Cranelift: x64 backend crashes compiling a standalone iadd_pairwise

[afonso360]

Hey, In #3272 @dheaton-arm implemented tests for a standalone iadd_pairwise instruction without a swiden_low/swiden_high/uwiden_low/uwiden_high instruction as an input. It looks like this case is not implemented in the x64 backend, instead we always perform an optimization and merge those previous instructions.

.clif Test Case

test run
target x86_64 machinst

function %iaddp_i8x16(i8x16, i8x16) -> i8x16 {
block0(v0: i8x16, v1: i8x16):
    v2 = iadd_pairwise v0, v1
    return v2
}
; run: %iaddp_i8x16([1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16], [17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32]) == [3 7 11 15 19 23 27 31 35 39 43 47 51 55 59 63]

function %iaddp_i16x8(i16x8, i16x8) -> i16x8 {
block0(v0: i16x8, v1: i16x8):
    v2 = iadd_pairwise v0, v1
    return v2
}
; run: %iaddp_i16x8([1 2 3 4 5 6 7 8], [100 99 98 97 96 95 94 93]) == [3 7 11 15 199 195 191 187]

function %iaddp_i32x4(i32x4, i32x4) -> i32x4 {
block0(v0: i32x4, v1: i32x4):
    v2 = iadd_pairwise v0, v1
    return v2
}
; run: %iaddp_i32x4([1 2 3 4], [5 6 7 8]) == [3 7 11 15]
; run: %iaddp_i32x4([4294967290 5 4294967290 5], [100 100 100 100]) == [4294967295 4294967295 200 200]

Steps to Reproduce

Run clif-util the-above.clif

Expected Results

Tests to pass

Actual Results

Cranelift crashes with:

thread 'worker #0' panicked at 'not implemented: Operands not supported for IaddPairwise', cranelift\codegen\src\isa\x64\lower.rs:5095:17
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace
FAIL .\filetests\filetests\runtests\test.clif: panicked in worker #0: not implemented: Operands not supported for IaddPairwise

Versions and Environment

Cranelift version or commit: main
Operating system: Windows
Architecture: x64

Extra Info

The test file above should be in-tree at cranelift/filetests/filetests/runtests/simd-iaddpairwise.clif, once #3272 is merged.

[cfallin]

cc @jlb6740 -- we discussed at the time (in #3209) that it was fine to merge without this more general case in order to make forward progress, but that we should eventually fill out the general case. Thanks for creating the issue to track this!

[alexcrichton]

This was implemented in #6561 for the x64 backend, with tests enabled there, so I'm going to close this