Prediction of GPU Kernel Performance for Matrix Multiplication

This repository contains the implementation of machine learning models used to predict the computational running time for multiplying two 2048x2048 matrices using a GPU OpenCL SGEMM kernel. The code is based on the research presented in the paper:

Prediction of the Running Time for Multiplying Two 2048x2048 Matrices using a GPU OpenCL SGEMM Kernel

Overview

Matrix multiplication is a fundamental operation in linear algebra, with applications in fields like machine learning, scientific computing, and engineering. This project leverages machine learning to predict the computational performance of a GPU-based SGEMM kernel for matrix multiplication, enabling engineers to evaluate GPU performance efficiently.

Key Features

• Dataset: 241,600 data points with 14 GPU configuration parameters.
• Machine Learning Models:
  • Linear Regression
  • Polynomial Regression
  • Decision Trees
  • Random Forests
  • K-Nearest Neighbors (KNN)
• Ensemble Learning: Combining Random Forest and KNN models.
• Performance Metrics: R², Mean Absolute Error (MAE), and Mean Squared Error (MSE).

Dataset

The dataset includes the following key parameters:

• Workgroup Sizes: MWG, NWG
• Tiling Sizes: KWG
• Local Memory Allocations: MDIMC, NDIMC, MDIMA, NDIMB
• Execution Patterns: STRM, STRN, SA, SB
• Vector Widths: VWM, VWN
• Kernel Shaping: KWI

The target variable is the logarithmic transformation of the average running time for matrix multiplication tasks.

Getting Started

Prerequisites

To run the code, you need:

• Python 3.8 or later
• Libraries:
  • NumPy
  • Pandas
  • Scikit-learn
  • Matplotlib
  • Seaborn

Install the required dependencies using:

pip install -r requirements.txt