C++ Expression Templates: Eigen, Blaze, Fastor, Armadillo & XTensor (2020)

Good article. Next time I ever want to revisit using C++ for scientific calculation again I'll keep in mind that Fastor has the fastest compilation time of all. Still, 2 seconds for just a benchmark code is a bit too much.

I remember doing this kind of finite difference stencil templating using plain C++. All the performance gains went out of the window as soon as the array size becomes dynamic.

> We now also have explicit SIMD vectorisation.. So say on an AVX512 enabled architecture you can compute the dot product of two vectors

I wonder will it be good? Computing dot product efficiently is harder than it seems, example: https://stackoverflow.com/a/59495197/126995 Eigen usually does stuff like that rather well.

Bonus points for measuring the compile times. Unfortunately matrix libraries are the types of things that really challenge C++, and the compiler devs haven't been able to bring the compile times down to reasonable levels yet.

It's really unfortunate that C++ keeps adding more functionality that makes one of its critical problems (horrible syntax debugging) worse. All of these libraries generate atrocious compiler errors for even trivial typos.

Compile times like these are why I avoid C++ for numerical programming these days.