home *** CD-ROM | disk | FTP | other *** search
- Path: hydra.zrz.TU-Berlin.DE!rawneiha
- From: rawneiha@hydra.zrz.TU-Berlin.DE (Philipp Boerker)
- Newsgroups: comp.sys.amiga.programmer
- Subject: Re: Texturemapping speedup
- Date: 4 Jan 1996 16:24:01 GMT
- Organization: Technical University Berlin, Germany
- Message-ID: <4cguv1$br2@brachio.zrz.TU-Berlin.DE>
- References: <john.hendrikx.40k4@grafix.xs4all.nl> <Rr3Px*AKe@yaps.rhein.de> <4bc7i9$5r7@maureen.teleport.com> <4be5k3$cli@brachio.zrz.TU-Berlin.DE> <4bf21s$egp@sunsystem5.informatik.tu-muenchen.de>
- NNTP-Posting-Host: hydra.zrz.tu-berlin.de
-
- fischerj@informatik.tu-muenchen.de (Juergen "Rally" Fischer) writes:
-
-
- >In article <4be5k3$cli@brachio.zrz.TU-Berlin.DE>, rawneiha@hercules.zrz.TU-Berlin.DE (Philipp Boerker) writes:
-
- >|> > I really have to desagree... flushing the inst cache is not a nono,
- >|> > is defenetly not expansive at all on a 020/030... BTW the loop we are
- >|> > talking about unrolled for a 230 might take >200 byte of the cache,
- >|> > A flush with have near ZERO impact on this loop. A texture map loop
- >|> > is not minimal.
- >|>
- >|> > Stephan
- >|> Again: you will have to clear datacaches as well on 040+!
-
- >why ? writeback thingy ?
-
- Yes. Stephan pointed out that one can disable caching for the modification
- and afterwards flush Inst Cache. Well, I never thought of that, still I won't
- start using SM (sadomaso) code, because I want my code to run on future
- cpus (emulations) as well.
-
-
- >|> And: talking about optimizations I might add that my polygon mapping
- >|> loop fits into the 256 bytes !
-
- >including 1st outer ? 2nd outer ?
-
- Input:
- Pointer to list of objects,
- pointer to chunkybuffer.
-
- Output:
- linearly mapped polygons.
-
- Performance:
- Estimated 15000 polygons per second on a 50 MHz 030, and some parts aren't
- optimized yet (C2P, rotation-projection)...
-
- Greets,
- Phil.
- ----------------------------
- grond/matrix
- rawneiha@sp.zrz.tu-berlin.de
- ----------------------------
-