Path: chuka.playstation.co.uk!news
From: Andrew Partington <andrew.partington@harbinger.com>
Newsgroups: scee.yaroze.freetalk.english
Subject: Re: negative scrolling on homebrew map routine - need help!!
Date: Wed, 21 Mar 2001 17:07:06 +0000
Organization: PlayStation Net Yaroze (SCEE)
Lines: 87
Message-ID: <3AB8DFBA.7BDE2A4E@harbinger.com>
References: <3AAF7BB2.B427AED3@harbinger.com> <98o38v$ij81@www.netyaroze-europe.com> <98p3qp$rms2@www.netyaroze-europe.com> <98t840$6f5@www.netyaroze-europe.com> <99057g$3ok7@www.netyaroze-europe.com> <3AB6404F.65C9AF1C@harbinger.com> <9961fn$7jl4@www.netyaroze-europe.com> <99624q$7jl5@www.netyaroze-europe.com> <9963bb$7jl6@www.netyaroze-europe.com> <997j6l$7jl8@www.netyaroze-europe.com> <3AB774C7.6B4D8F95@harbinger.com> <998bjc$7jl11@www.netyaroze-europe.com> <3AB8A751.A5D36CCF@harbinger.com>
NNTP-Posting-Host: manuk81.harbinger.co.uk
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
X-Mailer: Mozilla 4.7 [en] (WinNT; I)
X-Accept-Language: en

Hang on - I think I know why i'm having problems setting the stack to point at the
DCache:

I was browsing the member homepages and I came across Andreas Schrattenecker's page
- he did something similar, but he was using a couple of global variables to store
the stack pointer in, instead of trying to dump it straight on the stack like the
thick git that I am :-P

Andreas reckons he's getting a 400% speed increase under CW, negligable speed
increase under GCC - is GCC really that much better when it comes to producing
optimal code?


Cheers


Andy

Andrew Partington wrote:

> Thanks Peter, that worked 100%!  But now I can't see how many HSyncs/frame I am
> getting - is there any way I can get my VSync(0) value back using callbacks
> without causing another wait for vertical retrace?  (calling VSync with values
> greater than 0 doesn't seem to return anything useful)
>
> I also tried setting the stack to point at the DCache, as in the the NY FAQ.
> The given asm wouldn't work for CW though, so I tried to come up with something
> I thought was equivalent...
>
> volatile asm void SetStackToDCache()
> {
>     sw $29,0(sp)                        // Put R3000 register 29 (stack pointer)
> onto stack
>     lw $29,0x1f8003f0               // Point at the upper bound of DCache (since
> the stack grows downwards)
> }
>
> volatile asm void SetStackToR29()
> {
>     lw $29,0(sp)                        // Restore stack pointer
> }
>
> Trying this out, (not surprisingly perhaps, since the first item of the stack
> might have had some useful value in it) I got the black screen of death!  I know
> the DCache is only 1K but I don't think I was putting enough params on the stack
> to overflow it (unless setting the project to build with maximum optimisation
> has reorganised the code somewhat, or I may need a delay slot to be filled with
> nop after the load).
>
> Can anyone see what I am doing wrong?  Also, do I set the stack to point at the
> DCache as the very first thing in the initialiation code, or only before a
> time-critical procedure call (resetting it afterwards)?  Or would a better
> approach be to just dump structures into the DCache as needed without touching
> the stack pointer?
>
> Cheers everyone! [crikey, this thread's getting hefty innit!  Not that i'm
> complaining :)]
>
> Andy
>
> Peter Armstrong wrote:
>
> > "Andrew Partington" <andrew.partington@harbinger.com> wrote
> > >
> > > Last night I attempted using VSyncCallback() for the first time to attempt
> > to
> > > get more raster time (as explained in the Net Yaroze FAQ), but I couldn't
> > get it
> > > working [so I went away and had a look at some other optimisation measures
> > > instead, none of which seemed to have much effect :( ]
> > >
> > > I had something like:
> > >
> > > (in my globals section of my code)
> > > volatile int waitVSync = 0;
> >
> > There are problems accessing global variables from within the callback
> > function. Try declaring your variable as
> >
> > __declspec(bss) volatile int waitVSync = 0;
> >
> > This should work.  There's also another method, which involves restoring the
> > gp register value, check out the CW doc Targeting Net Yaroze, page 66/67 for
> > more info.
> >
> > Peter