Undefined reference to `pow' when the exponent is changed?

Question

I'm having some strange issues with pow(). If i do

return 44330*(1 - pow((float) pressure / PRESSURE0, 1/5.255F));

where pressure is an int32_t and PRESSURE0is a constant, I get an error saying undefined reference to `pow'. However, if I do

return 44330*(1 - pow((float) pressure / PRESSURE0, 1.0F));

it's fine. Am I doing something wrong?

Thanks

Answer 1

Generally, in order to use math functions like sqrt(), pow() etc. you need to link with the math library, libm (on gcc, use the -lm option).

However, the compiler is free to apply optimizations, especially when the arguments to the functions are constants. Consider the following two simplified functions:

double f1(int pressure) {
   return pow((float) pressure / PRESSURE0, 1/5.255F);
}

double f2(int pressure) {
   return pow((float) pressure / PRESSURE0, 1.0F);
}

This gets compiled into i386 code as follows (unnecessary assembler directives removed):

f1:
        pushl   %ebp
        movl    %esp, %ebp
        subl    $24, %esp

; (float) pressure / PRESSURE0
        fildl   8(%ebp)
        fldl    .LC0
        fdivrp  %st, %st(1)

; pow()
        fldl    .LC1
        fstpl   8(%esp)
        fstpl   (%esp)
        call    pow

        leave
        ret

f2:
        pushl   %ebp
        movl    %esp, %ebp

; (float) pressure / PRESSURE0
        fildl   8(%ebp)
        fldl    .LC0
        fdivrp  %st, %st(1)

; NO call to pow(x, 1.0), since it would not change the value of x

        popl    %ebp
        ret

In f2(), since the exponent to pow() is 1.0, there is no need to call the pow() function - the compiler detects this and removes the function call. Hence, if there are no other calls to any math functions in your code, it is not necessary to link against libm in that particular case.

See also

• literal constant vs variable in math library
• Do math functions of constant expressions get pre-calculated at compile time?