[Soot-list] Dacapo - Avrora Performance

[Steven Arzt]

Hi Naveen,

 

Can you run a profiler on the original code and on the transformed code to
see where the performance penalty comes from? Or decompile some of the
"worst" files and diff them? I would be highly interested in what happens
there and why things get so much slower. It would be great if you could do
some investigation there.

 

Best regards,

  Steven

 

Von: soot-list-bounces at CS.McGill.CA [mailto:soot-list-bounces at CS.McGill.CA]
Im Auftrag von Naveen Neelakandan
Gesendet: Mittwoch, 28. Oktober 2015 09:49
An: soot-list at CS.McGill.CA
Betreff: [Soot-list] Dacapo - Avrora Performance

 

Hello,

 

I am seeing a high performance penalty (~15% - 20%) for the Dacapo Avrora
benchmark when passed through soot.  Note that I am not doing any
transformations of my own in this case and this behaviour appears to happen
only for this benchmark. I was wondering if anyone else had a similar
experience with this benchmark?

 

The command that I am running to compile the benchmark through soot is

 

java -d64 -Xmx6G -cp soot-trunk.jar soot.Main -pp -w -allow-phantom-refs -cp
$compile_class_path -process-dir $process_dir -main-class Harness Harness
org.dacapo.harness.Avrora

 

where $compile_class_path and $process_dir refer to the same directory where
the contents of avrora-cvs-20091224.jar are extracted to.

 

Thanks,

Naveen

 

 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: https://mailman.CS.McGill.CA/pipermail/soot-list/attachments/20151030/bb0b8e72/attachment.html