[LAU] yoshimi battles the Intel Atom N270

[cal]

Poor yoshimi's been in quite a bit of bother over the last few weeks.
Having recently defended the stability projections for 0.060.1, it was
a bit embarrassing to learn that it might well eventually prove "stable",
but it was producing 3 or 4 incorrectly pitched notes per octave on some,
but not all 32 bit builds. 64 bit builds were fine.

With some tweaking, ie reverting a couple of clever little "optimisations",
we got it to the point where it was mostly fine, but not on eee netbook
thingies with an Intel Atom N270 cpu. Now that was a merry little chase,
still is. As things stand, I added to the signal handler, clearing all
floating point exceptions on receipt of SIGFPE. So far I've had one test
report that it's now pitching accurately on an Atom N270. I don't have
access to one, so more testing would be welcome. Even if it holds good
with further testing, I figure that's just a workaround, not quite a fix.

Beyond that I'd welcome theories or dissertations on just what the hell I
might be dealing with here.

cheers, Cal