Automatic matching of legacy code to heterogeneous APIs: An idiomatic approach

Philip Ginsbach; Toomas Remmelg; Michel Steuwer; Bruno Bodin; Christophe Dubach; Michael O'Boyle

doi:10.1145/3173162.3173182

Automatic matching of legacy code to heterogeneous APIs: An idiomatic approach

Philip Ginsbach, Toomas Remmelg, Michel Steuwer, Bruno Bodin, Christophe Dubach, Michael O'Boyle

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

Abstract / Description of output

Heterogeneous accelerators often disappoint. They provide the prospect of great performance, but only deliver it when using vendor specific optimized libraries or domain specific languages. This requires considerable legacy code modifications, hindering the adoption of heterogeneous computing. This paper develops a novel approach that automatically detects opportunities to exploit accelerators. We focus on calculations that are well supported by established APIs: sparse and dense linear algebra, stencils and generalized reductions and histograms.We call such opportunities idioms and use a custom constraint-based Idiom Description Language (IDL) to discover them within user code. Detected idioms are then mapped to BLAS libraries, cuSPARSE and clSPARSE and two DSLs: Halide and Lift. We implemented the approach in LLVM and evaluated it on the NAS and Parboil sequential C/C++ benchmarks, where we detect 60 idiom instances. In those cases where idioms are a significant part of the sequential execution time, we generate code that achieves 1.26× to over 20× speedup on integrated and external GPUs.

Original language	English
Title of host publication	23rd ACM International Conference on Architectural Support for Programming Languages and Operating Systems (ASPLOS'18)
Publisher	ACM
Pages	139-153
Number of pages	14
ISBN (Electronic)	978-1-4503-4911-6
DOIs	https://doi.org/10.1145/3173162.3173182
Publication status	Published - 19 Mar 2018
Event	The 23rd ACM International Conference on Architectural Support for Programming Languages and Operating Systems - Williamsburg, United States Duration: 24 Mar 2018 → 28 Mar 2018 https://www.asplos2018.org/

Publication series

Name	ACM SIGPLAN Notices
Publisher	ACM
Number	2
Volume	53
ISSN (Print)	0362-1340
ISSN (Electronic)	1558-1160

Conference

Conference	The 23rd ACM International Conference on Architectural Support for Programming Languages and Operating Systems
Abbreviated title	ASPLOS2018
Country/Territory	United States
City	Williamsburg
Period	24/03/18 → 28/03/18
Internet address	https://www.asplos2018.org/

Access to Document

10.1145/3173162.3173182

asplosAccepted author manuscript, 651 KB

https://dl.acm.org/citation.cfm?doid=3173162.3173182

Pattern Discovery and Program Shaping for Heterogeneous Manycore Systems
Cole, M., Franke, B. & O'Boyle, M.
EPSRC
1/07/17 → 31/12/20
Project: Research

Michael O'Boyle
- mobinf.ed.acuk
- School of Informatics - Personal Chair in Computer Science
- Institute for Computing Systems Architecture
- Computer Systems
Person: Academic: Research Active

Cite this

Ginsbach, P., Remmelg, T., Steuwer, M., Bodin, B., Dubach, C., & O'Boyle, M. (2018). Automatic matching of legacy code to heterogeneous APIs: An idiomatic approach. In 23rd ACM International Conference on Architectural Support for Programming Languages and Operating Systems (ASPLOS'18) (pp. 139-153). (ACM SIGPLAN Notices; Vol. 53, No. 2). ACM. https://doi.org/10.1145/3173162.3173182

@inproceedings{74bf9dba7d6146f3883b0f6fff11dec1,

title = "Automatic matching of legacy code to heterogeneous APIs: An idiomatic approach",

abstract = "Heterogeneous accelerators often disappoint. They provide the prospect of great performance, but only deliver it when using vendor specific optimized libraries or domain specific languages. This requires considerable legacy code modifications, hindering the adoption of heterogeneous computing. This paper develops a novel approach that automatically detects opportunities to exploit accelerators. We focus on calculations that are well supported by established APIs: sparse and dense linear algebra, stencils and generalized reductions and histograms.We call such opportunities idioms and use a custom constraint-based Idiom Description Language (IDL) to discover them within user code. Detected idioms are then mapped to BLAS libraries, cuSPARSE and clSPARSE and two DSLs: Halide and Lift. We implemented the approach in LLVM and evaluated it on the NAS and Parboil sequential C/C++ benchmarks, where we detect 60 idiom instances. In those cases where idioms are a significant part of the sequential execution time, we generate code that achieves 1.26× to over 20× speedup on integrated and external GPUs.",

author = "Philip Ginsbach and Toomas Remmelg and Michel Steuwer and Bruno Bodin and Christophe Dubach and Michael O'Boyle",

year = "2018",

month = mar,

day = "19",

doi = "10.1145/3173162.3173182",

language = "English",

series = "ACM SIGPLAN Notices",

publisher = "ACM",

number = "2",

pages = "139--153",

booktitle = "23rd ACM International Conference on Architectural Support for Programming Languages and Operating Systems (ASPLOS'18)",

note = "The 23rd ACM International Conference on Architectural Support for Programming Languages and Operating Systems, ASPLOS2018 ; Conference date: 24-03-2018 Through 28-03-2018",

url = "https://www.asplos2018.org/",

}

Ginsbach, P, Remmelg, T, Steuwer, M, Bodin, B, Dubach, C & O'Boyle, M 2018, Automatic matching of legacy code to heterogeneous APIs: An idiomatic approach. in 23rd ACM International Conference on Architectural Support for Programming Languages and Operating Systems (ASPLOS'18). ACM SIGPLAN Notices, no. 2, vol. 53, ACM, pp. 139-153, The 23rd ACM International Conference on Architectural Support for Programming Languages and Operating Systems, Williamsburg, United States, 24/03/18. https://doi.org/10.1145/3173162.3173182

Automatic matching of legacy code to heterogeneous APIs: An idiomatic approach. / Ginsbach, Philip; Remmelg, Toomas; Steuwer, Michel et al.
23rd ACM International Conference on Architectural Support for Programming Languages and Operating Systems (ASPLOS'18). ACM, 2018. p. 139-153 (ACM SIGPLAN Notices; Vol. 53, No. 2).

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

TY - GEN

T1 - Automatic matching of legacy code to heterogeneous APIs: An idiomatic approach

AU - Ginsbach, Philip

AU - Remmelg, Toomas

AU - Steuwer, Michel

AU - Bodin, Bruno

AU - Dubach, Christophe

AU - O'Boyle, Michael

PY - 2018/3/19

Y1 - 2018/3/19

N2 - Heterogeneous accelerators often disappoint. They provide the prospect of great performance, but only deliver it when using vendor specific optimized libraries or domain specific languages. This requires considerable legacy code modifications, hindering the adoption of heterogeneous computing. This paper develops a novel approach that automatically detects opportunities to exploit accelerators. We focus on calculations that are well supported by established APIs: sparse and dense linear algebra, stencils and generalized reductions and histograms.We call such opportunities idioms and use a custom constraint-based Idiom Description Language (IDL) to discover them within user code. Detected idioms are then mapped to BLAS libraries, cuSPARSE and clSPARSE and two DSLs: Halide and Lift. We implemented the approach in LLVM and evaluated it on the NAS and Parboil sequential C/C++ benchmarks, where we detect 60 idiom instances. In those cases where idioms are a significant part of the sequential execution time, we generate code that achieves 1.26× to over 20× speedup on integrated and external GPUs.

AB - Heterogeneous accelerators often disappoint. They provide the prospect of great performance, but only deliver it when using vendor specific optimized libraries or domain specific languages. This requires considerable legacy code modifications, hindering the adoption of heterogeneous computing. This paper develops a novel approach that automatically detects opportunities to exploit accelerators. We focus on calculations that are well supported by established APIs: sparse and dense linear algebra, stencils and generalized reductions and histograms.We call such opportunities idioms and use a custom constraint-based Idiom Description Language (IDL) to discover them within user code. Detected idioms are then mapped to BLAS libraries, cuSPARSE and clSPARSE and two DSLs: Halide and Lift. We implemented the approach in LLVM and evaluated it on the NAS and Parboil sequential C/C++ benchmarks, where we detect 60 idiom instances. In those cases where idioms are a significant part of the sequential execution time, we generate code that achieves 1.26× to over 20× speedup on integrated and external GPUs.

U2 - 10.1145/3173162.3173182

DO - 10.1145/3173162.3173182

M3 - Conference contribution

T3 - ACM SIGPLAN Notices

SP - 139

EP - 153

BT - 23rd ACM International Conference on Architectural Support for Programming Languages and Operating Systems (ASPLOS'18)

PB - ACM

T2 - The 23rd ACM International Conference on Architectural Support for Programming Languages and Operating Systems

Y2 - 24 March 2018 through 28 March 2018

ER -

Automatic matching of legacy code to heterogeneous APIs: An idiomatic approach

Abstract / Description of output

Publication series

Conference

Access to Document

Fingerprint

Projects

Pattern Discovery and Program Shaping for Heterogeneous Manycore Systems

Profiles

Michael O'Boyle

Cite this