This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
RE: Improving Alias Analysis
- From: "Sanjiv Kumar Gupta, Noida" <sanjivg at noida dot hcltech dot com>
- To: Diego Novillo <dnovillo at redhat dot com>
- Cc: gcc at gcc dot gnu dot org, rth at redhat dot com, Daniel Berlin <dberlin at dberlin dot org>
- Date: Mon, 21 Oct 2002 15:54:14 +0530
- Subject: RE: Improving Alias Analysis
>Again, points-to analysis is what you're looking for. The
>tree-alias-* files on tree-ssa branch will provide pointers to
>literature that covers this problem.
>Diego.
I built the tree-ssa-branch. It has the same problem. That may be
because aliasing ,GVN etc are in TODO list as of now. Also, it
implements Steengaard's algorithm based on store locations modeling
and type inference, which may well work as a tree optimizer but may
not suit my case of changing the exisiting RTL optimizer.
I also want to keep the aliasing improvment changes small and
focused enough to avoid any major design changes.
I am thinking in terms of approaches somewhat similiar to
http://citeseer.nj.nec.com/cache/papers/cs/1318/ftp:zSzzSzftp.cs.arizona.edu
zSzreportszSz1997zSzTR97-13.pdf/debray98alias.pdf
http://citeseer.nj.nec.com/cache/papers/cs/10403/ftp:zSzzSzftp.inria.frzSzIN
RIAzSzpublicationzSzpubli-pdfzSzRRzSzRR-3764.pdf/amme99data.pdf
As I mentioned earlier, one other obvious solution is to back traverse
the RTXs list and finding out if the two pointers differs by different
offsets to a common base. But I am looking for lighter solution.
Any ideas would be a great help!
Best Regards
Sanjiv