Date: Thu, 30 Nov 2017 22:17:13 +0100 From: Szabolcs Nagy <nsz@...t70.net> To: musl@...ts.openwall.com Subject: Re: remquo - underlying logic * Damian McGuckin <damianm@....com.au> [2017-12-01 07:29:05 +1100]: > It is on MY own TO-DO list. I was asking the question in case anybody had > done it before and found it useless, could think of major pitfalls, or had > any pearls of wisdom before we started. I will keep you posted. well, one of the difficulties is to get correct behaviour in case of fenv access (e.g. remquo has exact result in all rounding modes so intermediate results should be exact too, raising underflow correctly is sometimes nontrivial, getting the sign right when the result is zero in downward rounding mode sometimes needs special handling, etc), if you need to save/restore the fenv to achieve that then it's unlikely to be faster than using ints, otherwise floats are probably faster, but it's hard to tell.
Powered by blists - more mailing lists
Powered by Openwall GNU/*/Linux - Powered by OpenVZ