Analyzing Reduction Abstractions Capabilities

Tom Deakin; Simon N McIntosh-Smith; John Pennycook; Jason Sewall

doi:10.1109/P3HPC54578.2021.00007

Analyzing Reduction Abstractions Capabilities

Tom Deakin, Simon N McIntosh-Smith, John Pennycook, Jason Sewall

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

3 Citations (Scopus)

189 Downloads (Pure)

Abstract

Reductions are a common pattern in parallel programming, and every parallel programming language or framework has its own reduction abstraction with its own idiosyncrasies. These abstractions differ not only in their syntax, but also in their semantics and their ability to express certain types of reduction. Such differences may prevent specific combinations of abstraction and hardware platform from reaching high levels of performance, with consequences for portability and programmer productivity.

In this paper, we present a set of representative reduction benchmarks to explore the capabilities of five contemporary programming languages and frameworks -- OpenMP, Kokkos, RAJA, SYCL, and the oneAPI DPC++ Library (oneDPL) -- across a variety of hardware platforms, including CPUs and GPUs from multiple vendors. We discuss the advantages and disadvantages of each reduction abstraction, and conclude with recommendations to improve their design and implementation.

Original language	English
Title of host publication	2021 International Workshop on Performance, Portability and Productivity in HPC (P3HPC)
Publisher	IEEE Computer Society
ISBN (Electronic)	978-1-6654-2439-4
ISBN (Print)	978-1-6654-2440-0
DOIs	https://doi.org/10.1109/P3HPC54578.2021.00007
Publication status	Published - 14 Nov 2021
Event	2021 International Workshop on Performance, Portability and Productivity in HPC (P3HPC) - Duration: 14 Nov 2021 → 14 Nov 2021 https://p3hpc.org/workshop/2021/

Conference

Conference	2021 International Workshop on Performance, Portability and Productivity in HPC (P3HPC)
Abbreviated title	P3HPC
Period	14/11/21 → 14/11/21
Internet address	https://p3hpc.org/workshop/2021/

Access to Document

10.1109/P3HPC54578.2021.00007

Full-text PDF (accepted manuscript)
This is the accepted author manuscript (AAM). The final published version (version of record) is available online via IEEE at https://doi.org/10.1109/P3HPC54578.2021.00007. Please refer to any applicable terms of use of the publisher.
Accepted author manuscript, 288 KB

Handle.net

Persistent link

Cite this

@inproceedings{696a96fe7f7f487b9f34a6a991265722,

title = "Analyzing Reduction Abstractions Capabilities",

abstract = "Reductions are a common pattern in parallel programming, and every parallel programming language or framework has its own reduction abstraction with its own idiosyncrasies. These abstractions differ not only in their syntax, but also in their semantics and their ability to express certain types of reduction. Such differences may prevent specific combinations of abstraction and hardware platform from reaching high levels of performance, with consequences for portability and programmer productivity.In this paper, we present a set of representative reduction benchmarks to explore the capabilities of five contemporary programming languages and frameworks -- OpenMP, Kokkos, RAJA, SYCL, and the oneAPI DPC++ Library (oneDPL) -- across a variety of hardware platforms, including CPUs and GPUs from multiple vendors. We discuss the advantages and disadvantages of each reduction abstraction, and conclude with recommendations to improve their design and implementation.",

author = "Tom Deakin and McIntosh-Smith, {Simon N} and John Pennycook and Jason Sewall",

year = "2021",

month = nov,

day = "14",

doi = "10.1109/P3HPC54578.2021.00007",

language = "English",

isbn = "978-1-6654-2440-0",

booktitle = "2021 International Workshop on Performance, Portability and Productivity in HPC (P3HPC)",

publisher = "IEEE Computer Society",

address = "United States",

note = "2021 International Workshop on Performance, Portability and Productivity in HPC (P3HPC), P3HPC ; Conference date: 14-11-2021 Through 14-11-2021",

url = "https://p3hpc.org/workshop/2021/",

}

TY - GEN

T1 - Analyzing Reduction Abstractions Capabilities

AU - Deakin, Tom

AU - McIntosh-Smith, Simon N

AU - Pennycook, John

AU - Sewall, Jason

PY - 2021/11/14

Y1 - 2021/11/14

N2 - Reductions are a common pattern in parallel programming, and every parallel programming language or framework has its own reduction abstraction with its own idiosyncrasies. These abstractions differ not only in their syntax, but also in their semantics and their ability to express certain types of reduction. Such differences may prevent specific combinations of abstraction and hardware platform from reaching high levels of performance, with consequences for portability and programmer productivity.In this paper, we present a set of representative reduction benchmarks to explore the capabilities of five contemporary programming languages and frameworks -- OpenMP, Kokkos, RAJA, SYCL, and the oneAPI DPC++ Library (oneDPL) -- across a variety of hardware platforms, including CPUs and GPUs from multiple vendors. We discuss the advantages and disadvantages of each reduction abstraction, and conclude with recommendations to improve their design and implementation.

AB - Reductions are a common pattern in parallel programming, and every parallel programming language or framework has its own reduction abstraction with its own idiosyncrasies. These abstractions differ not only in their syntax, but also in their semantics and their ability to express certain types of reduction. Such differences may prevent specific combinations of abstraction and hardware platform from reaching high levels of performance, with consequences for portability and programmer productivity.In this paper, we present a set of representative reduction benchmarks to explore the capabilities of five contemporary programming languages and frameworks -- OpenMP, Kokkos, RAJA, SYCL, and the oneAPI DPC++ Library (oneDPL) -- across a variety of hardware platforms, including CPUs and GPUs from multiple vendors. We discuss the advantages and disadvantages of each reduction abstraction, and conclude with recommendations to improve their design and implementation.

U2 - 10.1109/P3HPC54578.2021.00007

DO - 10.1109/P3HPC54578.2021.00007

M3 - Conference Contribution (Conference Proceeding)

SN - 978-1-6654-2440-0

BT - 2021 International Workshop on Performance, Portability and Productivity in HPC (P3HPC)

PB - IEEE Computer Society

T2 - 2021 International Workshop on Performance, Portability and Productivity in HPC (P3HPC)

Y2 - 14 November 2021 through 14 November 2021

ER -

Analyzing Reduction Abstractions Capabilities

Abstract

Conference

Access to Document

Handle.net

Fingerprint

Cite this