Some parallelized linear algebra routines written in C++.
Meant to be a header-only library with no dependencies other than standard C++.
Tests will use googletest and benchmarks will use benchmark.
This repo requires OpenMP. As such, you need an OpenMP compliant compiler. See here for compilers that support OpenMP.
- Use C++ concepts to constrain template parameters to valid values. E.g, in order to place a type inside a matrix, you need to define addition, multiplication, subtraction, etc. for your type.
- Parallelize algorithms where possible. Lots of linear algebra algorithms are embarassingly parallel problems. Harness the simplicity of OpenMP to easily parallelize for loops, at the very least.
- Unit tests for all algorithms.
- Benchmarks for all algorithms, compared with the state of the art. I don't expect to write faster code, but I do hope to be in the same ball park, with hopefully more readable code than LAPACK.
- Easily include-able into other projects via a single (not sure yet) header.
- Simple looking code.
- Binary serialization/deserialization of matrices. This is so that data can be saved/loaded efficiently in terms of both space and time, similar to what
pickleis for Python applications.