Offloading C++17 Parallel STL on System Shared Virtual Memory Platforms

Pekka Jääskeläinen; John Glossner; Martin Jambor; Aleksi Tervo; Matti Rintala

Offloading C++17 Parallel STL on System Shared Virtual Memory Platforms

Pekka Jääskeläinen, John Glossner, Martin Jambor, Aleksi Tervo, Matti Rintala

Tutkimustuotos: Konferenssiartikkeli › Scientific › vertaisarvioitu

46 Lataukset (Pure)

Abstrakti

Shared virtual memory simplifies heterogeneous platform programming by enabling sharing of memory address pointers between heterogeneous devices in the platform. The most advanced implementations present a coherent view of memory to the programmer over the whole virtual address space of the process. From the point of view of data accesses, this System SVM (SSVM) enables the same programming paradigm in heterogeneous platforms as found in homogeneous platforms. C++ revision 17 adds its first features for explicit parallelism through its “Parallel Standard Template Library” (PSTL). This paper discusses the technical issues in offloading PSTL on heterogeneous platforms supporting SSVM and presents a working GCC-based proof-of-concept implementation. Initial benchmarking of the implementation on an AMD Carrizo platform shows speedups from 1.28X to 12.78X in comparison to host-only sequential STL execution.

Alkuperäiskieli	Englanti
Otsikko	High Performance Computing
Alaotsikko	ISC High Performance 2018 International Workshops, Frankfurt/Main, Germany, June 24 - 28, 2018, Revised Selected Papers
Kustantaja	Springer
Sivut	637-647
ISBN (elektroninen)	978-3-030-02465-9
Tila	Julkaistu - 28 kesäk. 2018
OKM-julkaisutyyppi	A4 Artikkeli konferenssijulkaisussa
Tapahtuma	ISC High Performance International Workshops - Kesto: 28 kesäk. 2018 → 28 kesäk. 2018

Julkaisusarja

Nimi	Theoretical Computer Science and General Issues
Vuosikerta	11203
ISSN (painettu)	0302-9743

Conference

Conference	ISC High Performance International Workshops
Ajanjakso	28/06/18 → 28/06/18

Julkaisufoorumi-taso

Jufo-taso 1

Pääsy asiakirjaan

ParOffloadAccepted author manuscript, 217 KB

http://urn.fi/URN:NBN:fi:tty-201901221134

Siteeraa tätä

Jääskeläinen, P., Glossner, J., Jambor, M., Tervo, A., & Rintala, M. (2018). Offloading C++17 Parallel STL on System Shared Virtual Memory Platforms. teoksessa High Performance Computing: ISC High Performance 2018 International Workshops, Frankfurt/Main, Germany, June 24 - 28, 2018, Revised Selected Papers (Sivut 637-647). (Theoretical Computer Science and General Issues; Vuosikerta 11203). Springer. http://urn.fi/URN:NBN:fi:tty-201901221134

@inproceedings{5c4b3f67ac4a4216b89a32844e49ffae,

title = "Offloading C++17 Parallel STL on System Shared Virtual Memory Platforms",

abstract = "Shared virtual memory simplifies heterogeneous platform programming by enabling sharing of memory address pointers between heterogeneous devices in the platform. The most advanced implementations present a coherent view of memory to the programmer over the whole virtual address space of the process. From the point of view of data accesses, this System SVM (SSVM) enables the same programming paradigm in heterogeneous platforms as found in homogeneous platforms. C++ revision 17 adds its first features for explicit parallelism through its “Parallel Standard Template Library” (PSTL). This paper discusses the technical issues in offloading PSTL on heterogeneous platforms supporting SSVM and presents a working GCC-based proof-of-concept implementation. Initial benchmarking of the implementation on an AMD Carrizo platform shows speedups from 1.28X to 12.78X in comparison to host-only sequential STL execution.",

author = "Pekka J{\"a}{\"a}skel{\"a}inen and John Glossner and Martin Jambor and Aleksi Tervo and Matti Rintala",

note = "jufoid=62555; ISC High Performance International Workshops ; Conference date: 28-06-2018 Through 28-06-2018",

year = "2018",

month = jun,

day = "28",

language = "English",

series = "Theoretical Computer Science and General Issues",

publisher = "Springer",

pages = "637--647",

booktitle = "High Performance Computing",

}

Jääskeläinen, P, Glossner, J, Jambor, M, Tervo, A & Rintala, M 2018, Offloading C++17 Parallel STL on System Shared Virtual Memory Platforms. julkaisussa High Performance Computing: ISC High Performance 2018 International Workshops, Frankfurt/Main, Germany, June 24 - 28, 2018, Revised Selected Papers. Theoretical Computer Science and General Issues, Vuosikerta 11203, Springer, Sivut 637-647, ISC High Performance International Workshops, 28/06/18. <http://urn.fi/URN:NBN:fi:tty-201901221134>

Offloading C++17 Parallel STL on System Shared Virtual Memory Platforms. / Jääskeläinen, Pekka; Glossner, John; Jambor, Martin et al.
High Performance Computing: ISC High Performance 2018 International Workshops, Frankfurt/Main, Germany, June 24 - 28, 2018, Revised Selected Papers. Springer, 2018. s. 637-647 (Theoretical Computer Science and General Issues; Vuosikerta 11203).

Tutkimustuotos: Konferenssiartikkeli › Scientific › vertaisarvioitu

TY - GEN

T1 - Offloading C++17 Parallel STL on System Shared Virtual Memory Platforms

AU - Jääskeläinen, Pekka

AU - Glossner, John

AU - Jambor, Martin

AU - Tervo, Aleksi

AU - Rintala, Matti

N1 - jufoid=62555

PY - 2018/6/28

Y1 - 2018/6/28

N2 - Shared virtual memory simplifies heterogeneous platform programming by enabling sharing of memory address pointers between heterogeneous devices in the platform. The most advanced implementations present a coherent view of memory to the programmer over the whole virtual address space of the process. From the point of view of data accesses, this System SVM (SSVM) enables the same programming paradigm in heterogeneous platforms as found in homogeneous platforms. C++ revision 17 adds its first features for explicit parallelism through its “Parallel Standard Template Library” (PSTL). This paper discusses the technical issues in offloading PSTL on heterogeneous platforms supporting SSVM and presents a working GCC-based proof-of-concept implementation. Initial benchmarking of the implementation on an AMD Carrizo platform shows speedups from 1.28X to 12.78X in comparison to host-only sequential STL execution.

AB - Shared virtual memory simplifies heterogeneous platform programming by enabling sharing of memory address pointers between heterogeneous devices in the platform. The most advanced implementations present a coherent view of memory to the programmer over the whole virtual address space of the process. From the point of view of data accesses, this System SVM (SSVM) enables the same programming paradigm in heterogeneous platforms as found in homogeneous platforms. C++ revision 17 adds its first features for explicit parallelism through its “Parallel Standard Template Library” (PSTL). This paper discusses the technical issues in offloading PSTL on heterogeneous platforms supporting SSVM and presents a working GCC-based proof-of-concept implementation. Initial benchmarking of the implementation on an AMD Carrizo platform shows speedups from 1.28X to 12.78X in comparison to host-only sequential STL execution.

M3 - Conference contribution

T3 - Theoretical Computer Science and General Issues

SP - 637

EP - 647

BT - High Performance Computing

PB - Springer

T2 - ISC High Performance International Workshops

Y2 - 28 June 2018 through 28 June 2018

ER -

Jääskeläinen P, Glossner J, Jambor M, Tervo A, Rintala M. Offloading C++17 Parallel STL on System Shared Virtual Memory Platforms. julkaisussa High Performance Computing: ISC High Performance 2018 International Workshops, Frankfurt/Main, Germany, June 24 - 28, 2018, Revised Selected Papers. Springer. 2018. s. 637-647. (Theoretical Computer Science and General Issues).

Offloading C++17 Parallel STL on System Shared Virtual Memory Platforms

Abstrakti

Julkaisusarja

Conference

Julkaisufoorumi-taso

Pääsy asiakirjaan

Sormenjälki

Siteeraa tätä