SparseAdapt: Runtime Control for Sparse Linear Algebra on a Reconfigurable Accelerator

Pal, Soham; Amarnath, Aporva; Feng, Siying; O’Boyle, Michael; Dreslinski, Ronald G.; Dubach, Christophe

doi:10.1145/3466752.3480134

Cited by 13 publications

(2 citation statements)

References 66 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…MINT [28] is a format converter widget that supports multiple sparse formats. Prior works Garg et al [7], coSPARSE [6] and SparseAdapt [24] propose frameworks for efficient sparse execution on CGRAs. However, to the best of our knowledge, this is the first work that proposes an accelerator for Sparse DNNs which exploits all the three dataflows.…”

Section: Related Workmentioning

confidence: 99%

Flexagon: A Multi-Dataflow Sparse-Sparse Matrix Multiplication Accelerator for Efficient DNN Processing

Muñoz-Martínez¹,

Garg²,

Abellán³

et al. 2023

Preprint

View full text Add to dashboard Cite

Sparsity is a growing trend in modern DNN models. Existing Sparse-Sparse Matrix Multiplication (SpMSpM) accelerators are tailored to a particular SpMSpM dataflow (i.e., Inner Product, Outer Product or Gustavson's), that determines their overall efficiency. We demonstrate that this static decision inherently results in a suboptimal dynamic solution. This is because different SpMSpM kernels show varying features (i.e., dimensions, sparsity pattern, sparsity degree), which makes each dataflow better suited to different data sets.In this work we present Flexagon, the first SpMSpM reconfigurable accelerator that is capable of performing SpM-SpM computation by using the particular dataflow that best matches each case. Flexagon accelerator is based on a novel Merger-Reduction Network (MRN) that unifies the concept of reducing and merging in the same substrate, increasing efficiency. Additionally, Flexagon also includes a 3-tier memory hierarchy, specifically tailored to the different access characteristics of the input and output compressed matrices. Using detailed cycle-level simulation of contemporary DNN models from a variety of application domains, we show that Flexagon achieves average performance benefits of 4.59×, 1.71×, and 1.35× with respect to the state-of-the-art SIGMA-like, Sparchlike and GAMMA-like accelerators (265% , 67% and 18%, respectively, in terms of average performance/area efficiency).

show abstract

Section: Related Workmentioning

confidence: 99%

Flexagon: A Multi-Dataflow Sparse-Sparse Matrix Multiplication Accelerator for Efficient DNN Processing

Muñoz-Martínez¹,

Garg²,

Abellán³

et al. 2023

Preprint

View full text Add to dashboard Cite

show abstract

“…Sparse accelerators: Sparsity has been a key optimization target for HPC and AI workloads. Numerous accelerators have been proposed for SpMM and SpGEMM acceleration [2], [5], [11], [15], [17], [18], [19], [23], [29], [32], [34], [35], [39], [40], [43], [50], [52], [53], [56], [60]. Though there are a lot of unique sparse accelerators, we believe that they can all be grouped into classes based on their computation dataflow.…”

Section: Related Workmentioning

confidence: 99%

Enabling Flexibility for Sparse Tensor Acceleration via Heterogeneity

Qin¹,

Garg²,

Bambhaniya³

et al. 2022

Preprint

View full text Add to dashboard Cite

Recently, numerous sparse hardware accelerators for Deep Neural Networks (DNNs), Graph Neural Networks (GNNs), and scientific computing applications have been proposed. A common characteristic among all of these accelerators is that they target tensor algebra (typically matrix multiplications); yet dozens of new accelerators are proposed for every new application. The motivation is that the size and sparsity of the workloads heavily influence which architecture is best for memory and computation efficiency. To satisfy the growing demand of efficient computations across a spectrum of workloads on large data-centers, we propose deploying a flexible 'heterogeneous' accelerator, which contains many 'sub-accelerators' (smaller specialized accelerators) working together. To this end, we propose: (1) HARD TACO, a quick and productive C++ to RTL design flow to generate many types of sub-accelerators for sparse and dense computations for fair design-space exploration, (2) AESPA, a heterogeneous sparse accelerator design template constructed with the sub-accelerators generated from HARD TACO, and (3) a suite of scheduling strategies to map tensor kernels onto heterogeneous sparse accelerators with high efficiency and utilization. AESPA with optimized scheduling achieves 1.96× higher performance, and 7.9× better energy delay product (EDP) than a Homogeneous EIE-like accelerator with our diverse workload suite.

show abstract

CoLeCTs: Cooperative Learning Classifier Tables for Resource Management in MPSoCs

Zyla,

Maurer,

Wild

et al. 2023

Lecture Notes in Computer Science

View full text Add to dashboard Cite

SparseAdapt: Runtime Control for Sparse Linear Algebra on a Reconfigurable Accelerator

Cited by 13 publications

References 66 publications

Flexagon: A Multi-Dataflow Sparse-Sparse Matrix Multiplication Accelerator for Efficient DNN Processing

Flexagon: A Multi-Dataflow Sparse-Sparse Matrix Multiplication Accelerator for Efficient DNN Processing

Enabling Flexibility for Sparse Tensor Acceleration via Heterogeneity

CoLeCTs: Cooperative Learning Classifier Tables for Resource Management in MPSoCs

Contact Info

Product

Resources

About