How data movement changes a destination register
This snippets is from "Computer Systems A Programmer’s Perspective, by Randal E. Bryant and David R. O’Hallaron"
As described, there are two different conventions regarding whether and how data movement instruc- tions modify the upper bytes of a destination register. This distinction is illustrated by the following code sequence:
1 movabsq $0x0011223344556677, %rax %rax = 0011223344556677
2 movb $-1, %al %rax = 00112233445566FF
3 movw $-1, %ax %rax = 001122334455FFFF
4 movl $-1, %eax %rax = 00000000FFFFFFFF
5 movq $-1, %rax %rax = FFFFFFFFFFFFFFFF
In the following discussion, we use hexadecimal notation. In the example, the instruction on line 1 initializes register %rax to the pattern 0011223344556677. The remaining instructions have immediate value −1 as their source values. Recall that the hexadecimal representation of −1 is of the form FF. . .F, where the number of F’s is twice the number of bytes in the representation. The movb instruction (line 2) therefore sets the low-order byte of %rax to FF, while the movw instruction (line 3) sets the low-order 2 bytes to FFFF, with the remaining bytes unchanged. The movl instruction (line 4) sets the low-order 4 bytes to FFFFFFFF, but it also sets the high-order 4 bytes to 00000000. Finally, the movq instruction (line 5) sets the complete register to FFFFFFFFFFFFFFFF.
some useful links