Re: Shared m68k packages (Was: CVS commit: src/sys/arch/m68k/m68k)

John Klos <[email protected]> Thu, 3 Apr 2025 08:49:39 +0000 (UTC)
Newsgroups gmane.os.netbsd.ports.m68k
Message-ID <[email protected]>
Hi,

> > What I was after was a uniform approach to all m68k (buggy lc040's included).
> >
> > It makes me wonder even why we build different packages for sun, amiga, x68k,
> > and mac68k also....when they are all m68k :(
> 
> I'm sure jklos@ can correct me if I'm mistaken :-p but in general
> pkgsrc builds share packages across all m68k platforms. If there was
> sufficient additional compute available it is more likely we would
> split builds by processor - eg '020 vs '040.

In my experience playing with -m680x0 flags, there aren't many instances 
where the performance difference is large enough to bother with different 
binaries.

OTOH, if resources weren't limited, -msoft-float compiled packages would 
make LC040s happy.

Or perhaps even better, binaries which have both soft float and floating 
point instructions which are chosen at run time would be good so we could 
have one set of binaries for everyone.

> There are differences mainly in /usr/include which will result in
> small differences in generated binaries, but in general It Does Not
> Matter which m68k platform was used to build the pkgsrc packages.

The m68k pkgsrc binary packages are compiled on a mix of m68060 Amigas and 
m68040 Macs, so any platform based differences because of /usr/include are 
well mixed in. I can't ever remember hearing of anyone having any issue 
running them, though.

> A potentially interesting task would be to reduce those differences
> for most (if not all) m68k userlands - probably starting with merging
> /usr/include and then onto moving the small set of port specific
> binaries to be triggered on m68k rather than the specific port :-p

That'd be good :)

John