Does a (standard) atomic wide pointer implementation exist?

[Dante-Broggi]

The ABA Problem can cause issues and is often resolved using wide atomic pointers. As Rust supports wide pointers for raw slices and trait objects, I thus ask:
Does a (standard) atomic wide (either pointer+usize or pointer+pointer) pointer implementation exist in the Rust community or stdlib?

[mdHMUpeyf8yluPfXI]

x86 doesn't have 128-bit atomics. There is a workaround by using AtomicPtr<Box<dyn Trait>>.

Version of AtomicPtr that can handle Unsized boxed Types

I am trying to consolidate my memory allocations but for that I need unsized types (aka FatPointers). Because they are unsized I need to Box them. Is there a Version of Box that stores the size at the allocation so that when I convert the Box into a raw Pointer I get a regular slim pointer that can be use with Atomic CAS and later restore the Box using the size stored at the allocation?

[SkiFire13]

@mdHMUpeyf8yluPfXI

x86 doesn't have 128-bit atomics.

This is not completly true. There are extensions that support them, for example CMPXCHG16B, which is for example required for running Windows 10.

Edit: there's core::arch::x86_64::cmpxchg16b and the portable-atomic crate for a wrapper over it.

@mdHMUpeyf8yluPfXI

There is a workaround by using AtomicPtr<Box<dyn Trait>>.

This is gonna be a pointer to a wide pointer to the actual data, which won't solve the problem.

[jseigh]

Probably not since the standard atomics are based on C++20 which doesn't support double wide compare and swap. x86-64 doesn't have atomic 128 bit loads and stores, not efficient ones anyway. So it doesn't fit into their standard notion of what methods atomics should support. This is unfortunate since some very basic lock-free data structures require this or a load-locked / store-conditional. Also you can't do things like AARC (actually atomic reference counting) which doesn't require ownership to copy or dereference. There's no technical reason you couldn't provide a special api for those cases but there you have it.

I once did a pre-C++11 AARC implementation once on x86-64 and powerpc with inline assembler where needed. I looked into porting to standard C++ atomics and gave up since double wide compare and swap wasn't supported. Not really a loss since reference counting is really non-performant. There are much better ways than using reference counting for shared storage management.

Joe Seigh

[ryanavella]

@SkiFire13

There are extensions that support them, for example CMPXCHG16B, which is for example required for running Windows 10.

And Windows 8 as well I believe. So if we ever drop support for Windows 7, all supported Windows targets would ostensibly have it.