Opt Versions Save

Added

• Higher verbosity levels, along with documentation

Fixed

• Robustness to divide-by-zero in pure GN solver.

Added

• Roadmap
• Partially-automated regression test script.
• Verbose-mode GPU memory usage reporting
• Improved documentation
• Improved error messages

Fixed

• #91: Graph energies with no unknown energies produce erroneous results

Added

• Changelog
• Semantic version number
• threadsPerBlock initialization parameter. Globally sets #threads/block
• Auto-detection of native double-precision floating point atomics

Fixed

• Longstanding code emission bug on graph domains

No API changes since v0.1.0.

Added

• Simple tests/

Fixed

• Memory leaks

This is the initial public release of Opt.

We hope you will try it out if you are looking to solve large-scale non-linear least squares problems fast using a GPU.