perf stat -e [event,...] COMMAND [ARGS] (Example usage) perf stat -e cycles:u,instructions:u,cache-references:u,cache-misses:u ./matmul (Some useful events) cycles:u - Total cycles (Be careful of what happens during CPU frequency scaling with respect to performance) instructions:u - Retired instructions cache-references:u - Usually, indicates LLC accesses, and it may include prefetch, coherence message depending on your CPU cache-misses:u - Usually, indicates LLC misses L1-dcache-load:u - L1 cache load accesses L1-dcache-load-misses:u - L1 cache load misses L1-dcache-stores:u - L1 cache store accesses L1-dcache-store-misses:u - L1 cache store misses