Vector sum v2

// 1.
static float4 operator+(const float4& A, const float4& B)
{
    float4 C;
    auto a = _mm_load_ps(&A.x);
    auto b = _mm_load_ps(&B.x);
    _mm_store_ps(&C.x, _mm_add_ps(a, b));
    return C;
}

//2.
static void summ(const float4* A, const float4* B, float4* C)
{
    __asm
    {
        lea eax, A;
        lea ebx, B;

        mov eax, [eax];
        mov ebx, [ebx];

        movaps xmm0, [eax];
        movaps xmm1, [ebx];
        addps xmm0, xmm1;
        movaps[edx], xmm0;

        mov C, edx;
    }
}

//3.
static void ssumm(const float4* A, const float4* B, float4* C)
{
    C->x = A->x + B->x;
    C->y = A->y + B->y;
    C->z = A->z + B->z;
    C->w = A->w + B->w;
}