Linus Torvalds (torvalds.delete@this.linux-foundation.org) on April 4, 2014 2:00 pm wrote:
>
> I suspect one reason is that most good chips already do so well on denormals that it just
> doesn't matter. I haven't timed it myself, but I thought both Intel and AMD have no penalty
> at all or only a slight slowdown these days on their main cores. So the "it slows things down
> enormously" doesn't even happen on most cores, it only happens on the small ones.
>
> Which makes a really rare problem be something that most compiler developers will never
> even see on the machines they use day-to-day. I can understand why they might not consider
> it a big deal. "Here's a nickel, kid, buy yourself a real computer".
>

That's incorrect.
Recent "big" Intel cores handle some cases of denormals with no penalty, but not all cases and even not all common cases.
It was discussed on RWT forum not long ago, but since the RWT forum lacks the search capability (is not it ironic that the forum called real world technologies is one of the very few forums on the whole www that has no search at all, not even most basic and primitive?) I can't find the reference.

IIRC, of two most common cases, i.e. multiplication of normal inputs that produces denormal result and addition one normal + denormal with normal result, the former is fast and the later is slow. But I can be wrong about exact details.