Re: GCC Status Report (2004-10-26)

[Mark Mitchell <mark at codesourcery dot com>]

Andrew MacLeod wrote:

> On Tue, 2004-10-26 at 09:27, Andrew MacLeod wrote:
>  
>
> > On Tue, 2004-10-26 at 03:28, Mark Mitchell wrote:
> >
> >        I have been informed that our SPEC benchmark results are looking
> >        pretty good, which matches our independent observations.  However,
> >        there are apparently regressions in perlbmk and mgrid.  Andy MacLeod
> >        has a potential fix for the perlbmk problem here:
> >        
> >          http://gcc.gnu.org/ml/gcc-patches/2004-10/msg00488.html
> >        
> >        I intend to review this patch, if nobody else gets to it.
> >        
> > I considered it a new optimization, hence didn't submit it for 4.0
> > review.  I've thought about it a couple of times since then, but never
> > got around to sending you a note :-) 
> >
> > It ought to be pretty safe if you want to add it to 4.0, it improves a
> > number of things, not just perl.   Probably should have been done a long
> > time ago, I just never thought of this solution before for some reason.
> > :-)
> >
> > I'll do a mainline version of it today, although I doubt its any
> > different than the tcb version.  I'm not aware of any out-of-ssa
> > variations between tcb and mainline.
> >
> >    
>
> Sorry, Ive been waylaid a couple of days.
>
> I bootstrapped and verified testsuite runs on a mainline build as of
> monday. The original patch applied with no changes. Diego has been
> running TCB with the patch for some time now on linux
> ppc/x86-64/ia64/x86 with no problems I am aware of.
>
> So it ought to be fine for 4.0 if you decide you want it.
>  
This patch is OK, with some fixes to the comments.

+ /* Return TRUE if S1 and S2 are equivilent copies. */

That should be "equivalent".

+    contain the stmnt sequence of copies.  */

s|stmt|same

+   /* If there isn't at least 2 edges, no sharing will happen.  */

s|isn't|aren't

+   /* ensure we have empty worklists.  */

s|ensure|Ensure that

! /* THis function will analyze the insertions which were performed on edges,

s|THis|This

--
Mark Mitchell
CodeSourcery, LLC
(916) 791-8304
mark@codesourcery.com