@article{oai:ipsj.ixsq.nii.ac.jp:00018294, author = {三輪, 忍 and 福山, 智久 and 嶋田, 創 and 五島, 正裕 and 中島, 康彦 and 森眞一郎 and 富田眞治 and Shinobu, Miwa and Tomohisa, Fukuyama and Hajime, Shimada and Masahiro, Goshima and Yashiko, Nakajima and Shin-ichiro and Mori, ShinjiTomita}, issue = {SIG12(ACS15)}, journal = {情報処理学会論文誌コンピューティングシステム(ACS)}, month = {Sep}, note = {PHT(Pattern History Table)における破壊的競合を抑制する方法にフィルタ機構がある.フィルタ機構では,強偏向の分岐命令はPHT を使用しないようにすることで,PHT の予測ミス率を低下させる.一方,パス情報が予測の手がかりになることが最近になって分かってきた.そこで本稿では,フィルタ機構においてパス情報を利用する手法を提案する.強偏向のパスはPHT を使用しないようにすることで,予測ミス率の低下を狙う.本手法をGlobal Perceptron Predictor に適用した場合,平均0.14%ミス率が低下した.特に,go においては0.7%のミス率低下が見られた.また,本手法をPath Based Predictor,Path Trace Predictor に適用した場合でも,go において0.6~0.7%のミス率低下が見られた., Branch filter mechanism is a method which reduces destructive aliasing on PHT (Pattern History Table). This improves the misprediction rate not to use PHT for the branches with strong tendencies. Otherwise, it proves recently that path traces are hint for branch predictions. So, we propose branch filter mechanism with path traces. It is supposed that this mechanism improves the misprediction rate not to use PHT for the path trace with strong tendencies. When this mechanism is impremented on Global Perceptron Predictor, the average misprediction rate is reduced by 0.14%. Especially, the misprediction rate of go is reduced by 0.7%. And, when proposal mechanism is implemented on Path Based Predictor, on Piecewise Linear Predictor, and on Path Trace Predictor, the misprediction rate of go is reduced by 0.6%~0.7%}, pages = {108--118}, title = {パス情報を用いた分岐フィルタ機構}, volume = {47}, year = {2006} }