Date: Tue, 31 Mar 2015 21:50:48 +0300 From: Konstantin Belousov <kostikbel@gmail.com> To: Bruce Evans <brde@optusnet.com.au> Cc: svn-src-head@freebsd.org, svn-src-all@freebsd.org, src-committers@freebsd.org, John Baldwin <jhb@freebsd.org> Subject: Re: svn commit: r280279 - head/sys/sys Message-ID: <20150331185048.GU2379@kib.kiev.ua> In-Reply-To: <20150331144411.R3908@besplex.bde.org> References: <201503201027.t2KAR6Ze053047@svn.freebsd.org> <20150322080015.O955@besplex.bde.org> <20150322093251.GY2379@kib.kiev.ua> <2526359.g5B2nXdKeQ@ralph.baldwin.cx> <20150330172434.GG2379@kib.kiev.ua> <20150331144411.R3908@besplex.bde.org>
next in thread | previous in thread | raw e-mail | index | archive | help
On Tue, Mar 31, 2015 at 03:49:28PM +1100, Bruce Evans wrote: > It looks a bit overengineered to me. A bit like my function pointers > for the bcopy() family on i386. bcopy() is a bulk operation, so in > theory you can do it much faster by selecting the best available > version at runtime. In practice, the gains were not large and are > too machine-dependent to maintain. It is even harder to get large > gains and maintain them by selecting individual instructions at runtime. > Yes, it is similar to bcopy. The difference in motivation is that the IFUNCs are for features, not for speed. The existing patch already demostrates this WRT self-snoop and different methods of saving FPU state.
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20150331185048.GU2379>