home *** CD-ROM | disk | FTP | other *** search
- Path: informatik.tu-muenchen.de!fischerj
- From: fischerj@informatik.tu-muenchen.de (Juergen "Rally" Fischer)
- Newsgroups: comp.sys.amiga.programmer
- Subject: Re: Demos on graphiccards
- Date: 2 Feb 1996 17:41:59 GMT
- Organization: Technische Universitaet Muenchen, Germany
- Distribution: world
- Message-ID: <4etid7$m36@sunsystem5.informatik.tu-muenchen.de>
- References: <xB53AMD0asz2@wizzcat.prima.ruhr.de> <4eavqk$da8@sunsystem5.informatik.tu-muenchen.de> <PETERM.96Jan31231828@tui.maths.irl.cri.nz>
- NNTP-Posting-Host: hphalle5.informatik.tu-muenchen.de
- Originator: fischerj@hphalle5.informatik.tu-muenchen.de
-
-
- In article <PETERM.96Jan31231828@tui.maths.irl.cri.nz>, peterm@maths.grace.cri.nz (Peter McGavin) writes:
- |> Organization: Industrial Research Ltd
- |> Lines: 23
- |> Distribution: world
- |> Message-ID: <PETERM.96Jan31231828@tui.maths.irl.cri.nz>
- |> References: <xB53AMD0asz2@wizzcat.prima.ruhr.de>
- |> <4eavqk$da8@sunsystem5.informatik.tu-muenchen.de>
- |> NNTP-Posting-Host: tui.grace.cri.nz
- |> In-reply-to: fischerj@informatik.tu-muenchen.de's message of 26 Jan 1996
- |> 16:34:28 GMT
- |>
- |> fischerj@informatik.tu-muenchen.de (Juergen "Rally" Fischer) writes:
- |> >well, supporting gfx-cards doesn't mean you need to fallback to
- |> >OS-routines for rendering. OS-rendering as option could speed up
- |> >on special hardware though.
- |>
- |> IMHO, it's the other way around. A demo must fall-back to high-level
- |> OS calls if it wants to support _all_ gfx-cards (including unknown and
- |> future cards). If a demo detects a known, particular card or custom
- |> chipset, then it could possibly go faster with an option to go right
- |> to the hardware.
-
- thats what I meant :)
- |>
- |> >|> See Gloom running systemfriendly !!
- |> >
- |> >Yes, that's nice. But it's too slow. Not because of sysfriendly, but
- |> >bacause of suboptimal c2p and suboptimal mapping.
- |>
- |> Feel free to write faster c2p routines than the ones I uploaded to
- |> Aminet game/demo/GloomC2P10.lha. Unfortunately the Gloom Deluxe Demo
- |> c2p interface doesn't seem to support async blitter, which would be
- |> faster on an '020. For CPU-only c2p, I think my routines would be
- |> hard to improve significantly (but I've been proved wrong before).
-
- that's it, not improvable, because gloom limits to cpu-only.
-
- Black magic made the turn from "we-aim-at-A1200-users-only-copperscreen"
- to "wow-we-use-the-OS!-quite-unefficient-cpu-only-c2p".
-
- But that's not the biggest prob, the problem is imho that
- the mapping is just too slow.
-
- For example non shaded floor can run in 16ish cycles inner loop
- (on chipset copper does floorshading). And wall is even faster,
- I guess 20ish cycles including shadingtable (020/030).
-
- I wish there was a clone coming much nearer to those theoretic max values.
-
- |> --
- |> Peter McGavin. (p.mcgavin@irl.cri.nz)
- ------------------------------------------------------------------------
- fischerj@Informatik.TU-Muenchen.DE (Juergen "Rally" Fischer) =:)
-