Hi, I’m into medical visualizations from 3D data. When doing precision tests of our system I’ve noticed some irritating effects.
Let’s say that I’ve got a 3D texture with 1x1x5 voxels (to keep it simple…) and the voxel data are 0 0.1 0.2 0.5 1. Doing a nearest neightbour texture lookup at 0.5 0.5 0.8 gives me the last voxel value, since I’m sampling exactly on the texel border. However, to get the value in the previous voxel (i.e. 0.5), I have to change the texture coordinate to something as low as around 0.7983.
Yes, it’s just as bad when using a power of two texture. I have an ATI HD2900xt card, and I’m about to try it on a Nvidia 8800 in a couple of minutes when my collegue gets back.
Ouch, I think we might have found an ATI issue here. I finally got around to try it on a computer with a Geforce 8800, and it worked exactly as you said Nico (and thus, exactly as I want).
My extremely slim test-application gives these results: