Yet Another Compressed Cache: a Low Cost Yet Effective Compressed Cache

Somayeh Sardashti (1) , André Seznec (2) , David A. Wood (1)

1 University of Wisconsin-Madison
2 ALF - Amdahl's Law is Forever

Cache memories play a critical role in bridging the latency, bandwidth, and energy gaps between cores and off-chip memory. However, caches frequently consume a significant fraction of a multicore chip’s area, and thus account for a significant fraction of its cost. Compression has the potential to improve the effective capacity of a cache, providing the performance and energy benefits of a larger cache while using less area. The design of a compressed cache must address two important issues: i) a low-latency, low-overhead compression algorithm that can represent a fixed-size cache block using fewer bits and ii) a cache organization that can efficiently store the resulting variable-size compressed blocks. This paper focuses on the latter issue. In this paper, we propose YACC (Yet Another Compressed Cache), a new compressed cache design that uses super-blocks to reduce tag overheads and variable-size blocks to reduce internal fragmentation, but eliminates two major sources of complexity in previous work—decoupled tag-data mapping and address skewing. YACC’s cache layout is similar to conventional caches, eliminating the back-pointers used to maintain a decoupled tag-data mapping and the extra decoders used to implement skewed associativity. An additional advantage of YACC is that it enables modern replacement mechanisms, such as RRIP. For our benchmark set, YACC performs comparably to the recently-proposed Skewed Compressed Cache (SCC) ‎[Sardashti et al. 2014], but with a simpler, more area efficient design without the complexity and overheads of skewing. Compared to a conventional uncompressed 8MB LLC, YACC improves performance by on average 8% and up to 26%, and reduces total energy by on average 6% and up to 20%. An 8MB YACC achieves approximately the same performance and energy improvements as a 16MB conventional cache at a much smaller silicon footprint, with 1.6% higher area than an 8MB conventional cache

Domaines: Architectures Matérielles [cs.AR]

Somayeh Sardashti, André Seznec, David A. Wood. Yet Another Compressed Cache: a Low Cost Yet Effective Compressed Cache . [Research Report] RR-8853, Inria. 2016, pp.23. ⟨hal-01270792⟩