aboutsummaryrefslogtreecommitdiffhomepage
path: root/src/effects/SkBlurMask.cpp
blob: ce3c504759764bd43479a091c8ec350ef265fd5f (plain)
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
/*
 * Copyright 2006 The Android Open Source Project
 *
 * Use of this source code is governed by a BSD-style license that can be
 * found in the LICENSE file.
 */


#include "SkBlurMask.h"
#include "SkMath.h"
#include "SkTemplates.h"
#include "SkEndian.h"


// This constant approximates the scaling done in the software path's
// "high quality" mode, in SkBlurMask::Blur() (1 / sqrt(3)).
// IMHO, it actually should be 1:  we blur "less" than we should do
// according to the CSS and canvas specs, simply because Safari does the same.
// Firefox used to do the same too, until 4.0 where they fixed it.  So at some
// point we should probably get rid of these scaling constants and rebaseline
// all the blur tests.
static const SkScalar kBLUR_SIGMA_SCALE = 0.57735f;

SkScalar SkBlurMask::ConvertRadiusToSigma(SkScalar radius) {
    return radius > 0 ? kBLUR_SIGMA_SCALE * radius + 0.5f : 0.0f;
}

SkScalar SkBlurMask::ConvertSigmaToRadius(SkScalar sigma) {
    return sigma > 0.5f ? (sigma - 0.5f) / kBLUR_SIGMA_SCALE : 0.0f;
}

#define UNROLL_SEPARABLE_LOOPS

/**
 * This function performs a box blur in X, of the given radius.  If the
 * "transpose" parameter is true, it will transpose the pixels on write,
 * such that X and Y are swapped. Reads are always performed from contiguous
 * memory in X, for speed. The destination buffer (dst) must be at least
 * (width + leftRadius + rightRadius) * height bytes in size.
 *
 * This is what the inner loop looks like before unrolling, and with the two
 * cases broken out separately (width < diameter, width >= diameter):
 *
 *      if (width < diameter) {
 *          for (int x = 0; x < width; ++x) {
 *              sum += *right++;
 *              *dptr = (sum * scale + half) >> 24;
 *              dptr += dst_x_stride;
 *          }
 *          for (int x = width; x < diameter; ++x) {
 *              *dptr = (sum * scale + half) >> 24;
 *              dptr += dst_x_stride;
 *          }
 *          for (int x = 0; x < width; ++x) {
 *              *dptr = (sum * scale + half) >> 24;
 *              sum -= *left++;
 *              dptr += dst_x_stride;
 *          }
 *      } else {
 *          for (int x = 0; x < diameter; ++x) {
 *              sum += *right++;
 *              *dptr = (sum * scale + half) >> 24;
 *              dptr += dst_x_stride;
 *          }
 *          for (int x = diameter; x < width; ++x) {
 *              sum += *right++;
 *              *dptr = (sum * scale + half) >> 24;
 *              sum -= *left++;
 *              dptr += dst_x_stride;
 *          }
 *          for (int x = 0; x < diameter; ++x) {
 *              *dptr = (sum * scale + half) >> 24;
 *              sum -= *left++;
 *              dptr += dst_x_stride;
 *          }
 *      }
 */
static int boxBlur(const uint8_t* src, int src_y_stride, uint8_t* dst,
                   int leftRadius, int rightRadius, int width, int height,
                   bool transpose)
{
    int diameter = leftRadius + rightRadius;
    int kernelSize = diameter + 1;
    int border = SkMin32(width, diameter);
    uint32_t scale = (1 << 24) / kernelSize;
    int new_width = width + SkMax32(leftRadius, rightRadius) * 2;
    int dst_x_stride = transpose ? height : 1;
    int dst_y_stride = transpose ? 1 : new_width;
    uint32_t half = 1 << 23;
    for (int y = 0; y < height; ++y) {
        uint32_t sum = 0;
        uint8_t* dptr = dst + y * dst_y_stride;
        const uint8_t* right = src + y * src_y_stride;
        const uint8_t* left = right;
        for (int x = 0; x < rightRadius - leftRadius; x++) {
            *dptr = 0;
            dptr += dst_x_stride;
        }
#define LEFT_BORDER_ITER \
            sum += *right++; \
            *dptr = (sum * scale + half) >> 24; \
            dptr += dst_x_stride;

        int x = 0;
#ifdef UNROLL_SEPARABLE_LOOPS
        for (; x < border - 16; x += 16) {
            LEFT_BORDER_ITER
            LEFT_BORDER_ITER
            LEFT_BORDER_ITER
            LEFT_BORDER_ITER
            LEFT_BORDER_ITER
            LEFT_BORDER_ITER
            LEFT_BORDER_ITER
            LEFT_BORDER_ITER
            LEFT_BORDER_ITER
            LEFT_BORDER_ITER
            LEFT_BORDER_ITER
            LEFT_BORDER_ITER
            LEFT_BORDER_ITER
            LEFT_BORDER_ITER
            LEFT_BORDER_ITER
            LEFT_BORDER_ITER
        }
#endif
        for (; x < border; ++x) {
            LEFT_BORDER_ITER
        }
#undef LEFT_BORDER_ITER
#define TRIVIAL_ITER \
            *dptr = (sum * scale + half) >> 24; \
            dptr += dst_x_stride;
        x = width;
#ifdef UNROLL_SEPARABLE_LOOPS
        for (; x < diameter - 16; x += 16) {
            TRIVIAL_ITER
            TRIVIAL_ITER
            TRIVIAL_ITER
            TRIVIAL_ITER
            TRIVIAL_ITER
            TRIVIAL_ITER
            TRIVIAL_ITER
            TRIVIAL_ITER
            TRIVIAL_ITER
            TRIVIAL_ITER
            TRIVIAL_ITER
            TRIVIAL_ITER
            TRIVIAL_ITER
            TRIVIAL_ITER
            TRIVIAL_ITER
            TRIVIAL_ITER
        }
#endif
        for (; x < diameter; ++x) {
            TRIVIAL_ITER
        }
#undef TRIVIAL_ITER
#define CENTER_ITER \
            sum += *right++; \
            *dptr = (sum * scale + half) >> 24; \
            sum -= *left++; \
            dptr += dst_x_stride;

        x = diameter;
#ifdef UNROLL_SEPARABLE_LOOPS
        for (; x < width - 16; x += 16) {
            CENTER_ITER
            CENTER_ITER
            CENTER_ITER
            CENTER_ITER
            CENTER_ITER
            CENTER_ITER
            CENTER_ITER
            CENTER_ITER
            CENTER_ITER
            CENTER_ITER
            CENTER_ITER
            CENTER_ITER
            CENTER_ITER
            CENTER_ITER
            CENTER_ITER
            CENTER_ITER
        }
#endif
        for (; x < width; ++x) {
            CENTER_ITER
        }
#undef CENTER_ITER
#define RIGHT_BORDER_ITER \
            *dptr = (sum * scale + half) >> 24; \
            sum -= *left++; \
            dptr += dst_x_stride;

        x = 0;
#ifdef UNROLL_SEPARABLE_LOOPS
        for (; x < border - 16; x += 16) {
            RIGHT_BORDER_ITER
            RIGHT_BORDER_ITER
            RIGHT_BORDER_ITER
            RIGHT_BORDER_ITER
            RIGHT_BORDER_ITER
            RIGHT_BORDER_ITER
            RIGHT_BORDER_ITER
            RIGHT_BORDER_ITER
            RIGHT_BORDER_ITER
            RIGHT_BORDER_ITER
            RIGHT_BORDER_ITER
            RIGHT_BORDER_ITER
            RIGHT_BORDER_ITER
            RIGHT_BORDER_ITER
            RIGHT_BORDER_ITER
            RIGHT_BORDER_ITER
        }
#endif
        for (; x < border; ++x) {
            RIGHT_BORDER_ITER
        }
#undef RIGHT_BORDER_ITER
        for (int x = 0; x < leftRadius - rightRadius; ++x) {
            *dptr = 0;
            dptr += dst_x_stride;
        }
        SkASSERT(sum == 0);
    }
    return new_width;
}

/**
 * This variant of the box blur handles blurring of non-integer radii.  It
 * keeps two running sums: an outer sum for the rounded-up kernel radius, and
 * an inner sum for the rounded-down kernel radius.  For each pixel, it linearly
 * interpolates between them.  In float this would be:
 *  outer_weight * outer_sum / kernelSize +
 *  (1.0 - outer_weight) * innerSum / (kernelSize - 2)
 *
 * This is what the inner loop looks like before unrolling, and with the two
 * cases broken out separately (width < diameter, width >= diameter):
 *
 *      if (width < diameter) {
 *          for (int x = 0; x < width; x++) {
 *              inner_sum = outer_sum;
 *              outer_sum += *right++;
 *              *dptr = (outer_sum * outer_scale + inner_sum * inner_scale + half) >> 24;
 *              dptr += dst_x_stride;
 *          }
 *          for (int x = width; x < diameter; ++x) {
 *              *dptr = (outer_sum * outer_scale + inner_sum * inner_scale + half) >> 24;
 *              dptr += dst_x_stride;
 *          }
 *          for (int x = 0; x < width; x++) {
 *              inner_sum = outer_sum - *left++;
 *              *dptr = (outer_sum * outer_scale + inner_sum * inner_scale + half) >> 24;
 *              dptr += dst_x_stride;
 *              outer_sum = inner_sum;
 *          }
 *      } else {
 *          for (int x = 0; x < diameter; x++) {
 *              inner_sum = outer_sum;
 *              outer_sum += *right++;
 *              *dptr = (outer_sum * outer_scale + inner_sum * inner_scale + half) >> 24;
 *              dptr += dst_x_stride;
 *          }
 *          for (int x = diameter; x < width; ++x) {
 *              inner_sum = outer_sum - *left;
 *              outer_sum += *right++;
 *              *dptr = (outer_sum * outer_scale + inner_sum * inner_scale + half) >> 24;
 *              dptr += dst_x_stride;
 *              outer_sum -= *left++;
 *          }
 *          for (int x = 0; x < diameter; x++) {
 *              inner_sum = outer_sum - *left++;
 *              *dptr = (outer_sum * outer_scale + inner_sum * inner_scale + half) >> 24;
 *              dptr += dst_x_stride;
 *              outer_sum = inner_sum;
 *          }
 *      }
 *  }
 *  return new_width;
 */

static int boxBlurInterp(const uint8_t* src, int src_y_stride, uint8_t* dst,
                         int radius, int width, int height,
                         bool transpose, uint8_t outer_weight)
{
    int diameter = radius * 2;
    int kernelSize = diameter + 1;
    int border = SkMin32(width, diameter);
    int inner_weight = 255 - outer_weight;
    outer_weight += outer_weight >> 7;
    inner_weight += inner_weight >> 7;
    uint32_t outer_scale = (outer_weight << 16) / kernelSize;
    uint32_t inner_scale = (inner_weight << 16) / (kernelSize - 2);
    uint32_t half = 1 << 23;
    int new_width = width + diameter;
    int dst_x_stride = transpose ? height : 1;
    int dst_y_stride = transpose ? 1 : new_width;
    for (int y = 0; y < height; ++y) {
        uint32_t outer_sum = 0, inner_sum = 0;
        uint8_t* dptr = dst + y * dst_y_stride;
        const uint8_t* right = src + y * src_y_stride;
        const uint8_t* left = right;
        int x = 0;

#define LEFT_BORDER_ITER \
            inner_sum = outer_sum; \
            outer_sum += *right++; \
            *dptr = (outer_sum * outer_scale + inner_sum * inner_scale + half) >> 24; \
            dptr += dst_x_stride;

#ifdef UNROLL_SEPARABLE_LOOPS
        for (;x < border - 16; x += 16) {
            LEFT_BORDER_ITER
            LEFT_BORDER_ITER
            LEFT_BORDER_ITER
            LEFT_BORDER_ITER
            LEFT_BORDER_ITER
            LEFT_BORDER_ITER
            LEFT_BORDER_ITER
            LEFT_BORDER_ITER
            LEFT_BORDER_ITER
            LEFT_BORDER_ITER
            LEFT_BORDER_ITER
            LEFT_BORDER_ITER
            LEFT_BORDER_ITER
            LEFT_BORDER_ITER
            LEFT_BORDER_ITER
            LEFT_BORDER_ITER
        }
#endif

        for (;x < border; ++x) {
            LEFT_BORDER_ITER
        }
#undef LEFT_BORDER_ITER
        for (int x = width; x < diameter; ++x) {
            *dptr = (outer_sum * outer_scale + inner_sum * inner_scale + half) >> 24;
            dptr += dst_x_stride;
        }
        x = diameter;

#define CENTER_ITER \
            inner_sum = outer_sum - *left; \
            outer_sum += *right++; \
            *dptr = (outer_sum * outer_scale + inner_sum * inner_scale + half) >> 24; \
            dptr += dst_x_stride; \
            outer_sum -= *left++;

#ifdef UNROLL_SEPARABLE_LOOPS
        for (; x < width - 16; x += 16) {
            CENTER_ITER
            CENTER_ITER
            CENTER_ITER
            CENTER_ITER
            CENTER_ITER
            CENTER_ITER
            CENTER_ITER
            CENTER_ITER
            CENTER_ITER
            CENTER_ITER
            CENTER_ITER
            CENTER_ITER
            CENTER_ITER
            CENTER_ITER
            CENTER_ITER
            CENTER_ITER
        }
#endif
        for (; x < width; ++x) {
            CENTER_ITER
        }
#undef CENTER_ITER

        #define RIGHT_BORDER_ITER \
            inner_sum = outer_sum - *left++; \
            *dptr = (outer_sum * outer_scale + inner_sum * inner_scale + half) >> 24; \
            dptr += dst_x_stride; \
            outer_sum = inner_sum;

        x = 0;
#ifdef UNROLL_SEPARABLE_LOOPS
        for (; x < border - 16; x += 16) {
            RIGHT_BORDER_ITER
            RIGHT_BORDER_ITER
            RIGHT_BORDER_ITER
            RIGHT_BORDER_ITER
            RIGHT_BORDER_ITER
            RIGHT_BORDER_ITER
            RIGHT_BORDER_ITER
            RIGHT_BORDER_ITER
            RIGHT_BORDER_ITER
            RIGHT_BORDER_ITER
            RIGHT_BORDER_ITER
            RIGHT_BORDER_ITER
            RIGHT_BORDER_ITER
            RIGHT_BORDER_ITER
            RIGHT_BORDER_ITER
            RIGHT_BORDER_ITER
        }
#endif
        for (; x < border; ++x) {
            RIGHT_BORDER_ITER
        }
#undef RIGHT_BORDER_ITER
        SkASSERT(outer_sum == 0 && inner_sum == 0);
    }
    return new_width;
}

static void get_adjusted_radii(SkScalar passRadius, int *loRadius, int *hiRadius)
{
    *loRadius = *hiRadius = SkScalarCeilToInt(passRadius);
    if (SkIntToScalar(*hiRadius) - passRadius > 0.5f) {
        *loRadius = *hiRadius - 1;
    }
}

#include "SkColorPriv.h"

static void merge_src_with_blur(uint8_t dst[], int dstRB,
                                const uint8_t src[], int srcRB,
                                const uint8_t blur[], int blurRB,
                                int sw, int sh) {
    dstRB -= sw;
    srcRB -= sw;
    blurRB -= sw;
    while (--sh >= 0) {
        for (int x = sw - 1; x >= 0; --x) {
            *dst = SkToU8(SkAlphaMul(*blur, SkAlpha255To256(*src)));
            dst += 1;
            src += 1;
            blur += 1;
        }
        dst += dstRB;
        src += srcRB;
        blur += blurRB;
    }
}

static void clamp_with_orig(uint8_t dst[], int dstRowBytes,
                            const uint8_t src[], int srcRowBytes,
                            int sw, int sh,
                            SkBlurStyle style) {
    int x;
    while (--sh >= 0) {
        switch (style) {
        case kSolid_SkBlurStyle:
            for (x = sw - 1; x >= 0; --x) {
                int s = *src;
                int d = *dst;
                *dst = SkToU8(s + d - SkMulDiv255Round(s, d));
                dst += 1;
                src += 1;
            }
            break;
        case kOuter_SkBlurStyle:
            for (x = sw - 1; x >= 0; --x) {
                if (*src) {
                    *dst = SkToU8(SkAlphaMul(*dst, SkAlpha255To256(255 - *src)));
                }
                dst += 1;
                src += 1;
            }
            break;
        default:
            SkDEBUGFAIL("Unexpected blur style here");
            break;
        }
        dst += dstRowBytes - sw;
        src += srcRowBytes - sw;
    }
}

///////////////////////////////////////////////////////////////////////////////

// we use a local function to wrap the class static method to work around
// a bug in gcc98
void SkMask_FreeImage(uint8_t* image);
void SkMask_FreeImage(uint8_t* image) {
    SkMask::FreeImage(image);
}

bool SkBlurMask::BoxBlur(SkMask* dst, const SkMask& src,
                         SkScalar sigma, SkBlurStyle style, SkBlurQuality quality,
                         SkIPoint* margin, bool force_quality) {

    if (src.fFormat != SkMask::kA8_Format) {
        return false;
    }

    // Force high quality off for small radii (performance)
    if (!force_quality && sigma <= SkIntToScalar(2)) {
        quality = kLow_SkBlurQuality;
    }

    SkScalar passRadius;
    if (kHigh_SkBlurQuality == quality) {
        // For the high quality path the 3 pass box blur kernel width is
        // 6*rad+1 while the full Gaussian width is 6*sigma.
        passRadius = sigma - (1/6.0f);
    } else {
        // For the low quality path we only attempt to cover 3*sigma of the
        // Gaussian blur area (1.5*sigma on each side). The single pass box
        // blur's kernel size is 2*rad+1.
        passRadius = 1.5f*sigma - 0.5f;
    }

    // highQuality: use three box blur passes as a cheap way
    // to approximate a Gaussian blur
    int passCount = (kHigh_SkBlurQuality == quality) ? 3 : 1;

    int rx = SkScalarCeilToInt(passRadius);
    int outerWeight = 255 - SkScalarRoundToInt((SkIntToScalar(rx) - passRadius) * 255);

    SkASSERT(rx >= 0);
    SkASSERT((unsigned)outerWeight <= 255);
    if (rx <= 0) {
        return false;
    }

    int ry = rx;    // only do square blur for now

    int padx = passCount * rx;
    int pady = passCount * ry;

    if (margin) {
        margin->set(padx, pady);
    }
    dst->fBounds.set(src.fBounds.fLeft - padx, src.fBounds.fTop - pady,
                     src.fBounds.fRight + padx, src.fBounds.fBottom + pady);

    dst->fRowBytes = dst->fBounds.width();
    dst->fFormat = SkMask::kA8_Format;
    dst->fImage = nullptr;

    if (src.fImage) {
        size_t dstSize = dst->computeImageSize();
        if (0 == dstSize) {
            return false;   // too big to allocate, abort
        }

        int             sw = src.fBounds.width();
        int             sh = src.fBounds.height();
        const uint8_t*  sp = src.fImage;
        uint8_t*        dp = SkMask::AllocImage(dstSize);
        SkAutoTCallVProc<uint8_t, SkMask_FreeImage> autoCall(dp);

        // build the blurry destination
        SkAutoTMalloc<uint8_t>  tmpBuffer(dstSize);
        uint8_t*                tp = tmpBuffer.get();
        int w = sw, h = sh;

        if (outerWeight == 255) {
            int loRadius, hiRadius;
            get_adjusted_radii(passRadius, &loRadius, &hiRadius);
            if (kHigh_SkBlurQuality == quality) {
                // Do three X blurs, with a transpose on the final one.
                w = boxBlur(sp, src.fRowBytes, tp, loRadius, hiRadius, w, h, false);
                w = boxBlur(tp, w,             dp, hiRadius, loRadius, w, h, false);
                w = boxBlur(dp, w,             tp, hiRadius, hiRadius, w, h, true);
                // Do three Y blurs, with a transpose on the final one.
                h = boxBlur(tp, h,             dp, loRadius, hiRadius, h, w, false);
                h = boxBlur(dp, h,             tp, hiRadius, loRadius, h, w, false);
                h = boxBlur(tp, h,             dp, hiRadius, hiRadius, h, w, true);
            } else {
                w = boxBlur(sp, src.fRowBytes, tp, rx, rx, w, h, true);
                h = boxBlur(tp, h,             dp, ry, ry, h, w, true);
            }
        } else {
            if (kHigh_SkBlurQuality == quality) {
                // Do three X blurs, with a transpose on the final one.
                w = boxBlurInterp(sp, src.fRowBytes, tp, rx, w, h, false, outerWeight);
                w = boxBlurInterp(tp, w,             dp, rx, w, h, false, outerWeight);
                w = boxBlurInterp(dp, w,             tp, rx, w, h, true, outerWeight);
                // Do three Y blurs, with a transpose on the final one.
                h = boxBlurInterp(tp, h,             dp, ry, h, w, false, outerWeight);
                h = boxBlurInterp(dp, h,             tp, ry, h, w, false, outerWeight);
                h = boxBlurInterp(tp, h,             dp, ry, h, w, true, outerWeight);
            } else {
                w = boxBlurInterp(sp, src.fRowBytes, tp, rx, w, h, true, outerWeight);
                h = boxBlurInterp(tp, h,             dp, ry, h, w, true, outerWeight);
            }
        }

        dst->fImage = dp;
        // if need be, alloc the "real" dst (same size as src) and copy/merge
        // the blur into it (applying the src)
        if (style == kInner_SkBlurStyle) {
            // now we allocate the "real" dst, mirror the size of src
            size_t srcSize = src.computeImageSize();
            if (0 == srcSize) {
                return false;   // too big to allocate, abort
            }
            dst->fImage = SkMask::AllocImage(srcSize);
            merge_src_with_blur(dst->fImage, src.fRowBytes,
                                sp, src.fRowBytes,
                                dp + passCount * (rx + ry * dst->fRowBytes),
                                dst->fRowBytes, sw, sh);
            SkMask::FreeImage(dp);
        } else if (style != kNormal_SkBlurStyle) {
            clamp_with_orig(dp + passCount * (rx + ry * dst->fRowBytes),
                            dst->fRowBytes, sp, src.fRowBytes, sw, sh, style);
        }
        (void)autoCall.release();
    }

    if (style == kInner_SkBlurStyle) {
        dst->fBounds = src.fBounds; // restore trimmed bounds
        dst->fRowBytes = src.fRowBytes;
    }

    return true;
}

/* Convolving a box with itself three times results in a piecewise
   quadratic function:

   0                              x <= -1.5
   9/8 + 3/2 x + 1/2 x^2   -1.5 < x <= -.5
   3/4 - x^2                -.5 < x <= .5
   9/8 - 3/2 x + 1/2 x^2    0.5 < x <= 1.5
   0                        1.5 < x

   Mathematica:

   g[x_] := Piecewise [ {
     {9/8 + 3/2 x + 1/2 x^2 ,  -1.5 < x <= -.5},
     {3/4 - x^2             ,   -.5 < x <= .5},
     {9/8 - 3/2 x + 1/2 x^2 ,   0.5 < x <= 1.5}
   }, 0]

   To get the profile curve of the blurred step function at the rectangle
   edge, we evaluate the indefinite integral, which is piecewise cubic:

   0                                        x <= -1.5
   9/16 + 9/8 x + 3/4 x^2 + 1/6 x^3   -1.5 < x <= -0.5
   1/2 + 3/4 x - 1/3 x^3              -.5 < x <= .5
   7/16 + 9/8 x - 3/4 x^2 + 1/6 x^3     .5 < x <= 1.5
   1                                  1.5 < x

   in Mathematica code:

   gi[x_] := Piecewise[ {
     { 0 , x <= -1.5 },
     { 9/16 + 9/8 x + 3/4 x^2 + 1/6 x^3, -1.5 < x <= -0.5 },
     { 1/2 + 3/4 x - 1/3 x^3          ,  -.5 < x <= .5},
     { 7/16 + 9/8 x - 3/4 x^2 + 1/6 x^3,   .5 < x <= 1.5}
   },1]
*/

static float gaussianIntegral(float x) {
    if (x > 1.5f) {
        return 0.0f;
    }
    if (x < -1.5f) {
        return 1.0f;
    }

    float x2 = x*x;
    float x3 = x2*x;

    if ( x > 0.5f ) {
        return 0.5625f - (x3 / 6.0f - 3.0f * x2 * 0.25f + 1.125f * x);
    }
    if ( x > -0.5f ) {
        return 0.5f - (0.75f * x - x3 / 3.0f);
    }
    return 0.4375f + (-x3 / 6.0f - 3.0f * x2 * 0.25f - 1.125f * x);
}

/*  ComputeBlurProfile allocates and fills in an array of floating
    point values between 0 and 255 for the profile signature of
    a blurred half-plane with the given blur radius.  Since we're
    going to be doing screened multiplications (i.e., 1 - (1-x)(1-y))
    all the time, we actually fill in the profile pre-inverted
    (already done 255-x).

    It's the responsibility of the caller to delete the
    memory returned in profile_out.
*/

uint8_t* SkBlurMask::ComputeBlurProfile(SkScalar sigma) {
    int size = SkScalarCeilToInt(6*sigma);

    int center = size >> 1;
    uint8_t* profile = new uint8_t[size];

    float invr = 1.f/(2*sigma);

    profile[0] = 255;
    for (int x = 1 ; x < size ; ++x) {
        float scaled_x = (center - x - .5f) * invr;
        float gi = gaussianIntegral(scaled_x);
        profile[x] = 255 - (uint8_t) (255.f * gi);
    }

    return profile;
}

// TODO MAYBE: Maintain a profile cache to avoid recomputing this for
// commonly used radii.  Consider baking some of the most common blur radii
// directly in as static data?

// Implementation adapted from Michael Herf's approach:
// http://stereopsis.com/shadowrect/

uint8_t SkBlurMask::ProfileLookup(const uint8_t *profile, int loc, int blurred_width, int sharp_width) {
    int dx = SkAbs32(((loc << 1) + 1) - blurred_width) - sharp_width; // how far are we from the original edge?
    int ox = dx >> 1;
    if (ox < 0) {
        ox = 0;
    }

    return profile[ox];
}

void SkBlurMask::ComputeBlurredScanline(uint8_t *pixels, const uint8_t *profile,
                                        unsigned int width, SkScalar sigma) {

    unsigned int profile_size = SkScalarCeilToInt(6*sigma);
    SkAutoTMalloc<uint8_t> horizontalScanline(width);

    unsigned int sw = width - profile_size;
    // nearest odd number less than the profile size represents the center
    // of the (2x scaled) profile
    int center = ( profile_size & ~1 ) - 1;

    int w = sw - center;

    for (unsigned int x = 0 ; x < width ; ++x) {
       if (profile_size <= sw) {
           pixels[x] = ProfileLookup(profile, x, width, w);
       } else {
           float span = float(sw)/(2*sigma);
           float giX = 1.5f - (x+.5f)/(2*sigma);
           pixels[x] = (uint8_t) (255 * (gaussianIntegral(giX) - gaussianIntegral(giX + span)));
       }
    }
}

bool SkBlurMask::BlurRect(SkScalar sigma, SkMask *dst,
                          const SkRect &src, SkBlurStyle style,
                          SkIPoint *margin, SkMask::CreateMode createMode) {
    int profile_size = SkScalarCeilToInt(6*sigma);

    int pad = profile_size/2;
    if (margin) {
        margin->set( pad, pad );
    }

    dst->fBounds.set(SkScalarRoundToInt(src.fLeft - pad),
                     SkScalarRoundToInt(src.fTop - pad),
                     SkScalarRoundToInt(src.fRight + pad),
                     SkScalarRoundToInt(src.fBottom + pad));

    dst->fRowBytes = dst->fBounds.width();
    dst->fFormat = SkMask::kA8_Format;
    dst->fImage = nullptr;

    int             sw = SkScalarFloorToInt(src.width());
    int             sh = SkScalarFloorToInt(src.height());

    if (createMode == SkMask::kJustComputeBounds_CreateMode) {
        if (style == kInner_SkBlurStyle) {
            dst->fBounds.set(SkScalarRoundToInt(src.fLeft),
                             SkScalarRoundToInt(src.fTop),
                             SkScalarRoundToInt(src.fRight),
                             SkScalarRoundToInt(src.fBottom)); // restore trimmed bounds
            dst->fRowBytes = sw;
        }
        return true;
    }

    std::unique_ptr<uint8_t[]> profile(ComputeBlurProfile(sigma));

    size_t dstSize = dst->computeImageSize();
    if (0 == dstSize) {
        return false;   // too big to allocate, abort
    }

    uint8_t*        dp = SkMask::AllocImage(dstSize);

    dst->fImage = dp;

    int dstHeight = dst->fBounds.height();
    int dstWidth = dst->fBounds.width();

    uint8_t *outptr = dp;

    SkAutoTMalloc<uint8_t> horizontalScanline(dstWidth);
    SkAutoTMalloc<uint8_t> verticalScanline(dstHeight);

    ComputeBlurredScanline(horizontalScanline, profile.get(), dstWidth, sigma);
    ComputeBlurredScanline(verticalScanline, profile.get(), dstHeight, sigma);

    for (int y = 0 ; y < dstHeight ; ++y) {
        for (int x = 0 ; x < dstWidth ; x++) {
            unsigned int maskval = SkMulDiv255Round(horizontalScanline[x], verticalScanline[y]);
            *(outptr++) = maskval;
        }
    }

    if (style == kInner_SkBlurStyle) {
        // now we allocate the "real" dst, mirror the size of src
        size_t srcSize = (size_t)(src.width() * src.height());
        if (0 == srcSize) {
            return false;   // too big to allocate, abort
        }
        dst->fImage = SkMask::AllocImage(srcSize);
        for (int y = 0 ; y < sh ; y++) {
            uint8_t *blur_scanline = dp + (y+pad)*dstWidth + pad;
            uint8_t *inner_scanline = dst->fImage + y*sw;
            memcpy(inner_scanline, blur_scanline, sw);
        }
        SkMask::FreeImage(dp);

        dst->fBounds.set(SkScalarRoundToInt(src.fLeft),
                         SkScalarRoundToInt(src.fTop),
                         SkScalarRoundToInt(src.fRight),
                         SkScalarRoundToInt(src.fBottom)); // restore trimmed bounds
        dst->fRowBytes = sw;

    } else if (style == kOuter_SkBlurStyle) {
        for (int y = pad ; y < dstHeight-pad ; y++) {
            uint8_t *dst_scanline = dp + y*dstWidth + pad;
            memset(dst_scanline, 0, sw);
        }
    } else if (style == kSolid_SkBlurStyle) {
        for (int y = pad ; y < dstHeight-pad ; y++) {
            uint8_t *dst_scanline = dp + y*dstWidth + pad;
            memset(dst_scanline, 0xff, sw);
        }
    }
    // normal and solid styles are the same for analytic rect blurs, so don't
    // need to handle solid specially.

    return true;
}

bool SkBlurMask::BlurRRect(SkScalar sigma, SkMask *dst,
                           const SkRRect &src, SkBlurStyle style,
                           SkIPoint *margin, SkMask::CreateMode createMode) {
    // Temporary for now -- always fail, should cause caller to fall back
    // to old path.  Plumbing just to land API and parallelize effort.

    return false;
}

// The "simple" blur is a direct implementation of separable convolution with a discrete
// gaussian kernel.  It's "ground truth" in a sense; too slow to be used, but very
// useful for correctness comparisons.

bool SkBlurMask::BlurGroundTruth(SkScalar sigma, SkMask* dst, const SkMask& src,
                                 SkBlurStyle style, SkIPoint* margin) {

    if (src.fFormat != SkMask::kA8_Format) {
        return false;
    }

    float variance = sigma * sigma;

    int windowSize = SkScalarCeilToInt(sigma*6);
    // round window size up to nearest odd number
    windowSize |= 1;

    SkAutoTMalloc<float> gaussWindow(windowSize);

    int halfWindow = windowSize >> 1;

    gaussWindow[halfWindow] = 1;

    float windowSum = 1;
    for (int x = 1 ; x <= halfWindow ; ++x) {
        float gaussian = expf(-x*x / (2*variance));
        gaussWindow[halfWindow + x] = gaussWindow[halfWindow-x] = gaussian;
        windowSum += 2*gaussian;
    }

    // leave the filter un-normalized for now; we will divide by the normalization
    // sum later;

    int pad = halfWindow;
    if (margin) {
        margin->set( pad, pad );
    }

    dst->fBounds = src.fBounds;
    dst->fBounds.outset(pad, pad);

    dst->fRowBytes = dst->fBounds.width();
    dst->fFormat = SkMask::kA8_Format;
    dst->fImage = nullptr;

    if (src.fImage) {

        size_t dstSize = dst->computeImageSize();
        if (0 == dstSize) {
            return false;   // too big to allocate, abort
        }

        int             srcWidth = src.fBounds.width();
        int             srcHeight = src.fBounds.height();
        int             dstWidth = dst->fBounds.width();

        const uint8_t*  srcPixels = src.fImage;
        uint8_t*        dstPixels = SkMask::AllocImage(dstSize);
        SkAutoTCallVProc<uint8_t, SkMask_FreeImage> autoCall(dstPixels);

        // do the actual blur.  First, make a padded copy of the source.
        // use double pad so we never have to check if we're outside anything

        int padWidth = srcWidth + 4*pad;
        int padHeight = srcHeight;
        int padSize = padWidth * padHeight;

        SkAutoTMalloc<uint8_t> padPixels(padSize);
        memset(padPixels, 0, padSize);

        for (int y = 0 ; y < srcHeight; ++y) {
            uint8_t* padptr = padPixels + y * padWidth + 2*pad;
            const uint8_t* srcptr = srcPixels + y * srcWidth;
            memcpy(padptr, srcptr, srcWidth);
        }

        // blur in X, transposing the result into a temporary floating point buffer.
        // also double-pad the intermediate result so that the second blur doesn't
        // have to do extra conditionals.

        int tmpWidth = padHeight + 4*pad;
        int tmpHeight = padWidth - 2*pad;
        int tmpSize = tmpWidth * tmpHeight;

        SkAutoTMalloc<float> tmpImage(tmpSize);
        memset(tmpImage, 0, tmpSize*sizeof(tmpImage[0]));

        for (int y = 0 ; y < padHeight ; ++y) {
            uint8_t *srcScanline = padPixels + y*padWidth;
            for (int x = pad ; x < padWidth - pad ; ++x) {
                float *outPixel = tmpImage + (x-pad)*tmpWidth + y + 2*pad; // transposed output
                uint8_t *windowCenter = srcScanline + x;
                for (int i = -pad ; i <= pad ; ++i) {
                    *outPixel += gaussWindow[pad+i]*windowCenter[i];
                }
                *outPixel /= windowSum;
            }
        }

        // blur in Y; now filling in the actual desired destination.  We have to do
        // the transpose again; these transposes guarantee that we read memory in
        // linear order.

        for (int y = 0 ; y < tmpHeight ; ++y) {
            float *srcScanline = tmpImage + y*tmpWidth;
            for (int x = pad ; x < tmpWidth - pad ; ++x) {
                float *windowCenter = srcScanline + x;
                float finalValue = 0;
                for (int i = -pad ; i <= pad ; ++i) {
                    finalValue += gaussWindow[pad+i]*windowCenter[i];
                }
                finalValue /= windowSum;
                uint8_t *outPixel = dstPixels + (x-pad)*dstWidth + y; // transposed output
                int integerPixel = int(finalValue + 0.5f);
                *outPixel = SkClampMax( SkClampPos(integerPixel), 255 );
            }
        }

        dst->fImage = dstPixels;
        // if need be, alloc the "real" dst (same size as src) and copy/merge
        // the blur into it (applying the src)
        if (style == kInner_SkBlurStyle) {
            // now we allocate the "real" dst, mirror the size of src
            size_t srcSize = src.computeImageSize();
            if (0 == srcSize) {
                return false;   // too big to allocate, abort
            }
            dst->fImage = SkMask::AllocImage(srcSize);
            merge_src_with_blur(dst->fImage, src.fRowBytes,
                srcPixels, src.fRowBytes,
                dstPixels + pad*dst->fRowBytes + pad,
                dst->fRowBytes, srcWidth, srcHeight);
            SkMask::FreeImage(dstPixels);
        } else if (style != kNormal_SkBlurStyle) {
            clamp_with_orig(dstPixels + pad*dst->fRowBytes + pad,
                dst->fRowBytes, srcPixels, src.fRowBytes, srcWidth, srcHeight, style);
        }
        (void)autoCall.release();
    }

    if (style == kInner_SkBlurStyle) {
        dst->fBounds = src.fBounds; // restore trimmed bounds
        dst->fRowBytes = src.fRowBytes;
    }

    return true;
}