Re: i386 extended vs. double precision math

[naddy@mips.inka.de (Christian Weisgerber)]

Bruce Fletcher <bruce@homeperformance.com> wrote:

> This is the section of code which I am having trouble
> with.  I don't know what platform this was written for,
> but apparently fpu_control.h is supposed to be a system
> header.
> 
>   /* FPU control. We MUST have only double precision
>     (not extended precision) */ 
>   #include <fpu_control.h>

Linux.

>   static fpu_control_t fpu_round_double =
>         (_FPU_DEFAULT & ~ _FPU_EXTENDED)|_FPU_DOUBLE;
>   static fpu_control_t fpu_init;
>   #define FPU_ROUND_DOUBLE  {_FPU_GETCW(fpu_init); \
>           _FPU_SETCW(fpu_round_double);} 
>   #define FPU_RESTORE       {_FPU_SETCW(fpu_init);}
> 
> Any idea how to get this running?

On FreeBSD/i386 this could be handled by fpgetprec(3) and fpsetprec(3),
however those seem to be missing from OpenBSD.  You'll need to
implement those yourself through asm{}.  That's ugly, but we're
firmly in MD territory here anyway.  You can probably get all that's
needed from the definition of the FreeBSD macros:

http://www.freebsd.org/cgi/cvsweb.cgi/src/sys/i386/include/floatingpoint.h?rev=1.10&content-type=text/x-cvsweb-markup

-- 
Christian "naddy" Weisgerber                          naddy@mips.inka.de