Getting started with gem5

Running gem5

• Interface is python3.
• gem5 python environment can be thought as python3 environment + m5 library + gem5 standard library.
  • There is no magic: the paths to the m5 library and gem5 standard library are compiled to the gem5 library (provide reference to the source code).
  • Scripts without using m5 library and gem5 library can be run by both gem5 and python.
• Demo of,
  • A simple script that can be run by both gem5 and python3.
  • A simple script that can only be run by gem5.
• The python script inputted to gem5 is usually referred as “system configuration”, “configuration script”, “python config”, etc.

Things to touch on

• “Model” vs “configuration”

First configuration example

gem5’s output

• m5out/
• stats.txt
  • Where the stats are.
  • Dumping stats: telling gem5 to emit all stats by appending the stat to the file stats.txt.
  • The stats will always be dumped at the end of simulations.
  • The stats can be manually dumped multiple times during a simulation, thus there could be multiple sets of stats in stats.txt.
    • The first N-1 ones are manually dumped.
    • The last one is dumped automatically.
  • Stats can be reset via m5.stats.reset() or m5_reset_stats()?
  • It is important to know which stats set belong to which stat dump command.
• config.ini, config.json, config.dot, config.dot.pdf, config.dot.svg,
  • How gem5 interpret the system configuration script.
  • Useful for sanity check: strongly recommended to check the outputted config file for each simulation.
• The difference between simulated and wallclock

Python reminder

This is where we give a brief overview of python.

• Briefly going over the basics of the language,
  • Types: int, string, dict, list, set, tuple.
  • Controls: for/while loops, if statements.
  • Functions.
• Detailed covering of frequently used features,
  • List comprehension: used for making a list of SimObject’s.
    • Appending a SimObject to a list is not allowed.
  • Iterator/Generator: “lazily” constructing a list of object,
    • Returning all objects in a list all at once vs returning the next object when queried.
    • Examples.
  • Importing modules,
    • Importing python standard modules.
    • Importing a python file as a module.
    • Importing m5 module, gem5 module (the gem5 module is known as the gem5 standard library).
  • Object oriented programming,
    • Base (parent) class, derived (child) class.
      • The child class inherits all functions and variables.
    • Overriding a function.
      • If the child class and the parent class provide a function with the same name, the child’s function will be called.
    • Class vs instance/object.
      • A class can be thought as a blueprint for an object of the same class.
    • Class variables vs object variables.
      • The class variables are not unique among objects.
      • The object variables are unique among objects.
    • Class functions vs object functions.
      • The class functions are not unique among objects.
      • The object functions are unique among objects.
    • Abstract functions, abstract classes.
      • abc library in python.
      • How to use @abstractmethod to define an abstract function.
      • A class is abstract if it contains at least one abstract functions.
      • Providing the guarantee (or interface) that all classes inherited from this class will have the abstract function defined.
  • m5 library,
    • m5.instatiate()
      • Tell gem5 to connect SimObject as specified in the script.
      • After this statement, the system configuration should not be modified.
    • m5.simulate()
      • Running the simulation.
      • Can specify how many instructions to simulate.
    • m5.stats.dump()
    • m5.stats.reset()
  • argparse
    • Parameterize the configuration script.
    • Example usage.
  • seaborn
    • Example usage of a barplot.

Example for Python reminder

• Parameterized input using argparse.
• Calling m5.stats.dump() once.
• Using python3 to parse the stats file and graphing results using seaborn.