[Linux-ia64] Itanium gets supercomputing software

Greg Lindahl lindahl at keyresearch.com
Thu Apr 17 16:33:00 EDT 2003

On Thu, Apr 17, 2003 at 09:56:48AM +0200, Alan Scheinine wrote:

>    I do not think there was a promise that getting efficiency would
> be easier with EPIC.  My understanding of the situation is that
> the logic of dynamic allocation of resources, that is, the various
> tricks done in silicon, could not scale to a large number of
> processing units on a chip.

That's not what I said. I said that getting more instructions per
cycle was what was supposed to be easier, and indeed, that means more
compiler complexity.

> Fifteen years ago I heard a talk in which
> it was claimed that compiler advances developed at universities
> arrive in commercial compilers after a delay of ten years.

That's an over-generalization. For example, a lot of compiler research
is done on the framework provided by Open64, which is SGI's compiler.
You can get research frameworks in which you can play with a
particular optimization idea, but if you want a research framework
which is already a really great compiler, Open64 is the only choice.

> Greg Lindahl wrote that "The ones [compiler people] I know hate EPIC
> with a passion".  Why?

They think it's a pig with lipstick.

> Do they say that the concept is wrong or
> is that problem that they cannot meet their deadlines because of
> the quantity of analysis that has been moved from the silicon to
> the compiler writer?

The lack of uptake in the marketplace meant that they had a couple of
extra years to do the compiler work, so deadlines weren't a problem.

Now in comparison, x86 chips are not very good compilation targets
either: trying to figure out how x86 instructions actually work after
they are translated into some unknown micro-ops isn't exactly easy.
But I suspect that a poll of compiler people would vote for x86 over

> By the way, it may be a good idea to develop more packages like
> Atlas and FFTW which optimize themselves based on the actual computer,
> since memory latency and other factors are variable.  But then,
> optimizing through experimentation takes a long time.

It is a good idea, but it's worth pointing out that Atlas and FFTW
work best on machines which are either out of order, where a bad
compiler isn't a problem, or on in-order machines with great
compilers. I'd love to hear about how well gcc does on ia-64 with
Atlas or FFTW.


Beowulf mailing list, Beowulf at beowulf.org
To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf

More information about the Beowulf mailing list