Sender: root AT meer DOT meer DOT net
Message-ID: <381B834C.A7D0A99B@netfall.com>
Date: Sat, 30 Oct 1999 16:46:21 -0700
From: Andrew Sharp <andy AT netfall DOT com>
Organization: Sharp Programmers
X-Mailer: Mozilla 4.61 [en] (X11; U; Linux 2.2.7 i686)
X-Accept-Language: en
MIME-Version: 1.0
To: pgcc AT delorie DOT com
Subject: Re: pgcc 2.95.1 vs pgcc 1.1.1 (1.1.1 wins !!!)
References: <381A4989 DOT 2AB8C071 AT cclinf DOT polito DOT it>
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Reply-To: pgcc AT delorie DOT com

I did a similar post a while back comparing output by pgcc for 2.95.1
and 2.91.6.  The floating point performance between the two has taken a
severe dive, but the integer/logic performance is up.  The latter
accounts for 95% of generated code, I will go out on a limb and claim,
but still the large drop in floating point performance is a bummer. 
This on a P2-333 system.  I used the Byte-benchmark, the source code for
which is commonly available on the net.

BTW, I have the exact same experience with bladeenc as below, but on my
P6-cored P2-333.  Can't use unroll-loops.  I've removed it from my
global config.cache and several problems cleared up in other programs.

a

Fox wrote:
> 
> Hello
> 
> I have an amd k6 233, Redhat 6.0 and kernel 2.2.13,
> just 2 days ago I have installed pgcc 2.95.1 (before I used pgcc 1.1.1).
> Than I have done some test:
> 1) bladeenc
>   compiling codec.c with an optimaztion higher than -O1 gives a
>   a working but bad encoder (but this is known even whith older
>   versions of pgcc and egcs).
>   compiling the rest of it (not the codec) whith -funroll-all-loops
>   makes bladeenc crash (whith pgcc 1.1.1 it didn't).
>   compiling whith -ffast-math gives a slow and bad working encoder
>   compiling bladeenc whithout -funroll-all-loops gives a fast
> well-working
>   bladeenc (faster the one compiled with pgcc 1.1.1 with
> -funroll-all-loops
>   enabled). So the flags I used for bladeenc :
>   -O6 -malign-functions=2 -malign-jumps=2 -mk6 -march=k6
> 2) XFree 3.3.5
>   compiling with pgcc 2.95.1 and with
>   -O6 -fno-strength-reduce -malign-functions=2 -malign-jumps=2 -mk6
> -march=k6
>   gives a not working X server (it crashes on startup),
>   compiling whith pgcc 1.1.1 and with
>   -O6 -fno-strength-reduce -malign-functions=2 -malign-jumps=2 -mamdk6
> -march=amdk6 -funroll-all-loops
>   gives a very fast and perfect Xserver (I have used it for mounths
> without a crash).
> 3) Mozilla M10
>   compiling with pgcc 2.95.1 and with
>   -O6 -fno-strength-reduce -malign-functions=2 -malign-jumps=2 -mk6
> -march=k6
>   gives a not working executable
>   compiling with pgcc 1.1.1 there is no way to get the build finished
> (due
>   to compiler internal errors)
> 4) WindowMaker 0.61.1
>   The one compiled with pgcc 2.95.1 is much faster than the one compiled
> with
>   pgcc 1.1.1.
> 5) The -fno-strength-reduce flag
>   The problems that there were in pgcc 1.1.1 with the
> -fno-strength-reduce flags
>   seem to be resolved in pgcc 2.95.1
> 
> My Conclusions:
> I find pgcc 1.1.1 much more reliable even if a little bit slower.
> 
> I hope you can tell me that I'm missing some "magic" flag that will
> make me compile Xfree and mozilla perfectly.
> If not I hope that this mail will help pgcc develop in some ways.
> Bye
>          Fox
> 
> P.S.
> Sorry for my bad english
> --
> -----------------------------
>         Fabio Volpe
>     fox AT cclinf DOT polito DOT it
>     fabiovolpe AT libero DOT it
>   via M. Ausiliatrice n. 32
>        Torino ITALY
>      Tel. 011/5224283
>        ICQ. 5214579
> -----------------------------