-
Notifications
You must be signed in to change notification settings - Fork 2
/
amaze_demosaic_RT.c
1412 lines (1212 loc) · 61.2 KB
/
amaze_demosaic_RT.c
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
////////////////////////////////////////////////////////////////
//
// AMaZE demosaic algorithm
// (Aliasing Minimization and Zipper Elimination)
//
// copyright (c) 2008-2010 Emil Martinec <ejmartin@uchicago.edu>
//
// incorporating ideas of Luis Sanz Rodrigues and Paul Lee
//
// original code dated: May 27, 2010, last update 4b77ef6013ae (Jan 16, 2014)
// https://code.google.com/p/rawtherapee/source/browse/rtengine/amaze_demosaic_RT.cc
// modified by a1ex for integration with cr2hdr
//
// amaze_interpolate_RT.cc is free software: you can redistribute it and/or modify
// it under the terms of the GNU General Public License as published by
// the Free Software Foundation, either version 3 of the License, or
// (at your option) any later version.
//
// 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, see <http://www.gnu.org/licenses/>.
//
////////////////////////////////////////////////////////////////
#include <stdio.h>
#include <stdlib.h>
#include <stdint.h>
#include <string.h>
#include <math.h>
#include <time.h>
#include "amaze-port.c"
void amaze_demosaic_RT(
float** rawData, /* holds preprocessed pixel values, rawData[i][j] corresponds to the ith row and jth column */
float** red, /* the interpolated red plane */
float** green, /* the interpolated green plane */
float** blue, /* the interpolated blue plane */
int winx, int winy, /* crop window for demosaicing */
int winw, int winh
)
{
printf ("AMaZE interpolation ...\n");
clock_t t1,t2;
t1 = clock();
#define HCLIP(x) x //is this still necessary???
//min(clip_pt,x)
int width=winw, height=winh;
const float clip_pt = 1/initialGain;
const float clip_pt8 = 0.8f/initialGain;
#define TS 160 // Tile size; the image is processed in square tiles to lower memory requirements and facilitate multi-threading
#define TSH 80 // half of Tile size
// local variables
//offset of R pixel within a Bayer quartet
int ex, ey;
//shifts of pointer value to access pixels in vertical and diagonal directions
static const int v1=TS, v2=2*TS, v3=3*TS, p1=-TS+1, p2=-2*TS+2, p3=-3*TS+3, m1=TS+1, m2=2*TS+2, m3=3*TS+3;
//tolerance to avoid dividing by zero
static const float eps=1e-5, epssq=1e-10; //tolerance to avoid dividing by zero
//adaptive ratios threshold
static const float arthresh=0.75;
//nyquist texture test threshold
static const float nyqthresh=0.5;
//gaussian on 5x5 quincunx, sigma=1.2
static const float gaussodd[4] = {0.14659727707323927f, 0.103592713382435f, 0.0732036125103057f, 0.0365543548389495f};
//gaussian on 5x5, sigma=1.2
static const float gaussgrad[6] = {0.07384411893421103f, 0.06207511968171489f, 0.0521818194747806f,
0.03687419286733595f, 0.03099732204057846f, 0.018413194161458882f};
//gaussian on 5x5 alt quincunx, sigma=1.5
static const float gausseven[2] = {0.13719494435797422f, 0.05640252782101291f};
//guassian on quincunx grid
static const float gquinc[4] = {0.169917f, 0.108947f, 0.069855f, 0.0287182f};
//~ volatile double progress = 0.0;
// %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
// Issue 1676
// Moved from inside the parallel section
struct s_hv {
float h;
float v;
};
//~ #pragma omp parallel
{
//position of top/left corner of the tile
int top, left;
// beginning of storage block for tile
char *buffer;
// green values
float (*rgbgreen);
// sum of square of horizontal gradient and square of vertical gradient
float (*delhvsqsum);
// gradient based directional weights for interpolation
float (*dirwts0);
float (*dirwts1);
// vertically interpolated color differences G-R, G-B
float (*vcd);
// horizontally interpolated color differences
float (*hcd);
// alternative vertical interpolation
float (*vcdalt);
// alternative horizontal interpolation
float (*hcdalt);
// square of average color difference
float (*cddiffsq);
// weight to give horizontal vs vertical interpolation
float (*hvwt);
// final interpolated color difference
float (*Dgrb)[TS*TSH];
// float (*Dgrb)[2];
// gradient in plus (NE/SW) direction
float (*delp);
// gradient in minus (NW/SE) direction
float (*delm);
// diagonal interpolation of R+B
float (*rbint);
// horizontal and vertical curvature of interpolated G (used to refine interpolation in Nyquist texture regions)
struct s_hv (*Dgrb2);
// difference between up/down interpolations of G
float (*dgintv);
// difference between left/right interpolations of G
float (*dginth);
// diagonal (plus) color difference R-B or G1-G2
// float (*Dgrbp1);
// diagonal (minus) color difference R-B or G1-G2
// float (*Dgrbm1);
float (*Dgrbsq1m);
float (*Dgrbsq1p);
// struct s_mp (*Dgrbsq1);
// square of diagonal color difference
// float (*Dgrbpsq1);
// square of diagonal color difference
// float (*Dgrbmsq1);
// tile raw data
float (*cfa);
// relative weight for combining plus and minus diagonal interpolations
float (*pmwt);
// interpolated color difference R-B in minus and plus direction
float (*rbm);
float (*rbp);
// nyquist texture flag 1=nyquist, 0=not nyquist
char (*nyquist);
#define CLF 1
// assign working space
buffer = (char *) calloc(22*sizeof(float)*TS*TS + sizeof(char)*TS*TSH+23*CLF*64 + 63, 1);
char *data;
data = (char*)( ( (uintptr_t)buffer + (uintptr_t)63) / 64 * 64);
//merror(buffer,"amaze_interpolate()");
rgbgreen = (float (*)) data; //pointers to array
delhvsqsum = (float (*)) ((char*)rgbgreen + sizeof(float)*TS*TS + CLF*64);
dirwts0 = (float (*)) ((char*)delhvsqsum + sizeof(float)*TS*TS + CLF*64);
dirwts1 = (float (*)) ((char*)dirwts0 + sizeof(float)*TS*TS + CLF*64);
vcd = (float (*)) ((char*)dirwts1 + sizeof(float)*TS*TS + CLF*64);
hcd = (float (*)) ((char*)vcd + sizeof(float)*TS*TS + CLF*64);
vcdalt = (float (*)) ((char*)hcd + sizeof(float)*TS*TS + CLF*64);
hcdalt = (float (*)) ((char*)vcdalt + sizeof(float)*TS*TS + CLF*64);
cddiffsq = (float (*)) ((char*)hcdalt + sizeof(float)*TS*TS + CLF*64);
hvwt = (float (*)) ((char*)cddiffsq + sizeof(float)*TS*TS + CLF*64);
Dgrb = (float (*)[TS*TSH]) ((char*)hvwt + sizeof(float)*TS*TSH + CLF*64);
delp = (float (*)) ((char*)Dgrb + sizeof(float)*TS*TS + CLF*64);
delm = (float (*)) ((char*)delp + sizeof(float)*TS*TSH + CLF*64);
rbint = (float (*)) ((char*)delm + sizeof(float)*TS*TSH + CLF*64);
Dgrb2 = (struct s_hv (*)) ((char*)rbint + sizeof(float)*TS*TSH + CLF*64);
dgintv = (float (*)) ((char*)Dgrb2 + sizeof(float)*TS*TS + CLF*64);
dginth = (float (*)) ((char*)dgintv + sizeof(float)*TS*TS + CLF*64);
Dgrbsq1m = (float (*)) ((char*)dginth + sizeof(float)*TS*TS + CLF*64);
Dgrbsq1p = (float (*)) ((char*)Dgrbsq1m + sizeof(float)*TS*TSH + CLF*64);
cfa = (float (*)) ((char*)Dgrbsq1p + sizeof(float)*TS*TSH + CLF*64);
pmwt = (float (*)) ((char*)cfa + sizeof(float)*TS*TS + CLF*64);
rbm = (float (*)) ((char*)pmwt + sizeof(float)*TS*TSH + CLF*64);
rbp = (float (*)) ((char*)rbm + sizeof(float)*TS*TSH + CLF*64);
nyquist = (char (*)) ((char*)rbp + sizeof(float)*TS*TSH + CLF*64);
#undef CLF
// %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
//determine GRBG coset; (ey,ex) is the offset of the R subarray
if (FC(0,0)==1) {//first pixel is G
if (FC(0,1)==0) {ey=0; ex=1;} else {ey=1; ex=0;}
} else {//first pixel is R or B
if (FC(0,0)==0) {ey=0; ex=0;} else {ey=1; ex=1;}
}
// Main algorithm: Tile loop
//#pragma omp parallel for shared(rawData,height,width,red,green,blue) private(top,left) schedule(dynamic)
//code is openmp ready; just have to pull local tile variable declarations inside the tile loop
// Issue 1676
// use collapse(2) to collapse the 2 loops to one large loop, so there is better scaling
//~ #pragma omp for schedule(dynamic) collapse(2) nowait
for (top=winy-16; top < winy+height; top += TS-32)
for (left=winx-16; left < winx+width; left += TS-32) {
memset(nyquist, 0, sizeof(char)*TS*TSH);
memset(rbint, 0, sizeof(float)*TS*TSH);
//location of tile bottom edge
int bottom = min(top+TS,winy+height+16);
//location of tile right edge
int right = min(left+TS, winx+width+16);
//tile width (=TS except for right edge of image)
int rr1 = bottom - top;
//tile height (=TS except for bottom edge of image)
int cc1 = right - left;
//tile vars
//counters for pixel location in the image
int row, col;
//min and max row/column in the tile
int rrmin, rrmax, ccmin, ccmax;
//counters for pixel location within the tile
int rr, cc;
//color index 0=R, 1=G, 2=B
int c;
//pointer counters within the tile
int indx, indx1;
//dummy indices
int i, j;
//color ratios in up/down/left/right directions
float cru, crd, crl, crr;
//adaptive weights for vertical/horizontal/plus/minus directions
float vwt, hwt, pwt, mwt;
//vertical and horizontal G interpolations
float Gintv, Ginth;
//G interpolated in vert/hor directions using adaptive ratios
float guar, gdar, glar, grar;
//G interpolated in vert/hor directions using Hamilton-Adams method
float guha, gdha, glha, grha;
//interpolated G from fusing left/right or up/down
float Ginthha, Gintvha;
//color difference (G-R or G-B) variance in up/down/left/right directions
float Dgrbvvaru, Dgrbvvard, Dgrbhvarl, Dgrbhvarr;
float uave, dave, lave, rave;
//color difference variances in vertical and horizontal directions
float vcdvar, hcdvar, vcdvar1, hcdvar1, hcdaltvar, vcdaltvar;
//adaptive interpolation weight using variance of color differences
float varwt; // 639 - 644
//adaptive interpolation weight using difference of left-right and up-down G interpolations
float diffwt; // 640 - 644
//alternative adaptive weight for combining horizontal/vertical interpolations
float hvwtalt; // 745 - 748
//interpolation of G in four directions
float gu, gd, gl, gr;
//variance of G in vertical/horizontal directions
float gvarh, gvarv;
//Nyquist texture test
float nyqtest; // 658 - 681
//accumulators for Nyquist texture interpolation
float sumh, sumv, sumsqh, sumsqv, areawt;
//color ratios in diagonal directions
float crse, crnw, crne, crsw;
//color differences in diagonal directions
float rbse, rbnw, rbne, rbsw;
//adaptive weights for combining diagonal interpolations
float wtse, wtnw, wtsw, wtne;
//alternate weight for combining diagonal interpolations
float pmwtalt; // 885 - 888
//variance of R-B in plus/minus directions
float rbvarm; // 843 - 848
// %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
// rgb from input CFA data
// rgb values should be floating point number between 0 and 1
// after white balance multipliers are applied
// a 16 pixel border is added to each side of the image
// bookkeeping for borders
if (top<winy) {rrmin=16;} else {rrmin=0;}
if (left<winx) {ccmin=16;} else {ccmin=0;}
if (bottom>(winy+height)) {rrmax=winy+height-top;} else {rrmax=rr1;}
if (right>(winx+width)) {ccmax=winx+width-left;} else {ccmax=cc1;}
#ifdef __SSE2__
const __m128 c65535v = _mm_set1_ps( 65535.0f );
__m128 tempv;
for (rr=rrmin; rr < rrmax; rr++){
for (row=rr+top, cc=ccmin; cc < ccmax-3; cc+=4) {
indx1=rr*TS+cc;
tempv = LVFU(rawData[row][cc+left]) / c65535v;
_mm_store_ps( &cfa[indx1], tempv );
_mm_store_ps( &rgbgreen[indx1], tempv );
}
for (; cc < ccmax; cc++) {
indx1=rr*TS+cc;
cfa[indx1] = (rawData[row][cc+left])/65535.0f;
if(FC(rr,cc)==1)
rgbgreen[indx1] = cfa[indx1];
}
}
// %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
//fill borders
if (rrmin>0) {
for (rr=0; rr<16; rr++)
for (cc=ccmin,row = 32-rr+top; cc<ccmax; cc++) {
cfa[rr*TS+cc] = (rawData[row][cc+left])/65535.0f;
if(FC(rr,cc)==1)
rgbgreen[rr*TS+cc] = cfa[rr*TS+cc];
}
}
if (rrmax<rr1) {
for (rr=0; rr<16; rr++)
for (cc=ccmin; cc<ccmax; cc+=4) {
indx1 = (rrmax+rr)*TS+cc;
tempv = LVFU(rawData[(winy+height-rr-2)][left+cc]) / c65535v;
_mm_store_ps( &cfa[indx1], tempv );
_mm_store_ps( &rgbgreen[indx1], tempv );
}
}
if (ccmin>0) {
for (rr=rrmin; rr<rrmax; rr++)
for (cc=0,row = rr + top; cc<16; cc++) {
cfa[rr*TS+cc] = (rawData[row][32-cc+left])/65535.0f;
if(FC(rr,cc)==1)
rgbgreen[rr*TS+cc] = cfa[rr*TS+cc];
}
}
if (ccmax<cc1) {
for (rr=rrmin; rr<rrmax; rr++)
for (cc=0; cc<16; cc++) {
cfa[rr*TS+ccmax+cc] = (rawData[(top+rr)][(winx+width-cc-2)])/65535.0f;
if(FC(rr,cc)==1)
rgbgreen[rr*TS+ccmax+cc] = cfa[rr*TS+ccmax+cc];
}
}
//also, fill the image corners
if (rrmin>0 && ccmin>0) {
for (rr=0; rr<16; rr++)
for (cc=0; cc<16; cc+=4) {
indx1 = (rr)*TS+cc;
tempv = LVFU(rawData[winy+32-rr][winx+32-cc]) / c65535v;
_mm_store_ps( &cfa[indx1], tempv );
_mm_store_ps( &rgbgreen[indx1], tempv );
}
}
if (rrmax<rr1 && ccmax<cc1) {
for (rr=0; rr<16; rr++)
for (cc=0; cc<16; cc+=4) {
indx1 = (rrmax+rr)*TS+ccmax+cc;
tempv = LVFU(rawData[(winy+height-rr-2)][(winx+width-cc-2)]) / c65535v;
_mm_storeu_ps( &cfa[indx1], tempv );
_mm_storeu_ps( &rgbgreen[indx1], tempv );
}
}
if (rrmin>0 && ccmax<cc1) {
for (rr=0; rr<16; rr++)
for (cc=0; cc<16; cc++) {
cfa[(rr)*TS+ccmax+cc] = (rawData[(winy+32-rr)][(winx+width-cc-2)])/65535.0f;
if(FC(rr,cc)==1)
rgbgreen[(rr)*TS+ccmax+cc] = cfa[(rr)*TS+ccmax+cc];
}
}
if (rrmax<rr1 && ccmin>0) {
for (rr=0; rr<16; rr++)
for (cc=0; cc<16; cc++) {
cfa[(rrmax+rr)*TS+cc] = (rawData[(winy+height-rr-2)][(winx+32-cc)])/65535.0f;
if(FC(rr,cc)==1)
rgbgreen[(rrmax+rr)*TS+cc] = cfa[(rrmax+rr)*TS+cc];
}
}
#else
for (rr=rrmin; rr < rrmax; rr++)
for (row=rr+top, cc=ccmin; cc < ccmax; cc++) {
indx1=rr*TS+cc;
cfa[indx1] = (rawData[row][cc+left])/65535.0f;
if(FC(rr,cc)==1)
rgbgreen[indx1] = cfa[indx1];
}
// %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
//fill borders
if (rrmin>0) {
for (rr=0; rr<16; rr++)
for (cc=ccmin,row = 32-rr+top; cc<ccmax; cc++) {
cfa[rr*TS+cc] = (rawData[row][cc+left])/65535.0f;
if(FC(rr,cc)==1)
rgbgreen[rr*TS+cc] = cfa[rr*TS+cc];
}
}
if (rrmax<rr1) {
for (rr=0; rr<16; rr++)
for (cc=ccmin; cc<ccmax; cc++) {
cfa[(rrmax+rr)*TS+cc] = (rawData[(winy+height-rr-2)][left+cc])/65535.0f;
if(FC(rr,cc)==1)
rgbgreen[(rrmax+rr)*TS+cc] = cfa[(rrmax+rr)*TS+cc];
}
}
if (ccmin>0) {
for (rr=rrmin; rr<rrmax; rr++)
for (cc=0,row = rr + top; cc<16; cc++) {
cfa[rr*TS+cc] = (rawData[row][32-cc+left])/65535.0f;
if(FC(rr,cc)==1)
rgbgreen[rr*TS+cc] = cfa[rr*TS+cc];
}
}
if (ccmax<cc1) {
for (rr=rrmin; rr<rrmax; rr++)
for (cc=0; cc<16; cc++) {
cfa[rr*TS+ccmax+cc] = (rawData[(top+rr)][(winx+width-cc-2)])/65535.0f;
if(FC(rr,cc)==1)
rgbgreen[rr*TS+ccmax+cc] = cfa[rr*TS+ccmax+cc];
}
}
//also, fill the image corners
if (rrmin>0 && ccmin>0) {
for (rr=0; rr<16; rr++)
for (cc=0; cc<16; cc++) {
cfa[(rr)*TS+cc] = (rawData[winy+32-rr][winx+32-cc])/65535.0f;
if(FC(rr,cc)==1)
rgbgreen[(rr)*TS+cc] = cfa[(rr)*TS+cc];
}
}
if (rrmax<rr1 && ccmax<cc1) {
for (rr=0; rr<16; rr++)
for (cc=0; cc<16; cc++) {
cfa[(rrmax+rr)*TS+ccmax+cc] = (rawData[(winy+height-rr-2)][(winx+width-cc-2)])/65535.0f;
if(FC(rr,cc)==1)
rgbgreen[(rrmax+rr)*TS+ccmax+cc] = cfa[(rrmax+rr)*TS+ccmax+cc];
}
}
if (rrmin>0 && ccmax<cc1) {
for (rr=0; rr<16; rr++)
for (cc=0; cc<16; cc++) {
cfa[(rr)*TS+ccmax+cc] = (rawData[(winy+32-rr)][(winx+width-cc-2)])/65535.0f;
if(FC(rr,cc)==1)
rgbgreen[(rr)*TS+ccmax+cc] = cfa[(rr)*TS+ccmax+cc];
}
}
if (rrmax<rr1 && ccmin>0) {
for (rr=0; rr<16; rr++)
for (cc=0; cc<16; cc++) {
cfa[(rrmax+rr)*TS+cc] = (rawData[(winy+height-rr-2)][(winx+32-cc)])/65535.0f;
if(FC(rr,cc)==1)
rgbgreen[(rrmax+rr)*TS+cc] = cfa[(rrmax+rr)*TS+cc];
}
}
#endif
//end of border fill
// %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
#ifdef __SSE2__
__m128 delhv,delvv;
const __m128 epsv = _mm_set1_ps( eps );
for (rr=2; rr < rr1-2; rr++) {
for (cc=0, indx=(rr)*TS+cc; cc < cc1; cc+=4, indx+=4) {
delhv = vabsf( LVFU( cfa[indx+1] ) - LVFU( cfa[indx-1] ) );
delvv = vabsf( LVF( cfa[indx+v1] ) - LVF( cfa[indx-v1] ) );
_mm_store_ps( &dirwts1[indx], epsv + vabsf( LVFU( cfa[indx+2] ) - LVF( cfa[indx] )) + vabsf( LVF( cfa[indx] ) - LVFU( cfa[indx-2] )) + delhv );
delhv = delhv * delhv;
_mm_store_ps( &dirwts0[indx], epsv + vabsf( LVF( cfa[indx+v2] ) - LVF( cfa[indx] )) + vabsf( LVF( cfa[indx] ) - LVF( cfa[indx-v2] )) + delvv );
delvv = delvv * delvv;
_mm_store_ps( &delhvsqsum[indx], delhv + delvv);
}
}
#else
// horizontal and vedrtical gradient
float delh,delv;
for (rr=2; rr < rr1-2; rr++)
for (cc=2, indx=(rr)*TS+cc; cc < cc1-2; cc++, indx++) {
delh = fabsf(cfa[indx+1]-cfa[indx-1]);
delv = fabsf(cfa[indx+v1]-cfa[indx-v1]);
dirwts0[indx] = eps+fabsf(cfa[indx+v2]-cfa[indx])+fabsf(cfa[indx]-cfa[indx-v2])+delv;
dirwts1[indx] = eps+fabsf(cfa[indx+2]-cfa[indx])+fabsf(cfa[indx]-cfa[indx-2])+delh;//+fabsf(cfa[indx+2]-cfa[indx-2]);
delhvsqsum[indx] = SQR(delh) + SQR(delv);
}
#endif
#ifdef __SSE2__
__m128 Dgrbsq1pv, Dgrbsq1mv,temp2v;
for (rr=6; rr < rr1-6; rr++){
if((FC(rr,2)&1)==0) {
for (cc=6, indx=(rr)*TS+cc; cc < cc1-6; cc+=8, indx+=8) {
tempv = LC2VFU(cfa[indx+1]);
Dgrbsq1pv = (SQRV(tempv-LC2VFU(cfa[indx+1-p1]))+SQRV(tempv-LC2VFU(cfa[indx+1+p1])));
_mm_storeu_ps( &delp[indx>>1], vabsf(LC2VFU(cfa[indx+p1])-LC2VFU(cfa[indx-p1])));
_mm_storeu_ps( &delm[indx>>1], vabsf(LC2VFU(cfa[indx+m1])-LC2VFU(cfa[indx-m1])));
Dgrbsq1mv = (SQRV(tempv-LC2VFU(cfa[indx+1-m1]))+SQRV(tempv-LC2VFU(cfa[indx+1+m1])));
_mm_storeu_ps( &Dgrbsq1m[indx>>1], Dgrbsq1mv );
_mm_storeu_ps( &Dgrbsq1p[indx>>1], Dgrbsq1pv );
}
}
else {
for (cc=6, indx=(rr)*TS+cc; cc < cc1-6; cc+=8, indx+=8) {
tempv = LC2VFU(cfa[indx]);
Dgrbsq1pv = (SQRV(tempv-LC2VFU(cfa[indx-p1]))+SQRV(tempv-LC2VFU(cfa[indx+p1])));
_mm_storeu_ps( &delp[indx>>1], vabsf(LC2VFU(cfa[indx+1+p1])-LC2VFU(cfa[indx+1-p1])));
_mm_storeu_ps( &delm[indx>>1], vabsf(LC2VFU(cfa[indx+1+m1])-LC2VFU(cfa[indx+1-m1])));
Dgrbsq1mv = (SQRV(tempv-LC2VFU(cfa[indx-m1]))+SQRV(tempv-LC2VFU(cfa[indx+m1])));
_mm_storeu_ps( &Dgrbsq1m[indx>>1], Dgrbsq1mv );
_mm_storeu_ps( &Dgrbsq1p[indx>>1], Dgrbsq1pv );
}
}
}
#else
for (rr=6; rr < rr1-6; rr++){
if((FC(rr,2)&1)==0) {
for (cc=6, indx=(rr)*TS+cc; cc < cc1-6; cc+=2, indx+=2) {
delp[indx>>1] = fabsf(cfa[indx+p1]-cfa[indx-p1]);
delm[indx>>1] = fabsf(cfa[indx+m1]-cfa[indx-m1]);
Dgrbsq1p[indx>>1]=(SQR(cfa[indx+1]-cfa[indx+1-p1])+SQR(cfa[indx+1]-cfa[indx+1+p1]));
Dgrbsq1m[indx>>1]=(SQR(cfa[indx+1]-cfa[indx+1-m1])+SQR(cfa[indx+1]-cfa[indx+1+m1]));
}
}
else {
for (cc=6, indx=(rr)*TS+cc; cc < cc1-6; cc+=2, indx+=2) {
Dgrbsq1p[indx>>1]=(SQR(cfa[indx]-cfa[indx-p1])+SQR(cfa[indx]-cfa[indx+p1]));
Dgrbsq1m[indx>>1]=(SQR(cfa[indx]-cfa[indx-m1])+SQR(cfa[indx]-cfa[indx+m1]));
delp[indx>>1] = fabsf(cfa[indx+1+p1]-cfa[indx+1-p1]);
delm[indx>>1] = fabsf(cfa[indx+1+m1]-cfa[indx+1-m1]);
}
}
}
#endif
// end of tile initialization
// %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
//interpolate vertical and horizontal color differences
#ifdef __SSE2__
__m128 sgnv,cruv,crdv,crlv,crrv,guhav,gdhav,glhav,grhav,hwtv,vwtv,Gintvhav,Ginthhav,guarv,gdarv,glarv,grarv;
vmask clipmask;
if( !(FC(4,4)&1) )
sgnv = _mm_set_ps( 1.0f, -1.0f, 1.0f, -1.0f );
else
sgnv = _mm_set_ps( -1.0f, 1.0f, -1.0f, 1.0f );
__m128 zd5v = _mm_set1_ps( 0.5f );
__m128 onev = _mm_set1_ps( 1.0f );
__m128 arthreshv = _mm_set1_ps( arthresh );
__m128 clip_pt8v = _mm_set1_ps( clip_pt8 );
for (rr=4; rr<rr1-4; rr++) {
sgnv = -sgnv;
for (cc=4,indx=rr*TS+cc; cc<cc1-7; cc+=4,indx+=4) {
//color ratios in each cardinal direction
cruv = LVF(cfa[indx-v1])*(LVF(dirwts0[indx-v2])+LVF(dirwts0[indx]))/(LVF(dirwts0[indx-v2])*(epsv+LVF(cfa[indx]))+LVF(dirwts0[indx])*(epsv+LVF(cfa[indx-v2])));
crdv = LVF(cfa[indx+v1])*(LVF(dirwts0[indx+v2])+LVF(dirwts0[indx]))/(LVF(dirwts0[indx+v2])*(epsv+LVF(cfa[indx]))+LVF(dirwts0[indx])*(epsv+LVF(cfa[indx+v2])));
crlv = LVFU(cfa[indx-1])*(LVFU(dirwts1[indx-2])+LVF(dirwts1[indx]))/(LVFU(dirwts1[indx-2])*(epsv+LVF(cfa[indx]))+LVF(dirwts1[indx])*(epsv+LVFU(cfa[indx-2])));
crrv = LVFU(cfa[indx+1])*(LVFU(dirwts1[indx+2])+LVF(dirwts1[indx]))/(LVFU(dirwts1[indx+2])*(epsv+LVF(cfa[indx]))+LVF(dirwts1[indx])*(epsv+LVFU(cfa[indx+2])));
guhav=LVF(cfa[indx-v1])+zd5v*(LVF(cfa[indx])-LVF(cfa[indx-v2]));
gdhav=LVF(cfa[indx+v1])+zd5v*(LVF(cfa[indx])-LVF(cfa[indx+v2]));
glhav=LVFU(cfa[indx-1])+zd5v*(LVF(cfa[indx])-LVFU(cfa[indx-2]));
grhav=LVFU(cfa[indx+1])+zd5v*(LVF(cfa[indx])-LVFU(cfa[indx+2]));
guarv = vself(vmaskf_lt(vabsf(onev-cruv), arthreshv), LVF(cfa[indx])*cruv, guhav);
gdarv = vself(vmaskf_lt(vabsf(onev-crdv), arthreshv), LVF(cfa[indx])*crdv, gdhav);
glarv = vself(vmaskf_lt(vabsf(onev-crlv), arthreshv), LVF(cfa[indx])*crlv, glhav);
grarv = vself(vmaskf_lt(vabsf(onev-crrv), arthreshv), LVF(cfa[indx])*crrv, grhav);
hwtv = LVFU(dirwts1[indx-1])/(LVFU(dirwts1[indx-1])+LVFU(dirwts1[indx+1]));
vwtv = LVF(dirwts0[indx-v1])/(LVF(dirwts0[indx+v1])+LVF(dirwts0[indx-v1]));
//interpolated G via adaptive weights of cardinal evaluations
Ginthhav = hwtv*grhav+(onev-hwtv)*glhav;
Gintvhav = vwtv*gdhav+(onev-vwtv)*guhav;
//interpolated color differences
_mm_store_ps( &hcdalt[indx], sgnv*(Ginthhav-LVF(cfa[indx])));
_mm_store_ps( &vcdalt[indx], sgnv*(Gintvhav-LVF(cfa[indx])));
clipmask = vorm( vorm( vmaskf_gt( LVF(cfa[indx]), clip_pt8v ), vmaskf_gt( Gintvhav, clip_pt8v ) ), vmaskf_gt( Ginthhav, clip_pt8v ));
guarv = vself( clipmask, guhav, guarv);
gdarv = vself( clipmask, gdhav, gdarv);
glarv = vself( clipmask, glhav, glarv);
grarv = vself( clipmask, grhav, grarv);
_mm_store_ps( &vcd[indx], vself( clipmask, LVF(vcdalt[indx]), sgnv*((vwtv*gdarv+(onev-vwtv)*guarv)-LVF(cfa[indx]))));
_mm_store_ps( &hcd[indx], vself( clipmask, LVF(hcdalt[indx]), sgnv*((hwtv*grarv+(onev-hwtv)*glarv)-LVF(cfa[indx]))));
//differences of interpolations in opposite directions
_mm_store_ps(&dgintv[indx],_mm_min_ps(SQRV(guhav-gdhav),SQRV(guarv-gdarv)));
_mm_store_ps(&dginth[indx],_mm_min_ps(SQRV(glhav-grhav),SQRV(glarv-grarv)));
}
}
#else
bool fcswitch;
for (rr=4; rr<rr1-4; rr++) {
for (cc=4,indx=rr*TS+cc,fcswitch = FC(rr,cc)&1; cc<cc1-4; cc++,indx++) {
//color ratios in each cardinal direction
cru = cfa[indx-v1]*(dirwts0[indx-v2]+dirwts0[indx])/(dirwts0[indx-v2]*(eps+cfa[indx])+dirwts0[indx]*(eps+cfa[indx-v2]));
crd = cfa[indx+v1]*(dirwts0[indx+v2]+dirwts0[indx])/(dirwts0[indx+v2]*(eps+cfa[indx])+dirwts0[indx]*(eps+cfa[indx+v2]));
crl = cfa[indx-1]*(dirwts1[indx-2]+dirwts1[indx])/(dirwts1[indx-2]*(eps+cfa[indx])+dirwts1[indx]*(eps+cfa[indx-2]));
crr = cfa[indx+1]*(dirwts1[indx+2]+dirwts1[indx])/(dirwts1[indx+2]*(eps+cfa[indx])+dirwts1[indx]*(eps+cfa[indx+2]));
guha=HCLIP(cfa[indx-v1])+xdiv2f(cfa[indx]-cfa[indx-v2]);
gdha=HCLIP(cfa[indx+v1])+xdiv2f(cfa[indx]-cfa[indx+v2]);
glha=HCLIP(cfa[indx-1])+xdiv2f(cfa[indx]-cfa[indx-2]);
grha=HCLIP(cfa[indx+1])+xdiv2f(cfa[indx]-cfa[indx+2]);
if (fabsf(1.0f-cru)<arthresh) {guar=cfa[indx]*cru;} else {guar=guha;}
if (fabsf(1.0f-crd)<arthresh) {gdar=cfa[indx]*crd;} else {gdar=gdha;}
if (fabsf(1.0f-crl)<arthresh) {glar=cfa[indx]*crl;} else {glar=glha;}
if (fabsf(1.0f-crr)<arthresh) {grar=cfa[indx]*crr;} else {grar=grha;}
hwt = dirwts1[indx-1]/(dirwts1[indx-1]+dirwts1[indx+1]);
vwt = dirwts0[indx-v1]/(dirwts0[indx+v1]+dirwts0[indx-v1]);
//interpolated G via adaptive weights of cardinal evaluations
Gintvha = vwt*gdha+(1.0f-vwt)*guha;
Ginthha = hwt*grha+(1.0f-hwt)*glha;
//interpolated color differences
if (fcswitch) {
vcd[indx] = cfa[indx]-(vwt*gdar+(1.0f-vwt)*guar);
hcd[indx] = cfa[indx]-(hwt*grar+(1.0f-hwt)*glar);
vcdalt[indx] = cfa[indx]-Gintvha;
hcdalt[indx] = cfa[indx]-Ginthha;
} else {
//interpolated color differences
vcd[indx] = (vwt*gdar+(1.0f-vwt)*guar)-cfa[indx];
hcd[indx] = (hwt*grar+(1.0f-hwt)*glar)-cfa[indx];
vcdalt[indx] = Gintvha-cfa[indx];
hcdalt[indx] = Ginthha-cfa[indx];
}
fcswitch = !fcswitch;
if (cfa[indx] > clip_pt8 || Gintvha > clip_pt8 || Ginthha > clip_pt8) {
//use HA if highlights are (nearly) clipped
guar=guha; gdar=gdha; glar=glha; grar=grha;
vcd[indx]=vcdalt[indx]; hcd[indx]=hcdalt[indx];
}
//differences of interpolations in opposite directions
dgintv[indx]=min(SQR(guha-gdha),SQR(guar-gdar));
dginth[indx]=min(SQR(glha-grha),SQR(glar-grar));
}
}
#endif
#ifdef __SSE2__
__m128 hcdvarv, vcdvarv;
__m128 hcdaltvarv,vcdaltvarv,hcdv,vcdv,hcdaltv,vcdaltv,sgn3v,Ginthv,Gintvv,hcdoldv,vcdoldv;
__m128 threev = _mm_set1_ps( 3.0f );
__m128 clip_ptv = _mm_set1_ps( clip_pt );
__m128 nsgnv;
vmask hcdmask, vcdmask,tempmask;
if( !(FC(4,4)&1) )
sgnv = _mm_set_ps( 1.0f, -1.0f, 1.0f, -1.0f );
else
sgnv = _mm_set_ps( -1.0f, 1.0f, -1.0f, 1.0f );
sgn3v = threev * sgnv;
for (rr=4; rr<rr1-4; rr++) {
nsgnv = sgnv;
sgnv = -sgnv;
sgn3v = -sgn3v;
for (cc=4,indx=rr*TS+cc,c=FC(rr,cc)&1; cc<cc1-4; cc+=4,indx+=4) {
hcdv = LVF( hcd[indx] );
hcdvarv = threev*(SQRV(LVFU(hcd[indx-2]))+SQRV(hcdv)+SQRV(LVFU(hcd[indx+2])))-SQRV(LVFU(hcd[indx-2])+hcdv+LVFU(hcd[indx+2]));
hcdaltv = LVF( hcdalt[indx] );
hcdaltvarv = threev*(SQRV(LVFU(hcdalt[indx-2]))+SQRV(hcdaltv)+SQRV(LVFU(hcdalt[indx+2])))-SQRV(LVFU(hcdalt[indx-2])+hcdaltv+LVFU(hcdalt[indx+2]));
vcdv = LVF( vcd[indx] );
vcdvarv = threev*(SQRV(LVF(vcd[indx-v2]))+SQRV(vcdv)+SQRV(LVF(vcd[indx+v2])))-SQRV(LVF(vcd[indx-v2])+vcdv+LVF(vcd[indx+v2]));
vcdaltv = LVF( vcdalt[indx] );
vcdaltvarv = threev*(SQRV(LVF(vcdalt[indx-v2]))+SQRV(vcdaltv)+SQRV(LVF(vcdalt[indx+v2])))-SQRV(LVF(vcdalt[indx-v2])+vcdaltv+LVF(vcdalt[indx+v2]));
//choose the smallest variance; this yields a smoother interpolation
hcdv = vself( vmaskf_lt( hcdaltvarv, hcdvarv ), hcdaltv, hcdv);
vcdv = vself( vmaskf_lt( vcdaltvarv, vcdvarv ), vcdaltv, vcdv);
Ginthv = sgnv * hcdv + LVF( cfa[indx] );
temp2v = sgn3v * hcdv;
hwtv = onev + temp2v / ( epsv + Ginthv + LVF( cfa[indx]));
hcdmask = vmaskf_gt( nsgnv * hcdv, ZEROV );
hcdoldv = hcdv;
tempv = nsgnv * (LVF(cfa[indx]) - ULIMV( Ginthv, LVFU(cfa[indx-1]), LVFU(cfa[indx+1]) ));
hcdv = vself( vmaskf_lt( (temp2v), -(LVF(cfa[indx])+Ginthv)), tempv, hwtv*hcdv + (onev - hwtv)*tempv);
hcdv = vself( hcdmask, hcdv, hcdoldv );
hcdv = vself( vmaskf_gt( Ginthv, clip_ptv), tempv, hcdv);
_mm_store_ps( &hcd[indx], hcdv);
Gintvv = sgnv * vcdv + LVF( cfa[indx] );
temp2v = sgn3v * vcdv;
vwtv = onev + temp2v / ( epsv + Gintvv + LVF( cfa[indx]));
vcdmask = vmaskf_gt( nsgnv * vcdv, ZEROV );
vcdoldv = vcdv;
tempv = nsgnv * (LVF(cfa[indx]) - ULIMV( Gintvv, LVF(cfa[indx-v1]), LVF(cfa[indx+v1]) ));
vcdv = vself( vmaskf_lt( (temp2v), -(LVF(cfa[indx])+Gintvv)), tempv, vwtv*vcdv + (onev - vwtv)*tempv);
vcdv = vself( vcdmask, vcdv, vcdoldv );
vcdv = vself( vmaskf_gt( Gintvv, clip_ptv), tempv, vcdv);
_mm_store_ps( &vcd[indx], vcdv);
_mm_storeu_ps(&cddiffsq[indx], SQRV(vcdv-hcdv));
}
}
#else
for (rr=4; rr<rr1-4; rr++) {
//for (cc=4+(FC(rr,2)&1),indx=rr*TS+cc,c=FC(rr,cc); cc<cc1-4; cc+=2,indx+=2) {
for (cc=4,indx=rr*TS+cc,c=FC(rr,cc)&1; cc<cc1-4; cc++,indx++) {
hcdvar =3.0f*(SQR(hcd[indx-2])+SQR(hcd[indx])+SQR(hcd[indx+2]))-SQR(hcd[indx-2]+hcd[indx]+hcd[indx+2]);
hcdaltvar =3.0f*(SQR(hcdalt[indx-2])+SQR(hcdalt[indx])+SQR(hcdalt[indx+2]))-SQR(hcdalt[indx-2]+hcdalt[indx]+hcdalt[indx+2]);
vcdvar =3.0f*(SQR(vcd[indx-v2])+SQR(vcd[indx])+SQR(vcd[indx+v2]))-SQR(vcd[indx-v2]+vcd[indx]+vcd[indx+v2]);
vcdaltvar =3.0f*(SQR(vcdalt[indx-v2])+SQR(vcdalt[indx])+SQR(vcdalt[indx+v2]))-SQR(vcdalt[indx-v2]+vcdalt[indx]+vcdalt[indx+v2]);
//choose the smallest variance; this yields a smoother interpolation
if (hcdaltvar<hcdvar) hcd[indx]=hcdalt[indx];
if (vcdaltvar<vcdvar) vcd[indx]=vcdalt[indx];
//bound the interpolation in regions of high saturation
if (c) {//G site
Ginth = -hcd[indx]+cfa[indx];//R or B
Gintv = -vcd[indx]+cfa[indx];//B or R
if (hcd[indx]>0) {
if (3.0f*hcd[indx] > (Ginth+cfa[indx])) {
hcd[indx]=-ULIM(Ginth,cfa[indx-1],cfa[indx+1])+cfa[indx];
} else {
hwt = 1.0f -3.0f*hcd[indx]/(eps+Ginth+cfa[indx]);
hcd[indx]=hwt*hcd[indx] + (1.0f-hwt)*(-ULIM(Ginth,cfa[indx-1],cfa[indx+1])+cfa[indx]);
}
}
if (vcd[indx]>0) {
if (3.0f*vcd[indx] > (Gintv+cfa[indx])) {
vcd[indx]=-ULIM(Gintv,cfa[indx-v1],cfa[indx+v1])+cfa[indx];
} else {
vwt = 1.0f -3.0f*vcd[indx]/(eps+Gintv+cfa[indx]);
vcd[indx]=vwt*vcd[indx] + (1.0f-vwt)*(-ULIM(Gintv,cfa[indx-v1],cfa[indx+v1])+cfa[indx]);
}
}
if (Ginth > clip_pt) hcd[indx]=-ULIM(Ginth,cfa[indx-1],cfa[indx+1])+cfa[indx];//for RT implementation
if (Gintv > clip_pt) vcd[indx]=-ULIM(Gintv,cfa[indx-v1],cfa[indx+v1])+cfa[indx];
//if (Ginth > pre_mul[c]) hcd[indx]=-ULIM(Ginth,cfa[indx-1],cfa[indx+1])+cfa[indx];//for dcraw implementation
//if (Gintv > pre_mul[c]) vcd[indx]=-ULIM(Gintv,cfa[indx-v1],cfa[indx+v1])+cfa[indx];
} else {//R or B site
Ginth = hcd[indx]+cfa[indx];//interpolated G
Gintv = vcd[indx]+cfa[indx];
if (hcd[indx]<0) {
if (3.0f*hcd[indx] < -(Ginth+cfa[indx])) {
hcd[indx]=ULIM(Ginth,cfa[indx-1],cfa[indx+1])-cfa[indx];
} else {
hwt = 1.0f +3.0f*hcd[indx]/(eps+Ginth+cfa[indx]);
hcd[indx]=hwt*hcd[indx] + (1.0f-hwt)*(ULIM(Ginth,cfa[indx-1],cfa[indx+1])-cfa[indx]);
}
}
if (vcd[indx]<0) {
if (3.0f*vcd[indx] < -(Gintv+cfa[indx])) {
vcd[indx]=ULIM(Gintv,cfa[indx-v1],cfa[indx+v1])-cfa[indx];
} else {
vwt = 1.0f +3.0f*vcd[indx]/(eps+Gintv+cfa[indx]);
vcd[indx]=vwt*vcd[indx] + (1.0f-vwt)*(ULIM(Gintv,cfa[indx-v1],cfa[indx+v1])-cfa[indx]);
}
}
if (Ginth > clip_pt) hcd[indx]=ULIM(Ginth,cfa[indx-1],cfa[indx+1])-cfa[indx];//for RT implementation
if (Gintv > clip_pt) vcd[indx]=ULIM(Gintv,cfa[indx-v1],cfa[indx+v1])-cfa[indx];
//if (Ginth > pre_mul[c]) hcd[indx]=ULIM(Ginth,cfa[indx-1],cfa[indx+1])-cfa[indx];//for dcraw implementation
//if (Gintv > pre_mul[c]) vcd[indx]=ULIM(Gintv,cfa[indx-v1],cfa[indx+v1])-cfa[indx];
cddiffsq[indx] = SQR(vcd[indx]-hcd[indx]);
}
c = !c;
}
}
#endif
#ifdef __SSE2__
__m128 uavev,davev,lavev,ravev,Dgrbvvaruv,Dgrbvvardv,Dgrbhvarlv,Dgrbhvarrv,varwtv,diffwtv,vcdvar1v,hcdvar1v;
__m128 epssqv = _mm_set1_ps( epssq );
vmask decmask;
for (rr=6; rr<rr1-6; rr++) {
for (cc=6+(FC(rr,2)&1),indx=rr*TS+cc; cc<cc1-6; cc+=8,indx+=8) {
//compute color difference variances in cardinal directions
tempv = LC2VFU(vcd[indx]);
uavev = tempv+LC2VFU(vcd[indx-v1])+LC2VFU(vcd[indx-v2])+LC2VFU(vcd[indx-v3]);
davev = tempv+LC2VFU(vcd[indx+v1])+LC2VFU(vcd[indx+v2])+LC2VFU(vcd[indx+v3]);
Dgrbvvaruv = SQRV(tempv-uavev)+SQRV(LC2VFU(vcd[indx-v1])-uavev)+SQRV(LC2VFU(vcd[indx-v2])-uavev)+SQRV(LC2VFU(vcd[indx-v3])-uavev);
Dgrbvvardv = SQRV(tempv-davev)+SQRV(LC2VFU(vcd[indx+v1])-davev)+SQRV(LC2VFU(vcd[indx+v2])-davev)+SQRV(LC2VFU(vcd[indx+v3])-davev);
hwtv = LC2VFU(dirwts1[indx-1])/(LC2VFU(dirwts1[indx-1])+LC2VFU(dirwts1[indx+1]));
vwtv = LC2VFU(dirwts0[indx-v1])/(LC2VFU(dirwts0[indx+v1])+LC2VFU(dirwts0[indx-v1]));
tempv = LC2VFU(hcd[indx]);
lavev = tempv+LC2VFU(hcd[indx-1])+LC2VFU(hcd[indx-2])+LC2VFU(hcd[indx-3]);
ravev = tempv+LC2VFU(hcd[indx+1])+LC2VFU(hcd[indx+2])+LC2VFU(hcd[indx+3]);
Dgrbhvarlv = SQRV(tempv-lavev)+SQRV(LC2VFU(hcd[indx-1])-lavev)+SQRV(LC2VFU(hcd[indx-2])-lavev)+SQRV(LC2VFU(hcd[indx-3])-lavev);
Dgrbhvarrv = SQRV(tempv-ravev)+SQRV(LC2VFU(hcd[indx+1])-ravev)+SQRV(LC2VFU(hcd[indx+2])-ravev)+SQRV(LC2VFU(hcd[indx+3])-ravev);
vcdvarv = epssqv+vwtv*Dgrbvvardv+(onev-vwtv)*Dgrbvvaruv;
hcdvarv = epssqv+hwtv*Dgrbhvarrv+(onev-hwtv)*Dgrbhvarlv;
//compute fluctuations in up/down and left/right interpolations of colors
Dgrbvvaruv = (LC2VFU(dgintv[indx]))+(LC2VFU(dgintv[indx-v1]))+(LC2VFU(dgintv[indx-v2]));
Dgrbvvardv = (LC2VFU(dgintv[indx]))+(LC2VFU(dgintv[indx+v1]))+(LC2VFU(dgintv[indx+v2]));
Dgrbhvarlv = (LC2VFU(dginth[indx]))+(LC2VFU(dginth[indx-1]))+(LC2VFU(dginth[indx-2]));
Dgrbhvarrv = (LC2VFU(dginth[indx]))+(LC2VFU(dginth[indx+1]))+(LC2VFU(dginth[indx+2]));
vcdvar1v = epssqv+vwtv*Dgrbvvardv+(onev-vwtv)*Dgrbvvaruv;
hcdvar1v = epssqv+hwtv*Dgrbhvarrv+(onev-hwtv)*Dgrbhvarlv;
//determine adaptive weights for G interpolation
varwtv=hcdvarv/(vcdvarv+hcdvarv);
diffwtv=hcdvar1v/(vcdvar1v+hcdvar1v);
//if both agree on interpolation direction, choose the one with strongest directional discrimination;
//otherwise, choose the u/d and l/r difference fluctuation weights
decmask = vandm( vmaskf_gt( (zd5v - varwtv) * (zd5v - diffwtv), ZEROV ), vmaskf_lt( vabsf( zd5v - diffwtv), vabsf( zd5v - varwtv) ) );
_mm_storeu_ps( &hvwt[indx>>1], vself( decmask, varwtv, diffwtv));
}
}
#else
for (rr=6; rr<rr1-6; rr++) {
for (cc=6+(FC(rr,2)&1),indx=rr*TS+cc; cc<cc1-6; cc+=2,indx+=2) {
//compute color difference variances in cardinal directions
uave = vcd[indx]+vcd[indx-v1]+vcd[indx-v2]+vcd[indx-v3];
dave = vcd[indx]+vcd[indx+v1]+vcd[indx+v2]+vcd[indx+v3];
lave = hcd[indx]+hcd[indx-1]+hcd[indx-2]+hcd[indx-3];
rave = hcd[indx]+hcd[indx+1]+hcd[indx+2]+hcd[indx+3];
Dgrbvvaru = SQR(vcd[indx]-uave)+SQR(vcd[indx-v1]-uave)+SQR(vcd[indx-v2]-uave)+SQR(vcd[indx-v3]-uave);
Dgrbvvard = SQR(vcd[indx]-dave)+SQR(vcd[indx+v1]-dave)+SQR(vcd[indx+v2]-dave)+SQR(vcd[indx+v3]-dave);
Dgrbhvarl = SQR(hcd[indx]-lave)+SQR(hcd[indx-1]-lave)+SQR(hcd[indx-2]-lave)+SQR(hcd[indx-3]-lave);
Dgrbhvarr = SQR(hcd[indx]-rave)+SQR(hcd[indx+1]-rave)+SQR(hcd[indx+2]-rave)+SQR(hcd[indx+3]-rave);
hwt = dirwts1[indx-1]/(dirwts1[indx-1]+dirwts1[indx+1]);
vwt = dirwts0[indx-v1]/(dirwts0[indx+v1]+dirwts0[indx-v1]);
vcdvar = epssq+vwt*Dgrbvvard+(1.0f-vwt)*Dgrbvvaru;
hcdvar = epssq+hwt*Dgrbhvarr+(1.0f-hwt)*Dgrbhvarl;
//compute fluctuations in up/down and left/right interpolations of colors
Dgrbvvaru = (dgintv[indx])+(dgintv[indx-v1])+(dgintv[indx-v2]);
Dgrbvvard = (dgintv[indx])+(dgintv[indx+v1])+(dgintv[indx+v2]);
Dgrbhvarl = (dginth[indx])+(dginth[indx-1])+(dginth[indx-2]);
Dgrbhvarr = (dginth[indx])+(dginth[indx+1])+(dginth[indx+2]);
vcdvar1 = epssq+vwt*Dgrbvvard+(1.0f-vwt)*Dgrbvvaru;
hcdvar1 = epssq+hwt*Dgrbhvarr+(1.0f-hwt)*Dgrbhvarl;
//determine adaptive weights for G interpolation
varwt=hcdvar/(vcdvar+hcdvar);
diffwt=hcdvar1/(vcdvar1+hcdvar1);
//if both agree on interpolation direction, choose the one with strongest directional discrimination;
//otherwise, choose the u/d and l/r difference fluctuation weights
if ((0.5-varwt)*(0.5-diffwt)>0 && fabsf(0.5-diffwt)<fabsf(0.5-varwt)) {hvwt[indx>>1]=varwt;} else {hvwt[indx>>1]=diffwt;}
//hvwt[indx]=varwt;
}
}
#endif
// %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
// Nyquist test
for (rr=6; rr<rr1-6; rr++)
for (cc=6+(FC(rr,2)&1),indx=rr*TS+cc; cc<cc1-6; cc+=2,indx+=2) {
//nyquist texture test: ask if difference of vcd compared to hcd is larger or smaller than RGGB gradients
nyqtest = (gaussodd[0]*cddiffsq[indx]+
gaussodd[1]*(cddiffsq[(indx-m1)]+cddiffsq[(indx+p1)]+
cddiffsq[(indx-p1)]+cddiffsq[(indx+m1)])+
gaussodd[2]*(cddiffsq[(indx-v2)]+cddiffsq[(indx-2)]+
cddiffsq[(indx+2)]+cddiffsq[(indx+v2)])+
gaussodd[3]*(cddiffsq[(indx-m2)]+cddiffsq[(indx+p2)]+
cddiffsq[(indx-p2)]+cddiffsq[(indx+m2)]));
nyqtest -= nyqthresh*(gaussgrad[0]*(delhvsqsum[indx])+
gaussgrad[1]*(delhvsqsum[indx-v1]+delhvsqsum[indx+1]+
delhvsqsum[indx-1]+delhvsqsum[indx+v1])+
gaussgrad[2]*(delhvsqsum[indx-m1]+delhvsqsum[indx+p1]+
delhvsqsum[indx-p1]+delhvsqsum[indx+m1])+
gaussgrad[3]*(delhvsqsum[indx-v2]+delhvsqsum[indx-2]+
delhvsqsum[indx+2]+delhvsqsum[indx+v2])+
gaussgrad[4]*(delhvsqsum[indx-2*TS-1]+delhvsqsum[indx-2*TS+1]+
delhvsqsum[indx-TS-2]+delhvsqsum[indx-TS+2]+
delhvsqsum[indx+TS-2]+delhvsqsum[indx+TS+2]+
delhvsqsum[indx+2*TS-1]+delhvsqsum[indx+2*TS+1])+
gaussgrad[5]*(delhvsqsum[indx-m2]+delhvsqsum[indx+p2]+
delhvsqsum[indx-p2]+delhvsqsum[indx+m2]));
if (nyqtest>0)
nyquist[indx>>1]=1;//nyquist=1 for nyquist region
}
unsigned int nyquisttemp;
for (rr=8; rr<rr1-8; rr++){
for (cc=8+(FC(rr,2)&1),indx=rr*TS+cc; cc<cc1-8; cc+=2,indx+=2) {
nyquisttemp=(nyquist[(indx-v2)>>1]+nyquist[(indx-m1)>>1]+nyquist[(indx+p1)>>1]+
nyquist[(indx-2)>>1]+nyquist[indx>>1]+nyquist[(indx+2)>>1]+
nyquist[(indx-p1)>>1]+nyquist[(indx+m1)>>1]+nyquist[(indx+v2)>>1]);
//if most of your neighbors are named Nyquist, it's likely that you're one too
if (nyquisttemp>4) nyquist[indx>>1]=1;
//or not
if (nyquisttemp<4) nyquist[indx>>1]=0;
}
}
// end of Nyquist test
// %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
// in areas of Nyquist texture, do area interpolation
for (rr=8; rr<rr1-8; rr++)
for (cc=8+(FC(rr,2)&1),indx=rr*TS+cc; cc<cc1-8; cc+=2,indx+=2) {
if (nyquist[indx>>1]) {
// %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
// area interpolation
sumh=sumv=sumsqh=sumsqv=areawt=0;
for (i=-6; i<7; i+=2)
for (j=-6; j<7; j+=2) {
indx1=(rr+i)*TS+cc+j;
if (nyquist[indx1>>1]) {
sumh += cfa[indx1]-xdiv2f(cfa[indx1-1]+cfa[indx1+1]);
sumv += cfa[indx1]-xdiv2f(cfa[indx1-v1]+cfa[indx1+v1]);
sumsqh += xdiv2f(SQR(cfa[indx1]-cfa[indx1-1])+SQR(cfa[indx1]-cfa[indx1+1]));
sumsqv += xdiv2f(SQR(cfa[indx1]-cfa[indx1-v1])+SQR(cfa[indx1]-cfa[indx1+v1]));
areawt +=1;
}
}
//horizontal and vertical color differences, and adaptive weight
hcdvar=epssq+fabsf(areawt*sumsqh-sumh*sumh);
vcdvar=epssq+fabsf(areawt*sumsqv-sumv*sumv);
hvwt[indx>>1]=hcdvar/(vcdvar+hcdvar);
// end of area interpolation
// %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
}
}
// %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
//populate G at R/B sites
for (rr=8; rr<rr1-8; rr++)
for (cc=8+(FC(rr,2)&1),indx=rr*TS+cc; cc<cc1-8; cc+=2,indx+=2) {
//first ask if one gets more directional discrimination from nearby B/R sites
hvwtalt = xdivf(hvwt[(indx-m1)>>1]+hvwt[(indx+p1)>>1]+hvwt[(indx-p1)>>1]+hvwt[(indx+m1)>>1],2);
// hvwtalt = 0.25*(hvwt[(indx-m1)>>1]+hvwt[(indx+p1)>>1]+hvwt[(indx-p1)>>1]+hvwt[(indx+m1)>>1]);
// vo=fabsf(0.5-hvwt[indx>>1]);
// ve=fabsf(0.5-hvwtalt);
if (fabsf(0.5-hvwt[indx>>1])<fabsf(0.5-hvwtalt)) {hvwt[indx>>1]=hvwtalt;}//a better result was obtained from the neighbors
// if (vo<ve) {hvwt[indx>>1]=hvwtalt;}//a better result was obtained from the neighbors
Dgrb[0][indx>>1] = (hcd[indx]*(1.0f-hvwt[indx>>1]) + vcd[indx]*hvwt[indx>>1]);//evaluate color differences
//if (hvwt[indx]<0.5) Dgrb[indx][0]=hcd[indx];
//if (hvwt[indx]>0.5) Dgrb[indx][0]=vcd[indx];
rgbgreen[indx] = cfa[indx] + Dgrb[0][indx>>1];//evaluate G (finally!)
//local curvature in G (preparation for nyquist refinement step)
if (nyquist[indx>>1]) {
Dgrb2[indx>>1].h = SQR(rgbgreen[indx] - xdiv2f(rgbgreen[indx-1]+rgbgreen[indx+1]));
Dgrb2[indx>>1].v = SQR(rgbgreen[indx] - xdiv2f(rgbgreen[indx-v1]+rgbgreen[indx+v1]));
} else {
Dgrb2[indx>>1].h = Dgrb2[indx>>1].v = 0;
}
}