• Home
• Projects
  • 3D Ram Design
  • Gleipnir
  • Gleipnir how-to
  • DineroIV how-to
  • DineroIV MultiCore
  • Scheduled Dataflow
• Other Project
• SimpleScalar
• Past Projects
• Retargetable Compiler for Hybrid Architectures
• Personnel
• Publications

DineroIV How-To Guide

DineroIV is a trace driven uni-processor cache simulator developed by Jan Edler and Mark D. Hill.

The simulator remains open source and free for redistribution for academic purposes. Please contact the authors regarding commercial use. Our modifications extend the original trace format to include the additional trace attributes provided by Gleipnir.

 contact: tjanjusic [at] unt [dot] edu 

Installing DineroIV

Get the latest version using subversion
username: guest, password: guest2011

 $ svn co http://csrl.unt.edu/svn/tools/DineroIVs/DineroIV_mod1 dineroIV \
--username=guest

cd into the directory and build it...

 $ cd dineroIV;  make; 

Run it

 pretty simple :) 

Running DineroIV

Running dinero is straightforward; however, take a minute to look at the -help output. It gives an idea of dineroIV's options.
Most of the cache options are highly customizible. The more common options are cache size, line size and associativity.

 
#!/bin/sh

#Script to run valgrind - perl_script - dinero

benchmark='/home/janjust/valgrind_test/tests/mibench'

dinero_param1='-l1-dsize 32768 -l1-dbsize 32 -l1-dassoc 4 -l1-drepl r -informat d'
dinero_param2='-l1-dsize 32768 -l1-dbsize 32 -l1-dassoc 1 -l1-drepl r -informat d'

p_dinero='/home/janjust/janjust_svn/DineroIV_modified'
script_path='/home/janjust/research/svn_janjust'

echo Start of Script

$p_dinero/dineroIV $dinero_param2 < ~/research/test_traces/testapp.valgrind.out > testapp.dinero.out

echo End of Script		
		

Dinero will produce an output file that contains access statistics for each function and each variable within that function. These statistics are used for visualizing the access behavior.

Visualizing the Behavior

Currently we do not have a GUI to visualize the access statistics; however, we use a set of scripts to convert dinero output into a GNUplot script. Final result is a graphical representation of total cache behavior per each function (see below).

The following figures depict the cache behavior for a variety of functions from the mibench benchmark suite. The y-axis represents the number of hits and misses on a particular cache line (x-axis). It becomes obvious that certain cache lines remain highly underutilized.