Compile with -fno-math-errno

[tkittel]

As explained here:

https://stackoverflow.com/questions/57673825/how-to-force-gcc-to-assume-that-a-floating-point-expression-is-non-negative/57674631#57674631

we should compile with -fno-math-errno for more efficient code (not sure if this only affects gcc, or also clang?). Obviously as a private compilation flag for libNCrystal only.

Specifically I discovered the need for this when I was trying to get simd auto-vectorisation to work on a large number of neutrons, and no matter when I simply could not get a simple loop like the following to vectorise:

for ( unsigned i = 0; i < n; ++i)
   v[i] = std::sqrt(x[i])

The problem is that the errno handling adds a branch inside the sqrt implementation. Using __builtin_sqrt or sqrt(fabs(x[i])) didn't work either. However compiling with -fno-math-errno did the trick.

[tkittel]

Clang accepts this flag, but it might be mostly for the case where it is using the gnu libc. In any case, it seems to do no harm for clang.

[tkittel]

This will be in the next release.