Re: What classes are analyzed by the points-to analysis?

[Derek Rayside <drayside@mit.edu>]

Jaime,

My experience is that the runtime of soot/spark is dominated by the kind
of output one requests, and not by the analysis it does (this was also
true of my own program analysis tool I wrote a few years ago).  For
example, on a fairly trivial benchmark, my running times for soot/spark
range from 1.5 minutes to 15 minutes, depending on what I ask for.  So try
asking for less output, and see how that runs.  In particular, it is
expensive to dump the pag, solution, and types (basically anything of the
form "dump x").

Also make sure that you're giving soot/spark plenty of memory: otherwise
it will grind away in gc, which can slow things down.  I usually use
something like -Xmx400M.

Good luck,
Derek.



On Mon, 8 Sep 2003, Jaime Spacco wrote:

> How does the Spark points-to analysis decide which classes to analyze?  I heard a
> rumor that it analyses every possible class that could have a static initializer
> called.  Is this true?  And if so, is there any way to limit the points-to
> analysis to just the methods and classes that the classloader needs to compile?
>
> We're getting pretty slow runtimes for a simple hello world-type application and
> we're looking for ways to speed it up.
>
> Thanks,
> 	-Jaime
>