I'm looking for a C++ (possibly C) library for performing computationally-intensive linear algebra operations on massive floating-point data - from the simplest computation of the norm of a large vector to matrix multiplication and inversion, linear equation solving etc. - but the twist is that it has to produce results with low errors due to the limited precision of floating-point representation.

So, specifically, I would expect use of Kahan and pairwise summation and other such methods (or "tricks").

Requirements: Libre license, open source, gratis, has few requirements of its own, usable on Linux.

Nice-to-have features: Geared towards modern C++ (C++11 and up), has primitives separate from the more heavyweight algorithms.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.