Translation of flexible-precision algorithms to production-ready C++

Determine the extent to which algorithms developed in symbolic high-level environments such as Matlab or NumPy, which support generic flexible precisions, can be translated one-to-one into production-ready C++ implementations without loss of correctness or performance.

Background

The paper notes that high-level environments (e.g., Matlab and NumPy) support flexible precisions enabling rapid methodological development. However, production HPC codes are typically written in C++ with fixed hardware-supported types, and template metaprogramming or bespoke libraries introduce practical complications.

The authors highlight uncertainty about how well such high-level, flexible-precision algorithms carry over to C++, motivating their proposed C++ annotations intended to ease mixed-precision and memory-centric development.

References

However, it remains unclear to which degree the developed algorithms translate one-to-one into production-ready C++ code.

— An extension of C++ with memory-centric specifications for HPC to reduce memory footprints and streamline MPI development (2406.06095 - Radtke et al., 10 Jun 2024) in Section 1 (Introduction)