Using a high-level parallel programming language for GPU-accelerated tomographic reconstruction

Mette Bjerg Lindhøj; Troels Henriksen; Lærke Pedersen; Jon Sporring

Using a high-level parallel programming language for GPU-accelerated tomographic reconstruction

Mette Bjerg Lindhøj, Troels Henriksen, Lærke Pedersen, Jon Sporring

Abstract

This paper aims to determine the usefulness of using a high-level parallel programming language for implementing parallel high-performance tomographic reconstruction algorithms. The purpose of this is to make it easier for researchers to implement advanced model-based iterative reconstruction algorithms for use at synchrotron facilities, while still taking advantage of hardware such as GPUs. To this end, we implement the forward- and back-projection in the programming language Futhark, and verify their applicability through an implementation of an algebraic reconstruction algorithm. We obtain promising performance results by use of algorithmic considerations instead of low-level optimizations. Finally, we demonstrate that the implementation makes it possible to prototype implementations of iterative reconstruction algorithms on a standard laptop while still obtaining good scaling towards highend GPUs.

Original language	English
Title of host publication	The 2019 International Conference on High Performance Computing & Simulation : HPCS 2019
Number of pages	6
Publication date	15 Jul 2019
Pages	27-32
Article number	3
Chapter	2
Publication status	Published - 15 Jul 2019

Cite this

Using a high-level parallel programming language for GPU-accelerated tomographic reconstruction. / Lindhøj, Mette Bjerg; Henriksen, Troels; Pedersen, Lærke et al.
The 2019 International Conference on High Performance Computing & Simulation: HPCS 2019. 2019. p. 27-32 3.

Research output: Chapter in Book/Report/Conference proceeding › Article in proceedings › Research › peer-review

@inproceedings{750c5aac32e34c61adbc388855cd1f08,

title = "Using a high-level parallel programming language for GPU-accelerated tomographic reconstruction",

abstract = "This paper aims to determine the usefulness of using a high-level parallel programming language for implementing parallel high-performance tomographic reconstruction algorithms. The purpose of this is to make it easier for researchers to implement advanced model-based iterative reconstruction algorithms for use at synchrotron facilities, while still taking advantage of hardware such as GPUs. To this end, we implement the forward- and back-projection in the programming language Futhark, and verify their applicability through an implementation of an algebraic reconstruction algorithm. We obtain promising performance results by use of algorithmic considerations instead of low-level optimizations. Finally, we demonstrate that the implementation makes it possible to prototype implementations of iterative reconstruction algorithms on a standard laptop while still obtaining good scaling towards highend GPUs.",

author = "Lindh{\o}j, {Mette Bjerg} and Troels Henriksen and L{\ae}rke Pedersen and Jon Sporring",

year = "2019",

month = jul,

day = "15",

language = "English",

pages = "27--32",

booktitle = "The 2019 International Conference on High Performance Computing & Simulation",

}

TY - GEN

T1 - Using a high-level parallel programming language for GPU-accelerated tomographic reconstruction

AU - Lindhøj, Mette Bjerg

AU - Henriksen, Troels

AU - Pedersen, Lærke

AU - Sporring, Jon

PY - 2019/7/15

Y1 - 2019/7/15

N2 - This paper aims to determine the usefulness of using a high-level parallel programming language for implementing parallel high-performance tomographic reconstruction algorithms. The purpose of this is to make it easier for researchers to implement advanced model-based iterative reconstruction algorithms for use at synchrotron facilities, while still taking advantage of hardware such as GPUs. To this end, we implement the forward- and back-projection in the programming language Futhark, and verify their applicability through an implementation of an algebraic reconstruction algorithm. We obtain promising performance results by use of algorithmic considerations instead of low-level optimizations. Finally, we demonstrate that the implementation makes it possible to prototype implementations of iterative reconstruction algorithms on a standard laptop while still obtaining good scaling towards highend GPUs.

AB - This paper aims to determine the usefulness of using a high-level parallel programming language for implementing parallel high-performance tomographic reconstruction algorithms. The purpose of this is to make it easier for researchers to implement advanced model-based iterative reconstruction algorithms for use at synchrotron facilities, while still taking advantage of hardware such as GPUs. To this end, we implement the forward- and back-projection in the programming language Futhark, and verify their applicability through an implementation of an algebraic reconstruction algorithm. We obtain promising performance results by use of algorithmic considerations instead of low-level optimizations. Finally, we demonstrate that the implementation makes it possible to prototype implementations of iterative reconstruction algorithms on a standard laptop while still obtaining good scaling towards highend GPUs.

M3 - Article in proceedings

SP - 27

EP - 32

BT - The 2019 International Conference on High Performance Computing & Simulation

ER -

Using a high-level parallel programming language for GPU-accelerated tomographic reconstruction

Abstract

Fingerprint

Cite this