[llvm] [Support/rpmalloc] Updated fake atomics with Interlocked functions (PR #148303)

[Alexandre Ganea via llvm-commits]

aganea wrote:

> All `Atomic access abstraction` related code has been removed from the upstream rpmalloc 2 years ago. The initial rpmalloc code in LLVM (since 2024) contained `Atomic access abstraction`. So these implementations were out of sync from the beginning.

Actually the rpmalloc fork in LLVM is [version 1.4.5](https://github.com/mjansson/rpmalloc/releases/tag/1.4.5) verbatim. What you might looking at is the develop branch which indeed had the "Atomic access abstraction" part removed from the code in this commit https://github.com/mjansson/rpmalloc/commit/ed405164f5a72fafe7f8f2e17952894686b9f9b0.

> 
> > Send a PR with your changes to https://github.com/mjansson/rpmalloc/pulls
> > Once that is approved and merged, we integrate the latest rpmalloc in LLVM.
> > Are you able to do that first step?
> 
> rpmalloc upstream does not have this problem any more, thus the fix is not needed there. We need it here till we will sync with rpmalloc upstream when LLVM will require Visual Studio 2022.

@mjansson Is there a possibility that you release a new version soon with the contents of the develop branch? There haven't been any releases for the past year. We're trying to see what avenues we should take with the change proposed in this PR. Ideally I'd like not to diverge from your repo. One option would be to integrate into LLVM [develop](https://github.com/mjansson/rpmalloc/tree/develop) at head.

https://github.com/llvm/llvm-project/pull/148303