strcat的几种实现及性能比较

 

一  原型说明

     strcat()为C语言标准库函数，用于字符串拼接。函数原型声明在string.h头文件中：

char *strcat(char *dest, const char *src);

     该函数将参数src所指字符串拷贝到参数dest所指字符串的结尾处(覆盖dest结尾处的'\0')并添加'\0'。其返回值为参数dest所指字符串的起始地址。注意，dest必须有足够的空间来容纳要拷贝的src字符串。

     本文将给出strcat函数的几种实现，并比较其执行效率。代码运行环境如下：

 

二  代码实现

2.1 汇编实现

char *AssemStrcat(char *pszDest, const char *pszSrc)
{
    int d0, d1, d2, d3;
    __asm__ __volatile__(
        "repne\n\t"
        "scasb\n\t"
        "decl %1\n"
        "1:\tlodsb\n\t"
        "stosb\n\t"
        "testb %%al,%%al\n\t"
        "jne 1b"
        : "=&S" (d0), "=&D" (d1), "=&a" (d2), "=&c" (d3)
        : "0" (pszSrc), "1" (pszDest), "2" (0), "3" (0xffffffff):"memory");
    return pszDest;
}

2.2 模拟C库

char *SimuStrcat(char *pszDest, const char *pszSrc)
{
    char *pszOrigDst = pszDest;

    while(*pszDest)
        pszDest++;
    while((*pszDest++ = *pszSrc++) != '\0')
        ;

    return pszOrigDst;
}

     因strcat的C标准库函数的实现方式未知，故使用SimuStrcat函数模拟标准库实现。

2.3 快速拼接

     由strcat函数的原型说明可知，每次调用时都必须扫描整个目的字符串(以寻找结束符)。这会降低该函数的执行效率，尤其是在频繁拼接时。

     FastStrcat函数每次返回拼接后的字符串尾部，即指向结束符所在位置。下次调用时便不再需要扫描整串，从而提高执行效率。

char *FastStrcat(char *pszDest, const char* pszSrc)
{
    while(*pszDest)
        pszDest++;
    while((*pszDest++ = *pszSrc++));
    return --pszDest;
}

     注意，第二个while循环若不加双层括号则会报”suggest parentheses around assignment used as truth value”的警告。因返回时对pszDest作减法运算，故单次执行时FastStrcat慢于SimuStrcat。

     为提高执行速度，本节函数实现均未对指针参数做判空处理。若兼求安全性，可在函数内使用assert断言指针合法性。Gcc编译器还对函数声明提供一种nonnull扩展属性，可在编译时进行有限的判空保护：

char *FastStrcat(char *pszDest, const char* pszSrc)__attribute__((__nonnull__(1, 2)));
char *NullPtr(void){return NULL;}
int main(void)
{
    char *pszBuf = NULL;
    FastStrcat(pszBuf, "Elizabeth ");
    FastStrcat(NullPtr(), "Elizabeth ");
 8     FastStrcat(NullPtr()?pszBuf:NULL, "Elizabeth ");
    FastStrcat(NULL, "Elizabeth ");
    return 0;
}

     其中，__nonnull__(1, 2)指示编译器对FastStrcat函数调用的第一个和第二个参数判空。

     编译结果如下：

[wangxiaoyuan_@localhost test1]$ gcc -Wall -o test test.c
test.c: In function 'main':
test.c:8: warning: null argument where non-null required (argument 1)
test.c:9: warning: null argument where non-null required (argument 1)

     可见，除非将指针参数显式地置空，否则nonnull机制也检测不到。此外，使用nonnull机制时，若打开编译优化选项-O2，则函数内关于指针参数的校验将被优化掉。

 

三  性能比较

     本节采用《Linux用户态程序计时方式详解》一文的TIME_ELAPSED()宏进行程序计时。

#define TIME_ELAPSED(codeToTime) do{ \
    struct timeval beginTime, endTime; \
    gettimeofday(&beginTime, NULL); \
    {codeToTime;} \
    gettimeofday(&endTime, NULL); \
    long secTime  = endTime.tv_sec - beginTime.tv_sec; \
    long usecTime = endTime.tv_usec - beginTime.tv_usec; \
    printf("[%s(%d)]Elapsed Time: SecTime = %lds, UsecTime = %ldus!\n", __FILE__, __LINE__, secTime, usecTime); \
}while(0)

     基于该宏，编写测试代码如下：

#ifdef strcat //检查标准库strcat是否用宏实现
    #warning strcat has been defined!
#endif 
int main(void)
{
    char szCatBuf[1000];
    szCatBuf[0] = '\0'; //字符串快速初始化

    char *pszBuf = szCatBuf;
    TIME_ELAPSED(
        pszBuf = FastStrcat(pszBuf, "Abraham, ");
        pszBuf = FastStrcat(pszBuf, "Alexander, ");
        pszBuf = FastStrcat(pszBuf, "Maximilian, ");
        pszBuf = FastStrcat(pszBuf, "Valentine ")
    );
    printf("  [FastStrcat]szCatBuf = %s\n", szCatBuf);

    szCatBuf[0] = '\0';
    TIME_ELAPSED(
        SimuStrcat(szCatBuf, "Abraham, ");
        SimuStrcat(szCatBuf, "Alexander, ");
        SimuStrcat(szCatBuf, "Maximilian, ");
        SimuStrcat(szCatBuf, "Valentine ")
    );
    printf("  [SimuStrcat]szCatBuf = %s\n", szCatBuf);

    szCatBuf[0] = '\0';
    TIME_ELAPSED(
        AssemStrcat(szCatBuf, "Abraham, ");
        AssemStrcat(szCatBuf, "Alexander, ");
        AssemStrcat(szCatBuf, "Maximilian, ");
        AssemStrcat(szCatBuf, "Valentine ")
    );
    printf("  [AssemStrcat]szCatBuf = %s\n", szCatBuf);

    szCatBuf[0] = '\0';
    TIME_ELAPSED(
        strcat(szCatBuf, "Abraham, ");
        strcat(szCatBuf, "Alexander, ");
        strcat(szCatBuf, "Maximilian, ");
        strcat(szCatBuf, "Valentine ")
    );
    printf("  [LibStrcat]szCatBuf = %s\n", szCatBuf);

    szCatBuf[0] = '\0';
    TIME_ELAPSED(
        sprintf(szCatBuf, "%s%s%s%s","Abraham, ", "Alexander, ", "Maximilian, ", "Valentine ")
    );
    printf("  [LibSprintf]szCatBuf = %s\n", szCatBuf);

    return 0;
}

     除上节三种strcat实现外，代码还对齐库实现及sprintf方式进行测试。测试结果如下：

[wangxiaoyuan_@localhost test1]$ gcc -Wall -o test test.c
[wangxiaoyuan_@localhost test1]$ ./test
[test.c(15)]Elapsed Time: SecTime = 0s, UsecTime = 2us!
  [FastStrcat]szCatBuf = Abraham, Alexander, Maximilian, Valentine 
[test.c(24)]Elapsed Time: SecTime = 0s, UsecTime = 2us!
  [SimuStrcat]szCatBuf = Abraham, Alexander, Maximilian, Valentine 
[test.c(33)]Elapsed Time: SecTime = 0s, UsecTime = 1us!
  [AssemStrcat]szCatBuf = Abraham, Alexander, Maximilian, Valentine 
[test.c(42)]Elapsed Time: SecTime = 0s, UsecTime = 1us!
  [LibStrcat]szCatBuf = Abraham, Alexander, Maximilian, Valentine 
[test.c(48)]Elapsed Time: SecTime = 0s, UsecTime = 6us!
  [LibSprintf]szCatBuf = Abraham, Alexander, Maximilian, Valentine

     因每次测试仅调用一次待测函数，故计时不太精准。对比另一次测试结果：

[wangxiaoyuan_@localhost test1]$ ./test
[test.c(15)]Elapsed Time: SecTime = 0s, UsecTime = 4us!
  [FastStrcat]szCatBuf = Abraham, Alexander, Maximilian, Valentine 
[test.c(24)]Elapsed Time: SecTime = 0s, UsecTime = 4us!
  [SimuStrcat]szCatBuf = Abraham, Alexander, Maximilian, Valentine 
[test.c(33)]Elapsed Time: SecTime = 0s, UsecTime = 3us!
  [AssemStrcat]szCatBuf = Abraham, Alexander, Maximilian, Valentine 
[test.c(42)]Elapsed Time: SecTime = 0s, UsecTime = 2us!
  [LibStrcat]szCatBuf = Abraham, Alexander, Maximilian, Valentine 
[test.c(48)]Elapsed Time: SecTime = 0s, UsecTime = 15us!
  [LibSprintf]szCatBuf = Abraham, Alexander, Maximilian, Valentine

     可见，单次执行时，strcat库函数最快，FastStrcat居中，sprintf库函数最慢。

     接着，比较批量执行时strcat库函数和FastStrcat的速度。测试代码如下：

int main(void)
{
#define ROUND    (unsigned short)1000
    char szCatBuf[sizeof("Elizabeth ")*ROUND];
    szCatBuf[0] = '\0';

    char *pszBuf = szCatBuf;
    TIME_ELAPSED(
        unsigned short wIdx = 0;
        for(; wIdx < ROUND; wIdx++)
            pszBuf = FastStrcat(pszBuf, "Elizabeth ");
    );

    szCatBuf[0] = '\0';
    TIME_ELAPSED(
        unsigned short wIdx = 0;
        for(; wIdx < ROUND; wIdx++)
            strcat(szCatBuf, "Elizabeth ");
    );

    return 0;
}

     测试结果如下：

[wangxiaoyuan_@localhost test1]$ gcc -Wall -o test test.c
[wangxiaoyuan_@localhost test1]$ ./test
[test.c(12)]Elapsed Time: SecTime = 0s, UsecTime = 99us!
[test.c(19)]Elapsed Time: SecTime = 0s, UsecTime = 3834us!

     可见，批量执行时，FastStrcat远快于strcat库函数。

 

四  总结

     单次拼接时，strcat库函数最快，FastStrcat居中，sprintf库函数最慢(若非格式化需要不建议用于字符串拼接)。

     频繁拼接时，FastStrcat相比strcat库函数具有明显的速度优势。