Question

I have two 32 bit integers i1,i2 which I need to convert to floats f1,f2 in such a way that their relative ordering is preserved (i.e. i1 < i2 => f1 < f2)

Will a reinterpret_cast do the trick? Is there some better way?

Answer 1

If the integer values are less than 2²⁴, just convert the values:

float f1 = i1, f2 = i2;

For larger values, you will lose precision and two distinct integers may convert to the same floating point value.

On the other hand, you could copy the bit pattern. If your floats are IEEE754, then this requires that the sign bits agree and that neither integer represents some form of NaN. (If the sign bits do not agree, you must beware of -0.f == +0.f:.) To copy the binary representation:

float f1;
std::copy(reinterpret_cast<const char*>(&i1),
          reinterpret_cast<const char*>(&i1) + 4,
          reinterpret_cast<char*>(&f1));