In article <200802151025.LAA0000059517%zel459.zel.kfa-juelich.de@localhost>, Matthias Drochner <M.Drochner%fz-juelich.de@localhost> wrote: > >agrier%poofygoof.com@localhost said: >> it is likely that the kernel floating point emulation is fine. > >It _was_ fine -- it was removed about a month ago. > >> the other >> alternative is to limit gcc to only generate floating point that the >> kernel can emulate > >This is how it used to be. There is the "-mno-fancy-math-387" >gcc option for exactly that purpose. It was set per default in gcc's >netbsd-aout configuration. Likely just an oversight that it got lost >in the transition to ELF. (see the MASK_NO_FANCY_MATH_387 flag >in gcc config files) > >best regards >Matthias I think that we should not penalize the majority of CPU's to support the few that need emulation. Building a soft-float userland seems to be the right solution because I don't think that anyone is planning to implement fsqrt and friends in the kernel. christos