Re: mips64 softfloat or hardfloat?

Michael <[email protected]> Wed, 5 Apr 2017 01:04:33 -0400
Newsgroups gmane.os.netbsd.ports.evbmips,gmane.os.netbsd.ports.mips.devel
Organization The NetBSD Foundation
Message-ID <20170405010433.25716917@blackbush>
Hello,

On Tue, 4 Apr 2017 11:18:20 -0500 (CDT)
"John D. Baker" <[email protected]> wrote:

> On Tue, 4 Apr 2017, Michael wrote:
> 
> > It's still soft float, although some parts of the tool chain seem to be
> > confused. Adding -Wa,-msoftfloat to CFLGS should do the trick.  
> 
> To be precise, adding:
> 
>   CFLAGS+=	-Wa,-msoft-float
> 
> seems to do the trick.  At present, I have restricted this to
> "pkgtools/cwrappers".  While the assembler warnings were treated as
> fatal for "cwrapper", a subsequent build of "pkgtools/digest" and
> "lang/perl5" reported the warnings but proceeded to build.

Yeah, I keep forgetting the exact number of dashes needed ;)
Either way, hardfloat mostly works, some kernel fp emulation support
for 64bit MIPS was not being compiled right, I fixed that a while ago
( you'd get SIGILLs on trunc.* instructions ). My main problem is how
to convince libc and friends to build softfloat support for 128bit
floats but use hardware otherwise ( like almost all other archs do ).
Building at least some pkgsrc stuff with -mhard-float actually works
even with a softfloat userland.

have fun
Michael