Compiler expert Matt Godbolt explains why x86 compilers prefer `xor eax, eax` over `sub eax, eax` for zeroing registers. This technique uses fewer bytes by avoiding a four-byte constant. The x86 architecture lacks a dedicated zero register, forcing compilers to generate explicit instructions. This choice stems from historical implementation quirks rather than functional necessity.

xor eax, eax and `sub eax, eax` both clear the register in one instruction, yet flag behavior differs subtly. `xor` leaves the auxiliary flag undefined, while `sub` explicitly clears it. Intel added detection for both patterns, renating the destination to an internal zero register. This optimization bypasses execution entirely and breaks dependency chains.

Stack Overflow concerns persist that other CPU manufacturers may only implement detection for `xor`, cementing its dominance. Once a minor advantage snowballed through compiler adoption into an industry standard. Raymond documents this evolution spanning decades on The Old New Thing. His analysis reveals how minor implementation details solidify into enduring programming conventions.