Date: Sun, 27 Mar 2016 23:44:13 +0300
From: "Kirill A. Shutemov" <kirill@shutemov.name>
To: George Spelvin <linux@horizon.com>
Cc: ebiggers3@gmail.com, linux-kernel@vger.kernel.org, linux-mm@kvack.org
Subject: Re: Bloat caused by unnecessary calls to compound_head()?
Message-ID: <20160327204413.GB9638@node.shutemov.name>
References: <20160327203304.9695.qmail@ns.horizon.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20160327203304.9695.qmail@ns.horizon.com>
User-Agent: Mutt/1.5.23.1 (2014-03-12)
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 656
Lines: 18

On Sun, Mar 27, 2016 at 04:33:04PM -0400, George Spelvin wrote:
> Could you just mark compound_head __pure?  That would tell the compiler
> that it's safe to re-use the return value as long as there is no memory
> mutation in between.
> 

Hm. It has some positive impact, but it's not dramatic. For instance,
mm/swap.o results:

add/remove: 0/0 grow/shrink: 0/2 up/down: 0/-43 (-43)
function                                     old     new   delta
__page_cache_release                         319     298     -21
release_pages                                722     700     -22

mark_page_accessed() problem was not fixed by that.

-- 
 Kirill A. Shutemov