[prev in list] [next in list] [prev in thread] [next in thread] 

List:       gcc
Subject:    Re: cache misses in gcc 3.3
From:       Daniel Berlin <dberlin () dberlin ! org>
Date:       2003-02-11 0:40:07
[Download RAW message or body]



On Mon, 10 Feb 2003, Richard Henderson wrote:

> On Mon, Feb 10, 2003 at 10:53:17AM -0700, law@redhat.com wrote:
> > While I expect there'll always be some constants or expressions that
> > only gcse will catch (specifically those created by the lowering process),
> > I expect that we can drop the memory tracking bits from gcse and maybe
> > simplify other stuff (like constant/copy propagation).
>
> Indeed, I would suspect that if there's much more to be gained except
> for address expressions exposed by lowering (particularly by stuff like
> symbol_ref -> high+lo_sum), then the tree-ssa optimizers aren't doing
> their job.
>
> I guess we'll see whether that guess is correct when we get that far.

Well, we currently don't lower array accesses + field accesses to pointer
arithmetic at the tree-ssa level.
We should do it after loop opts so the lowered expressions are easily
visible to SSAPRE.

>
>
>
> r~
>
[prev in list] [next in list] [prev in thread] [next in thread]