Search code examples
assemblyx86x86-64

Newbie on x86 asm - equivalence of mov a,[b+2] and add b,2|mov a,[b]

I'm wondering if the following

mov eax, [ebx+4]

is equivalent to

add ebx, 4
mov eax, [ebx]

is there a performance improvement or a reason why the first should be preferred over the latter, except for a readability and code styling issue?

Solution

  • Although "equivalent", there are differences as noted below and there are several reasons to prefer the mov eax,[ebx+4] over the add, mov:

    1. The address calculation ebx+4 is performed "for free" by the address calculation logic in the CPU
    2. It is only one instruction long, vs two
    3. It does not cause a dependency stall (depending on the CPU implementation) between the results of an ALU operation and a memory load
    4. It preserves the FLAGS register (condition code), the MOV doesn't alter EFLAGS whereas ADD does
    5. It is fewer bytes in memory
    6. It does not modify EBX, although depending on the context that might not be a good thing, maybe you want EBX to be updated?
    7. It frees up micro-architectural resources so that other instructions may execute in parallel with the MOV. Modern CPUs will execute the MOV (a load from memory) in a LOAD/STORE execution unit that can perform its own address calculation, and the ADD in a FIXED-POINT execution unit... if you do your address calculation in the MOV instruction this frees up one of the FIXED-POINT execution units to do something else in parallel.