Getting SIMD load/store instructions through llvmcall

TestSIMDrmw.jl

module TestSIMD5

typealias Uint64x2 NTuple{2, Uint64}
function ($)(x::Uint64x2, y::Uint64x2)
    Base.llvmcall("""%3 = xor <2 x i64> %1, %0
                     ret <2 x i64> %3""",
                  Uint64x2, (Uint64x2, Uint64x2), x, y)
end

unsafe_aligned_load(p::Ptr{Uint64}, i::Int) = unsafe_aligned_load(ptradd(p, (i-1)*sizeof(Uint64)*2))
function unsafe_aligned_load_2x(p::Ptr{Uint64})
    Base.llvmcall("""%2 = bitcast i64* %0 to <2 x i64>*
    				 %3 = load <2 x i64>* %2, align 16
                     ret <2 x i64> %3""",
                  Uint64x2, (Ptr{Uint64},), p)
end

unsafe_aligned_store!(p::Ptr{Uint64}, x::Uint64x2, i::Int) = unsafe_aligned_store!(ptradd(p, (i-1)*sizeof(Uint64)*2), x)
function unsafe_aligned_store_2x!(p::Ptr{Uint64}, x::Uint64x2)
    Base.llvmcall("""%3 = bitcast i64* %0 to <2 x i64>*
    				 store <2 x i64> %1, <2 x i64>* %3, align 16
                     ret void""",
                  Void, (Ptr{Uint64}, Uint64x2), p, x)
end

function rmw!{T}(dest::Ptr{T}, src::Ptr{T})
	s = unsafe_aligned_load_2x(src)
	d = unsafe_aligned_load_2x(dest)
    d $= s
    unsafe_aligned_store_2x!(dest, d)
end

code_native(rmw!, (Ptr{Uint64}, Ptr{Uint64}))

A, B = Uint64[1,2], Uint64[0x30,0x40]
rmw!(pointer(A), pointer(B))
@show A

end # module

I get the output

Source line: 30
    push    RBP
    mov RBP, RSP
    movaps  XMM0, XMMWORD PTR [RSI]
    xorps   XMM0, XMMWORD PTR [RDI]
    movaps  XMMWORD PTR [RDI], XMM0
Source line: 30
    pop RBP
    ret
A = Uint64[0x0000000000000031,0x0000000000000042]

so it seems to be working!