NetNews Offline 2/NetNews Offline Volume 2.iso/news/comp/sys/amiga/programmer/4462

Internet Message Format | 1996-08-05 | 10.9 KB

Path: informatik.tu-muenchen.de!fischerj
From: [e-mail] (Juergen "Rally" Fischer)
Newsgroups: comp.sys.amiga.programmer
Subject: Re: Amiga doesn`t need Pl
Date: 25 Feb 1996 01:05:56 GMT
Organization: Technische Universitaet Muenchen, Germany
Distribution: world
Message-ID: <4goclk$[e-mail]>
References: <[e-mail]>
NNTP-Posting-Host: hphalle5.informatik.tu-muenchen.de
X-Newsreader: TIN [version 1.2 PL2]

John Hendrikx ([e-mail]) wrote:
: In a message of 20 Feb 96 Juergen "rally" Fischer wrote to All:

:  >> > JrF> But when it comes to games, I say: AGA is still very usable today

:  >> >How can you say that?  What do you mean 'games'?  I think your definiti

:  JrF> DOOM. DESCENT. On a A1200 with megaturbo cpu, it'll run not much slower
:  JrF> like on VLB + megaturbo cpu. Just because those games still need some
:  JrF> frames even on a megaturbo.

: You're wrong.  On A1200 with mega-turbo you spend a *FIXED* amount of time
really ? what is wrong ?

: doing 'copying' (C2P).  No matter how fast your CPU is this time is always
: there, so if you use a 320x240 screen it will never go faster than a fixed FPS
: rate even if your mapping stuff takes 0 cycles.  This gets far worse for
: 640x480 screens (which is quite doable on a fast clone).

yes. so what ? lets assume a 320x160 screen running in 3 frames on a
A1200 turbo. 16.6 fps that is. The copy part is about half a frame
(assuming c2p is at about 6mb/sec. we're talking about fast cpus).

Now let's see what happens if copy is 4 times faster: 2.625 frames. 19fps.

wow. 19fps with a card at about 25mb/sec, 16.6 fps with AGA. so what ???

While 19fps IS smother, 16.6 fps is really not the "can't do DOOM"
everyone is talking about! think about it.

This numeric example should make things clear.

Yes, if your cpu can do DOOM in 1 frame, AGA IS a brake.
If your cpu rather does the speed of the given example, or even slower,
then AGA is NOT "the big doom problem".

: The 'copying' loop simply doesn't exist on the clones, just paste it wherever
well, doom runs well on my friends vlb, the 16bit vlb doing lot below
25mb/sec. See my example, makes things clear.

: You're thinking the wrong way, with megacpu you are actually likely to use LESS
: frames, so the more important AGA bandwidth becomes.  Or maybe you didn't

yes. but megacpus are used to do more than doom today. doom is 486.
a todays 1230 or 1240 should be able to do same (wasn't doom busy
waiting for vblank bit? I think so!)

megacpu is rather descent or above, those canyon-racing-games. back
at 3 frame animations again. AGA gets the big problem if you use the
megacpu for rotzoom...

: notice that with TextDemo the time needed for C2P goes UP relatively on faster
: CPU's?
I guess the c2p doesn't run the possible 6mb/sec on A1200 ;)
You must admit the problem is far less when using free c2p.
Yes, it will go up for faster cpus.

:  JrF> You rely on _programming crap_ in combination with low cpu.
: No, a game which is unplayable because it uses a fuzzy 2x2 160x120 display (or
: less!) and jerks like hell is CRAP.
fuzzy 2x2 due to low cpu, jerk due to low cpu. not AGA.

As I showed, 3 frame animations on A1200 are not that worse if running
AGA instead on card. So if your game jerks, i.e. rather 6 frames,
it can't be AGAs fault _by definition_. see the values.

: There are limits to what a programmer will do to speed up his game.  Tricks
: like Blitter assistance, BlitterScreen, Chunky Copper and so on all are
: incredibly hard to implement or distort the display or limit your game in some
: way:
aaah, that's what you refer to. not 6mb/sec. 3mb/sec. or maybe 1mb/sec? :->>
No wonder you tell it's not possible.

:  Blitter assistance: Requires non-interleaved bitmaps, impossible to
that's no problem.
:                      C2P a smaller part of the whole screen
that's no problem concering games like doom.

:                      on factors like CPU, fast-memory, etcetera...
The disadvantage is that you got to code more, right.
no technical argument.

:                      Try to get it to work on Intuition-screen and make
:                      your DOOM-clone support multiple window sizes.
closescreen. openscreen.

:       Chunky Copper: Crappy resolution, wastes lots of ChipRAM bandwidth
demo only.

:       BlitterScreen: Problems similair to Blitter assistance, but
:                      additionally you will need to turn up the contrast
:                      on your monitor, because the result of masking out

Wow that's the big problem ;) No user would accept putting up contrast,
even if it was the only method to get games smooth ;)
Ever seen a 1x1 blitterscreen ? ghost-look disappears!

The only true argument against: PAL/NTSC only, copper overtake.
both no problem on A1200+1084, which should be 90% of all users.

No argument to not support also 4pass versions and writeparr8...

: Any idea of how much C2P routines you could end up with when your game is done?
yes :)

:  Blitter+CPU C2P optimized for A1200 specifically
:  Blitter+CPU C2P optimized for 16-bit ChipRAM machines (ECS)
this is no argument. not only blitterscreen algos have to be
changed for ECS, also the 4passers. really no argument.
I wouldn't flame if there was only a AGA implementation of 3pass.
On ECS, 4 pass should be faster anyway and blitter is relative more
quick to fast->chip copy.

:  Blitter+CPU C2P optimized for 030 32-bit ChipRAM machines (4000/030)
:  CPU C2P optimized for 040 16-bit ChipRAM machines (ECS + 040)
dito.
:  CPU C2P optimized for 040 32-bit ChipRAM machines (4000/040)
this one exists.

you need cpu 1pass and maybe cpu 2passes. no cpu3passes needed for doom,
as those a cpu that can go beyond 25fps (the speed of blitter doing
fullscreen 1x1 for the remaining pass if cpu2pass) is a 040 -> 040cpu
4pass.

I already got 1pass. still got 2pass to write, ludde is said to have it
done free.

: and probably one which uses AKIKO, and of course for each of these routines a
FORGET AKIKO.

: version which does 2x1 displays, and versions for 16, 64 and 256 colors. Maybe
: also add ChunkyCopper and BlitterScreen C2P?  It looks like we've gone C2P
FORGET COPPER.

blitterscreen is only nessesary for fast clones, otherwise use 4pass.

: crazy...
Well, lot's of sources on aminet.

KEEP 6MB/sec and you get the specs I calculated!

: On top of that you will also want to support gfx-cards... and you need to write
already running on intuition-screens, it should be not that problem
to use writeparr8() ;)

: the rest of the game as well (but that's just a small part compared to all
: those C2P routines).  Have we reached the programmer's limit yet?

what ? writing doom is a small part compared to c2p ?
c2p is already written. doom not.

:  >> fully texturemapped/shaded games in two frames.

:  JrF> In two frames ? You must rely on a P133 system, with very good board &

: There is no Amiga with the same power as the P133.  And my 'average' (and 2
: year old $100) VLB Gfx card handles 15 MB/sec easily, more than enough to do
: 640x480 in 2 frames.

If 15mb/sec is more than enough to do 640x480 gfx, 6mb/sec surely are
no problem doing 320x160 ;) you contradict yourself :)

:  >> > JrF> Later AGA+ having 10 planes etc, would be a nice thing, yes.
:  JrF> 10bit is for example faster than 16bit, for your information.

: At what?  I bet it is slower than 16-bit for anything CPU calculated (ie,
: gouraud polygons, tmapping, rotating, and so on).

why do you think he told "10 planes would be slow" ? Because he thought
in AGA bandwidth. With enhanced bandwidth 10planes would be ok. For
a wb-screen, see the flame about planar and 2d operations ;)

ok, 10 planes would need a faster cpu. Maybe one of the faster PPCs
could do c2-10p in 15mb/sec ? :) If so this would outrule non-paletted
16bit displays (smother shading) and 24bit displays in terms of shding
speed (with 10 bit just use 4 for shading, doing shading with just
adding a register, no table-lookup -> more speed. Imagine descent real
gouraudshaded :).

:  >> >You forget that you still need a 68040 atleast to do DOOM even if you h
:  >> >super-fast Chunky card.  That's why there is no clone, the C2P problem

:  JrF> huh logic ? There is no clone because nobody programed it. A clone is
:  JrF> software, exisiting independent from the fact if a 040 exists.

: There is no (good) clone because it requires a 040 + fast Chunky gfx-card,
: period.  Caused of course by the fact that 040 + fast Chunky gfx-card is a rare
: combination found in the Amiga world.

"That's why there is no clone, the C2P problem...". What he wanted to say,
is a technical argument. You give me a marketing argument, which I am not
interested in now.

I still insist: AGA makes 19fps doom run 16.6fps. I guess users would be
very happy with 16.6 fps!

So if something keeps us from having 16.6fps doom, it's either the cpu
(how much could a 030-50 do when using same algo like doom ?) or
the programmers. not AGA. yes, the 16.6 fps doom would run 19fps on
a PCI gfx-card. I hope we all got this now...

:  >> >makes it worse though, on Amiga you'd require a 68060 to do fast DOOM (

:  JrF> the C2P problem on 040 is a PCer myth! C2P doesn't make it worse.

: Yes it does, see TextDemo.  The percentage of CPU time used for the C2P is
He said "require a 68060...". If 040 does the c2p already at bandwidth speed,
you don't need a 060. => the c2p is not the problem.

: NON-EXISTANT on the clones, because the 'fast-ram buffer' we use on Amiga is
: called 'the screen' on the clones.  No extra copying (or converting for that
: matter) needed.

really ? they render to gfx-card ram ?

:  >> >runs only 15-20 FPS on a 68060/50, 320x240x8 1x1, floors, walls, ceilin
:  >> >depressing).
:  JrF> huh ? first you tell me there is no clone and then you tell  YOU know
:  JrF> about how much it'll run on a 060 ? logic ?

: That's TextDemo 5.7x (unreleased version) someone tested for me.  15-20 FPS for
: a 68060/50 which is supposed to be 2-3 times as powerfull as a 486DX2/50 is
: quite depressing, considering that that 486 will do it at 30 FPS.  Now just

running DOOM... are you sure doom does as complicated mapping...
2-3 times... IMHO in that range it depends much more on the mem interface.
cache ? Reading textures is a quite mem intensive job, especially if you do
a table lookup for reach pixel (which is hopefully in cache sometimes).

: translate that to the slower Amiga's (ie, the ones only equipped with 030's and
: 040's).
well, imho there is nothing against using the 11cycle mapper for floor,
using coppershading. at least for 020/030.

12 cycles 320x160 on my 020-14 is 0.044sec...

: Grtz John

: -----------------------------------------------------------------------
:  [e-mail]   TextDemo/FastView/Etc... development
: -----------------------------------------------------------------------
: -- Via Xenolink 1.985B5, XenolinkUUCP 1.1
------------------------------------------------------------------------
   [e-mail] (Juergen "Rally" Fischer)   =:)