PASTA: A Parallel Sparse Tensor Algorithm Benchmark Suite

Abstract: Tensor methods have gained increasingly attention from various applications, including machine learning, quantum chemistry, healthcare analytics, social network analysis, data mining, and signal processing, to name a few. Sparse tensors and their algorithms become critical to further improve the performance of these methods and enhance the interpretability of their output. This work presents a sparse tensor algorithm benchmark suite (PASTA) for single- and multi-core CPUs. To the best of our knowledge, this is the first benchmark suite for sparse tensor world. PASTA targets on: 1) helping application users to evaluate different computer systems using its representative computational workloads; 2) providing insights to better utilize existed computer architecture and systems and inspiration for the future design. This benchmark suite is publicly released https://gitlab.com/tensorworld/pasta.

• The paper introduces PASTA, a benchmark suite for evaluating sparse tensor algorithms on single- and multi-core CPUs.
• It implements key kernels such as SpTTM, SpMM, SDDMM, and SpTV to cover essential tensor operations in various applications.
• The benchmark suite aids in evaluating performance and optimizing computer architectures for domains like machine learning and data analytics.

PASTA is a benchmark suite for sparse tensor algorithms on single- and multi-core CPUs (1902.03317). It is designed to help application users evaluate different computer systems using its representative computational workloads and to offer insights for improving the utilization of computer architectures and systems.

The benchmark suite includes several key sparse tensor algorithms. These kernels were chosen because of their prevalence in applications such as machine learning, data mining, healthcare analytics, social network analysis, signal processing and quantum chemistry.

Kernels:

• SpTTM (Sparse Tensor Times Matrix): This operation multiplies a sparse tensor by a dense matrix. It's a fundamental operation in many tensor decompositions.
• SpMM (Sparse Matrix-Matrix Multiplication): This kernel performs the multiplication of two sparse matrices. It is a key operation for many tensor algorithms.
• SDDMM (Sampled Dense-Dense Matrix Multiplication): Computes the element-wise product of two dense matrices, and the result is multiplied by a sparse matrix.
• SpTV (Sparse Tensor Times Vector): Calculates the product of a sparse tensor and a dense vector, crucial in operations involving tensor contractions with vectors.

PASTA is designed to offer insights for improving the utilization of existing computer architecture and systems and is publicly released at https://gitlab.com/tensorworld/pasta.

In summary, PASTA provides a comprehensive set of benchmarks for evaluating the performance of sparse tensor algorithms, aiding in system evaluation and optimization for tensor-based applications.