Date: Mon, 31 Mar 1997 18:25:49 +0300 (IDT)
From: Eli Zaretskii <eliz AT is DOT elta DOT co DOT il>
To: Dan Hirschberg <dan AT ics DOT uci DOT edu>
cc: djgpp AT delorie DOT com
Subject: Re: gcc optimize performance, V2.7.2.1 versus V2.6.0
In-Reply-To: <9703291145.aa19506@paris.ics.uci.edu>
Message-ID: <Pine.SUN.3.91.970331181732.17577N-100000@is>
MIME-Version: 1.0
Content-Type: TEXT/PLAIN; charset=US-ASCII


On Sat, 29 Mar 1997, Dan Hirschberg wrote:

> The same program compiled with the newer compiler had much worse
> performance (about 50% worse) on both my 486 and my Pentium Pro.
> The performance was even worse with -O2.  I was able to improve
> the performance somewhat by using -O1 and adding a few of the
> -f options (found by reading the info file on gcc), but still about
> 40% worse than under the 2.6.0 version.
> 
> Is this a common experience?

No.  People who tested this report that the optimization quality is the 
same, and sometimes slightly better in 2.7.2.1.

> Could this be a result of my program having particular characteristics?

Yes.  But first, be sure to read section 14.2 of the DJGPP FAQ list.  The 
effect of some optimizations switches is different between different GCC 
versions, and that section tells you how to choose the best combination 
for the new version.  

If the advice there don't help, I suggest reading section 14.5 which 
might give you ideas about possible causes of slow-down which are 
specific to your program.  If that doesn't unlock the problem either, 
tell more about what does your program do and in what environment (DOS? 
Windows? NT?) do you run the two different versions.  Profiling the 
program might also point to the problem spots.