-
Notifications
You must be signed in to change notification settings - Fork 42
/
Copy pathtiny_bvh_speedtest.cpp
1274 lines (1126 loc) · 45.4 KB
/
tiny_bvh_speedtest.cpp
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
#define TINYBVH_IMPLEMENTATION
#define INST_IDX_BITS 10 // reduces the size of the hit record to 16 bytes.
#include "tiny_bvh.h"
// 'screen resolution': see tiny_bvh_fenster.cpp; this program traces the
// same rays, but without visualization - just performance statistics.
#define SCRWIDTH 480
#define SCRHEIGHT 320
// GPU ray tracing
// #define ENABLE_OPENCL
#if 1
// tests to perform
// #define BUILD_MIDPOINT
#define BUILD_REFERENCE
#define BUILD_DOUBLE
#define BUILD_AVX
#define BUILD_NEON
#define BUILD_SBVH
#define REFIT_BVH2
#define REFIT_MBVH4
#define REFIT_MBVH8
#define TRAVERSE_2WAY_ST
#define TRAVERSE_ALT2WAY_ST
#define TRAVERSE_SOA2WAY_ST
#define TRAVERSE_4WAY
#define TRAVERSE_8WAY
#define TRAVERSE_2WAY_DBL
// #define TRAVERSE_CWBVH
#define TRAVERSE_2WAY_MT
#define TRAVERSE_2WAY_MT_PACKET
#define TRAVERSE_OPTIMIZED_ST
#define TRAVERSE_4WAY_OPTIMIZED
#define TRAVERSE_8WAY_OPTIMIZED
// #define EMBREE_BUILD // win64-only for now.
// #define EMBREE_TRAVERSE // win64-only for now.
// GPU rays: only if ENABLE_OPENCL is defined.
#define GPU_2WAY
#define GPU_4WAY
#define GPU_CWBVH
#else
// debug run
#define TRAVERSE_8WAY_OPTIMIZED
#endif
using namespace tinybvh;
#ifdef _MSC_VER
#include "Windows.h"
#endif
#ifdef _MSC_VER
#include "stdio.h" // for printf
#include "stdlib.h" // for rand
#else
#include <cstdio>
#endif
#ifdef _WIN32
#include <intrin.h> // for __cpuidex
#elif defined(__APPLE__) && defined(__MACH__)
// Keep ENABLE_OPENCL for APPLE
#elif defined ENABLE_OPENCL
#undef ENABLE_OPENCL
#endif
#if defined(__GNUC__) && defined(__x86_64__)
#include <cpuid.h>
#endif
#ifdef __EMSCRIPTEN__
#include <emscripten/version.h> // for __EMSCRIPTEN_major__, __EMSCRIPTEN_minor__
#endif
bvhvec4* triangles = 0;
#include <fstream>
int verts = 0;
float avgCost = 0;
float traceTime, buildTime, refitTime, * refDist = 0, * refDistFull = 0;
unsigned refOccluded[3] = {}, * refOccl[3] = {};
unsigned Nfull, Nsmall;
Ray* fullBatch[3], * smallBatch[3], * smallDiffuse[3];
Ray* shadowBatch[3];
#ifdef DOUBLE_PRECISION_SUPPORT
RayEx* doubleBatch[3];
#endif
// bvh layouts
BVH* bvh = new BVH();
BVH* ref_bvh = new BVH();
BVH_Verbose* bvh_verbose = 0;
BVH_Double* bvh_double = new BVH_Double();
BVH_SoA* bvh_soa = 0;
BVH_GPU* bvh_gpu = 0;
MBVH<4>* bvh4 = 0;
MBVH<8>* bvh8 = 0;
BVH4_CPU* bvh4_cpu = 0;
BVH4_GPU* bvh4_gpu = 0;
BVH8_CWBVH* cwbvh = 0;
BVH8_CPU* bvh8_cpu = 0;
enum { _DEFAULT = 1, _BVH, _VERBOSE, _DOUBLE, _SOA, _GPU2, _BVH4, _CPU4, _CPU4A, _CPU8, _GPU4, _BVH8, _CWBVH };
#if defined _WIN32 || defined _WIN64
#if defined EMBREE_BUILD || defined EMBREE_TRAVERSE
#include "embree4/rtcore.h"
static RTCScene embreeScene;
void embreeError( void* userPtr, enum RTCError error, const char* str )
{
printf( "error %d: %s\n", error, str );
}
#endif
#endif
#ifdef ENABLE_OPENCL
#define TINY_OCL_IMPLEMENTATION
#include "tiny_ocl.h"
#endif
float uniform_rand() { return (float)rand() / (float)RAND_MAX; }
void PrepareTest()
{
#ifdef _MSC_VER
// lock to a single core
SetThreadAffinityMask( GetCurrentThread(), 1 );
#endif
// clobber cashes to create a level playing field
static uint32_t* buffer = new uint32_t[8 * 1024 * 1024]; // 32MB should cover most CPU cashes
for (int p = 0, i = 0; i < 1000000; i++)
buffer[i]++, p = (p + 6353 /* prime */) & (8 * 1024 * 1024 - 1);
}
#include <chrono>
struct Timer
{
Timer() { reset(); }
float elapsed() const
{
auto t2 = std::chrono::high_resolution_clock::now();
return (float)std::chrono::duration_cast<std::chrono::duration<double>>(t2 - start).count();
}
void reset() { start = std::chrono::high_resolution_clock::now(); }
std::chrono::high_resolution_clock::time_point start;
};
float TestPrimaryRays( uint32_t layout, unsigned N, unsigned passes, float* avgCost = 0 )
{
// Primary rays: coherent batch of rays from a pinhole camera. One ray per
// pixel, organized in tiles to further increase coherence.
Timer t;
for (int view = 0; view < 3; view++)
{
Ray* batch = N == Nsmall ? smallBatch[view] : fullBatch[view];
for (unsigned i = 0; i < N; i++) batch[i].hit.t = 1e30f;
}
uint32_t travCost = 0;
for (unsigned pass = 0; pass < passes + 1; pass++)
{
uint32_t view = pass == 0 ? 0 : (3 - pass); // 0, 2, 1, 0
Ray* batch = N == Nsmall ? smallBatch[view] : fullBatch[view];
if (pass == 1)
{
Ray* batch = N == Nsmall ? smallBatch[0] : fullBatch[0];
for (unsigned i = 0; i < N; i++) batch[i].hit.t = 1e30f;
t.reset(); // first pass is cache warming
}
switch (layout)
{
case _BVH: for (unsigned i = 0; i < N; i++) travCost += bvh->Intersect( batch[i] ); break;
case _DEFAULT: for (unsigned i = 0; i < N; i++) travCost += ref_bvh->Intersect( batch[i] ); break;
case _GPU2: for (unsigned i = 0; i < N; i++) travCost += bvh_gpu->Intersect( batch[i] ); break;
case _GPU4: for (unsigned i = 0; i < N; i++) travCost += bvh4_gpu->Intersect( batch[i] ); break;
#ifdef BVH_USEAVX
case _CPU4: for (unsigned i = 0; i < N; i++) travCost += bvh4_cpu->Intersect( batch[i] ); break;
case _CWBVH: for (unsigned i = 0; i < N; i++) travCost += cwbvh->Intersect( batch[i] ); break;
case _SOA: for (unsigned i = 0; i < N; i++) travCost += bvh_soa->Intersect( batch[i] ); break;
case _CPU8: for (unsigned i = 0; i < N; i++) travCost += bvh8_cpu->Intersect( batch[i] ); break;
#endif
default: break;
};
}
if (avgCost) *avgCost = travCost / (float)(3 * N);
return t.elapsed() / passes;
}
float TestDiffuseRays( uint32_t layout, unsigned passes, float* avgCost = 0 )
{
// Diffuse rays: incoherent batch of rays resulting from a diffuse bounce.
Timer t;
for (int view = 0; view < 3; view++)
for (unsigned i = 0; i < Nsmall; i++) smallDiffuse[view][i].hit.t = 1e30f;
uint32_t travCost = 0;
for (unsigned pass = 0; pass < passes + 1; pass++)
{
uint32_t view = pass == 0 ? 0 : (3 - pass); // 0, 2, 1, 0
Ray* batch = smallDiffuse[view];
if (pass == 1)
{
for (unsigned i = 0; i < Nsmall; i++) smallDiffuse[0][i].hit.t = 1e30f;
t.reset(); // first pass is cache warming
}
switch (layout)
{
case _BVH: for (unsigned i = 0; i < Nsmall; i++) travCost += bvh->Intersect( batch[i] ); break;
case _DEFAULT: for (unsigned i = 0; i < Nsmall; i++) travCost += ref_bvh->Intersect( batch[i] ); break;
case _GPU2: for (unsigned i = 0; i < Nsmall; i++) travCost += bvh_gpu->Intersect( batch[i] ); break;
case _GPU4: for (unsigned i = 0; i < Nsmall; i++) travCost += bvh4_gpu->Intersect( batch[i] ); break;
#ifdef BVH_USEAVX
case _CPU4: for (unsigned i = 0; i < Nsmall; i++) travCost += bvh4_cpu->Intersect( batch[i] ); break;
case _CWBVH: for (unsigned i = 0; i < Nsmall; i++) travCost += cwbvh->Intersect( batch[i] ); break;
case _SOA: for (unsigned i = 0; i < Nsmall; i++) travCost += bvh_soa->Intersect( batch[i] ); break;
case _CPU8: for (unsigned i = 0; i < Nsmall; i++) travCost += bvh8_cpu->Intersect( batch[i] );
#endif
default: break;
};
}
if (avgCost) *avgCost = travCost / (float)(3 * Nsmall);
return t.elapsed() / passes;
}
#ifdef DOUBLE_PRECISION_SUPPORT
float TestPrimaryRaysEx( unsigned N, unsigned passes, float* avgCost = 0 )
{
// Primary rays: coherent batch of rays from a pinhole camera.
// Double-precision version.
Timer t;
for (int view = 0; view < 3; view++)
{
RayEx* batch = doubleBatch[view];
for (unsigned i = 0; i < N; i++) batch[i].hit.t = 1e30f;
}
uint32_t travCost = 0;
for (unsigned pass = 0; pass < passes + 1; pass++)
{
uint32_t view = pass == 0 ? 0 : (3 - pass); // 0, 2, 1, 0
RayEx* batch = doubleBatch[view];
if (pass == 1) t.reset(); // first pass is cache warming
for (unsigned i = 0; i < N; i++) travCost += bvh_double->Intersect( batch[i] );
}
if (avgCost) *avgCost = travCost / (float)(3 * N);
return t.elapsed() / passes;
}
void ValidateTraceResultEx( float* ref, unsigned N, unsigned line )
{
float refSum = 0;
double batchSum = 0;
for (unsigned i = 0; i < N; i += 4)
refSum += ref[i] == 1e30f ? 100 : ref[i],
batchSum += doubleBatch[0][i].hit.t == 1e300 ? 100 : doubleBatch[0][i].hit.t;
if (fabs( refSum - (float)batchSum ) / refSum < 0.0001f) return;
fprintf( stderr, "Validation failed on line %i.\n", line );
exit( 1 );
}
#endif
float TestShadowRays( uint32_t layout, unsigned N, unsigned passes )
{
// Shadow rays: coherent batch of rays from a single point to 'far away'. Shadow
// rays terminate on the first hit, and don't need sorted order. They also don't
// store intersection information, and are therefore expected to be faster than
// primary rays.
Timer t;
unsigned occluded = 0;
for (unsigned pass = 0; pass < passes + 1; pass++)
{
uint32_t view = pass == 0 ? 0 : (3 - pass); // 0, 2, 1, 0
Ray* batch = shadowBatch[view];
if (pass == 1) t.reset(); // first pass is cache warming
occluded = 0;
switch (layout)
{
case _DEFAULT: for (unsigned i = 0; i < N; i++) occluded += bvh->IsOccluded( batch[i] ); break;
#ifdef BVH_USEAVX
case _SOA: for (unsigned i = 0; i < N; i++) occluded += bvh_soa->IsOccluded( batch[i] ); break;
case _CPU4: for (unsigned i = 0; i < N; i++) occluded += bvh4_cpu->IsOccluded( batch[i] ); break;
#endif
case _GPU2: for (unsigned i = 0; i < N; i++) occluded += bvh_gpu->IsOccluded( batch[i] ); break;
#ifdef BVH_USEAVX2
case _CPU8: for (unsigned i = 0; i < N; i++) occluded += bvh8_cpu->IsOccluded( batch[i] ); break;
#endif
default: break;
}
}
// Shadow ray validation: The compacted triangle format used by some intersection
// kernels will lead to some diverging results. We check if no more than about
// 1/1000 checks differ. Shadow rays also use an origin offset, based on scene
// extend, to account for limited floating point accuracy.
if (abs( (int)occluded - (int)refOccluded[0] ) > 500) // allow some slack, we're using various tri intersectors
{
fprintf( stderr, "\nValidation for shadow rays failed (%i != %i).\n", (int)occluded, (int)refOccluded[0] );
// exit( 1 ); // don't terminate, just warn.
}
return t.elapsed() / passes;
}
void ValidateTraceResult( float* ref, unsigned N, unsigned line )
{
float refSum = 0, batchSum = 0;
Ray* batch = N == Nsmall ? smallBatch[0] : fullBatch[0];
for (unsigned i = 0; i < N; i += 4)
refSum += ref[i] == 1e30f ? 100 : ref[i],
batchSum += batch[i].hit.t == 1e30f ? 100 : batch[i].hit.t;
float diff = fabs( refSum - batchSum );
if (diff / refSum > 0.01f)
{
#if 1
printf( "!! Validation failed on line %i: %.1f != %.1f\n", line, refSum, batchSum );
#else
fprintf( stderr, "Validation failed on line %i - dumping img.raw.\n", line );
int step = (N == SCRWIDTH * SCRHEIGHT ? 1 : 16);
unsigned char pixel[SCRWIDTH * SCRHEIGHT];
for (unsigned i = 0, ty = 0; ty < SCRHEIGHT / 4; ty++) for (unsigned tx = 0; tx < SCRWIDTH / 4; tx++)
{
for (unsigned y = 0; y < 4; y++) for (unsigned x = 0; x < 4; x++, i += step)
{
float col = batch[i].hit.t == 1e30f ? 0 : batch[i].hit.t;
pixel[tx * 4 + x + (ty * 4 + y) * SCRWIDTH] = (unsigned char)((int)(col * 0.1f) & 255);
}
}
std::fstream s{ "img.raw", s.binary | s.out };
s.seekp( 0 );
s.write( (char*)&pixel, SCRWIDTH * SCRHEIGHT );
s.close();
exit( 1 );
#endif
}
}
// Multi-threading
#include <atomic>
#include <thread>
#include <vector>
static unsigned threadCount = std::thread::hardware_concurrency();
static std::atomic<int> batchIdx( 0 );
#if defined(TRAVERSE_2WAY_MT) || defined(ENABLE_OPENCL)
void IntersectBvhWorkerThread( int batchCount, Ray* fullBatch, int threadIdx )
{
int batch = threadIdx;
while (batch < batchCount)
{
const int batchStart = batch * 10000;
for (int i = 0; i < 10000; i++) bvh->Intersect( fullBatch[batchStart + i] );
batch = batchIdx++;
}
}
#endif
#ifdef TRAVERSE_2WAY_MT_PACKET
void IntersectBvh256WorkerThread( int batchCount, Ray* fullBatch, int threadIdx )
{
int batch = threadIdx;
while (batch < batchCount)
{
const int batchStart = batch * 30 * 256;
for (int i = 0; i < 30; i++) bvh->Intersect256Rays( fullBatch + batchStart + i * 256 );
batch = batchIdx++;
}
}
#endif
#ifdef BVH_USEAVX
void IntersectBvh256SSEWorkerThread( int batchCount, Ray* fullBatch, int threadIdx )
{
int batch = threadIdx;
while (batch < batchCount)
{
const int batchStart = batch * 30 * 256;
for (int i = 0; i < 30; i++) bvh->Intersect256RaysSSE( fullBatch + batchStart + i * 256 );
batch = batchIdx++;
}
}
#endif
int main()
{
int minor = TINY_BVH_VERSION_MINOR;
int major = TINY_BVH_VERSION_MAJOR;
int sub = TINY_BVH_VERSION_SUB;
printf( "tiny_bvh version %i.%i.%i performance statistics ", major, minor, sub );
// determine compiler
#ifdef _MSC_VER
printf( "(MSVC %i build)\n", _MSC_VER );
#elif defined __EMSCRIPTEN__
// EMSCRIPTEN needs to be before clang or gcc
printf( "(emcc %i.%i build)\n", __EMSCRIPTEN_major__, __EMSCRIPTEN_minor__ );
#elif defined __clang__
printf( "(clang %i.%i build)\n", __clang_major__, __clang_minor__ );
#elif defined __GNUC__
printf( "(gcc %i.%i build)\n", __GNUC__, __GNUC_MINOR__ );
#else
printf( "\n" );
#endif
// determine what CPU is running the tests.
#if (defined(__x86_64__) || defined(_M_X64)) && (defined (_WIN32) || defined(__GNUC__))
char model[64]{};
for (unsigned i = 0; i < 3; ++i)
{
#ifdef _WIN32
__cpuidex( (int*)(model + i * 16), i + 0x80000002, 0 );
#elif defined(__GNUC__)
__get_cpuid( i + 0x80000002,
(unsigned*)model + i * 4 + 0, (unsigned*)model + i * 4 + 1,
(unsigned*)model + i * 4 + 2, (unsigned*)model + i * 4 + 3 );
#endif
}
printf( "running on %s\n", model );
#endif
printf( "----------------------------------------------------------------\n" );
#ifdef ENABLE_OPENCL
// load and compile the OpenCL kernel code
// This also triggers OpenCL init and device identification.
tinyocl::Kernel ailalaine_kernel( "traverse.cl", "batch_ailalaine" );
tinyocl::Kernel gpu4way_kernel( "traverse.cl", "batch_gpu4way" );
tinyocl::Kernel cwbvh_kernel( "traverse.cl", "batch_cwbvh" );
printf( "----------------------------------------------------------------\n" );
#endif
// load raw vertex data for Crytek's Sponza
const std::string scene = "cryteksponza.bin";
std::string filename{ "./testdata/" };
filename += scene;
std::fstream s{ filename, s.binary | s.in };
s.seekp( 0 );
s.read( (char*)&verts, 4 );
printf( "Loading triangle data (%i tris).\n", verts );
verts *= 3, triangles = (bvhvec4*)tinybvh::malloc64( verts * sizeof( bvhvec4 ) );
s.read( (char*)triangles, verts * 16 );
// setup view pyramid for a pinhole camera:
// eye, p1 (top-left), p2 (top-right) and p3 (bottom-left)
bvhvec3 eyes[3] = {
bvhvec3( -15.24f, 21.5f, 2.54f ),
bvhvec3( -34, 5, 11.26f ),
bvhvec3( -1.3, 4.96, 12.28 )
}, eye = eyes[0];
bvhvec3 views[3] = {
tinybvh_normalize( bvhvec3( 0.826f, -0.438f, -0.356f ) ),
tinybvh_normalize( bvhvec3( 0.9427, 0.0292, -0.3324 ) ),
tinybvh_normalize( bvhvec3( -0.9886, 0.0507, -0.1419 ) )
}, view = views[0];
bvhvec3 right = tinybvh_normalize( tinybvh_cross( bvhvec3( 0, 1, 0 ), view ) );
bvhvec3 up = 0.8f * tinybvh_cross( view, right ), C = eye + 2 * view;
bvhvec3 p1 = C - right + up, p2 = C + right + up, p3 = C - right - up;
// generate primary rays in a cacheline-aligned buffer - and, for data locality:
// organized in 4x4 pixel tiles, 16 samples per pixel, so 256 rays per tile.
// one set for each camera position / direction.
for (int i = 0; i < 3; i++)
{
Nfull = Nsmall = 0;
fullBatch[i] = (Ray*)tinybvh::malloc64( SCRWIDTH * SCRHEIGHT * 16 * sizeof( Ray ) );
smallBatch[i] = (Ray*)tinybvh::malloc64( SCRWIDTH * SCRHEIGHT * 2 * sizeof( Ray ) );
smallDiffuse[i] = (Ray*)tinybvh::malloc64( SCRWIDTH * SCRHEIGHT * 2 * sizeof( Ray ) );
#ifdef DOUBLE_PRECISION_SUPPORT
doubleBatch[i] = (RayEx*)tinybvh::malloc64( SCRWIDTH * SCRHEIGHT * 2 * sizeof( RayEx ) );
#endif
for (int ty = 0; ty < SCRHEIGHT / 4; ty++) for (int tx = 0; tx < SCRWIDTH / 4; tx++)
{
for (int y = 0; y < 4; y++) for (int x = 0; x < 4; x++)
{
int pixel_x = tx * 4 + x;
int pixel_y = ty * 4 + y;
for (int s = 0; s < 16; s++) // 16 samples per pixel
{
float u = (float)(pixel_x * 4 + (s & 3)) / (SCRWIDTH * 4);
float v = (float)(pixel_y * 4 + (s >> 2)) / (SCRHEIGHT * 4);
bvhvec3 P = p1 + u * (p2 - p1) + v * (p3 - p1);
fullBatch[i][Nfull++] = Ray( eye, tinybvh_normalize( P - eye ) );
if ((s & 7) == 0)
{
smallBatch[i][Nsmall] = fullBatch[i][Nfull - 1];
#ifdef DOUBLE_PRECISION_SUPPORT
tinybvh::bvhdbl3 O = smallBatch[i][Nsmall].O;
tinybvh::bvhdbl3 D = smallBatch[i][Nsmall].D;
doubleBatch[i][Nsmall] = RayEx( O, D );
#endif
Nsmall++;
}
}
}
}
}
// T I N Y _ B V H P E R F O R M A N C E M E A S U R E M E N T S
Timer t;
// measure single-core bvh construction time - warming caches
printf( "BVH construction speed\n" );
printf( "warming caches... " );
bvh->Build( triangles, verts / 3 );
printf( "creating diffuse rays...\n" );
refDist = new float[Nsmall];
for (int i = 0; i < 3; i++) for (unsigned j = 0; j < Nsmall; j++)
{
const bvhvec3 O = smallBatch[i][j].O, D = smallBatch[i][j].D;
bvhvec3 I, R = tinybvh_normalize( bvhvec3( uniform_rand() - 0.5f, uniform_rand() - 0.5f, uniform_rand() - 0.5f ) );
Ray ray( O, D );
bvh->Intersect( ray );
if (i == 0) refDist[j] = ray.hit.t;
if (ray.hit.t < 100)
{
I = O + ray.hit.t * D;
uint32_t primIdx = ray.hit.prim;
int v0idx = primIdx * 3, v1idx = v0idx + 1, v2idx = v0idx + 2;
bvhvec3 v0 = triangles[v0idx], v1 = triangles[v1idx], v2 = triangles[v2idx];
bvhvec3 N = tinybvh_normalize( tinybvh_cross( v1 - v0, v2 - v0 ) );
if (tinybvh_dot( N, D ) > 0) N *= -1.0f;
if (tinybvh_dot( N, R ) < 0) R *= -1.0f;
}
else I = O + 20.0f * D;
smallDiffuse[i][j] = Ray( I + 0.001f * R, R );
}
#ifdef BUILD_MIDPOINT
// measure single-core bvh construction time - quick bvh builder
printf( "- quick bvh builder: " );
t.reset();
for (int pass = 0; pass < 3; pass++) bvh->BuildQuick( triangles, verts / 3 );
buildTime = t.elapsed() / 3.0f;
printf( "%7.2fms for %7i triangles ", buildTime * 1000.0f, verts / 3 );
printf( "- %6i nodes, SAH=%.2f\n", bvh->usedNodes, bvh->SAHCost() );
#endif
#ifdef BUILD_REFERENCE
// measure single-core bvh construction time - reference builder
printf( "- reference builder: " );
t.reset();
for (int pass = 0; pass < 3; pass++) bvh->Build( triangles, verts / 3 );
buildTime = t.elapsed() / 3.0f;
TestPrimaryRays( _BVH, Nsmall, 3, &avgCost );
printf( "%7.2fms for %7i triangles ", buildTime * 1000.0f, verts / 3 );
printf( "- %6i nodes, SAH=%.2f, rayCost=%.2f\n", bvh->usedNodes, bvh->SAHCost(), avgCost );
#endif
#if defined BUILD_DOUBLE && defined DOUBLE_PRECISION_SUPPORT
// measure single-core bvh construction time - double-precision builder
printf( "- 'double' builder: " );
t.reset();
tinybvh::bvhdbl3* triEx = (tinybvh::bvhdbl3*)tinybvh::malloc64( verts * sizeof( tinybvh::bvhdbl3 ) );
for (int i = 0; i < verts; i++)
triEx[i].x = (double)triangles[i].x,
triEx[i].y = (double)triangles[i].y,
triEx[i].z = (double)triangles[i].z;
bvh_double->Build( triEx, verts / 3 );
buildTime = t.elapsed();
TestPrimaryRaysEx( Nsmall, 3, &avgCost );
printf( "%7.2fms for %7i triangles ", buildTime * 1000.0f, verts / 3 );
printf( "- %6i nodes, SAH=%.2f, rayCost=%.2f\n", bvh->usedNodes, bvh->SAHCost(), avgCost );
#endif
#if defined BUILD_AVX && defined BVH_USEAVX
// measure single-core bvh construction time - AVX builder
printf( "- fast AVX builder: " );
t.reset();
for (int pass = 0; pass < 3; pass++) bvh->BuildAVX( triangles, verts / 3 );
buildTime = t.elapsed() / 3.0f;
TestPrimaryRays( _BVH, Nsmall, 3, &avgCost );
printf( "%7.2fms for %7i triangles ", buildTime * 1000.0f, verts / 3 );
printf( "- %6i nodes, SAH=%.2f, rayCost=%.2f\n", bvh->usedNodes, bvh->SAHCost(), avgCost );
#endif
#if defined BUILD_NEON && defined BVH_USENEON
// measure single-core bvh construction time - NEON builder
printf( "- fast NEON builder: " );
t.reset();
for (int pass = 0; pass < 3; pass++) bvh->BuildNEON( triangles, verts / 3 );
buildTime = t.elapsed() / 3.0f;
TestPrimaryRays( _BVH, Nsmall, 3, &avgCost );
printf( "%7.2fms for %7i triangles ", buildTime * 1000.0f, verts / 3 );
printf( "- %6i nodes, SAH=%.2f, rayCost=%.2f\n", bvh->usedNodes, bvh->SAHCost(), avgCost );
#endif
#ifdef BUILD_SBVH
// measure single-core bvh construction time - AVX builder
printf( "- HQ (SBVH) builder: " );
t.reset();
for (int pass = 0; pass < 2; pass++) bvh->BuildHQ( triangles, verts / 3 );
buildTime = t.elapsed() / 2.0f;
TestPrimaryRays( _BVH, Nsmall, 3, &avgCost );
printf( "%7.2fms for %7i triangles ", buildTime * 1000.0f, verts / 3 );
printf( "- %6i nodes, SAH=%.2f, rayCost=%.2f\n", bvh->usedNodes, bvh->SAHCost(), avgCost );
#endif
// measure single-core bvh construction time - warming caches
printf( "BVH refitting speed\n" );
#ifdef REFIT_BVH2
// measure single-core bvh refit time
printf( "- BVH2 refitting: " );
BVH tmpBVH;
tmpBVH.Build( triangles, verts / 3 );
float sahBefore = tmpBVH.SAHCost();
for (int pass = 0; pass < 10; pass++)
{
if (pass == 1) t.reset();
tmpBVH.Refit();
}
refitTime = t.elapsed() / 9.0f;
printf( "%7.2fms for %7i triangles ", refitTime * 1000.0f, verts / 3 );
printf( "- SAH=%.2f\n", tmpBVH.SAHCost() );
#endif
#ifdef REFIT_MBVH4
// measure single-core mbvh refit time
printf( "- BVH4 refitting: " );
MBVH<4> tmpBVH4;
tmpBVH4.Build( triangles, verts / 3 );
for (int pass = 0; pass < 10; pass++)
{
if (pass == 1) t.reset();
tmpBVH4.Refit();
}
refitTime = t.elapsed() / 9.0f;
printf( "%7.2fms for %7i triangles ", refitTime * 1000.0f, verts / 3 );
printf( "- SAH=%.2f\n", tmpBVH4.SAHCost() );
#endif
#ifdef REFIT_MBVH8
// measure single-core mbvh refit time
printf( "- BVH8 refitting: " );
MBVH<8> tmpBVH8;
tmpBVH8.Build( triangles, verts / 3 );
for (int pass = 0; pass < 10; pass++)
{
if (pass == 1) t.reset();
tmpBVH8.Refit();
}
refitTime = t.elapsed() / 9.0f;
printf( "%7.2fms for %7i triangles ", refitTime * 1000.0f, verts / 3 );
printf( "- SAH=%.2f\n", tmpBVH8.SAHCost() );
#endif
#if defined _WIN32 || defined _WIN64
#if defined EMBREE_BUILD || defined EMBREE_TRAVERSE
// convert data to correct format for Embree and build a BVH
printf( "- Embree builder: " );
RTCDevice embreeDevice = rtcNewDevice( NULL );
rtcSetDeviceErrorFunction( embreeDevice, embreeError, NULL );
embreeScene = rtcNewScene( embreeDevice );
RTCGeometry embreeGeom = rtcNewGeometry( embreeDevice, RTC_GEOMETRY_TYPE_TRIANGLE );
float* vertices = (float*)rtcSetNewGeometryBuffer( embreeGeom, RTC_BUFFER_TYPE_VERTEX, 0, RTC_FORMAT_FLOAT3, 3 * sizeof( float ), verts );
unsigned* indices = (unsigned*)rtcSetNewGeometryBuffer( embreeGeom, RTC_BUFFER_TYPE_INDEX, 0, RTC_FORMAT_UINT3, 3 * sizeof( unsigned ), verts / 3 );
for (int i = 0; i < verts; i++)
{
vertices[i * 3 + 0] = triangles[i].x, vertices[i * 3 + 1] = triangles[i].y;
vertices[i * 3 + 2] = triangles[i].z, indices[i] = i; // Note: not using shared vertices.
}
rtcSetGeometryBuildQuality( embreeGeom, RTC_BUILD_QUALITY_HIGH ); // max quality
// rtcSetGeometryBuildQuality( embreeGeom, RTC_BUILD_QUALITY_MEDIUM ); // default quality
rtcCommitGeometry( embreeGeom );
rtcAttachGeometry( embreeScene, embreeGeom );
rtcReleaseGeometry( embreeGeom );
rtcSetSceneBuildQuality( embreeScene, RTC_BUILD_QUALITY_HIGH );
// rtcSetSceneBuildQuality( embreeScene, RTC_BUILD_QUALITY_MEDIUM );
t.reset();
rtcCommitScene( embreeScene ); // assuming this is where (supposedly threaded) BVH build happens.
buildTime = t.elapsed();
printf( "%7.2fms for %7i triangles\n", buildTime * 1000.0f, verts / 3 );
#endif
#endif
// report CPU single ray, single-core performance
printf( "BVH traversal speed - single-threaded\n" );
ref_bvh->Build( triangles, verts / 3 );
// estimate correct shadow ray epsilon based on scene extends
tinybvh::bvhvec4 bmin( 1e30f ), bmax( -1e30f );
for (int i = 0; i < verts; i++)
bmin = tinybvh::tinybvh_min( bmin, triangles[i] ),
bmax = tinybvh::tinybvh_max( bmax, triangles[i] );
tinybvh::bvhvec3 e = bmax - bmin;
float maxExtent = tinybvh::tinybvh_max( tinybvh::tinybvh_max( e.x, e.y ), e.z );
float shadowEpsilon = maxExtent * 5e-7f;
// setup proper shadow ray batch
traceTime = TestPrimaryRays( _DEFAULT, Nsmall, 1 ); // just to generate intersection points
for (int view = 0; view < 3; view++)
{
shadowBatch[view] = (Ray*)tinybvh::malloc64( sizeof( Ray ) * Nsmall );
const tinybvh::bvhvec3 lightPos( 0, 0, 0 );
for (unsigned i = 0; i < Nsmall; i++)
{
float t = tinybvh::tinybvh_min( 1000.0f, smallBatch[view][i].hit.t );
bvhvec3 I = smallBatch[view][i].O + t * smallBatch[view][i].D;
bvhvec3 D = tinybvh_normalize( lightPos - I );
shadowBatch[view][i] = Ray( I + D * shadowEpsilon, D, tinybvh_length( lightPos - I ) - shadowEpsilon );
}
// get reference shadow ray query result
refOccluded[view] = 0, refOccl[view] = new unsigned[Nsmall];
for (unsigned i = 0; i < Nsmall; i++)
refOccluded[view] += (refOccl[view][i] = ref_bvh->IsOccluded( shadowBatch[view][i] ) ? 1 : 0);
}
#ifdef TRAVERSE_2WAY_ST
// WALD_32BYTE - Have this enabled at all times if validation is desired.
printf( "- BVH (plain) - primary: " );
PrepareTest();
traceTime = TestPrimaryRays( _DEFAULT, Nsmall, 3 );
// printf( "%4.2fM in %5.1fms (%7.2fMRays/s), ", (float)Nsmall * 1e-6f, traceTime * 1000, (float)Nsmall / traceTime * 1e-6f );
printf( "%7.2fMRays/s, ", (float)Nsmall / traceTime * 1e-6f );
traceTime = TestShadowRays( _DEFAULT, Nsmall, 3 );
// printf( "shadow: %5.1fms (%7.2fMRays/s)\n", traceTime * 1000, (float)Nsmall / traceTime * 1e-6f );
printf( "shadow: %7.2fMRays/s, ", (float)Nsmall / traceTime * 1e-6f );
traceTime = TestDiffuseRays( _DEFAULT, 3 );
printf( "diffuse: %7.2fMRays/s\n", (float)Nsmall / traceTime * 1e-6f );
#endif
#ifdef TRAVERSE_ALT2WAY_ST
// GPU
if (!bvh_gpu)
{
bvh_gpu = new BVH_GPU();
bvh_gpu->BuildHQ( triangles, verts / 3 );
}
printf( "- BVH_GPU - primary: " );
PrepareTest();
traceTime = TestPrimaryRays( _GPU2, Nsmall, 3 );
ValidateTraceResult( refDist, Nsmall, __LINE__ );
// printf( "%4.2fM rays in %5.1fms (%7.2fMRays/s), ", (float)Nsmall * 1e-6f, traceTime * 1000, (float)Nsmall / traceTime * 1e-6f );
printf( "%7.2fMRays/s, ", (float)Nsmall / traceTime * 1e-6f );
traceTime = TestShadowRays( _GPU2, Nsmall, 3 );
// printf( "shadow: %5.1fms (%7.2fMRays/s)\n", traceTime * 1000, (float)Nsmall / traceTime * 1e-6f );
printf( "shadow: %7.2fMRays/s, ", (float)Nsmall / traceTime * 1e-6f );
traceTime = TestDiffuseRays( _GPU2, 3 );
printf( "diffuse: %7.2fMRays/s\n", (float)Nsmall / traceTime * 1e-6f );
#endif
#if defined TRAVERSE_SOA2WAY_ST && defined BVH_USEAVX // BVH_SoA::IsOccluded is not available for NEON yet.
// SOA
if (!bvh_soa)
{
bvh_soa = new BVH_SoA();
bvh_soa->BuildHQ( triangles, verts / 3 );
}
printf( "- BVH_SOA - primary: " );
PrepareTest();
traceTime = TestPrimaryRays( _SOA, Nsmall, 3 );
ValidateTraceResult( refDist, Nsmall, __LINE__ );
// printf( "%4.2fM rays in %5.1fms (%7.2fMRays/s), ", (float)Nsmall * 1e-6f, traceTime * 1000, (float)Nsmall / traceTime * 1e-6f );
printf( "%7.2fMRays/s, ", (float)Nsmall / traceTime * 1e-6f );
traceTime = TestShadowRays( _SOA, Nsmall, 3 );
// printf( "shadow: %5.1fms (%7.2fMRays/s)\n", traceTime * 1000, (float)Nsmall / traceTime * 1e-6f );
printf( "shadow: %7.2fMRays/s, ", (float)Nsmall / traceTime * 1e-6f );
traceTime = TestDiffuseRays( _SOA, 3 );
printf( "diffuse: %7.2fMRays/s\n", (float)Nsmall / traceTime * 1e-6f );
#endif
#ifdef TRAVERSE_4WAY
// BVH4_CPU
if (!bvh4_cpu)
{
bvh4_cpu = new BVH4_CPU();
bvh4_cpu->BuildHQ( triangles, verts / 3 );
}
printf( "- BVH4_CPU - primary: " );
PrepareTest();
traceTime = TestPrimaryRays( _CPU4, Nsmall, 3 );
ValidateTraceResult( refDist, Nsmall, __LINE__ );
// printf( "%4.2fM rays in %5.1fms (%7.2fMRays/s), ", (float)Nsmall * 1e-6f, traceTime * 1000, (float)Nsmall / traceTime * 1e-6f );
printf( "%7.2fMRays/s, ", (float)Nsmall / traceTime * 1e-6f );
traceTime = TestShadowRays( _CPU4, Nsmall, 3 );
// printf( "shadow: %5.1fms (%7.2fMRays/s)\n", traceTime * 1000, (float)Nsmall / traceTime * 1e-6f );
printf( "shadow: %7.2fMRays/s, ", (float)Nsmall / traceTime * 1e-6f );
traceTime = TestDiffuseRays( _CPU4, 3 );
printf( "diffuse: %7.2fMRays/s\n", (float)Nsmall / traceTime * 1e-6f );
#endif
#if defined TRAVERSE_8WAY && defined BVH_USEAVX && defined BVH_USEAVX2
// BVH8_CPU
if (!bvh8_cpu)
{
bvh8_cpu = new BVH8_CPU();
bvh8_cpu->BuildHQ( triangles, verts / 3 );
}
printf( "- BVH8_CPU - primary: " );
PrepareTest();
traceTime = TestPrimaryRays( _CPU8, Nsmall, 3 );
ValidateTraceResult( refDist, Nsmall, __LINE__ );
// printf( "%4.2fM rays in %5.1fms (%7.2fMRays/s), ", (float)Nsmall * 1e-6f, traceTime * 1000, (float)Nsmall / traceTime * 1e-6f );
printf( "%7.2fMRays/s, ", (float)Nsmall / traceTime * 1e-6f );
traceTime = TestShadowRays( _CPU8, Nsmall, 3 );
// printf( "shadow: %5.1fms (%7.2fMRays/s)\n", traceTime * 1000, (float)Nsmall / traceTime * 1e-6f );
printf( "shadow: %7.2fMRays/s, ", (float)Nsmall / traceTime * 1e-6f );
traceTime = TestDiffuseRays( _CPU8, 3 );
printf( "diffuse: %7.2fMRays/s\n", (float)Nsmall / traceTime * 1e-6f );
#endif
#if defined TRAVERSE_2WAY_DBL && defined BUILD_DOUBLE && defined DOUBLE_PRECISION_SUPPORT
// double-precision Rays/BVH
printf( "- BVH_DOUBLE - primary: " );
PrepareTest();
traceTime = TestPrimaryRaysEx( Nsmall, 3 );
ValidateTraceResultEx( refDist, Nsmall, __LINE__ );
// printf( "%4.2fM rays in %5.1fms (%7.2fMRays/s)\n", (float)Nsmall * 1e-6f, traceTime * 1000, (float)Nsmall / traceTime * 1e-6f );
printf( "%7.2fMRays/s\n", (float)Nsmall / traceTime * 1e-6f );
#endif
#ifdef TRAVERSE_CWBVH
#ifdef BVH_USEAVX
// CWBVH - Not efficient on CPU.
if (!cwbvh)
{
cwbvh = new BVH8_CWBVH();
cwbvh->BuildHQ( triangles, verts / 3 );
}
printf( "- BVH8/CWBVH - primary: " );
traceTime = TestPrimaryRays( _CWBVH, Nsmall, 3 );
ValidateTraceResult( refDist, Nsmall, __LINE__ );
// printf( "%4.2fM rays in %5.1fms (%7.2fMRays/s)\n", (float)Nsmall * 1e-6f, traceTime * 1000, (float)Nsmall / traceTime * 1e-6f );
printf( "%7.2fMRays/s\n", (float)Nsmall / traceTime * 1e-6f );
#endif
#endif
#if defined TRAVERSE_OPTIMIZED_ST || defined TRAVERSE_4WAY_OPTIMIZED || defined TRAVERSE_8WAY_OPTIMIZED
printf( "Optimized BVH performance - Optimizing... " );
PrepareTest();
bvh->Build( triangles, verts / 3 );
float prevSAH = bvh->SAHCost();
if (!bvh_verbose)
{
bvh_verbose = new BVH_Verbose();
bvh_verbose->ConvertFrom( *bvh );
}
t.reset();
bvh->Optimize( 50, true );
TestPrimaryRays( _BVH, Nsmall, 3, &avgCost );
printf( "done (%.2fs). New: %i nodes, SAH=%.2f to %.2f, rayCost=%.2f\n", t.elapsed(), bvh->NodeCount(), prevSAH, bvh->SAHCost(), avgCost );
#endif
#ifdef TRAVERSE_OPTIMIZED_ST
#ifdef BVH_USEAVX
// ALT_SOA
delete bvh_soa;
// Building a BVH_SoA over an optimized BVH: Careful, do not delete the
// passed BVH; we use some of its data in the BVH_SoA.
bvh_soa = new BVH_SoA();
bvh_soa->ConvertFrom( *bvh );
printf( "- BVH_SOA - primary: " );
PrepareTest();
traceTime = TestPrimaryRays( _SOA, Nsmall, 3 );
ValidateTraceResult( refDist, Nsmall, __LINE__ );
// printf( "%4.2fM rays in %5.1fms (%7.2fMRays/s), ", (float)Nsmall * 1e-6f, traceTime * 1000, (float)Nsmall / traceTime * 1e-6f );
printf( "%7.2fMRays/s, ", (float)Nsmall / traceTime * 1e-6f );
traceTime = TestShadowRays( _SOA, Nsmall, 3 );
// printf( "shadow: %5.1fms (%7.2fMRays/s)\n", traceTime * 1000, (float)Nsmall / traceTime * 1e-6f );
printf( "shadow: %7.2fMRays/s, ", (float)Nsmall / traceTime * 1e-6f );
traceTime = TestDiffuseRays( _SOA, 3 );
printf( "diffuse: %7.2fMRays/s\n", (float)Nsmall / traceTime * 1e-6f );
#endif
#endif
#ifdef TRAVERSE_4WAY_OPTIMIZED
// BVH4_CPU
delete bvh4;
delete bvh4_cpu;
// Building a BVH4_CPU over an optimized BVH: Careful, do not delete the
// passed BVH; we use some of its data in the BVH4_CPU.
bvh4 = new MBVH<4>();
bvh4_cpu = new BVH4_CPU();
bvh4->ConvertFrom( *bvh );
bvh4_cpu->ConvertFrom( *bvh4 );
printf( "- BVH4_CPU - primary: " );
PrepareTest();
traceTime = TestPrimaryRays( _CPU4, Nsmall, 3 );
ValidateTraceResult( refDist, Nsmall, __LINE__ );
// printf( "%4.2fM rays in %5.1fms (%7.2fMRays/s), ", (float)Nsmall * 1e-6f, traceTime * 1000, (float)Nsmall / traceTime * 1e-6f );
printf( "%7.2fMRays/s, ", (float)Nsmall / traceTime * 1e-6f );
traceTime = TestShadowRays( _CPU4, Nsmall, 3 );
// printf( "shadow: %5.1fms (%7.2fMRays/s)\n", traceTime * 1000, (float)Nsmall / traceTime * 1e-6f );
printf( "shadow: %7.2fMRays/s, ", (float)Nsmall / traceTime * 1e-6f );
traceTime = TestDiffuseRays( _CPU4, 3 );
printf( "diffuse: %7.2fMRays/s\n", (float)Nsmall / traceTime * 1e-6f );
#endif
#ifdef TRAVERSE_8WAY_OPTIMIZED
// BVH8_CPU
delete bvh8_cpu;
bvh8_cpu = new BVH8_CPU();
bvh8_cpu->Build( triangles, verts / 3 );
bvh8_cpu->Optimize( 50, true );
printf( "- BVH8_CPU - primary: " );
PrepareTest();
traceTime = TestPrimaryRays( _CPU8, Nsmall, 3 );
ValidateTraceResult( refDist, Nsmall, __LINE__ );
// printf( "%4.2fM rays in %5.1fms (%7.2fMRays/s), ", (float)Nsmall * 1e-6f, traceTime * 1000, (float)Nsmall / traceTime * 1e-6f );
printf( "%7.2fMRays/s, ", (float)Nsmall / traceTime * 1e-6f );
traceTime = TestShadowRays( _CPU8, Nsmall, 3 );
// printf( "shadow: %5.1fms (%7.2fMRays/s)\n", traceTime * 1000, (float)Nsmall / traceTime * 1e-6f );
printf( "shadow: %7.2fMRays/s, ", (float)Nsmall / traceTime * 1e-6f );
traceTime = TestDiffuseRays( _CPU8, 3 );
printf( "diffuse: %7.2fMRays/s\n", (float)Nsmall / traceTime * 1e-6f );
#endif
#ifdef ENABLE_OPENCL
// report GPU performance
printf( "BVH traversal speed - GPU (OpenCL)\n" );
// calculate full res reference distances using threaded traversal on CPU.
const int batchCount = Nfull / 10000;
batchIdx = threadCount;
std::vector<std::thread> threads;
for (unsigned i = 0; i < Nfull; i++) fullBatch[0][i].hit.t = 1e30f;
for (uint32_t i = 0; i < threadCount; i++)
threads.emplace_back( &IntersectBvhWorkerThread, batchCount, fullBatch[0], i );
for (auto& thread : threads) thread.join();
refDistFull = new float[Nfull];
for (unsigned i = 0; i < Nfull; i++) refDistFull[i] = fullBatch[0][i].hit.t;
#ifdef GPU_2WAY
// trace the rays on GPU using OpenCL
printf( "- BVH_GPU - primary: " );
if (!bvh_gpu)
{
bvh_gpu = new BVH_GPU();
bvh_gpu->BuildHQ( triangles, verts / 3 );
}
// create OpenCL buffers for the BVH data calculated by tiny_bvh.h
tinyocl::Buffer gpuNodes( bvh_gpu->usedNodes * sizeof( BVH_GPU::BVHNode ), bvh_gpu->bvhNode );
tinyocl::Buffer idxData( bvh_gpu->idxCount * sizeof( unsigned ), bvh_gpu->bvh.primIdx );
tinyocl::Buffer triData( bvh_gpu->triCount * 3 * sizeof( tinybvh::bvhvec4 ), triangles );
// synchronize the host-side data to the gpu side
gpuNodes.CopyToDevice();
idxData.CopyToDevice();
triData.CopyToDevice();
// create rays and send them to the gpu side
tinyocl::Buffer rayData( Nfull * 64 /* sizeof( tinybvh::Ray ) */ );
// the size of the ray struct exceeds 64 bytes because of the large Intersection struct.
// Here we chop this off, since on the GPU side, the ray is precisely 64 bytes.
for (unsigned i = 0; i < Nfull; i++)
memcpy( (unsigned char*)rayData.GetHostPtr() + 64 * i, &fullBatch[0][i], 64 );
rayData.CopyToDevice();
// create an event to time the OpenCL kernel
cl_event event;
cl_ulong startTime, endTime;