Re: Merging mem-ssa into mainline

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

Hello,

> The branch has accumulated many changes over the last few months and
> there are several other changes that are not even in the branch.  In
> this message I will describe all the changes that are coming over.  I
> will be send the patches separately, but since all of them rely on
> infrastructure modifications to virtual operands and aliasing
> representation, it's best if they are committed in a single operation.

what about the changes to the optimizers that seem to need them (PRE,
store motion)?  It might be better idea to commit at least those
separately before.

Also, there are some changes below that IMHO would be more appropriate
to merge separately before the rest of the changes, to make the rest
easier to understand and follow:

> Removal of V_MUST_DEF
> ---------------------

This seems like mostly mechanical rewriting lot of "VMUSTDEF | VMAYDEF"
to "VDEF" on many places.

> New push/pop interface for modifying statements
> -----------------------------------------------

Similarly, most likely rewriting lot of update_stmt to a new form.

And some changes that look basically independent on mem-ssa, and
might be better merged either independently on it or after the merge
of the rest of mem-ssa (depending on whether they use mem-ssa
infrastructure significantly or not):

> DSE for aggregates
> ------------------

> Cleanups and speedup in alias analysis
> --------------------------------------

I know such a separation is more work for you, but I think it will
make it much easier to fix the breakages than if you commit the whole
bunch of changes at once.

Zdenek