Re: [PATCH] local-factoring on TREE-SSA

[Daniel Berlin <dberlin at dberlin dot org>]

> (Personally I think this whole transformation should not be done at the
> tree level, but instead on RTL.  What happened to that idea??)

Even if you decide the tree level is right, you'd be better off doing 
it in GVN-PRE anyway, where we have the infrastructure for doing it.

For example, GVN-PRE can already tell you what expressions have what 
values (and thus, whether two basic blocks compute the same values or 
contain the same statement), and where you can move them to.
Hoisting and sinking is thus a simple task, as the only thing you need 
to do is decide which you want to factor.

Looking through the patch, it looks like a straight conversion of some 
rtl pass to try to make it work on tree-ssa, without actually seeing 
whether it made sense to do it that way on an SSA form.
There has been much written about code hoisting and sinking in general, 
and code hoisting is very different (and much easier) on an SSA form 
than it is on a non-SSA form  like RTL.

I'd seriously suggest you take a second look about how you are going 
about doing this hoisting and factoring if you are going to do it on an 
SSA form.

--Dan