1 /*****************************************************************************
2 * encoder.c: top-level encoder functions
3 *****************************************************************************
4 * Copyright (C) 2003-2010 x264 project
6 * Authors: Laurent Aimar <fenrir@via.ecp.fr>
7 * Loren Merritt <lorenm@u.washington.edu>
8 * Fiona Glaser <fiona@x264.com>
10 * This program is free software; you can redistribute it and/or modify
11 * it under the terms of the GNU General Public License as published by
12 * the Free Software Foundation; either version 2 of the License, or
13 * (at your option) any later version.
15 * This program is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
18 * GNU General Public License for more details.
20 * You should have received a copy of the GNU General Public License
21 * along with this program; if not, write to the Free Software
22 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02111, USA.
24 * This program is also available under a commercial proprietary license.
25 * For more information, contact us at licensing@x264.com.
26 *****************************************************************************/
30 #include "common/common.h"
34 #include "ratecontrol.h"
35 #include "macroblock.h"
39 #include "common/visualize.h"
42 //#define DEBUG_MB_TYPE
44 #define bs_write_ue bs_write_ue_big
46 static int x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
47 x264_nal_t **pp_nal, int *pi_nal,
48 x264_picture_t *pic_out );
50 /****************************************************************************
52 ******************************* x264 libs **********************************
54 ****************************************************************************/
55 static float x264_psnr( int64_t i_sqe, int64_t i_size )
57 double f_mse = (double)i_sqe / (PIXEL_MAX*PIXEL_MAX * (double)i_size);
58 if( f_mse <= 0.0000000001 ) /* Max 100dB */
61 return -10.0 * log10( f_mse );
64 static float x264_ssim( float ssim )
66 return -10.0 * log10( 1 - ssim );
69 static void x264_frame_dump( x264_t *h )
71 FILE *f = fopen( h->param.psz_dump_yuv, "r+b" );
74 /* Write the frame in display order */
75 fseek( f, (uint64_t)h->fdec->i_frame * h->param.i_height * h->param.i_width * 3/2 * sizeof(pixel), SEEK_SET );
76 for( int y = 0; y < h->param.i_height; y++ )
77 fwrite( &h->fdec->plane[0][y*h->fdec->i_stride[0]], sizeof(pixel), h->param.i_width, f );
78 int cw = h->param.i_width>>1;
79 int ch = h->param.i_height>>1;
80 pixel *planeu = x264_malloc( cw*ch*2*sizeof(pixel) );
81 pixel *planev = planeu + cw*ch;
82 h->mc.plane_copy_deinterleave( planeu, cw, planev, cw, h->fdec->plane[1], h->fdec->i_stride[1], cw, ch );
83 fwrite( planeu, 1, cw*ch*2*sizeof(pixel), f );
89 /* Fill "default" values */
90 static void x264_slice_header_init( x264_t *h, x264_slice_header_t *sh,
91 x264_sps_t *sps, x264_pps_t *pps,
92 int i_idr_pic_id, int i_frame, int i_qp )
94 x264_param_t *param = &h->param;
96 /* First we fill all fields */
101 sh->i_last_mb = h->mb.i_mb_count - 1;
102 sh->i_pps_id = pps->i_id;
104 sh->i_frame_num = i_frame;
106 sh->b_mbaff = h->param.b_interlaced;
107 sh->b_field_pic = 0; /* no field support for now */
108 sh->b_bottom_field = 0; /* not yet used */
110 sh->i_idr_pic_id = i_idr_pic_id;
112 /* poc stuff, fixed later */
114 sh->i_delta_poc_bottom = 0;
115 sh->i_delta_poc[0] = 0;
116 sh->i_delta_poc[1] = 0;
118 sh->i_redundant_pic_cnt = 0;
120 h->mb.b_direct_auto_write = h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO
122 && ( h->param.rc.b_stat_write || !h->param.rc.b_stat_read );
124 if( !h->mb.b_direct_auto_read && sh->i_type == SLICE_TYPE_B )
126 if( h->fref1[0]->i_poc_l0ref0 == h->fref0[0]->i_poc )
128 if( h->mb.b_direct_auto_write )
129 sh->b_direct_spatial_mv_pred = ( h->stat.i_direct_score[1] > h->stat.i_direct_score[0] );
131 sh->b_direct_spatial_mv_pred = ( param->analyse.i_direct_mv_pred == X264_DIRECT_PRED_SPATIAL );
135 h->mb.b_direct_auto_write = 0;
136 sh->b_direct_spatial_mv_pred = 1;
139 /* else b_direct_spatial_mv_pred was read from the 2pass statsfile */
141 sh->b_num_ref_idx_override = 0;
142 sh->i_num_ref_idx_l0_active = 1;
143 sh->i_num_ref_idx_l1_active = 1;
145 sh->b_ref_pic_list_reordering_l0 = h->b_ref_reorder[0];
146 sh->b_ref_pic_list_reordering_l1 = h->b_ref_reorder[1];
148 /* If the ref list isn't in the default order, construct reordering header */
149 /* List1 reordering isn't needed yet */
150 if( sh->b_ref_pic_list_reordering_l0 )
152 int pred_frame_num = i_frame;
153 for( int i = 0; i < h->i_ref0; i++ )
155 int diff = h->fref0[i]->i_frame_num - pred_frame_num;
156 sh->ref_pic_list_order[0][i].idc = ( diff > 0 );
157 sh->ref_pic_list_order[0][i].arg = (abs(diff) - 1) & ((1 << sps->i_log2_max_frame_num) - 1);
158 pred_frame_num = h->fref0[i]->i_frame_num;
162 sh->i_cabac_init_idc = param->i_cabac_init_idc;
165 sh->i_qp_delta = i_qp - pps->i_pic_init_qp;
166 sh->b_sp_for_swidth = 0;
169 int deblock_thresh = i_qp + 2 * X264_MIN(param->i_deblocking_filter_alphac0, param->i_deblocking_filter_beta);
170 /* If effective qp <= 15, deblocking would have no effect anyway */
171 if( param->b_deblocking_filter && (h->mb.b_variable_qp || 15 < deblock_thresh ) )
172 sh->i_disable_deblocking_filter_idc = param->b_sliced_threads ? 2 : 0;
174 sh->i_disable_deblocking_filter_idc = 1;
175 sh->i_alpha_c0_offset = param->i_deblocking_filter_alphac0 << 1;
176 sh->i_beta_offset = param->i_deblocking_filter_beta << 1;
179 static void x264_slice_header_write( bs_t *s, x264_slice_header_t *sh, int i_nal_ref_idc )
183 assert( sh->i_first_mb % (2*sh->sps->i_mb_width) == 0 );
184 bs_write_ue( s, sh->i_first_mb >> 1 );
187 bs_write_ue( s, sh->i_first_mb );
189 bs_write_ue( s, sh->i_type + 5 ); /* same type things */
190 bs_write_ue( s, sh->i_pps_id );
191 bs_write( s, sh->sps->i_log2_max_frame_num, sh->i_frame_num & ((1<<sh->sps->i_log2_max_frame_num)-1) );
193 if( !sh->sps->b_frame_mbs_only )
195 bs_write1( s, sh->b_field_pic );
196 if( sh->b_field_pic )
197 bs_write1( s, sh->b_bottom_field );
200 if( sh->i_idr_pic_id >= 0 ) /* NAL IDR */
201 bs_write_ue( s, sh->i_idr_pic_id );
203 if( sh->sps->i_poc_type == 0 )
205 bs_write( s, sh->sps->i_log2_max_poc_lsb, sh->i_poc & ((1<<sh->sps->i_log2_max_poc_lsb)-1) );
206 if( sh->pps->b_pic_order && !sh->b_field_pic )
207 bs_write_se( s, sh->i_delta_poc_bottom );
209 else if( sh->sps->i_poc_type == 1 && !sh->sps->b_delta_pic_order_always_zero )
211 bs_write_se( s, sh->i_delta_poc[0] );
212 if( sh->pps->b_pic_order && !sh->b_field_pic )
213 bs_write_se( s, sh->i_delta_poc[1] );
216 if( sh->pps->b_redundant_pic_cnt )
217 bs_write_ue( s, sh->i_redundant_pic_cnt );
219 if( sh->i_type == SLICE_TYPE_B )
220 bs_write1( s, sh->b_direct_spatial_mv_pred );
222 if( sh->i_type == SLICE_TYPE_P || sh->i_type == SLICE_TYPE_B )
224 bs_write1( s, sh->b_num_ref_idx_override );
225 if( sh->b_num_ref_idx_override )
227 bs_write_ue( s, sh->i_num_ref_idx_l0_active - 1 );
228 if( sh->i_type == SLICE_TYPE_B )
229 bs_write_ue( s, sh->i_num_ref_idx_l1_active - 1 );
233 /* ref pic list reordering */
234 if( sh->i_type != SLICE_TYPE_I )
236 bs_write1( s, sh->b_ref_pic_list_reordering_l0 );
237 if( sh->b_ref_pic_list_reordering_l0 )
239 for( int i = 0; i < sh->i_num_ref_idx_l0_active; i++ )
241 bs_write_ue( s, sh->ref_pic_list_order[0][i].idc );
242 bs_write_ue( s, sh->ref_pic_list_order[0][i].arg );
247 if( sh->i_type == SLICE_TYPE_B )
249 bs_write1( s, sh->b_ref_pic_list_reordering_l1 );
250 if( sh->b_ref_pic_list_reordering_l1 )
252 for( int i = 0; i < sh->i_num_ref_idx_l1_active; i++ )
254 bs_write_ue( s, sh->ref_pic_list_order[1][i].idc );
255 bs_write_ue( s, sh->ref_pic_list_order[1][i].arg );
261 if( sh->pps->b_weighted_pred && sh->i_type == SLICE_TYPE_P )
263 /* pred_weight_table() */
264 bs_write_ue( s, sh->weight[0][0].i_denom );
265 bs_write_ue( s, sh->weight[0][1].i_denom );
266 for( int i = 0; i < sh->i_num_ref_idx_l0_active; i++ )
268 int luma_weight_l0_flag = !!sh->weight[i][0].weightfn;
269 int chroma_weight_l0_flag = !!sh->weight[i][1].weightfn || !!sh->weight[i][2].weightfn;
270 bs_write1( s, luma_weight_l0_flag );
271 if( luma_weight_l0_flag )
273 bs_write_se( s, sh->weight[i][0].i_scale );
274 bs_write_se( s, sh->weight[i][0].i_offset );
276 bs_write1( s, chroma_weight_l0_flag );
277 if( chroma_weight_l0_flag )
279 for( int j = 1; j < 3; j++ )
281 bs_write_se( s, sh->weight[i][j].i_scale );
282 bs_write_se( s, sh->weight[i][j].i_offset );
287 else if( sh->pps->b_weighted_bipred == 1 && sh->i_type == SLICE_TYPE_B )
292 if( i_nal_ref_idc != 0 )
294 if( sh->i_idr_pic_id >= 0 )
296 bs_write1( s, 0 ); /* no output of prior pics flag */
297 bs_write1( s, 0 ); /* long term reference flag */
301 bs_write1( s, sh->i_mmco_command_count > 0 ); /* adaptive_ref_pic_marking_mode_flag */
302 if( sh->i_mmco_command_count > 0 )
304 for( int i = 0; i < sh->i_mmco_command_count; i++ )
306 bs_write_ue( s, 1 ); /* mark short term ref as unused */
307 bs_write_ue( s, sh->mmco[i].i_difference_of_pic_nums - 1 );
309 bs_write_ue( s, 0 ); /* end command list */
314 if( sh->pps->b_cabac && sh->i_type != SLICE_TYPE_I )
315 bs_write_ue( s, sh->i_cabac_init_idc );
317 bs_write_se( s, sh->i_qp_delta ); /* slice qp delta */
319 if( sh->pps->b_deblocking_filter_control )
321 bs_write_ue( s, sh->i_disable_deblocking_filter_idc );
322 if( sh->i_disable_deblocking_filter_idc != 1 )
324 bs_write_se( s, sh->i_alpha_c0_offset >> 1 );
325 bs_write_se( s, sh->i_beta_offset >> 1 );
330 /* If we are within a reasonable distance of the end of the memory allocated for the bitstream, */
331 /* reallocate, adding an arbitrary amount of space (100 kilobytes). */
332 static int x264_bitstream_check_buffer( x264_t *h )
334 uint8_t *bs_bak = h->out.p_bitstream;
335 if( (h->param.b_cabac && (h->cabac.p_end - h->cabac.p < 2500)) ||
336 (h->out.bs.p_end - h->out.bs.p < 2500) )
338 h->out.i_bitstream += 100000;
339 CHECKED_MALLOC( h->out.p_bitstream, h->out.i_bitstream );
340 h->mc.memcpy_aligned( h->out.p_bitstream, bs_bak, (h->out.i_bitstream - 100000) & ~15 );
341 intptr_t delta = h->out.p_bitstream - bs_bak;
343 h->out.bs.p_start += delta;
344 h->out.bs.p += delta;
345 h->out.bs.p_end = h->out.p_bitstream + h->out.i_bitstream;
347 h->cabac.p_start += delta;
349 h->cabac.p_end = h->out.p_bitstream + h->out.i_bitstream;
351 for( int i = 0; i <= h->out.i_nal; i++ )
352 h->out.nal[i].p_payload += delta;
362 static void x264_encoder_thread_init( x264_t *h )
364 if( h->param.i_sync_lookahead )
365 x264_lower_thread_priority( 10 );
368 /* Misalign mask has to be set separately for each thread. */
369 if( h->param.cpu&X264_CPU_SSE_MISALIGN )
370 x264_cpu_mask_misalign_sse();
375 /****************************************************************************
377 ****************************************************************************
378 ****************************** External API*********************************
379 ****************************************************************************
381 ****************************************************************************/
383 static int x264_validate_parameters( x264_t *h )
387 if( !(x264_cpu_detect() & X264_CPU_SSE) )
389 x264_log( h, X264_LOG_ERROR, "your cpu does not support SSE1, but x264 was compiled with asm support\n");
391 if( !(x264_cpu_detect() & X264_CPU_MMXEXT) )
393 x264_log( h, X264_LOG_ERROR, "your cpu does not support MMXEXT, but x264 was compiled with asm support\n");
395 x264_log( h, X264_LOG_ERROR, "to run x264, recompile without asm support (configure --disable-asm)\n");
399 if( h->param.i_width <= 0 || h->param.i_height <= 0 )
401 x264_log( h, X264_LOG_ERROR, "invalid width x height (%dx%d)\n",
402 h->param.i_width, h->param.i_height );
406 if( h->param.i_width % 2 || h->param.i_height % 2 )
408 x264_log( h, X264_LOG_ERROR, "width or height not divisible by 2 (%dx%d)\n",
409 h->param.i_width, h->param.i_height );
412 int i_csp = h->param.i_csp & X264_CSP_MASK;
413 if( i_csp <= X264_CSP_NONE || i_csp >= X264_CSP_MAX )
415 x264_log( h, X264_LOG_ERROR, "invalid CSP (only I420/YV12/NV12 supported)\n" );
419 if( h->param.i_threads == X264_THREADS_AUTO )
420 h->param.i_threads = x264_cpu_num_processors() * (h->param.b_sliced_threads?2:3)/2;
421 h->param.i_threads = x264_clip3( h->param.i_threads, 1, X264_THREAD_MAX );
422 if( h->param.i_threads > 1 )
425 x264_log( h, X264_LOG_WARNING, "not compiled with pthread support!\n");
426 h->param.i_threads = 1;
428 /* Avoid absurdly small thread slices as they can reduce performance
429 * and VBV compliance. Capped at an arbitrary 4 rows per thread. */
430 if( h->param.b_sliced_threads )
432 int max_threads = (h->param.i_height+15)/16 / 4;
433 h->param.i_threads = X264_MIN( h->param.i_threads, max_threads );
437 h->param.b_sliced_threads = 0;
438 h->i_thread_frames = h->param.b_sliced_threads ? 1 : h->param.i_threads;
439 if( h->i_thread_frames > 1 )
440 h->param.nalu_process = NULL;
442 if( h->param.b_interlaced )
444 if( h->param.analyse.i_me_method >= X264_ME_ESA )
446 x264_log( h, X264_LOG_WARNING, "interlace + me=esa is not implemented\n" );
447 h->param.analyse.i_me_method = X264_ME_UMH;
449 if( h->param.analyse.i_weighted_pred > 0 )
451 x264_log( h, X264_LOG_WARNING, "interlace + weightp is not implemented\n" );
452 h->param.analyse.i_weighted_pred = X264_WEIGHTP_NONE;
456 /* Detect default ffmpeg settings and terminate with an error. */
459 score += h->param.analyse.i_me_range == 0;
460 score += h->param.rc.i_qp_step == 3;
461 score += h->param.i_keyint_max == 12;
462 score += h->param.rc.i_qp_min == 2;
463 score += h->param.rc.i_qp_max == 31;
464 score += h->param.rc.f_qcompress == 0.5;
465 score += fabs(h->param.rc.f_ip_factor - 1.25) < 0.01;
466 score += fabs(h->param.rc.f_pb_factor - 1.25) < 0.01;
467 score += h->param.analyse.inter == 0 && h->param.analyse.i_subpel_refine == 8;
470 x264_log( h, X264_LOG_ERROR, "broken ffmpeg default settings detected\n" );
471 x264_log( h, X264_LOG_ERROR, "use an encoding preset (vpre)\n" );
476 if( h->param.rc.i_rc_method < 0 || h->param.rc.i_rc_method > 2 )
478 x264_log( h, X264_LOG_ERROR, "no ratecontrol method specified\n" );
481 h->param.rc.f_rf_constant = x264_clip3f( h->param.rc.f_rf_constant, 0, QP_MAX );
482 h->param.rc.i_qp_constant = x264_clip3( h->param.rc.i_qp_constant, 0, QP_MAX );
483 if( h->param.rc.i_rc_method == X264_RC_CRF )
485 h->param.rc.i_qp_constant = h->param.rc.f_rf_constant;
486 h->param.rc.i_bitrate = 0;
488 if( (h->param.rc.i_rc_method == X264_RC_CQP || h->param.rc.i_rc_method == X264_RC_CRF)
489 && h->param.rc.i_qp_constant == 0 )
491 h->mb.b_lossless = 1;
492 h->param.i_cqm_preset = X264_CQM_FLAT;
493 h->param.psz_cqm_file = NULL;
494 h->param.rc.i_rc_method = X264_RC_CQP;
495 h->param.rc.f_ip_factor = 1;
496 h->param.rc.f_pb_factor = 1;
497 h->param.analyse.b_psnr = 0;
498 h->param.analyse.b_ssim = 0;
499 h->param.analyse.i_chroma_qp_offset = 0;
500 h->param.analyse.i_trellis = 0;
501 h->param.analyse.b_fast_pskip = 0;
502 h->param.analyse.i_noise_reduction = 0;
503 h->param.analyse.b_psy = 0;
504 h->param.i_bframe = 0;
505 /* 8x8dct is not useful at all in CAVLC lossless */
506 if( !h->param.b_cabac )
507 h->param.analyse.b_transform_8x8 = 0;
509 if( h->param.rc.i_rc_method == X264_RC_CQP )
511 float qp_p = h->param.rc.i_qp_constant;
512 float qp_i = qp_p - 6*log2f( h->param.rc.f_ip_factor );
513 float qp_b = qp_p + 6*log2f( h->param.rc.f_pb_factor );
514 h->param.rc.i_qp_min = x264_clip3( (int)(X264_MIN3( qp_p, qp_i, qp_b )), 0, QP_MAX );
515 h->param.rc.i_qp_max = x264_clip3( (int)(X264_MAX3( qp_p, qp_i, qp_b ) + .999), 0, QP_MAX );
516 h->param.rc.i_aq_mode = 0;
517 h->param.rc.b_mb_tree = 0;
519 h->param.rc.i_qp_max = x264_clip3( h->param.rc.i_qp_max, 0, QP_MAX );
520 h->param.rc.i_qp_min = x264_clip3( h->param.rc.i_qp_min, 0, h->param.rc.i_qp_max );
521 if( h->param.rc.i_vbv_buffer_size )
523 if( h->param.rc.i_rc_method == X264_RC_CQP )
525 x264_log( h, X264_LOG_WARNING, "VBV is incompatible with constant QP, ignored.\n" );
526 h->param.rc.i_vbv_max_bitrate = 0;
527 h->param.rc.i_vbv_buffer_size = 0;
529 else if( h->param.rc.i_vbv_max_bitrate == 0 )
531 if( h->param.rc.i_rc_method == X264_RC_ABR )
533 x264_log( h, X264_LOG_WARNING, "VBV maxrate unspecified, assuming CBR\n" );
534 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate;
538 x264_log( h, X264_LOG_WARNING, "VBV bufsize set but maxrate unspecified, ignored\n" );
539 h->param.rc.i_vbv_buffer_size = 0;
542 else if( h->param.rc.i_vbv_max_bitrate < h->param.rc.i_bitrate &&
543 h->param.rc.i_rc_method == X264_RC_ABR )
545 x264_log( h, X264_LOG_WARNING, "max bitrate less than average bitrate, assuming CBR\n" );
546 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate;
549 else if( h->param.rc.i_vbv_max_bitrate )
551 x264_log( h, X264_LOG_WARNING, "VBV maxrate specified, but no bufsize, ignored\n" );
552 h->param.rc.i_vbv_max_bitrate = 0;
555 if( h->param.b_interlaced && h->param.i_slice_max_size )
557 x264_log( h, X264_LOG_WARNING, "interlaced + slice-max-size is not implemented\n" );
558 h->param.i_slice_max_size = 0;
560 if( h->param.b_interlaced && h->param.i_slice_max_mbs )
562 x264_log( h, X264_LOG_WARNING, "interlaced + slice-max-mbs is not implemented\n" );
563 h->param.i_slice_max_mbs = 0;
565 int max_slices = (h->param.i_height+((16<<h->param.b_interlaced)-1))/(16<<h->param.b_interlaced);
566 if( h->param.b_sliced_threads )
567 h->param.i_slice_count = x264_clip3( h->param.i_threads, 0, max_slices );
570 h->param.i_slice_count = x264_clip3( h->param.i_slice_count, 0, max_slices );
571 h->param.i_slice_max_size = X264_MAX( h->param.i_slice_max_size, 0 );
572 h->param.i_slice_max_mbs = X264_MAX( h->param.i_slice_max_mbs, 0 );
573 if( h->param.i_slice_max_mbs || h->param.i_slice_max_size )
574 h->param.i_slice_count = 0;
577 h->param.i_frame_reference = x264_clip3( h->param.i_frame_reference, 1, X264_REF_MAX );
578 h->param.i_dpb_size = x264_clip3( h->param.i_dpb_size, 1, X264_REF_MAX );
579 h->param.i_keyint_max = x264_clip3( h->param.i_keyint_max, 1, X264_KEYINT_MAX_INFINITE );
580 if( h->param.i_scenecut_threshold < 0 )
581 h->param.i_scenecut_threshold = 0;
582 if( !h->param.analyse.i_subpel_refine && h->param.analyse.i_direct_mv_pred > X264_DIRECT_PRED_SPATIAL )
584 x264_log( h, X264_LOG_WARNING, "subme=0 + direct=temporal is not supported\n" );
585 h->param.analyse.i_direct_mv_pred = X264_DIRECT_PRED_SPATIAL;
587 h->param.i_bframe = x264_clip3( h->param.i_bframe, 0, X264_MIN( X264_BFRAME_MAX, h->param.i_keyint_max-1 ) );
588 h->param.i_open_gop = x264_clip3( h->param.i_open_gop, X264_OPEN_GOP_NONE, X264_OPEN_GOP_BLURAY );
589 if( h->param.i_keyint_max == 1 )
590 h->param.b_intra_refresh = 0;
591 h->param.i_bframe_bias = x264_clip3( h->param.i_bframe_bias, -90, 100 );
592 if( h->param.i_bframe <= 1 )
593 h->param.i_bframe_pyramid = X264_B_PYRAMID_NONE;
594 h->param.i_bframe_pyramid = x264_clip3( h->param.i_bframe_pyramid, X264_B_PYRAMID_NONE, X264_B_PYRAMID_NORMAL );
595 if( !h->param.i_bframe )
597 h->param.i_bframe_adaptive = X264_B_ADAPT_NONE;
598 h->param.analyse.i_direct_mv_pred = 0;
599 h->param.analyse.b_weighted_bipred = 0;
600 h->param.i_open_gop = X264_OPEN_GOP_NONE;
602 if( h->param.b_intra_refresh && h->param.i_bframe_pyramid == X264_B_PYRAMID_NORMAL )
604 x264_log( h, X264_LOG_WARNING, "b-pyramid normal + intra-refresh is not supported\n" );
605 h->param.i_bframe_pyramid = X264_B_PYRAMID_STRICT;
607 if( h->param.b_intra_refresh && (h->param.i_frame_reference > 1 || h->param.i_dpb_size > 1) )
609 x264_log( h, X264_LOG_WARNING, "ref > 1 + intra-refresh is not supported\n" );
610 h->param.i_frame_reference = 1;
611 h->param.i_dpb_size = 1;
613 if( h->param.b_intra_refresh && h->param.i_open_gop )
615 x264_log( h, X264_LOG_WARNING, "intra-refresh is not compatible with open-gop\n" );
616 h->param.i_open_gop = X264_OPEN_GOP_NONE;
618 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;
619 if( h->param.i_keyint_min == X264_KEYINT_MIN_AUTO )
620 h->param.i_keyint_min = X264_MIN( h->param.i_keyint_max / 10, fps );
621 h->param.i_keyint_min = x264_clip3( h->param.i_keyint_min, 1, h->param.i_keyint_max/2+1 );
622 h->param.rc.i_lookahead = x264_clip3( h->param.rc.i_lookahead, 0, X264_LOOKAHEAD_MAX );
624 int maxrate = X264_MAX( h->param.rc.i_vbv_max_bitrate, h->param.rc.i_bitrate );
625 float bufsize = maxrate ? (float)h->param.rc.i_vbv_buffer_size / maxrate : 0;
626 h->param.rc.i_lookahead = X264_MIN( h->param.rc.i_lookahead, X264_MAX( h->param.i_keyint_max, bufsize*fps ) );
629 if( !h->param.i_timebase_num || !h->param.i_timebase_den )
631 h->param.i_timebase_num = h->param.i_fps_den;
632 h->param.i_timebase_den = h->param.i_fps_num;
635 h->param.rc.f_qcompress = x264_clip3f( h->param.rc.f_qcompress, 0.0, 1.0 );
636 if( h->param.i_keyint_max == 1 || h->param.rc.f_qcompress == 1 )
637 h->param.rc.b_mb_tree = 0;
638 if( (!h->param.b_intra_refresh && h->param.i_keyint_max != X264_KEYINT_MAX_INFINITE) &&
639 !h->param.rc.i_lookahead && h->param.rc.b_mb_tree )
641 x264_log( h, X264_LOG_WARNING, "lookaheadless mb-tree requires intra refresh or infinite keyint\n" );
642 h->param.rc.b_mb_tree = 0;
644 if( h->param.rc.b_stat_read )
645 h->param.rc.i_lookahead = 0;
647 if( h->param.i_sync_lookahead < 0 )
648 h->param.i_sync_lookahead = h->param.i_bframe + 1;
649 h->param.i_sync_lookahead = X264_MIN( h->param.i_sync_lookahead, X264_LOOKAHEAD_MAX );
650 if( h->param.rc.b_stat_read || h->i_thread_frames == 1 )
651 h->param.i_sync_lookahead = 0;
653 h->param.i_sync_lookahead = 0;
656 h->param.i_deblocking_filter_alphac0 = x264_clip3( h->param.i_deblocking_filter_alphac0, -6, 6 );
657 h->param.i_deblocking_filter_beta = x264_clip3( h->param.i_deblocking_filter_beta, -6, 6 );
658 h->param.analyse.i_luma_deadzone[0] = x264_clip3( h->param.analyse.i_luma_deadzone[0], 0, 32 );
659 h->param.analyse.i_luma_deadzone[1] = x264_clip3( h->param.analyse.i_luma_deadzone[1], 0, 32 );
661 h->param.i_cabac_init_idc = x264_clip3( h->param.i_cabac_init_idc, 0, 2 );
663 if( h->param.i_cqm_preset < X264_CQM_FLAT || h->param.i_cqm_preset > X264_CQM_CUSTOM )
664 h->param.i_cqm_preset = X264_CQM_FLAT;
666 if( h->param.analyse.i_me_method < X264_ME_DIA ||
667 h->param.analyse.i_me_method > X264_ME_TESA )
668 h->param.analyse.i_me_method = X264_ME_HEX;
669 if( h->param.analyse.i_me_range < 4 )
670 h->param.analyse.i_me_range = 4;
671 if( h->param.analyse.i_me_range > 16 && h->param.analyse.i_me_method <= X264_ME_HEX )
672 h->param.analyse.i_me_range = 16;
673 if( h->param.analyse.i_me_method == X264_ME_TESA &&
674 (h->mb.b_lossless || h->param.analyse.i_subpel_refine <= 1) )
675 h->param.analyse.i_me_method = X264_ME_ESA;
676 h->param.analyse.i_subpel_refine = x264_clip3( h->param.analyse.i_subpel_refine, 0, 10 );
677 h->param.analyse.b_mixed_references = h->param.analyse.b_mixed_references && h->param.i_frame_reference > 1;
678 h->param.analyse.inter &= X264_ANALYSE_PSUB16x16|X264_ANALYSE_PSUB8x8|X264_ANALYSE_BSUB16x16|
679 X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
680 h->param.analyse.intra &= X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
681 if( !(h->param.analyse.inter & X264_ANALYSE_PSUB16x16) )
682 h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
683 if( !h->param.analyse.b_transform_8x8 )
685 h->param.analyse.inter &= ~X264_ANALYSE_I8x8;
686 h->param.analyse.intra &= ~X264_ANALYSE_I8x8;
688 h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -12, 12);
689 h->param.analyse.i_trellis = x264_clip3( h->param.analyse.i_trellis, 0, 2 );
690 if( !h->param.analyse.b_psy )
692 h->param.analyse.f_psy_rd = 0;
693 h->param.analyse.f_psy_trellis = 0;
695 if( !h->param.analyse.i_trellis )
696 h->param.analyse.f_psy_trellis = 0;
697 h->param.analyse.f_psy_rd = x264_clip3f( h->param.analyse.f_psy_rd, 0, 10 );
698 h->param.analyse.f_psy_trellis = x264_clip3f( h->param.analyse.f_psy_trellis, 0, 10 );
699 if( h->param.analyse.i_subpel_refine < 6 )
700 h->param.analyse.f_psy_rd = 0;
701 h->mb.i_psy_rd = FIX8( h->param.analyse.f_psy_rd );
702 /* Psy RDO increases overall quantizers to improve the quality of luma--this indirectly hurts chroma quality */
703 /* so we lower the chroma QP offset to compensate */
704 /* This can be triggered repeatedly on multiple calls to parameter_validate, but since encoding
705 * uses the pps chroma qp offset not the param chroma qp offset, this is not a problem. */
707 h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_rd < 0.25 ? 1 : 2;
708 h->mb.i_psy_trellis = FIX8( h->param.analyse.f_psy_trellis / 4 );
709 /* Psy trellis has a similar effect. */
710 if( h->mb.i_psy_trellis )
711 h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_trellis < 0.25 ? 1 : 2;
712 h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -12, 12);
713 h->param.rc.i_aq_mode = x264_clip3( h->param.rc.i_aq_mode, 0, 2 );
714 h->param.rc.f_aq_strength = x264_clip3f( h->param.rc.f_aq_strength, 0, 3 );
715 if( h->param.rc.f_aq_strength == 0 )
716 h->param.rc.i_aq_mode = 0;
717 /* MB-tree requires AQ to be on, even if the strength is zero. */
718 if( !h->param.rc.i_aq_mode && h->param.rc.b_mb_tree )
720 h->param.rc.i_aq_mode = 1;
721 h->param.rc.f_aq_strength = 0;
723 h->param.analyse.i_noise_reduction = x264_clip3( h->param.analyse.i_noise_reduction, 0, 1<<16 );
724 if( h->param.analyse.i_subpel_refine == 10 && (h->param.analyse.i_trellis != 2 || !h->param.rc.i_aq_mode) )
725 h->param.analyse.i_subpel_refine = 9;
728 const x264_level_t *l = x264_levels;
729 if( h->param.i_level_idc < 0 )
731 int maxrate_bak = h->param.rc.i_vbv_max_bitrate;
732 if( h->param.rc.i_rc_method == X264_RC_ABR && h->param.rc.i_vbv_buffer_size <= 0 )
733 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate * 2;
734 h->sps = h->sps_array;
735 x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
736 do h->param.i_level_idc = l->level_idc;
737 while( l[1].level_idc && x264_validate_levels( h, 0 ) && l++ );
738 h->param.rc.i_vbv_max_bitrate = maxrate_bak;
742 while( l->level_idc && l->level_idc != h->param.i_level_idc )
744 if( l->level_idc == 0 )
746 x264_log( h, X264_LOG_ERROR, "invalid level_idc: %d\n", h->param.i_level_idc );
750 if( h->param.analyse.i_mv_range <= 0 )
751 h->param.analyse.i_mv_range = l->mv_range >> h->param.b_interlaced;
753 h->param.analyse.i_mv_range = x264_clip3(h->param.analyse.i_mv_range, 32, 512 >> h->param.b_interlaced);
756 h->param.analyse.i_weighted_pred = x264_clip3( h->param.analyse.i_weighted_pred, 0, X264_WEIGHTP_SMART );
757 if( !h->param.analyse.i_weighted_pred && h->param.rc.b_mb_tree && h->param.analyse.b_psy && !h->param.b_interlaced )
758 h->param.analyse.i_weighted_pred = X264_WEIGHTP_FAKE;
760 if( h->i_thread_frames > 1 )
762 int r = h->param.analyse.i_mv_range_thread;
766 // half of the available space is reserved and divided evenly among the threads,
767 // the rest is allocated to whichever thread is far enough ahead to use it.
768 // reserving more space increases quality for some videos, but costs more time
769 // in thread synchronization.
770 int max_range = (h->param.i_height + X264_THREAD_HEIGHT) / h->i_thread_frames - X264_THREAD_HEIGHT;
773 r = X264_MAX( r, h->param.analyse.i_me_range );
774 r = X264_MIN( r, h->param.analyse.i_mv_range );
775 // round up to use the whole mb row
776 r2 = (r & ~15) + ((-X264_THREAD_HEIGHT) & 15);
779 x264_log( h, X264_LOG_DEBUG, "using mv_range_thread = %d\n", r2 );
780 h->param.analyse.i_mv_range_thread = r2;
783 if( h->param.rc.f_qblur < 0 )
784 h->param.rc.f_qblur = 0;
785 if( h->param.rc.f_complexity_blur < 0 )
786 h->param.rc.f_complexity_blur = 0;
788 h->param.i_sps_id &= 31;
790 if( h->param.i_log_level < X264_LOG_INFO )
792 h->param.analyse.b_psnr = 0;
793 h->param.analyse.b_ssim = 0;
796 if( h->param.b_interlaced )
797 h->param.b_pic_struct = 1;
799 if( h->param.i_nal_hrd && !h->param.rc.i_vbv_buffer_size )
801 x264_log( h, X264_LOG_WARNING, "NAL HRD parameters require VBV parameters\n" );
802 h->param.i_nal_hrd = X264_NAL_HRD_NONE;
805 if( h->param.i_nal_hrd == X264_NAL_HRD_CBR &&
806 (h->param.rc.i_bitrate != h->param.rc.i_vbv_max_bitrate || !h->param.rc.i_vbv_max_bitrate) )
808 x264_log( h, X264_LOG_WARNING, "CBR HRD requires constant bitrate\n" );
809 h->param.i_nal_hrd = X264_NAL_HRD_VBR;
812 /* ensure the booleans are 0 or 1 so they can be used in math */
813 #define BOOLIFY(x) h->param.x = !!h->param.x
815 BOOLIFY( b_constrained_intra );
816 BOOLIFY( b_deblocking_filter );
817 BOOLIFY( b_deterministic );
818 BOOLIFY( b_sliced_threads );
819 BOOLIFY( b_interlaced );
820 BOOLIFY( b_intra_refresh );
821 BOOLIFY( b_visualize );
823 BOOLIFY( b_repeat_headers );
825 BOOLIFY( b_vfr_input );
826 BOOLIFY( b_pic_struct );
827 BOOLIFY( b_fake_interlaced );
828 BOOLIFY( analyse.b_transform_8x8 );
829 BOOLIFY( analyse.b_weighted_bipred );
830 BOOLIFY( analyse.b_chroma_me );
831 BOOLIFY( analyse.b_mixed_references );
832 BOOLIFY( analyse.b_fast_pskip );
833 BOOLIFY( analyse.b_dct_decimate );
834 BOOLIFY( analyse.b_psy );
835 BOOLIFY( analyse.b_psnr );
836 BOOLIFY( analyse.b_ssim );
837 BOOLIFY( rc.b_stat_write );
838 BOOLIFY( rc.b_stat_read );
839 BOOLIFY( rc.b_mb_tree );
845 static void mbcmp_init( x264_t *h )
847 int satd = !h->mb.b_lossless && h->param.analyse.i_subpel_refine > 1;
848 memcpy( h->pixf.mbcmp, satd ? h->pixf.satd : h->pixf.sad_aligned, sizeof(h->pixf.mbcmp) );
849 memcpy( h->pixf.mbcmp_unaligned, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.mbcmp_unaligned) );
850 h->pixf.intra_mbcmp_x3_16x16 = satd ? h->pixf.intra_satd_x3_16x16 : h->pixf.intra_sad_x3_16x16;
851 h->pixf.intra_mbcmp_x3_8x8c = satd ? h->pixf.intra_satd_x3_8x8c : h->pixf.intra_sad_x3_8x8c;
852 h->pixf.intra_mbcmp_x3_8x8 = satd ? h->pixf.intra_sa8d_x3_8x8 : h->pixf.intra_sad_x3_8x8;
853 h->pixf.intra_mbcmp_x3_4x4 = satd ? h->pixf.intra_satd_x3_4x4 : h->pixf.intra_sad_x3_4x4;
854 satd &= h->param.analyse.i_me_method == X264_ME_TESA;
855 memcpy( h->pixf.fpelcmp, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.fpelcmp) );
856 memcpy( h->pixf.fpelcmp_x3, satd ? h->pixf.satd_x3 : h->pixf.sad_x3, sizeof(h->pixf.fpelcmp_x3) );
857 memcpy( h->pixf.fpelcmp_x4, satd ? h->pixf.satd_x4 : h->pixf.sad_x4, sizeof(h->pixf.fpelcmp_x4) );
860 static void x264_set_aspect_ratio( x264_t *h, x264_param_t *param, int initial )
863 if( param->vui.i_sar_width > 0 && param->vui.i_sar_height > 0 )
865 uint32_t i_w = param->vui.i_sar_width;
866 uint32_t i_h = param->vui.i_sar_height;
867 uint32_t old_w = h->param.vui.i_sar_width;
868 uint32_t old_h = h->param.vui.i_sar_height;
870 x264_reduce_fraction( &i_w, &i_h );
872 while( i_w > 65535 || i_h > 65535 )
878 x264_reduce_fraction( &i_w, &i_h );
880 if( i_w != old_w || i_h != old_h || initial )
882 h->param.vui.i_sar_width = 0;
883 h->param.vui.i_sar_height = 0;
884 if( i_w == 0 || i_h == 0 )
885 x264_log( h, X264_LOG_WARNING, "cannot create valid sample aspect ratio\n" );
888 x264_log( h, initial?X264_LOG_INFO:X264_LOG_DEBUG, "using SAR=%d/%d\n", i_w, i_h );
889 h->param.vui.i_sar_width = i_w;
890 h->param.vui.i_sar_height = i_h;
896 /****************************************************************************
898 ****************************************************************************/
899 x264_t *x264_encoder_open( x264_param_t *param )
903 int qp, i_slicetype_length;
905 CHECKED_MALLOCZERO( h, sizeof(x264_t) );
907 /* Create a copy of param */
908 memcpy( &h->param, param, sizeof(x264_param_t) );
910 if( param->param_free )
911 param->param_free( param );
913 if( x264_validate_parameters( h ) < 0 )
916 if( h->param.psz_cqm_file )
917 if( x264_cqm_parse_file( h, h->param.psz_cqm_file ) < 0 )
920 if( h->param.rc.psz_stat_out )
921 h->param.rc.psz_stat_out = strdup( h->param.rc.psz_stat_out );
922 if( h->param.rc.psz_stat_in )
923 h->param.rc.psz_stat_in = strdup( h->param.rc.psz_stat_in );
925 x264_set_aspect_ratio( h, &h->param, 1 );
927 x264_reduce_fraction( &h->param.i_fps_num, &h->param.i_fps_den );
928 x264_reduce_fraction( &h->param.i_timebase_num, &h->param.i_timebase_den );
934 uint64_t new_timebase_den = h->param.i_timebase_den;
935 if( h->param.b_dts_compress )
937 /* h->i_dts_compress_multiplier == h->frames.i_bframe_delay + 1 */
938 h->i_dts_compress_multiplier = h->param.i_bframe ? (h->param.i_bframe_pyramid ? 3 : 2) : 1;
939 if( h->i_dts_compress_multiplier != 1 )
941 new_timebase_den = h->param.i_timebase_den * h->i_dts_compress_multiplier;
942 x264_log( h, X264_LOG_DEBUG, "DTS compression changed timebase: %u/%u -> %u/%"PRIu64"\n",
943 h->param.i_timebase_num, h->param.i_timebase_den,
944 h->param.i_timebase_num, new_timebase_den );
948 h->i_dts_compress_multiplier = 1;
950 if( new_timebase_den * 2 > UINT32_MAX )
952 x264_log( h, X264_LOG_ERROR, "Effective timebase denominator %"PRIu64" exceeds H.264 maximum\n", new_timebase_den );
955 h->param.i_timebase_den = new_timebase_den;
957 h->sps = &h->sps_array[0];
958 x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
960 h->pps = &h->pps_array[0];
961 x264_pps_init( h->pps, h->param.i_sps_id, &h->param, h->sps );
963 x264_validate_levels( h, 1 );
965 h->chroma_qp_table = i_chroma_qp_table + 12 + h->pps->i_chroma_qp_index_offset;
967 if( x264_cqm_init( h ) < 0 )
970 h->mb.i_mb_width = h->sps->i_mb_width;
971 h->mb.i_mb_height = h->sps->i_mb_height;
972 h->mb.i_mb_count = h->mb.i_mb_width * h->mb.i_mb_height;
975 if( h->param.i_bframe_adaptive == X264_B_ADAPT_TRELLIS && !h->param.rc.b_stat_read )
976 h->frames.i_delay = X264_MAX(h->param.i_bframe,3)*4;
978 h->frames.i_delay = h->param.i_bframe;
979 if( h->param.rc.b_mb_tree || h->param.rc.i_vbv_buffer_size )
980 h->frames.i_delay = X264_MAX( h->frames.i_delay, h->param.rc.i_lookahead );
981 i_slicetype_length = h->frames.i_delay;
982 h->frames.i_delay += h->i_thread_frames - 1;
983 h->frames.i_delay += h->param.i_sync_lookahead;
984 h->frames.i_delay += h->param.b_vfr_input && (h->param.rc.i_rc_method == X264_RC_ABR || h->param.rc.b_stat_write
985 || h->param.rc.i_vbv_buffer_size);
986 h->frames.i_bframe_delay = h->param.i_bframe ? (h->param.i_bframe_pyramid ? 2 : 1) : 0;
988 h->frames.i_max_ref0 = h->param.i_frame_reference;
989 h->frames.i_max_ref1 = X264_MIN( h->sps->vui.i_num_reorder_frames, h->param.i_frame_reference );
990 h->frames.i_max_dpb = h->sps->vui.i_max_dec_frame_buffering;
991 h->frames.b_have_lowres = !h->param.rc.b_stat_read
992 && ( h->param.rc.i_rc_method == X264_RC_ABR
993 || h->param.rc.i_rc_method == X264_RC_CRF
994 || h->param.i_bframe_adaptive
995 || h->param.i_scenecut_threshold
996 || h->param.rc.b_mb_tree
997 || h->param.analyse.i_weighted_pred == X264_WEIGHTP_SMART );
998 h->frames.b_have_lowres |= h->param.rc.b_stat_read && h->param.rc.i_vbv_buffer_size > 0;
999 h->frames.b_have_sub8x8_esa = !!(h->param.analyse.inter & X264_ANALYSE_PSUB8x8);
1001 h->frames.i_last_idr =
1002 h->frames.i_last_keyframe = - h->param.i_keyint_max;
1003 h->frames.i_input = 0;
1004 h->frames.i_largest_pts = h->frames.i_second_largest_pts = -1;
1005 h->frames.i_poc_last_open_gop = -1;
1007 CHECKED_MALLOCZERO( h->frames.unused[0], (h->frames.i_delay + 3) * sizeof(x264_frame_t *) );
1008 /* Allocate room for max refs plus a few extra just in case. */
1009 CHECKED_MALLOCZERO( h->frames.unused[1], (h->i_thread_frames + X264_REF_MAX + 4) * sizeof(x264_frame_t *) );
1010 CHECKED_MALLOCZERO( h->frames.current, (h->param.i_sync_lookahead + h->param.i_bframe
1011 + h->i_thread_frames + 3) * sizeof(x264_frame_t *) );
1012 if( h->param.analyse.i_weighted_pred > 0 )
1013 CHECKED_MALLOCZERO( h->frames.blank_unused, h->i_thread_frames * 4 * sizeof(x264_frame_t *) );
1016 h->i_cpb_delay = h->i_coded_fields = h->i_disp_fields = h->i_prev_duration = 0;
1017 h->i_disp_fields_last_frame = -1;
1020 /* init CPU functions */
1021 x264_predict_16x16_init( h->param.cpu, h->predict_16x16 );
1022 x264_predict_8x8c_init( h->param.cpu, h->predict_8x8c );
1023 x264_predict_8x8_init( h->param.cpu, h->predict_8x8, &h->predict_8x8_filter );
1024 x264_predict_4x4_init( h->param.cpu, h->predict_4x4 );
1025 if( !h->param.b_cabac )
1026 x264_init_vlc_tables();
1027 x264_pixel_init( h->param.cpu, &h->pixf );
1028 x264_dct_init( h->param.cpu, &h->dctf );
1029 x264_zigzag_init( h->param.cpu, &h->zigzagf, h->param.b_interlaced );
1030 x264_mc_init( h->param.cpu, &h->mc );
1031 x264_quant_init( h, h->param.cpu, &h->quantf );
1032 x264_deblock_init( h->param.cpu, &h->loopf );
1033 x264_bitstream_init( h->param.cpu, &h->bsf );
1034 x264_dct_init_weights();
1038 p = buf + sprintf( buf, "using cpu capabilities:" );
1039 for( int i = 0; x264_cpu_names[i].flags; i++ )
1041 if( !strcmp(x264_cpu_names[i].name, "SSE2")
1042 && h->param.cpu & (X264_CPU_SSE2_IS_FAST|X264_CPU_SSE2_IS_SLOW) )
1044 if( !strcmp(x264_cpu_names[i].name, "SSE3")
1045 && (h->param.cpu & X264_CPU_SSSE3 || !(h->param.cpu & X264_CPU_CACHELINE_64)) )
1047 if( !strcmp(x264_cpu_names[i].name, "SSE4.1")
1048 && (h->param.cpu & X264_CPU_SSE42) )
1050 if( (h->param.cpu & x264_cpu_names[i].flags) == x264_cpu_names[i].flags
1051 && (!i || x264_cpu_names[i].flags != x264_cpu_names[i-1].flags) )
1052 p += sprintf( p, " %s", x264_cpu_names[i].name );
1055 p += sprintf( p, " none!" );
1056 x264_log( h, X264_LOG_INFO, "%s\n", buf );
1058 for( qp = h->param.rc.i_qp_min; qp <= h->param.rc.i_qp_max; qp++ )
1059 if( x264_analyse_init_costs( h, qp ) )
1061 if( x264_analyse_init_costs( h, X264_LOOKAHEAD_QP ) )
1064 static const uint16_t cost_mv_correct[7] = { 24, 47, 95, 189, 379, 757, 1515 };
1065 /* Checks for known miscompilation issues. */
1066 if( h->cost_mv[x264_lambda_tab[X264_LOOKAHEAD_QP]][2013] != cost_mv_correct[BIT_DEPTH-8] )
1068 x264_log( h, X264_LOG_ERROR, "MV cost test failed: x264 has been miscompiled!\n" );
1072 /* Must be volatile or else GCC will optimize it out. */
1073 volatile int temp = 392;
1074 if( x264_clz( temp ) != 23 )
1076 x264_log( h, X264_LOG_ERROR, "CLZ test failed: x264 has been miscompiled!\n" );
1077 #if ARCH_X86 || ARCH_X86_64
1078 x264_log( h, X264_LOG_ERROR, "Are you attempting to run an SSE4a-targeted build on a CPU that\n" );
1079 x264_log( h, X264_LOG_ERROR, "doesn't support it?\n" );
1085 h->out.i_bitstream = X264_MAX( 1000000, h->param.i_width * h->param.i_height * 4
1086 * ( h->param.rc.i_rc_method == X264_RC_ABR ? pow( 0.95, h->param.rc.i_qp_min )
1087 : pow( 0.95, h->param.rc.i_qp_constant ) * X264_MAX( 1, h->param.rc.f_ip_factor )));
1089 CHECKED_MALLOC( h->nal_buffer, h->out.i_bitstream * 3/2 + 4 );
1090 h->nal_buffer_size = h->out.i_bitstream * 3/2 + 4;
1092 if( h->param.i_threads > 1 &&
1093 x264_threadpool_init( &h->threadpool, h->param.i_threads, (void*)x264_encoder_thread_init, h ) )
1097 for( int i = 1; i < h->param.i_threads + !!h->param.i_sync_lookahead; i++ )
1098 CHECKED_MALLOC( h->thread[i], sizeof(x264_t) );
1100 for( int i = 0; i < h->param.i_threads; i++ )
1102 int init_nal_count = h->param.i_slice_count + 3;
1103 int allocate_threadlocal_data = !h->param.b_sliced_threads || !i;
1107 if( allocate_threadlocal_data )
1109 h->thread[i]->fdec = x264_frame_pop_unused( h, 1 );
1110 if( !h->thread[i]->fdec )
1114 h->thread[i]->fdec = h->thread[0]->fdec;
1116 CHECKED_MALLOC( h->thread[i]->out.p_bitstream, h->out.i_bitstream );
1117 /* Start each thread with room for init_nal_count NAL units; it'll realloc later if needed. */
1118 CHECKED_MALLOC( h->thread[i]->out.nal, init_nal_count*sizeof(x264_nal_t) );
1119 h->thread[i]->out.i_nals_allocated = init_nal_count;
1121 if( allocate_threadlocal_data && x264_macroblock_cache_allocate( h->thread[i] ) < 0 )
1125 if( x264_lookahead_init( h, i_slicetype_length ) )
1128 for( int i = 0; i < h->param.i_threads; i++ )
1129 if( x264_macroblock_thread_allocate( h->thread[i], 0 ) < 0 )
1132 if( x264_ratecontrol_new( h ) < 0 )
1135 if( h->param.i_nal_hrd )
1137 x264_log( h, X264_LOG_DEBUG, "HRD bitrate: %i bits/sec\n", h->sps->vui.hrd.i_bit_rate_unscaled );
1138 x264_log( h, X264_LOG_DEBUG, "CPB size: %i bits\n", h->sps->vui.hrd.i_cpb_size_unscaled );
1141 if( h->param.psz_dump_yuv )
1143 /* create or truncate the reconstructed video file */
1144 FILE *f = fopen( h->param.psz_dump_yuv, "w" );
1147 x264_log( h, X264_LOG_ERROR, "dump_yuv: can't write to %s\n", h->param.psz_dump_yuv );
1150 else if( !x264_is_regular_file( f ) )
1152 x264_log( h, X264_LOG_ERROR, "dump_yuv: incompatible with non-regular file %s\n", h->param.psz_dump_yuv );
1158 const char *profile = h->sps->i_profile_idc == PROFILE_BASELINE ? "Baseline" :
1159 h->sps->i_profile_idc == PROFILE_MAIN ? "Main" :
1160 h->sps->i_profile_idc == PROFILE_HIGH ? "High" :
1161 h->sps->i_profile_idc == PROFILE_HIGH10 ? "High 10" :
1162 "High 4:4:4 Predictive";
1164 snprintf( level, sizeof(level), "%d.%d", h->sps->i_level_idc/10, h->sps->i_level_idc%10 );
1165 if( h->sps->i_level_idc == 9 || ( h->sps->i_level_idc == 11 && h->sps->b_constraint_set3 ) )
1166 strcpy( level, "1b" );
1168 if( h->sps->i_profile_idc < PROFILE_HIGH10 )
1170 x264_log( h, X264_LOG_INFO, "profile %s, level %s\n",
1175 x264_log( h, X264_LOG_INFO, "profile %s, level %s, bit depth %d\n",
1176 profile, level, BIT_DEPTH );
1185 /****************************************************************************
1186 * x264_encoder_reconfig:
1187 ****************************************************************************/
1188 int x264_encoder_reconfig( x264_t *h, x264_param_t *param )
1190 int rc_reconfig = 0;
1191 h = h->thread[h->thread[0]->i_thread_phase];
1192 x264_set_aspect_ratio( h, param, 0 );
1193 #define COPY(var) h->param.var = param->var
1194 COPY( i_frame_reference ); // but never uses more refs than initially specified
1195 COPY( i_bframe_bias );
1196 if( h->param.i_scenecut_threshold )
1197 COPY( i_scenecut_threshold ); // can't turn it on or off, only vary the threshold
1198 COPY( b_deblocking_filter );
1199 COPY( i_deblocking_filter_alphac0 );
1200 COPY( i_deblocking_filter_beta );
1201 COPY( analyse.inter );
1202 COPY( analyse.intra );
1203 COPY( analyse.i_direct_mv_pred );
1204 /* Scratch buffer prevents me_range from being increased for esa/tesa */
1205 if( h->param.analyse.i_me_method < X264_ME_ESA || param->analyse.i_me_range < h->param.analyse.i_me_range )
1206 COPY( analyse.i_me_range );
1207 COPY( analyse.i_noise_reduction );
1208 /* We can't switch out of subme=0 during encoding. */
1209 if( h->param.analyse.i_subpel_refine )
1210 COPY( analyse.i_subpel_refine );
1211 COPY( analyse.i_trellis );
1212 COPY( analyse.b_chroma_me );
1213 COPY( analyse.b_dct_decimate );
1214 COPY( analyse.b_fast_pskip );
1215 COPY( analyse.b_mixed_references );
1216 COPY( analyse.f_psy_rd );
1217 COPY( analyse.f_psy_trellis );
1218 // can only twiddle these if they were enabled to begin with:
1219 if( h->param.analyse.i_me_method >= X264_ME_ESA || param->analyse.i_me_method < X264_ME_ESA )
1220 COPY( analyse.i_me_method );
1221 if( h->param.analyse.i_me_method >= X264_ME_ESA && !h->frames.b_have_sub8x8_esa )
1222 h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
1223 if( h->pps->b_transform_8x8_mode )
1224 COPY( analyse.b_transform_8x8 );
1225 if( h->frames.i_max_ref1 > 1 )
1226 COPY( i_bframe_pyramid );
1227 COPY( i_slice_max_size );
1228 COPY( i_slice_max_mbs );
1229 COPY( i_slice_count );
1231 /* VBV can't be turned on if it wasn't on to begin with */
1232 if( h->param.rc.i_vbv_max_bitrate > 0 && h->param.rc.i_vbv_buffer_size > 0 &&
1233 param->rc.i_vbv_max_bitrate > 0 && param->rc.i_vbv_buffer_size > 0 )
1235 COPY( rc.i_vbv_max_bitrate );
1236 COPY( rc.i_vbv_buffer_size );
1237 COPY( rc.i_bitrate );
1240 if( h->param.rc.f_rf_constant != param->rc.f_rf_constant )
1242 COPY( rc.f_rf_constant );
1245 if( h->param.rc.f_rf_constant_max != param->rc.f_rf_constant_max )
1247 COPY( rc.f_rf_constant_max );
1255 int ret = x264_validate_parameters( h );
1257 /* Supported reconfiguration options (1-pass only):
1261 * bitrate (CBR only) */
1262 if( !ret && rc_reconfig )
1263 x264_ratecontrol_init_reconfigurable( h, 0 );
1268 /****************************************************************************
1269 * x264_encoder_parameters:
1270 ****************************************************************************/
1271 void x264_encoder_parameters( x264_t *h, x264_param_t *param )
1273 memcpy( param, &h->thread[h->i_thread_phase]->param, sizeof(x264_param_t) );
1276 /* internal usage */
1277 static void x264_nal_start( x264_t *h, int i_type, int i_ref_idc )
1279 x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1281 nal->i_ref_idc = i_ref_idc;
1282 nal->i_type = i_type;
1283 nal->b_long_startcode = 1;
1286 nal->p_payload= &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8];
1289 /* if number of allocated nals is not enough, re-allocate a larger one. */
1290 static int x264_nal_check_buffer( x264_t *h )
1292 if( h->out.i_nal >= h->out.i_nals_allocated )
1294 x264_nal_t *new_out = x264_malloc( sizeof(x264_nal_t) * (h->out.i_nals_allocated*2) );
1297 memcpy( new_out, h->out.nal, sizeof(x264_nal_t) * (h->out.i_nals_allocated) );
1298 x264_free( h->out.nal );
1299 h->out.nal = new_out;
1300 h->out.i_nals_allocated *= 2;
1305 static int x264_nal_end( x264_t *h )
1307 x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1308 nal->i_payload = &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8] - nal->p_payload;
1309 if( h->param.nalu_process )
1310 h->param.nalu_process( h, nal );
1313 return x264_nal_check_buffer( h );
1316 static int x264_encoder_encapsulate_nals( x264_t *h, int start )
1318 int nal_size = 0, previous_nal_size = 0;
1320 if( h->param.nalu_process )
1322 for( int i = start; i < h->out.i_nal; i++ )
1323 nal_size += h->out.nal[i].i_payload;
1327 for( int i = 0; i < start; i++ )
1328 previous_nal_size += h->out.nal[i].i_payload;
1330 for( int i = start; i < h->out.i_nal; i++ )
1331 nal_size += h->out.nal[i].i_payload;
1333 /* Worst-case NAL unit escaping: reallocate the buffer if it's too small. */
1334 if( h->nal_buffer_size < nal_size * 3/2 + h->out.i_nal * 4 )
1336 uint8_t *buf = x264_malloc( nal_size * 2 + h->out.i_nal * 4 );
1339 if( previous_nal_size )
1340 memcpy( buf, h->nal_buffer, previous_nal_size );
1341 x264_free( h->nal_buffer );
1342 h->nal_buffer = buf;
1345 uint8_t *nal_buffer = h->nal_buffer + previous_nal_size;
1347 for( int i = start; i < h->out.i_nal; i++ )
1349 h->out.nal[i].b_long_startcode = !i || h->out.nal[i].i_type == NAL_SPS || h->out.nal[i].i_type == NAL_PPS;
1350 x264_nal_encode( h, nal_buffer, &h->out.nal[i] );
1351 nal_buffer += h->out.nal[i].i_payload;
1356 return nal_buffer - (h->nal_buffer + previous_nal_size);
1359 /****************************************************************************
1360 * x264_encoder_headers:
1361 ****************************************************************************/
1362 int x264_encoder_headers( x264_t *h, x264_nal_t **pp_nal, int *pi_nal )
1365 /* init bitstream context */
1367 bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
1369 /* Write SEI, SPS and PPS. */
1371 /* generate sequence parameters */
1372 x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
1373 x264_sps_write( &h->out.bs, h->sps );
1374 if( x264_nal_end( h ) )
1377 /* generate picture parameters */
1378 x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
1379 x264_pps_write( &h->out.bs, h->pps );
1380 if( x264_nal_end( h ) )
1383 /* identify ourselves */
1384 x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
1385 if( x264_sei_version_write( h, &h->out.bs ) )
1387 if( x264_nal_end( h ) )
1390 frame_size = x264_encoder_encapsulate_nals( h, 0 );
1393 *pi_nal = h->out.i_nal;
1394 *pp_nal = &h->out.nal[0];
1400 /* Check to see whether we have chosen a reference list ordering different
1401 * from the standard's default. */
1402 static inline void x264_reference_check_reorder( x264_t *h )
1404 /* The reorder check doesn't check for missing frames, so just
1405 * force a reorder if one of the reference list is corrupt. */
1406 for( int i = 0; h->frames.reference[i]; i++ )
1407 if( h->frames.reference[i]->b_corrupt )
1409 h->b_ref_reorder[0] = 1;
1412 for( int i = 0; i < h->i_ref0 - 1; i++ )
1413 /* P and B-frames use different default orders. */
1414 if( h->sh.i_type == SLICE_TYPE_P ? h->fref0[i]->i_frame_num < h->fref0[i+1]->i_frame_num
1415 : h->fref0[i]->i_poc < h->fref0[i+1]->i_poc )
1417 h->b_ref_reorder[0] = 1;
1422 /* return -1 on failure, else return the index of the new reference frame */
1423 int x264_weighted_reference_duplicate( x264_t *h, int i_ref, const x264_weight_t *w )
1427 x264_frame_t *newframe;
1428 if( i <= 1 ) /* empty list, definitely can't duplicate frame */
1431 newframe = x264_frame_pop_blank_unused( h );
1433 //FIXME: probably don't need to copy everything
1434 *newframe = *h->fref0[i_ref];
1435 newframe->i_reference_count = 1;
1436 newframe->orig = h->fref0[i_ref];
1437 newframe->b_duplicate = 1;
1438 memcpy( h->fenc->weight[j], w, sizeof(h->fenc->weight[i]) );
1440 /* shift the frames to make space for the dupe. */
1441 h->b_ref_reorder[0] = 1;
1442 if( h->i_ref0 < X264_REF_MAX )
1444 h->fref0[X264_REF_MAX-1] = NULL;
1445 x264_frame_unshift( &h->fref0[j], newframe );
1450 static void x264_weighted_pred_init( x264_t *h )
1452 /* for now no analysis and set all weights to nothing */
1453 for( int i_ref = 0; i_ref < h->i_ref0; i_ref++ )
1454 h->fenc->weighted[i_ref] = h->fref0[i_ref]->filtered[0];
1456 // FIXME: This only supports weighting of one reference frame
1457 // and duplicates of that frame.
1458 h->fenc->i_lines_weighted = 0;
1460 for( int i_ref = 0; i_ref < (h->i_ref0 << h->sh.b_mbaff); i_ref++ )
1461 for( int i = 0; i < 3; i++ )
1462 h->sh.weight[i_ref][i].weightfn = NULL;
1465 if( h->sh.i_type != SLICE_TYPE_P || h->param.analyse.i_weighted_pred <= 0 )
1468 int i_padv = PADV << h->param.b_interlaced;
1471 int buffer_next = 0;
1472 //FIXME: when chroma support is added, move this into loop
1473 h->sh.weight[0][1].weightfn = h->sh.weight[0][2].weightfn = NULL;
1474 h->sh.weight[0][1].i_denom = h->sh.weight[0][2].i_denom = 0;
1475 for( int j = 0; j < h->i_ref0; j++ )
1477 if( h->fenc->weight[j][0].weightfn )
1479 h->sh.weight[j][0] = h->fenc->weight[j][0];
1480 // if weight is useless, don't write it to stream
1481 if( h->sh.weight[j][0].i_scale == 1<<h->sh.weight[j][0].i_denom && h->sh.weight[j][0].i_offset == 0 )
1482 h->sh.weight[j][0].weightfn = NULL;
1488 h->sh.weight[0][0].i_denom = denom = h->sh.weight[j][0].i_denom;
1489 assert( x264_clip3( denom, 0, 7 ) == denom );
1491 assert( h->sh.weight[j][0].i_denom == denom );
1492 assert( x264_clip3( h->sh.weight[j][0].i_scale, 0, 127 ) == h->sh.weight[j][0].i_scale );
1493 assert( x264_clip3( h->sh.weight[j][0].i_offset, -128, 127 ) == h->sh.weight[j][0].i_offset );
1494 h->fenc->weighted[j] = h->mb.p_weight_buf[buffer_next++] +
1495 h->fenc->i_stride[0] * i_padv + PADH;
1499 //scale full resolution frame
1500 if( h->sh.weight[j][0].weightfn && h->param.i_threads == 1 )
1502 pixel *src = h->fref0[j]->filtered[0] - h->fref0[j]->i_stride[0]*i_padv - PADH;
1503 pixel *dst = h->fenc->weighted[j] - h->fenc->i_stride[0]*i_padv - PADH;
1504 int stride = h->fenc->i_stride[0];
1505 int width = h->fenc->i_width[0] + PADH*2;
1506 int height = h->fenc->i_lines[0] + i_padv*2;
1507 x264_weight_scale_plane( h, dst, stride, src, stride, width, height, &h->sh.weight[j][0] );
1508 h->fenc->i_lines_weighted = height;
1512 h->sh.weight[0][0].i_denom = 0;
1515 static inline void x264_reference_build_list( x264_t *h, int i_poc )
1519 /* build ref list 0/1 */
1520 h->mb.pic.i_fref[0] = h->i_ref0 = 0;
1521 h->mb.pic.i_fref[1] = h->i_ref1 = 0;
1522 if( h->sh.i_type == SLICE_TYPE_I )
1525 for( int i = 0; h->frames.reference[i]; i++ )
1527 if( h->frames.reference[i]->b_corrupt )
1529 if( h->frames.reference[i]->i_poc < i_poc )
1530 h->fref0[h->i_ref0++] = h->frames.reference[i];
1531 else if( h->frames.reference[i]->i_poc > i_poc )
1532 h->fref1[h->i_ref1++] = h->frames.reference[i];
1535 /* Order ref0 from higher to lower poc */
1539 for( int i = 0; i < h->i_ref0 - 1; i++ )
1541 if( h->fref0[i]->i_poc < h->fref0[i+1]->i_poc )
1543 XCHG( x264_frame_t*, h->fref0[i], h->fref0[i+1] );
1550 if( h->sh.i_mmco_remove_from_end )
1551 for( int i = h->i_ref0-1; i >= h->i_ref0 - h->sh.i_mmco_remove_from_end; i-- )
1553 int diff = h->i_frame_num - h->fref0[i]->i_frame_num;
1554 h->sh.mmco[h->sh.i_mmco_command_count].i_poc = h->fref0[i]->i_poc;
1555 h->sh.mmco[h->sh.i_mmco_command_count++].i_difference_of_pic_nums = diff;
1558 /* Order ref1 from lower to higher poc (bubble sort) for B-frame */
1562 for( int i = 0; i < h->i_ref1 - 1; i++ )
1564 if( h->fref1[i]->i_poc > h->fref1[i+1]->i_poc )
1566 XCHG( x264_frame_t*, h->fref1[i], h->fref1[i+1] );
1573 x264_reference_check_reorder( h );
1575 h->i_ref1 = X264_MIN( h->i_ref1, h->frames.i_max_ref1 );
1576 h->i_ref0 = X264_MIN( h->i_ref0, h->frames.i_max_ref0 );
1577 h->i_ref0 = X264_MIN( h->i_ref0, h->param.i_frame_reference ); // if reconfig() has lowered the limit
1579 /* add duplicates */
1580 if( h->fenc->i_type == X264_TYPE_P )
1583 if( h->param.analyse.i_weighted_pred == X264_WEIGHTP_SMART )
1586 w[1].weightfn = w[2].weightfn = NULL;
1587 if( h->param.rc.b_stat_read )
1588 x264_ratecontrol_set_weights( h, h->fenc );
1590 if( !h->fenc->weight[0][0].weightfn )
1592 h->fenc->weight[0][0].i_denom = 0;
1593 SET_WEIGHT( w[0], 1, 1, 0, -1 );
1594 idx = x264_weighted_reference_duplicate( h, 0, w );
1598 if( h->fenc->weight[0][0].i_scale == 1<<h->fenc->weight[0][0].i_denom )
1600 SET_WEIGHT( h->fenc->weight[0][0], 1, 1, 0, h->fenc->weight[0][0].i_offset );
1602 x264_weighted_reference_duplicate( h, 0, weight_none );
1603 if( h->fenc->weight[0][0].i_offset > -128 )
1605 w[0] = h->fenc->weight[0][0];
1607 h->mc.weight_cache( h, &w[0] );
1608 idx = x264_weighted_reference_duplicate( h, 0, w );
1612 else if( h->param.analyse.i_weighted_pred == X264_WEIGHTP_BLIND )
1614 //weighted offset=-1
1616 SET_WEIGHT( w[0], 1, 1, 0, -1 );
1617 h->fenc->weight[0][0].i_denom = 0;
1618 w[1].weightfn = w[2].weightfn = NULL;
1619 idx = x264_weighted_reference_duplicate( h, 0, w );
1621 h->mb.ref_blind_dupe = idx;
1624 assert( h->i_ref0 + h->i_ref1 <= X264_REF_MAX );
1625 h->mb.pic.i_fref[0] = h->i_ref0;
1626 h->mb.pic.i_fref[1] = h->i_ref1;
1629 static void x264_fdec_filter_row( x264_t *h, int mb_y, int b_inloop )
1631 /* mb_y is the mb to be encoded next, not the mb to be filtered here */
1632 int b_hpel = h->fdec->b_kept_as_ref;
1633 int b_deblock = h->sh.i_disable_deblocking_filter_idc != 1;
1634 int b_end = mb_y == h->i_threadslice_end;
1635 int b_measure_quality = 1;
1636 int min_y = mb_y - (1 << h->sh.b_mbaff);
1637 int b_start = min_y == h->i_threadslice_start;
1638 int max_y = b_end ? h->i_threadslice_end : mb_y;
1639 b_deblock &= b_hpel || h->param.psz_dump_yuv;
1640 if( h->param.b_sliced_threads && b_start && min_y && !b_inloop )
1642 b_deblock = 0; /* We already deblocked on the inloop pass. */
1643 b_measure_quality = 0; /* We already measured quality on the inloop pass. */
1645 if( mb_y & h->sh.b_mbaff )
1647 if( min_y < h->i_threadslice_start )
1651 for( int y = min_y; y < max_y; y += (1 << h->sh.b_mbaff) )
1652 x264_frame_deblock_row( h, y );
1656 int end = mb_y == h->mb.i_mb_height;
1657 x264_frame_expand_border( h, h->fdec, min_y, end );
1658 if( h->param.analyse.i_subpel_refine )
1660 x264_frame_filter( h, h->fdec, min_y, end );
1661 x264_frame_expand_border_filtered( h, h->fdec, min_y, end );
1665 if( h->i_thread_frames > 1 && h->fdec->b_kept_as_ref )
1666 x264_frame_cond_broadcast( h->fdec, mb_y*16 + (b_end ? 10000 : -(X264_THREAD_HEIGHT << h->sh.b_mbaff)) );
1668 min_y = min_y*16 - 8 * !b_start;
1669 max_y = b_end ? X264_MIN( h->i_threadslice_end*16 , h->param.i_height ) : mb_y*16 - 8;
1671 if( b_measure_quality )
1673 if( h->param.analyse.b_psnr )
1675 uint64_t ssd_y = x264_pixel_ssd_wxh( &h->pixf,
1676 h->fdec->plane[0] + min_y * h->fdec->i_stride[0], h->fdec->i_stride[0],
1677 h->fenc->plane[0] + min_y * h->fenc->i_stride[0], h->fenc->i_stride[0],
1678 h->param.i_width, max_y-min_y );
1679 uint64_t ssd_uv = x264_pixel_ssd_nv12( &h->pixf,
1680 h->fdec->plane[1] + (min_y>>1) * h->fdec->i_stride[1], h->fdec->i_stride[1],
1681 h->fenc->plane[1] + (min_y>>1) * h->fenc->i_stride[1], h->fenc->i_stride[1],
1682 h->param.i_width>>1, (max_y-min_y)>>1 );
1683 h->stat.frame.i_ssd[0] += ssd_y;
1684 h->stat.frame.i_ssd[1] += (uint32_t)ssd_uv;
1685 h->stat.frame.i_ssd[2] += ssd_uv>>32;
1688 if( h->param.analyse.b_ssim )
1691 /* offset by 2 pixels to avoid alignment of ssim blocks with dct blocks,
1692 * and overlap by 4 */
1693 min_y += b_start ? 2 : -6;
1694 h->stat.frame.f_ssim +=
1695 x264_pixel_ssim_wxh( &h->pixf,
1696 h->fdec->plane[0] + 2+min_y*h->fdec->i_stride[0], h->fdec->i_stride[0],
1697 h->fenc->plane[0] + 2+min_y*h->fenc->i_stride[0], h->fenc->i_stride[0],
1698 h->param.i_width-2, max_y-min_y, h->scratch_buffer );
1703 static inline int x264_reference_update( x264_t *h )
1705 if( !h->fdec->b_kept_as_ref )
1707 if( h->i_thread_frames > 1 )
1709 x264_frame_push_unused( h, h->fdec );
1710 h->fdec = x264_frame_pop_unused( h, 1 );
1717 /* apply mmco from previous frame. */
1718 for( int i = 0; i < h->sh.i_mmco_command_count; i++ )
1719 for( int j = 0; h->frames.reference[j]; j++ )
1720 if( h->frames.reference[j]->i_poc == h->sh.mmco[i].i_poc )
1721 x264_frame_push_unused( h, x264_frame_shift( &h->frames.reference[j] ) );
1723 /* move frame in the buffer */
1724 x264_frame_push( h->frames.reference, h->fdec );
1725 if( h->frames.reference[h->sps->i_num_ref_frames] )
1726 x264_frame_push_unused( h, x264_frame_shift( h->frames.reference ) );
1727 h->fdec = x264_frame_pop_unused( h, 1 );
1733 static inline void x264_reference_reset( x264_t *h )
1735 while( h->frames.reference[0] )
1736 x264_frame_push_unused( h, x264_frame_pop( h->frames.reference ) );
1741 static inline void x264_reference_hierarchy_reset( x264_t *h )
1744 int b_hasdelayframe = 0;
1746 /* look for delay frames -- chain must only contain frames that are disposable */
1747 for( int i = 0; h->frames.current[i] && IS_DISPOSABLE( h->frames.current[i]->i_type ); i++ )
1748 b_hasdelayframe |= h->frames.current[i]->i_coded
1749 != h->frames.current[i]->i_frame + h->sps->vui.i_num_reorder_frames;
1751 /* This function must handle b-pyramid and clear frames for open-gop */
1752 if( h->param.i_bframe_pyramid != X264_B_PYRAMID_STRICT && !b_hasdelayframe && h->frames.i_poc_last_open_gop == -1 )
1755 /* Remove last BREF. There will never be old BREFs in the
1756 * dpb during a BREF decode when pyramid == STRICT */
1757 for( ref = 0; h->frames.reference[ref]; ref++ )
1759 if( ( h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT
1760 && h->frames.reference[ref]->i_type == X264_TYPE_BREF )
1761 || ( h->frames.reference[ref]->i_poc < h->frames.i_poc_last_open_gop
1762 && h->sh.i_type != SLICE_TYPE_B ) )
1764 int diff = h->i_frame_num - h->frames.reference[ref]->i_frame_num;
1765 h->sh.mmco[h->sh.i_mmco_command_count].i_difference_of_pic_nums = diff;
1766 h->sh.mmco[h->sh.i_mmco_command_count++].i_poc = h->frames.reference[ref]->i_poc;
1767 x264_frame_push_unused( h, x264_frame_shift( &h->frames.reference[ref] ) );
1768 h->b_ref_reorder[0] = 1;
1773 /* Prepare room in the dpb for the delayed display time of the later b-frame's */
1774 if( h->param.i_bframe_pyramid )
1775 h->sh.i_mmco_remove_from_end = X264_MAX( ref + 2 - h->frames.i_max_dpb, 0 );
1778 static inline void x264_slice_init( x264_t *h, int i_nal_type, int i_global_qp )
1780 /* ------------------------ Create slice header ----------------------- */
1781 if( i_nal_type == NAL_SLICE_IDR )
1783 x264_slice_header_init( h, &h->sh, h->sps, h->pps, h->i_idr_pic_id, h->i_frame_num, i_global_qp );
1786 h->i_idr_pic_id ^= 1;
1790 x264_slice_header_init( h, &h->sh, h->sps, h->pps, -1, h->i_frame_num, i_global_qp );
1792 h->sh.i_num_ref_idx_l0_active = h->i_ref0 <= 0 ? 1 : h->i_ref0;
1793 h->sh.i_num_ref_idx_l1_active = h->i_ref1 <= 0 ? 1 : h->i_ref1;
1794 if( h->sh.i_num_ref_idx_l0_active != h->pps->i_num_ref_idx_l0_default_active ||
1795 (h->sh.i_type == SLICE_TYPE_B && h->sh.i_num_ref_idx_l1_active != h->pps->i_num_ref_idx_l1_default_active) )
1797 h->sh.b_num_ref_idx_override = 1;
1801 h->fdec->i_frame_num = h->sh.i_frame_num;
1803 if( h->sps->i_poc_type == 0 )
1805 h->sh.i_poc = h->fdec->i_poc;
1806 if( h->param.b_interlaced )
1808 h->sh.i_delta_poc_bottom = h->param.b_tff ? 1 : -1;
1809 if( h->sh.i_delta_poc_bottom == -1 )
1810 h->sh.i_poc = h->fdec->i_poc + 1;
1813 h->sh.i_delta_poc_bottom = 0;
1815 else if( h->sps->i_poc_type == 1 )
1817 /* FIXME TODO FIXME */
1821 /* Nothing to do ? */
1824 x264_macroblock_slice_init( h );
1827 static int x264_slice_write( x264_t *h )
1830 int mb_xy, i_mb_x, i_mb_y;
1831 int i_skip_bak = 0; /* Shut up GCC. */
1833 x264_cabac_t cabac_bak;
1834 uint8_t cabac_prevbyte_bak = 0; /* Shut up GCC. */
1835 int mv_bits_bak = 0;
1836 int tex_bits_bak = 0;
1837 /* Assume no more than 3 bytes of NALU escaping.
1838 * NALUs other than the first use a 3-byte startcode. */
1839 int overhead_guess = (NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal)) + 3;
1840 int slice_max_size = h->param.i_slice_max_size > 0 ? (h->param.i_slice_max_size-overhead_guess)*8 : INT_MAX;
1841 int starting_bits = bs_pos(&h->out.bs);
1842 int b_deblock = h->sh.i_disable_deblocking_filter_idc != 1;
1843 int b_hpel = h->fdec->b_kept_as_ref;
1844 b_deblock &= b_hpel || h->param.psz_dump_yuv;
1845 bs_realign( &h->out.bs );
1848 x264_nal_start( h, h->i_nal_type, h->i_nal_ref_idc );
1849 h->out.nal[h->out.i_nal].i_first_mb = h->sh.i_first_mb;
1852 x264_macroblock_thread_init( h );
1854 /* If this isn't the first slice in the threadslice, set the slice QP
1855 * equal to the last QP in the previous slice for more accurate
1856 * CABAC initialization. */
1857 if( h->sh.i_first_mb != h->i_threadslice_start * h->mb.i_mb_width )
1859 h->sh.i_qp = h->mb.i_last_qp;
1860 h->sh.i_qp_delta = h->sh.i_qp - h->pps->i_pic_init_qp;
1863 x264_slice_header_write( &h->out.bs, &h->sh, h->i_nal_ref_idc );
1864 if( h->param.b_cabac )
1866 /* alignment needed */
1867 bs_align_1( &h->out.bs );
1870 x264_cabac_context_init( &h->cabac, h->sh.i_type, x264_clip3( h->sh.i_qp-QP_BD_OFFSET, 0, 51 ), h->sh.i_cabac_init_idc );
1871 x264_cabac_encode_init ( &h->cabac, h->out.bs.p, h->out.bs.p_end );
1873 h->mb.i_last_qp = h->sh.i_qp;
1874 h->mb.i_last_dqp = 0;
1876 i_mb_y = h->sh.i_first_mb / h->mb.i_mb_width;
1877 i_mb_x = h->sh.i_first_mb % h->mb.i_mb_width;
1880 while( (mb_xy = i_mb_x + i_mb_y * h->mb.i_mb_width) <= h->sh.i_last_mb )
1882 int mb_spos = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
1884 if( x264_bitstream_check_buffer( h ) )
1887 if( h->param.i_slice_max_size > 0 )
1889 mv_bits_bak = h->stat.frame.i_mv_bits;
1890 tex_bits_bak = h->stat.frame.i_tex_bits;
1891 /* We don't need the contexts because flushing the CABAC encoder has no context
1892 * dependency and macroblocks are only re-encoded in the case where a slice is
1893 * ended (and thus the content of all contexts are thrown away). */
1894 if( h->param.b_cabac )
1896 memcpy( &cabac_bak, &h->cabac, offsetof(x264_cabac_t, f8_bits_encoded) );
1897 /* x264's CABAC writer modifies the previous byte during carry, so it has to be
1899 cabac_prevbyte_bak = h->cabac.p[-1];
1904 i_skip_bak = i_skip;
1908 if( i_mb_x == 0 && !h->mb.b_reencode_mb )
1909 x264_fdec_filter_row( h, i_mb_y, 1 );
1912 x264_macroblock_cache_load( h, i_mb_x, i_mb_y );
1914 x264_macroblock_analyse( h );
1916 /* encode this macroblock -> be careful it can change the mb type to P_SKIP if needed */
1917 x264_macroblock_encode( h );
1919 if( h->param.b_cabac )
1921 if( mb_xy > h->sh.i_first_mb && !(h->sh.b_mbaff && (i_mb_y&1)) )
1922 x264_cabac_encode_terminal( &h->cabac );
1924 if( IS_SKIP( h->mb.i_type ) )
1925 x264_cabac_mb_skip( h, 1 );
1928 if( h->sh.i_type != SLICE_TYPE_I )
1929 x264_cabac_mb_skip( h, 0 );
1930 x264_macroblock_write_cabac( h, &h->cabac );
1935 if( IS_SKIP( h->mb.i_type ) )
1939 if( h->sh.i_type != SLICE_TYPE_I )
1941 bs_write_ue( &h->out.bs, i_skip ); /* skip run */
1944 x264_macroblock_write_cavlc( h );
1948 int total_bits = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
1949 int mb_size = total_bits - mb_spos;
1951 /* We'll just re-encode this last macroblock if we go over the max slice size. */
1952 if( total_bits - starting_bits > slice_max_size && !h->mb.b_reencode_mb )
1954 if( mb_xy != h->sh.i_first_mb )
1956 h->stat.frame.i_mv_bits = mv_bits_bak;
1957 h->stat.frame.i_tex_bits = tex_bits_bak;
1958 if( h->param.b_cabac )
1960 memcpy( &h->cabac, &cabac_bak, offsetof(x264_cabac_t, f8_bits_encoded) );
1961 h->cabac.p[-1] = cabac_prevbyte_bak;
1966 i_skip = i_skip_bak;
1968 h->mb.b_reencode_mb = 1;
1969 h->sh.i_last_mb = mb_xy-1;
1974 h->sh.i_last_mb = mb_xy;
1975 h->mb.b_reencode_mb = 0;
1979 h->mb.b_reencode_mb = 0;
1982 if( h->param.b_visualize )
1983 x264_visualize_mb( h );
1987 x264_macroblock_cache_save( h );
1989 /* accumulate mb stats */
1990 h->stat.frame.i_mb_count[h->mb.i_type]++;
1992 int b_intra = IS_INTRA( h->mb.i_type );
1993 if( h->param.i_log_level >= X264_LOG_INFO || h->param.rc.b_stat_write )
1995 if( !b_intra && !IS_SKIP( h->mb.i_type ) && !IS_DIRECT( h->mb.i_type ) )
1997 if( h->mb.i_partition != D_8x8 )
1998 h->stat.frame.i_mb_partition[h->mb.i_partition] += 4;
2000 for( int i = 0; i < 4; i++ )
2001 h->stat.frame.i_mb_partition[h->mb.i_sub_partition[i]] ++;
2002 if( h->param.i_frame_reference > 1 )
2003 for( int i_list = 0; i_list <= (h->sh.i_type == SLICE_TYPE_B); i_list++ )
2004 for( int i = 0; i < 4; i++ )
2006 int i_ref = h->mb.cache.ref[i_list][ x264_scan8[4*i] ];
2008 h->stat.frame.i_mb_count_ref[i_list][i_ref] ++;
2013 if( h->param.i_log_level >= X264_LOG_INFO )
2015 if( h->mb.i_cbp_luma | h->mb.i_cbp_chroma )
2017 int cbpsum = (h->mb.i_cbp_luma&1) + ((h->mb.i_cbp_luma>>1)&1)
2018 + ((h->mb.i_cbp_luma>>2)&1) + (h->mb.i_cbp_luma>>3);
2019 h->stat.frame.i_mb_cbp[!b_intra + 0] += cbpsum;
2020 h->stat.frame.i_mb_cbp[!b_intra + 2] += !!h->mb.i_cbp_chroma;
2021 h->stat.frame.i_mb_cbp[!b_intra + 4] += h->mb.i_cbp_chroma >> 1;
2023 if( h->mb.i_cbp_luma && !b_intra )
2025 h->stat.frame.i_mb_count_8x8dct[0] ++;
2026 h->stat.frame.i_mb_count_8x8dct[1] += h->mb.b_transform_8x8;
2028 if( b_intra && h->mb.i_type != I_PCM )
2030 if( h->mb.i_type == I_16x16 )
2031 h->stat.frame.i_mb_pred_mode[0][h->mb.i_intra16x16_pred_mode]++;
2032 else if( h->mb.i_type == I_8x8 )
2033 for( int i = 0; i < 16; i += 4 )
2034 h->stat.frame.i_mb_pred_mode[1][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
2035 else //if( h->mb.i_type == I_4x4 )
2036 for( int i = 0; i < 16; i++ )
2037 h->stat.frame.i_mb_pred_mode[2][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
2038 h->stat.frame.i_mb_pred_mode[3][x264_mb_pred_mode8x8c_fix[h->mb.i_chroma_pred_mode]]++;
2042 /* calculate deblock strength values (actual deblocking is done per-row along with hpel) */
2045 int mvy_limit = 4 >> h->sh.b_mbaff;
2046 uint8_t (*bs)[4][4] = h->deblock_strength[h->mb.i_mb_y&1][h->mb.i_mb_x];
2047 x264_macroblock_cache_load_deblock( h );
2048 if( IS_INTRA( h->mb.type[h->mb.i_mb_xy] ) )
2049 memset( bs, 3, 2*4*4*sizeof(uint8_t) );
2051 h->loopf.deblock_strength( h->mb.cache.non_zero_count, h->mb.cache.ref, h->mb.cache.mv,
2052 bs, mvy_limit, h->sh.i_type == SLICE_TYPE_B );
2055 x264_ratecontrol_mb( h, mb_size );
2059 i_mb_x += i_mb_y & 1;
2060 i_mb_y ^= i_mb_x < h->mb.i_mb_width;
2064 if( i_mb_x == h->mb.i_mb_width )
2070 h->out.nal[h->out.i_nal].i_last_mb = h->sh.i_last_mb;
2072 if( h->param.b_cabac )
2074 x264_cabac_encode_flush( h, &h->cabac );
2075 h->out.bs.p = h->cabac.p;
2080 bs_write_ue( &h->out.bs, i_skip ); /* last skip run */
2081 /* rbsp_slice_trailing_bits */
2082 bs_rbsp_trailing( &h->out.bs );
2083 bs_flush( &h->out.bs );
2085 if( x264_nal_end( h ) )
2088 if( h->sh.i_last_mb == (h->i_threadslice_end * h->mb.i_mb_width - 1) )
2090 h->stat.frame.i_misc_bits = bs_pos( &h->out.bs )
2091 + (h->out.i_nal*NALU_OVERHEAD * 8)
2092 - h->stat.frame.i_tex_bits
2093 - h->stat.frame.i_mv_bits;
2094 x264_fdec_filter_row( h, h->i_threadslice_end, 1 );
2100 static void x264_thread_sync_context( x264_t *dst, x264_t *src )
2105 // reference counting
2106 for( x264_frame_t **f = src->frames.reference; *f; f++ )
2107 (*f)->i_reference_count++;
2108 for( x264_frame_t **f = dst->frames.reference; *f; f++ )
2109 x264_frame_push_unused( src, *f );
2110 src->fdec->i_reference_count++;
2111 x264_frame_push_unused( src, dst->fdec );
2113 // copy everything except the per-thread pointers and the constants.
2114 memcpy( &dst->i_frame, &src->i_frame, offsetof(x264_t, mb.type) - offsetof(x264_t, i_frame) );
2115 dst->param = src->param;
2116 dst->stat = src->stat;
2119 static void x264_thread_sync_stat( x264_t *dst, x264_t *src )
2123 memcpy( &dst->stat.i_frame_count, &src->stat.i_frame_count, sizeof(dst->stat) - sizeof(dst->stat.frame) );
2126 static void *x264_slices_write( x264_t *h )
2128 int i_slice_num = 0;
2129 int last_thread_mb = h->sh.i_last_mb;
2132 if( h->param.b_visualize )
2133 if( x264_visualize_init( h ) )
2138 memset( &h->stat.frame, 0, sizeof(h->stat.frame) );
2139 h->mb.b_reencode_mb = 0;
2140 while( h->sh.i_first_mb <= last_thread_mb )
2142 h->sh.i_last_mb = last_thread_mb;
2143 if( h->param.i_slice_max_mbs )
2144 h->sh.i_last_mb = h->sh.i_first_mb + h->param.i_slice_max_mbs - 1;
2145 else if( h->param.i_slice_count && !h->param.b_sliced_threads )
2147 int height = h->mb.i_mb_height >> h->param.b_interlaced;
2148 int width = h->mb.i_mb_width << h->param.b_interlaced;
2150 h->sh.i_last_mb = (height * i_slice_num + h->param.i_slice_count/2) / h->param.i_slice_count * width - 1;
2152 h->sh.i_last_mb = X264_MIN( h->sh.i_last_mb, last_thread_mb );
2153 if( x264_stack_align( x264_slice_write, h ) )
2155 h->sh.i_first_mb = h->sh.i_last_mb + 1;
2159 if( h->param.b_visualize )
2161 x264_visualize_show( h );
2162 x264_visualize_close( h );
2169 static int x264_threaded_slices_write( x264_t *h )
2171 /* set first/last mb and sync contexts */
2172 for( int i = 0; i < h->param.i_threads; i++ )
2174 x264_t *t = h->thread[i];
2177 t->param = h->param;
2178 memcpy( &t->i_frame, &h->i_frame, offsetof(x264_t, rc) - offsetof(x264_t, i_frame) );
2180 int height = h->mb.i_mb_height >> h->param.b_interlaced;
2181 t->i_threadslice_start = ((height * i + h->param.i_slice_count/2) / h->param.i_threads) << h->param.b_interlaced;
2182 t->i_threadslice_end = ((height * (i+1) + h->param.i_slice_count/2) / h->param.i_threads) << h->param.b_interlaced;
2183 t->sh.i_first_mb = t->i_threadslice_start * h->mb.i_mb_width;
2184 t->sh.i_last_mb = t->i_threadslice_end * h->mb.i_mb_width - 1;
2187 x264_stack_align( x264_analyse_weight_frame, h, h->mb.i_mb_height*16 + 16 );
2189 x264_threads_distribute_ratecontrol( h );
2192 for( int i = 0; i < h->param.i_threads; i++ )
2194 x264_threadpool_run( h->threadpool, (void*)x264_slices_write, h->thread[i] );
2195 h->thread[i]->b_thread_active = 1;
2197 for( int i = 0; i < h->param.i_threads; i++ )
2199 h->thread[i]->b_thread_active = 0;
2200 if( (intptr_t)x264_threadpool_wait( h->threadpool, h->thread[i] ) )
2204 /* Go back and fix up the hpel on the borders between slices. */
2205 for( int i = 1; i < h->param.i_threads; i++ )
2207 x264_fdec_filter_row( h->thread[i], h->thread[i]->i_threadslice_start + 1, 0 );
2209 x264_fdec_filter_row( h->thread[i], h->thread[i]->i_threadslice_start + 2, 0 );
2212 x264_threads_merge_ratecontrol( h );
2214 for( int i = 1; i < h->param.i_threads; i++ )
2216 x264_t *t = h->thread[i];
2217 for( int j = 0; j < t->out.i_nal; j++ )
2219 h->out.nal[h->out.i_nal] = t->out.nal[j];
2221 x264_nal_check_buffer( h );
2223 /* All entries in stat.frame are ints except for ssd/ssim. */
2224 for( int j = 0; j < (offsetof(x264_t,stat.frame.i_ssd) - offsetof(x264_t,stat.frame.i_mv_bits)) / sizeof(int); j++ )
2225 ((int*)&h->stat.frame)[j] += ((int*)&t->stat.frame)[j];
2226 for( int j = 0; j < 3; j++ )
2227 h->stat.frame.i_ssd[j] += t->stat.frame.i_ssd[j];
2228 h->stat.frame.f_ssim += t->stat.frame.f_ssim;
2234 void x264_encoder_intra_refresh( x264_t *h )
2236 h = h->thread[h->i_thread_phase];
2237 h->b_queued_intra_refresh = 1;
2240 int x264_encoder_invalidate_reference( x264_t *h, int64_t pts )
2242 if( h->param.i_bframe )
2244 x264_log( h, X264_LOG_ERROR, "x264_encoder_invalidate_reference is not supported with B-frames enabled\n" );
2247 if( h->param.b_intra_refresh )
2249 x264_log( h, X264_LOG_ERROR, "x264_encoder_invalidate_reference is not supported with intra refresh enabled\n" );
2252 h = h->thread[h->i_thread_phase];
2253 if( pts >= h->i_last_idr_pts )
2255 for( int i = 0; h->frames.reference[i]; i++ )
2256 if( pts <= h->frames.reference[i]->i_pts )
2257 h->frames.reference[i]->b_corrupt = 1;
2258 if( pts <= h->fdec->i_pts )
2259 h->fdec->b_corrupt = 1;
2264 /****************************************************************************
2265 * x264_encoder_encode:
2266 * XXX: i_poc : is the poc of the current given picture
2267 * i_frame : is the number of the frame being coded
2268 * ex: type frame poc
2276 ****************************************************************************/
2277 int x264_encoder_encode( x264_t *h,
2278 x264_nal_t **pp_nal, int *pi_nal,
2279 x264_picture_t *pic_in,
2280 x264_picture_t *pic_out )
2282 x264_t *thread_current, *thread_prev, *thread_oldest;
2283 int i_nal_type, i_nal_ref_idc, i_global_qp;
2284 int overhead = NALU_OVERHEAD;
2286 if( h->i_thread_frames > 1 )
2288 thread_prev = h->thread[ h->i_thread_phase ];
2289 h->i_thread_phase = (h->i_thread_phase + 1) % h->i_thread_frames;
2290 thread_current = h->thread[ h->i_thread_phase ];
2291 thread_oldest = h->thread[ (h->i_thread_phase + 1) % h->i_thread_frames ];
2292 x264_thread_sync_context( thread_current, thread_prev );
2293 x264_thread_sync_ratecontrol( thread_current, thread_prev, thread_oldest );
2302 if( h->i_thread_frames == 1 && h->param.cpu&X264_CPU_SSE_MISALIGN )
2303 x264_cpu_mask_misalign_sse();
2306 // ok to call this before encoding any frames, since the initial values of fdec have b_kept_as_ref=0
2307 if( x264_reference_update( h ) )
2309 h->fdec->i_lines_completed = -1;
2315 /* ------------------- Setup new frame from picture -------------------- */
2316 if( pic_in != NULL )
2318 /* 1: Copy the picture to a frame and move it to a buffer */
2319 x264_frame_t *fenc = x264_frame_pop_unused( h, 0 );
2323 if( x264_frame_copy_picture( h, fenc, pic_in ) < 0 )
2326 if( h->param.i_width != 16 * h->mb.i_mb_width ||
2327 h->param.i_height != 16 * h->mb.i_mb_height )
2328 x264_frame_expand_border_mod16( h, fenc );
2330 fenc->i_frame = h->frames.i_input++;
2332 if( fenc->i_frame == 0 )
2333 h->frames.i_first_pts = fenc->i_pts;
2334 if( h->frames.i_bframe_delay && fenc->i_frame == h->frames.i_bframe_delay )
2335 h->frames.i_bframe_delay_time = fenc->i_pts - h->frames.i_first_pts;
2337 if( h->param.b_vfr_input && fenc->i_pts <= h->frames.i_largest_pts )
2338 x264_log( h, X264_LOG_WARNING, "non-strictly-monotonic PTS\n" );
2340 h->frames.i_second_largest_pts = h->frames.i_largest_pts;
2341 h->frames.i_largest_pts = fenc->i_pts;
2343 if( (fenc->i_pic_struct < PIC_STRUCT_AUTO) || (fenc->i_pic_struct > PIC_STRUCT_TRIPLE) )
2344 fenc->i_pic_struct = PIC_STRUCT_AUTO;
2346 if( fenc->i_pic_struct == PIC_STRUCT_AUTO )
2348 int b_interlaced = fenc->param ? fenc->param->b_interlaced : h->param.b_interlaced;
2351 int b_tff = fenc->param ? fenc->param->b_tff : h->param.b_tff;
2352 fenc->i_pic_struct = b_tff ? PIC_STRUCT_TOP_BOTTOM : PIC_STRUCT_BOTTOM_TOP;
2355 fenc->i_pic_struct = PIC_STRUCT_PROGRESSIVE;
2358 if( h->param.rc.b_mb_tree && h->param.rc.b_stat_read )
2360 if( x264_macroblock_tree_read( h, fenc, pic_in->prop.quant_offsets ) )
2364 x264_stack_align( x264_adaptive_quant_frame, h, fenc, pic_in->prop.quant_offsets );
2366 if( pic_in->prop.quant_offsets_free )
2367 pic_in->prop.quant_offsets_free( pic_in->prop.quant_offsets );
2369 if( h->frames.b_have_lowres )
2370 x264_frame_init_lowres( h, fenc );
2372 /* 2: Place the frame into the queue for its slice type decision */
2373 x264_lookahead_put_frame( h, fenc );
2375 if( h->frames.i_input <= h->frames.i_delay + 1 - h->i_thread_frames )
2377 /* Nothing yet to encode, waiting for filling of buffers */
2378 pic_out->i_type = X264_TYPE_AUTO;
2384 /* signal kills for lookahead thread */
2385 x264_pthread_mutex_lock( &h->lookahead->ifbuf.mutex );
2386 h->lookahead->b_exit_thread = 1;
2387 x264_pthread_cond_broadcast( &h->lookahead->ifbuf.cv_fill );
2388 x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
2392 /* 3: The picture is analyzed in the lookahead */
2393 if( !h->frames.current[0] )
2394 x264_lookahead_get_frames( h );
2396 if( !h->frames.current[0] && x264_lookahead_is_empty( h ) )
2397 return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
2399 /* ------------------- Get frame to be encoded ------------------------- */
2400 /* 4: get picture to encode */
2401 h->fenc = x264_frame_shift( h->frames.current );
2402 if( h->i_frame == h->i_thread_frames - 1 )
2403 h->i_reordered_pts_delay = h->fenc->i_reordered_pts;
2404 if( h->fenc->param )
2406 x264_encoder_reconfig( h, h->fenc->param );
2407 if( h->fenc->param->param_free )
2408 h->fenc->param->param_free( h->fenc->param );
2411 if( !IS_X264_TYPE_I( h->fenc->i_type ) )
2413 int valid_refs_left = 0;
2414 for( int i = 0; h->frames.reference[i]; i++ )
2415 if( !h->frames.reference[i]->b_corrupt )
2417 /* No valid reference frames left: force an IDR. */
2418 if( !valid_refs_left )
2420 h->fenc->b_keyframe = 1;
2421 h->fenc->i_type = X264_TYPE_IDR;
2425 if( h->fenc->b_keyframe )
2427 h->frames.i_last_keyframe = h->fenc->i_frame;
2428 if( h->fenc->i_type == X264_TYPE_IDR )
2431 h->frames.i_last_idr = h->fenc->i_frame;
2434 h->sh.i_mmco_command_count =
2435 h->sh.i_mmco_remove_from_end = 0;
2436 h->b_ref_reorder[0] =
2437 h->b_ref_reorder[1] = 0;
2439 h->fenc->i_poc = 2 * ( h->fenc->i_frame - X264_MAX( h->frames.i_last_idr, 0 ) );
2441 /* ------------------- Setup frame context ----------------------------- */
2442 /* 5: Init data dependent of frame type */
2443 if( h->fenc->i_type == X264_TYPE_IDR )
2445 /* reset ref pictures */
2446 i_nal_type = NAL_SLICE_IDR;
2447 i_nal_ref_idc = NAL_PRIORITY_HIGHEST;
2448 h->sh.i_type = SLICE_TYPE_I;
2449 x264_reference_reset( h );
2450 h->frames.i_poc_last_open_gop = -1;
2452 else if( h->fenc->i_type == X264_TYPE_I )
2454 i_nal_type = NAL_SLICE;
2455 i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
2456 h->sh.i_type = SLICE_TYPE_I;
2457 x264_reference_hierarchy_reset( h );
2458 if( h->param.i_open_gop )
2459 h->frames.i_poc_last_open_gop = h->fenc->b_keyframe ? h->fenc->i_poc : -1;
2461 else if( h->fenc->i_type == X264_TYPE_P )
2463 i_nal_type = NAL_SLICE;
2464 i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
2465 h->sh.i_type = SLICE_TYPE_P;
2466 x264_reference_hierarchy_reset( h );
2467 h->frames.i_poc_last_open_gop = -1;
2469 else if( h->fenc->i_type == X264_TYPE_BREF )
2471 i_nal_type = NAL_SLICE;
2472 i_nal_ref_idc = h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT ? NAL_PRIORITY_LOW : NAL_PRIORITY_HIGH;
2473 h->sh.i_type = SLICE_TYPE_B;
2474 x264_reference_hierarchy_reset( h );
2478 i_nal_type = NAL_SLICE;
2479 i_nal_ref_idc = NAL_PRIORITY_DISPOSABLE;
2480 h->sh.i_type = SLICE_TYPE_B;
2483 h->fdec->i_type = h->fenc->i_type;
2484 h->fdec->i_frame = h->fenc->i_frame;
2485 h->fenc->b_kept_as_ref =
2486 h->fdec->b_kept_as_ref = i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE && h->param.i_keyint_max > 1;
2488 h->fdec->i_pts = h->fenc->i_pts *= h->i_dts_compress_multiplier;
2489 if( h->frames.i_bframe_delay )
2491 int64_t *prev_reordered_pts = thread_current->frames.i_prev_reordered_pts;
2492 if( h->i_frame <= h->frames.i_bframe_delay )
2494 if( h->i_dts_compress_multiplier == 1 )
2495 h->fdec->i_dts = h->fenc->i_reordered_pts - h->frames.i_bframe_delay_time;
2498 /* DTS compression */
2499 if( h->i_frame == 1 )
2500 thread_current->frames.i_init_delta = (h->fenc->i_reordered_pts - h->frames.i_first_pts) * h->i_dts_compress_multiplier;
2501 h->fdec->i_dts = h->i_frame * thread_current->frames.i_init_delta / h->i_dts_compress_multiplier + h->frames.i_first_pts * h->i_dts_compress_multiplier;
2505 h->fdec->i_dts = prev_reordered_pts[ (h->i_frame - h->frames.i_bframe_delay) % h->frames.i_bframe_delay ];
2506 prev_reordered_pts[ h->i_frame % h->frames.i_bframe_delay ] = h->fenc->i_reordered_pts * h->i_dts_compress_multiplier;
2509 h->fdec->i_dts = h->fenc->i_reordered_pts;
2510 if( h->fenc->i_type == X264_TYPE_IDR )
2511 h->i_last_idr_pts = h->fdec->i_pts;
2513 /* ------------------- Init ----------------------------- */
2514 /* build ref list 0/1 */
2515 x264_reference_build_list( h, h->fdec->i_poc );
2517 /* ---------------------- Write the bitstream -------------------------- */
2518 /* Init bitstream context */
2519 if( h->param.b_sliced_threads )
2521 for( int i = 0; i < h->param.i_threads; i++ )
2523 bs_init( &h->thread[i]->out.bs, h->thread[i]->out.p_bitstream, h->thread[i]->out.i_bitstream );
2524 h->thread[i]->out.i_nal = 0;
2529 bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
2533 if( h->param.b_aud )
2537 if( h->sh.i_type == SLICE_TYPE_I )
2539 else if( h->sh.i_type == SLICE_TYPE_P )
2541 else if( h->sh.i_type == SLICE_TYPE_B )
2546 x264_nal_start( h, NAL_AUD, NAL_PRIORITY_DISPOSABLE );
2547 bs_write( &h->out.bs, 3, pic_type );
2548 bs_rbsp_trailing( &h->out.bs );
2549 if( x264_nal_end( h ) )
2551 overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2554 h->i_nal_type = i_nal_type;
2555 h->i_nal_ref_idc = i_nal_ref_idc;
2557 if( h->param.b_intra_refresh )
2559 if( IS_X264_TYPE_I( h->fenc->i_type ) )
2561 h->fdec->i_frames_since_pir = 0;
2562 h->b_queued_intra_refresh = 0;
2563 /* PIR is currently only supported with ref == 1, so any intra frame effectively refreshes
2564 * the whole frame and counts as an intra refresh. */
2565 h->fdec->f_pir_position = h->mb.i_mb_width;
2567 else if( h->fenc->i_type == X264_TYPE_P )
2569 int pocdiff = (h->fdec->i_poc - h->fref0[0]->i_poc)/2;
2570 float increment = X264_MAX( ((float)h->mb.i_mb_width-1) / h->param.i_keyint_max, 1 );
2571 h->fdec->f_pir_position = h->fref0[0]->f_pir_position;
2572 h->fdec->i_frames_since_pir = h->fref0[0]->i_frames_since_pir + pocdiff;
2573 if( h->fdec->i_frames_since_pir >= h->param.i_keyint_max ||
2574 (h->b_queued_intra_refresh && h->fdec->f_pir_position + 0.5 >= h->mb.i_mb_width) )
2576 h->fdec->f_pir_position = 0;
2577 h->fdec->i_frames_since_pir = 0;
2578 h->b_queued_intra_refresh = 0;
2579 h->fenc->b_keyframe = 1;
2581 h->fdec->i_pir_start_col = h->fdec->f_pir_position+0.5;
2582 h->fdec->f_pir_position += increment * pocdiff;
2583 h->fdec->i_pir_end_col = h->fdec->f_pir_position+0.5;
2584 /* If our intra refresh has reached the right side of the frame, we're done. */
2585 if( h->fdec->i_pir_end_col >= h->mb.i_mb_width - 1 )
2586 h->fdec->f_pir_position = h->mb.i_mb_width;
2590 if( h->fenc->b_keyframe )
2592 /* Write SPS and PPS */
2593 if( h->param.b_repeat_headers )
2595 /* generate sequence parameters */
2596 x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
2597 x264_sps_write( &h->out.bs, h->sps );
2598 if( x264_nal_end( h ) )
2600 overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
2602 /* generate picture parameters */
2603 x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
2604 x264_pps_write( &h->out.bs, h->pps );
2605 if( x264_nal_end( h ) )
2607 overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
2610 /* buffering period sei is written in x264_encoder_frame_end */
2613 /* write extra sei */
2614 for( int i = 0; i < h->fenc->extra_sei.num_payloads; i++ )
2616 x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2617 x264_sei_write( &h->out.bs, h->fenc->extra_sei.payloads[i].payload, h->fenc->extra_sei.payloads[i].payload_size,
2618 h->fenc->extra_sei.payloads[i].payload_type );
2619 if( x264_nal_end( h ) )
2621 overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2622 if( h->fenc->extra_sei.sei_free && h->fenc->extra_sei.payloads[i].payload )
2623 h->fenc->extra_sei.sei_free( h->fenc->extra_sei.payloads[i].payload );
2626 if( h->fenc->extra_sei.sei_free && h->fenc->extra_sei.payloads )
2627 h->fenc->extra_sei.sei_free( h->fenc->extra_sei.payloads );
2629 if( h->fenc->b_keyframe )
2631 if( h->param.b_repeat_headers && h->fenc->i_frame == 0 )
2633 /* identify ourself */
2634 x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2635 if( x264_sei_version_write( h, &h->out.bs ) )
2637 if( x264_nal_end( h ) )
2639 overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2642 if( h->fenc->i_type != X264_TYPE_IDR )
2644 int time_to_recovery = h->param.i_open_gop ? 0 : X264_MIN( h->mb.i_mb_width - 1, h->param.i_keyint_max ) + h->param.i_bframe - 1;
2645 x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2646 x264_sei_recovery_point_write( h, &h->out.bs, time_to_recovery );
2648 overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2652 /* generate sei pic timing */
2653 if( h->sps->vui.b_pic_struct_present || h->sps->vui.b_nal_hrd_parameters_present )
2655 x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2656 x264_sei_pic_timing_write( h, &h->out.bs );
2657 if( x264_nal_end( h ) )
2659 overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2662 /* Init the rate control */
2663 /* FIXME: Include slice header bit cost. */
2664 x264_ratecontrol_start( h, h->fenc->i_qpplus1, overhead*8 );
2665 i_global_qp = x264_ratecontrol_qp( h );
2667 pic_out->i_qpplus1 =
2668 h->fdec->i_qpplus1 = i_global_qp + 1;
2670 if( h->param.rc.b_stat_read && h->sh.i_type != SLICE_TYPE_I )
2672 x264_reference_build_list_optimal( h );
2673 x264_reference_check_reorder( h );
2677 h->fdec->i_poc_l0ref0 = h->fref0[0]->i_poc;
2679 if( h->sh.i_type == SLICE_TYPE_B )
2680 x264_macroblock_bipred_init( h );
2682 /*------------------------- Weights -------------------------------------*/
2683 x264_weighted_pred_init( h );
2685 /* ------------------------ Create slice header ----------------------- */
2686 x264_slice_init( h, i_nal_type, i_global_qp );
2688 if( i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE )
2692 h->i_threadslice_start = 0;
2693 h->i_threadslice_end = h->mb.i_mb_height;
2694 if( h->i_thread_frames > 1 )
2696 x264_threadpool_run( h->threadpool, (void*)x264_slices_write, h );
2697 h->b_thread_active = 1;
2699 else if( h->param.b_sliced_threads )
2701 if( x264_threaded_slices_write( h ) )
2705 if( (intptr_t)x264_slices_write( h ) )
2708 return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
2711 static int x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
2712 x264_nal_t **pp_nal, int *pi_nal,
2713 x264_picture_t *pic_out )
2715 char psz_message[80];
2717 if( h->b_thread_active )
2719 h->b_thread_active = 0;
2720 if( (intptr_t)x264_threadpool_wait( h->threadpool, h ) )
2725 pic_out->i_type = X264_TYPE_AUTO;
2730 /* generate sei buffering period and insert it into place */
2731 if( h->fenc->b_keyframe && h->sps->vui.b_nal_hrd_parameters_present )
2733 x264_hrd_fullness( h );
2734 x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2735 x264_sei_buffering_period_write( h, &h->out.bs );
2736 if( x264_nal_end( h ) )
2738 /* buffering period sei must follow AUD, SPS and PPS and precede all other SEIs */
2740 while( h->out.nal[idx].i_type == NAL_AUD ||
2741 h->out.nal[idx].i_type == NAL_SPS ||
2742 h->out.nal[idx].i_type == NAL_PPS )
2744 x264_nal_t nal_tmp = h->out.nal[h->out.i_nal-1];
2745 memmove( &h->out.nal[idx+1], &h->out.nal[idx], (h->out.i_nal-idx-1)*sizeof(x264_nal_t) );
2746 h->out.nal[idx] = nal_tmp;
2749 int frame_size = x264_encoder_encapsulate_nals( h, 0 );
2751 /* Set output picture properties */
2752 pic_out->i_type = h->fenc->i_type;
2754 pic_out->b_keyframe = h->fenc->b_keyframe;
2756 pic_out->i_pts = h->fdec->i_pts;
2757 pic_out->i_dts = h->fdec->i_dts;
2759 if( pic_out->i_pts < pic_out->i_dts )
2760 x264_log( h, X264_LOG_WARNING, "invalid DTS: PTS is less than DTS\n" );
2762 pic_out->img.i_csp = X264_CSP_NV12;
2763 pic_out->img.i_plane = h->fdec->i_plane;
2764 for( int i = 0; i < 2; i++ )
2766 pic_out->img.i_stride[i] = h->fdec->i_stride[i];
2767 // FIXME This breaks the API when pixel != uint8_t.
2768 pic_out->img.plane[i] = h->fdec->plane[i];
2771 x264_frame_push_unused( thread_current, h->fenc );
2773 /* ---------------------- Update encoder state ------------------------- */
2777 if( x264_ratecontrol_end( h, frame_size * 8, &filler ) < 0 )
2780 pic_out->hrd_timing = h->fenc->hrd_timing;
2785 overhead = (FILLER_OVERHEAD - h->param.b_annexb);
2786 if( h->param.i_slice_max_size && filler > h->param.i_slice_max_size )
2788 int next_size = filler - h->param.i_slice_max_size;
2789 int overflow = X264_MAX( overhead - next_size, 0 );
2790 f = h->param.i_slice_max_size - overhead - overflow;
2793 f = X264_MAX( 0, filler - overhead );
2795 x264_nal_start( h, NAL_FILLER, NAL_PRIORITY_DISPOSABLE );
2796 x264_filler_write( h, &h->out.bs, f );
2797 if( x264_nal_end( h ) )
2799 int total_size = x264_encoder_encapsulate_nals( h, h->out.i_nal-1 );
2800 frame_size += total_size;
2801 filler -= total_size;
2804 /* End bitstream, set output */
2805 *pi_nal = h->out.i_nal;
2806 *pp_nal = h->out.nal;
2810 x264_noise_reduction_update( h );
2812 /* ---------------------- Compute/Print statistics --------------------- */
2813 x264_thread_sync_stat( h, h->thread[0] );
2816 h->stat.i_frame_count[h->sh.i_type]++;
2817 h->stat.i_frame_size[h->sh.i_type] += frame_size;
2818 h->stat.f_frame_qp[h->sh.i_type] += h->fdec->f_qp_avg_aq;
2820 for( int i = 0; i < X264_MBTYPE_MAX; i++ )
2821 h->stat.i_mb_count[h->sh.i_type][i] += h->stat.frame.i_mb_count[i];
2822 for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
2823 h->stat.i_mb_partition[h->sh.i_type][i] += h->stat.frame.i_mb_partition[i];
2824 for( int i = 0; i < 2; i++ )
2825 h->stat.i_mb_count_8x8dct[i] += h->stat.frame.i_mb_count_8x8dct[i];
2826 for( int i = 0; i < 6; i++ )
2827 h->stat.i_mb_cbp[i] += h->stat.frame.i_mb_cbp[i];
2828 for( int i = 0; i < 4; i++ )
2829 for( int j = 0; j < 13; j++ )
2830 h->stat.i_mb_pred_mode[i][j] += h->stat.frame.i_mb_pred_mode[i][j];
2831 if( h->sh.i_type != SLICE_TYPE_I )
2832 for( int i_list = 0; i_list < 2; i_list++ )
2833 for( int i = 0; i < X264_REF_MAX*2; i++ )
2834 h->stat.i_mb_count_ref[h->sh.i_type][i_list][i] += h->stat.frame.i_mb_count_ref[i_list][i];
2835 if( h->sh.i_type == SLICE_TYPE_P )
2837 h->stat.i_consecutive_bframes[h->fdec->i_frame - h->fref0[0]->i_frame - 1]++;
2838 if( h->param.analyse.i_weighted_pred == X264_WEIGHTP_SMART )
2839 for( int i = 0; i < 3; i++ )
2840 for( int j = 0; j < h->i_ref0; j++ )
2841 if( h->sh.weight[0][i].i_denom != 0 )
2843 h->stat.i_wpred[i]++;
2847 if( h->sh.i_type == SLICE_TYPE_B )
2849 h->stat.i_direct_frames[ h->sh.b_direct_spatial_mv_pred ] ++;
2850 if( h->mb.b_direct_auto_write )
2852 //FIXME somewhat arbitrary time constants
2853 if( h->stat.i_direct_score[0] + h->stat.i_direct_score[1] > h->mb.i_mb_count )
2854 for( int i = 0; i < 2; i++ )
2855 h->stat.i_direct_score[i] = h->stat.i_direct_score[i] * 9/10;
2856 for( int i = 0; i < 2; i++ )
2857 h->stat.i_direct_score[i] += h->stat.frame.i_direct_score[i];
2861 psz_message[0] = '\0';
2862 if( h->param.analyse.b_psnr )
2865 h->stat.frame.i_ssd[0],
2866 h->stat.frame.i_ssd[1],
2867 h->stat.frame.i_ssd[2],
2870 h->stat.i_ssd_global[h->sh.i_type] += ssd[0] + ssd[1] + ssd[2];
2871 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 );
2872 h->stat.f_psnr_mean_y[h->sh.i_type] += x264_psnr( ssd[0], h->param.i_width * h->param.i_height );
2873 h->stat.f_psnr_mean_u[h->sh.i_type] += x264_psnr( ssd[1], h->param.i_width * h->param.i_height / 4 );
2874 h->stat.f_psnr_mean_v[h->sh.i_type] += x264_psnr( ssd[2], h->param.i_width * h->param.i_height / 4 );
2876 snprintf( psz_message, 80, " PSNR Y:%5.2f U:%5.2f V:%5.2f",
2877 x264_psnr( ssd[0], h->param.i_width * h->param.i_height ),
2878 x264_psnr( ssd[1], h->param.i_width * h->param.i_height / 4),
2879 x264_psnr( ssd[2], h->param.i_width * h->param.i_height / 4) );
2882 if( h->param.analyse.b_ssim )
2884 double ssim_y = h->stat.frame.f_ssim
2885 / (((h->param.i_width-6)>>2) * ((h->param.i_height-6)>>2));
2886 h->stat.f_ssim_mean_y[h->sh.i_type] += ssim_y;
2887 snprintf( psz_message + strlen(psz_message), 80 - strlen(psz_message),
2888 " SSIM Y:%.5f", ssim_y );
2890 psz_message[79] = '\0';
2892 x264_log( h, X264_LOG_DEBUG,
2893 "frame=%4d QP=%.2f NAL=%d Slice:%c Poc:%-3d I:%-4d P:%-4d SKIP:%-4d size=%d bytes%s\n",
2895 h->fdec->f_qp_avg_aq,
2897 h->sh.i_type == SLICE_TYPE_I ? 'I' : (h->sh.i_type == SLICE_TYPE_P ? 'P' : 'B' ),
2899 h->stat.frame.i_mb_count_i,
2900 h->stat.frame.i_mb_count_p,
2901 h->stat.frame.i_mb_count_skip,
2905 // keep stats all in one place
2906 x264_thread_sync_stat( h->thread[0], h );
2907 // for the use of the next frame
2908 x264_thread_sync_stat( thread_current, h );
2910 #ifdef DEBUG_MB_TYPE
2912 static const char mb_chars[] = { 'i', 'i', 'I', 'C', 'P', '8', 'S',
2913 'D', '<', 'X', 'B', 'X', '>', 'B', 'B', 'B', 'B', '8', 'S' };
2914 for( int mb_xy = 0; mb_xy < h->mb.i_mb_width * h->mb.i_mb_height; mb_xy++ )
2916 if( h->mb.type[mb_xy] < X264_MBTYPE_MAX && h->mb.type[mb_xy] >= 0 )
2917 fprintf( stderr, "%c ", mb_chars[ h->mb.type[mb_xy] ] );
2919 fprintf( stderr, "? " );
2921 if( (mb_xy+1) % h->mb.i_mb_width == 0 )
2922 fprintf( stderr, "\n" );
2927 /* Remove duplicates, must be done near the end as breaks h->fref0 array
2928 * by freeing some of its pointers. */
2929 for( int i = 0; i < h->i_ref0; i++ )
2930 if( h->fref0[i] && h->fref0[i]->b_duplicate )
2932 x264_frame_push_blank_unused( h, h->fref0[i] );
2936 if( h->param.psz_dump_yuv )
2937 x264_frame_dump( h );
2942 static void x264_print_intra( int64_t *i_mb_count, double i_count, int b_print_pcm, char *intra )
2944 intra += sprintf( intra, "I16..4%s: %4.1f%% %4.1f%% %4.1f%%",
2945 b_print_pcm ? "..PCM" : "",
2946 i_mb_count[I_16x16]/ i_count,
2947 i_mb_count[I_8x8] / i_count,
2948 i_mb_count[I_4x4] / i_count );
2950 sprintf( intra, " %4.1f%%", i_mb_count[I_PCM] / i_count );
2953 /****************************************************************************
2954 * x264_encoder_close:
2955 ****************************************************************************/
2956 void x264_encoder_close ( x264_t *h )
2958 int64_t i_yuv_size = 3 * h->param.i_width * h->param.i_height / 2;
2959 int64_t i_mb_count_size[2][7] = {{0}};
2961 int b_print_pcm = h->stat.i_mb_count[SLICE_TYPE_I][I_PCM]
2962 || h->stat.i_mb_count[SLICE_TYPE_P][I_PCM]
2963 || h->stat.i_mb_count[SLICE_TYPE_B][I_PCM];
2965 x264_lookahead_delete( h );
2967 if( h->param.i_threads > 1 )
2968 x264_threadpool_delete( h->threadpool );
2969 if( h->i_thread_frames > 1 )
2971 for( int i = 0; i < h->i_thread_frames; i++ )
2972 if( h->thread[i]->b_thread_active )
2974 assert( h->thread[i]->fenc->i_reference_count == 1 );
2975 x264_frame_delete( h->thread[i]->fenc );
2978 x264_t *thread_prev = h->thread[h->i_thread_phase];
2979 x264_thread_sync_ratecontrol( h, thread_prev, h );
2980 x264_thread_sync_ratecontrol( thread_prev, thread_prev, h );
2981 h->i_frame = thread_prev->i_frame + 1 - h->i_thread_frames;
2985 /* Slices used and PSNR */
2986 for( int i = 0; i < 3; i++ )
2988 static const uint8_t slice_order[] = { SLICE_TYPE_I, SLICE_TYPE_P, SLICE_TYPE_B };
2989 int i_slice = slice_order[i];
2991 if( h->stat.i_frame_count[i_slice] > 0 )
2993 const int i_count = h->stat.i_frame_count[i_slice];
2994 if( h->param.analyse.b_psnr )
2996 x264_log( h, X264_LOG_INFO,
2997 "frame %c:%-5d Avg QP:%5.2f size:%6.0f PSNR Mean Y:%5.2f U:%5.2f V:%5.2f Avg:%5.2f Global:%5.2f\n",
2998 slice_type_to_char[i_slice],
3000 h->stat.f_frame_qp[i_slice] / i_count,
3001 (double)h->stat.i_frame_size[i_slice] / i_count,
3002 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,
3003 h->stat.f_psnr_average[i_slice] / i_count,
3004 x264_psnr( h->stat.i_ssd_global[i_slice], i_count * i_yuv_size ) );
3008 x264_log( h, X264_LOG_INFO,
3009 "frame %c:%-5d Avg QP:%5.2f size:%6.0f\n",
3010 slice_type_to_char[i_slice],
3012 h->stat.f_frame_qp[i_slice] / i_count,
3013 (double)h->stat.i_frame_size[i_slice] / i_count );
3017 if( h->param.i_bframe && h->stat.i_frame_count[SLICE_TYPE_P] )
3021 // weight by number of frames (including the P-frame) that are in a sequence of N B-frames
3022 for( int i = 0; i <= h->param.i_bframe; i++ )
3023 den += (i+1) * h->stat.i_consecutive_bframes[i];
3024 for( int i = 0; i <= h->param.i_bframe; i++ )
3025 p += sprintf( p, " %4.1f%%", 100. * (i+1) * h->stat.i_consecutive_bframes[i] / den );
3026 x264_log( h, X264_LOG_INFO, "consecutive B-frames:%s\n", buf );
3029 for( int i_type = 0; i_type < 2; i_type++ )
3030 for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
3032 if( i == D_DIRECT_8x8 ) continue; /* direct is counted as its own type */
3033 i_mb_count_size[i_type][x264_mb_partition_pixel_table[i]] += h->stat.i_mb_partition[i_type][i];
3037 if( h->stat.i_frame_count[SLICE_TYPE_I] > 0 )
3039 int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_I];
3040 double i_count = h->stat.i_frame_count[SLICE_TYPE_I] * h->mb.i_mb_count / 100.0;
3041 x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
3042 x264_log( h, X264_LOG_INFO, "mb I %s\n", buf );
3044 if( h->stat.i_frame_count[SLICE_TYPE_P] > 0 )
3046 int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_P];
3047 double i_count = h->stat.i_frame_count[SLICE_TYPE_P] * h->mb.i_mb_count / 100.0;
3048 int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_P];
3049 x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
3050 x264_log( h, X264_LOG_INFO,
3051 "mb P %s P16..4: %4.1f%% %4.1f%% %4.1f%% %4.1f%% %4.1f%% skip:%4.1f%%\n",
3053 i_mb_size[PIXEL_16x16] / (i_count*4),
3054 (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
3055 i_mb_size[PIXEL_8x8] / (i_count*4),
3056 (i_mb_size[PIXEL_8x4] + i_mb_size[PIXEL_4x8]) / (i_count*4),
3057 i_mb_size[PIXEL_4x4] / (i_count*4),
3058 i_mb_count[P_SKIP] / i_count );
3060 if( h->stat.i_frame_count[SLICE_TYPE_B] > 0 )
3062 int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_B];
3063 double i_count = h->stat.i_frame_count[SLICE_TYPE_B] * h->mb.i_mb_count / 100.0;
3064 double i_mb_list_count;
3065 int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_B];
3066 int64_t list_count[3] = {0}; /* 0 == L0, 1 == L1, 2 == BI */
3067 x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
3068 for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
3069 for( int j = 0; j < 2; j++ )
3071 int l0 = x264_mb_type_list_table[i][0][j];
3072 int l1 = x264_mb_type_list_table[i][1][j];
3074 list_count[l1+l0*l1] += h->stat.i_mb_count[SLICE_TYPE_B][i] * 2;
3076 list_count[0] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L0_8x8];
3077 list_count[1] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L1_8x8];
3078 list_count[2] += h->stat.i_mb_partition[SLICE_TYPE_B][D_BI_8x8];
3079 i_mb_count[B_DIRECT] += (h->stat.i_mb_partition[SLICE_TYPE_B][D_DIRECT_8x8]+2)/4;
3080 i_mb_list_count = (list_count[0] + list_count[1] + list_count[2]) / 100.0;
3081 x264_log( h, X264_LOG_INFO,
3082 "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",
3084 i_mb_size[PIXEL_16x16] / (i_count*4),
3085 (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
3086 i_mb_size[PIXEL_8x8] / (i_count*4),
3087 i_mb_count[B_DIRECT] / i_count,
3088 i_mb_count[B_SKIP] / i_count,
3089 list_count[0] / i_mb_list_count,
3090 list_count[1] / i_mb_list_count,
3091 list_count[2] / i_mb_list_count );
3094 x264_ratecontrol_summary( h );
3096 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 )
3098 #define SUM3(p) (p[SLICE_TYPE_I] + p[SLICE_TYPE_P] + p[SLICE_TYPE_B])
3099 #define SUM3b(p,o) (p[SLICE_TYPE_I][o] + p[SLICE_TYPE_P][o] + p[SLICE_TYPE_B][o])
3100 int64_t i_i8x8 = SUM3b( h->stat.i_mb_count, I_8x8 );
3101 int64_t i_intra = i_i8x8 + SUM3b( h->stat.i_mb_count, I_4x4 )
3102 + SUM3b( h->stat.i_mb_count, I_16x16 );
3103 int64_t i_all_intra = i_intra + SUM3b( h->stat.i_mb_count, I_PCM);
3104 const int i_count = h->stat.i_frame_count[SLICE_TYPE_I] +
3105 h->stat.i_frame_count[SLICE_TYPE_P] +
3106 h->stat.i_frame_count[SLICE_TYPE_B];
3107 int64_t i_mb_count = (int64_t)i_count * h->mb.i_mb_count;
3108 float fps = (float) h->param.i_fps_num / h->param.i_fps_den;
3110 /* duration algorithm fails with one frame */
3111 if( !h->param.b_vfr_input || i_count == 1 )
3112 f_bitrate = fps * SUM3(h->stat.i_frame_size) / i_count / 125;
3115 float duration = (float)(2 * h->frames.i_largest_pts - h->frames.i_second_largest_pts - h->frames.i_first_pts)
3116 * h->i_dts_compress_multiplier * h->param.i_timebase_num / h->param.i_timebase_den;
3117 f_bitrate = SUM3(h->stat.i_frame_size) / duration / 125;
3120 if( h->pps->b_transform_8x8_mode )
3123 if( h->stat.i_mb_count_8x8dct[0] )
3124 sprintf( buf, " inter:%.1f%%", 100. * h->stat.i_mb_count_8x8dct[1] / h->stat.i_mb_count_8x8dct[0] );
3125 x264_log( h, X264_LOG_INFO, "8x8 transform intra:%.1f%%%s\n", 100. * i_i8x8 / i_intra, buf );
3128 if( (h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO ||
3129 (h->stat.i_direct_frames[0] && h->stat.i_direct_frames[1]))
3130 && h->stat.i_frame_count[SLICE_TYPE_B] )
3132 x264_log( h, X264_LOG_INFO, "direct mvs spatial:%.1f%% temporal:%.1f%%\n",
3133 h->stat.i_direct_frames[1] * 100. / h->stat.i_frame_count[SLICE_TYPE_B],
3134 h->stat.i_direct_frames[0] * 100. / h->stat.i_frame_count[SLICE_TYPE_B] );
3138 if( i_mb_count != i_all_intra )
3139 sprintf( buf, " inter: %.1f%% %.1f%% %.1f%%",
3140 h->stat.i_mb_cbp[1] * 100.0 / ((i_mb_count - i_all_intra)*4),
3141 h->stat.i_mb_cbp[3] * 100.0 / ((i_mb_count - i_all_intra) ),
3142 h->stat.i_mb_cbp[5] * 100.0 / ((i_mb_count - i_all_intra)) );
3143 x264_log( h, X264_LOG_INFO, "coded y,uvDC,uvAC intra: %.1f%% %.1f%% %.1f%%%s\n",
3144 h->stat.i_mb_cbp[0] * 100.0 / (i_all_intra*4),
3145 h->stat.i_mb_cbp[2] * 100.0 / (i_all_intra ),
3146 h->stat.i_mb_cbp[4] * 100.0 / (i_all_intra ), buf );
3148 int64_t fixed_pred_modes[4][9] = {{0}};
3149 int64_t sum_pred_modes[4] = {0};
3150 for( int i = 0; i <= I_PRED_16x16_DC_128; i++ )
3152 fixed_pred_modes[0][x264_mb_pred_mode16x16_fix[i]] += h->stat.i_mb_pred_mode[0][i];
3153 sum_pred_modes[0] += h->stat.i_mb_pred_mode[0][i];
3155 if( sum_pred_modes[0] )
3156 x264_log( h, X264_LOG_INFO, "i16 v,h,dc,p: %2.0f%% %2.0f%% %2.0f%% %2.0f%%\n",
3157 fixed_pred_modes[0][0] * 100.0 / sum_pred_modes[0],
3158 fixed_pred_modes[0][1] * 100.0 / sum_pred_modes[0],
3159 fixed_pred_modes[0][2] * 100.0 / sum_pred_modes[0],
3160 fixed_pred_modes[0][3] * 100.0 / sum_pred_modes[0] );
3161 for( int i = 1; i <= 2; i++ )
3163 for( int j = 0; j <= I_PRED_8x8_DC_128; j++ )
3165 fixed_pred_modes[i][x264_mb_pred_mode4x4_fix(j)] += h->stat.i_mb_pred_mode[i][j];
3166 sum_pred_modes[i] += h->stat.i_mb_pred_mode[i][j];
3168 if( sum_pred_modes[i] )
3169 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,
3170 fixed_pred_modes[i][0] * 100.0 / sum_pred_modes[i],
3171 fixed_pred_modes[i][1] * 100.0 / sum_pred_modes[i],
3172 fixed_pred_modes[i][2] * 100.0 / sum_pred_modes[i],
3173 fixed_pred_modes[i][3] * 100.0 / sum_pred_modes[i],
3174 fixed_pred_modes[i][4] * 100.0 / sum_pred_modes[i],
3175 fixed_pred_modes[i][5] * 100.0 / sum_pred_modes[i],
3176 fixed_pred_modes[i][6] * 100.0 / sum_pred_modes[i],
3177 fixed_pred_modes[i][7] * 100.0 / sum_pred_modes[i],
3178 fixed_pred_modes[i][8] * 100.0 / sum_pred_modes[i] );
3180 for( int i = 0; i <= I_PRED_CHROMA_DC_128; i++ )
3182 fixed_pred_modes[3][x264_mb_pred_mode8x8c_fix[i]] += h->stat.i_mb_pred_mode[3][i];
3183 sum_pred_modes[3] += h->stat.i_mb_pred_mode[3][i];
3185 if( sum_pred_modes[3] )
3186 x264_log( h, X264_LOG_INFO, "i8c dc,h,v,p: %2.0f%% %2.0f%% %2.0f%% %2.0f%%\n",
3187 fixed_pred_modes[3][0] * 100.0 / sum_pred_modes[3],
3188 fixed_pred_modes[3][1] * 100.0 / sum_pred_modes[3],
3189 fixed_pred_modes[3][2] * 100.0 / sum_pred_modes[3],
3190 fixed_pred_modes[3][3] * 100.0 / sum_pred_modes[3] );
3192 if( h->param.analyse.i_weighted_pred == X264_WEIGHTP_SMART && h->stat.i_frame_count[SLICE_TYPE_P] > 0 )
3193 x264_log( h, X264_LOG_INFO, "Weighted P-Frames: Y:%.1f%%\n",
3194 h->stat.i_wpred[0] * 100.0 / h->stat.i_frame_count[SLICE_TYPE_P] );
3196 for( int i_list = 0; i_list < 2; i_list++ )
3197 for( int i_slice = 0; i_slice < 2; i_slice++ )
3202 for( int i = 0; i < X264_REF_MAX*2; i++ )
3203 if( h->stat.i_mb_count_ref[i_slice][i_list][i] )
3205 i_den += h->stat.i_mb_count_ref[i_slice][i_list][i];
3210 for( int i = 0; i <= i_max; i++ )
3211 p += sprintf( p, " %4.1f%%", 100. * h->stat.i_mb_count_ref[i_slice][i_list][i] / i_den );
3212 x264_log( h, X264_LOG_INFO, "ref %c L%d:%s\n", "PB"[i_slice], i_list, buf );
3215 if( h->param.analyse.b_ssim )
3217 float ssim = SUM3( h->stat.f_ssim_mean_y ) / i_count;
3218 x264_log( h, X264_LOG_INFO, "SSIM Mean Y:%.7f (%6.3fdb)\n", ssim, x264_ssim( ssim ) );
3220 if( h->param.analyse.b_psnr )
3222 x264_log( h, X264_LOG_INFO,
3223 "PSNR Mean Y:%6.3f U:%6.3f V:%6.3f Avg:%6.3f Global:%6.3f kb/s:%.2f\n",
3224 SUM3( h->stat.f_psnr_mean_y ) / i_count,
3225 SUM3( h->stat.f_psnr_mean_u ) / i_count,
3226 SUM3( h->stat.f_psnr_mean_v ) / i_count,
3227 SUM3( h->stat.f_psnr_average ) / i_count,
3228 x264_psnr( SUM3( h->stat.i_ssd_global ), i_count * i_yuv_size ),
3232 x264_log( h, X264_LOG_INFO, "kb/s:%.2f\n", f_bitrate );
3236 x264_ratecontrol_delete( h );
3239 if( h->param.rc.psz_stat_out )
3240 free( h->param.rc.psz_stat_out );
3241 if( h->param.rc.psz_stat_in )
3242 free( h->param.rc.psz_stat_in );
3244 x264_cqm_delete( h );
3245 x264_free( h->nal_buffer );
3246 x264_analyse_free_costs( h );
3248 if( h->i_thread_frames > 1)
3249 h = h->thread[h->i_thread_phase];
3252 x264_frame_delete_list( h->frames.unused[0] );
3253 x264_frame_delete_list( h->frames.unused[1] );
3254 x264_frame_delete_list( h->frames.current );
3255 x264_frame_delete_list( h->frames.blank_unused );
3259 for( int i = h->param.i_threads - 1; i >= 0; i-- )
3261 x264_frame_t **frame;
3263 if( !h->param.b_sliced_threads || i == 0 )
3265 for( frame = h->thread[i]->frames.reference; *frame; frame++ )
3267 assert( (*frame)->i_reference_count > 0 );
3268 (*frame)->i_reference_count--;
3269 if( (*frame)->i_reference_count == 0 )
3270 x264_frame_delete( *frame );
3272 frame = &h->thread[i]->fdec;
3273 assert( (*frame)->i_reference_count > 0 );
3274 (*frame)->i_reference_count--;
3275 if( (*frame)->i_reference_count == 0 )
3276 x264_frame_delete( *frame );
3277 x264_macroblock_cache_free( h->thread[i] );
3279 x264_macroblock_thread_free( h->thread[i], 0 );
3280 x264_free( h->thread[i]->out.p_bitstream );
3281 x264_free( h->thread[i]->out.nal);
3282 x264_free( h->thread[i] );
3286 /****************************************************************************
3287 * x264_encoder_delayed_frames:
3288 ****************************************************************************/
3289 int x264_encoder_delayed_frames( x264_t *h )
3291 int delayed_frames = 0;
3292 if( h->i_thread_frames > 1 )
3294 for( int i = 0; i < h->i_thread_frames; i++ )
3295 delayed_frames += h->thread[i]->b_thread_active;
3296 h = h->thread[h->i_thread_phase];
3298 for( int i = 0; h->frames.current[i]; i++ )
3300 x264_pthread_mutex_lock( &h->lookahead->ofbuf.mutex );
3301 x264_pthread_mutex_lock( &h->lookahead->ifbuf.mutex );
3302 x264_pthread_mutex_lock( &h->lookahead->next.mutex );
3303 delayed_frames += h->lookahead->ifbuf.i_size + h->lookahead->next.i_size + h->lookahead->ofbuf.i_size;
3304 x264_pthread_mutex_unlock( &h->lookahead->next.mutex );
3305 x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
3306 x264_pthread_mutex_unlock( &h->lookahead->ofbuf.mutex );
3307 return delayed_frames;