Autoparallelization

[Zdenek Dvorak]

Hello,

> Zdenek Dvorak wrote on 10/03/06 09:44:
> 
> > I thought about this and considered it a bit easier to start by
> > targetting libgomp directly, at least for the initial implementation.
> >
> OK, but it's really not the right direction.  If there are things to fix
> in the OMP IL, then we will fix them.  Otherwise, we will fall in the
> usual trap of a misdesigned implementation that lives on because
> it-works-and-who-really-has-time-to-fix-it-up.

in the IL, there is nothing really missing.  Only a massive rewrite of
omp-low.c would be necessary, to make it all work on SSA and probably
also preserve loop structures.

I still do not like this approach -- it implies extra
compile-time cost of running another omp_lower pass after
parallelization, and forces us to modify lot of code that
parallelization does not really need.  Why do you prefer
this to just sharing the parts that are used by both omp lowering
and parallelization?

Zdenke