Indexing Vectors with Vec{N,Int32} generates invalid LLVM assembly

[favba]

MWE:

julia> i = Vec(map(Int32,(1,2,3,4))...)
<4 x Int32>[1, 2, 3, 4]

julia> a = rand(4)
4-element Vector{Float64}:
 0.5626535341100131
 0.7054219085896695
 0.6418954581789188
 0.693063317804356

julia> a[i]
ERROR: Failed to parse LLVM assembly: 
<string>:3:14: error: invalid cast opcode for cast from '<4 x i32>' to '<4 x i64>'
%2 = bitcast <4 x i32> %0 to <4 x i64>
             ^

Stacktrace:
 [1] macro expansion
   @ ~/.julia/packages/SIMD/0q83J/src/LLVM_intrinsics.jl:770 [inlined]
 [2] inttoptr
   @ ~/.julia/packages/SIMD/0q83J/src/LLVM_intrinsics.jl:762 [inlined]
 [3] convert
   @ ~/.julia/packages/SIMD/0q83J/src/simdvec.jl:27 [inlined]
 [4] +
   @ ~/.julia/packages/SIMD/0q83J/src/simdvec.jl:315 [inlined]
 [5] getindex(::Vector{Float64}, ::Vec{4, Int32})
   @ SIMD ~/.julia/packages/SIMD/0q83J/src/arrayops.jl:317
 [6] top-level scope
   @ REPL[21]:1

Explicitly converting to Int64 works as expected:

julia> a[convert(Vec{4,Int64}, i)]
<4 x Float64>[0.5626535341100131, 0.7054219085896695, 0.6418954581789188, 0.693063317804356]

I expected this conversion to be done automatically by the Base.getindex method defined in SIMD.jl