From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from eggs.gnu.org ([2001:4830:134:3::10]:57879) by lists.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1bcWcZ-0003hg-KL for qemu-devel@nongnu.org; Wed, 24 Aug 2016 07:45:28 -0400 Received: from Debian-exim by eggs.gnu.org with spam-scanned (Exim 4.71) (envelope-from ) id 1bcWcY-0002ks-MG for qemu-devel@nongnu.org; Wed, 24 Aug 2016 07:45:27 -0400 Date: Wed, 24 Aug 2016 07:45:15 -0400 (EDT) From: Paolo Bonzini Message-ID: <2076948386.5196666.1472039115921.JavaMail.zimbra@redhat.com> In-Reply-To: References: <1472012279-20581-1-git-send-email-rth@twiddle.net> <20160824083457.GA2032@work-vm> MIME-Version: 1.0 Content-Type: text/plain; charset=utf-8 Content-Transfer-Encoding: 7bit Subject: Re: [Qemu-devel] [PATCH 0/7] Improve buffer_is_zero List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , To: Peter Maydell Cc: Adam Richter , "Dr. David Alan Gilbert" , Richard Henderson , Vijay Kilari , Liang Li , QEMU Developers , qemu-arm > On 24 August 2016 at 11:26, Adam Richter wrote: > > I hope that that consideration, combined with the micro-costs to > > readability and portability of using and ELF specific and perhaps > > currently GCC specific feature might tip the balance against the > > savings of a level of function call indirection that I assume the use > > of ifunc was intended to provide. > > It doesn't actually save a level of indirection -- if you single step > through an ifunc call it goes via some ELF section. The thing it > does save is that you don't pay the cost of figuring out the right > ifunc to use on this system at startup, but only when the ifunc call > path is first used. That's useful for a big thing like glibc which > might have lots of ifuncs and not want to pay a big startup cost, > but for QEMU there's really no need given we only have one... It does save a level of indirection after the first call AFAIK, but it shouldn't be measurable. Paolo