Re: [patch] Speed up # of iterations analysis

[Zdenek Dvorak <rakdver at atrey dot karlin dot mff dot cuni dot cz>]

Heello,

> > > Does your analysis still properly determine that the second loop has 50
> > > iterations with that removed?
> > 
> > no.
> > 
> 
> I don't think we should be reducing the number of loops we can determine
> the count for without a *ton* of benchmarks.
> 
> A lot of advanced optimizations we are starting to implement
> (parallelization, better data dependence) depend heavily on having
> countable loops to be able to either generate good code (vectorization),
> or work at all (data dependence).
> 
> Especially if we are only saving 1% compile time.

1% compile time is a lot (it is more than most of the optimizations
passes needs).  But OK, I will just propose moving this analysis to
-O3, at least for now.

Zdenek