Tuning block size for QR factorization on CPU-GPU hybrid systems

Yaohung M. Tsai; Weichung Wang; Ray-Bing Chen

doi:10.1109/MCSoC.2012.32

Tuning block size for QR factorization on CPU-GPU hybrid systems

Yaohung M. Tsai, Weichung Wang, Ray-Bing Chen

Department of Statistics

Research output: Contribution to conference › Paper › peer-review

7 Citations (Scopus)

Abstract

In CPU-GPU hybrid systems, the QR factorization in MAGMA results in CPU idle due to the fixed block size. To improve the computational efficiency of MAGMA QR factorization, we propose a variable block size auto-tuning scheme on CPU-GPU hybrid systems. First, we fit the CPU and GPU costs in MAGMA QR factorization via two independent regression models as CPU and GPU performance models. Next, we propose a block size optimization scheme to tune the block size adaptively and therefore to minimize a cost objective function. The cost objective function is designed to balance the workloads between CPU and GPU based on the performance models. Finally, several numerical results demonstrate the performance gains due to the novel QR factorization algorithm.

Original language	English
Pages	205-211
Number of pages	7
DOIs	https://doi.org/10.1109/MCSoC.2012.32
Publication status	Published - 2012 Dec 1
Event	2012 IEEE 6th International Symposium on Embedded Multi-Core Systems on Chips, MCSoC 2012 - Aizu-Wakamatsu, Fukushima, Japan Duration: 2012 Sept 20 → 2012 Sept 22

Other

Other	2012 IEEE 6th International Symposium on Embedded Multi-Core Systems on Chips, MCSoC 2012
Country/Territory	Japan
City	Aizu-Wakamatsu, Fukushima
Period	12-09-20 → 12-09-22

All Science Journal Classification (ASJC) codes

Hardware and Architecture
Electrical and Electronic Engineering

Access to Document

10.1109/MCSoC.2012.32

Cite this

@conference{fc3f7d7f8bfc454ca9bfd12aeb87b29f,

title = "Tuning block size for QR factorization on CPU-GPU hybrid systems",

abstract = "In CPU-GPU hybrid systems, the QR factorization in MAGMA results in CPU idle due to the fixed block size. To improve the computational efficiency of MAGMA QR factorization, we propose a variable block size auto-tuning scheme on CPU-GPU hybrid systems. First, we fit the CPU and GPU costs in MAGMA QR factorization via two independent regression models as CPU and GPU performance models. Next, we propose a block size optimization scheme to tune the block size adaptively and therefore to minimize a cost objective function. The cost objective function is designed to balance the workloads between CPU and GPU based on the performance models. Finally, several numerical results demonstrate the performance gains due to the novel QR factorization algorithm.",

author = "Tsai, {Yaohung M.} and Weichung Wang and Ray-Bing Chen",

year = "2012",

month = dec,

day = "1",

doi = "10.1109/MCSoC.2012.32",

language = "English",

pages = "205--211",

note = "2012 IEEE 6th International Symposium on Embedded Multi-Core Systems on Chips, MCSoC 2012 ; Conference date: 20-09-2012 Through 22-09-2012",

}

TY - CONF

T1 - Tuning block size for QR factorization on CPU-GPU hybrid systems

AU - Tsai, Yaohung M.

AU - Wang, Weichung

AU - Chen, Ray-Bing

PY - 2012/12/1

Y1 - 2012/12/1

N2 - In CPU-GPU hybrid systems, the QR factorization in MAGMA results in CPU idle due to the fixed block size. To improve the computational efficiency of MAGMA QR factorization, we propose a variable block size auto-tuning scheme on CPU-GPU hybrid systems. First, we fit the CPU and GPU costs in MAGMA QR factorization via two independent regression models as CPU and GPU performance models. Next, we propose a block size optimization scheme to tune the block size adaptively and therefore to minimize a cost objective function. The cost objective function is designed to balance the workloads between CPU and GPU based on the performance models. Finally, several numerical results demonstrate the performance gains due to the novel QR factorization algorithm.

AB - In CPU-GPU hybrid systems, the QR factorization in MAGMA results in CPU idle due to the fixed block size. To improve the computational efficiency of MAGMA QR factorization, we propose a variable block size auto-tuning scheme on CPU-GPU hybrid systems. First, we fit the CPU and GPU costs in MAGMA QR factorization via two independent regression models as CPU and GPU performance models. Next, we propose a block size optimization scheme to tune the block size adaptively and therefore to minimize a cost objective function. The cost objective function is designed to balance the workloads between CPU and GPU based on the performance models. Finally, several numerical results demonstrate the performance gains due to the novel QR factorization algorithm.

UR - http://www.scopus.com/inward/record.url?scp=84872539240&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84872539240&partnerID=8YFLogxK

U2 - 10.1109/MCSoC.2012.32

DO - 10.1109/MCSoC.2012.32

M3 - Paper

AN - SCOPUS:84872539240

SP - 205

EP - 211

T2 - 2012 IEEE 6th International Symposium on Embedded Multi-Core Systems on Chips, MCSoC 2012

Y2 - 20 September 2012 through 22 September 2012

ER -

Tuning block size for QR factorization on CPU-GPU hybrid systems

Abstract

Other

All Science Journal Classification (ASJC) codes

Access to Document

Other files and links

Fingerprint

Cite this