Date: Sat, 28 Jul 2012 21:40:04 -0500 From: Stephen Montgomery-Smith <stephen@missouri.edu> To: Bruce Evans <brde@optusnet.com.au> Cc: freebsd-bugs@FreeBSD.org, Stephen Montgomery-Smith <stephen@FreeBSD.org> Subject: Re: bin/170206: complex arcsinh, log, etc. Message-ID: <5014A284.2060204@missouri.edu> In-Reply-To: <20120729121443.M1008@besplex.bde.org> References: <201207281550.q6SFoBdC073014@freefall.freebsd.org> <20120729121443.M1008@besplex.bde.org>
next in thread | previous in thread | raw e-mail | index | archive | help
On 07/28/2012 09:31 PM, Bruce Evans wrote: > On Sat, 28 Jul 2012, Stephen Montgomery-Smith wrote: > >> OK. This clog really seems to work. >> >> x*x + y*y - 1 is computed with a ULP less than 0.8. The rest of the >> errors seem to be due to the implementation of log1p. The ULP of the >> final answer seems to be never bigger than a little over 2. > > I really don't like this version. If we want bignums and slowness, we > can use lib[g]mp or even exec a mostly-interactive calculator (bc for > portability, closed-source bigware for unportability). > > This version would have been useful in development to verify that the > doubled or tripled floating point was as exact as intended. Now writing > it in your favourite interactive calculator is easiest, except for the > problem of comparing the results. I can understand your reticence. I'll let you work some more on clog, and I will concentrate on the catrig stuff. But did you see that I messed up my earlier error analysis of log1p(1+x)? Also, if I had a log1pl, I would have used that.
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?5014A284.2060204>