Re: [autovect] [patch] and aliasing question

[Ira Rosen <IRAR at il dot ibm dot com>]

Daniel Berlin <dberlin@dberlin.org> wrote on 25/05/2005 15:59:01:

> > But there is no flow-sensitive alias information (as I understand from
> > alias dump files).
> For the pointers you are concerned about?

There is no flow-sensitive information at all in alias dump files
(attached).

>
> >  And according to type memory tag, the pointer "points-to
> > anything".
>
> Which does?
>
> TMT's still have a may-alias set, and i'm pretty sure we don't have
> TMT's that point to anything.
>
> I'm going to need to see dump files :)

vect-77v.c (attached)
ib_16 doesn't have a TMT

>
> >  Can I somehow apply flow-sensitive alias analysis?
>
> Rerun may_alias before the pass.
>

But will it give me flow-sensitive info?

>
> >
> > Another problem is that not all symbols have a type memory tag, e.g.,
> > symbols inserted by PRE
> Only pointers have type memory tags, and PRE shouldn't be inserting
> pointer dereferences right now, so it really shouldn't matter, unless
> there are some weird corner cases i'm not thinking about.
> Again, do you have some testcases/dumpfiles where PRE is creating
> something that needs a TMT?
>
vect-101.c (pre and alias dumpfiles attached):
pretmp.13 is a pointer.


(See attached file: vect-101.c.t46.pre)(See attached file: vect-77v.c)(See
attached file: vect-77v.c.t44.alias5)(See attached file: vect-101.c)(See
attached file: vect-101.c.t44.alias5)

Attachment: vect-101.c.t46.pre
Description: Lotus Freelance

Attachment: vect-77v.c
Description: Binary data

Attachment: vect-77v.c.t44.alias5
Description: Binary data

Attachment: vect-101.c
Description: Binary data

Attachment: vect-101.c.t44.alias5
Description: Binary data