Date:	Wed, 21 Jul 1999 17:23:18 +0300 (IDT)
From:	Eli Zaretskii <eliz AT is DOT elta DOT co DOT il>
X-Sender:	eliz AT is
To:	"Mark E." <snowball3 AT bigfoot DOT com>
cc:	djgpp-workers AT delorie DOT com
Subject:	Re: .align directives in libc.a
In-Reply-To:	<199907211310.NAA41654@out5.ibm.net>
Message-ID:	<Pine.SUN.3.91.990721171710.429J-100000@is>
MIME-Version:	1.0
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 Wed, 21 Jul 1999, Mark E. wrote:

> > Then I suggest to aim at having the next release of Binutils use
> > 16-byte section alignment for DJGPP.  Do you agree?
> 
> If changing the alignment will help programs run faster, then I'm all for 
> it. 

We had a thread here some time ago about this, and a conclusion was that 
boosting to 16-byte alignment removed random slow-down of programs 
compiled with different versions of GCC and Binutils, when run on Pentium.
This issue is an annoying source of FAQs since Binutils 2.6 (see section 
14.3 in the FAQ).  v2.02 solved part of this by aligning the stack on 
8-byte boundary.  It would be nice if we could finally close this issue 
for good, at least until Intel come out with a chip that fetches on 
32-byte boundary ;-)

Also, if somebody wants to squeeze the most out of an inner loop, they 
would carefully align the jump targets and entry points to 16-byte 
boundary, as Intel suggests.  However, if the subsections aren't 16-byte 
aligned, they will never be able to get what they want, not with COFF.

- Raw text -