NetNews Offline 2/NetNews Offline Volume 2.iso/news/comp/sys/amiga/programmer/1412

Internet Message Format | 1996-08-05 | 4.3 KB

Path: informatik.tu-muenchen.de!fischerj
From: [e-mail] (Juergen "Rally" Fischer)
Newsgroups: comp.sys.amiga.programmer
Subject: Re: TMapping again!
Date: 18 Jan 1996 19:05:49 GMT
Organization: Technische Universitaet Muenchen, Germany
Distribution: world
Message-ID: <4dm5md$[e-mail]>
References: <4d0ou6$[e-mail]> <Z31Wx*[e-mail]> <4d42di$[e-mail]> <4d5lvi$[e-mail]> <4d6v0t$[e-mail]> <4djj0t$[e-mail]> <4dlo7d$[e-mail]>
NNTP-Posting-Host: hphalle5.informatik.tu-muenchen.de
Originator: [e-mail]


In article <4dlo7d$[e-mail]>, [e-mail] (Philipp Boerker) writes:
|> Organization: Technical University of Berlin, Germany
|> Lines: 71
|> Message-ID: <4dlo7d$[e-mail]>
|> References: <4d0ou6$[e-mail]> <Z31Wx*[e-mail]> <4d42di$[e-mail]> <4d5lvi$[e-mail]> <4d6v0t$[e-mail]> <4djj0t$[e-mail]>
|> NNTP-Posting-Host: hydra.zrz.tu-berlin.de
|> Mime-Version: 1.0
|> Content-Type: text/plain; charset=iso-8859-1
|> Content-Transfer-Encoding: 8bit
|>
|> [e-mail] (Juergen "Rally" Fischer) writes:
|>
|>
|> >|>  AAAAAA..........BBBBBCCCCCCDDDDD
|> >|>
|> >|>  A = x fraction
|> >|>  B = y integer
|> >|>  C = y fraction
|> >|>  D = x integer
|> >|>  . = zero or more precision for x
|> >|>
|> >|>  as you see the example above work...
|>
|> >uhm has anyone actually seen the mapper work ? IMHO the fact that
|> >CCCCCC y fraction also causes an adress, this could be used imho
|> >for smoother mapping!
|>
|> >I guess you normally use 32 equal pixels
|>
|> I think that you rather have the same 32x32 64 times in your
|> 256x256 texture. This will cause the y fraction to be ignored.
|>
|> > but if you'd use 32 values
|> >beeing interpolated between 2 pix of the "real 32x32 map", then it
|> >should give you interpolation for free!
|>
|> hm.
:)

|>
|> >|> : Yes, I agree. Sharon, one of the coders of matrix, is actually a
|> >|> : PC coder but his polygon engine on a 33 MHz 486 is slower than
|> >|> : our (Sharon, Grond, Skyphos all/matrix) engine on a 25 MHz 030!
|> >|> : That's the power of a large register set!
|>
|> >well is you polyengine still faster using 8 planes ? ;)
|>
|> I was refering to 8bit.
|>
|> >he got a non-local-bus vga ? ;)
|>
|> Local bus...

ooh =:o

well hm... polygon is just writing 32bit values to vram.

The 030 will do this at on AGA 7MB/sec. Or do you refer to
A3000+gfx-card ? ;)

average VLBs installed in a 486-33 should have about(?) 15mb/sec speed.
mhm. is the PC version writing bytewise ?

how much fps ? is it mem overwritten multiple times (fastmem render ?).
then fastmem speeds cares.

are you sure the PC version is as optmized?

|>
|> >It depends on the routine if a large register set is needed.
|> >inner loops can live with x86 chipset, but when it comes to
|> >2st outer, the L1 caches have to play the role of a larger
|> >register set.
|>
|> But if you have real large register sets and superscalar processor
|> design you maybe can do 2 rasterlines at once or more likely to frames
|> for i-glasses...
??

|>
|>
|> >my friend always argues that you could see the cache as big register
|> >set on intel
|>
|> Then ask him how he does a
|> "shift left with mask out and insert" rdest,rsource1,rsource2
|> within 1 cycle from datacache...

can 680x0 do that ? x86 operations with datacache as destination need
33 cycles AFAIK. add.l D0,var or in x86 add eax,var

|>
|> >, but I disagree, a mapper loads so much from mem that
|> >your variables in cache are bombed by the texture data
|>
|> Hm, I don't think that you will set so many pixels, that you can
|> fill 8k cache before you execute outerloop a second time...

doesn't care how much pix are read in innerloop, scanning a 64k pic
will destroy cached instructions. don't know how fatal it is.

|>
|> Greets,
|> Phil.
|> ----------------------------
|> grond/matrix
|> [e-mail]
|> ----------------------------
|>
------------------------------------------------------------------------
   [e-mail] (Juergen "Rally" Fischer)   =:)