Date:	Mon, 3 Sep 2001 15:23:51 +0300 (WET)
From:	Andris Pavenis <pavenis AT lanet DOT lv>
X-Sender:	pavenis AT ieva06
To:	Juan Manuel Guerrero <ST001906 AT HRZ1 DOT HRZ DOT TU-Darmstadt DOT De>
Cc:	Eli Zaretskii <eliz AT is DOT elta DOT co DOT il>, djgpp-workers AT delorie DOT com
Subject:	Re: gcc301 difficulty
In-Reply-To:	<61357C1227C@HRZ1.hrz.tu-darmstadt.de>
Message-ID:	<Pine.A41.4.05.10109031514350.20338-100000@ieva06>
MIME-Version:	1.0
Reply-To:	djgpp-workers AT delorie DOT com
Errors-To:	nobody AT delorie DOT com
X-Mailing-List:	djgpp-workers AT delorie DOT com
X-Unsubscribes-To:	listserv AT delorie DOT com

On Mon, 3 Sep 2001, Juan Manuel Guerrero wrote:

> The reason for the failure seems to be the processor.
> I am using a system composed by:
> MB:  Tyan Trinity 100AT
> uP:  SGS-Thomson ST6x86 P166+ (FSB 66MHz, CPU Multiplicator: 2x)
> RAM: 32MB (fast page) + 32MB (edo)
> 
> Please note that the uP is certainly *not* overclocked.
> The only way to get gcc-301 working is to disable the internal
> CPU cache (using BIOS settings). Disabling on-board L2 cache has
> no influence. I have reduced FSB to 60MHz and reduced CPU Multiplicator
> to 1x but it had no positive influence. I have substituted the board with
> a Chaintech 5IFM but I have got no positive results. This old
> board allows FSB of 55 MHz and 50 MHz. Lowering FSB seems to have
> no influence and the only way to get gcc working with a cyrix cpu
> (the cpu has no MMX instructions) seems to be to disable the
> internal cache. Of course, this measure is not acceptable at all
> due to the impressive slow-down.
> I am very sorprised about all this. I have been using gcc30b.zip
> for a month without any difficulty. I expected gcc301b.zip
> to be only a minor bug fix release and not to cause such
> difficulties.

I don't know. Binaries of gcc-3.0 for DJGPP were built using current CVS 
version of DJGPP runtime. Binaries of gcc-3.0.1 were built with slightly
modified djdev203 (Win2k related patches from Charles Sandmann, but I
simply replaced related object files in libc.a with ones I got from
clio.rice.edu, maybe I should compile them myself next time ...)

However such effect with L1 cache seems strange anyway

> 
> Btw, I have tried to recompile gcc301s.zip. This seems impossible
> using win95 (4.00.950). The genattrib program always dies with
> the message that virtual memory has become exhausted. (DPMI
> memory available=48726KB, DPMI swap space available=16106KB).
> Am I missing something here? Can win95 not be used to compile gcc301?

Win95 is not guilty. You have not enough RAM as genattrtab is very 
memory hungry. Try setting memory for DOS box to 65535K and see whether
it helps. I built gcc-3.0.1 on PIII machine with 128Mb of memory, so I
used default (auto) and didn't get any problems

There are some existing patches which makes genattrtab to use less 
memory but these patches are not yet in official sources

Andris

- Raw text -