METAL: Caching Multi-level Indexes in Domain-Specific Architectures

No ratings

Presented at ASPLOS 2024 by

Anagha Molakalmur Anil Kumar

Aditya Prasanna

Jonathan Balkind

Arrvindh Shriraman

Abstract: State-of-the-art domain specific architectures (DSAs) work with sparse data, and need hardware support for index data-structures [31, 43, 57, 61]. Indexes are more space-efficient for sparse-data, and reduce DRAM bandwidth, if data reuse can be managed. However, indexes exhibit dynamic accesses, chase pointers, and need to walk-and-search. This inflates the working set and thrashes the cache. We observe that the cache organization itself is responsible for this behavior. We develop METAL, a portable caching idiom that enables DSAs to employ index data-structures. METAL decouples reuse of the index metadata from data reuse, and optimizes it independently. We propose two ideas: i) IX-Cache: A cache that leverages range tags to short-circuits index walks, and reduces the working set. IX-cache helps capture the trade-off between wider index nodes that maximize reach vs those that are closer to leaf and minimize walk latency. ii) Reuse Patterns: An interface to explicitly manage the cache. Patterns orchestrate cache insertions and bypass as we dynamically traverse different index regions. METAL improves performance vs. streaming DSAs by 7.8×, address-caches by 4.1×, and state-of-the-art DSA-cache [50] by 2.4×. We reduce DRAM energy by 1.6× vs. prior state-of-the-art.