Date: Tue, 9 Mar 1999 17:43:32 +0100
To: "pgcc AT delorie DOT com" <pgcc AT delorie DOT com>
Subject: Re: Optimizations questions
Message-ID: <19990309174332.D2217@cerebro.laendle>
Mail-Followup-To: "pgcc AT delorie DOT com" <pgcc AT delorie DOT com>
References: <36E48B4C DOT 86B84AD1 AT netplus DOT net>
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
In-Reply-To: <36E48B4C.86B84AD1@netplus.net>; from Steve Bergman on Mon, Mar 08, 1999 at 08:45:32PM -0600
X-Operating-System: Linux version 2.2.2 (marc AT cerebro) (gcc driver version pgcc-2.93.04 19990131 (gcc2 ss-980929 experimental) executing gcc version 2.7.2.3) 
From: Marc Lehmann <pcg AT goof DOT com>
Reply-To: pgcc AT delorie DOT com
X-Mailing-List: pgcc AT delorie DOT com
X-Unsubscribes-To: listserv AT delorie DOT com
Precedence: bulk

On Mon, Mar 08, 1999 at 08:45:32PM -0600, Steve Bergman wrote:
> I'm getting ready to recompile my RH5.2 system and 2.2.2 kernel.  I have an
> AMDK6-2/300.  Would "-O6 -march=amdk6 -mcpu=amdk6" be reasonable options to set
> in rpmrc?  What is the status of "-fstrength-reduce" and "-funroll-all-loops"

strength-reduce is fixed. it was never broken enough to warrant turning it
off in the first place.

loop unrolling has a few bugs in pgcc-1.1.1 (as well as in egcs), most of
these seem to be gone in the snapshots.

> these days?  Are they stable enough to compile my whole system with?

loop-unrolling + 1.1.1 is probably not.

> a bit unclear about the optimizations table in the faq.  When an option is
> marked "U" but is shown to be invoked at, say, level -O4, what does that mean? 

That there are known cases of it generating bad code.

> At the top of the table it says it "must be explicitly enabled in its -foption
> form".  Does that mean that you have to specify both -O4 and -foption to use
> that optimization?  It seems a bit ambiguous.

Well, that means it belongs into that optimization level, but _currently_
it is disabled. Some versions have it enabled, so I consider this useful
information nevertheless.

> I should add that this system is my own home system, and although I do most of
> my work on it, it is not a "critical production system" by any means.  I can
> afford to be at least a bit adventuresome, in the interest of performance and
> testing. (At home, no one can here you scream...  ;-)  )

Then just use the highest optimization level and experiment. Its a bit
inconvinient when your libc breaks, though... In any case, while I can't
guarantee it, most problems that you can expect will not kill your data.

--  
      -----==-                                             |
      ----==-- _                                           |
      ---==---(_)__  __ ____  __       Marc Lehmann      +--
      --==---/ / _ \/ // /\ \/ /       pcg AT goof DOT com      |e|
      -=====/_/_//_/\_,_/ /_/\_\       XX11-RIPE         --+
    The choice of a GNU generation                       |
                                                         |