Version 2.6 Beta

This is a beta version for integration testing of the log.

Version 2.5.1

This minor release fixes any compilation bug from gcc version 10. It also has some code cleanup from when I overused iso_c_bindings.

Version 2.5

Features:

• Ability to gather a matrix on to each process locally.
• Ability to change a matrix to a fixed sparsity pattern to facilitate conversion.

Bug Fixes:

• Some bug fixes related to the formatting of matrix market output.

Miscellaneous:

• Automatic Code Linting
• Some cleaning up of constants

Version 2.4.3 Beta

Another beta for integration testing.

Version 2.4.2 Beta

Another beta for integration.

Version 2.4.1 Beta

A beta release for checking integration.

Version 2.4

The main improvement in 2.4 is related to dense matrix operations. As a backup plan, NTPoly can convert to dense matrix multiplication when it determines that a block is not very sparse. I noticed that there were some potential bugs in this code when using older compilers. I also discovered that the performance was sub-optimal because we were still transposing those matrices using the sparse data structures. Version 2.4 fixes those issues, and in general improves the code.

As another minor improvement, I've set the matrix maps module to allow optional arguments.

Finally, we've fully switched from Travis-CI to Github actions, and I've removed any legacy files that were only there for Travis.

Version 2.3.2

Version 2.3.2 contains some fixes.

Bug Fixes:

• @vyu16 found some memory that was not being deallocated.

Miscellaneous:

• Build system fixes for newer versions of OpenMPI.
• Continuous integration simplified.
• Code cleanup with the flake8 tool.

Version 2.3.1

Version 2.3.1 contains some fixes and minor improvements.

Bug Fixes:

• Major fix to the implementation of the purification based extrapolation
  technique (thanks to Victor Yu).

Features:

• Rubennson's Scale and Fold purification technique.

Miscellaneous:

• cmake now automatically handles the use of position independent code.
• More options for cmake.
• Improvements to the CI system.

I would note that the Scale and Fold technique requires knowledge about the
homo and lumo band locations, which makes it less easy to use. However, the
performance is good, so I think this should at least be included as a
benchmark to compare new methods against.

I hope that this new version of NTPoly is easier to compile. I am also planning
on adding it to the SPACK package manager soon.

Spack Test Release

More extensive cmake options (#123)

* More extensive cmake options.

* Better warning message.

* Removed custom test target

Older cmake versions complain about this usage.