Subject: Re: port-i386/3510: libm build on i386 w/i387 should be easier
To: William O Ferry <WOFerryemail@example.com>
From: J.T. Conklin <firstname.lastname@example.org>
Date: 04/18/1997 11:22:29
BTW, are there any other places in the NetBSD sources that I need to
uncomment? Does this mean I've been using the FPU emulation code all
this time? I don't have MATH_EMULATE enabled in my kernel.
No, you've been using your floating point unit all along. What you
haven't been using is assembly language versions of the core
exponential, trancendental, etc. functions.
I wouldn't worry about it. The assembly language versions were
written by an idiot and the C versions were written by experts who
actually know a thing about floating point and numerical analysis.
The asm versions are known to return sub-optimal results around
Fortunately, the idiot knows about them and hopefully he'll get around
to fixing the problems. Until then, I don't recommend anyone use the