delorie.com/archives/browse.cgi   search  
Mail Archives: djgpp-workers/1999/09/02/08:25:13

Message-ID: <19990902132931.A28343@tabor.ta.jcu.cz>
Date: Thu, 2 Sep 1999 13:29:31 +0200
From: Jan Hubicka <hubicka AT tabor DOT ta DOT jcu DOT cz>
To: djgpp-workers AT delorie DOT com
Subject: Re: gcc-2.95.1
References: <B0000100246 AT stargate DOT astr DOT lu DOT lv>
Mime-Version: 1.0
X-Mailer: Mutt 0.93i
In-Reply-To: <B0000100246@stargate.astr.lu.lv>; from pavenis@lanet.lv on Thu, Sep 02, 1999 at 12:48:44PM +0300
Reply-To: djgpp-workers AT delorie DOT com
X-Mailing-List: djgpp-workers AT delorie DOT com
X-Unsubscribes-To: listserv AT delorie DOT com

On Thu, Sep 02, 1999 at 12:48:44PM +0300, pavenis AT lanet DOT lv wrote:
> Hi!
> 
> Did some benchmarking of gcc-2.95.1 configured both with and 
> without --haifa-enable on 4 diffent CPUs:
> 
> 	- i486 DX266   - no evident difference between 2 compilers (with
>                                    and without --enable-haifa)
> 	- Pentium 200MMX -  the same
> 	- K6-2 300 - HAIFA enabled compiler generates about 20% faster
> 	                     code for -O0 and -O2. For -O3 difference is smaller or
>                              is absent in some conditions
>         - Pentium 2 350MHz - no evident difference
> 	
> So it remains unclear whether I should use --enable-haifa for binaries 
> I'll upload to ftp.delorie.com
Enabling haifa on x86 platforms is quite dangerous IMO. My K6 code is tuned
for it, but others don't. In larger tests the results are hit/miss and gcc
maitainers decided to stay with the old stable choice.

I would suggest you to wait for next gcc release (that don't contain normal
scheduler anyway)

Honza
> 
> Andris

- Raw text -


  webmaster     delorie software   privacy  
  Copyright © 2019   by DJ Delorie     Updated Jul 2019