From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from hetzner.pbcl.net (mail.pbcl.net [88.198.119.4]) by mail.openembedded.org (Postfix) with ESMTP id CB227606E8 for ; Fri, 29 May 2015 22:24:40 +0000 (UTC) Received: from blundell.swaffham-prior.co.uk ([91.216.112.25] helo=e130.local) by hetzner.pbcl.net with esmtpsa (TLS1.2:RSA_AES_128_CBC_SHA1:128) (Exim 4.80) (envelope-from ) id 1YyShh-0006uY-9h; Sat, 30 May 2015 00:24:37 +0200 Message-ID: <1432938270.2461.46.camel@pbcl.net> From: Phil Blundell To: Paul Eggleton Date: Fri, 29 May 2015 23:24:30 +0100 In-Reply-To: <1528283.0UTgDEqc8E@peggleto-mobl.ger.corp.intel.com> References: <20150528084706.GC2383@jama> <5567C29C.3030307@windriver.com> <1528283.0UTgDEqc8E@peggleto-mobl.ger.corp.intel.com> X-Mailer: Evolution 3.12.9-1+b1 Mime-Version: 1.0 Cc: openembedded-core@lists.openembedded.org Subject: Re: [PATCH 4/5] libav: upgrade to 9.18 X-BeenThere: openembedded-core@lists.openembedded.org X-Mailman-Version: 2.1.12 Precedence: list List-Id: Patches and discussions about the oe-core layer List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Fri, 29 May 2015 22:24:41 -0000 Content-Type: text/plain; charset="UTF-8" Content-Transfer-Encoding: 7bit On Fri, 2015-05-29 at 16:28 +0100, Paul Eggleton wrote: > Just for background, the reason I disabled the textrel check for x86 in > libav.inc was that I was able to determine based on quick research that > upstream deliberately doesn't enable -fPIC for x86 (32-bit) because apparently > it doesn't really work there. I honestly didn't check what the situation was > on 32-bit ARM; I probably should have done that at the time. PIC is particularly problematic on x86 because it ties up %ebx. This represents 25% of the available general purpose registers which can cause a measurable performance loss because of increased spilling, and in the particular case of libav there is a lot of hand-written assembly in the codecs which expects to be able to clobber that register. It certainly used to be the case that this code wouldn't compile at all under -fpic. The performance thing does apply to ARM as well but to a slightly lesser extent. The second problem probably doesn't affect ARM at all. But, personally, I would disable the textrel check and build without -fpic on all platforms. p.