On the use of small 2D convolutions on GPUs

S.A.H. Al Umairy; I.D. Setija; M.C. Beurden, van; H.J. Sips; A.S. Amesfoort, van

On the use of small 2D convolutions on GPUs

S.A.H. Al Umairy, I.D. Setija, M.C. Beurden, van, H.J. Sips, A.S. Amesfoort, van

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

2 Downloads (Pure)

Abstract

Computing many small 2D convolutions using FFTs is a basis for a large number of applications in many domains in science and engineering, among them electromagnetic di??raction modeling in physics. The GPU architecture seems to be a suitable architecture to accelerate these convolutions, but reaching high application performance requires substantial development time and non-portable optimizations. In this work, we present the techniques, performance results and considerations to accelerate small 2D convolutions using CUDA, and compare performance to a multi-threaded CPU implementation. To improve programmability and performance of applications that make heavy use of small convolutions, we argue that two improvements to software and hardware are needed: FFT libraries must be extended with a single convolution function and communication bandwidth between CPU and GPU needs to be drastically improved.

Original language	English
Title of host publication	Proceedings of the First Workshop on Applications for Multi and Many Core Processors, A4MMC 2010, held in conjunction with ISCA 2010, 19 June 2010, St. Malo, France
Place of Publication	Z.pl.
Publisher	s.n.
Pages	52-64
Publication status	Published - 2010

Fingerprint Dive into the research topics of 'On the use of small 2D convolutions on GPUs'. Together they form a unique fingerprint.

Cite this

@inproceedings{559810672cae480cb11684c75ab25836,

title = "On the use of small 2D convolutions on GPUs",

abstract = "Computing many small 2D convolutions using FFTs is a basis for a large number of applications in many domains in science and engineering, among them electromagnetic di??raction modeling in physics. The GPU architecture seems to be a suitable architecture to accelerate these convolutions, but reaching high application performance requires substantial development time and non-portable optimizations. In this work, we present the techniques, performance results and considerations to accelerate small 2D convolutions using CUDA, and compare performance to a multi-threaded CPU implementation. To improve programmability and performance of applications that make heavy use of small convolutions, we argue that two improvements to software and hardware are needed: FFT libraries must be extended with a single convolution function and communication bandwidth between CPU and GPU needs to be drastically improved.",

author = "{Al Umairy}, S.A.H. and I.D. Setija and {Beurden, van}, M.C. and H.J. Sips and {Amesfoort, van}, A.S.",

year = "2010",

language = "English",

pages = "52--64",

booktitle = "Proceedings of the First Workshop on Applications for Multi and Many Core Processors, A4MMC 2010, held in conjunction with ISCA 2010, 19 June 2010, St. Malo, France",

publisher = "s.n.",

}

On the use of small 2D convolutions on GPUs. / Al Umairy, S.A.H.; Setija, I.D.; Beurden, van, M.C.; Sips, H.J.; Amesfoort, van, A.S.

Proceedings of the First Workshop on Applications for Multi and Many Core Processors, A4MMC 2010, held in conjunction with ISCA 2010, 19 June 2010, St. Malo, France. Z.pl. : s.n., 2010. p. 52-64.

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

TY - GEN

T1 - On the use of small 2D convolutions on GPUs

AU - Al Umairy, S.A.H.

AU - Setija, I.D.

AU - Beurden, van, M.C.

AU - Sips, H.J.

AU - Amesfoort, van, A.S.

PY - 2010

Y1 - 2010

N2 - Computing many small 2D convolutions using FFTs is a basis for a large number of applications in many domains in science and engineering, among them electromagnetic di??raction modeling in physics. The GPU architecture seems to be a suitable architecture to accelerate these convolutions, but reaching high application performance requires substantial development time and non-portable optimizations. In this work, we present the techniques, performance results and considerations to accelerate small 2D convolutions using CUDA, and compare performance to a multi-threaded CPU implementation. To improve programmability and performance of applications that make heavy use of small convolutions, we argue that two improvements to software and hardware are needed: FFT libraries must be extended with a single convolution function and communication bandwidth between CPU and GPU needs to be drastically improved.

AB - Computing many small 2D convolutions using FFTs is a basis for a large number of applications in many domains in science and engineering, among them electromagnetic di??raction modeling in physics. The GPU architecture seems to be a suitable architecture to accelerate these convolutions, but reaching high application performance requires substantial development time and non-portable optimizations. In this work, we present the techniques, performance results and considerations to accelerate small 2D convolutions using CUDA, and compare performance to a multi-threaded CPU implementation. To improve programmability and performance of applications that make heavy use of small convolutions, we argue that two improvements to software and hardware are needed: FFT libraries must be extended with a single convolution function and communication bandwidth between CPU and GPU needs to be drastically improved.

M3 - Conference contribution

SP - 52

EP - 64

BT - Proceedings of the First Workshop on Applications for Multi and Many Core Processors, A4MMC 2010, held in conjunction with ISCA 2010, 19 June 2010, St. Malo, France

PB - s.n.

CY - Z.pl.

ER -