simd_op_check used to be a single monolithic test, which decided which architecture to run based on the value of HL_TARGET. We recently split the tests up into architecture-specific variants, but the logic for the right host config is still there, meaning that running (say) simd_op_check_arm on x86 does... nothing at all.

The reasonable default should probably instead be to test one (or more) likely targets for that test (eg perhaps x86 should test several simd variants), and just skip the jit testing: it is still valuable to verify that the expected instructions are being generated.