Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!cs.utexas.edu!uunet!intercon!amanda@intercon.com From: amanda@intercon.com (Amanda Walker) Newsgroups: comp.sys.mac.programmer Subject: Re: How to Cosine: some results Message-ID: <1545@intercon.com> Date: 11 Nov 89 19:27:50 GMT References: <2256@hudson.acc.virginia.edu> Sender: news@intercon.com Reply-To: amanda@intercon.com (Amanda Walker) Lines: 25 Well, you look like you're off to a good start. Actually measuring execution times is something that a lot of people don't do as often as they should :-). Anyway, there are a couple of comments I have: - Even in a tight memory situation, a 4Kb table is not much space, and the quadratic interpolation technique is significantly enough faster than anything else (for this purpose) that I'd go with it. That being said, I'm also a firm believer in using hardware to do what it can. The 68882 is indeed faster than the 68881, and as you noticed, Motorola does good stuff with silicon. The increased accuracy of floating point may actually make up for the speed difference. You also gain automatically when the FPU gets faster, which is nice. Then again, it won't run an a Mac Plus or SE. Depends on the specific application... - Among the reasons that matrix multiplication is so popular in 3D graphics work are: (a) you can concatenate transformations (rotation, translation, scaling, etc.) in any order and still end up with one matrix that you then multiply against your points, (b) if you build a hardware matrix multiplier, having a single task ("multiply this point by this matrix") makes the design *much* simpler, and (c) using homogeneous coordinates is a nifty mathematical trick :-). I wish you luck with your project, -- Amanda Walker