Optimization through recomputation in the polyhedral model

M.H. Jongen; L.J.W. Waeijen; R. Jordans; L. Jozwiak; H. Corporaal

Optimization through recomputation in the polyhedral model

M.H. Jongen, L.J.W. Waeijen, R. Jordans, L. Jozwiak, H. Corporaal

Electronic Systems

Onderzoeksoutput: Hoofdstuk in Boek/Rapport/Congresprocedure › Conferentiebijdrage › Academic › peer review

233 Downloads (Pure)

Samenvatting

Many modern (mobile) systems involve memory intensive computations. External memory accesses are costly when it comes to the execution time and energy consumption of a program. To overcome this, we usually apply tiling to improve data locality and data reuse in internal memories. In the research reported in this paper we add the possibility to recompute data rather than storing temporary results, and demonstrate that this can have a positive e ect on the overall application performance.
To achieve this we represented recomputation in the Polyhedral model by extending Polly. We experimentally veri ed the e ectiveness of recomputation on a pair of Convolutional Neural Network layers, when applying loop tiling, loop fusion, and recompute.

Originele taal-2	Engels
Titel	Eighth International Workshop on Polyhedral Compilation Techniques
Subtitel	In conjunction with HiPEAC 2018
Aantal pagina's	9
Status	Gepubliceerd - 22 jan. 2018
Evenement	8th International Workshop on Polyhedral Compilation Techniques - Manchester, Verenigd Koninkrijk Duur: 23 jan. 2018 → 23 jan. 2018 Congresnummer: 2018 http://impact.gforge.inria.fr/impact2018

Workshop

Workshop	8th International Workshop on Polyhedral Compilation Techniques
Verkorte titel	IMPACT
Land/Regio	Verenigd Koninkrijk
Stad	Manchester
Periode	23/01/18 → 23/01/18
Internet adres	http://impact.gforge.inria.fr/impact2018

Duurzame ontwikkelingsdoelstellingen van de VN

Deze output draagt bij aan de volgende duurzame ontwikkelingsdoelstelling(en)

Toegang tot document

submittedDefinitieve gepubliceerde versie, 948 KBLicentie: Onbepaald

Citeer dit

@inproceedings{e8d63c02f2c847308155000184cc6d16,

title = "Optimization through recomputation in the polyhedral model",

abstract = "Many modern (mobile) systems involve memory intensive computations. External memory accesses are costly when it comes to the execution time and energy consumption of a program. To overcome this, we usually apply tiling to improve data locality and data reuse in internal memories. In the research reported in this paper we add the possibility to recompute data rather than storing temporary results, and demonstrate that this can have a positive e ect on the overall application performance. To achieve this we represented recomputation in the Polyhedral model by extending Polly. We experimentally veri ed the e ectiveness of recomputation on a pair of Convolutional Neural Network layers, when applying loop tiling, loop fusion, and recompute.",

author = "M.H. Jongen and L.J.W. Waeijen and R. Jordans and L. Jozwiak and H. Corporaal",

year = "2018",

month = jan,

day = "22",

language = "English",

booktitle = "Eighth International Workshop on Polyhedral Compilation Techniques",

note = "8th International Workshop on Polyhedral Compilation Techniques (IMPACT 2018), January 23, 2018, Manchester, UK, IMPACT ; Conference date: 23-01-2018 Through 23-01-2018",

url = "http://impact.gforge.inria.fr/impact2018",

}

TY - GEN

T1 - Optimization through recomputation in the polyhedral model

AU - Jongen, M.H.

AU - Waeijen, L.J.W.

AU - Jordans, R.

AU - Jozwiak, L.

AU - Corporaal, H.

N1 - Conference code: 2018

PY - 2018/1/22

Y1 - 2018/1/22

N2 - Many modern (mobile) systems involve memory intensive computations. External memory accesses are costly when it comes to the execution time and energy consumption of a program. To overcome this, we usually apply tiling to improve data locality and data reuse in internal memories. In the research reported in this paper we add the possibility to recompute data rather than storing temporary results, and demonstrate that this can have a positive e ect on the overall application performance. To achieve this we represented recomputation in the Polyhedral model by extending Polly. We experimentally veri ed the e ectiveness of recomputation on a pair of Convolutional Neural Network layers, when applying loop tiling, loop fusion, and recompute.

AB - Many modern (mobile) systems involve memory intensive computations. External memory accesses are costly when it comes to the execution time and energy consumption of a program. To overcome this, we usually apply tiling to improve data locality and data reuse in internal memories. In the research reported in this paper we add the possibility to recompute data rather than storing temporary results, and demonstrate that this can have a positive e ect on the overall application performance. To achieve this we represented recomputation in the Polyhedral model by extending Polly. We experimentally veri ed the e ectiveness of recomputation on a pair of Convolutional Neural Network layers, when applying loop tiling, loop fusion, and recompute.

M3 - Conference contribution

BT - Eighth International Workshop on Polyhedral Compilation Techniques

T2 - 8th International Workshop on Polyhedral Compilation Techniques (IMPACT 2018), January 23, 2018, Manchester, UK

Y2 - 23 January 2018 through 23 January 2018

ER -

Optimization through recomputation in the polyhedral model

Samenvatting

Workshop

Duurzame ontwikkelingsdoelstellingen van de VN

Toegang tot document

Vingerafdruk

Citeer dit