Slow SSE code for vector-on-axis test

IsAxial.cpp

#include <stdio.h>
#include <tchar.h>
#include <emmintrin.h>
#include <xmmintrin.h>
#include <smmintrin.h>
#include <iostream>
#include <chrono>

__declspec(noinline)
bool is_axial(__m128 vec)
{
	auto r = _mm_castps_si128(_mm_cmpeq_ps(vec, _mm_setzero_ps()));
	auto r1 = _mm_xor_si128(r, _mm_setr_epi32(0, -1, -1, -1));
	auto r2 = _mm_xor_si128(r, _mm_setr_epi32(-1, 0, -1, -1));
	auto r3 = _mm_xor_si128(r, _mm_setr_epi32(-1, -1, 0, -1));
	auto m = _mm_setr_epi32(_mm_movemask_epi8(r1), _mm_movemask_epi8(r2), _mm_movemask_epi8(r3), -1);
	auto n = _mm_cmpeq_epi32(m, _mm_setzero_si128());
	return !_mm_testz_si128(n, n);
}

//__declspec(noinline)
bool is_axial2(__m128 vec)
{
	auto r = _mm_castps_si128(_mm_cmpeq_ps(vec, _mm_setzero_ps()));
	r = _mm_hadd_epi32(r, r);
	r = _mm_hadd_epi32(r, r);
	return _mm_cvtsi128_si32(r) + 3 == 0;
}

//__declspec(noinline)
bool is_axial6(float x, float y, float z)
{
	auto r = _mm_castps_si128(_mm_cmpeq_ps(_mm_setr_ps(x, y, z, 0.f), _mm_setzero_ps()));
	r = _mm_hadd_epi32(r, r);
	r = _mm_hadd_epi32(r, r);
	return _mm_cvtsi128_si32(r) + 3 == 0;
}

//__declspec(noinline)
bool is_axial3(float x, float y, float z)
{
	return ((x != 0.0f) + (y != 0.0f) + (z != 0.0f)) == 1;
}

//__declspec(noinline)
bool is_axial4(float x, float y, float z)
{
	return x == 0.f && y == 0.f && z != 0.f ||
		   x == 0.f && y != 0.f && z == 0.f ||
		   x != 0.f && y == 0.f && z == 0.f;
}

//__declspec(noinline)
bool is_axial5(float x, float y, float z)
{
	return x == 0.f && (y == 0.f && z != 0.f  ||
						y != 0.f && z == 0.f) ||
		   x != 0.f && y == 0.f && z == 0.f;
}

bool is_axial7(float x, float y, float z)
{
	if (x == 0.f)
		if (y == 0.f)
			return z != 0.f;
		else
			return z == 0.f;
	return y == 0.f && z == 0.f;
}

using namespace std;

int main()
{
	int temp = 0;
	auto start_time = chrono::high_resolution_clock::now();
	for (int i = -500; i < 500; i++)
	for (int j = -500; j < 500; j++)
	for (int k = -500; k < 500; k++)
		//temp += is_axial2(_mm_setr_ps(i & 1, j & 1, k & 1, 0.f));
		temp += is_axial7(i & 1, j & 1, k & 1);
	auto end_time = chrono::high_resolution_clock::now();
	cout << chrono::duration_cast<chrono::milliseconds>(end_time - start_time).count() << endl;
	cout << temp;

	return temp;

	printf("%d\n", is_axial2(_mm_setr_ps(0.f, 0.f, 1.f, 0.f)));
	printf("%d\n", is_axial2(_mm_setr_ps(1.f, 1.f, 0.f, 0.f)));
	printf("%d\n", is_axial2(_mm_setr_ps(1.f, 1.f, 1.f, 0.f)));
	printf("%d\n", is_axial2(_mm_setr_ps(0.f, 1.f, 0.f, 0.f)));
	printf("%d\n", is_axial2(_mm_setr_ps(1.f, 0.f, 0.f, 0.f)));
	printf("%d\n", is_axial2(_mm_setr_ps(0.f, 0.f, 0.f, 0.f)));
}