Thanks! I merged and updated the new Fortran data into the notebook plots. With actual large-Nx data, the Julia unrolled code is in a dead heat with Fortran. I also fixed up your two other issues. BTW, your icon looks an awful lot like a modified Larks’ Tongues in Aspic album cover. King Crimson fan?
That’d be great, please do. I have the benchmark algorithm with real-to-complex FFTs implemented in C++ and Fortran already. Both are pretty simple modifications. Python can do it with Numpy, though I haven’t coded it yet. Not sure about Matlab. Showing the source modifications and timings for this in all languages will be another good comparison.
Thanks, I will look this up. My fluids code (www.channelflow.org) implements a number of higher-order methods in a modular way; probably the best is 4th-order semi-implicit backwards differentiation. I’ve been meaning to learn more about exponential integrators. For this benchmark I did CNAB2 to avoid coding more complex methods from scratch in N languages, especially the painful languages. But it could be a good illustration of how Julia makes complex algorithms much easier to code.