A Pattern-Based SpGEMM Library for Multi-Core and Many-Core Architectures

Xie, Zhen; Tan, Guangming; Liu, Weifeng; Sun, Ninghui

doi:10.1109/tpds.2021.3090328

Cited by 18 publications

(7 citation statements)

References 48 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Previous research (Xie et al, 2021) trains a model using matrix features and thumbnails to predict the SpGEMM algorithm with the best performance on input matrices which they called MatNet, but it has several limitations and drawbacks. (a) The model utilized matrix thumbnails as one of its features for training and prediction.…”

Section: Motivationmentioning

confidence: 99%

See 1 more Smart Citation

Predicting optimal sparse general matrix-matrix multiplication algorithm on GPUs

Wei,

Wang,

Chang

et al. 2024

The International Journal of High Performance Computing Applica

View full text Add to dashboard Cite

Sparse General Matrix-Matrix Multiplication (SpGEMM) has played an important role in a number of applications. So far, many efficient algorithms have been proposed to improve the performance of SpGEMM on GPUs. However, the performance of each algorithm for matrices of different structures varies a lot. There is no algorithm that can achieve the optimal performance of SpGEMM computation on all matrices. In this article, we design a machine learning based approach for predicting the optimal SpGEMM algorithm on input matrices. By extracting features from input matrices, we utilize LightGBM and XGBoost to train different lightweight models. The models are capable of predicting the best performing algorithm with low inference overhead and high accuracy for the given input matrices. We also investigate the impact of tree depth on model accuracy and inference overhead. Our evaluation shows that an increase in tree depth leads to a corresponding increase in prediction accuracy, reaching a maximum of approximately 85%, while resulting in increased inference overhead of approximately 10 µs. Compared with the state-of-the-art algorithms on three GPU platforms, our method achieves better overall performance.

show abstract

Section: Motivationmentioning

confidence: 99%

“…For results accumulation, there are sparse accumulators and dense accumulators (Patwary et al, 2015). Although there have been many algorithms to solve the above issues, the experiment (Xie et al, 2021) has proved that there is no algorithm that can achieve the best performance of SpGEMM over all matrices.…”

Section: Introductionmentioning

confidence: 99%

Predicting optimal sparse general matrix-matrix multiplication algorithm on GPUs

Wei,

Wang,

Chang

et al. 2024

The International Journal of High Performance Computing Applica

View full text Add to dashboard Cite

show abstract

“…They also have proposed a threshold to switch between a dense and a sparse representation of each tile and designed a tiled algorithm for SpMV [37]. Finally, Xie et al presented a novel AI-based approach for SpGEMM [44]. Although the algorithms on which they rely all use the Gustavson method, they investigate different matrix storage formats and algorithms and train a custom deep-learning network which chooses the best solution for given input matrices and target architecture.…”

Section: Related Workmentioning

confidence: 99%

Optimization of SpGEMM with Risc-V vector instructions

Fèvre¹,

Casas²

2023

Preprint

View full text Add to dashboard Cite

The Sparse GEneral Matrix-Matrix multiplication (SpGEMM) 𝐶 = 𝐴 × 𝐵 is a fundamental routine extensively used in domains like machine learning or graph analytics. Despite its relevance, the efficient execution of SpGEMM on vector architectures is a relatively unexplored topic. The most recent algorithm to run SpGEMM on these architectures is based on the SParse Accumulator (SPA) approach, and it is relatively efficient for sparse matrices featuring several tens of nonzero coefficients per column as it computes 𝐶 columns one by one. However, when dealing with matrices containing just a few non-zero coefficients per column, the state-of-the-art algorithm is not able to fully exploit long vector architectures when computing the SpGEMM kernel.To overcome this issue we propose the SPA paRallel with Sorting (SPARS) algorithm, which computes in parallel several 𝐶 columns among other optimizations, and the HASH algorithm, which uses dynamically sized hash tables to store intermediate output values. To combine the efficiency of SPA for relatively dense matrix blocks with the high performance that SPARS and HASH deliver for very sparse matrix blocks we propose H-SPA(𝑡) and H-HASH(𝑡), which dynamically switch between different algorithms. H-SPA(𝑡) and H-HASH(𝑡) obtain 1.24× and 1.57× average speed-ups with respect to SPA respectively, over a set of 40 sparse matrices obtained from the SuiteSparse Matrix Collection [19]. For the 22 most sparse matrices, H-SPA(𝑡) and H-HASH(𝑡) deliver 1.42× and 1.99× average speed-ups respectively.

show abstract

“…Compared to sparse matrix-vector multiplication (SpMV), the more challenging SpGEMM has been covered to a lesser extent in the literature so far -as pointed out by Winter et al [29]. Recently, however, the optimization of static SpGEMM algorithms for specific parallel architectures is on the rise, e. g., for multithreaded CPUs [19,13] GPUs [29], CPU/GPU combinations [11,30], and other accelerators [19]. A possible reason for this spike could be the use of SpGEMM in deep learning with sparse DNNs, as described in Ref.…”

Section: Related Workmentioning

confidence: 99%

Fast Dynamic Updates and Dynamic SpGEMM on MPI-Distributed Graphs

Grinten¹,

Custers²,

Thanh³

et al. 2022

Preprint

View full text Add to dashboard Cite

Sparse matrix multiplication (SpGEMM) is a fundamental kernel used in many diverse application areas, both numerical and discrete. Recently, SpGEMM has received growing attention regarding implementations for specific (parallel) architectures. Yet, this concerns only the static problem, where both input matrices do not change. In many applications, however, matrices (or their corresponding graphs) change over time. Although recomputing from scratch is very expensive, we are not aware of any dynamic SpGEMM algorithms in the literature. In this paper, we thus propose a batch-dynamic algorithm for MPI-based parallel computing. Building on top of a distributed graph/matrix data structure that allows for fast updates, our dynamic SpGEMM reduces the communication volume significantly. It does so by exploiting that updates change far fewer matrix entries than there are non-zeros in the input operands. Our experiments with popular benchmark graphs show that our approach pays off. For batches of insertions or removals of matrix entries, our dynamic SpGEMM is substantially faster than the static state-of-the-art algorithm in CombBLAS 2.0 and in CTF.

show abstract

A Pattern-Based SpGEMM Library for Multi-Core and Many-Core Architectures

Cited by 18 publications

References 48 publications

Predicting optimal sparse general matrix-matrix multiplication algorithm on GPUs

Predicting optimal sparse general matrix-matrix multiplication algorithm on GPUs

Optimization of SpGEMM with Risc-V vector instructions

Fast Dynamic Updates and Dynamic SpGEMM on MPI-Distributed Graphs

Contact Info

Product

Resources

About