SSE 加速运算例子详解：乘法、加法、平方、最小值、最大值、与操作

版权声明：本文为博主原创文章，未经博主允许不得转载。欢迎交流，QQ：896922782，微信：15058133936 https://blog.csdn.net/zhubenfulovepoem/article/details/51204862

SSE(Streaming SIMD Extensions)是英特尔在AMD的3D Now!发布一年之后，在其计算机芯片Pentium III中引入的指令集，是MMX的超集。AMD后来在Athlon XP中加入了对这个指令集的支持。这个指令集增加了对8个128位寄存器XMM0-XMM7的支持，每个寄存器可以存储4个单精度浮点数。使用这些寄存器的程序必须使用FXSAVE和FXRSTR指令来保持和恢复状态。但是在Pentium III对SSE的实现中，浮点数寄存器又一次被新的指令集占用了，但是这一次切换运算模式不是必要的了，只是SSE和浮点数指令不能同时进入CPU的处理线而已。

库文件说明

#ifndef __METHOD
#define __METHOD


void ScaleValue1(float *pArray, DWORD dwCount, float fScale);//乘法
void ScaleValue2(float *pArray, DWORD dwCount, float fScale);

void Add1(float *pArray, DWORD dwCount, float fScale);//加法
void Add2(float *pArray, DWORD dwCount, float fScale);

void Sqrt1(float *pArray, DWORD dwCount, float fScale);//平方
void Sqrt2(float *pArray, DWORD dwCount, float fScale);

void Min1(float *pArray, DWORD dwCount, float fScale);//最小值
void Min2(float *pArray, DWORD dwCount, float fScale);//最小值

void Max1(float *pArray, DWORD dwCount, float fScale);//最小值
void Max2(float *pArray, DWORD dwCount, float fScale);//最小值

void And1(float *pArray, DWORD dwCount, float fScale);//与操作
void And2(float *pArray, DWORD dwCount, float fScale);//与操作

#endif

#include <xmmintrin.h>
#include <Windows.h>
#include <math.h>

void ScaleValue1(float *pArray, DWORD dwCount, float fScale)//乘法
{
    DWORD dwGroupCount = dwCount/4;
    __m128 e_Scale = _mm_set_ps1(fScale);//设置所有4个值为同一值

    for (DWORD i=0; i<dwGroupCount; i++)
    {
        *(__m128*)(pArray + i*4) = _mm_mul_ps( *(__m128*)(pArray + i*4),e_Scale);
    }
}

void ScaleValue2(float *pArray, DWORD dwCount, float fScale)
{
    for (DWORD i =0; i<dwCount; i++)
    {
        pArray[i] *= fScale;
    }
}

void Add1(float *pArray, DWORD dwCount, float fScale)//加法
{
    DWORD dwGroupCount = dwCount/4;
    __m128 e_Scale = _mm_set_ps1(fScale);//设置所有4个值为同一值

    for (DWORD i=0; i<dwGroupCount; i++)
    {
        *(__m128*)(pArray + i*4) = _mm_add_ps( *(__m128*)(pArray + i*4),e_Scale);
    }
}

void Add2(float *pArray, DWORD dwCount, float fScale)
{
    for (DWORD i =0; i<dwCount; i++)
    {
        pArray[i] += fScale;
    }
}

void Sqrt1(float *pArray, DWORD dwCount, float fScale)//平方
{
    DWORD dwGroupCount = dwCount/4;
    __m128 e_Scale = _mm_set_ps1(fScale);//设置所有4个值为同一值

    for (DWORD i=0; i<dwGroupCount; i++)
    {
        *(__m128*)(pArray + i*4) = _mm_sqrt_ps(e_Scale);
    }
}

void Sqrt2(float *pArray, DWORD dwCount, float fScale)
{
    for (DWORD i =0; i<dwCount; i++)
    {
        pArray[i] = sqrt(fScale);
    }
}

void Min1(float *pArray, DWORD dwCount, float fScale)//最小值
{
    DWORD dwGroupCount = dwCount/4;
    __m128 e_Scale = _mm_set_ps1(fScale);//设置所有4个值为同一值

    for (DWORD i=0; i<dwGroupCount; i++)
    {
        *(__m128*)(pArray + i*4) = _mm_min_ps( *(__m128*)(pArray + i*4),e_Scale);
    }
}

void Min2(float *pArray, DWORD dwCount, float fScale)
{
    for (DWORD i =0; i<dwCount; i++)
    {
        pArray[i] = (pArray[i]>fScale? fScale : pArray[i]);
    }
}

void Max1(float *pArray, DWORD dwCount, float fScale)//最大值
{
    DWORD dwGroupCount = dwCount/4;
    __m128 e_Scale = _mm_set_ps1(fScale);//设置所有4个值为同一值

    for (DWORD i=0; i<dwGroupCount; i++)
    {
        *(__m128*)(pArray + i*4) = _mm_max_ps( *(__m128*)(pArray + i*4),e_Scale);
    }
}

void Max2(float *pArray, DWORD dwCount, float fScale)
{
    for (DWORD i =0; i<dwCount; i++)
    {
        pArray[i] = (pArray[i]<fScale? fScale : pArray[i]);
    }
}

void And1(float *pArray, DWORD dwCount, float fScale)//与操作
{
    DWORD dwGroupCount = dwCount/4;
    __m128 e_Scale = _mm_set_ps1(fScale);//设置所有4个值为同一值

    for (DWORD i=0; i<dwGroupCount; i++)
    {
        *(__m128*)(pArray + i*4) = _mm_and_ps( *(__m128*)(pArray + i*4),e_Scale);
    }
}

void And2(float *pArray, DWORD dwCount, float fScale)
{
    for (DWORD i =0; i<dwCount; i++)
    {
        pArray[i] = (int)(pArray[i]) & (int)(fScale);
    }
}

采用SSE和不采用SSE的数学计算操作速度对比：

#include <xmmintrin.h>
#include <Windows.h>
#include <iostream>

#include "Method.h"

using namespace std;

#define ARRAYCOUNT 1000

#define COUNTSIZE 10000

class CTimer
{
public:
    __forceinline CTimer(void)
    {
        QueryPerformanceFrequency(&m_Frequency);// 获取时钟周期
        QueryPerformanceCounter(&m_StartCount);// 获取时钟计数
    }

    __forceinline void Reset(void)
    {
        QueryPerformanceCounter(&m_StartCount);
    }

    __forceinline double End(void)
    {
        QueryPerformanceCounter(&m_EndCount);
        return ( m_EndCount.QuadPart - m_StartCount.QuadPart )*1000/m_Frequency.QuadPart;
    }


private:
    LARGE_INTEGER m_Frequency;
    LARGE_INTEGER m_StartCount;
    LARGE_INTEGER m_EndCount;
};



int __cdecl main()
{
    float __declspec(align(16))Array[ARRAYCOUNT];
    //__declspec(align(16))做为数组定义的修释符，这表示该数组是以16字节为边界对齐的，
    //因为SSE指令只能支持这种格式的内存数据
    memset(Array, 0, sizeof(float)*ARRAYCOUNT);
    CTimer t;
    double dTime;

    //乘法
    cout<<"乘法:"<<endl;
    t.Reset();
    for (int i=0; i<COUNTSIZE; i++)
    {
        ScaleValue1(Array, ARRAYCOUNT, 1000.0f);
    }
    dTime = t.End();

    cout<<"Use SSE: "<<dTime<<"毫秒"<<endl;


    t.Reset();
    for (int i=0; i<COUNTSIZE; i++)
    {
        ScaleValue2(Array, ARRAYCOUNT, 1000.0f);
    }
    dTime = t.End();

    cout<<"Not Use SSE: "<<dTime<<"毫秒"<<endl;


//加法
    cout<<"加法："<<endl;
    t.Reset();
    for (int i=0; i<COUNTSIZE; i++)
    {
        Add1(Array, ARRAYCOUNT, 1000.0f);
    }
    dTime = t.End();

    cout<<"Use SSE: "<<dTime<<"毫秒"<<endl;


    t.Reset();
    for (int i=0; i<COUNTSIZE; i++)
    {
        Add2(Array, ARRAYCOUNT, 1000.0f);
    }
    dTime = t.End();

    cout<<"Not Use SSE: "<<dTime<<"毫秒"<<endl;


    //平方
    cout<<"平方："<<endl;
    t.Reset();
    for (int i=0; i<COUNTSIZE; i++)
    {
        Sqrt1(Array, ARRAYCOUNT, 1000.0f);
    }
    dTime = t.End();

    cout<<"Use SSE: "<<dTime<<"毫秒"<<endl;

    t.Reset();
    for (int i=0; i<COUNTSIZE; i++)
    {
        Sqrt2(Array, ARRAYCOUNT, 1000.0f);
    }
    dTime = t.End();

    cout<<"Not Use SSE: "<<dTime<<"毫秒"<<endl;


    //最小值
    cout<<"最小值："<<endl;
    t.Reset();
    for (int i=0; i<COUNTSIZE; i++)
    {
        Min1(Array, ARRAYCOUNT, 1000.0f);
    }
    dTime = t.End();

    cout<<"Use SSE: "<<dTime<<"毫秒"<<endl;

    t.Reset();
    for (int i=0; i<COUNTSIZE; i++)
    {
        Min2(Array, ARRAYCOUNT, 1000.0f);
    }
    dTime = t.End();

    cout<<"Not Use SSE: "<<dTime<<"毫秒"<<endl;


    //最大值
    cout<<"最大值："<<endl;
    t.Reset();
    for (int i=0; i<COUNTSIZE; i++)
    {
        Max1(Array, ARRAYCOUNT, 1000.0f);
    }
    dTime = t.End();

    cout<<"Use SSE: "<<dTime<<"毫秒"<<endl;

    t.Reset();
    for (int i=0; i<COUNTSIZE; i++)
    {
        Max2(Array, ARRAYCOUNT, 1000.0f);
    }
    dTime = t.End();

    cout<<"Not Use SSE: "<<dTime<<"毫秒"<<endl;


    //与操作
    cout<<"与操作："<<endl;
    t.Reset();
    for (int i=0; i<COUNTSIZE; i++)
    {
        And1(Array, ARRAYCOUNT, 1000.0f);
    }
    dTime = t.End();

    cout<<"Use SSE: "<<dTime<<"毫秒"<<endl;

    t.Reset();
    for (int i=0; i<COUNTSIZE; i++)
    {
        And2(Array, ARRAYCOUNT, 1000.0f);
    }
    dTime = t.End();

    cout<<"Not Use SSE: "<<dTime<<"毫秒"<<endl;

    system("pause");

    return 0;

}