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
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
|
/*
* Copyright 2011-2021 Blender Foundation
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
#include "integrator/render_scheduler.h"
#include "session/session.h"
#include "session/tile.h"
#include "util/log.h"
#include "util/math.h"
#include "util/time.h"
CCL_NAMESPACE_BEGIN
/* --------------------------------------------------------------------
* Render scheduler.
*/
RenderScheduler::RenderScheduler(TileManager &tile_manager, const SessionParams ¶ms)
: headless_(params.headless),
background_(params.background),
pixel_size_(params.pixel_size),
tile_manager_(tile_manager),
default_start_resolution_divider_(pixel_size_ * 8)
{
use_progressive_noise_floor_ = !background_;
}
void RenderScheduler::set_need_schedule_cryptomatte(bool need_schedule_cryptomatte)
{
need_schedule_cryptomatte_ = need_schedule_cryptomatte;
}
void RenderScheduler::set_need_schedule_rebalance(bool need_schedule_rebalance)
{
need_schedule_rebalance_works_ = need_schedule_rebalance;
}
bool RenderScheduler::is_background() const
{
return background_;
}
void RenderScheduler::set_denoiser_params(const DenoiseParams ¶ms)
{
denoiser_params_ = params;
}
void RenderScheduler::set_adaptive_sampling(const AdaptiveSampling &adaptive_sampling)
{
adaptive_sampling_ = adaptive_sampling;
}
bool RenderScheduler::is_adaptive_sampling_used() const
{
return adaptive_sampling_.use;
}
void RenderScheduler::set_start_sample(int start_sample)
{
start_sample_ = start_sample;
}
int RenderScheduler::get_start_sample() const
{
return start_sample_;
}
void RenderScheduler::set_num_samples(int num_samples)
{
num_samples_ = num_samples;
}
int RenderScheduler::get_num_samples() const
{
return num_samples_;
}
void RenderScheduler::set_sample_offset(int sample_offset)
{
sample_offset_ = sample_offset;
}
int RenderScheduler::get_sample_offset() const
{
return sample_offset_;
}
void RenderScheduler::set_time_limit(double time_limit)
{
time_limit_ = time_limit;
}
double RenderScheduler::get_time_limit() const
{
return time_limit_;
}
int RenderScheduler::get_rendered_sample() const
{
DCHECK_GT(get_num_rendered_samples(), 0);
return start_sample_ + get_num_rendered_samples() - 1;
}
int RenderScheduler::get_num_rendered_samples() const
{
return state_.num_rendered_samples;
}
void RenderScheduler::reset(const BufferParams &buffer_params, int num_samples, int sample_offset)
{
buffer_params_ = buffer_params;
update_start_resolution_divider();
set_num_samples(num_samples);
set_start_sample(sample_offset);
set_sample_offset(sample_offset);
/* In background mode never do lower resolution render preview, as it is not really supported
* by the software. */
if (background_) {
state_.resolution_divider = 1;
}
else {
/* NOTE: Divide by 2 because of the way how scheduling works: it advances resolution divider
* first and then initialized render work. */
state_.resolution_divider = start_resolution_divider_ * 2;
}
state_.num_rendered_samples = 0;
state_.last_display_update_time = 0.0;
state_.last_display_update_sample = -1;
state_.last_rebalance_time = 0.0;
state_.num_rebalance_requested = 0;
state_.num_rebalance_changes = 0;
state_.last_rebalance_changed = false;
state_.need_rebalance_at_next_work = false;
/* TODO(sergey): Choose better initial value. */
/* NOTE: The adaptive sampling settings might not be available here yet. */
state_.adaptive_sampling_threshold = 0.4f;
state_.last_work_tile_was_denoised = false;
state_.tile_result_was_written = false;
state_.postprocess_work_scheduled = false;
state_.full_frame_work_scheduled = false;
state_.full_frame_was_written = false;
state_.path_trace_finished = false;
state_.start_render_time = 0.0;
state_.end_render_time = 0.0;
state_.time_limit_reached = false;
state_.occupancy_num_samples = 0;
state_.occupancy = 1.0f;
first_render_time_.path_trace_per_sample = 0.0;
first_render_time_.denoise_time = 0.0;
first_render_time_.display_update_time = 0.0;
path_trace_time_.reset();
denoise_time_.reset();
adaptive_filter_time_.reset();
display_update_time_.reset();
rebalance_time_.reset();
}
void RenderScheduler::reset_for_next_tile()
{
reset(buffer_params_, num_samples_, sample_offset_);
}
bool RenderScheduler::render_work_reschedule_on_converge(RenderWork &render_work)
{
/* Move to the next resolution divider. Assume adaptive filtering is not needed during
* navigation. */
if (state_.resolution_divider != pixel_size_) {
return false;
}
if (render_work_reschedule_on_idle(render_work)) {
return true;
}
state_.path_trace_finished = true;
bool denoiser_delayed, denoiser_ready_to_display;
render_work.tile.denoise = work_need_denoise(denoiser_delayed, denoiser_ready_to_display);
render_work.display.update = work_need_update_display(denoiser_delayed);
render_work.display.use_denoised_result = denoiser_ready_to_display;
return false;
}
bool RenderScheduler::render_work_reschedule_on_idle(RenderWork &render_work)
{
if (!use_progressive_noise_floor_) {
return false;
}
/* Move to the next resolution divider. Assume adaptive filtering is not needed during
* navigation. */
if (state_.resolution_divider != pixel_size_) {
return false;
}
if (adaptive_sampling_.use) {
if (state_.adaptive_sampling_threshold > adaptive_sampling_.threshold) {
state_.adaptive_sampling_threshold = max(state_.adaptive_sampling_threshold / 2,
adaptive_sampling_.threshold);
render_work.adaptive_sampling.threshold = state_.adaptive_sampling_threshold;
render_work.adaptive_sampling.reset = true;
return true;
}
}
return false;
}
void RenderScheduler::render_work_reschedule_on_cancel(RenderWork &render_work)
{
VLOG(3) << "Schedule work for cancel.";
/* Un-schedule samples: they will not be rendered and should not be counted. */
state_.num_rendered_samples -= render_work.path_trace.num_samples;
const bool has_rendered_samples = get_num_rendered_samples() != 0;
/* Reset all fields of the previous work, canceling things like adaptive sampling filtering and
* denoising.
* However, need to preserve write requests, since those will not be possible to recover and
* writes are only to happen once. */
const bool tile_write = render_work.tile.write;
const bool full_write = render_work.full.write;
render_work = RenderWork();
render_work.tile.write = tile_write;
render_work.full.write = full_write;
/* Do not write tile if it has zero samples it it, treat it similarly to all other tiles which
* got canceled. */
if (!state_.tile_result_was_written && has_rendered_samples) {
render_work.tile.write = true;
}
if (!state_.full_frame_was_written) {
render_work.full.write = true;
}
/* Update current tile, but only if any sample was rendered.
* Allows to have latest state of tile visible while full buffer is being processed.
*
* Note that if there are no samples in the current tile its render buffer might have pixels
* remained from previous state.
*
* If the full result was written, then there is no way any updates were made to the render
* buffers. And the buffers might have been freed from the device, so display update is not
* possible. */
if (has_rendered_samples && !state_.full_frame_was_written) {
render_work.display.update = true;
}
}
bool RenderScheduler::done() const
{
if (state_.resolution_divider != pixel_size_) {
return false;
}
if (state_.path_trace_finished || state_.time_limit_reached) {
return true;
}
return get_num_rendered_samples() >= num_samples_;
}
RenderWork RenderScheduler::get_render_work()
{
check_time_limit_reached();
const double time_now = time_dt();
if (done()) {
RenderWork render_work;
render_work.resolution_divider = state_.resolution_divider;
if (!set_postprocess_render_work(&render_work)) {
set_full_frame_render_work(&render_work);
}
if (!render_work) {
state_.end_render_time = time_now;
}
update_state_for_render_work(render_work);
return render_work;
}
RenderWork render_work;
if (state_.resolution_divider != pixel_size_) {
state_.resolution_divider = max(state_.resolution_divider / 2, pixel_size_);
state_.num_rendered_samples = 0;
state_.last_display_update_sample = -1;
}
render_work.resolution_divider = state_.resolution_divider;
render_work.path_trace.start_sample = get_start_sample_to_path_trace();
render_work.path_trace.num_samples = get_num_samples_to_path_trace();
render_work.path_trace.sample_offset = get_sample_offset();
render_work.init_render_buffers = (render_work.path_trace.start_sample == get_start_sample());
/* NOTE: Rebalance scheduler requires current number of samples to not be advanced forward. */
render_work.rebalance = work_need_rebalance();
/* NOTE: Advance number of samples now, so that filter and denoising check can see that all the
* samples are rendered. */
state_.num_rendered_samples += render_work.path_trace.num_samples;
render_work.adaptive_sampling.filter = work_need_adaptive_filter();
render_work.adaptive_sampling.threshold = work_adaptive_threshold();
render_work.adaptive_sampling.reset = false;
bool denoiser_delayed, denoiser_ready_to_display;
render_work.tile.denoise = work_need_denoise(denoiser_delayed, denoiser_ready_to_display);
render_work.tile.write = done();
render_work.display.update = work_need_update_display(denoiser_delayed);
render_work.display.use_denoised_result = denoiser_ready_to_display;
if (done()) {
set_postprocess_render_work(&render_work);
}
update_state_for_render_work(render_work);
return render_work;
}
void RenderScheduler::update_state_for_render_work(const RenderWork &render_work)
{
const double time_now = time_dt();
if (render_work.rebalance) {
state_.last_rebalance_time = time_now;
++state_.num_rebalance_requested;
}
/* A fallback display update time, for the case there is an error of display update, or when
* there is no display at all. */
if (render_work.display.update) {
state_.last_display_update_time = time_now;
state_.last_display_update_sample = state_.num_rendered_samples;
}
state_.last_work_tile_was_denoised = render_work.tile.denoise;
state_.tile_result_was_written |= render_work.tile.write;
state_.full_frame_was_written |= render_work.full.write;
}
bool RenderScheduler::set_postprocess_render_work(RenderWork *render_work)
{
if (state_.postprocess_work_scheduled) {
return false;
}
state_.postprocess_work_scheduled = true;
bool any_scheduled = false;
if (need_schedule_cryptomatte_) {
render_work->cryptomatte.postprocess = true;
any_scheduled = true;
}
if (denoiser_params_.use && !state_.last_work_tile_was_denoised) {
render_work->tile.denoise = !tile_manager_.has_multiple_tiles();
any_scheduled = true;
}
if (!state_.tile_result_was_written) {
render_work->tile.write = true;
any_scheduled = true;
}
/* Force update. */
any_scheduled = true;
if (any_scheduled) {
render_work->display.update = true;
}
return any_scheduled;
}
void RenderScheduler::set_full_frame_render_work(RenderWork *render_work)
{
if (state_.full_frame_work_scheduled) {
return;
}
if (!tile_manager_.has_multiple_tiles()) {
/* There is only single tile, so all work has been performed already. */
return;
}
if (!tile_manager_.done()) {
/* There are still tiles to be rendered. */
return;
}
if (state_.full_frame_was_written) {
return;
}
state_.full_frame_work_scheduled = true;
render_work->full.write = true;
}
/* Knowing time which it took to complete a task at the current resolution divider approximate how
* long it would have taken to complete it at a final resolution. */
static double approximate_final_time(const RenderWork &render_work, double time)
{
if (render_work.resolution_divider == 1) {
return time;
}
const double resolution_divider_sq = render_work.resolution_divider *
render_work.resolution_divider;
return time * resolution_divider_sq;
}
void RenderScheduler::report_work_begin(const RenderWork &render_work)
{
/* Start counting render time when rendering samples at their final resolution.
*
* NOTE: The work might have the path trace part be all zero: this happens when a post-processing
* work is scheduled after the path tracing. Checking for just a start sample doesn't work here
* because it might be wrongly 0. Check for whether path tracing is actually happening as it is
* expected to happen in the first work. */
if (render_work.resolution_divider == pixel_size_ && render_work.path_trace.num_samples != 0 &&
render_work.path_trace.start_sample == get_start_sample()) {
state_.start_render_time = time_dt();
}
}
void RenderScheduler::report_path_trace_time(const RenderWork &render_work,
double time,
bool is_cancelled)
{
path_trace_time_.add_wall(time);
if (is_cancelled) {
return;
}
const double final_time_approx = approximate_final_time(render_work, time);
if (work_is_usable_for_first_render_estimation(render_work)) {
first_render_time_.path_trace_per_sample = final_time_approx /
render_work.path_trace.num_samples;
}
if (work_report_reset_average(render_work)) {
path_trace_time_.reset_average();
}
path_trace_time_.add_average(final_time_approx, render_work.path_trace.num_samples);
VLOG(4) << "Average path tracing time: " << path_trace_time_.get_average() << " seconds.";
}
void RenderScheduler::report_path_trace_occupancy(const RenderWork &render_work, float occupancy)
{
state_.occupancy_num_samples = render_work.path_trace.num_samples;
state_.occupancy = occupancy;
VLOG(4) << "Measured path tracing occupancy: " << occupancy;
}
void RenderScheduler::report_adaptive_filter_time(const RenderWork &render_work,
double time,
bool is_cancelled)
{
adaptive_filter_time_.add_wall(time);
if (is_cancelled) {
return;
}
const double final_time_approx = approximate_final_time(render_work, time);
if (work_report_reset_average(render_work)) {
adaptive_filter_time_.reset_average();
}
adaptive_filter_time_.add_average(final_time_approx, render_work.path_trace.num_samples);
VLOG(4) << "Average adaptive sampling filter time: " << adaptive_filter_time_.get_average()
<< " seconds.";
}
void RenderScheduler::report_denoise_time(const RenderWork &render_work, double time)
{
denoise_time_.add_wall(time);
const double final_time_approx = approximate_final_time(render_work, time);
if (work_is_usable_for_first_render_estimation(render_work)) {
first_render_time_.denoise_time = final_time_approx;
}
if (work_report_reset_average(render_work)) {
denoise_time_.reset_average();
}
denoise_time_.add_average(final_time_approx);
VLOG(4) << "Average denoising time: " << denoise_time_.get_average() << " seconds.";
}
void RenderScheduler::report_display_update_time(const RenderWork &render_work, double time)
{
display_update_time_.add_wall(time);
const double final_time_approx = approximate_final_time(render_work, time);
if (work_is_usable_for_first_render_estimation(render_work)) {
first_render_time_.display_update_time = final_time_approx;
}
if (work_report_reset_average(render_work)) {
display_update_time_.reset_average();
}
display_update_time_.add_average(final_time_approx);
VLOG(4) << "Average display update time: " << display_update_time_.get_average() << " seconds.";
/* Move the display update moment further in time, so that logic which checks when last update
* did happen have more reliable point in time (without path tracing and denoising parts of the
* render work). */
state_.last_display_update_time = time_dt();
}
void RenderScheduler::report_rebalance_time(const RenderWork &render_work,
double time,
bool balance_changed)
{
rebalance_time_.add_wall(time);
if (work_report_reset_average(render_work)) {
rebalance_time_.reset_average();
}
rebalance_time_.add_average(time);
if (balance_changed) {
++state_.num_rebalance_changes;
}
state_.last_rebalance_changed = balance_changed;
VLOG(4) << "Average rebalance time: " << rebalance_time_.get_average() << " seconds.";
}
string RenderScheduler::full_report() const
{
const double render_wall_time = state_.end_render_time - state_.start_render_time;
const int num_rendered_samples = get_num_rendered_samples();
string result = "\nRender Scheduler Summary\n\n";
{
string mode;
if (headless_) {
mode = "Headless";
}
else if (background_) {
mode = "Background";
}
else {
mode = "Interactive";
}
result += "Mode: " + mode + "\n";
}
result += "Resolution: " + to_string(buffer_params_.width) + "x" +
to_string(buffer_params_.height) + "\n";
result += "\nAdaptive sampling:\n";
result += " Use: " + string_from_bool(adaptive_sampling_.use) + "\n";
if (adaptive_sampling_.use) {
result += " Step: " + to_string(adaptive_sampling_.adaptive_step) + "\n";
result += " Min Samples: " + to_string(adaptive_sampling_.min_samples) + "\n";
result += " Threshold: " + to_string(adaptive_sampling_.threshold) + "\n";
}
result += "\nDenoiser:\n";
result += " Use: " + string_from_bool(denoiser_params_.use) + "\n";
if (denoiser_params_.use) {
result += " Type: " + string(denoiserTypeToHumanReadable(denoiser_params_.type)) + "\n";
result += " Start Sample: " + to_string(denoiser_params_.start_sample) + "\n";
string passes = "Color";
if (denoiser_params_.use_pass_albedo) {
passes += ", Albedo";
}
if (denoiser_params_.use_pass_normal) {
passes += ", Normal";
}
result += " Passes: " + passes + "\n";
}
if (state_.num_rebalance_requested) {
result += "\nRebalancer:\n";
result += " Number of requested rebalances: " + to_string(state_.num_rebalance_requested) +
"\n";
result += " Number of performed rebalances: " + to_string(state_.num_rebalance_changes) +
"\n";
}
result += "\nTime (in seconds):\n";
result += string_printf(" %20s %20s %20s\n", "", "Wall", "Average");
result += string_printf(" %20s %20f %20f\n",
"Path Tracing",
path_trace_time_.get_wall(),
path_trace_time_.get_average());
if (adaptive_sampling_.use) {
result += string_printf(" %20s %20f %20f\n",
"Adaptive Filter",
adaptive_filter_time_.get_wall(),
adaptive_filter_time_.get_average());
}
if (denoiser_params_.use) {
result += string_printf(
" %20s %20f %20f\n", "Denoiser", denoise_time_.get_wall(), denoise_time_.get_average());
}
result += string_printf(" %20s %20f %20f\n",
"Display Update",
display_update_time_.get_wall(),
display_update_time_.get_average());
if (state_.num_rebalance_requested) {
result += string_printf(" %20s %20f %20f\n",
"Rebalance",
rebalance_time_.get_wall(),
rebalance_time_.get_average());
}
const double total_time = path_trace_time_.get_wall() + adaptive_filter_time_.get_wall() +
denoise_time_.get_wall() + display_update_time_.get_wall();
result += "\n Total: " + to_string(total_time) + "\n";
result += string_printf(
"\nRendered %d samples in %f seconds\n", num_rendered_samples, render_wall_time);
/* When adaptive sampling is used the average time becomes meaningless, because different samples
* will likely render different number of pixels. */
if (!adaptive_sampling_.use) {
result += string_printf("Average time per sample: %f seconds\n",
render_wall_time / num_rendered_samples);
}
return result;
}
double RenderScheduler::guess_display_update_interval_in_seconds() const
{
return guess_display_update_interval_in_seconds_for_num_samples(state_.num_rendered_samples);
}
double RenderScheduler::guess_display_update_interval_in_seconds_for_num_samples(
int num_rendered_samples) const
{
double update_interval = guess_display_update_interval_in_seconds_for_num_samples_no_limit(
num_rendered_samples);
if (time_limit_ != 0.0 && state_.start_render_time != 0.0) {
const double remaining_render_time = max(0.0,
time_limit_ - (time_dt() - state_.start_render_time));
update_interval = min(update_interval, remaining_render_time);
}
return update_interval;
}
/* TODO(sergey): This is just a quick implementation, exact values might need to be tweaked based
* on a more careful experiments with viewport rendering. */
double RenderScheduler::guess_display_update_interval_in_seconds_for_num_samples_no_limit(
int num_rendered_samples) const
{
/* TODO(sergey): Need a decision on whether this should be using number of samples rendered
* within the current render session, or use absolute number of samples with the start sample
* taken into account. It will depend on whether the start sample offset clears the render
* buffer. */
if (state_.need_rebalance_at_next_work) {
return 0.1;
}
if (state_.last_rebalance_changed) {
return 0.2;
}
if (headless_) {
/* In headless mode do rare updates, so that the device occupancy is high, but there are still
* progress messages printed to the logs. */
return 30.0;
}
if (background_) {
if (num_rendered_samples < 32) {
return 1.0;
}
return 2.0;
}
/* Render time and number of samples rendered are used to figure out the display update interval.
* Render time is used to allow for fast display updates in the first few seconds of rendering
* on fast devices. Number of samples rendered is used to allow for potentially quicker display
* updates on slow devices during the first few samples. */
const double render_time = path_trace_time_.get_wall();
if (render_time < 1) {
return 0.1;
}
if (render_time < 2) {
return 0.25;
}
if (render_time < 4) {
return 0.5;
}
if (render_time < 8 || num_rendered_samples < 32) {
return 1.0;
}
return 2.0;
}
int RenderScheduler::calculate_num_samples_per_update() const
{
const double time_per_sample_average = path_trace_time_.get_average();
const double num_samples_in_second = pixel_size_ * pixel_size_ / time_per_sample_average;
const double update_interval_in_seconds = guess_display_update_interval_in_seconds();
return max(int(num_samples_in_second * update_interval_in_seconds), 1);
}
int RenderScheduler::get_start_sample_to_path_trace() const
{
return start_sample_ + state_.num_rendered_samples;
}
/* Round number of samples to the closest power of two.
* Rounding might happen to higher or lower value depending on which one is closer. Such behavior
* allows to have number of samples to be power of two without diverging from the planned number of
* samples too much. */
static inline uint round_num_samples_to_power_of_2(const uint num_samples)
{
if (num_samples == 1) {
return 1;
}
if (is_power_of_two(num_samples)) {
return num_samples;
}
const uint num_samples_up = next_power_of_two(num_samples);
const uint num_samples_down = num_samples_up - (num_samples_up >> 1);
const uint delta_up = num_samples_up - num_samples;
const uint delta_down = num_samples - num_samples_down;
if (delta_up <= delta_down) {
return num_samples_up;
}
return num_samples_down;
}
int RenderScheduler::get_num_samples_to_path_trace() const
{
if (state_.resolution_divider != pixel_size_) {
return get_num_samples_during_navigation(state_.resolution_divider);
}
/* Always start full resolution render with a single sample. Gives more instant feedback to
* artists, and allows to gather information for a subsequent path tracing works. Do it in the
* headless mode as well, to give some estimate of how long samples are taking. */
if (state_.num_rendered_samples == 0) {
return 1;
}
const int num_samples_per_update = calculate_num_samples_per_update();
const int path_trace_start_sample = get_start_sample_to_path_trace();
/* Round number of samples to a power of two, so that division of path states into tiles goes in
* a more integer manner.
* This might make it so updates happens more rarely due to rounding up. In the test scenes this
* is not huge deal because it is not seen that more than 8 samples can be rendered between
* updates. If that becomes a problem we can add some extra rules like never allow to round up
* more than N samples. */
const int num_samples_pot = round_num_samples_to_power_of_2(num_samples_per_update);
const int max_num_samples_to_render = start_sample_ + num_samples_ - path_trace_start_sample;
int num_samples_to_render = min(num_samples_pot, max_num_samples_to_render);
/* When enough statistics is available and doing an offline rendering prefer to keep device
* occupied. */
if (state_.occupancy_num_samples && (background_ || headless_)) {
/* Keep occupancy at about 0.5 (this is more of an empirical figure which seems to match scenes
* with good performance without forcing occupancy to be higher). */
int num_samples_to_occupy = state_.occupancy_num_samples;
if (state_.occupancy < 0.5f) {
num_samples_to_occupy = lround(state_.occupancy_num_samples * 0.7f / state_.occupancy);
}
/* When time limit is used clamp the calculated number of samples to keep occupancy.
* This is because time limit causes the last render iteration to happen with less number of
* samples, which conflicts with the occupancy (lower number of samples causes lower
* occupancy, also the calculation is based on number of previously rendered samples).
*
* When time limit is not used the number of samples per render iteration is either increasing
* or stays the same, so there is no need to clamp number of samples calculated for occupancy.
*/
if (time_limit_ != 0.0 && state_.start_render_time != 0.0) {
const double remaining_render_time = max(
0.0, time_limit_ - (time_dt() - state_.start_render_time));
const double time_per_sample_average = path_trace_time_.get_average();
const double predicted_render_time = num_samples_to_occupy * time_per_sample_average;
if (predicted_render_time > remaining_render_time) {
num_samples_to_occupy = lround(num_samples_to_occupy *
(remaining_render_time / predicted_render_time));
}
}
num_samples_to_render = max(num_samples_to_render,
min(num_samples_to_occupy, max_num_samples_to_render));
}
/* If adaptive sampling is not use, render as many samples per update as possible, keeping the
* device fully occupied, without much overhead of display updates. */
if (!adaptive_sampling_.use) {
return num_samples_to_render;
}
/* TODO(sergey): Add extra "clamping" here so that none of the filtering points is missing. This
* is to ensure that the final render is pixel-matched regardless of how many samples per second
* compute device can do. */
return adaptive_sampling_.align_samples(path_trace_start_sample, num_samples_to_render);
}
int RenderScheduler::get_num_samples_during_navigation(int resolution_divider) const
{
/* Special trick for fast navigation: schedule multiple samples during fast navigation
* (which will prefer to use lower resolution to keep up with refresh rate). This gives more
* usable visual feedback for artists. There are a couple of tricks though. */
if (is_denoise_active_during_update()) {
/* When denoising is used during navigation prefer using a higher resolution with less samples
* (scheduling less samples here will make it so the resolution_divider calculation will use a
* lower value for the divider). This is because both OpenImageDenoiser and OptiX denoiser
* give visually better results on a higher resolution image with less samples. */
return 1;
}
if (resolution_divider <= pixel_size_) {
/* When resolution divider is at or below pixel size, schedule one sample. This doesn't effect
* the sample count at this resolution division, but instead assists in the calculation of
* the resolution divider. */
return 1;
}
if (resolution_divider == pixel_size_ * 2) {
/* When resolution divider is the previous step to the final resolution, schedule two samples.
* This is so that rendering on lower resolution does not exceed time that it takes to render
* first sample at the full resolution. */
return 2;
}
/* Always render 4 samples, even if scene is configured for less.
* The idea here is to have enough information on the screen. Resolution divider of 2 allows us
* to have 4 time extra samples, so overall worst case timing is the same as the final resolution
* at one sample. */
return 4;
}
bool RenderScheduler::work_need_adaptive_filter() const
{
return adaptive_sampling_.need_filter(get_rendered_sample());
}
float RenderScheduler::work_adaptive_threshold() const
{
if (!use_progressive_noise_floor_) {
return adaptive_sampling_.threshold;
}
return max(state_.adaptive_sampling_threshold, adaptive_sampling_.threshold);
}
bool RenderScheduler::work_need_denoise(bool &delayed, bool &ready_to_display)
{
delayed = false;
ready_to_display = true;
if (!denoiser_params_.use) {
/* Denoising is disabled, no need to scheduler work for it. */
return false;
}
/* When multiple tiles are used the full frame will be denoised.
* Avoid per-tile denoising to save up render time. */
if (tile_manager_.has_multiple_tiles()) {
return false;
}
if (done()) {
/* Always denoise at the last sample. */
return true;
}
if (background_) {
/* Background render, only denoise when rendering the last sample. */
/* TODO(sergey): Follow similar logic to viewport, giving an overview of how final denoised
* image looks like even for the background rendering. */
return false;
}
/* Viewport render. */
/* Navigation might render multiple samples at a lower resolution. Those are not to be counted as
* final samples. */
const int num_samples_finished = state_.resolution_divider == pixel_size_ ?
state_.num_rendered_samples :
1;
/* Immediately denoise when we reach the start sample or last sample. */
if (num_samples_finished == denoiser_params_.start_sample ||
num_samples_finished == num_samples_) {
return true;
}
/* Do not denoise until the sample at which denoising should start is reached. */
if (num_samples_finished < denoiser_params_.start_sample) {
ready_to_display = false;
return false;
}
/* Avoid excessive denoising in viewport after reaching a certain sample count and render time.
*/
/* TODO(sergey): Consider making time interval and sample configurable. */
delayed = (path_trace_time_.get_wall() > 4 && num_samples_finished >= 20 &&
(time_dt() - state_.last_display_update_time) < 1.0);
return !delayed;
}
bool RenderScheduler::work_need_update_display(const bool denoiser_delayed)
{
if (headless_) {
/* Force disable display update in headless mode. There will be nothing to display the
* in-progress result. */
return false;
}
if (denoiser_delayed) {
/* If denoiser has been delayed the display can not be updated as it will not contain
* up-to-date state of the render result. */
return false;
}
if (!adaptive_sampling_.use) {
/* When adaptive sampling is not used the work is scheduled in a way that they keep render
* device busy for long enough, so that the display update can happen right after the
* rendering. */
return true;
}
if (done() || state_.last_display_update_sample == -1) {
/* Make sure an initial and final results of adaptive sampling is communicated ot the display.
*/
return true;
}
/* For the development purposes of adaptive sampling it might be very useful to see all updates
* of active pixels after convergence check. However, it would cause a slowdown for regular usage
* users. Possibly, make it a debug panel option to allow rapid update to ease development
* without need to re-compiled. */
// if (work_need_adaptive_filter()) {
// return true;
// }
/* When adaptive sampling is used, its possible that only handful of samples of a very simple
* scene will be scheduled to a powerful device (in order to not "miss" any of filtering points).
* We take care of skipping updates here based on when previous display update did happen. */
const double update_interval = guess_display_update_interval_in_seconds_for_num_samples(
state_.last_display_update_sample);
return (time_dt() - state_.last_display_update_time) > update_interval;
}
bool RenderScheduler::work_need_rebalance()
{
/* This is the minimum time, as the rebalancing can not happen more often than the path trace
* work. */
static const double kRebalanceIntervalInSeconds = 1;
if (!need_schedule_rebalance_works_) {
return false;
}
if (state_.resolution_divider != pixel_size_) {
/* Don't rebalance at a non-final resolution divider. Some reasons for this:
* - It will introduce unnecessary during navigation.
* - Per-render device timing information is not very reliable yet. */
return false;
}
if (state_.num_rendered_samples == 0) {
state_.need_rebalance_at_next_work = true;
return false;
}
if (state_.need_rebalance_at_next_work) {
state_.need_rebalance_at_next_work = false;
return true;
}
if (state_.last_rebalance_changed) {
return true;
}
return (time_dt() - state_.last_rebalance_time) > kRebalanceIntervalInSeconds;
}
void RenderScheduler::update_start_resolution_divider()
{
if (start_resolution_divider_ == 0) {
/* Resolution divider has never been calculated before: use default resolution, so that we have
* somewhat good initial behavior, giving a chance to collect real numbers. */
start_resolution_divider_ = default_start_resolution_divider_;
VLOG(3) << "Initial resolution divider is " << start_resolution_divider_;
return;
}
if (first_render_time_.path_trace_per_sample == 0.0) {
/* Not enough information to calculate better resolution, keep the existing one. */
return;
}
const double desired_update_interval_in_seconds =
guess_viewport_navigation_update_interval_in_seconds();
const double actual_time_per_update = first_render_time_.path_trace_per_sample +
first_render_time_.denoise_time +
first_render_time_.display_update_time;
/* Allow some percent of tolerance, so that if the render time is close enough to the higher
* resolution we prefer to use it instead of going way lower resolution and time way below the
* desired one. */
const int resolution_divider_for_update = calculate_resolution_divider_for_time(
desired_update_interval_in_seconds * 1.4, actual_time_per_update);
/* TODO(sergey): Need to add hysteresis to avoid resolution divider bouncing around when actual
* render time is somewhere on a boundary between two resolutions. */
/* Never increase resolution to higher than the pixel size (which is possible if the scene is
* simple and compute device is fast). */
start_resolution_divider_ = max(resolution_divider_for_update, pixel_size_);
VLOG(3) << "Calculated resolution divider is " << start_resolution_divider_;
}
double RenderScheduler::guess_viewport_navigation_update_interval_in_seconds() const
{
if (is_denoise_active_during_update()) {
/* Use lower value than the non-denoised case to allow having more pixels to reconstruct the
* image from. With the faster updates and extra compute required the resolution becomes too
* low to give usable feedback. */
/* NOTE: Based on performance of OpenImageDenoiser on CPU. For OptiX denoiser or other denoiser
* on GPU the value might need to become lower for faster navigation. */
return 1.0 / 12.0;
}
/* For the best match with the Blender's viewport the refresh ratio should be 60fps. This will
* avoid "jelly" effects. However, on a non-trivial scenes this can only be achieved with high
* values of the resolution divider which does not give very pleasant updates during navigation.
* Choose less frequent updates to allow more noise-free and higher resolution updates. */
/* TODO(sergey): Can look into heuristic which will allow to have 60fps if the resolution divider
* is not too high. Alternatively, synchronize Blender's overlays updates to Cycles updates. */
return 1.0 / 30.0;
}
bool RenderScheduler::is_denoise_active_during_update() const
{
if (!denoiser_params_.use) {
return false;
}
if (denoiser_params_.start_sample > 1) {
return false;
}
return true;
}
bool RenderScheduler::work_is_usable_for_first_render_estimation(const RenderWork &render_work)
{
return render_work.resolution_divider == pixel_size_ &&
render_work.path_trace.start_sample == start_sample_;
}
bool RenderScheduler::work_report_reset_average(const RenderWork &render_work)
{
/* When rendering at a non-final resolution divider time average is not very useful because it
* will either bias average down (due to lower render times on the smaller images) or will give
* incorrect result when trying to estimate time which would have spent on the final resolution.
*
* So we only accumulate average for the latest resolution divider which was rendered. */
return render_work.resolution_divider != pixel_size_;
}
void RenderScheduler::check_time_limit_reached()
{
if (time_limit_ == 0.0) {
/* No limit is enforced. */
return;
}
if (state_.start_render_time == 0.0) {
/* Rendering did not start yet. */
return;
}
const double current_time = time_dt();
if (current_time - state_.start_render_time < time_limit_) {
/* Time limit is not reached yet. */
return;
}
state_.time_limit_reached = true;
state_.end_render_time = current_time;
}
/* --------------------------------------------------------------------
* Utility functions.
*/
int RenderScheduler::calculate_resolution_divider_for_time(double desired_time, double actual_time)
{
/* TODO(sergey): There should a non-iterative analytical formula here. */
int resolution_divider = 1;
/* This algorithm iterates through resolution dividers until a divider is found that achieves
* the desired render time. A limit of default_start_resolution_divider_ is put in place as the
* maximum resolution divider to avoid an unreadable viewport due to a low resolution.
* pre_resolution_division_samples and post_resolution_division_samples are used in this
* calculation to better predict the performance impact of changing resolution divisions as
* the sample count can also change between resolution divisions. */
while (actual_time > desired_time && resolution_divider < default_start_resolution_divider_) {
int pre_resolution_division_samples = get_num_samples_during_navigation(resolution_divider);
resolution_divider = resolution_divider * 2;
int post_resolution_division_samples = get_num_samples_during_navigation(resolution_divider);
actual_time /= 4.0 * pre_resolution_division_samples / post_resolution_division_samples;
}
return resolution_divider;
}
int calculate_resolution_divider_for_resolution(int width, int height, int resolution)
{
if (resolution == INT_MAX) {
return 1;
}
int resolution_divider = 1;
while (width * height > resolution * resolution) {
width = max(1, width / 2);
height = max(1, height / 2);
resolution_divider <<= 1;
}
return resolution_divider;
}
int calculate_resolution_for_divider(int width, int height, int resolution_divider)
{
const int pixel_area = width * height;
const int resolution = lround(sqrt(pixel_area));
return resolution / resolution_divider;
}
CCL_NAMESPACE_END
|