18-548/15-548
Tools Information

Carnegie Mellon University, Fall 1998

• Dinero -- cache simulator
• Atom -- instrumentation tool
• Hennessy & Patterson tool set

Dinero

We are using a modified copy of dinero. Specifically, this dinero is not limited to 32-bit data lengths, and can process 64-bit address traces when run on a 64-bit processor.

Documentation:

• Browseable Man Page
• Programmer notes -- doc.h

Program Files:

• Executable for Alpha
• ... or use dinero.tar to perform an installation of your own
• Dinero source directory
• Source file for man page: dinero.1 and formatted man page dinero.cat

Dinero Example:

• Example dinero trace input file (note that comments are permitted after the trace address on each line).
• Execute this script to see if dinero is running (assumes that there is an executable called "dinero" in the src directory, which will the case if you've done a "make" from within the src directory).

Processing Scripts:

• Generic Dinero output processing script in Perl (proc_dinero.pl). To use, capture a set of dinero runs in a single output file, then take that output file and feed it to this Perl script -- you will get one line of output for each dinero run with the conditions of the run, cache miss rates, and traffic ratio. Edit the print statements to suit (you don't really need to know any Perl to do this). To execute the Perl script, just chmod +x on the file and treat it as you would a shell script.
• A U-cache size & U-cache miss ratio example of the Perl script (proc_umiss.pl)

Multi-Level Caches & Dinero

• Dinero can be used to simulate multi-level caches with appropriate use of pipes and an intermediate filter to process bus-output records. We provide a starting point that you can tailor to your needs. To try it out, create a scratch directory and copy the following to files into it. Then do a "source 3level" and watch what happens.
  • AWK program for processing bus output records
  • CSH script for running a complete 3-level cache example (you'll want to customize this for your purposes; requires the preceding awk program to be available).

Atom

Atom is a generalized instrumentation tool that runs on Digital's Alpha workstations. We are using a very limited subset of its capabilities to create dinero traces for the class homeworks, but you are welcome to use its full capabilities for your projects.

Documentation:

• Browseable Man Page (or, you can type "man atom" from the Unix command line)
• Atom user manual (.pdf | .ps)
• Atom reference manual (.pdf | .ps)
• If you're curious, also see Digital WRL Technical Report 94.2 (local .pdf available) and
  Digital WRL Technical Note 44 (local .pdf available)

Basic Atom Example:

This is a very basic example of running atom. Assume you want to instrument the program test.c to generate dinero traces (other tools at /usr/ccs/lib/cmplrs/atom/tools are available with atom as well). To use atom, first you have to compile it with at least the following cc command line (assume you're compiling test.c):

cc -O -Wl,-r -non_shared test.c -o test.rr

Then you'll need an analysis file and an instrumentation file. If you want to generate dinero traces, use instr.c and analy.c provided with the course materials. Or, you can skip these files and use one of the atom toolkits with the -tool option. If you're using the dinero files, use the command line:

atom test.rr instr.c analy.c -o test.atom

Then use this instrumented program with dinero, such as:

test.atom | dinero -b16 -u16K

If you're using atom with a complicated compilation, it's probably enough to set the CFLAGS variable in the make file to the appropriate compiler flags above (-O -Wl,-r -non_shared) and re-name the compiled output file to a ".rr" extension to run through atom. Atom may or may not work with gcc or g++ -- you're on your own there.

Caution: when running an "atom-ized" program instrumented for dinero, the normal stdout characters from the program will still be generated -- so you may want to modify analy.c to permit you to grep/awk/perl the dinero information from the normal program lines.

A specialized set of Atom instrumentation:

A specialized set of Atom instrumentation is used in one of the homeworks. This instrumentation only measures data loads & stores (not instruction fetches). Furthermore it has the capability to instrument a single procedure within a larger program by setting a text string within the instrumentation file. These files are used in the manner of the previous example.

• Data load & store-only analysis file.
• Single-procedure instrumentation file.

Hennessy & Patterson Tool Set

The standard software distribution for the H&P textbook is at ftp://max.stanford.edu/pub/hennessy-patterson.software. (Also, check out related tools from other sources at http://www.mkp.com/books_catalog/ca/hp2e_res.htm)

However, to make life easier for students of this course, the un-tarred files are available on afs. These files are provided as-is for your convenience if you care to explore concepts beyond those taught in the course. The course staff cannot assume responsibility for maintaining these files or the data/tools in them. The files are stored at: /afs/ece/class/ee742/hp_tools/, and include a readme file with some interesting control characters in it. The subdirectories include:

• Benchmarks of gcc, tex, spice
• The original distribution of dinero (not modified for 64-bit operation)
• A distribution of DLX
• And two sub-directories of dinero input trace files.

18-548/15-548 home page