Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!ut-emx!walt.cc.utexas.edu!greg From: greg@walt.cc.utexas.edu (Greg Harp) Newsgroups: comp.sys.amiga Subject: Re: Compression Message-ID: <37922@ut-emx.uucp> Date: 5 Oct 90 06:01:30 GMT References: <3266@orbit.cts.com> <1990Oct3.215038.8863@zorch.SF-Bay.ORG> Sender: news@ut-emx.uucp Reply-To: greg@walt.cc.utexas.edu (Greg Harp) Organization: The University of Texas at Austin, Austin, Texas Lines: 49 In article <1990Oct3.215038.8863@zorch.SF-Bay.ORG> xanthian@zorch.SF-Bay.ORG (Kent Paul Dolan) writes: >koleman@pnet51.orb.mn.org (Kurt Koller) writes: >>Why don't the people that write the LHArc, Zip, etc stuff for the Amiga >>compile a Floating-Point version and include that as well? If this has been >>done, where can I find it? >> >>Kurt "Koleman" Koller - amdahl!bungia!orbit!pnet51!koleman >Maybe I'm being a bit dense, but since I make data compression one of my >specialties, I don't think so: what in the world has "floating point" >to do with data compression? Moreover, if there were some way to convert >the present, in their essence integer based, algorithms to floating point, >why would anyone want to slow them down that much? > >Kent, the man from xanth. > I think that Kurt asks his question because he's looking for a compression program that uses the an FPU if available. From my experience with several of the more popular compression algorithms, I think I can safely say that there's no use for floating point operations in data compression. I can see one obscure, usage-dependant case where the user would be compressing floating-point data, where a delta-Y algorithm might be useful. (A delta-Y algorithm basically works on the assumption that common _differences_ can exist between consecutive numbers. Imagine a x-y graph of a data set, with x being the position in the data and the y being the value. Common patterns can sometimes exist in the differences between the current and previous y value. The patterns can then be compressed using your favorite algorithm. Digitized audio data can be compressed like this, since specific values don't normally repeat, but _differences_ sometimes do.) HOWEVER, since most data compression occurs on files, which (of course) consist of BYTES (be they graphics, text, floating point numbers, code, or whatever) flops are pretty useless. (I'd be rather tickled if someone proved me wrong here by showing me a floating point compression algorithm.) If you want speed in data compression, get a fast hard drive and use LZ. :-) :-) :-) Greg -- Disclaimer: "Who me? Surely you must be mistaken!" _ _ "The lunatic is in the hall. The lunatics are in my hall. AMIGA! //// The paper holds their folded faces to the floor, //// And every day the paperboy brings more." -- Pink Floyd _ _ //// \\\\//// Greg Harp greg@ccwf.cc.utexas.edu \\XX//