The toolkit is moving towards supporting quad-precision (via the __float128 GNU quadmath library), and possibly to higher precision via the GMP library.
Moving to quad precision involves several issues
- treatment of constants, eg pi, sqrt2, etc need float128 support
- BLAS/LAPACK library, it looks like a good candidate is MPACK
- EXPOKIT needs porting to __float128. Or maybe use a different library, see eg http://people.sc.fsu.edu/~jburkardt/f_src/matrix_exponential/matrix_exponential.html
- ARPACK needs porting to __float128. Probably a big job, but maybe someone has already done it?
- Handling adjustable parameters, for example eigensolver tolerances, and default values.
