SSE Sort

/*
Sort 4 floats in SSE vector using sorting network and return indices of moved values
*/
inline __m128i v4_sort(__m128& v)
{
	auto i = _mm_castsi128_ps(_mm_set_epi32(3, 2, 1, 0));
	const auto mask = _mm_castsi128_ps(_mm_set1_epi32(0xFFFFFFFC));

	// Place indices in lower 2 bits of mantissa
	v = _mm_or_ps(_mm_and_ps(v, mask), i);

	// Simple sorting network for n=4
	// First pass
	auto temp = _mm_shuffle_ps(v, v, _MM_SHUFFLE(1, 0, 3, 2));
	auto cmp = _mm_cmplt_ps(v, temp);

	cmp = _mm_shuffle_ps(cmp, cmp, _MM_SHUFFLE(1, 0, 1, 0));
	auto temp2 = _mm_blendv_ps(temp, v, cmp);

	// Second pass
	temp = _mm_shuffle_ps(temp2, temp2, _MM_SHUFFLE(2, 3, 0, 1));
	cmp = _mm_cmplt_ps(temp2, temp);

	cmp = _mm_shuffle_ps(cmp, cmp, _MM_SHUFFLE(2, 0, 2, 0));

	temp2 = _mm_blendv_ps(temp, temp2, cmp);

	// Third pass
	temp = _mm_shuffle_ps(temp2, temp2, _MM_SHUFFLE(3, 1, 2, 0));
	cmp = _mm_cmplt_ps(temp2, temp);

	cmp = _mm_shuffle_ps(cmp, cmp, _MM_SHUFFLE(3, 1, 1, 0));

	v = _mm_blendv_ps(temp, temp2, cmp);

	// Remove indices from input and return them
	auto ret = _mm_castps_si128(_mm_andnot_ps(mask, v));

	return ret;
}