Exploiting dynamic sparse matrices for performance portable linear algebra operations

Stylianou, Christodoulos; Weiland, Michele

doi:10.1109/p3hpc56579.2022.00010

Cited by 4 publications

(3 citation statements)

References 18 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Morpheus [11] is a C++ library that supports the runtime switching of sparse matrix storage formats through Dynam-icMatrix, a single dynamic "abstract" format. Morpheus provides a transparent mechanism that can efficiently switch to the different formats supported by the DynamicMatrix and it currently supports the six formats mentioned in Section II-B.…”

Section: Morpheusmentioning

confidence: 99%

See 1 more Smart Citation

Optimizing Sparse Linear Algebra Through Automatic Format Selection and Machine Learning

Stylianou¹,

Weiland²

2023

Preprint

View full text Add to dashboard Cite

Sparse matrices are an integral part of scientific simulations. As hardware evolves new sparse matrix storage formats are proposed aiming to exploit optimizations specific to the new hardware. In the era of heterogeneous computing, users often are required to use multiple formats for their applications to remain optimal across the different available hardware, resulting in larger development times and maintenance overhead. A potential solution to this problem is the use of a lightweight auto-tuner driven by Machine Learning (ML) that would select for the user an optimal format from a pool of available formats that will match the characteristics of the sparsity pattern, target hardware and operation to execute.In this paper, we introduce Morpheus-Oracle, a library that provides a lightweight ML auto-tuner capable of accurately predicting the optimal format across multiple backends, targeting the major HPC architectures aiming to eliminate any format selection input by the end-user. From more than 2000 reallife matrices, we achieve an average classification accuracy and balanced accuracy of 92.63% and 80.22% respectively across the available systems. The adoption of the auto-tuner results in average speedup of 1.1× on CPUs and 1.5× to 8× on NVIDIA and AMD GPUs, with maximum speedups reaching up to 7× and 1000× respectively.

show abstract

Section: Morpheusmentioning

confidence: 99%

“…Since the matrix is generally unknown during compile-time, this exploitation can only be done at runtime. Previous efforts [11]- [14] provide abstractions and mechanisms that effectively allow for runtime switching to the different formats that are supported.…”

Section: Introductionmentioning

confidence: 99%

Optimizing Sparse Linear Algebra Through Automatic Format Selection and Machine Learning

Stylianou¹,

Weiland²

2023

Preprint

View full text Add to dashboard Cite

show abstract

“…S PARSE matrix-vector product (SpMV) is a central part of many numerical algorithms and its performance can have a very big impact on the performance of scientific and engineering applications [1], [2]. There are a lot of various sparse matrix storage formats and sophisticated techniques for developing efficient implementations of SpMV that utilize the underlying hardware of modern multicore CPUs and GPUs [3], [4], [5], [6], [7], [8], [9]. Unfortunately, these methods are rather complicated and usually depend on particular computer architecture, thus developing efficient and portable sparse matrix source code is still a challenge.…”

Section: Introductionmentioning

confidence: 99%

Performance of Portable Sparse Matrix-Vector Product Implemented Using OpenACC

Stec,

Stpiczynski

2023

Annals of Computer Science and Information Systems

View full text Add to dashboard Cite

The aim of this paper is to study the performance of OpenACC implementations of sparse matrix-vector product for several storage formats: CSR, ELL, JAD, pJAD, and BSR, achieved on Intel CPU and NVIDIA GPU platforms to compare them with the performance of SpMV implementations using the BSR storage format provided by Intel MKL and NVIDIA cuSPARSE libraries. Numerical experiments show that vendorprovided BSR is the best format for CPUs but in the case of GPUs, the pJAD storage format allows to achieve better performance.

show abstract