From: thinman@netcom.com (Technically Sweet)
Subject: Re: TECH: A discussion update for newcomers: distributed worlds
Date: Thu, 23 Jul 92 13:17:55 PDT



On mip-maps: they are successively condensed versions of a pixmap.
Simple version: a coarse-grain version that's an average of (say) four
pixels, and a larger pixmap that has the rest of each pixel.  To build
higher and higher res versions of an image, you add in the successive
deltas.  Since deltas have lower ranges than the full pixel, you can
store deltas in smaller units, say 8 bits for an 24-bit pixel.

This gives a small representation of a coarse version of a pixmap, but
takes more and more work to get a larger version.  Real-time texture
mapping requires a fixed amount of time per pixel.  As a target
surface gets larger the mip-map technique requires an escalating
number of operations per pixel.

I'm against it; you should just sample the original at larger
intervals to get real-time texture mapping.  When the surface is
moving, the inaccuracy of bad sampling will be less noticeable.  When
a surface stops, you can build a better version in background as a 2D
polygon and use that for the surface since you don't have to do any
sampling in the refresh, just a 2d->2d polygon copy.  This is needed
now for doing text.

Lance