"xor %eax, %eax" is even smaller (2 bytes vs. 3) and does the same thing since operations on the 32-bit subregisters are defined to zero the upper half [1]. Thi

"xor %eax, %eax" is even smaller (2 bytes vs. 3) and does the same thing since operations on the 32-bit subregisters are defined to zero the upper half [1]. This is what compilers tend to do when they want to zero a register. [1] This is not true for 8- or 16-bit subregisters, causing all kinds of fun issues (see "partial register stalls", also discussed in Agner Fog's materials).