-
Notifications
You must be signed in to change notification settings - Fork 2
/
FFT3DFilter.cpp
3833 lines (3541 loc) · 153 KB
/
FFT3DFilter.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
/*
FFT3DFilter plugin for Avisynth 2.6 and Avisynth + - 3D Frequency Domain filter
Copyright(C)2004-2006 A.G.Balakhnin aka Fizick, bag@hotmail.ru, http://avisynth.org.ru
This program is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License version 2 as published by
the Free Software Foundation.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License
along with this program; if not, write to the Free Software
Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
Plugin uses external FFTW library version 3 (http://www.fftw.org)
as Windows binary DLL (compiled with gcc under MinGW by Alessio Massaro),
which support for threads and have AMD K7 (3dNow!) support in addition to SSE/SSE2.
It may be downloaded from ftp://ftp.fftw.org/pub/fftw/fftw3win32mingw.zip
You must put FFTW3.DLL file from this package to some directory in path
(for example, C:\WINNT\System32).
The algorithm is based on the 3D IIR/3D Frequency Domain Filter from:
MOTION PICTURE RESTORATION. by Anil Christopher Kokaram. Ph.D. Thesis. May 1993.
http://www.mee.tcd.ie/~ack/papers/a4ackphd.ps.gz
Version 0.1, 23 November 2004 - initial
Version 0.2, 3 December 2004 - add beta parameter of noise margin
Version 0.3, 21 December 2004 - add bt parameter of temporal size
Version 0.4, 16 January 2005 - algorithm optimized for speed for bt=2 (now default),
mode bt=3 is temporary disabled, changed default bw=bh=32, filtered region now centered.
Version 0.5, 28 January 2005 - added YUY2 support
Version 0.6, 29 January 2005 - added Kalman filter mode for bt=0, ratio parameter
Version 0.7, 30 January 2005 - re-enabled Wiener filter mode with 3 frames (bt=3)
Version 0.8, 05 February2005 - added option to sharpen, and bt=-1
Version 0.8.1, 6 February2005 - skip sharpening of the lowest frequencies to prevent parasitic lines near border
Version 0.8.2, February 15, 2005 - added internal buffer to process whole frame (borders included) for any bw, bh (a little slower)
Version 0.8.3, March 16, 2005 - fixed sharpen mode (bt=-1) for YUY2
Version 0.8.4, April 3, 2005 - delayed FFTW3.DLL loading
Version 0.9 - April 3,2005 - variable overlapping size
Version 0.9.1 - April 7,2005 - some assembler 3DNow! optimization for mode bt=3
Version 0.9.2 - April 10,2005 - some assembler 3DNow! optimization for mode bt=2,
option measure=true is now default as more fast
Version 0.9.3 - April 24,2005 - bug fixed for bt=2 with 3DNow; bt=3 now default;
modifyed sharpen to horizontal only (still experimental)
Version 1.0 - June 22, 2005 - improved edges processing (by padding);
added svr parameter to control vertical sharpening
Version 1.0.1 - July 05, 2005 - fixed bug for YUY2 chroma planes
Version 1.1 - July 8,2005 - improved sharpen mode to prevent grid artifactes and to limit sharpening,
added parameters smin, smax; renamed parameter ratio to kratio.
Version 1.2 - July 12, 2005 - changed parameters defaults (bw=bh=48, ow=bw/3, oh=bh/3) to prevent grid artifactes
Version 1.3 - July 20, 2005 - added interlaced mode
Version 1.3.1 - July 21, 2005 - fixed bug for YUY2 interlaced
Version 1.4 - July 23, 2005 - corrected neutral level for chroma processing, added wintype to decrease grid artefactes
Version 1.5 - July 26, 2005 - added noise pattern method and its parameters pframe, px, py, pshow, pcutoff, pfactor
Version 1.5.1 - July 29, 2005 - fixed bug with pshow
Version 1.5.2 - July 31, 2005 - fixed bug with Kalman mode (bt=0) for Athlon (introduced in v1.5)
Version 1.6 - August 01, 2005 - added mode bt=4; optimized SSE version for bt=2,3
Version 1.7 - August 29, 2005 - added SSE version for for sharpen and pattern modes bt=2,3 ; restuctured code, GPL v2
Version 1.8 - September 6, 2005 - improved internal fft cache; added degrid=0; changed wintype=0
Version 1.8.1 - October 26, 2005 - fixed bug with sharpen>0 AND degrid>0 for bt not equal 1.
Version 1.8.2 - November 04, 2005 - really set default degrid=1.0 (was = 0)
Version 1.8.3 - November 28, 2005 - fixed bug with first frame for Kalman YV12 (thanks to Tsp)
Version 1.8.4 - November 29, 2005 - added multiplane modes plane=3,4
Version 1.8.5 - 4 December 2005 - fixed bug with memory leakage (thanks to tsp).
Version 1.9 - April 25, 2006 - added dehalo options; corrected sharpen mode a little;
re-enabled 3DNow and SSE optimization for degrid=0; added SSE optimization for bt=3,-1 with degrid>0 (faster by 15%)
Version 1.9.1 - May 10, 2006 - added SSE optimization for bt=4 with degrid>0 (faster by 30%)
Version 1.9.2 - September 6, 2006 - added new mode bt=5
Version 2.0.0 - November 6, 2006 - added motion compensation mc parameter, window reorganized, multi-cpu
Version 2.1.0 - January 17, 2007 - removed motion compensation mc parameter
Version 2.1.1 - February 19, 2007 - fixed bug with bw not mod 4 (restored v1.9.2 window method)
Version 2.2 - February 25, 2015 - martin53: made AviSynth 2.6 ready, FFT3dFilter2_VersionNumber function, error msgs
Version 2.3 February 21, 2017 pinterf
- apply current avs+ headers
- 10-16 bits and 32 bit float colorspace support in AVS+
- Planar RGB support
- look for libfftw3f-3.dll first, then fftw3.dll
- inline asm ignored on x64 builds
- pre-check: if plane to process for greyscale is U and/or V return original clip
- auto register MT mode for avs+: MT_SERIALIZED
Version 2.4 June 01, 2017 pinterf
- intrinsics bt=0: void ApplyKalman_SSE2_simd(fftwf_complex *outcur, fftwf_complex *outLast,
- intrinsics bt=2, degrid=0, pfactor=0: void ApplyWiener3D2_SSE_simd(fftwf_complex *outcur, fftwf_complex *outprev,// bt=2, degrid=0, pfactor=0
- intrinsics bt=3 sharpen=0/1 dehalo=0/1: void do_Sharpen_degrid_SSE_simd(fftwf_complex *outcur, int outwidth, int outpitch, int bh,
- intrinsics bt=3: void ApplyWiener3D3_degrid_SSE_simd(fftwf_complex *outcur, fftwf_complex *outprev, ...
- Adaptive MT: MT_SERIALIZED for bt==0 (temporal), MT_MULTI_INSTANCE for others
- Copy Alpha plane if exists
- reentrancy checks against bad multithreading usage
Version 2.5 July 02, 2018 pinterf
- 32bit Float YUV: Chroma center to 0.0 instead of 0.5, to match Avisynth+ r2728-
Version 2.6 January 31, 2019 pinterf
- use 0.5 rounding before converting back to the integer pixel domain.
Version 2.7 November 30, 2020 pinterf
- make fft3w calls thread safe
- preserve frame properties for avs+
Version 2.7 December 1, 2020 pinterf
- fix chroma for 32 bit float formats
*/
#define VERSION_NUMBER 2.10
#include <avisynth.h>
#include <stdio.h>
#include <stdint.h>
#ifdef _WIN32
#define NOMINMAX
#include <Windows.h>
#endif
#include "math.h"
#include "fftwlite.h"
#include "info.h"
#include <emmintrin.h>
#include <mmintrin.h>
#include <algorithm>
#include <atomic>
#include <mutex>
// FFTW is not thread-safe, need to guard around its functions (except fftw_execute).
// http://www.fftw.org/fftw3_doc/Thread-safety.html#Thread-safety
static std::mutex fftw_mutex; // defined as static
// declarations of filtering functions:
#ifndef X86_64
// 3DNow
void ApplyWiener3D4_3DNow(fftwf_complex *outcur, fftwf_complex *outprev2, fftwf_complex *outprev, fftwf_complex *outnext, int outwidth, int outpitch, int bh, int howmanyblocks, float sigmaSquaredNoiseNormed, float beta);
#endif
void ApplyKalman_SSE2_simd(fftwf_complex *outcur, fftwf_complex *outLast, fftwf_complex *covar, fftwf_complex *covarProcess, int outwidth, int outpitch, int bh, int howmanyblocks, float covarNoiseNormed, float kratio2);
// SSE
void ApplyWiener3D2_SSE_simd(fftwf_complex *outcur, fftwf_complex *outprev, int outwidth, int outpitch, int bh, int howmanyblocks, float sigmaSquaredNoiseNormed, float beta);
void ApplyWiener3D4_SSE_simd(fftwf_complex* outcur, fftwf_complex* outprev2, fftwf_complex* outprev, fftwf_complex* outnext, int outwidth, int outpitch, int bh, int howmanyblocks, float sigmaSquaredNoiseNormed, float beta);
void ApplyPattern3D2_SSE(fftwf_complex *outcur, fftwf_complex *outprev, int outwidth, int outpitch, int bh, int howmanyblocks, float * pattern3d, float beta);
void ApplyWiener3D3_SSE(fftwf_complex *outcur, fftwf_complex *outprev, fftwf_complex *outnext, int outwidth, int outpitch, int bh, int howmanyblocks, float sigmaSquaredNoiseNormed, float beta);
void ApplyPattern3D3_SSE(fftwf_complex *outcur, fftwf_complex *outprev, fftwf_complex *outnext, int outwidth, int outpitch, int bh, int howmanyblocks, float *pattern3d, float beta);
void Sharpen_SSE(fftwf_complex *outcur, int outwidth, int outpitch, int bh, int howmanyblocks, float sharpen, float sigmaSquaredSharpenMin, float sigmaSquaredSharpenMax, float *wsharpen);
// C
void ApplyWiener2D_C(fftwf_complex *out, int outwidth, int outpitch, int bh, int howmanyblocks, float sigmaSquaredNoiseNormed, float beta, float sharpen, float sigmaSquaredSharpenMin, float sigmaSquaredSharpenMax, float *wsharpen, float dehalo, float *wdehalo, float ht2n);
void ApplyPattern2D_C(fftwf_complex *outcur, int outwidth, int outpitch, int bh, int howmanyblocks, float pfactor, float *pattern2d0, float beta);
void ApplyWiener3D2_C(fftwf_complex *outcur, fftwf_complex *outprev, int outwidth, int outpitch, int bh, int howmanyblocks, float sigmaSquaredNoiseNormed, float beta);
void ApplyPattern3D2_C(fftwf_complex *outcur, fftwf_complex *outprev, int outwidth, int outpitch, int bh, int howmanyblocks, float *pattern3d, float beta);
void ApplyWiener3D3_C(fftwf_complex *out, fftwf_complex *outprev, fftwf_complex *outnext, int outwidth, int outpitch, int bh, int howmanyblocks, float sigmaSquaredNoiseNormed, float beta);
void ApplyPattern3D3_C(fftwf_complex *out, fftwf_complex *outprev, fftwf_complex *outnext, int outwidth, int outpitch, int bh, int howmanyblocks, float *pattern3d, float beta);
void ApplyWiener3D4_C(fftwf_complex *out, fftwf_complex *outprev2, fftwf_complex *outprev, fftwf_complex *outnext, int outwidth, int outpitch, int bh, int howmanyblocks, float sigmaSquaredNoiseNormed, float beta);
void ApplyPattern3D4_C(fftwf_complex *out, fftwf_complex *outprev2, fftwf_complex *outprev, fftwf_complex *outnext, int outwidth, int outpitch, int bh, int howmanyblocks, float* pattern3d, float beta);
void ApplyWiener3D5_C(fftwf_complex *out, fftwf_complex *outprev2, fftwf_complex *outprev, fftwf_complex *outnext, fftwf_complex *outnext2, int outwidth, int outpitch, int bh, int howmanyblocks, float sigmaSquaredNoiseNormed, float beta);
void ApplyPattern3D5_C(fftwf_complex *out, fftwf_complex *outprev2, fftwf_complex *outprev, fftwf_complex *outnext, fftwf_complex *outnext2, int outwidth, int outpitch, int bh, int howmanyblocks, float* pattern3d, float beta);
void ApplyKalmanPattern_C(fftwf_complex *outcur, fftwf_complex *outLast, fftwf_complex *covar, fftwf_complex *covarProcess, int outwidth, int outpitch, int bh, int howmanyblocks, float *covarNoiseNormed, float kratio2);
void ApplyKalman_C(fftwf_complex *outcur, fftwf_complex *outLast, fftwf_complex *covar, fftwf_complex *covarProcess, int outwidth, int outpitch, int bh, int howmanyblocks, float covarNoiseNormed, float kratio2);
void Sharpen_C(fftwf_complex *outcur, int outwidth, int outpitch, int bh, int howmanyblocks, float sharpen, float sigmaSquaredSharpenMin, float sigmaSquaredSharpenMax, float *wsharpen, float dehalo, float *wdehalo, float ht2n);
// degrid_C
void ApplyWiener2D_degrid_C(fftwf_complex *out, int outwidth, int outpitch, int bh, int howmanyblocks, float sigmaSquaredNoiseNormed, float beta, float sharpen, float sigmaSquaredSharpenMin, float sigmaSquaredSharpenMax, float *wsharpen, float degrid, fftwf_complex *gridsample, float dehalo, float *wdehalo, float ht2n);
void ApplyWiener3D2_degrid_C(fftwf_complex *outcur, fftwf_complex *outprev, int outwidth, int outpitch, int bh, int howmanyblocks, float sigmaSquaredNoiseNormed, float beta, float degrid, fftwf_complex *gridsample);
void ApplyWiener3D3_degrid_C(fftwf_complex *outcur, fftwf_complex *outprev, fftwf_complex *outnext, int outwidth, int outpitch, int bh, int howmanyblocks, float sigmaSquaredNoiseNormed, float beta, float degrid, fftwf_complex *gridsample);
void ApplyWiener3D4_degrid_C(fftwf_complex *out, fftwf_complex *outprev2, fftwf_complex *outprev, fftwf_complex *outnext, int outwidth, int outpitch, int bh, int howmanyblocks, float sigmaSquaredNoiseNormed, float beta, float degrid, fftwf_complex *gridsample);
void ApplyWiener3D5_degrid_C(fftwf_complex *out, fftwf_complex *outprev2, fftwf_complex *outprev, fftwf_complex *outnext, fftwf_complex *outnext2, int outwidth, int outpitch, int bh, int howmanyblocks, float sigmaSquaredNoiseNormed, float beta, float degrid, fftwf_complex *gridsample);
void Sharpen_degrid_C(fftwf_complex *outcur, int outwidth, int outpitch, int bh, int howmanyblocks, float sharpen, float sigmaSquaredSharpenMin, float sigmaSquaredSharpenMax, float *wsharpen, float degrid, fftwf_complex *gridsample, float dehalo, float *wdehalo, float ht2n);
void ApplyPattern2D_degrid_C(fftwf_complex *outcur, int outwidth, int outpitch, int bh, int howmanyblocks, float pfactor, float *pattern2d0, float beta, float degrid, fftwf_complex *gridsample);
void ApplyPattern3D2_degrid_C(fftwf_complex *outcur, fftwf_complex *outprev, int outwidth, int outpitch, int bh, int howmanyblocks, float *pattern3d, float beta, float degrid, fftwf_complex *gridsample);
void ApplyPattern3D3_degrid_C(fftwf_complex *out, fftwf_complex *outprev, fftwf_complex *outnext, int outwidth, int outpitch, int bh, int howmanyblocks, float *pattern3d, float beta, float degrid, fftwf_complex *gridsample);
void ApplyPattern3D4_degrid_C(fftwf_complex *out, fftwf_complex *outprev2, fftwf_complex *outprev, fftwf_complex *outnext, int outwidth, int outpitch, int bh, int howmanyblocks, float* pattern3d, float beta, float degrid, fftwf_complex *gridsample);
void ApplyPattern3D5_degrid_C(fftwf_complex *out, fftwf_complex *outprev2, fftwf_complex *outprev, fftwf_complex *outnext, fftwf_complex *outnext2, int outwidth, int outpitch, int bh, int howmanyblocks, float* pattern3d, float beta, float degrid, fftwf_complex *gridsample);
// degrid_SSE
void ApplyWiener3D3_degrid_SSE(fftwf_complex *outcur, fftwf_complex *outprev, fftwf_complex *outnext, int outwidth, int outpitch, int bh, int howmanyblocks, float sigmaSquaredNoiseNormed, float beta, float degrid, fftwf_complex *gridsample);
void ApplyWiener3D3_degrid_SSE_simd(fftwf_complex *outcur, fftwf_complex *outprev, fftwf_complex *outnext, int outwidth, int outpitch, int bh, int howmanyblocks, float sigmaSquaredNoiseNormed, float beta, float degrid, fftwf_complex *gridsample);
void ApplyPattern3D3_degrid_SSE(fftwf_complex *out, fftwf_complex *outprev, fftwf_complex *outnext, int outwidth, int outpitch, int bh, int howmanyblocks, float *pattern3d, float beta, float degrid, fftwf_complex *gridsample);
void Sharpen_degrid_SSE_simd(fftwf_complex *outcur, int outwidth, int outpitch, int bh, int howmanyblocks, float sharpen, float sigmaSquaredSharpenMin, float sigmaSquaredSharpenMax, float *wsharpen, float degrid, fftwf_complex *gridsample, float dehalo, float *wdehalo, float ht2n);
void ApplyWiener3D4_degrid_SSE(fftwf_complex *outcur, fftwf_complex *outprev2, fftwf_complex *outprev, fftwf_complex *outnext, int outwidth, int outpitch, int bh, int howmanyblocks, float sigmaSquaredNoiseNormed, float beta, float degrid, fftwf_complex *gridsample);
void ApplyPattern3D4_degrid_SSE(fftwf_complex *outcur, fftwf_complex *outprev2, fftwf_complex *outprev, fftwf_complex *outnext, int outwidth, int outpitch, int bh, int howmanyblocks, float *pattern3d, float beta, float degrid, fftwf_complex *gridsample);
//-------------------------------------------------------------------------------------------
void ApplyWiener2D(fftwf_complex *out, int outwidth, int outpitch, int bh, int howmanyblocks, float sigmaSquaredNoiseNormed,
float beta, float sharpen, float sigmaSquaredSharpenMin, float sigmaSquaredSharpenMax, float *wsharpen, float dehalo, float *wdehalo, float ht2n, int CPUFlags)
{
ApplyWiener2D_C(out, outwidth, outpitch, bh, howmanyblocks, sigmaSquaredNoiseNormed, beta, sharpen, sigmaSquaredSharpenMin, sigmaSquaredSharpenMax, wsharpen, dehalo, wdehalo, ht2n);
}
//-------------------------------------------------------------------------------------------
void ApplyWiener3D2(fftwf_complex *outcur, fftwf_complex *outprev, int outwidth, int outpitch, int bh, int howmanyblocks, float sigmaSquaredNoiseNormed, float beta, int CPUFlags)
{
if (CPUFlags & CPUF_SSE2) // 170302 simd, SSE2
ApplyWiener3D2_SSE_simd(outcur, outprev, outwidth, outpitch, bh, howmanyblocks, sigmaSquaredNoiseNormed, beta);
else
ApplyWiener3D2_C(outcur, outprev, outwidth, outpitch, bh, howmanyblocks, sigmaSquaredNoiseNormed, beta);
}
//-------------------------------------------------------------------------------------------
void ApplyPattern3D2(fftwf_complex *outcur, fftwf_complex *outprev, int outwidth, int outpitch, int bh, int howmanyblocks, float *pattern3d, float beta, int CPUFlags)
{
#ifndef X86_64
if (CPUFlags & CPUF_SSE)
ApplyPattern3D2_SSE(outcur, outprev, outwidth, outpitch, bh, howmanyblocks, pattern3d, beta);
else
#endif
ApplyPattern3D2_C(outcur, outprev, outwidth, outpitch, bh, howmanyblocks, pattern3d, beta);
}
//-------------------------------------------------------------------------------------------
void ApplyWiener3D3(fftwf_complex *out, fftwf_complex *outprev, fftwf_complex *outnext, int outwidth, int outpitch, int bh, int howmanyblocks, float sigmaSquaredNoiseNormed, float beta, int CPUFlags)
{
#ifndef X86_64
if (CPUFlags & CPUF_SSE)
ApplyWiener3D3_SSE(out, outprev, outnext, outwidth, outpitch, bh, howmanyblocks, sigmaSquaredNoiseNormed, beta);
else
#endif
ApplyWiener3D3_C(out, outprev, outnext, outwidth, outpitch, bh, howmanyblocks, sigmaSquaredNoiseNormed, beta);
}
//-------------------------------------------------------------------------------------------
void ApplyWiener3D3_degrid(fftwf_complex *out, fftwf_complex *outprev, fftwf_complex *outnext, int outwidth, int outpitch, int bh, int howmanyblocks, float sigmaSquaredNoiseNormed, float beta, float degrid, fftwf_complex *gridsample, int CPUFlags)
{
if (CPUFlags & CPUF_SSE)
ApplyWiener3D3_degrid_SSE_simd(out, outprev, outnext, outwidth, outpitch, bh, howmanyblocks, sigmaSquaredNoiseNormed, beta, degrid, gridsample);
else
ApplyWiener3D3_degrid_C(out, outprev, outnext, outwidth, outpitch, bh, howmanyblocks, sigmaSquaredNoiseNormed, beta, degrid, gridsample);
}
//-------------------------------------------------------------------------------------------
void ApplyWiener3D4_degrid(fftwf_complex *out, fftwf_complex *outprev2, fftwf_complex *outprev, fftwf_complex *outnext, int outwidth, int outpitch, int bh, int howmanyblocks, float sigmaSquaredNoiseNormed, float beta, float degrid, fftwf_complex *gridsample, int CPUFlags)
{
#ifndef X86_64
if (CPUFlags & CPUF_SSE)
ApplyWiener3D4_degrid_SSE(out, outprev2, outprev, outnext, outwidth, outpitch, bh, howmanyblocks, sigmaSquaredNoiseNormed, beta, degrid, gridsample);
else
#endif
ApplyWiener3D4_degrid_C(out, outprev2, outprev, outnext, outwidth, outpitch, bh, howmanyblocks, sigmaSquaredNoiseNormed, beta, degrid, gridsample);
}
//-------------------------------------------------------------------------------------------
void ApplyPattern2D(fftwf_complex *outcur, int outwidth, int outpitch, int bh, int howmanyblocks, float pfactor, float *pattern2d0, float beta, int CPUFlags)
{
ApplyPattern2D_C(outcur, outwidth, outpitch, bh, howmanyblocks, pfactor, pattern2d0, beta);
}
//-------------------------------------------------------------------------------------------
void ApplyPattern3D3(fftwf_complex *out, fftwf_complex *outprev, fftwf_complex *outnext, int outwidth, int outpitch, int bh, int howmanyblocks, float *pattern3d, float beta, int CPUFlags)
{
#ifndef X86_64
if (CPUFlags & CPUF_SSE)
ApplyPattern3D3_SSE(out, outprev, outnext, outwidth, outpitch, bh, howmanyblocks, pattern3d, beta);
else
#endif
ApplyPattern3D3_C(out, outprev, outnext, outwidth, outpitch, bh, howmanyblocks, pattern3d, beta);
}
//-------------------------------------------------------------------------------------------
void ApplyPattern3D3_degrid(fftwf_complex *out, fftwf_complex *outprev, fftwf_complex *outnext, int outwidth, int outpitch, int bh, int howmanyblocks, float *pattern3d, float beta, float degrid, fftwf_complex *gridsample, int CPUFlags)
{
#ifndef X86_64
if (CPUFlags & CPUF_SSE)
ApplyPattern3D3_degrid_SSE(out, outprev, outnext, outwidth, outpitch, bh, howmanyblocks, pattern3d, beta, degrid, gridsample);
else
#endif
ApplyPattern3D3_degrid_C(out, outprev, outnext, outwidth, outpitch, bh, howmanyblocks, pattern3d, beta, degrid, gridsample);
}
//-------------------------------------------------------------------------------------------
void ApplyPattern3D4_degrid(fftwf_complex *out, fftwf_complex *outprev2, fftwf_complex *outprev, fftwf_complex *outnext, int outwidth, int outpitch, int bh, int howmanyblocks, float *pattern3d, float beta, float degrid, fftwf_complex *gridsample, int CPUFlags)
{
#ifndef X86_64
if (CPUFlags & CPUF_SSE)
ApplyPattern3D4_degrid_SSE(out, outprev2, outprev, outnext, outwidth, outpitch, bh, howmanyblocks, pattern3d, beta, degrid, gridsample);
else
#endif
ApplyPattern3D4_degrid_C(out, outprev2, outprev, outnext, outwidth, outpitch, bh, howmanyblocks, pattern3d, beta, degrid, gridsample);
}
//-------------------------------------------------------------------------------------------
void ApplyWiener3D4(fftwf_complex *out, fftwf_complex *outprev2, fftwf_complex *outprev, fftwf_complex *outnext, int outwidth, int outpitch, int bh, int howmanyblocks, float sigmaSquaredNoiseNormed, float beta, int CPUFlags)
{
if (CPUFlags & CPUF_SSE2)
ApplyWiener3D4_SSE_simd(out, outprev2, outprev, outnext, outwidth, outpitch, bh, howmanyblocks, sigmaSquaredNoiseNormed, beta);
else
ApplyWiener3D4_C(out, outprev2, outprev, outnext, outwidth, outpitch, bh, howmanyblocks, sigmaSquaredNoiseNormed, beta);
}
//-------------------------------------------------------------------------------------------
void ApplyPattern3D4(fftwf_complex *out, fftwf_complex *outprev2, fftwf_complex *outprev, fftwf_complex *outnext, int outwidth, int outpitch, int bh, int howmanyblocks, float* pattern3d, float beta, int CPUFlags)
{
ApplyPattern3D4_C(out, outprev2, outprev, outnext, outwidth, outpitch, bh, howmanyblocks, pattern3d, beta);
}
//-------------------------------------------------------------------------------------------
void ApplyKalmanPattern(fftwf_complex *outcur, fftwf_complex *outLast, fftwf_complex *covar, fftwf_complex *covarProcess, int outwidth, int outpitch, int bh, int howmanyblocks, float *covarNoiseNormed, float kratio2, int CPUFlags)
{
ApplyKalmanPattern_C(outcur, outLast, covar, covarProcess, outwidth, outpitch, bh, howmanyblocks, covarNoiseNormed, kratio2);
}
//-------------------------------------------------------------------------------------------
void ApplyKalman(fftwf_complex *outcur, fftwf_complex *outLast, fftwf_complex *covar, fftwf_complex *covarProcess, int outwidth, int outpitch, int bh, int howmanyblocks, float covarNoiseNormed, float kratio2, int CPUFlags)
{
// bt=0
// moved to SSE2 simd (though only 8 bytes internal working mode)
if (CPUFlags & CPUF_SSE2)
ApplyKalman_SSE2_simd(outcur, outLast, covar, covarProcess, outwidth, outpitch, bh, howmanyblocks, covarNoiseNormed, kratio2);
else
ApplyKalman_C(outcur, outLast, covar, covarProcess, outwidth, outpitch, bh, howmanyblocks, covarNoiseNormed, kratio2);
}
//-------------------------------------------------------------------------------------------
void Sharpen(fftwf_complex *outcur, int outwidth, int outpitch, int bh, int howmanyblocks, float sharpen, float sigmaSquaredSharpenMin, float sigmaSquaredSharpenMax, float *wsharpen, float dehalo, float *wdehalo, float ht2n, int CPUFlags)
{
#ifndef X86_64
if ((CPUFlags & CPUF_SSE) && dehalo == 0)
Sharpen_SSE(outcur, outwidth, outpitch, bh, howmanyblocks, sharpen, sigmaSquaredSharpenMin, sigmaSquaredSharpenMax, wsharpen);
else
#endif
Sharpen_C(outcur, outwidth, outpitch, bh, howmanyblocks, sharpen, sigmaSquaredSharpenMin, sigmaSquaredSharpenMax, wsharpen, dehalo, wdehalo, ht2n);
}
//-------------------------------------------------------------------------------------------
void Sharpen_degrid(fftwf_complex *outcur, int outwidth, int outpitch, int bh, int howmanyblocks, float sharpen, float sigmaSquaredSharpenMin, float sigmaSquaredSharpenMax, float *wsharpen, float degrid, fftwf_complex *gridsample, float dehalo, float *wdehalo, float ht2n, int CPUFlags)
{
if ((CPUFlags & CPUF_SSE2))
Sharpen_degrid_SSE_simd(outcur, outwidth, outpitch, bh, howmanyblocks, sharpen, sigmaSquaredSharpenMin, sigmaSquaredSharpenMax, wsharpen, degrid, gridsample, dehalo, wdehalo, ht2n);
else
Sharpen_degrid_C(outcur, outwidth, outpitch, bh, howmanyblocks, sharpen, sigmaSquaredSharpenMin, sigmaSquaredSharpenMax, wsharpen, degrid, gridsample, dehalo, wdehalo, ht2n);
}
//-------------------------------------------------------------------------------------------
//-------------------------------------------------------------------
void fill_complex(fftwf_complex *plane, int outsize, float realvalue, float imgvalue)
{
// it is not fast, but called only in constructor
int w;
for (w = 0; w < outsize; w++) {
plane[w][0] = realvalue;
plane[w][1] = imgvalue;
}
}
//-------------------------------------------------------------------
void SigmasToPattern(float sigma, float sigma2, float sigma3, float sigma4, int bh, int outwidth, int outpitch, float norm, float *pattern2d)
{
// it is not fast, but called only in constructor
float sigmacur;
float ft2 = sqrt(0.5f) / 2; // frequency for sigma2
float ft3 = sqrt(0.5f) / 4; // frequency for sigma3
for (int h = 0; h < bh; h++)
{
for (int w = 0; w < outwidth; w++)
{
float fy = (bh - 2.0f*abs(h - bh / 2)) / bh; // normalized to 1
float fx = (w*1.0f) / outwidth; // normalized to 1
float f = sqrt((fx*fx + fy*fy)*0.5f); // normalized to 1
if (f < ft3)
{ // low frequencies
sigmacur = sigma4 + (sigma3 - sigma4)*f / ft3;
}
else if (f < ft2)
{ // middle frequencies
sigmacur = sigma3 + (sigma2 - sigma3)*(f - ft3) / (ft2 - ft3);
}
else
{// high frequencies
sigmacur = sigma + (sigma2 - sigma)*(1 - f) / (1 - ft2);
}
pattern2d[w] = sigmacur*sigmacur / norm;
}
pattern2d += outpitch;
}
}
//-------------------------------------------------------------------------------------------
class FFT3DFilter : public GenericVideoFilter {
// FFT3DFilter defines the name of your filter class.
// This name is only used internally, and does not affect the name of your filter or similar.
// This filter extends GenericVideoFilter, which incorporates basic functionality.
// All functions present in the filter must also be present here.
// parameters
float sigma; // noise level (std deviation) for high frequncies ***
float beta; // relative noise margin for Wiener filter
int plane; // color plane
int bw;// block width
int bh;// block height
int bt;// block size along time (mumber of frames), =0 for Kalman, >0 for Wiener
int ow; // overlap width - v.0.9
int oh; // overlap height - v.0.9
float kratio; // threshold to sigma ratio for Kalman filter
float sharpen; // sharpen factor (0 to 1 and above)
float scutoff; // sharpen cufoff frequency (relative to max) - v1.7
float svr; // sharpen vertical ratio (0 to 1 and above) - v.1.0
float smin; // minimum limit for sharpen (prevent noise amplifying) - v.1.1 ***
float smax; // maximum limit for sharpen (prevent oversharping) - v.1.1 ***
bool measure; // fft optimal method
bool interlaced;
int wintype; // window type
int pframe; // noise pattern frame number
int px; // noise pattern window x-position
int py; // noise pattern window y-position
bool pshow; // show noise pattern
float pcutoff; // pattern cutoff frequency (relative to max)
float pfactor; // noise pattern denoise strength
float sigma2; // noise level for middle frequencies ***
float sigma3; // noise level for low frequencies ***
float sigma4; // noise level for lowest (zero) frequencies ***
float degrid; // decrease grid
float dehalo; // remove halo strength - v.1.9
float hr; // halo radius - v1.9
float ht; // halo threshold - v1.9
// bool mc; // motion compensation - v2.0
int ncpu; // number of threads - v2.0
int multiplane; // multiplane value
// additional parameterss
float *in;
fftwf_complex *out, *outprev, *outnext, *outtemp, *outprev2, *outnext2;
fftwf_complex *outrez, *gridsample; //v1.8
fftwf_plan plan, planinv, plan1;
int nox, noy;
int outwidth;
int outpitch; //v.1.7
int outsize;
int howmanyblocks;
int ndim[2];
int inembed[2];
int onembed[2];
float *wanxl; // analysis
float *wanxr;
float *wanyl;
float *wanyr;
float *wsynxl; // synthesis
float *wsynxr;
float *wsynyl;
float *wsynyr;
float *wsharpen;
float *wdehalo;
int nlast;// frame number at last step, PF: multithread warning, used for cacheing when sequential access detected
int btcurlast; //v1.7 to prevent multiple Pattern2Dto3D for the same btcurrent. btcurrent can change and may differ from bt for e.g. first/last frame
fftwf_complex *outLast, *covar, *covarProcess;
float sigmaSquaredNoiseNormed;
float sigmaSquaredNoiseNormed2D;
float sigmaNoiseNormed2D;
float sigmaMotionNormed;
float sigmaSquaredSharpenMinNormed;
float sigmaSquaredSharpenMaxNormed;
float ht2n; // halo threshold squared normed
float norm; // normalization factor
BYTE *coverbuf; // block buffer covering the frame without remainders (with sufficient width and heigth)
int coverwidth;
int coverheight;
int coverpitch;
int mirw; // mirror width for padding
int mirh; // mirror height for padding
float *mean;
float *pwin;
float *pattern2d;
float *pattern3d;
bool isPatternSet;
float psigma;
char *messagebuf;
FFTFunctionPointers fftfp;
/*
// added in v.0.9 for delayed FFTW3.DLL loading
HINSTANCE hinstLib;
fftwf_malloc_proc fftwf_malloc;
fftwf_free_proc fftwf_free;
fftwf_plan_many_dft_r2c_proc fftwf_plan_many_dft_r2c;
fftwf_plan_many_dft_c2r_proc fftwf_plan_many_dft_c2r;
fftwf_destroy_plan_proc fftwf_destroy_plan;
fftwf_execute_dft_r2c_proc fftwf_execute_dft_r2c;
fftwf_execute_dft_c2r_proc fftwf_execute_dft_c2r;
fftwf_init_threads_proc fftwf_init_threads;
fftwf_plan_with_nthreads_proc fftwf_plan_with_nthreads;
*/
int CPUFlags;
// avs+
int pixelsize;
int bits_per_pixel;
int planes[4]; // prefilled PLANAR_Y/PLANAR_U/PLANAR_V/PLANAR_A or PLANAR_G/PLANAR_B/PLANAR_R
fftwf_complex ** cachefft; //v1.8
int * cachewhat;//v1.8
int cachesize;//v1.8
int _instance_id; // debug unique id
std::atomic<bool> reentrancy_check;
bool has_at_least_v8; // frame property support
// float *fullwinan; // disabled in v2.2.1, return to v1.9.2 method
// float *fullwinsyn;
//void FFT3DFilter::InitOverlapPlane(float * inp, const BYTE *srcp, int src_pitch, int planeBase);
template<typename pixel_t, int bits_per_pixel, bool chroma>
void do_InitOverlapPlane(float * inp, const BYTE *srcp, int src_pitch);
void InitOverlapPlane(float * inp, const BYTE *srcp, int src_pitch, bool chroma);
template<typename pixel_t, int bits_per_pixel, bool chroma>
void do_DecodeOverlapPlane(float *in, float norm, BYTE *dstp, int dst_pitch);
void DecodeOverlapPlane(float *in, float norm, BYTE *dstp, int dst_pitch, bool chroma);
// void FFT3DFilter::InitFullWin(float * inp0, float *wanxl, float *wanxr, float *wanyl, float *wanyr);
// void FFT3DFilter::InitOverlapPlaneWin(float * inp0, const BYTE *srcp0, int src_pitch, int planeBase, float * fullwin);
public:
// This defines that these functions are present in your class.
// These functions must be that same as those actually implemented.
// Since the functions are "public" they are accessible to other classes.
// Otherwise they can only be called from functions within the class itself.
FFT3DFilter(PClip _child, float _sigma, float _beta, int _plane, int _bw, int _bh, int _bt, int _ow, int _oh,
float _kratio, float _sharpen, float _scutoff, float _svr, float _smin, float _smax,
bool _measure, bool _interlaced, int _wintype,
int _pframe, int _px, int _py, bool _pshow, float _pcutoff, float _pfactor,
float _sigma2, float _sigma3, float _sigma4, float _degrid,
float _dehalo, float _hr, float _ht, int _ncpu, int _multiplane, IScriptEnvironment* env);
// This is the constructor. It does not return any value, and is always used,
// when an instance of the class is created.
// Since there is no code in this, this is the definition.
~FFT3DFilter();
// The is the destructor definition. This is called when the filter is destroyed.
PVideoFrame __stdcall GetFrame(int n, IScriptEnvironment* env);
// This is the function that AviSynth calls to get a given frame.
// So when this functions gets called, the filter is supposed to return frame n.
// Auto register AVS+ mode: serialized
int __stdcall SetCacheHints(int cachehints, int frame_range) override {
return cachehints == CACHE_GET_MTMODE ? (bt==0 ? MT_SERIALIZED : MT_MULTI_INSTANCE) : 0;
}
};
//-------------------------------------------------------------------
// The following is the implementation
// of the defined functions.
//Here is the acutal constructor code used
FFT3DFilter::FFT3DFilter(PClip _child, float _sigma, float _beta, int _plane, int _bw, int _bh, int _bt, int _ow, int _oh,
float _kratio, float _sharpen, float _scutoff, float _svr, float _smin, float _smax,
bool _measure, bool _interlaced, int _wintype,
int _pframe, int _px, int _py, bool _pshow, float _pcutoff, float _pfactor,
float _sigma2, float _sigma3, float _sigma4, float _degrid,
float _dehalo, float _hr, float _ht, int _ncpu, int _multiplane, IScriptEnvironment* env) :
GenericVideoFilter(_child), sigma(_sigma), beta(_beta), plane(_plane), bw(_bw), bh(_bh), bt(_bt), ow(_ow), oh(_oh),
kratio(_kratio), sharpen(_sharpen), scutoff(_scutoff), svr(_svr), smin(_smin), smax(_smax),
measure(_measure), interlaced(_interlaced), wintype(_wintype),
pframe(_pframe), px(_px), py(_py), pshow(_pshow), pcutoff(_pcutoff), pfactor(_pfactor),
sigma2(_sigma2), sigma3(_sigma3), sigma4(_sigma4), degrid(_degrid),
dehalo(_dehalo), hr(_hr), ht(_ht), ncpu(_ncpu), multiplane(_multiplane) {
// This is the implementation of the constructor.
// The child clip (source clip) is inherited by the GenericVideoFilter,
// where the following variables gets defined:
// PClip child; // Contains the source clip.
// VideoInfo vi; // Contains videoinfo on the source clip.
static int id = 0; _instance_id = id++;
reentrancy_check = false;
_RPT1(0, "FFT3DFilter.Create instance_id=%d\n", _instance_id);
// Check frame property support
has_at_least_v8 = true;
try { env->CheckVersion(8); }
catch (const AvisynthError&) { has_at_least_v8 = false; }
int i, j;
pixelsize = vi.ComponentSize();
bits_per_pixel = vi.BitsPerComponent();
float factor;
if (pixelsize == 1) factor = 1.0f;
else if (pixelsize == 2) factor = float(1 << (bits_per_pixel-8));
else // float
factor = 1 / 255.0f;
sigma = sigma * factor;
sigma2 = sigma2 * factor;
sigma3 = sigma3 * factor;
sigma4 = sigma4 * factor;
smin = smin * factor;
smax = smax * factor;
int planes_y[4] = { PLANAR_Y, PLANAR_U, PLANAR_V, PLANAR_A };
int planes_r[4] = { PLANAR_G, PLANAR_B, PLANAR_R, PLANAR_A };
int *current_planes = (vi.IsYUV() || vi.IsYUVA()) ? planes_y : planes_r;
for (i = 0; i < 4; i++)
planes[i] = current_planes[i];
#ifndef X86_64
_mm_empty(); // _asm emms;
#endif
// if (bw%2 !=0 ) env->ThrowError("FFT3DFilter: Block width must be even"); // I forget why even, so removed in v 1.2
// if (bh%2 !=0 ) env->ThrowError("FFT3DFilter: Block height must be even");
if (ow * 2 > bw) env->ThrowError("FFT3DFilter: Must not be 2*ow > bw");
if (oh * 2 > bh) env->ThrowError("FFT3DFilter: Must not be 2*oh > bh");
if (ow < 0) ow = bw / 3; // changed from bw/4 to bw/3 in v.1.2
if (oh < 0) oh = bh / 3; // changed from bh/4 to bh/3 in v.1.2
if (bt < -1 || bt >5) env->ThrowError("FFT3DFilter: bt must be -1(Sharpen), 0(Kalman), 1,2,3,4,5(Wiener)");
/*
(Parameter bt = 1)
2D(spatial) Wiener filter for spectrum data. Use current frame data only.
Reduce weak frequencies(with small power spectral density) by optimal Wiener filter with some
given noise value. Sharpening and denoising are simultaneous in this mode.
(Parameter bt = 2)
3D Wiener filter for spectrum data.
Add third dimension to FFT by using previous and current frame data.
Reduce weak frequencies (with small power spectral density) by optimal Wiener filter with some
given noise value.
(Parameter bt = 3)
Also 3D Wiener filter for spectrum data with previous, current and next frame data.
(Parameter bt = 4)
Also 3D Wiener filter for spectrum data with two previous, current and next frame data.
(Parameter bt = 5)
Also 3D Wiener filter for spectrum data with two previous, current and two next frames data.
(Parameter bt = 0)
Temporal Kalman filter for spectrum data.
Use all previous frames data to get estimation of cleaned current data with optimal recursive
data process algorithm.
The filter starts work with small(= 1) gain(degree of noise reducing), and than gradually(in frames sequence) increases the gain
if inter - frame local spectrum(noise) variation is small.
So, Kalman filter can provide stronger noise reduction than Wiener filter.
The Kalman filter gain is limited by some given noise value.
The local gain(and filter work) is reset to 1 when local variation exceeds the given threshold
(due to motion, scene change, etc).
So, the Kalman filter output is history - dependent(on frame taken as a start filtered frame).
PF: bt==0 have to be MT_SERIALIZED because it maintains internal history
*/
// plane: 0 - luma(Y), 1 - chroma U, 2 - chroma V
// multiplanes are handled in FFT3DFilterMulti constructor: 3 - chroma planes U and V, 4 - both luma and chroma(default = 0)
if (vi.IsPlanar()) // also for grey
{
int avs_plane = planes[plane];
bool greyOrRgb = vi.IsY() || vi.IsRGB();
int xRatioShift = (greyOrRgb) ? 0 : vi.GetPlaneWidthSubsampling(avs_plane);
int yRatioShift = (greyOrRgb) ? 0 : vi.GetPlaneHeightSubsampling(avs_plane);
nox = ((vi.width >> xRatioShift) - ow + (bw - ow - 1)) / (bw - ow);
noy = ((vi.height >> yRatioShift) - oh + (bh - oh - 1)) / (bh - oh);
/*
if (plane == 0)
{ // Y
nox = (vi.width - ow + (bw - ow - 1)) / (bw - ow); //removed mirrors (added below) in v.1.2
noy = (vi.height - oh + (bh - oh - 1)) / (bh - oh);
}
else if (plane == 1 || plane == 2) // U,V
{
nox = ((vi.width >> vi.GetPlaneWidthSubsampling(1 << plane)) - ow + (bw - ow - 1)) / (bw - ow);
noy = ((vi.height >> vi.GetPlaneHeightSubsampling(1 << plane)) - oh + (bh - oh - 1)) / (bh - oh);
}
*/
}
/* handled in planar
else if (vi.IsY8())
{
if (plane == 0)
{ // Y
nox = (vi.width - ow + (bw - ow - 1)) / (bw - ow);
noy = (vi.height - oh + (bh - oh - 1)) / (bh - oh);
}
}*/
else if (vi.IsYUY2())
{
if (plane == 0)
{ // Y
nox = (vi.width - ow + (bw - ow - 1)) / (bw - ow);
noy = (vi.height - oh + (bh - oh - 1)) / (bh - oh);
}
else if (plane == 1 || plane == 2) // U,V
{
nox = (vi.width / 2 - ow + (bw - ow - 1)) / (bw - ow);
noy = (vi.height - oh + (bh - oh - 1)) / (bh - oh);
}
else
env->ThrowError("FFT3DFilter: internal plane must be 0,1,2");
}
else
env->ThrowError("FFT3DFilter: video must be planar or YUY2");
// padding by 1 block per side
nox += 2;
noy += 2;
mirw = bw - ow; // set mirror size as block interval
mirh = bh - oh;
if (beta < 1)
env->ThrowError("FFT3DFilter: beta must be not less 1.0");
int istat;
try {
fftfp.load();
}
catch (const std::exception& e)
{
throw AvisynthError(e.what());
}
coverwidth = nox*(bw - ow) + ow;
coverheight = noy*(bh - oh) + oh;
coverpitch = ((coverwidth + 7) / 8) * 8; // align to 8 elements. Pitch is element-granularity. For byte pitch, multiply is by pixelsize
coverbuf = (BYTE*)malloc(coverheight*coverpitch*pixelsize);
{
std::lock_guard<std::mutex> lock(fftw_mutex);
int insize = bw * bh * nox * noy;
in = (float*)fftfp.fftwf_malloc(sizeof(float) * insize);
outwidth = bw / 2 + 1; // width (pitch) of complex fft block
outpitch = ((outwidth + 1) / 2) * 2; // must be even for SSE - v1.7
outsize = outpitch * bh * nox * noy; // replace outwidth to outpitch here and below in v1.7
// out = (fftwf_complex *)fftwf_malloc(sizeof(fftwf_complex) * outsize);
// if (bt >= 2)
// outprev = (fftwf_complex *)fftwf_malloc(sizeof(fftwf_complex) * outsize);
// if (bt >= 3)
// outnext = (fftwf_complex *)fftwf_malloc(sizeof(fftwf_complex) * outsize);
// if (bt >= 4)
// outprev2 = (fftwf_complex *)fftwf_malloc(sizeof(fftwf_complex) * outsize);
if (bt == 0) // Kalman
{
outLast = (fftwf_complex*)fftfp.fftwf_malloc(sizeof(fftwf_complex) * outsize);
covar = (fftwf_complex*)fftfp.fftwf_malloc(sizeof(fftwf_complex) * outsize);
covarProcess = (fftwf_complex*)fftfp.fftwf_malloc(sizeof(fftwf_complex) * outsize);
}
outrez = (fftwf_complex*)fftfp.fftwf_malloc(sizeof(fftwf_complex) * outsize); //v1.8
gridsample = (fftwf_complex*)fftfp.fftwf_malloc(sizeof(fftwf_complex) * outsize); //v1.8
// fft cache - added in v1.8
cachesize = bt + 2;
cachewhat = (int*)malloc(sizeof(int) * cachesize);
cachefft = (fftwf_complex**)fftfp.fftwf_malloc(sizeof(fftwf_complex*) * cachesize);
for (i = 0; i < cachesize; i++)
{
cachefft[i] = (fftwf_complex*)fftfp.fftwf_malloc(sizeof(fftwf_complex) * outsize);
cachewhat[i] = -1; // init as notexistant
}
}
int planFlags;
// use FFTW_ESTIMATE or FFTW_MEASURE (more optimal plan, but with time calculation at load stage)
if (measure)
planFlags = FFTW_MEASURE;
else
planFlags = FFTW_ESTIMATE;
int rank = 2; // 2d
ndim[0] = bh; // size of block along height
ndim[1] = bw; // size of block along width
int istride = 1;
int ostride = 1;
int idist = bw*bh;
int odist = outpitch*bh;// v1.7 (was outwidth)
inembed[0] = bh;
inembed[1] = bw;
onembed[0] = bh;
onembed[1] = outpitch;// v1.7 (was outwidth)
howmanyblocks = nox*noy;
// *inembed = NULL;
// *onembed = NULL;
{
std::lock_guard<std::mutex> lock(fftw_mutex);
if (fftfp.has_threading())
fftfp.fftwf_init_threads();
if (ncpu > 1 && fftfp.has_threading()) {
fftfp.fftwf_plan_with_nthreads(ncpu);
}
plan = fftfp.fftwf_plan_many_dft_r2c(rank, ndim, howmanyblocks,
in, inembed, istride, idist, outrez, onembed, ostride, odist, planFlags);
if (plan == NULL)
env->ThrowError("FFT3DFilter: FFTW plan error");
planinv = fftfp.fftwf_plan_many_dft_c2r(rank, ndim, howmanyblocks,
outrez, onembed, ostride, odist, in, inembed, istride, idist, planFlags);
if (planinv == NULL)
env->ThrowError("FFT3DFilter: FFTW plan error");
if (fftfp.has_threading())
fftfp.fftwf_plan_with_nthreads(1);
}
wanxl = (float*)malloc(ow * sizeof(float));
wanxr = (float*)malloc(ow * sizeof(float));
wanyl = (float*)malloc(oh * sizeof(float));
wanyr = (float*)malloc(oh * sizeof(float));
wsynxl = (float*)malloc(ow * sizeof(float));
wsynxr = (float*)malloc(ow * sizeof(float));
wsynyl = (float*)malloc(oh * sizeof(float));
wsynyr = (float*)malloc(oh * sizeof(float));
{
std::lock_guard<std::mutex> lock(fftw_mutex);
wsharpen = (float*)fftfp.fftwf_malloc(bh * outpitch * sizeof(float));
wdehalo = (float*)fftfp.fftwf_malloc(bh * outpitch * sizeof(float));
}
// define analysis and synthesis windows
// combining window (analize mult by synthesis) is raised cosine (Hanning)
float pi = 3.1415926535897932384626433832795f;
if (wintype == 0) // window type
{ // , used in all version up to 1.3
// half-cosine, the same for analysis and synthesis
// define analysis windows
for (i = 0; i < ow; i++)
{
wanxl[i] = cosf(pi*(i - ow + 0.5f) / (ow * 2)); // left analize window (half-cosine)
wanxr[i] = cosf(pi*(i + 0.5f) / (ow * 2)); // right analize window (half-cosine)
}
for (i = 0; i < oh; i++)
{
wanyl[i] = cosf(pi*(i - oh + 0.5f) / (oh * 2));
wanyr[i] = cosf(pi*(i + 0.5f) / (oh * 2));
}
// use the same windows for synthesis too.
for (i = 0; i < ow; i++)
{
wsynxl[i] = wanxl[i]; // left window (half-cosine)
wsynxr[i] = wanxr[i]; // right window (half-cosine)
}
for (i = 0; i < oh; i++)
{
wsynyl[i] = wanyl[i];
wsynyr[i] = wanyr[i];
}
}
else if (wintype == 1) // added in v.1.4
{
// define analysis windows as more flat (to decrease grid)
for (i = 0; i < ow; i++)
{
wanxl[i] = sqrt(cosf(pi*(i - ow + 0.5f) / (ow * 2)));
wanxr[i] = sqrt(cosf(pi*(i + 0.5f) / (oh * 2)));
}
for (i = 0; i < oh; i++)
{
wanyl[i] = sqrt(cosf(pi*(i - oh + 0.5f) / (oh * 2)));
wanyr[i] = sqrt(cosf(pi*(i + 0.5f) / (oh * 2)));
}
// define synthesis as supplenent to rised cosine (Hanning)
for (i = 0; i < ow; i++)
{
wsynxl[i] = wanxl[i] * wanxl[i] * wanxl[i]; // left window
wsynxr[i] = wanxr[i] * wanxr[i] * wanxr[i]; // right window
}
for (i = 0; i < oh; i++)
{
wsynyl[i] = wanyl[i] * wanyl[i] * wanyl[i];
wsynyr[i] = wanyr[i] * wanyr[i] * wanyr[i];
}
}
else // (wintype==2) - added in v.1.4
{
// define analysis windows as flat (to prevent grid)
for (i = 0; i < ow; i++)
{
wanxl[i] = 1;
wanxr[i] = 1;
}
for (i = 0; i < oh; i++)
{
wanyl[i] = 1;
wanyr[i] = 1;
}
// define synthesis as rised cosine (Hanning)
for (i = 0; i < ow; i++)
{
wsynxl[i] = cosf(pi*(i - ow + 0.5f) / (ow * 2));
wsynxl[i] = wsynxl[i] * wsynxl[i];// left window (rised cosine)
wsynxr[i] = cosf(pi*(i + 0.5f) / (ow * 2));
wsynxr[i] = wsynxr[i] * wsynxr[i]; // right window (falled cosine)
}
for (i = 0; i < oh; i++)
{
wsynyl[i] = cosf(pi*(i - oh + 0.5f) / (oh * 2));
wsynyl[i] = wsynyl[i] * wsynyl[i];
wsynyr[i] = cosf(pi*(i + 0.5f) / (oh * 2));
wsynyr[i] = wsynyr[i] * wsynyr[i];
}
}
// window for sharpen
for (j = 0; j < bh; j++)
{
int dj = j;
if (j >= bh / 2)
dj = bh - j;
float d2v = float(dj*dj)*(svr*svr) / ((bh / 2)*(bh / 2)); // v1.7
for (i = 0; i < outwidth; i++)
{
float d2 = d2v + float(i*i) / ((bw / 2)*(bw / 2)); // distance_2 - v1.7
wsharpen[i] = 1 - exp(-d2 / (2 * scutoff*scutoff));
}
wsharpen += outpitch;
}
wsharpen -= outpitch*bh; // restore pointer
// window for dehalo - added in v1.9
float wmax = 0;
for (j = 0; j < bh; j++)
{
int dj = j;
if (j >= bh / 2)
dj = bh - j;
float d2v = float(dj*dj)*(svr*svr) / ((bh / 2)*(bh / 2));
for (i = 0; i < outwidth; i++)
{
float d2 = d2v + float(i*i) / ((bw / 2)*(bw / 2)); // squared distance in frequency domain
float d1 = sqrt(d2);
wdehalo[i] = exp(-0.7f*d2*hr*hr) - exp(-d2*hr*hr); // some window with max around 1/hr, small at low and high frequencies
if (wdehalo[i] > wmax) wmax = wdehalo[i]; // for normalization
}
wdehalo += outpitch;
}
wdehalo -= outpitch*bh; // restore pointer
for (j = 0; j < bh; j++)
{
for (i = 0; i < outwidth; i++)
{
wdehalo[i] /= wmax; // normalize
}
wdehalo += outpitch;
}
wdehalo -= outpitch*bh; // restore pointer
// init nlast
nlast = -999; // init as nonexistant
btcurlast = -999; // init as nonexistant
norm = 1.0f / (bw*bh); // do not forget set FFT normalization factor
sigmaSquaredNoiseNormed2D = sigma*sigma / norm;
sigmaNoiseNormed2D = sigma / sqrtf(norm);
sigmaMotionNormed = sigma*kratio / sqrtf(norm);
sigmaSquaredSharpenMinNormed = smin*smin / norm;
sigmaSquaredSharpenMaxNormed = smax*smax / norm;
ht2n = ht*ht / norm; // halo threshold squared and normed - v1.9
// init Kalman
if (bt == 0) // Kalman
{
fill_complex(outLast, outsize, 0, 0);
fill_complex(covar, outsize, sigmaSquaredNoiseNormed2D, sigmaSquaredNoiseNormed2D); // fixed bug in v.1.1
fill_complex(covarProcess, outsize, sigmaSquaredNoiseNormed2D, sigmaSquaredNoiseNormed2D);// fixed bug in v.1.1
}
CPUFlags = env->GetCPUFlags(); //re-enabled in v.1.9
mean = (float*)malloc(nox*noy * sizeof(float));
pwin = (float*)malloc(bh*outpitch * sizeof(float)); // pattern window array
float fw2, fh2;
for (j = 0; j < bh; j++)
{
if (j < bh / 2)
fh2 = (j*2.0f / bh)*(j*2.0f / bh);
else
fh2 = ((bh - 1 - j)*2.0f / bh)*((bh - 1 - j)*2.0f / bh);
for (i = 0; i < outwidth; i++)
{
fw2 = (i*2.0f / bw)*(j*2.0f / bw);
pwin[i] = (fh2 + fw2) / (fh2 + fw2 + pcutoff*pcutoff);
}
pwin += outpitch;
}
pwin -= outpitch*bh; // restore pointer
{
std::lock_guard<std::mutex> lock(fftw_mutex);
pattern2d = (float*)fftfp.fftwf_malloc(bh * outpitch * sizeof(float)); // noise pattern window array
pattern3d = (float*)fftfp.fftwf_malloc(bh * outpitch * sizeof(float)); // noise pattern window array
}
if ((sigma2 != sigma || sigma3 != sigma || sigma4 != sigma) && pfactor == 0)
{// we have different sigmas, so create pattern from sigmas
SigmasToPattern(sigma, sigma2, sigma3, sigma4, bh, outwidth, outpitch, norm, pattern2d);
isPatternSet = true;
pfactor = 1;
}