AVR-Chat

>
>        
>        clr             sum
>        sbrc            tmp,0
>        inc             sum
>        sbrc            tmp,1
>        inc             sum
>        sbrc            tmp,2
>        inc             sum
>        sbrc            tmp,3
>        inc             sum

That's one cycle longer (but smaller code) than the lookup table.

>        Another way:
>
>        clr             sum
>        lsr             tmp
>        adc             sum, tmp
>        lsr             tmp
>        lsr             tmp
>        adc             sum, tmp

Ok, that is better, 6 cycles, and no table.