; this takes two bytes, it is slower on some processors but faster or the same on others
label:
    loop label

; this takes three bytes but is slower on some processors
label:
    dec ecx
    jnz label

; this is takes five bytes and is potentially faster than the above
label:
    sub ecx,1
    jnz label

; there is also a two-byte jecxz instruction but no jecxnz