I noticed the bad floating point performance several weeks ago. Upon
analysing the code that gcc produces, I found that no transcendental
FPU functions were generated (e.g. fcos, fsincos). The FPU was only
used for multiplication/division. I then got mathlib-2.1-386bsd.tar.Z
and installed it as a replacement for libm.a. Works like a charm.
The FPU performance increased by a factor of 6 for my application.
Only caveat: cbrt (cubic root) is missing in mathlib-2.1-386bsd.tar.Z,
you will have to add it manually. If you omit this, XFree-2.0 won't
compile.


      Lex Wennmacher                   
                                       
Radioastronomisches Institut           email: lwennmacher@astro.uni-bonn.de 
   der Universitaet Bonn                      AIBONN::WENNMACHER     (SPAN)  
     Auf dem Huegel 71                 phone: +49 228 73-3645 
        D-53121 Bonn                   FAX:   +49 228 73-3672 
          Germany                      .esrevinu rorrim a morf eb ot mees uoY  

------------------------------------------------------------------------------