SSE命令示例代码（转换、加载、置位、存储）

1.   SSE的转换操作

// 测试SSE的转换操作
void TestSSEConvert()
{
	// 浮点数组
	__declspec(align(16)) float A[4] = { 5.35, 10.70, 16.05, 21.40 };
	__declspec(align(16)) float B[4] = { 5.75, 11.50, 17.25, -25.3 };

	__m128 *p = (__m128 *)A;
	__m128 *q = (__m128 *)B;
	__m128  a = p[0];
	__m128  b = q[0];
	printf("a: (%.2f, %.2f, %.2f, %.2f)\t b: (%.2f, %.2f, %.2f, %.2f)\n\n", 
		p->m128_f32[0], p->m128_f32[1], p->m128_f32[2], p->m128_f32[3], 
		q->m128_f32[0], q->m128_f32[1], q->m128_f32[2], q->m128_f32[3]);

	int v1 = _mm_cvtss_si32(a);
	int v2 = _mm_cvtss_si32(b);
	int v3 = _mm_cvttss_si32(a);
	int v4 = _mm_cvttss_si32(b);
	printf("(v1, v2, v3, v4): (%d, %d, %d, %d)\n\n", v1, v2, v3, v4);

	//__int64 v5 = _mm_cvtss_si64(a);
	//__int64 v6 = _mm_cvttss_si64(a);
	__m64 v5 = _mm_cvtps_pi32(a);
	__m64 v6 = _mm_cvttps_pi32(a);
	printf("v5: (%d, %d)\t v6: (%d, %d)\n\n", v5.m64_i32[0], v5.m64_i32[1], v6.m64_i32[0], v6.m64_i32[1]);

	//__m128 v7 = _mm_cvtsi64_ss(a, 123);
	__m128 v7 = _mm_cvtsi32_ss(a, 123);
	__m128 v8 = _mm_cvtpi32_ps(a, v5);
	printf("v7: (%.2f, %.2f, %.2f, %.2f)\n\n", v7.m128_f32[0], v7.m128_f32[1], v7.m128_f32[2], v7.m128_f32[3]);
	printf("v8: (%.2f, %.2f, %.2f, %.2f)\n\n", v8.m128_f32[0], v8.m128_f32[1], v8.m128_f32[2], v8.m128_f32[3]);

	__m64  v9  = _mm_cvtps_pi16(b);
	__m128 v10 = _mm_cvtpi16_ps(v9);
	__m128 v11 = _mm_cvtpu16_ps(v9);
	printf("v9: (%d, %d, %d, %d)\n\n", v9.m64_i16[0], v9.m64_i16[1], v9.m64_i16[2], v9.m64_i16[3]);
	printf("v10: (%.2f, %.2f, %.2f, %.2f)\n\n", v10.m128_f32[0], v10.m128_f32[1], v10.m128_f32[2], v10.m128_f32[3]);
	printf("v11: (%.2f, %.2f, %.2f, %.2f)\n\n", v11.m128_f32[0], v11.m128_f32[1], v11.m128_f32[2], v11.m128_f32[3]);

	__m64 temp;
	temp.m64_i8[0] = -1;
	temp.m64_i8[1] = 2;
	temp.m64_i8[2] = -3;
	temp.m64_i8[3] = 4;
	__m128 v12 = _mm_cvtpi8_ps(temp);
	__m128 v13 = _mm_cvtpu8_ps(temp);
	printf("v12: (%.2f, %.2f, %.2f, %.2f)\n\n", v12.m128_f32[0], v12.m128_f32[1], v12.m128_f32[2], v12.m128_f32[3]);
	printf("v13: (%.2f, %.2f, %.2f, %.2f)\n\n", v13.m128_f32[0], v13.m128_f32[1], v13.m128_f32[2], v13.m128_f32[3]);

	__m64 a1, a2;
	a1.m64_i32[0] = -120;
	a1.m64_i32[1] = 120;
	a2.m64_i32[0] = 256;
	a2.m64_i32[1] = -256;
	__m128 v14 = _mm_cvtpi32x2_ps(a1, a2);
	__m64  v15 = _mm_cvtps_pi16(v14);
	printf("v14: (%.2f, %.2f, %.2f, %.2f)\n\n", v14.m128_f32[0], v14.m128_f32[1], v14.m128_f32[2], v14.m128_f32[3]);
	printf("v15: (%d, %d, %d, %d)\n\n", v15.m64_i16[0], v15.m64_i16[1], v15.m64_i16[2], v15.m64_i16[3]);

	float f = _mm_cvtss_f32(a);
	printf("_mm_cvtss_f32(a) = %.2f\n\n", f);

	__m64 v16 = _mm_cvtps_pi8(v14);
	printf("v16: (%d, %d, %d, %d)\n\n", v16.m64_i8[0], v16.m64_i8[1], v16.m64_i8[2], v16.m64_i8[3]);
}

   测试结果:

   

   Remark:  由于一些奇怪的原因，上面出现 -1.#J 的地方是显示错误，分别纠正如下：

     1. v7   中的  -1.#J 实为 21.40

     2. v10  中的  -1.#J 实为 -25.00

     3. v12  中的  -1.#J 实为 4.00

     4. v14 中的 -1.#J 实为 -256.00

2.  SSE的加载操作

// 测试SSE加载操作
void TestSSELoad()
{
	// 浮点数组
	__declspec(align(16)) float A[2] = { 5.35, 10.70 };
	const __m64 *p = (const __m64 *)A;
	printf("(p0, p1): (%.2f, %.2f)\n\n", p->m64_f32[0], p->m64_f32[1]);

	__m128 a;
	a.m128_f32[0] = 80.45;
	a.m128_f32[1] = 85.55;
	a.m128_f32[2] = 90.65;
	a.m128_f32[3] = 95.75;

	__m128 v1 = _mm_loadh_pi(a, p);
	__m128 v2 = _mm_loadl_pi(a, p);
	printf("v1: (%.2f, %.2f, %.2f, %.2f)\n\n", v1.m128_f32[0], v1.m128_f32[1], v1.m128_f32[2], v1.m128_f32[3]);
	printf("v2: (%.2f, %.2f, %.2f, %.2f)\n\n", v2.m128_f32[0], v2.m128_f32[1], v2.m128_f32[2], v2.m128_f32[3]);

	const float f  = 100.85;
	const float *q = &f;
	__m128 v3 = _mm_load_ss(q);
	__m128 v4 = _mm_load1_ps(q);
	printf("v3: (%.2f, %.2f, %.2f, %.2f)\n\n", v3.m128_f32[0], v3.m128_f32[1], v3.m128_f32[2], v3.m128_f32[3]);
	printf("v4: (%.2f, %.2f, %.2f, %.2f)\n\n", v4.m128_f32[0], v4.m128_f32[1], v4.m128_f32[2], v4.m128_f32[3]);

	__declspec(align(16)) const float r[4] = {1.11, 2.22, 3.33, 4.44};
	__m128 v5 = _mm_loadu_ps(r);
	__m128 v6 = _mm_loadr_ps(r);		// 注意此处地址r必须16字节对齐
	printf("v5: (%.2f, %.2f, %.2f, %.2f)\n\n", v5.m128_f32[0], v5.m128_f32[1], v5.m128_f32[2], v5.m128_f32[3]);
	printf("v6: (%.2f, %.2f, %.2f, %.2f)\n\n", v6.m128_f32[0], v6.m128_f32[1], v6.m128_f32[2], v6.m128_f32[3]);
}

    测试结果:

   

3.   SSE的置位操作

// 测试SSE置位操作
void TestSSESet()
{
	__m128 v0 = _mm_setzero_ps();
	__m128 v1 = _mm_set_ss(3.55);
	__m128 v2 = _mm_set1_ps(3.55);
	__m128 v3 = _mm_set_ps(1.11, 2.22, 3.33, 4.44);
	__m128 v4 = _mm_setr_ps(1.11, 2.22, 3.33, 4.44);
	printf("v0: (%.2f, %.2f, %.2f, %.2f)\n\n", v0.m128_f32[0], v0.m128_f32[1], v0.m128_f32[2], v0.m128_f32[3]);
	printf("v1: (%.2f, %.2f, %.2f, %.2f)\n\n", v1.m128_f32[0], v1.m128_f32[1], v1.m128_f32[2], v1.m128_f32[3]);
	printf("v2: (%.2f, %.2f, %.2f, %.2f)\n\n", v2.m128_f32[0], v2.m128_f32[1], v2.m128_f32[2], v2.m128_f32[3]);
	printf("v3: (%.2f, %.2f, %.2f, %.2f)\n\n", v3.m128_f32[0], v3.m128_f32[1], v3.m128_f32[2], v3.m128_f32[3]);
	printf("v4: (%.2f, %.2f, %.2f, %.2f)\n\n", v4.m128_f32[0], v4.m128_f32[1], v4.m128_f32[2], v4.m128_f32[3]);
}

测试结果:

 

4.   SSE的存储操作

// 测试SSE存储操作
void TestSSEStore()
{
	__m128 a;
	a.m128_f32[0] = 80.45;
	a.m128_f32[1] = 85.55;
	a.m128_f32[2] = 90.65;
	a.m128_f32[3] = 95.75;

	__declspec(align(16)) float A[2] = { 0 };
	__declspec(align(16)) float B[2] = { 0 };
	__m64 *p = (__m64 *)A;
	__m64 *q = (__m64 *)B;
	_mm_storeh_pi(p, a);
	_mm_storel_pi(q, a);
	printf("(p0, p1): (%.2f, %.2f)\n\n", p->m64_f32[0], p->m64_f32[1]);
	printf("(q0, q1): (%.2f, %.2f)\n\n", q->m64_f32[0], q->m64_f32[1]);

	__declspec(align(16)) float f[4] = { 0 };
	_mm_store_ss(&f[0], a);
	printf("f[0]: %.2f\n\n", f[0]);

	_mm_store1_ps(f, a);
	printf("f: (%.2f, %.2f, %.2f, %.2f)\n\n", f[0], f[1], f[2], f[3]);

	_mm_store_ps(f, a);
	printf("f: (%.2f, %.2f, %.2f, %.2f)\n\n", f[0], f[1], f[2], f[3]);

	float g[4] = { 0 };
	_mm_storeu_ps(g, a);
	printf("g: (%.2f, %.2f, %.2f, %.2f)\n\n", f[0], f[1], f[2], f[3]);

	_mm_storer_ps(f, a);
	printf("f: (%.2f, %.2f, %.2f, %.2f)\n\n", f[0], f[1], f[2], f[3]);
}

   测试结果: