Date: Sun, 4 Jul 1999 15:44:29 +0300 (IDT)
From: Eli Zaretskii <eliz AT is DOT elta DOT co DOT il>
X-Sender: eliz AT is
To: Bill Currie <bill AT taniwha DOT org>
cc: djgpp-workers AT delorie DOT com
Subject: Re: .align directives in libc.a
In-Reply-To: <377C5986.1B33420B@taniwha.org>
Message-ID: <Pine.SUN.3.91.990704154405.13333I-100000@is>
MIME-Version: 1.0
Content-Type: TEXT/PLAIN; charset=US-ASCII
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
Precedence: bulk


On Fri, 2 Jul 1999, Bill Currie wrote:

> > > Intel manuals seem to recommend 16-byte alignment, both for code and
> > > for data.
> > 
> > And that's normally the best. MSVC uses 16 and 32 (based on some strange
> > criteria).
> 
> Isn't 32 bytes the size of a cache line on 486+?

The cache line size is 32 bytes, but the prefetch buffers fetch on
16-byte boundaries (or at least that's what Intel manuals say).  Intel
explicitly recommends to align on 16-byte boundaries for the large
data types and targets of jumps.  The only exception is large arrays
and data structures (longer than 32 bytes) for which 32-byte alignment
is recommended.