Benchmarking and Extending SYCL Hierarchical Parallelism

Tom Deakin; Simon N McIntosh-Smith; Aksel Alpay; Vincent Heuveline

doi:10.1109/HiPar54615.2021.00007

Benchmarking and Extending SYCL Hierarchical Parallelism

Tom Deakin, Simon N McIntosh-Smith, Aksel Alpay, Vincent Heuveline

Microelectronics

Research output: Chapter in Book/Report/Conference proceeding › Conference Contribution (Conference Proceeding)

11 Citations (Scopus)

509 Downloads (Pure)

Abstract

SYCL is an open-standard, parallel programming model for programming heterogeneous devices from Khronos. It allows single-source programming of diverse attached devices in a cross-platform manner in modern C++. SYCL provides different layers of parallel abstractions, including Same Instruction Multiple Thread (SIMT) kernels, data-parallel loop concurrency and hierarchical parallelism. We then discuss Scoped Parallelism as an extension to the existing Hierarchical Parallelism in SYCL, and discuss the advantages and disadvantages of these models from the perspective of the programmer and an implementer of SYCL. In this paper, we compare writing benchmark programs using SIMT kernel, hierarchical parallelism and scoped parallelism paradigms, and present results running on a high-performance CPU and GPU.

Original language	English
Title of host publication	Workshop on Hierarchical Parallelism for Exascale Computing
Publisher	IEEE Computer Society
ISBN (Electronic)	978-1-6654-1132-5
ISBN (Print)	978-1-6654-1133-2
DOIs	https://doi.org/10.1109/HiPar54615.2021.00007
Publication status	E-pub ahead of print - 24 Dec 2021

Keywords

Concurrent computing
Parallel programming
Exascale computing
Conferences
Graphics processing units
C++ languages
Parallel processing

Access to Document

10.1109/HiPar54615.2021.00007

Full-text PDF (author’s accepted manuscript)
This is the author accepted manuscript (AAM). The final published version (version of record) is available online via IEEE at 10.1109/HiPar54615.2021.00007. Please refer to any applicable terms of use of the publisher.
Accepted author manuscript, 216 KB

Handle.net

Persistent link

Cite this

@inproceedings{a7cdba58e8b84674be7f1d23dca96cdc,

title = "Benchmarking and Extending SYCL Hierarchical Parallelism",

abstract = "SYCL is an open-standard, parallel programming model for programming heterogeneous devices from Khronos. It allows single-source programming of diverse attached devices in a cross-platform manner in modern C++. SYCL provides different layers of parallel abstractions, including Same Instruction Multiple Thread (SIMT) kernels, data-parallel loop concurrency and hierarchical parallelism. We then discuss Scoped Parallelism as an extension to the existing Hierarchical Parallelism in SYCL, and discuss the advantages and disadvantages of these models from the perspective of the programmer and an implementer of SYCL. In this paper, we compare writing benchmark programs using SIMT kernel, hierarchical parallelism and scoped parallelism paradigms, and present results running on a high-performance CPU and GPU.",

keywords = "Concurrent computing, Parallel programming, Exascale computing, Conferences, Graphics processing units, C++ languages, Parallel processing",

author = "Tom Deakin and McIntosh-Smith, \{Simon N\} and Aksel Alpay and Vincent Heuveline",

year = "2021",

month = dec,

day = "24",

doi = "10.1109/HiPar54615.2021.00007",

language = "English",

isbn = "978-1-6654-1133-2",

booktitle = "Workshop on Hierarchical Parallelism for Exascale Computing",

publisher = "IEEE Computer Society",

address = "United States",

}

TY - GEN

T1 - Benchmarking and Extending SYCL Hierarchical Parallelism

AU - Deakin, Tom

AU - McIntosh-Smith, Simon N

AU - Alpay, Aksel

AU - Heuveline, Vincent

PY - 2021/12/24

Y1 - 2021/12/24

N2 - SYCL is an open-standard, parallel programming model for programming heterogeneous devices from Khronos. It allows single-source programming of diverse attached devices in a cross-platform manner in modern C++. SYCL provides different layers of parallel abstractions, including Same Instruction Multiple Thread (SIMT) kernels, data-parallel loop concurrency and hierarchical parallelism. We then discuss Scoped Parallelism as an extension to the existing Hierarchical Parallelism in SYCL, and discuss the advantages and disadvantages of these models from the perspective of the programmer and an implementer of SYCL. In this paper, we compare writing benchmark programs using SIMT kernel, hierarchical parallelism and scoped parallelism paradigms, and present results running on a high-performance CPU and GPU.

AB - SYCL is an open-standard, parallel programming model for programming heterogeneous devices from Khronos. It allows single-source programming of diverse attached devices in a cross-platform manner in modern C++. SYCL provides different layers of parallel abstractions, including Same Instruction Multiple Thread (SIMT) kernels, data-parallel loop concurrency and hierarchical parallelism. We then discuss Scoped Parallelism as an extension to the existing Hierarchical Parallelism in SYCL, and discuss the advantages and disadvantages of these models from the perspective of the programmer and an implementer of SYCL. In this paper, we compare writing benchmark programs using SIMT kernel, hierarchical parallelism and scoped parallelism paradigms, and present results running on a high-performance CPU and GPU.

KW - Concurrent computing

KW - Parallel programming

KW - Exascale computing

KW - Conferences

KW - Graphics processing units

KW - C++ languages

KW - Parallel processing

U2 - 10.1109/HiPar54615.2021.00007

DO - 10.1109/HiPar54615.2021.00007

M3 - Conference Contribution (Conference Proceeding)

SN - 978-1-6654-1133-2

BT - Workshop on Hierarchical Parallelism for Exascale Computing

PB - IEEE Computer Society

ER -

Benchmarking and Extending SYCL Hierarchical Parallelism

Abstract

Keywords

Access to Document

Handle.net

Fingerprint

Cite this