Re: [Bug-gnubg] Possible error in SSE sigmoid calculation

[Philippe Michel]

I've commited the fix, including the fast-math related ifdef below. As a 
result of the latter, the code is now sligtly slower by default but one 
can always use -ffast-math everywhere and recoup that and some more.

I didn't run serious speed or non-regression tests, but it looks like this 
option is well worth using.


> The differences between nosse and ssecvtt are due, I think, to the usage of a 
> reciprocal instead of a true division in 1/(1+exp(x)). I suppose this was 
> found acceptable at the time and would dissappear if everything was compiled 
> with the -ffast-math flag. Maybe one could use something like
>
> #ifdef __FAST_MATH__
> ... use reciprocal intrinsic
> #else
> ... use divide intrinsic
> #endif