Re: scm_bits_t / scm_ubits_t

[Rob Browning]

Marius Vollmer <address@hidden> writes:

> Yeah, it's tempting to use a union, but I think we had performance
> problems with it.  It didn't nearly optimize as well as using an
> integral type.  Or something.

I recall a discussion about that, but if a union would be a Right
Thing, then we might want to check back on the performance every now
and then.  Sooner or later the compilers ought to catch up, unless
there's a Hard Problem here.

Which reminds me.  If we don't have it already, it'd be kinda nice to
start accumulating code (if we can think of any) in a "make benchmark"
target -- or perhaps that should be a separate top-level module.  This
would be good for regression testing.

-- 
Rob Browning <address@hidden> PGP=E80E0D04F521A094 532B97F5D64E3930