unisa-hpc / synergy-models

SYnergy Models

OS Requirements

These models have been tested on Ubuntu 20.04, 22.04, and RHEL 8.1.

Hardware Requirements

• Single-node experiments: at least one NVIDIA GPU is required.
• Multi-node experiments: a cluster with NVIDIA GPUs is required, equipped with the provided NVGPUFREQ SLURM plugin.

Software Requirements

• DPC++ (Intel/LLVM) 2022-09
  • Install using the Getting Started Guide
• Clang and LLVM 15
  • Install using the LLVM automatic installation script
  • The opt-15 tool must be available in the system PATH
• CUDA Toolkit (tested with CUDA 11.8)
• Python 3
  • Install with sudo apt install python3
  • Required packages: scikit-learn>=0.24, pandas, numpy, matplotlib, paretoset
    • Install with pip install "scikit-learn>=0.24" pandas numpy matplotlib paretoset
• cmake 3.17 or later
  • Install with sudo apt install cmake
  • Alternatively, download the latest stable release
  • Check that cmake version is >= 3.17 using cmake --version

Required for launching the benchmarks on a cluster:

• NVGPUFREQ SLURM plugin
  • Follow the instruction in the readme file of the repository

How to use this repository

This repository is divided in four directories:

• passes, it contains the source code of the compiler passes used to extract the code features
• training-dataset, it contains all the scripts required to generate the data on which the models are trained
• modeling, it contains the modeling training script that can also be used to predict frequency values for new samples

First, the training dataset must be built. Instructions are in the training-dataset/ folder.

There are two scripts that must be used in the workflow, and reside in the root directory of the repository. The extract_features.sh script utilizes the LLVM pass in the passes folder to extract features from a target application.

• The directory of the source code can be specified through the --dir command-line argument
• Header files needed for the pass can be included through the --include command-line argument

Finally, the models can be used to predict the frequencies for each target. The predict.sh script contained in the modeling/ folder must be launched to do so. the predicted frequencies will be in the modeling/predictions/ folder.