home *** CD-ROM | disk | FTP | other *** search
- Path: nntp.teleport.com!sschaem
- From: sschaem@teleport.com (Stephan Schaem)
- Newsgroups: comp.sys.amiga.programmer
- Subject: Re: TMapping again!
- Date: 1 Mar 1996 20:49:11 GMT
- Organization: Teleport - Portland's Public Access (503) 220-1016
- Distribution: world
- Message-ID: <4h7ns7$3n9@maureen.teleport.com>
- References: <38232577@kone.fipnet.fi> <4gl167$r6s@oreig.uji.es> <38232663@kone.fipnet.fi> <4h45p8$ilk@sunsystem5.informatik.tu-muenchen.de>
- NNTP-Posting-Host: julie.teleport.com
- X-Newsreader: TIN [version 1.2 PL2]
-
- Juergen "Rally" Fischer (fischerj@informatik.tu-muenchen.de) wrote:
-
- : In article <38232663@kone.fipnet.fi>, "Jyrki Saarinen" <jsaarinen@kone.fipnet.fi> writes:
- : |>
- : |> > What is faster? Dividing each 16 pixels or using forward differencing?
- : |>
- : |> Using more than one add per pixel? Quadratic or cubic?
- : |> Well, I have some formulas for quadratic but they dont work.
- : |>
- : |> > Also, why is it that nobody (I think) uses z-constant texture mapping?
- : |>
- : |> I guess it is hard to do correctly without any holes between
- : |> polygons.
-
- : But it's possible, my friend did it :) Had to play around a lot with
- : bresenham theory.
- : Well, writing non-horiz in mem should cause no slowdown ?
- : Don't know how much instructions he needed for that...
-
- some demo on PC uses constant Z, the end part of 'escape' from mwb
- uses that.
- The idea for constant Z involve stepping in a 2d steping table. but
- actually that table is 1 dimentional. very similare to using
- precalculated steping for the texture , but you use to for
- the destination buffer VS source. same thing
-
- REPEAT 16
- move.w d1,d4
- move.b d0,d4
- move.l d4,a0
- addx.l d2,d0
- addx.l d3,d1
- adda.w (a2)+,a1 ; for constant Z
- move.b (a0),(a1)
- ENDR
-
- (dont bother reordering the inst, its just an example)
- Humm, 80cycles for 16 pixels (5*16 for the extra add)
- It might be that the potential speed of the methode after all is:
-
- 1) constant Z
- 2) quadratic interpolation with subdivition (full forward diferencing)
- 3) 1/z with subdivition
-
- With constant Z also giving the best visual result... (If handled
- as it should, not ignoring subpixel value.)
-
- Stephan
-