Why x86 compilers prefer xor over sub for zeroing registers: encoding efficiency and convention

Summary

Matt Godbolt's article explores why x86 compilers prefer the `xor eax, eax` instruction over `sub eax, eax` for zeroing out registers. The key reason is that `xor` is the most compact encoding on x86, being several bytes shorter than `mov eax, 0` since it avoids encoding a four-byte constant. The article notes that while both `xor` and `sub` could theoretically work, `xor` became the standard idiom due to historical convention, encoding efficiency, and compiler optimization preferences.