Re: Fw: [PATCH] ipa aliasing analysis

[Steven Bosscher <stevenb at suse dot de>]

On Monday 02 May 2005 09:04, Olga Golovanevsky wrote:
> On Monday 1 May 2005 13:59, Steven Bosscher wrote:
> > > http://www.cs.wisc.edu/~cs701-1/PAPERS/CooperKennedyPOPL89.pdf
> >
> > Unfortunately this link is broken :-(  I can't find another PS/PDF
> > reference...
>
> Sorry, old link. I did not find neither. Here is my copy:
>  (See attached file: FastIPAACooperKennedy.pdf)

Thanks!

> > I don't understand what you mean here, are you saying that alias info
> > is calculated but not used yet?
>
> Yes, exactly. It's calculated but did non saved for use of
> subsequent optimization passes. The reason is that
> I have not found correct way to save it. Do you know one?

Hm, no :-(

Say you'd have something like this:

void foo (int *x, int *y)
{
  int i;
  for (i=0; i < 10; i++)
    x[i] = y[i];
}

In tree-ssa-alias.c, add_pointed_to_expr sets the SSA_NAMEs for x
and y to points-to-anything immediately the first time it sees them.
I don't see how you could explain to it that x and y may point to
anything but not to each other.

The tree AA guru's are Diego and Dan, maybe they have ideas...


> So actually we have two possibilities here: to have three
> separate files for ipaa, ipcp and ipa, or, alternatively, to have
> all functionality in one file (what we have right now).
> What would you prefer?

I would prefer three files.

Gr.
Steven