CareerCup

I got almost the same but 16777222 (C# VS2010).
Results depend on how float numbers stored in the memory, but I don't know how exactly explain this. For double type you'll get infinite loop.

Floating point numbers are stored in base 2 "scientific notation". Once x gets large enough (about 2**24, since floats have 23 bits of significand), x is the closest float to the real number x + 1. The exact behavior depends on rounding modes and how floating point is implemented internally.

Doubles have the same problem, except you need to have the patience to get to about 2**53...

Do u mean the values achieved their max thats y they didn't increase??
as per IEEE the float representation is:
bit 31-->Sign bit
bit 30-23-->Exponent
bit 22-00--> Mantissa
In my case Exponent is 11000 and mantissa is 1. No overflow.

Not the max; the exponent could be larger. However, with respect to that exponent, the
"ones" bit of the significand (formerly the mantissa) is one past the end and thus is thrown away.

I agree -- the code should terminate when adding 1.0 can't change the mantissa of f1 (because the exponent has gotten "pretty large"). Overflowing f1 isn't required