On Feb 13, 2008 4:28 PM, Greg Ewing <[EMAIL PROTECTED]> wrote: > There's also > > 4. Scale the texture image so that it has power-of-2 sizes > > which may work well enough for photographic images and such > like where you're not too concerned about the appearance of > fine details. > I've heard that approach suggested before, but In my opinion, that approach is either inferior or equivalent in every respect to approach 1 that I mentioned (putting the image in a portion of a power of 2 texture, and draw from a subset of the texture)
The reasons why are: 1. They both take up exactly the same amount of video memory 2. Performance is worse for the scaled one because when drawing only the original image pixels from a subset of the texture the video card only has to lookup and filter those original pixels, but when the image was scaled now the video card has to lookup and filter additional pixels (it would tend to be around 2x the mem lookup) 3. Quality is worse for the scaled one, especially when rendering at the original image size - even if you got a super awesome up scale (like blackman or something) going into the texture, you are still getting a bilinear filter on the rendering 4. Time to load images to the card is (ever so slightly) worse with the scaled one, cause you have to do the scale 5. You have to write/utilize more code in the scaling case For that reason, I would say there are only 3 sensible approaches to the images to power-of-2 texture mapping problem, that I am aware of.
