Fix minor wrong assumption and bug in BlurTests
- It was assumed that the pixels `blurRadius` distance away from the
blur epicenter are not affected. In a correct Gaussian blur, there
will be some minor changes to these pixels too. Double the radius
to get to 4 sigmas which should not have any visible distortion on
an 8-bit display.
- The comparison for "affected" pixels was covering a horizontal region
of `(cx - blurRadius, cx + blurRadius]` with a step of `stepSize`, ie.
half-open interval, because of a less-than-or-equal comparison. Fix
it to a fully-open interval.
Test: atest BlurTests # (with and without b/185365391 fixes patched in)
Bug: 193861404
Bug: 185365391
Change-Id: Ide9a3fc5ee95ca73d9f2e63c349c5ba758fda049
1 file changed
