v0.2.0

Release v0.2.0: New FactorizationSettings-API

The new factorization settings API lets you select settings during
numerical factorization.

Currently, only a single setting is exposed, pivot_tolerance,
but the API lays the groundwork for adding arbitrary configuration
in a backward-compatible way.

Full Changelog: v0.1.0...v0.2.0

v0.1.0

First version of Leopard.

Quality metrics

No known correctness problems

The algorithm should give correct results for all valid inputs, and give detailed error information for any invalid input.
If you find any problem, file an issue.

Competitive performance

The algorithm still has some way to go with respect to pure numerical performance. This release's focus has been getting something out the door that gives the correct result for any input with a relatively stable API.
The algorithm is asymptotically faster than a dense method for any reasonably sized sparse matrix.
It is not as fast as the best paid solvers available, but on par with eg. the open-source UMFPACK.

API stability

This is the first release of Leopard. The API has been reworked several times in the run-up to this first release, but it
would be surprising if there will not be some breaking changes in subsequent revisions, given the library's early state.

Portability

The library is compiled for an old GLIBC and should link on almost any modern Linux distro. The library has no system requirements except libc, libgcc, and pthread.

TODOS:

• Windows support: if you need this, please file an issue and I will prioritize it.
• 32-bit architecture build: If you need this, please file an issue and I will prioritize it.
• Exotic instruction sets for better performance, eg. AVX: If you need this, please file an issue and I will prioritize it.
• Better runtime performance: This can be realized through a multitude of optimizations. There are still low hanging fruit with regards to this.