List in C/C++ backend

[czgdp1807]

Implementing 2.ii in #1077 (comment).

Closes #1077

Benchmark - #857 (comment) (see list04 row). Our generated C code (with no optimisation flags enabled for C compiler) is faster than un-optimised (no -O3 flag) executable generated clang++.

[czgdp1807]

@certik This is ready to review. Feel free to review the diff in total and let me know if any change is to be made. Otherwise I will clean up the git history and merge.

[certik]

Thanks for implementing it!

I think this is fine to merge.

As an improvement in subsequent PRs, the C++ version should just use std::vector for lists, as that is the natural data structure that people would use in C++, pretty much equivalent to our list in ASR.

[czgdp1807]

As an improvement in subsequent PRs, the C++ version should just use std::vector for lists, as that is the natural data structure that people would use in C++, pretty much equivalent to our list in ASR.

I think we should expose this with something like --use-stl for C++ backend. The scene is that our implementation is very competitive and light weight (no templates because we already know the types so we directly create the correctly typed lists and hence we don't depend on templating support of C++). So, it should be left to the user to decide whether they want to use our implementations for container data structures or STL implementations.

[certik]

Yes, making it configurable is even better.

[certik]

How do the list benchmarks behave? Are we now getting the same performance in C/C++ as in LLVM, and if not, why not?

[czgdp1807]

I don’t know how to enable -O3 optimisation for C compiler in integration tests. Please let me know and I will share the benchmark data with you.

[certik]

I would execute the benchmarks by hand. Just compile it by hand and enable all optimization options.