About memcpy(). Here is sample file. Compile it to assembler:

 gcc -O2 -S -masm=intel fnv.c

With memcpy() main loop is compiled to:

.L3:
    mov esi, DWORD PTR [ebx]
    imul    eax, eax, 1000003
    add ebx, 4
    xor eax, esi
    sub ecx, 1
    mov DWORD PTR [esp+24], esi
    jne .L3

With per-byte copy it is compiled to:

.L3:
    mov dl, BYTE PTR [ecx]
    imul    eax, eax, 1000003
    sub ebp, 1
    movzx   ebx, BYTE PTR [ecx+1]
    movzx   edi, BYTE PTR [ecx+2]
    movzx   esi, BYTE PTR [ecx+3]
    add ecx, 4
    mov dh, bl
    sal edi, 16
    movzx   edx, dx
    sal esi, 24
    or  edx, edi
    or  edx, esi
    xor eax, edx
    cmp ebp, -1
    jne .L3