23 February 2010
22 February 2010
Proebsting's Law: Compiler Advances Double Computing Power Every 18 Years
Also paper by Nicholas FitzRoy-Dale.
15 February 2010
09 February 2010
There's one easy way to make SpecInt2006 relevant, with
no changes what-so-ever.
Go to the spec site, and instead of asking for the baseline
and result numbers, just query by 403.gcc instead. Voila!
You now have meaningful and comparable results.
That's actually been true for at least the last two decades.
Nobody has ever broken gcc as a benchmark, and it continues
to be the one benchmark that actually gives a good view of
real-life integer performance.
Almost everything else is about small snippets of code
that can be optimized to hell and back.