NNTP-Posting-Host: 172.16.26.200
Newsgroups: comp.os.msdos.djgpp
Date: Tue, 17 Aug 1999 14:43:06 +0200
Message-ID: <0E16861EE7BCD111BE9400805FE6841F0AF6D76B@c1s5x001.cor.srvfarm.origin-it.com>
From: bas DOT hamstraNOSPAM AT wxs DOT nl (Bas Hamstra)
Subject: Re: New gcc 2.95
References: <7oqfvr$qpu$1 AT reader1 DOT wxs DOT nl> <Pine DOT SUN DOT 3 DOT 91 DOT 990812105820 DOT 9373R-100000 AT is> <19990815170554 DOT A17468 AT tabor DOT ta DOT jcu DOT cz>
X-Newsreader: Forte Free Agent 1.0.82
Lines: 62
To: djgpp AT delorie DOT com
DJ-Gateway: from newsgroup comp.os.msdos.djgpp
Reply-To: djgpp AT delorie DOT com

I think you are right about the bad luck. I found this:

					gcc 2.81	gcc 2.95	CBuilder 4.0		

- old version of chess program 		123a		112a		132a	
- new version of chess program 		-		550b		500b

a = Positions searched per Second (x1000)
b = Times a sec that all legal moves in a pos are generated (x1000)

The new version of the program uses a lot of 64 bit integers.

The strange thing is for program a) gcc 2.95 is slowest, while for b)
it is fastest. Borland other way round.

So it could very well be that gcc 2.95 is overall faster than 2.81.

optimation options used: 
-o6 -fomit_frame_pointer

>On Thu, Aug 12, 1999 at 10:58:39AM +0300, Eli Zaretskii wrote:
>> 
>> On Wed, 11 Aug 1999, "Bas Hamstra" <bas.hamstra wrote:
>> 
>> > I donwloaded some components of GCC 2.95 (binaries of gcc, gxx and the libs,
>> > I believe) and copied them onto the old djgpp package (latest official
>> > version). It works, but now my programs run slower, regardless of the
>> > optimation options.
>There are many changes in 2.95, so it is possible that you have just bad luck
>and some of new features hurts your program. At the average 2.95 is a win
>(pretty noticeable).

>I would recommend you to try -mpreferred-stack-boundary=2 option. It turns off
>stack alignment code and makes executable smaller, function calling faster
>and may reduce performance of fp code, if some of the temporaries spilled to
>stack in internal loop gets missaligned.

>Honza


Jan Hubicka <hubicka AT tabor DOT ta DOT jcu DOT cz> wrote:

>On Thu, Aug 12, 1999 at 10:58:39AM +0300, Eli Zaretskii wrote:
>> 
>> On Wed, 11 Aug 1999, "Bas Hamstra" <bas.hamstra wrote:
>> 
>> > I donwloaded some components of GCC 2.95 (binaries of gcc, gxx and the libs,
>> > I believe) and copied them onto the old djgpp package (latest official
>> > version). It works, but now my programs run slower, regardless of the
>> > optimation options.
>There are many changes in 2.95, so it is possible that you have just bad luck
>and some of new features hurts your program. At the average 2.95 is a win
>(pretty noticeable).

>I would recommend you to try -mpreferred-stack-boundary=2 option. It turns off
>stack alignment code and makes executable smaller, function calling faster
>and may reduce performance of fp code, if some of the temporaries spilled to
>stack in internal loop gets missaligned.

>Honza