Re: [Help-gsl] [OT] double vs long double

[David Schaich]

Processor speed is one issue, memory access is another.  If moving from 
double to long double could make the difference between leaving certain 
things in cache vs. having to fetch them from the disk, you could see a 
significant performance hit.

If you don't have any pressing need to use long double as opposed to 
double, I would encourage keeping things simple and making the most 
efficient use of available resources.

Cheers,
David Schaich
physics.bu.edu/~schaich


Marco Maggi wrote:
> Ciao,
>
> I am reorganising my code that uses GSL vectors and matrices;
> I have to choose a scalar number type and I am wondering if,
> given modern processors speed, it makes sense to just use "long
> double" rather than "double" (when available).  I have no
> special
> needs, my code is to be used in general purpose matrix
> computations.
>
> I know that many GSL algorithms, like ODE, are implemented only
> for "double", but when moving values in/out of matrices from/to
> algorithms a conversion can take place.
>
> We do not use 16 bits "int" anymore, so are we not there yet
> for "long double"?