1 /*****************************************************************************
2 * encoder.c: top-level encoder functions
3 *****************************************************************************
4 * Copyright (C) 2003-2012 x264 project
6 * Authors: Laurent Aimar <fenrir@via.ecp.fr>
7 * Loren Merritt <lorenm@u.washington.edu>
8 * Fiona Glaser <fiona@x264.com>
10 * This program is free software; you can redistribute it and/or modify
11 * it under the terms of the GNU General Public License as published by
12 * the Free Software Foundation; either version 2 of the License, or
13 * (at your option) any later version.
15 * This program is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
18 * GNU General Public License for more details.
20 * You should have received a copy of the GNU General Public License
21 * along with this program; if not, write to the Free Software
22 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02111, USA.
24 * This program is also available under a commercial proprietary license.
25 * For more information, contact us at licensing@x264.com.
26 *****************************************************************************/
28 #include "common/common.h"
32 #include "ratecontrol.h"
33 #include "macroblock.h"
37 #include "common/visualize.h"
40 //#define DEBUG_MB_TYPE
42 #define bs_write_ue bs_write_ue_big
44 static int x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
45 x264_nal_t **pp_nal, int *pi_nal,
46 x264_picture_t *pic_out );
48 /****************************************************************************
50 ******************************* x264 libs **********************************
52 ****************************************************************************/
53 static double x264_psnr( double sqe, double size )
55 double mse = sqe / (PIXEL_MAX*PIXEL_MAX * size);
56 if( mse <= 0.0000000001 ) /* Max 100dB */
59 return -10.0 * log10( mse );
62 static double x264_ssim( double ssim )
64 double inv_ssim = 1 - ssim;
65 if( inv_ssim <= 0.0000000001 ) /* Max 100dB */
68 return -10.0 * log10( inv_ssim );
71 static void x264_frame_dump( x264_t *h )
73 FILE *f = fopen( h->param.psz_dump_yuv, "r+b" );
77 /* Write the frame in display order */
78 int frame_size = FRAME_SIZE( h->param.i_height * h->param.i_width * sizeof(pixel) );
79 fseek( f, (uint64_t)h->fdec->i_frame * frame_size, SEEK_SET );
80 for( int p = 0; p < (CHROMA444 ? 3 : 1); p++ )
81 for( int y = 0; y < h->param.i_height; y++ )
82 fwrite( &h->fdec->plane[p][y*h->fdec->i_stride[p]], sizeof(pixel), h->param.i_width, f );
85 int cw = h->param.i_width>>1;
86 int ch = h->param.i_height>>CHROMA_V_SHIFT;
87 pixel *planeu = x264_malloc( (cw*ch*2+32)*sizeof(pixel) );
88 pixel *planev = planeu + cw*ch + 16;
89 h->mc.plane_copy_deinterleave( planeu, cw, planev, cw, h->fdec->plane[1], h->fdec->i_stride[1], cw, ch );
90 fwrite( planeu, 1, cw*ch*sizeof(pixel), f );
91 fwrite( planev, 1, cw*ch*sizeof(pixel), f );
97 /* Fill "default" values */
98 static void x264_slice_header_init( x264_t *h, x264_slice_header_t *sh,
99 x264_sps_t *sps, x264_pps_t *pps,
100 int i_idr_pic_id, int i_frame, int i_qp )
102 x264_param_t *param = &h->param;
104 /* First we fill all fields */
109 sh->i_last_mb = h->mb.i_mb_count - 1;
110 sh->i_pps_id = pps->i_id;
112 sh->i_frame_num = i_frame;
114 sh->b_mbaff = PARAM_INTERLACED;
115 sh->b_field_pic = 0; /* no field support for now */
116 sh->b_bottom_field = 0; /* not yet used */
118 sh->i_idr_pic_id = i_idr_pic_id;
120 /* poc stuff, fixed later */
122 sh->i_delta_poc_bottom = 0;
123 sh->i_delta_poc[0] = 0;
124 sh->i_delta_poc[1] = 0;
126 sh->i_redundant_pic_cnt = 0;
128 h->mb.b_direct_auto_write = h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO
130 && ( h->param.rc.b_stat_write || !h->param.rc.b_stat_read );
132 if( !h->mb.b_direct_auto_read && sh->i_type == SLICE_TYPE_B )
134 if( h->fref[1][0]->i_poc_l0ref0 == h->fref[0][0]->i_poc )
136 if( h->mb.b_direct_auto_write )
137 sh->b_direct_spatial_mv_pred = ( h->stat.i_direct_score[1] > h->stat.i_direct_score[0] );
139 sh->b_direct_spatial_mv_pred = ( param->analyse.i_direct_mv_pred == X264_DIRECT_PRED_SPATIAL );
143 h->mb.b_direct_auto_write = 0;
144 sh->b_direct_spatial_mv_pred = 1;
147 /* else b_direct_spatial_mv_pred was read from the 2pass statsfile */
149 sh->b_num_ref_idx_override = 0;
150 sh->i_num_ref_idx_l0_active = 1;
151 sh->i_num_ref_idx_l1_active = 1;
153 sh->b_ref_pic_list_reordering[0] = h->b_ref_reorder[0];
154 sh->b_ref_pic_list_reordering[1] = h->b_ref_reorder[1];
156 /* If the ref list isn't in the default order, construct reordering header */
157 for( int list = 0; list < 2; list++ )
159 if( sh->b_ref_pic_list_reordering[list] )
161 int pred_frame_num = i_frame;
162 for( int i = 0; i < h->i_ref[list]; i++ )
164 int diff = h->fref[list][i]->i_frame_num - pred_frame_num;
165 sh->ref_pic_list_order[list][i].idc = ( diff > 0 );
166 sh->ref_pic_list_order[list][i].arg = (abs(diff) - 1) & ((1 << sps->i_log2_max_frame_num) - 1);
167 pred_frame_num = h->fref[list][i]->i_frame_num;
172 sh->i_cabac_init_idc = param->i_cabac_init_idc;
174 sh->i_qp = SPEC_QP(i_qp);
175 sh->i_qp_delta = sh->i_qp - pps->i_pic_init_qp;
176 sh->b_sp_for_swidth = 0;
179 int deblock_thresh = i_qp + 2 * X264_MIN(param->i_deblocking_filter_alphac0, param->i_deblocking_filter_beta);
180 /* If effective qp <= 15, deblocking would have no effect anyway */
181 if( param->b_deblocking_filter && (h->mb.b_variable_qp || 15 < deblock_thresh ) )
182 sh->i_disable_deblocking_filter_idc = param->b_sliced_threads ? 2 : 0;
184 sh->i_disable_deblocking_filter_idc = 1;
185 sh->i_alpha_c0_offset = param->i_deblocking_filter_alphac0 << 1;
186 sh->i_beta_offset = param->i_deblocking_filter_beta << 1;
189 static void x264_slice_header_write( bs_t *s, x264_slice_header_t *sh, int i_nal_ref_idc )
193 int first_x = sh->i_first_mb % sh->sps->i_mb_width;
194 int first_y = sh->i_first_mb / sh->sps->i_mb_width;
195 assert( (first_y&1) == 0 );
196 bs_write_ue( s, (2*first_x + sh->sps->i_mb_width*(first_y&~1) + (first_y&1)) >> 1 );
199 bs_write_ue( s, sh->i_first_mb );
201 bs_write_ue( s, sh->i_type + 5 ); /* same type things */
202 bs_write_ue( s, sh->i_pps_id );
203 bs_write( s, sh->sps->i_log2_max_frame_num, sh->i_frame_num & ((1<<sh->sps->i_log2_max_frame_num)-1) );
205 if( !sh->sps->b_frame_mbs_only )
207 bs_write1( s, sh->b_field_pic );
208 if( sh->b_field_pic )
209 bs_write1( s, sh->b_bottom_field );
212 if( sh->i_idr_pic_id >= 0 ) /* NAL IDR */
213 bs_write_ue( s, sh->i_idr_pic_id );
215 if( sh->sps->i_poc_type == 0 )
217 bs_write( s, sh->sps->i_log2_max_poc_lsb, sh->i_poc & ((1<<sh->sps->i_log2_max_poc_lsb)-1) );
218 if( sh->pps->b_pic_order && !sh->b_field_pic )
219 bs_write_se( s, sh->i_delta_poc_bottom );
222 if( sh->pps->b_redundant_pic_cnt )
223 bs_write_ue( s, sh->i_redundant_pic_cnt );
225 if( sh->i_type == SLICE_TYPE_B )
226 bs_write1( s, sh->b_direct_spatial_mv_pred );
228 if( sh->i_type == SLICE_TYPE_P || sh->i_type == SLICE_TYPE_B )
230 bs_write1( s, sh->b_num_ref_idx_override );
231 if( sh->b_num_ref_idx_override )
233 bs_write_ue( s, sh->i_num_ref_idx_l0_active - 1 );
234 if( sh->i_type == SLICE_TYPE_B )
235 bs_write_ue( s, sh->i_num_ref_idx_l1_active - 1 );
239 /* ref pic list reordering */
240 if( sh->i_type != SLICE_TYPE_I )
242 bs_write1( s, sh->b_ref_pic_list_reordering[0] );
243 if( sh->b_ref_pic_list_reordering[0] )
245 for( int i = 0; i < sh->i_num_ref_idx_l0_active; i++ )
247 bs_write_ue( s, sh->ref_pic_list_order[0][i].idc );
248 bs_write_ue( s, sh->ref_pic_list_order[0][i].arg );
253 if( sh->i_type == SLICE_TYPE_B )
255 bs_write1( s, sh->b_ref_pic_list_reordering[1] );
256 if( sh->b_ref_pic_list_reordering[1] )
258 for( int i = 0; i < sh->i_num_ref_idx_l1_active; i++ )
260 bs_write_ue( s, sh->ref_pic_list_order[1][i].idc );
261 bs_write_ue( s, sh->ref_pic_list_order[1][i].arg );
267 if( sh->pps->b_weighted_pred && sh->i_type == SLICE_TYPE_P )
269 /* pred_weight_table() */
270 bs_write_ue( s, sh->weight[0][0].i_denom );
271 bs_write_ue( s, sh->weight[0][1].i_denom );
272 for( int i = 0; i < sh->i_num_ref_idx_l0_active; i++ )
274 int luma_weight_l0_flag = !!sh->weight[i][0].weightfn;
275 int chroma_weight_l0_flag = !!sh->weight[i][1].weightfn || !!sh->weight[i][2].weightfn;
276 bs_write1( s, luma_weight_l0_flag );
277 if( luma_weight_l0_flag )
279 bs_write_se( s, sh->weight[i][0].i_scale );
280 bs_write_se( s, sh->weight[i][0].i_offset );
282 bs_write1( s, chroma_weight_l0_flag );
283 if( chroma_weight_l0_flag )
285 for( int j = 1; j < 3; j++ )
287 bs_write_se( s, sh->weight[i][j].i_scale );
288 bs_write_se( s, sh->weight[i][j].i_offset );
293 else if( sh->pps->b_weighted_bipred == 1 && sh->i_type == SLICE_TYPE_B )
298 if( i_nal_ref_idc != 0 )
300 if( sh->i_idr_pic_id >= 0 )
302 bs_write1( s, 0 ); /* no output of prior pics flag */
303 bs_write1( s, 0 ); /* long term reference flag */
307 bs_write1( s, sh->i_mmco_command_count > 0 ); /* adaptive_ref_pic_marking_mode_flag */
308 if( sh->i_mmco_command_count > 0 )
310 for( int i = 0; i < sh->i_mmco_command_count; i++ )
312 bs_write_ue( s, 1 ); /* mark short term ref as unused */
313 bs_write_ue( s, sh->mmco[i].i_difference_of_pic_nums - 1 );
315 bs_write_ue( s, 0 ); /* end command list */
320 if( sh->pps->b_cabac && sh->i_type != SLICE_TYPE_I )
321 bs_write_ue( s, sh->i_cabac_init_idc );
323 bs_write_se( s, sh->i_qp_delta ); /* slice qp delta */
325 if( sh->pps->b_deblocking_filter_control )
327 bs_write_ue( s, sh->i_disable_deblocking_filter_idc );
328 if( sh->i_disable_deblocking_filter_idc != 1 )
330 bs_write_se( s, sh->i_alpha_c0_offset >> 1 );
331 bs_write_se( s, sh->i_beta_offset >> 1 );
336 /* If we are within a reasonable distance of the end of the memory allocated for the bitstream, */
337 /* reallocate, adding an arbitrary amount of space. */
338 static int x264_bitstream_check_buffer( x264_t *h )
340 uint8_t *bs_bak = h->out.p_bitstream;
341 int max_row_size = (2500 << SLICE_MBAFF) * h->mb.i_mb_width;
342 if( (h->param.b_cabac && (h->cabac.p_end - h->cabac.p < max_row_size)) ||
343 (h->out.bs.p_end - h->out.bs.p < max_row_size) )
345 h->out.i_bitstream += max_row_size;
346 CHECKED_MALLOC( h->out.p_bitstream, h->out.i_bitstream );
347 h->mc.memcpy_aligned( h->out.p_bitstream, bs_bak, (h->out.i_bitstream - max_row_size) & ~15 );
348 intptr_t delta = h->out.p_bitstream - bs_bak;
350 h->out.bs.p_start += delta;
351 h->out.bs.p += delta;
352 h->out.bs.p_end = h->out.p_bitstream + h->out.i_bitstream;
354 h->cabac.p_start += delta;
356 h->cabac.p_end = h->out.p_bitstream + h->out.i_bitstream;
358 for( int i = 0; i <= h->out.i_nal; i++ )
359 h->out.nal[i].p_payload += delta;
369 static void x264_encoder_thread_init( x264_t *h )
371 if( h->param.i_sync_lookahead )
372 x264_lower_thread_priority( 10 );
375 /* Misalign mask has to be set separately for each thread. */
376 if( h->param.cpu&X264_CPU_SSE_MISALIGN )
377 x264_cpu_mask_misalign_sse();
382 /****************************************************************************
384 ****************************************************************************
385 ****************************** External API*********************************
386 ****************************************************************************
388 ****************************************************************************/
390 static int x264_validate_parameters( x264_t *h, int b_open )
394 if( b_open && !(x264_cpu_detect() & X264_CPU_SSE) )
396 x264_log( h, X264_LOG_ERROR, "your cpu does not support SSE1, but x264 was compiled with asm support\n");
398 if( b_open && !(x264_cpu_detect() & X264_CPU_MMX2) )
400 x264_log( h, X264_LOG_ERROR, "your cpu does not support MMXEXT, but x264 was compiled with asm support\n");
402 x264_log( h, X264_LOG_ERROR, "to run x264, recompile without asm support (configure --disable-asm)\n");
408 h->param.b_interlaced = !!PARAM_INTERLACED;
410 if( h->param.b_interlaced )
412 x264_log( h, X264_LOG_ERROR, "not compiled with interlaced support\n" );
417 if( h->param.i_width <= 0 || h->param.i_height <= 0 )
419 x264_log( h, X264_LOG_ERROR, "invalid width x height (%dx%d)\n",
420 h->param.i_width, h->param.i_height );
424 int i_csp = h->param.i_csp & X264_CSP_MASK;
425 #if X264_CHROMA_FORMAT
426 if( CHROMA_FORMAT != CHROMA_420 && i_csp >= X264_CSP_I420 && i_csp <= X264_CSP_NV12 )
428 x264_log( h, X264_LOG_ERROR, "not compiled with 4:2:0 support\n" );
431 else if( CHROMA_FORMAT != CHROMA_422 && i_csp >= X264_CSP_I422 && i_csp <= X264_CSP_NV16 )
433 x264_log( h, X264_LOG_ERROR, "not compiled with 4:2:2 support\n" );
436 else if( CHROMA_FORMAT != CHROMA_444 && i_csp >= X264_CSP_I444 && i_csp <= X264_CSP_RGB )
438 x264_log( h, X264_LOG_ERROR, "not compiled with 4:4:4 support\n" );
442 if( i_csp <= X264_CSP_NONE || i_csp >= X264_CSP_MAX )
444 x264_log( h, X264_LOG_ERROR, "invalid CSP (only I420/YV12/NV12/I422/YV16/NV16/I444/YV24/BGR/BGRA/RGB supported)\n" );
448 if( i_csp < X264_CSP_I444 && h->param.i_width % 2 )
450 x264_log( h, X264_LOG_ERROR, "width not divisible by 2 (%dx%d)\n",
451 h->param.i_width, h->param.i_height );
455 if( i_csp < X264_CSP_I422 && PARAM_INTERLACED && h->param.i_height % 4 )
457 x264_log( h, X264_LOG_ERROR, "height not divisible by 4 (%dx%d)\n",
458 h->param.i_width, h->param.i_height );
462 if( (i_csp < X264_CSP_I422 || PARAM_INTERLACED) && h->param.i_height % 2 )
464 x264_log( h, X264_LOG_ERROR, "height not divisible by 2 (%dx%d)\n",
465 h->param.i_width, h->param.i_height );
469 if( (h->param.crop_rect.i_left + h->param.crop_rect.i_right ) >= h->param.i_width ||
470 (h->param.crop_rect.i_top + h->param.crop_rect.i_bottom) >= h->param.i_height )
472 x264_log( h, X264_LOG_ERROR, "invalid crop-rect %u,%u,%u,%u\n", h->param.crop_rect.i_left,
473 h->param.crop_rect.i_top, h->param.crop_rect.i_right, h->param.crop_rect.i_bottom );
477 if( h->param.i_threads == X264_THREADS_AUTO )
478 h->param.i_threads = x264_cpu_num_processors() * (h->param.b_sliced_threads?2:3)/2;
479 if( h->param.i_threads > 1 )
482 x264_log( h, X264_LOG_WARNING, "not compiled with thread support!\n");
483 h->param.i_threads = 1;
485 /* Avoid absurdly small thread slices as they can reduce performance
486 * and VBV compliance. Capped at an arbitrary 4 rows per thread. */
487 if( h->param.b_sliced_threads )
489 int max_threads = (h->param.i_height+15)/16 / 4;
490 h->param.i_threads = X264_MIN( h->param.i_threads, max_threads );
493 h->param.i_threads = x264_clip3( h->param.i_threads, 1, X264_THREAD_MAX );
494 if( h->param.i_threads == 1 )
495 h->param.b_sliced_threads = 0;
496 h->i_thread_frames = h->param.b_sliced_threads ? 1 : h->param.i_threads;
497 if( h->i_thread_frames > 1 )
498 h->param.nalu_process = NULL;
500 h->param.i_keyint_max = x264_clip3( h->param.i_keyint_max, 1, X264_KEYINT_MAX_INFINITE );
501 if( h->param.i_keyint_max == 1 )
503 h->param.b_intra_refresh = 0;
504 h->param.analyse.i_weighted_pred = 0;
507 h->param.i_frame_packing = x264_clip3( h->param.i_frame_packing, -1, 5 );
509 /* Detect default ffmpeg settings and terminate with an error. */
513 score += h->param.analyse.i_me_range == 0;
514 score += h->param.rc.i_qp_step == 3;
515 score += h->param.i_keyint_max == 12;
516 score += h->param.rc.i_qp_min == 2;
517 score += h->param.rc.i_qp_max == 31;
518 score += h->param.rc.f_qcompress == 0.5;
519 score += fabs(h->param.rc.f_ip_factor - 1.25) < 0.01;
520 score += fabs(h->param.rc.f_pb_factor - 1.25) < 0.01;
521 score += h->param.analyse.inter == 0 && h->param.analyse.i_subpel_refine == 8;
524 x264_log( h, X264_LOG_ERROR, "broken ffmpeg default settings detected\n" );
525 x264_log( h, X264_LOG_ERROR, "use an encoding preset (e.g. -vpre medium)\n" );
526 x264_log( h, X264_LOG_ERROR, "preset usage: -vpre <speed> -vpre <profile>\n" );
527 x264_log( h, X264_LOG_ERROR, "speed presets are listed in x264 --help\n" );
528 x264_log( h, X264_LOG_ERROR, "profile is optional; x264 defaults to high\n" );
533 if( h->param.rc.i_rc_method < 0 || h->param.rc.i_rc_method > 2 )
535 x264_log( h, X264_LOG_ERROR, "no ratecontrol method specified\n" );
538 h->param.rc.f_rf_constant = x264_clip3f( h->param.rc.f_rf_constant, -QP_BD_OFFSET, 51 );
539 h->param.rc.f_rf_constant_max = x264_clip3f( h->param.rc.f_rf_constant_max, -QP_BD_OFFSET, 51 );
540 h->param.rc.i_qp_constant = x264_clip3( h->param.rc.i_qp_constant, 0, QP_MAX );
541 h->param.analyse.i_subpel_refine = x264_clip3( h->param.analyse.i_subpel_refine, 0, 11 );
542 h->param.rc.f_ip_factor = X264_MAX( h->param.rc.f_ip_factor, 0.01f );
543 h->param.rc.f_pb_factor = X264_MAX( h->param.rc.f_pb_factor, 0.01f );
544 if( h->param.rc.i_rc_method == X264_RC_CRF )
546 h->param.rc.i_qp_constant = h->param.rc.f_rf_constant + QP_BD_OFFSET;
547 h->param.rc.i_bitrate = 0;
549 if( (h->param.rc.i_rc_method == X264_RC_CQP || h->param.rc.i_rc_method == X264_RC_CRF)
550 && h->param.rc.i_qp_constant == 0 )
552 h->mb.b_lossless = 1;
553 h->param.i_cqm_preset = X264_CQM_FLAT;
554 h->param.psz_cqm_file = NULL;
555 h->param.rc.i_rc_method = X264_RC_CQP;
556 h->param.rc.f_ip_factor = 1;
557 h->param.rc.f_pb_factor = 1;
558 h->param.analyse.b_psnr = 0;
559 h->param.analyse.b_ssim = 0;
560 h->param.analyse.i_chroma_qp_offset = 0;
561 h->param.analyse.i_trellis = 0;
562 h->param.analyse.b_fast_pskip = 0;
563 h->param.analyse.i_noise_reduction = 0;
564 h->param.analyse.b_psy = 0;
565 h->param.i_bframe = 0;
566 /* 8x8dct is not useful without RD in CAVLC lossless */
567 if( !h->param.b_cabac && h->param.analyse.i_subpel_refine < 6 )
568 h->param.analyse.b_transform_8x8 = 0;
570 if( h->param.rc.i_rc_method == X264_RC_CQP )
572 float qp_p = h->param.rc.i_qp_constant;
573 float qp_i = qp_p - 6*log2f( h->param.rc.f_ip_factor );
574 float qp_b = qp_p + 6*log2f( h->param.rc.f_pb_factor );
575 h->param.rc.i_qp_min = x264_clip3( (int)(X264_MIN3( qp_p, qp_i, qp_b )), 0, QP_MAX );
576 h->param.rc.i_qp_max = x264_clip3( (int)(X264_MAX3( qp_p, qp_i, qp_b ) + .999), 0, QP_MAX );
577 h->param.rc.i_aq_mode = 0;
578 h->param.rc.b_mb_tree = 0;
579 h->param.rc.i_bitrate = 0;
581 h->param.rc.i_qp_max = x264_clip3( h->param.rc.i_qp_max, 0, QP_MAX );
582 h->param.rc.i_qp_min = x264_clip3( h->param.rc.i_qp_min, 0, h->param.rc.i_qp_max );
583 h->param.rc.i_qp_step = x264_clip3( h->param.rc.i_qp_step, 2, QP_MAX );
584 h->param.rc.i_bitrate = x264_clip3( h->param.rc.i_bitrate, 0, 2000000 );
585 h->param.rc.i_vbv_buffer_size = x264_clip3( h->param.rc.i_vbv_buffer_size, 0, 2000000 );
586 h->param.rc.i_vbv_max_bitrate = x264_clip3( h->param.rc.i_vbv_max_bitrate, 0, 2000000 );
587 h->param.rc.f_vbv_buffer_init = x264_clip3f( h->param.rc.f_vbv_buffer_init, 0, 2000000 );
588 if( h->param.rc.i_vbv_buffer_size )
590 if( h->param.rc.i_rc_method == X264_RC_CQP )
592 x264_log( h, X264_LOG_WARNING, "VBV is incompatible with constant QP, ignored.\n" );
593 h->param.rc.i_vbv_max_bitrate = 0;
594 h->param.rc.i_vbv_buffer_size = 0;
596 else if( h->param.rc.i_vbv_max_bitrate == 0 )
598 if( h->param.rc.i_rc_method == X264_RC_ABR )
600 x264_log( h, X264_LOG_WARNING, "VBV maxrate unspecified, assuming CBR\n" );
601 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate;
605 x264_log( h, X264_LOG_WARNING, "VBV bufsize set but maxrate unspecified, ignored\n" );
606 h->param.rc.i_vbv_buffer_size = 0;
609 else if( h->param.rc.i_vbv_max_bitrate < h->param.rc.i_bitrate &&
610 h->param.rc.i_rc_method == X264_RC_ABR )
612 x264_log( h, X264_LOG_WARNING, "max bitrate less than average bitrate, assuming CBR\n" );
613 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate;
616 else if( h->param.rc.i_vbv_max_bitrate )
618 x264_log( h, X264_LOG_WARNING, "VBV maxrate specified, but no bufsize, ignored\n" );
619 h->param.rc.i_vbv_max_bitrate = 0;
622 h->param.i_slice_max_size = X264_MAX( h->param.i_slice_max_size, 0 );
623 h->param.i_slice_max_mbs = X264_MAX( h->param.i_slice_max_mbs, 0 );
625 int max_slices = (h->param.i_height+((16<<PARAM_INTERLACED)-1))/(16<<PARAM_INTERLACED);
626 if( h->param.b_sliced_threads )
627 h->param.i_slice_count = x264_clip3( h->param.i_threads, 0, max_slices );
630 h->param.i_slice_count = x264_clip3( h->param.i_slice_count, 0, max_slices );
631 if( h->param.i_slice_max_mbs || h->param.i_slice_max_size )
632 h->param.i_slice_count = 0;
635 if( h->param.b_bluray_compat )
637 h->param.i_bframe_pyramid = X264_MIN( X264_B_PYRAMID_STRICT, h->param.i_bframe_pyramid );
638 h->param.i_bframe = X264_MIN( h->param.i_bframe, 3 );
640 h->param.i_nal_hrd = X264_MAX( h->param.i_nal_hrd, X264_NAL_HRD_VBR );
641 h->param.i_slice_max_size = 0;
642 h->param.i_slice_max_mbs = 0;
643 h->param.b_intra_refresh = 0;
644 h->param.i_frame_reference = X264_MIN( h->param.i_frame_reference, 6 );
645 h->param.i_dpb_size = X264_MIN( h->param.i_dpb_size, 6 );
646 /* Don't use I-frames, because Blu-ray treats them the same as IDR. */
647 h->param.i_keyint_min = 1;
648 /* Due to the proliferation of broken players that don't handle dupes properly. */
649 h->param.analyse.i_weighted_pred = X264_MIN( h->param.analyse.i_weighted_pred, X264_WEIGHTP_SIMPLE );
650 if( h->param.b_fake_interlaced )
651 h->param.b_pic_struct = 1;
654 h->param.i_frame_reference = x264_clip3( h->param.i_frame_reference, 1, X264_REF_MAX );
655 h->param.i_dpb_size = x264_clip3( h->param.i_dpb_size, 1, X264_REF_MAX );
656 if( h->param.i_scenecut_threshold < 0 )
657 h->param.i_scenecut_threshold = 0;
658 h->param.analyse.i_direct_mv_pred = x264_clip3( h->param.analyse.i_direct_mv_pred, X264_DIRECT_PRED_NONE, X264_DIRECT_PRED_AUTO );
659 if( !h->param.analyse.i_subpel_refine && h->param.analyse.i_direct_mv_pred > X264_DIRECT_PRED_SPATIAL )
661 x264_log( h, X264_LOG_WARNING, "subme=0 + direct=temporal is not supported\n" );
662 h->param.analyse.i_direct_mv_pred = X264_DIRECT_PRED_SPATIAL;
664 h->param.i_bframe = x264_clip3( h->param.i_bframe, 0, X264_MIN( X264_BFRAME_MAX, h->param.i_keyint_max-1 ) );
665 h->param.i_bframe_bias = x264_clip3( h->param.i_bframe_bias, -90, 100 );
666 if( h->param.i_bframe <= 1 )
667 h->param.i_bframe_pyramid = X264_B_PYRAMID_NONE;
668 h->param.i_bframe_pyramid = x264_clip3( h->param.i_bframe_pyramid, X264_B_PYRAMID_NONE, X264_B_PYRAMID_NORMAL );
669 h->param.i_bframe_adaptive = x264_clip3( h->param.i_bframe_adaptive, X264_B_ADAPT_NONE, X264_B_ADAPT_TRELLIS );
670 if( !h->param.i_bframe )
672 h->param.i_bframe_adaptive = X264_B_ADAPT_NONE;
673 h->param.analyse.i_direct_mv_pred = 0;
674 h->param.analyse.b_weighted_bipred = 0;
675 h->param.b_open_gop = 0;
677 if( h->param.b_intra_refresh && h->param.i_bframe_pyramid == X264_B_PYRAMID_NORMAL )
679 x264_log( h, X264_LOG_WARNING, "b-pyramid normal + intra-refresh is not supported\n" );
680 h->param.i_bframe_pyramid = X264_B_PYRAMID_STRICT;
682 if( h->param.b_intra_refresh && (h->param.i_frame_reference > 1 || h->param.i_dpb_size > 1) )
684 x264_log( h, X264_LOG_WARNING, "ref > 1 + intra-refresh is not supported\n" );
685 h->param.i_frame_reference = 1;
686 h->param.i_dpb_size = 1;
688 if( h->param.b_intra_refresh && h->param.b_open_gop )
690 x264_log( h, X264_LOG_WARNING, "intra-refresh is not compatible with open-gop\n" );
691 h->param.b_open_gop = 0;
693 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;
694 if( h->param.i_keyint_min == X264_KEYINT_MIN_AUTO )
695 h->param.i_keyint_min = X264_MIN( h->param.i_keyint_max / 10, fps );
696 h->param.i_keyint_min = x264_clip3( h->param.i_keyint_min, 1, h->param.i_keyint_max/2+1 );
697 h->param.rc.i_lookahead = x264_clip3( h->param.rc.i_lookahead, 0, X264_LOOKAHEAD_MAX );
699 int maxrate = X264_MAX( h->param.rc.i_vbv_max_bitrate, h->param.rc.i_bitrate );
700 float bufsize = maxrate ? (float)h->param.rc.i_vbv_buffer_size / maxrate : 0;
701 h->param.rc.i_lookahead = X264_MIN( h->param.rc.i_lookahead, X264_MAX( h->param.i_keyint_max, bufsize*fps ) );
704 if( !h->param.i_timebase_num || !h->param.i_timebase_den || !(h->param.b_vfr_input || h->param.b_pulldown) )
706 h->param.i_timebase_num = h->param.i_fps_den;
707 h->param.i_timebase_den = h->param.i_fps_num;
710 h->param.rc.f_qcompress = x264_clip3f( h->param.rc.f_qcompress, 0.0, 1.0 );
711 if( h->param.i_keyint_max == 1 || h->param.rc.f_qcompress == 1 )
712 h->param.rc.b_mb_tree = 0;
713 if( (!h->param.b_intra_refresh && h->param.i_keyint_max != X264_KEYINT_MAX_INFINITE) &&
714 !h->param.rc.i_lookahead && h->param.rc.b_mb_tree )
716 x264_log( h, X264_LOG_WARNING, "lookaheadless mb-tree requires intra refresh or infinite keyint\n" );
717 h->param.rc.b_mb_tree = 0;
719 if( h->param.rc.b_stat_read )
720 h->param.rc.i_lookahead = 0;
722 if( h->param.i_sync_lookahead < 0 )
723 h->param.i_sync_lookahead = h->param.i_bframe + 1;
724 h->param.i_sync_lookahead = X264_MIN( h->param.i_sync_lookahead, X264_LOOKAHEAD_MAX );
725 if( h->param.rc.b_stat_read || h->i_thread_frames == 1 )
726 h->param.i_sync_lookahead = 0;
728 h->param.i_sync_lookahead = 0;
731 h->param.i_deblocking_filter_alphac0 = x264_clip3( h->param.i_deblocking_filter_alphac0, -6, 6 );
732 h->param.i_deblocking_filter_beta = x264_clip3( h->param.i_deblocking_filter_beta, -6, 6 );
733 h->param.analyse.i_luma_deadzone[0] = x264_clip3( h->param.analyse.i_luma_deadzone[0], 0, 32 );
734 h->param.analyse.i_luma_deadzone[1] = x264_clip3( h->param.analyse.i_luma_deadzone[1], 0, 32 );
736 h->param.i_cabac_init_idc = x264_clip3( h->param.i_cabac_init_idc, 0, 2 );
738 if( h->param.i_cqm_preset < X264_CQM_FLAT || h->param.i_cqm_preset > X264_CQM_CUSTOM )
739 h->param.i_cqm_preset = X264_CQM_FLAT;
741 if( h->param.analyse.i_me_method < X264_ME_DIA ||
742 h->param.analyse.i_me_method > X264_ME_TESA )
743 h->param.analyse.i_me_method = X264_ME_HEX;
744 h->param.analyse.i_me_range = x264_clip3( h->param.analyse.i_me_range, 4, 1024 );
745 if( h->param.analyse.i_me_range > 16 && h->param.analyse.i_me_method <= X264_ME_HEX )
746 h->param.analyse.i_me_range = 16;
747 if( h->param.analyse.i_me_method == X264_ME_TESA &&
748 (h->mb.b_lossless || h->param.analyse.i_subpel_refine <= 1) )
749 h->param.analyse.i_me_method = X264_ME_ESA;
750 h->param.analyse.b_mixed_references = h->param.analyse.b_mixed_references && h->param.i_frame_reference > 1;
751 h->param.analyse.inter &= X264_ANALYSE_PSUB16x16|X264_ANALYSE_PSUB8x8|X264_ANALYSE_BSUB16x16|
752 X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
753 h->param.analyse.intra &= X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
754 if( !(h->param.analyse.inter & X264_ANALYSE_PSUB16x16) )
755 h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
756 if( !h->param.analyse.b_transform_8x8 )
758 h->param.analyse.inter &= ~X264_ANALYSE_I8x8;
759 h->param.analyse.intra &= ~X264_ANALYSE_I8x8;
761 h->param.analyse.i_trellis = x264_clip3( h->param.analyse.i_trellis, 0, 2 );
762 h->param.rc.i_aq_mode = x264_clip3( h->param.rc.i_aq_mode, 0, 2 );
763 h->param.rc.f_aq_strength = x264_clip3f( h->param.rc.f_aq_strength, 0, 3 );
764 if( h->param.rc.f_aq_strength == 0 )
765 h->param.rc.i_aq_mode = 0;
767 if( h->param.i_log_level < X264_LOG_INFO )
769 h->param.analyse.b_psnr = 0;
770 h->param.analyse.b_ssim = 0;
772 /* Warn users trying to measure PSNR/SSIM with psy opts on. */
773 if( b_open && (h->param.analyse.b_psnr || h->param.analyse.b_ssim) )
777 if( h->param.analyse.b_psy )
779 s = h->param.analyse.b_psnr ? "psnr" : "ssim";
780 x264_log( h, X264_LOG_WARNING, "--%s used with psy on: results will be invalid!\n", s );
782 else if( !h->param.rc.i_aq_mode && h->param.analyse.b_ssim )
784 x264_log( h, X264_LOG_WARNING, "--ssim used with AQ off: results will be invalid!\n" );
787 else if( h->param.rc.i_aq_mode && h->param.analyse.b_psnr )
789 x264_log( h, X264_LOG_WARNING, "--psnr used with AQ on: results will be invalid!\n" );
793 x264_log( h, X264_LOG_WARNING, "--tune %s should be used if attempting to benchmark %s!\n", s, s );
796 if( !h->param.analyse.b_psy )
798 h->param.analyse.f_psy_rd = 0;
799 h->param.analyse.f_psy_trellis = 0;
801 h->param.analyse.f_psy_rd = x264_clip3f( h->param.analyse.f_psy_rd, 0, 10 );
802 h->param.analyse.f_psy_trellis = x264_clip3f( h->param.analyse.f_psy_trellis, 0, 10 );
803 h->mb.i_psy_rd = h->param.analyse.i_subpel_refine >= 6 ? FIX8( h->param.analyse.f_psy_rd ) : 0;
804 h->mb.i_psy_trellis = h->param.analyse.i_trellis ? FIX8( h->param.analyse.f_psy_trellis / 4 ) : 0;
805 h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -32, 32);
806 /* In 4:4:4 mode, chroma gets twice as much resolution, so we can halve its quality. */
807 if( b_open && i_csp >= X264_CSP_I444 && i_csp < X264_CSP_BGR && h->param.analyse.b_psy )
808 h->param.analyse.i_chroma_qp_offset += 6;
809 /* Psy RDO increases overall quantizers to improve the quality of luma--this indirectly hurts chroma quality */
810 /* so we lower the chroma QP offset to compensate */
811 if( b_open && h->mb.i_psy_rd )
812 h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_rd < 0.25 ? 1 : 2;
813 /* Psy trellis has a similar effect. */
814 if( b_open && h->mb.i_psy_trellis )
815 h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_trellis < 0.25 ? 1 : 2;
816 h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -12, 12);
817 /* MB-tree requires AQ to be on, even if the strength is zero. */
818 if( !h->param.rc.i_aq_mode && h->param.rc.b_mb_tree )
820 h->param.rc.i_aq_mode = 1;
821 h->param.rc.f_aq_strength = 0;
823 h->param.analyse.i_noise_reduction = x264_clip3( h->param.analyse.i_noise_reduction, 0, 1<<16 );
824 if( h->param.analyse.i_subpel_refine >= 10 && (h->param.analyse.i_trellis != 2 || !h->param.rc.i_aq_mode) )
825 h->param.analyse.i_subpel_refine = 9;
828 const x264_level_t *l = x264_levels;
829 if( h->param.i_level_idc < 0 )
831 int maxrate_bak = h->param.rc.i_vbv_max_bitrate;
832 if( h->param.rc.i_rc_method == X264_RC_ABR && h->param.rc.i_vbv_buffer_size <= 0 )
833 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate * 2;
834 x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
835 do h->param.i_level_idc = l->level_idc;
836 while( l[1].level_idc && x264_validate_levels( h, 0 ) && l++ );
837 h->param.rc.i_vbv_max_bitrate = maxrate_bak;
841 while( l->level_idc && l->level_idc != h->param.i_level_idc )
843 if( l->level_idc == 0 )
845 x264_log( h, X264_LOG_ERROR, "invalid level_idc: %d\n", h->param.i_level_idc );
849 if( h->param.analyse.i_mv_range <= 0 )
850 h->param.analyse.i_mv_range = l->mv_range >> PARAM_INTERLACED;
852 h->param.analyse.i_mv_range = x264_clip3(h->param.analyse.i_mv_range, 32, 512 >> PARAM_INTERLACED);
855 h->param.analyse.i_weighted_pred = x264_clip3( h->param.analyse.i_weighted_pred, X264_WEIGHTP_NONE, X264_WEIGHTP_SMART );
857 if( PARAM_INTERLACED )
859 if( h->param.analyse.i_me_method >= X264_ME_ESA )
861 x264_log( h, X264_LOG_WARNING, "interlace + me=esa is not implemented\n" );
862 h->param.analyse.i_me_method = X264_ME_UMH;
864 if( h->param.analyse.i_weighted_pred > 0 )
866 x264_log( h, X264_LOG_WARNING, "interlace + weightp is not implemented\n" );
867 h->param.analyse.i_weighted_pred = X264_WEIGHTP_NONE;
871 if( !h->param.analyse.i_weighted_pred && h->param.rc.b_mb_tree && h->param.analyse.b_psy )
872 h->param.analyse.i_weighted_pred = X264_WEIGHTP_FAKE;
874 if( h->i_thread_frames > 1 )
876 int r = h->param.analyse.i_mv_range_thread;
880 // half of the available space is reserved and divided evenly among the threads,
881 // the rest is allocated to whichever thread is far enough ahead to use it.
882 // reserving more space increases quality for some videos, but costs more time
883 // in thread synchronization.
884 int max_range = (h->param.i_height + X264_THREAD_HEIGHT) / h->i_thread_frames - X264_THREAD_HEIGHT;
887 r = X264_MAX( r, h->param.analyse.i_me_range );
888 r = X264_MIN( r, h->param.analyse.i_mv_range );
889 // round up to use the whole mb row
890 r2 = (r & ~15) + ((-X264_THREAD_HEIGHT) & 15);
893 x264_log( h, X264_LOG_DEBUG, "using mv_range_thread = %d\n", r2 );
894 h->param.analyse.i_mv_range_thread = r2;
897 if( h->param.rc.f_rate_tolerance < 0 )
898 h->param.rc.f_rate_tolerance = 0;
899 if( h->param.rc.f_qblur < 0 )
900 h->param.rc.f_qblur = 0;
901 if( h->param.rc.f_complexity_blur < 0 )
902 h->param.rc.f_complexity_blur = 0;
904 h->param.i_sps_id &= 31;
906 if( PARAM_INTERLACED )
907 h->param.b_pic_struct = 1;
909 h->param.i_nal_hrd = x264_clip3( h->param.i_nal_hrd, X264_NAL_HRD_NONE, X264_NAL_HRD_CBR );
911 if( h->param.i_nal_hrd && !h->param.rc.i_vbv_buffer_size )
913 x264_log( h, X264_LOG_WARNING, "NAL HRD parameters require VBV parameters\n" );
914 h->param.i_nal_hrd = X264_NAL_HRD_NONE;
917 if( h->param.i_nal_hrd == X264_NAL_HRD_CBR &&
918 (h->param.rc.i_bitrate != h->param.rc.i_vbv_max_bitrate || !h->param.rc.i_vbv_max_bitrate) )
920 x264_log( h, X264_LOG_WARNING, "CBR HRD requires constant bitrate\n" );
921 h->param.i_nal_hrd = X264_NAL_HRD_VBR;
924 if( h->param.psz_dump_yuv )
925 h->param.b_full_recon = 1;
927 /* ensure the booleans are 0 or 1 so they can be used in math */
928 #define BOOLIFY(x) h->param.x = !!h->param.x
930 BOOLIFY( b_constrained_intra );
931 BOOLIFY( b_deblocking_filter );
932 BOOLIFY( b_deterministic );
933 BOOLIFY( b_sliced_threads );
934 BOOLIFY( b_interlaced );
935 BOOLIFY( b_intra_refresh );
936 BOOLIFY( b_visualize );
938 BOOLIFY( b_repeat_headers );
940 BOOLIFY( b_vfr_input );
941 BOOLIFY( b_pulldown );
943 BOOLIFY( b_pic_struct );
944 BOOLIFY( b_fake_interlaced );
945 BOOLIFY( b_open_gop );
946 BOOLIFY( b_bluray_compat );
947 BOOLIFY( b_full_recon );
948 BOOLIFY( analyse.b_transform_8x8 );
949 BOOLIFY( analyse.b_weighted_bipred );
950 BOOLIFY( analyse.b_chroma_me );
951 BOOLIFY( analyse.b_mixed_references );
952 BOOLIFY( analyse.b_fast_pskip );
953 BOOLIFY( analyse.b_dct_decimate );
954 BOOLIFY( analyse.b_psy );
955 BOOLIFY( analyse.b_psnr );
956 BOOLIFY( analyse.b_ssim );
957 BOOLIFY( rc.b_stat_write );
958 BOOLIFY( rc.b_stat_read );
959 BOOLIFY( rc.b_mb_tree );
965 static void mbcmp_init( x264_t *h )
967 int satd = !h->mb.b_lossless && h->param.analyse.i_subpel_refine > 1;
968 memcpy( h->pixf.mbcmp, satd ? h->pixf.satd : h->pixf.sad_aligned, sizeof(h->pixf.mbcmp) );
969 memcpy( h->pixf.mbcmp_unaligned, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.mbcmp_unaligned) );
970 h->pixf.intra_mbcmp_x3_16x16 = satd ? h->pixf.intra_satd_x3_16x16 : h->pixf.intra_sad_x3_16x16;
971 h->pixf.intra_mbcmp_x3_8x16c = satd ? h->pixf.intra_satd_x3_8x16c : h->pixf.intra_sad_x3_8x16c;
972 h->pixf.intra_mbcmp_x3_8x8c = satd ? h->pixf.intra_satd_x3_8x8c : h->pixf.intra_sad_x3_8x8c;
973 h->pixf.intra_mbcmp_x3_8x8 = satd ? h->pixf.intra_sa8d_x3_8x8 : h->pixf.intra_sad_x3_8x8;
974 h->pixf.intra_mbcmp_x3_4x4 = satd ? h->pixf.intra_satd_x3_4x4 : h->pixf.intra_sad_x3_4x4;
975 h->pixf.intra_mbcmp_x9_4x4 = h->param.b_cpu_independent || h->mb.b_lossless ? NULL
976 : satd ? h->pixf.intra_satd_x9_4x4 : h->pixf.intra_sad_x9_4x4;
977 h->pixf.intra_mbcmp_x9_8x8 = h->param.b_cpu_independent || h->mb.b_lossless ? NULL
978 : satd ? h->pixf.intra_sa8d_x9_8x8 : h->pixf.intra_sad_x9_8x8;
979 satd &= h->param.analyse.i_me_method == X264_ME_TESA;
980 memcpy( h->pixf.fpelcmp, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.fpelcmp) );
981 memcpy( h->pixf.fpelcmp_x3, satd ? h->pixf.satd_x3 : h->pixf.sad_x3, sizeof(h->pixf.fpelcmp_x3) );
982 memcpy( h->pixf.fpelcmp_x4, satd ? h->pixf.satd_x4 : h->pixf.sad_x4, sizeof(h->pixf.fpelcmp_x4) );
985 static void chroma_dsp_init( x264_t *h )
987 memcpy( h->luma2chroma_pixel, x264_luma2chroma_pixel[CHROMA_FORMAT], sizeof(h->luma2chroma_pixel) );
989 switch( CHROMA_FORMAT )
992 memcpy( h->predict_chroma, h->predict_8x8c, sizeof(h->predict_chroma) );
993 h->mc.prefetch_fenc = h->mc.prefetch_fenc_420;
994 h->loopf.deblock_chroma[0] = h->loopf.deblock_h_chroma_420;
995 h->loopf.deblock_chroma_intra[0] = h->loopf.deblock_h_chroma_420_intra;
996 h->loopf.deblock_chroma_mbaff = h->loopf.deblock_chroma_420_mbaff;
997 h->loopf.deblock_chroma_intra_mbaff = h->loopf.deblock_chroma_420_intra_mbaff;
998 h->pixf.intra_mbcmp_x3_chroma = h->pixf.intra_mbcmp_x3_8x8c;
999 h->quantf.coeff_last[DCT_CHROMA_DC] = h->quantf.coeff_last4;
1000 h->quantf.coeff_level_run[DCT_CHROMA_DC] = h->quantf.coeff_level_run4;
1003 memcpy( h->predict_chroma, h->predict_8x16c, sizeof(h->predict_chroma) );
1004 h->mc.prefetch_fenc = h->mc.prefetch_fenc_422;
1005 h->loopf.deblock_chroma[0] = h->loopf.deblock_h_chroma_422;
1006 h->loopf.deblock_chroma_intra[0] = h->loopf.deblock_h_chroma_422_intra;
1007 h->loopf.deblock_chroma_mbaff = h->loopf.deblock_chroma_422_mbaff;
1008 h->loopf.deblock_chroma_intra_mbaff = h->loopf.deblock_chroma_422_intra_mbaff;
1009 h->pixf.intra_mbcmp_x3_chroma = h->pixf.intra_mbcmp_x3_8x16c;
1010 h->quantf.coeff_last[DCT_CHROMA_DC] = h->quantf.coeff_last8;
1011 h->quantf.coeff_level_run[DCT_CHROMA_DC] = h->quantf.coeff_level_run8;
1014 h->mc.prefetch_fenc = h->mc.prefetch_fenc_422; /* FIXME: doesn't cover V plane */
1015 h->loopf.deblock_chroma_mbaff = h->loopf.deblock_luma_mbaff;
1016 h->loopf.deblock_chroma_intra_mbaff = h->loopf.deblock_luma_intra_mbaff;
1021 static void x264_set_aspect_ratio( x264_t *h, x264_param_t *param, int initial )
1024 if( param->vui.i_sar_width > 0 && param->vui.i_sar_height > 0 )
1026 uint32_t i_w = param->vui.i_sar_width;
1027 uint32_t i_h = param->vui.i_sar_height;
1028 uint32_t old_w = h->param.vui.i_sar_width;
1029 uint32_t old_h = h->param.vui.i_sar_height;
1031 x264_reduce_fraction( &i_w, &i_h );
1033 while( i_w > 65535 || i_h > 65535 )
1039 x264_reduce_fraction( &i_w, &i_h );
1041 if( i_w != old_w || i_h != old_h || initial )
1043 h->param.vui.i_sar_width = 0;
1044 h->param.vui.i_sar_height = 0;
1045 if( i_w == 0 || i_h == 0 )
1046 x264_log( h, X264_LOG_WARNING, "cannot create valid sample aspect ratio\n" );
1049 x264_log( h, initial?X264_LOG_INFO:X264_LOG_DEBUG, "using SAR=%d/%d\n", i_w, i_h );
1050 h->param.vui.i_sar_width = i_w;
1051 h->param.vui.i_sar_height = i_h;
1053 x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
1058 /****************************************************************************
1059 * x264_encoder_open:
1060 ****************************************************************************/
1061 x264_t *x264_encoder_open( x264_param_t *param )
1065 int qp, i_slicetype_length;
1067 CHECKED_MALLOCZERO( h, sizeof(x264_t) );
1069 /* Create a copy of param */
1070 memcpy( &h->param, param, sizeof(x264_param_t) );
1072 if( param->param_free )
1073 param->param_free( param );
1075 if( x264_threading_init() )
1077 x264_log( h, X264_LOG_ERROR, "unable to initialize threading\n" );
1081 if( x264_validate_parameters( h, 1 ) < 0 )
1084 if( h->param.psz_cqm_file )
1085 if( x264_cqm_parse_file( h, h->param.psz_cqm_file ) < 0 )
1088 if( h->param.rc.psz_stat_out )
1089 h->param.rc.psz_stat_out = strdup( h->param.rc.psz_stat_out );
1090 if( h->param.rc.psz_stat_in )
1091 h->param.rc.psz_stat_in = strdup( h->param.rc.psz_stat_in );
1093 x264_reduce_fraction( &h->param.i_fps_num, &h->param.i_fps_den );
1094 x264_reduce_fraction( &h->param.i_timebase_num, &h->param.i_timebase_den );
1099 h->i_idr_pic_id = 0;
1101 if( (uint64_t)h->param.i_timebase_den * 2 > UINT32_MAX )
1103 x264_log( h, X264_LOG_ERROR, "Effective timebase denominator %u exceeds H.264 maximum\n", h->param.i_timebase_den );
1107 x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
1108 x264_pps_init( h->pps, h->param.i_sps_id, &h->param, h->sps );
1110 x264_set_aspect_ratio( h, &h->param, 1 );
1112 x264_validate_levels( h, 1 );
1114 h->chroma_qp_table = i_chroma_qp_table + 12 + h->pps->i_chroma_qp_index_offset;
1116 if( x264_cqm_init( h ) < 0 )
1119 h->mb.i_mb_width = h->sps->i_mb_width;
1120 h->mb.i_mb_height = h->sps->i_mb_height;
1121 h->mb.i_mb_count = h->mb.i_mb_width * h->mb.i_mb_height;
1123 h->mb.chroma_h_shift = CHROMA_FORMAT == CHROMA_420 || CHROMA_FORMAT == CHROMA_422;
1124 h->mb.chroma_v_shift = CHROMA_FORMAT == CHROMA_420;
1126 /* Adaptive MBAFF and subme 0 are not supported as we require halving motion
1127 * vectors during prediction, resulting in hpel mvs.
1128 * The chosen solution is to make MBAFF non-adaptive in this case. */
1129 h->mb.b_adaptive_mbaff = PARAM_INTERLACED && h->param.analyse.i_subpel_refine;
1132 if( h->param.i_bframe_adaptive == X264_B_ADAPT_TRELLIS && !h->param.rc.b_stat_read )
1133 h->frames.i_delay = X264_MAX(h->param.i_bframe,3)*4;
1135 h->frames.i_delay = h->param.i_bframe;
1136 if( h->param.rc.b_mb_tree || h->param.rc.i_vbv_buffer_size )
1137 h->frames.i_delay = X264_MAX( h->frames.i_delay, h->param.rc.i_lookahead );
1138 i_slicetype_length = h->frames.i_delay;
1139 h->frames.i_delay += h->i_thread_frames - 1;
1140 h->frames.i_delay += h->param.i_sync_lookahead;
1141 h->frames.i_delay += h->param.b_vfr_input;
1142 h->frames.i_bframe_delay = h->param.i_bframe ? (h->param.i_bframe_pyramid ? 2 : 1) : 0;
1144 h->frames.i_max_ref0 = h->param.i_frame_reference;
1145 h->frames.i_max_ref1 = X264_MIN( h->sps->vui.i_num_reorder_frames, h->param.i_frame_reference );
1146 h->frames.i_max_dpb = h->sps->vui.i_max_dec_frame_buffering;
1147 h->frames.b_have_lowres = !h->param.rc.b_stat_read
1148 && ( h->param.rc.i_rc_method == X264_RC_ABR
1149 || h->param.rc.i_rc_method == X264_RC_CRF
1150 || h->param.i_bframe_adaptive
1151 || h->param.i_scenecut_threshold
1152 || h->param.rc.b_mb_tree
1153 || h->param.analyse.i_weighted_pred );
1154 h->frames.b_have_lowres |= h->param.rc.b_stat_read && h->param.rc.i_vbv_buffer_size > 0;
1155 h->frames.b_have_sub8x8_esa = !!(h->param.analyse.inter & X264_ANALYSE_PSUB8x8);
1157 h->frames.i_last_idr =
1158 h->frames.i_last_keyframe = - h->param.i_keyint_max;
1159 h->frames.i_input = 0;
1160 h->frames.i_largest_pts = h->frames.i_second_largest_pts = -1;
1161 h->frames.i_poc_last_open_gop = -1;
1163 CHECKED_MALLOCZERO( h->frames.unused[0], (h->frames.i_delay + 3) * sizeof(x264_frame_t *) );
1164 /* Allocate room for max refs plus a few extra just in case. */
1165 CHECKED_MALLOCZERO( h->frames.unused[1], (h->i_thread_frames + X264_REF_MAX + 4) * sizeof(x264_frame_t *) );
1166 CHECKED_MALLOCZERO( h->frames.current, (h->param.i_sync_lookahead + h->param.i_bframe
1167 + h->i_thread_frames + 3) * sizeof(x264_frame_t *) );
1168 if( h->param.analyse.i_weighted_pred > 0 )
1169 CHECKED_MALLOCZERO( h->frames.blank_unused, h->i_thread_frames * 4 * sizeof(x264_frame_t *) );
1170 h->i_ref[0] = h->i_ref[1] = 0;
1171 h->i_cpb_delay = h->i_coded_fields = h->i_disp_fields = 0;
1172 h->i_prev_duration = ((uint64_t)h->param.i_fps_den * h->sps->vui.i_time_scale) / ((uint64_t)h->param.i_fps_num * h->sps->vui.i_num_units_in_tick);
1173 h->i_disp_fields_last_frame = -1;
1176 /* init CPU functions */
1177 x264_predict_16x16_init( h->param.cpu, h->predict_16x16 );
1178 x264_predict_8x8c_init( h->param.cpu, h->predict_8x8c );
1179 x264_predict_8x16c_init( h->param.cpu, h->predict_8x16c );
1180 x264_predict_8x8_init( h->param.cpu, h->predict_8x8, &h->predict_8x8_filter );
1181 x264_predict_4x4_init( h->param.cpu, h->predict_4x4 );
1182 x264_pixel_init( h->param.cpu, &h->pixf );
1183 x264_dct_init( h->param.cpu, &h->dctf );
1184 x264_zigzag_init( h->param.cpu, &h->zigzagf_progressive, &h->zigzagf_interlaced );
1185 memcpy( &h->zigzagf, PARAM_INTERLACED ? &h->zigzagf_interlaced : &h->zigzagf_progressive, sizeof(h->zigzagf) );
1186 x264_mc_init( h->param.cpu, &h->mc );
1187 x264_quant_init( h, h->param.cpu, &h->quantf );
1188 x264_deblock_init( h->param.cpu, &h->loopf, PARAM_INTERLACED );
1189 x264_bitstream_init( h->param.cpu, &h->bsf );
1190 if( h->param.b_cabac )
1191 x264_cabac_init( h );
1193 x264_stack_align( x264_cavlc_init, h );
1196 chroma_dsp_init( h );
1198 p = buf + sprintf( buf, "using cpu capabilities:" );
1199 for( int i = 0; x264_cpu_names[i].flags; i++ )
1201 if( !strcmp(x264_cpu_names[i].name, "SSE2")
1202 && h->param.cpu & (X264_CPU_SSE2_IS_FAST|X264_CPU_SSE2_IS_SLOW) )
1204 if( !strcmp(x264_cpu_names[i].name, "SSE3")
1205 && (h->param.cpu & X264_CPU_SSSE3 || !(h->param.cpu & X264_CPU_CACHELINE_64)) )
1207 if( !strcmp(x264_cpu_names[i].name, "SSE4.1")
1208 && (h->param.cpu & X264_CPU_SSE42) )
1210 if( (h->param.cpu & x264_cpu_names[i].flags) == x264_cpu_names[i].flags
1211 && (!i || x264_cpu_names[i].flags != x264_cpu_names[i-1].flags) )
1212 p += sprintf( p, " %s", x264_cpu_names[i].name );
1215 p += sprintf( p, " none!" );
1216 x264_log( h, X264_LOG_INFO, "%s\n", buf );
1218 float *logs = x264_analyse_prepare_costs( h );
1221 for( qp = X264_MIN( h->param.rc.i_qp_min, QP_MAX_SPEC ); qp <= h->param.rc.i_qp_max; qp++ )
1222 if( x264_analyse_init_costs( h, logs, qp ) )
1224 if( x264_analyse_init_costs( h, logs, X264_LOOKAHEAD_QP ) )
1228 static const uint16_t cost_mv_correct[7] = { 24, 47, 95, 189, 379, 757, 1515 };
1229 /* Checks for known miscompilation issues. */
1230 if( h->cost_mv[X264_LOOKAHEAD_QP][2013] != cost_mv_correct[BIT_DEPTH-8] )
1232 x264_log( h, X264_LOG_ERROR, "MV cost test failed: x264 has been miscompiled!\n" );
1236 /* Must be volatile or else GCC will optimize it out. */
1237 volatile int temp = 392;
1238 if( x264_clz( temp ) != 23 )
1240 x264_log( h, X264_LOG_ERROR, "CLZ test failed: x264 has been miscompiled!\n" );
1241 #if ARCH_X86 || ARCH_X86_64
1242 x264_log( h, X264_LOG_ERROR, "Are you attempting to run an SSE4a-targeted build on a CPU that\n" );
1243 x264_log( h, X264_LOG_ERROR, "doesn't support it?\n" );
1249 h->out.i_bitstream = X264_MAX( 1000000, h->param.i_width * h->param.i_height * 4
1250 * ( h->param.rc.i_rc_method == X264_RC_ABR ? pow( 0.95, h->param.rc.i_qp_min )
1251 : pow( 0.95, h->param.rc.i_qp_constant ) * X264_MAX( 1, h->param.rc.f_ip_factor )));
1253 h->nal_buffer_size = h->out.i_bitstream * 3/2 + 4;
1254 CHECKED_MALLOC( h->nal_buffer, h->nal_buffer_size );
1256 if( h->param.i_threads > 1 &&
1257 x264_threadpool_init( &h->threadpool, h->param.i_threads, (void*)x264_encoder_thread_init, h ) )
1261 for( int i = 1; i < h->param.i_threads + !!h->param.i_sync_lookahead; i++ )
1262 CHECKED_MALLOC( h->thread[i], sizeof(x264_t) );
1264 for( int i = 0; i < h->param.i_threads; i++ )
1266 int init_nal_count = h->param.i_slice_count + 3;
1267 int allocate_threadlocal_data = !h->param.b_sliced_threads || !i;
1271 if( allocate_threadlocal_data )
1273 h->thread[i]->fdec = x264_frame_pop_unused( h, 1 );
1274 if( !h->thread[i]->fdec )
1278 h->thread[i]->fdec = h->thread[0]->fdec;
1280 CHECKED_MALLOC( h->thread[i]->out.p_bitstream, h->out.i_bitstream );
1281 /* Start each thread with room for init_nal_count NAL units; it'll realloc later if needed. */
1282 CHECKED_MALLOC( h->thread[i]->out.nal, init_nal_count*sizeof(x264_nal_t) );
1283 h->thread[i]->out.i_nals_allocated = init_nal_count;
1285 if( allocate_threadlocal_data && x264_macroblock_cache_allocate( h->thread[i] ) < 0 )
1289 if( x264_lookahead_init( h, i_slicetype_length ) )
1292 for( int i = 0; i < h->param.i_threads; i++ )
1293 if( x264_macroblock_thread_allocate( h->thread[i], 0 ) < 0 )
1296 if( x264_ratecontrol_new( h ) < 0 )
1299 if( h->param.i_nal_hrd )
1301 x264_log( h, X264_LOG_DEBUG, "HRD bitrate: %i bits/sec\n", h->sps->vui.hrd.i_bit_rate_unscaled );
1302 x264_log( h, X264_LOG_DEBUG, "CPB size: %i bits\n", h->sps->vui.hrd.i_cpb_size_unscaled );
1305 if( h->param.psz_dump_yuv )
1307 /* create or truncate the reconstructed video file */
1308 FILE *f = fopen( h->param.psz_dump_yuv, "w" );
1311 x264_log( h, X264_LOG_ERROR, "dump_yuv: can't write to %s\n", h->param.psz_dump_yuv );
1314 else if( !x264_is_regular_file( f ) )
1316 x264_log( h, X264_LOG_ERROR, "dump_yuv: incompatible with non-regular file %s\n", h->param.psz_dump_yuv );
1322 const char *profile = h->sps->i_profile_idc == PROFILE_BASELINE ? "Constrained Baseline" :
1323 h->sps->i_profile_idc == PROFILE_MAIN ? "Main" :
1324 h->sps->i_profile_idc == PROFILE_HIGH ? "High" :
1325 h->sps->i_profile_idc == PROFILE_HIGH10 ? (h->sps->b_constraint_set3 == 1 ? "High 10 Intra" : "High 10") :
1326 h->sps->i_profile_idc == PROFILE_HIGH422 ? (h->sps->b_constraint_set3 == 1 ? "High 4:2:2 Intra" : "High 4:2:2") :
1327 h->sps->b_constraint_set3 == 1 ? "High 4:4:4 Intra" : "High 4:4:4 Predictive";
1329 snprintf( level, sizeof(level), "%d.%d", h->sps->i_level_idc/10, h->sps->i_level_idc%10 );
1330 if( h->sps->i_level_idc == 9 || ( h->sps->i_level_idc == 11 && h->sps->b_constraint_set3 &&
1331 (h->sps->i_profile_idc >= PROFILE_BASELINE && h->sps->i_profile_idc <= PROFILE_EXTENDED) ) )
1332 strcpy( level, "1b" );
1334 if( h->sps->i_profile_idc < PROFILE_HIGH10 )
1336 x264_log( h, X264_LOG_INFO, "profile %s, level %s\n",
1341 static const char * const subsampling[4] = { "4:0:0", "4:2:0", "4:2:2", "4:4:4" };
1342 x264_log( h, X264_LOG_INFO, "profile %s, level %s, %s %d-bit\n",
1343 profile, level, subsampling[CHROMA_FORMAT], BIT_DEPTH );
1352 /****************************************************************************
1353 * x264_encoder_reconfig:
1354 ****************************************************************************/
1355 int x264_encoder_reconfig( x264_t *h, x264_param_t *param )
1357 int rc_reconfig = 0;
1358 h = h->thread[h->thread[0]->i_thread_phase];
1359 x264_set_aspect_ratio( h, param, 0 );
1360 #define COPY(var) h->param.var = param->var
1361 COPY( i_frame_reference ); // but never uses more refs than initially specified
1362 COPY( i_bframe_bias );
1363 if( h->param.i_scenecut_threshold )
1364 COPY( i_scenecut_threshold ); // can't turn it on or off, only vary the threshold
1365 COPY( b_deblocking_filter );
1366 COPY( i_deblocking_filter_alphac0 );
1367 COPY( i_deblocking_filter_beta );
1368 COPY( i_frame_packing );
1369 COPY( analyse.inter );
1370 COPY( analyse.intra );
1371 COPY( analyse.i_direct_mv_pred );
1372 /* Scratch buffer prevents me_range from being increased for esa/tesa */
1373 if( h->param.analyse.i_me_method < X264_ME_ESA || param->analyse.i_me_range < h->param.analyse.i_me_range )
1374 COPY( analyse.i_me_range );
1375 COPY( analyse.i_noise_reduction );
1376 /* We can't switch out of subme=0 during encoding. */
1377 if( h->param.analyse.i_subpel_refine )
1378 COPY( analyse.i_subpel_refine );
1379 COPY( analyse.i_trellis );
1380 COPY( analyse.b_chroma_me );
1381 COPY( analyse.b_dct_decimate );
1382 COPY( analyse.b_fast_pskip );
1383 COPY( analyse.b_mixed_references );
1384 COPY( analyse.f_psy_rd );
1385 COPY( analyse.f_psy_trellis );
1387 // can only twiddle these if they were enabled to begin with:
1388 if( h->param.analyse.i_me_method >= X264_ME_ESA || param->analyse.i_me_method < X264_ME_ESA )
1389 COPY( analyse.i_me_method );
1390 if( h->param.analyse.i_me_method >= X264_ME_ESA && !h->frames.b_have_sub8x8_esa )
1391 h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
1392 if( h->pps->b_transform_8x8_mode )
1393 COPY( analyse.b_transform_8x8 );
1394 if( h->frames.i_max_ref1 > 1 )
1395 COPY( i_bframe_pyramid );
1396 COPY( i_slice_max_size );
1397 COPY( i_slice_max_mbs );
1398 COPY( i_slice_count );
1401 /* VBV can't be turned on if it wasn't on to begin with */
1402 if( h->param.rc.i_vbv_max_bitrate > 0 && h->param.rc.i_vbv_buffer_size > 0 &&
1403 param->rc.i_vbv_max_bitrate > 0 && param->rc.i_vbv_buffer_size > 0 )
1405 rc_reconfig |= h->param.rc.i_vbv_max_bitrate != param->rc.i_vbv_max_bitrate;
1406 rc_reconfig |= h->param.rc.i_vbv_buffer_size != param->rc.i_vbv_buffer_size;
1407 rc_reconfig |= h->param.rc.i_bitrate != param->rc.i_bitrate;
1408 COPY( rc.i_vbv_max_bitrate );
1409 COPY( rc.i_vbv_buffer_size );
1410 COPY( rc.i_bitrate );
1412 rc_reconfig |= h->param.rc.f_rf_constant != param->rc.f_rf_constant;
1413 rc_reconfig |= h->param.rc.f_rf_constant_max != param->rc.f_rf_constant_max;
1414 COPY( rc.f_rf_constant );
1415 COPY( rc.f_rf_constant_max );
1420 int ret = x264_validate_parameters( h, 0 );
1422 /* Supported reconfiguration options (1-pass only):
1426 * bitrate (CBR only) */
1427 if( !ret && rc_reconfig )
1428 x264_ratecontrol_init_reconfigurable( h, 0 );
1433 /****************************************************************************
1434 * x264_encoder_parameters:
1435 ****************************************************************************/
1436 void x264_encoder_parameters( x264_t *h, x264_param_t *param )
1438 memcpy( param, &h->thread[h->i_thread_phase]->param, sizeof(x264_param_t) );
1441 /* internal usage */
1442 static void x264_nal_start( x264_t *h, int i_type, int i_ref_idc )
1444 x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1446 nal->i_ref_idc = i_ref_idc;
1447 nal->i_type = i_type;
1448 nal->b_long_startcode = 1;
1451 nal->p_payload= &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8];
1454 /* if number of allocated nals is not enough, re-allocate a larger one. */
1455 static int x264_nal_check_buffer( x264_t *h )
1457 if( h->out.i_nal >= h->out.i_nals_allocated )
1459 x264_nal_t *new_out = x264_malloc( sizeof(x264_nal_t) * (h->out.i_nals_allocated*2) );
1462 memcpy( new_out, h->out.nal, sizeof(x264_nal_t) * (h->out.i_nals_allocated) );
1463 x264_free( h->out.nal );
1464 h->out.nal = new_out;
1465 h->out.i_nals_allocated *= 2;
1470 static int x264_nal_end( x264_t *h )
1472 x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1473 uint8_t *end = &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8];
1474 nal->i_payload = end - nal->p_payload;
1475 /* nal_escape_mmx reads past the end of the input.
1476 * While undefined padding wouldn't actually affect the output, it makes valgrind unhappy. */
1477 memset( end, 0xff, 32 );
1478 if( h->param.nalu_process )
1479 h->param.nalu_process( h, nal );
1482 return x264_nal_check_buffer( h );
1485 static int x264_encoder_encapsulate_nals( x264_t *h, int start )
1487 int nal_size = 0, previous_nal_size = 0;
1489 if( h->param.nalu_process )
1491 for( int i = start; i < h->out.i_nal; i++ )
1492 nal_size += h->out.nal[i].i_payload;
1496 for( int i = 0; i < start; i++ )
1497 previous_nal_size += h->out.nal[i].i_payload;
1499 for( int i = start; i < h->out.i_nal; i++ )
1500 nal_size += h->out.nal[i].i_payload;
1502 /* Worst-case NAL unit escaping: reallocate the buffer if it's too small. */
1503 int necessary_size = nal_size * 3/2 + h->out.i_nal * 4;
1504 if( h->nal_buffer_size < necessary_size )
1506 h->nal_buffer_size = necessary_size * 2;
1507 uint8_t *buf = x264_malloc( h->nal_buffer_size );
1510 if( previous_nal_size )
1511 memcpy( buf, h->nal_buffer, previous_nal_size );
1512 x264_free( h->nal_buffer );
1513 h->nal_buffer = buf;
1516 uint8_t *nal_buffer = h->nal_buffer + previous_nal_size;
1518 for( int i = start; i < h->out.i_nal; i++ )
1520 h->out.nal[i].b_long_startcode = !i || h->out.nal[i].i_type == NAL_SPS || h->out.nal[i].i_type == NAL_PPS;
1521 x264_nal_encode( h, nal_buffer, &h->out.nal[i] );
1522 nal_buffer += h->out.nal[i].i_payload;
1527 return nal_buffer - (h->nal_buffer + previous_nal_size);
1530 /****************************************************************************
1531 * x264_encoder_headers:
1532 ****************************************************************************/
1533 int x264_encoder_headers( x264_t *h, x264_nal_t **pp_nal, int *pi_nal )
1536 /* init bitstream context */
1538 bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
1540 /* Write SEI, SPS and PPS. */
1542 /* generate sequence parameters */
1543 x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
1544 x264_sps_write( &h->out.bs, h->sps );
1545 if( x264_nal_end( h ) )
1548 /* generate picture parameters */
1549 x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
1550 x264_pps_write( &h->out.bs, h->sps, h->pps );
1551 if( x264_nal_end( h ) )
1554 /* identify ourselves */
1555 x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
1556 if( x264_sei_version_write( h, &h->out.bs ) )
1558 if( x264_nal_end( h ) )
1561 frame_size = x264_encoder_encapsulate_nals( h, 0 );
1562 if( frame_size < 0 )
1566 *pi_nal = h->out.i_nal;
1567 *pp_nal = &h->out.nal[0];
1573 /* Check to see whether we have chosen a reference list ordering different
1574 * from the standard's default. */
1575 static inline void x264_reference_check_reorder( x264_t *h )
1577 /* The reorder check doesn't check for missing frames, so just
1578 * force a reorder if one of the reference list is corrupt. */
1579 for( int i = 0; h->frames.reference[i]; i++ )
1580 if( h->frames.reference[i]->b_corrupt )
1582 h->b_ref_reorder[0] = 1;
1585 for( int list = 0; list <= (h->sh.i_type == SLICE_TYPE_B); list++ )
1586 for( int i = 0; i < h->i_ref[list] - 1; i++ )
1588 int framenum_diff = h->fref[list][i+1]->i_frame_num - h->fref[list][i]->i_frame_num;
1589 int poc_diff = h->fref[list][i+1]->i_poc - h->fref[list][i]->i_poc;
1590 /* P and B-frames use different default orders. */
1591 if( h->sh.i_type == SLICE_TYPE_P ? framenum_diff > 0 : list == 1 ? poc_diff < 0 : poc_diff > 0 )
1593 h->b_ref_reorder[list] = 1;
1599 /* return -1 on failure, else return the index of the new reference frame */
1600 int x264_weighted_reference_duplicate( x264_t *h, int i_ref, const x264_weight_t *w )
1602 int i = h->i_ref[0];
1604 x264_frame_t *newframe;
1605 if( i <= 1 ) /* empty list, definitely can't duplicate frame */
1608 //Duplication is only used in X264_WEIGHTP_SMART
1609 if( h->param.analyse.i_weighted_pred != X264_WEIGHTP_SMART )
1612 /* Duplication is a hack to compensate for crappy rounding in motion compensation.
1613 * With high bit depth, it's not worth doing, so turn it off except in the case of
1614 * unweighted dupes. */
1615 if( BIT_DEPTH > 8 && w != x264_weight_none )
1618 newframe = x264_frame_pop_blank_unused( h );
1622 //FIXME: probably don't need to copy everything
1623 *newframe = *h->fref[0][i_ref];
1624 newframe->i_reference_count = 1;
1625 newframe->orig = h->fref[0][i_ref];
1626 newframe->b_duplicate = 1;
1627 memcpy( h->fenc->weight[j], w, sizeof(h->fenc->weight[i]) );
1629 /* shift the frames to make space for the dupe. */
1630 h->b_ref_reorder[0] = 1;
1631 if( h->i_ref[0] < X264_REF_MAX )
1633 h->fref[0][X264_REF_MAX-1] = NULL;
1634 x264_frame_unshift( &h->fref[0][j], newframe );
1639 static void x264_weighted_pred_init( x264_t *h )
1641 /* for now no analysis and set all weights to nothing */
1642 for( int i_ref = 0; i_ref < h->i_ref[0]; i_ref++ )
1643 h->fenc->weighted[i_ref] = h->fref[0][i_ref]->filtered[0][0];
1645 // FIXME: This only supports weighting of one reference frame
1646 // and duplicates of that frame.
1647 h->fenc->i_lines_weighted = 0;
1649 for( int i_ref = 0; i_ref < (h->i_ref[0] << SLICE_MBAFF); i_ref++ )
1650 for( int i = 0; i < 3; i++ )
1651 h->sh.weight[i_ref][i].weightfn = NULL;
1654 if( h->sh.i_type != SLICE_TYPE_P || h->param.analyse.i_weighted_pred <= 0 )
1657 int i_padv = PADV << PARAM_INTERLACED;
1659 int weightplane[2] = { 0, 0 };
1660 int buffer_next = 0;
1661 for( int i = 0; i < 3; i++ )
1663 for( int j = 0; j < h->i_ref[0]; j++ )
1665 if( h->fenc->weight[j][i].weightfn )
1667 h->sh.weight[j][i] = h->fenc->weight[j][i];
1668 // if weight is useless, don't write it to stream
1669 if( h->sh.weight[j][i].i_scale == 1<<h->sh.weight[j][i].i_denom && h->sh.weight[j][i].i_offset == 0 )
1670 h->sh.weight[j][i].weightfn = NULL;
1673 if( !weightplane[!!i] )
1675 weightplane[!!i] = 1;
1676 h->sh.weight[0][!!i].i_denom = denom = h->sh.weight[j][i].i_denom;
1677 assert( x264_clip3( denom, 0, 7 ) == denom );
1680 assert( h->sh.weight[j][i].i_denom == denom );
1683 h->fenc->weighted[j] = h->mb.p_weight_buf[buffer_next++] + h->fenc->i_stride[0] * i_padv + PADH;
1684 //scale full resolution frame
1685 if( h->param.i_threads == 1 )
1687 pixel *src = h->fref[0][j]->filtered[0][0] - h->fref[0][j]->i_stride[0]*i_padv - PADH;
1688 pixel *dst = h->fenc->weighted[j] - h->fenc->i_stride[0]*i_padv - PADH;
1689 int stride = h->fenc->i_stride[0];
1690 int width = h->fenc->i_width[0] + PADH*2;
1691 int height = h->fenc->i_lines[0] + i_padv*2;
1692 x264_weight_scale_plane( h, dst, stride, src, stride, width, height, &h->sh.weight[j][0] );
1693 h->fenc->i_lines_weighted = height;
1701 if( weightplane[1] )
1702 for( int i = 0; i < h->i_ref[0]; i++ )
1704 if( h->sh.weight[i][1].weightfn && !h->sh.weight[i][2].weightfn )
1706 h->sh.weight[i][2].i_scale = 1 << h->sh.weight[0][1].i_denom;
1707 h->sh.weight[i][2].i_offset = 0;
1709 else if( h->sh.weight[i][2].weightfn && !h->sh.weight[i][1].weightfn )
1711 h->sh.weight[i][1].i_scale = 1 << h->sh.weight[0][1].i_denom;
1712 h->sh.weight[i][1].i_offset = 0;
1716 if( !weightplane[0] )
1717 h->sh.weight[0][0].i_denom = 0;
1718 if( !weightplane[1] )
1719 h->sh.weight[0][1].i_denom = 0;
1720 h->sh.weight[0][2].i_denom = h->sh.weight[0][1].i_denom;
1723 static inline int x264_reference_distance( x264_t *h, x264_frame_t *frame )
1725 if( h->param.i_frame_packing == 5 )
1726 return abs((h->fenc->i_frame&~1) - (frame->i_frame&~1)) +
1727 ((h->fenc->i_frame&1) != (frame->i_frame&1));
1729 return abs(h->fenc->i_frame - frame->i_frame);
1732 static inline void x264_reference_build_list( x264_t *h, int i_poc )
1736 /* build ref list 0/1 */
1737 h->mb.pic.i_fref[0] = h->i_ref[0] = 0;
1738 h->mb.pic.i_fref[1] = h->i_ref[1] = 0;
1739 if( h->sh.i_type == SLICE_TYPE_I )
1742 for( int i = 0; h->frames.reference[i]; i++ )
1744 if( h->frames.reference[i]->b_corrupt )
1746 if( h->frames.reference[i]->i_poc < i_poc )
1747 h->fref[0][h->i_ref[0]++] = h->frames.reference[i];
1748 else if( h->frames.reference[i]->i_poc > i_poc )
1749 h->fref[1][h->i_ref[1]++] = h->frames.reference[i];
1752 /* Order reference lists by distance from the current frame. */
1753 for( int list = 0; list < 2; list++ )
1755 h->fref_nearest[list] = h->fref[list][0];
1759 for( int i = 0; i < h->i_ref[list] - 1; i++ )
1761 if( list ? h->fref[list][i+1]->i_poc < h->fref_nearest[list]->i_poc
1762 : h->fref[list][i+1]->i_poc > h->fref_nearest[list]->i_poc )
1763 h->fref_nearest[list] = h->fref[list][i+1];
1764 if( x264_reference_distance( h, h->fref[list][i] ) > x264_reference_distance( h, h->fref[list][i+1] ) )
1766 XCHG( x264_frame_t*, h->fref[list][i], h->fref[list][i+1] );
1774 if( h->sh.i_mmco_remove_from_end )
1775 for( int i = h->i_ref[0]-1; i >= h->i_ref[0] - h->sh.i_mmco_remove_from_end; i-- )
1777 int diff = h->i_frame_num - h->fref[0][i]->i_frame_num;
1778 h->sh.mmco[h->sh.i_mmco_command_count].i_poc = h->fref[0][i]->i_poc;
1779 h->sh.mmco[h->sh.i_mmco_command_count++].i_difference_of_pic_nums = diff;
1782 x264_reference_check_reorder( h );
1784 h->i_ref[1] = X264_MIN( h->i_ref[1], h->frames.i_max_ref1 );
1785 h->i_ref[0] = X264_MIN( h->i_ref[0], h->frames.i_max_ref0 );
1786 h->i_ref[0] = X264_MIN( h->i_ref[0], h->param.i_frame_reference ); // if reconfig() has lowered the limit
1788 /* For Blu-ray compliance, don't reference frames outside of the minigop. */
1789 if( IS_X264_TYPE_B( h->fenc->i_type ) && h->param.b_bluray_compat )
1790 h->i_ref[0] = X264_MIN( h->i_ref[0], IS_X264_TYPE_B( h->fref[0][0]->i_type ) + 1 );
1792 /* add duplicates */
1793 if( h->fenc->i_type == X264_TYPE_P )
1796 if( h->param.analyse.i_weighted_pred >= X264_WEIGHTP_SIMPLE )
1799 w[1].weightfn = w[2].weightfn = NULL;
1800 if( h->param.rc.b_stat_read )
1801 x264_ratecontrol_set_weights( h, h->fenc );
1803 if( !h->fenc->weight[0][0].weightfn )
1805 h->fenc->weight[0][0].i_denom = 0;
1806 SET_WEIGHT( w[0], 1, 1, 0, -1 );
1807 idx = x264_weighted_reference_duplicate( h, 0, w );
1811 if( h->fenc->weight[0][0].i_scale == 1<<h->fenc->weight[0][0].i_denom )
1813 SET_WEIGHT( h->fenc->weight[0][0], 1, 1, 0, h->fenc->weight[0][0].i_offset );
1815 x264_weighted_reference_duplicate( h, 0, x264_weight_none );
1816 if( h->fenc->weight[0][0].i_offset > -128 )
1818 w[0] = h->fenc->weight[0][0];
1820 h->mc.weight_cache( h, &w[0] );
1821 idx = x264_weighted_reference_duplicate( h, 0, w );
1825 h->mb.ref_blind_dupe = idx;
1828 assert( h->i_ref[0] + h->i_ref[1] <= X264_REF_MAX );
1829 h->mb.pic.i_fref[0] = h->i_ref[0];
1830 h->mb.pic.i_fref[1] = h->i_ref[1];
1833 static void x264_fdec_filter_row( x264_t *h, int mb_y, int b_inloop )
1835 /* mb_y is the mb to be encoded next, not the mb to be filtered here */
1836 int b_hpel = h->fdec->b_kept_as_ref;
1837 int b_deblock = h->sh.i_disable_deblocking_filter_idc != 1;
1838 int b_end = mb_y == h->i_threadslice_end;
1839 int b_measure_quality = 1;
1840 int min_y = mb_y - (1 << SLICE_MBAFF);
1841 int b_start = min_y == h->i_threadslice_start;
1842 /* Even in interlaced mode, deblocking never modifies more than 4 pixels
1843 * above each MB, as bS=4 doesn't happen for the top of interlaced mbpairs. */
1844 int minpix_y = min_y*16 - 4 * !b_start;
1845 int maxpix_y = mb_y*16 - 4 * !b_end;
1846 b_deblock &= b_hpel || h->param.b_full_recon;
1847 if( h->param.b_sliced_threads && b_start && min_y && !b_inloop )
1849 b_deblock = 0; /* We already deblocked on the inloop pass. */
1850 b_measure_quality = 0; /* We already measured quality on the inloop pass. */
1852 if( mb_y & SLICE_MBAFF )
1854 if( min_y < h->i_threadslice_start )
1858 for( int y = min_y; y < mb_y; y += (1 << SLICE_MBAFF) )
1859 x264_frame_deblock_row( h, y );
1861 /* FIXME: Prediction requires different borders for interlaced/progressive mc,
1862 * but the actual image data is equivalent. For now, maintain this
1863 * consistency by copying deblocked pixels between planes. */
1864 if( PARAM_INTERLACED )
1865 for( int p = 0; p < h->fdec->i_plane; p++ )
1866 for( int i = minpix_y>>(CHROMA_V_SHIFT && p); i < maxpix_y>>(CHROMA_V_SHIFT && p); i++ )
1867 memcpy( h->fdec->plane_fld[p] + i*h->fdec->i_stride[p],
1868 h->fdec->plane[p] + i*h->fdec->i_stride[p],
1869 h->mb.i_mb_width*16*sizeof(pixel) );
1873 int end = mb_y == h->mb.i_mb_height;
1874 x264_frame_expand_border( h, h->fdec, min_y, end );
1875 if( h->param.analyse.i_subpel_refine )
1877 x264_frame_filter( h, h->fdec, min_y, end );
1878 x264_frame_expand_border_filtered( h, h->fdec, min_y, end );
1883 for( int i = 0; i < 3; i++ )
1885 XCHG( pixel *, h->intra_border_backup[0][i], h->intra_border_backup[3][i] );
1886 XCHG( pixel *, h->intra_border_backup[1][i], h->intra_border_backup[4][i] );
1889 if( h->i_thread_frames > 1 && h->fdec->b_kept_as_ref )
1890 x264_frame_cond_broadcast( h->fdec, mb_y*16 + (b_end ? 10000 : -(X264_THREAD_HEIGHT << SLICE_MBAFF)) );
1892 if( b_measure_quality )
1894 maxpix_y = X264_MIN( maxpix_y, h->param.i_height );
1895 if( h->param.analyse.b_psnr )
1897 for( int p = 0; p < (CHROMA444 ? 3 : 1); p++ )
1898 h->stat.frame.i_ssd[p] += x264_pixel_ssd_wxh( &h->pixf,
1899 h->fdec->plane[p] + minpix_y * h->fdec->i_stride[p], h->fdec->i_stride[p],
1900 h->fenc->plane[p] + minpix_y * h->fenc->i_stride[p], h->fenc->i_stride[p],
1901 h->param.i_width, maxpix_y-minpix_y );
1904 uint64_t ssd_u, ssd_v;
1905 int v_shift = CHROMA_V_SHIFT;
1906 x264_pixel_ssd_nv12( &h->pixf,
1907 h->fdec->plane[1] + (minpix_y>>v_shift) * h->fdec->i_stride[1], h->fdec->i_stride[1],
1908 h->fenc->plane[1] + (minpix_y>>v_shift) * h->fenc->i_stride[1], h->fenc->i_stride[1],
1909 h->param.i_width>>1, (maxpix_y-minpix_y)>>v_shift, &ssd_u, &ssd_v );
1910 h->stat.frame.i_ssd[1] += ssd_u;
1911 h->stat.frame.i_ssd[2] += ssd_v;
1915 if( h->param.analyse.b_ssim )
1919 /* offset by 2 pixels to avoid alignment of ssim blocks with dct blocks,
1920 * and overlap by 4 */
1921 minpix_y += b_start ? 2 : -6;
1922 h->stat.frame.f_ssim +=
1923 x264_pixel_ssim_wxh( &h->pixf,
1924 h->fdec->plane[0] + 2+minpix_y*h->fdec->i_stride[0], h->fdec->i_stride[0],
1925 h->fenc->plane[0] + 2+minpix_y*h->fenc->i_stride[0], h->fenc->i_stride[0],
1926 h->param.i_width-2, maxpix_y-minpix_y, h->scratch_buffer, &ssim_cnt );
1927 h->stat.frame.i_ssim_cnt += ssim_cnt;
1932 static inline int x264_reference_update( x264_t *h )
1934 if( !h->fdec->b_kept_as_ref )
1936 if( h->i_thread_frames > 1 )
1938 x264_frame_push_unused( h, h->fdec );
1939 h->fdec = x264_frame_pop_unused( h, 1 );
1946 /* apply mmco from previous frame. */
1947 for( int i = 0; i < h->sh.i_mmco_command_count; i++ )
1948 for( int j = 0; h->frames.reference[j]; j++ )
1949 if( h->frames.reference[j]->i_poc == h->sh.mmco[i].i_poc )
1950 x264_frame_push_unused( h, x264_frame_shift( &h->frames.reference[j] ) );
1952 /* move frame in the buffer */
1953 x264_frame_push( h->frames.reference, h->fdec );
1954 if( h->frames.reference[h->sps->i_num_ref_frames] )
1955 x264_frame_push_unused( h, x264_frame_shift( h->frames.reference ) );
1956 h->fdec = x264_frame_pop_unused( h, 1 );
1962 static inline void x264_reference_reset( x264_t *h )
1964 while( h->frames.reference[0] )
1965 x264_frame_push_unused( h, x264_frame_pop( h->frames.reference ) );
1970 static inline void x264_reference_hierarchy_reset( x264_t *h )
1973 int b_hasdelayframe = 0;
1975 /* look for delay frames -- chain must only contain frames that are disposable */
1976 for( int i = 0; h->frames.current[i] && IS_DISPOSABLE( h->frames.current[i]->i_type ); i++ )
1977 b_hasdelayframe |= h->frames.current[i]->i_coded
1978 != h->frames.current[i]->i_frame + h->sps->vui.i_num_reorder_frames;
1980 /* This function must handle b-pyramid and clear frames for open-gop */
1981 if( h->param.i_bframe_pyramid != X264_B_PYRAMID_STRICT && !b_hasdelayframe && h->frames.i_poc_last_open_gop == -1 )
1984 /* Remove last BREF. There will never be old BREFs in the
1985 * dpb during a BREF decode when pyramid == STRICT */
1986 for( ref = 0; h->frames.reference[ref]; ref++ )
1988 if( ( h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT
1989 && h->frames.reference[ref]->i_type == X264_TYPE_BREF )
1990 || ( h->frames.reference[ref]->i_poc < h->frames.i_poc_last_open_gop
1991 && h->sh.i_type != SLICE_TYPE_B ) )
1993 int diff = h->i_frame_num - h->frames.reference[ref]->i_frame_num;
1994 h->sh.mmco[h->sh.i_mmco_command_count].i_difference_of_pic_nums = diff;
1995 h->sh.mmco[h->sh.i_mmco_command_count++].i_poc = h->frames.reference[ref]->i_poc;
1996 x264_frame_push_unused( h, x264_frame_shift( &h->frames.reference[ref] ) );
1997 h->b_ref_reorder[0] = 1;
2002 /* Prepare room in the dpb for the delayed display time of the later b-frame's */
2003 if( h->param.i_bframe_pyramid )
2004 h->sh.i_mmco_remove_from_end = X264_MAX( ref + 2 - h->frames.i_max_dpb, 0 );
2007 static inline void x264_slice_init( x264_t *h, int i_nal_type, int i_global_qp )
2009 /* ------------------------ Create slice header ----------------------- */
2010 if( i_nal_type == NAL_SLICE_IDR )
2012 x264_slice_header_init( h, &h->sh, h->sps, h->pps, h->i_idr_pic_id, h->i_frame_num, i_global_qp );
2015 h->i_idr_pic_id ^= 1;
2019 x264_slice_header_init( h, &h->sh, h->sps, h->pps, -1, h->i_frame_num, i_global_qp );
2021 h->sh.i_num_ref_idx_l0_active = h->i_ref[0] <= 0 ? 1 : h->i_ref[0];
2022 h->sh.i_num_ref_idx_l1_active = h->i_ref[1] <= 0 ? 1 : h->i_ref[1];
2023 if( h->sh.i_num_ref_idx_l0_active != h->pps->i_num_ref_idx_l0_default_active ||
2024 (h->sh.i_type == SLICE_TYPE_B && h->sh.i_num_ref_idx_l1_active != h->pps->i_num_ref_idx_l1_default_active) )
2026 h->sh.b_num_ref_idx_override = 1;
2030 if( h->fenc->i_type == X264_TYPE_BREF && h->param.b_bluray_compat && h->sh.i_mmco_command_count )
2033 h->sh_backup = h->sh;
2036 h->fdec->i_frame_num = h->sh.i_frame_num;
2038 if( h->sps->i_poc_type == 0 )
2040 h->sh.i_poc = h->fdec->i_poc;
2041 if( PARAM_INTERLACED )
2043 h->sh.i_delta_poc_bottom = h->param.b_tff ? 1 : -1;
2044 h->sh.i_poc += h->sh.i_delta_poc_bottom == -1;
2047 h->sh.i_delta_poc_bottom = 0;
2048 h->fdec->i_delta_poc[0] = h->sh.i_delta_poc_bottom == -1;
2049 h->fdec->i_delta_poc[1] = h->sh.i_delta_poc_bottom == 1;
2053 /* Nothing to do ? */
2056 x264_macroblock_slice_init( h );
2062 uint8_t cabac_prevbyte;
2065 x264_frame_stat_t stat;
2068 int field_decoding_flag;
2071 static ALWAYS_INLINE void x264_bitstream_backup( x264_t *h, x264_bs_bak_t *bak, int i_skip, int full )
2075 bak->stat = h->stat.frame;
2076 bak->last_qp = h->mb.i_last_qp;
2077 bak->last_dqp = h->mb.i_last_dqp;
2078 bak->field_decoding_flag = h->mb.field_decoding_flag;
2082 bak->stat.i_mv_bits = h->stat.frame.i_mv_bits;
2083 bak->stat.i_tex_bits = h->stat.frame.i_tex_bits;
2085 /* In the per-MB backup, we don't need the contexts because flushing the CABAC
2086 * encoder has no context dependency and in this case, a slice is ended (and
2087 * thus the content of all contexts are thrown away). */
2088 if( h->param.b_cabac )
2091 memcpy( &bak->cabac, &h->cabac, sizeof(x264_cabac_t) );
2093 memcpy( &bak->cabac, &h->cabac, offsetof(x264_cabac_t, f8_bits_encoded) );
2094 /* x264's CABAC writer modifies the previous byte during carry, so it has to be
2096 bak->cabac_prevbyte = h->cabac.p[-1];
2100 bak->bs = h->out.bs;
2105 static ALWAYS_INLINE void x264_bitstream_restore( x264_t *h, x264_bs_bak_t *bak, int *skip, int full )
2109 h->stat.frame = bak->stat;
2110 h->mb.i_last_qp = bak->last_qp;
2111 h->mb.i_last_dqp = bak->last_dqp;
2112 h->mb.field_decoding_flag = bak->field_decoding_flag;
2116 h->stat.frame.i_mv_bits = bak->stat.i_mv_bits;
2117 h->stat.frame.i_tex_bits = bak->stat.i_tex_bits;
2119 if( h->param.b_cabac )
2122 memcpy( &h->cabac, &bak->cabac, sizeof(x264_cabac_t) );
2124 memcpy( &h->cabac, &bak->cabac, offsetof(x264_cabac_t, f8_bits_encoded) );
2125 h->cabac.p[-1] = bak->cabac_prevbyte;
2129 h->out.bs = bak->bs;
2134 static int x264_slice_write( x264_t *h )
2137 int mb_xy, i_mb_x, i_mb_y;
2138 /* NALUs other than the first use a 3-byte startcode.
2139 * Add one extra byte for the rbsp, and one more for the final CABAC putbyte.
2140 * Then add an extra 5 bytes just in case, to account for random NAL escapes and
2141 * other inaccuracies. */
2142 int overhead_guess = (NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal)) + 1 + h->param.b_cabac + 5;
2143 int slice_max_size = h->param.i_slice_max_size > 0 ? (h->param.i_slice_max_size-overhead_guess)*8 : 0;
2144 int back_up_bitstream = slice_max_size || (!h->param.b_cabac && h->sps->i_profile_idc < PROFILE_HIGH);
2145 int starting_bits = bs_pos(&h->out.bs);
2146 int b_deblock = h->sh.i_disable_deblocking_filter_idc != 1;
2147 int b_hpel = h->fdec->b_kept_as_ref;
2148 int orig_last_mb = h->sh.i_last_mb;
2149 uint8_t *last_emu_check;
2150 x264_bs_bak_t bs_bak[2];
2151 b_deblock &= b_hpel || h->param.b_full_recon;
2152 bs_realign( &h->out.bs );
2155 x264_nal_start( h, h->i_nal_type, h->i_nal_ref_idc );
2156 h->out.nal[h->out.i_nal].i_first_mb = h->sh.i_first_mb;
2159 x264_macroblock_thread_init( h );
2161 /* If this isn't the first slice in the threadslice, set the slice QP
2162 * equal to the last QP in the previous slice for more accurate
2163 * CABAC initialization. */
2164 if( h->sh.i_first_mb != h->i_threadslice_start * h->mb.i_mb_width )
2166 h->sh.i_qp = h->mb.i_last_qp;
2167 h->sh.i_qp_delta = h->sh.i_qp - h->pps->i_pic_init_qp;
2170 x264_slice_header_write( &h->out.bs, &h->sh, h->i_nal_ref_idc );
2171 if( h->param.b_cabac )
2173 /* alignment needed */
2174 bs_align_1( &h->out.bs );
2177 x264_cabac_context_init( h, &h->cabac, h->sh.i_type, x264_clip3( h->sh.i_qp-QP_BD_OFFSET, 0, 51 ), h->sh.i_cabac_init_idc );
2178 x264_cabac_encode_init ( &h->cabac, h->out.bs.p, h->out.bs.p_end );
2179 last_emu_check = h->cabac.p;
2182 last_emu_check = h->out.bs.p;
2183 h->mb.i_last_qp = h->sh.i_qp;
2184 h->mb.i_last_dqp = 0;
2185 h->mb.field_decoding_flag = 0;
2187 i_mb_y = h->sh.i_first_mb / h->mb.i_mb_width;
2188 i_mb_x = h->sh.i_first_mb % h->mb.i_mb_width;
2193 mb_xy = i_mb_x + i_mb_y * h->mb.i_mb_width;
2194 int mb_spos = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
2198 if( x264_bitstream_check_buffer( h ) )
2200 if( !(i_mb_y & SLICE_MBAFF) && h->param.rc.i_vbv_buffer_size )
2201 x264_bitstream_backup( h, &bs_bak[1], i_skip, 1 );
2202 if( !h->mb.b_reencode_mb )
2203 x264_fdec_filter_row( h, i_mb_y, 1 );
2206 if( !(i_mb_y & SLICE_MBAFF) && back_up_bitstream )
2207 x264_bitstream_backup( h, &bs_bak[0], i_skip, 0 );
2209 if( PARAM_INTERLACED )
2211 if( h->mb.b_adaptive_mbaff )
2215 /* FIXME: VSAD is fast but fairly poor at choosing the best interlace type. */
2216 h->mb.b_interlaced = x264_field_vsad( h, i_mb_x, i_mb_y );
2217 memcpy( &h->zigzagf, MB_INTERLACED ? &h->zigzagf_interlaced : &h->zigzagf_progressive, sizeof(h->zigzagf) );
2218 if( !MB_INTERLACED && (i_mb_y+2) == h->mb.i_mb_height )
2219 x264_expand_border_mbpair( h, i_mb_x, i_mb_y );
2222 h->mb.field[mb_xy] = MB_INTERLACED;
2227 x264_macroblock_cache_load_interlaced( h, i_mb_x, i_mb_y );
2229 x264_macroblock_cache_load_progressive( h, i_mb_x, i_mb_y );
2231 x264_macroblock_analyse( h );
2233 /* encode this macroblock -> be careful it can change the mb type to P_SKIP if needed */
2235 x264_macroblock_encode( h );
2237 if( h->param.b_cabac )
2239 if( mb_xy > h->sh.i_first_mb && !(SLICE_MBAFF && (i_mb_y&1)) )
2240 x264_cabac_encode_terminal( &h->cabac );
2242 if( IS_SKIP( h->mb.i_type ) )
2243 x264_cabac_mb_skip( h, 1 );
2246 if( h->sh.i_type != SLICE_TYPE_I )
2247 x264_cabac_mb_skip( h, 0 );
2248 x264_macroblock_write_cabac( h, &h->cabac );
2253 if( IS_SKIP( h->mb.i_type ) )
2257 if( h->sh.i_type != SLICE_TYPE_I )
2259 bs_write_ue( &h->out.bs, i_skip ); /* skip run */
2262 x264_macroblock_write_cavlc( h );
2263 /* If there was a CAVLC level code overflow, try again at a higher QP. */
2264 if( h->mb.b_overflow )
2266 h->mb.i_chroma_qp = h->chroma_qp_table[++h->mb.i_qp];
2267 h->mb.i_skip_intra = 0;
2268 h->mb.b_skip_mc = 0;
2269 h->mb.b_overflow = 0;
2270 x264_bitstream_restore( h, &bs_bak[0], &i_skip, 0 );
2276 int total_bits = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
2277 int mb_size = total_bits - mb_spos;
2279 if( slice_max_size && (!SLICE_MBAFF || (i_mb_y&1)) )
2281 /* Count the skip run, just in case. */
2282 if( !h->param.b_cabac )
2283 total_bits += bs_size_ue_big( i_skip );
2284 /* Check for escape bytes. */
2285 uint8_t *end = h->param.b_cabac ? h->cabac.p : h->out.bs.p;
2286 for( ; last_emu_check < end - 2; last_emu_check++ )
2287 if( last_emu_check[0] == 0 && last_emu_check[1] == 0 && last_emu_check[2] <= 3 )
2289 slice_max_size -= 8;
2292 /* We'll just re-encode this last macroblock if we go over the max slice size. */
2293 if( total_bits - starting_bits > slice_max_size && !h->mb.b_reencode_mb )
2295 if( mb_xy-SLICE_MBAFF*h->mb.i_mb_stride != h->sh.i_first_mb )
2297 x264_bitstream_restore( h, &bs_bak[0], &i_skip, 0 );
2298 h->mb.b_reencode_mb = 1;
2301 // set to bottom of previous mbpair
2303 h->sh.i_last_mb = mb_xy-1+h->mb.i_mb_stride*(!(i_mb_y&1));
2305 h->sh.i_last_mb = (i_mb_y-2+!(i_mb_y&1))*h->mb.i_mb_stride + h->mb.i_mb_width - 1;
2308 h->sh.i_last_mb = mb_xy-1;
2312 h->sh.i_last_mb = mb_xy;
2315 h->mb.b_reencode_mb = 0;
2318 if( h->param.b_visualize )
2319 x264_visualize_mb( h );
2323 x264_macroblock_cache_save( h );
2325 if( x264_ratecontrol_mb( h, mb_size ) < 0 )
2327 x264_bitstream_restore( h, &bs_bak[1], &i_skip, 1 );
2328 h->mb.b_reencode_mb = 1;
2330 i_mb_y = i_mb_y - SLICE_MBAFF;
2331 h->mb.i_mb_prev_xy = i_mb_y * h->mb.i_mb_stride - 1;
2332 h->sh.i_last_mb = orig_last_mb;
2336 /* accumulate mb stats */
2337 h->stat.frame.i_mb_count[h->mb.i_type]++;
2339 int b_intra = IS_INTRA( h->mb.i_type );
2340 int b_skip = IS_SKIP( h->mb.i_type );
2341 if( h->param.i_log_level >= X264_LOG_INFO || h->param.rc.b_stat_write )
2343 if( !b_intra && !b_skip && !IS_DIRECT( h->mb.i_type ) )
2345 if( h->mb.i_partition != D_8x8 )
2346 h->stat.frame.i_mb_partition[h->mb.i_partition] += 4;
2348 for( int i = 0; i < 4; i++ )
2349 h->stat.frame.i_mb_partition[h->mb.i_sub_partition[i]] ++;
2350 if( h->param.i_frame_reference > 1 )
2351 for( int i_list = 0; i_list <= (h->sh.i_type == SLICE_TYPE_B); i_list++ )
2352 for( int i = 0; i < 4; i++ )
2354 int i_ref = h->mb.cache.ref[i_list][ x264_scan8[4*i] ];
2356 h->stat.frame.i_mb_count_ref[i_list][i_ref] ++;
2361 if( h->param.i_log_level >= X264_LOG_INFO )
2363 if( h->mb.i_cbp_luma | h->mb.i_cbp_chroma )
2367 for( int i = 0; i < 4; i++ )
2368 if( h->mb.i_cbp_luma & (1 << i) )
2369 for( int p = 0; p < 3; p++ )
2372 int nnz8x8 = M16( &h->mb.cache.non_zero_count[x264_scan8[s8]+0] )
2373 | M16( &h->mb.cache.non_zero_count[x264_scan8[s8]+8] );
2374 h->stat.frame.i_mb_cbp[!b_intra + p*2] += !!nnz8x8;
2379 int cbpsum = (h->mb.i_cbp_luma&1) + ((h->mb.i_cbp_luma>>1)&1)
2380 + ((h->mb.i_cbp_luma>>2)&1) + (h->mb.i_cbp_luma>>3);
2381 h->stat.frame.i_mb_cbp[!b_intra + 0] += cbpsum;
2382 h->stat.frame.i_mb_cbp[!b_intra + 2] += !!h->mb.i_cbp_chroma;
2383 h->stat.frame.i_mb_cbp[!b_intra + 4] += h->mb.i_cbp_chroma >> 1;
2386 if( h->mb.i_cbp_luma && !b_intra )
2388 h->stat.frame.i_mb_count_8x8dct[0] ++;
2389 h->stat.frame.i_mb_count_8x8dct[1] += h->mb.b_transform_8x8;
2391 if( b_intra && h->mb.i_type != I_PCM )
2393 if( h->mb.i_type == I_16x16 )
2394 h->stat.frame.i_mb_pred_mode[0][h->mb.i_intra16x16_pred_mode]++;
2395 else if( h->mb.i_type == I_8x8 )
2396 for( int i = 0; i < 16; i += 4 )
2397 h->stat.frame.i_mb_pred_mode[1][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
2398 else //if( h->mb.i_type == I_4x4 )
2399 for( int i = 0; i < 16; i++ )
2400 h->stat.frame.i_mb_pred_mode[2][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
2401 h->stat.frame.i_mb_pred_mode[3][x264_mb_chroma_pred_mode_fix[h->mb.i_chroma_pred_mode]]++;
2403 h->stat.frame.i_mb_field[b_intra?0:b_skip?2:1] += MB_INTERLACED;
2406 /* calculate deblock strength values (actual deblocking is done per-row along with hpel) */
2408 x264_macroblock_deblock_strength( h );
2410 if( mb_xy == h->sh.i_last_mb )
2415 i_mb_x += i_mb_y & 1;
2416 i_mb_y ^= i_mb_x < h->mb.i_mb_width;
2420 if( i_mb_x == h->mb.i_mb_width )
2426 h->out.nal[h->out.i_nal].i_last_mb = h->sh.i_last_mb;
2428 if( h->param.b_cabac )
2430 x264_cabac_encode_flush( h, &h->cabac );
2431 h->out.bs.p = h->cabac.p;
2436 bs_write_ue( &h->out.bs, i_skip ); /* last skip run */
2437 /* rbsp_slice_trailing_bits */
2438 bs_rbsp_trailing( &h->out.bs );
2439 bs_flush( &h->out.bs );
2441 if( x264_nal_end( h ) )
2444 if( h->sh.i_last_mb == (h->i_threadslice_end * h->mb.i_mb_width - 1) )
2446 h->stat.frame.i_misc_bits = bs_pos( &h->out.bs )
2447 + (h->out.i_nal*NALU_OVERHEAD * 8)
2448 - h->stat.frame.i_tex_bits
2449 - h->stat.frame.i_mv_bits;
2450 x264_fdec_filter_row( h, h->i_threadslice_end, 1 );
2456 static void x264_thread_sync_context( x264_t *dst, x264_t *src )
2461 // reference counting
2462 for( x264_frame_t **f = src->frames.reference; *f; f++ )
2463 (*f)->i_reference_count++;
2464 for( x264_frame_t **f = dst->frames.reference; *f; f++ )
2465 x264_frame_push_unused( src, *f );
2466 src->fdec->i_reference_count++;
2467 x264_frame_push_unused( src, dst->fdec );
2469 // copy everything except the per-thread pointers and the constants.
2470 memcpy( &dst->i_frame, &src->i_frame, offsetof(x264_t, mb.type) - offsetof(x264_t, i_frame) );
2471 dst->param = src->param;
2472 dst->stat = src->stat;
2473 dst->pixf = src->pixf;
2476 static void x264_thread_sync_stat( x264_t *dst, x264_t *src )
2480 memcpy( &dst->stat.i_frame_count, &src->stat.i_frame_count, sizeof(dst->stat) - sizeof(dst->stat.frame) );
2483 static void *x264_slices_write( x264_t *h )
2485 int i_slice_num = 0;
2486 int last_thread_mb = h->sh.i_last_mb;
2489 if( h->param.b_visualize )
2490 if( x264_visualize_init( h ) )
2495 memset( &h->stat.frame, 0, sizeof(h->stat.frame) );
2496 h->mb.b_reencode_mb = 0;
2497 while( h->sh.i_first_mb + SLICE_MBAFF*h->mb.i_mb_stride <= last_thread_mb )
2499 h->sh.i_last_mb = last_thread_mb;
2500 if( h->param.i_slice_max_mbs )
2504 // convert first to mbaff form, add slice-max-mbs, then convert back to normal form
2505 int last_mbaff = 2*(h->sh.i_first_mb % h->mb.i_mb_width)
2506 + h->mb.i_mb_width*(h->sh.i_first_mb / h->mb.i_mb_width)
2507 + h->param.i_slice_max_mbs - 1;
2508 int last_x = (last_mbaff % (2*h->mb.i_mb_width))/2;
2509 int last_y = (last_mbaff / (2*h->mb.i_mb_width))*2 + 1;
2510 h->sh.i_last_mb = last_x + h->mb.i_mb_stride*last_y;
2513 h->sh.i_last_mb = h->sh.i_first_mb + h->param.i_slice_max_mbs - 1;
2515 else if( h->param.i_slice_count && !h->param.b_sliced_threads )
2517 int height = h->mb.i_mb_height >> PARAM_INTERLACED;
2518 int width = h->mb.i_mb_width << PARAM_INTERLACED;
2520 h->sh.i_last_mb = (height * i_slice_num + h->param.i_slice_count/2) / h->param.i_slice_count * width - 1;
2522 h->sh.i_last_mb = X264_MIN( h->sh.i_last_mb, last_thread_mb );
2523 if( x264_stack_align( x264_slice_write, h ) )
2525 h->sh.i_first_mb = h->sh.i_last_mb + 1;
2526 // if i_first_mb is not the last mb in a row then go to the next mb in MBAFF order
2527 if( SLICE_MBAFF && h->sh.i_first_mb % h->mb.i_mb_width )
2528 h->sh.i_first_mb -= h->mb.i_mb_stride;
2532 if( h->param.b_visualize )
2534 x264_visualize_show( h );
2535 x264_visualize_close( h );
2542 static int x264_threaded_slices_write( x264_t *h )
2544 /* set first/last mb and sync contexts */
2545 for( int i = 0; i < h->param.i_threads; i++ )
2547 x264_t *t = h->thread[i];
2550 t->param = h->param;
2551 memcpy( &t->i_frame, &h->i_frame, offsetof(x264_t, rc) - offsetof(x264_t, i_frame) );
2553 int height = h->mb.i_mb_height >> PARAM_INTERLACED;
2554 t->i_threadslice_start = ((height * i + h->param.i_slice_count/2) / h->param.i_threads) << PARAM_INTERLACED;
2555 t->i_threadslice_end = ((height * (i+1) + h->param.i_slice_count/2) / h->param.i_threads) << PARAM_INTERLACED;
2556 t->sh.i_first_mb = t->i_threadslice_start * h->mb.i_mb_width;
2557 t->sh.i_last_mb = t->i_threadslice_end * h->mb.i_mb_width - 1;
2560 x264_stack_align( x264_analyse_weight_frame, h, h->mb.i_mb_height*16 + 16 );
2562 x264_threads_distribute_ratecontrol( h );
2565 for( int i = 0; i < h->param.i_threads; i++ )
2567 x264_threadpool_run( h->threadpool, (void*)x264_slices_write, h->thread[i] );
2568 h->thread[i]->b_thread_active = 1;
2570 for( int i = 0; i < h->param.i_threads; i++ )
2572 h->thread[i]->b_thread_active = 0;
2573 if( (intptr_t)x264_threadpool_wait( h->threadpool, h->thread[i] ) )
2577 /* Go back and fix up the hpel on the borders between slices. */
2578 for( int i = 1; i < h->param.i_threads; i++ )
2580 x264_fdec_filter_row( h->thread[i], h->thread[i]->i_threadslice_start + 1, 0 );
2582 x264_fdec_filter_row( h->thread[i], h->thread[i]->i_threadslice_start + 2, 0 );
2585 x264_threads_merge_ratecontrol( h );
2587 for( int i = 1; i < h->param.i_threads; i++ )
2589 x264_t *t = h->thread[i];
2590 for( int j = 0; j < t->out.i_nal; j++ )
2592 h->out.nal[h->out.i_nal] = t->out.nal[j];
2594 x264_nal_check_buffer( h );
2596 /* All entries in stat.frame are ints except for ssd/ssim. */
2597 for( int j = 0; j < (offsetof(x264_t,stat.frame.i_ssd) - offsetof(x264_t,stat.frame.i_mv_bits)) / sizeof(int); j++ )
2598 ((int*)&h->stat.frame)[j] += ((int*)&t->stat.frame)[j];
2599 for( int j = 0; j < 3; j++ )
2600 h->stat.frame.i_ssd[j] += t->stat.frame.i_ssd[j];
2601 h->stat.frame.f_ssim += t->stat.frame.f_ssim;
2602 h->stat.frame.i_ssim_cnt += t->stat.frame.i_ssim_cnt;
2608 void x264_encoder_intra_refresh( x264_t *h )
2610 h = h->thread[h->i_thread_phase];
2611 h->b_queued_intra_refresh = 1;
2614 int x264_encoder_invalidate_reference( x264_t *h, int64_t pts )
2616 if( h->param.i_bframe )
2618 x264_log( h, X264_LOG_ERROR, "x264_encoder_invalidate_reference is not supported with B-frames enabled\n" );
2621 if( h->param.b_intra_refresh )
2623 x264_log( h, X264_LOG_ERROR, "x264_encoder_invalidate_reference is not supported with intra refresh enabled\n" );
2626 h = h->thread[h->i_thread_phase];
2627 if( pts >= h->i_last_idr_pts )
2629 for( int i = 0; h->frames.reference[i]; i++ )
2630 if( pts <= h->frames.reference[i]->i_pts )
2631 h->frames.reference[i]->b_corrupt = 1;
2632 if( pts <= h->fdec->i_pts )
2633 h->fdec->b_corrupt = 1;
2638 /****************************************************************************
2639 * x264_encoder_encode:
2640 * XXX: i_poc : is the poc of the current given picture
2641 * i_frame : is the number of the frame being coded
2642 * ex: type frame poc
2650 ****************************************************************************/
2651 int x264_encoder_encode( x264_t *h,
2652 x264_nal_t **pp_nal, int *pi_nal,
2653 x264_picture_t *pic_in,
2654 x264_picture_t *pic_out )
2656 x264_t *thread_current, *thread_prev, *thread_oldest;
2657 int i_nal_type, i_nal_ref_idc, i_global_qp;
2658 int overhead = NALU_OVERHEAD;
2660 if( h->i_thread_frames > 1 )
2662 thread_prev = h->thread[ h->i_thread_phase ];
2663 h->i_thread_phase = (h->i_thread_phase + 1) % h->i_thread_frames;
2664 thread_current = h->thread[ h->i_thread_phase ];
2665 thread_oldest = h->thread[ (h->i_thread_phase + 1) % h->i_thread_frames ];
2666 x264_thread_sync_context( thread_current, thread_prev );
2667 x264_thread_sync_ratecontrol( thread_current, thread_prev, thread_oldest );
2676 if( h->param.cpu&X264_CPU_SSE_MISALIGN )
2677 x264_cpu_mask_misalign_sse();
2680 // ok to call this before encoding any frames, since the initial values of fdec have b_kept_as_ref=0
2681 if( x264_reference_update( h ) )
2683 h->fdec->i_lines_completed = -1;
2689 /* ------------------- Setup new frame from picture -------------------- */
2690 if( pic_in != NULL )
2692 /* 1: Copy the picture to a frame and move it to a buffer */
2693 x264_frame_t *fenc = x264_frame_pop_unused( h, 0 );
2697 if( x264_frame_copy_picture( h, fenc, pic_in ) < 0 )
2700 if( h->param.i_width != 16 * h->mb.i_mb_width ||
2701 h->param.i_height != 16 * h->mb.i_mb_height )
2702 x264_frame_expand_border_mod16( h, fenc );
2704 fenc->i_frame = h->frames.i_input++;
2706 if( fenc->i_frame == 0 )
2707 h->frames.i_first_pts = fenc->i_pts;
2708 if( h->frames.i_bframe_delay && fenc->i_frame == h->frames.i_bframe_delay )
2709 h->frames.i_bframe_delay_time = fenc->i_pts - h->frames.i_first_pts;
2711 if( h->param.b_vfr_input && fenc->i_pts <= h->frames.i_largest_pts )
2712 x264_log( h, X264_LOG_WARNING, "non-strictly-monotonic PTS\n" );
2714 h->frames.i_second_largest_pts = h->frames.i_largest_pts;
2715 h->frames.i_largest_pts = fenc->i_pts;
2717 if( (fenc->i_pic_struct < PIC_STRUCT_AUTO) || (fenc->i_pic_struct > PIC_STRUCT_TRIPLE) )
2718 fenc->i_pic_struct = PIC_STRUCT_AUTO;
2720 if( fenc->i_pic_struct == PIC_STRUCT_AUTO )
2723 int b_interlaced = fenc->param ? fenc->param->b_interlaced : h->param.b_interlaced;
2725 int b_interlaced = 0;
2729 int b_tff = fenc->param ? fenc->param->b_tff : h->param.b_tff;
2730 fenc->i_pic_struct = b_tff ? PIC_STRUCT_TOP_BOTTOM : PIC_STRUCT_BOTTOM_TOP;
2733 fenc->i_pic_struct = PIC_STRUCT_PROGRESSIVE;
2736 if( h->param.rc.b_mb_tree && h->param.rc.b_stat_read )
2738 if( x264_macroblock_tree_read( h, fenc, pic_in->prop.quant_offsets ) )
2742 x264_stack_align( x264_adaptive_quant_frame, h, fenc, pic_in->prop.quant_offsets );
2744 if( pic_in->prop.quant_offsets_free )
2745 pic_in->prop.quant_offsets_free( pic_in->prop.quant_offsets );
2747 if( h->frames.b_have_lowres )
2748 x264_frame_init_lowres( h, fenc );
2750 /* 2: Place the frame into the queue for its slice type decision */
2751 x264_lookahead_put_frame( h, fenc );
2753 if( h->frames.i_input <= h->frames.i_delay + 1 - h->i_thread_frames )
2755 /* Nothing yet to encode, waiting for filling of buffers */
2756 pic_out->i_type = X264_TYPE_AUTO;
2762 /* signal kills for lookahead thread */
2763 x264_pthread_mutex_lock( &h->lookahead->ifbuf.mutex );
2764 h->lookahead->b_exit_thread = 1;
2765 x264_pthread_cond_broadcast( &h->lookahead->ifbuf.cv_fill );
2766 x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
2770 /* 3: The picture is analyzed in the lookahead */
2771 if( !h->frames.current[0] )
2772 x264_lookahead_get_frames( h );
2774 if( !h->frames.current[0] && x264_lookahead_is_empty( h ) )
2775 return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
2777 /* ------------------- Get frame to be encoded ------------------------- */
2778 /* 4: get picture to encode */
2779 h->fenc = x264_frame_shift( h->frames.current );
2780 if( h->i_frame == h->i_thread_frames - 1 )
2781 h->i_reordered_pts_delay = h->fenc->i_reordered_pts;
2782 if( h->fenc->param )
2784 x264_encoder_reconfig( h, h->fenc->param );
2785 if( h->fenc->param->param_free )
2786 h->fenc->param->param_free( h->fenc->param );
2789 if( !IS_X264_TYPE_I( h->fenc->i_type ) )
2791 int valid_refs_left = 0;
2792 for( int i = 0; h->frames.reference[i]; i++ )
2793 if( !h->frames.reference[i]->b_corrupt )
2795 /* No valid reference frames left: force an IDR. */
2796 if( !valid_refs_left )
2798 h->fenc->b_keyframe = 1;
2799 h->fenc->i_type = X264_TYPE_IDR;
2803 if( h->fenc->b_keyframe )
2805 h->frames.i_last_keyframe = h->fenc->i_frame;
2806 if( h->fenc->i_type == X264_TYPE_IDR )
2809 h->frames.i_last_idr = h->fenc->i_frame;
2812 h->sh.i_mmco_command_count =
2813 h->sh.i_mmco_remove_from_end = 0;
2814 h->b_ref_reorder[0] =
2815 h->b_ref_reorder[1] = 0;
2817 h->fenc->i_poc = 2 * ( h->fenc->i_frame - X264_MAX( h->frames.i_last_idr, 0 ) );
2819 /* ------------------- Setup frame context ----------------------------- */
2820 /* 5: Init data dependent of frame type */
2821 if( h->fenc->i_type == X264_TYPE_IDR )
2823 /* reset ref pictures */
2824 i_nal_type = NAL_SLICE_IDR;
2825 i_nal_ref_idc = NAL_PRIORITY_HIGHEST;
2826 h->sh.i_type = SLICE_TYPE_I;
2827 x264_reference_reset( h );
2828 h->frames.i_poc_last_open_gop = -1;
2830 else if( h->fenc->i_type == X264_TYPE_I )
2832 i_nal_type = NAL_SLICE;
2833 i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
2834 h->sh.i_type = SLICE_TYPE_I;
2835 x264_reference_hierarchy_reset( h );
2836 if( h->param.b_open_gop )
2837 h->frames.i_poc_last_open_gop = h->fenc->b_keyframe ? h->fenc->i_poc : -1;
2839 else if( h->fenc->i_type == X264_TYPE_P )
2841 i_nal_type = NAL_SLICE;
2842 i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
2843 h->sh.i_type = SLICE_TYPE_P;
2844 x264_reference_hierarchy_reset( h );
2845 h->frames.i_poc_last_open_gop = -1;
2847 else if( h->fenc->i_type == X264_TYPE_BREF )
2849 i_nal_type = NAL_SLICE;
2850 i_nal_ref_idc = h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT ? NAL_PRIORITY_LOW : NAL_PRIORITY_HIGH;
2851 h->sh.i_type = SLICE_TYPE_B;
2852 x264_reference_hierarchy_reset( h );
2856 i_nal_type = NAL_SLICE;
2857 i_nal_ref_idc = NAL_PRIORITY_DISPOSABLE;
2858 h->sh.i_type = SLICE_TYPE_B;
2861 h->fdec->i_type = h->fenc->i_type;
2862 h->fdec->i_frame = h->fenc->i_frame;
2863 h->fenc->b_kept_as_ref =
2864 h->fdec->b_kept_as_ref = i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE && h->param.i_keyint_max > 1;
2866 h->fdec->i_pts = h->fenc->i_pts;
2867 if( h->frames.i_bframe_delay )
2869 int64_t *prev_reordered_pts = thread_current->frames.i_prev_reordered_pts;
2870 h->fdec->i_dts = h->i_frame > h->frames.i_bframe_delay
2871 ? prev_reordered_pts[ (h->i_frame - h->frames.i_bframe_delay) % h->frames.i_bframe_delay ]
2872 : h->fenc->i_reordered_pts - h->frames.i_bframe_delay_time;
2873 prev_reordered_pts[ h->i_frame % h->frames.i_bframe_delay ] = h->fenc->i_reordered_pts;
2876 h->fdec->i_dts = h->fenc->i_reordered_pts;
2877 if( h->fenc->i_type == X264_TYPE_IDR )
2878 h->i_last_idr_pts = h->fdec->i_pts;
2880 /* ------------------- Init ----------------------------- */
2881 /* build ref list 0/1 */
2882 x264_reference_build_list( h, h->fdec->i_poc );
2884 /* ---------------------- Write the bitstream -------------------------- */
2885 /* Init bitstream context */
2886 if( h->param.b_sliced_threads )
2888 for( int i = 0; i < h->param.i_threads; i++ )
2890 bs_init( &h->thread[i]->out.bs, h->thread[i]->out.p_bitstream, h->thread[i]->out.i_bitstream );
2891 h->thread[i]->out.i_nal = 0;
2896 bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
2900 if( h->param.b_aud )
2904 if( h->sh.i_type == SLICE_TYPE_I )
2906 else if( h->sh.i_type == SLICE_TYPE_P )
2908 else if( h->sh.i_type == SLICE_TYPE_B )
2913 x264_nal_start( h, NAL_AUD, NAL_PRIORITY_DISPOSABLE );
2914 bs_write( &h->out.bs, 3, pic_type );
2915 bs_rbsp_trailing( &h->out.bs );
2916 if( x264_nal_end( h ) )
2918 overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2921 h->i_nal_type = i_nal_type;
2922 h->i_nal_ref_idc = i_nal_ref_idc;
2924 if( h->param.b_intra_refresh )
2926 if( IS_X264_TYPE_I( h->fenc->i_type ) )
2928 h->fdec->i_frames_since_pir = 0;
2929 h->b_queued_intra_refresh = 0;
2930 /* PIR is currently only supported with ref == 1, so any intra frame effectively refreshes
2931 * the whole frame and counts as an intra refresh. */
2932 h->fdec->f_pir_position = h->mb.i_mb_width;
2934 else if( h->fenc->i_type == X264_TYPE_P )
2936 int pocdiff = (h->fdec->i_poc - h->fref[0][0]->i_poc)/2;
2937 float increment = X264_MAX( ((float)h->mb.i_mb_width-1) / h->param.i_keyint_max, 1 );
2938 h->fdec->f_pir_position = h->fref[0][0]->f_pir_position;
2939 h->fdec->i_frames_since_pir = h->fref[0][0]->i_frames_since_pir + pocdiff;
2940 if( h->fdec->i_frames_since_pir >= h->param.i_keyint_max ||
2941 (h->b_queued_intra_refresh && h->fdec->f_pir_position + 0.5 >= h->mb.i_mb_width) )
2943 h->fdec->f_pir_position = 0;
2944 h->fdec->i_frames_since_pir = 0;
2945 h->b_queued_intra_refresh = 0;
2946 h->fenc->b_keyframe = 1;
2948 h->fdec->i_pir_start_col = h->fdec->f_pir_position+0.5;
2949 h->fdec->f_pir_position += increment * pocdiff;
2950 h->fdec->i_pir_end_col = h->fdec->f_pir_position+0.5;
2951 /* If our intra refresh has reached the right side of the frame, we're done. */
2952 if( h->fdec->i_pir_end_col >= h->mb.i_mb_width - 1 )
2954 h->fdec->f_pir_position = h->mb.i_mb_width;
2955 h->fdec->i_pir_end_col = h->mb.i_mb_width - 1;
2960 if( h->fenc->b_keyframe )
2962 /* Write SPS and PPS */
2963 if( h->param.b_repeat_headers )
2965 /* generate sequence parameters */
2966 x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
2967 x264_sps_write( &h->out.bs, h->sps );
2968 if( x264_nal_end( h ) )
2970 overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
2972 /* generate picture parameters */
2973 x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
2974 x264_pps_write( &h->out.bs, h->sps, h->pps );
2975 if( x264_nal_end( h ) )
2977 overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
2980 /* when frame threading is used, buffering period sei is written in x264_encoder_frame_end */
2981 if( h->i_thread_frames == 1 && h->sps->vui.b_nal_hrd_parameters_present )
2983 x264_hrd_fullness( h );
2984 x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2985 x264_sei_buffering_period_write( h, &h->out.bs );
2986 if( x264_nal_end( h ) )
2988 overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
2992 /* write extra sei */
2993 for( int i = 0; i < h->fenc->extra_sei.num_payloads; i++ )
2995 x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2996 x264_sei_write( &h->out.bs, h->fenc->extra_sei.payloads[i].payload, h->fenc->extra_sei.payloads[i].payload_size,
2997 h->fenc->extra_sei.payloads[i].payload_type );
2998 if( x264_nal_end( h ) )
3000 overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
3001 if( h->fenc->extra_sei.sei_free && h->fenc->extra_sei.payloads[i].payload )
3002 h->fenc->extra_sei.sei_free( h->fenc->extra_sei.payloads[i].payload );
3005 if( h->fenc->extra_sei.sei_free && h->fenc->extra_sei.payloads )
3006 h->fenc->extra_sei.sei_free( h->fenc->extra_sei.payloads );
3008 if( h->fenc->b_keyframe )
3010 if( h->param.b_repeat_headers && h->fenc->i_frame == 0 )
3012 /* identify ourself */
3013 x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3014 if( x264_sei_version_write( h, &h->out.bs ) )
3016 if( x264_nal_end( h ) )
3018 overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
3021 if( h->fenc->i_type != X264_TYPE_IDR )
3023 int time_to_recovery = h->param.b_open_gop ? 0 : X264_MIN( h->mb.i_mb_width - 1, h->param.i_keyint_max ) + h->param.i_bframe - 1;
3024 x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3025 x264_sei_recovery_point_write( h, &h->out.bs, time_to_recovery );
3026 if( x264_nal_end( h ) )
3028 overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
3031 if ( h->param.i_frame_packing >= 0 )
3033 x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3034 x264_sei_frame_packing_write( h, &h->out.bs );
3035 if( x264_nal_end( h ) )
3037 overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
3041 /* generate sei pic timing */
3042 if( h->sps->vui.b_pic_struct_present || h->sps->vui.b_nal_hrd_parameters_present )
3044 x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3045 x264_sei_pic_timing_write( h, &h->out.bs );
3046 if( x264_nal_end( h ) )
3048 overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
3051 /* As required by Blu-ray. */
3052 if( !IS_X264_TYPE_B( h->fenc->i_type ) && h->b_sh_backup )
3055 x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3056 x264_sei_dec_ref_pic_marking_write( h, &h->out.bs );
3057 if( x264_nal_end( h ) )
3059 overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
3062 if( h->fenc->b_keyframe && h->param.b_intra_refresh )
3063 h->i_cpb_delay_pir_offset = h->fenc->i_cpb_delay;
3065 /* Init the rate control */
3066 /* FIXME: Include slice header bit cost. */
3067 x264_ratecontrol_start( h, h->fenc->i_qpplus1, overhead*8 );
3068 i_global_qp = x264_ratecontrol_qp( h );
3070 pic_out->i_qpplus1 =
3071 h->fdec->i_qpplus1 = i_global_qp + 1;
3073 if( h->param.rc.b_stat_read && h->sh.i_type != SLICE_TYPE_I )
3075 x264_reference_build_list_optimal( h );
3076 x264_reference_check_reorder( h );
3080 h->fdec->i_poc_l0ref0 = h->fref[0][0]->i_poc;
3082 /* ------------------------ Create slice header ----------------------- */
3083 x264_slice_init( h, i_nal_type, i_global_qp );
3085 /*------------------------- Weights -------------------------------------*/
3086 if( h->sh.i_type == SLICE_TYPE_B )
3087 x264_macroblock_bipred_init( h );
3089 x264_weighted_pred_init( h );
3091 if( i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE )
3095 h->i_threadslice_start = 0;
3096 h->i_threadslice_end = h->mb.i_mb_height;
3097 if( h->i_thread_frames > 1 )
3099 x264_threadpool_run( h->threadpool, (void*)x264_slices_write, h );
3100 h->b_thread_active = 1;
3102 else if( h->param.b_sliced_threads )
3104 if( x264_threaded_slices_write( h ) )
3108 if( (intptr_t)x264_slices_write( h ) )
3111 return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
3114 static int x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
3115 x264_nal_t **pp_nal, int *pi_nal,
3116 x264_picture_t *pic_out )
3118 char psz_message[80];
3120 if( h->b_thread_active )
3122 h->b_thread_active = 0;
3123 if( (intptr_t)x264_threadpool_wait( h->threadpool, h ) )
3128 pic_out->i_type = X264_TYPE_AUTO;
3133 /* generate buffering period sei and insert it into place */
3134 if( h->i_thread_frames > 1 && h->fenc->b_keyframe && h->sps->vui.b_nal_hrd_parameters_present )
3136 x264_hrd_fullness( h );
3137 x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3138 x264_sei_buffering_period_write( h, &h->out.bs );
3139 if( x264_nal_end( h ) )
3141 /* buffering period sei must follow AUD, SPS and PPS and precede all other SEIs */
3143 while( h->out.nal[idx].i_type == NAL_AUD ||
3144 h->out.nal[idx].i_type == NAL_SPS ||
3145 h->out.nal[idx].i_type == NAL_PPS )
3147 x264_nal_t nal_tmp = h->out.nal[h->out.i_nal-1];
3148 memmove( &h->out.nal[idx+1], &h->out.nal[idx], (h->out.i_nal-idx-1)*sizeof(x264_nal_t) );
3149 h->out.nal[idx] = nal_tmp;
3152 int frame_size = x264_encoder_encapsulate_nals( h, 0 );
3153 if( frame_size < 0 )
3156 /* Set output picture properties */
3157 pic_out->i_type = h->fenc->i_type;
3159 pic_out->b_keyframe = h->fenc->b_keyframe;
3160 pic_out->i_pic_struct = h->fenc->i_pic_struct;
3162 pic_out->i_pts = h->fdec->i_pts;
3163 pic_out->i_dts = h->fdec->i_dts;
3165 if( pic_out->i_pts < pic_out->i_dts )
3166 x264_log( h, X264_LOG_WARNING, "invalid DTS: PTS is less than DTS\n" );
3168 pic_out->opaque = h->fenc->opaque;
3170 pic_out->img.i_csp = h->fdec->i_csp;
3172 pic_out->img.i_csp |= X264_CSP_HIGH_DEPTH;
3174 pic_out->img.i_plane = h->fdec->i_plane;
3175 for( int i = 0; i < pic_out->img.i_plane; i++ )
3177 pic_out->img.i_stride[i] = h->fdec->i_stride[i] * sizeof(pixel);
3178 pic_out->img.plane[i] = (uint8_t*)h->fdec->plane[i];
3181 x264_frame_push_unused( thread_current, h->fenc );
3183 /* ---------------------- Update encoder state ------------------------- */
3187 if( x264_ratecontrol_end( h, frame_size * 8, &filler ) < 0 )
3190 pic_out->hrd_timing = h->fenc->hrd_timing;
3195 overhead = (FILLER_OVERHEAD - h->param.b_annexb);
3196 if( h->param.i_slice_max_size && filler > h->param.i_slice_max_size )
3198 int next_size = filler - h->param.i_slice_max_size;
3199 int overflow = X264_MAX( overhead - next_size, 0 );
3200 f = h->param.i_slice_max_size - overhead - overflow;
3203 f = X264_MAX( 0, filler - overhead );
3205 x264_nal_start( h, NAL_FILLER, NAL_PRIORITY_DISPOSABLE );
3206 x264_filler_write( h, &h->out.bs, f );
3207 if( x264_nal_end( h ) )
3209 int total_size = x264_encoder_encapsulate_nals( h, h->out.i_nal-1 );
3210 if( total_size < 0 )
3212 frame_size += total_size;
3213 filler -= total_size;
3216 /* End bitstream, set output */
3217 *pi_nal = h->out.i_nal;
3218 *pp_nal = h->out.nal;
3222 x264_noise_reduction_update( h );
3224 /* ---------------------- Compute/Print statistics --------------------- */
3225 x264_thread_sync_stat( h, h->thread[0] );
3228 h->stat.i_frame_count[h->sh.i_type]++;
3229 h->stat.i_frame_size[h->sh.i_type] += frame_size;
3230 h->stat.f_frame_qp[h->sh.i_type] += h->fdec->f_qp_avg_aq;
3232 for( int i = 0; i < X264_MBTYPE_MAX; i++ )
3233 h->stat.i_mb_count[h->sh.i_type][i] += h->stat.frame.i_mb_count[i];
3234 for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
3235 h->stat.i_mb_partition[h->sh.i_type][i] += h->stat.frame.i_mb_partition[i];
3236 for( int i = 0; i < 2; i++ )
3237 h->stat.i_mb_count_8x8dct[i] += h->stat.frame.i_mb_count_8x8dct[i];
3238 for( int i = 0; i < 6; i++ )
3239 h->stat.i_mb_cbp[i] += h->stat.frame.i_mb_cbp[i];
3240 for( int i = 0; i < 4; i++ )
3241 for( int j = 0; j < 13; j++ )
3242 h->stat.i_mb_pred_mode[i][j] += h->stat.frame.i_mb_pred_mode[i][j];
3243 if( h->sh.i_type != SLICE_TYPE_I )
3244 for( int i_list = 0; i_list < 2; i_list++ )
3245 for( int i = 0; i < X264_REF_MAX*2; i++ )
3246 h->stat.i_mb_count_ref[h->sh.i_type][i_list][i] += h->stat.frame.i_mb_count_ref[i_list][i];
3247 for( int i = 0; i < 3; i++ )
3248 h->stat.i_mb_field[i] += h->stat.frame.i_mb_field[i];
3249 if( h->sh.i_type == SLICE_TYPE_P && h->param.analyse.i_weighted_pred >= X264_WEIGHTP_SIMPLE )
3251 h->stat.i_wpred[0] += !!h->sh.weight[0][0].weightfn;
3252 h->stat.i_wpred[1] += !!h->sh.weight[0][1].weightfn || !!h->sh.weight[0][2].weightfn;
3254 if( h->sh.i_type == SLICE_TYPE_B )
3256 h->stat.i_direct_frames[ h->sh.b_direct_spatial_mv_pred ] ++;
3257 if( h->mb.b_direct_auto_write )
3259 //FIXME somewhat arbitrary time constants
3260 if( h->stat.i_direct_score[0] + h->stat.i_direct_score[1] > h->mb.i_mb_count )
3261 for( int i = 0; i < 2; i++ )
3262 h->stat.i_direct_score[i] = h->stat.i_direct_score[i] * 9/10;
3263 for( int i = 0; i < 2; i++ )
3264 h->stat.i_direct_score[i] += h->stat.frame.i_direct_score[i];
3268 h->stat.i_consecutive_bframes[h->fenc->i_bframes]++;
3270 psz_message[0] = '\0';
3271 double dur = h->fenc->f_duration;
3272 h->stat.f_frame_duration[h->sh.i_type] += dur;
3273 if( h->param.analyse.b_psnr )
3277 h->stat.frame.i_ssd[0],
3278 h->stat.frame.i_ssd[1],
3279 h->stat.frame.i_ssd[2],
3281 int luma_size = h->param.i_width * h->param.i_height;
3282 int chroma_size = CHROMA_SIZE( luma_size );
3283 pic_out->prop.f_psnr[0] = x264_psnr( ssd[0], luma_size );
3284 pic_out->prop.f_psnr[1] = x264_psnr( ssd[1], chroma_size );
3285 pic_out->prop.f_psnr[2] = x264_psnr( ssd[2], chroma_size );
3286 pic_out->prop.f_psnr_avg = x264_psnr( ssd[0] + ssd[1] + ssd[2], luma_size + chroma_size*2 );
3288 h->stat.f_ssd_global[h->sh.i_type] += dur * (ssd[0] + ssd[1] + ssd[2]);
3289 h->stat.f_psnr_average[h->sh.i_type] += dur * pic_out->prop.f_psnr_avg;
3290 h->stat.f_psnr_mean_y[h->sh.i_type] += dur * pic_out->prop.f_psnr[0];
3291 h->stat.f_psnr_mean_u[h->sh.i_type] += dur * pic_out->prop.f_psnr[1];
3292 h->stat.f_psnr_mean_v[h->sh.i_type] += dur * pic_out->prop.f_psnr[2];
3294 snprintf( psz_message, 80, " PSNR Y:%5.2f U:%5.2f V:%5.2f", pic_out->prop.f_psnr[0],
3295 pic_out->prop.f_psnr[1],
3296 pic_out->prop.f_psnr[2] );
3299 if( h->param.analyse.b_ssim )
3301 pic_out->prop.f_ssim = h->stat.frame.f_ssim / h->stat.frame.i_ssim_cnt;
3302 h->stat.f_ssim_mean_y[h->sh.i_type] += pic_out->prop.f_ssim * dur;
3303 snprintf( psz_message + strlen(psz_message), 80 - strlen(psz_message),
3304 " SSIM Y:%.5f", pic_out->prop.f_ssim );
3306 psz_message[79] = '\0';
3308 x264_log( h, X264_LOG_DEBUG,
3309 "frame=%4d QP=%.2f NAL=%d Slice:%c Poc:%-3d I:%-4d P:%-4d SKIP:%-4d size=%d bytes%s\n",
3311 h->fdec->f_qp_avg_aq,
3313 h->sh.i_type == SLICE_TYPE_I ? 'I' : (h->sh.i_type == SLICE_TYPE_P ? 'P' : 'B' ),
3315 h->stat.frame.i_mb_count_i,
3316 h->stat.frame.i_mb_count_p,
3317 h->stat.frame.i_mb_count_skip,
3321 // keep stats all in one place
3322 x264_thread_sync_stat( h->thread[0], h );
3323 // for the use of the next frame
3324 x264_thread_sync_stat( thread_current, h );
3326 #ifdef DEBUG_MB_TYPE
3328 static const char mb_chars[] = { 'i', 'i', 'I', 'C', 'P', '8', 'S',
3329 'D', '<', 'X', 'B', 'X', '>', 'B', 'B', 'B', 'B', '8', 'S' };
3330 for( int mb_xy = 0; mb_xy < h->mb.i_mb_width * h->mb.i_mb_height; mb_xy++ )
3332 if( h->mb.type[mb_xy] < X264_MBTYPE_MAX && h->mb.type[mb_xy] >= 0 )
3333 fprintf( stderr, "%c ", mb_chars[ h->mb.type[mb_xy] ] );
3335 fprintf( stderr, "? " );
3337 if( (mb_xy+1) % h->mb.i_mb_width == 0 )
3338 fprintf( stderr, "\n" );
3343 /* Remove duplicates, must be done near the end as breaks h->fref0 array
3344 * by freeing some of its pointers. */
3345 for( int i = 0; i < h->i_ref[0]; i++ )
3346 if( h->fref[0][i] && h->fref[0][i]->b_duplicate )
3348 x264_frame_push_blank_unused( h, h->fref[0][i] );
3352 if( h->param.psz_dump_yuv )
3353 x264_frame_dump( h );
3359 static void x264_print_intra( int64_t *i_mb_count, double i_count, int b_print_pcm, char *intra )
3361 intra += sprintf( intra, "I16..4%s: %4.1f%% %4.1f%% %4.1f%%",
3362 b_print_pcm ? "..PCM" : "",
3363 i_mb_count[I_16x16]/ i_count,
3364 i_mb_count[I_8x8] / i_count,
3365 i_mb_count[I_4x4] / i_count );
3367 sprintf( intra, " %4.1f%%", i_mb_count[I_PCM] / i_count );
3370 /****************************************************************************
3371 * x264_encoder_close:
3372 ****************************************************************************/
3373 void x264_encoder_close ( x264_t *h )
3375 int64_t i_yuv_size = FRAME_SIZE( h->param.i_width * h->param.i_height );
3376 int64_t i_mb_count_size[2][7] = {{0}};
3378 int b_print_pcm = h->stat.i_mb_count[SLICE_TYPE_I][I_PCM]
3379 || h->stat.i_mb_count[SLICE_TYPE_P][I_PCM]
3380 || h->stat.i_mb_count[SLICE_TYPE_B][I_PCM];
3382 x264_lookahead_delete( h );
3384 if( h->param.i_threads > 1 )
3385 x264_threadpool_delete( h->threadpool );
3386 if( h->i_thread_frames > 1 )
3388 for( int i = 0; i < h->i_thread_frames; i++ )
3389 if( h->thread[i]->b_thread_active )
3391 assert( h->thread[i]->fenc->i_reference_count == 1 );
3392 x264_frame_delete( h->thread[i]->fenc );
3395 x264_t *thread_prev = h->thread[h->i_thread_phase];
3396 x264_thread_sync_ratecontrol( h, thread_prev, h );
3397 x264_thread_sync_ratecontrol( thread_prev, thread_prev, h );
3398 h->i_frame = thread_prev->i_frame + 1 - h->i_thread_frames;
3402 /* Slices used and PSNR */
3403 for( int i = 0; i < 3; i++ )
3405 static const uint8_t slice_order[] = { SLICE_TYPE_I, SLICE_TYPE_P, SLICE_TYPE_B };
3406 int i_slice = slice_order[i];
3408 if( h->stat.i_frame_count[i_slice] > 0 )
3410 int i_count = h->stat.i_frame_count[i_slice];
3411 double dur = h->stat.f_frame_duration[i_slice];
3412 if( h->param.analyse.b_psnr )
3414 x264_log( h, X264_LOG_INFO,
3415 "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",
3416 slice_type_to_char[i_slice],
3418 h->stat.f_frame_qp[i_slice] / i_count,
3419 (double)h->stat.i_frame_size[i_slice] / i_count,
3420 h->stat.f_psnr_mean_y[i_slice] / dur, h->stat.f_psnr_mean_u[i_slice] / dur, h->stat.f_psnr_mean_v[i_slice] / dur,
3421 h->stat.f_psnr_average[i_slice] / dur,
3422 x264_psnr( h->stat.f_ssd_global[i_slice], dur * i_yuv_size ) );
3426 x264_log( h, X264_LOG_INFO,
3427 "frame %c:%-5d Avg QP:%5.2f size:%6.0f\n",
3428 slice_type_to_char[i_slice],
3430 h->stat.f_frame_qp[i_slice] / i_count,
3431 (double)h->stat.i_frame_size[i_slice] / i_count );
3435 if( h->param.i_bframe && h->stat.i_frame_count[SLICE_TYPE_B] )
3439 // weight by number of frames (including the I/P-frames) that are in a sequence of N B-frames
3440 for( int i = 0; i <= h->param.i_bframe; i++ )
3441 den += (i+1) * h->stat.i_consecutive_bframes[i];
3442 for( int i = 0; i <= h->param.i_bframe; i++ )
3443 p += sprintf( p, " %4.1f%%", 100. * (i+1) * h->stat.i_consecutive_bframes[i] / den );
3444 x264_log( h, X264_LOG_INFO, "consecutive B-frames:%s\n", buf );
3447 for( int i_type = 0; i_type < 2; i_type++ )
3448 for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
3450 if( i == D_DIRECT_8x8 ) continue; /* direct is counted as its own type */
3451 i_mb_count_size[i_type][x264_mb_partition_pixel_table[i]] += h->stat.i_mb_partition[i_type][i];
3455 if( h->stat.i_frame_count[SLICE_TYPE_I] > 0 )
3457 int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_I];
3458 double i_count = h->stat.i_frame_count[SLICE_TYPE_I] * h->mb.i_mb_count / 100.0;
3459 x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
3460 x264_log( h, X264_LOG_INFO, "mb I %s\n", buf );
3462 if( h->stat.i_frame_count[SLICE_TYPE_P] > 0 )
3464 int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_P];
3465 double i_count = h->stat.i_frame_count[SLICE_TYPE_P] * h->mb.i_mb_count / 100.0;
3466 int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_P];
3467 x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
3468 x264_log( h, X264_LOG_INFO,
3469 "mb P %s P16..4: %4.1f%% %4.1f%% %4.1f%% %4.1f%% %4.1f%% skip:%4.1f%%\n",
3471 i_mb_size[PIXEL_16x16] / (i_count*4),
3472 (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
3473 i_mb_size[PIXEL_8x8] / (i_count*4),
3474 (i_mb_size[PIXEL_8x4] + i_mb_size[PIXEL_4x8]) / (i_count*4),
3475 i_mb_size[PIXEL_4x4] / (i_count*4),
3476 i_mb_count[P_SKIP] / i_count );
3478 if( h->stat.i_frame_count[SLICE_TYPE_B] > 0 )
3480 int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_B];
3481 double i_count = h->stat.i_frame_count[SLICE_TYPE_B] * h->mb.i_mb_count / 100.0;
3482 double i_mb_list_count;
3483 int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_B];
3484 int64_t list_count[3] = {0}; /* 0 == L0, 1 == L1, 2 == BI */
3485 x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
3486 for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
3487 for( int j = 0; j < 2; j++ )
3489 int l0 = x264_mb_type_list_table[i][0][j];
3490 int l1 = x264_mb_type_list_table[i][1][j];
3492 list_count[l1+l0*l1] += h->stat.i_mb_count[SLICE_TYPE_B][i] * 2;
3494 list_count[0] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L0_8x8];
3495 list_count[1] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L1_8x8];
3496 list_count[2] += h->stat.i_mb_partition[SLICE_TYPE_B][D_BI_8x8];
3497 i_mb_count[B_DIRECT] += (h->stat.i_mb_partition[SLICE_TYPE_B][D_DIRECT_8x8]+2)/4;
3498 i_mb_list_count = (list_count[0] + list_count[1] + list_count[2]) / 100.0;
3499 sprintf( buf + strlen(buf), " B16..8: %4.1f%% %4.1f%% %4.1f%% direct:%4.1f%% skip:%4.1f%%",
3500 i_mb_size[PIXEL_16x16] / (i_count*4),
3501 (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
3502 i_mb_size[PIXEL_8x8] / (i_count*4),
3503 i_mb_count[B_DIRECT] / i_count,
3504 i_mb_count[B_SKIP] / i_count );
3505 if( i_mb_list_count != 0 )
3506 sprintf( buf + strlen(buf), " L0:%4.1f%% L1:%4.1f%% BI:%4.1f%%",
3507 list_count[0] / i_mb_list_count,
3508 list_count[1] / i_mb_list_count,
3509 list_count[2] / i_mb_list_count );
3510 x264_log( h, X264_LOG_INFO, "mb B %s\n", buf );
3513 x264_ratecontrol_summary( h );
3515 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 )
3517 #define SUM3(p) (p[SLICE_TYPE_I] + p[SLICE_TYPE_P] + p[SLICE_TYPE_B])
3518 #define SUM3b(p,o) (p[SLICE_TYPE_I][o] + p[SLICE_TYPE_P][o] + p[SLICE_TYPE_B][o])
3519 int64_t i_i8x8 = SUM3b( h->stat.i_mb_count, I_8x8 );
3520 int64_t i_intra = i_i8x8 + SUM3b( h->stat.i_mb_count, I_4x4 )
3521 + SUM3b( h->stat.i_mb_count, I_16x16 );
3522 int64_t i_all_intra = i_intra + SUM3b( h->stat.i_mb_count, I_PCM);
3523 int64_t i_skip = SUM3b( h->stat.i_mb_count, P_SKIP )
3524 + SUM3b( h->stat.i_mb_count, B_SKIP );
3525 const int i_count = h->stat.i_frame_count[SLICE_TYPE_I] +
3526 h->stat.i_frame_count[SLICE_TYPE_P] +
3527 h->stat.i_frame_count[SLICE_TYPE_B];
3528 int64_t i_mb_count = (int64_t)i_count * h->mb.i_mb_count;
3529 int64_t i_inter = i_mb_count - i_skip - i_intra;
3530 const double duration = h->stat.f_frame_duration[SLICE_TYPE_I] +
3531 h->stat.f_frame_duration[SLICE_TYPE_P] +
3532 h->stat.f_frame_duration[SLICE_TYPE_B];
3533 float f_bitrate = SUM3(h->stat.i_frame_size) / duration / 125;
3535 if( PARAM_INTERLACED )
3537 char *fieldstats = buf;
3540 fieldstats += sprintf( fieldstats, " inter:%.1f%%", h->stat.i_mb_field[1] * 100.0 / i_inter );
3542 fieldstats += sprintf( fieldstats, " skip:%.1f%%", h->stat.i_mb_field[2] * 100.0 / i_skip );
3543 x264_log( h, X264_LOG_INFO, "field mbs: intra: %.1f%%%s\n",
3544 h->stat.i_mb_field[0] * 100.0 / i_intra, buf );
3547 if( h->pps->b_transform_8x8_mode )
3550 if( h->stat.i_mb_count_8x8dct[0] )
3551 sprintf( buf, " inter:%.1f%%", 100. * h->stat.i_mb_count_8x8dct[1] / h->stat.i_mb_count_8x8dct[0] );
3552 x264_log( h, X264_LOG_INFO, "8x8 transform intra:%.1f%%%s\n", 100. * i_i8x8 / i_intra, buf );
3555 if( (h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO ||
3556 (h->stat.i_direct_frames[0] && h->stat.i_direct_frames[1]))
3557 && h->stat.i_frame_count[SLICE_TYPE_B] )
3559 x264_log( h, X264_LOG_INFO, "direct mvs spatial:%.1f%% temporal:%.1f%%\n",
3560 h->stat.i_direct_frames[1] * 100. / h->stat.i_frame_count[SLICE_TYPE_B],
3561 h->stat.i_direct_frames[0] * 100. / h->stat.i_frame_count[SLICE_TYPE_B] );
3565 int csize = CHROMA444 ? 4 : 1;
3566 if( i_mb_count != i_all_intra )
3567 sprintf( buf, " inter: %.1f%% %.1f%% %.1f%%",
3568 h->stat.i_mb_cbp[1] * 100.0 / ((i_mb_count - i_all_intra)*4),
3569 h->stat.i_mb_cbp[3] * 100.0 / ((i_mb_count - i_all_intra)*csize),
3570 h->stat.i_mb_cbp[5] * 100.0 / ((i_mb_count - i_all_intra)*csize) );
3571 x264_log( h, X264_LOG_INFO, "coded y,%s,%s intra: %.1f%% %.1f%% %.1f%%%s\n",
3572 CHROMA444?"u":"uvDC", CHROMA444?"v":"uvAC",
3573 h->stat.i_mb_cbp[0] * 100.0 / (i_all_intra*4),
3574 h->stat.i_mb_cbp[2] * 100.0 / (i_all_intra*csize),
3575 h->stat.i_mb_cbp[4] * 100.0 / (i_all_intra*csize), buf );
3577 int64_t fixed_pred_modes[4][9] = {{0}};
3578 int64_t sum_pred_modes[4] = {0};
3579 for( int i = 0; i <= I_PRED_16x16_DC_128; i++ )
3581 fixed_pred_modes[0][x264_mb_pred_mode16x16_fix[i]] += h->stat.i_mb_pred_mode[0][i];
3582 sum_pred_modes[0] += h->stat.i_mb_pred_mode[0][i];
3584 if( sum_pred_modes[0] )
3585 x264_log( h, X264_LOG_INFO, "i16 v,h,dc,p: %2.0f%% %2.0f%% %2.0f%% %2.0f%%\n",
3586 fixed_pred_modes[0][0] * 100.0 / sum_pred_modes[0],
3587 fixed_pred_modes[0][1] * 100.0 / sum_pred_modes[0],
3588 fixed_pred_modes[0][2] * 100.0 / sum_pred_modes[0],
3589 fixed_pred_modes[0][3] * 100.0 / sum_pred_modes[0] );
3590 for( int i = 1; i <= 2; i++ )
3592 for( int j = 0; j <= I_PRED_8x8_DC_128; j++ )
3594 fixed_pred_modes[i][x264_mb_pred_mode4x4_fix(j)] += h->stat.i_mb_pred_mode[i][j];
3595 sum_pred_modes[i] += h->stat.i_mb_pred_mode[i][j];
3597 if( sum_pred_modes[i] )
3598 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,
3599 fixed_pred_modes[i][0] * 100.0 / sum_pred_modes[i],
3600 fixed_pred_modes[i][1] * 100.0 / sum_pred_modes[i],
3601 fixed_pred_modes[i][2] * 100.0 / sum_pred_modes[i],
3602 fixed_pred_modes[i][3] * 100.0 / sum_pred_modes[i],
3603 fixed_pred_modes[i][4] * 100.0 / sum_pred_modes[i],
3604 fixed_pred_modes[i][5] * 100.0 / sum_pred_modes[i],
3605 fixed_pred_modes[i][6] * 100.0 / sum_pred_modes[i],
3606 fixed_pred_modes[i][7] * 100.0 / sum_pred_modes[i],
3607 fixed_pred_modes[i][8] * 100.0 / sum_pred_modes[i] );
3609 for( int i = 0; i <= I_PRED_CHROMA_DC_128; i++ )
3611 fixed_pred_modes[3][x264_mb_chroma_pred_mode_fix[i]] += h->stat.i_mb_pred_mode[3][i];
3612 sum_pred_modes[3] += h->stat.i_mb_pred_mode[3][i];
3614 if( sum_pred_modes[3] && !CHROMA444 )
3615 x264_log( h, X264_LOG_INFO, "i8c dc,h,v,p: %2.0f%% %2.0f%% %2.0f%% %2.0f%%\n",
3616 fixed_pred_modes[3][0] * 100.0 / sum_pred_modes[3],
3617 fixed_pred_modes[3][1] * 100.0 / sum_pred_modes[3],
3618 fixed_pred_modes[3][2] * 100.0 / sum_pred_modes[3],
3619 fixed_pred_modes[3][3] * 100.0 / sum_pred_modes[3] );
3621 if( h->param.analyse.i_weighted_pred >= X264_WEIGHTP_SIMPLE && h->stat.i_frame_count[SLICE_TYPE_P] > 0 )
3622 x264_log( h, X264_LOG_INFO, "Weighted P-Frames: Y:%.1f%% UV:%.1f%%\n",
3623 h->stat.i_wpred[0] * 100.0 / h->stat.i_frame_count[SLICE_TYPE_P],
3624 h->stat.i_wpred[1] * 100.0 / h->stat.i_frame_count[SLICE_TYPE_P] );
3626 for( int i_list = 0; i_list < 2; i_list++ )
3627 for( int i_slice = 0; i_slice < 2; i_slice++ )
3632 for( int i = 0; i < X264_REF_MAX*2; i++ )
3633 if( h->stat.i_mb_count_ref[i_slice][i_list][i] )
3635 i_den += h->stat.i_mb_count_ref[i_slice][i_list][i];
3640 for( int i = 0; i <= i_max; i++ )
3641 p += sprintf( p, " %4.1f%%", 100. * h->stat.i_mb_count_ref[i_slice][i_list][i] / i_den );
3642 x264_log( h, X264_LOG_INFO, "ref %c L%d:%s\n", "PB"[i_slice], i_list, buf );
3645 if( h->param.analyse.b_ssim )
3647 float ssim = SUM3( h->stat.f_ssim_mean_y ) / duration;
3648 x264_log( h, X264_LOG_INFO, "SSIM Mean Y:%.7f (%6.3fdb)\n", ssim, x264_ssim( ssim ) );
3650 if( h->param.analyse.b_psnr )
3652 x264_log( h, X264_LOG_INFO,
3653 "PSNR Mean Y:%6.3f U:%6.3f V:%6.3f Avg:%6.3f Global:%6.3f kb/s:%.2f\n",
3654 SUM3( h->stat.f_psnr_mean_y ) / duration,
3655 SUM3( h->stat.f_psnr_mean_u ) / duration,
3656 SUM3( h->stat.f_psnr_mean_v ) / duration,
3657 SUM3( h->stat.f_psnr_average ) / duration,
3658 x264_psnr( SUM3( h->stat.f_ssd_global ), duration * i_yuv_size ),
3662 x264_log( h, X264_LOG_INFO, "kb/s:%.2f\n", f_bitrate );
3666 x264_ratecontrol_delete( h );
3669 if( h->param.rc.psz_stat_out )
3670 free( h->param.rc.psz_stat_out );
3671 if( h->param.rc.psz_stat_in )
3672 free( h->param.rc.psz_stat_in );
3674 x264_cqm_delete( h );
3675 x264_free( h->nal_buffer );
3676 x264_analyse_free_costs( h );
3678 if( h->i_thread_frames > 1)
3679 h = h->thread[h->i_thread_phase];
3682 x264_frame_delete_list( h->frames.unused[0] );
3683 x264_frame_delete_list( h->frames.unused[1] );
3684 x264_frame_delete_list( h->frames.current );
3685 x264_frame_delete_list( h->frames.blank_unused );
3689 for( int i = 0; i < h->i_thread_frames; i++ )
3690 if( h->thread[i]->b_thread_active )
3691 for( int j = 0; j < h->thread[i]->i_ref[0]; j++ )
3692 if( h->thread[i]->fref[0][j] && h->thread[i]->fref[0][j]->b_duplicate )
3693 x264_frame_delete( h->thread[i]->fref[0][j] );
3695 for( int i = h->param.i_threads - 1; i >= 0; i-- )
3697 x264_frame_t **frame;
3699 if( !h->param.b_sliced_threads || i == 0 )
3701 for( frame = h->thread[i]->frames.reference; *frame; frame++ )
3703 assert( (*frame)->i_reference_count > 0 );
3704 (*frame)->i_reference_count--;
3705 if( (*frame)->i_reference_count == 0 )
3706 x264_frame_delete( *frame );
3708 frame = &h->thread[i]->fdec;
3711 assert( (*frame)->i_reference_count > 0 );
3712 (*frame)->i_reference_count--;
3713 if( (*frame)->i_reference_count == 0 )
3714 x264_frame_delete( *frame );
3716 x264_macroblock_cache_free( h->thread[i] );
3718 x264_macroblock_thread_free( h->thread[i], 0 );
3719 x264_free( h->thread[i]->out.p_bitstream );
3720 x264_free( h->thread[i]->out.nal);
3721 x264_free( h->thread[i] );
3725 int x264_encoder_delayed_frames( x264_t *h )
3727 int delayed_frames = 0;
3728 if( h->i_thread_frames > 1 )
3730 for( int i = 0; i < h->i_thread_frames; i++ )
3731 delayed_frames += h->thread[i]->b_thread_active;
3732 h = h->thread[h->i_thread_phase];
3734 for( int i = 0; h->frames.current[i]; i++ )
3736 x264_pthread_mutex_lock( &h->lookahead->ofbuf.mutex );
3737 x264_pthread_mutex_lock( &h->lookahead->ifbuf.mutex );
3738 x264_pthread_mutex_lock( &h->lookahead->next.mutex );
3739 delayed_frames += h->lookahead->ifbuf.i_size + h->lookahead->next.i_size + h->lookahead->ofbuf.i_size;
3740 x264_pthread_mutex_unlock( &h->lookahead->next.mutex );
3741 x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
3742 x264_pthread_mutex_unlock( &h->lookahead->ofbuf.mutex );
3743 return delayed_frames;
3746 int x264_encoder_maximum_delayed_frames( x264_t *h )
3748 return h->frames.i_delay;