Target-specific tuning is where compilers earn benchmarks: choosing lea over add chains, aligning loop tops to 16 bytes, and scheduling for the pipeline are why -march=native matters. GCC and LLVM each carry per-microarchitecture cost models — thousands of lines for Skylake alone — doing at scale what you are doing here by hand.