VOBLA: A vehicle for optimized basic linear algebra

Ulysse Beaugnon; Alexey Kravets; Sven Van Haastregt; Riyadh Baghdadi; David Tweed; Javed Absar; Anton Lokhmotov

doi:10.1145/2597809.2597818

VOBLA: A vehicle for optimized basic linear algebra

Ulysse Beaugnon, Alexey Kravets, Sven Van Haastregt, Riyadh Baghdadi, David Tweed, Javed Absar, Anton Lokhmotov

Computer Science

Research output: Contribution to journal › Article › peer-review

Abstract

We present VOBLA, a domain-specific language designed for programming linear algebra libraries. VOBLA is compiled to PENCIL, a domain independent intermediate language designed for efficient mapping to accelerator architectures such as GPGPUs. PENCIL is compiled to efficient, platform-specific OpenCL code using techniques based on the polyhedral model. This approach addresses both the programmer productivity and performance portability concerns associated with accelerator programming.

We demonstrate our approach by using VOBLA to implement a BLAS library. We have evaluated the performance of OpenCL code generated using our compilation flow on ARM Mali, AMD Radeon, and AMD Opteron platforms. The generated code is currently on average 1:9× slower than highly hand-optimized OpenCL code, but on average 8:1× faster than straightforward OpenCL code. Given that the VOBLA coding takes significantly less effort compared to hand-optimizing OpenCL code, we believe our approach leads to improved productivity and performance portability. Copyright is held by the owner/author(s).

Original language	English (US)
Pages (from-to)	115-124
Number of pages	10
Journal	ACM SIGPLAN Notices
Volume	49
Issue number	5
DOIs	https://doi.org/10.1145/2597809.2597818
State	Published - May 5 2014

Keywords

BLAS
Domain-specific language
GPU
Linear algebra
Parallel
Sparse matrix

ASJC Scopus subject areas

General Computer Science

Access to Document

10.1145/2597809.2597818

Cite this

@article{1f5c9d8290444308a27c65b31d0b5eb4,

title = "VOBLA: A vehicle for optimized basic linear algebra",

abstract = "We present VOBLA, a domain-specific language designed for programming linear algebra libraries. VOBLA is compiled to PENCIL, a domain independent intermediate language designed for efficient mapping to accelerator architectures such as GPGPUs. PENCIL is compiled to efficient, platform-specific OpenCL code using techniques based on the polyhedral model. This approach addresses both the programmer productivity and performance portability concerns associated with accelerator programming.We demonstrate our approach by using VOBLA to implement a BLAS library. We have evaluated the performance of OpenCL code generated using our compilation flow on ARM Mali, AMD Radeon, and AMD Opteron platforms. The generated code is currently on average 1:9× slower than highly hand-optimized OpenCL code, but on average 8:1× faster than straightforward OpenCL code. Given that the VOBLA coding takes significantly less effort compared to hand-optimizing OpenCL code, we believe our approach leads to improved productivity and performance portability. Copyright is held by the owner/author(s).",

keywords = "BLAS, Domain-specific language, GPU, Linear algebra, Parallel, Sparse matrix",

author = "Ulysse Beaugnon and Alexey Kravets and {Van Haastregt}, Sven and Riyadh Baghdadi and David Tweed and Javed Absar and Anton Lokhmotov",

year = "2014",

month = may,

day = "5",

doi = "10.1145/2597809.2597818",

language = "English (US)",

volume = "49",

pages = "115--124",

journal = "ACM SIGPLAN Notices",

issn = "1523-2867",

publisher = "Association for Computing Machinery (ACM)",

number = "5",

}

TY - JOUR

T1 - VOBLA

T2 - A vehicle for optimized basic linear algebra

AU - Beaugnon, Ulysse

AU - Kravets, Alexey

AU - Van Haastregt, Sven

AU - Baghdadi, Riyadh

AU - Tweed, David

AU - Absar, Javed

AU - Lokhmotov, Anton

PY - 2014/5/5

Y1 - 2014/5/5

N2 - We present VOBLA, a domain-specific language designed for programming linear algebra libraries. VOBLA is compiled to PENCIL, a domain independent intermediate language designed for efficient mapping to accelerator architectures such as GPGPUs. PENCIL is compiled to efficient, platform-specific OpenCL code using techniques based on the polyhedral model. This approach addresses both the programmer productivity and performance portability concerns associated with accelerator programming.We demonstrate our approach by using VOBLA to implement a BLAS library. We have evaluated the performance of OpenCL code generated using our compilation flow on ARM Mali, AMD Radeon, and AMD Opteron platforms. The generated code is currently on average 1:9× slower than highly hand-optimized OpenCL code, but on average 8:1× faster than straightforward OpenCL code. Given that the VOBLA coding takes significantly less effort compared to hand-optimizing OpenCL code, we believe our approach leads to improved productivity and performance portability. Copyright is held by the owner/author(s).

AB - We present VOBLA, a domain-specific language designed for programming linear algebra libraries. VOBLA is compiled to PENCIL, a domain independent intermediate language designed for efficient mapping to accelerator architectures such as GPGPUs. PENCIL is compiled to efficient, platform-specific OpenCL code using techniques based on the polyhedral model. This approach addresses both the programmer productivity and performance portability concerns associated with accelerator programming.We demonstrate our approach by using VOBLA to implement a BLAS library. We have evaluated the performance of OpenCL code generated using our compilation flow on ARM Mali, AMD Radeon, and AMD Opteron platforms. The generated code is currently on average 1:9× slower than highly hand-optimized OpenCL code, but on average 8:1× faster than straightforward OpenCL code. Given that the VOBLA coding takes significantly less effort compared to hand-optimizing OpenCL code, we believe our approach leads to improved productivity and performance portability. Copyright is held by the owner/author(s).

KW - BLAS

KW - Domain-specific language

KW - GPU

KW - Linear algebra

KW - Parallel

KW - Sparse matrix

UR - http://www.scopus.com/inward/record.url?scp=84907027213&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84907027213&partnerID=8YFLogxK

U2 - 10.1145/2597809.2597818

DO - 10.1145/2597809.2597818

M3 - Article

AN - SCOPUS:84907027213

SN - 1523-2867

VL - 49

SP - 115

EP - 124

JO - ACM SIGPLAN Notices

JF - ACM SIGPLAN Notices

IS - 5

ER -

VOBLA: A vehicle for optimized basic linear algebra

Abstract

Keywords

ASJC Scopus subject areas

Access to Document

Other files and links

Fingerprint

Cite this