Uncomment wasm32 bitmask instruction

[Daniel-Liu-c0deb0t]

#874 introduced some iNNxMM_bitmask instructions that were supposed to be supported in LLVM 11, but they were commented out. Since Rust is already on LLVM 11 (https://blog.rust-lang.org/2020/10/08/Rust-1.47.html), this PR uncomments those instructions so they can be used.

[rust-highfive]

r? @Amanieu

(rust-highfive has picked a reviewer for you, use r? to override)

[Amanieu]

cc @alexcrichton: This looks good to me but I want to run it by you first since it's WASM-related.

[Daniel-Liu-c0deb0t]

Saw some other commented out lines for the abs instructions and i64x2_mul, so I uncommented those as well.

[Daniel-Liu-c0deb0t]

The tests for WASM don't seem to be running. Is there a problem with them?

[Amanieu]

I believe it was commented it in the CI configuration because it was failing and never got re-enabled. You can try re-enabling it.

[Daniel-Liu-c0deb0t]

I'll have to play around with enabling and updating the CI config and I'll likely open another PR for that.

Edit: PR at #1006
If wasmtime is not able to handle the new WASM SIMD instructions, then we should try wasmer.

[Amanieu]

Can you rebase to see if it passes WASM CI?

[alexcrichton]

Looks great to me!

FWIW wasm simd and LLVM are still sort of constantly moving targets. The simd proposal for wasm is nearing the next phase of stabilization, however, so I hope that it will settle down in the near future. Once everything settles down again on the wasm side I plan to make another pass through this module so sync with the current state of the spec (which has renamed, added, and tweaked various instructions)

For now though looks great!

[Daniel-Liu-c0deb0t]

The reason why I want bitmask right now is because I want to use it in a project I'm working on.

Running the CI (wasmtime) results in weird cranelift errors that look like:

thread '<unnamed>' panicked at 'ABI argument v4 should be in a register', cranelift/codegen/src/regalloc/coloring.rs:409:21

Not sure what to do, as I'm not familiar with this.

Edit: tried wasmer, but it isn't up to date as the bitmask operations aren't supported. Latest cranelift version seems to support them, though (https://github.com/bytecodealliance/wasmtime/blob/09b976e1d53a05150c7b8acd36a82b34cec787b3/cranelift/wasm/src/code_translator.rs#L1639), so I'm back to trying wasmtime. Regardless, i64x2_bitmask doesn't seem to be supported, so I commented it out.

[Daniel-Liu-c0deb0t]

Running tests with the latest wasmtime dev release results in these errors:

thread '<unnamed>' panicked at 'ABI argument v5 should be in a register', cranelift/codegen/src/regalloc/coloring.rs:409:21
thread '<unnamed>' panicked at 'not implemented: Currently only SIMD instructions are mapped to their return type; the following instruction is not mapped: V128AnyTrue', cranelift/wasm/src/code_translator.rs:2688:14

@alexcrichton any idea what is going on?

[alexcrichton]

The SIMD support in Wasmtime is still experimental and not fully fleshed out, so you're probably hitting a bug of a case that isn't implemented yet or just a bug in the implementation. If possible could you minimize the case and report it to rust-lang/rust?

[Daniel-Liu-c0deb0t]

Another potential thing that should be fixed: missing target feature attributes for some WASM SIMD functions, which affects inlining. Though this should be less of a problem once rust-lang/rust#74320 is fixed.

[bors]

☔ The latest upstream changes (presumably 8724eb4) made this pull request unmergeable. Please resolve the merge conflicts.

[alexcrichton]

I believe that wasm simd stuff is all up to date by this point, so I think this can be closed now?