W-Cycle SVD: A Multilevel Algorithm for Batched SVD on GPUs

As a basic matrix factorization operation, Singular Value Decomposition (SVD) is widely used in diverse domains. In real-world applications, the computational bottleneck of matrix factorization is on small matrices, and many GPU-accelerated batched SVD algorithms have been developed recently for hig...

Celý popis

Uloženo v:
Podrobná bibliografie
Vydáno v:International Conference for High Performance Computing, Networking, Storage and Analysis (Online) s. 1 - 16
Hlavní autoři: Xiao, Junmin, Pang, Yunfei, Xue, Qing, Shui, Chaoyang, Meng, Ke, Ma, Hui, Li, Mingyi, Zhang, Xiaoyang, Tan, Guangming
Médium: Konferenční příspěvek
Jazyk:angličtina
Vydáno: IEEE 01.11.2022
Témata:
Convergence
Data assimilation
GPU
High performance computing
Jacobian matrices
Kernel
Matrix decomposition
Singular value decomposition
ISSN:2167-4337
On-line přístup:Získat plný text
Tagy: Přidat tag
Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!
Popis
Shrnutí:As a basic matrix factorization operation, Singular Value Decomposition (SVD) is widely used in diverse domains. In real-world applications, the computational bottleneck of matrix factorization is on small matrices, and many GPU-accelerated batched SVD algorithms have been developed recently for higher performance. However, these algorithms failed to achieve both high data locality and convergence speed, because they are size-sensitive. In this work, we propose a novel W-cycle SVD to accelerate the batched one-sided Jacobi SVD on GPUs. The W-cycle SVD, which is size-oblivious, successfully exploits the data reuse and ensures the optimal convergence speed for batched SVD. Further, we present the efficient batched kernel design, and propose a tailoring strategy based on auto-tuning to improve the batched matrix multiplication in SVDs. The evaluation demonstrates that the proposed algorithm achieves 2.6∼10.2× speedup over the state-of-the-art cuSOLVER. In a real-world data assimilation application, our algorithm achieves 2.73∼3.09× speedup compared with MAGMA.
ISSN:2167-4337
DOI:10.1109/SC41404.2022.00087