Reproduce “LizTail: Unifying Shape Adaptation and Spatial Parallelism with Segment-Reconfigurable Systolic Arrays” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Gang Liu, Chen Yang, Junyan Hu, Jianguo Chen, Lixin Duan

LizTail adapts a systolic array to matrix shapes and available parallelism by trimming segments, joining them into longer paths, or assigning them to concurrent subarrays. In architecture-level simulations against a fixed TPU-like array, it reports geometric-mean speedups of 5.02× across three CNN inference workloads and 25.67× across three LLM autoregressive decoding workloads. These results are modeled speedups, not deployed-chip measurements.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.