流数据解析中高位地址转换的性能分析

在流数据（文件、网络）解析中，常常碰到低位高址与低位低址的转换问题，心血来潮，分析了一下各种转换方法的性能。

废话少说，直接上代码：

#include "windows.h"

inline DWORD SwapDWORD_0(DWORD in)
{
    return (in << 24) | ((in & 0x0000FF00) << 8) | ((in & 0x00FF0000) >> 8) | (in >> 24);
}

inline DWORD SwapDWORD_1(DWORD in)
{
    BYTE* pb = (BYTE*)(&in);
    BYTE temp;
    temp = pb[0]; pb[0] = pb[3]; pb[3] = temp;
    temp = pb[1]; pb[1] = pb[2]; pb[2] = temp;

    return in;
}

inline DWORD SwapDWORD_2(DWORD in)
{
    BYTE* pb = (BYTE*)(&in);
    BYTE temp;

    for(int i = 0, j = 3; i < j; ++i,--j){
        temp = pb[i]; pb[i] = pb[j]; pb[j] = temp;
    }

    return in;
}

inline DWORD SwapDWORD_3(DWORD in)
{
DWORD out;
BYTE* pin = (BYTE*)(&in);
BYTE* pout = (BYTE*)(&out);

pout[0] = pin[3];
pout[1] = pin[2];
pout[2] = pin[1];
pout[3] = pin[0];

return out;
}

inline DWORD SwapDWORD_4(DWORD in)
{
DWORD out;
BYTE* pin = (BYTE*)(&in);
BYTE* pout = (BYTE*)(&out);

for(int i = 0; i < 4; ++i){
pout[i] = pin[3-i];
}

return out;
}

///////////////////////////////////////////////////////////////////////////////////////
inline void SwapDWORD_R0(DWORD& in)
{
in = (in << 24) | ((in & 0x0000FF00) << 8) | ((in & 0x00FF0000) >> 8) | (in >> 24);
}

inline void SwapDWORD_R1(DWORD& in)
{
BYTE* pb = (BYTE*)(&in);
BYTE temp;
temp = pb[0]; pb[0] = pb[3]; pb[3] = temp;
temp = pb[1]; pb[1] = pb[2]; pb[2] = temp;
}

inline void SwapDWORD_R2(DWORD& in)
{
BYTE* pb = (BYTE*)(&in);
BYTE temp;

for(int i = 0, j = 3; i < j; ++i,--j){
temp = pb[i]; pb[i] = pb[j]; pb[j] = temp;
}
}

int _tmain(int argc, _TCHAR* argv[])
{
int loopCount = 10000000;
int i;

DWORD start, end;
DWORD dw;

dw = 0x01020304;
start = ::GetTickCount();
for(i = 0; i < loopCount; i ++){
dw = SwapDWORD_0(dw);
}
end = ::GetTickCount();
printf("SwapDWORD_0 takes: %d - 0x%08x\n", end - start, dw);

dw = 0x11223344;
start = ::GetTickCount();
for(i = 0; i < loopCount; i ++){
dw = SwapDWORD_1(dw);
}
end = ::GetTickCount();
printf("SwapDWORD_1 takes: %d - 0x%08x\n", end - start, dw);

dw = 0xaa223344;
start = ::GetTickCount();
for(i = 0; i < loopCount; i ++){
dw = SwapDWORD_2(dw);
}
end = ::GetTickCount();
printf("SwapDWORD_2 takes: %d - 0x%08x\n", end - start, dw);

dw = 0x1122bb44;
start = ::GetTickCount();
for(i = 0; i < loopCount; i ++){
dw = SwapDWORD_3(dw);
}
end = ::GetTickCount();
printf("SwapDWORD_3 takes: %d - 0x%08x\n", end - start, dw);

dw = 0xEE223344;
start = ::GetTickCount();
for(i = 0; i < loopCount; i ++){
dw = SwapDWORD_4(dw);
}
end = ::GetTickCount();
printf("SwapDWORD_4 takes: %d - 0x%08x\n", end - start, dw);

printf("----------------------------------------------\n");

start = ::GetTickCount();
for(i = 0; i < loopCount; i ++){
SwapDWORD_R0(dw);
}
end = ::GetTickCount();
printf("SwapDWORD_R0 takes: %d - 0x%08x\n", end - start, dw);

dw = 0x11223344;
start = ::GetTickCount();
for(i = 0; i < loopCount; i ++){
SwapDWORD_R1(dw);
}
end = ::GetTickCount();
printf("SwapDWORD_R1 takes: %d - 0x%08x\n", end - start, dw);

dw = 0xaa223344;
start = ::GetTickCount();
for(i = 0; i < loopCount; i ++){
SwapDWORD_R2(dw);
}
end = ::GetTickCount();
printf("SwapDWORD_R2 takes: %d - 0x%08x\n", end - start, dw);

return 0;
}

运行结果：

SwapDWORD_0 takes: 47ms - 0x01020304
SwapDWORD_1 takes: 218ms - 0x11223344
SwapDWORD_2 takes: 234ms - 0xaa223344
SwapDWORD_3 takes: 203ms - 0x1122bb44
SwapDWORD_4 takes: 203ms - 0xee223344
-----------------------------------
SwapDWORD_R0 takes: 62ms - 0xee223344
SwapDWORD_R1 takes: 16ms - 0x11223344
SwapDWORD_R2 takes: 47ms - 0xaa223344

结论：

对于原址转换，SwapDWORD_R1最快，对于转换到其他变量里的，位操作最快。