Date: Thu, 28 Nov 1996 16:03:16 -0500 (EST) From: "John S. Dyson" <toor@dyson.iquest.net> To: joerg_wunsch@uriah.heep.sax.de Cc: freebsd-current@FreeBSD.org Subject: Re: inlining (Was: users of "ft" tapes...) Message-ID: <199611282103.QAA03782@dyson.iquest.net> In-Reply-To: <199611282005.VAA16339@uriah.heep.sax.de> from "J Wunsch" at Nov 28, 96 09:05:02 pm
next in thread | previous in thread | raw e-mail | index | archive | help
> As Poul-Henning Kamp wrote: > > > >I have been thinking about un-inlining spls. This saves 29K out of > > >1096K text. It may even save some time > > > I generally use the rule of thumb that unless the text-size is > > smaller as a result, then inlining is wrong. > > This is hard to achieve. I think inlining is first aimed at > optimizing for speed, not for size. > Sometimes speed == size. Esp, on Intel architectures with small caches. I would like to know if there is going to be an impact on VM perf (which is much of what people see as perf on a machine, running various programs, and one reason why NT is so dog slow), before un-inlining the spl's. If it is neutral (and that is very likely), I would be for de-inlining also. Note that we actually had a net microbenchmark slowdown with some of my recent VM changes, but that had to be weighed against a speedup on large systems (by 2x-3x.) Most of the slowdown is probably due to the larger data structures (and larger cache footprint.) Alas, I am just not able to figure out how to make some of those damned things smaller, but if anyone does have any ideas, PLEASE PLEASE let me know. John
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199611282103.QAA03782>