Date: Tue, 02 Dec 1997 21:34:52 +0100 From: Poul-Henning Kamp <phk@critter.freebsd.dk> To: current@FreeBSD.ORG Subject: Re: cvs commit: src/sys/kern vfs_aio.c Message-ID: <3286.881094892@critter.freebsd.dk> In-Reply-To: Your message of "Tue, 02 Dec 1997 11:42:21 PST." <199712021942.LAA16742@kithrup.com>
next in thread | previous in thread | raw e-mail | index | archive | help
In message <199712021942.LAA16742@kithrup.com>, Sean Eric Fagan writes: >>We save the time it takes to push the &retval on the stack on ALL syscalls. > >Versus the time it take sto copy the extra two words around when creating >the proc structure, versus the extra size of the proc structure, versus the >extra work now being done to deal with it, etc. the difference between you and me on this is that I have measured the actual difference, you are writing on the back of an envelope. >It takes less than 5 cycles to push the address. That adds up to some >degree when dealing with the layers of calls -- that I can see as a savings. >But it's insignificant, or should be -- unless you have real numbers to back >this up. If it adds up to 0.1% of an average system call time, I would be >highly surprised. I do have numbers. Overall improvement after this change: 0.01 % >Sorry, that is *still* not a compelling argument. If you want to make >things more efficient, there are some things you can do with FPU emulation >that can save a bunch of clock cycles off; or you could think about >rewriting how the trap handler itself is done -- that could be improved >quite a bit, I think. But it'd probably require more assembly programming. Well, why don't you look at that, and leave me to do what I have in my pipeline, OK ? -- Poul-Henning Kamp FreeBSD coreteam member phk@FreeBSD.ORG "Real hackers run -current on their laptop."
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?3286.881094892>