Offload expose device selection

[Sa4dUs]

get amount of available devices with omp_get_num_devices and add arg to the intrinsic for device selection

r? @ZuseZ4

[bjorn3]

Does it actually need to be a rust intrinsic? Does LLVM give omp_get_num_devices any special behavior or is it just a regular runtime call that we could directly call from rust code too?

[ZuseZ4]

@bjorn3 Yes, you could call all of our openmp/offload libraries directly, especially here it's easy enough. Some of the other ones we use for data transfer and kernel launches are just a bit more involved to set up, so we do it in the compiler.

The two problems of exposing it are that you'd need to link in the GCC or LLVM libraries yourself, if you build and use the Offload setup, we do it for users. Also cranelift doesn't provide it, with an intrinsic we could have better fallback behaviour.

[rustbot]

Some changes occurred to the intrinsics. Make sure the CTFE / Miri interpreter
gets adapted for the changes, if necessary.

cc @rust-lang/miri, @RalfJung, @oli-obk, @lcnr

[ZuseZ4]

1. We should probably land the offload macro first, we're starting to have a lot of arguments which can be annoying without it. Also reduces the churn on tests.
2. Let's discuss how much effort it would be to directly account for CPU offload.

[rustbot]

This PR was rebased onto a different main commit. Here's a range-diff highlighting what actually changed.

Rebasing is a normal part of keeping PRs up to date, so no action is needed—this note is just to help reviewers.

[ZuseZ4]

The code should work on the technical side, but I feel like exposing the -1 as "any device" to the frontend is a little too low-level.
How about in the macro, we only allow device >= 0 if a user wants to target a specific GPU and use -1 as the fallback, if it's left unspecified by the user?

Also, could you add a check to make sure that if we select GPU X, that this GPU is also available? Right now I often forget to load rocm on a server, run the binary, and have it fail with some obscure llvm error since offload fails (no gpu available) but we prohibit falling back to the CPU. It would be nice to get a more meaningfull error message on the Rust side already.

[ZuseZ4]

Left just a nit, other than that, can you add the later cleanup commits into the first three? Otherwise names seem also good now, so lgtm.

[ZuseZ4]

We should later add a way to give users information about the different GPU devices. Servers just have 4/8x identical GPUs, but the majority of end-users will probably have a tiny igpu and a large dgpu.
But for now things like offloading to the CPU seem to be in higher demand, so let's do that first.

@bors r+ rollup

[rust-bors]

📌 Commit 3777642 has been approved by ZuseZ4

It is now in the queue for this repository.

🌲 The tree is currently closed for pull requests below priority 10. This pull request will be tested once the tree is reopened.

Reason for tree closure: Main branch is broken