Date: Tue, 28 May 2013 21:12:00 +1000 (EST) From: Bruce Evans <brde@optusnet.com.au> To: David Schultz <das@FreeBSD.org> Cc: Diane Bruce <db@db.net>, Bruce Evans <brde@optusnet.com.au>, John Baldwin <jhb@FreeBSD.org>, David Chisnall <theraven@FreeBSD.org>, Stephen Montgomery-Smith <stephen@missouri.edu>, freebsd-numerics@FreeBSD.org, Steve Kargl <sgk@troutmask.apl.washington.edu>, Peter Jeremy <peter@rulingia.com>, Warner Losh <imp@bsdimp.com> Subject: Re: Use of C99 extra long double math functions after r236148 Message-ID: <20130528205441.U2294@besplex.bde.org> In-Reply-To: <20130528081921.GB13594@zim.MIT.EDU> References: <501204AD.30605@missouri.edu> <20120727032611.GB25690@server.rulingia.com> <20120728125824.GA26553@server.rulingia.com> <501460BB.30806@missouri.edu> <20120728231300.GA20741@server.rulingia.com> <50148F02.4020104@missouri.edu> <20120729222706.GA29048@server.rulingia.com> <5015BB9F.90807@missouri.edu> <20130528043205.GA3282@zim.MIT.EDU> <20130528155933.V1298@besplex.bde.org> <20130528081921.GB13594@zim.MIT.EDU>
next in thread | previous in thread | raw e-mail | index | archive | help
On Tue, 28 May 2013, David Schultz wrote: > On Tue, May 28, 2013, Bruce Evans wrote: >> >> I think I tested "inline" on the large functions (just 2) and found >> it useful for efficiency. This is like inline on large trig support >> functions being useful. The inline parts are duplicated once per >> C99-API function, and often the caller only uses on C99-API function. >> Actually, the large inlines are not duplicated that much. cacosh() >> and casinh() are just wrappers that call cacos() and casin(), >> respectively. There is no inlining for the last 2 (even larger) >> functions. The overhead for the wrappers is noticeable, but more >> inlining didn't seem to reduce it much. > > If you did benchmarks to show that using inline is worthwhile > despite the cache pressure, then it's fine with me. I had assumed > that it was added without much thought. I retested. Inlining the big function do_hard_work() helps for gcc on amd64 (about 5% faster), but makes no significant difference for clang. The previous testing was mostly with gcc. Bruce
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20130528205441.U2294>