TY - GEN
T1 - KEDM
T2 - 5th Practice and Experience in Advanced Research Computing Conference: Evolution Across All Dimensions, PEARC 2021
AU - Takahashi, Keichi
AU - Watanakeesuntorn, Wassapon
AU - Ichikawa, Kohei
AU - Park, Joseph
AU - Takano, Ryousei
AU - Haga, Jason
AU - Sugihara, George
AU - Pao, Gerald M.
N1 - Funding Information:
This work was partly supported by JSPS KAKENHI Grant Number JP20K19808 (KT) and an Innovation grant by the Kavli Institute for Brain and Mind (GMP). The authors would like to thank Dominic R. W. Burrows at the MRC Centre for Neurodevelopmental Disorders, Institute of Psychiatry, Psychology and Neuroscience, King’s College London, London, UK for providing the F1 dataset used in the performance evaluation.
Publisher Copyright:
© 2021 ACM.
PY - 2021/7/17
Y1 - 2021/7/17
N2 - Empirical Dynamic Modeling (EDM) is a state-of-the-art non-linear time-series analysis framework. Despite its wide applicability, EDM was not scalable to large datasets due to its expensive computational cost. To overcome this obstacle, researchers have attempted and succeeded in accelerating EDM from both algorithmic and implementational aspects. In previous work, we developed a massively parallel implementation of EDM targeting HPC systems (mpEDM). However, mpEDM maintains different backends for different architectures. This design becomes a burden in the increasingly diversifying HPC systems, when porting to new hardware. In this paper, we design and develop a performance-portable implementation of EDM based on the Kokkos performance portability framework (kEDM), which runs on both CPUs and GPUs while based on a single codebase. Furthermore, we optimize individual kernels specifically for EDM computation, and use real-world datasets to demonstrate up to 5.5 × speedup compared to mpEDM in convergent cross mapping computation.
AB - Empirical Dynamic Modeling (EDM) is a state-of-the-art non-linear time-series analysis framework. Despite its wide applicability, EDM was not scalable to large datasets due to its expensive computational cost. To overcome this obstacle, researchers have attempted and succeeded in accelerating EDM from both algorithmic and implementational aspects. In previous work, we developed a massively parallel implementation of EDM targeting HPC systems (mpEDM). However, mpEDM maintains different backends for different architectures. This design becomes a burden in the increasingly diversifying HPC systems, when porting to new hardware. In this paper, we design and develop a performance-portable implementation of EDM based on the Kokkos performance portability framework (kEDM), which runs on both CPUs and GPUs while based on a single codebase. Furthermore, we optimize individual kernels specifically for EDM computation, and use real-world datasets to demonstrate up to 5.5 × speedup compared to mpEDM in convergent cross mapping computation.
KW - Empirical Dynamic Modeling
KW - GPU
KW - High Performance Computing
KW - Kokkos
KW - Performance Portability
UR - http://www.scopus.com/inward/record.url?scp=85111029383&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85111029383&partnerID=8YFLogxK
U2 - 10.1145/3437359.3465571
DO - 10.1145/3437359.3465571
M3 - Conference contribution
AN - SCOPUS:85111029383
T3 - ACM International Conference Proceeding Series
BT - PEARC 2021 - Practice and Experience in Advanced Research Computing 2021
PB - Association for Computing Machinery
Y2 - 19 July 2021 through 22 July 2021
ER -