Re: [tcb] Incremental SSA updates

[Jeffrey A Law <law at redhat dot com>]

On Wed, 2005-03-30 at 14:49 -0500, Kazu Hirata wrote:
> Hi Jeff,
> 
> > If you're not walking the statements, then you're missing roughly 40%
> > of the jump threading opportunities, possibly more.  That's why I say
> > it's pointless to compare against mainline, you need to compare against
> > the patch in the threading meta-bug.
> 
> By the way, what's the status of your patch?  IIRC, you were squeezing
> out a few remaining (latent?) bugs outside DOM?  I am aware of your
> patch, but too much patching and comparisons makes me dizzy.
There's some SPEC weirdness which I want to understand.  I find it hard
to believe that better jump threading is producing a 20% improvement in
PERL for example.  I want to understand that effect as well as a couple
regressions (which may or may not be caused by ivopts instability).

We also need a better limiter -- otherwise the new one will happily
unroll loops when the backedge always leads to a threadable block.
This can lead to compile-time explosions in a couple different ways.


Jeff