Read the problem as: 

integrate cdfN(f(a,b,x))*pdfGamma(c,x) over the positive reals  with f(x) =
= a/sqrt(x)+b*sqrt(x), cdfN is the cumulative normal distribution, pdfGamma the
density of the gamma distribution cdfGamma (called P in Abramowitz & Stegun).

The limits of f depend on the signs of a and b, so the first argument will
reach 0 or 1. Or 1/2. Now estimate where to replace it by that constant (and
justify why ... but i guess for mj's solution one would have to as well).

So for being 'close to 0' one switches to a cdfGamma and for the infinite
tail use an adaptive integration (instead of cutting of) which stops if the
integrand is decreasing and becomes small.

For c = time/nu getting large (i.e. beyond 170, where most implementations
will quitt - but then impl volatility is almost constant) one can switch to
asymptotics (which i did without extreme care).

For a=0 or b=0 one can even write down 'easy' explicite solutions in terms of
hypergeometric functions, which i ignored as well as i ignored the numerical
case sigma --> 0.