This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Fwd: IPA branch and IPA MOD/REF projects?


Forwarding back because i hit reply instead of reply-to-all

---------- Forwarded message ----------
From: Daniel Berlin <dberlin@dberlin.org>
Date: Sep 21, 2006 11:22 PM
Subject: Re: IPA branch and IPA MOD/REF projects?
To: Mark Mitchell <mark@codesourcery.com>


On 9/21/06, Mark Mitchell <mark@codesourcery.com> wrote:
Danny, Jan --

How (if at all) to the IPA branch and IPA MOD/REF projects interact?  In
particular, does the MOD/REF code use any of the functionality from the
IPA branch

No, it doesn't.


? If not, should it?

So, doing on SSA would make it more flow sensitive than it is now, but most of the symbols it is going to answer questions about are symbols that we don't rename into SSA anyway (globals, etc). Thus local SSA will give some precision improvement, but not much.

That said, the analysis phase, since it is based on the same
constraint analyzer as points-to analysis (I actually haven't updated
the wiki page, but this part is already done ahead of schedule), knows
how to deal with both SSA and non-SSA.

The first IPA analysis I expect to get real benefit from IPA + local
SSA would be an IPA points-to analyzer, because it would be able to
answer the questions we currently compute locally, and recompute
multiple times.  It could also do the analysis part  incrementally if
we still wanted to do the recomputation we do now.

Because of Google's lack of an assignment, my time to explore this has
been limited to outside of work (and I'm currently concentrating on
the 4.2 p1 aliasing regressions we have).


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]