Exploiting historical data: Pruning autotuning spaces and
estimating the number of tuning steps

J 2020

Exploiting historical data: Pruning autotuning spaces and estimating the number of tuning steps

OĽHA, Jaroslav, Jana HOZZOVÁ, Jan FOUSEK and Jiří FILIPOVIČ

Basic information

Original name

Exploiting historical data: Pruning autotuning spaces and estimating the number of tuning steps

Authors

OĽHA, Jaroslav (703 Slovakia, guarantor, belonging to the institution), Jana HOZZOVÁ (703 Slovakia, belonging to the institution), Jan FOUSEK (203 Czech Republic, belonging to the institution) and Jiří FILIPOVIČ (203 Czech Republic, belonging to the institution)

Edition

CONCURRENCY AND COMPUTATION-PRACTICE & EXPERIENCE, HOBOKEN, WILEY, 2020, 1532-0626

Other information

Language

English

Type of outcome

Článek v odborném periodiku

Field of Study

10201 Computer sciences, information science, bioinformatics

Country of publisher

United States of America

Confidentiality degree

není předmětem státního či obchodního tajemství

References:

URL

Impact factor

Impact factor: 1.536

RIV identification code

RIV/00216224:14610/20:00116267

Organization unit

Institute of Computer Science

DOI

http://dx.doi.org/10.1002/cpe.5962

UT WoS

000557422400001

Keywords in English

autotuning; prediction of tuning cost; sensitivity analysis; tuning space pruning

Abstract

V originále

Autotuning, the practice of automatic tuning of applications to provide performance portability, has received increased attention in the research community, especially in high performance computing. Ensuring high performance on a variety of hardware usually means modifications to the code, often via different values of a selected set of parameters, such as tiling size, loop unrolling factor, or data layout. However, the search space of all possible combinations of these parameters can be large, which can result in cases where the benefits of autotuning are outweighed by its cost, especially with dynamic tuning. Therefore, estimating the tuning time in advance or shortening the tuning time is very important in dynamic tuning applications. We have found that certain properties of tuning spaces do not vary much when hardware is changed. In this article, we demonstrate that it is possible to use historical data to reliably predict the number of tuning steps that is necessary to find a well-performing configuration and to reduce the size of the tuning space. We evaluate our hypotheses on a number of HPC benchmarks written in CUDA and OpenCL, using several different generations of GPUs and CPUs.

Links

EF16_013/0001802, research and development project

Name: CERIT Scientific Cloud

LM2015085, research and development project

Name: CERIT Scientific Cloud (Acronym: CERIT-SC)

Investor: Ministry of Education, Youth and Sports of the CR, CERIT Scientific Cloud

MUNI/A/1050/2019, interní kód MU

Name: Rozsáhlé výpočetní systémy: modely, aplikace a verifikace IX (Acronym: SV-FI MAV IX)

Investor: Masaryk University, Category A

Citovat

OĽHA, Jaroslav, Jana HOZZOVÁ, Jan FOUSEK and Jiří FILIPOVIČ. Exploiting historical data: Pruning autotuning spaces and estimating the number of tuning steps. CONCURRENCY AND COMPUTATION-PRACTICE & EXPERIENCE. HOBOKEN: WILEY, 2020, vol. 32, No 21, p. 1-15. ISSN 1532-0626. Available from: https://dx.doi.org/10.1002/cpe.5962.

@article{1675386,
   author = {Oľha, Jaroslav and Hozzová, Jana and Fousek, Jan and Filipovič, Jiří},
   article_location = {HOBOKEN},
   article_number = {21},
   doi = {http://dx.doi.org/10.1002/cpe.5962},
   keywords = {autotuning; prediction of tuning cost; sensitivity analysis; tuning space pruning},
   language = {eng},
   issn = {1532-0626},
   journal = {CONCURRENCY AND COMPUTATION-PRACTICE & EXPERIENCE},
   title = {Exploiting historical data: Pruning autotuning spaces and estimating the number of tuning steps},
   url = {https://onlinelibrary.wiley.com/doi/10.1002/cpe.5962},
   volume = {32},
   year = {2020}
}

TY  - JOUR
ID  - 1675386
AU  - Oľha, Jaroslav - Hozzová, Jana - Fousek, Jan - Filipovič, Jiří
PY  - 2020
TI  - Exploiting historical data: Pruning autotuning spaces and estimating the number of tuning steps
JF  - CONCURRENCY AND COMPUTATION-PRACTICE & EXPERIENCE
VL  - 32
IS  - 21
SP  - 1-15
EP  - 1-15
PB  - WILEY
SN  - 15320626
KW  - autotuning
KW  - prediction of tuning cost
KW  - sensitivity analysis
KW  - tuning space pruning
UR  - https://onlinelibrary.wiley.com/doi/10.1002/cpe.5962
L2  - https://onlinelibrary.wiley.com/doi/10.1002/cpe.5962
N2  - Autotuning, the practice of automatic tuning of applications to provide performance portability, has received increased attention in the research community, especially in high performance computing. Ensuring high performance on a variety of hardware usually means modifications to the code, often via different values of a selected set of parameters, such as tiling size, loop unrolling factor, or data layout. However, the search space of all possible combinations of these parameters can be large, which can result in cases where the benefits of autotuning are outweighed by its cost, especially with dynamic tuning. Therefore, estimating the tuning time in advance or shortening the tuning time is very important in dynamic tuning applications. We have found that certain properties of tuning spaces do not vary much when hardware is changed. In this article, we demonstrate that it is possible to use historical data to reliably predict the number of tuning steps that is necessary to find a well-performing configuration and to reduce the size of the tuning space. We evaluate our hypotheses on a number of HPC benchmarks written in CUDA and OpenCL, using several different generations of GPUs and CPUs.
ER  -

OĽHA, Jaroslav, Jana HOZZOVÁ, Jan FOUSEK and Jiří FILIPOVIČ. Exploiting historical data: Pruning autotuning spaces and estimating the number of tuning steps. \textit{CONCURRENCY AND COMPUTATION-PRACTICE \&{} EXPERIENCE}. HOBOKEN: WILEY, 2020, vol.~32, No~21, p.~1-15. ISSN~1532-0626. Available from: https://dx.doi.org/10.1002/cpe.5962.

Detailed Information on Publication Record