Loading...
Please wait, while we are loading the content...
Similar Documents
Exploiting Historical Context in Irregular Data Prefetching
| Content Provider | Semantic Scholar |
|---|---|
| Author | Putman, Houston P. |
| Copyright Year | 2016 |
| Abstract | In this paper we propose a new irregular data prefetcher, Temporal Sparse Distributed Memory, which builds on the address correlation and PC localization. Address correlation, learning correlation between pairs of addresses, is the inspiration behind the current leader in irregular data prefetching. TSDM distinguishes itself from current address correlation techniques by including the correlation information from a stream of addresses in the prediction step. This combination of correlation information replicates the historical context of the stream. We show that the use of historical context gives TSDM a significant advantage over conventional address correlation on micro-benchmarks that exhibit high levels of aliasing. TSDM fares slightly better on irregular benchmarks than the current irregular prefetching leader, and it performs similarly when run on a restricted hardware budget. |
| File Format | PDF HTM / HTML |
| Alternate Webpage(s) | https://apps.cs.utexas.edu/apps/sites/default/files/tech_reports/TR-2235.pdf |
| Language | English |
| Access Restriction | Open |
| Content Type | Text |
| Resource Type | Article |
