Vectorized dot product of float arrays in C# and C++

vdp.asm

; C++ core loop, key intrinsic is _mm_dp_ps
00007fff`548b10d5 0f100c0a        movups  xmm1,xmmword ptr [rdx+rcx]  ; LOOP
00007fff`548b10d9 0f1011          movups  xmm2,xmmword ptr [rcx]
00007fff`548b10dc 4883c110        add     rcx,10h
00007fff`548b10e0 660f3a40d1f1    dpps    xmm2,xmm1,0F1h
00007fff`548b10e6 f30f58c2        addss   xmm0,xmm2
00007fff`548b10ea 4983e801        sub     r8,1
00007fff`548b10ee 75e5            jne     00007fff`548b10d5  ; LOOP

; C# core loop, key intrinsic is Vector.Dot(Vector<float>, Vector<float>);
; runs 70% slower than C++ version
00007ffe`fc663686 448d5003        lea     r10d,[rax+3]	; LOOP
00007ffe`fc66368a 453bd0          cmp     r10d,r8d
00007ffe`fc66368d 7337            jae     RANGE_FAIL
00007ffe`fc66368f 0f104c8110      movups  xmm1,xmmword ptr [rcx+rax*4+10h]
00007ffe`fc663694 453bd1          cmp     r10d,r9d
00007ffe`fc663697 732d            jae     RANGE_FAIL
00007ffe`fc663699 0f10548210      movups  xmm2,xmmword ptr [rdx+rax*4+10h]
00007ffe`fc66369e 0f59ca          mulps   xmm1,xmm2
00007ffe`fc6636a1 0f28d9          movaps  xmm3,xmm1
00007ffe`fc6636a4 0fc6dbb1        shufps  xmm3,xmm3,0B1h
00007ffe`fc6636a8 0f58cb          addps   xmm1,xmm3
00007ffe`fc6636ab 0f28d9          movaps  xmm3,xmm1
00007ffe`fc6636ae 0fc6db1b        shufps  xmm3,xmm3,1Bh
00007ffe`fc6636b2 0f58cb          addps   xmm1,xmm3
00007ffe`fc6636b5 f30f58c1        addss   xmm0,xmm1
00007ffe`fc6636b9 83c004          add     eax,4
00007ffe`fc6636bc 443bc0          cmp     r8d,eax
00007ffe`fc6636bf 7fc5            jg      00007ffe`fc663686  ; LOOP

vdp.cpp

extern "C" __declspec(dllexport) float vectorized_native_same_type_float(float* a, float* b, int length)
{
	float aux = 0;
	for (int i = 0; i < length; i += 4)
	{
		__m128 va = _mm_loadu_ps(a + i);
		__m128 vb = _mm_loadu_ps(b + i);
		__m128 dp = _mm_dp_ps(va, vb, 0xF1);
		aux += dp.m128_f32[0];
	}
	return aux;
}

vdp.cs

static float VectorizedSameTypeFloat(float[] a, float[] b)
{
    float aux = 0;
    int vecSize = Vector<float>.Count;
    for (int i = 0; i < a.Length; i += vecSize)
    {
        Vector<float> va = new Vector<float>(a, i);
        Vector<float> vb = new Vector<float>(b, i);

        aux += Vector.Dot(va, vb);
    }
    return aux;
}