How to Be Fast and Not Furious: Looking Under the Hood of CPU Cache Prefetching
Title
How to Be Fast and Not Furious: Looking Under the Hood of CPU Cache Prefetching
Authors
Roland Kühn, Jan Mühlig, and Jens Teubner
Accepted at
Data Management on Modern Hardware (DaMoN), Santiago, Chile, June 2024
Abstract
Software-based prefetching is an effective method for tolerating one of the most formidable barriers encountered by data processing systems: memory latency. Although the idea appears simple—just inform the CPU about upcoming data accesses—the intricacies of its implementation remain insufficiently understood. Existing works demonstrate how algorithms need to be prepared for prefetching, yet they often overlook the limitations and hardware implications of bringing data into the cache hierarchy. In this paper, we thoroughly examine software-based prefetching by delving into its implementation and identifying pitfalls across various platforms. Furthermore, we provide actionable insights and recommendations for developers seeking to boost their applications through this technique.
Download
Publication Log
May 2024
- camera ready (PDF)
March 2024
Submission to DaMoN 2024 (accept)
- submission (PDF)
- reviews (accept, strong accept, accept)