double (64-bits) and __int64 (which is allowed here, since timus have
MSVC++ compiler) are not enough to hold fibonacci numbers which are
generated during the computation. :(
I've got AC with Pascal, since it's extended have 80-bit precision.
The main problem in the fact that MSVC++ have not 80-bit type at all!
even long double which is 80 bit in Borland C++ is 64bit here (the
same trouble was in 1114).

But if u've got AC with C++ , feel free to send me your solution on
__my_inbox@mail.ru

There exist a solution that doesn't use such long sunbers. The only
test when you can obtyain them is when there are all the zeroes.

Another way is to use the binary-serach like algorithm