[ast-optimizer-branch] Goto/Break/Continue/Switch elimination

[Diego Novillo]

On Thu, 2002-05-02 at 14:16, Richard Henderson wrote:

> I wonder if there's some way to write the tree optimizations such that
> they use an abstract cfg.  If goto elimination was "easy", then we use
> a set of cfg routines that sort-of generate the cfg on the fly as we
> examine it.  Otherwise we do the full graph thing.
> 
The usage I've seen advertised for goto/break elimination passes is
things like parallelizing loops.  We already have a generic CFG builder
that deals with anything you throw at it.

I don't think we should always apply this transformation.  However, it
may happen that for some transformations we want/need to deal with a
simpler CFG.  In those cases we can call the goto elimination pass.  As
usual, we should try to balance the code bloat vs performance vs
compilation speed.


Diego.