Perf Counters

ELI5 - The Vibe Check

Perf counters are the CPU's own built-in scorekeepers, tiny hardware registers that silently count things like cache misses, branch mispredictions, and instructions executed, without slowing your program down at all. Tools like Linux's perf just read these numbers straight off the chip, so you get an answer to is this slow because of cache misses or because of the algorithm without any guessing.

Real Talk

Performance counters (hardware performance counters) are special-purpose registers built into the CPU that count low-level microarchitectural events like cache misses, branch mispredictions, instructions retired, and cycles stalled. Profiling tools like Linux perf, VTune, or perf_event_open read these counters directly from hardware, giving near-zero-overhead insight into exactly why code is slow at the CPU level.

Show Me The Code

perf stat -e cache-misses,branch-misses,instructions ./my_program

When You'll Hear This

"perf stat showed a huge cache-miss rate, that's the false sharing you suspected." / "Hardware counters told us it was branch mispredictions, not the algorithm."