isnan() and isinf() in C++

If it wasn't optimized, you wouldn't need the volatile temp.

Well, GCC has a pragma (or attribute?) that can enable/disable various command line options per function. This includes the optimizer attributes, so you could apply -O0 to a specific function.

Of course, GCC will also always have a working isnan(), so that doesn't help.

Quote:

---

Originally Posted by cpjust

Maybe it would be good to add a #pragma to the C/C++ standard to do something like this:

Code:

---

#pragma no_optimize_begin
bool myisnan(double var)
{
    volatile double temp = var;
    return temp != temp;
}
#pragma no_optimize_end

---

---

By definition, pragmas are compiler and/or operating system specific, and therefore no particular particular pragmas will ever be added to the C/C++ standards.

Actually, C99 has standard pragmas in the form

#pragma STDC pragma_name

However, a no-optimization directive would imply that there is a standard implementation technique for each platform.

Another possibility is of course to use integer bit operations to do exactly what I expect the isnan function actually does:
Check if all the bits in exponent is 1 (although in real life, since the exponent is offset, it actually has the highest bit cleared).

http://en.wikipedia.org/wiki/IEEE_fl...point_standard
http://en.wikipedia.org/wiki/NaN

--
Mats