wasmtime / PR #12222 Partial support for no_std in cranel... · git-wasmtime

Stream: git-wasmtime

Topic: wasmtime / PR #12222 Partial support for no_std in cranel...

Wasmtime GitHub notifications bot (Jan 05 2026 at 23:00):

the-ssd edited PR #12222.

Wasmtime GitHub notifications bot (Jan 05 2026 at 23:00):

the-ssd created PR review comment:

https://github.com/rust-lang/compiler-builtins/pull/935

Wasmtime GitHub notifications bot (Jan 05 2026 at 23:00):

the-ssd submitted PR review.

Wasmtime GitHub notifications bot (Jan 05 2026 at 23:01):

the-ssd edited PR review comment.

Wasmtime GitHub notifications bot (Jan 05 2026 at 23:01):

the-ssd edited PR #12222:

Closes #1158, allows for cranelift to be used in a no_std environment, like a kernel.

There is already #9007, which stalled.

And I am at a point I think it makes sense to ask someone what do you want to be done about FxHashMap and OnceLock. There are alternative crates that implement it, like this one (also has a comparison chart).

What is left to do:

[x] isle codegen generates code that depends on no_std, solved in the previous PR (https://github.com/bytecodealliance/wasmtime/pull/9007/files#diff-0496ab91e0634653bd3d4295c7aa90e3f06b0c5f462332db0997a22f294cbf94)

[x] separate PRs for: cranelift_assembler_x86 (#12235), moving large parts to core and alloc (#12237), and isle (#12236)

[ ] floating point issues

[x] FxHashMap

[ ] OnceLock (sort of)

[x] CI

[ ] libm: roundeven has a typo in it's implementation https://github.com/rust-lang/compiler-builtins/pull/1053, so a hack is used, waiting for libm to be published to crates.io

Wasmtime GitHub notifications bot (Jan 05 2026 at 23:01):

the-ssd edited PR #12222:

Closes #1158, allows for cranelift to be used in a no_std environment, like a kernel.

There is already #9007, which stalled.

And I am at a point I think it makes sense to ask someone what do you want to be done about FxHashMap and OnceLock. There are alternative crates that implement it, like this one (also has a comparison chart).

What is left to do:

[x] isle codegen generates code that depends on no_std, solved in the previous PR (https://github.com/bytecodealliance/wasmtime/pull/9007/files#diff-0496ab91e0634653bd3d4295c7aa90e3f06b0c5f462332db0997a22f294cbf94)

[x] separate PRs for: cranelift_assembler_x86 (#12235), moving large parts to core and alloc (#12237), and isle (#12236)

[x] FxHashMap

[ ] floating point issues

[ ] OnceLock (sort of)

[x] CI

[ ] libm: roundeven has a typo in it's implementation https://github.com/rust-lang/compiler-builtins/pull/1053, so a hack is used, waiting for libm to be published to crates.io

Wasmtime GitHub notifications bot (Jan 05 2026 at 23:01):

the-ssd edited PR #12222:

Closes #1158, allows for cranelift to be used in a no_std environment, like a kernel.

There is already #9007, which stalled.

And I am at a point I think it makes sense to ask someone what do you want to be done about FxHashMap and OnceLock. There are alternative crates that implement it, like this one (also has a comparison chart).

What is left to do:

[x] isle codegen generates code that depends on no_std, solved in the previous PR (https://github.com/bytecodealliance/wasmtime/pull/9007/files#diff-0496ab91e0634653bd3d4295c7aa90e3f06b0c5f462332db0997a22f294cbf94)

[x] separate PRs for: cranelift_assembler_x86 (#12235), moving large parts to core and alloc (#12237), and isle (#12236)

[x] FxHashMap

[ ] floating point issues

[ ] OnceLock (sort of)

[ ] CI

[ ] libm: roundeven has a typo in it's implementation https://github.com/rust-lang/compiler-builtins/pull/1053, so a hack is used, waiting for libm to be published to crates.io

Wasmtime GitHub notifications bot (Jan 05 2026 at 23:04):

the-ssd edited PR #12222:

Used to close #1158, allows for cranelift to be used in a no_std environment, like a kernel.

There is already #9007, which stalled.

And I am at a point I think it makes sense to ask someone what do you want to be done about FxHashMap and OnceLock. There are alternative crates that implement it, like this one (also has a comparison chart).

What is left to do:

[x] isle codegen generates code that depends on no_std, solved in the previous PR (https://github.com/bytecodealliance/wasmtime/pull/9007/files#diff-0496ab91e0634653bd3d4295c7aa90e3f06b0c5f462332db0997a22f294cbf94)

[x] separate PRs for: cranelift_assembler_x86 (#12235), moving large parts to core and alloc (#12237), and isle (#12236)

[x] FxHashMap

[ ] floating point issues

[ ] OnceLock (sort of)

[ ] CI

[ ] libm: roundeven has a typo in it's implementation https://github.com/rust-lang/compiler-builtins/pull/1053, so a hack is used, waiting for libm to be published to crates.io

Wasmtime GitHub notifications bot (Jan 05 2026 at 23:31):

alexcrichton submitted PR review.

Wasmtime GitHub notifications bot (Jan 05 2026 at 23:31):

alexcrichton created PR review comment:

As a heads up, I've sent https://github.com/bytecodealliance/regalloc2/pull/250 which would obsolete this, but this is fine to leave as-is in the meantime.

Wasmtime GitHub notifications bot (Jan 05 2026 at 23:32):

alexcrichton submitted PR review.

Wasmtime GitHub notifications bot (Jan 05 2026 at 23:32):

alexcrichton created PR review comment:

cc @cfallin should this perhaps continue to skip this value? Or panic! since this should be unreachable? Or is 0 ok to leave in?

Wasmtime GitHub notifications bot (Jan 05 2026 at 23:33):

alexcrichton submitted PR review.

Wasmtime GitHub notifications bot (Jan 05 2026 at 23:33):

alexcrichton created PR review comment:

We unfortunately can't use git deps in Wasmtime for published crates, and we can't land dependencies on internal features, even if temporary, in case that ends up accidentally making its way to a release branch. Leaving as-is is ok for now, but it might be possible to do some macro/float trickery to avoid blocking on a compiler-builtins update for this

Wasmtime GitHub notifications bot (Jan 06 2026 at 13:14):

cfallin submitted PR review.

Wasmtime GitHub notifications bot (Jan 06 2026 at 13:14):

cfallin created PR review comment:

Yes, if we don't have unwind then we don't have the appropriate metadata to allow a native debugger to interpret frames anyway, so we should continue. I'm a bit curious how this built in a non-unwind build before -- or was that broken because std always enabled unwind?

Wasmtime GitHub notifications bot (Jan 06 2026 at 13:15):

the-ssd submitted PR review.

Wasmtime GitHub notifications bot (Jan 06 2026 at 13:15):

the-ssd created PR review comment:

It was broken

Wasmtime GitHub notifications bot (Jan 07 2026 at 09:39):

cfallin submitted PR review.

Wasmtime GitHub notifications bot (Jan 07 2026 at 09:39):

cfallin created PR review comment:

@the-ssd Updating to version = "0.13.4" should now pull in Alex's change.

Wasmtime GitHub notifications bot (Jan 07 2026 at 09:40):

cfallin created PR review comment:

@the-ssd OK, yep, let's do a continue in this case.

Wasmtime GitHub notifications bot (Jan 07 2026 at 09:40):

cfallin submitted PR review.

Wasmtime GitHub notifications bot (Jan 07 2026 at 09:51):

cfallin submitted PR review.

Wasmtime GitHub notifications bot (Jan 07 2026 at 09:51):

cfallin created PR review comment:

@the-ssd if you're willing to do a little more hacking, jameysharp's idea here is probably the best way to remove the need to cache the MachineEnv at all (in a separate set of PRs of course). It would involve a change to RA2's API then an update to Cranelift to use statically-defined MachineEnvs. I have specific ideas on how to consume a PRegSet for the register iterator on the RA2 side if you're interested -- I'll file an issue over there.

Wasmtime GitHub notifications bot (Jan 07 2026 at 10:00):

cfallin submitted PR review.

Wasmtime GitHub notifications bot (Jan 07 2026 at 10:00):

cfallin created PR review comment:

I wrote up bytecodealliance/regalloc2#252 for this.

Wasmtime GitHub notifications bot (Jan 07 2026 at 14:00):

the-ssd edited PR review comment.

Wasmtime GitHub notifications bot (Jan 07 2026 at 14:03):

the-ssd updated PR #12222.

Wasmtime GitHub notifications bot (Jan 07 2026 at 14:03):

the-ssd submitted PR review.

Wasmtime GitHub notifications bot (Jan 07 2026 at 14:03):

the-ssd created PR review comment:

updated

Wasmtime GitHub notifications bot (Jan 07 2026 at 14:06):

the-ssd submitted PR review.

Wasmtime GitHub notifications bot (Jan 07 2026 at 14:06):

the-ssd created PR review comment:

I've removed that feature and instead wrote this

// TODO: when libm v0.2.16 is published line below can be used instead
// Self::with_float(Libm::<$float_ty>::roundeven(self.$as_float()))
Self::with_float(self.$as_float().round_ties_even())

Wasmtime GitHub notifications bot (Jan 07 2026 at 14:06):

the-ssd updated PR #12222.

Wasmtime GitHub notifications bot (Jan 07 2026 at 14:14):

the-ssd updated PR #12222.

Wasmtime GitHub notifications bot (Jan 07 2026 at 14:21):

the-ssd submitted PR review.

Wasmtime GitHub notifications bot (Jan 07 2026 at 14:21):

the-ssd created PR review comment:

Can you check if I did it correctly?
It just looks a bit wrong, like there has to be a better solution

Wasmtime GitHub notifications bot (Jan 07 2026 at 16:26):

alexcrichton updated PR #12222.

Wasmtime GitHub notifications bot (Jan 07 2026 at 16:26):

alexcrichton created PR review comment:

Looks good to me!

Wasmtime GitHub notifications bot (Jan 07 2026 at 16:26):

alexcrichton submitted PR review.

Wasmtime GitHub notifications bot (Jan 07 2026 at 16:26):

alexcrichton has enabled auto merge for PR #12222.

Wasmtime GitHub notifications bot (Jan 07 2026 at 17:03):

alexcrichton merged PR #12222.

Last updated: Jan 09 2026 at 13:15 UTC