__m128 aa32 = _mm_set_ps(0.f, 1.f, 2.f, 3.f);
__m128 bb32 = _mm_set_ps(5.f, 6.f, 7.f, 8.f);
__m128 cc32 = _mm_shuffle_ps(aa32, bb32, _MM_SHUFFLE(0, 1, 2, 3));

__m256d aa64 = _mm256_set_pd(0.0, 1.0, 2.0, 3.0);
__m256d bb64 = _mm256_set_pd(5.0, 6.0, 7.0, 8.0);
__m256d cc64 = _mm256_shuffle_pd(aa64, bb64, _MM_SHUFFLE(0, 1, 2, 3));


cc32 = { 0.00000000, 1.00000000, 7.00000000, 8.00000000 }
cc64 = { 2.0000000000000000, 7.0000000000000000, 1.0000000000000000, 5.0000000000000000	 }