Cranelift: load.f32x2 unimplemented on x86_64

[bjorn3]

Thanks for filing an issue! Please fill out the TODOs below.

.clif Test Case

function u0:0(f32x2) -> f32x2 system_v {
    ss0 = explicit_slot 16
    ss1 = explicit_slot 16
    sig0 = (f32) -> f32 system_v
    fn0 = u0:1 sig0 ; "cosf"

block0(v0: f32x2):
    stack_store v0, ss1

; _0 = simd_fcos::<f32x2>(move _1)
    v1 = stack_load.f32 ss1
    v2 = call fn0(v1) ; lib_call cosf
    stack_store v2, ss0
    v3 = stack_load.f32 ss1+4
    v4 = call fn0(v3) ; lib_call cosf
    stack_store v4, ss0+4

    v5 = stack_load.f32x2 ss0
    return v5
}

Steps to Reproduce

• Compile on x86_64

Expected Results

Compiles

Actual Results

thread '<unnamed>' panicked at /home/bjorn/.cargo/registry/src/index.crates.io-6f17d22bba15001f/cranelift-codegen-0.105.2/src/machinst/lower.rs:766:21:
should be implemented in ISLE: inst = `v5 = load.f32x2 notrap v11`, type = `Some(types::F32X2)`
stack backtrace:
   0: rust_begin_unwind
             at /rustc/46b180ec2452d388c5d9c14009442e2e0beb01d7/library/std/src/panicking.rs:645:5
   1: core::panicking::panic_fmt
             at /rustc/46b180ec2452d388c5d9c14009442e2e0beb01d7/library/core/src/panicking.rs:72:14
   2: cranelift_codegen::machinst::compile::compile
   3: <cranelift_codegen::isa::x64::X64Backend as cranelift_codegen::isa::TargetIsa>::compile_function
   4: cranelift_codegen::context::Context::compile_stencil
   5: cranelift_codegen::context::Context::compile_and_emit
   6: <cranelift_object::backend::ObjectModule as cranelift_module::module::Module>::define_function_with_control_plane
   7: cranelift_module::module::Module::define_function
   8: rustc_codegen_cranelift::base::compile_fn

Versions and Environment

Cranelift version or commit: Cranelift 1.105.2

Operating system: N/A

Architecture: x86_64

Extra Info

Anything else you'd like to add?

[bjorn3]

I haven't checked if store.f32x2 works. load.f32x2 does work on AArch64.

[cfallin]

Thanks for reporting! I suspect you'll find a bunch more holes with "narrow SIMD" (64-bit vectors) on x86 -- this mostly exists only on aarch64 for now. As with other missing implementations, we'd be happy to take a PR from anyone and this could make a decent starter issue.