From mboxrd@z Thu Jan 1 00:00:00 1970 From: Jesse Barnes Subject: Re: [PATCH] drm/i915: Add OACONTROL to the command parser register whitelist. Date: Fri, 16 May 2014 13:12:49 -0700 Message-ID: <20140516131249.5e696f70@jbarnes-desktop> References: <20140326182605.GA28133@bdvolkin-ubuntu-desktop> <20140327075721.GA26878@phenom.ffwll.local> <20140327155708.GA4690@bdvolkin-ubuntu-desktop> <53349971.3000806@whitecape.org> <5334B2C4.5000702@whitecape.org> <20140516120545.23ddf6d7@jbarnes-desktop> <20140516192050.GA4495@nuc-i3427.alporthouse.com> <20140516123408.75e68aed@jbarnes-desktop> <20140516194930.GB4495@nuc-i3427.alporthouse.com> Mime-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit Return-path: Received: from gproxy3-pub.mail.unifiedlayer.com (gproxy3-pub.mail.unifiedlayer.com [69.89.30.42]) by gabe.freedesktop.org (Postfix) with SMTP id 0B55B6E229 for ; Fri, 16 May 2014 13:12:44 -0700 (PDT) In-Reply-To: <20140516194930.GB4495@nuc-i3427.alporthouse.com> List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: intel-gfx-bounces@lists.freedesktop.org Sender: "Intel-gfx" To: Chris Wilson Cc: Daniel Vetter , "intel-gfx@lists.freedesktop.org" List-Id: intel-gfx@lists.freedesktop.org On Fri, 16 May 2014 20:49:30 +0100 Chris Wilson wrote: > On Fri, May 16, 2014 at 12:34:08PM -0700, Jesse Barnes wrote: > > On Fri, 16 May 2014 20:20:50 +0100 > > Chris Wilson wrote: > > > We haven't even fixed the major regression from enabling FBC. What's > > > another massive slowdown? > > > > I thought you had that fixed in the X driver by avoiding front buffer > > rendering altogether. If that's the case we just need an ioctl to opt > > out of front buffer tracking, right? Presumably that flag would follow > > the current DRM_MASTER process... > > No. All rendering triggers FBC updates. Ville has patches to fix the > majority of that with only nuking FBC when front buffer rendering. (Note > that games aren't usually affected by this because FBC is disabled by > pageflipping.) The overhead could probably be reduced further by periodic > nuking the FBC like (ideally) PSR. And yes X can avoid front buffer > rendering altogether. The remaining challenge is to know when to enable it > (the kernel doesn't give us any information about FBC or PSR after setting a > mode) and when not, i.e. there is already a pageflipping compositor. I thought there was a control bit for when the nuke occurred? If not, yeah I guess we have to go with a sw approach... -- Jesse Barnes, Intel Open Source Technology Center