Re: MKSOFTFLOAT for evbppc

[Simon Burge <simonb%NetBSD.org@localhost>]

Mouse wrote:

> > On a P2020, a build.sh distribution for evbppc took 6.2% less time on
> > a softfloat userland .vs. hardfloat userland with kernel-emulation.
> 
> I'm not surprised there's a difference; they're basically doing the
> same thing, except one of them is paying trap-to-the-kernel overhead.
> 
> I'm surprised the difference is that great; a build of the world must
> be doing a lot more floating-point than I realized.  Did you do
> multiple runs to find out how the difference compares to the noise?

I mentioned awk in my previous message.  Every "number" in awk is a
floating point number - there's no concept of integers.  From memory
(and this was over a decade ago!), the awk component of a kernel build
was the significant difference between hardfloat and softfloat.

Cheers,
Simon.