I rebuilt the whole tree with -mtune=nocona, and I
it seems like this may be the way to go (instead of, for example,
-march=nocona) for EMT64 processors.  I have to run some benchmarks,
but what inspired me to try this was looking at the table in
override_options() in dist/gcc4/gcc/config/i386.c.

I also think we should build the kernel with all of the -mno-sse,
-mno-mmx etc. flags present.

Any thoughts?

--Blair