XOR'ing a register with itself is the idiom for zeroing it out. Why not sub?

XOR reg, reg zeroes a register faster than SUB reg, reg by 1 CPU cycle on most x86 architectures, a micro-optimization that still ignites debate among assembly die-hards. The technique leverages a dedicated zeroing idiom recognized by modern processors like Intel's Skylake, which eliminates dependency stalls, outperforming #2's discrete transistor approach in raw efficiency. With 38 points and 20 comments, this thread exemplifies the arcane processor knowledge that keeps low-level programming alive.