This is the mail archive of the gcc@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]

Re: denormals/subnormals are heading for extinction


On Thu, 16 Aug 2001 dewar@gnat.com wrote:

> <<I would like to add that even if denormals are usually encounted
> infrequently enough that any ineffeciencies with them are minor, there is
> still a big cost. Implementation complexity. Having to add and waste the
> transistors to deal with them in the first place.
> >>
>
> Well modern chips have millions of transistors to waste these days, so
> that is a weak argument at this stage.
>

Then why are hardware folks so annoyed and duking it and trying to get rid
of them? If it was as easy as you seem to proclaim, then tell me why the
hardware people don't like denorms?


> <<If the gains and costs of denormals are neglegible, leave them out.
> They're a headache you don't want to design for.
> >>
>
> But the gains are by no means negligible. Remember that this subject has
> been duked out with substantial opposition from the hardware folks, and
> the IEEE standard (remember IEEE is a group with substantial hardware
> input) decided that denormals are worthwhile for good reasons. The fact
> that Scott believes otherwise, without good documentation, is not very
> convincing.
>

I am unqualified to judge one way or another. Nor did I.

You'd origionally claimed that you hadn't seen a software program that
was slowed down by slow denormal computation. Which may or may not be
true. But just having denormals in a chip is an extra feature, with a cost
in engineer time and design complexity.

Thus, the decision isn't whether they slow down a program or not, but are
they worth the effort to implement.

You seemed to be pretending that there is no cost to denormals. There is.


Scott



--
No DVD movie will ever enter the public domain, nor will any CD. The last CD
and the last DVD will have moldered away decades before they leave copyright.
This is not encouraging the creation of knowledge in the public domain.




Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]