Search

Artificial intelligence (AI) application domains consist of a mix of tensor operations with high and low arithmetic intensities (aka reuse). Hierarchical (i.e. compute along multiple levels of memory hierarchy) and heterogeneous (multiple different sub-accelerators) accelerators are emerging as a popular way to process mixed reuse workloads, and workloads which consist of tensor operators with diverse shapes. However, the space of hierarchical and/or heterogeneous processors (HHP's) is relatively under-explored. Prior works have proposed custom architectures to take advantage of heterogeneity to have multiple sub-accelerators that are efficient for different operator shapes. In this work, we propose HARP, a taxonomy to classify various hierarchical and heterogeneous accelerators and use the it to study the impact of heterogeneity at various levels in the architecture. HARP taxonomy captures various ways in which HHP's can be conceived, ranging from B100 cores with an "intra-node heterogeneity" between SM and tensor core to NeuPIM with cross-depth heterogeneity which occurs at different levels of memory hierarchy. We use Timeloop mapper to find the best mapping for sub-accelerators and also modify the Timeloop cost model to extend it to model hierarchical and heterogeneous accelerators.

Because of the recent trends in Deep Neural Networks (DNN) models being memory-bound, inter-operator pipelining for DNN accelerators is emerging as a promising optimization. Inter-operator pipelining reduces costly on-chip global memory and off-chip memory accesses by forwarding the output of a layer as the input of the next layer within the compute array, which is proven to be an effective optimization by previous works. However, the design space of inter-operator pipelining is huge, and the space is not yet fully explored. In particular, identifying the right depth and granularity of pipelining (or no pipelining at all) is significantly dependent on the layer shapes and data volumes of weights and activations, and these are different even within a domain. Moreover, works divide the substrate into large chunks and map one layer onto each chunk, which requires communicating halfway through or through the global buffer. However, for fine-grained inter-operation pipelining, placing the corresponding consumer of the next layer tile close to the producer tile of the current layer is a better way to exploit fine-grained spatial reuse. In order to support variable number of layers (ie the right depth) and support multiple spatial organizations of layers (in accordance with the pipelining granularity) on the substrate, we propose PipeOrgan, a new class of spatial data organization strategy for energy efficient and congestion-free communication between the PEs for various pipeline depth and granularity. PipeOrgan takes advantage of flexible spatial organization and can allocate layers to PEs based on the granularity of pipelining. We also propose changes to the conventional mesh topology to improve the performance of coarse-grained allocation. PipeOrgan achieves 1.95x performance improvement over the state-of-the-art pipelined dataflow on XR-bench workloads.

HPC applications are critical in various scientific domains ranging from molecular dynamics to chemistry to fluid dynamics. Conjugate Gradient (CG) is a popular application kernel used in iterative linear HPC solvers and has applications in numerous scientific domains. However, the HPCG benchmark shows that the peformance achieved by Top500 HPC systems on CG is a small fraction of the performance achieved on Linpack. While CG applications have significant portions of computations that are dense and sparse matrix multiplications, skewed SpMMs/GEMMs in the HPC solvers have poor arithmetic intensities which makes their execution highly memory bound unlike GEMMs in DNNs which have high arithmetic intensity. The problem of low intensity individual skewed GEMMs also exists in various emerging workloads from other domains like Graph Neural Networks, Transformers etc. In this work we identify various reuse opportunities between the tensors in these solver applications to extract reuse in the entire Directed Acyclic Graph of the tensor operations rather than individual tensor operations. These opportunities essentially depend on the dimensions of the tensors and the structure of the tensor dependency graph. We propose a systematic methodology to determine various kinds of reuse opportunities in the graph of operations and determine the loop order and tiling in the interdependent operations. As a result, we propose a novel mapping strategy GOGETA that improves reuse of HPC applications on spatial accelerators. We also propose a data organization strategy in the buffer. Our mapping achieves geomean 6.7x reduction in memory accesses.

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 Gustavsons), 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 SpMSpM 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 4.59x, 1.71x, and 1.35x with respect to the state-of-the-art SIGMA-like, Sparch-like and GAMMA-like accelerators (265% , 67% and 18%, respectively, in terms of average performance/area efficiency)., Comment: To appear on ASPLOS 2023

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 1.96X higher performance, and 7.9X better energy-delay product (EDP) than a Homogeneous EIE-like accelerator with our diverse workload suite.

Graph Neural Networks (GNNs) have garnered a lot of recent interest because of their success in learning representations from graph-structured data across several critical applications in cloud and HPC. Owing to their unique compute and memory characteristics that come from an interplay between dense and sparse phases of computations, the emergence of reconfigurable dataflow (aka spatial) accelerators offers promise for acceleration by mapping optimized dataflows (i.e., computation order and parallelism) for both phases. The goal of this work is to characterize and understand the design-space of dataflow choices for running GNNs on spatial accelerators in order for mappers or design-space exploration tools to optimize the dataflow based on the workload. Specifically, we propose a taxonomy to describe all possible choices for mapping the dense and sparse phases of GNN inference, spatially and temporally over a spatial accelerator, capturing both the intra-phase dataflow and the inter-phase (pipelined) dataflow. Using this taxonomy, we do deep-dives into the cost and benefits of several dataflows and perform case studies on implications of hardware parameters for dataflows and value of flexibility to support pipelined execution., Comment: Accepted for publication at the 36th IEEE International Parallel & Distributed Processing Symposium (IPDPS 2022)