Date: Wed, 16 Oct 2002 01:25:30 +1000 (EST) From: Bruce Evans <bde@zeta.org.au> To: Steve Kargl <sgk@troutmask.apl.washington.edu> Cc: Tim Robbins <tjr@FreeBSD.ORG>, Kris Kennaway <kris@obsecurity.org>, Andrew Gallatin <gallatin@cs.duke.edu>, Nate Lawson <nate@root.org>, <current@FreeBSD.ORG> Subject: Re: X problems & 5.0... -RELEASE? Message-ID: <20021016005506.Y992-100000@gamplex.bde.org> In-Reply-To: <20021015044028.GA17862@troutmask.apl.washington.edu>
next in thread | previous in thread | raw e-mail | index | archive | help
On Mon, 14 Oct 2002, Steve Kargl wrote: > On Tue, Oct 15, 2002 at 12:39:26PM +1000, Tim Robbins wrote: > > On Mon, Oct 14, 2002 at 01:00:46PM -0700, Kris Kennaway wrote: > > > > [...] > > > Did anyone test -current with the various FP test suites people posted > > > about last week? I run some of them occasionally (last ran ucbtest on June 18), but they are perfectly irrelevant to this problem since they don't use signals in any significant way. > > Yes. I ran paranoia from http://cm.bell-labs.com/netlib/paranoia/ and > > found that FP arithmethic is satisfactory when -O is not used, and no > > -march or -mcpu options are used. > > > > However, compiling with -O causes a lot of failures. > > Are you using an intel cpu? You need to add -ffloat-store > to get the correct results due to the 80 bit FPU registers. > Otherwise, intermediate results are stored with higher > precision. This is a well-known bug in gcc. Intermediate results may be computed with higher precision. paranioa knows this and uses lots of assignments to clip the results to double (or single) precision, aa is required to work by C standards (very fuzzily in C90 and very clearly in C99). This doesn't actually work in the i386 gcc. The -ffloat-store hack works around this and gives much the same pessimizations as would a standards conforming compiler. Bruce To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-current" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20021016005506.Y992-100000>