Performance-aware component composition for GPU-based systems.

Format:

Book

Author/Creator:

Dastgeer, Usman.

Series:

Linköping Studies in Science and Technology : Dissertations

Language:

English

Subjects (All):

Graphics processing units.

Software engineering.

Physical Description:

1 online resource (258 pages)

Edition:

1st ed.

Other Title:

Linköping Studies in Science and Technology

Place of Publication:

Linköping : Linköping University Electronic Press, 2014.

Summary:

This thesis by Usman Dastgeer addresses the challenges of efficiently programming GPU-based systems, focusing on performance and portability. It explores three innovative approaches: a skeleton programming library (SkePU) for high-level abstraction, a composition tool utilizing metadata from XML files, and a framework leveraging source-code annotations and analysis. Each approach offers unique advantages, demonstrating the effectiveness of component programming techniques in optimizing GPU computations. The work is supported by experimental evaluations on various GPU-based systems, showcasing improvements in performance across different applications. This research is supported by EU FP7 projects and highlights the ongoing evolution and optimization of heterogeneous computing systems. Generated by AI.

Contents:

Front

Abstract

Acknowledgements

1.1 Motivation

1.2 Component-based approach

1.3 Optimized composition problem

1.4 Our work

1.5 Contributions

1.6 List of publications

1.7 Research Method

1.8 Thesis outline

2.1 Component-based software engineering

2.2 Skeleton programming

2.3 Runtime scheduling and selection using a runtime system

2.4 Multicore and OpenMP

2.5 Programming NVIDIA GPUs

Chapter 3

3.1 SkePU library

3.2 Implementation selection

3.3 Runtime support for hybrid execution

3.4 Evaluation

3.5 Case study on MapOverlap2D on CUDA

3.6 Summary

Chapter 4

4.1 Introduction

4.2 Memory management in SkePU

4.3 Evaluation

4.4 Containers with StarPU runtime system

4.5 SkePU program execution model

4.6 Summary

5.1 PEPPHER Component model

5.2 Composition tool

5.3 Prototype implementation

5.4 Composition example

5.5 Evaluation

5.6 Summary

Chapter 6

6.1 GCF Component Model

6.2 Global Composition Framework

6.3 Evaluation

6.4 Global composition

6.5 Summary

7.1 Three approaches

7.2 Concluding remarks

8.1 Skeleton programming

8.2 Programming approaches for GPU-based systems

8.3 Component Models in HPC and Grid Com-puting

9.1 SkePU extensions

9.2 PEPPHER Composition tool

9.3 Global Composition Framework (GCF).

Notes:

Description based on publisher supplied metadata and other sources.

Part of the metadata in this record was created by AI, based on the text of the resource.

ISBN:

9789175193830

9175193833

OCLC:

927227438