Improving performance of floating-point operations with SIMD

Question

I have this function to compute a piece of array of double\'s:

void avx2_mul_64_block(double& sum, double* lhs, double* rhs) noexcept
{
    __m256i accumulato