Just explain that in slightly more detail and I will raise a support ticket with QNX.
Basically fpclassify is returning FP_NORMAL rather than FP_SUBNORMAL when passed a denormalised number (at least that's my assumption based on the results you reported). The simple test case below should allow you to check this, Regards, John. P.S. It's probably not a good idea to post headers that aren't open-source to a public list mailing list ;-) #include <assert.h> #include <math.h> #include <limits> template <class T> void test(T t) { T denorm; // denorm_min should be either zero or a denormalised number: denorm = std::numeric_limits<T>::denorm_min(); if(denorm != static_cast<T>(0.0)) assert(fpclassify(denorm) == FP_SUBNORMAL); // any number less than min() should be demormal or zero: denorm = std::numeric_limits<T>::min() / 2; if(denorm != static_cast<T>(0.0)) assert(fpclassify(denorm) == FP_SUBNORMAL); } int main() { test(1.0F); test(1.0); test(1.0L); }