-
Notifications
You must be signed in to change notification settings - Fork 29
/
Copy pathUT_BVHImpl.h
1676 lines (1563 loc) · 68.4 KB
/
UT_BVHImpl.h
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) 2018 Side Effects Software Inc.
*
* Permission is hereby granted, free of charge, to any person obtaining a copy
* of this software and associated documentation files (the "Software"), to deal
* in the Software without restriction, including without limitation the rights
* to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
* copies of the Software, and to permit persons to whom the Software is
* furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice shall be included in all
* copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
* AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
* OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
* SOFTWARE.
*
* COMMENTS:
* Bounding Volume Hierarchy (BVH) implementation.
* The main file is UT_BVH.h; this file is separate so that
* files that don't actually need to call functions on the BVH
* won't have unnecessary headers and functions included.
*/
#pragma once
#ifndef __HDK_UT_BVHImpl_h__
#define __HDK_UT_BVHImpl_h__
#include "UT_BVH.h"
#include "UT_Array.h"
#include "UT_FixedVector.h"
#include "UT_ParallelUtil.h"
#include "UT_SmallArray.h"
#include "SYS_Types.h"
#include <algorithm>
namespace HDK_Sample {
namespace UT {
template<typename T,uint NAXES>
SYS_FORCE_INLINE bool utBoxExclude(const UT::Box<T,NAXES>& box) noexcept {
bool has_nan_or_inf = !SYSisFinite(box[0][0]);
has_nan_or_inf |= !SYSisFinite(box[0][1]);
for (uint axis = 1; axis < NAXES; ++axis)
{
has_nan_or_inf |= !SYSisFinite(box[axis][0]);
has_nan_or_inf |= !SYSisFinite(box[axis][1]);
}
return has_nan_or_inf;
}
template<uint NAXES>
SYS_FORCE_INLINE bool utBoxExclude(const UT::Box<fpreal32,NAXES>& box) noexcept {
const int32 *pboxints = reinterpret_cast<const int32*>(&box);
// Fast check for NaN or infinity: check if exponent bits are 0xFF.
bool has_nan_or_inf = ((pboxints[0] & 0x7F800000) == 0x7F800000);
has_nan_or_inf |= ((pboxints[1] & 0x7F800000) == 0x7F800000);
for (uint axis = 1; axis < NAXES; ++axis)
{
has_nan_or_inf |= ((pboxints[2*axis] & 0x7F800000) == 0x7F800000);
has_nan_or_inf |= ((pboxints[2*axis + 1] & 0x7F800000) == 0x7F800000);
}
return has_nan_or_inf;
}
template<typename T,uint NAXES>
SYS_FORCE_INLINE T utBoxCenter(const UT::Box<T,NAXES>& box, uint axis) noexcept {
const T* v = box.vals[axis];
return v[0] + v[1];
}
template<typename T>
struct ut_BoxCentre {
constexpr static uint scale = 2;
};
template<typename T,uint NAXES,bool INSTANTIATED>
SYS_FORCE_INLINE T utBoxExclude(const UT_FixedVector<T,NAXES,INSTANTIATED>& position) noexcept {
bool has_nan_or_inf = !SYSisFinite(position[0]);
for (uint axis = 1; axis < NAXES; ++axis)
has_nan_or_inf |= !SYSisFinite(position[axis]);
return has_nan_or_inf;
}
template<uint NAXES,bool INSTANTIATED>
SYS_FORCE_INLINE bool utBoxExclude(const UT_FixedVector<fpreal32,NAXES,INSTANTIATED>& position) noexcept {
const int32 *ppositionints = reinterpret_cast<const int32*>(&position);
// Fast check for NaN or infinity: check if exponent bits are 0xFF.
bool has_nan_or_inf = ((ppositionints[0] & 0x7F800000) == 0x7F800000);
for (uint axis = 1; axis < NAXES; ++axis)
has_nan_or_inf |= ((ppositionints[axis] & 0x7F800000) == 0x7F800000);
return has_nan_or_inf;
}
template<typename T,uint NAXES,bool INSTANTIATED>
SYS_FORCE_INLINE T utBoxCenter(const UT_FixedVector<T,NAXES,INSTANTIATED>& position, uint axis) noexcept {
return position[axis];
}
template<typename T,uint NAXES,bool INSTANTIATED>
struct ut_BoxCentre<UT_FixedVector<T,NAXES,INSTANTIATED>> {
constexpr static uint scale = 1;
};
template<typename BOX_TYPE,typename SRC_INT_TYPE,typename INT_TYPE>
INT_TYPE utExcludeNaNInfBoxIndices(const BOX_TYPE* boxes, SRC_INT_TYPE* indices, INT_TYPE& nboxes) noexcept
{
constexpr INT_TYPE PARALLEL_THRESHOLD = 65536;
INT_TYPE ntasks = 1;
if (nboxes >= PARALLEL_THRESHOLD)
{
INT_TYPE nprocessors = UT_Thread::getNumProcessors();
ntasks = (nprocessors > 1) ? SYSmin(4*nprocessors, nboxes/(PARALLEL_THRESHOLD/2)) : 1;
}
if (ntasks == 1)
{
// Serial: easy case; just loop through.
const SRC_INT_TYPE* indices_end = indices + nboxes;
// Loop through forward once
SRC_INT_TYPE* psrc_index = indices;
for (; psrc_index != indices_end; ++psrc_index)
{
const bool exclude = utBoxExclude(boxes[*psrc_index]);
if (exclude)
break;
}
if (psrc_index == indices_end)
return 0;
// First NaN or infinite box
SRC_INT_TYPE* nan_start = psrc_index;
for (++psrc_index; psrc_index != indices_end; ++psrc_index)
{
const bool exclude = utBoxExclude(boxes[*psrc_index]);
if (!exclude)
{
*nan_start = *psrc_index;
++nan_start;
}
}
nboxes = nan_start-indices;
return indices_end - nan_start;
}
// Parallel: hard case.
// 1) Collapse each of ntasks chunks and count number of items to exclude
// 2) Accumulate number of items to exclude.
// 3) If none, return.
// 4) Copy non-NaN chunks
UT_SmallArray<INT_TYPE> nexcluded;
nexcluded.setSizeNoInit(ntasks);
UTparallelFor(UT_BlockedRange<INT_TYPE>(0,ntasks), [boxes,indices,ntasks,nboxes,&nexcluded](const UT_BlockedRange<INT_TYPE>& r)
{
for (INT_TYPE taski = r.begin(), task_end = r.end(); taski < task_end; ++taski)
{
SRC_INT_TYPE* indices_start = indices + (taski*exint(nboxes))/ntasks;
const SRC_INT_TYPE* indices_end = indices + ((taski+1)*exint(nboxes))/ntasks;
SRC_INT_TYPE* psrc_index = indices_start;
for (; psrc_index != indices_end; ++psrc_index)
{
const bool exclude = utBoxExclude(boxes[*psrc_index]);
if (exclude)
break;
}
if (psrc_index == indices_end)
{
nexcluded[taski] = 0;
continue;
}
// First NaN or infinite box
SRC_INT_TYPE* nan_start = psrc_index;
for (++psrc_index; psrc_index != indices_end; ++psrc_index)
{
const bool exclude = utBoxExclude(boxes[*psrc_index]);
if (!exclude)
{
*nan_start = *psrc_index;
++nan_start;
}
}
nexcluded[taski] = indices_end - nan_start;
}
}, 0, 1);
// Accumulate
INT_TYPE total_excluded = nexcluded[0];
for (INT_TYPE taski = 1; taski < ntasks; ++taski)
{
total_excluded += nexcluded[taski];
}
if (total_excluded == 0)
return 0;
// TODO: Parallelize this part, if it's a bottleneck and we care about cases with NaNs or infinities.
INT_TYPE taski = 0;
while (nexcluded[taski] == 0)
{
++taski;
}
SRC_INT_TYPE* dest_indices = indices + ((taski+1)*exint(nboxes))/ntasks - nexcluded[taski];
SRC_INT_TYPE* dest_end = indices + nboxes - total_excluded;
for (++taski; taski < ntasks && dest_indices < dest_end; ++taski)
{
const SRC_INT_TYPE* psrc_index = indices + (taski*exint(nboxes))/ntasks;
const SRC_INT_TYPE* psrc_end = indices + ((taski+1)*exint(nboxes))/ntasks - nexcluded[taski];
INT_TYPE count = psrc_end - psrc_index;
// Note should be memmove as it is overlapping.
memmove(dest_indices, psrc_index, sizeof(SRC_INT_TYPE)*count);
dest_indices += count;
}
nboxes -= total_excluded;
return total_excluded;
}
template<uint N>
template<BVH_Heuristic H,typename T,uint NAXES,typename BOX_TYPE,typename SRC_INT_TYPE>
void BVH<N>::init(const BOX_TYPE* boxes, const INT_TYPE nboxes, SRC_INT_TYPE* indices, bool reorder_indices, INT_TYPE max_items_per_leaf) noexcept {
Box<T,NAXES> axes_minmax;
computeFullBoundingBox(axes_minmax, boxes, nboxes, indices);
init<H>(axes_minmax, boxes, nboxes, indices, reorder_indices, max_items_per_leaf);
}
template<uint N>
template<BVH_Heuristic H,typename T,uint NAXES,typename BOX_TYPE,typename SRC_INT_TYPE>
void BVH<N>::init(Box<T,NAXES> axes_minmax, const BOX_TYPE* boxes, INT_TYPE nboxes, SRC_INT_TYPE* indices, bool reorder_indices, INT_TYPE max_items_per_leaf) noexcept {
// Clear the tree in advance to save memory.
myRoot.reset();
if (nboxes == 0) {
myNumNodes = 0;
return;
}
UT_Array<INT_TYPE> local_indices;
if (!indices) {
local_indices.setSizeNoInit(nboxes);
indices = local_indices.array();
createTrivialIndices(indices, nboxes);
}
// Exclude any boxes with NaNs or infinities by shifting down indices
// over the bad box indices and updating nboxes.
INT_TYPE nexcluded = utExcludeNaNInfBoxIndices(boxes, indices, nboxes);
if (nexcluded != 0) {
if (nboxes == 0) {
myNumNodes = 0;
return;
}
computeFullBoundingBox(axes_minmax, boxes, nboxes, indices);
}
UT_Array<Node> nodes;
// Preallocate an overestimate of the number of nodes needed.
nodes.setCapacity(nodeEstimate(nboxes));
nodes.setSize(1);
if (reorder_indices)
initNodeReorder<H>(nodes, nodes[0], axes_minmax, boxes, indices, nboxes, 0, max_items_per_leaf);
else
initNode<H>(nodes, nodes[0], axes_minmax, boxes, indices, nboxes);
// If capacity is more than 12.5% over the size, rellocate.
if (8*nodes.capacity() > 9*nodes.size()) {
nodes.setCapacity(nodes.size());
}
// Steal ownership of the array from the UT_Array
myRoot.reset(nodes.array());
myNumNodes = nodes.size();
nodes.unsafeClearData();
}
template<uint N>
template<typename LOCAL_DATA,typename FUNCTORS>
void BVH<N>::traverse(
FUNCTORS &functors,
LOCAL_DATA* data_for_parent) const noexcept
{
if (!myRoot)
return;
// NOTE: The root is always index 0.
traverseHelper(0, INT_TYPE(-1), functors, data_for_parent);
}
template<uint N>
template<typename LOCAL_DATA,typename FUNCTORS>
void BVH<N>::traverseHelper(
INT_TYPE nodei,
INT_TYPE parent_nodei,
FUNCTORS &functors,
LOCAL_DATA* data_for_parent) const noexcept
{
const Node &node = myRoot[nodei];
bool descend = functors.pre(nodei, data_for_parent);
if (!descend)
return;
LOCAL_DATA local_data[N];
INT_TYPE s;
for (s = 0; s < N; ++s) {
const INT_TYPE node_int = node.child[s];
if (Node::isInternal(node_int)) {
if (node_int == Node::EMPTY) {
// NOTE: Anything after this will be empty too, so we can break.
break;
}
traverseHelper(Node::getInternalNum(node_int), nodei, functors, &local_data[s]);
}
else {
functors.item(node_int, nodei, local_data[s]);
}
}
// NOTE: s is now the number of non-empty entries in this node.
functors.post(nodei, parent_nodei, data_for_parent, s, local_data);
}
template<uint N>
template<typename LOCAL_DATA,typename FUNCTORS>
void BVH<N>::traverseParallel(
INT_TYPE parallel_threshold,
FUNCTORS& functors,
LOCAL_DATA* data_for_parent) const noexcept
{
if (!myRoot)
return;
// NOTE: The root is always index 0.
traverseParallelHelper(0, INT_TYPE(-1), parallel_threshold, myNumNodes, functors, data_for_parent);
}
template<uint N>
template<typename LOCAL_DATA,typename FUNCTORS>
void BVH<N>::traverseParallelHelper(
INT_TYPE nodei,
INT_TYPE parent_nodei,
INT_TYPE parallel_threshold,
INT_TYPE next_node_id,
FUNCTORS& functors,
LOCAL_DATA* data_for_parent) const noexcept
{
const Node &node = myRoot[nodei];
bool descend = functors.pre(nodei, data_for_parent);
if (!descend)
return;
// To determine the number of nodes in a child's subtree, we take the next
// node ID minus the current child's node ID.
INT_TYPE next_nodes[N];
INT_TYPE nnodes[N];
INT_TYPE nchildren = N;
INT_TYPE nparallel = 0;
// s is currently unsigned, so we check s < N for bounds check.
// The s >= 0 check is in case s ever becomes signed, and should be
// automatically removed by the compiler for unsigned s.
for (INT_TYPE s = N-1; (std::is_signed<INT_TYPE>::value ? (s >= 0) : (s < N)); --s) {
const INT_TYPE node_int = node.child[s];
if (node_int == Node::EMPTY) {
--nchildren;
continue;
}
next_nodes[s] = next_node_id;
if (Node::isInternal(node_int)) {
// NOTE: This depends on BVH<N>::initNode appending the child nodes
// in between their content, instead of all at once.
INT_TYPE child_node_id = Node::getInternalNum(node_int);
nnodes[s] = next_node_id - child_node_id;
next_node_id = child_node_id;
}
else {
nnodes[s] = 0;
}
nparallel += (nnodes[s] >= parallel_threshold);
}
LOCAL_DATA local_data[N];
if (nparallel >= 2) {
// Do any non-parallel ones first
if (nparallel < nchildren) {
for (INT_TYPE s = 0; s < N; ++s) {
if (nnodes[s] >= parallel_threshold) {
continue;
}
const INT_TYPE node_int = node.child[s];
if (Node::isInternal(node_int)) {
if (node_int == Node::EMPTY) {
// NOTE: Anything after this will be empty too, so we can break.
break;
}
traverseHelper(Node::getInternalNum(node_int), nodei, functors, &local_data[s]);
}
else {
functors.item(node_int, nodei, local_data[s]);
}
}
}
// Now do the parallel ones
UTparallelFor(UT_BlockedRange<INT_TYPE>(0,nparallel), [this,nodei,&node,&nnodes,&next_nodes,¶llel_threshold,&functors,&local_data](const UT_BlockedRange<INT_TYPE>& r) {
for (INT_TYPE taski = r.begin(); taski < r.end(); ++taski) {
INT_TYPE parallel_count = 0;
// NOTE: The check for s < N is just so that the compiler can
// (hopefully) figure out that it can fully unroll the loop.
INT_TYPE s;
for (s = 0; s < N; ++s) {
if (nnodes[s] < parallel_threshold) {
continue;
}
if (parallel_count == taski) {
break;
}
++parallel_count;
}
const INT_TYPE node_int = node.child[s];
if (Node::isInternal(node_int)) {
UT_ASSERT_MSG_P(node_int != Node::EMPTY, "Empty entries should have been excluded above.");
traverseParallelHelper(Node::getInternalNum(node_int), nodei, parallel_threshold, next_nodes[s], functors, &local_data[s]);
}
else {
functors.item(node_int, nodei, local_data[s]);
}
}
}, 0, 1);
}
else {
// All in serial
for (INT_TYPE s = 0; s < N; ++s) {
const INT_TYPE node_int = node.child[s];
if (Node::isInternal(node_int)) {
if (node_int == Node::EMPTY) {
// NOTE: Anything after this will be empty too, so we can break.
break;
}
traverseHelper(Node::getInternalNum(node_int), nodei, functors, &local_data[s]);
}
else {
functors.item(node_int, nodei, local_data[s]);
}
}
}
functors.post(nodei, parent_nodei, data_for_parent, nchildren, local_data);
}
template<uint N>
template<typename LOCAL_DATA,typename FUNCTORS>
void BVH<N>::traverseVector(
FUNCTORS &functors,
LOCAL_DATA* data_for_parent) const noexcept
{
if (!myRoot)
return;
// NOTE: The root is always index 0.
traverseVectorHelper(0, INT_TYPE(-1), functors, data_for_parent);
}
template<uint N>
template<typename LOCAL_DATA,typename FUNCTORS>
void BVH<N>::traverseVectorHelper(
INT_TYPE nodei,
INT_TYPE parent_nodei,
FUNCTORS &functors,
LOCAL_DATA* data_for_parent) const noexcept
{
const Node &node = myRoot[nodei];
INT_TYPE descend = functors.pre(nodei, data_for_parent);
if (!descend)
return;
LOCAL_DATA local_data[N];
INT_TYPE s;
for (s = 0; s < N; ++s) {
if ((descend>>s) & 1) {
const INT_TYPE node_int = node.child[s];
if (Node::isInternal(node_int)) {
if (node_int == Node::EMPTY) {
// NOTE: Anything after this will be empty too, so we can break.
descend &= (INT_TYPE(1)<<s)-1;
break;
}
traverseVectorHelper(Node::getInternalNum(node_int), nodei, functors, &local_data[s]);
}
else {
functors.item(node_int, nodei, local_data[s]);
}
}
}
// NOTE: s is now the number of non-empty entries in this node.
functors.post(nodei, parent_nodei, data_for_parent, s, local_data, descend);
}
template<uint N>
template<typename SRC_INT_TYPE>
void BVH<N>::createTrivialIndices(SRC_INT_TYPE* indices, const INT_TYPE n) noexcept {
constexpr INT_TYPE PARALLEL_THRESHOLD = 65536;
INT_TYPE ntasks = 1;
if (n >= PARALLEL_THRESHOLD) {
INT_TYPE nprocessors = UT_Thread::getNumProcessors();
ntasks = (nprocessors > 1) ? SYSmin(4*nprocessors, n/(PARALLEL_THRESHOLD/2)) : 1;
}
if (ntasks == 1) {
for (INT_TYPE i = 0; i < n; ++i) {
indices[i] = i;
}
}
else {
UTparallelFor(UT_BlockedRange<INT_TYPE>(0,ntasks), [indices,ntasks,n](const UT_BlockedRange<INT_TYPE>& r) {
for (INT_TYPE taski = r.begin(), taskend = r.end(); taski != taskend; ++taski) {
INT_TYPE start = (taski * exint(n))/ntasks;
INT_TYPE end = ((taski+1) * exint(n))/ntasks;
for (INT_TYPE i = start; i != end; ++i) {
indices[i] = i;
}
}
}, 0, 1);
}
}
template<uint N>
template<typename T,uint NAXES,typename BOX_TYPE,typename SRC_INT_TYPE>
void BVH<N>::computeFullBoundingBox(Box<T,NAXES>& axes_minmax, const BOX_TYPE* boxes, const INT_TYPE nboxes, SRC_INT_TYPE* indices) noexcept {
if (!nboxes) {
axes_minmax.initBounds();
return;
}
INT_TYPE ntasks = 1;
if (nboxes >= 2*4096) {
INT_TYPE nprocessors = UT_Thread::getNumProcessors();
ntasks = (nprocessors > 1) ? SYSmin(4*nprocessors, nboxes/4096) : 1;
}
if (ntasks == 1) {
Box<T,NAXES> box;
if (indices) {
box.initBounds(boxes[indices[0]]);
for (INT_TYPE i = 1; i < nboxes; ++i) {
box.combine(boxes[indices[i]]);
}
}
else {
box.initBounds(boxes[0]);
for (INT_TYPE i = 1; i < nboxes; ++i) {
box.combine(boxes[i]);
}
}
axes_minmax = box;
}
else {
// Combine boxes in parallel, into just a few boxes
UT_SmallArray<Box<T,NAXES>> parallel_boxes;
parallel_boxes.setSize(ntasks);
UTparallelFor(UT_BlockedRange<INT_TYPE>(0,ntasks), [¶llel_boxes,ntasks,boxes,nboxes,indices](const UT_BlockedRange<INT_TYPE>& r) {
for (INT_TYPE taski = r.begin(), end = r.end(); taski < end; ++taski) {
const INT_TYPE startbox = (taski*uint64(nboxes))/ntasks;
const INT_TYPE endbox = ((taski+1)*uint64(nboxes))/ntasks;
Box<T,NAXES> box;
if (indices) {
box.initBounds(boxes[indices[startbox]]);
for (INT_TYPE i = startbox+1; i < endbox; ++i) {
box.combine(boxes[indices[i]]);
}
}
else {
box.initBounds(boxes[startbox]);
for (INT_TYPE i = startbox+1; i < endbox; ++i) {
box.combine(boxes[i]);
}
}
parallel_boxes[taski] = box;
}
}, 0, 1);
// Combine parallel_boxes
Box<T,NAXES> box = parallel_boxes[0];
for (INT_TYPE i = 1; i < ntasks; ++i) {
box.combine(parallel_boxes[i]);
}
axes_minmax = box;
}
}
template<uint N>
template<BVH_Heuristic H,typename T,uint NAXES,typename BOX_TYPE,typename SRC_INT_TYPE>
void BVH<N>::initNode(UT_Array<Node>& nodes, Node &node, const Box<T,NAXES>& axes_minmax, const BOX_TYPE* boxes, SRC_INT_TYPE* indices, INT_TYPE nboxes) noexcept {
if (nboxes <= N) {
// Fits in one node
for (INT_TYPE i = 0; i < nboxes; ++i) {
node.child[i] = indices[i];
}
for (INT_TYPE i = nboxes; i < N; ++i) {
node.child[i] = Node::EMPTY;
}
return;
}
SRC_INT_TYPE* sub_indices[N+1];
Box<T,NAXES> sub_boxes[N];
if (N == 2) {
sub_indices[0] = indices;
sub_indices[2] = indices+nboxes;
split<H>(axes_minmax, boxes, indices, nboxes, sub_indices[1], &sub_boxes[0]);
}
else {
multiSplit<H>(axes_minmax, boxes, indices, nboxes, sub_indices, sub_boxes);
}
// Count the number of nodes to run in parallel and fill in single items in this node
INT_TYPE nparallel = 0;
static constexpr INT_TYPE PARALLEL_THRESHOLD = 1024;
for (INT_TYPE i = 0; i < N; ++i) {
INT_TYPE sub_nboxes = sub_indices[i+1]-sub_indices[i];
if (sub_nboxes == 1) {
node.child[i] = sub_indices[i][0];
}
else if (sub_nboxes >= PARALLEL_THRESHOLD) {
++nparallel;
}
}
// NOTE: Child nodes of this node need to be placed just before the nodes in
// their corresponding subtree, in between the subtrees, because
// traverseParallel uses the difference between the child node IDs
// to determine the number of nodes in the subtree.
// Recurse
if (nparallel >= 2) {
// Do the parallel ones first, so that they can be inserted in the right place.
// Although the choice may seem somewhat arbitrary, we need the results to be
// identical whether we choose to parallelize or not, and in case we change the
// threshold later.
UT_SmallArray<UT_Array<Node>> parallel_nodes;
parallel_nodes.setSize(nparallel);
UT_SmallArray<Node> parallel_parent_nodes;
parallel_parent_nodes.setSize(nparallel);
UTparallelFor(UT_BlockedRange<INT_TYPE>(0,nparallel), [¶llel_nodes,¶llel_parent_nodes,&sub_indices,boxes,&sub_boxes](const UT_BlockedRange<INT_TYPE>& r) {
for (INT_TYPE taski = r.begin(), end = r.end(); taski < end; ++taski) {
// First, find which child this is
INT_TYPE counted_parallel = 0;
INT_TYPE sub_nboxes;
INT_TYPE childi;
for (childi = 0; childi < N; ++childi) {
sub_nboxes = sub_indices[childi+1]-sub_indices[childi];
if (sub_nboxes >= PARALLEL_THRESHOLD) {
if (counted_parallel == taski) {
break;
}
++counted_parallel;
}
}
UT_ASSERT_P(counted_parallel == taski);
UT_Array<Node>& local_nodes = parallel_nodes[taski];
// Preallocate an overestimate of the number of nodes needed.
// At worst, we could have only 2 children in every leaf, and
// then above that, we have a geometric series with r=1/N and a=(sub_nboxes/2)/N
// The true worst case might be a little worst than this, but
// it's probably fairly unlikely.
local_nodes.setCapacity(nodeEstimate(sub_nboxes));
Node& parent_node = parallel_parent_nodes[taski];
// We'll have to fix the internal node numbers in parent_node and local_nodes later
initNode<H>(local_nodes, parent_node, sub_boxes[childi], boxes, sub_indices[childi], sub_nboxes);
}
}, 0, 1);
INT_TYPE counted_parallel = 0;
for (INT_TYPE i = 0; i < N; ++i) {
INT_TYPE sub_nboxes = sub_indices[i+1]-sub_indices[i];
if (sub_nboxes != 1) {
INT_TYPE local_nodes_start = nodes.size();
node.child[i] = Node::markInternal(local_nodes_start);
if (sub_nboxes >= PARALLEL_THRESHOLD) {
// First, adjust the root child node
Node child_node = parallel_parent_nodes[counted_parallel];
++local_nodes_start;
for (INT_TYPE childi = 0; childi < N; ++childi) {
INT_TYPE child_child = child_node.child[childi];
if (Node::isInternal(child_child) && child_child != Node::EMPTY) {
child_child += local_nodes_start;
child_node.child[childi] = child_child;
}
}
// Make space in the array for the sub-child nodes
const UT_Array<Node>& local_nodes = parallel_nodes[counted_parallel];
++counted_parallel;
INT_TYPE n = local_nodes.size();
nodes.bumpCapacity(local_nodes_start + n);
nodes.setSizeNoInit(local_nodes_start + n);
nodes[local_nodes_start-1] = child_node;
}
else {
nodes.bumpCapacity(local_nodes_start + 1);
nodes.setSizeNoInit(local_nodes_start + 1);
initNode<H>(nodes, nodes[local_nodes_start], sub_boxes[i], boxes, sub_indices[i], sub_nboxes);
}
}
}
// Now, adjust and copy all sub-child nodes that were made in parallel
adjustParallelChildNodes<PARALLEL_THRESHOLD>(nparallel, nodes, node, parallel_nodes.array(), sub_indices);
}
else {
for (INT_TYPE i = 0; i < N; ++i) {
INT_TYPE sub_nboxes = sub_indices[i+1]-sub_indices[i];
if (sub_nboxes != 1) {
INT_TYPE local_nodes_start = nodes.size();
node.child[i] = Node::markInternal(local_nodes_start);
nodes.bumpCapacity(local_nodes_start + 1);
nodes.setSizeNoInit(local_nodes_start + 1);
initNode<H>(nodes, nodes[local_nodes_start], sub_boxes[i], boxes, sub_indices[i], sub_nboxes);
}
}
}
}
template<uint N>
template<BVH_Heuristic H,typename T,uint NAXES,typename BOX_TYPE,typename SRC_INT_TYPE>
void BVH<N>::initNodeReorder(UT_Array<Node>& nodes, Node &node, const Box<T,NAXES>& axes_minmax, const BOX_TYPE* boxes, SRC_INT_TYPE* indices, INT_TYPE nboxes, const INT_TYPE indices_offset, const INT_TYPE max_items_per_leaf) noexcept {
if (nboxes <= N) {
// Fits in one node
for (INT_TYPE i = 0; i < nboxes; ++i) {
node.child[i] = indices_offset+i;
}
for (INT_TYPE i = nboxes; i < N; ++i) {
node.child[i] = Node::EMPTY;
}
return;
}
SRC_INT_TYPE* sub_indices[N+1];
Box<T,NAXES> sub_boxes[N];
if (N == 2) {
sub_indices[0] = indices;
sub_indices[2] = indices+nboxes;
split<H>(axes_minmax, boxes, indices, nboxes, sub_indices[1], &sub_boxes[0]);
}
else {
multiSplit<H>(axes_minmax, boxes, indices, nboxes, sub_indices, sub_boxes);
}
// Move any children with max_items_per_leaf or fewer indices before any children with more,
// for better cache coherence when we're accessing data in a corresponding array.
INT_TYPE nleaves = 0;
UT_SmallArray<SRC_INT_TYPE> leaf_indices;
SRC_INT_TYPE leaf_sizes[N];
INT_TYPE sub_nboxes0 = sub_indices[1]-sub_indices[0];
if (sub_nboxes0 <= max_items_per_leaf) {
leaf_sizes[0] = sub_nboxes0;
for (int j = 0; j < sub_nboxes0; ++j)
leaf_indices.append(sub_indices[0][j]);
++nleaves;
}
INT_TYPE sub_nboxes1 = sub_indices[2]-sub_indices[1];
if (sub_nboxes1 <= max_items_per_leaf) {
leaf_sizes[nleaves] = sub_nboxes1;
for (int j = 0; j < sub_nboxes1; ++j)
leaf_indices.append(sub_indices[1][j]);
++nleaves;
}
for (INT_TYPE i = 2; i < N; ++i) {
INT_TYPE sub_nboxes = sub_indices[i+1]-sub_indices[i];
if (sub_nboxes <= max_items_per_leaf) {
leaf_sizes[nleaves] = sub_nboxes;
for (int j = 0; j < sub_nboxes; ++j)
leaf_indices.append(sub_indices[i][j]);
++nleaves;
}
}
if (nleaves > 0) {
// NOTE: i < N condition is because INT_TYPE is unsigned.
// i >= 0 condition is in case INT_TYPE is changed to signed.
INT_TYPE move_distance = 0;
INT_TYPE index_move_distance = 0;
for (INT_TYPE i = N-1; (std::is_signed<INT_TYPE>::value ? (i >= 0) : (i < N)); --i) {
INT_TYPE sub_nboxes = sub_indices[i+1]-sub_indices[i];
if (sub_nboxes <= max_items_per_leaf) {
++move_distance;
index_move_distance += sub_nboxes;
}
else if (move_distance > 0) {
SRC_INT_TYPE *start_src_index = sub_indices[i];
for (SRC_INT_TYPE *src_index = sub_indices[i+1]-1; src_index >= start_src_index; --src_index) {
src_index[index_move_distance] = src_index[0];
}
sub_indices[i+move_distance] = sub_indices[i]+index_move_distance;
}
}
index_move_distance = 0;
for (INT_TYPE i = 0; i < nleaves; ++i) {
INT_TYPE sub_nboxes = leaf_sizes[i];
sub_indices[i] = indices+index_move_distance;
for (int j = 0; j < sub_nboxes; ++j)
indices[index_move_distance+j] = leaf_indices[index_move_distance+j];
index_move_distance += sub_nboxes;
}
}
// Count the number of nodes to run in parallel and fill in single items in this node
INT_TYPE nparallel = 0;
static constexpr INT_TYPE PARALLEL_THRESHOLD = 1024;
for (INT_TYPE i = 0; i < N; ++i) {
INT_TYPE sub_nboxes = sub_indices[i+1]-sub_indices[i];
if (sub_nboxes <= max_items_per_leaf) {
node.child[i] = indices_offset+(sub_indices[i]-sub_indices[0]);
}
else if (sub_nboxes >= PARALLEL_THRESHOLD) {
++nparallel;
}
}
// NOTE: Child nodes of this node need to be placed just before the nodes in
// their corresponding subtree, in between the subtrees, because
// traverseParallel uses the difference between the child node IDs
// to determine the number of nodes in the subtree.
// Recurse
if (nparallel >= 2) {
// Do the parallel ones first, so that they can be inserted in the right place.
// Although the choice may seem somewhat arbitrary, we need the results to be
// identical whether we choose to parallelize or not, and in case we change the
// threshold later.
UT_SmallArray<UT_Array<Node>,4*sizeof(UT_Array<Node>)> parallel_nodes;
parallel_nodes.setSize(nparallel);
UT_SmallArray<Node,4*sizeof(Node)> parallel_parent_nodes;
parallel_parent_nodes.setSize(nparallel);
UTparallelFor(UT_BlockedRange<INT_TYPE>(0,nparallel), [¶llel_nodes,¶llel_parent_nodes,&sub_indices,boxes,&sub_boxes,indices_offset,max_items_per_leaf](const UT_BlockedRange<INT_TYPE>& r) {
for (INT_TYPE taski = r.begin(), end = r.end(); taski < end; ++taski) {
// First, find which child this is
INT_TYPE counted_parallel = 0;
INT_TYPE sub_nboxes;
INT_TYPE childi;
for (childi = 0; childi < N; ++childi) {
sub_nboxes = sub_indices[childi+1]-sub_indices[childi];
if (sub_nboxes >= PARALLEL_THRESHOLD) {
if (counted_parallel == taski) {
break;
}
++counted_parallel;
}
}
UT_ASSERT_P(counted_parallel == taski);
UT_Array<Node>& local_nodes = parallel_nodes[taski];
// Preallocate an overestimate of the number of nodes needed.
// At worst, we could have only 2 children in every leaf, and
// then above that, we have a geometric series with r=1/N and a=(sub_nboxes/2)/N
// The true worst case might be a little worst than this, but
// it's probably fairly unlikely.
local_nodes.setCapacity(nodeEstimate(sub_nboxes));
Node& parent_node = parallel_parent_nodes[taski];
// We'll have to fix the internal node numbers in parent_node and local_nodes later
initNodeReorder<H>(local_nodes, parent_node, sub_boxes[childi], boxes, sub_indices[childi], sub_nboxes,
indices_offset+(sub_indices[childi]-sub_indices[0]), max_items_per_leaf);
}
}, 0, 1);
INT_TYPE counted_parallel = 0;
for (INT_TYPE i = 0; i < N; ++i) {
INT_TYPE sub_nboxes = sub_indices[i+1]-sub_indices[i];
if (sub_nboxes > max_items_per_leaf) {
INT_TYPE local_nodes_start = nodes.size();
node.child[i] = Node::markInternal(local_nodes_start);
if (sub_nboxes >= PARALLEL_THRESHOLD) {
// First, adjust the root child node
Node child_node = parallel_parent_nodes[counted_parallel];
++local_nodes_start;
for (INT_TYPE childi = 0; childi < N; ++childi) {
INT_TYPE child_child = child_node.child[childi];
if (Node::isInternal(child_child) && child_child != Node::EMPTY) {
child_child += local_nodes_start;
child_node.child[childi] = child_child;
}
}
// Make space in the array for the sub-child nodes
const UT_Array<Node>& local_nodes = parallel_nodes[counted_parallel];
++counted_parallel;
INT_TYPE n = local_nodes.size();
nodes.bumpCapacity(local_nodes_start + n);
nodes.setSizeNoInit(local_nodes_start + n);
nodes[local_nodes_start-1] = child_node;
}
else {
nodes.bumpCapacity(local_nodes_start + 1);
nodes.setSizeNoInit(local_nodes_start + 1);
initNodeReorder<H>(nodes, nodes[local_nodes_start], sub_boxes[i], boxes, sub_indices[i], sub_nboxes,
indices_offset+(sub_indices[i]-sub_indices[0]), max_items_per_leaf);
}
}
}
// Now, adjust and copy all sub-child nodes that were made in parallel
adjustParallelChildNodes<PARALLEL_THRESHOLD>(nparallel, nodes, node, parallel_nodes.array(), sub_indices);
}
else {
for (INT_TYPE i = 0; i < N; ++i) {
INT_TYPE sub_nboxes = sub_indices[i+1]-sub_indices[i];
if (sub_nboxes > max_items_per_leaf) {
INT_TYPE local_nodes_start = nodes.size();
node.child[i] = Node::markInternal(local_nodes_start);
nodes.bumpCapacity(local_nodes_start + 1);
nodes.setSizeNoInit(local_nodes_start + 1);
initNodeReorder<H>(nodes, nodes[local_nodes_start], sub_boxes[i], boxes, sub_indices[i], sub_nboxes,
indices_offset+(sub_indices[i]-sub_indices[0]), max_items_per_leaf);
}
}
}
}
template<uint N>
template<BVH_Heuristic H,typename T,uint NAXES,typename BOX_TYPE,typename SRC_INT_TYPE>
void BVH<N>::multiSplit(const Box<T,NAXES>& axes_minmax, const BOX_TYPE* boxes, SRC_INT_TYPE* indices, INT_TYPE nboxes, SRC_INT_TYPE* sub_indices[N+1], Box<T,NAXES> sub_boxes[N]) noexcept {
sub_indices[0] = indices;
sub_indices[2] = indices+nboxes;
split<H>(axes_minmax, boxes, indices, nboxes, sub_indices[1], &sub_boxes[0]);
if (N == 2) {
return;
}
if (H == BVH_Heuristic::MEDIAN_MAX_AXIS) {
SRC_INT_TYPE* sub_indices_startend[2*N];
Box<T,NAXES> sub_boxes_unsorted[N];
sub_boxes_unsorted[0] = sub_boxes[0];
sub_boxes_unsorted[1] = sub_boxes[1];
sub_indices_startend[0] = sub_indices[0];
sub_indices_startend[1] = sub_indices[1];
sub_indices_startend[2] = sub_indices[1];
sub_indices_startend[3] = sub_indices[2];
for (INT_TYPE nsub = 2; nsub < N; ++nsub) {
SRC_INT_TYPE* selected_start = sub_indices_startend[0];
SRC_INT_TYPE* selected_end = sub_indices_startend[1];
Box<T,NAXES> sub_box = sub_boxes_unsorted[0];
// Shift results back.
for (INT_TYPE i = 0; i < nsub-1; ++i) {
sub_indices_startend[2*i ] = sub_indices_startend[2*i+2];
sub_indices_startend[2*i+1] = sub_indices_startend[2*i+3];
}
for (INT_TYPE i = 0; i < nsub-1; ++i) {
sub_boxes_unsorted[i] = sub_boxes_unsorted[i-1];
}
// Do the split
split<H>(sub_box, boxes, selected_start, selected_end-selected_start, sub_indices_startend[2*nsub-1], &sub_boxes_unsorted[nsub]);
sub_indices_startend[2*nsub-2] = selected_start;
sub_indices_startend[2*nsub] = sub_indices_startend[2*nsub-1];
sub_indices_startend[2*nsub+1] = selected_end;
// Sort pointers so that they're in the correct order
sub_indices[N] = indices+nboxes;
for (INT_TYPE i = 0; i < N; ++i) {
SRC_INT_TYPE* prev_pointer = (i != 0) ? sub_indices[i-1] : nullptr;
SRC_INT_TYPE* min_pointer = nullptr;
Box<T,NAXES> box;
for (INT_TYPE j = 0; j < N; ++j) {
SRC_INT_TYPE* cur_pointer = sub_indices_startend[2*j];
if ((cur_pointer > prev_pointer) && (!min_pointer || (cur_pointer < min_pointer))) {
min_pointer = cur_pointer;
box = sub_boxes_unsorted[j];
}
}
UT_ASSERT_P(min_pointer);
sub_indices[i] = min_pointer;
sub_boxes[i] = box;
}
}
}
else {
T sub_box_areas[N];
sub_box_areas[0] = unweightedHeuristic<H>(sub_boxes[0]);
sub_box_areas[1] = unweightedHeuristic<H>(sub_boxes[1]);
for (INT_TYPE nsub = 2; nsub < N; ++nsub) {
// Choose which one to split
INT_TYPE split_choice = INT_TYPE(-1);
T max_heuristic;
for (INT_TYPE i = 0; i < nsub; ++i) {
const INT_TYPE index_count = (sub_indices[i+1]-sub_indices[i]);
if (index_count > 1) {
const T heuristic = sub_box_areas[i]*index_count;
if (split_choice == INT_TYPE(-1) || heuristic > max_heuristic) {
split_choice = i;
max_heuristic = heuristic;
}
}
}
UT_ASSERT_MSG_P(split_choice != INT_TYPE(-1), "There should always be at least one that can be split!");
SRC_INT_TYPE* selected_start = sub_indices[split_choice];
SRC_INT_TYPE* selected_end = sub_indices[split_choice+1];
// Shift results over; we can skip the one we selected.
for (INT_TYPE i = nsub; i > split_choice; --i) {
sub_indices[i+1] = sub_indices[i];
}
for (INT_TYPE i = nsub-1; i > split_choice; --i) {