[Soot-list] Why Spark in soot 2.2.4 slows down

[Chris Pickett]

Hi, just to chime in here... Rich and Clark and I are submitting a 
camera ready copy in the next week or so that uses Spark and the 
reviewers asked us to include compilation times... so if/when this gets 
fixed, please say something on the list, as we'd like to rerun the 
transformations.  (Yes, maybe I should be fixing it if it matters to me.)

Chris

Eric Bodden wrote:
> Ondrej could you do the same run with hprof enabled? I think with such
> a significant difference, this should easily show up in a profile.
> 
> Eric
> 
> On 22/06/07, Ondrej Lhotak <olhotak at uwaterloo.ca> wrote:
>> Strange, I also notice a big slowdown.
>>
>> Also, Soot 2.2.4 gives significantly different answers than 2.2.3.
>>
>> If I use the pre-jimplify switch, which separates Spark time from
>> Jimplification time, I get the following Spark times on soot-c:
>> 2.2.3:  15.3s
>> 2.2.4: 113.9s
>>
>> Something *very* weird is going on.
>>
>> I'm not aware of any recent changes to Spark. Has anyone been working on
>> Spark?
>>
>> I would like to do a binary search through the Subversion revisions
>> between releases 2.2.3 and 2.2.4 to determine the cause of this.
>> Unfortunately, the internet is dead here at the moment, so I can't do
>> that right now (and who knows when this e-mail will even go out). I'll
>> be travelling all next week, so I can't investigate the problem until
>> later. If someone manages to track it down, I'd be interested to hear
>> what the cause is.
>>
>> On Thu, Jun 21, 2007 at 08:39:25AM -0400, Eric Bodden wrote:
>> > Nope. No such observation here. 50s sounds pretty fast to me. Are you
>> > sure it's doing the right thing? If so, maybe could you email us you
>> > test case and command line?
>> >
>> > Eric
>> >
>> > On 21/06/07, Joe Qian <owqian at gmail.com> wrote:
>> > >Hi everyone,
>> > >
>> > >       Recently I moved from soot 2.2.3 to soot 2.2.4. But when 
>> testing old
>> > >programs (also with JDK1.4.1), i found the Spark points-to analysis 
>> run
>> > >much slower than before.  For my small test case, the old version 
>> of spark
>> > >run about 50s, while the new spark version have to run about 200s.
>> > >I have no idea about this situation. Does any one else have similar
>> > >experience?
>> > >
>> > >       Thanks
>> > >
>> > >Joe
>> > >_______________________________________________
>> > >Soot-list mailing list
>> > >Soot-list at sable.mcgill.ca
>> > >http://mailman.cs.mcgill.ca/mailman/listinfo/soot-list
>> > >
>> > >
>> >
>> >
>> > --
>> > Eric Bodden
>> > Sable Research Group
>> > McGill University, Montréal, Canada
>> > _______________________________________________
>> > Soot-list mailing list
>> > Soot-list at sable.mcgill.ca
>> > http://mailman.cs.mcgill.ca/mailman/listinfo/soot-list
>> >
>>
> 
>