On Tue, 23 Aug 2005 16:54:01 +0900 Carsten Haitzler (The Rasterman) <[EMAIL PROTECTED]> bubbled:
> On Tue, 23 Aug 2005 09:51:26 +0200 Martin Zwickel > <[EMAIL PROTECTED]> babbled: > > > On Tue, 23 Aug 2005 10:30:28 +0900 > > Carsten Haitzler (The Rasterman) <[EMAIL PROTECTED]> bubbled: > > > > > On Mon, 22 Aug 2005 20:53:07 +0200 ilLogict <[EMAIL PROTECTED]> > > > babbled: > > > > > > > Hello! > > > > > > > > I always had the same problem on my box. But on amd64, this is > > > > set as gcc's default. Do anyone know how to bypass it so that it > > > > won't be used? > > > > > > man gcc. as for the sse fp math... it (from memory) makes gcc use > > > the sse instruction set to do floating point math operations. this > > > is likely a bad thing as it 1. means using the "mmx unit" and > > > swapping out of fp mode and thus you have a cycle hit with an emms > > > coming back in - as the fp ops are short and not all over this > > > isn't much worth it. i suspect there is some conversion rounding > > > cumulative error. but as i dont have an amd64 i cant tell you much > > > more than what i suspect. :) > > > > Maybe you can disable the option in the specs file?! > > that is not our job - to dictate build flag policy. that's the job of > the builder. :) we provide a SAMPLe .spec file for building rpm's - > it's nto a final thing. it's intended to help packagers build off. Ahhh, sorry Carsten. I haven't meant you! Replied to the wrong mail, Of course I meant only ilLogict that he may disable this flag in his local gcc specs file... -- MyExcuse: The Usenet news is out of date Martin Zwickel <[EMAIL PROTECTED]> Research & Development TechnoTrend AG <http://www.technotrend.de>
pgpEGQ6FFxwjV.pgp
Description: PGP signature
