Search arXiv⌕ Search
Home/Archive/arXiv

arXiv · 2312.12766

IOPS: An Unified SpMM Accelerator Based on Inner-Outer-Hybrid Product

Wenhao Sun·Wendi Sun·Song Chen·Yi Kang
View original source ↗Read source document ↗Export metadata ↓

Abstract

Sparse matrix multiplication (SpMM) is widely applied to numerous domains, such as graph processing, machine learning, and data analytics. However, inner product based SpMM induces redundant zero-element computing for mismatched nonzero operands, while outer product based approach lacks input reuse across Process Elements (PEs) and poor output locality for accumulating partial sum (psum) matrices. Besides, current works only focus on sparse-sparse matrix multiplication (SSMM) or sparse-dense matrix multiplication (SDMM), rarely performing efficiently for both. To address these problems, this paper proposes an unified SpMM accelerator, called IOPS, hybridizing inner with outer products. It reuses the input matrix among PEs with inner product dataflow, and removes zero-element calculations with outer product approach in each PE, which can efficiently process SSMM and SDMM. Moreover, an address mapping method is designed to accumulate the irregular sparse psum matrices, reducing the latency and DRAM access of psum accumulating. Furthermore, an adaptive partition strategy is proposed to tile the input matrices based on their sparsity ratios, effectively utilizing the storage of architecture and reducing DRAM access. Compared with the SSMM accelerator, SpArch, we achieve 1.7x~6.3x energy efficiency and 1.2x~4.4x resource efficiency, with 1.4x~2.1x DRAM access saving.

Explore related subjects

cs.AR

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Wenhao Sun, Wendi Sun, Song Chen, Yi Kang. 2023-12-20. IOPS: An Unified SpMM Accelerator Based on Inner-Outer-Hybrid Product. https://arxiv.org/abs/2312.12766

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related papers

arXiv2026

Exploiting Decompression Latency for Covert Channels in Inter-Line-Compressed LLCs

The recently proposed XOR cache is an inter-line-compressed last-level cache (LLC) that leverages the data-inclusion relationship between the private caches and the LLC, compressing two cache lines into one by XORing them. The architecture relies on the cache coherence protocol for data decompression. In this paper, we demonstrate that this mechanism - specifically the latency asymmetry between a cache hit on an uncompressed vs. compressed line - introduces microarchitectural vulnerabilities. Based on this observation, we propose a covert channel attack targeting the XOR cache. A colluding sender controls the receiver's access latency by triggering decompression through targeted write requests to partner cache lines. By exploiting the data-dependent compression behavior of the XOR cache, the sender and receiver establish the channel using pre-agreed data values. The channel achieves higher bandwidth than the Prime+Probe baseline for two reasons: first, each bit is encoded in the compression state of an individual line rather than the occupancy of a cache set, so a single set carries multiple bits; second, each bit is resolved by manipulating coherence-protocol state rather than forcing shared-cache evictions, so it costs fewer LLC accesses and demand misses than Prime+Probe. Full-system simulations show a bandwidth of 2.9 Mbps at an observed 0.98% bit-error rate (BER) over 50,000 transmitted bits, 13.1 times the bandwidth of Prime+Probe under the same sub-1%-BER selection rule.

cs.AR↗
arXiv2026

Mamba-Family State-Space Model Kernels on a Programmable CGLA

Edge and embedded inference is constrained by power and data movement. Mamba-family state-space models replace attention with sequence-linear recurrence, but their inference path combines dense projections, short-reduction SSD kernels, and recurrent-state updates. This paper maps these kernel groups onto IMAX, a programmable CPU-Grounded Linear Array (CGLA), and measures them from kernel execution to token-level integration. Projection kernels match the long-reduction IMAX pipeline, whereas SSD Step-1 is limited by short reductions and kernel-boundary overheads. Mamba-130M token-level integration identifies projection GEMV as the decode bottleneck. These results show that programmable CGLAs fit long-reduction projection kernels, while SSD and decode-time projection support require boundary reduction and persistent-weight execution.

cs.AR↗
arXiv2026

Energy-Oriented CGLA Mapping of a Memory-Polynomial Digital Predistortion Kernel

Memory-polynomial digital predistortion (DPD) evaluates a small fixed coefficient set over a sliding input history, so its reduction step is a complex-MAC workload with local reuse. We map this DPD reduction kernel onto In-Memory Accelerator eXtension (IMAX), a programmable CPU-Grounded Linear Array (CGLA) composed of a one-dimensional processing-element/local-memory pipeline. For a (P,M)=(5,5) odd-order memory-polynomial instance, the mapping keeps the 120 B coefficient set in local memory, advances the five-tap history over 1024-sample tiles, and realizes the 15 order-delay terms as a 33-stage streaming complex-MAC reduction. The evaluation measures kernel latency and modeled energy. All measured paths use the same single-precision complex workload of 32 sequences, each with 2048 complex samples, across an IMAX FPGA prototype, a CUDA implementation on an RTX 4090 system, and an ARM-NEON implementation on Jetson AGX Orin. With this 1024-sample tile configuration, the IMAX FPGA prototype reports 20.201 ms end-to-end latency and 1.948 ms kernel-only latency. Using the previously reported 28 nm IMAX frequency and power model, the projected IMAX configuration gives 3.14 ms end-to-end latency and 0.34 ms kernel-only latency. The RTX 4090 baseline has the lowest end-to-end latency at 0.484 ms. Under model-based platform power accounting and the stated power assumptions, the projected IMAX configuration gives 169.1 times smaller modeled end-to-end energy per batch than the RTX 4090 baseline. This value uses platform power assumptions rather than workload-dependent runtime power or a direct silicon power measurement. A controlled synthetic PA-model validation checks that the same 15-term form improves test-set NMSE by 26.1 dB and ACLR by 26.0 dB. These results characterize the mapped memory-polynomial DPD reduction on IMAX for the evaluated tile configuration and power model.

cs.AR↗