_bitstream += 100000;
328 CHECKED_MALLOC( h->out.p_bitstream, h->out.i_bitstream );
329 h->mc.memcpy_aligned( h->out.p_bitstream, bs_bak, (h->out.i_bitstream - 100000) & ~15 );
330 intptr_t delta = h->out.p_bitstream - bs_bak;
332 h->out.bs.p_start += delta;
333 h->out.bs.p += delta;
334 h->out.bs.p_end = h->out.p_bitstream + h->out.i_bitstream;
336 h->cabac.p_start += delta;
338 h->cabac.p_end = h->out.p_bitstream + h->out.i_bitstream;
340 for( int i = 0; i <= h->out.i_nal; i++ )
341 h->out.nal[i].p_payload += delta;
350 /****************************************************************************
352 ****************************************************************************
353 ****************************** External API*********************************
354 ****************************************************************************
356 ****************************************************************************/
358 static int x264_validate_parameters( x264_t *h )
361 if( !(x264_cpu_detect() & X264_CPU_SSE) )
363 x264_log( h, X264_LOG_ERROR, "your cpu does not support SSE1, but x264 was compiled with asm support\n");
364 x264_log( h, X264_LOG_ERROR, "to run x264, recompile without asm support (configure --disable-asm)\n");
368 if( h->param.i_width <= 0 || h->param.i_height <= 0 )
370 x264_log( h, X264_LOG_ERROR, "invalid width x height (%dx%d)\n",
371 h->param.i_width, h->param.i_height );
375 if( h->param.i_width % 2 || h->param.i_height % 2 )
377 x264_log( h, X264_LOG_ERROR, "width or height not divisible by 2 (%dx%d)\n",
378 h->param.i_width, h->param.i_height );
381 int i_csp = h->param.i_csp & X264_CSP_MASK;
382 if( i_csp != X264_CSP_I420 && i_csp != X264_CSP_YV12 )
384 x264_log( h, X264_LOG_ERROR, "invalid CSP (only I420/YV12 supported)\n" );
388 if( h->param.i_threads == X264_THREADS_AUTO )
389 h->param.i_threads = x264_cpu_num_processors() * (h->param.b_sliced_threads?2:3)/2;
390 h->param.i_threads = x264_clip3( h->param.i_threads, 1, X264_THREAD_MAX );
391 if( h->param.i_threads > 1 )
394 x264_log( h, X264_LOG_WARNING, "not compiled with pthread support!\n");
395 h->param.i_threads = 1;
397 /* Avoid absurdly small thread slices as they can reduce performance
398 * and VBV compliance. Capped at an arbitrary 4 rows per thread. */
399 if( h->param.b_sliced_threads )
401 int max_threads = (h->param.i_height+15)/16 / 4;
402 h->param.i_threads = X264_MIN( h->param.i_threads, max_threads );
406 h->param.b_sliced_threads = 0;
407 h->i_thread_frames = h->param.b_sliced_threads ? 1 : h->param.i_threads;
409 if( h->param.b_interlaced )
411 if( h->param.analyse.i_me_method >= X264_ME_ESA )
413 x264_log( h, X264_LOG_WARNING, "interlace + me=esa is not implemented\n" );
414 h->param.analyse.i_me_method = X264_ME_UMH;
416 if( h->param.analyse.i_weighted_pred > 0 )
418 x264_log( h, X264_LOG_WARNING, "interlace + weightp is not implemented\n" );
419 h->param.analyse.i_weighted_pred = X264_WEIGHTP_NONE;
423 /* Detect default ffmpeg settings and terminate with an error. */
426 score += h->param.analyse.i_me_range == 0;
427 score += h->param.rc.i_qp_step == 3;
428 score += h->param.i_keyint_max == 12;
429 score += h->param.rc.i_qp_min == 2;
430 score += h->param.rc.i_qp_max == 31;
431 score += h->param.rc.f_qcompress == 0.5;
432 score += fabs(h->param.rc.f_ip_factor - 1.25) < 0.01;
433 score += fabs(h->param.rc.f_pb_factor - 1.25) < 0.01;
434 score += h->param.analyse.inter == 0 && h->param.analyse.i_subpel_refine == 8;
437 x264_log( h, X264_LOG_ERROR, "broken ffmpeg default settings detected\n" );
438 x264_log( h, X264_LOG_ERROR, "use an encoding preset (vpre)\n" );
443 if( h->param.rc.i_rc_method < 0 || h->param.rc.i_rc_method > 2 )
445 x264_log( h, X264_LOG_ERROR, "no ratecontrol method specified\n" );
448 h->param.rc.f_rf_constant = x264_clip3f( h->param.rc.f_rf_constant, 0, 51 );
449 h->param.rc.i_qp_constant = x264_clip3( h->param.rc.i_qp_constant, 0, 51 );
450 if( h->param.rc.i_rc_method == X264_RC_CRF )
452 h->param.rc.i_qp_constant = h->param.rc.f_rf_constant;
453 h->param.rc.i_bitrate = 0;
455 if( (h->param.rc.i_rc_method == X264_RC_CQP || h->param.rc.i_rc_method == X264_RC_CRF)
456 && h->param.rc.i_qp_constant == 0 )
458 h->mb.b_lossless = 1;
459 h->param.i_cqm_preset = X264_CQM_FLAT;
460 h->param.psz_cqm_file = NULL;
461 h->param.rc.i_rc_method = X264_RC_CQP;
462 h->param.rc.f_ip_factor = 1;
463 h->param.rc.f_pb_factor = 1;
464 h->param.analyse.b_psnr = 0;
465 h->param.analyse.b_ssim = 0;
466 h->param.analyse.i_chroma_qp_offset = 0;
467 h->param.analyse.i_trellis = 0;
468 h->param.analyse.b_fast_pskip = 0;
469 h->param.analyse.i_noise_reduction = 0;
470 h->param.analyse.b_psy = 0;
471 h->param.i_bframe = 0;
472 /* 8x8dct is not useful at all in CAVLC lossless */
473 if( !h->param.b_cabac )
474 h->param.analyse.b_transform_8x8 = 0;
476 if( h->param.rc.i_rc_method == X264_RC_CQP )
478 float qp_p = h->param.rc.i_qp_constant;
479 float qp_i = qp_p - 6*log2f( h->param.rc.f_ip_factor );
480 float qp_b = qp_p + 6*log2f( h->param.rc.f_pb_factor );
481 h->param.rc.i_qp_min = x264_clip3( (int)(X264_MIN3( qp_p, qp_i, qp_b )), 0, 51 );
482 h->param.rc.i_qp_max = x264_clip3( (int)(X264_MAX3( qp_p, qp_i, qp_b ) + .999), 0, 51 );
483 h->param.rc.i_aq_mode = 0;
484 h->param.rc.b_mb_tree = 0;
486 h->param.rc.i_qp_max = x264_clip3( h->param.rc.i_qp_max, 0, 51 );
487 h->param.rc.i_qp_min = x264_clip3( h->param.rc.i_qp_min, 0, h->param.rc.i_qp_max );
488 if( h->param.rc.i_vbv_buffer_size )
490 if( h->param.rc.i_rc_method == X264_RC_CQP )
492 x264_log( h, X264_LOG_WARNING, "VBV is incompatible with constant QP, ignored.\n" );
493 h->param.rc.i_vbv_max_bitrate = 0;
494 h->param.rc.i_vbv_buffer_size = 0;
496 else if( h->param.rc.i_vbv_max_bitrate == 0 )
498 if( h->param.rc.i_rc_method == X264_RC_ABR )
500 x264_log( h, X264_LOG_WARNING, "VBV maxrate unspecified, assuming CBR\n" );
501 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate;
505 x264_log( h, X264_LOG_WARNING, "VBV bufsize set but maxrate unspecified, ignored\n" );
506 h->param.rc.i_vbv_buffer_size = 0;
509 else if( h->param.rc.i_vbv_max_bitrate < h->param.rc.i_bitrate &&
510 h->param.rc.i_rc_method == X264_RC_ABR )
512 x264_log( h, X264_LOG_WARNING, "max bitrate less than average bitrate, assuming CBR\n" );
513 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate;
516 else if( h->param.rc.i_vbv_max_bitrate )
518 x264_log( h, X264_LOG_WARNING, "VBV maxrate specified, but no bufsize, ignored\n" );
519 h->param.rc.i_vbv_max_bitrate = 0;
522 if( h->param.b_interlaced && h->param.i_slice_max_size )
524 x264_log( h, X264_LOG_WARNING, "interlaced + slice-max-size is not implemented\n" );
525 h->param.i_slice_max_size = 0;
527 if( h->param.b_interlaced && h->param.i_slice_max_mbs )
529 x264_log( h, X264_LOG_WARNING, "interlaced + slice-max-mbs is not implemented\n" );
530 h->param.i_slice_max_mbs = 0;
532 int max_slices = (h->param.i_height+((16<<h->param.b_interlaced)-1))/(16<<h->param.b_interlaced);
533 if( h->param.b_sliced_threads )
534 h->param.i_slice_count = x264_clip3( h->param.i_threads, 0, max_slices );
537 h->param.i_slice_count = x264_clip3( h->param.i_slice_count, 0, max_slices );
538 h->param.i_slice_max_size = X264_MAX( h->param.i_slice_max_size, 0 );
539 h->param.i_slice_max_mbs = X264_MAX( h->param.i_slice_max_mbs, 0 );
540 if( h->param.i_slice_max_mbs || h->param.i_slice_max_size )
541 h->param.i_slice_count = 0;
544 h->param.i_frame_reference = x264_clip3( h->param.i_frame_reference, 1, 16 );
545 if( h->param.i_keyint_max <= 0 )
546 h->param.i_keyint_max = 1;
547 if( h->param.i_scenecut_threshold < 0 )
548 h->param.i_scenecut_threshold = 0;
549 if( !h->param.analyse.i_subpel_refine && h->param.analyse.i_direct_mv_pred > X264_DIRECT_PRED_SPATIAL )
551 x264_log( h, X264_LOG_WARNING, "subme=0 + direct=temporal is not supported\n" );
552 h->param.analyse.i_direct_mv_pred = X264_DIRECT_PRED_SPATIAL;
554 h->param.i_bframe = x264_clip3( h->param.i_bframe, 0, X264_BFRAME_MAX );
555 if( h->param.i_keyint_max == 1 )
557 h->param.i_bframe = 0;
558 h->param.b_intra_refresh = 0;
560 h->param.i_bframe_bias = x264_clip3( h->param.i_bframe_bias, -90, 100 );
561 if( h->param.i_bframe <= 1 )
562 h->param.i_bframe_pyramid = X264_B_PYRAMID_NONE;
563 h->param.i_bframe_pyramid = x264_clip3( h->param.i_bframe_pyramid, X264_B_PYRAMID_NONE, X264_B_PYRAMID_NORMAL );
564 if( !h->param.i_bframe )
566 h->param.i_bframe_adaptive = X264_B_ADAPT_NONE;
567 h->param.analyse.i_direct_mv_pred = 0;
568 h->param.analyse.b_weighted_bipred = 0;
570 if( h->param.b_intra_refresh && h->param.i_bframe_pyramid == X264_B_PYRAMID_NORMAL )
572 x264_log( h, X264_LOG_WARNING, "b-pyramid normal + intra-refresh is not supported\n" );
573 h->param.i_bframe_pyramid = X264_B_PYRAMID_STRICT;
575 if( h->param.b_intra_refresh && h->param.i_frame_reference > 1 )
577 x264_log( h, X264_LOG_WARNING, "ref > 1 + intra-refresh is not supported\n" );
578 h->param.i_frame_reference = 1;
580 if( h->param.i_keyint_min == X264_KEYINT_MIN_AUTO )
581 h->param.i_keyint_min = h->param.i_keyint_max / 10;
582 h->param.i_keyint_min = x264_clip3( h->param.i_keyint_min, 1, h->param.i_keyint_max/2+1 );
583 h->param.rc.i_lookahead = x264_clip3( h->param.rc.i_lookahead, 0, X264_LOOKAHEAD_MAX );
585 int maxrate = X264_MAX( h->param.rc.i_vbv_max_bitrate, h->param.rc.i_bitrate );
586 float bufsize = maxrate ? (float)h->param.rc.i_vbv_buffer_size / maxrate : 0;
587 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;
588 h->param.rc.i_lookahead = X264_MIN( h->param.rc.i_lookahead, X264_MAX( h->param.i_keyint_max, bufsize*fps ) );
591 if( !h->param.i_timebase_num || !h->param.i_timebase_den )
593 h->param.i_timebase_num = h->param.i_fps_den;
594 h->param.i_timebase_den = h->param.i_fps_num;
597 h->param.rc.f_qcompress = x264_clip3f( h->param.rc.f_qcompress, 0.0, 1.0 );
598 if( !h->param.rc.i_lookahead || h->param.i_keyint_max == 1 || h->param.rc.f_qcompress == 1 )
599 h->param.rc.b_mb_tree = 0;
600 if( h->param.rc.b_stat_read )
601 h->param.rc.i_lookahead = 0;
603 if( h->param.i_sync_lookahead )
604 h->param.i_sync_lookahead = x264_clip3( h->param.i_sync_lookahead, h->i_thread_frames + h->param.i_bframe, X264_LOOKAHEAD_MAX );
605 if( h->param.rc.b_stat_read || h->i_thread_frames == 1 )
606 h->param.i_sync_lookahead = 0;
608 h->param.i_sync_lookahead = 0;
611 h->param.i_deblocking_filter_alphac0 = x264_clip3( h->param.i_deblocking_filter_alphac0, -6, 6 );
612 h->param.i_deblocking_filter_beta = x264_clip3( h->param.i_deblocking_filter_beta, -6, 6 );
613 h->param.analyse.i_luma_deadzone[0] = x264_clip3( h->param.analyse.i_luma_deadzone[0], 0, 32 );
614 h->param.analyse.i_luma_deadzone[1] = x264_clip3( h->param.analyse.i_luma_deadzone[1], 0, 32 );
616 h->param.i_cabac_init_idc = x264_clip3( h->param.i_cabac_init_idc, 0, 2 );
618 if( h->param.i_cqm_preset < X264_CQM_FLAT || h->param.i_cqm_preset > X264_CQM_CUSTOM )
619 h->param.i_cqm_preset = X264_CQM_FLAT;
621 if( h->param.analyse.i_me_method < X264_ME_DIA ||
622 h->param.analyse.i_me_method > X264_ME_TESA )
623 h->param.analyse.i_me_method = X264_ME_HEX;
624 if( h->param.analyse.i_me_range < 4 )
625 h->param.analyse.i_me_range = 4;
626 if( h->param.analyse.i_me_range > 16 && h->param.analyse.i_me_method <= X264_ME_HEX )
627 h->param.analyse.i_me_range = 16;
628 if( h->param.analyse.i_me_method == X264_ME_TESA &&
629 (h->mb.b_lossless || h->param.analyse.i_subpel_refine <= 1) )
630 h->param.analyse.i_me_method = X264_ME_ESA;
631 h->param.analyse.i_subpel_refine = x264_clip3( h->param.analyse.i_subpel_refine, 0, 10 );
632 h->param.analyse.b_mixed_references = h->param.analyse.b_mixed_references && h->param.i_frame_reference > 1;
633 h->param.analyse.inter &= X264_ANALYSE_PSUB16x16|X264_ANALYSE_PSUB8x8|X264_ANALYSE_BSUB16x16|
634 X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
635 h->param.analyse.intra &= X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
636 if( !(h->param.analyse.inter & X264_ANALYSE_PSUB16x16) )
637 h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
638 if( !h->param.analyse.b_transform_8x8 )
640 h->param.analyse.inter &= ~X264_ANALYSE_I8x8;
641 h->param.analyse.intra &= ~X264_ANALYSE_I8x8;
643 h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -12, 12);
644 if( !h->param.b_cabac )
645 h->param.analyse.i_trellis = 0;
646 h->param.analyse.i_trellis = x264_clip3( h->param.analyse.i_trellis, 0, 2 );
647 if( !h->param.analyse.b_psy )
649 h->param.analyse.f_psy_rd = 0;
650 h->param.analyse.f_psy_trellis = 0;
652 if( !h->param.analyse.i_trellis )
653 h->param.analyse.f_psy_trellis = 0;
654 h->param.analyse.f_psy_rd = x264_clip3f( h->param.analyse.f_psy_rd, 0, 10 );
655 h->param.analyse.f_psy_trellis = x264_clip3f( h->param.analyse.f_psy_trellis, 0, 10 );
656 if( h->param.analyse.i_subpel_refine < 6 )
657 h->param.analyse.f_psy_rd = 0;
658 h->mb.i_psy_rd = FIX8( h->param.analyse.f_psy_rd );
659 /* Psy RDO increases overall quantizers to improve the quality of luma--this indirectly hurts chroma quality */
660 /* so we lower the chroma QP offset to compensate */
661 /* This can be triggered repeatedly on multiple calls to parameter_validate, but since encoding
662 * uses the pps chroma qp offset not the param chroma qp offset, this is not a problem. */
664 h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_rd < 0.25 ? 1 : 2;
665 h->mb.i_psy_trellis = FIX8( h->param.analyse.f_psy_trellis / 4 );
666 /* Psy trellis has a similar effect. */
667 if( h->mb.i_psy_trellis )
668 h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_trellis < 0.25 ? 1 : 2;
669 h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -12, 12);
670 h->param.rc.i_aq_mode = x264_clip3( h->param.rc.i_aq_mode, 0, 2 );
671 h->param.rc.f_aq_strength = x264_clip3f( h->param.rc.f_aq_strength, 0, 3 );
672 if( h->param.rc.f_aq_strength == 0 )
673 h->param.rc.i_aq_mode = 0;
674 /* MB-tree requires AQ to be on, even if the strength is zero. */
675 if( !h->param.rc.i_aq_mode && h->param.rc.b_mb_tree )
677 h->param.rc.i_aq_mode = 1;
678 h->param.rc.f_aq_strength = 0;
680 h->param.analyse.i_noise_reduction = x264_clip3( h->param.analyse.i_noise_reduction, 0, 1<<16 );
681 if( h->param.analyse.i_subpel_refine == 10 && (h->param.analyse.i_trellis != 2 || !h->param.rc.i_aq_mode) )
682 h->param.analyse.i_subpel_refine = 9;
685 const x264_level_t *l = x264_levels;
686 if( h->param.i_level_idc < 0 )
688 int maxrate_bak = h->param.rc.i_vbv_max_bitrate;
689 if( h->param.rc.i_rc_method == X264_RC_ABR && h->param.rc.i_vbv_buffer_size <= 0 )
690 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate * 2;
691 h->sps = h->sps_array;
692 x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
693 do h->param.i_level_idc = l->level_idc;
694 while( l[1].level_idc && x264_validate_levels( h, 0 ) && l++ );
695 h->param.rc.i_vbv_max_bitrate = maxrate_bak;
699 while( l->level_idc && l->level_idc != h->param.i_level_idc )
701 if( l->level_idc == 0 )
703 x264_log( h, X264_LOG_ERROR, "invalid level_idc: %d\n", h->param.i_level_idc );
707 if( h->param.analyse.i_mv_range <= 0 )
708 h->param.analyse.i_mv_range = l->mv_range >> h->param.b_interlaced;
710 h->param.analyse.i_mv_range = x264_clip3(h->param.analyse.i_mv_range, 32, 512 >> h->param.b_interlaced);
713 h->param.analyse.i_weighted_pred = x264_clip3( h->param.analyse.i_weighted_pred, 0, X264_WEIGHTP_SMART );
714 if( !h->param.analyse.i_weighted_pred && h->param.rc.b_mb_tree && h->param.analyse.b_psy && !h->param.b_interlaced )
715 h->param.analyse.i_weighted_pred = X264_WEIGHTP_FAKE;
717 if( h->i_thread_frames > 1 )
719 int r = h->param.analyse.i_mv_range_thread;
723 // half of the available space is reserved and divided evenly among the threads,
724 // the rest is allocated to whichever thread is far enough ahead to use it.
725 // reserving more space increases quality for some videos, but costs more time
726 // in thread synchronization.
727 int max_range = (h->param.i_height + X264_THREAD_HEIGHT) / h->i_thread_frames - X264_THREAD_HEIGHT;
730 r = X264_MAX( r, h->param.analyse.i_me_range );
731 r = X264_MIN( r, h->param.analyse.i_mv_range );
732 // round up to use the whole mb row
733 r2 = (r & ~15) + ((-X264_THREAD_HEIGHT) & 15);
736 x264_log( h, X264_LOG_DEBUG, "using mv_range_thread = %d\n", r2 );
737 h->param.analyse.i_mv_range_thread = r2;
740 if( h->param.rc.f_qblur < 0 )
741 h->param.rc.f_qblur = 0;
742 if( h->param.rc.f_complexity_blur < 0 )
743 h->param.rc.f_complexity_blur = 0;
745 h->param.i_sps_id &= 31;
747 if( h->param.i_log_level < X264_LOG_INFO )
749 h->param.analyse.b_psnr = 0;
750 h->param.analyse.b_ssim = 0;
753 if( h->param.b_interlaced )
754 h->param.b_pic_struct = 1;
756 if( h->param.i_nal_hrd && !h->param.rc.i_vbv_buffer_size )
758 x264_log( h, X264_LOG_WARNING, "NAL HRD parameters require VBV parameters\n" );
759 h->param.i_nal_hrd = X264_NAL_HRD_NONE;
762 if( h->param.i_nal_hrd == X264_NAL_HRD_CBR &&
763 (h->param.rc.i_bitrate != h->param.rc.i_vbv_max_bitrate || !h->param.rc.i_vbv_max_bitrate) )
765 x264_log( h, X264_LOG_WARNING, "CBR HRD requires constant bitrate\n" );
766 h->param.i_nal_hrd = X264_NAL_HRD_VBR;
769 /* ensure the booleans are 0 or 1 so they can be used in math */
770 #define BOOLIFY(x) h->param.x = !!h->param.x
772 BOOLIFY( b_constrained_intra );
773 BOOLIFY( b_deblocking_filter );
774 BOOLIFY( b_deterministic );
775 BOOLIFY( b_sliced_threads );
776 BOOLIFY( b_interlaced );
777 BOOLIFY( b_intra_refresh );
778 BOOLIFY( b_visualize );
780 BOOLIFY( b_repeat_headers );
782 BOOLIFY( b_vfr_input );
783 BOOLIFY( b_pic_struct );
784 BOOLIFY( analyse.b_transform_8x8 );
785 BOOLIFY( analyse.b_weighted_bipred );
786 BOOLIFY( analyse.b_chroma_me );
787 BOOLIFY( analyse.b_mixed_references );
788 BOOLIFY( analyse.b_fast_pskip );
789 BOOLIFY( analyse.b_dct_decimate );
790 BOOLIFY( analyse.b_psy );
791 BOOLIFY( analyse.b_psnr );
792 BOOLIFY( analyse.b_ssim );
793 BOOLIFY( rc.b_stat_write );
794 BOOLIFY( rc.b_stat_read );
795 BOOLIFY( rc.b_mb_tree );
801 static void mbcmp_init( x264_t *h )
803 int satd = !h->mb.b_lossless && h->param.analyse.i_subpel_refine > 1;
804 memcpy( h->pixf.mbcmp, satd ? h->pixf.satd : h->pixf.sad_aligned, sizeof(h->pixf.mbcmp) );
805 memcpy( h->pixf.mbcmp_unaligned, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.mbcmp_unaligned) );
806 h->pixf.intra_mbcmp_x3_16x16 = satd ? h->pixf.intra_satd_x3_16x16 : h->pixf.intra_sad_x3_16x16;
807 h->pixf.intra_mbcmp_x3_8x8c = satd ? h->pixf.intra_satd_x3_8x8c : h->pixf.intra_sad_x3_8x8c;
808 h->pixf.intra_mbcmp_x3_4x4 = satd ? h->pixf.intra_satd_x3_4x4 : h->pixf.intra_sad_x3_4x4;
809 satd &= h->param.analyse.i_me_method == X264_ME_TESA;
810 memcpy( h->pixf.fpelcmp, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.fpelcmp) );
811 memcpy( h->pixf.fpelcmp_x3, satd ? h->pixf.satd_x3 : h->pixf.sad_x3, sizeof(h->pixf.fpelcmp_x3) );
812 memcpy( h->pixf.fpelcmp_x4, satd ? h->pixf.satd_x4 : h->pixf.sad_x4, sizeof(h->pixf.fpelcmp_x4) );
815 static void x264_set_aspect_ratio( x264_t *h, x264_param_t *param, int initial )
818 if( param->vui.i_sar_width > 0 && param->vui.i_sar_height > 0 )
820 uint32_t i_w = param->vui.i_sar_width;
821 uint32_t i_h = param->vui.i_sar_height;
822 uint32_t old_w = h->param.vui.i_sar_width;
823 uint32_t old_h = h->param.vui.i_sar_height;
825 x264_reduce_fraction( &i_w, &i_h );
827 while( i_w > 65535 || i_h > 65535 )
833 x264_reduce_fraction( &i_w, &i_h );
835 if( i_w != old_w || i_h != old_h || initial )
837 h->param.vui.i_sar_width = 0;
838 h->param.vui.i_sar_height = 0;
839 if( i_w == 0 || i_h == 0 )
840 x264_log( h, X264_LOG_WARNING, "cannot create valid sample aspect ratio\n" );
843 x264_log( h, initial?X264_LOG_INFO:X264_LOG_DEBUG, "using SAR=%d/%d\n", i_w, i_h );
844 h->param.vui.i_sar_width = i_w;
845 h->param.vui.i_sar_height = i_h;
851 /****************************************************************************
853 ****************************************************************************/
854 x264_t *x264_encoder_open( x264_param_t *param )
858 int qp, i_slicetype_length;
860 CHECKED_MALLOCZERO( h, sizeof(x264_t) );
862 /* Create a copy of param */
863 memcpy( &h->param, param, sizeof(x264_param_t) );
865 if( param->param_free )
866 param->param_free( param );
868 if( x264_validate_parameters( h ) < 0 )
871 if( h->param.psz_cqm_file )
872 if( x264_cqm_parse_file( h, h->param.psz_cqm_file ) < 0 )
875 if( h->param.rc.psz_stat_out )
876 h->param.rc.psz_stat_out = strdup( h->param.rc.psz_stat_out );
877 if( h->param.rc.psz_stat_in )
878 h->param.rc.psz_stat_in = strdup( h->param.rc.psz_stat_in );
880 x264_set_aspect_ratio( h, &h->param, 1 );
882 x264_reduce_fraction( &h->param.i_fps_num, &h->param.i_fps_den );
883 x264_reduce_fraction( &h->param.i_timebase_num, &h->param.i_timebase_den );
889 uint64_t new_timebase_den = h->param.i_timebase_den;
890 if( h->param.b_dts_compress )
892 /* h->i_dts_compress_multiplier == h->frames.i_bframe_delay + 1 */
893 h->i_dts_compress_multiplier = h->param.i_bframe ? (h->param.i_bframe_pyramid ? 3 : 2) : 1;
894 if( h->i_dts_compress_multiplier != 1 )
896 new_timebase_den = h->param.i_timebase_den * h->i_dts_compress_multiplier;
897 x264_log( h, X264_LOG_DEBUG, "DTS compresion changed timebase: %u/%u -> %u/%"PRIu64"\n",
898 h->param.i_timebase_num, h->param.i_timebase_den,
899 h->param.i_timebase_num, new_timebase_den );
903 h->i_dts_compress_multiplier = 1;
905 if( new_timebase_den * 2 > UINT32_MAX )
907 x264_log( h, X264_LOG_ERROR, "Effective timebase denominator %"PRIu64" exceeds H.264 maximum\n", new_timebase_den );
910 h->param.i_timebase_den = new_timebase_den;
912 h->sps = &h->sps_array[0];
913 x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
915 h->pps = &h->pps_array[0];
916 x264_pps_init( h->pps, h->param.i_sps_id, &h->param, h->sps );
918 x264_validate_levels( h, 1 );
920 h->chroma_qp_table = i_chroma_qp_table + 12 + h->pps->i_chroma_qp_index_offset;
922 if( x264_cqm_init( h ) < 0 )
925 h->mb.i_mb_count = h->sps->i_mb_width * h->sps->i_mb_height;
928 if( h->param.i_bframe_adaptive == X264_B_ADAPT_TRELLIS )
929 h->frames.i_delay = X264_MAX(h->param.i_bframe,3)*4;
931 h->frames.i_delay = h->param.i_bframe;
932 if( h->param.rc.b_mb_tree || h->param.rc.i_vbv_buffer_size )
933 h->frames.i_delay = X264_MAX( h->frames.i_delay, h->param.rc.i_lookahead );
934 i_slicetype_length = h->frames.i_delay;
935 h->frames.i_delay += h->i_thread_frames - 1;
936 h->frames.i_delay = X264_MIN( h->frames.i_delay, X264_LOOKAHEAD_MAX );
937 h->frames.i_delay += h->param.i_sync_lookahead;
938 h->frames.i_delay += h->param.b_vfr_input && (h->param.rc.i_rc_method == X264_RC_ABR || h->param.rc.b_stat_write
939 || h->param.rc.i_vbv_buffer_size);
940 h->frames.i_bframe_delay = h->param.i_bframe ? (h->param.i_bframe_pyramid ? 2 : 1) : 0;
942 h->frames.i_max_ref0 = h->param.i_frame_reference;
943 h->frames.i_max_ref1 = X264_MIN( h->sps->vui.i_num_reorder_frames, h->param.i_frame_reference );
944 h->frames.i_max_dpb = h->sps->vui.i_max_dec_frame_buffering;
945 h->frames.b_have_lowres = !h->param.rc.b_stat_read
946 && ( h->param.rc.i_rc_method == X264_RC_ABR
947 || h->param.rc.i_rc_method == X264_RC_CRF
948 || h->param.i_bframe_adaptive
949 || h->param.i_scenecut_threshold
950 || h->param.rc.b_mb_tree
951 || h->param.analyse.i_weighted_pred == X264_WEIGHTP_SMART );
952 h->frames.b_have_lowres |= h->param.rc.b_stat_read && h->param.rc.i_vbv_buffer_size > 0;
953 h->frames.b_have_sub8x8_esa = !!(h->param.analyse.inter & X264_ANALYSE_PSUB8x8);
955 h->frames.i_last_keyframe = - h->param.i_keyint_max;
956 h->frames.i_input = 0;
957 h->frames.i_largest_pts = h->frames.i_second_largest_pts = -1;
959 CHECKED_MALLOCZERO( h->frames.unused[0], (h->frames.i_delay + 3) * sizeof(x264_frame_t *) );
960 /* Allocate room for max refs plus a few extra just in case. */
961 CHECKED_MALLOCZERO( h->frames.unused[1], (h->i_thread_frames + 20) * sizeof(x264_frame_t *) );
962 CHECKED_MALLOCZERO( h->frames.current, (h->param.i_sync_lookahead + h->param.i_bframe
963 + h->i_thread_frames + 3) * sizeof(x264_frame_t *) );
964 if( h->param.analyse.i_weighted_pred > 0 )
965 CHECKED_MALLOCZERO( h->frames.blank_unused, h->i_thread_frames * 4 * sizeof(x264_frame_t *) );
968 h->i_cpb_delay = h->i_coded_fields = h->i_disp_fields = h->i_prev_duration = 0;
969 h->i_disp_fields_last_frame = -1;
972 /* init CPU functions */
973 x264_predict_16x16_init( h->param.cpu, h->predict_16x16 );
974 x264_predict_8x8c_init( h->param.cpu, h->predict_8x8c );
975 x264_predict_8x8_init( h->param.cpu, h->predict_8x8, &h->predict_8x8_filter );
976 x264_predict_4x4_init( h->param.cpu, h->predict_4x4 );
977 if( !h->param.b_cabac )
978 x264_init_vlc_tables();
979 x264_pixel_init( h->param.cpu, &h->pixf );
980 x264_dct_init( h->param.cpu, &h->dctf );
981 x264_zigzag_init( h->param.cpu, &h->zigzagf, h->param.b_interlaced );
982 x264_mc_init( h->param.cpu, &h->mc );
983 x264_quant_init( h, h->param.cpu, &h->quantf );
984 x264_deblock_init( h->param.cpu, &h->loopf );
985 x264_dct_init_weights();
989 p = buf + sprintf( buf, "using cpu capabilities:" );
990 for( int i = 0; x264_cpu_names[i].flags; i++ )
992 if( !strcmp(x264_cpu_names[i].name, "SSE2")
993 && h->param.cpu & (X264_CPU_SSE2_IS_FAST|X264_CPU_SSE2_IS_SLOW) )
995 if( !strcmp(x264_cpu_names[i].name, "SSE3")
996 && (h->param.cpu & X264_CPU_SSSE3 || !(h->param.cpu & X264_CPU_CACHELINE_64)) )
998 if( !strcmp(x264_cpu_names[i].name, "SSE4.1")
999 && (h->param.cpu & X264_CPU_SSE42) )
1001 if( (h->param.cpu & x264_cpu_names[i].flags) == x264_cpu_names[i].flags
1002 && (!i || x264_cpu_names[i].flags != x264_cpu_names[i-1].flags) )
1003 p += sprintf( p, " %s", x264_cpu_names[i].name );
1006 p += sprintf( p, " none!" );
1007 x264_log( h, X264_LOG_INFO, "%s\n", buf );
1009 for( qp = h->param.rc.i_qp_min; qp <= h->param.rc.i_qp_max; qp++ )
1010 if( x264_analyse_init_costs( h, qp ) )
1012 if( x264_analyse_init_costs( h, X264_LOOKAHEAD_QP ) )
1015 /* Checks for known miscompilation issues. */
1016 if( h->cost_mv[1][2013] != 24 )
1018 x264_log( h, X264_LOG_ERROR, "MV cost test failed: x264 has been miscompiled!\n" );
1022 /* Must be volatile or else GCC will optimize it out. */
1023 volatile int temp = 392;
1024 if( x264_clz( temp ) != 23 )
1026 x264_log( h, X264_LOG_ERROR, "CLZ test failed: x264 has been miscompiled!\n" );
1027 #if defined(ARCH_X86) || defined(ARCH_X86_64)
1028 x264_log( h, X264_LOG_ERROR, "Are you attempting to run an SSE4a-targeted build on a CPU that\n" );
1029 x264_log( h, X264_LOG_ERROR, "doesn't support it?\n" );
1035 h->out.i_bitstream = X264_MAX( 1000000, h->param.i_width * h->param.i_height * 4
1036 * ( h->param.rc.i_rc_method == X264_RC_ABR ? pow( 0.95, h->param.rc.i_qp_min )
1037 : pow( 0.95, h->param.rc.i_qp_constant ) * X264_MAX( 1, h->param.rc.f_ip_factor )));
1039 CHECKED_MALLOC( h->nal_buffer, h->out.i_bitstream * 3/2 + 4 );
1040 h->nal_buffer_size = h->out.i_bitstream * 3/2 + 4;
1043 for( int i = 1; i < h->param.i_threads + !!h->param.i_sync_lookahead; i++ )
1044 CHECKED_MALLOC( h->thread[i], sizeof(x264_t) );
1046 if( x264_lookahead_init( h, i_slicetype_length ) )
1049 for( int i = 0; i < h->param.i_threads; i++ )
1051 int init_nal_count = h->param.i_slice_count + 3;
1052 int allocate_threadlocal_data = !h->param.b_sliced_threads || !i;
1056 if( allocate_threadlocal_data )
1058 h->thread[i]->fdec = x264_frame_pop_unused( h, 1 );
1059 if( !h->thread[i]->fdec )
1063 h->thread[i]->fdec = h->thread[0]->fdec;
1065 CHECKED_MALLOC( h->thread[i]->out.p_bitstream, h->out.i_bitstream );
1066 /* Start each thread with room for init_nal_count NAL units; it'll realloc later if needed. */
1067 CHECKED_MALLOC( h->thread[i]->out.nal, init_nal_count*sizeof(x264_nal_t) );
1068 h->thread[i]->out.i_nals_allocated = init_nal_count;
1070 if( allocate_threadlocal_data && x264_macroblock_cache_allocate( h->thread[i] ) < 0 )
1074 for( int i = 0; i < h->param.i_threads; i++ )
1075 if( x264_macroblock_thread_allocate( h->thread[i], 0 ) < 0 )
1078 if( x264_ratecontrol_new( h ) < 0 )
1081 if( h->param.i_nal_hrd )
1083 x264_log( h, X264_LOG_DEBUG, "HRD bitrate: %i bits/sec\n", h->sps->vui.hrd.i_bit_rate_unscaled );
1084 x264_log( h, X264_LOG_DEBUG, "CPB size: %i bits\n", h->sps->vui.hrd.i_cpb_size_unscaled );
1087 if( h->param.psz_dump_yuv )
1089 /* create or truncate the reconstructed video file */
1090 FILE *f = fopen( h->param.psz_dump_yuv, "w" );
1093 x264_log( h, X264_LOG_ERROR, "dump_yuv: can't write to %s\n", h->param.psz_dump_yuv );
1096 else if( !x264_is_regular_file( f ) )
1098 x264_log( h, X264_LOG_ERROR, "dump_yuv: incompatible with non-regular file %s\n", h->param.psz_dump_yuv );
1104 x264_log( h, X264_LOG_INFO, "profile %s, level %d.%d\n",
1105 h->sps->i_profile_idc == PROFILE_BASELINE ? "Baseline" :
1106 h->sps->i_profile_idc == PROFILE_MAIN ? "Main" :
1107 h->sps->i_profile_idc == PROFILE_HIGH ? "High" :
1108 "High 4:4:4 Predictive", h->sps->i_level_idc/10, h->sps->i_level_idc%10 );
1116 /****************************************************************************
1117 * x264_encoder_reconfig:
1118 ****************************************************************************/
1119 int x264_encoder_reconfig( x264_t *h, x264_param_t *param )
1121 int rc_reconfig = 0;
1122 h = h->thread[h->thread[0]->i_thread_phase];
1123 x264_set_aspect_ratio( h, param, 0 );
1124 #define COPY(var) h->param.var = param->var
1125 COPY( i_frame_reference ); // but never uses more refs than initially specified
1126 COPY( i_bframe_bias );
1127 if( h->param.i_scenecut_threshold )
1128 COPY( i_scenecut_threshold ); // can't turn it on or off, only vary the threshold
1129 COPY( b_deblocking_filter );
1130 COPY( i_deblocking_filter_alphac0 );
1131 COPY( i_deblocking_filter_beta );
1132 COPY( analyse.inter );
1133 COPY( analyse.intra );
1134 COPY( analyse.i_direct_mv_pred );
1135 /* Scratch buffer prevents me_range from being increased for esa/tesa */
1136 if( h->param.analyse.i_me_method < X264_ME_ESA || param->analyse.i_me_range < h->param.analyse.i_me_range )
1137 COPY( analyse.i_me_range );
1138 COPY( analyse.i_noise_reduction );
1139 /* We can't switch out of subme=0 during encoding. */
1140 if( h->param.analyse.i_subpel_refine )
1141 COPY( analyse.i_subpel_refine );
1142 COPY( analyse.i_trellis );
1143 COPY( analyse.b_chroma_me );
1144 COPY( analyse.b_dct_decimate );
1145 COPY( analyse.b_fast_pskip );
1146 COPY( analyse.b_mixed_references );
1147 COPY( analyse.f_psy_rd );
1148 COPY( analyse.f_psy_trellis );
1149 // can only twiddle these if they were enabled to begin with:
1150 if( h->param.analyse.i_me_method >= X264_ME_ESA || param->analyse.i_me_method < X264_ME_ESA )
1151 COPY( analyse.i_me_method );
1152 if( h->param.analyse.i_me_method >= X264_ME_ESA && !h->frames.b_have_sub8x8_esa )
1153 h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
1154 if( h->pps->b_transform_8x8_mode )
1155 COPY( analyse.b_transform_8x8 );
1156 if( h->frames.i_max_ref1 > 1 )
1157 COPY( i_bframe_pyramid );
1158 COPY( i_slice_max_size );
1159 COPY( i_slice_max_mbs );
1160 COPY( i_slice_count );
1162 /* VBV can't be turned on if it wasn't on to begin with */
1163 if( h->param.rc.i_vbv_max_bitrate > 0 && h->param.rc.i_vbv_buffer_size > 0 &&
1164 param->rc.i_vbv_max_bitrate > 0 && param->rc.i_vbv_buffer_size > 0 )
1166 COPY( rc.i_vbv_max_bitrate );
1167 COPY( rc.i_vbv_buffer_size );
1168 COPY( rc.i_bitrate );
1171 if( h->param.rc.f_rf_constant != param->rc.f_rf_constant )
1173 COPY( rc.f_rf_constant );
1176 if( h->param.rc.f_rf_constant_max != param->rc.f_rf_constant_max )
1178 COPY( rc.f_rf_constant_max );
1186 int ret = x264_validate_parameters( h );
1188 /* Supported reconfiguration options (1-pass only):
1192 * bitrate (CBR only) */
1193 if( !ret && rc_reconfig )
1194 x264_ratecontrol_init_reconfigurable( h, 0 );
1199 /****************************************************************************
1200 * x264_encoder_parameters:
1201 ****************************************************************************/
1202 void x264_encoder_parameters( x264_t *h, x264_param_t *param )
1204 memcpy( param, &h->thread[h->i_thread_phase]->param, sizeof(x264_param_t) );
1207 /* internal usage */
1208 static void x264_nal_start( x264_t *h, int i_type, int i_ref_idc )
1210 x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1212 nal->i_ref_idc = i_ref_idc;
1213 nal->i_type = i_type;
1216 nal->p_payload= &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8];
1219 /* if number of allocated nals is not enough, re-allocate a larger one. */
1220 static int x264_nal_check_buffer( x264_t *h )
1222 if( h->out.i_nal >= h->out.i_nals_allocated )
1224 x264_nal_t *new_out = x264_malloc( sizeof(x264_nal_t) * (h->out.i_nals_allocated*2) );
1227 memcpy( new_out, h->out.nal, sizeof(x264_nal_t) * (h->out.i_nals_allocated) );
1228 x264_free( h->out.nal );
1229 h->out.nal = new_out;
1230 h->out.i_nals_allocated *= 2;
1235 static int x264_nal_end( x264_t *h )
1237 x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1238 nal->i_payload = &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8] - nal->p_payload;
1241 return x264_nal_check_buffer( h );
1244 static int x264_encoder_encapsulate_nals( x264_t *h, int start )
1246 int nal_size = 0, previous_nal_size = 0;
1248 for( int i = 0; i < start; i++ )
1249 previous_nal_size += h->out.nal[i].i_payload;
1251 for( int i = start; i < h->out.i_nal; i++ )
1252 nal_size += h->out.nal[i].i_payload;
1254 /* Worst-case NAL unit escaping: reallocate the buffer if it's too small. */
1255 if( h->nal_buffer_size < nal_size * 3/2 + h->out.i_nal * 4 )
1257 uint8_t *buf = x264_malloc( nal_size * 2 + h->out.i_nal * 4 );
1260 if( previous_nal_size )
1261 memcpy( buf, h->nal_buffer, previous_nal_size );
1262 x264_free( h->nal_buffer );
1263 h->nal_buffer = buf;
1266 uint8_t *nal_buffer = h->nal_buffer + previous_nal_size;
1268 for( int i = start; i < h->out.i_nal; i++ )
1270 int long_startcode = !i || h->out.nal[i].i_type == NAL_SPS || h->out.nal[i].i_type == NAL_PPS;
1271 int size = x264_nal_encode( nal_buffer, &h->out.nal[i], h->param.b_annexb, long_startcode );
1272 h->out.nal[i].i_payload = size;
1273 h->out.nal[i].p_payload = nal_buffer;
1277 return nal_buffer - (h->nal_buffer + previous_nal_size);
1280 /****************************************************************************
1281 * x264_encoder_headers:
1282 ****************************************************************************/
1283 int x264_encoder_headers( x264_t *h, x264_nal_t **pp_nal, int *pi_nal )
1286 /* init bitstream context */
1288 bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
1290 /* Write SEI, SPS and PPS. */
1292 /* generate sequence parameters */
1293 x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
1294 x264_sps_write( &h->out.bs, h->sps );
1295 if( x264_nal_end( h ) )
1298 /* generate picture parameters */
1299 x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
1300 x264_pps_write( &h->out.bs, h->pps );
1301 if( x264_nal_end( h ) )
1304 /* identify ourselves */
1305 x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
1306 if( x264_sei_version_write( h, &h->out.bs ) )
1308 if( x264_nal_end( h ) )
1311 frame_size = x264_encoder_encapsulate_nals( h, 0 );
1314 *pi_nal = h->out.i_nal;
1315 *pp_nal = &h->out.nal[0];
1321 /* Check to see whether we have chosen a reference list ordering different
1322 * from the standard's default. */
1323 static inline void x264_reference_check_reorder( x264_t *h )
1325 for( int i = 0; i < h->i_ref0 - 1; i++ )
1326 /* P and B-frames use different default orders. */
1327 if( h->sh.i_type == SLICE_TYPE_P ? h->fref0[i]->i_frame_num < h->fref0[i+1]->i_frame_num
1328 : h->fref0[i]->i_poc < h->fref0[i+1]->i_poc )
1330 h->b_ref_reorder[0] = 1;
1335 /* return -1 on failure, else return the index of the new reference frame */
1336 int x264_weighted_reference_duplicate( x264_t *h, int i_ref, const x264_weight_t *w )
1340 x264_frame_t *newframe;
1341 if( i <= 1 ) /* empty list, definitely can't duplicate frame */
1344 /* Find a place to insert the duplicate in the reference list. */
1345 for( j = 0; j < i; j++ )
1346 if( h->fref0[i_ref]->i_frame != h->fref0[j]->i_frame )
1348 /* found a place, after j, make sure there is not already a duplicate there */
1349 if( j == i-1 || ( h->fref0[i_ref]->i_frame != h->fref0[j+1]->i_frame ) )
1353 if( j == i ) /* No room in the reference list for the duplicate. */
1357 newframe = x264_frame_pop_blank_unused( h );
1359 //FIXME: probably don't need to copy everything
1360 *newframe = *h->fref0[i_ref];
1361 newframe->i_reference_count = 1;
1362 newframe->orig = h->fref0[i_ref];
1363 newframe->b_duplicate = 1;
1364 memcpy( h->fenc->weight[j], w, sizeof(h->fenc->weight[i]) );
1366 /* shift the frames to make space for the dupe. */
1367 h->b_ref_reorder[0] = 1;
1368 if( h->i_ref0 < 16 )
1370 h->fref0[15] = NULL;
1371 x264_frame_unshift( &h->fref0[j], newframe );
1376 static void x264_weighted_pred_init( x264_t *h )
1378 /* for now no analysis and set all weights to nothing */
1379 for( int i_ref = 0; i_ref < h->i_ref0; i_ref++ )
1380 h->fenc->weighted[i_ref] = h->fref0[i_ref]->filtered[0];
1382 // FIXME: This only supports weighting of one reference frame
1383 // and duplicates of that frame.
1384 h->fenc->i_lines_weighted = 0;
1386 for( int i_ref = 0; i_ref < (h->i_ref0 << h->sh.b_mbaff); i_ref++ )
1387 for( int i = 0; i < 3; i++ )
1388 h->sh.weight[i_ref][i].weightfn = NULL;
1391 if( h->sh.i_type != SLICE_TYPE_P || h->param.analyse.i_weighted_pred <= 0 )
1394 int i_padv = PADV << h->param.b_interlaced;
1397 int buffer_next = 0;
1398 //FIXME: when chroma support is added, move this into loop
1399 h->sh.weight[0][1].weightfn = h->sh.weight[0][2].weightfn = NULL;
1400 h->sh.weight[0][1].i_denom = h->sh.weight[0][2].i_denom = 0;
1401 for( int j = 0; j < h->i_ref0; j++ )
1403 if( h->fenc->weight[j][0].weightfn )
1405 h->sh.weight[j][0] = h->fenc->weight[j][0];
1406 // if weight is useless, don't write it to stream
1407 if( h->sh.weight[j][0].i_scale == 1<<h->sh.weight[j][0].i_denom && h->sh.weight[j][0].i_offset == 0 )
1408 h->sh.weight[j][0].weightfn = NULL;
1414 h->sh.weight[0][0].i_denom = denom = h->sh.weight[j][0].i_denom;
1415 assert( x264_clip3( denom, 0, 7 ) == denom );
1417 assert( h->sh.weight[j][0].i_denom == denom );
1418 assert( x264_clip3( h->sh.weight[j][0].i_scale, 0, 127 ) == h->sh.weight[j][0].i_scale );
1419 assert( x264_clip3( h->sh.weight[j][0].i_offset, -128, 127 ) == h->sh.weight[j][0].i_offset );
1420 h->fenc->weighted[j] = h->mb.p_weight_buf[buffer_next++] +
1421 h->fenc->i_stride[0] * i_padv + PADH;
1425 //scale full resolution frame
1426 if( h->sh.weight[j][0].weightfn && h->param.i_threads == 1 )
1428 uint8_t *src = h->fref0[j]->filtered[0] - h->fref0[j]->i_stride[0]*i_padv - PADH;
1429 uint8_t *dst = h->fenc->weighted[j] - h->fenc->i_stride[0]*i_padv - PADH;
1430 int stride = h->fenc->i_stride[0];
1431 int width = h->fenc->i_width[0] + PADH*2;
1432 int height = h->fenc->i_lines[0] + i_padv*2;
1433 x264_weight_scale_plane( h, dst, stride, src, stride, width, height, &h->sh.weight[j][0] );
1434 h->fenc->i_lines_weighted = height;
1438 h->sh.weight[0][0].i_denom = 0;
1441 static inline void x264_reference_build_list( x264_t *h, int i_poc )
1445 /* build ref list 0/1 */
1446 h->mb.pic.i_fref[0] = h->i_ref0 = 0;
1447 h->mb.pic.i_fref[1] = h->i_ref1 = 0;
1448 if( h->sh.i_type == SLICE_TYPE_I )
1451 for( int i = 0; h->frames.reference[i]; i++ )
1453 if( h->frames.reference[i]->i_poc < i_poc )
1454 h->fref0[h->i_ref0++] = h->frames.reference[i];
1455 else if( h->frames.reference[i]->i_poc > i_poc )
1456 h->fref1[h->i_ref1++] = h->frames.reference[i];
1459 /* Order ref0 from higher to lower poc */
1463 for( int i = 0; i < h->i_ref0 - 1; i++ )
1465 if( h->fref0[i]->i_poc < h->fref0[i+1]->i_poc )
1467 XCHG( x264_frame_t*, h->fref0[i], h->fref0[i+1] );
1474 if( h->sh.i_mmco_remove_from_end )
1475 for( int i = h->i_ref0-1; i >= h->i_ref0 - h->sh.i_mmco_remove_from_end; i-- )
1477 int diff = h->i_frame_num - h->fref0[i]->i_frame_num;
1478 h->sh.mmco[h->sh.i_mmco_command_count].i_poc = h->fref0[i]->i_poc;
1479 h->sh.mmco[h->sh.i_mmco_command_count++].i_difference_of_pic_nums = diff;
1482 /* Order ref1 from lower to higher poc (bubble sort) for B-frame */
1486 for( int i = 0; i < h->i_ref1 - 1; i++ )
1488 if( h->fref1[i]->i_poc > h->fref1[i+1]->i_poc )
1490 XCHG( x264_frame_t*, h->fref1[i], h->fref1[i+1] );
1497 x264_reference_check_reorder( h );
1499 h->i_ref1 = X264_MIN( h->i_ref1, h->frames.i_max_ref1 );
1500 h->i_ref0 = X264_MIN( h->i_ref0, h->frames.i_max_ref0 );
1501 h->i_ref0 = X264_MIN( h->i_ref0, h->param.i_frame_reference ); // if reconfig() has lowered the limit
1503 /* add duplicates */
1504 if( h->fenc->i_type == X264_TYPE_P )
1507 if( h->param.analyse.i_weighted_pred == X264_WEIGHTP_SMART )
1510 w[1].weightfn = w[2].weightfn = NULL;
1511 if( h->param.rc.b_stat_read )
1512 x264_ratecontrol_set_weights( h, h->fenc );
1514 if( !h->fenc->weight[0][0].weightfn )
1516 h->fenc->weight[0][0].i_denom = 0;
1517 SET_WEIGHT( w[0], 1, 1, 0, -1 );
1518 idx = x264_weighted_reference_duplicate( h, 0, w );
1522 if( h->fenc->weight[0][0].i_scale == 1<<h->fenc->weight[0][0].i_denom )
1524 SET_WEIGHT( h->fenc->weight[0][0], 1, 1, 0, h->fenc->weight[0][0].i_offset );
1526 x264_weighted_reference_duplicate( h, 0, weight_none );
1527 if( h->fenc->weight[0][0].i_offset > -128 )
1529 w[0] = h->fenc->weight[0][0];
1531 h->mc.weight_cache( h, &w[0] );
1532 idx = x264_weighted_reference_duplicate( h, 0, w );
1536 else if( h->param.analyse.i_weighted_pred == X264_WEIGHTP_BLIND )
1538 //weighted offset=-1
1540 SET_WEIGHT( w[0], 1, 1, 0, -1 );
1541 h->fenc->weight[0][0].i_denom = 0;
1542 w[1].weightfn = w[2].weightfn = NULL;
1543 idx = x264_weighted_reference_duplicate( h, 0, w );
1545 h->mb.ref_blind_dupe = idx;
1548 assert( h->i_ref0 + h->i_ref1 <= 16 );
1549 h->mb.pic.i_fref[0] = h->i_ref0;
1550 h->mb.pic.i_fref[1] = h->i_ref1;
1553 static void x264_fdec_filter_row( x264_t *h, int mb_y, int b_inloop )
1555 /* mb_y is the mb to be encoded next, not the mb to be filtered here */
1556 int b_hpel = h->fdec->b_kept_as_ref;
1557 int b_deblock = h->sh.i_disable_deblocking_filter_idc != 1;
1558 int b_end = mb_y == h->i_threadslice_end;
1559 int b_measure_quality = 1;
1560 int min_y = mb_y - (1 << h->sh.b_mbaff);
1561 int b_start = min_y == h->i_threadslice_start;
1562 int max_y = b_end ? h->i_threadslice_end : mb_y;
1563 b_deblock &= b_hpel || h->param.psz_dump_yuv;
1564 if( h->param.b_sliced_threads && b_start && min_y && !b_inloop )
1566 b_deblock = 0; /* We already deblocked on the inloop pass. */
1567 b_measure_quality = 0; /* We already measured quality on the inloop pass. */
1569 if( mb_y & h->sh.b_mbaff )
1571 if( min_y < h->i_threadslice_start )
1574 if( !b_end && b_inloop )
1575 for( int j = 0; j <= h->sh.b_mbaff; j++ )
1576 for( int i = 0; i < 3; i++ )
1578 memcpy( h->intra_border_backup[j][i],
1579 h->fdec->plane[i] + ((mb_y*16 >> !!i) + j - 1 - h->sh.b_mbaff) * h->fdec->i_stride[i],
1580 h->sps->i_mb_width*16 >> !!i );
1584 for( int y = min_y; y < max_y; y += (1 << h->sh.b_mbaff) )
1585 x264_frame_deblock_row( h, y );
1589 int end = mb_y == h->sps->i_mb_height;
1590 x264_frame_expand_border( h, h->fdec, min_y, end );
1591 if( h->param.analyse.i_subpel_refine )
1593 x264_frame_filter( h, h->fdec, min_y, end );
1594 x264_frame_expand_border_filtered( h, h->fdec, min_y, end );
1598 if( h->i_thread_frames > 1 && h->fdec->b_kept_as_ref )
1599 x264_frame_cond_broadcast( h->fdec, mb_y*16 + (b_end ? 10000 : -(X264_THREAD_HEIGHT << h->sh.b_mbaff)) );
1601 min_y = min_y*16 - 8 * !b_start;
1602 max_y = b_end ? X264_MIN( h->i_threadslice_end*16 , h->param.i_height ) : mb_y*16 - 8;
1604 if( b_measure_quality )
1606 if( h->param.analyse.b_psnr )
1607 for( int i = 0; i < 3; i++ )
1608 h->stat.frame.i_ssd[i] +=
1609 x264_pixel_ssd_wxh( &h->pixf,
1610 h->fdec->plane[i] + (min_y>>!!i) * h->fdec->i_stride[i], h->fdec->i_stride[i],
1611 h->fenc->plane[i] + (min_y>>!!i) * h->fenc->i_stride[i], h->fenc->i_stride[i],
1612 h->param.i_width >> !!i, (max_y-min_y) >> !!i );
1614 if( h->param.analyse.b_ssim )
1617 /* offset by 2 pixels to avoid alignment of ssim blocks with dct blocks,
1618 * and overlap by 4 */
1619 min_y += b_start ? 2 : -6;
1620 h->stat.frame.f_ssim +=
1621 x264_pixel_ssim_wxh( &h->pixf,
1622 h->fdec->plane[0] + 2+min_y*h->fdec->i_stride[0], h->fdec->i_stride[0],
1623 h->fenc->plane[0] + 2+min_y*h->fenc->i_stride[0], h->fenc->i_stride[0],
1624 h->param.i_width-2, max_y-min_y, h->scratch_buffer );
1629 static inline int x264_reference_update( x264_t *h )
1631 if( !h->fdec->b_kept_as_ref )
1633 if( h->i_thread_frames > 1 )
1635 x264_frame_push_unused( h, h->fdec );
1636 h->fdec = x264_frame_pop_unused( h, 1 );
1643 /* apply mmco from previous frame. */
1644 for( int i = 0; i < h->sh.i_mmco_command_count; i++ )
1645 for( int j = 0; h->frames.reference[j]; j++ )
1646 if( h->frames.reference[j]->i_poc == h->sh.mmco[i].i_poc )
1647 x264_frame_push_unused( h, x264_frame_shift( &h->frames.reference[j] ) );
1649 /* move frame in the buffer */
1650 x264_frame_push( h->frames.reference, h->fdec );
1651 if( h->frames.reference[h->sps->i_num_ref_frames] )
1652 x264_frame_push_unused( h, x264_frame_shift( h->frames.reference ) );
1653 h->fdec = x264_frame_pop_unused( h, 1 );
1659 static inline void x264_reference_reset( x264_t *h )
1661 while( h->frames.reference[0] )
1662 x264_frame_push_unused( h, x264_frame_pop( h->frames.reference ) );
1667 static inline void x264_reference_hierarchy_reset( x264_t *h )
1670 int b_hasdelayframe = 0;
1671 if( !h->param.i_bframe_pyramid )
1674 /* look for delay frames -- chain must only contain frames that are disposable */
1675 for( int i = 0; h->frames.current[i] && IS_DISPOSABLE( h->frames.current[i]->i_type ); i++ )
1676 b_hasdelayframe |= h->frames.current[i]->i_coded
1677 != h->frames.current[i]->i_frame + h->sps->vui.i_num_reorder_frames;
1679 if( h->param.i_bframe_pyramid != X264_B_PYRAMID_STRICT && !b_hasdelayframe )
1682 /* Remove last BREF. There will never be old BREFs in the
1683 * dpb during a BREF decode when pyramid == STRICT */
1684 for( ref = 0; h->frames.reference[ref]; ref++ )
1686 if( h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT
1687 && h->frames.reference[ref]->i_type == X264_TYPE_BREF )
1689 int diff = h->i_frame_num - h->frames.reference[ref]->i_frame_num;
1690 h->sh.mmco[h->sh.i_mmco_command_count].i_difference_of_pic_nums = diff;
1691 h->sh.mmco[h->sh.i_mmco_command_count++].i_poc = h->frames.reference[ref]->i_poc;
1692 x264_frame_push_unused( h, x264_frame_pop( h->frames.reference ) );
1693 h->b_ref_reorder[0] = 1;
1698 /* Prepare to room in the dpb for the delayed display time of the later b-frame's */
1699 h->sh.i_mmco_remove_from_end = X264_MAX( ref + 2 - h->frames.i_max_dpb, 0 );
1702 static inline void x264_slice_init( x264_t *h, int i_nal_type, int i_global_qp )
1704 /* ------------------------ Create slice header ----------------------- */
1705 if( i_nal_type == NAL_SLICE_IDR )
1707 x264_slice_header_init( h, &h->sh, h->sps, h->pps, h->i_idr_pic_id, h->i_frame_num, i_global_qp );
1710 h->i_idr_pic_id = ( h->i_idr_pic_id + 1 ) % 65536;
1714 x264_slice_header_init( h, &h->sh, h->sps, h->pps, -1, h->i_frame_num, i_global_qp );
1716 h->sh.i_num_ref_idx_l0_active = h->i_ref0 <= 0 ? 1 : h->i_ref0;
1717 h->sh.i_num_ref_idx_l1_active = h->i_ref1 <= 0 ? 1 : h->i_ref1;
1718 if( h->sh.i_num_ref_idx_l0_active != h->pps->i_num_ref_idx_l0_default_active ||
1719 (h->sh.i_type == SLICE_TYPE_B && h->sh.i_num_ref_idx_l1_active != h->pps->i_num_ref_idx_l1_default_active) )
1721 h->sh.b_num_ref_idx_override = 1;
1725 h->fdec->i_frame_num = h->sh.i_frame_num;
1727 if( h->sps->i_poc_type == 0 )
1729 h->sh.i_poc = h->fdec->i_poc;
1730 if( h->param.b_interlaced )
1732 h->sh.i_delta_poc_bottom = h->param.b_tff ? 1 : -1;
1733 if( h->sh.i_delta_poc_bottom == -1 )
1734 h->sh.i_poc = h->fdec->i_poc + 1;
1737 h->sh.i_delta_poc_bottom = 0;
1739 else if( h->sps->i_poc_type == 1 )
1741 /* FIXME TODO FIXME */
1745 /* Nothing to do ? */
1748 x264_macroblock_slice_init( h );
1751 static int x264_slice_write( x264_t *h )
1754 int mb_xy, i_mb_x, i_mb_y;
1755 int i_skip_bak = 0; /* Shut up GCC. */
1757 x264_cabac_t cabac_bak;
1758 uint8_t cabac_prevbyte_bak = 0; /* Shut up GCC. */
1759 /* Assume no more than 3 bytes of NALU escaping.
1760 * NALUs other than the first use a 3-byte startcode. */
1761 int overhead_guess = (NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal)) + 3;
1762 int slice_max_size = h->param.i_slice_max_size > 0 ? (h->param.i_slice_max_size-overhead_guess)*8 : INT_MAX;
1763 int starting_bits = bs_pos(&h->out.bs);
1764 bs_realign( &h->out.bs );
1767 x264_nal_start( h, h->i_nal_type, h->i_nal_ref_idc );
1770 x264_macroblock_thread_init( h );
1772 /* If this isn't the first slice in the threadslice, set the slice QP
1773 * equal to the last QP in the previous slice for more accurate
1774 * CABAC initialization. */
1775 if( h->sh.i_first_mb != h->i_threadslice_start * h->sps->i_mb_width )
1777 h->sh.i_qp = h->mb.i_last_qp;
1778 h->sh.i_qp_delta = h->sh.i_qp - h->pps->i_pic_init_qp;
1781 x264_slice_header_write( &h->out.bs, &h->sh, h->i_nal_ref_idc );
1782 if( h->param.b_cabac )
1784 /* alignment needed */
1785 bs_align_1( &h->out.bs );
1788 x264_cabac_context_init( &h->cabac, h->sh.i_type, h->sh.i_qp, h->sh.i_cabac_init_idc );
1789 x264_cabac_encode_init ( &h->cabac, h->out.bs.p, h->out.bs.p_end );
1791 h->mb.i_last_qp = h->sh.i_qp;
1792 h->mb.i_last_dqp = 0;
1794 i_mb_y = h->sh.i_first_mb / h->sps->i_mb_width;
1795 i_mb_x = h->sh.i_first_mb % h->sps->i_mb_width;
1798 while( (mb_xy = i_mb_x + i_mb_y * h->sps->i_mb_width) <= h->sh.i_last_mb )
1800 int mb_spos = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
1801 if( h->param.i_slice_max_size > 0 )
1803 /* We don't need the contexts because flushing the CABAC encoder has no context
1804 * dependency and macroblocks are only re-encoded in the case where a slice is
1805 * ended (and thus the content of all contexts are thrown away). */
1806 if( h->param.b_cabac )
1808 memcpy( &cabac_bak, &h->cabac, offsetof(x264_cabac_t, f8_bits_encoded) );
1809 /* x264's CABAC writer modifies the previous byte during carry, so it has to be
1811 cabac_prevbyte_bak = h->cabac.p[-1];
1816 i_skip_bak = i_skip;
1820 if( i_mb_x == 0 && !h->mb.b_reencode_mb )
1821 x264_fdec_filter_row( h, i_mb_y, 1 );
1824 x264_macroblock_cache_load( h, i_mb_x, i_mb_y );
1826 x264_macroblock_analyse( h );
1828 /* encode this macroblock -> be careful it can change the mb type to P_SKIP if needed */
1829 x264_macroblock_encode( h );
1831 if( x264_bitstream_check_buffer( h ) )
1834 if( h->param.b_cabac )
1836 if( mb_xy > h->sh.i_first_mb && !(h->sh.b_mbaff && (i_mb_y&1)) )
1837 x264_cabac_encode_terminal( &h->cabac );
1839 if( IS_SKIP( h->mb.i_type ) )
1840 x264_cabac_mb_skip( h, 1 );
1843 if( h->sh.i_type != SLICE_TYPE_I )
1844 x264_cabac_mb_skip( h, 0 );
1845 x264_macroblock_write_cabac( h, &h->cabac );
1850 if( IS_SKIP( h->mb.i_type ) )
1854 if( h->sh.i_type != SLICE_TYPE_I )
1856 bs_write_ue( &h->out.bs, i_skip ); /* skip run */
1859 x264_macroblock_write_cavlc( h );
1863 int total_bits = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
1864 int mb_size = total_bits - mb_spos;
1866 /* We'll just re-encode this last macroblock if we go over the max slice size. */
1867 if( total_bits - starting_bits > slice_max_size && !h->mb.b_reencode_mb )
1869 if( mb_xy != h->sh.i_first_mb )
1871 if( h->param.b_cabac )
1873 memcpy( &h->cabac, &cabac_bak, offsetof(x264_cabac_t, f8_bits_encoded) );
1874 h->cabac.p[-1] = cabac_prevbyte_bak;
1879 i_skip = i_skip_bak;
1881 h->mb.b_reencode_mb = 1;
1882 h->sh.i_last_mb = mb_xy-1;
1887 h->sh.i_last_mb = mb_xy;
1888 h->mb.b_reencode_mb = 0;
1892 h->mb.b_reencode_mb = 0;
1894 #ifdef HAVE_VISUALIZE
1895 if( h->param.b_visualize )
1896 x264_visualize_mb( h );
1900 x264_macroblock_cache_save( h );
1902 /* accumulate mb stats */
1903 h->stat.frame.i_mb_count[h->mb.i_type]++;
1905 if( !IS_INTRA(h->mb.i_type) && !IS_SKIP(h->mb.i_type) && !IS_DIRECT(h->mb.i_type) )
1907 if( h->mb.i_partition != D_8x8 )
1908 h->stat.frame.i_mb_partition[h->mb.i_partition] += 4;
1910 for( int i = 0; i < 4; i++ )
1911 h->stat.frame.i_mb_partition[h->mb.i_sub_partition[i]] ++;
1912 if( h->param.i_frame_reference > 1 )
1913 for( int i_list = 0; i_list <= (h->sh.i_type == SLICE_TYPE_B); i_list++ )
1914 for( int i = 0; i < 4; i++ )
1916 int i_ref = h->mb.cache.ref[i_list][ x264_scan8[4*i] ];
1918 h->stat.frame.i_mb_count_ref[i_list][i_ref] ++;
1922 if( h->param.i_log_level >= X264_LOG_INFO )
1924 if( h->mb.i_cbp_luma || h->mb.i_cbp_chroma )
1926 int cbpsum = (h->mb.i_cbp_luma&1) + ((h->mb.i_cbp_luma>>1)&1)
1927 + ((h->mb.i_cbp_luma>>2)&1) + (h->mb.i_cbp_luma>>3);
1928 int b_intra = IS_INTRA(h->mb.i_type);
1929 h->stat.frame.i_mb_cbp[!b_intra + 0] += cbpsum;
1930 h->stat.frame.i_mb_cbp[!b_intra + 2] += h->mb.i_cbp_chroma >= 1;
1931 h->stat.frame.i_mb_cbp[!b_intra + 4] += h->mb.i_cbp_chroma == 2;
1933 if( h->mb.i_cbp_luma && !IS_INTRA(h->mb.i_type) )
1935 h->stat.frame.i_mb_count_8x8dct[0] ++;
1936 h->stat.frame.i_mb_count_8x8dct[1] += h->mb.b_transform_8x8;
1938 if( IS_INTRA(h->mb.i_type) && h->mb.i_type != I_PCM )
1940 if( h->mb.i_type == I_16x16 )
1941 h->stat.frame.i_mb_pred_mode[0][h->mb.i_intra16x16_pred_mode]++;
1942 else if( h->mb.i_type == I_8x8 )
1943 for( int i = 0; i < 16; i += 4 )
1944 h->stat.frame.i_mb_pred_mode[1][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
1945 else //if( h->mb.i_type == I_4x4 )
1946 for( int i = 0; i < 16; i++ )
1947 h->stat.frame.i_mb_pred_mode[2][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
1951 x264_ratecontrol_mb( h, mb_size );
1955 i_mb_x += i_mb_y & 1;
1956 i_mb_y ^= i_mb_x < h->sps->i_mb_width;
1960 if( i_mb_x == h->sps->i_mb_width )
1967 if( h->param.b_cabac )
1969 x264_cabac_encode_flush( h, &h->cabac );
1970 h->out.bs.p = h->cabac.p;
1975 bs_write_ue( &h->out.bs, i_skip ); /* last skip run */
1976 /* rbsp_slice_trailing_bits */
1977 bs_rbsp_trailing( &h->out.bs );
1978 bs_flush( &h->out.bs );
1980 if( x264_nal_end( h ) )
1983 if( h->sh.i_last_mb == (h->i_threadslice_end * h->sps->i_mb_width - 1) )
1985 h->stat.frame.i_misc_bits = bs_pos( &h->out.bs )
1986 + (h->out.i_nal*NALU_OVERHEAD * 8)
1987 - h->stat.frame.i_tex_bits
1988 - h->stat.frame.i_mv_bits;
1989 x264_fdec_filter_row( h, h->i_threadslice_end, 1 );
1995 static void x264_thread_sync_context( x264_t *dst, x264_t *src )
2000 // reference counting
2001 for( x264_frame_t **f = src->frames.reference; *f; f++ )
2002 (*f)->i_reference_count++;
2003 for( x264_frame_t **f = dst->frames.reference; *f; f++ )
2004 x264_frame_push_unused( src, *f );
2005 src->fdec->i_reference_count++;
2006 x264_frame_push_unused( src, dst->fdec );
2008 // copy everything except the per-thread pointers and the constants.
2009 memcpy( &dst->i_frame, &src->i_frame, offsetof(x264_t, mb.type) - offsetof(x264_t, i_frame) );
2010 dst->param = src->param;
2011 dst->stat = src->stat;
2014 static void x264_thread_sync_stat( x264_t *dst, x264_t *src )
2018 memcpy( &dst->stat.i_frame_count, &src->stat.i_frame_count, sizeof(dst->stat) - sizeof(dst->stat.frame) );
2021 static void *x264_slices_write( x264_t *h )
2023 int i_slice_num = 0;
2024 int last_thread_mb = h->sh.i_last_mb;
2025 if( h->param.i_sync_lookahead )
2026 x264_lower_thread_priority( 10 );
2029 /* Misalign mask has to be set separately for each thread. */
2030 if( h->param.cpu&X264_CPU_SSE_MISALIGN )
2031 x264_cpu_mask_misalign_sse();
2034 #ifdef HAVE_VISUALIZE
2035 if( h->param.b_visualize )
2036 if( x264_visualize_init( h ) )
2041 memset( &h->stat.frame, 0, sizeof(h->stat.frame) );
2042 h->mb.b_reencode_mb = 0;
2043 while( h->sh.i_first_mb <= last_thread_mb )
2045 h->sh.i_last_mb = last_thread_mb;
2046 if( h->param.i_slice_max_mbs )
2047 h->sh.i_last_mb = h->sh.i_first_mb + h->param.i_slice_max_mbs - 1;
2048 else if( h->param.i_slice_count && !h->param.b_sliced_threads )
2050 int height = h->sps->i_mb_height >> h->param.b_interlaced;
2051 int width = h->sps->i_mb_width << h->param.b_interlaced;
2053 h->sh.i_last_mb = (height * i_slice_num + h->param.i_slice_count/2) / h->param.i_slice_count * width - 1;
2055 h->sh.i_last_mb = X264_MIN( h->sh.i_last_mb, last_thread_mb );
2056 if( x264_stack_align( x264_slice_write, h ) )
2058 h->sh.i_first_mb = h->sh.i_last_mb + 1;
2061 #ifdef HAVE_VISUALIZE
2062 if( h->param.b_visualize )
2064 x264_visualize_show( h );
2065 x264_visualize_close( h );
2072 static int x264_threaded_slices_write( x264_t *h )
2075 /* set first/last mb and sync contexts */
2076 for( int i = 0; i < h->param.i_threads; i++ )
2078 x264_t *t = h->thread[i];
2081 t->param = h->param;
2082 memcpy( &t->i_frame, &h->i_frame, offsetof(x264_t, rc) - offsetof(x264_t, i_frame) );
2084 int height = h->sps->i_mb_height >> h->param.b_interlaced;
2085 t->i_threadslice_start = ((height * i + h->param.i_slice_count/2) / h->param.i_threads) << h->param.b_interlaced;
2086 t->i_threadslice_end = ((height * (i+1) + h->param.i_slice_count/2) / h->param.i_threads) << h->param.b_interlaced;
2087 t->sh.i_first_mb = t->i_threadslice_start * h->sps->i_mb_width;
2088 t->sh.i_last_mb = t->i_threadslice_end * h->sps->i_mb_width - 1;
2091 x264_stack_align( x264_analyse_weight_frame, h, h->sps->i_mb_height*16 + 16 );
2093 x264_threads_distribute_ratecontrol( h );
2096 for( int i = 0; i < h->param.i_threads; i++ )
2098 if( x264_pthread_create( &h->thread[i]->thread_handle, NULL, (void*)x264_slices_write, (void*)h->thread[i] ) )
2100 h->thread[i]->b_thread_active = 1;
2102 for( int i = 0; i < h->param.i_threads; i++ )
2104 x264_pthread_join( h->thread[i]->thread_handle, &ret );
2105 h->thread[i]->b_thread_active = 0;
2107 return (intptr_t)ret;
2110 /* Go back and fix up the hpel on the borders between slices. */
2111 for( int i = 1; i < h->param.i_threads; i++ )
2112 x264_fdec_filter_row( h->thread[i], h->thread[i]->i_threadslice_start + 1, 0 );
2114 x264_threads_merge_ratecontrol( h );
2116 for( int i = 1; i < h->param.i_threads; i++ )
2118 x264_t *t = h->thread[i];
2119 for( int j = 0; j < t->out.i_nal; j++ )
2121 h->out.nal[h->out.i_nal] = t->out.nal[j];
2123 x264_nal_check_buffer( h );
2125 /* All entries in stat.frame are ints except for ssd/ssim. */
2126 for( int j = 0; j < (offsetof(x264_t,stat.frame.i_ssd) - offsetof(x264_t,stat.frame.i_mv_bits)) / sizeof(int); j++ )
2127 ((int*)&h->stat.frame)[j] += ((int*)&t->stat.frame)[j];
2128 for( int j = 0; j < 3; j++ )
2129 h->stat.frame.i_ssd[j] += t->stat.frame.i_ssd[j];
2130 h->stat.frame.f_ssim += t->stat.frame.f_ssim;
2136 /****************************************************************************
2137 * x264_encoder_encode:
2138 * XXX: i_poc : is the poc of the current given picture
2139 * i_frame : is the number of the frame being coded
2140 * ex: type frame poc
2148 ****************************************************************************/
2149 int x264_encoder_encode( x264_t *h,
2150 x264_nal_t **pp_nal, int *pi_nal,
2151 x264_picture_t *pic_in,
2152 x264_picture_t *pic_out )
2154 x264_t *thread_current, *thread_prev, *thread_oldest;
2155 int i_nal_type, i_nal_ref_idc, i_global_qp;
2156 int overhead = NALU_OVERHEAD;
2158 if( h->i_thread_frames > 1 )
2160 thread_prev = h->thread[ h->i_thread_phase ];
2161 h->i_thread_phase = (h->i_thread_phase + 1) % h->i_thread_frames;
2162 thread_current = h->thread[ h->i_thread_phase ];
2163 thread_oldest = h->thread[ (h->i_thread_phase + 1) % h->i_thread_frames ];
2164 x264_thread_sync_context( thread_current, thread_prev );
2165 x264_thread_sync_ratecontrol( thread_current, thread_prev, thread_oldest );
2174 // ok to call this before encoding any frames, since the initial values of fdec have b_kept_as_ref=0
2175 if( x264_reference_update( h ) )
2177 h->fdec->i_lines_completed = -1;
2183 /* ------------------- Setup new frame from picture -------------------- */
2184 if( pic_in != NULL )
2186 /* 1: Copy the picture to a frame and move it to a buffer */
2187 x264_frame_t *fenc = x264_frame_pop_unused( h, 0 );
2191 if( x264_frame_copy_picture( h, fenc, pic_in ) < 0 )
2194 if( h->param.i_width != 16 * h->sps->i_mb_width ||
2195 h->param.i_height != 16 * h->sps->i_mb_height )
2196 x264_frame_expand_border_mod16( h, fenc );
2198 fenc->i_frame = h->frames.i_input++;
2200 if( h->frames.i_bframe_delay && fenc->i_frame == h->frames.i_bframe_delay )
2201 h->frames.i_bframe_delay_time = fenc->i_pts;
2203 if( h->param.b_vfr_input && fenc->i_pts <= h->frames.i_largest_pts )
2204 x264_log( h, X264_LOG_WARNING, "non-strictly-monotonic PTS\n" );
2206 h->frames.i_second_largest_pts = h->frames.i_largest_pts;
2207 h->frames.i_largest_pts = fenc->i_pts;
2209 if( (fenc->i_pic_struct < PIC_STRUCT_AUTO) || (fenc->i_pic_struct > PIC_STRUCT_TRIPLE) )
2210 fenc->i_pic_struct = PIC_STRUCT_AUTO;
2212 if( fenc->i_pic_struct == PIC_STRUCT_AUTO )
2214 int b_interlaced = fenc->param ? fenc->param->b_interlaced : h->param.b_interlaced;
2217 int b_tff = fenc->param ? fenc->param->b_tff : h->param.b_tff;
2218 fenc->i_pic_struct = b_tff ? PIC_STRUCT_TOP_BOTTOM : PIC_STRUCT_BOTTOM_TOP;
2221 fenc->i_pic_struct = PIC_STRUCT_PROGRESSIVE;
2224 if( h->frames.b_have_lowres )
2226 if( h->param.analyse.i_weighted_pred == X264_WEIGHTP_FAKE || h->param.analyse.i_weighted_pred == X264_WEIGHTP_SMART )
2227 x264_weight_plane_analyse( h, fenc );
2228 x264_frame_init_lowres( h, fenc );
2231 if( h->param.rc.b_mb_tree && h->param.rc.b_stat_read )
2233 if( x264_macroblock_tree_read( h, fenc ) )
2236 else if( h->param.rc.i_aq_mode )
2237 x264_adaptive_quant_frame( h, fenc );
2239 /* 2: Place the frame into the queue for its slice type decision */
2240 x264_lookahead_put_frame( h, fenc );
2242 if( h->frames.i_input <= h->frames.i_delay + 1 - h->i_thread_frames )
2244 /* Nothing yet to encode, waiting for filling of buffers */
2245 pic_out->i_type = X264_TYPE_AUTO;
2251 /* signal kills for lookahead thread */
2252 x264_pthread_mutex_lock( &h->lookahead->ifbuf.mutex );
2253 h->lookahead->b_exit_thread = 1;
2254 x264_pthread_cond_broadcast( &h->lookahead->ifbuf.cv_fill );
2255 x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
2259 /* 3: The picture is analyzed in the lookahead */
2260 if( !h->frames.current[0] )
2261 x264_lookahead_get_frames( h );
2263 if( !h->frames.current[0] && x264_lookahead_is_empty( h ) )
2264 return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
2266 /* ------------------- Get frame to be encoded ------------------------- */
2267 /* 4: get picture to encode */
2268 h->fenc = x264_frame_shift( h->frames.current );
2269 if( h->i_frame == 0 )
2270 h->first_pts = h->fenc->i_reordered_pts;
2271 if( h->fenc->param )
2273 x264_encoder_reconfig( h, h->fenc->param );
2274 if( h->fenc->param->param_free )
2275 h->fenc->param->param_free( h->fenc->param );
2278 if( h->fenc->b_keyframe )
2280 h->frames.i_last_keyframe = h->fenc->i_frame;
2281 if( h->fenc->i_type == X264_TYPE_IDR )
2284 h->sh.i_mmco_command_count =
2285 h->sh.i_mmco_remove_from_end = 0;
2286 h->b_ref_reorder[0] =
2287 h->b_ref_reorder[1] = 0;
2289 /* ------------------- Setup frame context ----------------------------- */
2290 /* 5: Init data dependent of frame type */
2291 if( h->fenc->i_type == X264_TYPE_IDR )
2293 /* reset ref pictures */
2294 i_nal_type = NAL_SLICE_IDR;
2295 i_nal_ref_idc = NAL_PRIORITY_HIGHEST;
2296 h->sh.i_type = SLICE_TYPE_I;
2297 x264_reference_reset( h );
2299 else if( h->fenc->i_type == X264_TYPE_I )
2301 i_nal_type = NAL_SLICE;
2302 i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
2303 h->sh.i_type = SLICE_TYPE_I;
2304 x264_reference_hierarchy_reset( h );
2306 else if( h->fenc->i_type == X264_TYPE_P )
2308 i_nal_type = NAL_SLICE;
2309 i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
2310 h->sh.i_type = SLICE_TYPE_P;
2311 x264_reference_hierarchy_reset( h );
2313 else if( h->fenc->i_type == X264_TYPE_BREF )
2315 i_nal_type = NAL_SLICE;
2316 i_nal_ref_idc = h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT ? NAL_PRIORITY_LOW : NAL_PRIORITY_HIGH;
2317 h->sh.i_type = SLICE_TYPE_B;
2318 x264_reference_hierarchy_reset( h );
2322 i_nal_type = NAL_SLICE;
2323 i_nal_ref_idc = NAL_PRIORITY_DISPOSABLE;
2324 h->sh.i_type = SLICE_TYPE_B;
2328 h->fenc->i_poc = 2 * (h->fenc->i_frame - h->frames.i_last_keyframe);
2329 h->fdec->i_type = h->fenc->i_type;
2330 h->fdec->i_frame = h->fenc->i_frame;
2331 h->fenc->b_kept_as_ref =
2332 h->fdec->b_kept_as_ref = i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE && h->param.i_keyint_max > 1;
2336 /* ------------------- Init ----------------------------- */
2337 /* build ref list 0/1 */
2338 x264_reference_build_list( h, h->fdec->i_poc );
2340 /* ---------------------- Write the bitstream -------------------------- */
2341 /* Init bitstream context */
2342 if( h->param.b_sliced_threads )
2344 for( int i = 0; i < h->param.i_threads; i++ )
2346 bs_init( &h->thread[i]->out.bs, h->thread[i]->out.p_bitstream, h->thread[i]->out.i_bitstream );
2347 h->thread[i]->out.i_nal = 0;
2352 bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
2356 if( h->param.b_aud )
2360 if( h->sh.i_type == SLICE_TYPE_I )
2362 else if( h->sh.i_type == SLICE_TYPE_P )
2364 else if( h->sh.i_type == SLICE_TYPE_B )
2369 x264_nal_start( h, NAL_AUD, NAL_PRIORITY_DISPOSABLE );
2370 bs_write( &h->out.bs, 3, pic_type );
2371 bs_rbsp_trailing( &h->out.bs );
2372 if( x264_nal_end( h ) )
2374 overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2377 h->i_nal_type = i_nal_type;
2378 h->i_nal_ref_idc = i_nal_ref_idc;
2380 if( h->param.b_intra_refresh && h->fenc->i_type == X264_TYPE_P )
2382 int pocdiff = (h->fdec->i_poc - h->fref0[0]->i_poc)/2;
2383 float increment = X264_MAX( ((float)h->sps->i_mb_width-1) / h->param.i_keyint_max, 1 );
2384 int max_position = (int)(increment * h->param.i_keyint_max);
2385 if( IS_X264_TYPE_I( h->fref0[0]->i_type ) )
2386 h->fdec->f_pir_position = 0;
2389 h->fdec->f_pir_position = h->fref0[0]->f_pir_position;
2390 if( h->fdec->f_pir_position+0.5 >= max_position )
2392 h->fdec->f_pir_position = 0;
2393 h->fenc->b_keyframe = 1;
2396 h->fdec->i_pir_start_col = h->fdec->f_pir_position+0.5;
2397 h->fdec->f_pir_position += increment * pocdiff;
2398 h->fdec->i_pir_end_col = h->fdec->f_pir_position+0.5;
2401 if( h->fenc->b_keyframe )
2403 /* Write SPS and PPS */
2404 if( h->param.b_repeat_headers )
2406 /* generate sequence parameters */
2407 x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
2408 x264_sps_write( &h->out.bs, h->sps );
2409 if( x264_nal_end( h ) )
2411 overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
2413 /* generate picture parameters */
2414 x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
2415 x264_pps_write( &h->out.bs, h->pps );
2416 if( x264_nal_end( h ) )
2418 overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
2421 /* buffering period sei is written in x264_encoder_frame_end */
2423 if( h->param.b_repeat_headers && h->fenc->i_frame == 0 )
2425 /* identify ourself */
2426 x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2427 if( x264_sei_version_write( h, &h->out.bs ) )
2429 if( x264_nal_end( h ) )
2431 overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2434 if( h->fenc->i_type != X264_TYPE_IDR )
2436 int time_to_recovery = X264_MIN( h->sps->i_mb_width - 1, h->param.i_keyint_max ) + h->param.i_bframe;
2437 x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2438 x264_sei_recovery_point_write( h, &h->out.bs, time_to_recovery );
2440 overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2444 /* generate sei pic timing */
2445 if( h->sps->vui.b_pic_struct_present || h->sps->vui.b_nal_hrd_parameters_present )
2447 x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2448 x264_sei_pic_timing_write( h, &h->out.bs );
2449 if( x264_nal_end( h ) )
2451 overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2454 /* Init the rate control */
2455 /* FIXME: Include slice header bit cost. */
2456 x264_ratecontrol_start( h, h->fenc->i_qpplus1, overhead*8 );
2457 i_global_qp = x264_ratecontrol_qp( h );
2459 pic_out->i_qpplus1 =
2460 h->fdec->i_qpplus1 = i_global_qp + 1;
2462 if( h->param.rc.b_stat_read && h->sh.i_type != SLICE_TYPE_I )
2464 x264_reference_build_list_optimal( h );
2465 x264_reference_check_reorder( h );
2469 h->fdec->i_poc_l0ref0 = h->fref0[0]->i_poc;
2471 if( h->sh.i_type == SLICE_TYPE_B )
2472 x264_macroblock_bipred_init( h );
2474 /*------------------------- Weights -------------------------------------*/
2475 x264_weighted_pred_init( h );
2477 /* ------------------------ Create slice header ----------------------- */
2478 x264_slice_init( h, i_nal_type, i_global_qp );
2480 if( i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE )
2484 h->i_threadslice_start = 0;
2485 h->i_threadslice_end = h->sps->i_mb_height;
2486 if( h->i_thread_frames > 1 )
2488 if( x264_pthread_create( &h->thread_handle, NULL, (void*)x264_slices_write, h ) )
2490 h->b_thread_active = 1;
2492 else if( h->param.b_sliced_threads )
2494 if( x264_threaded_slices_write( h ) )
2498 if( (intptr_t)x264_slices_write( h ) )
2501 return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
2504 static int x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
2505 x264_nal_t **pp_nal, int *pi_nal,
2506 x264_picture_t *pic_out )
2508 char psz_message[80];
2510 if( h->b_thread_active )
2513 x264_pthread_join( h->thread_handle, &ret );
2514 h->b_thread_active = 0;
2516 return (intptr_t)ret;
2520 pic_out->i_type = X264_TYPE_AUTO;
2525 /* generate sei buffering period and insert it into place */
2526 if( h->fenc->b_keyframe && h->sps->vui.b_nal_hrd_parameters_present )
2528 h->initial_cpb_removal_delay = x264_hrd_fullness( h );
2530 x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2531 x264_sei_buffering_period_write( h, &h->out.bs );
2532 if( x264_nal_end( h ) )
2534 /* buffering period sei must follow AUD, SPS and PPS and precede all other SEIs */
2536 while( h->out.nal[idx].i_type == NAL_AUD ||
2537 h->out.nal[idx].i_type == NAL_SPS ||
2538 h->out.nal[idx].i_type == NAL_PPS )
2540 x264_nal_t nal_tmp = h->out.nal[h->out.i_nal-1];
2541 memmove( &h->out.nal[idx+1], &h->out.nal[idx], (h->out.i_nal-idx-1)*sizeof(x264_nal_t) );
2542 h->out.nal[idx] = nal_tmp;
2545 int frame_size = x264_encoder_encapsulate_nals( h, 0 );
2547 /* Set output picture properties */
2548 pic_out->i_type = h->fenc->i_type;
2550 pic_out->b_keyframe = h->fenc->b_keyframe;
2552 pic_out->i_pts = h->fenc->i_pts *= h->i_dts_compress_multiplier;
2553 if( h->frames.i_bframe_delay )
2555 int64_t *prev_reordered_pts = thread_current->frames.i_prev_reordered_pts;
2556 if( h->i_frame <= h->frames.i_bframe_delay )
2558 if( h->i_dts_compress_multiplier == 1 )
2559 pic_out->i_dts = h->fenc->i_reordered_pts - h->frames.i_bframe_delay_time;
2562 /* DTS compression */
2563 if( h->i_frame == 1 )
2564 thread_current->frames.i_init_delta = h->fenc->i_reordered_pts * h->i_dts_compress_multiplier;
2565 pic_out->i_dts = h->i_frame * thread_current->frames.i_init_delta / h->i_dts_compress_multiplier;
2569 pic_out->i_dts = prev_reordered_pts[ (h->i_frame - h->frames.i_bframe_delay) % h->frames.i_bframe_delay ];
2570 prev_reordered_pts[ h->i_frame % h->frames.i_bframe_delay ] = h->fenc->i_reordered_pts * h->i_dts_compress_multiplier;
2573 pic_out->i_dts = h->fenc->i_reordered_pts;
2574 if( pic_out->i_pts < pic_out->i_dts )
2575 x264_log( h, X264_LOG_WARNING, "invalid DTS: PTS is less than DTS\n" );
2577 pic_out->img.i_plane = h->fdec->i_plane;
2578 for( int i = 0; i < 3; i++ )
2580 pic_out->img.i_stride[i] = h->fdec->i_stride[i];
2581 pic_out->img.plane[i] = h->fdec->plane[i];
2584 x264_frame_push_unused( thread_current, h->fenc );
2586 /* ---------------------- Update encoder state ------------------------- */
2590 if( x264_ratecontrol_end( h, frame_size * 8, &filler ) < 0 )
2593 pic_out->hrd_timing = h->fenc->hrd_timing;
2598 overhead = (FILLER_OVERHEAD - h->param.b_annexb);
2599 if( h->param.i_slice_max_size && filler > h->param.i_slice_max_size )
2601 int next_size = filler - h->param.i_slice_max_size;
2602 int overflow = X264_MAX( overhead - next_size, 0 );
2603 f = h->param.i_slice_max_size - overhead - overflow;
2606 f = X264_MAX( 0, filler - overhead );
2608 x264_nal_start( h, NAL_FILLER, NAL_PRIORITY_DISPOSABLE );
2609 x264_filler_write( h, &h->out.bs, f );
2610 if( x264_nal_end( h ) )
2612 int total_size = x264_encoder_encapsulate_nals( h, h->out.i_nal-1 );
2613 frame_size += total_size;
2614 filler -= total_size;
2617 /* End bitstream, set output */
2618 *pi_nal = h->out.i_nal;
2619 *pp_nal = h->out.nal;
2623 x264_noise_reduction_update( h );
2625 /* ---------------------- Compute/Print statistics --------------------- */
2626 x264_thread_sync_stat( h, h->thread[0] );
2629 h->stat.i_frame_count[h->sh.i_type]++;
2630 h->stat.i_frame_size[h->sh.i_type] += frame_size;
2631 h->stat.f_frame_qp[h->sh.i_type] += h->fdec->f_qp_avg_aq;
2633 for( int i = 0; i < X264_MBTYPE_MAX; i++ )
2634 h->stat.i_mb_count[h->sh.i_type][i] += h->stat.frame.i_mb_count[i];
2635 for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
2636 h->stat.i_mb_partition[h->sh.i_type][i] += h->stat.frame.i_mb_partition[i];
2637 for( int i = 0; i < 2; i++ )
2638 h->stat.i_mb_count_8x8dct[i] += h->stat.frame.i_mb_count_8x8dct[i];
2639 for( int i = 0; i < 6; i++ )
2640 h->stat.i_mb_cbp[i] += h->stat.frame.i_mb_cbp[i];
2641 for( int i = 0; i < 3; i++ )
2642 for( int j = 0; j < 13; j++ )
2643 h->stat.i_mb_pred_mode[i][j] += h->stat.frame.i_mb_pred_mode[i][j];
2644 if( h->sh.i_type != SLICE_TYPE_I )
2645 for( int i_list = 0; i_list < 2; i_list++ )
2646 for( int i = 0; i < 32; i++ )
2647 h->stat.i_mb_count_ref[h->sh.i_type][i_list][i] += h->stat.frame.i_mb_count_ref[i_list][i];
2648 if( h->sh.i_type == SLICE_TYPE_P )
2650 h->stat.i_consecutive_bframes[h->fdec->i_frame - h->fref0[0]->i_frame - 1]++;
2651 if( h->param.analyse.i_weighted_pred == X264_WEIGHTP_SMART )
2652 for( int i = 0; i < 3; i++ )
2653 for( int j = 0; j < h->i_ref0; j++ )
2654 if( h->sh.weight[0][i].i_denom != 0 )
2656 h->stat.i_wpred[i]++;
2660 if( h->sh.i_type == SLICE_TYPE_B )
2662 h->stat.i_direct_frames[ h->sh.b_direct_spatial_mv_pred ] ++;
2663 if( h->mb.b_direct_auto_write )
2665 //FIXME somewhat arbitrary time constants
2666 if( h->stat.i_direct_score[0] + h->stat.i_direct_score[1] > h->mb.i_mb_count )
2667 for( int i = 0; i < 2; i++ )
2668 h->stat.i_direct_score[i] = h->stat.i_direct_score[i] * 9/10;
2669 for( int i = 0; i < 2; i++ )
2670 h->stat.i_direct_score[i] += h->stat.frame.i_direct_score[i];
2674 psz_message[0] = '\0';
2675 if( h->param.analyse.b_psnr )
2678 h->stat.frame.i_ssd[0],
2679 h->stat.frame.i_ssd[1],
2680 h->stat.frame.i_ssd[2],
2683 h->stat.i_ssd_global[h->sh.i_type] += ssd[0] + ssd[1] + ssd[2];
2684 h->stat.f_psnr_average[h->sh.i_type] += x264_psnr( ssd[0] + ssd[1] + ssd[2], 3 * h->param.i_width * h->param.i_height / 2 );
2685 h->stat.f_psnr_mean_y[h->sh.i_type] += x264_psnr( ssd[0], h->param.i_width * h->param.i_height );
2686 h->stat.f_psnr_mean_u[h->sh.i_type] += x264_psnr( ssd[1], h->param.i_width * h->param.i_height / 4 );
2687 h->stat.f_psnr_mean_v[h->sh.i_type] += x264_psnr( ssd[2], h->param.i_width * h->param.i_height / 4 );
2689 snprintf( psz_message, 80, " PSNR Y:%5.2f U:%5.2f V:%5.2f",
2690 x264_psnr( ssd[0], h->param.i_width * h->param.i_height ),
2691 x264_psnr( ssd[1], h->param.i_width * h->param.i_height / 4),
2692 x264_psnr( ssd[2], h->param.i_width * h->param.i_height / 4) );
2695 if( h->param.analyse.b_ssim )
2697 double ssim_y = h->stat.frame.f_ssim
2698 / (((h->param.i_width-6)>>2) * ((h->param.i_height-6)>>2));
2699 h->stat.f_ssim_mean_y[h->sh.i_type] += ssim_y;
2700 snprintf( psz_message + strlen(psz_message), 80 - strlen(psz_message),
2701 " SSIM Y:%.5f", ssim_y );
2703 psz_message[79] = '\0';
2705 x264_log( h, X264_LOG_DEBUG,
2706 "frame=%4d QP=%.2f NAL=%d Slice:%c Poc:%-3d I:%-4d P:%-4d SKIP:%-4d size=%d bytes%s\n",
2708 h->fdec->f_qp_avg_aq,
2710 h->sh.i_type == SLICE_TYPE_I ? 'I' : (h->sh.i_type == SLICE_TYPE_P ? 'P' : 'B' ),
2712 h->stat.frame.i_mb_count_i,
2713 h->stat.frame.i_mb_count_p,
2714 h->stat.frame.i_mb_count_skip,
2718 // keep stats all in one place
2719 x264_thread_sync_stat( h->thread[0], h );
2720 // for the use of the next frame
2721 x264_thread_sync_stat( thread_current, h );
2723 #ifdef DEBUG_MB_TYPE
2725 static const char mb_chars[] = { 'i', 'i', 'I', 'C', 'P', '8', 'S',
2726 'D', '<', 'X', 'B', 'X', '>', 'B', 'B', 'B', 'B', '8', 'S' };
2727 for( int mb_xy = 0; mb_xy < h->sps->i_mb_width * h->sps->i_mb_height; mb_xy++ )
2729 if( h->mb.type[mb_xy] < X264_MBTYPE_MAX && h->mb.type[mb_xy] >= 0 )
2730 fprintf( stderr, "%c ", mb_chars[ h->mb.type[mb_xy] ] );
2732 fprintf( stderr, "? " );
2734 if( (mb_xy+1) % h->sps->i_mb_width == 0 )
2735 fprintf( stderr, "\n" );
2740 /* Remove duplicates, must be done near the end as breaks h->fref0 array
2741 * by freeing some of its pointers. */
2742 for( int i = 0; i < h->i_ref0; i++ )
2743 if( h->fref0[i] && h->fref0[i]->b_duplicate )
2745 x264_frame_push_blank_unused( h, h->fref0[i] );
2749 if( h->param.psz_dump_yuv )
2750 x264_frame_dump( h );
2755 static void x264_print_intra( int64_t *i_mb_count, double i_count, int b_print_pcm, char *intra )
2757 intra += sprintf( intra, "I16..4%s: %4.1f%% %4.1f%% %4.1f%%",
2758 b_print_pcm ? "..PCM" : "",
2759 i_mb_count[I_16x16]/ i_count,
2760 i_mb_count[I_8x8] / i_count,
2761 i_mb_count[I_4x4] / i_count );
2763 sprintf( intra, " %4.1f%%", i_mb_count[I_PCM] / i_count );
2766 /****************************************************************************
2767 * x264_encoder_close:
2768 ****************************************************************************/
2769 void x264_encoder_close ( x264_t *h )
2771 int64_t i_yuv_size = 3 * h->param.i_width * h->param.i_height / 2;
2772 int64_t i_mb_count_size[2][7] = {{0}};
2774 int b_print_pcm = h->stat.i_mb_count[SLICE_TYPE_I][I_PCM]
2775 || h->stat.i_mb_count[SLICE_TYPE_P][I_PCM]
2776 || h->stat.i_mb_count[SLICE_TYPE_B][I_PCM];
2778 x264_lookahead_delete( h );
2780 if( h->param.i_threads > 1 )
2782 // don't strictly have to wait for the other threads, but it's simpler than canceling them
2783 for( int i = 0; i < h->param.i_threads; i++ )
2784 if( h->thread[i]->b_thread_active )
2785 x264_pthread_join( h->thread[i]->thread_handle, NULL );
2786 if( h->i_thread_frames > 1 )
2788 for( int i = 0; i < h->i_thread_frames; i++ )
2789 if( h->thread[i]->b_thread_active )
2791 assert( h->thread[i]->fenc->i_reference_count == 1 );
2792 x264_frame_delete( h->thread[i]->fenc );
2795 x264_t *thread_prev = h->thread[h->i_thread_phase];
2796 x264_thread_sync_ratecontrol( h, thread_prev, h );
2797 x264_thread_sync_ratecontrol( thread_prev, thread_prev, h );
2798 h->i_frame = thread_prev->i_frame + 1 - h->i_thread_frames;
2803 /* Slices used and PSNR */
2804 for( int i = 0; i < 5; i++ )
2806 static const int slice_order[] = { SLICE_TYPE_I, SLICE_TYPE_SI, SLICE_TYPE_P, SLICE_TYPE_SP, SLICE_TYPE_B };
2807 static const char *slice_name[] = { "P", "B", "I", "SP", "SI" };
2808 int i_slice = slice_order[i];
2810 if( h->stat.i_frame_count[i_slice] > 0 )
2812 const int i_count = h->stat.i_frame_count[i_slice];
2813 if( h->param.analyse.b_psnr )
2815 x264_log( h, X264_LOG_INFO,
2816 "frame %s:%-5d Avg QP:%5.2f size:%6.0f PSNR Mean Y:%5.2f U:%5.2f V:%5.2f Avg:%5.2f Global:%5.2f\n",
2817 slice_name[i_slice],
2819 h->stat.f_frame_qp[i_slice] / i_count,
2820 (double)h->stat.i_frame_size[i_slice] / i_count,
2821 h->stat.f_psnr_mean_y[i_slice] / i_count, h->stat.f_psnr_mean_u[i_slice] / i_count, h->stat.f_psnr_mean_v[i_slice] / i_count,
2822 h->stat.f_psnr_average[i_slice] / i_count,
2823 x264_psnr( h->stat.i_ssd_global[i_slice], i_count * i_yuv_size ) );
2827 x264_log( h, X264_LOG_INFO,
2828 "frame %s:%-5d Avg QP:%5.2f size:%6.0f\n",
2829 slice_name[i_slice],
2831 h->stat.f_frame_qp[i_slice] / i_count,
2832 (double)h->stat.i_frame_size[i_slice] / i_count );
2836 if( h->param.i_bframe && h->stat.i_frame_count[SLICE_TYPE_P] )
2840 // weight by number of frames (including the P-frame) that are in a sequence of N B-frames
2841 for( int i = 0; i <= h->param.i_bframe; i++ )
2842 den += (i+1) * h->stat.i_consecutive_bframes[i];
2843 for( int i = 0; i <= h->param.i_bframe; i++ )
2844 p += sprintf( p, " %4.1f%%", 100. * (i+1) * h->stat.i_consecutive_bframes[i] / den );
2845 x264_log( h, X264_LOG_INFO, "consecutive B-frames:%s\n", buf );
2848 for( int i_type = 0; i_type < 2; i_type++ )
2849 for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
2851 if( i == D_DIRECT_8x8 ) continue; /* direct is counted as its own type */
2852 i_mb_count_size[i_type][x264_mb_partition_pixel_table[i]] += h->stat.i_mb_partition[i_type][i];
2856 if( h->stat.i_frame_count[SLICE_TYPE_I] > 0 )
2858 int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_I];
2859 double i_count = h->stat.i_frame_count[SLICE_TYPE_I] * h->mb.i_mb_count / 100.0;
2860 x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
2861 x264_log( h, X264_LOG_INFO, "mb I %s\n", buf );
2863 if( h->stat.i_frame_count[SLICE_TYPE_P] > 0 )
2865 int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_P];
2866 double i_count = h->stat.i_frame_count[SLICE_TYPE_P] * h->mb.i_mb_count / 100.0;
2867 int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_P];
2868 x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
2869 x264_log( h, X264_LOG_INFO,
2870 "mb P %s P16..4: %4.1f%% %4.1f%% %4.1f%% %4.1f%% %4.1f%% skip:%4.1f%%\n",
2872 i_mb_size[PIXEL_16x16] / (i_count*4),
2873 (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
2874 i_mb_size[PIXEL_8x8] / (i_count*4),
2875 (i_mb_size[PIXEL_8x4] + i_mb_size[PIXEL_4x8]) / (i_count*4),
2876 i_mb_size[PIXEL_4x4] / (i_count*4),
2877 i_mb_count[P_SKIP] / i_count );
2879 if( h->stat.i_frame_count[SLICE_TYPE_B] > 0 )
2881 int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_B];
2882 double i_count = h->stat.i_frame_count[SLICE_TYPE_B] * h->mb.i_mb_count / 100.0;
2883 double i_mb_list_count;
2884 int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_B];
2885 int64_t list_count[3] = {0}; /* 0 == L0, 1 == L1, 2 == BI */
2886 x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
2887 for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
2888 for( int j = 0; j < 2; j++ )
2890 int l0 = x264_mb_type_list_table[i][0][j];
2891 int l1 = x264_mb_type_list_table[i][1][j];
2893 list_count[l1+l0*l1] += h->stat.i_mb_count[SLICE_TYPE_B][i] * 2;
2895 list_count[0] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L0_8x8];
2896 list_count[1] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L1_8x8];
2897 list_count[2] += h->stat.i_mb_partition[SLICE_TYPE_B][D_BI_8x8];
2898 i_mb_count[B_DIRECT] += (h->stat.i_mb_partition[SLICE_TYPE_B][D_DIRECT_8x8]+2)/4;
2899 i_mb_list_count = (list_count[0] + list_count[1] + list_count[2]) / 100.0;
2900 x264_log( h, X264_LOG_INFO,
2901 "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",
2903 i_mb_size[PIXEL_16x16] / (i_count*4),
2904 (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
2905 i_mb_size[PIXEL_8x8] / (i_count*4),
2906 i_mb_count[B_DIRECT] / i_count,
2907 i_mb_count[B_SKIP] / i_count,
2908 list_count[0] / i_mb_list_count,
2909 list_count[1] / i_mb_list_count,
2910 list_count[2] / i_mb_list_count );
2913 x264_ratecontrol_summary( h );
2915 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 )
2917 #define SUM3(p) (p[SLICE_TYPE_I] + p[SLICE_TYPE_P] + p[SLICE_TYPE_B])
2918 #define SUM3b(p,o) (p[SLICE_TYPE_I][o] + p[SLICE_TYPE_P][o] + p[SLICE_TYPE_B][o])
2919 int64_t i_i8x8 = SUM3b( h->stat.i_mb_count, I_8x8 );
2920 int64_t i_intra = i_i8x8 + SUM3b( h->stat.i_mb_count, I_4x4 )
2921 + SUM3b( h->stat.i_mb_count, I_16x16 );
2922 int64_t i_all_intra = i_intra + SUM3b( h->stat.i_mb_count, I_PCM);
2923 const int i_count = h->stat.i_frame_count[SLICE_TYPE_I] +
2924 h->stat.i_frame_count[SLICE_TYPE_P] +
2925 h->stat.i_frame_count[SLICE_TYPE_B];
2926 int64_t i_mb_count = i_count * h->mb.i_mb_count;
2927 float fps = (float) h->param.i_fps_num / h->param.i_fps_den;
2929 /* duration algorithm fails with one frame */
2931 f_bitrate = fps * SUM3(h->stat.i_frame_size) / i_count / 125;
2934 float duration = (float)(2 * h->frames.i_largest_pts - h->frames.i_second_largest_pts) * h->param.i_timebase_num / h->param.i_timebase_den;
2935 f_bitrate = SUM3(h->stat.i_frame_size) / duration / 125;
2938 if( h->pps->b_transform_8x8_mode )
2941 if( h->stat.i_mb_count_8x8dct[0] )
2942 sprintf( buf, " inter:%.1f%%", 100. * h->stat.i_mb_count_8x8dct[1] / h->stat.i_mb_count_8x8dct[0] );
2943 x264_log( h, X264_LOG_INFO, "8x8 transform intra:%.1f%%%s\n", 100. * i_i8x8 / i_intra, buf );
2946 if( (h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO ||
2947 (h->stat.i_direct_frames[0] && h->stat.i_direct_frames[1]))
2948 && h->stat.i_frame_count[SLICE_TYPE_B] )
2950 x264_log( h, X264_LOG_INFO, "direct mvs spatial:%.1f%% temporal:%.1f%%\n",
2951 h->stat.i_direct_frames[1] * 100. / h->stat.i_frame_count[SLICE_TYPE_B],
2952 h->stat.i_direct_frames[0] * 100. / h->stat.i_frame_count[SLICE_TYPE_B] );
2956 if( i_mb_count != i_all_intra )
2957 sprintf( buf, " inter: %.1f%% %.1f%% %.1f%%",
2958 h->stat.i_mb_cbp[1] * 100.0 / ((i_mb_count - i_all_intra)*4),
2959 h->stat.i_mb_cbp[3] * 100.0 / ((i_mb_count - i_all_intra) ),
2960 h->stat.i_mb_cbp[5] * 100.0 / ((i_mb_count - i_all_intra)) );
2961 x264_log( h, X264_LOG_INFO, "coded y,uvDC,uvAC intra: %.1f%% %.1f%% %.1f%%%s\n",
2962 h->stat.i_mb_cbp[0] * 100.0 / (i_all_intra*4),
2963 h->stat.i_mb_cbp[2] * 100.0 / (i_all_intra ),
2964 h->stat.i_mb_cbp[4] * 100.0 / (i_all_intra ), buf );
2966 int64_t fixed_pred_modes[3][9] = {{0}};
2967 int64_t sum_pred_modes[3] = {0};
2968 for( int i = 0; i <= I_PRED_16x16_DC_128; i++ )
2970 fixed_pred_modes[0][x264_mb_pred_mode16x16_fix[i]] += h->stat.i_mb_pred_mode[0][i];
2971 sum_pred_modes[0] += h->stat.i_mb_pred_mode[0][i];
2973 if( sum_pred_modes[0] )
2974 x264_log( h, X264_LOG_INFO, "i16 v,h,dc,p: %2.0f%% %2.0f%% %2.0f%% %2.0f%%\n",
2975 fixed_pred_modes[0][0] * 100.0 / sum_pred_modes[0],
2976 fixed_pred_modes[0][1] * 100.0 / sum_pred_modes[0],
2977 fixed_pred_modes[0][2] * 100.0 / sum_pred_modes[0],
2978 fixed_pred_modes[0][3] * 100.0 / sum_pred_modes[0] );
2979 for( int i = 1; i <= 2; i++ )
2981 for( int j = 0; j <= I_PRED_8x8_DC_128; j++ )
2983 fixed_pred_modes[i][x264_mb_pred_mode4x4_fix(j)] += h->stat.i_mb_pred_mode[i][j];
2984 sum_pred_modes[i] += h->stat.i_mb_pred_mode[i][j];
2986 if( sum_pred_modes[i] )
2987 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,
2988 fixed_pred_modes[i][0] * 100.0 / sum_pred_modes[i],
2989 fixed_pred_modes[i][1] * 100.0 / sum_pred_modes[i],
2990 fixed_pred_modes[i][2] * 100.0 / sum_pred_modes[i],
2991 fixed_pred_modes[i][3] * 100.0 / sum_pred_modes[i],
2992 fixed_pred_modes[i][4] * 100.0 / sum_pred_modes[i],
2993 fixed_pred_modes[i][5] * 100.0 / sum_pred_modes[i],
2994 fixed_pred_modes[i][6] * 100.0 / sum_pred_modes[i],
2995 fixed_pred_modes[i][7] * 100.0 / sum_pred_modes[i],
2996 fixed_pred_modes[i][8] * 100.0 / sum_pred_modes[i] );
2999 if( h->param.analyse.i_weighted_pred == X264_WEIGHTP_SMART && h->stat.i_frame_count[SLICE_TYPE_P] > 0 )
3000 x264_log( h, X264_LOG_INFO, "Weighted P-Frames: Y:%.1f%%\n",
3001 h->stat.i_wpred[0] * 100.0 / h->stat.i_frame_count[SLICE_TYPE_P] );
3003 for( int i_list = 0; i_list < 2; i_list++ )
3004 for( int i_slice = 0; i_slice < 2; i_slice++ )
3009 for( int i = 0; i < 32; i++ )
3010 if( h->stat.i_mb_count_ref[i_slice][i_list][i] )
3012 i_den += h->stat.i_mb_count_ref[i_slice][i_list][i];
3017 for( int i = 0; i <= i_max; i++ )
3018 p += sprintf( p, " %4.1f%%", 100. * h->stat.i_mb_count_ref[i_slice][i_list][i] / i_den );
3019 x264_log( h, X264_LOG_INFO, "ref %c L%d:%s\n", "PB"[i_slice], i_list, buf );
3022 if( h->param.analyse.b_ssim )
3024 x264_log( h, X264_LOG_INFO,
3025 "SSIM Mean Y:%.7f\n",
3026 SUM3( h->stat.f_ssim_mean_y ) / i_count );
3028 if( h->param.analyse.b_psnr )
3030 x264_log( h, X264_LOG_INFO,
3031 "PSNR Mean Y:%6.3f U:%6.3f V:%6.3f Avg:%6.3f Global:%6.3f kb/s:%.2f\n",
3032 SUM3( h->stat.f_psnr_mean_y ) / i_count,
3033 SUM3( h->stat.f_psnr_mean_u ) / i_count,
3034 SUM3( h->stat.f_psnr_mean_v ) / i_count,
3035 SUM3( h->stat.f_psnr_average ) / i_count,
3036 x264_psnr( SUM3( h->stat.i_ssd_global ), i_count * i_yuv_size ),
3040 x264_log( h, X264_LOG_INFO, "kb/s:%.2f\n", f_bitrate );
3044 x264_ratecontrol_delete( h );
3047 if( h->param.rc.psz_stat_out )
3048 free( h->param.rc.psz_stat_out );
3049 if( h->param.rc.psz_stat_in )
3050 free( h->param.rc.psz_stat_in );
3052 x264_cqm_delete( h );
3053 x264_free( h->nal_buffer );
3054 x264_analyse_free_costs( h );
3056 if( h->i_thread_frames > 1)
3057 h = h->thread[h->i_thread_phase];
3060 x264_frame_delete_list( h->frames.unused[0] );
3061 x264_frame_delete_list( h->frames.unused[1] );
3062 x264_frame_delete_list( h->frames.current );
3063 x264_frame_delete_list( h->frames.blank_unused );
3067 for( int i = h->param.i_threads - 1; i >= 0; i-- )
3069 x264_frame_t **frame;
3071 if( !h->param.b_sliced_threads || i == 0 )
3073 for( frame = h->thread[i]->frames.reference; *frame; frame++ )
3075 assert( (*frame)->i_reference_count > 0 );
3076 (*frame)->i_reference_count--;
3077 if( (*frame)->i_reference_count == 0 )
3078 x264_frame_delete( *frame );
3080 frame = &h->thread[i]->fdec;
3081 assert( (*frame)->i_reference_count > 0 );
3082 (*frame)->i_reference_count--;
3083 if( (*frame)->i_reference_count == 0 )
3084 x264_frame_delete( *frame );
3085 x264_macroblock_cache_free( h->thread[i] );
3087 x264_macroblock_thread_free( h->thread[i], 0 );
3088 x264_free( h->thread[i]->out.p_bitstream );
3089 x264_free( h->thread[i]->out.nal);
3090 x264_free( h->thread[i] );
3094 /****************************************************************************
3095 * x264_encoder_delayed_frames:
3096 ****************************************************************************/
3097 int x264_encoder_delayed_frames( x264_t *h )
3099 int delayed_frames = 0;
3100 if( h->i_thread_frames > 1 )
3102 for( int i = 0; i < h->i_thread_frames; i++ )
3103 delayed_frames += h->thread[i]->b_thread_active;
3104 h = h->thread[h->i_thread_phase];
3106 for( int i = 0; h->frames.current[i]; i++ )
3108 x264_pthread_mutex_lock( &h->lookahead->ofbuf.mutex );
3109 x264_pthread_mutex_lock( &h->lookahead->ifbuf.mutex );
3110 x264_pthread_mutex_lock( &h->lookahead->next.mutex );
3111 delayed_frames += h->lookahead->ifbuf.i_size + h->lookahead->next.i_size + h->lookahead->ofbuf.i_size;
3112 x264_pthread_mutex_unlock( &h->lookahead->next.mutex );
3113 x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
3114 x264_pthread_mutex_unlock( &h->lookahead->ofbuf.mutex );
3115 return delayed_frames;