| Creator |
|
|
|
|
|
|
|
|
|
| Language |
|
| Publisher |
|
|
|
| Date |
|
| Source Title |
|
| Source Title |
|
| First Page |
|
| Last Page |
|
| Publication Type |
|
| Access Rights |
|
| Rights |
|
| Related DOI |
|
|
|
|
|
|
|
| Related URI |
|
|
|
|
|
|
|
| Relation |
|
|
|
|
|
|
|
| Abstract |
マイクロプロセッサと主記憶との動作周波数差は，年々拡大する一方である．両者の周波数差は，マ イクロプロセッサの性能阻害要因であり，一般的にメモリ・ウォール問題と呼ばれる．本稿では，メモ リ・ウォール問題の解決策のうち，再計算に基づくメモリ・システム高性能化手法CCC (Computing Centric Computation) を提案する．CCC は，キャッシュ・ミス頻発命令を実行する代わりに再...計 算を行なうことで，主記憶へのアクセス回数を削減する．本稿では，性能向上が得られる可能性があ るかどうかに関する予備的な評価を行なった．評価対象ベンチマークにおいて最大45.3% の実行サ イクル数削減率を達成した．
In recent years, the performance gap between microprocessor speed and main memory latency has been increasing. This problem prevents higher throughput improvements and is well-known in the literature as the Memory-Wall Problem (MWP). This paper proposes a new method to minimize the MWP effect by means of [re-computation]. The basic idea is to replace frequently cache-missed loads (or delinquent loads) with a piece of code that regenerates the missed value (recomputation code). This method can be used to reduce the number of main memory accesses and consequently alliviate the MWP. From the experiments, one can obtain up to 45.3% reduction on computation time for SPEC2000 benchmark programs.show more
|
Export Mendeley
miwa04_1