Subject: Re: soft-float builds
To: None <thorpej@zembu.com>
From: Simon Burge <simonb@wasabisystems.com>
List: tech-toolchain
Date: 05/31/2001 12:10:38
Jason R Thorpe wrote:

> On Wed, May 30, 2001 at 10:22:08PM +1000, Simon Burge wrote:
> 
>  > What I was thinking was something along the lines of a
>  > 
>  > 	MKSOFTFLOAT=	yes|no
>  > 
>  > make variable, which would do things like add "-msoft-float" to the CC
>  > and CXX (and FC?) variables, and be used in the libc/arch/* makefiles
>  > (or even just libc/Makefile) to enable softfloat support in libc.
>  > 
>  > Any comments on this, or better ways to do it?
> 
> That sounds fine.

Cool.  One "aye" and no "nay"s so far.

>  > This raises a further question about whether or not we should have a
>  > separate MACHINE_ARCH for softfloat-built userlands and what it should
>  > be called (soft<arch>, soft-<arch>, ...).  Any ideas here will be
>  > appreciated...
> 
> Yes, esp. for e.g. PowerPC, where it changes the ABI for e.g. stdarg.  Do
> you have kernel FP emulation for FPU-less PowerPCs?  If so, then you can
> still run FPU-using PowerPC code by emulating the register accesses, but
> that'll be slow...

We have kernel FP emulation, but yes, it's not quite as fast :)  Early
userland work was done with macppc binaries.

> Actually, are there ELF header bits that indiciate "uses FPU"?  I know
> there are for MIPS...

Do you have a reference for that?  We don't seem to have any flags in
our source tree for those (that I can see)...

Simon.
--
Simon Burge                            <simonb@wasabisystems.com>
NetBSD CDs, Support and Service:    http://www.wasabisystems.com/