Variable Expansion Optimization

[Sebastian Pop]

On Mon, Aug 16, 2004 at 12:31:08PM +0200, Giovanni Bajo wrote:
> 
> So, it looks like we can get rid of the old loop unroller if we merge the LNO
> strength reduction & prefetch passes for 3.5. Let's see what Mark decides about

One of the reasons Zdenek thought that it was not a good idea to
remove the rtl based loop unroller is because duplicating code earlier
will just add overhead on all the optimizers and expanders.  

This same issue arises when considering the transmission of data
attached to the loops structure to the back-end, for instance the data
dependence graph.  One of the solutions for avoiding to update the DDG
along with the loop unrolling was to have the unroller just before the
construction of the DDG.

So there are pros and cons for having the unroller only on trees.

Sebastian