Date: Thu, 17 Sep 2015 21:40:26 +0200 From: Nikola Pajkovsky <n.pajkovsky@gmail.com> To: "O. Hartmann" <ohartman@zedat.fu-berlin.de> Cc: =?utf-8?Q?Jean-S=C3=A9bastien_P=C3=A9dron?= <dumbbell@FreeBSD.org>, freebsd-current@freebsd.org Subject: Re: Intel Haswell support - Any updates? Message-ID: <87io78ltqt.fsf@gooddata.com> In-Reply-To: <20150917212228.75ec5471.ohartman@zedat.fu-berlin.de> (O. Hartmann's message of "Thu, 17 Sep 2015 21:22:28 %2B0200") References: <55FA7324.5010603@icloud.com> <55FAF19F.9090701@FreeBSD.org> <20150917212228.75ec5471.ohartman@zedat.fu-berlin.de>
next in thread | previous in thread | raw e-mail | index | archive | help
"O. Hartmann" <ohartman@zedat.fu-berlin.de> writes: > Am Thu, 17 Sep 2015 19:00:15 +0200 > Jean-S=C3=A9bastien P=C3=A9dron <dumbbell@FreeBSD.org> schrieb: > >> Hi everyone! >>=20 >> I'm very sorry I didn't communicate at all on the i915 update project. >>=20 >> So here is a status update: since this morning, the driver builds fine. >> I'm currently attending the XDC (X.Org Developers Conference) and don't >> have an Intel laptop to test with me. However, Johannes Dieterich (also >> attending the conference) offerred his help, so we will do that today. >>=20 >> Obviously, do not expect something stable in the coming couple days. >> Thank you for your patience :) >>=20 >> To answer various questions in this thread: >>=20 >> Why does it take so much time to update? Once Konstantin committed his >> i915 update, I was busy with non-FreeBSD activities until last July, >> when I slowly started back to work on i915. My goal is to reduce the >> diff with Linux as much as possible. But, as opposed to OpenBSD and >> DragonFlyBSD, we do not use a Linux compatibility layer which would >> dramatically ease our life. > > My concerns are speed and performance. Isn't any kind of layer consuming = performance - > sometimes worse, sometimes negligible. But anyway, HPC isn't a FreeBSD do= main, so ... Look at the linux spinlock layer in ofed/include/linux/spinlock.h #define spin_lock(_l) mtx_lock(&(_l)->m) #define spin_unlock(_l) mtx_unlock(&(_l)->m) #define spin_trylock(_l) mtx_trylock(&(_l)->m) means, that using spinlock linux layer does not have any performance impact. I haven't read all ofed code, but most of that is just bunch of macros and renaming stuff to use linux code without changes and no performance impact. --=20 Nikola
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?87io78ltqt.fsf>