This is the mail archive of the
binutils@sourceware.org
mailing list for the binutils project.
RE: Enable 32-bit Alignment
- From: "Menezes, Evandro" <evandro dot menezes at amd dot com>
- To: "H.J. Lu" <hjl at lucon dot org>
- Cc: binutils at sourceware dot org, "Meissner, Michael" <michael dot meissner at amd dot com>, "Harle, Christophe" <christophe dot harle at amd dot com>, "rajagopal, dwarak" <dwarak dot rajagopal at amd dot com>
- Date: Wed, 4 Jul 2007 11:06:42 -0500
- Subject: RE: Enable 32-bit Alignment
- References: <1E52B51E700FAF42ADA0B9432F7CCCD2490910@SAUSEXMB1.amd.com> <20070704010231.GA2213@lucon.org>
HJ,
> Can you add nops-4.d and x86-64-nops-4.d?
Sure.
> I'd like to see some further changes:
>
> 1. For N byte alignment padding up to 15 bytes:
> a. On 32bit,
> i. Use (N - 1) 66 prefixes + 90 by default.
> ii. When Pentium Pro or newer is specified, use (N - X) 66
> prefixes + new nop insn with X bytes.
> iii. For Pentum or older, use the old padding.
If #2 is for Ppro or newer and #3, for P5 or older, when does #1 come into play? #1 does seem to be detrimental on 486 and older.
Anyways, I wasn't able to find good information on older processors (before Ppro), so I'd rather leave them as is.
> b. On 64bit, use (N - X) 66 prefixes + new nop insn with X bytes.
I.e., no change, right?
> 2. For alignment padding between 16 and 30 bytes, use 2 x (<
> 15byte nops).
> 3. For 31 byte alignment padding, use 2 x (15byte nops) + 1byte nop.
I disagree. The idea is to keep the number of NOPs down and by doing so there would be some cases that would result in one more NOP than in the way I proposed in my patch. Other than neatness, is there any reason you're proposing this?
Thanks,
--
_______________________________________________________
Evandro Menezes AMD Austin, TX