Some time ago I ported sparc64 strlen to sparc.  It's fatter: 144
bytes vs 28 bytes of the current one.  It is faster then the old one
for strings of 5 bytes that starts word aligned and faster for strings
of 8 bytes and longer regardless of alignemnt (actual numbers variy
+/-1 from cpu to cpu).  It's about 2x faster for strings that are
about 20+ bytes long.  Asymptotically it's about 2.5x faster.

The new strlen and a small profiling harness are at:

    http://snark.ptc.spbu.ru/~uwe/volatile/strlen.tgz

I wonder if it's worth comitting.

SY, Uwe
-- 
uwe@ptc.spbu.ru                         |       Zu Grunde kommen
http://www.ptc.spbu.ru/~uwe/            |       Ist zu Grunde gehen