[petsc-users] petscprintf and __float128

[Barry Smith]

> On Sep 24, 2015, at 9:24 PM, Gideon Simpson <gideon.simpson at gmail.com> wrote:
> 
> I saw that Barry posted something about this in the dev listserv last December, but there wasn’t any follow up.  Right now, on an OS X based machine with gcc compilers, if I do a PetscPrintf with a __float128, with %g, it outputs 0.  Is there a way to get around this, other than using quadmath_snprintf?

   There are only two ways: caste with a (double)   or use the quadmath_snprintf. When printing the residual norms it is fine to just cast to doubles to look at convergence you don't need to do anything beyond the usual -snes_monitor -ksp_monitor etc (warning only PETSc LU solvers support __float128 so you can't use Superlu or MUMPs for the direct solvers). 

  I found quadmath_snprintf cumbersome to use in PETSc and not supported on all systems so didn't try to use it in PETSc.

  Barry

> 
> 
> -gideon
>