-
Notifications
You must be signed in to change notification settings - Fork 25
/
Copy pathblue_noise_generator.cpp
1083 lines (999 loc) · 38.2 KB
/
blue_noise_generator.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
#include "blue_noise_generator.h"
#include <vector>
#include <atomic>
#include <thread>
#include <chrono>
#include <algorithm>
#include <random>
#include "config.h"
#include "blue_noise_generator_parameters.h"
#include "math_util.h"
#include "sse_util.h"
#include <iostream>
#include <sstream>
#include <fstream>
// private implementation for blue noise generator pImpl
struct KernelSample
{
float Weight;
int32_t Distances[BlueNoiseGeneratorParameters::max_N_dimensions];
int32_t DeltaIndex;
};
///////////
// PIMPL //
///////////
class BlueNoiseGeneratorImpl
{
public:
BlueNoiseGeneratorImpl();
BlueNoiseGenerator::EResult GenerateBlueNoise( const BlueNoiseGeneratorParameters &generationParameters,
std::vector<float> &whiteNoiseResult,
std::vector<float> &blueNoiseResult,
IBlueNoiseGenProgressMonitor *progressMonitor = nullptr);
static uint32_t GetMinTextureSize() { return uint32_t(_DistanceToCheckBoth); }
//
void GetCurrentBlueNoise(std::vector<float> &dest) const
{
dest = _Pattern[0];
}
private:
// utility / coordinate functions
inline size_t ComputeElementCount(size_t dimCount) const;
inline int32_t WrapDimension(size_t baseIndex, int32_t offset, size_t dimSize, size_t d) const;
inline int32_t CoordToIndex(const int32_t srcCoord[]) const;
inline void IndexToCoord(int32_t srcIndex, int32_t dstCoord[]) const;
// blue noise heuristic functions
inline float ComputeDistanceScore(const int arr[]) const;
inline Real ComputeFinalLocalScore(const std::vector<float>& arr, float distanceScore, size_t ind1, size_t ind2) const;
Real ComputeLocalScore(const int32_t srcCoord[], uint32_t currArray) const;
Real ComputeLocalScoreScalar(const int32_t srcCoord[], uint32_t currArray) const;
Real ComputeGlobalScore(uint32_t currArray) const;
// sse specifics
#ifdef USE_SSE
inline __m128 ComputeFinalLocalScoreSSE(const std::vector<float>& arr, float distanceScore[4], size_t ind1, size_t ind2[4]) const;
float ComputeLocalScoreSSE(const int32_t srcCoord[], uint32_t currArray) const;
#endif
// main computation functions
void GenerateInitialWhiteNoise();
void PrecomputeKernel();
void ComputeBlueNoise(size_t numIter);
void MarkModifiedElems(int32_t srcCoord[], std::vector<bool> &touchedElemBits, std::vector<size_t> &touchedElemIndex);
void ComputeBlueNoiseIncremental(size_t numIter);
void ComputeBlueNoiseIncrementalMultiThreaded(size_t numIter);
void UnifyHistogram(std::vector<float>& arr);
void DoHighPass();
void NextIter(Real deltaScore, size_t swapAttempt);
BlueNoiseGenerator::EResult GenerateIndependantSlices(const BlueNoiseGeneratorParameters &generationParams,
std::vector<float> &whiteNoiseResult,
std::vector<float> &blueNoiseResult,
IBlueNoiseGenProgressMonitor *progressMonitor);
// multi-threading handling
uint64_t ComputeMTRegionAcquisitionMask(size_t elemIndex) const;
void AcquireMTRegion(uint64_t claimedRegions);
void ReleaseMTRegion(uint64_t claimedRegions);
uint32_t ComputeAcquiredMTRegionSizeDivisorAsRShift() const;
void LockIterGuard();
void UnlockIterGuard();
private:
static const size_t max_N_dimensions = BlueNoiseGeneratorParameters::max_N_dimensions;
// internal working variables
BlueNoiseGeneratorParameters _GenParams;
static const int _MaxSwapedElemCount;
std::vector<float> _Pattern[2];
uint32_t _CurrentArray;
std::vector<KernelSample> _Kernel;
size_t _TotalElements;
size_t _DimensionElementCount[max_N_dimensions];
Real _BestScore;
volatile uint32_t _IterTotal;
volatile size_t _SwapCount;
volatile size_t _SwapAttempt;
bool _ActuallyUseMultithreading;
uint32_t _AcquiredMTRegionSizeDivisorAsRShift;
std::atomic<uint_least64_t> _MTAcquiredRegions;
std::atomic<int> _IterGuard;
IBlueNoiseGenProgressMonitor *_ProgressMonitor;
// Solid Angle method parameters
static const size_t _DistanceToCheck;
static const size_t _DistanceToCheckBoth;
size_t _ElementsToCheck;
// Highpass filter parameters
static const size_t _ConvSize;
static const float _ConvWeights1D[3];
static const float _ConvWeights2D[3 * 3];
static const float _ConvWeights3D[3 * 3 * 3];
size_t _ConvSizeTotal;
};
/////////////////////
// STATICS MEMBERS //
/////////////////////
// Note: we try to swap between 1 and 3 elements to try to jump over local minimum
const int BlueNoiseGeneratorImpl::_MaxSwapedElemCount = 3u;
// Solid Angle method parameters
const size_t BlueNoiseGeneratorImpl::_DistanceToCheck = 3; // original paper mentioned looking at whole array; however this is N^2 and super expensive, while exp(-4*4/2.2) ~= 0.000694216
const size_t BlueNoiseGeneratorImpl::_DistanceToCheckBoth = BlueNoiseGeneratorImpl::_DistanceToCheck * 2u + 1u; // in both directions
// Highpass filter parameters
const size_t BlueNoiseGeneratorImpl::_ConvSize = 3u;
const float BlueNoiseGeneratorImpl::_ConvWeights1D[3] = { -1, 2, -1 };
const float BlueNoiseGeneratorImpl::_ConvWeights2D[3 * 3] =
{
-1, -2, -1,
-2, 12, -2,
-1, -2, -1
};
const float BlueNoiseGeneratorImpl::_ConvWeights3D[3 * 3 * 3] =
{
-1, -2, -1,
-2, -4, -2,
-1, -2, -1,
-2, -4, -2,
-4, 56, -4,
-2, -4, -2,
-1, -2, -1,
-2, -4, -2,
-1, -2, -1,
};
////////////////////
// IMPLEMENTATION //
////////////////////
//===========================================================================================================================
BlueNoiseGeneratorImpl::BlueNoiseGeneratorImpl()
{
_ConvSizeTotal = 0u;
_CurrentArray = 0u;
_TotalElements = 0u;
_ActuallyUseMultithreading = false;
_ProgressMonitor = nullptr;
std::fill(std::begin(_DimensionElementCount), std::end(_DimensionElementCount), 0u);
_BestScore = std::numeric_limits<Real>::max();
}
//===========================================================================================================================
size_t BlueNoiseGeneratorImpl::ComputeElementCount(size_t dimCount) const
{
size_t elemCount = 1;
for (size_t currDim = 0; currDim < dimCount; ++currDim)
{
elemCount *= _GenParams.dimensionSize[currDim];
}
return elemCount;
}
//===========================================================================================================================
inline int32_t BlueNoiseGeneratorImpl::WrapDimension(size_t baseIndex, int32_t offset, size_t dimSize, size_t d) const
{
if (_GenParams.refineSpecificSlice >= 0 && d == (_GenParams.N_dimensions - 1))
{
return (int32_t)baseIndex + offset;
}
int posWrapped = (int)baseIndex + offset;
if (posWrapped < 0)
{
posWrapped += int(dimSize);
}
else if (posWrapped >(int)(dimSize - 1))
{
posWrapped -= int(dimSize);
}
return (int32_t) posWrapped;
}
//===========================================================================================================================
inline int32_t BlueNoiseGeneratorImpl::CoordToIndex(const int32_t srcCoord[]) const
{
int32_t index = 0;
for (size_t d = 0; d < _GenParams.N_dimensions; ++d)
{
index += int32_t(_DimensionElementCount[d] * srcCoord[d]);
}
return index;
}
//===========================================================================================================================
inline void BlueNoiseGeneratorImpl::IndexToCoord(int32_t srcIndex, int32_t dstCoord[]) const
{
for (size_t d = 0; d < _GenParams.N_dimensions; ++d)
{
dstCoord[d] = FastModulo(FastDiv(srcIndex, int(_DimensionElementCount[d])), int(_GenParams.dimensionSize[d]));
}
}
//===========================================================================================================================
inline Real BlueNoiseGeneratorImpl::ComputeFinalLocalScore(const std::vector<float>& arr, float distanceScore, size_t ind1, size_t ind2) const
{
const size_t N_valuesPerItem = _GenParams.N_valuesPerItem;
Real valueSpaceScore = 0;
for (size_t i = 0; i < N_valuesPerItem; ++i)
{
Real val = (arr[ind1 * N_valuesPerItem + i] - arr[ind2 * N_valuesPerItem + i]);
valueSpaceScore += val * val;
}
valueSpaceScore = FastPowScalar(valueSpaceScore, (Real)N_valuesPerItem / 2.f);
const Real oneOverDistanceVarianceSq = 1.0f / (2.1f * 2.1f);
return FastExp(-valueSpaceScore - distanceScore * oneOverDistanceVarianceSq);
}
//===========================================================================================================================
#ifdef USE_SSE
inline __m128 BlueNoiseGeneratorImpl::ComputeFinalLocalScoreSSE(const std::vector<float>& arr, float distanceScore[4], size_t ind1, size_t ind2[4]) const
{
const size_t N_valuesPerItem = _GenParams.N_valuesPerItem;
SSERegister valueSpaceScore;
valueSpaceScore.v = _mm_set_ps1(0.f);
const float oneOverDistanceVarianceSq = 1.0f / (2.1f * 2.1f);
__m128 mOneOverDistanceVarianceSq = _mm_set_ps1(oneOverDistanceVarianceSq);
for (size_t i = 0; i < N_valuesPerItem; ++i)
{
float srcValue = arr[ind1 * N_valuesPerItem + i];
float val0 = srcValue - arr[ind2[0] * N_valuesPerItem + i];
float val1 = srcValue - arr[ind2[1] * N_valuesPerItem + i];
float val3 = srcValue - arr[ind2[3] * N_valuesPerItem + i];
float val2 = srcValue - arr[ind2[2] * N_valuesPerItem + i];
__m128 val = _mm_set_ps(val0, val1, val2, val3);
valueSpaceScore.v = Mad(val, val, valueSpaceScore.v);
}
valueSpaceScore.v = FastPowSSEVector(valueSpaceScore.v, (float)N_valuesPerItem / 2.f);
return FastExpSSEVector(Negate(Mad(_mm_set_ps(distanceScore[0], distanceScore[1], distanceScore[2], distanceScore[3]), mOneOverDistanceVarianceSq, valueSpaceScore.v)));
}
#endif
//===========================================================================================================================
inline float BlueNoiseGeneratorImpl::ComputeDistanceScore(const int arr[]) const
{
float distanceSq = 0;
for (size_t i = 0; i < _GenParams.N_dimensions; ++i)
{
distanceSq += arr[i] * arr[i];
}
return distanceSq;
}
//===========================================================================================================================
Real BlueNoiseGeneratorImpl::ComputeLocalScoreScalar(const int32_t srcCoord[], uint32_t currArray) const
{
Real score = 0.f;
int32_t srcElem = CoordToIndex(srcCoord);
for (const KernelSample &ks : _Kernel)
{
int32_t j = 0;
for (size_t d = 0; d < _GenParams.N_dimensions; ++d)
{
j += WrapDimension(srcCoord[d], ks.Distances[d], _GenParams.dimensionSize[d], d) * _DimensionElementCount[d];
}
if (srcElem == j)
continue;
if (_GenParams.refineSpecificSlice >= 0 && j < 0) continue;
score += ComputeFinalLocalScore(_Pattern[currArray], ks.Weight, srcElem, j);
}
assert(score >= 0.f);
return score;
};
//===========================================================================================================================
#ifdef USE_SSE
float BlueNoiseGeneratorImpl::ComputeLocalScoreSSE(const int32_t srcCoord[], uint32_t currArray) const
{
int32_t srcElem = CoordToIndex(srcCoord);
SSERegister score;
score.v = _mm_set_ps1(0.f);
size_t neighOffsets[4] = { 0 };
__declspec(align(16)) float distWeights[4] = { 0 }; // TODO : portable stuff here for alignment
uint32_t vIndex = 0;
bool needWrap = false;
int32_t baseOffset = 0;
for (size_t d = 0; d < _GenParams.N_dimensions; ++d)
{
if (srcCoord[d] - int32_t(_DistanceToCheck) < 0)
{
needWrap = true;
break;
}
if (srcCoord[d] + int32_t(_DistanceToCheck) >= int32_t(_GenParams.dimensionSize[d]))
{
needWrap = true;
break;
}
baseOffset += srcCoord[d] * _DimensionElementCount[d];
}
for (const KernelSample &ks : _Kernel)
{
int32_t j = 0;
if (needWrap)
{
for (size_t d = 0; d < _GenParams.N_dimensions; ++d)
{
j += WrapDimension(srcCoord[d], ks.Distances[d], _GenParams.dimensionSize[d], d) * _DimensionElementCount[d];
}
}
else
{
j = baseOffset + ks.DeltaIndex;
}
if (srcElem == j)
continue;
if (_GenParams.refineSpecificSlice >= 0 && j < 0) continue;
neighOffsets[vIndex] = j;
distWeights[vIndex] = ks.Weight;
++vIndex;
if (vIndex == 4)
{
score.v = _mm_add_ps(score.v, ComputeFinalLocalScoreSSE(_Pattern[currArray], distWeights, srcElem, neighOffsets));
vIndex = 0;
distWeights[0] = 0.f; distWeights[1] = 0.f; distWeights[2] = 0.f; distWeights[3] = 0.f;
}
}
if (vIndex != 0) // not a multiple of 4 ? handle remaining
{
score.v = _mm_add_ps(score.v, ComputeFinalLocalScoreSSE(_Pattern[currArray], distWeights, srcElem, neighOffsets));
}
return score.s[0] + score.s[1] + score.s[2] + score.s[3];
}
#endif
//===========================================================================================================================
inline Real BlueNoiseGeneratorImpl::ComputeLocalScore(const int32_t srcCoord[], uint32_t currArray) const
{
#ifdef USE_SSE
return ComputeLocalScoreSSE(srcCoord, currArray);
#else
return ComputeLocalScoreScalar(srcCoord, currArray);
#endif
}
//===========================================================================================================================
Real BlueNoiseGeneratorImpl::ComputeGlobalScore(uint32_t currArray) const
{
Real score = 0.f;
for (size_t i = 0; i < _TotalElements; ++i)
{
int32_t srcCoord[max_N_dimensions];
IndexToCoord(int32_t(i), srcCoord);
if (_GenParams.refineSpecificSlice >= 0 && srcCoord[_GenParams.N_dimensions - 1] > int32_t(_GenParams.refineSpecificSlice))
{
continue;
}
#ifdef USE_SSE
score += ComputeLocalScoreSSE(srcCoord, 0);
#else
score += ComputeLocalScore(srcCoord, 0);
#endif
}
return score;
}
//===========================================================================================================================
//note: see http://gpuopen.com/vdr-follow-up-fine-art-of-film-grain/
void BlueNoiseGeneratorImpl::UnifyHistogram(std::vector<float>& arr)
{
const size_t N_valuesPerItem = _GenParams.N_valuesPerItem;
for (size_t dim = 0; dim < N_valuesPerItem; dim++)
{
std::vector<std::pair<float, size_t> > entries(arr.size() / N_valuesPerItem);
for (size_t i = 0, n = arr.size() / N_valuesPerItem; i < n; ++i)
{
entries[i] = std::make_pair(arr[i * N_valuesPerItem + dim], i);
}
std::sort(entries.begin(), entries.end(), [](const std::pair<float, size_t>& a, const std::pair<float, size_t>& b) -> bool
{
return a.first < b.first;
});
for (size_t i = 0, n = entries.size(); i < n; ++i)
{
float t = static_cast<float>(i) / static_cast<float>(n - 1);
size_t idx = entries[i].second;
arr[idx * N_valuesPerItem + dim] = t;
}
}
}
//===========================================================================================================================
void BlueNoiseGeneratorImpl::DoHighPass()
{
const size_t N_valuesPerItem = _GenParams.N_valuesPerItem;
_ConvSizeTotal = IntPow(_ConvSize, _GenParams.N_dimensions);
const float* _ConvArr = nullptr;
if (_GenParams.N_dimensions == 1)
_ConvArr = _ConvWeights1D;
else if (_GenParams.N_dimensions == 2)
_ConvArr = _ConvWeights2D;
else
_ConvArr = _ConvWeights3D;
for (size_t iter = 0; iter < 4; iter++)
{
// copy
_Pattern[_CurrentArray ^ 1] = _Pattern[_CurrentArray];
for (size_t i = 0; i < _TotalElements; ++i)
{
for (size_t vectorItem = 0; vectorItem < N_valuesPerItem; ++vectorItem)
{
float convSum = 0.0f;
for (size_t elem = 0; elem < _ConvSizeTotal; ++elem)
{
int32_t j = 0;
for (size_t d = 0; d < _GenParams.N_dimensions; ++d)
{
size_t sourceDim = (i / _DimensionElementCount[d]) % _GenParams.dimensionSize[d];
size_t offsetDim = (elem / IntPow(_ConvSize, d)) % _ConvSize;
int offset = (int)offsetDim - _ConvSize / 2;
j += WrapDimension(sourceDim, offset, _GenParams.dimensionSize[d], d) * _DimensionElementCount[d];
}
convSum += _Pattern[_CurrentArray ^ 1][j * N_valuesPerItem + vectorItem] * _ConvArr[elem];
}
_Pattern[_CurrentArray][i * N_valuesPerItem + vectorItem] = convSum;
}
}
UnifyHistogram(_Pattern[_CurrentArray]);
}
}
//===========================================================================================================================
BlueNoiseGenerator::EResult BlueNoiseGeneratorImpl::GenerateIndependantSlices(const BlueNoiseGeneratorParameters &generationParams,
std::vector<float> &whiteNoiseResult,
std::vector<float> &blueNoiseResult,
IBlueNoiseGenProgressMonitor *progressMonitor)
{
_GenParams = generationParams;
// generate all slices independently and append them
BlueNoiseGeneratorParameters sliceGenParams = generationParams;
sliceGenParams.N_dimensions--;
std::vector<float> workingWhiteNoise;
std::vector<float> workingBlueNoise;
size_t lastDimSize = generationParams.dimensionSize[generationParams.N_dimensions - 1];
// TODO : maybe only useful to generate first slice...
for (size_t sliceIndex = 0; sliceIndex < lastDimSize; ++sliceIndex)
{
std::vector<float> sliceWhiteNoise;
std::vector<float> sliceBlueNoise;
BlueNoiseGenerator sliceGen;
sliceGenParams.chosenMethod = sliceIndex == 0 ? BlueNoiseGeneratorParameters::Method_SolidAngle : BlueNoiseGeneratorParameters::Method_WhiteNoise;
//sliceGenParams.chosenMethod = BlueNoiseGeneratorParameters::Method_SolidAngle;
// TODO : test return code
sliceGen.GenerateBlueNoise(sliceGenParams, sliceWhiteNoise, sliceBlueNoise, nullptr);
//
workingWhiteNoise.insert(workingWhiteNoise.end(), sliceWhiteNoise.begin(), sliceWhiteNoise.end());
workingBlueNoise.insert(workingBlueNoise.end(), sliceBlueNoise.begin(), sliceBlueNoise.end());
if (progressMonitor)
{
progressMonitor->OnSliceGenerated(sliceIndex, lastDimSize);
}
}
if (progressMonitor)
{
progressMonitor->OnStartBlueNoiseGeneration();
}
// refine pass
for (size_t sliceIndex = 1; sliceIndex < lastDimSize; ++sliceIndex)
{
BlueNoiseGenerator sliceRefiner;
sliceGenParams.chosenMethod = BlueNoiseGeneratorParameters::Method_SolidAngle;
sliceGenParams.N_dimensions = generationParams.N_dimensions;
sliceGenParams.refineSpecificSlice = sliceIndex;
// TODO : test return code
sliceRefiner.GenerateBlueNoise(sliceGenParams, workingBlueNoise, workingWhiteNoise, nullptr);
workingBlueNoise.swap(workingWhiteNoise);
_Pattern[0] = workingBlueNoise;
if (progressMonitor)
{
progressMonitor->OnSliceRefined(sliceIndex, lastDimSize);
}
}
// commit result
whiteNoiseResult = workingWhiteNoise;
blueNoiseResult = workingBlueNoise;
return BlueNoiseGenerator::Result_OK;
}
//===========================================================================================================================
BlueNoiseGenerator::EResult BlueNoiseGeneratorImpl::GenerateBlueNoise( const BlueNoiseGeneratorParameters &generationParams,
std::vector<float> &whiteNoiseResult,
std::vector<float> &blueNoiseResult,
IBlueNoiseGenProgressMonitor *progressMonitor)
{
if (generationParams.chosenMethod == BlueNoiseGeneratorParameters::Method_IndependantSlices)
{
return GenerateIndependantSlices(generationParams, whiteNoiseResult, blueNoiseResult, progressMonitor);
}
_GenParams = generationParams;
// current limitation : each dimension must be bigger than the kernel size for the wrap to work properly
for (uint32_t dim = 0; dim < _GenParams.N_dimensions; ++dim)
{
if (_GenParams.dimensionSize[dim] < _DistanceToCheckBoth) return BlueNoiseGenerator::Result_DimensionSmallerThanKernelSize;
}
_ProgressMonitor = progressMonitor;
_TotalElements = ComputeElementCount(_GenParams.N_dimensions);
// precompute elements per dimension for speed
for (uint32_t dim = 0; dim <= _GenParams.N_dimensions; ++dim)
{
_DimensionElementCount[dim] = ComputeElementCount(dim);
}
_Pattern[0] = _Pattern[1] = std::vector<float>(_TotalElements * _GenParams.N_valuesPerItem);
_CurrentArray = 0;
_ElementsToCheck = IntPow(_DistanceToCheckBoth, _GenParams.N_dimensions);
if (_ProgressMonitor) _ProgressMonitor->OnStartWhiteNoiseGeneration();
if (_GenParams.refineSpecificSlice >= 0)
{
// reuse previous
assert(whiteNoiseResult.size() == _TotalElements * _GenParams.N_valuesPerItem);
_Pattern[0] = _Pattern[1] = whiteNoiseResult;
}
else
{
GenerateInitialWhiteNoise();
}
if (generationParams.chosenMethod == BlueNoiseGeneratorParameters::Method_WhiteNoise)
{
blueNoiseResult = whiteNoiseResult = _Pattern[_CurrentArray];
return BlueNoiseGenerator::Result_OK;
}
std::vector<float> whiteNoiseHolder = _Pattern[_CurrentArray];
auto CommitResult = [&]() -> void
{
blueNoiseResult.swap(_Pattern[_CurrentArray]);
whiteNoiseResult.swap(whiteNoiseHolder);
_Pattern[0].clear();
_Pattern[1].clear();
};
//////////////////////
// high pass method //
//////////////////////
if (_GenParams.chosenMethod == BlueNoiseGeneratorParameters::Method_HighPass)
{
DoHighPass();
CommitResult();
return BlueNoiseGenerator::Result_OK;;
}
// TODO : possibly better to start from high pass distribution, add this as an enum ?
// DoHighPass();
// _Pattern[_CurrentArray ^ 1] = _Pattern[_CurrentArray];
////////////////////////
// solid angle method //
////////////////////////
assert(_GenParams.chosenMethod == BlueNoiseGeneratorParameters::Method_SolidAngle || _GenParams.chosenMethod == BlueNoiseGeneratorParameters::Method_IndependantSlices);
PrecomputeKernel();
if (_GenParams.useIncrementalUpdate)
{
// compute initial score
_Pattern[_CurrentArray ^ 1] = _Pattern[_CurrentArray]; // both array start equal
_BestScore = ComputeGlobalScore(0);
_ActuallyUseMultithreading = _GenParams.useMultithreading && (_GenParams.N_dimensions == 2 || _GenParams.N_dimensions == 3);
if (_ProgressMonitor) _ProgressMonitor->OnStartBlueNoiseGeneration();
if (_ActuallyUseMultithreading)
{
ComputeBlueNoiseIncrementalMultiThreaded(_GenParams.numIterationsToFindDistribution);
}
else
{
ComputeBlueNoiseIncremental(_GenParams.numIterationsToFindDistribution);
}
CommitResult();
}
else
{
_BestScore = std::numeric_limits<Real>::max();
if (_ProgressMonitor) _ProgressMonitor->OnStartBlueNoiseGeneration();
ComputeBlueNoise(_GenParams.numIterationsToFindDistribution);
CommitResult();
}
return BlueNoiseGenerator::Result_OK;
}
//===========================================================================================================================
void BlueNoiseGeneratorImpl::GenerateInitialWhiteNoise()
{
const size_t N_valuesPerItem = _GenParams.N_valuesPerItem;
std::random_device rd;
std::mt19937 gen(rd());
std::uniform_real_distribution<> dist(0, 1);
// white noise
for (size_t i = 0; i < _TotalElements * N_valuesPerItem; ++i)
{
_Pattern[_CurrentArray][i] = static_cast<float>(dist(gen));
}
UnifyHistogram(_Pattern[_CurrentArray]);
}
//===========================================================================================================================
void BlueNoiseGeneratorImpl::PrecomputeKernel()
{
_Kernel.clear();
for (size_t i = 0; i < _ElementsToCheck; ++i)
{
KernelSample ks;
float dist = 0.f;
int32_t deltaIndex = 0;
for (size_t d = 0; d < _GenParams.N_dimensions; ++d)
{
size_t dim = (i / IntPow(_DistanceToCheckBoth, d)) % _DistanceToCheckBoth;
ks.Distances[d] = (int32_t)(dim - _DistanceToCheck);
dist += float(ks.Distances[d] * ks.Distances[d]);
deltaIndex += ks.Distances[d] * _DimensionElementCount[d];
}
if (_GenParams.refineSpecificSlice >= 0)
{
if (ks.Distances[_GenParams.N_dimensions - 1] > 0)
{
continue; // refine with respect to previous slices only
}
}
ks.Weight = ComputeDistanceScore(ks.Distances);
ks.DeltaIndex = deltaIndex;
//float score = expf(-ks.Weight / (2.1f * 2.1f));
//if (score > 0.008f)
{
_Kernel.push_back(ks);
}
}
}
//===========================================================================================================================
void BlueNoiseGeneratorImpl::MarkModifiedElems(int32_t srcCoord[], std::vector<bool> &touchedElemBits, std::vector<size_t> &touchedElemIndex)
{
Real score = 0.f;
for (const KernelSample &ks : _Kernel)
{
int32_t j = 0;
for (size_t d = 0; d < _GenParams.N_dimensions; ++d)
{
j += WrapDimension(srcCoord[d], ks.Distances[d], _GenParams.dimensionSize[d], d) * _DimensionElementCount[d];
}
if (_GenParams.refineSpecificSlice >= 0 && j < 0) continue;
if (touchedElemBits[j] == false)
{
touchedElemIndex.push_back(j);
touchedElemBits[j] = true;
}
}
}
//===========================================================================================================================
void BlueNoiseGeneratorImpl::ComputeBlueNoise(size_t numIter)
{
_IterTotal = 0;
_SwapCount = 0;
_SwapAttempt = 0;
std::random_device rd;
std::mt19937 gen(rd());
std::uniform_real_distribution<> dist(0, 1);
std::uniform_int_distribution<> distInt(1, _MaxSwapedElemCount);
std::uniform_int_distribution<> distSwap(0, (int)(_TotalElements - 1));
// version with global score update
for (size_t iter = 0; iter < numIter; ++iter)
{
// copy
_Pattern[_CurrentArray ^ 1] = _Pattern[_CurrentArray];
uint32_t num_swaps = distInt(gen);
for (size_t i = 0; i < num_swaps; ++i)
{
size_t from = distSwap(gen);
size_t to = distSwap(gen);
while (from == to)
to = distSwap(gen);
for (size_t vecDim = 0; vecDim < _GenParams.N_valuesPerItem; ++vecDim)
{
std::swap(_Pattern[_CurrentArray][from * _GenParams.N_valuesPerItem + vecDim], _Pattern[_CurrentArray][to * _GenParams.N_valuesPerItem + vecDim]);
}
}
const Real score = ComputeGlobalScore(_CurrentArray);
_SwapAttempt += size_t(num_swaps);
if (score < _BestScore)
{
_SwapCount += size_t(num_swaps);
_BestScore = score;
}
else
{
// swap back
_CurrentArray ^= 1;
}
++_IterTotal;
if (_ProgressMonitor)
{
_ProgressMonitor->OnProgress(_IterTotal, _BestScore, _SwapCount, _SwapAttempt);
}
}
}
//===========================================================================================================================
void BlueNoiseGeneratorImpl::ComputeBlueNoiseIncrementalMultiThreaded(size_t numIter)
{
_MTAcquiredRegions.store(0ul);
_IterGuard.store(0);
_IterTotal = 0;
_SwapCount = 0;
_SwapAttempt = 0;
_AcquiredMTRegionSizeDivisorAsRShift = ComputeAcquiredMTRegionSizeDivisorAsRShift();
size_t numThread = std::thread::hardware_concurrency() - 2;
numThread = std::max(size_t(1u), numThread);
std::vector<std::shared_ptr<std::thread> > threads;
auto DoWork = [&](size_t numIter) -> void
{
this->ComputeBlueNoiseIncremental(numIter);
};
volatile bool finished = false;
// monitoring thread
auto MonitorThreadFunc = [&]() -> void
{
if (_ProgressMonitor)
{
while (!finished)
{
LockIterGuard();
_ProgressMonitor->OnProgress(_IterTotal, _BestScore, _SwapCount, _SwapAttempt);
UnlockIterGuard();
std::this_thread::sleep_for(std::chrono::milliseconds(20)); // check once in a while
}
}
};
std::thread monitorThread(MonitorThreadFunc);
for (size_t threadIndex = 0u; threadIndex < numThread; ++threadIndex)
{
threads.push_back(std::shared_ptr<std::thread>(new std::thread(DoWork, (numIter + numThread - 1) / numThread)));
}
for (auto& th : threads)
{
th->join();
}
finished = true;
monitorThread.join();
threads.clear();
}
//===========================================================================================================================
uint32_t BlueNoiseGeneratorImpl::ComputeAcquiredMTRegionSizeDivisorAsRShift() const
{
size_t cubicSize = 0;
uint32_t acquiredRegionSizeShift = 0u; // how many bit to shift a coord to create the region acquisition mask
for (uint32_t dim = 0; dim < _GenParams.N_dimensions; ++dim)
{
cubicSize = std::max(cubicSize, _GenParams.dimensionSize[dim]);
}
cubicSize = NextPow2(uint32_t(cubicSize));
if (_GenParams.N_dimensions == 2)
{
acquiredRegionSizeShift = (cubicSize <= 8) ? 0 : uint32_t(log2(float(cubicSize / 8)));
}
else
{
assert(_GenParams.N_dimensions == 3);
acquiredRegionSizeShift = (cubicSize <= 4) ? 0 : uint32_t(log2(float(cubicSize / 4)));
}
return acquiredRegionSizeShift;
}
//===========================================================================================================================
uint64_t BlueNoiseGeneratorImpl::ComputeMTRegionAcquisitionMask(size_t elemIndex) const
{
int32_t coord[max_N_dimensions];
IndexToCoord(int32_t(elemIndex), coord);
uint64_t claimedRegions(0u);
for (const KernelSample &ks : _Kernel)
{
int32_t bitIndex = 0;
if (_GenParams.N_dimensions == 2)
{
const int32_t wrappedDim[] =
{
WrapDimension(coord[0], ks.Distances[0], _GenParams.dimensionSize[0], 0),
WrapDimension(coord[1], ks.Distances[1], _GenParams.dimensionSize[1], 1)
};
if (wrappedDim[1] >= 0)
{
// claim part of 8 x 8 grid
bitIndex = (wrappedDim[0] >> _AcquiredMTRegionSizeDivisorAsRShift) +
8 * (wrappedDim[1] >> _AcquiredMTRegionSizeDivisorAsRShift);
if (bitIndex >= 0) // may happen with incremental refine of slice
{
assert(bitIndex < 64u);
claimedRegions |= uint64_t(1u) << bitIndex;
}
}
}
else
{
assert(_GenParams.N_dimensions == 3);
// claim part of 4 x 4 x 4 grid
const int32_t wrappedDim[] =
{
WrapDimension(coord[0], ks.Distances[0], _GenParams.dimensionSize[0], 0),
WrapDimension(coord[1], ks.Distances[1], _GenParams.dimensionSize[1], 1),
WrapDimension(coord[2], ks.Distances[2], _GenParams.dimensionSize[2], 2)
};
if (wrappedDim[2] >= 0)
{
bitIndex = (wrappedDim[0] >> _AcquiredMTRegionSizeDivisorAsRShift) +
4 * (wrappedDim[1] >> _AcquiredMTRegionSizeDivisorAsRShift) +
16 * (wrappedDim[2] >> _AcquiredMTRegionSizeDivisorAsRShift);
if (bitIndex >= 0) // may happen with incremental refine of slice
{
assert(bitIndex < 64u);
claimedRegions |= uint64_t(1u) << bitIndex;
}
}
}
}
return claimedRegions;
}
//===========================================================================================================================
void BlueNoiseGeneratorImpl::AcquireMTRegion(uint64_t claimedRegions)
{
while (true)
{
uint64_t previousBits = _MTAcquiredRegions.fetch_or(claimedRegions);
if ((previousBits & claimedRegions) == uint64_t(0)) break; // all acquired at once ?
uint64_t partialAcquiredBits = (~previousBits) & claimedRegions;
// not all regions could be acquired simultaneously... release and spin lock
_MTAcquiredRegions.fetch_and(~partialAcquiredBits);
}
}
//===========================================================================================================================
void BlueNoiseGeneratorImpl::ReleaseMTRegion(uint64_t claimedRegions)
{
uint64_t oldBits = _MTAcquiredRegions.fetch_and(~claimedRegions);
assert((oldBits & claimedRegions) == claimedRegions);
}
//===========================================================================================================================
void BlueNoiseGeneratorImpl::LockIterGuard()
{
while (true)
{
int oldValue = _IterGuard.fetch_add(1);
if (oldValue == 0) break;
_IterGuard.fetch_add(-1);
}
}
//===========================================================================================================================
void BlueNoiseGeneratorImpl::UnlockIterGuard()
{
_IterGuard.fetch_add(-1);
}
//===========================================================================================================================
void BlueNoiseGeneratorImpl::NextIter(Real deltaScore, uint32_t swapAttempt)
{
if (_ActuallyUseMultithreading)
{
LockIterGuard();
}
_BestScore += deltaScore;
++_IterTotal;
_SwapAttempt += swapAttempt;
_SwapCount += deltaScore < 0.f ? swapAttempt : 0;
if (_ActuallyUseMultithreading)
{
UnlockIterGuard();
}
}
//===========================================================================================================================
void BlueNoiseGeneratorImpl::ComputeBlueNoiseIncremental(size_t numIter)
{
_IterTotal = 0;
_SwapCount = 0;
_SwapAttempt = 0;
std::vector<bool> touchedElemBits(_TotalElements, false);
std::vector<size_t> touchedElemIndex;
std::random_device rd;
std::mt19937 gen(rd());
std::uniform_real_distribution<> dist(0, 1);
std::uniform_int_distribution<> distInt(1, _MaxSwapedElemCount);
int32_t totalElement = _TotalElements;
size_t indexOffset = 0;
if (_GenParams.refineSpecificSlice >= 0)
{
totalElement = ComputeElementCount(_GenParams.N_dimensions - 1);
indexOffset = ComputeElementCount(_GenParams.N_dimensions - 1) * _GenParams.refineSpecificSlice;
}
std::uniform_int_distribution<> distSwap(0, (int)(totalElement - 1));
//std::uniform_int_distribution<> distSwapDelta(-7, 7);
//std::uniform_int_distribution<> distSwapDeltaDepth(-2, 2);
for (size_t iter = 0; iter < numIter; ++iter)
{
uint32_t num_swaps = _ActuallyUseMultithreading ? 1u : distInt(gen);
size_t swapedElemIndex[_MaxSwapedElemCount * 2];
size_t from[_MaxSwapedElemCount];
size_t to[_MaxSwapedElemCount];
uint64_t claimedRegions;
while (true)
{
claimedRegions = uint64_t(0u);
// compute the swaps, possibly building a bitfield or the regions that need to be acquired (for multi-threaded version)
for (size_t i = 0; i < num_swaps; ++i)
{
from[i] = distSwap(gen) + indexOffset;
to[i] = distSwap(gen) + indexOffset;
while (from[i] == to[i])
to[i] = distSwap(gen) + indexOffset;
// update claimed region bitfield
if (_ActuallyUseMultithreading)
{
claimedRegions |= ComputeMTRegionAcquisitionMask(from[i]);
claimedRegions |= ComputeMTRegionAcquisitionMask(to[i]);
}
}
if (!_ActuallyUseMultithreading) break;
// because acquisition is very costly, diminish the chance of collision by doing a test
// with no atomic
if ((claimedRegions & _MTAcquiredRegions.load()) == uint64_t(0u)) break;
}
// acquire the region to be swapped
if (_ActuallyUseMultithreading)
{
AcquireMTRegion(claimedRegions);
// no they are acquired, be can do read-write op without colliding with another thread :)
}
// do the actual swaps
for (size_t i = 0; i < num_swaps; ++i)
{
swapedElemIndex[2 * i] = from[i];
swapedElemIndex[2 * i + 1] = to[i];
// mark region where score must be recomputed
int32_t toCoord[max_N_dimensions];
IndexToCoord(int32_t(to[i]), toCoord);
int32_t fromCoord[max_N_dimensions];
IndexToCoord(int32_t(from[i]), fromCoord);
MarkModifiedElems(toCoord, touchedElemBits, touchedElemIndex);
MarkModifiedElems(fromCoord, touchedElemBits, touchedElemIndex);
for (size_t vecDim = 0; vecDim < _GenParams.N_valuesPerItem; ++vecDim)
{
std::swap(_Pattern[0][from[i] * _GenParams.N_valuesPerItem + vecDim], _Pattern[0][to[i] * _GenParams.N_valuesPerItem + vecDim]);
}
}
// compute score delta
// we do it in local neighbourhood only, because elements further than 3 / 4 cell from here do not contribute significantly
Real scoreToRemove = 0.f;
for (size_t elemIndex : touchedElemIndex)
{
int32_t elemCoord[max_N_dimensions];
IndexToCoord(int32_t(elemIndex), elemCoord);
scoreToRemove += ComputeLocalScore(elemCoord, 1);
}