Performance and Reproducibility Assessment of Quantum Dissipative Dynamics Framework: A Comparative Study of Fortran Compilers, MKL, and FFTW

Abstract

This paper seeks to assess the performance, energy consumption, and repeatability of the framework Quantum Dissipative Dynamics (QDD). We have observed some trouble with repeatability and reproducibility in such programs. QDD uses the Math Kernel Library (MKL) and the Fastest Fourier Transform in the West (FFTW) library to compute the Discrete Fourier Transform (DFT), in addition to the new Intel Fortran compiler compared to well-established ones, such as gfortran and the former ifort. Our findings indicate that gfortran, despite being open source, exhibits commendable performance when compared to the Intel compilers for this application. In our case, the new ifx compiler does not appear to offer significant benefits in performance over its predecessors. Additionally, our results suggest that MKL outperforms FFTW in terms of computational speed. Regarding energy consumption, there is minimal difference among the options, supporting the notion that faster execution is more energy-efficient. This paper provides performance comparisons and recommendations aimed at enhancing the repeatability and reproducibility of scientific computing experiments. In addition, we found that, by default, FFTW sometimes lacks determinism, which compromises the repeatability essential for debugging. We found a compromise between speed and reliable results for our configuration with QDD. The combination of gfortran and MKL is the one performing the best, contrary to what was expected.