From mboxrd@z Thu Jan 1 00:00:00 1970 From: Avi Kivity Subject: Re: [PATCH] Lazy FPU for SVM Date: Tue, 24 Apr 2007 17:32:58 +0300 Message-ID: <462E151A.7070306@qumranet.com> References: <462C2048.4040700@us.ibm.com> <462C574E.2030304@qumranet.com> <462CC380.5020405@codemonkey.ws> <462DDE52.5040303@qumranet.com> <462E0D52.5000707@codemonkey.ws> Mime-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit Cc: kvm-devel To: Anthony Liguori Return-path: In-reply-to: <462E0D52.5000707-rdkfGonbjUSkNkDKm+mE6A@public.gmane.org> List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Sender: kvm-devel-bounces-5NWGOfrQmneRv+LV9MX5uipxlwaOVQ5f@public.gmane.org Errors-To: kvm-devel-bounces-5NWGOfrQmneRv+LV9MX5uipxlwaOVQ5f@public.gmane.org List-Id: kvm.vger.kernel.org Anthony Liguori wrote: > Avi Kivity wrote: >> Anthony Liguori wrote: >> >> [updated patch] >> >> Applied. >> >> >>> Hard to say exactly because of the noise. I did two runs of 4 >>> test/vmexit: >>> >>> Before 4091, 4194, 4559, 4439 >>> After: 3979, 4324, 3918, 3910 >>> >>> So there's definitely a speedup, but probably only 100-200 cycles. >>> >>> >> >> Running on an idle, headless server with the vcpu pinned reduces >> variability to a few cycles in the after case. Strangely, the >> variability is high in the before case. I get a similar disappointing >> result of 100-200 cycles. >> > > I previously measured how long fx_save/fx_restore took and it was > right around 150 cycles on AMD. Well, we're calling it twice per vmexit. A save/restore pair is 313 cycles on a Woodcrest; two of those make 626 out of ~4000 cycles should get us a nice 15% gain. -- Do not meddle in the internals of kernels, for they are subtle and quick to panic. ------------------------------------------------------------------------- This SF.net email is sponsored by DB2 Express Download DB2 Express C - the FREE version of DB2 express and take control of your XML. No limits. Just data. Click to get it now. http://sourceforge.net/powerbar/db2/