Hierarchical-Cache-Simulator

This repository contains a modular C++ implementation of three level cache hierarchy for multi-core processors. It provides a wide range of functionalities such as different choice of replacement policy and different architectures. For eg. One can keep separate L1/L2 cache for data and instruction or unified cache.

How to use

• First generate address trace of a program. To understand how to do this, visit this repository. We will assume that trace is stored in a file called addrtrace.out
• Specify your architecture by appropriately modifying the code simulate.cpp
• Compile using g++ -std=c++11 simulate.cpp L1Cache.cpp L2Cache.cpp -o simulate
• Run ./simulate {}addrtrace.out > LLCtrace.out' Now you have trace for LLC in the file LLCtrace.out`.
• Change specifications of LLC according to your need
• Compile using g++ -std=c++11 llc_simulation.cpp LLC.cpp CE_Belady.cpp -o llc
• Run ./llc LLCtrace.out LLCtrace_hitfile.out LLCtrace_sharefile.out LLCtrace_reusefile.out These arguments will be explained later.

Description of different files/folders

• CE_Belady.cpp, CE_Belady.h contains definitions and declarations of cache entry required for running Belady's cache replacement policy

• L1Cache.cpp, L1Cache.h contains definitions and declarations of L1 Cache

• L2Cache.cpp, L2Cache.h contains definitions and declarations of L2 Cache

• LLC.cpp, LLC.h contains definitions and declarations of Last level Cache

• constants.h contains declarations of few constanst used in the program

• belady.cpp contains the commented preprocessing code needed for efficient implementation of Belady's policy

• simulate.cpp contains the code which specifies the whole architecture. Number of cores, how L1 and L2 cache are related to each other, what is their size(size = num_sets * associativity * size of Cache Block). Note that size of cache block is assumed to be 64B. This code finally simulates the address trace generated by PIN tool and generates trace for LLC

• llc-simulation.cpp contains the code which specifies the properties of LLC. It takes the trace generated by simulate.cpp as input and runs it through the LLC, performs analysis on it.

• l1l2_run.py contains the driver script for running simulate.cpp

• llc-run.py contains the driver script for running llc-simulation.cpp

• llc-lru.c is a C implementation of LRU algorithm

• llc-belady.c is a C implementation of Belady Algorithm

• llc-pc.c is the implementation in which we calculate the probability of getting next hit based on PC. We maintain hit count

• llc-prob.c is the implementation in which we calculate the probability of getting next hit based on memory location accessed. We maintain hit count

• llc-prob-share.c is the implementation in which we calculate the probability of getting next hit based on memory location accessed. We maintain hit count separately for shared cache blocks and non-shared cache blocks

• prob-bel.c Contains code that runs belady's algorithm and count based probability algorithm simulatenously. Can be used to see where the algorithm makes poor choice in replacement decision

• The folder analysis contains few codes to analyze the results obtained after LLC simulation

Note that, I am collecting hit, reuse and sharing statistics of cache blocks in LLC. One can very easily modify/add code in L1Cache.cpp, L2Cache.cpp, LLC.cpp to collect different statistics. Moreover, one can add new replacement policies. The arguments in LLC.cpp are the names of files in which output should be written.

<*> This code has been tested on Ubuntu machine and works without any issue. If you have any question, comment, doubt, feel free to get in touch with Abhishek Kumar[[email protected]]