Proebsting's Law

[Hakim Cassimally]

Proebsting's page on this gives some links to other articles,
including one with a defense of compiler research.  To summarize what
I understood (very little), Proebsting's estimate is actually rather
generous to the compiler researchers; but even so, there is value in
the research, because of changing processor technology - an enormous
amount of effort has to be expended simply to keep the status quo in
compiler speed, never mind improving it.

osfameron

On 13/12/05, James Davis <jamesd at jml.net> wrote:
> Simon Wistow wrote:
>
> > I claim the following simple experiment supports this depressing claim.
>
> Is this meant to be a troll by Proebsting? I apologise for falling for
> it if it is, but I've not seen such a badly written argument for a while :-)
>
> > Run the benchmarks both with and without optimizations enabled. The ratio of of those numbers represents the
> > entirety of the contribution of compiler optimizations to speeding up
> > those benchmarks.
>
> False, even with optimizations disabled I'm sure that modern compilers
> don't suddenly turn into the compiler technology of 35 years ago if only
> because languages and target architectures have changed.
>
> > Let's assume that this ratio is about 4X for typical
> > real-world applications
>
> A figure pulled from a hat?
>
> > and let's further assume that compiler
> > optimization work has been going on for about 36 years.
>
> Again, pulled from a hat?
>
> > These
> > assumptions lead to the conclusion that compiler optimization advances
> > double computing power every 18 years. QED.
>
> Given two figures I've shown no basis for I'm claiming something.
>
> > This means that while hardware computing horsepower increases at roughly
> > 60%/year, compiler optimizations contribute only 4%. Basically, compiler
> > optimization work makes only marginal contributions.
>
> I'll admit this is probably true despite the numbers. Code produced from
> a 35 year old compiler run on a modern machine probably runs faster than
> the output of a modern compiler on a 35 year old machine.
>
> > Perhaps this means Programming Language Research should be concentrating
> > on something other than optimizations. Perhaps programmer productivity
> > is a more fruitful arena.
>
> Isn't this is Amdahl's Law (or is at least a corollary to it)?. It does
> also ignore some interesting points.
>
> - If your program runs for 101 minutes and you've rented 100 minutes of
> time on a shared computing resource 4% optimization is very important to
> you.
>
> - You have that 4% optimization available now and it costs you only
> compilation time. You might not wish to wait a few months for faster
> hardware.
>
> - That computer scientists might like to study programming languages and
> compilers for their own sake amongst various other reasons not including
> producing faster code. More insight into computer science might be found
> in an optimization that produces the tiniest of performance improvement
> by a researcher with no interest in programmer productivity.
>
> James
>
>
> --
> http://www.freecharity.org.uk/ - Free hosting for charities
> http://jamesd.ukgeeks.co.uk/
>