I’m not a math wiz but the skew seems pretty cheap to me. You really think it would save that much time not to do it? I would guess that it might take more time to do it without the skew because the contribution calculation is now harder and not the same for every corner. I also have nothing to back this up but for instance in 2D your tidy little circles become more like ellipses for two of the three corners. However if you figure out anything cool I’d be happy to hear about it :-) I use simplex noise for everything.

Hi,

I’ve been working with a project requiring a good noise-function once again. I implemented the basic simplex-noise instead of perlin-noise this time. I’m attracted by a fact that there should be no visible artifacts (The main problem with every perlin-noise i’ve written)… I checked the reference implementation at:http://staffwww.itn.liu.se/\~stegu/simplexnoise/simplexnoise.pdf and used it as a base for my implementation. I removed floating points and static, constant gradients and replaced them with full-random - gradients…

When i was reimplementing the noise-function. After i transformed the original coordinates into “hypercube”-space… i spent some time trying to do the interpolation in this space directly. Instead of transforming the corners back to the real space. I managed to implement the noise with linear interpolation….. The problem is, no matter what i tried. I couldn’t make it look any better (than linear). I have a strong feeling that the whole function could be implemented without backward-transform (skew)… I don’t have any mathematical base to say so though. If this is possible, the simplex-noise implementations would increase significantly in speed.

I was wondering that maybe someone have been fighting with the same issue? It would be enough for me to know if it can, or if it cannot be done.

Thanks, Tuomo