windows下sse性能对比

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 __METHODvoid 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
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23

#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);    }}
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117

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 10000class 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;}