Avoid use of "register" as optimization hint

[Roland McGrath]

> I'd presumed that the code had been copied in when GMP and glibc were both 
> under the same version of the LGPL (as they were for a long time) and so 
> hadn't involved any special permission; I hadn't realised that there was a 
> relicensing from GPL to LGPL involved when the code was added to glibc.

The beginning of version control history for libc is 1995-02-18, by which
time we already had code that was from a prerelease of what would be GMP
2.0 and that was under LGPLv2 (as was libc at that time).

So further archeology requires finding tarballs from the earlier era (which
are not on ftp.gnu.org any more), or trusting inferences we can make from
old ChangeLog entries, or trusting my memory (which is known to be unreliable).

The initial use of GMP code in libc has a ChangeLog entry from 1993-12-24.
The newest GMP release at that time was 1.3.2, which was under GPLv2.  This
matches my memory of writing scripts to massage GMP sources to replace
their license text.  But I also had something like that for code that lived
in /gd/gnu/lib (an informal predecessor to today's gnulib) such as getopt,
so maybe I'm conflating the memories.


Thanks,
Roland