Repository navigation
CoherencyLatency v1.0.0 — macOS / Windows
基于 CXWorld/MicroBenchX v1.0.5 中 CoherencyLatency 的多平台移植版本。
非官方移植,与 CapFrameX / CXWorld 没有隶属或背书关系。
- CoherencyLatency-macOS-v1.0.0.zip — macOS 10.7 Lion 或更高版本,解压后直接运行;首次打开被 Gatekeeper 拦截,请右键选择“打开”,或执行
xattr -dr com.apple.quarantine CoherencyLatency。 - CoherencyLatency-Windows-v1.0.0.exe — Windows Vista SP2 / 7 / 8 / 8.1 / 10 / 11,单文件启动器,静态 CRT,无需安装 VC++ 运行库。
保留原版 bounce / owned 两种模式、命令行参数和分号 CSV 输出;修复了原版的最小值统计 bug,矩阵内存从 O(n2 × runs) 降到 O(n2)。
性能优化:工作线程每个核心对只创建一次、跨 run 复用(原版每次 run 重建线程),计时改为整数高精度时钟,进度条改为每个核心对刷新一次(原版每行才刷新一次,多核机器上可能长时间停在 0%)。4/8/16/32 槽位的短测试耗时约为原版的 39–55%,峰值内存稳定在约 4.2 MB,不再随 runs 增长。测量方法与输出格式未变。
SHA256:
CoherencyLatency-macOS-v1.0.0.zip:72386be585d58ffd89f5bc4270c48dd3a7f86c7218e84580ac88dc5680c21248CoherencyLatency-Windows-v1.0.0.exe:a3bbd10808a40bcff3ca91821098d3985d5a5ce50802d473c31f885c000a6cfe
English
Multi-platform ports of CoherencyLatency from MicroBenchX v1.0.5.
This is an unofficial port, not affiliated with or endorsed by CapFrameX or CXWorld.
- CoherencyLatency-macOS-v1.0.0.zip — macOS 10.7 Lion or later; unzip and run. If Gatekeeper blocks the first launch, right-click and choose Open, or run
xattr -dr com.apple.quarantine CoherencyLatency. - CoherencyLatency-Windows-v1.0.0.exe — Windows Vista SP2 / 7 / 8 / 8.1 / 10 / 11; single-file launcher, static CRT, no Visual C++ runtime required.
The original bounce / owned modes, command-line options and semicolon-separated CSV output are preserved. The port fixes the original minimum-statistic bug and reduces matrix memory from O(n2 × runs) to O(n2).
Performance in this build: worker threads are created once per pair and reused across runs (the original rebuilt them for every run), timing uses an integer high-resolution clock, and the progress bar now refreshes after every core pair (the original only refreshed once per row, so it could sit at 0% for a long time on many-core machines). Short tests at 4/8/16/32 slots take roughly 39–55% of the original wall time, and peak working set stays flat at about 4.2 MB instead of growing with runs. Measurement method and output format are unchanged.