Re: [tree-ssa] [PATCH]: Loop invariant code motion for trees

[Daniel Berlin <dberlin at dberlin dot org>]

On Oct 18, 2003, at 10:13 PM, Diego Novillo wrote:

> On Sat, 2003-10-18 at 22:05, Daniel Berlin wrote:
>
> > The fact that we want to perform sinking also is not a good reason not
> > to accept the code. It's not like it wasn't written in a way that
> > sinking isn't allowed, it's just not performed.
> IIRC, there were two things I had problems with: (1) the hack to work
> around some cfgloop limitations,

Which is necessary currently for *any* loop stuff.

I believe Zdenek either has the same hack in the tree-ssa-cfg branch, 
or he made the code able to make forwarder blocks, but I can't remember 
which and i'm not up to looking it up ATM.

> and (2) another hack that would
> explicitly look for unrenamed variables inside a statement.  At the
> time, I thought that #2 should be unnecessary, but it's been a while.

Errr, how else would we detect these statements?
At the time, as i pointed out in follow ups, has_volatile_ops didn't 
cut it and was causing bootstrap failures where just ignoring 
statements with unrenamed variables worked fine everywhere 
(bootstrapped and regtested on three architectures).

> Could you post the patch once again?
Sorry, as I've said, i don't have a tree with it anymore.
I SPEC'd it and it wasn't a noticeable gain, so it wasn't worth arguing 
about.
>