Implement properly-rounded, arbitrary-precision floating-point under a permissive license
Similar to #4, there is a need for a permissively-licensed, header-only C++ implementation of arbitrary-precision, properly-rounded floating-point arithmetic (e.g., as a replacement for MPFR).
In addition to licensing concerns, two performance can be addressed beyond current copy-left libraries:
- Add support for a stack of scalars in each active precision so that no memory allocations are required when using temporary variables. This can have order-of-magnitude performance implications.
- Add support for memory pooling/packing so that arrays with uniform precision can be created with a single memory allocation.
A good test case for arbitrary-precision sparse-direct factorizations is solving a Linear Program for computing Delsarte bounds in coding theory.