[avr-libc-dev] What is the necessity of _FFS() ?

[Dmitry K.]

What is the reason of '_FFS' appearance in the last Avr-libc CVS?


Originaly 'ffs' function was defined as:

    #define ffs(x) (__builtin_constant_p (x) \
        ? ..experession with (x)..           \
        : ffs(x) )

Such definition excludes a call of function if the argument is a
constant known during compilation. In this case after processing
expression the compiler will substitute the ready answer instead
of a call of function.

You can make own functions on the basis of 'ffs', they will possess
too such property, for example:

    #define N   8

    /* Return number of tail zeroes. (reverse to _BV) */
    static inline int ntz (int x)
    {
        return ffs(x) ? ffs(x) - 1 : 16;
    }

    int foo (int x)
    {
        return ntz(x) + ntz(N);
    }

will produce for function 'foo':
    foo:
        rcall ffs
        sbiw r24,0
        breq .L2
        sbiw r24,1
        rjmp .L40
    .L2:
        ldi r24,lo8(16)
        ldi r25,hi8(16)
    .L40:
        adiw r24,3
        ret

On the other hand, you can quickly replace a constant with a variable,
it will not lead to monstrous swelling of a code (it sometimes helps at
debugging).  In the above example substitution of:

    //#define N   8
    extern int N;

will produce second call of 'ffs' (with N variable).

I do not see any reason to introduce new function '_FFS'.

Thanks,
Dmitry.