More on gimplify:create_tmp_var

[Richard Henderson]

On Wed, Jun 09, 2004 at 08:28:40PM -0400, Richard Kenner wrote:
> Also note that we have to able to propagate this down to RTL in MEM_EXPR, so
> we need to keep the chain intact.

Not necessarily.  The tree-level optimizers have *significantly*
better variable-level aliasing information to draw from.  Stuff
that can't be represented in MEM_EXPR at all.  In order to preserve
this at the rtl level, we're considering a partitioning scheme
that can pass down *everything* via alias sets.

In short, I'm wouldn't worry about it.



r~