VectorPU: A Generic and Efficient Data-container and Component Model for Transparent Data Transfer on GPU-based Heterogeneous Systems

Lu Li; Christoph Kessler

doi:10.1145/3029580.3029582

VectorPU: A Generic and Efficient Data-container and Component Model for Transparent Data Transfer on GPU-based Heterogeneous Systems

Lu Li, Christoph Kessler

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

Abstract

We present VectorPU, a C++ based programming framework providing high-level and efficient unified memory access on heterogeneous systems, in particular GPU-based systems. VectorPU consists of a light-weight runtime library providing a generic, "smart" data-container abstraction for transparent software caching of array operands with programmable memory coherence, and a light-weight component model realized by macro-based data access annotations. VectorPU thereby enables a flexible unified memory view with data transfer and device memory management abstracted away from programmers, while keeping the efficiency of expert-written code with manual data movement and memory management. We provide a prototype of VectorPU for (CUDA) GPU-based systems, and show that it can achieve 1.40× to 13.29× speedup over good quality code using Nvidia's Unified Memory by experiments on several machines ranging from laptops to supercomputer nodes, with Kepler and Maxwell GPUs. We also show the expressiveness and wide applicability of VectorPU, and its low overhead and equal efficiency compared to expert-written code.

Original language	English
Title of host publication	Proceedings of the 8th Workshop and 6th Workshop on Parallel Programming and Run-Time Management Techniques for Many-core Architectures and Design Tools and Architectures for Multicore Embedded Computing Platforms
Place of Publication	New York, NY, USA
Publisher	ACM
Pages	7-12
Number of pages	6
ISBN (Print)	978-1-4503-4877-5
DOIs	https://doi.org/10.1145/3029580.3029582
Publication status	Published - 25 Jan 2017
Event	8th Workshop on Parallel Programming and Run-Time Management Techniques for Many-core Architectures and 6th Workshop on Design Tools and Architectures for Multicore Embedded Computing Platforms - Stockholm, Sweden Duration: 28 Jan 2017 → … http://conferences.microlab.ntua.gr/ditam-parma2017/

Publication series

Name	PARMA-DITAM '17
Publisher	ACM

Conference

Conference	8th Workshop on Parallel Programming and Run-Time Management Techniques for Many-core Architectures and 6th Workshop on Design Tools and Architectures for Multicore Embedded Computing Platforms
Abbreviated title	PARMA-DITAM 2017
Country/Territory	Sweden
City	Stockholm
Period	28/01/17 → …
Internet address	http://conferences.microlab.ntua.gr/ditam-parma2017/

Keywords / Materials (for Non-textual outputs)

Coherence Management, Flow Signature, GPGPU, Programming Framework, Run-time System, VectorPU

Access to Document

10.1145/3029580.3029582

https://dl.acm.org/citation.cfm?id=3029582

Cite this

Li, L., & Kessler, C. (2017). VectorPU: A Generic and Efficient Data-container and Component Model for Transparent Data Transfer on GPU-based Heterogeneous Systems. In Proceedings of the 8th Workshop and 6th Workshop on Parallel Programming and Run-Time Management Techniques for Many-core Architectures and Design Tools and Architectures for Multicore Embedded Computing Platforms (pp. 7-12). (PARMA-DITAM '17). ACM. https://doi.org/10.1145/3029580.3029582

Li, Lu ; Kessler, Christoph. / VectorPU: A Generic and Efficient Data-container and Component Model for Transparent Data Transfer on GPU-based Heterogeneous Systems. Proceedings of the 8th Workshop and 6th Workshop on Parallel Programming and Run-Time Management Techniques for Many-core Architectures and Design Tools and Architectures for Multicore Embedded Computing Platforms. New York, NY, USA : ACM, 2017. pp. 7-12 (PARMA-DITAM '17).

@inproceedings{7681ae21a1924efc991a6e902edf62e0,

title = "VectorPU: A Generic and Efficient Data-container and Component Model for Transparent Data Transfer on GPU-based Heterogeneous Systems",

abstract = "We present VectorPU, a C++ based programming framework providing high-level and efficient unified memory access on heterogeneous systems, in particular GPU-based systems. VectorPU consists of a light-weight runtime library providing a generic, {"}smart{"} data-container abstraction for transparent software caching of array operands with programmable memory coherence, and a light-weight component model realized by macro-based data access annotations. VectorPU thereby enables a flexible unified memory view with data transfer and device memory management abstracted away from programmers, while keeping the efficiency of expert-written code with manual data movement and memory management. We provide a prototype of VectorPU for (CUDA) GPU-based systems, and show that it can achieve 1.40× to 13.29× speedup over good quality code using Nvidia's Unified Memory by experiments on several machines ranging from laptops to supercomputer nodes, with Kepler and Maxwell GPUs. We also show the expressiveness and wide applicability of VectorPU, and its low overhead and equal efficiency compared to expert-written code.",

keywords = "Coherence Management, Flow Signature, GPGPU, Programming Framework, Run-time System, VectorPU",

author = "Lu Li and Christoph Kessler",

year = "2017",

month = jan,

day = "25",

doi = "10.1145/3029580.3029582",

language = "English",

isbn = "978-1-4503-4877-5",

series = "PARMA-DITAM '17",

publisher = "ACM",

pages = "7--12",

booktitle = "Proceedings of the 8th Workshop and 6th Workshop on Parallel Programming and Run-Time Management Techniques for Many-core Architectures and Design Tools and Architectures for Multicore Embedded Computing Platforms",

note = "8th Workshop on Parallel Programming and Run-Time Management Techniques for Many-core Architectures and 6th Workshop on Design Tools and Architectures for Multicore Embedded Computing Platforms, PARMA-DITAM 2017 ; Conference date: 28-01-2017",

url = "http://conferences.microlab.ntua.gr/ditam-parma2017/",

}

Li, L & Kessler, C 2017, VectorPU: A Generic and Efficient Data-container and Component Model for Transparent Data Transfer on GPU-based Heterogeneous Systems. in Proceedings of the 8th Workshop and 6th Workshop on Parallel Programming and Run-Time Management Techniques for Many-core Architectures and Design Tools and Architectures for Multicore Embedded Computing Platforms. PARMA-DITAM '17, ACM, New York, NY, USA, pp. 7-12, 8th Workshop on Parallel Programming and Run-Time Management Techniques for Many-core Architectures and 6th Workshop on Design Tools and Architectures for Multicore Embedded Computing Platforms, Stockholm, Sweden, 28/01/17. https://doi.org/10.1145/3029580.3029582

VectorPU: A Generic and Efficient Data-container and Component Model for Transparent Data Transfer on GPU-based Heterogeneous Systems. / Li, Lu; Kessler, Christoph.
Proceedings of the 8th Workshop and 6th Workshop on Parallel Programming and Run-Time Management Techniques for Many-core Architectures and Design Tools and Architectures for Multicore Embedded Computing Platforms. New York, NY, USA: ACM, 2017. p. 7-12 (PARMA-DITAM '17).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

TY - GEN

T1 - VectorPU: A Generic and Efficient Data-container and Component Model for Transparent Data Transfer on GPU-based Heterogeneous Systems

AU - Li, Lu

AU - Kessler, Christoph

PY - 2017/1/25

Y1 - 2017/1/25

N2 - We present VectorPU, a C++ based programming framework providing high-level and efficient unified memory access on heterogeneous systems, in particular GPU-based systems. VectorPU consists of a light-weight runtime library providing a generic, "smart" data-container abstraction for transparent software caching of array operands with programmable memory coherence, and a light-weight component model realized by macro-based data access annotations. VectorPU thereby enables a flexible unified memory view with data transfer and device memory management abstracted away from programmers, while keeping the efficiency of expert-written code with manual data movement and memory management. We provide a prototype of VectorPU for (CUDA) GPU-based systems, and show that it can achieve 1.40× to 13.29× speedup over good quality code using Nvidia's Unified Memory by experiments on several machines ranging from laptops to supercomputer nodes, with Kepler and Maxwell GPUs. We also show the expressiveness and wide applicability of VectorPU, and its low overhead and equal efficiency compared to expert-written code.

AB - We present VectorPU, a C++ based programming framework providing high-level and efficient unified memory access on heterogeneous systems, in particular GPU-based systems. VectorPU consists of a light-weight runtime library providing a generic, "smart" data-container abstraction for transparent software caching of array operands with programmable memory coherence, and a light-weight component model realized by macro-based data access annotations. VectorPU thereby enables a flexible unified memory view with data transfer and device memory management abstracted away from programmers, while keeping the efficiency of expert-written code with manual data movement and memory management. We provide a prototype of VectorPU for (CUDA) GPU-based systems, and show that it can achieve 1.40× to 13.29× speedup over good quality code using Nvidia's Unified Memory by experiments on several machines ranging from laptops to supercomputer nodes, with Kepler and Maxwell GPUs. We also show the expressiveness and wide applicability of VectorPU, and its low overhead and equal efficiency compared to expert-written code.

KW - Coherence Management, Flow Signature, GPGPU, Programming Framework, Run-time System, VectorPU

U2 - 10.1145/3029580.3029582

DO - 10.1145/3029580.3029582

M3 - Conference contribution

SN - 978-1-4503-4877-5

T3 - PARMA-DITAM '17

SP - 7

EP - 12

BT - Proceedings of the 8th Workshop and 6th Workshop on Parallel Programming and Run-Time Management Techniques for Many-core Architectures and Design Tools and Architectures for Multicore Embedded Computing Platforms

PB - ACM

CY - New York, NY, USA

T2 - 8th Workshop on Parallel Programming and Run-Time Management Techniques for Many-core Architectures and 6th Workshop on Design Tools and Architectures for Multicore Embedded Computing Platforms

Y2 - 28 January 2017

ER -

Li L, Kessler C. VectorPU: A Generic and Efficient Data-container and Component Model for Transparent Data Transfer on GPU-based Heterogeneous Systems. In Proceedings of the 8th Workshop and 6th Workshop on Parallel Programming and Run-Time Management Techniques for Many-core Architectures and Design Tools and Architectures for Multicore Embedded Computing Platforms. New York, NY, USA: ACM. 2017. p. 7-12. (PARMA-DITAM '17). doi: 10.1145/3029580.3029582

VectorPU: A Generic and Efficient Data-container and Component Model for Transparent Data Transfer on GPU-based Heterogeneous Systems

Abstract

Publication series

Conference

Keywords / Materials (for Non-textual outputs)

Access to Document

Fingerprint

Cite this