What compiler optimization flags actually enable under the hood:

• -O2: Enables standard optimizations (instruction scheduling, register allocation, inlining, common subexpression elimination).
• -O3: Adds aggressive loop unrolling, SIMD auto-vectorization, and function cloning (can increase binary code size).
• -Os / -Oz: Prioritizes code compactness, disabling loop unrolling to maximize CPU L1 Instruction Cache hits.
• -Ofast: Enables -O3 plus -ffast-math (breaks IEEE 754 floating point standard compliance and NaN/Infinity guarantees!).

Warning: Never use -Ofast in financial, cryptographic, or physics simulation software because -ffast-math assumes floats are associative and ignores NaN checks.

In many instruction-cache bound servers, -Os actually outperforms -O3 due to fewer L1i cache misses!