KAPLA: Pragmatic Representation and Fast Solving of Scalable NN Accelerator Dataflow

TL;DR

KAPLA introduces a pragmatic, tensor-centric dataflow representation and a fast solver for scalable neural network accelerators, enabling efficient exploration of optimized dataflow schemes with minimal energy overhead.

Contribution

This work presents a comprehensive dataflow representation and a novel fast solver, KAPLA, for scalable NN accelerators, significantly improving design exploration speed and solution quality.

Findings

KAPLA achieves within 2.2% and 7.7% energy overheads for training and inference dataflows.

KAPLA outperforms random and machine-learning-based approaches in optimization quality.

KAPLA provides orders of magnitude faster search speed compared to exhaustive methods.

Abstract

Dataflow scheduling decisions are of vital importance to neural network (NN) accelerators. Recent scalable NN accelerators support a rich set of advanced dataflow techniques. The problems of comprehensively representing and quickly finding optimized dataflow schemes thus become significantly more complicated and challenging. In this work, we first propose comprehensive and pragmatic dataflow representations for temporal and spatial scheduling on scalable multi-node NN architectures. An informal hierarchical taxonomy highlights the tight coupling across different levels of the dataflow space as the major difficulty for fast design exploration. A set of formal tensor-centric directives accurately express various inter-layer and intra-layer schemes, and allow for quickly determining their validity and efficiency. We then build a generic, optimized, and fast dataflow solver, KAPLA, which makes use of the pragmatic directives to explore the design space with effective validity check and efficiency estimation. KAPLA decouples the upper inter-layer level for fast pruning, and solves the lower intra-layer schemes with a novel bottom-up cost descending method. KAPLA achieves within only 2.2% and 7.7% energy overheads on the result dataflow for training and inference, respectively, compared to the exhaustively searched optimal schemes. It also outperforms random and machine-learning-based approaches, with more optimized results and orders of magnitude faster search speedup.