[PPL-devel] floating point behaviour

[Roberto Bagnara]

On 06/15/2010 09:36 PM, piotrm wrote:
> Apologies if this is not the right place to inquire about this, cannot
> seem to find any other outlet:

Hi piotrm,

you should check out the library documentation.

> Is it correct behavior for floating point calculations to be done
> differently once ppl.hh is included in a c++ program as well as the
> printing of floats? Here is an example of what I'm referring to:
>
> #include <cstdio>
>
> int main() {
> double a = 0.50000000000000011102230246251565404236316680908203125;
> double b = 0.5f / a;
> printf("%0.60f %f %0.60f %f\n", a, a, b, b);
> return 0;
> }
>
> This prints (for me)
>
> 0.500000000000000111022302462515654042363166809082031250000000 0.500000
> 0.999999999999999777955395074968691915273666381835937500000000 1.000000
>
> If I include ppl, as in below:
>
> #include <cstdio>
> #include "ppl.hh"
>
> int main() {
> double a = 0.50000000000000011102230246251565404236316680908203125;
> double b = 0.5f / a;
> printf("%0.60f %f %0.60f %f\n", a, a, b, b);
> return 0;
> }
>
> I get the following
>
> 0.500000000000000111022302462515654042363166809082031250000000 0.500000
> 0.999999999999999888977697537484345957636833190917968750000000 0.:00000
>
> Notice the strange way the b is printed the second time. I also found
> instances where numbers that are close to 0.01 get printed as 0.00:000
>
> Is this perhaps an installation issue on my part? I noticed this using
> the OCaml interface, that is, use of ppl changed the behavior of the
> float types in OCaml in the same way the behavior is changed in C++ above.

See the documentation of

   void Parma_Polyhedra_Library::set_rounding_for_PPL()
   void Parma_Polyhedra_Library::restore_pre_PPL_rounding()
   void Parma_Polyhedra_Library::initialize()
   void Parma_Polyhedra_Library::finalize()

Use also `man libppl' to see the role played by the
PPL_NO_AUTOMATIC_INITIALIZATION macro.
All the best,

    Roberto

-- 
Prof. Roberto Bagnara
Applied Formal Methods Laboratory
Department of Mathematics, University of Parma, Italy
http://www.cs.unipr.it/~bagnara/
mailto:bagnara at cs.unipr.it