This is the mail archive of the
binutils@sourceware.org
mailing list for the binutils project.
Re: Enable 32-bit Alignment
- From: Alan Modra <amodra at bigpond dot net dot au>
- To: "Menezes, Evandro" <evandro dot menezes at amd dot com>
- Cc: binutils at sourceware dot org, Michael Meissner <michael dot meissner at sbosexmb1 dot amd dot com>, Christophe Harle <christophe dot harle at txexmtae dot amd dot com>
- Date: Sat, 30 Jun 2007 13:17:42 +0930
- Subject: Re: Enable 32-bit Alignment
- References: <1E52B51E700FAF42ADA0B9432F7CCCD24907E7@SAUSEXMB1.amd.com>
On Fri, Jun 29, 2007 at 03:57:01PM -0500, Menezes, Evandro wrote:
> If it's not too late in the release cycle, I'd appreciate if this patch
> enabling optimal code alignment up to 32-byte boundaries, used by GCC
> when tuning for AMD Barcelona, were applied. It also does away with the
> special NOP sequences for AMD Opteron in 64-bit code.
In the current i386_align_code, see comment "Adjust jump offset" in
block handling 16-bit mode. Now that larger nop sequences are
allowed, I think you should similarly adjust a branch (if used) in
32-bit mode. Also, your code sequence for 31 byte nop padding will
not be very good. You'll get two 15 byte nops and a single nop.
Please rethink the way you have implemented this.
--
Alan Modra
IBM OzLabs - Linux Technology Centre