ArchXplore

Table of Contents

• About
• Getting Started
• Usage
• Contributing

About

ArchXplore is a simulation infrastructure for research and development in the field of computer architecture. It provides a platform to simulate and analyze microarchitecture designs, and to explore the impact of different design choices on system performance.

Compared to traditional simulation tools, ArchXplore offers several unique features:

• Configurability: ArchXplore provides a flexible and powerful way to configure, composite modules and build topologies with auto-complete in IDEs.

• Performance: ArchXplore leverages dynamic binary translation(QEMU) to achieve high performance. It also supports fine-grained parallel simulation for multi-threaded applications.

• Accuracy: ArchXplore provides the ability to trade off performance for accuracy. With sparta, a discrete event simulation framework, it is possible to simulate microarchitectures with high fidelity and high precision.

Getting Started

Prerequisites

Ubuntu 22.04 is recommended.

sudo apt install ninja-build libglib2.0-dev

Installing

• If conda is not installed, install it using the following command:

wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh

bash Miniconda3-latest-Linux-x86_64.sh 

• Install jq and yq using the following commands:

conda install -c conda-forge jq yq

• Clone the repository:

git clone https://github.com/ArchitectXplore/ArchXplore.git

cd ArchXplore

• Build conda environment:

./ext/map/scripts/create_conda_env.sh <Your environment name> run 

conda activate <Your environment name>

• Build ArchXplore:

mkdir build && cd build

cmake ..

make -j$(nproc)

Usage

• Start inter process communication server:

./IPCService &

• Demonstrate the usage of ArchXplore by running a simple CPU simulation:

./ArchXplore ../configs/simpleCPU.py

• The output should be similar to the following:

Host time elapsed(s):  0.8125679176300764
Guest time elapsed(s):  0.001
Total instructions executed:  43248512
Million instructions per second:  53.22448876167542

Features

• Support python configuration files for simulation.
• Support for multi-threads simulation.
• Support for multi-processes simulation.
• Integrate DRAMSim3 for cycle accurate memory modeling.
• Design modeling framework for coherent caches.
• Design modeling framework for network on chip(NoC).
• Top-down analysis framework for microarchitecture exploration.
• Pthread/OpenMP API instrumentation for synchronization-aware multi-core simulation.