Decoupled Compressed Cache: Exploiting Spatial Locality for Energy Optimization

Abstract

Listen

Translate article

Last-level caches (LLCs) play a crucial role in reducing multicore system energy by filtering out expensive accesses to main memory. Cache compression can increase effective LLC capacity and reduce misses. However, previous designs limit compression benefits caused by internal fragmentation, limited tags, and energy-expensive recompaction when a block's size changes. The authors propose decoupled compressed cache (DCC), which uses decoupled superblocks and noncontiguous sub-block allocation to decrease tag overhead without increasing internal fragmentation and to eliminate the need for energy-expensive recompaction. They also demonstrate a practical design based on a recent commercial LLC design.