| カテゴリ | 重要度 | ステータス | 解決状況 | 登録日時 | 最終更新 |
|---|---|---|---|---|---|
| 低 | 仮承認 | 未処理 | 2025-07-17 02:48 | 2025-07-17 21:57 | |
| テスター | hap | 担当者 | ソース | ||
| バージョン | 0.278 | 発生バージョン | 修正バージョン | ||
| 修正コミット | プルリク | ||||
| フラグ | |||||
| セット | |||||
| セット詳細 | |||||
| 概 要 | x86 CPUコアのREP命令が遅すぎる。 | ||||
| 詳 細 | x86 CPU cores: REP opcodes are too slow with tight scheduler intervals. REP opcodes are interruptable, and therefore, MAME checks for icount>0 while doing the REP loop. However, this also means that if icount<=0 due to another reason than an interrupt, it will redo the opcode from the start, instead of continuing the internal loop. This adds a couple of extra cycles and memory accesses that don't happen on the real CPU. If for example so-called perfect quantum is used, REP opcodes will be much slower (in other words, counter-intuitively, perfect quantum will lower timing accuracy for these opcodes). Applies to: cpu/i86 (i86, i186, i286) cpu/i386, probably cpu/nec - bug: REP opcodes interruptability unemulated - fixed for 0.279 and also fixed the issue described in this report cpu/v30mz - bug: REP opcodes interruptability unemulated | ||||
| 再現手順 | Unknown if there currently are any symptoms due to this bug. | ||||
| 追加情報 | |||||
| 添付ファイル | |||||