[llvm-commits] building a Dependence Graph

[Preston Briggs]

On Wed, Oct 17, 2012 at 6:47 PM, Hal Finkel <hfinkel at anl.gov> wrote:

> ----- Original Message -----
> > From: "Preston Briggs" <preston.briggs at gmail.com>
> > To: llvm-commits at cs.uiuc.edu
> > Sent: Wednesday, October 17, 2012 7:04:20 PM
> > Subject: [llvm-commits] building a Dependence Graph
> >
> >
> > Attached is a small patch that begins the definition of a new
> > function pass, DependenceGraph, that will build a dependence graph
> > for an entire function.
>
> What will time time/space complexity of this be?
>
> Thanks again,
> Hal
>
>
In the worst case, O(n^2), where n is the number of memory references
(loads, stores, calls), but that's very pessimistic. I discuss many ways to
reduce the time/space impact of the dependence graph
here<https://sites.google.com/site/parallelizationforllvm/building-the-dependence-graph>
.

We did something similar for the Tera compiler, 20 years ago, running on
workstations of the day, and it was practical. Nevertheless, in the context
of LLVM, I expect it will an expensive analysis, perhaps only suitable for
certain demanding applications.

Preston


> > I'm trying to sneak up on the problem in
> > stages, rather than blatting out one big chunk of code, hoping to
> > make it easier to review.
> >
> >
> > This new pass depends on the DependenceAnalysis pass to answer
> > questions about the dependence between pairs of instructions. There
> > are two separate passes because there's a useful separation of
> > concerns. A dependence exists if two conditions are met:
> >
> >
> >     1. Two instructions reference the same memory location, and
> >     2. There is a flow of control leading from one instruction to the
> >     other.
> >
> >
> > DependenceAnalysis attacks the first condition; DependenceGraph
> > attacks the second.
> >
> >
> > Someday it may be useful to write a different pass, called something
> > like LoopDependenceGraph, that would build a dependence graph for a
> > single loop nest. Such a pass could still rely on the
> > DependenceAnalysis pass.
> >
> >
> > Thanks,
> > Preston
> >
> >
> > _______________________________________________
> > llvm-commits mailing list
> > llvm-commits at cs.uiuc.edu
> > http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
> >
>
> --
> Hal Finkel
> Postdoctoral Appointee
> Leadership Computing Facility
> Argonne National Laboratory
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20121017/ac2f9b78/attachment.html>