The definitive answer is always to check the .cnf file for the relevant benchmark.

You can do this trivially by pulling the relevant cnf file out of the dacapo jar:

   % unzip dacapo-9.12-bach.jar cnf/avrora.cnf
   Archive:  [...]/dacapo-9.12-bach.jar
     inflating: cnf/antlr.cnf           

or grab cnf files for all benchmarks:

   %  unzip dacapo-9.12-bach.jar "*.cnf"

Then you can see precisely what command line arguments are passed to the benchmark for each of the size settings (as well as other info about the benchmarks).   For example, if you look at sunflow.cnf, you'll see that in each case the benchmark gets just an integer argument, which is 32, 256 and 512 for the small, default and large cases respectively.  If you look at fop.cnf you'll see that there are only small and default cases, and these respectively get fop to process the "readme" and "test" test files.  And so on...


From: Zhijia Zhao [zzhao@cs.wm.edu]
Sent: Saturday, June 22, 2013 12:55 AM
To: dacapobench-researchers@lists.sourceforge.net
Subject: [dacapobench-researchers] Input differences between three inputs (small, default and large)

Hi All,

I would like to know the how the three inputs are different from each other. 

After reading the Dacapo website, I found this: 
"For some benchmarks default and large workloads are identical"

But how different they are between the small input and the default one?

If you have any reference related, that would be great!