I am hoping for portable SIMD so much. But I still think that often a manually rolled SIMD will be faster.

Also the state of SIMD in Cranelift is also very WIP. They pretty much just support a subset of 128bit vectors with some rare exceptions.