wasmtime: Implement global.{get,set} for externref globals

[fitzgen]

We use libcalls to implement these -- unlike table.{get,set}, for which we
create inline JIT fast paths -- because no known toolchain actually uses
externref globals.

Part of #929

[github-actions]

Subscribe to Label Action

cc @peterhuene

Details

This issue or pull request has been labeled: "wasmtime:api"

Thus the following users have been cc'd because of the following labels:

• peterhuene: wasmtime:api

To subscribe or unsubscribe from this label, edit the .github/subscribe-to-label.json configuration file.

Learn more.

[alexcrichton]

Nice, and even more tests passing!

I think it's fine that these aren't fast right-out-the-gate, but I wonder if perhaps a little refactoring in the code generator could make this PR even smaller and also faster at the same time? It seems that table.get and global.get are basically the same codegen after they determine the address they're loading from. Similarly table.set and global.set are the same once they've got the address.

Do you think it'd be possible to refactor the translation of table.set and table.get to be shared with these instructions? Ideally just a small portion of it would be chopped off to go elsewhere (which would take in an address) and then the instructions would all otherwise codegen exactly the same way.

If this looks harder than I expect though it seems fine to have out-of-line intrinsics for now and optimize later as necessary.

Other than that, could you also add some API tests which exercises get/set for funcref/anyref globals? It looks like there's get for externref but at the wasmtime API layer I don't think there's much other coverage?

[fitzgen]

My thinking was that, because no toolchains emit externref globals, it wasn't worth the effort to make the inline JIT fast path generation code more generic. This would also require some cranelift-wasm interface changes.

Do you think it is worth the effort?

[alexcrichton]

Nah I don't think it's worth it if it requires larger changes. I just thought that it might be simple to implement today if the code was extracted from table.{get,set}, but if that's not the case then this seems fine to me!

[fitzgen]

@alexcrichton good call on adding tests for the global API with funcref and externref, that actually required a bit of plumbing in the trampoline module. Do you want to take a look at the new commit? Thanks!

[alexcrichton]

👍