Johannes Demel

If we provide a `deinit` function, we should also spend some thought on how to make initialization thread safe. Unless, we decide that this is a user problem and VOLK...

The windows build failed. I was too optimistic about C11 support in MSVC in 2021... `stdatomic.h` is not available.

> In free, you also just read the atomic variable. Shouldn't this be an atomic set to make sure that only one thread frees? Actually, I tried to use `threads.h`...

> Since you do not want to introduce library init calls, the only option is probably thread-local storage. You do not want to acquire mutexes or read atomic variables on...

> > Fixes #440 > > We should distinguish between the two different issues (albeit their resolution is somewhat linked): > > 1. #440 is about a “memory leak” which...

I think I was able to resolve all those thread safety concern and add the option to release the struct. Though, I use `threads.h` to get a mutex. This seems...

We start our CMakeLists.txt with ``` set(CMAKE_C_STANDARD 11) ``` I'd expect that this makes compilers without C11 standard available fail. I stumbled over a code section in VOLK that checks...

> I only have a rough understanding, but is this assignment to the global variable (i.e., the global function pointer) synchronized? No, this assignment is unsynchronized. It follows an "ancient"...

Where should this be documented? Maybe in the project README?

Our current structure looks like this - apps - Python and C++ - cmake - CMake stuff obviously - docs - Doxygen - gen - Python: Build system - include/volk...