perf is a Linux profiling tool that reads hardware performance counters from the CPU. Measure cycles, instructions, cache misses, branches—everything the CPU tracks natively. Answer the fundamental question: where is the program spending time, and why is it slow?

Unlike guessing, perf gives ground truth from the hardware. Low overhead (1-5%), no recompilation needed, and the data is reliable. In HPC work, perf is the first tool to reach for when something is slower than expected.

$ perf stat ./program
Performance counter stats for './program':
     4,102.38 msec task-clock              # 3.98 CPUs utilized
    16,847,203,412 cycles                  # 4.107 GHz
    12,334,901,088 instructions            # 0.73 insn per cycle
       401,292,771 cache-misses            # 33.33% of cache refs

IPC (instructions per cycle) of 0.73 on a modern superscalar CPU means the pipeline is stalled. A 33% cache miss rate means it's waiting for memory. Two numbers tell you everything you need to know to start optimizing.

1. Basics
2. perf stat
3. perf record
4. perf report
5. perf annotate
6. Events
7. Metrics
8. Flame graphs
9. Sampling
10. Scripting