Re: Limiting whole-program analysis

[Venkatesh Prasad Ranganath <vranganath@sbcglobal.net>]

Ondrej Lhotak wrote:

> On Thu, Jun 03, 2004 at 12:46:22AM +0200, Helge Jensen wrote:
>  
>
> > Helge Jensen wrote:
> >
> >    
> >
> > > Helge Jensen wrote:
> > >
> > >      
> > >
> > > > What should I do? where is it described what I should do to stop the 
> > > > transitive traversal? any hints appreciated.
> > > >        
> > I'm wondering about the performance of the Points-to analysis, do these 
> > timings (on a P4M-1700Mhz, 1G ram) seem reasonable to you?
> >    
>
> Yes, these are consistent with what I'm seeing.
>
> I tend to run Spark with the class library from the 1.3.1 JDK, which in
> this case ends up running about 6 times faster. I'm not quite sure what
> it is about the 1.4 class library that takes so long.
>
> Profiling the code, I see that over half the time is being spent
> resolving virtual method calls.
>
> Ondrej
>
>  
My understanding based on CC03 paper (on SPARK) is that SPARK is used to 
calculate Points-to information and construct a precise call graph, if 
configured.  So, when the below numbers indicate SPARK takes 140+ 
seconds, I am wondering if it's just to calculate points-to information 
(as in variable X points to M, N, and O abstract allocation objects) or 
is something more being calculated?  If it's just points-to information 
then is there way of viewing points-to information to compare it with 
information generated from other points-to analysis implementation?  
Also, is there a way to view call graph information for purpose of 
comparison?

waiting for reply,

> > Analyzing the program "A.java":
> >
> >   public class A { public static void main(String[] args) { } }
> >
> > I get the outputs:
> >
> > $ java -Xmx400m soot.Main -w -p cg safe-forname:false -p cg.spark 
> > on,field-based:true -app A -src-prec java
> > Soot started on Thu Jun 03 00:34:26 CEST 2004
> > [Call Graph] For information on where the call graph may be incomplete, 
> > use the verbose option to the cg phase.
> > [Spark] Pointer Assignment Graph in 6.8 seconds.
> > [Spark] Type masks in 0.8 seconds.
> > [Spark] Pointer Graph simplified in 0.0 seconds.
> > [Spark] Propagation in 146.1 seconds.
> > [Spark] Solution found in 146.1 seconds.
> > Transforming A...
> > Soot finished on Thu Jun 03 00:37:05 CEST 2004
> > Soot has run for 2 min. 39 sec.
> >
> > $ java -Xmx400m soot.Main -w -p cg.spark on -app A -src-prec java
> > Soot started on Thu Jun 03 00:40:30 CEST 2004
> > [Call Graph] For information on where the call graph may be incomplete, 
> > use the verbose option to the cg phase.
> > [Spark] Pointer Assignment Graph in 10.3 seconds.
> > [Spark] Type masks in 1.3 seconds.
> > [Spark] Pointer Graph simplified in 0.0 seconds.
> > [Spark] Propagation in 151.6 seconds.
> > [Spark] Solution found in 151.6 seconds.
> > Transforming A...
> > Soot finished on Thu Jun 03 00:43:24 CEST 2004
> > Soot has run for 2 min. 54 sec.
> >
> > --
> > Helge
> >
> >    
>
>  


--

Venkatesh Prasad Ranganath,
Dept. Computing and Information Science,
Kansas State University, US.
web: http://www.cis.ksu.edu/~rvprasad