To: Theo Schlossnagle <jesus@omniti.com>
Cc: Tom Lane <tgl@sss.pgh.pa.us>
Cc: Bruce Momjian <pgman@candle.pha.pa.us>, pgsql-bugs@postgresql.org,
	josh.berkus@sun.com, robert.lor@sun.com
Subject: Re: BUG #2401: spinlocks not available on amd64
In-Reply-To: <4447A288.6020409@omniti.com>
User-Agent: Gnus/5.11 (Gnus v5.11) Emacs/22.1 (gnu/linux)
References: <4447A288.6020409@omniti.com>
From: Gregory Stark <stark@mit.edu>
Date: Wed, 24 Jun 2009 16:27:02 +0100
Message-ID: <877hz1abzt.fsf_-_@oxford.xeocode.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii


Theo Schlossnagle wrote:

> Tom Lane wrote:
>
>> There is no reason for the i386 or AMD64 code to be different from what's
>> already tested on Linux --- the hardware's the same and the OS surely
>> doesn't make a difference at this level.
>
> On linux you use gcc, which allows for inline assembly. So, the code is
> already very different.

How does this interact with binary builds such as rpms? If someone installs an
amd64 binary on an x86 machine or vice versa does this assembly do the right
thing at all? Does it perform slowly?

Ideally we would compile both and pick the right one at run-time but that
might have annoying overhead if there's a branch before every pg_atomic_cas
call.

Perhaps a minimal thing to do would be to detect a mismatch on startup and log
a message about it.

-- 
  Gregory Stark
  http://mit.edu/~gsstark/resume.pdf