It's not in the standard. Some implementations choose to put a #define of PI or M_PI in cmath or math.h, sometimes requiring you to specify #define _USE_MATH_DEFINES yourself.
When I did my very first programming course, it was suggested to use 4*atan(1). This works in any programming language and gives PI to the precision of the particular machine.
The main drawback is that the result is assigned to a variable, not a const.
Mostly I tend to hard-code it as a const as recommended above, using a value such as 3.14159265358979323846264338328, or one of the following:
1 2 3
constlongdouble PI = 3.141592653589793238L;
constdouble PI = 3.141592653589793;
constfloat PI = 3.1415927;