nx_perf_counters.nx

nx_perf_counters.nx -- bits-up hardware performance counters via perf_event_open, PURE NishiLang (no libc, no `perf` binary, no C). The STATE-INDEPENDENT energy path: instruction / cycle / cache counts do NOT care whether the node is charging or on AC -- they are the CPU's own activity, read straight from the silicon's performance-monitoring unit. They feed a calibrated per-event energy model that yields real joules on ANY hardware in ANY power state. (Battery-discharge was a weak fallback; THIS is the S-class answer -- a real measurement everywhere.) Research (Cardinal #4, real sources): event-driven / PMC-based power modeling -- Bellosa, "The Benefits of Event-Driven Energy Accounting," ACM SIGOPS EW 2000; Isci & Martonosi, "Runtime Power Monitoring using PMCs," MICRO 2003; Bircher & John, "Complete System Power Estimation using PMCs," ISPASS 2007; per-operation energy figures from Horowitz, "Computing's Energy Problem," ISSCC 2014. perf_event_open(2) is the Linux interface; the attr struct is built byte-exact.