Re: PATCH C++: speed up templates

[Ziemowit Laski <zlaski at apple dot com>]

On Wednesday, Feb 19, 2003, at 11:36 US/Pacific, Ziemowit Laski wrote:

> On Wednesday, Feb 19, 2003, at 10:52 US/Pacific, Matt Austern wrote:
>
> > On Tuesday, February 18, 2003, at 01:08 PM, Mark Mitchell wrote:
> >
> > > --On Tuesday, February 18, 2003 12:27:42 PM -0800 Matt Austern 
> > > <austern at apple dot com> wrote:
> > > > Is that true even in 3.3?  Remember, it's the 3.3 release branch 
> > > > where we
> > > > found the regression.
> > >
> > > No, it's not going to be true for 3.3 -- which is why I approved 
> > > Mike's patch there. :-)
> >
> > Suppose we take a more radical approach, and just back out your patch 
> > entirely from the 3.3 release branch.  What would be the >> consequences?
>
> In this particular case, the consequences should be nil.  As I 
> understand Mark's November 8 patch, it merely tries to... improve 
> performance by storing the visited tree nodes in a hash table. :-) :-) 
> :-)  So one option is definitely to just back it out (along with the 
> 'htab_delete' fix that was added later).  The other is to use the hash 
> table functionality already supported by walk_tree().  I was not aware 
> of this functionality (and, it would appear, neither was Mark) until 
> Devang pointed it out to me yesterday.  I'm currently running a test 
> utilizing this.  If it offers further performance, I'll offer it as 
> part of the "revert patch" patch.

Ok, I tried utilizing the walk_tree() hash table functionality, and it 
slows things down even more than Mark's original patch. :-) :-)  So 
much for that.

I'll post the revert patch & ChangeLog shortly.  Stay tuned.

--Zem
--------------------------------------------------------------
Ziemowit Laski                 1 Infinite Loop, MS 301-2K
Mac OS X Compiler Group        Cupertino, CA USA  95014-2083
Apple Computer, Inc.           +1.408.974.6229  Fax .5477