Optimal Optimizations: A Very Preliminary Report

[Scott Robert Ladd]

Ian Lance Taylor wrote:
> Considering the 18% improvement over -O3, this clearly suggests that
> at -O42 gcc should automatically compile code with several different
> options and compare the results by running them through a CPU pipeline
> simulator...

Give me ASCI Purple, and I'll give it a try! ;) Heck, give me ASCI 
Purple, and I'll rule the world! Whaah-ha-ha-ha-ha!

Ahem.

The actual improvement was over 38%, BTW. Yes, I'm compulsive. ;)

> Perhaps compilation is not the data analysis problem I always thought
> it was, but is rather more of a directed search problem.

Ultimately, I think it is a combination of the two approaches. More than 
one compilation of a function may be efficacious in different 
circumstances within the same program -- and data analysis of decision 
might refine the results from a directed search .

Whether such is practical (or necessary) with current hardware is open 
for discussion.

-- 
Scott Robert Ladd
Coyote Gulch Productions (http://www.coyotegulch.com)
Software Invention for High-Performance Computing