Leveraging Hardware Caches for Memoization

  • Authors:
    Guowei Zhang (MIT), Daniel Sanchez (MIT)
    Publication ID:
    P092835
    Publication Type:
    Paper
    Received Date:
    13-Nov-2017
    Last Edit Date:
    13-Nov-2017
    Research:
    2384.007 (University of Michigan)

Abstract

Memoization improves performance and saves energy by caching and reusing the outputs of repetitive computations. Prior work has proposed software and hardware memoization techniques, but both have significant drawbacks. Software memoization suffers from high runtime overheads, and is thus limited to long computations. Conventional hardware memoization techniques achieve low overheads and can memoize short functions, but they rely on large, special-purpose memoization caches that waste significant area and energy.

We propose MCACHE, a hardware technique that leverages data caches for memoization. MCACHE stores memoization tables in memory, and allows them to share cache capacity with normal program data. MCACHE introduces ISA and pipeline extensions to accelerate memoization operations, bridging the gap between software and conventional hardware techniques. Simulation results show that MCACHE improves performance by up to 21×, outperforms software memoization by up to 2.2×, and achieves similar or superior performance over conventional hardware techniques without any dedicated storage.

4819 Emperor Blvd, Suite 300 Durham, NC 27703 Voice: (919) 941-9400 Fax: (919) 941-9450

Important Information for the SRC website. This site uses cookies to store information on your computer. By continuing to use our site, you consent to our cookies. If you are not happy with the use of these cookies, please review our Cookie Policy to learn how they can be disabled. By disabling cookies, some features of the site will not work.