Parallel computation DAG / Shared Futures?

It's not. It can be used on older versions of macos, but it still relies on Darwin's os_unfair_lock or OSSpinLock.
I see the Mutex implementation for Linux uses futex probably to avoid notifying the kernel when possible.
So I guess to make that solution cross-platform you'd need to reimplement these ll_ prefixed functions and make sure that the size of LowLevelLock matches the size of your underlying primitive.