dolfin team mailing list archive

[Anders Logg <question97399@xxxxxxxxxxxxxxxxxxxxx>]

Question #97399 on DOLFIN changed:
https://answers.launchpad.net/dolfin/+question/97399

    Status: Open => Answered

Anders Logg proposed the following answer:
On Thu, Jan 14, 2010 at 05:42:17PM -0000, Murtazo Nazarov wrote:
> Question #97399 on DOLFIN changed:
> https://answers.launchpad.net/dolfin/+question/97399
>
>     Status: Answered => Open
>
> Murtazo Nazarov is still having a problem:
> Garth Wells wrote:
> > Question #97399 on DOLFIN changed:
> > https://answers.launchpad.net/dolfin/+question/97399
> >
> >     Status: Open => Answered
> >
> > Garth Wells proposed the following answer:
> >
> > Murtazo Nazarov wrote:
> >
> >> New question #97399 on DOLFIN:
> >> https://answers.launchpad.net/dolfin/+question/97399
> >>
> >>
> >> In example in /demo/ode/stiff/cpp/Timedependent1.h I tried to change function f(u,t) into
> >>
> >>
> >>   void f(const real* u, real t, real* y)
> >>   {
> >>     y[0] = -1000.0 * u[0] + cos(t);
> >>   }
> >>
> >> But I get a compilation error:
> >>
> >> In file included from main.cpp:12:
> >> TestProblem1.h: In member function 'virtual void TestProblem1::f(const dolfin::real*, dolfin::real, dolfin::real*)':
> >> TestProblem1.h:27: error: cannot convert 'dolfin::real' to 'double' for argument '1' to 'double cos(double)'
> >> scons: *** [main.o] Error 1
> >>
> >> It seems dolfin::real is not double by default for the latest dolfin. Is it bug or I can fix it easily in my code?
> >> Any suggestion?
> >>
> >>
> >
> > Use 'double'.
> >
> > Garth
> >
>
> I tried with 'double'.
>
>   void u0(double* u)
>   {
>     u[0] = 1.0;
>   }
>
>   void f(const double* u, double t, double* y)
>   {
>     y[0] = -1000.0 * u[0] + cos(t);
>   }
>
> Then I got similar error
>
> main.cpp:64: error: cannot declare variable 'test_problem' to be of
> abstract type 'TestProblem1'
> TestProblem1.h:12: note:   because the following virtual functions are
> pure within 'TestProblem1':
> /afs/nada.kth.se/home/x/u18gh7kx/data/Work/fenics/build/include/dolfin/ode/ODE.h:83:
> note:     virtual void dolfin::ODE::u0(dolfin::real*)
>
> The I tried like this:
>
>   void u0(real* u)
>   {
>     u[0] = 1.0;
>   }
>
>   void f(const real* u, double t, double* y)
>   {
>     y[0] = -1000.0 * u[0] + cos(t);
>   }
>
> Still error:
>
> In file included from main.cpp:12:
> TestProblem1.h: In member function 'void TestProblem1::f(const
> dolfin::real*, double, double*)':
> TestProblem1.h:27: error: cannot convert '__gmp_expr<__mpf_struct [1],
> __gmp_binary_expr<__gmp_expr<__mpf_struct [1], __gmp_binary_expr<double,
> __gmp_expr<__mpf_struct [1], __mpf_struct [1]>, __gmp_binary_multiplies>
>  >, double, __gmp_binary_plus> >' to 'double' in assignment
> scons: *** [main.o] Error 1

Looks like you compiled DOLFIN with GMP (GNU multiprecision) and then
you need to use real (not double) in the ODE solvers.

What happens above is that cos() is not defined for reals (only for
doubles). So you need to either compile DOLFIN without GMP, or
use

  y[0] = -1000.0 * u[0] + cos(to_double(t));

Which option you want depends on if you need extended precision. My
guess is you don't so just add enableGmp=no when compiling.

--
Anders

-- 
You received this question notification because you are a member of
DOLFIN Team, which is an answer contact for DOLFIN.