Date: Wed, 16 Jul 2003 11:17:58 +1000 (EST) From: Bruce Evans <bde@zeta.org.au> To: David Schultz <das@FreeBSD.org> Cc: cvs-all@FreeBSD.org Subject: Re: cvs commit: src/contrib/gcc/config/i386 freebsd.h Message-ID: <20030716110439.M13786@gamplex.bde.org> In-Reply-To: <20030715060049.GA33856@HAL9000.homeunix.com> References: <200307121933.h6CJXYsP069815@repoman.freebsd.org> <20030715060049.GA33856@HAL9000.homeunix.com>
next in thread | previous in thread | raw e-mail | index | archive | help
On Mon, 14 Jul 2003, David Schultz wrote: > On Sat, Jul 12, 2003, Alexander Kabaev wrote: > > kan 2003/07/12 12:33:34 PDT > > > > FreeBSD src repository > > > > Modified files: > > contrib/gcc/config/i386 freebsd.h > > Log: > > Do not override FP rounding options for 32bit platforms. amd64 doesn't > > need this. > > You mean for 64-bit platforms, right? Does this mean that gcc's > problems with FP precision and constant evaluation have been > fixed, at least for amd64? I understand there have been some > improvements in this area in gcc 3.3. Right. It just means that the override is broken for the TARGET_64BIT case. amd64 seems to need the code for the same reasons as i386 (the npx control word is set to round to 53-bit precision). The override informs gcc of this, but it apparently causes problems for the 64-bit case. This seems to be the only improvement in this area. i386.md still bogusly says that the i387 has non-widening FP operations (SF + SF -> SF, etc.), so casts and assignments don't narrow widened FP results on i386's. Bruce
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20030716110439.M13786>