Optimizing GPU Kernels for Irregular Batch Workloads: A Case Study for Cholesky Factorization

Ahmad Abdelfattah; Azzam Haidar; Stanimire Tomov; Jack Dongarra

doi:10.1109/HPEC.2018.8547576

Optimizing GPU Kernels for Irregular Batch Workloads: A Case Study for Cholesky Factorization

Ahmad Abdelfattah, Azzam Haidar, Stanimire Tomov, Jack Dongarra

Computer Science and Mathematics Div

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

2 Scopus citations

Abstract

This paper introduces several frameworks for the design and implementation of high performance GPU kernels that target batch workloads with irregular sizes. Such workloads are ubiquitous in many scientific applications, including sparse direct solvers, astrophysics, and quantum chemistry. The paper addresses two main categories of frameworks, taking the Cholesky factorization as a case study. The first uses host-side kernel launches, and the second uses device-side launches. Within each category, different design options are introduced, with an emphasis on the advantages and the disadvantages of each approach. Our best performing design outperforms the state-of-the-art CPU implementation, scoring up to 4.7× speedup in double precision on a Pascal P100 GPU.

Original language	English
Title of host publication	2018 IEEE High Performance Extreme Computing Conference, HPEC 2018
Publisher	Institute of Electrical and Electronics Engineers Inc.
ISBN (Electronic)	9781538659892
DOIs	https://doi.org/10.1109/HPEC.2018.8547576
State	Published - Nov 26 2018
Event	2018 IEEE High Performance Extreme Computing Conference, HPEC 2018 - Waltham, United States Duration: Sep 25 2018 → Sep 27 2018

Publication series

Name	2018 IEEE High Performance Extreme Computing Conference, HPEC 2018

Conference

Conference	2018 IEEE High Performance Extreme Computing Conference, HPEC 2018
Country/Territory	United States
City	Waltham
Period	09/25/18 → 09/27/18

Funding

This work is partially supported by NSF grant No. OAC-1740250 and CSR 1514286, NVIDIA, and by the Exascale Computing Project (17-SC-20-SC). This work is partially supported by NSF grant No. OAC-1740250 and CSR 1514286, NVIDIA, and by the Exascale Computing Project (17-SC-20-SC)

Keywords

Batch Linear Algebra
GPU Computing
Matrix Factorization

Access to Document

10.1109/HPEC.2018.8547576

Cite this

Abdelfattah, A., Haidar, A., Tomov, S., & Dongarra, J. (2018). Optimizing GPU Kernels for Irregular Batch Workloads: A Case Study for Cholesky Factorization. In 2018 IEEE High Performance Extreme Computing Conference, HPEC 2018 Article 8547576 (2018 IEEE High Performance Extreme Computing Conference, HPEC 2018). Institute of Electrical and Electronics Engineers Inc.. https://doi.org/10.1109/HPEC.2018.8547576

@inproceedings{03d9b185701148caa94b6ec7827fd380,

title = "Optimizing GPU Kernels for Irregular Batch Workloads: A Case Study for Cholesky Factorization",

abstract = "This paper introduces several frameworks for the design and implementation of high performance GPU kernels that target batch workloads with irregular sizes. Such workloads are ubiquitous in many scientific applications, including sparse direct solvers, astrophysics, and quantum chemistry. The paper addresses two main categories of frameworks, taking the Cholesky factorization as a case study. The first uses host-side kernel launches, and the second uses device-side launches. Within each category, different design options are introduced, with an emphasis on the advantages and the disadvantages of each approach. Our best performing design outperforms the state-of-the-art CPU implementation, scoring up to 4.7× speedup in double precision on a Pascal P100 GPU.",

keywords = "Batch Linear Algebra, GPU Computing, Matrix Factorization",

author = "Ahmad Abdelfattah and Azzam Haidar and Stanimire Tomov and Jack Dongarra",

note = "Publisher Copyright: {\textcopyright} 2018 IEEE.; 2018 IEEE High Performance Extreme Computing Conference, HPEC 2018 ; Conference date: 25-09-2018 Through 27-09-2018",

year = "2018",

month = nov,

day = "26",

doi = "10.1109/HPEC.2018.8547576",

language = "English",

series = "2018 IEEE High Performance Extreme Computing Conference, HPEC 2018",

publisher = "Institute of Electrical and Electronics Engineers Inc.",

booktitle = "2018 IEEE High Performance Extreme Computing Conference, HPEC 2018",

}

Abdelfattah, A, Haidar, A, Tomov, S & Dongarra, J 2018, Optimizing GPU Kernels for Irregular Batch Workloads: A Case Study for Cholesky Factorization. in 2018 IEEE High Performance Extreme Computing Conference, HPEC 2018., 8547576, 2018 IEEE High Performance Extreme Computing Conference, HPEC 2018, Institute of Electrical and Electronics Engineers Inc., 2018 IEEE High Performance Extreme Computing Conference, HPEC 2018, Waltham, United States, 09/25/18. https://doi.org/10.1109/HPEC.2018.8547576

Optimizing GPU Kernels for Irregular Batch Workloads: A Case Study for Cholesky Factorization. / Abdelfattah, Ahmad; Haidar, Azzam; Tomov, Stanimire et al.
2018 IEEE High Performance Extreme Computing Conference, HPEC 2018. Institute of Electrical and Electronics Engineers Inc., 2018. 8547576 (2018 IEEE High Performance Extreme Computing Conference, HPEC 2018).

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

TY - GEN

T1 - Optimizing GPU Kernels for Irregular Batch Workloads

T2 - 2018 IEEE High Performance Extreme Computing Conference, HPEC 2018

AU - Abdelfattah, Ahmad

AU - Haidar, Azzam

AU - Tomov, Stanimire

AU - Dongarra, Jack

PY - 2018/11/26

Y1 - 2018/11/26

N2 - This paper introduces several frameworks for the design and implementation of high performance GPU kernels that target batch workloads with irregular sizes. Such workloads are ubiquitous in many scientific applications, including sparse direct solvers, astrophysics, and quantum chemistry. The paper addresses two main categories of frameworks, taking the Cholesky factorization as a case study. The first uses host-side kernel launches, and the second uses device-side launches. Within each category, different design options are introduced, with an emphasis on the advantages and the disadvantages of each approach. Our best performing design outperforms the state-of-the-art CPU implementation, scoring up to 4.7× speedup in double precision on a Pascal P100 GPU.

AB - This paper introduces several frameworks for the design and implementation of high performance GPU kernels that target batch workloads with irregular sizes. Such workloads are ubiquitous in many scientific applications, including sparse direct solvers, astrophysics, and quantum chemistry. The paper addresses two main categories of frameworks, taking the Cholesky factorization as a case study. The first uses host-side kernel launches, and the second uses device-side launches. Within each category, different design options are introduced, with an emphasis on the advantages and the disadvantages of each approach. Our best performing design outperforms the state-of-the-art CPU implementation, scoring up to 4.7× speedup in double precision on a Pascal P100 GPU.

KW - Batch Linear Algebra

KW - GPU Computing

KW - Matrix Factorization

UR - http://www.scopus.com/inward/record.url?scp=85060088912&partnerID=8YFLogxK

U2 - 10.1109/HPEC.2018.8547576

DO - 10.1109/HPEC.2018.8547576

M3 - Conference contribution

AN - SCOPUS:85060088912

T3 - 2018 IEEE High Performance Extreme Computing Conference, HPEC 2018

BT - 2018 IEEE High Performance Extreme Computing Conference, HPEC 2018

PB - Institute of Electrical and Electronics Engineers Inc.

Y2 - 25 September 2018 through 27 September 2018

ER -

Abdelfattah A, Haidar A, Tomov S, Dongarra J. Optimizing GPU Kernels for Irregular Batch Workloads: A Case Study for Cholesky Factorization. In 2018 IEEE High Performance Extreme Computing Conference, HPEC 2018. Institute of Electrical and Electronics Engineers Inc. 2018. 8547576. (2018 IEEE High Performance Extreme Computing Conference, HPEC 2018). doi: 10.1109/HPEC.2018.8547576

Optimizing GPU Kernels for Irregular Batch Workloads: A Case Study for Cholesky Factorization

Abstract

Publication series

Conference

Funding

Keywords

Access to Document

Other files and links

Fingerprint

Cite this