Assembly Language for Beginners

3.11. INLINE FUNCTIONS

push edi
mov edi, DWORD PTR _outbuf$[esp+4]
add edi, 10
mov ecx, 32
rep movsd
pop edi
pop esi
ret 0
_memcpy_128 ENDP

When copying 123 bytes, 30 32-bit words are copied first usingMOVSD(that’s 120 bytes), then 2 bytes are
copied usingMOVSW, then one more byte usingMOVSB.

Listing 3.48: Optimizing MSVC 2010

_inbuf$ = 8 ; size = 4
_outbuf$ = 12 ; size = 4
_memcpy_123 PROC
push esi
mov esi, DWORD PTR _inbuf$[esp]
push edi
mov edi, DWORD PTR _outbuf$[esp+4]
add edi, 10
mov ecx, 30
rep movsd
movsw
movsb
pop edi
pop esi
ret 0
_memcpy_123 ENDP

GCC uses one big universal functions, that works for any block size:

Listing 3.49: Optimizing GCC 4.8.1

memcpy_123:
.LFB3:
push edi
mov eax, 123
push esi
mov edx, DWORD PTR [esp+16]
mov esi, DWORD PTR [esp+12]
lea edi, [edx+10]
test edi, 1
jne .L24
test edi, 2
jne .L25
.L7:
mov ecx, eax
xor edx, edx
shr ecx, 2
test al, 2
rep movsd
je .L8
movzx edx, WORD PTR [esi]
mov WORD PTR [edi], dx
mov edx, 2
.L8:
test al, 1
je .L5
movzx eax, BYTE PTR [esi+edx]
mov BYTE PTR [edi+edx], al
.L5:
pop esi
pop edi
ret
.L24:
movzx eax, BYTE PTR [esi]
lea edi, [edx+11]
add esi, 1