Build and simulate

Table of contents

1. Build gem5(click me)
   1. Setup on Ubuntu 20.04 (gem5 >= v21.0)
   2. Setup on Ubuntu 18.04 (gem5 >= v21.0)
   3. Using Docker
   4. Build
2. Write an insteresting app(sieve)
3. Use gem5

At this part, you will learn installation and simulation of gem5

Build gem5(click me)

You can choose either of three ways below.

Setup on Ubuntu 20.04 (gem5 >= v21.0)

sudo apt install build-essential git m4 scons zlib1g zlib1g-dev \
    libprotobuf-dev protobuf-compiler libprotoc-dev libgoogle-perftools-dev \
    python3-dev python-is-python3 libboost-all-dev pkg-config

Setup on Ubuntu 18.04 (gem5 >= v21.0)

sudo apt install build-essential git m4 scons zlib1g zlib1g-dev \
    libprotobuf-dev protobuf-compiler libprotoc-dev libgoogle-perftools-dev \
    python3-dev python libboost-all-dev pkg-config

Using Docker

• Install Docker

For windows, install wsl first by using

wsl --install 

then download Docker Desktop for Windows and install.

wsl is windows subsystem for linux. You can do most of the things you can do on Linux on WSL. To integrate docker with wsl through setting=>Resources=>WSL Integration

For other systems, reference here.

• Build gem5 with Docker

First,obtain gem5 image

docker pull gcr.io/gem5-test/ubuntu-20.04_all-dependencies:v22-0

Due to some reason, you may fail to pull that image, you can change gcr.io to gcr.lank8s.cn

Then, launch the container

docker run -itd -v <gem5 directiory>:/gem5 <image>

using docker images to get info about <image>

using git clone -b v22.0.0.2 https://github.com/gem5/gem5.git to get gem5

Then,connect to the container

docker exec -it <container> /bin/bash

using docker ps to get info about <container>

Build

To get gem5

git clone -b v22.0.0.2 https://github.com/gem5/gem5.git

Within the root of the gem5 directory, gem5 can be built with SCons using:

scons build/X86/gem5.opt -j <cpu-nums>

Write an insteresting app(sieve)

Write a program that outputs one single integer at the end the number of prime numbers <= N(at default N = 100,000,000) . Compile your program as a static binary. Note that your program must achieve O(N) complexity.

Use gem5

• to simulate the program in gem5, these tutorials may be helpful

  • create a simple configuration script

  • using the default configuration scripts

• Run your sieve program in gem5

  • choose an appropriate input size

  You should use something large enough that the application is interesting, but not too large that gem5 takes more than 10 minutes to execute a simulation.

  • change the CPU model from X86TimingSimpleCPU to X86MinorCPU.

  gem5 won’t compile X86MinorCPU by default. You need to add some modifications. gem5 use CPU_MODELS as a parameter in the past. Try to search the keyword CPU_MODELS and see what you can find out. If you search CPU_MODELS and you will see it in release-note.md. It tells that you should modify under arch dir. To simplify the exp, we give the hint that you should modify src/arch/x86/X86CPU.py .

  • Vary the CPU clock from 1GHz to 2GHz (in steps of 1GHz) with both CPU models.

  • Change the memory configuration from DDR3_1600_8x8 to DDR3_2133_8x8 (DDR3 with a faster clock)

a. In each output, does system.cpu.numCycles times system.clk_domain.clock equals simTicks ? Why ?

b. A single-cycle processor executes one instruction per clock cycle. A classic five-stage pipelined processor, on average, executes less than one instruction per clock cycle (due to branching). So why are processors on the market pipelined?