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: TMapping again!
- Date: 5 Mar 1996 13:53:03 GMT
- Organization: Technische Universitaet Muenchen, Germany
- Distribution: world
- Message-ID: <4hhgvv$50g@sunsystem5.informatik.tu-muenchen.de>
- References: <38232567@kone.fipnet.fi> <4h45g4$ije@sunsystem5.informatik.tu-muenchen.de> <38232795@kone.fipnet.fi>
- NNTP-Posting-Host: hphalle6g.informatik.tu-muenchen.de
- Originator: fischerj@hphalle6g.informatik.tu-muenchen.de
-
-
- In article <38232795@kone.fipnet.fi>, "Jyrki Saarinen" <jsaarinen@kone.fipnet.fi> writes:
- |>
- |> > |> Hmm. Not nice in a game for example where several textures
- |> > |> are needed. But a nice idea anyway!
- |> >
- |> > nice ? Cooooooool!
- |> > Cool idea to pack the shading value into the texture adress!
- |>
- |> Yeah but problematic when we want several textures.
- well, 8 times more mem consumption. or 4 times.
-
- needing more mem but getting speed.
- at least worth an option, as not much to code!
-
- |>
- |> > add.l a2,d2 0...00BBBbbbbbbbbbbbbbbbb
- |> > add.l a3,d1 x...xxxxxYYYYYYYYyyyyyyyy
- |> > addx.l d3,d0 0...0000000000000XXXXXXXX
- |> > move.l d2,d6
- |> > move.w d1,d6
- |> > move.b d0,d6
- |> > move.b 0(a0,d6.l),(a1)+
- |> >
- |> > 20 cycles gouraud mapping :)
- |>
- |> Isnt that 22?
-
- oki, let's assume (An,dn.l),(an)+ is 9 cycles. 8 is to optimistic.
-
- would be 21 :)
-
- |>
- |> Why not:
- |>
- |> add.l a2,d2 0...00BBBbbbbbbbbbbbbbbbb
- |> add.l a3,d1 x...xxxxxYYYYYYYYyyyyyyyy
- |> addx.l d3,d0 0...0000000000000XXXXXXXX
- |> move.l d2,d6
- hi word contains ...0BBB
-
- |> move.w d1,d6
- |> move.b d0,d6
- |> move.l d6,a0
- hi word contains ...0BBB
-
- |> move.b (a0),(a1)+
- hi word contains ...0BBB
- You wanna read textures from chipmem ? ;->
-
- |>
- |> 21 cycles.. ;)
- mhm a (an),(an)+ is 6 cycles, so it would have been 20 :)
-
- anyway imho very fast! A game that would fit into a std A1200 could
- run with precalced textures on A1201.
-
- many people tell you need a turbo anyway, but most turbos
- got 4mb mem, so why sparing mem ? ;)
-
- |>
- |> > 8 shadingsteps are 512k, not too much even for a A1200 game imho
- |> > (well, people tell me they aim for higher machines anyway,
- |> > so the mem consumption can't be considered so bad ;)
- |>
- |> 8 intensity levels..? ;) And only one texture.. Maybe
- |> good for a 40k intro but not for a game.
-
- no, 4 textures if you use 128x128. 16 if you use 64x64!
- 64 if you use 2 mb mem. lots of turbos (which you seem to require
- anyway) got this. 50% mapping speedup!
-
- |>
- |> > I would really think about using this routine, it's really LOT quicker.
- |> > You can put multiple textures into a given 64k range btw. but those
- |> > can't be "repeart-mapped" uh "multiplexed" ? how you call it if x gets
- |> >รก256 and then _plop_, 0 again on the left side ? Is this called bump-mapping
- |>
- |> Repeating?
- uhm :) yes :) how to explain :\
- if you scan over the border of a 256x256 texture, you get into it
- again due to the 256x256 adressing. holding multiple textures in
- a 256x256 area will result in getting into a different (wrong) texture
- when leaving texture border (in source), so no "repeating" :)
-
- |>
- |> > Isn't it enough that average ami got less cpu than clones ?
- |> > why do slower algos than known ?
- |> > You can live with it as you got that 040/60, but average ami... ;)
- |>
- |> The aim is about 030/50 where our project would run nicely in
- |> 320x128, that is about 20fps.
-
- yep and on 030 the non-table-shader is lot faaaster :)
- would maybe give 320x160 1x1 in 20fps :)
-
- I'd like to see it as option.
-
- |>
- |> -- _
- |> a Stellar programmer _ //
- |> "Amiga - back for the future" \X/
- ------------------------------------------------------------------------
- fischerj@Informatik.TU-Muenchen.DE (Juergen "Rally" Fischer) =:)
-
-