Speculative prefetching: report

[Jan Hubicka]

> >  Before pushing it to mainline I would like see the
> >quadratic bottlenect of bsi_for_stmt resolved.  I don't know where the
> >discussion surrounding Zdenek's patch went but I guess without that we
> >can fix it by keeping around BSIs instead of direct statement pointers
> >in the interface that is straightforward change.
> 
> No, I tried that and could not get it to work.  The problem was that 
> once
> you've expanded one tree nearby stored bsi's become useless.

Hmm, now I remember Zdenek mentioning it...
OK, I guess it is best to keep the code as it is and wait if we get
better infrastructure for this later.  I don't see how to easilly
reorganize the code to not depend on having pointers on several places
in insn stream...

Honza
> 
> Besides, storing 3-word bsi's all over the place is not a good thing.