You could also theoretically just change the type in SumCtx to be __float128 instead of double or LONGDOUBLE_TYPE