-
Notifications
You must be signed in to change notification settings - Fork 54
/
firstpass.c
3480 lines (2988 loc) · 126 KB
/
firstpass.c
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
/*
* Copyright (c) 2016, Alliance for Open Media. All rights reserved
*
* This source code is subject to the terms of the BSD 2 Clause License and
* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
* was not distributed with this source code in the LICENSE file, you can
* obtain it at www.aomedia.org/license/software. If the Alliance for Open
* Media Patent License 1.0 was not distributed with this source code in the
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
#include <limits.h>
#include <math.h>
#include <stdio.h>
#include "config/aom_dsp_rtcd.h"
#include "config/aom_scale_rtcd.h"
#include "aom_dsp/aom_dsp_common.h"
#include "aom_mem/aom_mem.h"
#include "aom_ports/mem.h"
#include "aom_ports/system_state.h"
#include "aom_scale/aom_scale.h"
#include "aom_scale/yv12config.h"
#include "aom_dsp/variance.h"
#include "av1/common/entropymv.h"
#include "av1/common/quant_common.h"
#include "av1/common/reconinter.h" // av1_setup_dst_planes()
#include "av1/common/txb_common.h"
#include "av1/encoder/aq_variance.h"
#include "av1/encoder/av1_quantize.h"
#include "av1/encoder/block.h"
#include "av1/encoder/dwt.h"
#include "av1/encoder/encodeframe.h"
#include "av1/encoder/encodemb.h"
#include "av1/encoder/encodemv.h"
#include "av1/encoder/encoder.h"
#include "av1/encoder/extend.h"
#include "av1/encoder/firstpass.h"
#include "av1/encoder/mcomp.h"
#include "av1/encoder/rd.h"
#include "av1/encoder/reconinter_enc.h"
#define OUTPUT_FPF 0
#define ARF_STATS_OUTPUT 0
#define GROUP_ADAPTIVE_MAXQ 1
#define BOOST_BREAKOUT 12.5
#define BOOST_FACTOR 12.5
#define FACTOR_PT_LOW 0.70
#define FACTOR_PT_HIGH 0.90
#define FIRST_PASS_Q 10.0
#define GF_MAX_BOOST 90.0
#define INTRA_MODE_PENALTY 1024
#define KF_MIN_FRAME_BOOST 80.0
#define KF_MAX_FRAME_BOOST 128.0
#define MIN_ARF_GF_BOOST 240
#define MIN_DECAY_FACTOR 0.01
#define MIN_KF_BOOST 300
#define NEW_MV_MODE_PENALTY 32
#define DARK_THRESH 64
#define DEFAULT_GRP_WEIGHT 1.0
#define RC_FACTOR_MIN 0.75
#define RC_FACTOR_MAX 1.75
#define MIN_FWD_KF_INTERVAL 8
#define NCOUNT_INTRA_THRESH 8192
#define NCOUNT_INTRA_FACTOR 3
#define NCOUNT_FRAME_II_THRESH 5.0
#define DOUBLE_DIVIDE_CHECK(x) ((x) < 0 ? (x)-0.000001 : (x) + 0.000001)
#if ARF_STATS_OUTPUT
unsigned int arf_count = 0;
#endif
// Resets the first pass file to the given position using a relative seek from
// the current position.
static void reset_fpf_position(TWO_PASS *p, const FIRSTPASS_STATS *position) {
p->stats_in = position;
}
// Read frame stats at an offset from the current position.
static const FIRSTPASS_STATS *read_frame_stats(const TWO_PASS *p, int offset) {
if ((offset >= 0 && p->stats_in + offset >= p->stats_in_end) ||
(offset < 0 && p->stats_in + offset < p->stats_in_start)) {
return NULL;
}
return &p->stats_in[offset];
}
static int input_stats(TWO_PASS *p, FIRSTPASS_STATS *fps) {
if (p->stats_in >= p->stats_in_end) return EOF;
*fps = *p->stats_in;
++p->stats_in;
return 1;
}
static void output_stats(FIRSTPASS_STATS *stats,
struct aom_codec_pkt_list *pktlist) {
struct aom_codec_cx_pkt pkt;
pkt.kind = AOM_CODEC_STATS_PKT;
pkt.data.twopass_stats.buf = stats;
pkt.data.twopass_stats.sz = sizeof(FIRSTPASS_STATS);
aom_codec_pkt_list_add(pktlist, &pkt);
// TEMP debug code
#if OUTPUT_FPF
{
FILE *fpfile;
fpfile = fopen("firstpass.stt", "a");
fprintf(fpfile,
"%12.0lf %12.4lf %12.0lf %12.0lf %12.0lf %12.4lf %12.4lf"
"%12.4lf %12.4lf %12.4lf %12.4lf %12.4lf %12.4lf %12.4lf %12.4lf"
"%12.4lf %12.4lf %12.0lf %12.0lf %12.0lf %12.4lf %12.4lf\n",
stats->frame, stats->weight, stats->intra_error, stats->coded_error,
stats->sr_coded_error, stats->pcnt_inter, stats->pcnt_motion,
stats->pcnt_second_ref, stats->pcnt_neutral, stats->intra_skip_pct,
stats->inactive_zone_rows, stats->inactive_zone_cols, stats->MVr,
stats->mvr_abs, stats->MVc, stats->mvc_abs, stats->MVrv,
stats->MVcv, stats->mv_in_out_count, stats->new_mv_count,
stats->count, stats->duration);
fclose(fpfile);
}
#endif
}
#if CONFIG_FP_MB_STATS
static void output_fpmb_stats(uint8_t *this_frame_mb_stats, int stats_size,
struct aom_codec_pkt_list *pktlist) {
struct aom_codec_cx_pkt pkt;
pkt.kind = AOM_CODEC_FPMB_STATS_PKT;
pkt.data.firstpass_mb_stats.buf = this_frame_mb_stats;
pkt.data.firstpass_mb_stats.sz = stats_size * sizeof(*this_frame_mb_stats);
aom_codec_pkt_list_add(pktlist, &pkt);
}
#endif
static void zero_stats(FIRSTPASS_STATS *section) {
section->frame = 0.0;
section->weight = 0.0;
section->intra_error = 0.0;
section->frame_avg_wavelet_energy = 0.0;
section->coded_error = 0.0;
section->sr_coded_error = 0.0;
section->pcnt_inter = 0.0;
section->pcnt_motion = 0.0;
section->pcnt_second_ref = 0.0;
section->pcnt_neutral = 0.0;
section->intra_skip_pct = 0.0;
section->inactive_zone_rows = 0.0;
section->inactive_zone_cols = 0.0;
section->MVr = 0.0;
section->mvr_abs = 0.0;
section->MVc = 0.0;
section->mvc_abs = 0.0;
section->MVrv = 0.0;
section->MVcv = 0.0;
section->mv_in_out_count = 0.0;
section->new_mv_count = 0.0;
section->count = 0.0;
section->duration = 1.0;
}
static void accumulate_stats(FIRSTPASS_STATS *section,
const FIRSTPASS_STATS *frame) {
section->frame += frame->frame;
section->weight += frame->weight;
section->intra_error += frame->intra_error;
section->frame_avg_wavelet_energy += frame->frame_avg_wavelet_energy;
section->coded_error += frame->coded_error;
section->sr_coded_error += frame->sr_coded_error;
section->pcnt_inter += frame->pcnt_inter;
section->pcnt_motion += frame->pcnt_motion;
section->pcnt_second_ref += frame->pcnt_second_ref;
section->pcnt_neutral += frame->pcnt_neutral;
section->intra_skip_pct += frame->intra_skip_pct;
section->inactive_zone_rows += frame->inactive_zone_rows;
section->inactive_zone_cols += frame->inactive_zone_cols;
section->MVr += frame->MVr;
section->mvr_abs += frame->mvr_abs;
section->MVc += frame->MVc;
section->mvc_abs += frame->mvc_abs;
section->MVrv += frame->MVrv;
section->MVcv += frame->MVcv;
section->mv_in_out_count += frame->mv_in_out_count;
section->new_mv_count += frame->new_mv_count;
section->count += frame->count;
section->duration += frame->duration;
}
static void subtract_stats(FIRSTPASS_STATS *section,
const FIRSTPASS_STATS *frame) {
section->frame -= frame->frame;
section->weight -= frame->weight;
section->intra_error -= frame->intra_error;
section->frame_avg_wavelet_energy -= frame->frame_avg_wavelet_energy;
section->coded_error -= frame->coded_error;
section->sr_coded_error -= frame->sr_coded_error;
section->pcnt_inter -= frame->pcnt_inter;
section->pcnt_motion -= frame->pcnt_motion;
section->pcnt_second_ref -= frame->pcnt_second_ref;
section->pcnt_neutral -= frame->pcnt_neutral;
section->intra_skip_pct -= frame->intra_skip_pct;
section->inactive_zone_rows -= frame->inactive_zone_rows;
section->inactive_zone_cols -= frame->inactive_zone_cols;
section->MVr -= frame->MVr;
section->mvr_abs -= frame->mvr_abs;
section->MVc -= frame->MVc;
section->mvc_abs -= frame->mvc_abs;
section->MVrv -= frame->MVrv;
section->MVcv -= frame->MVcv;
section->mv_in_out_count -= frame->mv_in_out_count;
section->new_mv_count -= frame->new_mv_count;
section->count -= frame->count;
section->duration -= frame->duration;
}
// Calculate the linear size relative to a baseline of 1080P
#define BASE_SIZE 2073600.0 // 1920x1080
static double get_linear_size_factor(const AV1_COMP *cpi) {
const double this_area = cpi->initial_width * cpi->initial_height;
return pow(this_area / BASE_SIZE, 0.5);
}
// Calculate an active area of the image that discounts formatting
// bars and partially discounts other 0 energy areas.
#define MIN_ACTIVE_AREA 0.5
#define MAX_ACTIVE_AREA 1.0
static double calculate_active_area(const AV1_COMP *cpi,
const FIRSTPASS_STATS *this_frame) {
double active_pct;
active_pct =
1.0 -
((this_frame->intra_skip_pct / 2) +
((this_frame->inactive_zone_rows * 2) / (double)cpi->common.mb_rows));
return fclamp(active_pct, MIN_ACTIVE_AREA, MAX_ACTIVE_AREA);
}
// Calculate a modified Error used in distributing bits between easier and
// harder frames.
#define ACT_AREA_CORRECTION 0.5
static double calculate_modified_err(const AV1_COMP *cpi,
const TWO_PASS *twopass,
const AV1EncoderConfig *oxcf,
const FIRSTPASS_STATS *this_frame) {
const FIRSTPASS_STATS *const stats = &twopass->total_stats;
const double av_weight = stats->weight / stats->count;
const double av_err = (stats->coded_error * av_weight) / stats->count;
double modified_error =
av_err * pow(this_frame->coded_error * this_frame->weight /
DOUBLE_DIVIDE_CHECK(av_err),
oxcf->two_pass_vbrbias / 100.0);
// Correction for active area. Frames with a reduced active area
// (eg due to formatting bars) have a higher error per mb for the
// remaining active MBs. The correction here assumes that coding
// 0.5N blocks of complexity 2X is a little easier than coding N
// blocks of complexity X.
modified_error *=
pow(calculate_active_area(cpi, this_frame), ACT_AREA_CORRECTION);
return fclamp(modified_error, twopass->modified_error_min,
twopass->modified_error_max);
}
// This function returns the maximum target rate per frame.
static int frame_max_bits(const RATE_CONTROL *rc,
const AV1EncoderConfig *oxcf) {
int64_t max_bits = ((int64_t)rc->avg_frame_bandwidth *
(int64_t)oxcf->two_pass_vbrmax_section) /
100;
if (max_bits < 0)
max_bits = 0;
else if (max_bits > rc->max_frame_bandwidth)
max_bits = rc->max_frame_bandwidth;
return (int)max_bits;
}
void av1_init_first_pass(AV1_COMP *cpi) {
zero_stats(&cpi->twopass.total_stats);
}
void av1_end_first_pass(AV1_COMP *cpi) {
output_stats(&cpi->twopass.total_stats, cpi->output_pkt_list);
}
static aom_variance_fn_t get_block_variance_fn(BLOCK_SIZE bsize) {
switch (bsize) {
case BLOCK_8X8: return aom_mse8x8;
case BLOCK_16X8: return aom_mse16x8;
case BLOCK_8X16: return aom_mse8x16;
default: return aom_mse16x16;
}
}
static unsigned int get_prediction_error(BLOCK_SIZE bsize,
const struct buf_2d *src,
const struct buf_2d *ref) {
unsigned int sse;
const aom_variance_fn_t fn = get_block_variance_fn(bsize);
fn(src->buf, src->stride, ref->buf, ref->stride, &sse);
return sse;
}
static aom_variance_fn_t highbd_get_block_variance_fn(BLOCK_SIZE bsize,
int bd) {
switch (bd) {
default:
switch (bsize) {
case BLOCK_8X8: return aom_highbd_8_mse8x8;
case BLOCK_16X8: return aom_highbd_8_mse16x8;
case BLOCK_8X16: return aom_highbd_8_mse8x16;
default: return aom_highbd_8_mse16x16;
}
break;
case 10:
switch (bsize) {
case BLOCK_8X8: return aom_highbd_10_mse8x8;
case BLOCK_16X8: return aom_highbd_10_mse16x8;
case BLOCK_8X16: return aom_highbd_10_mse8x16;
default: return aom_highbd_10_mse16x16;
}
break;
case 12:
switch (bsize) {
case BLOCK_8X8: return aom_highbd_12_mse8x8;
case BLOCK_16X8: return aom_highbd_12_mse16x8;
case BLOCK_8X16: return aom_highbd_12_mse8x16;
default: return aom_highbd_12_mse16x16;
}
break;
}
}
static unsigned int highbd_get_prediction_error(BLOCK_SIZE bsize,
const struct buf_2d *src,
const struct buf_2d *ref,
int bd) {
unsigned int sse;
const aom_variance_fn_t fn = highbd_get_block_variance_fn(bsize, bd);
fn(src->buf, src->stride, ref->buf, ref->stride, &sse);
return sse;
}
// Refine the motion search range according to the frame dimension
// for first pass test.
static int get_search_range(const AV1_COMP *cpi) {
int sr = 0;
const int dim = AOMMIN(cpi->initial_width, cpi->initial_height);
while ((dim << sr) < MAX_FULL_PEL_VAL) ++sr;
return sr;
}
static void first_pass_motion_search(AV1_COMP *cpi, MACROBLOCK *x,
const MV *ref_mv, MV *best_mv,
int *best_motion_err) {
MACROBLOCKD *const xd = &x->e_mbd;
MV tmp_mv = kZeroMv;
MV ref_mv_full = { ref_mv->row >> 3, ref_mv->col >> 3 };
int num00, tmp_err, n;
const BLOCK_SIZE bsize = xd->mi[0]->sb_type;
aom_variance_fn_ptr_t v_fn_ptr = cpi->fn_ptr[bsize];
const int new_mv_mode_penalty = NEW_MV_MODE_PENALTY;
int step_param = 3;
int further_steps = (MAX_MVSEARCH_STEPS - 1) - step_param;
const int sr = get_search_range(cpi);
step_param += sr;
further_steps -= sr;
// Override the default variance function to use MSE.
v_fn_ptr.vf = get_block_variance_fn(bsize);
if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
v_fn_ptr.vf = highbd_get_block_variance_fn(bsize, xd->bd);
}
// Center the initial step/diamond search on best mv.
tmp_err = cpi->diamond_search_sad(x, &cpi->ss_cfg, &ref_mv_full, &tmp_mv,
step_param, x->sadperbit16, &num00,
&v_fn_ptr, ref_mv);
if (tmp_err < INT_MAX)
tmp_err = av1_get_mvpred_var(x, &tmp_mv, ref_mv, &v_fn_ptr, 1);
if (tmp_err < INT_MAX - new_mv_mode_penalty) tmp_err += new_mv_mode_penalty;
if (tmp_err < *best_motion_err) {
*best_motion_err = tmp_err;
*best_mv = tmp_mv;
}
// Carry out further step/diamond searches as necessary.
n = num00;
num00 = 0;
while (n < further_steps) {
++n;
if (num00) {
--num00;
} else {
tmp_err = cpi->diamond_search_sad(x, &cpi->ss_cfg, &ref_mv_full, &tmp_mv,
step_param + n, x->sadperbit16, &num00,
&v_fn_ptr, ref_mv);
if (tmp_err < INT_MAX)
tmp_err = av1_get_mvpred_var(x, &tmp_mv, ref_mv, &v_fn_ptr, 1);
if (tmp_err < INT_MAX - new_mv_mode_penalty)
tmp_err += new_mv_mode_penalty;
if (tmp_err < *best_motion_err) {
*best_motion_err = tmp_err;
*best_mv = tmp_mv;
}
}
}
}
static BLOCK_SIZE get_bsize(const AV1_COMMON *cm, int mb_row, int mb_col) {
if (mi_size_wide[BLOCK_16X16] * mb_col + mi_size_wide[BLOCK_8X8] <
cm->mi_cols) {
return mi_size_wide[BLOCK_16X16] * mb_row + mi_size_wide[BLOCK_8X8] <
cm->mi_rows
? BLOCK_16X16
: BLOCK_16X8;
} else {
return mi_size_wide[BLOCK_16X16] * mb_row + mi_size_wide[BLOCK_8X8] <
cm->mi_rows
? BLOCK_8X16
: BLOCK_8X8;
}
}
static int find_fp_qindex(aom_bit_depth_t bit_depth) {
int i;
for (i = 0; i < QINDEX_RANGE; ++i)
if (av1_convert_qindex_to_q(i, bit_depth) >= FIRST_PASS_Q) break;
if (i == QINDEX_RANGE) i--;
return i;
}
static void set_first_pass_params(AV1_COMP *cpi) {
AV1_COMMON *const cm = &cpi->common;
if (!cpi->refresh_alt_ref_frame &&
(cm->current_video_frame == 0 || (cpi->frame_flags & FRAMEFLAGS_KEY))) {
cm->frame_type = KEY_FRAME;
} else {
cm->frame_type = INTER_FRAME;
}
// Do not use periodic key frames.
cpi->rc.frames_to_key = INT_MAX;
}
static double raw_motion_error_stdev(int *raw_motion_err_list,
int raw_motion_err_counts) {
int64_t sum_raw_err = 0;
double raw_err_avg = 0;
double raw_err_stdev = 0;
if (raw_motion_err_counts == 0) return 0;
int i;
for (i = 0; i < raw_motion_err_counts; i++) {
sum_raw_err += raw_motion_err_list[i];
}
raw_err_avg = (double)sum_raw_err / raw_motion_err_counts;
for (i = 0; i < raw_motion_err_counts; i++) {
raw_err_stdev += (raw_motion_err_list[i] - raw_err_avg) *
(raw_motion_err_list[i] - raw_err_avg);
}
// Calculate the standard deviation for the motion error of all the inter
// blocks of the 0,0 motion using the last source
// frame as the reference.
raw_err_stdev = sqrt(raw_err_stdev / raw_motion_err_counts);
return raw_err_stdev;
}
#define UL_INTRA_THRESH 50
#define INVALID_ROW -1
void av1_first_pass(AV1_COMP *cpi, const struct lookahead_entry *source) {
int mb_row, mb_col;
MACROBLOCK *const x = &cpi->td.mb;
AV1_COMMON *const cm = &cpi->common;
const SequenceHeader *const seq_params = &cm->seq_params;
const int num_planes = av1_num_planes(cm);
MACROBLOCKD *const xd = &x->e_mbd;
TileInfo tile;
struct macroblock_plane *const p = x->plane;
struct macroblockd_plane *const pd = xd->plane;
const PICK_MODE_CONTEXT *ctx =
&cpi->td.pc_root[MAX_MIB_SIZE_LOG2 - MIN_MIB_SIZE_LOG2]->none;
int i;
int recon_yoffset, recon_uvoffset;
int64_t intra_error = 0;
int64_t frame_avg_wavelet_energy = 0;
int64_t coded_error = 0;
int64_t sr_coded_error = 0;
int sum_mvr = 0, sum_mvc = 0;
int sum_mvr_abs = 0, sum_mvc_abs = 0;
int64_t sum_mvrs = 0, sum_mvcs = 0;
int mvcount = 0;
int intercount = 0;
int second_ref_count = 0;
const int intrapenalty = INTRA_MODE_PENALTY;
double neutral_count;
int intra_skip_count = 0;
int image_data_start_row = INVALID_ROW;
int new_mv_count = 0;
int sum_in_vectors = 0;
MV lastmv = kZeroMv;
TWO_PASS *twopass = &cpi->twopass;
int recon_y_stride, recon_uv_stride, uv_mb_height;
YV12_BUFFER_CONFIG *const lst_yv12 = get_ref_frame_buffer(cpi, LAST_FRAME);
YV12_BUFFER_CONFIG *gld_yv12 = get_ref_frame_buffer(cpi, GOLDEN_FRAME);
YV12_BUFFER_CONFIG *const new_yv12 = get_frame_new_buffer(cm);
const YV12_BUFFER_CONFIG *first_ref_buf = lst_yv12;
double intra_factor;
double brightness_factor;
BufferPool *const pool = cm->buffer_pool;
const int qindex = find_fp_qindex(seq_params->bit_depth);
const int mb_scale = mi_size_wide[BLOCK_16X16];
int *raw_motion_err_list;
int raw_motion_err_counts = 0;
CHECK_MEM_ERROR(
cm, raw_motion_err_list,
aom_calloc(cm->mb_rows * cm->mb_cols, sizeof(*raw_motion_err_list)));
// First pass code requires valid last and new frame buffers.
assert(new_yv12 != NULL);
assert(frame_is_intra_only(cm) || (lst_yv12 != NULL));
#if CONFIG_FP_MB_STATS
if (cpi->use_fp_mb_stats) {
av1_zero_array(cpi->twopass.frame_mb_stats_buf, cpi->initial_mbs);
}
#endif
aom_clear_system_state();
xd->mi = cm->mi_grid_visible;
xd->mi[0] = cm->mi;
x->e_mbd.mi[0]->sb_type = BLOCK_16X16;
intra_factor = 0.0;
brightness_factor = 0.0;
neutral_count = 0.0;
set_first_pass_params(cpi);
av1_set_quantizer(cm, qindex);
av1_setup_block_planes(&x->e_mbd, seq_params->subsampling_x,
seq_params->subsampling_y, num_planes);
av1_setup_src_planes(x, cpi->source, 0, 0, num_planes);
av1_setup_dst_planes(xd->plane, seq_params->sb_size, new_yv12, 0, 0, 0,
num_planes);
if (!frame_is_intra_only(cm)) {
av1_setup_pre_planes(xd, 0, first_ref_buf, 0, 0, NULL, num_planes);
}
xd->mi = cm->mi_grid_visible;
xd->mi[0] = cm->mi;
// Don't store luma on the fist pass since chroma is not computed
xd->cfl.store_y = 0;
av1_frame_init_quantizer(cpi);
for (i = 0; i < num_planes; ++i) {
p[i].coeff = ctx->coeff[i];
p[i].qcoeff = ctx->qcoeff[i];
pd[i].dqcoeff = ctx->dqcoeff[i];
p[i].eobs = ctx->eobs[i];
p[i].txb_entropy_ctx = ctx->txb_entropy_ctx[i];
}
av1_init_mv_probs(cm);
av1_init_lv_map(cm);
av1_initialize_rd_consts(cpi);
// Tiling is ignored in the first pass.
av1_tile_init(&tile, cm, 0, 0);
recon_y_stride = new_yv12->y_stride;
recon_uv_stride = new_yv12->uv_stride;
uv_mb_height = 16 >> (new_yv12->y_height > new_yv12->uv_height);
for (mb_row = 0; mb_row < cm->mb_rows; ++mb_row) {
MV best_ref_mv = kZeroMv;
// Reset above block coeffs.
xd->up_available = (mb_row != 0);
recon_yoffset = (mb_row * recon_y_stride * 16);
recon_uvoffset = (mb_row * recon_uv_stride * uv_mb_height);
// Set up limit values for motion vectors to prevent them extending
// outside the UMV borders.
x->mv_limits.row_min = -((mb_row * 16) + BORDER_MV_PIXELS_B16);
x->mv_limits.row_max =
((cm->mb_rows - 1 - mb_row) * 16) + BORDER_MV_PIXELS_B16;
for (mb_col = 0; mb_col < cm->mb_cols; ++mb_col) {
int this_error;
const int use_dc_pred = (mb_col || mb_row) && (!mb_col || !mb_row);
const BLOCK_SIZE bsize = get_bsize(cm, mb_row, mb_col);
double log_intra;
int level_sample;
#if CONFIG_FP_MB_STATS
const int mb_index = mb_row * cm->mb_cols + mb_col;
#endif
aom_clear_system_state();
const int idx_str = xd->mi_stride * mb_row * mb_scale + mb_col * mb_scale;
xd->mi = cm->mi_grid_visible + idx_str;
xd->mi[0] = cm->mi + idx_str;
xd->plane[0].dst.buf = new_yv12->y_buffer + recon_yoffset;
xd->plane[1].dst.buf = new_yv12->u_buffer + recon_uvoffset;
xd->plane[2].dst.buf = new_yv12->v_buffer + recon_uvoffset;
xd->left_available = (mb_col != 0);
xd->mi[0]->sb_type = bsize;
xd->mi[0]->ref_frame[0] = INTRA_FRAME;
set_mi_row_col(xd, &tile, mb_row * mb_scale, mi_size_high[bsize],
mb_col * mb_scale, mi_size_wide[bsize], cm->mi_rows,
cm->mi_cols);
set_plane_n4(xd, mi_size_wide[bsize], mi_size_high[bsize], num_planes);
// Do intra 16x16 prediction.
xd->mi[0]->segment_id = 0;
xd->lossless[xd->mi[0]->segment_id] = (qindex == 0);
xd->mi[0]->mode = DC_PRED;
xd->mi[0]->tx_size =
use_dc_pred ? (bsize >= BLOCK_16X16 ? TX_16X16 : TX_8X8) : TX_4X4;
av1_encode_intra_block_plane(cpi, x, bsize, 0, 0, mb_row * 2, mb_col * 2);
this_error = aom_get_mb_ss(x->plane[0].src_diff);
// Keep a record of blocks that have almost no intra error residual
// (i.e. are in effect completely flat and untextured in the intra
// domain). In natural videos this is uncommon, but it is much more
// common in animations, graphics and screen content, so may be used
// as a signal to detect these types of content.
if (this_error < UL_INTRA_THRESH) {
++intra_skip_count;
} else if ((mb_col > 0) && (image_data_start_row == INVALID_ROW)) {
image_data_start_row = mb_row;
}
if (seq_params->use_highbitdepth) {
switch (seq_params->bit_depth) {
case AOM_BITS_8: break;
case AOM_BITS_10: this_error >>= 4; break;
case AOM_BITS_12: this_error >>= 8; break;
default:
assert(0 &&
"seq_params->bit_depth should be AOM_BITS_8, "
"AOM_BITS_10 or AOM_BITS_12");
return;
}
}
aom_clear_system_state();
log_intra = log(this_error + 1.0);
if (log_intra < 10.0)
intra_factor += 1.0 + ((10.0 - log_intra) * 0.05);
else
intra_factor += 1.0;
if (seq_params->use_highbitdepth)
level_sample = CONVERT_TO_SHORTPTR(x->plane[0].src.buf)[0];
else
level_sample = x->plane[0].src.buf[0];
if ((level_sample < DARK_THRESH) && (log_intra < 9.0))
brightness_factor += 1.0 + (0.01 * (DARK_THRESH - level_sample));
else
brightness_factor += 1.0;
// Intrapenalty below deals with situations where the intra and inter
// error scores are very low (e.g. a plain black frame).
// We do not have special cases in first pass for 0,0 and nearest etc so
// all inter modes carry an overhead cost estimate for the mv.
// When the error score is very low this causes us to pick all or lots of
// INTRA modes and throw lots of key frames.
// This penalty adds a cost matching that of a 0,0 mv to the intra case.
this_error += intrapenalty;
// Accumulate the intra error.
intra_error += (int64_t)this_error;
int stride = x->plane[0].src.stride;
uint8_t *buf = x->plane[0].src.buf;
for (int r8 = 0; r8 < 2; ++r8)
for (int c8 = 0; c8 < 2; ++c8) {
int hbd = xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH;
frame_avg_wavelet_energy += av1_haar_ac_sad_8x8_uint8_input(
buf + c8 * 8 + r8 * 8 * stride, stride, hbd);
}
#if CONFIG_FP_MB_STATS
if (cpi->use_fp_mb_stats) {
// initialization
cpi->twopass.frame_mb_stats_buf[mb_index] = 0;
}
#endif
// Set up limit values for motion vectors to prevent them extending
// outside the UMV borders.
x->mv_limits.col_min = -((mb_col * 16) + BORDER_MV_PIXELS_B16);
x->mv_limits.col_max =
((cm->mb_cols - 1 - mb_col) * 16) + BORDER_MV_PIXELS_B16;
if (!frame_is_intra_only(cm)) { // Do a motion search
int tmp_err, motion_error, raw_motion_error;
// Assume 0,0 motion with no mv overhead.
MV mv = kZeroMv, tmp_mv = kZeroMv;
struct buf_2d unscaled_last_source_buf_2d;
xd->plane[0].pre[0].buf = first_ref_buf->y_buffer + recon_yoffset;
if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
motion_error = highbd_get_prediction_error(
bsize, &x->plane[0].src, &xd->plane[0].pre[0], xd->bd);
} else {
motion_error = get_prediction_error(bsize, &x->plane[0].src,
&xd->plane[0].pre[0]);
}
// Compute the motion error of the 0,0 motion using the last source
// frame as the reference. Skip the further motion search on
// reconstructed frame if this error is small.
unscaled_last_source_buf_2d.buf =
cpi->unscaled_last_source->y_buffer + recon_yoffset;
unscaled_last_source_buf_2d.stride =
cpi->unscaled_last_source->y_stride;
if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
raw_motion_error = highbd_get_prediction_error(
bsize, &x->plane[0].src, &unscaled_last_source_buf_2d, xd->bd);
} else {
raw_motion_error = get_prediction_error(bsize, &x->plane[0].src,
&unscaled_last_source_buf_2d);
}
// TODO(pengchong): Replace the hard-coded threshold
if (raw_motion_error > 25) {
// Test last reference frame using the previous best mv as the
// starting point (best reference) for the search.
first_pass_motion_search(cpi, x, &best_ref_mv, &mv, &motion_error);
// If the current best reference mv is not centered on 0,0 then do a
// 0,0 based search as well.
if (!is_zero_mv(&best_ref_mv)) {
tmp_err = INT_MAX;
first_pass_motion_search(cpi, x, &kZeroMv, &tmp_mv, &tmp_err);
if (tmp_err < motion_error) {
motion_error = tmp_err;
mv = tmp_mv;
}
}
// Search in an older reference frame.
if ((cm->current_video_frame > 1) && gld_yv12 != NULL) {
// Assume 0,0 motion with no mv overhead.
int gf_motion_error;
xd->plane[0].pre[0].buf = gld_yv12->y_buffer + recon_yoffset;
if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
gf_motion_error = highbd_get_prediction_error(
bsize, &x->plane[0].src, &xd->plane[0].pre[0], xd->bd);
} else {
gf_motion_error = get_prediction_error(bsize, &x->plane[0].src,
&xd->plane[0].pre[0]);
}
first_pass_motion_search(cpi, x, &kZeroMv, &tmp_mv,
&gf_motion_error);
if (gf_motion_error < motion_error && gf_motion_error < this_error)
++second_ref_count;
// Reset to last frame as reference buffer.
xd->plane[0].pre[0].buf = first_ref_buf->y_buffer + recon_yoffset;
xd->plane[1].pre[0].buf = first_ref_buf->u_buffer + recon_uvoffset;
xd->plane[2].pre[0].buf = first_ref_buf->v_buffer + recon_uvoffset;
// In accumulating a score for the older reference frame take the
// best of the motion predicted score and the intra coded error
// (just as will be done for) accumulation of "coded_error" for
// the last frame.
if (gf_motion_error < this_error)
sr_coded_error += gf_motion_error;
else
sr_coded_error += this_error;
} else {
sr_coded_error += motion_error;
}
} else {
sr_coded_error += motion_error;
}
// Start by assuming that intra mode is best.
best_ref_mv.row = 0;
best_ref_mv.col = 0;
#if CONFIG_FP_MB_STATS
if (cpi->use_fp_mb_stats) {
// intra predication statistics
cpi->twopass.frame_mb_stats_buf[mb_index] = 0;
cpi->twopass.frame_mb_stats_buf[mb_index] |= FPMB_DCINTRA_MASK;
cpi->twopass.frame_mb_stats_buf[mb_index] |= FPMB_MOTION_ZERO_MASK;
if (this_error > FPMB_ERROR_LARGE_TH) {
cpi->twopass.frame_mb_stats_buf[mb_index] |= FPMB_ERROR_LARGE_MASK;
} else if (this_error < FPMB_ERROR_SMALL_TH) {
cpi->twopass.frame_mb_stats_buf[mb_index] |= FPMB_ERROR_SMALL_MASK;
}
}
#endif
if (motion_error <= this_error) {
aom_clear_system_state();
// Keep a count of cases where the inter and intra were very close
// and very low. This helps with scene cut detection for example in
// cropped clips with black bars at the sides or top and bottom.
if (((this_error - intrapenalty) * 9 <= motion_error * 10) &&
(this_error < (2 * intrapenalty))) {
neutral_count += 1.0;
// Also track cases where the intra is not much worse than the inter
// and use this in limiting the GF/arf group length.
} else if ((this_error > NCOUNT_INTRA_THRESH) &&
(this_error < (NCOUNT_INTRA_FACTOR * motion_error))) {
neutral_count +=
(double)motion_error / DOUBLE_DIVIDE_CHECK((double)this_error);
}
mv.row *= 8;
mv.col *= 8;
this_error = motion_error;
xd->mi[0]->mode = NEWMV;
xd->mi[0]->mv[0].as_mv = mv;
xd->mi[0]->tx_size = TX_4X4;
xd->mi[0]->ref_frame[0] = LAST_FRAME;
xd->mi[0]->ref_frame[1] = NONE_FRAME;
av1_build_inter_predictors_sby(cm, xd, mb_row * mb_scale,
mb_col * mb_scale, NULL, bsize);
av1_encode_sby_pass1(cm, x, bsize);
sum_mvr += mv.row;
sum_mvr_abs += abs(mv.row);
sum_mvc += mv.col;
sum_mvc_abs += abs(mv.col);
sum_mvrs += mv.row * mv.row;
sum_mvcs += mv.col * mv.col;
++intercount;
best_ref_mv = mv;
#if CONFIG_FP_MB_STATS
if (cpi->use_fp_mb_stats) {
// inter predication statistics
cpi->twopass.frame_mb_stats_buf[mb_index] = 0;
cpi->twopass.frame_mb_stats_buf[mb_index] &= ~FPMB_DCINTRA_MASK;
cpi->twopass.frame_mb_stats_buf[mb_index] |= FPMB_MOTION_ZERO_MASK;
if (this_error > FPMB_ERROR_LARGE_TH) {
cpi->twopass.frame_mb_stats_buf[mb_index] |=
FPMB_ERROR_LARGE_MASK;
} else if (this_error < FPMB_ERROR_SMALL_TH) {
cpi->twopass.frame_mb_stats_buf[mb_index] |=
FPMB_ERROR_SMALL_MASK;
}
}
#endif
if (!is_zero_mv(&mv)) {
++mvcount;
#if CONFIG_FP_MB_STATS
if (cpi->use_fp_mb_stats) {
cpi->twopass.frame_mb_stats_buf[mb_index] &=
~FPMB_MOTION_ZERO_MASK;
// check estimated motion direction
if (mv.col > 0 && mv.col >= abs(mv.row)) {
// right direction
cpi->twopass.frame_mb_stats_buf[mb_index] |=
FPMB_MOTION_RIGHT_MASK;
} else if (mv.row < 0 && abs(mv.row) >= abs(mv.col)) {
// up direction
cpi->twopass.frame_mb_stats_buf[mb_index] |=
FPMB_MOTION_UP_MASK;
} else if (mv.col < 0 && abs(mv.col) >= abs(mv.row)) {
// left direction
cpi->twopass.frame_mb_stats_buf[mb_index] |=
FPMB_MOTION_LEFT_MASK;
} else {
// down direction
cpi->twopass.frame_mb_stats_buf[mb_index] |=
FPMB_MOTION_DOWN_MASK;
}
}
#endif
// Non-zero vector, was it different from the last non zero vector?
if (!is_equal_mv(&mv, &lastmv)) ++new_mv_count;
lastmv = mv;
// Does the row vector point inwards or outwards?
if (mb_row < cm->mb_rows / 2) {
if (mv.row > 0)
--sum_in_vectors;
else if (mv.row < 0)
++sum_in_vectors;
} else if (mb_row > cm->mb_rows / 2) {
if (mv.row > 0)
++sum_in_vectors;
else if (mv.row < 0)
--sum_in_vectors;
}
// Does the col vector point inwards or outwards?
if (mb_col < cm->mb_cols / 2) {
if (mv.col > 0)
--sum_in_vectors;
else if (mv.col < 0)
++sum_in_vectors;
} else if (mb_col > cm->mb_cols / 2) {
if (mv.col > 0)
++sum_in_vectors;
else if (mv.col < 0)
--sum_in_vectors;
}
}
}
raw_motion_err_list[raw_motion_err_counts++] = raw_motion_error;
} else {
sr_coded_error += (int64_t)this_error;
}
coded_error += (int64_t)this_error;
// Adjust to the next column of MBs.
x->plane[0].src.buf += 16;
x->plane[1].src.buf += uv_mb_height;
x->plane[2].src.buf += uv_mb_height;
recon_yoffset += 16;
recon_uvoffset += uv_mb_height;
}
// Adjust to the next row of MBs.
x->plane[0].src.buf += 16 * x->plane[0].src.stride - 16 * cm->mb_cols;
x->plane[1].src.buf +=
uv_mb_height * x->plane[1].src.stride - uv_mb_height * cm->mb_cols;
x->plane[2].src.buf +=
uv_mb_height * x->plane[1].src.stride - uv_mb_height * cm->mb_cols;
aom_clear_system_state();
}
const double raw_err_stdev =
raw_motion_error_stdev(raw_motion_err_list, raw_motion_err_counts);
aom_free(raw_motion_err_list);
// Clamp the image start to rows/2. This number of rows is discarded top
// and bottom as dead data so rows / 2 means the frame is blank.
if ((image_data_start_row > cm->mb_rows / 2) ||
(image_data_start_row == INVALID_ROW)) {
image_data_start_row = cm->mb_rows / 2;
}
// Exclude any image dead zone
if (image_data_start_row > 0) {
intra_skip_count =
AOMMAX(0, intra_skip_count - (image_data_start_row * cm->mb_cols * 2));
}
{
FIRSTPASS_STATS fps;
// The minimum error here insures some bit allocation to frames even
// in static regions. The allocation per MB declines for larger formats
// where the typical "real" energy per MB also falls.
// Initial estimate here uses sqrt(mbs) to define the min_err, where the
// number of mbs is proportional to the image area.
const int num_mbs = (cpi->oxcf.resize_mode != RESIZE_NONE)
? cpi->initial_mbs
: cpi->common.MBs;
const double min_err = 200 * sqrt(num_mbs);
intra_factor = intra_factor / (double)num_mbs;
brightness_factor = brightness_factor / (double)num_mbs;
fps.weight = intra_factor * brightness_factor;
fps.frame = cm->current_video_frame;
fps.coded_error = (double)(coded_error >> 8) + min_err;
fps.sr_coded_error = (double)(sr_coded_error >> 8) + min_err;